merge from trunkglxmisc-3-0-0-20000607

208 files changed, 35610 insertions, 3745 deletions

diff --git a/xc/config/cf/FreeBSD.cf b/xc/config/cf/FreeBSD.cf
index 1043ed8bf..c92902d37 100644
--- a/xc/config/cf/FreeBSD.cf
+++ b/xc/config/cf/FreeBSD.cf
@@ -82,7 +82,7 @@ XCOMM operating system:  OSName (OSMajorVersion./**/OSMinorVersion./**/OSTeenyVe
 #if defined(UseInstalled)
 #define DefaultCCOptions	/**/
 #else
-#define DefaultCCOptions	-ansi -pedantic -Dasm=__asm GccWarningOptions
+/*#define DefaultCCOptions	-ansi -pedantic -Dasm=__asm GccWarningOptions*/
 #endif
 #ifndef ExtraLibraries
 /* support for multi-byte locales is in libxpg4 rather than libc */
@@ -464,7 +464,7 @@ install.man::								@@\
  * A hack to work around an optimisation problem with the compiler on
  * FreeBSD 4.0-current in late 1999/early 2000.
  */
-#if OSMajorVersion == 4 && OSMinorVersion == 0 && OSTeenyVersion == 0
+#if (OSMajorVersion == 4 || OSMajorVersion == 5) && OSMinorVersion == 0 && OSTeenyVersion == 0
 #define FreeBSDGccOptBug
 #endif
 
diff --git a/xc/config/cf/host.def b/xc/config/cf/host.def
index 8b3406a77..8b164f632 100644
--- a/xc/config/cf/host.def
+++ b/xc/config/cf/host.def
@@ -1,8 +1,8 @@
-#define ProjectRoot /usr/XF86-glxmisc
+
 #define DefaultGcc2i386Opt -O2
 #define LibraryCDebugFlags -O2
 #define BuildServersOnly YES
-#define XF86CardDrivers vga tdfx i810 mga glint r128
+#define XF86CardDrivers vga tdfx i810 mga r128
 #define LinuxDistribution LinuxRedHat
 #define GccWarningOptions -Wall -Wpointer-arith -Wstrict-prototypes \
                           -Wmissing-prototypes -Wmissing-declarations \
@@ -17,9 +17,16 @@
 /* #define GlxBuiltInTdfx YES */
 /* #define GlxBuiltInI810 YES */
 /* #define GlxBuiltInMga YES */
+/* #define GlxBuiltInR128 YES */
 /* #define DoLoadableServer NO */
-#define SharedLibFont NO
 
+/* Optionally turn this on to change the place where you install the build */
+/* #define ProjectRoot /usr/XF86-main */
+
+/* Optionally turn this on to force the kernel modules to build */
+/* #define BuildXF86DRM YES */
+
+#define SharedLibFont NO
 #define XnestServer NO
 #define XVirtualFramebufferServer NO
 #define XprtServer NO
diff --git a/xc/config/cf/xfree86.cf b/xc/config/cf/xfree86.cf
index 77e5e454d..5f5991c77 100644
--- a/xc/config/cf/xfree86.cf
+++ b/xc/config/cf/xfree86.cf
@@ -512,12 +512,21 @@ IPLAN2P8_DEFS = -DUSE_IPLAN2P8
 #  ifndef GlxBuiltInTdfx
 #    define GlxBuiltInTdfx         NO
 #  endif
+#  ifndef GlxBuiltInMga
+#    define GlxBuiltInMga          NO
+#  endif
+#  ifndef GlxBuiltInI810
+#    define GlxBuiltInI810         NO
+#  endif
+#  ifndef GlxBuiltInR128
+#    define GlxBuiltInR128         NO
+#  endif
 
-#  if GlxBuiltInTdfx
+#  if GlxBuiltInTdfx || GlxBuiltInMga || GlxBuiltInI810 || GlxBuiltInR128
 #    define GlxDriverUsesMesa      YES
 #  endif
 
-#  if GlxBuiltInGamma || GlxBuiltInMesa || GlxBuiltInTdfx
+#  if GlxBuiltInGamma || GlxBuiltInTdfx || GlxBuiltInMga || GlxBuiltInI810 || GlxBuiltInR128 || GlxBuiltInMesa
 #    define GlxUseBuiltInDRIDriver YES
 #    define DRIDynLoadDefines /**/
 #  else
diff --git a/xc/include/GL/glx.h b/xc/include/GL/glx.h
index 367ce0a51..e4712b95f 100644
--- a/xc/include/GL/glx.h
+++ b/xc/include/GL/glx.h
@@ -1,26 +1,38 @@
 #ifndef __GLX_glx_h__
 #define __GLX_glx_h__
 
-/* $XFree86: xc/include/GL/glx.h,v 1.5 2000/03/02 16:07:29 martin Exp $ */
 /*
-** The contents of this file are subject to the GLX Public License Version 1.0
-** (the "License"). You may not use this file except in compliance with the
-** License. You may obtain a copy of the License at Silicon Graphics, Inc.,
-** attn: Legal Services, 2011 N. Shoreline Blvd., Mountain View, CA 94043
-** or at http://www.sgi.com/software/opensource/glx/license.html.
-**
-** Software distributed under the License is distributed on an "AS IS"
-** basis. ALL WARRANTIES ARE DISCLAIMED, INCLUDING, WITHOUT LIMITATION, ANY
-** IMPLIED WARRANTIES OF MERCHANTABILITY, OF FITNESS FOR A PARTICULAR
-** PURPOSE OR OF NON- INFRINGEMENT. See the License for the specific
-** language governing rights and limitations under the License.
-**
-** The Original Software is GLX version 1.2 source code, released February,
-** 1999. The developer of the Original Software is Silicon Graphics, Inc.
-** Those portions of the Subject Software created by Silicon Graphics, Inc.
-** are Copyright (c) 1991-9 Silicon Graphics, Inc. All Rights Reserved.
-**
-** $SGI$
+** License Applicability. Except to the extent portions of this file are
+** made subject to an alternative license as permitted in the SGI Free
+** Software License B, Version 1.0 (the "License"), the contents of this
+** file are subject only to the provisions of the License. You may not use
+** this file except in compliance with the License. You may obtain a copy
+** of the License at Silicon Graphics, Inc., attn: Legal Services, 1600
+** Amphitheatre Parkway, Mountain View, CA 94043-1351, or at:
+** 
+** http://oss.sgi.com/projects/FreeB
+** 
+** Note that, as provided in the License, the Software is distributed on an
+** "AS IS" basis, with ALL EXPRESS AND IMPLIED WARRANTIES AND CONDITIONS
+** DISCLAIMED, INCLUDING, WITHOUT LIMITATION, ANY IMPLIED WARRANTIES AND
+** CONDITIONS OF MERCHANTABILITY, SATISFACTORY QUALITY, FITNESS FOR A
+** PARTICULAR PURPOSE, AND NON-INFRINGEMENT.
+** 
+** Original Code. The Original Code is: OpenGL Sample Implementation,
+** Version 1.2.1, released January 26, 2000, developed by Silicon Graphics,
+** Inc. The Original Code is Copyright (c) 1991-2000 Silicon Graphics, Inc.
+** Copyright in any portions created by third parties is as indicated
+** elsewhere herein. All Rights Reserved.
+** 
+** Additional Notice Provisions: The application programming interfaces
+** established by SGI in conjunction with the Original Code are The
+** OpenGL(R) Graphics System: A Specification (Version 1.2.1), released
+** April 1, 1999; The OpenGL(R) Graphics System Utility Library (Version
+** 1.3), released November 4, 1998; and OpenGL(R) Graphics with the X
+** Window System(R) (Version 1.3), released October 19, 1998. This software
+** was created using the OpenGL(R) version 1.2.1 Sample Implementation
+** published by SGI, but has not been independently verified as being
+** compliant with the OpenGL(R) version 1.2.1 Specification.
 */
 
 #include <X11/Xlib.h>
@@ -39,19 +51,19 @@ extern "C" {
 typedef XID GLXContextID;
 typedef XID GLXPixmap;
 typedef XID GLXDrawable;
-/* GLX 1.3 */
-typedef XID GLXFBConfigID;
-typedef XID GLXPfuffer;
-typedef XID GLXWindow;
 typedef XID GLXPbuffer;
-typedef XID GLXFBConfig;
-
+typedef XID GLXWindow;
+typedef XID GLXFBConfigID;
 
 /*
 ** GLXContext is a pointer to opaque data.
-**/
+*/
 typedef struct __GLXcontextRec *GLXContext;
 
+/*
+** GLXFBConfig is a pointer to opaque data.
+*/
+typedef struct __GLXFBConfigRec *GLXFBConfig;
 
 /************************************************************************/
 
@@ -76,34 +88,57 @@ extern const char * glXGetClientString (Display *dpy, int name );
 extern const char * glXQueryServerString (Display *dpy, int screen, int name );
 extern const char * glXQueryExtensionsString (Display *dpy, int screen );
 
-/* GLX 1.3 */
-extern GLXFBConfig glXChooseFBConfig (Display *dpy, int screen, const int *attribList, int *nitems);
+/* New for GLX 1.3 */
+extern GLXFBConfig * glXGetFBConfigs (Display *dpy, int screen, int *nelements);
+extern GLXFBConfig * glXChooseFBConfig (Display *dpy, int screen, const int *attrib_list, int *nelements);
 extern int glXGetFBConfigAttrib (Display *dpy, GLXFBConfig config, int attribute, int *value);
 extern XVisualInfo * glXGetVisualFromFBConfig (Display *dpy, GLXFBConfig config);
-extern GLXWindow glXCreateWindow (Display *dpy, GLXFBConfig config, Window win, const int *attribList);
-extern void glXDestroyWindow (Display *dpy, GLXWindow window);
-extern GLXPixmap glXCreatePixmap (Display *dpy, GLXFBConfig config,Pixmap pixmap, const int *attribList);
+extern GLXWindow glXCreateWindow (Display *dpy, GLXFBConfig config, Window win, const int *attrib_list);
+extern void glXDestroyWindow (Display *dpy, GLXWindow win);
+extern GLXPixmap glXCreatePixmap (Display *dpy, GLXFBConfig config, Pixmap pixmap, const int *attrib_list);
 extern void glXDestroyPixmap (Display *dpy, GLXPixmap pixmap);
-extern GLXPbuffer glXCreatePbuffer (Display *dpy, GLXFBConfig config, const int *attribList);
+extern GLXPbuffer glXCreatePbuffer (Display *dpy, GLXFBConfig config, const int *attrib_list);
 extern void glXDestroyPbuffer (Display *dpy, GLXPbuffer pbuf);
 extern void glXQueryDrawable (Display *dpy, GLXDrawable draw, int attribute, unsigned int *value);
-extern GLXContext glXCreateNewContext (Display *dpy, GLXFBConfig config, int renderType, GLXContext shareList, Bool direct);
-extern Bool glXMakeContextCurrent (Display *dpy, GLXDrawable draw, GLXDrawable read, GLXContext ctx);
+extern GLXContext glXCreateNewContext (Display *dpy, GLXFBConfig config, int render_type, GLXContext share_list, Bool direct);
+extern Bool glXMakeContextCurrent (Display *display, GLXDrawable draw, GLXDrawable read, GLXContext ctx);
 extern GLXDrawable glXGetCurrentReadDrawable (void);
+extern Display * glXGetCurrentDisplay (void);
 extern int glXQueryContext (Display *dpy, GLXContext ctx, int attribute, int *value);
-extern void glXSelectEvent (Display *dpy, GLXDrawable drawable, unsigned long mask);
-extern void glXGetSelectedEvent (Display *dpy, GLXDrawable drawable, unsigned long *mask);
+extern void glXSelectEvent (Display *dpy, GLXDrawable draw, unsigned long event_mask);
+extern void glXGetSelectedEvent (Display *dpy, GLXDrawable draw, unsigned long *event_mask);
 
-/* Extensions */
-extern Display * glXGetCurrentDisplay (void);
+/*** SGI GLX extensions */
 extern GLXContextID glXGetContextIDEXT (const GLXContext ctx);
 extern GLXDrawable glXGetCurrentDrawableEXT (void);
 extern GLXContext glXImportContextEXT (Display *dpy, GLXContextID contextID);
 extern void glXFreeContextEXT (Display *dpy, GLXContext ctx);
 extern int glXQueryContextInfoEXT (Display *dpy, GLXContext ctx, int attribute, int *value);
-extern void (*glXGetProcAddressARB(const GLubyte *procName))();
 
+extern void (*glXGetProcAddressARB(const GLubyte *procName))( void );
+
+/*** Should these go here, or in another header? */
+/*
+** GLX Events
+*/
+typedef struct {
+    int event_type;		/* GLX_DAMAGED or GLX_SAVED */
+    int draw_type;		/* GLX_WINDOW or GLX_PBUFFER */
+    unsigned long serial;	/* # of last request processed by server */
+    Bool send_event;		/* true if this came for SendEvent request */
+    Display *display;		/* display the event was read from */
+    GLXDrawable drawable;	/* XID of Drawable */
+    unsigned int buffer_mask;	/* mask indicating which buffers are affected */
+    unsigned int aux_buffer;	/* which aux buffer was affected */
+    int x, y;
+    int width, height;
+    int count;			/* if nonzero, at least this many more */
+} GLXPbufferClobberEvent;
 
+typedef union __GLXEvent {
+    GLXPbufferClobberEvent glxpbufferclobber;
+    long pad[24];
+} GLXEvent;
 
 #ifdef __cplusplus
 }
diff --git a/xc/include/GL/glxtokens.h b/xc/include/GL/glxtokens.h
index 5a3e219c3..4efa20a71 100644
--- a/xc/include/GL/glxtokens.h
+++ b/xc/include/GL/glxtokens.h
@@ -1,26 +1,38 @@
 #ifndef __GLX_glxtokens_h__
 #define __GLX_glxtokens_h__
 
-/* $XFree86: xc/include/GL/glxtokens.h,v 1.3 2000/02/15 07:13:24 martin Exp $ */
 /*
-** The contents of this file are subject to the GLX Public License Version 1.0
-** (the "License"). You may not use this file except in compliance with the
-** License. You may obtain a copy of the License at Silicon Graphics, Inc.,
-** attn: Legal Services, 2011 N. Shoreline Blvd., Mountain View, CA 94043
-** or at http://www.sgi.com/software/opensource/glx/license.html.
-**
-** Software distributed under the License is distributed on an "AS IS"
-** basis. ALL WARRANTIES ARE DISCLAIMED, INCLUDING, WITHOUT LIMITATION, ANY
-** IMPLIED WARRANTIES OF MERCHANTABILITY, OF FITNESS FOR A PARTICULAR
-** PURPOSE OR OF NON- INFRINGEMENT. See the License for the specific
-** language governing rights and limitations under the License.
-**
-** The Original Software is GLX version 1.2 source code, released February,
-** 1999. The developer of the Original Software is Silicon Graphics, Inc.
-** Those portions of the Subject Software created by Silicon Graphics, Inc.
-** are Copyright (c) 1991-9 Silicon Graphics, Inc. All Rights Reserved.
-**
-** $SGI$
+** License Applicability. Except to the extent portions of this file are
+** made subject to an alternative license as permitted in the SGI Free
+** Software License B, Version 1.0 (the "License"), the contents of this
+** file are subject only to the provisions of the License. You may not use
+** this file except in compliance with the License. You may obtain a copy
+** of the License at Silicon Graphics, Inc., attn: Legal Services, 1600
+** Amphitheatre Parkway, Mountain View, CA 94043-1351, or at:
+** 
+** http://oss.sgi.com/projects/FreeB
+** 
+** Note that, as provided in the License, the Software is distributed on an
+** "AS IS" basis, with ALL EXPRESS AND IMPLIED WARRANTIES AND CONDITIONS
+** DISCLAIMED, INCLUDING, WITHOUT LIMITATION, ANY IMPLIED WARRANTIES AND
+** CONDITIONS OF MERCHANTABILITY, SATISFACTORY QUALITY, FITNESS FOR A
+** PARTICULAR PURPOSE, AND NON-INFRINGEMENT.
+** 
+** Original Code. The Original Code is: OpenGL Sample Implementation,
+** Version 1.2.1, released January 26, 2000, developed by Silicon Graphics,
+** Inc. The Original Code is Copyright (c) 1991-2000 Silicon Graphics, Inc.
+** Copyright in any portions created by third parties is as indicated
+** elsewhere herein. All Rights Reserved.
+** 
+** Additional Notice Provisions: The application programming interfaces
+** established by SGI in conjunction with the Original Code are The
+** OpenGL(R) Graphics System: A Specification (Version 1.2.1), released
+** April 1, 1999; The OpenGL(R) Graphics System Utility Library (Version
+** 1.3), released November 4, 1998; and OpenGL(R) Graphics with the X
+** Window System(R) (Version 1.3), released October 19, 1998. This software
+** was created using the OpenGL(R) version 1.2.1 Sample Implementation
+** published by SGI, but has not been independently verified as being
+** compliant with the OpenGL(R) version 1.2.1 Specification.
 */
 
 #ifdef __cplusplus
@@ -29,9 +41,10 @@ extern "C" {
 
 #define GLX_VERSION_1_1 1
 #define GLX_VERSION_1_2 1
+#define GLX_VERSION_1_3 1
 
 /*
-** Names for attributes to glXGetConfig.
+** Visual Config Attributes (glXGetConfig, glXGetFBConfigAttrib)
 */
 #define GLX_USE_GL		1	/* support GLX rendering */
 #define GLX_BUFFER_SIZE		2	/* depth of the color buffer */
@@ -39,7 +52,7 @@ extern "C" {
 #define GLX_RGBA		4	/* true if RGBA mode */
 #define GLX_DOUBLEBUFFER	5	/* double buffering supported */
 #define GLX_STEREO		6	/* stereo buffering supported */
-#define GLX_AUX_BUFFERS 	7	/* number of aux buffers */
+#define GLX_AUX_BUFFERS		7	/* number of aux buffers */
 #define GLX_RED_SIZE		8	/* number of red component bits */
 #define GLX_GREEN_SIZE		9	/* number of green component bits */
 #define GLX_BLUE_SIZE		10	/* number of blue component bits */
@@ -50,6 +63,25 @@ extern "C" {
 #define GLX_ACCUM_GREEN_SIZE	15	/* number of green accum bits */
 #define GLX_ACCUM_BLUE_SIZE	16	/* number of blue accum bits */
 #define GLX_ACCUM_ALPHA_SIZE	17	/* number of alpha accum bits */
+/*
+** FBConfig-specific attributes
+*/
+#define GLX_X_VISUAL_TYPE		0x22
+#define GLX_CONFIG_CAVEAT		0x20	/* Like visual_info VISUAL_CAVEAT_EXT */
+#define GLX_TRANSPARENT_TYPE		0x23
+#define GLX_TRANSPARENT_INDEX_VALUE	0x24
+#define GLX_TRANSPARENT_RED_VALUE	0x25
+#define GLX_TRANSPARENT_GREEN_VALUE	0x26
+#define GLX_TRANSPARENT_BLUE_VALUE	0x27
+#define GLX_TRANSPARENT_ALPHA_VALUE	0x28
+#define GLX_DRAWABLE_TYPE		0x8010
+#define GLX_RENDER_TYPE			0x8011
+#define GLX_X_RENDERABLE		0x8012
+#define GLX_FBCONFIG_ID			0x8013
+#define GLX_MAX_PBUFFER_WIDTH		0x8016
+#define GLX_MAX_PBUFFER_HEIGHT		0x8017
+#define GLX_MAX_PBUFFER_PIXELS		0x8018
+#define GLX_VISUAL_ID			0x800B
 
 /*
 ** Error return values from glXGetConfig.  Success is indicated by
@@ -59,9 +91,85 @@ extern "C" {
 #define GLX_BAD_ATTRIBUTE	2	/* attribute to get is bad */
 #define GLX_NO_EXTENSION	3	/* no glx extension on server */
 #define GLX_BAD_VISUAL		4	/* visual # not known by GLX */
-#define GLX_BAD_CONTEXT         5
-#define GLX_BAD_VALUE           6
-#define GLX_BAD_ENUM            7
+#define GLX_BAD_CONTEXT		5	/* returned only by import_context EXT? */
+#define GLX_BAD_VALUE		6	/* returned only by glXSwapIntervalSGI? */
+#define GLX_BAD_ENUM		7	/* unused? */
+
+/* FBConfig attribute values */
+
+/*
+** Generic "don't care" value for glX ChooseFBConfig attributes (except
+** GLX_LEVEL)
+*/
+#define GLX_DONT_CARE			0xFFFFFFFF
+
+/* GLX_RENDER_TYPE bits */
+#define GLX_RGBA_BIT			0x00000001
+#define GLX_COLOR_INDEX_BIT		0x00000002
+
+/* GLX_DRAWABLE_TYPE bits */
+#define GLX_WINDOW_BIT			0x00000001
+#define GLX_PIXMAP_BIT			0x00000002
+#define GLX_PBUFFER_BIT			0x00000004
+
+/* GLX_CONFIG_CAVEAT attribute values */
+#define GLX_NONE			0x8000
+#define GLX_SLOW_CONFIG			0x8001
+#define GLX_NON_CONFORMANT_CONFIG	0x800D
+
+/* GLX_X_VISUAL_TYPE attribute values */
+#define GLX_TRUE_COLOR			0x8002
+#define GLX_DIRECT_COLOR		0x8003
+#define GLX_PSEUDO_COLOR		0x8004
+#define GLX_STATIC_COLOR		0x8005
+#define GLX_GRAY_SCALE			0x8006
+#define GLX_STATIC_GRAY			0x8007
+
+/* GLX_TRANSPARENT_TYPE attribute values */
+/* #define GLX_NONE			   0x8000 */
+#define GLX_TRANSPARENT_RGB		0x8008
+#define GLX_TRANSPARENT_INDEX		0x8009
+
+/* glXCreateGLXPbuffer attributes */
+#define GLX_PRESERVED_CONTENTS		0x801B
+#define GLX_LARGEST_PBUFFER		0x801C
+#define GLX_PBUFFER_HEIGHT		0x8040	/* New for GLX 1.3 */
+#define GLX_PBUFFER_WIDTH		0x8041	/* New for GLX 1.3 */
+
+/* glXQueryGLXPBuffer attributes */
+#define GLX_WIDTH			0x801D
+#define GLX_HEIGHT			0x801E
+#define GLX_EVENT_MASK			0x801F
+
+/* glXCreateNewContext render_type attribute values */
+#define GLX_RGBA_TYPE			0x8014
+#define GLX_COLOR_INDEX_TYPE		0x8015
+
+/* glXQueryContext attributes */
+/* #define GLX_FBCONFIG_ID		  0x8013 */
+/* #define GLX_RENDER_TYPE		  0x8011 */
+#define GLX_SCREEN			0x800C
+
+/* glXSelectEvent event mask bits */
+#define GLX_PBUFFER_CLOBBER_MASK	0x08000000
+
+/* GLXPbufferClobberEvent event_type values */
+#define GLX_DAMAGED			0x8020
+#define GLX_SAVED			0x8021
+
+/* GLXPbufferClobberEvent draw_type values */
+#define GLX_WINDOW			0x8022
+#define GLX_PBUFFER			0x8023
+
+/* GLXPbufferClobberEvent buffer_mask bits */
+#define GLX_FRONT_LEFT_BUFFER_BIT	0x00000001
+#define GLX_FRONT_RIGHT_BUFFER_BIT	0x00000002
+#define GLX_BACK_LEFT_BUFFER_BIT	0x00000004
+#define GLX_BACK_RIGHT_BUFFER_BIT	0x00000008
+#define GLX_AUX_BUFFERS_BIT		0x00000010
+#define GLX_DEPTH_BUFFER_BIT		0x00000020
+#define GLX_STENCIL_BUFFER_BIT		0x00000040
+#define GLX_ACCUM_BUFFER_BIT		0x00000080
 
 /*
 ** Extension return values from glXGetConfig.  These are also
@@ -71,27 +179,27 @@ extern "C" {
 #define GLX_X_VISUAL_TYPE_EXT	0x22	/* visual_info extension type */
 #define GLX_TRANSPARENT_TYPE_EXT 0x23	/* visual_info extension */
 #define GLX_TRANSPARENT_INDEX_VALUE_EXT 0x24	/* visual_info extension */
-#define GLX_TRANSPARENT_RED_VALUE_EXT   0x25	/* visual_info extension */
+#define GLX_TRANSPARENT_RED_VALUE_EXT	0x25	/* visual_info extension */
 #define GLX_TRANSPARENT_GREEN_VALUE_EXT 0x26	/* visual_info extension */
-#define GLX_TRANSPARENT_BLUE_VALUE_EXT  0x27	/* visual_info extension */
+#define GLX_TRANSPARENT_BLUE_VALUE_EXT	0x27	/* visual_info extension */
 #define GLX_TRANSPARENT_ALPHA_VALUE_EXT 0x28	/* visual_info extension */
 
 /* Property values for visual_type */
-#define GLX_TRUE_COLOR_EXT     	0x8002
-#define GLX_DIRECT_COLOR_EXT    0x8003  
-#define GLX_PSEUDO_COLOR_EXT    0x8004 
-#define GLX_STATIC_COLOR_EXT    0x8005
-#define GLX_GRAY_SCALE_EXT    	0x8006  
-#define GLX_STATIC_GRAY_EXT    	0x8007 
+#define GLX_TRUE_COLOR_EXT	0x8002
+#define GLX_DIRECT_COLOR_EXT	0x8003
+#define GLX_PSEUDO_COLOR_EXT	0x8004
+#define GLX_STATIC_COLOR_EXT	0x8005
+#define GLX_GRAY_SCALE_EXT	0x8006
+#define GLX_STATIC_GRAY_EXT	0x8007
 
 /* Property values for transparent pixel */
-#define GLX_NONE_EXT     	0x8000     
-#define GLX_TRANSPARENT_RGB_EXT    	0x8008    
-#define GLX_TRANSPARENT_INDEX_EXT    	0x8009    
+#define GLX_NONE_EXT		0x8000
+#define GLX_TRANSPARENT_RGB_EXT		0x8008
+#define GLX_TRANSPARENT_INDEX_EXT	0x8009
 
 /* Property values for visual_rating */
-#define GLX_VISUAL_CAVEAT_EXT    	0x20  /* visual_rating extension type */
-#define GLX_SLOW_VISUAL_EXT		0x8001   
+#define GLX_VISUAL_CAVEAT_EXT		0x20  /* visual_rating extension type */
+#define GLX_SLOW_VISUAL_EXT		0x8001
 #define GLX_NON_CONFORMANT_VISUAL_EXT	0x800D
 
 /*
@@ -109,11 +217,12 @@ extern "C" {
 #define GLX_SCREEN_EXT		0x800C	/* screen number */
 
 /* GLX Extension Strings */
-#define GLX_EXT_import_context  1
-#define GLX_EXT_visual_info     1
-#define GLX_EXT_visual_rating   1
+#define GLX_EXT_import_context	1
+#define GLX_EXT_visual_info	1
+#define GLX_EXT_visual_rating	1
 #define GLX_ARB_get_proc_address 1
 
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/xc/lib/GL/Imakefile b/xc/lib/GL/Imakefile
index dfc8231d2..c882109e1 100644
--- a/xc/lib/GL/Imakefile
+++ b/xc/lib/GL/Imakefile
@@ -134,6 +134,22 @@ REQUIREDLIBS += -lglide3x
 
   GLXSUBDIRS = glx dri mesa/dri mesa/include/GL mesa/src
 
+#elif GlxBuiltInR128
+
+    R128OBJS = mesa/src/drv/r128/?*.o
+   R128UOBJS = mesa/src/drv/r128/unshared/?*.o
+   R128DOBJS = mesa/src/drv/r128/debugger/?*.o
+   R128POBJS = mesa/src/drv/r128/profiled/?*.o
+   R128DONES = mesa/src/drv/r128/DONE
+
+     DRVOBJS = $(R128OBJS)  $(MESAOBJS)  $(DRIMESAOBJS)  $(DRMOBJS)
+    DRVUOBJS = $(R128UOBJS) $(MESAUOBJS) $(DRIMESAUOBJS) $(DRMUOBJS)
+    DRVDOBJS = $(R128DOBJS) $(MESADOBJS) $(DRIMESADOBJS) $(DRMDOBJS)
+    DRVPOBJS = $(R128POBJS) $(MESAPOBJS) $(DRIMESAPOBJS) $(DRMPOBJS)
+    DRVDONES = $(R128DONES) $(MESADONES) $(DRIMESADONES) $(DRMDONES)
+
+  GLXSUBDIRS = glx dri mesa/dri mesa/include/GL mesa/src
+
 #elif GlxBuiltInMesa
 
 #ifndef GlxDriverUsesMesa
diff --git a/xc/lib/GL/dri/XF86dri.c b/xc/lib/GL/dri/XF86dri.c
index 51615556a..dd0ee56ff 100644
--- a/xc/lib/GL/dri/XF86dri.c
+++ b/xc/lib/GL/dri/XF86dri.c
@@ -276,7 +276,7 @@ Bool XF86DRIGetClientDriverName(dpy, screen, ddxDriverMajorVersion,
     *ddxDriverPatchVersion = rep.ddxDriverPatchVersion;
 
     if (rep.length) {
-        if (!(*clientDriverName = (char *)Xcalloc(rep.length + 1, 1))) {
+        if (!(*clientDriverName = (char *)Xcalloc(rep.clientDriverNameLength + 1, 1))) {
             _XEatData(dpy, ((rep.clientDriverNameLength+3) & ~3));
             return False;
         }
diff --git a/xc/lib/GL/dri/drm/Imakefile b/xc/lib/GL/dri/drm/Imakefile
index 18619461b..42b6202d6 100644
--- a/xc/lib/GL/dri/drm/Imakefile
+++ b/xc/lib/GL/dri/drm/Imakefile
@@ -12,12 +12,17 @@ ALLOC_DEFINES = -DMALLOC_0_RETURNS_NULL
 
       DEFINES = $(ALLOC_DEFINES)
      INCLUDES = -I$(XLIBSRC) -I$(EXTINCSRC) -I.
-         SRCS = xf86drm.c xf86drmHash.c xf86drmRandom.c xf86drmSL.c
-         OBJS = xf86drm.o xf86drmHash.o xf86drmRandom.o xf86drmSL.o
+         SRCS = xf86drm.c xf86drmHash.c xf86drmRandom.c xf86drmSL.c \
+		xf86drmR128.c
+         OBJS = xf86drm.o xf86drmHash.o xf86drmRandom.o xf86drmSL.o \
+		xf86drmR128.o
 
 #if defined(LinuxArchitecture)
 OS_SUBDIR = linux
 #endif
+#if defined(FreeBSDArchitecture)
+OS_SUBDIR = bsd
+#endif
 
 LinkSourceFile(xf86drm.c,$(XF86OSSRC)/$(OS_SUBDIR)/drm)
 LinkSourceFile(xf86drmHash.c,$(XF86OSSRC)/$(OS_SUBDIR)/drm)
@@ -25,6 +30,11 @@ LinkSourceFile(xf86drmRandom.c,$(XF86OSSRC)/$(OS_SUBDIR)/drm)
 LinkSourceFile(xf86drmSL.c,$(XF86OSSRC)/$(OS_SUBDIR)/drm)
 LinkSourceFile(xf86drm.h,$(XF86OSSRC))
 LinkSourceFile(drm.h,$(XF86OSSRC)/$(OS_SUBDIR)/drm/kernel)
+LinkSourceFile(mga_drm.h,$(XF86OSSRC)/$(OS_SUBDIR)/drm/kernel)
+LinkSourceFile(i810_drm.h,$(XF86OSSRC)/$(OS_SUBDIR)/drm/kernel)
+LinkSourceFile(xf86drmR128.c,$(XF86OSSRC)/$(OS_SUBDIR)/drm)
+LinkSourceFile(xf86drmR128.h,$(XF86OSSRC))
+LinkSourceFile(r128_drm.h,$(XF86OSSRC)/$(OS_SUBDIR)/drm/kernel)
 
 #include <Library.tmpl>
 
diff --git a/xc/lib/GL/mesa/src/Imakefile b/xc/lib/GL/mesa/src/Imakefile
index 7d9b1d8e8..1d42b4950 100644
--- a/xc/lib/GL/mesa/src/Imakefile
+++ b/xc/lib/GL/mesa/src/Imakefile
@@ -348,8 +348,12 @@ LinkSourceFile(zoom.h, ../../../../extras/Mesa/src)
 #ifdef i386Architecture
      ASM_SRCS = 
      ASM_OBJS = 
+#ifdef MesaUse3DNow
+     ASM_DEFS = -DUSE_MMX_ASM -DUSE_X86_ASM -DUSE_3DNOW_ASM
+#else
      ASM_DEFS = -DUSE_MMX_ASM -DUSE_X86_ASM
 #endif
+#endif
 
       DEFINES = $(ALLOC_DEFINES) GlxDefines $(TDFX_DEFS) $(ASM_DEFS)
      INCLUDES = -I$(XLIBSRC) -I$(EXTINCSRC) -I../include -I../../dri -I. -I../../../../include
diff --git a/xc/lib/GL/mesa/src/X86/Imakefile b/xc/lib/GL/mesa/src/X86/Imakefile
index 858f0cd73..53402ac67 100644
--- a/xc/lib/GL/mesa/src/X86/Imakefile
+++ b/xc/lib/GL/mesa/src/X86/Imakefile
@@ -16,7 +16,6 @@ LinkSourceFile(x86.h, ../../../../../extras/Mesa/src/X86)
 LinkSourceFile(x86a.S, ../../../../../extras/Mesa/src/X86)
 LinkSourceFile(vertex.S, ../../../../../extras/Mesa/src/X86)
 LinkSourceFile(x86flatregs.m4, ../../../../../extras/Mesa/src/X86)
-LinkSourceFile(x86a.S.m4, ../../../../../extras/Mesa/src/X86)
 
 LinkSourceFile(mmx.h, ../../../../../extras/Mesa/src/X86)
 LinkSourceFile(mmx_blend.S, ../../../../../extras/Mesa/src/X86)
@@ -55,9 +54,7 @@ XCOMM We'll learn at runtime whether 3dNow, MMX, etc are really present.
 
       MMX_DEFS = -DUSE_MMX_ASM
 
-
-XCOMM Disabling 3DNow code for the time being
-#if 0
+#ifdef MesaUse3DNow
     3DNOW_SRCS = 3dnow.c 3dnow_norm_raw.S 3dnow_xform_masked1.S \
 		 3dnow_xform_masked2.S 3dnow_xform_masked3.S \
 		 3dnow_xform_masked4.S 3dnow_xform_raw1.S \
@@ -73,6 +70,7 @@ XCOMM Disabling 3DNow code for the time being
     3DNOW_DEFS = -DUSE_3DNOW_ASM
 #endif
 
+
 #endif
 
       DEFINES = $(ALLOC_DEFINES) GlxDefines -DFX $(X86_DEFS) $(MMX_DEFS) $(3DNOW_DEFS)
@@ -96,7 +94,7 @@ STD_CPP_DEFINES = StandardDefines $(PROJECT_DEFINES)
 SubdirLibraryRule($(OBJS))
 NormalLintTarget($(SRCS))
 
-#ifdef HAVE_3DNOW
+#ifdef MesaUse3DNow
 ObjectFromAsmSource(3dnow_norm_raw, NullParameter)
 ObjectFromAsmSource(3dnow_xform_masked1, NullParameter)
 ObjectFromAsmSource(3dnow_xform_masked2, NullParameter)
@@ -109,9 +107,7 @@ ObjectFromAsmSource(3dnow_xform_raw4, NullParameter)
 ObjectFromAsmSource(vertex_3dnow, NullParameter)
 #endif
 
-#ifdef HAVE_MMX
 ObjectFromAsmSource(mmx_blend, NullParameter)
-#endif
 
 ObjectFromAsmSource(common_x86asm, NullParameter)
 ObjectFromAsmSource(vertex, NullParameter)
@@ -119,6 +115,3 @@ ObjectFromAsmSource(x86a, NullParameter)
 
 DependTarget()
 
-x86a.S: x86flatregs.m4
-x86a.S: x86a.S.m4
-	m4 $< >$@
diff --git a/xc/lib/GL/mesa/src/drv/Imakefile b/xc/lib/GL/mesa/src/drv/Imakefile
index fadb29bb2..ed30e7196 100644
--- a/xc/lib/GL/mesa/src/drv/Imakefile
+++ b/xc/lib/GL/mesa/src/drv/Imakefile
@@ -16,6 +16,15 @@ DRIVER += gamma
 #if GlxBuiltInTdfx
 DRIVER += tdfx
 #endif
+#if GlxBuiltInMga
+DRIVER += common mga
+#endif
+#if GlxBuiltInI810
+DRIVER += common i810
+#endif
+#if GlxBuiltInR128
+DRIVER += r128
+#endif
 
 SUBDIRS = $(DRIVER)
 #else
@@ -23,6 +32,10 @@ SUBDIRS += gamma
 #if HasGlide3
 SUBDIRS += tdfx
 #endif
+SUBDIRS += common
+SUBDIRS += mga
+SUBDIRS += i810
+SUBDIRS += r128
 #endif
 
 MakeSubdirs($(SUBDIRS))
diff --git a/xc/lib/GL/mesa/src/drv/common/Imakefile b/xc/lib/GL/mesa/src/drv/common/Imakefile
index 5c3d40dd7..d47b662e8 100644
--- a/xc/lib/GL/mesa/src/drv/common/Imakefile
+++ b/xc/lib/GL/mesa/src/drv/common/Imakefile
@@ -22,13 +22,12 @@ MESA_INCLUDES = -I. -I.. -I../../include
 
 
       DEFINES = $(ALLOC_DEFINES) $(DRI_DEFINES)
-     INCLUDES = -I$(XLIBSRC) -I$(EXTINCSRC) $(MESA_INCLUDES) $(DRI_INCLUDES) \
-		-I/usr/include/glide
+     INCLUDES = -I$(XLIBSRC) -I$(EXTINCSRC) $(MESA_INCLUDES) $(DRI_INCLUDES)
       DRISRCS = hwlog.c mm.c
       DRIOBJS = hwlog.o mm.o
 
-	 SRCS = $(DRISRCS) 
-	 OBJS = $(DRIOBJS) 
+         SRCS = $(DRISRCS) 
+         OBJS = $(DRIOBJS) 
 
 #if !GlxUseBuiltInDRIDriver
 #undef DoNormalLib NormalLibGlx
@@ -44,7 +43,8 @@ LibraryObjectRule()
 SubdirLibraryRule($(OBJS))
 NormalLintTarget($(SRCS))
 
-#if !GlxUseBuiltInDRIDriver
+XCOMM #if !GlxUseBuiltInDRIDriver
+#if 0
 LIBNAME = i810_dri.so
 ALL_OBJS = $(OBJS)
 ALL_DEPS = DONE
diff --git a/xc/lib/GL/mesa/src/drv/gamma/Imakefile b/xc/lib/GL/mesa/src/drv/gamma/Imakefile
index 05e35561e..0e9833177 100644
--- a/xc/lib/GL/mesa/src/drv/gamma/Imakefile
+++ b/xc/lib/GL/mesa/src/drv/gamma/Imakefile
@@ -1,4 +1,5 @@
-XCOMM $XFree86: xc/lib/GL/mesa/src/drv/gamma/Imakefile,v 1.8 2000/03/02 16:07:35 martin Exp $
+
+#include <Threads.tmpl>
 
 #define DoNormalLib NormalLibGlx
 #define DoSharedLib SharedLibGlx
@@ -6,28 +7,35 @@ XCOMM $XFree86: xc/lib/GL/mesa/src/drv/gamma/Imakefile,v 1.8 2000/03/02 16:07:35
 #define DoDebugLib DebugLibGlx
 #define DoProfileLib ProfileLibGlx
 
-LinkSourceFile(xmesaP.h, ../../../../../../extras/Mesa/src/X)
-LinkSourceFile(glapi.h, ../../../../../../extras/Mesa/src)
-
-
 #if Malloc0ReturnsNull
 ALLOC_DEFINES = -DMALLOC_0_RETURNS_NULL
 #endif
 
 #if BuildXF86DRI
-  DRI_DEFINES = GlxDefines
+  DRI_DEFINES = GlxDefines -DDRIVERTS
  DRI_INCLUDES = -I../../../../dri -I../../../../glx \
 		-I../../../dri \
 		-I$(TOP)/include -I$(TOP)/include/GL \
 		-I$(XF86OSSRC) -I$(SERVERSRC)/GL/dri \
 		-I$(XF86DRIVERSRC)/glint \
-		-I../../../include -I../.. -I../../X
+		-I../../../include -I../.. -I../../X -I../common \
+	        -I$(XF86OSSRC)/linux/drm/kernel
 #endif
 
 MESA_INCLUDES = -I. -I.. -I../../include
 
+
+
       DEFINES = $(ALLOC_DEFINES) $(DRI_DEFINES)
-     INCLUDES = -I$(XLIBSRC) -I$(EXTINCSRC) $(MESA_INCLUDES) $(DRI_INCLUDES)
+     INCLUDES = -I$(XLIBSRC) -I$(EXTINCSRC) $(MESA_INCLUDES) $(DRI_INCLUDES) 
+
+#if 0
+	LOSRC = ../../../../lowpc.c
+	LOOBJ = ../../../../lowpc.o
+
+	HISRC = ../../../../highpc.c
+	HIOBJ = ../../../../highpc.o
+#endif
 
       DRISRCS = ../../../dri/dri_mesa.c \
 		../../../../dri/dri_tmm.c
@@ -46,16 +54,216 @@ MESA_INCLUDES = -I. -I.. -I../../include
 		../../../../dri/drm/xf86drmSL.o
 
     GAMMASRCS = gamma_gl.c gamma_xmesa.c gamma_init.c gamma_matrix.c \
-                gamma_inithw.c gamma_texture.c
+                gamma_inithw.c gamma_texture.c /* gamma_dlist.c */
 
     GAMMAOBJS = gamma_gl.o gamma_xmesa.o gamma_init.o gamma_matrix.o \
-                gamma_inithw.o gamma_texture.o
+                gamma_inithw.o gamma_texture.o /* gamma_dlist.o */
+
+     MESASRCS = ../../aatriangle.c \
+		../../accum.c \
+		../../alpha.c \
+		../../alphabuf.c \
+		../../attrib.c \
+		../../bbox.c \
+		../../bitmap.c \
+		../../blend.c \
+		../../buffers.c \
+		../../clip.c \
+		../../colortab.c \
+		../../config.c \
+		../../context.c \
+		../../copypix.c \
+		../../cva.c \
+		../../debug_xform.c \
+		../../depth.c \
+		../../dlist.c \
+		../../drawpix.c \
+		../../enable.c \
+		../../enums.c \
+		../../eval.c \
+		../../extensions.c \
+		../../feedback.c \
+		../../fog.c \
+		../../get.c \
+		../../glapi.c \
+		../../glapinoop.c \
+		../../glthread.c \
+		../../hash.c \
+		../../image.c \
+		../../imaging.c \
+		../../light.c \
+		../../lines.c \
+		../../logic.c \
+		../../masking.c \
+		../../matrix.c \
+		../../mem.c \
+		../../mmath.c \
+		../../pb.c \
+		../../pipeline.c \
+		../../pixel.c \
+		../../pixeltex.c \
+		../../points.c \
+		../../polygon.c \
+		../../quads.c \
+		../../rastpos.c \
+		../../readpix.c \
+		../../rect.c \
+		../../scissor.c \
+		../../shade.c \
+		../../span.c \
+		../../stages.c \
+		../../state.c \
+		../../stencil.c \
+		../../teximage.c \
+		../../texobj.c \
+		../../texstate.c \
+		../../texture.c \
+		../../texutil.c \
+		../../translate.c \
+		../../triangle.c \
+		../../varray.c \
+		../../vb.c \
+		../../vbcull.c \
+		../../vbfill.c \
+		../../vbindirect.c \
+		../../vbrender.c \
+		../../vbxform.c \
+		../../vector.c \
+		../../vertices.c \
+		../../winpos.c \
+		../../xform.c \
+		../../zoom.c \
+		../../X86/common_x86.c
+
+     MESAOBJS = ../../aatriangle.o \
+		../../accum.o \
+		../../alpha.o \
+		../../alphabuf.o \
+		../../attrib.o \
+		../../bbox.o \
+		../../bitmap.o \
+		../../blend.o \
+		../../buffers.o \
+		../../clip.o \
+		../../colortab.o \
+		../../config.o \
+		../../context.o \
+		../../copypix.o \
+		../../cva.o \
+		../../debug_xform.o \
+		../../depth.o \
+		../../dlist.o \
+		../../drawpix.o \
+		../../enable.o \
+		../../enums.o \
+		../../eval.o \
+		../../extensions.o \
+		../../feedback.o \
+		../../fog.o \
+		../../get.o \
+		../../hash.o \
+		../../hint.o \
+		../../image.o \
+		../../imaging.o \
+		../../light.o \
+		../../lines.o \
+		../../logic.o \
+		../../masking.o \
+		../../matrix.o \
+		../../mem.o \
+		../../mmath.o \
+		../../pb.o \
+		../../pipeline.o \
+		../../pixel.o \
+		../../pixeltex.o \
+		../../points.o \
+		../../polygon.o \
+		../../quads.o \
+		../../rastpos.o \
+		../../readpix.o \
+		../../rect.o \
+		../../scissor.o \
+		../../shade.o \
+		../../span.o \
+		../../stages.o \
+		../../state.o \
+		../../stencil.o \
+		../../teximage.o \
+		../../texobj.o \
+		../../texstate.o \
+		../../texture.o \
+		../../texutil.o \
+		../../translate.o \
+		../../triangle.o \
+		../../varray.o \
+		../../vb.o \
+		../../vbcull.o \
+		../../vbfill.o \
+		../../vbindirect.o \
+		../../vbrender.o \
+		../../vbxform.o \
+		../../vector.o \
+		../../vertices.o \
+		../../winpos.o \
+		../../xform.o \
+		../../zoom.o
+
+#ifdef i386Architecture
+     X86_SRCS = ../../X86/x86.c \
+		../../X86/x86a.S \
+		../../X86/common_x86.c \
+		../../X86/common_x86asm.S \
+		../../X86/vertex.S
 
-XCOMM	 SRCS = $(DRISRCS) $(DRMSRCS) $(GAMMASRCS)
-XCOMM	 OBJS = $(DRIOBJS) $(DRMOBJS) $(GAMMAOBJS)
+     X86_OBJS = ../../X86/x86.o \
+		../../X86/x86a.o \
+		../../X86/common_x86.o \
+		../../X86/common_x86asm.o \
+		../../X86/vertex.o
 
-	 SRCS = $(GAMMASRCS)
-	 OBJS = $(DRIOBJS) $(DRMOBJS) $(GAMMAOBJS)
+     MMX_SRCS = ../../X86/mmx_blend.S
+
+     MMX_OBJS = ../../X86/mmx_blend.o
+
+XCOMM Disabling 3Dnow code for the time being.
+#if 0
+   3DNOW_SRCS = ../../X86/3dnow.c \
+		../../X86/3dnow_norm_raw.S \
+		../../X86/3dnow_xform_masked1.S \
+		../../X86/3dnow_xform_masked2.S \
+		../../X86/3dnow_xform_masked3.S \
+		../../X86/3dnow_xform_masked4.S \
+		../../X86/3dnow_xform_raw1.S \
+		../../X86/3dnow_xform_raw2.S \
+		../../X86/3dnow_xform_raw3.S \
+		../../X86/3dnow_xform_raw4.S \
+		../../X86/vertex_3dnow.S
+
+   3DNOW_OBJS = ../../X86/3dnow.o \
+		../../X86/3dnow_norm_raw.o \
+		../../X86/3dnow_xform_masked1.o \
+		../../X86/3dnow_xform_masked2.o \
+		../../X86/3dnow_xform_masked3.o \
+		../../X86/3dnow_xform_masked4.o \
+		../../X86/3dnow_xform_raw1.o \
+		../../X86/3dnow_xform_raw2.o \
+		../../X86/3dnow_xform_raw3.o \
+		../../X86/3dnow_xform_raw4.o \
+		../../X86/vertex_3dnow.o
+#endif
+
+#endif
+
+      ASMSRCS = $(X86_SRCS) $(MMX_SRCS) $(3DNOW_SRCS)
+      ASMOBJS = $(X86_OBJS) $(MMX_OBJS) $(3DNOW_OBJS)
+
+	 SRCS = $(LOWSRC) $(DRISRCS) $(DRMSRCS) $(MESASRCS) $(ASMSRCS) $(GAMMASRCS) $(HISRC)
+	 OBJS = $(LOWOBJ) $(DRIOBJS) $(DRMOBJS) $(MESAOBJS) $(ASMOBJS) $(GAMMAOBJS) $(HIOBJ)
+
+REQUIREDLIBS += -lm
+#if !GlxBuiltInMga
+REQUIREDLIBS += -L../../../.. -lGL
+#endif
 
 
 #if !GlxUseBuiltInDRIDriver
diff --git a/xc/lib/GL/mesa/src/drv/gamma/gamma_gl.c b/xc/lib/GL/mesa/src/drv/gamma/gamma_gl.c
index d72c642cf..61e59a7a9 100644
--- a/xc/lib/GL/mesa/src/drv/gamma/gamma_gl.c
+++ b/xc/lib/GL/mesa/src/drv/gamma/gamma_gl.c
@@ -30,13 +30,16 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  * Authors:
  *   Kevin E. Martin <kevin@precisioninsight.com>
  *   Brian Paul <brian@precisioninsight.com>
+ *   Alan Hourihane <Alan.Hourihane@btinternet.com>
  */
 
 #ifdef GLX_DIRECT_RENDERING
 
+#include <Xarch.h>
 #include <math.h>
 #include "gamma_gl.h"
 #include "gamma_init.h"
+#include "glint_dri.h"
 #ifdef RANDOMIZE_COLORS
 #include <stdlib.h>
 #endif
@@ -114,7 +117,7 @@ void _gamma_Begin(GLenum mode)
     DEBUG_GLCMDS(("Begin: %04x\n", (int)mode));
 
     if ((gCCPriv->Begin & B_PrimType_Mask) != B_PrimType_Null) {
-	/* ERROR!!! */
+        DEBUG_ERROR(("Begin: Error\n"));
 	return;
     }
 
@@ -177,8 +180,9 @@ void _gamma_BindTexture(GLenum target, GLuint texture)
     gCCPriv->curTexObj = gammaTOFind(texture);
 
     /* Make the new texture images resident */
-    if (!driTMMMakeImagesResident(gCCPriv->tmm, MIPMAP_LEVELS,
-				  gCCPriv->curTexObj->image, addrs)) {
+    if (driTMMMakeImagesResident(gCCPriv->tmm, MIPMAP_LEVELS,
+				  gCCPriv->curTexObj->image, addrs) < 0) {
+    	DEBUG_ERROR(("BindTexture: unable\n"));
 	/* NOT_DONE: Handle error */
     }
 
@@ -358,8 +362,10 @@ void _gamma_CallLists(GLsizei n, GLenum type, const GLvoid *lists)
 
 void _gamma_Clear(GLbitfield mask)
 {
+    int temp;
     unsigned int depth = 0;
     int do_clear = 0;
+    GLINTDRIPtr gDRIPriv = (GLINTDRIPtr)gCC->driScreenPriv->pDevPriv;
 #ifdef DO_VALIDATE
     __DRIscreenPrivate *driScrnPriv = gCC->driScreenPriv;
 #endif
@@ -437,11 +443,12 @@ void _gamma_Clear(GLbitfield mask)
 	gCCPriv->FrameCount &= 0xff;
 #endif
 
+	temp = (gCCPriv->LBReadMode & LBPartialProdMask) | LBWindowOriginBot;
+	/* UGH - move this later ! */
+	if (gDRIPriv->numMXDevices == 2) temp |= LBScanLineInt2;
+
 	CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
-	WRITE(gCCPriv->buf, LBReadMode, 
-	      ((gCCPriv->LBReadMode & LBPartialProdMask) |
-	       LBScanLineInt2 |
-	       LBWindowOriginBot));
+	WRITE(gCCPriv->buf, LBReadMode, temp);
 
 	/* Force FCP to be written */
 	CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
@@ -627,34 +634,48 @@ void _gamma_ClipPlane(GLenum plane, const GLdouble *equation)
 
 void _gamma_Color3b(GLbyte red, GLbyte green, GLbyte blue)
 {
+    GLfloat r,g,b;
+
     DEBUG_GLCMDS(("Color3b: %d %d %d\n", red, green, blue));
+  
+    r = BYTE_TO_FLOAT(red);
+    g = BYTE_TO_FLOAT(green);
+    b = BYTE_TO_FLOAT(blue);
+
+    _gamma_Color3f(r,g,b);
 }
 
 void _gamma_Color3bv(const GLbyte *v)
 {
+    GLfloat p[3];
+
     DEBUG_GLCMDS(("Color3bv: %d %d %d\n", v[0], v[1], v[2]));
+
+    p[0] = BYTE_TO_FLOAT(v[0]);
+    p[1] = BYTE_TO_FLOAT(v[1]);
+    p[2] = BYTE_TO_FLOAT(v[2]);
+
+    _gamma_Color3fv(p);
 }
 
 void _gamma_Color3d(GLdouble red, GLdouble green, GLdouble blue)
 {
     DEBUG_GLCMDS(("Color3d: %f %f %f\n", red, green, blue));
+
+    _gamma_Color3f((GLfloat)red,(GLfloat)green,(GLfloat)blue);
 }
 
 void _gamma_Color3dv(const GLdouble *v)
 {
     DEBUG_GLCMDS(("Color3dv: %f %f %f\n", v[0], v[1], v[2]));
+
+    _gamma_Color3fv((GLfloat*)v);
 }
 
 void _gamma_Color3f(GLfloat red, GLfloat green, GLfloat blue)
 {
     DEBUG_GLCMDS(("Color3f: %f %f %f\n", red, green, blue));
 
-#ifdef RANDOMIZE_COLORS
-    red = (random() / (double)RAND_MAX);
-    green = (random() / (double)RAND_MAX);
-    blue = (random() / (double)RAND_MAX);
-#endif
-
     CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
     WRITEF(gCCPriv->buf, Cb,  blue);
     WRITEF(gCCPriv->buf, Cg,  green);
@@ -665,107 +686,188 @@ void _gamma_Color3fv(const GLfloat *v)
 {
     DEBUG_GLCMDS(("Color3fv: %f %f %f\n", v[0], v[1], v[2]));
 
-#ifdef RANDOMIZE_COLORS
-    {
-	float r, g, b;
-	r = (random() / (double)RAND_MAX);
-	g = (random() / (double)RAND_MAX);
-	b = (random() / (double)RAND_MAX);
-	CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
-	WRITEF(gCCPriv->buf, Cb,  b);
-	WRITEF(gCCPriv->buf, Cg,  g);
-	WRITEF(gCCPriv->buf, Cr3, r);
-    }
-#else
     CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
     WRITEF(gCCPriv->buf, Cb,  v[2]);
     WRITEF(gCCPriv->buf, Cg,  v[1]);
     WRITEF(gCCPriv->buf, Cr3, v[0]);
-#endif
 }
 
 void _gamma_Color3i(GLint red, GLint green, GLint blue)
 {
+    GLfloat r,g,b;
+
     DEBUG_GLCMDS(("Color3i: %d %d %d\n", (int)red, (int)green, (int)blue));
+
+    r = INT_TO_FLOAT(red);
+    g = INT_TO_FLOAT(green);
+    b = INT_TO_FLOAT(blue);
+
+    _gamma_Color3f(r,g,b);
 }
 
 void _gamma_Color3iv(const GLint *v)
 {
+    GLfloat p[3];
+
     DEBUG_GLCMDS(("Color3iv: %d %d %d\n", (int)v[0], (int)v[1], (int)v[2]));
+
+    p[0] = INT_TO_FLOAT(v[0]);
+    p[1] = INT_TO_FLOAT(v[1]);
+    p[2] = INT_TO_FLOAT(v[2]);
+
+    _gamma_Color3fv(p);
 }
 
 void _gamma_Color3s(GLshort red, GLshort green, GLshort blue)
 {
+    GLfloat r,g,b;
+
     DEBUG_GLCMDS(("Color3s: %d %d %d\n", red, green, blue));
+
+    r = SHORT_TO_FLOAT(red);
+    g = SHORT_TO_FLOAT(green);
+    b = SHORT_TO_FLOAT(blue);
+
+    _gamma_Color3f(r,g,b);
 }
 
 void _gamma_Color3sv(const GLshort *v)
 {
+    GLfloat p[3];
+
     DEBUG_GLCMDS(("Color3sv: %d %d %d\n", v[0], v[1], v[2]));
+
+    p[0] = SHORT_TO_FLOAT(v[0]);
+    p[1] = SHORT_TO_FLOAT(v[1]);
+    p[2] = SHORT_TO_FLOAT(v[2]);
+
+    _gamma_Color3fv(p);
 }
 
 void _gamma_Color3ub(GLubyte red, GLubyte green, GLubyte blue)
 {
+    GLuint c;
+
     DEBUG_GLCMDS(("Color3ub: %d %d %d\n", red, green, blue));
+
+    c = (blue << 16) | (green << 8) | red;
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+    WRITE(gCCPriv->buf, PackedColor3,  c);
 }
 
 void _gamma_Color3ubv(const GLubyte *v)
 {
+    GLuint c;
+
     DEBUG_GLCMDS(("Color3ubv: %d %d %d\n", v[0], v[1], v[2]));
+
+    c = (v[2] << 16) | (v[1] << 8) | v[0];
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+    WRITE(gCCPriv->buf, PackedColor3,  c);
 }
 
 void _gamma_Color3ui(GLuint red, GLuint green, GLuint blue)
 {
+    GLfloat r,g,b;
+
     DEBUG_GLCMDS(("Color3ui: %d %d %d\n",
 		  (unsigned int)red, (unsigned int)green, (unsigned int)blue));
+
+    r = UINT_TO_FLOAT(red);
+    g = UINT_TO_FLOAT(green);
+    b = UINT_TO_FLOAT(blue);
+
+    _gamma_Color3f(r,g,b);
 }
 
 void _gamma_Color3uiv(const GLuint *v)
 {
+    GLfloat p[3];
+
     DEBUG_GLCMDS(("Color3uiv: %d %d %d\n",
 		  (unsigned int)v[0], (unsigned int)v[1], (unsigned int)v[2]));
+
+    p[0] = UINT_TO_FLOAT(v[0]);
+    p[1] = UINT_TO_FLOAT(v[1]);
+    p[2] = UINT_TO_FLOAT(v[2]);
+
+    _gamma_Color3fv(p);
 }
 
 void _gamma_Color3us(GLushort red, GLushort green, GLushort blue)
 {
+    GLuint c;
+
     DEBUG_GLCMDS(("Color3us: %d %d %d\n", red, green, blue));
+
+    c = (USHORT_TO_UBYTE(blue) << 16) | (USHORT_TO_UBYTE(green) << 8) | 
+	USHORT_TO_UBYTE(red);
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+    WRITE(gCCPriv->buf, PackedColor3,  c);
 }
 
 void _gamma_Color3usv(const GLushort *v)
 {
+    GLuint c;
+
     DEBUG_GLCMDS(("Color3usv: %d %d %d\n", v[0], v[1], v[2]));
+
+    c = (USHORT_TO_UBYTE(v[2]) << 16) | (USHORT_TO_UBYTE(v[1]) << 8) | 
+	USHORT_TO_UBYTE(v[0]);
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+    WRITE(gCCPriv->buf, PackedColor3,  c);
 }
 
 void _gamma_Color4b(GLbyte red, GLbyte green, GLbyte blue, GLbyte alpha)
 {
+    GLfloat r,g,b,a;
+
     DEBUG_GLCMDS(("Color4b: %d %d %d %d\n", red, green, blue, alpha));
+
+    r = BYTE_TO_FLOAT(red);
+    g = BYTE_TO_FLOAT(green);
+    b = BYTE_TO_FLOAT(blue);
+    a = BYTE_TO_FLOAT(alpha);
+
+    _gamma_Color4f(r,g,b,a);
 }
 
 void _gamma_Color4bv(const GLbyte *v)
 {
+    GLfloat p[4];
+
     DEBUG_GLCMDS(("Color4bv: %d %d %d %d\n", v[0], v[1], v[2], v[3]));
+
+    p[0] = BYTE_TO_FLOAT(v[0]);
+    p[1] = BYTE_TO_FLOAT(v[1]);
+    p[2] = BYTE_TO_FLOAT(v[2]);
+    p[3] = BYTE_TO_FLOAT(v[3]);
+
+    _gamma_Color4fv(p);
 }
 
 void _gamma_Color4d(GLdouble red, GLdouble green, GLdouble blue, GLdouble alpha)
 {
     DEBUG_GLCMDS(("Color4d: %f %f %f %f\n", red, green, blue, alpha));
+
+    _gamma_Color4f((GLfloat)red,(GLfloat)green,(GLfloat)blue,(GLfloat)alpha);
 }
 
 void _gamma_Color4dv(const GLdouble *v)
 {
     DEBUG_GLCMDS(("Color4dv: %f %f %f %f\n", v[0], v[1], v[2], v[3]));
+
+    _gamma_Color4fv((GLfloat*)v);
 }
 
 void _gamma_Color4f(GLfloat red, GLfloat green, GLfloat blue, GLfloat alpha)
 {
     DEBUG_GLCMDS(("Color4f: %f %f %f %f\n", red, green, blue, alpha));
 
-#ifdef RANDOMIZE_COLORS
-    red = (random() / (double)RAND_MAX);
-    green = (random() / (double)RAND_MAX);
-    blue = (random() / (double)RAND_MAX);
-#endif
-
     CHECK_DMA_BUFFER(gCC, gCCPriv, 4);
     WRITEF(gCCPriv->buf, Ca,  alpha);
     WRITEF(gCCPriv->buf, Cb,  blue);
@@ -777,37 +879,41 @@ void _gamma_Color4fv(const GLfloat *v)
 {
     DEBUG_GLCMDS(("Color4fv: %f %f %f %f\n", v[0], v[1], v[2], v[3]));
 
-#ifdef RANDOMIZE_COLORS
-    {
-	float r, g, b;
-	r = (random() / (double)RAND_MAX);
-	g = (random() / (double)RAND_MAX);
-	b = (random() / (double)RAND_MAX);
-	CHECK_DMA_BUFFER(gCC, gCCPriv, 4);
-	WRITEF(gCCPriv->buf, Ca,  v[3]);
-	WRITEF(gCCPriv->buf, Cb,  b);
-	WRITEF(gCCPriv->buf, Cg,  g);
-	WRITEF(gCCPriv->buf, Cr3, r);
-    }
-#else
     CHECK_DMA_BUFFER(gCC, gCCPriv, 4);
     WRITEF(gCCPriv->buf, Ca,  v[3]);
     WRITEF(gCCPriv->buf, Cb,  v[2]);
     WRITEF(gCCPriv->buf, Cg,  v[1]);
     WRITEF(gCCPriv->buf, Cr4, v[0]);
-#endif
 }
 
 void _gamma_Color4i(GLint red, GLint green, GLint blue, GLint alpha)
 {
+    GLfloat r,g,b,a;
+
     DEBUG_GLCMDS(("Color4i: %d %d %d %d\n", (int)red, (int)green, (int)blue,
 		  (int)alpha));
+
+    r = INT_TO_FLOAT(red);
+    g = INT_TO_FLOAT(green);
+    b = INT_TO_FLOAT(blue);
+    a = INT_TO_FLOAT(alpha);
+
+    _gamma_Color4f(r,g,b,a);
 }
 
 void _gamma_Color4iv(const GLint *v)
 {
+    GLfloat p[4];
+
     DEBUG_GLCMDS(("Color4iv: %d %d %d %d\n", (int)v[0], (int)v[1], (int)v[2],
 		  (int)v[3]));
+
+    p[0] = INT_TO_FLOAT(v[0]);
+    p[1] = INT_TO_FLOAT(v[1]);
+    p[2] = INT_TO_FLOAT(v[2]);
+    p[3] = INT_TO_FLOAT(v[3]);
+
+    _gamma_Color4fv(p);
 }
 
 void _gamma_Color4s(GLshort red, GLshort green, GLshort blue, GLshort alpha)
@@ -826,11 +932,7 @@ void _gamma_Color4ub(GLubyte red, GLubyte green, GLubyte blue, GLubyte alpha)
 
     DEBUG_GLCMDS(("Color4ub: %d %d %d %d\n", red, green, blue, alpha));
 
-#ifdef RANDOMIZE_COLORS
-    c = (random() / (double)RAND_MAX) * 16777216;
-#else
     c = (alpha << 24) | (blue << 16) | (green << 8) | red;
-#endif
 
     CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
     WRITE(gCCPriv->buf, PackedColor4,  c);
@@ -842,17 +944,7 @@ void _gamma_Color4ubv(const GLubyte *v)
 
     DEBUG_GLCMDS(("Color4ubv: %d %d %d %d\n", v[0], v[1], v[2], v[3]));
 
-#ifdef RANDOMIZE_COLORS
-    c = (random() / (double)RAND_MAX) * 16777216;
-#else
-/* NOT_DONE: Is there a standard define for endianness? */
-#define IS_LITTLE_ENDIAN 1
-#if IS_LITTLE_ENDIAN
-    c = *((GLuint *)v);
-#else
     c = (v[3] << 24) | (v[2] << 16) | (v[1] << 8) | v[0];
-#endif
-#endif
 
     CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
     WRITE(gCCPriv->buf, PackedColor4,  c);
@@ -890,6 +982,43 @@ void _gamma_ColorMask(GLboolean red, GLboolean green, GLboolean blue, GLboolean
 void _gamma_ColorMaterial(GLenum face, GLenum mode)
 {
     DEBUG_GLCMDS(("ColorMaterial: %04x %04x\n", (int)face, (int)mode));
+
+    gCCPriv->MaterialMode &= ~MaterialModeEnable;
+    gCCPriv->ColorMaterialMode = ColorMaterialModeDisable;
+
+    switch (face) {
+    case GL_FRONT:
+	gCCPriv->ColorMaterialMode |= ColorMaterialModeFront;
+	break;
+    case GL_BACK:
+	gCCPriv->ColorMaterialMode |= ColorMaterialModeBack;
+	break;
+    case GL_FRONT_AND_BACK:
+	gCCPriv->ColorMaterialMode |= ColorMaterialModeFrontAndBack;
+	break;
+    }
+
+    switch (mode) {
+    case GL_AMBIENT:
+	gCCPriv->ColorMaterialMode |= ColorMaterialModeAmbient;
+	break;
+    case GL_EMISSION:
+	gCCPriv->ColorMaterialMode |= ColorMaterialModeEmission;
+	break;
+    case GL_DIFFUSE:
+	gCCPriv->ColorMaterialMode |= ColorMaterialModeDiffuse;
+	break;
+    case GL_SPECULAR:
+	gCCPriv->ColorMaterialMode |= ColorMaterialModeSpecular;
+	break;
+    case GL_AMBIENT_AND_DIFFUSE:
+	gCCPriv->ColorMaterialMode |= ColorMaterialModeAmbAndDiff;
+	break;
+    }
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 2);
+    WRITE(gCCPriv->buf, ColorMaterialMode, gCCPriv->ColorMaterialMode);
+    WRITE(gCCPriv->buf, MaterialMode, gCCPriv->MaterialMode);
 }
 
 void _gamma_ColorPointer(GLint size, GLenum type, GLsizei stride, const GLvoid *pointer)
@@ -938,9 +1067,6 @@ void _gamma_CullFace(GLenum mode)
 
     gCCPriv->GeometryMode &= ~GM_PolyCullMask;
 
-#ifdef CULL_ALL_PRIMS
-    gCCPriv->GeometryMode |= GM_PolyCullBoth;
-#else
     switch (mode) {
     case GL_FRONT:
 	gCCPriv->GeometryMode |= GM_PolyCullFront;
@@ -955,7 +1081,7 @@ void _gamma_CullFace(GLenum mode)
 	/* ERROR!! */
 	break;
     }
-#endif
+
     CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
     WRITE(gCCPriv->buf, GeometryMode, gCCPriv->GeometryMode);
 }
@@ -980,7 +1106,8 @@ void _gamma_DeleteTextures(GLsizei n, const GLuint *textures)
 
     for (i = 0; i < n; i++) {
 	gammaTexObj *t = gammaTOFind(textures[i]);
-	if (!driTMMDeleteImages(gCCPriv->tmm, MIPMAP_LEVELS, t->image)) {
+	if (driTMMDeleteImages(gCCPriv->tmm, MIPMAP_LEVELS, t->image) < 0) {
+    	    DEBUG_ERROR(("DeleteTextures: unable\n"));
 	    /* NOT_DONE: Handle error */
 	}
 	gammaTODelete(textures[i]);
@@ -1068,11 +1195,9 @@ void _gamma_Disable(GLenum cap)
 
     switch (cap) {
     case GL_CULL_FACE:
-#ifdef CULL_ALL_PRIMS
 	gCCPriv->GeometryMode &= ~GM_PolyCullEnable;
 	CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
 	WRITE(gCCPriv->buf, GeometryMode, gCCPriv->GeometryMode);
-#endif
 	break;
     case GL_DEPTH_TEST:
 	if (gCCPriv->Flags & GAMMA_DEPTH_BUFFER) {
@@ -1106,6 +1231,35 @@ void _gamma_Disable(GLenum cap)
 	WRITE(gCCPriv->buf, FBReadMode, (gCCPriv->FBReadMode |
 					 gCCPriv->AB_FBReadMode));
 	break;
+    case GL_COLOR_MATERIAL:
+	gCCPriv->ColorMaterialMode &= ~ColorMaterialModeEnable;
+	CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+	WRITE(gCCPriv->buf, ColorMaterialMode, gCCPriv->ColorMaterialMode);
+	break;
+    case GL_FOG:
+	gCCPriv->Begin &= ~B_FogEnable;
+	gCCPriv->GeometryMode &= ~GM_FogEnable;
+	gCCPriv->DeltaMode &= ~DM_FogEnable;
+	CHECK_DMA_BUFFER(gCC, gCCPriv, 2);
+	WRITE(gCCPriv->buf, GeometryMode, gCCPriv->GeometryMode);
+	WRITE(gCCPriv->buf, DeltaMode, gCCPriv->DeltaMode);
+	WRITE(gCCPriv->buf, FogMode, FogModeDisable);
+	break;
+    case GL_LIGHTING:
+	gCCPriv->LightingMode &= ~LightingModeEnable;
+	CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+	WRITE(gCCPriv->buf, LightingMode, gCCPriv->LightingMode);
+	break;
+    case GL_LIGHT0:
+	gCCPriv->Light0Mode &= ~LNM_On;
+	CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+	WRITE(gCCPriv->buf, Light0Mode, gCCPriv->Light0Mode);
+	break;
+    case GL_LIGHT1:
+	gCCPriv->Light1Mode &= ~LNM_On;
+	CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+	WRITE(gCCPriv->buf, Light1Mode, gCCPriv->Light1Mode);
+	break;
     case GL_TEXTURE_2D:
 	gCCPriv->Texture2DEnabled = GL_FALSE;
 	gCCPriv->Begin                         &= ~B_TextureEnable;
@@ -1227,6 +1381,35 @@ void _gamma_Enable(GLenum cap)
 	WRITE(gCCPriv->buf, FBReadMode, (gCCPriv->FBReadMode |
 					 gCCPriv->AB_FBReadMode));
 	break;
+    case GL_COLOR_MATERIAL:
+	gCCPriv->ColorMaterialMode |= ColorMaterialModeEnable;
+	CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+	WRITE(gCCPriv->buf, ColorMaterialMode, gCCPriv->ColorMaterialMode);
+	break;
+    case GL_FOG:
+	gCCPriv->Begin |= B_FogEnable;
+	gCCPriv->GeometryMode |= GM_FogEnable;
+	gCCPriv->DeltaMode |= DM_FogEnable;
+	CHECK_DMA_BUFFER(gCC, gCCPriv, 2);
+	WRITE(gCCPriv->buf, GeometryMode, gCCPriv->GeometryMode);
+	WRITE(gCCPriv->buf, DeltaMode, gCCPriv->DeltaMode);
+	WRITE(gCCPriv->buf, FogMode, FogModeEnable);
+	break;
+    case GL_LIGHTING:
+	gCCPriv->LightingMode |= LightingModeEnable | 16<<6;
+	CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+	WRITE(gCCPriv->buf, LightingMode, gCCPriv->LightingMode);
+	break;
+    case GL_LIGHT0:
+	gCCPriv->Light0Mode |= LNM_On;
+	CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+	WRITE(gCCPriv->buf, Light0Mode, gCCPriv->Light0Mode);
+	break;
+    case GL_LIGHT1:
+	gCCPriv->Light1Mode |= LNM_On;
+	CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+	WRITE(gCCPriv->buf, Light1Mode, gCCPriv->Light1Mode);
+	break;
     case GL_TEXTURE_2D:
 	gCCPriv->Texture2DEnabled = GL_TRUE;
 #ifndef TURN_OFF_TEXTURES
@@ -1294,34 +1477,42 @@ void _gamma_EvalCoord1d(GLdouble u)
 
 void _gamma_EvalCoord1dv(const GLdouble *u)
 {
+    DEBUG_GLCMDS(("EvalCoord1dv: %f\n", *u));
 }
 
 void _gamma_EvalCoord1f(GLfloat u)
 {
+    DEBUG_GLCMDS(("EvalCoord1f: %f\n", u));
 }
 
 void _gamma_EvalCoord1fv(const GLfloat *u)
 {
+    DEBUG_GLCMDS(("EvalCoord1fv: %f\n", *u));
 }
 
 void _gamma_EvalCoord2d(GLdouble u, GLdouble v)
 {
+    DEBUG_GLCMDS(("EvalCoord2d: %f %f\n", u, v));
 }
 
 void _gamma_EvalCoord2dv(const GLdouble *u)
 {
+    DEBUG_GLCMDS(("EvalCoord2dv: %f %f\n", u[0], u[1]));
 }
 
 void _gamma_EvalCoord2f(GLfloat u, GLfloat v)
 {
+    DEBUG_GLCMDS(("EvalCoord2f: %f %f\n", u, v));
 }
 
 void _gamma_EvalCoord2fv(const GLfloat *u)
 {
+    DEBUG_GLCMDS(("EvalCoord1fv: %f %f\n", u[0], u[1]));
 }
 
 void _gamma_EvalMesh1(GLenum mode, GLint i1, GLint i2)
 {
+    DEBUG_GLCMDS(("EvalMesh1: %d %d %d\n", mode, i1, i2));
 }
 
 void _gamma_EvalMesh2(GLenum mode, GLint i1, GLint i2, GLint j1, GLint j2)
@@ -1358,16 +1549,55 @@ void _gamma_Flush(void)
 void _gamma_Fogf(GLenum pname, GLfloat param)
 {
     DEBUG_GLCMDS(("Fogf: %04x %f\n", (int)pname, param));
+
+    switch (pname) {
+    case GL_FOG_DENSITY:
+    	CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+	WRITEF(gCCPriv->buf, FogDensity, param);
+	break;
+    }
 }
 
 void _gamma_Fogfv(GLenum pname, const GLfloat *params)
 {
+    int color;
+
     DEBUG_GLCMDS(("Fogfv: %04x %f\n", (int)pname, *params));
+
+    switch (pname) {
+    case GL_FOG_COLOR:
+	color = ((int)params[3]<<24) | ((int)params[2]<<16) |
+	 	((int)params[1]<<8)  | ((int)params[0]);
+    	CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+	WRITE(gCCPriv->buf, FogColor, color);
+	break;
+    }
 }
 
 void _gamma_Fogi(GLenum pname, GLint param)
 {
     DEBUG_GLCMDS(("Fogi: %04x %d\n", (int)pname, (int)param));
+
+    gCCPriv->GeometryMode &= ~GM_FogMask;
+
+    switch (pname) {
+    case GL_FOG_MODE:
+	switch (param) {
+	case GL_EXP:
+	    gCCPriv->GeometryMode |= GM_FogExp;
+	    break;
+	case GL_EXP2:
+	    gCCPriv->GeometryMode |= GM_FogExpSquared;
+	    break;
+	case GL_LINEAR:
+	    gCCPriv->GeometryMode |= GM_FogLinear;
+	    break;
+	}
+ 	break;
+    }
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+    WRITE(gCCPriv->buf, GeometryMode, gCCPriv->GeometryMode);
 }
 
 void _gamma_Fogiv(GLenum pname, const GLint *params)
@@ -1378,6 +1608,16 @@ void _gamma_Fogiv(GLenum pname, const GLint *params)
 void _gamma_FrontFace(GLenum mode)
 {
     DEBUG_GLCMDS(("FrontFace: %04x\n", (int)mode));
+
+    gCCPriv->GeometryMode &= ~GM_FFMask;
+
+    if (mode == GL_CCW)
+	gCCPriv->GeometryMode |= GM_FrontFaceCCW;
+    else
+	gCCPriv->GeometryMode |= GM_FrontFaceCW;
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+    WRITE(gCCPriv->buf, GeometryMode, gCCPriv->GeometryMode);
 }
 
 void _gamma_Frustum(GLdouble left, GLdouble right, GLdouble bottom, GLdouble top, GLdouble zNear, GLdouble zFar)
@@ -1463,14 +1703,24 @@ void _gamma_GetFloatv(GLenum val, GLfloat *f)
 void _gamma_GetIntegerv(GLenum val, GLint *i)
 {
     DEBUG_GLCMDS(("GetIntegerv: %04x\n", (int)val));
+
+    switch (val) {
+    case GL_MAX_TEXTURE_SIZE:
+	*i = 2048;
+	break;
+    }
 }
 
 void _gamma_GetLightfv(GLenum light, GLenum pname, GLfloat *params)
 {
+    DEBUG_GLCMDS(("GetLightfv: %04x %04x %f\n",
+		  (int)light, (int)pname, *params));
 }
 
 void _gamma_GetLightiv(GLenum light, GLenum pname, GLint *params)
 {
+    DEBUG_GLCMDS(("GetLightiv: %04x %04x %d\n",
+		  (int)light, (int)pname, *params));
 }
 
 void _gamma_GetMapdv(GLenum target, GLenum query, GLdouble *v)
@@ -1654,6 +1904,12 @@ GLboolean _gamma_IsEnabled(GLenum cap)
 {
     DEBUG_GLCMDS(("IsEnabled: %04x\n", (int)cap));
 
+    switch (cap) {
+    case GL_LIGHTING:
+	return ((gCCPriv->LightingMode & LightingModeEnable)?GL_TRUE:GL_FALSE);
+        break;
+    }
+
     return GL_TRUE;
 }
 
@@ -1673,10 +1929,44 @@ GLboolean _gamma_IsTexture(GLuint texture)
 
 void _gamma_LightModelf(GLenum pname, GLfloat param)
 {
+    DEBUG_GLCMDS(("LightModelf: %04x %f\n",
+		  (int)pname, param));
 }
 
 void _gamma_LightModelfv(GLenum pname, const GLfloat *params)
 {
+    DEBUG_GLCMDS(("LightModelfv: %04x %f\n",
+		  (int)pname, *params));
+
+    switch (pname) {
+    case GL_LIGHT_MODEL_AMBIENT:
+	CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
+	/* We don't do alpha */
+	WRITEF(gCCPriv->buf, SceneAmbientColorBlue, params[2]);
+	WRITEF(gCCPriv->buf, SceneAmbientColorGreen, params[1]);
+	WRITEF(gCCPriv->buf, SceneAmbientColorRed, params[0]);
+	break;
+    case GL_LIGHT_MODEL_LOCAL_VIEWER:
+	if ((int)params[0] != 0)
+	    gCCPriv->LightingMode |= LightingModeLocalViewer;
+	else
+	    gCCPriv->LightingMode &= ~LightingModeLocalViewer;
+	CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+	WRITE(gCCPriv->buf, LightingMode, gCCPriv->LightingMode);
+	break;
+    case GL_LIGHT_MODEL_TWO_SIDE:
+	if ((int)params[0] != 0) {
+	    gCCPriv->LightingMode |= LightingModeTwoSides;
+	    gCCPriv->MaterialMode |= MaterialModeTwoSides;
+	} else {
+	    gCCPriv->LightingMode &= ~LightingModeTwoSides;
+	    gCCPriv->MaterialMode &= ~MaterialModeTwoSides;
+	}
+	CHECK_DMA_BUFFER(gCC, gCCPriv, 2);
+	WRITE(gCCPriv->buf, LightingMode, gCCPriv->LightingMode);
+	WRITE(gCCPriv->buf, MaterialMode, gCCPriv->MaterialMode);
+	break;
+    }
 }
 
 void _gamma_LightModeli(GLenum pname, GLint param)
@@ -1689,10 +1979,173 @@ void _gamma_LightModeliv(GLenum pname, const GLint *params)
 
 void _gamma_Lightf(GLenum light, GLenum pname, GLfloat param)
 {
+    DEBUG_GLCMDS(("Lightf: %04x %04x %f\n",
+		  (int)light, (int)pname, param));
 }
 
 void _gamma_Lightfv(GLenum light, GLenum pname, const GLfloat *params)
 {
+    GLfloat l,x,y,z;
+    DEBUG_GLCMDS(("Lightfv: %04x %04x %f\n",
+		  (int)light, (int)pname, *params));
+
+    switch(light) {
+    case GL_LIGHT0:
+	switch (pname) {
+	case GL_AMBIENT:
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
+	    /* We don't do alpha */
+	    WRITEF(gCCPriv->buf, Light0AmbientIntensityBlue, params[2]);
+	    WRITEF(gCCPriv->buf, Light0AmbientIntensityGreen, params[1]);
+	    WRITEF(gCCPriv->buf, Light0AmbientIntensityRed, params[0]);
+	    break;
+	case GL_DIFFUSE:
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
+	    /* We don't do alpha */
+	    WRITEF(gCCPriv->buf, Light0DiffuseIntensityBlue, params[2]);
+	    WRITEF(gCCPriv->buf, Light0DiffuseIntensityGreen, params[1]);
+	    WRITEF(gCCPriv->buf, Light0DiffuseIntensityRed, params[0]);
+	    break;
+	case GL_SPECULAR:
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
+	    /* We don't do alpha */
+	    WRITEF(gCCPriv->buf, Light0SpecularIntensityBlue, params[2]);
+	    WRITEF(gCCPriv->buf, Light0SpecularIntensityGreen, params[1]);
+	    WRITEF(gCCPriv->buf, Light0SpecularIntensityRed, params[0]);
+	    break;
+	case GL_POSITION:
+    	    /* Normalize <x,y,z> */
+	    x = params[0]; y = params[1]; z = params[2];
+	    l = sqrt(x*x + y*y + z*z);
+	    x /= l;
+	    y /= l;
+	    z /= l;
+	    if ((int)params[3] != 0) {
+		gCCPriv->Light0Mode |= Light0ModeAttenuation;
+		gCCPriv->Light0Mode |= Light0ModeLocal;
+	    } else {
+		gCCPriv->Light0Mode &= ~Light0ModeAttenuation;
+		gCCPriv->Light0Mode &= ~Light0ModeLocal;
+	    }
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 5);
+	    WRITE(gCCPriv->buf, Light0Mode, gCCPriv->Light0Mode);
+	    WRITEF(gCCPriv->buf, Light0PositionW, params[3]);
+	    WRITEF(gCCPriv->buf, Light0PositionZ, z);
+	    WRITEF(gCCPriv->buf, Light0PositionY, y);
+	    WRITEF(gCCPriv->buf, Light0PositionX, x);
+	    break;
+	case GL_SPOT_DIRECTION:
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
+	    WRITEF(gCCPriv->buf, Light0SpotlightDirectionZ, params[3]);
+	    WRITEF(gCCPriv->buf, Light0SpotlightDirectionY, params[2]);
+	    WRITEF(gCCPriv->buf, Light0SpotlightDirectionX, params[1]);
+	    /* WRITEF(gCCPriv->buf, Light0SpotlightDirectionW, params[0]); */
+	    break;
+	case GL_SPOT_EXPONENT:
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+	    WRITEF(gCCPriv->buf, Light0SpotlightExponent, params[0]);
+	    break;
+	case GL_SPOT_CUTOFF:
+	    if ((int)params[0] != -1) 
+		gCCPriv->Light0Mode |= Light0ModeSpotLight;
+	    else
+		gCCPriv->Light0Mode &= ~Light0ModeSpotLight;
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 2);
+	    WRITE(gCCPriv->buf, Light0Mode, gCCPriv->Light0Mode);
+	    WRITEF(gCCPriv->buf, Light0CosSpotlightCutoffAngle, params[0]);
+	    break;
+	case GL_CONSTANT_ATTENUATION:
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+	    WRITEF(gCCPriv->buf, Light0ConstantAttenuation, params[0]);
+	    break;
+	case GL_LINEAR_ATTENUATION:
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+	    WRITEF(gCCPriv->buf, Light0LinearAttenuation, params[0]);
+	    break;
+	case GL_QUADRATIC_ATTENUATION:
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+	    WRITEF(gCCPriv->buf, Light0QuadraticAttenuation, params[0]);
+	    break;
+	}
+	break;
+    case GL_LIGHT1:
+	switch (pname) {
+	case GL_AMBIENT:
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
+	    /* We don't do alpha */
+	    WRITEF(gCCPriv->buf, Light1AmbientIntensityBlue, params[2]);
+	    WRITEF(gCCPriv->buf, Light1AmbientIntensityGreen, params[1]);
+	    WRITEF(gCCPriv->buf, Light1AmbientIntensityRed, params[0]);
+	    break;
+	case GL_DIFFUSE:
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
+	    /* We don't do alpha */
+	    WRITEF(gCCPriv->buf, Light1DiffuseIntensityBlue, params[2]);
+	    WRITEF(gCCPriv->buf, Light1DiffuseIntensityGreen, params[1]);
+	    WRITEF(gCCPriv->buf, Light1DiffuseIntensityRed, params[0]);
+	    break;
+	case GL_SPECULAR:
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
+	    /* We don't do alpha */
+	    WRITEF(gCCPriv->buf, Light1SpecularIntensityBlue, params[2]);
+	    WRITEF(gCCPriv->buf, Light1SpecularIntensityGreen, params[1]);
+	    WRITEF(gCCPriv->buf, Light1SpecularIntensityRed, params[0]);
+	    break;
+	case GL_POSITION:
+    	    /* Normalize <x,y,z> */
+	    x = params[0]; y = params[1]; z = params[2];
+	    l = sqrt(x*x + y*y + z*z);
+	    x /= l;
+	    y /= l;
+	    z /= l;
+	    if ((int)params[3] != 0) {
+		gCCPriv->Light1Mode |= Light1ModeAttenuation;
+		gCCPriv->Light1Mode |= Light1ModeLocal;
+	    } else {
+		gCCPriv->Light0Mode &= ~Light0ModeAttenuation;
+		gCCPriv->Light0Mode &= ~Light0ModeLocal;
+	    }
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 5);
+	    WRITE(gCCPriv->buf, Light1Mode, gCCPriv->Light1Mode);
+	    WRITEF(gCCPriv->buf, Light1PositionW, params[3]);
+	    WRITEF(gCCPriv->buf, Light1PositionZ, z);
+	    WRITEF(gCCPriv->buf, Light1PositionY, y);
+	    WRITEF(gCCPriv->buf, Light1PositionX, x);
+	    break;
+	case GL_SPOT_DIRECTION:
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
+	    WRITEF(gCCPriv->buf, Light1SpotlightDirectionZ, params[3]);
+	    WRITEF(gCCPriv->buf, Light1SpotlightDirectionY, params[2]);
+	    WRITEF(gCCPriv->buf, Light1SpotlightDirectionX, params[1]);
+	    /* WRITEF(gCCPriv->buf, Light1SpotlightDirectionW, params[0]); */
+	    break;
+	case GL_SPOT_EXPONENT:
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+	    WRITEF(gCCPriv->buf, Light1SpotlightExponent, params[0]);
+	    break;
+	case GL_SPOT_CUTOFF:
+	    if ((int)params[0] != -1) 
+		gCCPriv->Light1Mode |= Light1ModeSpotLight;
+	    else
+		gCCPriv->Light1Mode &= ~Light1ModeSpotLight;
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 2);
+	    WRITE(gCCPriv->buf, Light1Mode, gCCPriv->Light1Mode);
+	    WRITEF(gCCPriv->buf, Light1CosSpotlightCutoffAngle, params[0]);
+	    break;
+	case GL_CONSTANT_ATTENUATION:
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+	    WRITEF(gCCPriv->buf, Light1ConstantAttenuation, params[0]);
+	    break;
+	case GL_LINEAR_ATTENUATION:
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+	    WRITEF(gCCPriv->buf, Light1LinearAttenuation, params[0]);
+	    break;
+	case GL_QUADRATIC_ATTENUATION:
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+	    WRITEF(gCCPriv->buf, Light1QuadraticAttenuation, params[0]);
+	    break;
+	}
+    }
 }
 
 void _gamma_Lighti(GLenum light, GLenum pname, GLint param)
@@ -1817,6 +2270,104 @@ void _gamma_Materialfv(GLenum face, GLenum pname, const GLfloat *params)
 {
     DEBUG_GLCMDS(("Materialfv: %04x %04x %f\n",
 		  (int)face, (int)pname, *params));
+
+    gCCPriv->MaterialMode |= MaterialModeEnable;
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+    WRITE(gCCPriv->buf, MaterialMode, gCCPriv->MaterialMode);
+
+    if ((face == GL_FRONT) || (face == GL_FRONT_AND_BACK)) {
+	switch (pname) {
+	case GL_AMBIENT:
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
+	    WRITEF(gCCPriv->buf, FrontAmbientColorBlue, params[2]);
+	    WRITEF(gCCPriv->buf, FrontAmbientColorGreen, params[1]);
+	    WRITEF(gCCPriv->buf, FrontAmbientColorRed, params[0]);
+	    break;
+	case GL_DIFFUSE:
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 4);
+	    WRITEF(gCCPriv->buf, FrontAlpha, params[3]);
+	    WRITEF(gCCPriv->buf, FrontDiffuseColorBlue, params[2]);
+	    WRITEF(gCCPriv->buf, FrontDiffuseColorGreen, params[1]);
+	    WRITEF(gCCPriv->buf, FrontDiffuseColorRed, params[0]);
+	    break;
+	case GL_SPECULAR:
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
+	    WRITEF(gCCPriv->buf, FrontSpecularColorBlue, params[2]);
+	    WRITEF(gCCPriv->buf, FrontSpecularColorGreen, params[1]);
+	    WRITEF(gCCPriv->buf, FrontSpecularColorRed, params[0]);
+	    break;
+	case GL_EMISSION:
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
+	    WRITEF(gCCPriv->buf, FrontEmissiveColorBlue, params[2]);
+	    WRITEF(gCCPriv->buf, FrontEmissiveColorGreen, params[1]);
+	    WRITEF(gCCPriv->buf, FrontEmissiveColorRed, params[0]);
+	    break;
+	case GL_SHININESS:
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+	    WRITEF(gCCPriv->buf, FrontSpecularExponent, params[0]);
+	    break;
+	case GL_AMBIENT_AND_DIFFUSE:
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 7);
+	    WRITEF(gCCPriv->buf, FrontAmbientColorBlue, params[2]);
+	    WRITEF(gCCPriv->buf, FrontAmbientColorGreen, params[1]);
+	    WRITEF(gCCPriv->buf, FrontAmbientColorRed, params[0]);
+	    WRITEF(gCCPriv->buf, FrontAlpha, params[3]);
+	    WRITEF(gCCPriv->buf, FrontDiffuseColorBlue, params[2]);
+	    WRITEF(gCCPriv->buf, FrontDiffuseColorGreen, params[1]);
+	    WRITEF(gCCPriv->buf, FrontDiffuseColorRed, params[0]);
+	    break;
+	case GL_COLOR_INDEXES:
+	    /* NOT_DONE */
+	    break;
+	}
+    }
+
+    if ((face == GL_BACK) || (face == GL_FRONT_AND_BACK)) {
+	switch (pname) {
+	case GL_AMBIENT:
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
+	WRITEF(gCCPriv->buf, BackAmbientColorBlue, params[2]);
+	WRITEF(gCCPriv->buf, BackAmbientColorGreen, params[1]);
+	WRITEF(gCCPriv->buf, BackAmbientColorRed, params[0]);
+	break;
+    case GL_DIFFUSE:
+	CHECK_DMA_BUFFER(gCC, gCCPriv, 4);
+	WRITEF(gCCPriv->buf, BackAlpha, params[3]);
+	WRITEF(gCCPriv->buf, BackDiffuseColorBlue, params[2]);
+	WRITEF(gCCPriv->buf, BackDiffuseColorGreen, params[1]);
+	WRITEF(gCCPriv->buf, BackDiffuseColorRed, params[0]);
+	break;
+    case GL_SPECULAR:
+	CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
+	WRITEF(gCCPriv->buf, BackSpecularColorBlue, params[2]);
+	WRITEF(gCCPriv->buf, BackSpecularColorGreen, params[1]);
+	WRITEF(gCCPriv->buf, BackSpecularColorRed, params[0]);
+	break;
+    case GL_EMISSION:
+	CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
+	WRITEF(gCCPriv->buf, BackEmissiveColorBlue, params[2]);
+	WRITEF(gCCPriv->buf, BackEmissiveColorGreen, params[1]);
+	WRITEF(gCCPriv->buf, BackEmissiveColorRed, params[0]);
+	break;
+    case GL_SHININESS:
+	CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+	WRITEF(gCCPriv->buf, BackSpecularExponent, params[0]);
+	break;
+    case GL_AMBIENT_AND_DIFFUSE:
+	CHECK_DMA_BUFFER(gCC, gCCPriv, 7);
+	WRITEF(gCCPriv->buf, BackAmbientColorBlue, params[2]);
+	WRITEF(gCCPriv->buf, BackAmbientColorGreen, params[1]);
+	WRITEF(gCCPriv->buf, BackAmbientColorRed, params[0]);
+	WRITEF(gCCPriv->buf, BackAlpha, params[3]);
+	WRITEF(gCCPriv->buf, BackDiffuseColorBlue, params[2]);
+	WRITEF(gCCPriv->buf, BackDiffuseColorGreen, params[1]);
+	WRITEF(gCCPriv->buf, BackDiffuseColorRed, params[0]);
+	break;
+    case GL_COLOR_INDEXES:
+	/* NOT_DONE */
+	break;
+    }
+    }
 }
 
 void _gamma_Materiali(GLenum face, GLenum pname, GLint param)
@@ -1837,9 +2388,9 @@ void _gamma_MatrixMode(GLenum mode)
 
     switch (mode) {
     case GL_TEXTURE:
-	/* Eanble the Texture transform */
+	/* Enable the Texture transform */
 	CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
-	WRITE(gCCPriv->buf, TransformModeOr, 0x00000010);
+	WRITE(gCCPriv->buf, TransformModeOr, XM_XformTexture);
     case GL_MODELVIEW:
     case GL_PROJECTION:
         gCCPriv->MatrixMode = mode;
@@ -1878,52 +2429,114 @@ void _gamma_NewList(GLuint list, GLenum mode)
 
 void _gamma_Normal3b(GLbyte nx, GLbyte ny, GLbyte nz)
 {
+    GLfloat x,y,z;
+
     DEBUG_GLCMDS(("Normal3b: %d %d %d\n", nx, ny, nz));
+
+    x = BYTE_TO_FLOAT(nx);
+    y = BYTE_TO_FLOAT(ny);
+    z = BYTE_TO_FLOAT(nz);
+   
+    _gamma_Normal3f(x,y,z);
 }
 
 void _gamma_Normal3bv(const GLbyte *v)
 {
+    GLfloat p[3];
+
     DEBUG_GLCMDS(("Normal3bv: %d %d %d\n", v[0], v[1], v[2]));
+
+    p[0] = BYTE_TO_FLOAT(v[0]);
+    p[1] = BYTE_TO_FLOAT(v[1]);
+    p[2] = BYTE_TO_FLOAT(v[2]);
+   
+    _gamma_Normal3fv(p);
 }
 
 void _gamma_Normal3d(GLdouble nx, GLdouble ny, GLdouble nz)
 {
     DEBUG_GLCMDS(("Normal3d: %f %f %f\n", nx, ny, nz));
+
+    _gamma_Normal3f((GLfloat)nx,(GLfloat)ny,(GLfloat)nz);
 }
 
 void _gamma_Normal3dv(const GLdouble *v)
 {
     DEBUG_GLCMDS(("Normal3dv: %f %f %f\n", v[0], v[1], v[2]));
+
+    _gamma_Normal3fv((GLfloat*)v);
 }
 
 void _gamma_Normal3f(GLfloat nx, GLfloat ny, GLfloat nz)
 {
     DEBUG_GLCMDS(("Normal3f: %f %f %f\n", nx, ny, nz));
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
+    WRITEF(gCCPriv->buf, Nz, nz);
+    WRITEF(gCCPriv->buf, Ny, ny);
+    WRITEF(gCCPriv->buf, Nx, nx);
 }
 
 void _gamma_Normal3fv(const GLfloat *v)
 {
     DEBUG_GLCMDS(("Normal3fv: %f %f %f\n", v[0], v[1], v[2]));
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
+    WRITEF(gCCPriv->buf, Nz, v[2]);
+    WRITEF(gCCPriv->buf, Ny, v[1]);
+    WRITEF(gCCPriv->buf, Nx, v[0]);
 }
 
 void _gamma_Normal3i(GLint nx, GLint ny, GLint nz)
 {
+    GLfloat x,y,z;
+
     DEBUG_GLCMDS(("Normal3i: %d %d %d\n", (int)nx, (int)ny, (int)nz));
+
+    x = INT_TO_FLOAT(nx);
+    y = INT_TO_FLOAT(ny);
+    z = INT_TO_FLOAT(nz);
+
+    _gamma_Normal3f(x,y,z);
 }
 
 void _gamma_Normal3iv(const GLint *v)
 {
+    GLfloat p[3];
+
     DEBUG_GLCMDS(("Normal3iv: %d %d %d\n", (int)v[0], (int)v[1], (int)v[2]));
+
+    p[0] = INT_TO_FLOAT(v[0]);
+    p[1] = INT_TO_FLOAT(v[1]);
+    p[2] = INT_TO_FLOAT(v[2]);
+
+    _gamma_Normal3fv(p);
 }
 
 void _gamma_Normal3s(GLshort nx, GLshort ny, GLshort nz)
 {
+    GLfloat x,y,z;
+
     DEBUG_GLCMDS(("Normal3s: %d %d %d\n", nx, ny, nz));
+
+    x = SHORT_TO_FLOAT(nx);
+    y = SHORT_TO_FLOAT(ny);
+    z = SHORT_TO_FLOAT(nz);
+
+    _gamma_Normal3f(x,y,z);
 }
 
 void _gamma_Normal3sv(const GLshort *v)
 {
+    GLfloat p[3];
+
     DEBUG_GLCMDS(("Normal3sv: %d %d %d\n", v[0], v[1], v[2]));
+
+    p[0] = SHORT_TO_FLOAT(v[0]);
+    p[1] = SHORT_TO_FLOAT(v[1]);
+    p[2] = SHORT_TO_FLOAT(v[2]);
+
+    _gamma_Normal3fv(p);
 }
 
 void _gamma_NormalPointer(GLenum type, GLsizei stride, const GLvoid *pointer)
@@ -2020,6 +2633,56 @@ void _gamma_PointSize(GLfloat size)
 void _gamma_PolygonMode(GLenum face, GLenum mode)
 {
     DEBUG_GLCMDS(("PolygonMode: %04x %04x\n", (int)face, (int)mode));
+
+    gCCPriv->GeometryMode &= ~GM_FB_PolyMask;
+
+    switch (mode) {
+    case GL_FILL:
+	switch (face) {
+	case GL_FRONT:
+	    gCCPriv->GeometryMode |= GM_FrontPolyFill;
+	    break;
+	case GL_BACK:
+	    gCCPriv->GeometryMode |= GM_BackPolyFill;
+	    break;
+	case GL_FRONT_AND_BACK:
+	    gCCPriv->GeometryMode |= GM_FrontPolyFill;
+	    gCCPriv->GeometryMode |= GM_BackPolyFill;
+	    break;
+	}
+	break;
+    case GL_LINE:
+	switch (face) {
+	case GL_FRONT:
+	    gCCPriv->GeometryMode |= GM_FrontPolyLine;
+	    break;
+	case GL_BACK:
+	    gCCPriv->GeometryMode |= GM_BackPolyLine;
+	    break;
+	case GL_FRONT_AND_BACK:
+	    gCCPriv->GeometryMode |= GM_FrontPolyLine;
+	    gCCPriv->GeometryMode |= GM_BackPolyLine;
+	    break;
+	}
+	break;
+    case GL_POINT:
+	switch (face) {
+	case GL_FRONT:
+	    gCCPriv->GeometryMode |= GM_FrontPolyPoint;
+	    break;
+	case GL_BACK:
+	    gCCPriv->GeometryMode |= GM_BackPolyPoint;
+	    break;
+	case GL_FRONT_AND_BACK:
+	    gCCPriv->GeometryMode |= GM_FrontPolyPoint;
+	    gCCPriv->GeometryMode |= GM_BackPolyPoint;
+	    break;
+	}
+	break;
+    }
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+    WRITE(gCCPriv->buf, GeometryMode, gCCPriv->GeometryMode);
 }
 
 void _gamma_PolygonOffset(GLfloat factor, GLfloat units)
@@ -2272,6 +2935,13 @@ void _gamma_Rectdv(const GLdouble *v1, const GLdouble *v2)
 void _gamma_Rectf(GLfloat x1, GLfloat y1, GLfloat x2, GLfloat y2)
 {
     DEBUG_GLCMDS(("Rectf: %f %f %f %f\n", x1, y1, x2, y2));
+ 
+    _gamma_Begin(GL_POLYGON);
+    _gamma_Vertex2f(x1,y1);
+    _gamma_Vertex2f(x2,y1);
+    _gamma_Vertex2f(x2,y2);
+    _gamma_Vertex2f(x1,y2);
+    _gamma_End();
 }
 
 void _gamma_Rectfv(const GLfloat *v1, const GLfloat *v2)
@@ -2678,6 +3348,33 @@ void _gamma_TexEnvfv(GLenum target, GLenum pname, const GLfloat *params)
 {
     DEBUG_GLCMDS(("TexEnvfv: %04x %04x %f\n",
 		  (int)target, (int)pname, *params));
+
+    if (target != GL_TEXTURE_ENV) {
+	/* ERROR !! */
+    }
+
+    switch (pname) {
+    case GL_TEXTURE_ENV_MODE:
+	gCCPriv->curTexObj->TextureColorMode &= ~TCM_ApplicationMask;
+	switch ((int)params[0]) {
+	case GL_MODULATE:
+	    gCCPriv->curTexObj->TextureColorMode |= TCM_Modulate;
+	    break;
+	case GL_DECAL:
+	    gCCPriv->curTexObj->TextureColorMode |= TCM_Decal;
+	    break;
+	case GL_BLEND:
+	    gCCPriv->curTexObj->TextureColorMode |= TCM_Blend;
+	    break;
+	case GL_REPLACE:
+	    gCCPriv->curTexObj->TextureColorMode |= TCM_Replace;
+	    break;
+	}
+    }
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+    WRITE(gCCPriv->buf, TextureColorMode,
+	  gCCPriv->curTexObj->TextureColorMode);
 }
 
 void _gamma_TexEnvi(GLenum target, GLenum pname, GLint param)
@@ -2745,10 +3442,12 @@ void _gamma_TexImage2D(GLenum target, GLint level, GLint components, GLsizei wid
 
     if (target == GL_TEXTURE_1D) {
 	/* NOT_DONE */
+	DEBUG_ERROR(("TexImage2D: 1D requested - ERROR\n"));
     } else if (target == GL_TEXTURE_2D) {
 	/* NOT_DONE: The follow are not currently supported... */
-	if (border != 0 || format != GL_RGBA || type != GL_UNSIGNED_BYTE ||
+	if (border != 0 /* || format != GL_RGBA */ || type != GL_UNSIGNED_BYTE ||
 	    (components != 3 && components != 4)) {
+	    DEBUG_ERROR(("TexImage2D: 2D op not supported - ERROR\n"));
 	    return;
 	}
 
@@ -2790,7 +3489,11 @@ void _gamma_TexImage2D(GLenum target, GLint level, GLint components, GLsizei wid
 		break;
 	    }
 
+#if X_BYTE_ORDER == X_LITTLE_ENDIAN
 	    t->TextureFormat = (TF_LittleEndian |
+#else
+	    t->TextureFormat = (TF_BigEndian |
+#endif
 				TF_ColorOrder_BGR |
 				TF_Compnents_4 |
 				TF_OutputFmt_Texel);
@@ -2804,14 +3507,17 @@ void _gamma_TexImage2D(GLenum target, GLint level, GLint components, GLsizei wid
 	t->image[level] = driTMMInsertImage(gCCPriv->tmm,
 					    width, height, 1<<l2d,
 					    image, NULL);
+
 	if (!t->image[level]) {
 	    /* NOT_DONE: Handle error */
+	    DEBUG_ERROR(("TexImage2D: unable1\n"));
 	}
 
 	/* Make the new image resident (and all of the other mipmaps) */
-	if (!driTMMMakeImagesResident(gCCPriv->tmm, MIPMAP_LEVELS,
-				      t->image, addrs)) {
+	if (driTMMMakeImagesResident(gCCPriv->tmm, MIPMAP_LEVELS,
+				      t->image, addrs) < 0) {
 	    /* NOT_DONE: Handle error */
+	    DEBUG_ERROR(("TexImage2D: unable2\n"));
 	}
 
 	for (i = 0; i < MIPMAP_LEVELS; i++)
@@ -3010,6 +3716,123 @@ void _gamma_TexParameterfv(GLenum target, GLenum pname, const GLfloat *params)
 {
     DEBUG_GLCMDS(("TexParameterfv: %04x %04x %f\n",
 		  (int)target, (int)pname, *params));
+
+    if (target == GL_TEXTURE_1D) {
+	/* NOT_DONE */
+    } else if (target == GL_TEXTURE_2D) {
+	switch (pname) {
+	case GL_TEXTURE_MAG_FILTER:
+	    gCCPriv->curTexObj2D->TextureReadMode &= ~TRM_Mag_Mask;
+	    switch ((int)params[0]) {
+	    case GL_NEAREST:
+		gCCPriv->curTexObj2D->TextureReadMode |=
+		    TRM_Mag_Nearest;
+		break;
+	    case GL_LINEAR:
+		gCCPriv->curTexObj2D->TextureReadMode |=
+		    TRM_Mag_Linear;
+		break;
+	    default:
+		break;
+	    }
+
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
+	    WRITE(gCCPriv->buf, TextureReadMode,
+		  gCCPriv->curTexObj2D->TextureReadMode);
+	    break;
+	case GL_TEXTURE_MIN_FILTER:
+	    gCCPriv->curTexObj2D->TextureReadMode    &= ~TRM_Min_Mask;
+	    gCCPriv->curTexObj2D->TextureReadMode    |= TRM_MipMapEnable;
+	    gCCPriv->curTexObj2D->TextureAddressMode |= TAM_LODEnable;
+	    switch ((int)params[0]) {
+	    case GL_NEAREST:
+		gCCPriv->curTexObj2D->TextureReadMode |=
+		    TRM_Min_Nearest;
+		gCCPriv->curTexObj2D->TextureReadMode    &= ~TRM_MipMapEnable;
+		gCCPriv->curTexObj2D->TextureAddressMode &= ~TAM_LODEnable;
+		break;
+	    case GL_LINEAR:
+		gCCPriv->curTexObj2D->TextureReadMode |=
+		    TRM_Min_Linear;
+		gCCPriv->curTexObj2D->TextureReadMode    &= ~TRM_MipMapEnable;
+		gCCPriv->curTexObj2D->TextureAddressMode &= ~TAM_LODEnable;
+		break;
+	    case GL_NEAREST_MIPMAP_NEAREST:
+		gCCPriv->curTexObj2D->TextureReadMode |=
+		    TRM_Min_NearestMMNearest;
+		break;
+	    case GL_LINEAR_MIPMAP_NEAREST:
+		gCCPriv->curTexObj2D->TextureReadMode |=
+		    TRM_Min_NearestMMLinear;
+		break;
+	    case GL_NEAREST_MIPMAP_LINEAR:
+		gCCPriv->curTexObj2D->TextureReadMode |=
+		    TRM_Min_LinearMMNearest;
+		break;
+	    case GL_LINEAR_MIPMAP_LINEAR:
+		gCCPriv->curTexObj2D->TextureReadMode |=
+		    TRM_Min_LinearMMLinear;
+		break;
+	    default:
+		break;
+	    }
+
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 2);
+	    WRITE(gCCPriv->buf, TextureReadMode,
+		  gCCPriv->curTexObj2D->TextureReadMode);
+	    WRITE(gCCPriv->buf, TextureAddressMode,
+		  gCCPriv->curTexObj2D->TextureAddressMode);
+	    break;
+	case GL_TEXTURE_WRAP_S:
+	    gCCPriv->curTexObj2D->TextureAddressMode &= ~TAM_SWrap_Mask;
+	    gCCPriv->curTexObj2D->TextureReadMode    &= ~TRM_UWrap_Mask;
+	    switch ((int)params[0]) {
+	    case GL_CLAMP:
+		gCCPriv->curTexObj2D->TextureAddressMode |= TAM_SWrap_Clamp;
+		gCCPriv->curTexObj2D->TextureReadMode    |= TRM_UWrap_Clamp;
+		break;
+	    case GL_REPEAT:
+		gCCPriv->curTexObj2D->TextureAddressMode |= TAM_SWrap_Repeat;
+		gCCPriv->curTexObj2D->TextureReadMode    |= TRM_UWrap_Repeat;
+		break;
+	    default:
+		break;
+	    }
+
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 2);
+	    WRITE(gCCPriv->buf, TextureReadMode,
+		  gCCPriv->curTexObj2D->TextureReadMode);
+	    WRITE(gCCPriv->buf, TextureAddressMode,
+		  gCCPriv->curTexObj2D->TextureAddressMode);
+	    break;
+	case GL_TEXTURE_WRAP_T:
+	    gCCPriv->curTexObj2D->TextureAddressMode &= ~TAM_TWrap_Mask;
+	    gCCPriv->curTexObj2D->TextureReadMode    &= ~TRM_VWrap_Mask;
+	    switch ((int)params[0]) {
+	    case GL_CLAMP:
+		gCCPriv->curTexObj2D->TextureAddressMode |= TAM_TWrap_Clamp;
+		gCCPriv->curTexObj2D->TextureReadMode    |= TRM_VWrap_Clamp;
+		break;
+	    case GL_REPEAT:
+		gCCPriv->curTexObj2D->TextureAddressMode |= TAM_TWrap_Repeat;
+		gCCPriv->curTexObj2D->TextureReadMode    |= TRM_VWrap_Repeat;
+		break;
+	    default:
+		break;
+	    }
+
+	    CHECK_DMA_BUFFER(gCC, gCCPriv, 2);
+	    WRITE(gCCPriv->buf, TextureReadMode,
+		  gCCPriv->curTexObj2D->TextureReadMode);
+	    WRITE(gCCPriv->buf, TextureAddressMode,
+		  gCCPriv->curTexObj2D->TextureAddressMode);
+	    break;
+	default:
+	    break;
+	}
+    } else {
+	/* ERROR !! */
+    }
 }
 
 void _gamma_TexParameteri(GLenum target, GLenum pname, GLint param)
@@ -3054,8 +3877,8 @@ void _gamma_TexSubImage2D(GLenum target, GLint level, GLint xoffset, GLint yoffs
 	** format for the image that it is replacing.
 	*/
 
-	if (!driTMMSubImage(gCCPriv->tmm, gCCPriv->curTexObj2D->image[level],
-			    xoffset, yoffset, width, height, image)) {
+	if (driTMMSubImage(gCCPriv->tmm, gCCPriv->curTexObj2D->image[level],
+			    xoffset, yoffset, width, height, image) < 0) {
 	    /* NOT_DONE: Handle error */
 	}
     } else {
@@ -3089,7 +3912,7 @@ void _gamma_Translatef(GLfloat x, GLfloat y, GLfloat z)
     GLfloat m[16];
     int i;
 
-    DEBUG_GLCMDS(("Tranlatef: %f %f %f\n", x, y, z));
+    DEBUG_GLCMDS(("Translatef: %f %f %f\n", x, y, z));
 
     for (i = 0; i < 16; i++)
 	if (i % 5 == 0)
@@ -3108,26 +3931,25 @@ void _gamma_Translatef(GLfloat x, GLfloat y, GLfloat z)
 void _gamma_Vertex2d(GLdouble x, GLdouble y)
 {
     DEBUG_GLCMDS(("Vertex2d: %f %f\n", x, y));
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 2);
+    WRITEF(gCCPriv->buf, Vy,  (GLfloat)y);
+    WRITEF(gCCPriv->buf, Vx2, (GLfloat)x);
 }
 
 void _gamma_Vertex2dv(const GLdouble *v)
 {
     DEBUG_GLCMDS(("Vertex2dv: %f %f\n", v[0], v[1]));
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 2);
+    WRITEF(gCCPriv->buf, Vy,  (GLfloat)v[1]);
+    WRITEF(gCCPriv->buf, Vx2, (GLfloat)v[0]);
 }
 
 void _gamma_Vertex2f(GLfloat x, GLfloat y)
 {
     DEBUG_GLCMDS(("Vertex2f: %f %f\n", x, y));
 
-#ifdef RANDOMIZE_COLORS
-    {
-	GLuint c;
-	c = (random() / (double)RAND_MAX) * 16777216;
-	CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
-	WRITE(gCCPriv->buf, PackedColor4,  c);
-    }
-#endif
-
     CHECK_DMA_BUFFER(gCC, gCCPriv, 2);
     WRITEF(gCCPriv->buf, Vy,  y);
     WRITEF(gCCPriv->buf, Vx2, x);
@@ -3136,51 +3958,72 @@ void _gamma_Vertex2f(GLfloat x, GLfloat y)
 void _gamma_Vertex2fv(const GLfloat *v)
 {
     DEBUG_GLCMDS(("Vertex2fv: %f %f\n", v[0], v[1]));
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 2);
+    WRITEF(gCCPriv->buf, Vy,  v[1]);
+    WRITEF(gCCPriv->buf, Vx2, v[0]);
 }
 
 void _gamma_Vertex2i(GLint x, GLint y)
 {
     DEBUG_GLCMDS(("Vertex2i: %d %d\n", (int)x, (int)y));
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 2);
+    WRITEF(gCCPriv->buf, Vy,  (GLfloat)y);
+    WRITEF(gCCPriv->buf, Vx2, (GLfloat)x);
 }
 
 void _gamma_Vertex2iv(const GLint *v)
 {
     DEBUG_GLCMDS(("Vertex2iv: %d %d\n", (int)v[0], (int)v[1]));
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 2);
+    WRITEF(gCCPriv->buf, Vy,  (GLfloat)v[1]);
+    WRITEF(gCCPriv->buf, Vx2, (GLfloat)v[0]);
 }
 
 void _gamma_Vertex2s(GLshort x, GLshort y)
 {
     DEBUG_GLCMDS(("Vertex2s: %d %d\n", x, y));
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 2);
+    WRITEF(gCCPriv->buf, Vy,  (GLfloat)y);
+    WRITEF(gCCPriv->buf, Vx2, (GLfloat)x);
 }
 
 void _gamma_Vertex2sv(const GLshort *v)
 {
     DEBUG_GLCMDS(("Vertex2sv: %d %d\n", v[0], v[1]));
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 2);
+    WRITEF(gCCPriv->buf, Vy,  (GLfloat)v[1]);
+    WRITEF(gCCPriv->buf, Vx2, (GLfloat)v[0]);
 }
 
 void _gamma_Vertex3d(GLdouble x, GLdouble y, GLdouble z)
 {
     DEBUG_GLCMDS(("Vertex3d: %f %f %f\n", x, y, z));
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
+    WRITEF(gCCPriv->buf, Vz,  (GLfloat)z);
+    WRITEF(gCCPriv->buf, Vy,  (GLfloat)y);
+    WRITEF(gCCPriv->buf, Vx3, (GLfloat)x);
 }
 
 void _gamma_Vertex3dv(const GLdouble *v)
 {
     DEBUG_GLCMDS(("Vertex2fv: %f %f %f\n", v[0], v[1], v[2]));
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
+    WRITEF(gCCPriv->buf, Vz,  (GLfloat)v[2]);
+    WRITEF(gCCPriv->buf, Vy,  (GLfloat)v[1]);
+    WRITEF(gCCPriv->buf, Vx3, (GLfloat)v[0]);
 }
 
 void _gamma_Vertex3f(GLfloat x, GLfloat y, GLfloat z)
 {
     DEBUG_GLCMDS(("Vertex3f: %f %f %f\n", x, y, z));
 
-#ifdef RANDOMIZE_COLORS
-    {
-	GLuint c;
-	c = (random() / (double)RAND_MAX) * 16777216;
-	CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
-	WRITE(gCCPriv->buf, PackedColor4,  c);
-    }
-#endif
-
     CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
     WRITEF(gCCPriv->buf, Vz,  z);
     WRITEF(gCCPriv->buf, Vy,  y);
@@ -3191,15 +4034,6 @@ void _gamma_Vertex3fv(const GLfloat *v)
 {
     DEBUG_GLCMDS(("Vertex3fv: %f %f %f\n", v[0], v[1], v[2]));
 
-#ifdef RANDOMIZE_COLORS
-    {
-	GLuint c;
-	c = (random() / (double)RAND_MAX) * 16777216;
-	CHECK_DMA_BUFFER(gCC, gCCPriv, 1);
-	WRITE(gCCPriv->buf, PackedColor4,  c);
-    }
-#endif
-
     CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
     WRITEF(gCCPriv->buf, Vz,  v[2]);
     WRITEF(gCCPriv->buf, Vy,  v[1]);
@@ -3209,62 +4043,130 @@ void _gamma_Vertex3fv(const GLfloat *v)
 void _gamma_Vertex3i(GLint x, GLint y, GLint z)
 {
     DEBUG_GLCMDS(("Vertex3i: %d %d %d\n", (int)x, (int)y, (int)z));
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
+    WRITEF(gCCPriv->buf, Vz,  (GLfloat)z);
+    WRITEF(gCCPriv->buf, Vy,  (GLfloat)y);
+    WRITEF(gCCPriv->buf, Vx3, (GLfloat)x);
 }
 
 void _gamma_Vertex3iv(const GLint *v)
 {
     DEBUG_GLCMDS(("Vertex3iv: %d %d %d\n", (int)v[0], (int)v[1], (int)v[2]));
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
+    WRITEF(gCCPriv->buf, Vz,  (GLfloat)v[2]);
+    WRITEF(gCCPriv->buf, Vy,  (GLfloat)v[1]);
+    WRITEF(gCCPriv->buf, Vx3, (GLfloat)v[0]);
 }
 
 void _gamma_Vertex3s(GLshort x, GLshort y, GLshort z)
 {
     DEBUG_GLCMDS(("Vertex3s: %d %d %d\n", x, y, z));
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
+    WRITEF(gCCPriv->buf, Vz,  (GLfloat)z);
+    WRITEF(gCCPriv->buf, Vy,  (GLfloat)y);
+    WRITEF(gCCPriv->buf, Vx3, (GLfloat)x);
 }
 
 void _gamma_Vertex3sv(const GLshort *v)
 {
     DEBUG_GLCMDS(("Vertex3sv: %d %d %d\n", v[0], v[1], v[2]));
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 3);
+    WRITEF(gCCPriv->buf, Vz,  (GLfloat)v[2]);
+    WRITEF(gCCPriv->buf, Vy,  (GLfloat)v[1]);
+    WRITEF(gCCPriv->buf, Vx3, (GLfloat)v[0]);
 }
 
 void _gamma_Vertex4d(GLdouble x, GLdouble y, GLdouble z, GLdouble w)
 {
     DEBUG_GLCMDS(("Vertex4d: %f %f %f %f\n", x, y, z, w));
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 4);
+    WRITEF(gCCPriv->buf, Vw, (GLfloat)w);
+    WRITEF(gCCPriv->buf, Vz, (GLfloat)z);
+    WRITEF(gCCPriv->buf, Vy, (GLfloat)y);
+    WRITEF(gCCPriv->buf, Vx4,(GLfloat)x);
 }
 
 void _gamma_Vertex4dv(const GLdouble *v)
 {
     DEBUG_GLCMDS(("Vertex4dv: %f %f %f %f\n", v[0], v[1], v[2], v[3]));
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 4);
+    WRITEF(gCCPriv->buf, Vw,  (GLfloat)v[3]);
+    WRITEF(gCCPriv->buf, Vz,  (GLfloat)v[2]);
+    WRITEF(gCCPriv->buf, Vy,  (GLfloat)v[1]);
+    WRITEF(gCCPriv->buf, Vx4, (GLfloat)v[0]);
 }
 
 void _gamma_Vertex4f(GLfloat x, GLfloat y, GLfloat z, GLfloat w)
 {
     DEBUG_GLCMDS(("Vertex4f: %f %f %f %f\n", x, y, z, w));
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 4);
+    WRITEF(gCCPriv->buf, Vw,  w);
+    WRITEF(gCCPriv->buf, Vz,  z);
+    WRITEF(gCCPriv->buf, Vy,  y);
+    WRITEF(gCCPriv->buf, Vx4, x);
 }
 
 void _gamma_Vertex4fv(const GLfloat *v)
 {
     DEBUG_GLCMDS(("Vertex4fv: %f %f %f %f\n", v[0], v[1], v[2], v[3]));
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 4);
+    WRITEF(gCCPriv->buf, Vw,  v[3]);
+    WRITEF(gCCPriv->buf, Vz,  v[2]);
+    WRITEF(gCCPriv->buf, Vy,  v[1]);
+    WRITEF(gCCPriv->buf, Vx4, v[0]);
 }
 
 void _gamma_Vertex4i(GLint x, GLint y, GLint z, GLint w)
 {
     DEBUG_GLCMDS(("Vertex4i: %d %d %d %d\n", (int)x, (int)y, (int)z, (int)w));
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 4);
+    WRITEF(gCCPriv->buf, Vw,  (GLfloat)w);
+    WRITEF(gCCPriv->buf, Vz,  (GLfloat)z);
+    WRITEF(gCCPriv->buf, Vy,  (GLfloat)y);
+    WRITEF(gCCPriv->buf, Vx4, (GLfloat)x);
 }
 
 void _gamma_Vertex4iv(const GLint *v)
 {
     DEBUG_GLCMDS(("Vertex4iv: %d %d %d %d\n",
 		  (int)v[0], (int)v[1], (int)v[2], (int)v[3]));
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 4);
+    WRITEF(gCCPriv->buf, Vw,  (GLfloat)v[3]);
+    WRITEF(gCCPriv->buf, Vz,  (GLfloat)v[2]);
+    WRITEF(gCCPriv->buf, Vy,  (GLfloat)v[1]);
+    WRITEF(gCCPriv->buf, Vx4, (GLfloat)v[0]);
 }
 
 void _gamma_Vertex4s(GLshort x, GLshort y, GLshort z, GLshort w)
 {
     DEBUG_GLCMDS(("Vertex4s: %d %d %d %d\n", x, y, z, w));
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 4);
+    WRITEF(gCCPriv->buf, Vw, (GLfloat)w);
+    WRITEF(gCCPriv->buf, Vz, (GLfloat)z);
+    WRITEF(gCCPriv->buf, Vy, (GLfloat)y);
+    WRITEF(gCCPriv->buf, Vx4,(GLfloat)x);
 }
 
 void _gamma_Vertex4sv(const GLshort *v)
 {
     DEBUG_GLCMDS(("Vertex4sv: %d %d %d %d\n", v[0], v[1], v[2], v[3]));
+
+    CHECK_DMA_BUFFER(gCC, gCCPriv, 4);
+    WRITEF(gCCPriv->buf, Vw,  (GLfloat)v[3]);
+    WRITEF(gCCPriv->buf, Vz,  (GLfloat)v[2]);
+    WRITEF(gCCPriv->buf, Vy,  (GLfloat)v[1]);
+    WRITEF(gCCPriv->buf, Vx4, (GLfloat)v[0]);
 }
 
 void _gamma_VertexPointer(GLint size, GLenum type, GLsizei stride, const GLvoid *pointer)
@@ -3296,10 +4198,13 @@ void _gamma_Viewport(GLint x, GLint y, GLsizei width, GLsizei height)
     oy = y + sy;
 
     CHECK_DMA_BUFFER(gCC, gCCPriv, 4);
-    WRITEF(gCCPriv->buf, ViewPortOffsetX, ox);
-    WRITEF(gCCPriv->buf, ViewPortOffsetY, oy);
     WRITEF(gCCPriv->buf, ViewPortScaleX,  sx);
     WRITEF(gCCPriv->buf, ViewPortScaleY,  sy);
+    WRITEF(gCCPriv->buf, ViewPortOffsetX, ox);
+    WRITEF(gCCPriv->buf, ViewPortOffsetY, oy);
+#if 1 /* Err - this shouldn't be needed, but something isn't flushing */
+    FLUSH_DMA_BUFFER(gCC,gCCPriv);
+#endif
 }
 
 
diff --git a/xc/lib/GL/mesa/src/drv/gamma/gamma_init.h b/xc/lib/GL/mesa/src/drv/gamma/gamma_init.h
index ce804d018..e8fd5a0df 100644
--- a/xc/lib/GL/mesa/src/drv/gamma/gamma_init.h
+++ b/xc/lib/GL/mesa/src/drv/gamma/gamma_init.h
@@ -43,7 +43,6 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "gamma_regs.h"
 #include "gamma_macros.h"
 #include "gamma_texture.h"
-#include "xmesaP.h"
 
 typedef struct {
     int           regionCount;       /* Count of register regions */
@@ -109,6 +108,12 @@ typedef struct {
     int                 AB_FBReadMode;
     int                 AB_FBReadMode_Save;
     int                 DeltaMode;
+    int			ColorMaterialMode;
+    int			MaterialMode;
+    int			LightingMode;
+    int			Light0Mode;
+    int			Light1Mode;
+    int			ScissorMode;
     int                 Window; /* GID part probably should be in draw priv */
 
     gammaTexObj        *curTexObj;
@@ -140,8 +145,8 @@ extern void       gammaLoadHWMatrix(void);
 extern void       gammaInitHW(gammaContextPrivate *gcp);
 
 extern float                IdentityMatrix[16];
-extern XMesaContext         nullCC;
-extern XMesaContext         gCC;
+extern __DRIcontextPrivate *nullCC;
+extern __DRIcontextPrivate *gCC;
 extern gammaContextPrivate *gCCPriv;
 
 #endif
diff --git a/xc/lib/GL/mesa/src/drv/gamma/gamma_inithw.c b/xc/lib/GL/mesa/src/drv/gamma/gamma_inithw.c
index edb1d3da0..87d68278b 100644
--- a/xc/lib/GL/mesa/src/drv/gamma/gamma_inithw.c
+++ b/xc/lib/GL/mesa/src/drv/gamma/gamma_inithw.c
@@ -29,27 +29,32 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 /*
  * Authors:
  *   Kevin E. Martin <kevin@precisioninsight.com>
+ *   Alan Hourihane <Alan.Hourihane@btinternet.com>
  *
  */
 
 #ifdef GLX_DIRECT_RENDERING
 
 #include "gamma_init.h"
+#include "glint_dri.h"
 
 void gammaInitHW(gammaContextPrivate *gcp)
 {
     __DRIscreenPrivate *driScrnPriv = gcp->gammaScrnPriv->driScrnPriv;
+    GLINTDRIPtr         gDRIPriv = (GLINTDRIPtr)driScrnPriv->pDevPriv;
 
-    /* Set up each MX's ScanLineOwnership for OpenGL */
-    CHECK_DMA_BUFFER(nullCC, gcp, 4);
-    WRITE(gcp->buf, BroadcastMask, 1);
-    WRITE(gcp->buf, ScanLineOwnership, 5); /* Use bottom left as [0,0] */
-    WRITE(gcp->buf, BroadcastMask, 2);
-    WRITE(gcp->buf, ScanLineOwnership, 1); /* Use bottom left as [0,0] */
+    if (gDRIPriv->numMXDevices == 2) {
+	/* Set up each MX's ScanLineOwnership for OpenGL */
+	CHECK_DMA_BUFFER(nullCC, gcp, 4);
+	WRITE(gcp->buf, BroadcastMask, 1);
+	WRITE(gcp->buf, ScanLineOwnership, 5); /* Use bottom left as [0,0] */
+	WRITE(gcp->buf, BroadcastMask, 2);
+	WRITE(gcp->buf, ScanLineOwnership, 1); /* Use bottom left as [0,0] */
 
-    /* Broadcast to both MX's */
-    CHECK_DMA_BUFFER(nullCC, gcp, 1);
-    WRITE(gcp->buf, BroadcastMask, 3);
+	/* Broadcast to both MX's */
+	CHECK_DMA_BUFFER(nullCC, gcp, 1);
+	WRITE(gcp->buf, BroadcastMask, 3);
+    }
 
     /* Set MXs to known state */
     CHECK_DMA_BUFFER(nullCC, gcp, 27);
@@ -82,6 +87,7 @@ void gammaInitHW(gammaContextPrivate *gcp)
     WRITE(gcp->buf, PixelSize, 0);
 
     /* Set Gamma to known state */
+    CHECK_DMA_BUFFER(nullCC, gcp, 10);
     WRITE(gcp->buf, TriangleMode, 0);
     WRITE(gcp->buf, GeometryMode, 0);
     WRITE(gcp->buf, NormalizeMode, 0);
@@ -136,7 +142,10 @@ void gammaInitHW(gammaContextPrivate *gcp)
     WRITE(gcp->buf, AlphaTestMode, gcp->AlphaTestMode);
     WRITE(gcp->buf, AlphaBlendMode, gcp->AlphaBlendMode);
     WRITE(gcp->buf, DitherMode, DitherModeEnable | DM_ColorOrder_RGB);
-    WRITE(gcp->buf, RasterizerMode, RM_MultiGLINT | RM_BiasCoordNearHalf);
+    if (gDRIPriv->numMXDevices == 2)
+    	WRITE(gcp->buf, RasterizerMode, RM_MultiGLINT | RM_BiasCoordNearHalf);
+    else
+    	WRITE(gcp->buf, RasterizerMode, RM_BiasCoordNearHalf);
     WRITE(gcp->buf, GLINTWindow, gcp->Window);
     WRITE(gcp->buf, FastClearDepth, 0xffffffff);
     WRITE(gcp->buf, GLINTDepth, 0xffffffff);
@@ -242,7 +251,7 @@ void gammaInitHW(gammaContextPrivate *gcp)
     CHECK_DMA_BUFFER(nullCC, gcp, 5);
     WRITE(gcp->buf, GeometryMode, gcp->GeometryMode);
     WRITE(gcp->buf, NormalizeMode, NormalizeModeDisable);
-    WRITE(gcp->buf, LightingMode, LightingModeDisable);
+    WRITE(gcp->buf, LightingMode, gcp->LightingMode);
     WRITE(gcp->buf, ColorMaterialMode, ColorMaterialModeDisable);
     WRITE(gcp->buf, MaterialMode, MaterialModeDisable);
 
@@ -251,18 +260,18 @@ void gammaInitHW(gammaContextPrivate *gcp)
     WRITE(gcp->buf, BackSpecularExponent,  0); /* fixed point */
 
     CHECK_DMA_BUFFER(nullCC, gcp, 29);
-    WRITEF(gcp->buf, FrontAmbientColorRed,    0.0);
-    WRITEF(gcp->buf, FrontAmbientColorGreen,  0.0);
-    WRITEF(gcp->buf, FrontAmbientColorBlue,   0.0);
-    WRITEF(gcp->buf, BackAmbientColorRed,     0.0);
-    WRITEF(gcp->buf, BackAmbientColorGreen,   0.0);
-    WRITEF(gcp->buf, BackAmbientColorBlue,    0.0);
-    WRITEF(gcp->buf, FrontDiffuseColorRed,    0.0);
-    WRITEF(gcp->buf, FrontDiffuseColorGreen,  0.0);
-    WRITEF(gcp->buf, FrontDiffuseColorBlue,   0.0);
-    WRITEF(gcp->buf, BackDiffuseColorRed,     0.0);
-    WRITEF(gcp->buf, BackDiffuseColorGreen,   0.0);
-    WRITEF(gcp->buf, BackDiffuseColorBlue,    0.0);
+    WRITEF(gcp->buf, FrontAmbientColorRed,    0.2);
+    WRITEF(gcp->buf, FrontAmbientColorGreen,  0.2);
+    WRITEF(gcp->buf, FrontAmbientColorBlue,   0.2);
+    WRITEF(gcp->buf, BackAmbientColorRed,     0.2);
+    WRITEF(gcp->buf, BackAmbientColorGreen,   0.2);
+    WRITEF(gcp->buf, BackAmbientColorBlue,    0.2);
+    WRITEF(gcp->buf, FrontDiffuseColorRed,    0.8);
+    WRITEF(gcp->buf, FrontDiffuseColorGreen,  0.8);
+    WRITEF(gcp->buf, FrontDiffuseColorBlue,   0.8);
+    WRITEF(gcp->buf, BackDiffuseColorRed,     0.8);
+    WRITEF(gcp->buf, BackDiffuseColorGreen,   0.8);
+    WRITEF(gcp->buf, BackDiffuseColorBlue,    0.8);
     WRITEF(gcp->buf, FrontSpecularColorRed,   0.0);
     WRITEF(gcp->buf, FrontSpecularColorGreen, 0.0);
     WRITEF(gcp->buf, FrontSpecularColorBlue,  0.0);
@@ -275,11 +284,11 @@ void gammaInitHW(gammaContextPrivate *gcp)
     WRITEF(gcp->buf, BackEmissiveColorRed,    0.0);
     WRITEF(gcp->buf, BackEmissiveColorGreen,  0.0);
     WRITEF(gcp->buf, BackEmissiveColorBlue,   0.0);
-    WRITEF(gcp->buf, SceneAmbientColorRed,    0.0);
-    WRITEF(gcp->buf, SceneAmbientColorGreen,  0.0);
-    WRITEF(gcp->buf, SceneAmbientColorBlue,   0.0);
-    WRITEF(gcp->buf, FrontAlpha,              0.0);
-    WRITEF(gcp->buf, BackAlpha,               0.0);
+    WRITEF(gcp->buf, SceneAmbientColorRed,    0.2);
+    WRITEF(gcp->buf, SceneAmbientColorGreen,  0.2);
+    WRITEF(gcp->buf, SceneAmbientColorBlue,   0.2);
+    WRITEF(gcp->buf, FrontAlpha,              1.0);
+    WRITEF(gcp->buf, BackAlpha,               1.0);
 
     CHECK_DMA_BUFFER(nullCC, gcp, 8);
     WRITE(gcp->buf, PointMode, (PM_AntialiasDisable |
@@ -319,28 +328,28 @@ void gammaInitHW(gammaContextPrivate *gcp)
     WRITE(gcp->buf, Light14Mode, LNM_Off);
     WRITE(gcp->buf, Light15Mode, LNM_Off);
 
-    CHECK_DMA_BUFFER(nullCC, gcp, 21);
-    WRITEF(gcp->buf, FrontAmbientColorRed,    1.0);
-    WRITEF(gcp->buf, FrontAmbientColorGreen,  1.0);
-    WRITEF(gcp->buf, FrontAmbientColorBlue,   1.0);
-    WRITEF(gcp->buf, BackAmbientColorRed,     1.0);
-    WRITEF(gcp->buf, BackAmbientColorGreen,   1.0);
-    WRITEF(gcp->buf, BackAmbientColorBlue,    1.0);
-    WRITEF(gcp->buf, FrontDiffuseColorRed,    1.0);
-    WRITEF(gcp->buf, FrontDiffuseColorGreen,  1.0);
-    WRITEF(gcp->buf, FrontDiffuseColorBlue,   1.0);
-    WRITEF(gcp->buf, BackDiffuseColorRed,     1.0);
-    WRITEF(gcp->buf, BackDiffuseColorGreen,   1.0);
-    WRITEF(gcp->buf, BackDiffuseColorBlue,    1.0);
-    WRITEF(gcp->buf, FrontSpecularColorRed,   0.0);
-    WRITEF(gcp->buf, FrontSpecularColorGreen, 0.0);
-    WRITEF(gcp->buf, FrontSpecularColorBlue,  0.0);
-    WRITEF(gcp->buf, BackSpecularColorRed,    0.0);
-    WRITEF(gcp->buf, BackSpecularColorGreen,  0.0);
-    WRITEF(gcp->buf, BackSpecularColorBlue,   0.0);
-    WRITEF(gcp->buf, SceneAmbientColorRed,    0.0);
-    WRITEF(gcp->buf, SceneAmbientColorGreen,  0.0);
-    WRITEF(gcp->buf, SceneAmbientColorBlue,   0.0);
+    CHECK_DMA_BUFFER(nullCC, gcp, 22);
+    WRITEF(gcp->buf, Light0AmbientIntensityBlue, 0.0);
+    WRITEF(gcp->buf, Light0AmbientIntensityGreen, 0.0);
+    WRITEF(gcp->buf, Light0AmbientIntensityRed, 0.0);
+    WRITEF(gcp->buf, Light0DiffuseIntensityBlue, 1.0);
+    WRITEF(gcp->buf, Light0DiffuseIntensityGreen, 1.0);
+    WRITEF(gcp->buf, Light0DiffuseIntensityRed, 1.0);
+    WRITEF(gcp->buf, Light0SpecularIntensityBlue, 1.0);
+    WRITEF(gcp->buf, Light0SpecularIntensityGreen, 1.0);
+    WRITEF(gcp->buf, Light0SpecularIntensityRed, 1.0);
+    WRITEF(gcp->buf, Light0SpotlightDirectionZ, 0.0);
+    WRITEF(gcp->buf, Light0SpotlightDirectionY, 0.0);
+    WRITEF(gcp->buf, Light0SpotlightDirectionX, -1.0);
+    WRITEF(gcp->buf, Light0SpotlightExponent, 0.0);
+    WRITEF(gcp->buf, Light0PositionZ, 0.0);
+    WRITEF(gcp->buf, Light0PositionY, 0.0);
+    WRITEF(gcp->buf, Light0PositionX, 1.0);
+    WRITEF(gcp->buf, Light0PositionW, 0.0);
+    WRITEF(gcp->buf, Light0CosSpotlightCutoffAngle, -1.0);
+    WRITEF(gcp->buf, Light0ConstantAttenuation, 1.0);
+    WRITEF(gcp->buf, Light0LinearAttenuation,   0.0);
+    WRITEF(gcp->buf, Light0QuadraticAttenuation,0.0);
 
     CHECK_DMA_BUFFER(nullCC, gcp, 6);
     WRITEF(gcp->buf, ViewPortScaleX,  (gcp->w)/2.0);
diff --git a/xc/lib/GL/mesa/src/drv/gamma/gamma_matrix.c b/xc/lib/GL/mesa/src/drv/gamma/gamma_matrix.c
index f6c166609..17220fbb4 100644
--- a/xc/lib/GL/mesa/src/drv/gamma/gamma_matrix.c
+++ b/xc/lib/GL/mesa/src/drv/gamma/gamma_matrix.c
@@ -163,9 +163,7 @@ void gammaLoadHWMatrix(void)
 	WRITEF(gCCPriv->buf, ModelViewMatrix13, gCCPriv->ModelView[13]);
 	WRITEF(gCCPriv->buf, ModelViewMatrix14, gCCPriv->ModelView[14]);
 	WRITEF(gCCPriv->buf, ModelViewMatrix15, gCCPriv->ModelView[15]);
-	/* Fall through to load ModelViewProjectionMatrix */
 #endif
-
     case GL_PROJECTION:
 	CHECK_DMA_BUFFER(gCC, gCCPriv, 16);
 	WRITEF(gCCPriv->buf,
diff --git a/xc/lib/GL/mesa/src/drv/gamma/gamma_regs.h b/xc/lib/GL/mesa/src/drv/gamma/gamma_regs.h
index 3dd4b8783..b13dbdba5 100644
--- a/xc/lib/GL/mesa/src/drv/gamma/gamma_regs.h
+++ b/xc/lib/GL/mesa/src/drv/gamma/gamma_regs.h
@@ -163,6 +163,10 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define FBWriteModeEnable             0x00000001
 #define FBW_UploadColorData           0x00000008
 
+/* FogMode */
+#define FogModeDisable                0x00000000
+#define FogModeEnable                 0x00000001
+
 /* LBWriteMode */
 #define LBWriteModeDisable            0x00000000
 #define LBWriteModeEnable             0x00000001
@@ -391,14 +395,17 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define GM_FogLinear                  0x00000000
 #define GM_FogExp                     0x00000004
 #define GM_FogExpSquared              0x00000008
+#define GM_FogMask                    0x0000000C
 #define GM_FrontPolyPoint             0x00000000
 #define GM_FrontPolyLine              0x00000010
 #define GM_FrontPolyFill              0x00000020
 #define GM_BackPolyPoint              0x00000000
 #define GM_BackPolyLine               0x00000040
 #define GM_BackPolyFill               0x00000080
+#define GM_FB_PolyMask                0x000000F0
 #define GM_FrontFaceCW                0x00000000
 #define GM_FrontFaceCCW               0x00000100
+#define GM_FFMask                     0x00000100
 #define GM_PolyCullDisable            0x00000000
 #define GM_PolyCullEnable             0x00000200
 #define GM_PolyCullFront              0x00000000
@@ -480,18 +487,44 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 /* LightingMode */
 #define LightingModeDisable           0x00000000
 #define LightingModeEnable            0x00000001
+#define LightingModeTwoSides          0x00000002
+#define LightingModeLocalViewer       0x00000008
+
+/* Light0Mode */
+#define Light0ModeDisable             0x00000000
+#define Light0ModeEnable              0x00000001
+#define Light0ModeSpotLight           0x00000002
+#define Light0ModeAttenuation         0x00000004
+#define Light0ModeLocal               0x00000008
+
+/* Light0Mode */
+#define Light1ModeDisable             0x00000000
+#define Light1ModeEnable              0x00000001
+#define Light1ModeSpotLight           0x00000002
+#define Light1ModeAttenuation         0x00000004
+#define Light1ModeLocal               0x00000008
 
 /* ColorMaterialMode */
 #define ColorMaterialModeDisable      0x00000000
 #define ColorMaterialModeEnable       0x00000001
+#define ColorMaterialModeFront        0x00000000
+#define ColorMaterialModeBack         0x00000002
+#define ColorMaterialModeFrontAndBack 0x00000004
+#define ColorMaterialModeEmission     0x00000000
+#define ColorMaterialModeAmbient      0x00000008
+#define ColorMaterialModeDiffuse      0x00000010
+#define ColorMaterialModeSpecular     0x00000018
+#define ColorMaterialModeAmbAndDiff   0x00000020
+#define ColorMaterialModeMask         0x0000003e
 
 /* MaterialMode */
 #define MaterialModeDisable           0x00000000
 #define MaterialModeEnable            0x00000001
+#define MaterialModeTwoSides          0x00000080
 
 /* DeltaMode */
 #define DM_Target300SX                0x00000000
-#define DM_Target300TXMX              0x00000001
+#define DM_Target500TXMX              0x00000001
 #define DM_Depth16                    0x00000004
 #define DM_Depth24                    0x00000008
 #define DM_Depth32                    0x0000000c
diff --git a/xc/lib/GL/mesa/src/drv/gamma/gamma_texture.c b/xc/lib/GL/mesa/src/drv/gamma/gamma_texture.c
index 14a993969..ffa437ed2 100644
--- a/xc/lib/GL/mesa/src/drv/gamma/gamma_texture.c
+++ b/xc/lib/GL/mesa/src/drv/gamma/gamma_texture.c
@@ -34,6 +34,7 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 #ifdef GLX_DIRECT_RENDERING
 
+#include <Xarch.h>
 #include <X11/Xlibint.h>
 #include "gamma_init.h"
 #include <string.h>
@@ -74,7 +75,11 @@ void gammaTOInit(gammaTexObj *t)
 
     t->TextureFilterMode = (TextureFilterModeDisable);
 
+#if X_BYTE_ORDER == X_LITTLE_ENDIAN
     t->TextureFormat = (TF_LittleEndian |
+#else
+    t->TextureFormat = (TF_BigEndian |
+#endif
 			TF_16Bit_565 |
 			TF_ColorOrder_RGB |
 			TF_Compnents_4 |
diff --git a/xc/lib/GL/mesa/src/drv/gamma/gamma_xmesa.c b/xc/lib/GL/mesa/src/drv/gamma/gamma_xmesa.c
index 450b081a3..b199bc073 100644
--- a/xc/lib/GL/mesa/src/drv/gamma/gamma_xmesa.c
+++ b/xc/lib/GL/mesa/src/drv/gamma/gamma_xmesa.c
@@ -30,6 +30,7 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  * Authors:
  *   Kevin E. Martin <kevin@precisioninsight.com>
  *   Brian Paul <brian@precisioninsight.com>
+ *   Alan Hourihane <Alan.Hourihane@btinternet.com>
  */
 
 #ifdef GLX_DIRECT_RENDERING
@@ -38,6 +39,7 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "gamma_init.h"
 #include "gamma_gl.h"
 #include "glapi.h"
+#include "glint_dri.h"
 #include "context.h"
 #include "mmath.h"
 
@@ -144,6 +146,7 @@ GLboolean XMesaCreateContext( Display *dpy,
     gammaContextPrivate *cPriv;
     __DRIscreenPrivate *driScrnPriv = driContextPriv->driScreenPriv;
     gammaScreenPrivate *gPriv = (gammaScreenPrivate *)driScrnPriv->private;
+    GLINTDRIPtr         gDRIPriv = (GLINTDRIPtr)driScrnPriv->pDevPriv;
 
     if (!Dispatch) {
        GLuint size = _glapi_get_dispatch_table_size() * sizeof(GLvoid *);
@@ -194,27 +197,26 @@ GLboolean XMesaCreateContext( Display *dpy,
 		cPriv->ModelViewProj[i] =
 		cPriv->Texture[i] = 0.0;
 
-    /*
-    ** NOT_DONE: The 0xe4 in LBReadMode and FBReadMode refers to the
-    ** partial products for the 640x480 mode.  We will need to look up
-    ** the partial products in a table (c.f., glint_driver.c) to support
-    ** other FB sizes.
-    */
     cPriv->LBReadMode = (LBReadSrcDisable |
 			 LBReadDstDisable |
 			 LBDataTypeDefault |
 			 LBWindowOriginBot |
-			 LBScanLineInt2 |
-			 0xe4); /* NOT_DONE: 640x480 partial products */
+			 gDRIPriv->pprod);
     cPriv->FBReadMode = (FBReadSrcDisable |
 			 FBReadDstDisable |
 			 FBDataTypeDefault |
 			 FBWindowOriginBot |
-			 FBScanLineInt2 |
-			 0xe4); /* NOT_DONE: 640x480 partial products */
-
-    cPriv->FBWindowBase = driScrnPriv->fbWidth * (driScrnPriv->fbHeight/2 - 1);
-    cPriv->LBWindowBase = driScrnPriv->fbWidth * (driScrnPriv->fbHeight/2 - 1);
+			 gDRIPriv->pprod);
+ 
+    if (gDRIPriv->numMXDevices == 2) {
+	cPriv->LBReadMode |= LBScanLineInt2;
+	cPriv->FBReadMode |= FBScanLineInt2;
+    	cPriv->FBWindowBase =driScrnPriv->fbWidth*(driScrnPriv->fbHeight/2 - 1);
+    	cPriv->LBWindowBase =driScrnPriv->fbWidth*(driScrnPriv->fbHeight/2 - 1);
+    } else {
+    	cPriv->FBWindowBase = driScrnPriv->fbWidth * driScrnPriv->fbHeight;
+    	cPriv->LBWindowBase = driScrnPriv->fbWidth * driScrnPriv->fbHeight;
+    }
 
     cPriv->Begin = (B_AreaStippleDisable |
 		    B_LineStippleDisable |
@@ -306,7 +308,7 @@ GLboolean XMesaCreateContext( Display *dpy,
 			      4, 1,
 			      gammaTOLoad,
 			      gammaTOLoadSub);
-	
+
     cPriv->curTexObj = gammaTOFind(0);
     cPriv->curTexObj1D = cPriv->curTexObj;
     cPriv->curTexObj2D = cPriv->curTexObj;
@@ -334,7 +336,7 @@ GLboolean XMesaCreateContext( Display *dpy,
 
     cPriv->DeltaMode = (DM_SubPixlCorrectionEnable |
 			DM_SmoothShadingEnable |
-			DM_Target300TXMX);
+			DM_Target500TXMX);
 
     switch (cPriv->DepthSize) {
     case 16:
@@ -352,6 +354,14 @@ GLboolean XMesaCreateContext( Display *dpy,
 
     cPriv->gammaScrnPriv = gPriv;
 
+    cPriv->LightingMode = LightingModeDisable;
+    cPriv->Light0Mode = LNM_Off;
+    cPriv->Light1Mode = LNM_Off;
+
+    cPriv->MaterialMode = MaterialModeDisable;
+
+    cPriv->ScissorMode = UserScissorDisable | ScreenScissorDisable;
+
     driContextPriv->driverPrivate = cPriv;
 
     /* Initialize the HW to a known state */
@@ -491,7 +501,6 @@ GLboolean XMesaMakeCurrent(__DRIcontextPrivate *driContextPriv,
                            __DRIdrawablePrivate *driDrawPriv,
                            __DRIdrawablePrivate *driReadPriv)
 {
-   
     if (driContextPriv) {
 	gCC     = driContextPriv;
 	gCCPriv = (gammaContextPrivate *) driContextPriv->driverPrivate;
@@ -503,6 +512,8 @@ GLboolean XMesaMakeCurrent(__DRIcontextPrivate *driContextPriv,
 	WRITE(gCCPriv->buf, GLINTWindow, gCCPriv->Window);
 
         _glapi_set_dispatch(Dispatch);
+
+	_gamma_Viewport(0, 0, driDrawPriv->w, driDrawPriv->h);
     } else {
 	gCC     = NULL;
 	gCCPriv = NULL;
diff --git a/xc/lib/GL/mesa/src/drv/i810/Imakefile b/xc/lib/GL/mesa/src/drv/i810/Imakefile
index 7260846fd..6b72c80e4 100644
--- a/xc/lib/GL/mesa/src/drv/i810/Imakefile
+++ b/xc/lib/GL/mesa/src/drv/i810/Imakefile
@@ -1,4 +1,6 @@
 
+#include <Threads.tmpl>
+
 #define DoNormalLib NormalLibGlx
 #define DoSharedLib SharedLibGlx
 #define DoExtraLib SharedLibGlx
@@ -9,13 +11,22 @@
 ALLOC_DEFINES = -DMALLOC_0_RETURNS_NULL
 #endif
 
+#if defined(LinuxArchitecture)
+OS_SUBDIR = linux
+#endif
+#if defined(FreeBSDArchitecture)
+OS_SUBDIR = bsd
+#endif
+
 #if BuildXF86DRI
   DRI_DEFINES = GlxDefines -DDRIVERTS
  DRI_INCLUDES = -I../../../../dri -I../../../../glx \
+		-I../../../dri \
 		-I$(TOP)/include -I$(TOP)/include/GL \
 		-I$(XF86OSSRC) -I$(SERVERSRC)/GL/dri \
 		-I$(XF86DRIVERSRC)/i810  \
-		-I../../../include -I../.. -I../common -I../../X
+		-I../../../include -I../.. -I../common -I../../X \
+	        -I$(XF86OSSRC)/$(OS_SUBDIR)/drm/kernel
 #endif
 
 MESA_INCLUDES = -I. -I.. -I../../include
@@ -23,20 +34,244 @@ MESA_INCLUDES = -I. -I.. -I../../include
 
 
       DEFINES = $(ALLOC_DEFINES) $(DRI_DEFINES)
-     INCLUDES = -I$(XLIBSRC) -I$(EXTINCSRC) $(MESA_INCLUDES) $(DRI_INCLUDES) \
-		-I/usr/include/glide
-      DRISRCS = i810_xmesa.c i810clear.c	\
-		i810dd.c i810depth.c i810dma.c i810ring.c		\
-		i810pipeline.c i810span.c i810state.c i810swap.c	\
-		i810tex.c i810tris.c i810vb.c i810fastpath.c
+     INCLUDES = -I$(XLIBSRC) -I$(EXTINCSRC) $(MESA_INCLUDES) $(DRI_INCLUDES) 
+
+      DRISRCS = ../../../dri/dri_mesa.c \
+		../../../../dri/dri_tmm.c
+
+      DRIOBJS = ../../../dri/dri_mesa.o \
+		../../../../dri/dri_tmm.o
+
+      DRMSRCS = ../../../../dri/drm/xf86drm.c \
+		../../../../dri/drm/xf86drmHash.c \
+		../../../../dri/drm/xf86drmRandom.c \
+		../../../../dri/drm/xf86drmSL.c
+
+      DRMOBJS = ../../../../dri/drm/xf86drm.o \
+		../../../../dri/drm/xf86drmHash.o \
+		../../../../dri/drm/xf86drmRandom.o \
+		../../../../dri/drm/xf86drmSL.o
+
+
+     I810SRCS = i810_xmesa.c	\
+		i810dd.c  	\
+		i810pipeline.c i810span.c i810state.c	\
+		i810tex.c i810tris.c i810vb.c i810fastpath.c i810ioctl.c
+
+     I810OBJS = i810_xmesa.o \
+		i810dd.o  		\
+		i810pipeline.o i810span.o i810state.o	\
+		i810tex.o i810tris.o i810vb.o i810fastpath.o i810ioctl.o
+
+     MESASRCS = ../../aatriangle.c \
+		../../accum.c \
+		../../alpha.c \
+		../../alphabuf.c \
+		../../attrib.c \
+		../../bbox.c \
+		../../bitmap.c \
+		../../blend.c \
+		../../buffers.c \
+		../../clip.c \
+		../../colortab.c \
+		../../config.c \
+		../../context.c \
+		../../copypix.c \
+		../../cva.c \
+		../../debug_xform.c \
+		../../depth.c \
+		../../dlist.c \
+		../../drawpix.c \
+		../../enable.c \
+		../../enums.c \
+		../../eval.c \
+		../../extensions.c \
+		../../feedback.c \
+		../../fog.c \
+		../../get.c \
+		../../glapi.c \
+		../../glapinoop.c \
+		../../glthread.c \
+		../../hash.c \
+		../../image.c \
+		../../imaging.c \
+		../../light.c \
+		../../lines.c \
+		../../logic.c \
+		../../masking.c \
+		../../matrix.c \
+		../../mem.c \
+		../../mmath.c \
+		../../pb.c \
+		../../pipeline.c \
+		../../pixel.c \
+		../../pixeltex.c \
+		../../points.c \
+		../../polygon.c \
+		../../quads.c \
+		../../rastpos.c \
+		../../readpix.c \
+		../../rect.c \
+		../../scissor.c \
+		../../shade.c \
+		../../span.c \
+		../../stages.c \
+		../../state.c \
+		../../stencil.c \
+		../../teximage.c \
+		../../texobj.c \
+		../../texstate.c \
+		../../texture.c \
+		../../texutil.c \
+		../../translate.c \
+		../../triangle.c \
+		../../varray.c \
+		../../vb.c \
+		../../vbcull.c \
+		../../vbfill.c \
+		../../vbindirect.c \
+		../../vbrender.c \
+		../../vbxform.c \
+		../../vector.c \
+		../../vertices.c \
+		../../winpos.c \
+		../../xform.c \
+		../../zoom.c \
+		../../X86/common_x86.c
+
+     MESAOBJS = ../../aatriangle.o \
+		../../accum.o \
+		../../alpha.o \
+		../../alphabuf.o \
+		../../attrib.o \
+		../../bbox.o \
+		../../bitmap.o \
+		../../blend.o \
+		../../buffers.o \
+		../../clip.o \
+		../../colortab.o \
+		../../config.o \
+		../../context.o \
+		../../copypix.o \
+		../../cva.o \
+		../../debug_xform.o \
+		../../depth.o \
+		../../dlist.o \
+		../../drawpix.o \
+		../../enable.o \
+		../../enums.o \
+		../../eval.o \
+		../../extensions.o \
+		../../feedback.o \
+		../../fog.o \
+		../../get.o \
+		../../hash.o \
+		../../hint.o \
+		../../image.o \
+		../../imaging.o \
+		../../light.o \
+		../../lines.o \
+		../../logic.o \
+		../../masking.o \
+		../../matrix.o \
+		../../mem.o \
+		../../mmath.o \
+		../../pb.o \
+		../../pipeline.o \
+		../../pixel.o \
+		../../pixeltex.o \
+		../../points.o \
+		../../polygon.o \
+		../../quads.o \
+		../../rastpos.o \
+		../../readpix.o \
+		../../rect.o \
+		../../scissor.o \
+		../../shade.o \
+		../../span.o \
+		../../stages.o \
+		../../state.o \
+		../../stencil.o \
+		../../teximage.o \
+		../../texobj.o \
+		../../texstate.o \
+		../../texture.o \
+		../../texutil.o \
+		../../translate.o \
+		../../triangle.o \
+		../../varray.o \
+		../../vb.o \
+		../../vbcull.o \
+		../../vbfill.o \
+		../../vbindirect.o \
+		../../vbrender.o \
+		../../vbxform.o \
+		../../vector.o \
+		../../vertices.o \
+		../../winpos.o \
+		../../xform.o \
+		../../zoom.o
 
-      DRIOBJS = i810_xmesa.o i810clear.o	\
-		i810dd.o i810depth.o i810dma.o i810ring.o		\
-		i810pipeline.o i810span.o i810state.o i810swap.o	\
-		i810tex.o i810tris.o i810vb.o i810fastpath.o
+#ifdef i386Architecture
+     X86_SRCS = ../../X86/x86.c \
+		../../X86/x86a.S \
+		../../X86/common_x86.c \
+		../../X86/common_x86asm.S \
+		../../X86/vertex.S
+
+     X86_OBJS = ../../X86/x86.o \
+		../../X86/x86a.o \
+		../../X86/common_x86.o \
+		../../X86/common_x86asm.o \
+		../../X86/vertex.o
+
+     MMX_SRCS = ../../X86/mmx_blend.S
+
+     MMX_OBJS = ../../X86/mmx_blend.o
+
+XCOMM Disabling 3Dnow code for the time being.
+#if 0
+   3DNOW_SRCS = ../../X86/3dnow.c \
+		../../X86/3dnow_norm_raw.S \
+		../../X86/3dnow_xform_masked1.S \
+		../../X86/3dnow_xform_masked2.S \
+		../../X86/3dnow_xform_masked3.S \
+		../../X86/3dnow_xform_masked4.S \
+		../../X86/3dnow_xform_raw1.S \
+		../../X86/3dnow_xform_raw2.S \
+		../../X86/3dnow_xform_raw3.S \
+		../../X86/3dnow_xform_raw4.S \
+		../../X86/vertex_3dnow.S
+
+   3DNOW_OBJS = ../../X86/3dnow.o \
+		../../X86/3dnow_norm_raw.o \
+		../../X86/3dnow_xform_masked1.o \
+		../../X86/3dnow_xform_masked2.o \
+		../../X86/3dnow_xform_masked3.o \
+		../../X86/3dnow_xform_masked4.o \
+		../../X86/3dnow_xform_raw1.o \
+		../../X86/3dnow_xform_raw2.o \
+		../../X86/3dnow_xform_raw3.o \
+		../../X86/3dnow_xform_raw4.o \
+		../../X86/vertex_3dnow.o
+#endif
+
+#endif
+
+      ASMSRCS = $(X86_SRCS) $(MMX_SRCS) $(3DNOW_SRCS)
+      ASMOBJS = $(X86_OBJS) $(MMX_OBJS) $(3DNOW_OBJS)
+
+      COMMONSRCS = ../common/mm.c ../common/hwlog.c
+      COMMONOBJS = ../common/mm.o ../common/hwlog.o
+
+         SRCS = $(DRISRCS) $(DRMSRCS) $(MESASRCS) $(ASMSRCS) $(COMMONSRCS) $(I810SRCS) 
+         OBJS = $(DRIOBJS) $(DRMOBJS) $(MESAOBJS) $(ASMOBJS) $(COMMONOBJS) $(I810OBJS)
+
+REQUIREDLIBS += -lm
+#if !GlxBuiltInI810
+REQUIREDLIBS += -L../../../.. -lGL
+#endif
 
-	 SRCS = $(DRISRCS) 
-	 OBJS = $(DRIOBJS) 
 
 #if !GlxUseBuiltInDRIDriver
 #undef DoNormalLib NormalLibGlx
@@ -57,7 +292,7 @@ LIBNAME = i810_dri.so
 ALL_OBJS = $(OBJS)
 ALL_DEPS = DONE
 SharedDepModuleTarget($(LIBNAME),$(ALL_DEPS),$(ALL_OBJS))
-InstallDynamicModule($(LIBNAME),$(MODULEDIR),.)
+InstallDynamicModule($(LIBNAME),$(MODULEDIR)/dri,.)
 #endif
 
 DependTarget()
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810_3d_reg.h b/xc/lib/GL/mesa/src/drv/i810/i810_3d_reg.h
index 8e6942497..a4b947b30 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810_3d_reg.h
+++ b/xc/lib/GL/mesa/src/drv/i810/i810_3d_reg.h
@@ -108,6 +108,8 @@
 #define LCS_UPDATE_LINEWIDTH    (0x1<<15)
 #define LCS_LINEWIDTH_MASK      (0x7<<12)
 #define LCS_LINEWIDTH_SHIFT           12
+#define LCS_LINEWIDTH_0_5       (0x1<<12)
+#define LCS_LINEWIDTH_1_0       (0x2<<12)
 #define LCS_UPDATE_ALPHA_INTERP (0x1<<11)
 #define LCS_ALPHA_FLAT          (0x0<<10)
 #define LCS_ALPHA_INTERP        (0x1<<10)
@@ -616,9 +618,6 @@ typedef struct {
 #define DV_PF_555           (0x1<<8)
 #define DV_PF_565           (0x2<<8)
 
-
-
-
 #define GFX_OP_ANTIALIAS         ((0x3<<29)|(0x6<<24))
 #define AA_UPDATE_EDGEFLAG       (1<<13)
 #define AA_ENABLE_EDGEFLAG       (1<<12)
@@ -641,84 +640,16 @@ typedef struct {
 #define ST1_ENABLE               (1<<16)
 #define ST1_MASK                 (0xffff)
 
+#define I810_SET_FIELD( var, mask, value ) (var &= ~(mask), var |= value)
 
-/* Indices into buf.Setup where various bits of state are mirrored per
- * context and per buffer.  These can be fired at the card as a unit,
- * or in a piecewise fashion as required.
- */
-
-/* Destbuffer state 
- *    - backbuffer linear offset and pitch -- invarient in the current dri
- *    - zbuffer linear offset and pitch -- also invarient
- *    - drawing origin in back and depth buffers.
- *
- * Keep the depth/back buffer state here to acommodate private buffers
- * in the future.
- */
-#define I810_DESTREG_DI0  0	/* CMD_OP_DESTBUFFER_INFO (2 dwords) */
-#define I810_DESTREG_DI1  1
-#define I810_DESTREG_ZB0  2	/* CMD_OP_Z_BUFFER_INFO */
-#define I810_DESTREG_ZB1  3	/* CMD_OP_Z_BUFFER_INFO */
-#define I810_DESTREG_DV0  4	/* GFX_OP_DESTBUFFER_VARS (2 dwords) */
-#define I810_DESTREG_DV1  5
-#define I810_DESTREG_DR0  6	/* GFX_OP_DRAWRECT_INFO (4 dwords) */
-#define I810_DESTREG_DR1  7
-#define I810_DESTREG_DR2  8
-#define I810_DESTREG_DR3  9
-#define I810_DESTREG_DR4  10
-
-#define I810_DEST_SETUP_SIZE (11+1)
-
-/* Context state
- */
-#define I810_CTXREG_VF    0	/* GFX_OP_VERTEX_FMT */
-#define I810_CTXREG_MT    1	/* GFX_OP_MAP_TEXELS */
-#define I810_CTXREG_MC0   2	/* GFX_OP_MAP_COLOR_STAGES - stage 0 */
-#define I810_CTXREG_MC1   3	/* GFX_OP_MAP_COLOR_STAGES - stage 1 */
-#define I810_CTXREG_MC2   4	/* GFX_OP_MAP_COLOR_STAGES - stage 2 */
-#define I810_CTXREG_MA0   5	/* GFX_OP_MAP_ALPHA_STAGES - stage 0 */
-#define I810_CTXREG_MA1   6	/* GFX_OP_MAP_ALPHA_STAGES - stage 1 */
-#define I810_CTXREG_MA2   7	/* GFX_OP_MAP_ALPHA_STAGES - stage 2 */
-#define I810_CTXREG_SDM   8	/* GFX_OP_SRC_DEST_MONO */
-#define I810_CTXREG_CF0   9	/* GFX_OP_COLOR_FACTOR */
-#define I810_CTXREG_CF1   10	
-#define I810_CTXREG_FOG   11	/* GFX_OP_FOG_COLOR */
-#define I810_CTXREG_B1    12	/* GFX_OP_BOOL_1 */
-#define I810_CTXREG_B2    13	/* GFX_OP_BOOL_2 */
-#define I810_CTXREG_LCS   14	/* GFX_OP_LINEWIDTH_CULL_SHADE_MODE */
-#define I810_CTXREG_PV    15	/* GFX_OP_PV_RULE -- Invarient! */
-#define I810_CTXREG_ZA    16	/* GFX_OP_ZBIAS_ALPHAFUNC */
-#define I810_CTXREG_ST0   17    /* GFX_OP_STIPPLE */
-#define I810_CTXREG_ST1   18
-#define I810_CTXREG_AA    19	/* GFX_OP_ANTIALIAS */
-#define I810_CTX_SETUP_SIZE (20) /* pad to qword */
-
-
-/* Cliprect state - keep seperate from context as it is used for
- *                  drawing triangles to the shared backbuffer, and
- *                  changes more frequently than the 'normal' context.
- */
-#define I810_CLIPREG_SCI0  0 	/* GFX_OP_SCISSOR_INFO (3 dwords) */
-#define I810_CLIPREG_SCI1  1	
-#define I810_CLIPREG_SCI2  2	
-#define I810_CLIPREG_SC    3	/* GFX_OP_SCISSOR */
-
-#define I810_CLIP_SETUP_SIZE 4
-
-/* Texture state (per tex_buffer)
- */
-#define I810_TEXREG_MI0  0	/* GFX_OP_MAP_INFO (4 dwords) */
-#define I810_TEXREG_MI1  1	
-#define I810_TEXREG_MI2  2	
-#define I810_TEXREG_MI3  3	
-#define I810_TEXREG_MF   4	/* GFX_OP_MAP_FILTER */
-#define I810_TEXREG_MLC  5	/* GFX_OP_MAP_LOD_CTL */
-#define I810_TEXREG_MLL  6	/* GFX_OP_MAP_LOD_LIMITS */
-#define I810_TEXREG_MCS  7	/* GFX_OP_MAP_COORD_SETS ??? */
-
-#define I810_TEX_SETUP_SIZE 8
+#define I810PACKCOLOR4444(r,g,b,a) \
+  ((((a) & 0xf0) << 8) | (((r) & 0xf0) << 4) | ((g) & 0xf0) | ((b) >> 4))
 
+#define I810PACKCOLOR1555(r,g,b,a) \
+  ((((r) & 0xf8) << 7) | (((g) & 0xf8) << 2) | (((b) & 0xf8) >> 3) | \
+    ((a) ? 0x8000 : 0))
 
-#define I810_SET_FIELD( var, mask, value ) (var &= ~(mask), var |= value)
+#define I810PACKCOLOR565(r,g,b) \
+  ((((r) & 0xf8) << 8) | (((g) & 0xfc) << 3) | (((b) & 0xf8) >> 3))
 
 #endif
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810_xmesa.c b/xc/lib/GL/mesa/src/drv/i810/i810_xmesa.c
index 7545a325f..8b439e52c 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810_xmesa.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810_xmesa.c
@@ -47,7 +47,6 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "i810tex.h"
 #include "i810span.h"
 #include "i810tris.h"
-#include "i810swap.h"
 #include "i810pipeline.h"
 #include "i810ioctl.h"
 
@@ -74,8 +73,40 @@ int I810_DEBUG = (0
 
 static i810ContextPtr      i810Ctx = 0;
 
-i810Glx_t	i810glx;
 
+/* These functions are accessed externally to the driver:
+ *
+ * XMesaInitDriver
+ * XMesaResetDriver
+ * XMesaCreateVisual
+ * XMesaDestroyVisual
+ * XMesaCreateContext 
+ * XMesaDestroyContext
+ * XMesaCreateWindowBuffer
+ * XMesaCreatePixmapBuffer
+ * XMesaDestroyBuffer
+ * XMesaSwapBuffers
+ * XMesaMakeCurrent
+ *
+ */
+
+
+static drmBufMapPtr i810_create_empty_buffers(void)
+{
+   drmBufMapPtr retval;
+   
+   retval = (drmBufMapPtr)Xmalloc(sizeof(drmBufMap));
+   if(retval == NULL) return NULL;
+   memset(retval, 0, sizeof(drmBufMap));
+   retval->list = (drmBufPtr)Xmalloc(sizeof(drmBuf) * I810_DMA_BUF_NR);
+   if(retval->list == NULL) {
+      Xfree(retval);
+      return NULL;
+   }
+   memset(retval->list, 0, sizeof(drmBuf) * I810_DMA_BUF_NR);
+   fprintf(stderr, "retval : %p, retval->list : %p\n", retval, retval->list);
+   return retval;
+}
 
 GLboolean XMesaInitDriver(__DRIscreenPrivate *sPriv)
 {
@@ -148,13 +179,13 @@ GLboolean XMesaInitDriver(__DRIscreenPrivate *sPriv)
       fprintf(stderr, "Tex heap size %x, granularity %x bytes\n",
 	      i810Screen->textureSize, 1<<(i810Screen->logTextureGranularity));
 
-   i810Screen->bufs = drmMapBufs(sPriv->fd);
-   if (!i810Screen->bufs) {
+   i810Screen->bufs = i810_create_empty_buffers();
+   if(i810Screen->bufs == NULL)
+   {
       Xfree(i810Screen);
-      sPriv->private = NULL;
       return GL_FALSE;
    }
-
+    
    i810Screen->back.handle = gDRIPriv->backbuffer;
    i810Screen->back.size = gDRIPriv->backbufferSize;
 
@@ -198,11 +229,6 @@ GLboolean XMesaInitDriver(__DRIscreenPrivate *sPriv)
    }
 
 
-   /* Ditch i810glx in favor of i810Screen?
-    */
-   memset(&i810glx, 0, sizeof(i810glx));
-   i810glx.texVirtual = i810Screen->tex.map;
-   
    i810DDFastPathInit();
    i810DDTrifuncInit();
    i810DDSetupInit();
@@ -274,13 +300,13 @@ GLboolean XMesaCreateContext( Display *dpy, GLvisual *mesaVis,
     * in memory at once.
     */
    if (i810Screen->textureSize < 2*1024*1024) {
-      ctx->Const.MaxTextureLevels = 8;
+      ctx->Const.MaxTextureLevels = 9;
       ctx->Const.MaxTextureSize = 1<<8;
    } else if (i810Screen->textureSize < 8*1024*1024) {
-      ctx->Const.MaxTextureLevels = 9;
+      ctx->Const.MaxTextureLevels = 10;
       ctx->Const.MaxTextureSize = 1<<9;     
    } else {
-      ctx->Const.MaxTextureLevels = 10;
+      ctx->Const.MaxTextureLevels = 11;
       ctx->Const.MaxTextureSize = 1<<10;
    }      
 
@@ -415,23 +441,6 @@ void XMesaSwapBuffers(__DRIdrawablePrivate *driDrawPriv)
    i810SwapBuffers(imesa);
 }
 
-static void i810InitClipRects( i810ContextPtr imesa )
-{
-   switch (imesa->numClipRects) {
-   case 0:
-      imesa->ClipSetup[I810_CLIPREG_SC] = ( GFX_OP_SCISSOR |
-					    SC_UPDATE_SCISSOR );
-      imesa->ClipSetup[I810_CLIPREG_SCI1] = 0;
-      imesa->ClipSetup[I810_CLIPREG_SCI2] = 0;
-      break;
-   case 1:
-      break;
-   default:
-      break;
-   }
-   imesa->dirty |= I810_EMIT_CLIPRECT;
-}
-
 
 
 void i810XMesaSetFrontClipRects( i810ContextPtr imesa )
@@ -440,12 +449,12 @@ void i810XMesaSetFrontClipRects( i810ContextPtr imesa )
 
    imesa->numClipRects = dPriv->numClipRects;
    imesa->pClipRects = dPriv->pClipRects;
+   imesa->dirty |= I810_UPLOAD_CLIPRECTS;
    imesa->drawX = dPriv->x;
    imesa->drawY = dPriv->y;
 
    imesa->drawMap = imesa->driScreen->pFB;
    i810EmitDrawingRectangle( imesa );
-   i810InitClipRects( imesa );
 }
 
 
@@ -477,7 +486,7 @@ void i810XMesaSetBackClipRects( i810ContextPtr imesa )
 
    imesa->drawMap = imesa->i810Screen->back.map;
    i810EmitDrawingRectangle( imesa );
-   i810InitClipRects( imesa );
+   imesa->dirty |= I810_UPLOAD_CLIPRECTS;
 
    if (I810_DEBUG & DEBUG_VERBOSE_DRI)
       for (i = 0 ; i < imesa->numClipRects ; i++) 
@@ -519,29 +528,27 @@ GLboolean XMesaUnbindContext(__DRIcontextPrivate *driContextPriv)
 }
 
 
-/* This looks buggy to me - the 'b' variable isn't used anywhere...
- * Hmm - It seems that the drawable is already hooked in to
- * driDrawablePriv.  
- */
 GLboolean XMesaMakeCurrent(__DRIcontextPrivate *driContextPriv,
-                           __DRIdrawablePrivate *driDrawPriv,
-                           __DRIdrawablePrivate *driReadPriv)
+			   __DRIdrawablePrivate *driDrawPriv,
+			   __DRIdrawablePrivate *driReadPriv)
 {
    if (driContextPriv) {
       i810Ctx = (i810ContextPtr) driContextPriv->driverPrivate;
-
-      gl_make_current2(i810Ctx->glCtx, driDrawPriv->mesaBuffer, driReadPriv->mesaBuffer);
-
+      
+      gl_make_current2(i810Ctx->glCtx, driDrawPriv->mesaBuffer, 
+		       driReadPriv->mesaBuffer);
+      
+      
       i810Ctx->driDrawable = driDrawPriv;
       i810Ctx->dirty = ~0;
-   
+      
       i810XMesaWindowMoved( i810Ctx );
-
+      
       if (!i810Ctx->glCtx->Viewport.Width)
 	 gl_Viewport(i810Ctx->glCtx, 0, 0, driDrawPriv->w, driDrawPriv->h);
-
    }
-   else {
+   else 
+   {
       gl_make_current(0,0);
       i810Ctx = NULL;
    }
@@ -588,10 +595,11 @@ void i810GetLock( i810ContextPtr imesa, GLuint flags )
     * more broken than usual.
     */
    if (sarea->ctxOwner != me) {
-      i810glx.c_ctxlost++;
-      imesa->dirty |= I810_UPLOAD_CTX;
-      imesa->dirty |= I810_EMIT_CLIPRECT;
-      imesa->dirty |= I810_UPLOAD_BUFFERS;
+      imesa->dirty |= (I810_UPLOAD_CTX |
+		       I810_UPLOAD_CLIPRECTS |
+		       I810_UPLOAD_BUFFERS |
+		       I810_UPLOAD_TEX0 |
+		       I810_UPLOAD_TEX1);
       sarea->ctxOwner = me;
    }
 
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810clear.c b/xc/lib/GL/mesa/src/drv/i810/i810clear.c
deleted file mode 100644
index 5ba264059..000000000
--- a/xc/lib/GL/mesa/src/drv/i810/i810clear.c
+++ /dev/null
@@ -1,193 +0,0 @@
-
-
-#include "types.h"
-#include "vbrender.h"
-#include "i810log.h"
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <unistd.h>
-
-#include "mm.h"
-#include "i810lib.h"
-#include "i810dd.h"
-#include "i810clear.h"
-#include "i810state.h"
-#include "i810tris.h"
-
-
-
-/* Clear the depthbuffer.  Always uses the auxillary cliprects and
- * origin from 'dPriv'.
- */
-static void i810_clear_depthbuffer( GLcontext *ctx, 
-				    GLboolean all,
-				    GLint cx, GLint cy, 
-				    GLint cwidth, GLint cheight ) 
-{
-   i810ContextPtr imesa = I810_CONTEXT( ctx );
-   GLuint zval = (GLuint) (ctx->Depth.Clear * DEPTH_SCALE);
-   i810ScreenPrivate *i810Screen = imesa->i810Screen;
-   __DRIdrawablePrivate *dPriv = imesa->driDrawable;
-
-/*     int _nc = dPriv->numAuxClipRects; */
-/*     XF86DRIClipRectPtr box = dPriv->pAuxClipRects; */
-
-   int _nc = imesa->numClipRects;
-   XF86DRIClipRectPtr box = imesa->pClipRects;
-
-   cy = dPriv->h-cy-cheight;
-   cx += dPriv->auxX;
-   cy += dPriv->auxY;
-         
-   while (_nc--) {	 
-      GLint x = box[_nc].x1;
-      GLint y = box[_nc].y1;
-      GLint width = box[_nc].x2 - x;
-      GLint height = box[_nc].y2 - y;
-
-      if (!all) {
-	 if (x < cx) width -= cx - x, x = cx; 
-	 if (y < cy) height -= cy - y, y = cy;
-	 if (x + width > cx + cwidth) width = cx + cwidth - x;
-	 if (y + height > cy + cheight) height = cy + cheight - y;
-	 if (width <= 0) continue;
-	 if (height <= 0) continue;
-      }
-
-      if (I810_DEBUG&DEBUG_VERBOSE_2D)
-	 fprintf(stderr, "clear depth rect %d,%d %dx%d\n", 
-		 (int) x, (int) y, (int) width, (int) height);
-
-      {
-	 int start = (i810Screen->depthOffset + 
-		      y * i810Screen->auxPitch + 
-		      x * 2);
-	    
-	 BEGIN_BATCH(imesa, 6);
-   
-	 OUT_BATCH( BR00_BITBLT_CLIENT | BR00_OP_COLOR_BLT | 0x3 );
-	 OUT_BATCH( BR13_SOLID_PATTERN | 
-		    (0xF0 << 16) |
-		    i810Screen->auxPitch );
-	 OUT_BATCH( (height << 16) | (width * 2));
-	 OUT_BATCH( start );
-	 OUT_BATCH( zval );
-	 OUT_BATCH( 0 );
-
-	 ADVANCE_BATCH();   
-      }
-   }
-}
-
-
-/* Clear the current drawbuffer.  Checks 'imesa' to determine which set of
- * cliprects and origin to use.
- */
-static void i810_clear_colorbuffer( GLcontext *ctx, 
-				    GLboolean all,
-				    GLint cx, GLint cy, 
-				    GLint cwidth, GLint cheight ) 
-{
-   i810ContextPtr imesa = I810_CONTEXT( ctx );
-   GLushort c = imesa->ClearColor;
-   i810ScreenPrivate *i810Screen = imesa->i810Screen;
-   __DRIdrawablePrivate *dPriv = imesa->driDrawable;
-
-   int _nc = imesa->numClipRects;
-   XF86DRIClipRectPtr box = imesa->pClipRects;
-
-   cy = dPriv->h-cy-cheight;
-   cx += imesa->drawX;
-   cy += imesa->drawY;
-         
-   while (_nc--) {	 
-      GLint x = box[_nc].x1;
-      GLint y = box[_nc].y1;
-      GLint width = box[_nc].x2 - x;
-      GLint height = box[_nc].y2 - y;
-
-      if (!all) {
-	 if (x < cx) width -= cx - x, x = cx; 
-	 if (y < cy) height -= cy - y, y = cy;
-	 if (x + width > cx + cwidth) width = cx + cwidth - x;
-	 if (y + height > cy + cheight) height = cy + cheight - y;
-	 if (width <= 0) continue;
-	 if (height <= 0) continue;
-      }
-
-      if (I810_DEBUG&DEBUG_VERBOSE_2D)
-	 fprintf(stderr, "clear color rect %d,%d %dx%d\n", 
-		 (int) x, (int) y, (int) width, (int) height);
-
-
-      {
-	 int start = (imesa->drawOffset +  
-		      y * i810Screen->auxPitch + 
-		      x * i810Screen->cpp);
-	    
-	 BEGIN_BATCH( imesa, 6 );
-	    
-	 OUT_BATCH( BR00_BITBLT_CLIENT | BR00_OP_COLOR_BLT | 0x3 );
-	 OUT_BATCH( BR13_SOLID_PATTERN | 
-		    (0xF0 << 16) |
-		    i810Screen->auxPitch );
-	 OUT_BATCH( (height << 16) | (width * i810Screen->cpp));
-	 OUT_BATCH( start );
-	 OUT_BATCH( c );
-	 OUT_BATCH( 0 );
-
-	 ADVANCE_BATCH();
-      }
-   }
-}
-
-
-
-
-/*
- * i810Clear
- *
- * Clear the color and/or depth buffers.  If 'all' is GL_TRUE, clear
- * whole buffer, otherwise clear the region defined by the remaining
- * parameters.  
- */
-GLbitfield i810Clear( GLcontext *ctx, GLbitfield mask, GLboolean all,
-		      GLint cx, GLint cy, GLint cwidth, GLint cheight ) 
-{
-   i810ContextPtr imesa = I810_CONTEXT( ctx );
-
-   if (I810_DEBUG&DEBUG_VERBOSE_API)
-      fprintf(stderr, "i810Clear( %d, %d, %d, %d, %d )\n",
-	      (int)mask, (int)cx, (int)cy, (int)cwidth, (int)cheight );
-	
-   LOCK_HARDWARE(imesa);
-
-   {
-      BEGIN_BATCH( imesa, 2 );   
-      OUT_BATCH( INST_PARSER_CLIENT | INST_OP_FLUSH );
-      OUT_BATCH( 0 );
-      ADVANCE_BATCH();      
-   }
-
-   if (mask & GL_COLOR_BUFFER_BIT) {
-      i810_clear_colorbuffer( ctx, all, cx, cy, cwidth, cheight );
-      mask &= ~GL_COLOR_BUFFER_BIT;
-   }
-
-   if ( (mask & GL_DEPTH_BUFFER_BIT) && ctx->Depth.Mask ) {
-      i810_clear_depthbuffer( ctx, all, cx, cy, cwidth, cheight );
-      mask &= ~GL_DEPTH_BUFFER_BIT;
-   }      
-
-   {
-      BEGIN_BATCH( imesa, 2 );   
-      OUT_BATCH( INST_PARSER_CLIENT | INST_OP_FLUSH );
-      OUT_BATCH( 0 );
-      ADVANCE_BATCH();      
-   }
-
-   UNLOCK_HARDWARE( imesa );
-   return mask;
-}
-
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810clear.h b/xc/lib/GL/mesa/src/drv/i810/i810clear.h
deleted file mode 100644
index aedbf84d9..000000000
--- a/xc/lib/GL/mesa/src/drv/i810/i810clear.h
+++ /dev/null
@@ -1,7 +0,0 @@
-#ifndef _I810_CLEAR_H
-#define _I810_CLEAR_H
-
-extern GLbitfield i810Clear( GLcontext *ctx, GLbitfield mask, GLboolean all,
-			    GLint x, GLint y, GLint width, GLint height );
-
-#endif
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810context.h b/xc/lib/GL/mesa/src/drv/i810/i810context.h
index 9f316f6fd..629abf0b8 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810context.h
+++ b/xc/lib/GL/mesa/src/drv/i810/i810context.h
@@ -28,24 +28,32 @@
 
 typedef struct i810_context_t i810Context;
 typedef struct i810_context_t *i810ContextPtr;
+typedef struct i810_texture_object_t *i810TextureObjectPtr;
 
 #include <X11/Xlibint.h>
 #include "dri_tmm.h"
 #include "dri_mesaint.h"
 #include "dri_mesa.h"
-#include "xmesaP.h"
 
 #include "types.h"
 #include "i810_init.h"
-#include "i810_sarea.h"
+#include "drm.h"
+#include "mm.h"
+#include "i810log.h"
 
 #include "i810tex.h"
 #include "i810vb.h"
 
 
-
-#define I810_FALLBACK_TEXTURE   0x1
-#define I810_FALLBACK_BUFFER    0x2
+/* Reasons to fallback on all primitives.  (see also
+ * imesa->IndirectTriangles).  
+ */
+#define I810_FALLBACK_TEXTURE        0x1
+#define I810_FALLBACK_DRAW_BUFFER    0x2
+#define I810_FALLBACK_READ_BUFFER    0x4
+#define I810_FALLBACK_COLORMASK      0x8  
+#define I810_FALLBACK_STIPPLE        0x10  
+#define I810_FALLBACK_SPECULAR       0x20 
 
 
 
@@ -53,17 +61,6 @@ typedef struct i810_context_t *i810ContextPtr;
  */
 #define I810_NEW_TEXTURE 0x1
 
-/* for i810ctx.dirty - manage driver->hw state changes, including
- * lost contexts.
- */
-#define I810_REQUIRE_QUIESCENT 0x1
-#define I810_UPLOAD_TEX0IMAGE  0x2
-#define I810_UPLOAD_TEX1IMAGE  0x4
-#define I810_UPLOAD_CTX        0x8
-#define I810_UPLOAD_BUFFERS    0x10
-#define I810_REFRESH_RING      0x20
-#define I810_EMIT_CLIPRECT     0x40
-
 
 typedef void (*i810_interp_func)( GLfloat t, 
 				  GLfloat *result,
@@ -88,7 +85,6 @@ struct i810_context_t {
     */
    GLuint Setup[I810_CTX_SETUP_SIZE];
    GLuint BufferSetup[I810_DEST_SETUP_SIZE];
-   GLuint ClipSetup[I810_CLIP_SETUP_SIZE];
    
 
    /* Support for CVA and the fast paths.
@@ -125,24 +121,38 @@ struct i810_context_t {
 
    /* DRI stuff
     */
+   drmBufPtr  vertex_dma_buffer;
+   GLuint vertex_prim;
+
    GLframebuffer *glBuffer;
    
+   /* Two flags to keep track of fallbacks.
+    */
+   GLuint IndirectTriangles;
    GLuint Fallback;
+
+
    GLuint needClip;
 
    /* These refer to the current draw (front vs. back) buffer:
     */
-   int drawOffset;		/* draw buffer address in agp space */
+   char *drawMap;		/* draw buffer address in virtual mem */
+   char *readMap;	
    int drawX;			/* origin of drawable in draw buffer */
    int drawY;
    GLuint numClipRects;		/* cliprects for that buffer */
    XF86DRIClipRectPtr pClipRects;
 
-
    int lastSwap;
+   int secondLastSwap;
    int texAge;
+   int ctxAge;
+   int dirtyAge;
+   int any_contend;		/* throttle me harder */
 
-   XF86DRIClipRectRec draw_rect;
+   int scissor;
+   drm_clip_rect_t draw_rect;
+   drm_clip_rect_t scissor_rect;
 
    drmContext hHWContext;
    drmLock *driHwLock;
@@ -152,7 +162,7 @@ struct i810_context_t {
    __DRIdrawablePrivate *driDrawable;
    __DRIscreenPrivate *driScreen;
    i810ScreenPrivate *i810Screen; 
-   I810SAREAPriv *sarea;
+   drm_i810_sarea_t *sarea;
 };
 
 
@@ -161,7 +171,7 @@ struct i810_context_t {
  */
 #define I810_DEBUG 0   
 #ifndef I810_DEBUG
-/*  #warning "Debugging enabled - expect reduced performance" */
+#warning "Debugging enabled - expect reduced performance"
 extern int I810_DEBUG;
 #endif
 
@@ -176,6 +186,7 @@ extern int I810_DEBUG;
 #define DEBUG_VALIDATE_RING  0x800
 #define DEBUG_VERBOSE_LRU    0x1000
 #define DEBUG_VERBOSE_DRI    0x2000
+#define DEBUG_VERBOSE_IOCTL  0x4000
 
 
 
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810dd.c b/xc/lib/GL/mesa/src/drv/i810/i810dd.c
index c1fbda734..583eb630c 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810dd.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810dd.c
@@ -29,8 +29,6 @@
 #include <stdio.h>
 
 #include "mm.h"
-#include "i810lib.h"
-#include "i810clear.h"
 #include "i810dd.h"
 #include "i810log.h"
 #include "i810state.h"
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810dma.h b/xc/lib/GL/mesa/src/drv/i810/i810dma.h
deleted file mode 100644
index 9d472719a..000000000
--- a/xc/lib/GL/mesa/src/drv/i810/i810dma.h
+++ /dev/null
@@ -1,186 +0,0 @@
-/*
-  GLX Hardware Device Driver for Intel i810
-  Copyright (C) 1999 Keith Whitwell <keithw@precisioninsight.com>
-  
-  This program is free software; you can redistribute it and/or modify
-  it under the terms of the GNU General Public License as published by
-  the Free Software Foundation; either version 2 of the License, or
-  any later version.
-  
-  This program is distributed in the hope that it will be useful,
-  but WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-  GNU General Public License for more details.
-  
-  You should have received a copy of the GNU General Public License
-  along with this program; if not, write to the Free Software
-  Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
-  
- original by Jeff Hartmann <slicer@ionet.net>
-  6/16/99: rewrite by John Carmack <johnc@idsoftware.com>
-  Oct 99: port to i180 by Keith Whitwell <keithw@precisioninsight.com>
-*/
-
-#ifndef I810DMA_H
-#define I810DMA_H
-
-#include "i810lib.h"
-#include "mm.h"
-
-/* a flush command will guarantee that all data added to the dma buffer
-is on its way to the card, and will eventually complete with no more
-intervention.
-*/
-void i810DmaFlush( i810ContextPtr imesa );
-
-/* the overflow function is called when a block can't be allocated
-in the current dma buffer.  It flushes the current buffer and
-records some information */
-void i810DmaOverflow(int newDwords);
-
-/* a finish command will guarantee that all dma commands have actually
-been consumed by the card.  Note that there may still be a couple primitives
-that have not yet been executed out of the internal FIFO, so this does not
-guarantee that the drawing engine is idle. */
-void i810DmaFinish( i810ContextPtr imesa );
-
-
-
-typedef struct {
-   unsigned long Start;
-   unsigned long End;
-   unsigned long Size;
-} I810MemRange;
-
-
-
-typedef struct i810_batch_buffer {
-   I810MemRange mem;
-   char *virtual_start;
-   int head;
-   int space;
-   int additional_space;
-   int texture_age;
-} i810BatchBuffer;
-
-
-
-#define I810_USE_BATCH 0
-#if I810_USE_BATCH
-
-#define BEGIN_BATCH( imesa, n )						\
-   unsigned int outbatch;						\
-   volatile char *virt;							\
-   if (I810_DEBUG & DEBUG_VERBOSE_RING)					\
-      fprintf(stderr,							\
-      "BEGIN_BATCH(%d) in %s\n"						\
-      "(spc left %d/%ld, head %x vstart %p start %lx)\n",		\
-      n, __FUNCTION__, i810glx.dma_buffer->space,			\
-      i810glx.dma_buffer->mem.Size - i810glx.dma_buffer->head,		\
-      i810glx.dma_buffer->head,						\
-      i810glx.dma_buffer->virtual_start,				\
-      i810glx.dma_buffer->mem.Start );					\
-   if (i810glx.dma_buffer->space < n*4)					\
-       i810DmaOverflow(n);						\
-   outbatch = i810glx.dma_buffer->head;					\
-   virt = i810glx.dma_buffer->virtual_start;			
-
-
-#define OUT_BATCH(val) {					\
-   *(volatile unsigned int *)(virt + outbatch) = val;		\
-   if (I810_DEBUG & DEBUG_VERBOSE_RING)				\
-      fprintf(stderr, "OUT_BATCH %x: %x\n", (int)(outbatch/4), (int)(val));	\
-   outbatch += 4;						\
-}
-   
-
-#define ADVANCE_BATCH() {						\
-   if (I810_DEBUG & DEBUG_VERBOSE_RING)					\
-      fprintf(stderr, "ADVANCE_BATCH(%f) in %s\n",			\
-                  (outbatch - i810glx.dma_buffer->head) / 4.0,		\
-                  __FUNCTION__);					\
-   i810glx.dma_buffer->space -= outbatch - i810glx.dma_buffer->head;	\
-   i810glx.dma_buffer->head = outbatch;					\
-}
-
-#define FINISH_PRIM()
-
-#else
-
-
-/* As an alternate path to dma, we can export the registers (and hence
- * ringbuffer) to the client.  For security we should alloc these
- * things in agp memory which isn't exported to the client.  But then
- * again, the client could blit over such things if it wanted to hang
- * the system.
- *
- * Malicious clients aren't really delt with by the DRI.
- */
-
-
-extern void _I810RefreshLpRing( i810ContextPtr imesa, int update );
-extern int _I810WaitLpRing( i810ContextPtr imesa, int n, int timeout_usec );
-extern int _I810Sync( i810ContextPtr imesa );
-
-
-#define OUT_RING(n) {					\
-   if (I810_DEBUG & DEBUG_VERBOSE_RING)			\
-      fprintf(stderr, "OUT_RING %x: %x\n", outring, (int)(n));	\
-   if (!(I810_DEBUG & DEBUG_NO_OUTRING)) {		\
-      *(volatile unsigned int *)(virt + outring) = n;	\
-      outring += 4;					\
-      outring &= ringmask;				\
-   }							\
-}
-
-#define ADVANCE_LP_RING() {			\
-    i810glx.LpRing.tail = outring;		\
-    OUTREG(LP_RING + RING_TAIL, outring);	\
-}
-
-#define BEGIN_LP_RING(imesa, n)						\
-   unsigned int outring, ringmask;					\
-   volatile char *virt;							\
-   if ((I810_DEBUG&DEBUG_ALWAYS_SYNC) && n>2) _I810Sync( imesa );	\
-   if (i810glx.LpRing.space < n*4) _I810WaitLpRing( imesa, n*4, 0);	\
-   i810glx.LpRing.space -= n*4;						\
-   if (I810_DEBUG & DEBUG_VERBOSE_RING) 				\
-      fprintf(stderr, "BEGIN_LP_RING %d in %s\n", n, __FUNCTION__);	\
-   if (I810_DEBUG & DEBUG_VALIDATE_RING) { 				\
-      CARD32 tail = INREG(LP_RING+RING_TAIL);				\
-      if (tail != i810glx.LpRing.tail) {				\
-	 fprintf(stderr, "tail %x pI810->LpRing.tail %x\n",		\
-		    (int) tail, (int) i810glx.LpRing.tail);				\
-         exit(1);							\
-      }									\
-   }									\
-   i810glx.LpRing.synced = 0;						\
-   outring = i810glx.LpRing.tail;					\
-   ringmask = i810glx.LpRing.tail_mask;					\
-   virt = i810glx.LpRing.virtual_start;			
-
-
-#define EXTRAAA \
-
-#define INREG(addr)         *(volatile CARD32 *)(i810glx.MMIOBase + (addr))
-#define INREG16(addr)       *(volatile CARD16 *)(i810glx.MMIOBase + (addr))
-#define INREG8(addr)        *(volatile CARD8  *)(i810glx.MMIOBase + (addr))
-
-#define OUTREG(addr, val) do {					\
-   if (I810_DEBUG&DEBUG_VERBOSE_OUTREG)				\
-      fprintf(stderr, "OUTREG(%x, %x)\n", addr, val);		\
-   if (!(I810_DEBUG&DEBUG_NO_OUTREG))				\
-      *(volatile CARD32 *)(i810glx.MMIOBase + (addr)) = (val);	\
-} while (0)
-
-
-#define BEGIN_BATCH(imesa, n) BEGIN_LP_RING(imesa, n)
-#define ADVANCE_BATCH() ADVANCE_LP_RING()
-#define OUT_BATCH(val) OUT_RING(val)
-#define FINISH_PRIM() OUTREG(LP_RING + RING_TAIL, i810glx.LpRing.tail)
-
-#endif
-
-
-
-#endif
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810ioctl.c b/xc/lib/GL/mesa/src/drv/i810/i810ioctl.c
new file mode 100644
index 000000000..b4e86cff2
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/i810/i810ioctl.c
@@ -0,0 +1,477 @@
+#include <stdio.h>
+#include <unistd.h>
+
+
+#include "types.h"
+#include "pb.h"
+#include "dd.h"
+
+#include "mm.h"
+#include "i810context.h"
+#include "i810log.h"
+#include "i810ioctl.h"
+
+#include "drm.h"
+#include <sys/ioctl.h>
+
+static drmBufPtr i810_get_buffer_ioctl( i810ContextPtr imesa )
+{
+   drm_i810_dma_t dma;
+   drmBufPtr buf;
+   int retcode;
+   
+   if (I810_DEBUG&DEBUG_VERBOSE_IOCTL)
+      fprintf(stderr,  "Getting dma buffer\n");
+
+   while (1) {
+      retcode = ioctl(imesa->driFd, DRM_IOCTL_I810_GETBUF, &dma);
+
+      if (dma.granted == 1 && retcode == 0) 
+	 break;
+
+      if (I810_DEBUG&DEBUG_VERBOSE_IOCTL)
+	 fprintf(stderr, "Retcode : %d, granted : %d\n", retcode, dma.granted);
+
+      ioctl(imesa->driFd, DRM_IOCTL_I810_FLUSH);
+   }
+
+   if (I810_DEBUG&DEBUG_VERBOSE_IOCTL)
+      fprintf(stderr, 
+	      "imesa->i810Screen->bufs->list : %p, "
+	      "dma.request_idx : %d\n", 
+	      imesa->i810Screen->bufs->list, dma.request_idx);
+
+   buf = &(imesa->i810Screen->bufs->list[dma.request_idx]);
+   buf->idx = dma.request_idx;
+   buf->used = 0;
+   buf->total = dma.request_size;
+   buf->address = (drmAddress)dma.virtual;
+   return buf;
+}
+
+
+
+#define DEPTH_SCALE ((1<<16)-1)
+
+GLbitfield i810Clear( GLcontext *ctx, GLbitfield mask, GLboolean all,
+		      GLint cx, GLint cy, GLint cw, GLint ch ) 
+{
+   i810ContextPtr imesa = I810_CONTEXT( ctx );
+   __DRIdrawablePrivate *dPriv = imesa->driDrawable;
+   const GLuint colorMask = *((GLuint *) &ctx->Color.ColorMask);
+   drm_i810_clear_t clear;
+   int i;
+
+   clear.flags = 0;
+   clear.clear_color = imesa->ClearColor;
+   clear.clear_depth = (GLuint) (ctx->Depth.Clear * DEPTH_SCALE);
+
+   FLUSH_BATCH( imesa );
+	
+   if ((mask & DD_FRONT_LEFT_BIT) && colorMask == ~0) {
+      clear.flags |= I810_FRONT;
+      mask &= ~DD_FRONT_LEFT_BIT;
+   }
+
+   if ((mask & DD_BACK_LEFT_BIT) && colorMask == ~0) {
+      clear.flags |= I810_BACK;
+      mask &= ~DD_BACK_LEFT_BIT;
+   }
+
+   if ((mask & DD_DEPTH_BIT) && ctx->Depth.Mask) {
+      clear.flags |= I810_DEPTH;
+      mask &= ~DD_DEPTH_BIT;
+   }
+
+   if (!clear.flags)
+      return mask;
+
+   LOCK_HARDWARE( imesa );
+
+   /* flip top to bottom */
+   cy = dPriv->h-cy-ch;
+   cx += imesa->drawX;
+   cy += imesa->drawY;
+
+   if (I810_DEBUG&DEBUG_VERBOSE_IOCTL)
+      fprintf(stderr, "Clear, bufs %x nbox %d\n", 
+	      (int)clear.flags, (int)imesa->numClipRects);
+
+   for (i = 0 ; i < imesa->numClipRects ; ) 
+   { 	 
+      int nr = MIN2(i + I810_NR_SAREA_CLIPRECTS, imesa->numClipRects);
+      XF86DRIClipRectRec *box = imesa->pClipRects;	 
+      drm_clip_rect_t *b = imesa->sarea->boxes;
+      int n = 0;
+
+      if (!all) {
+	 for ( ; i < nr ; i++) {
+	    GLint x = box[i].x1;
+	    GLint y = box[i].y1;
+	    GLint w = box[i].x2 - x;
+	    GLint h = box[i].y2 - y;
+
+	    if (x < cx) w -= cx - x, x = cx; 
+	    if (y < cy) h -= cy - y, y = cy;
+	    if (x + w > cx + cw) w = cx + cw - x;
+	    if (y + h > cy + ch) h = cy + ch - y;
+	    if (w <= 0) continue;
+	    if (h <= 0) continue;
+
+	    b->x1 = x;
+	    b->y1 = y;
+	    b->x2 = x + w;
+	    b->y2 = y + h;
+	    b++;
+	    n++;
+	 }
+      } else {
+	 for ( ; i < nr ; i++) {
+	    *b++ = *(drm_clip_rect_t *)&box[i];
+	    n++;
+	 }
+      }
+
+      imesa->sarea->nbox = n;
+      ioctl(imesa->driFd, DRM_IOCTL_I810_CLEAR, &clear);
+   }
+
+   UNLOCK_HARDWARE( imesa );
+   imesa->dirty |= I810_UPLOAD_CLIPRECTS;
+
+   return mask;
+}
+
+
+
+
+/*
+ * Copy the back buffer to the front buffer. 
+ */
+void i810SwapBuffers( i810ContextPtr imesa ) 
+{
+   __DRIdrawablePrivate *dPriv = imesa->driDrawable;
+   XF86DRIClipRectPtr pbox;
+   int nbox;
+   int i;
+   int tmp;
+
+   FLUSH_BATCH( imesa );
+   LOCK_HARDWARE( imesa );
+   
+   pbox = dPriv->pClipRects;
+   nbox = dPriv->numClipRects;
+
+   for (i = 0 ; i < nbox ; )
+   {
+      int nr = MIN2(i + I810_NR_SAREA_CLIPRECTS, dPriv->numClipRects);
+      XF86DRIClipRectRec *b = (XF86DRIClipRectRec *)imesa->sarea->boxes;
+
+      imesa->sarea->nbox = nr - i;
+
+      for ( ; i < nr ; i++) 
+	 *b++ = pbox[i];
+
+      ioctl(imesa->driFd, DRM_IOCTL_I810_SWAP);
+   }
+
+   tmp = GET_ENQUEUE_AGE(imesa);
+   UNLOCK_HARDWARE( imesa );
+
+   if (GET_DISPATCH_AGE(imesa) < imesa->lastSwap)
+      i810WaitAge(imesa, imesa->lastSwap);
+
+   imesa->lastSwap = tmp;
+   imesa->dirty |= I810_UPLOAD_CLIPRECTS;
+}
+
+
+
+
+
+
+/* This waits for *everybody* to finish rendering -- overkill.
+ */
+void i810DmaFinish( i810ContextPtr imesa  ) 
+{
+   FLUSH_BATCH( imesa );
+
+   if (imesa->sarea->last_quiescent != imesa->sarea->last_enqueue) {
+     
+      if (I810_DEBUG&DEBUG_VERBOSE_IOCTL) 
+	 fprintf(stderr, "i810DmaFinish\n");
+
+      LOCK_HARDWARE( imesa );
+      i810RegetLockQuiescent( imesa );
+      UNLOCK_HARDWARE( imesa );
+      imesa->sarea->last_quiescent = imesa->sarea->last_enqueue;
+   }
+}
+
+
+void i810RegetLockQuiescent( i810ContextPtr imesa  ) 
+{
+   if (imesa->sarea->last_quiescent != imesa->sarea->last_enqueue) {
+      if (I810_DEBUG&DEBUG_VERBOSE_IOCTL)
+	 fprintf(stderr, "i810RegetLockQuiescent\n");
+
+      drmUnlock(imesa->driFd, imesa->hHWContext);
+      i810GetLock( imesa, DRM_LOCK_QUIESCENT ); 
+      imesa->sarea->last_quiescent = imesa->sarea->last_enqueue;
+   }
+}
+
+void i810WaitAgeLocked( i810ContextPtr imesa, int age  ) 
+{
+   int i = 0;
+
+
+   while (++i < 500000 && GET_DISPATCH_AGE(imesa) < age) {
+      ioctl(imesa->driFd, DRM_IOCTL_I810_GETAGE);
+   }
+
+   if (GET_DISPATCH_AGE(imesa) < age) {
+      if (0)
+	 fprintf(stderr, "wait locked %d %d\n", age, GET_DISPATCH_AGE(imesa));
+      ioctl(imesa->driFd, DRM_IOCTL_I810_FLUSH);
+   }
+}
+
+
+void i810WaitAge( i810ContextPtr imesa, int age  ) 
+{
+   int i = 0;
+
+   while (++i < 500000 && GET_DISPATCH_AGE(imesa) < age) {
+      ioctl(imesa->driFd, DRM_IOCTL_I810_GETAGE);
+   }
+
+   if (GET_DISPATCH_AGE(imesa) >= age)
+      return;
+
+   i = 0;
+   while (++i < 1000 && GET_DISPATCH_AGE(imesa) < age) {
+      ioctl(imesa->driFd, DRM_IOCTL_I810_GETAGE);
+      usleep(1000);
+   }
+   
+   /* To be effective at letting other clients at the hardware,
+    * particularly the X server which regularly needs quiescence to
+    * touch the framebuffer, we really need to sleep *beyond* the
+    * point where our last buffer clears the hardware.  
+    */
+   if (imesa->any_contend) {
+      usleep(3000); 
+   }
+
+   imesa->any_contend = 0;
+
+   if (GET_DISPATCH_AGE(imesa) < age) {
+      LOCK_HARDWARE(imesa);
+      if (GET_DISPATCH_AGE(imesa) < age) 
+	 ioctl(imesa->driFd, DRM_IOCTL_I810_FLUSH);
+      UNLOCK_HARDWARE(imesa);
+   }
+}
+
+
+
+void i810FlushVertices( i810ContextPtr imesa ) 
+{
+   if (!imesa->vertex_dma_buffer) return;
+
+   LOCK_HARDWARE( imesa );
+   i810FlushVerticesLocked( imesa );
+   UNLOCK_HARDWARE( imesa );
+}
+
+
+static int intersect_rect( drm_clip_rect_t *out,
+			    drm_clip_rect_t *a,
+			    drm_clip_rect_t *b )
+{
+   *out = *a;
+   if (b->x1 > out->x1) out->x1 = b->x1;
+   if (b->y1 > out->y1) out->y1 = b->y1;
+   if (b->x2 < out->x2) out->x2 = b->x2;
+   if (b->y2 < out->y2) out->y2 = b->y2;
+   if (out->x1 >= out->x2) return 0;
+   if (out->y1 >= out->y2) return 0;
+   return 1;
+}
+
+
+static void age_imesa( i810ContextPtr imesa, int age )
+{
+   if (imesa->CurrentTexObj[0]) imesa->CurrentTexObj[0]->age = age;
+   if (imesa->CurrentTexObj[1]) imesa->CurrentTexObj[1]->age = age;
+}
+
+void i810FlushVerticesLocked( i810ContextPtr imesa )
+{
+   drm_clip_rect_t *pbox = (drm_clip_rect_t *)imesa->pClipRects;
+   int nbox = imesa->numClipRects;
+   drmBufPtr buffer = imesa->vertex_dma_buffer;
+   drm_i810_vertex_t vertex;
+   int i;
+
+   if (I810_DEBUG&DEBUG_VERBOSE_IOCTL)
+      fprintf(stderr, "i810FlushVerticesLocked, buf->used %d\n", 
+	      buffer->used);
+
+   if (!buffer)
+      return;
+
+   if (imesa->dirty & ~I810_UPLOAD_CLIPRECTS)
+      i810EmitHwStateLocked( imesa );
+
+   if (I810_DEBUG&DEBUG_VERBOSE_IOCTL)
+      fprintf(stderr, "i810FlushVerticesLocked, used %d\n",
+	      buffer->used);
+   
+   imesa->vertex_dma_buffer = 0;
+
+   vertex.idx = buffer->idx;
+   vertex.used = buffer->used;
+   vertex.discard = 0;
+
+   if (!nbox)
+      vertex.used = 0;
+
+   if (nbox > I810_NR_SAREA_CLIPRECTS)
+      imesa->dirty |= I810_UPLOAD_CLIPRECTS;
+
+   imesa->sarea->vertex_prim = imesa->vertex_prim;
+
+   if (!nbox || !(imesa->dirty & I810_UPLOAD_CLIPRECTS)) 
+   {
+      if (nbox == 1) 
+	 imesa->sarea->nbox = 0;
+      else
+	 imesa->sarea->nbox = nbox;
+
+      if (I810_DEBUG&DEBUG_VERBOSE_IOCTL)
+	 fprintf(stderr, "DRM_IOCTL_I810_VERTEX CASE1 nbox %d used %d\n", 
+		 nbox, vertex.used);
+
+      vertex.discard = 1;
+      ioctl(imesa->driFd, DRM_IOCTL_I810_VERTEX, &vertex);
+      age_imesa(imesa, imesa->sarea->last_enqueue);
+   }  
+   else 
+   {
+      for (i = 0 ; i < nbox ; )
+      {
+	 int nr = MIN2(i + I810_NR_SAREA_CLIPRECTS, nbox);
+	 drm_clip_rect_t *b = imesa->sarea->boxes;
+
+	 if (imesa->scissor) {
+	    imesa->sarea->nbox = 0;
+	 
+	    for ( ; i < nr ; i++) {
+	       b->x1 = pbox[i].x1 - imesa->drawX;
+	       b->y1 = pbox[i].y1 - imesa->drawY;
+	       b->x2 = pbox[i].x2 - imesa->drawX;
+	       b->y2 = pbox[i].y2 - imesa->drawY;
+
+	       if (intersect_rect(b, b, &imesa->scissor_rect)) {
+		  imesa->sarea->nbox++;
+		  b++;
+	       }
+	    }
+
+	    /* Culled?
+	     */
+	    if (!imesa->sarea->nbox) {
+	       if (nr < nbox) continue;
+	       vertex.used = 0;
+	    }
+	 } else {
+	    imesa->sarea->nbox = nr - i;
+	    for ( ; i < nr ; i++, b++) {
+	       b->x1 = pbox[i].x1 - imesa->drawX;
+	       b->y1 = pbox[i].y1 - imesa->drawY;
+	       b->x2 = pbox[i].x2 - imesa->drawX;
+	       b->y2 = pbox[i].y2 - imesa->drawY;
+	    }
+	 }
+	 
+	 /* Finished with the buffer?
+	  */
+	 if (nr == nbox) 
+	    vertex.discard = 1;
+
+     	 if (I810_DEBUG&DEBUG_VERBOSE_IOCTL)
+	    fprintf(stderr, "DRM_IOCTL_I810_VERTEX nbox %d used %d\n", 
+		    nbox, vertex.used);
+
+	 ioctl(imesa->driFd, DRM_IOCTL_I810_VERTEX, &vertex);
+	 age_imesa(imesa, imesa->sarea->last_enqueue);
+      }
+   }
+
+   imesa->dirty = 0;
+   if (I810_DEBUG&DEBUG_VERBOSE_IOCTL)
+      fprintf(stderr, "finished i810FlushVerticesLocked\n");
+}
+
+
+GLuint *i810AllocDwords( i810ContextPtr imesa, int dwords, GLuint prim )
+{
+   GLuint *start;
+
+   if (!imesa->vertex_dma_buffer) 
+   {
+      if (I810_DEBUG&DEBUG_VERBOSE_IOCTL)
+	 fprintf(stderr, "i810AllocPrimitiveVerts -- get buf\n");
+      LOCK_HARDWARE(imesa);
+      imesa->vertex_dma_buffer = i810_get_buffer_ioctl( imesa );
+      imesa->vertex_dma_buffer->used = 4;
+      imesa->vertex_prim = prim;
+      UNLOCK_HARDWARE(imesa);
+   } 
+   else if (imesa->vertex_prim != prim ||
+	    imesa->vertex_dma_buffer->used + dwords * 4 > 
+	    imesa->vertex_dma_buffer->total) 
+   {
+      if (I810_DEBUG&DEBUG_VERBOSE_IOCTL)
+	 fprintf(stderr, "i810AllocPrimitiveVerts -- flush\n");
+      i810FlushVertices( imesa );
+      LOCK_HARDWARE(imesa);
+      imesa->vertex_dma_buffer = i810_get_buffer_ioctl( imesa );
+      imesa->vertex_dma_buffer->used = 4;
+      imesa->vertex_prim = prim;
+      UNLOCK_HARDWARE(imesa);
+   }
+
+
+   if (0)
+      fprintf(stderr, "i810AllocPrimitiveVerts %d, buf %d, used %d\n",
+	      dwords, imesa->vertex_dma_buffer->idx,
+	      imesa->vertex_dma_buffer->used);
+      
+   start = (GLuint *)((char *)imesa->vertex_dma_buffer->address + 
+		      imesa->vertex_dma_buffer->used);
+
+   imesa->vertex_dma_buffer->used += dwords * 4;
+
+   return start;
+}
+
+static void i810DDFlush( GLcontext *ctx )
+{
+   i810ContextPtr imesa = I810_CONTEXT( ctx );
+   FLUSH_BATCH( imesa );
+}
+
+static void i810DDFinish( GLcontext *ctx  ) 
+{
+   i810ContextPtr imesa = I810_CONTEXT( ctx );
+   i810DmaFinish( imesa );
+}
+
+void i810DDInitIoctlFuncs( GLcontext *ctx )
+{
+   ctx->Driver.Flush = i810DDFlush;
+   ctx->Driver.Finish = i810DDFinish;
+}
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810ioctl.h b/xc/lib/GL/mesa/src/drv/i810/i810ioctl.h
new file mode 100644
index 000000000..004b723b4
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/i810/i810ioctl.h
@@ -0,0 +1,36 @@
+#ifndef MGA_IOCTL_H
+#define MGA_IOCTL_H
+
+#include "i810context.h"
+
+
+GLuint *i810AllocDwords( i810ContextPtr imesa, int dwords, GLuint prim );
+
+void i810GetGeneralDmaBufferLocked( i810ContextPtr mmesa ); 
+
+void i810FlushVertices( i810ContextPtr mmesa ); 
+void i810FlushVerticesLocked( i810ContextPtr mmesa );
+
+void i810FlushGeneralLocked( i810ContextPtr imesa );
+void i810WaitAgeLocked( i810ContextPtr imesa, int age );
+void i810WaitAge( i810ContextPtr imesa, int age );
+
+void i810DmaFinish( i810ContextPtr imesa );
+
+void i810RegetLockQuiescent( i810ContextPtr imesa );
+
+void i810DDInitIoctlFuncs( GLcontext *ctx );
+
+void i810SwapBuffers( i810ContextPtr imesa );
+
+GLbitfield i810Clear( GLcontext *ctx, GLbitfield mask, GLboolean all,
+		      GLint cx, GLint cy, GLint cw, GLint ch );
+
+#define FLUSH_BATCH(imesa) do {						\
+        if (I810_DEBUG&DEBUG_VERBOSE_IOCTL)  				\
+              fprintf(stderr, "FLUSH_BATCH in %s\n", __FUNCTION__);	\
+	if (imesa->vertex_dma_buffer) i810FlushVertices(imesa);		\
+} while (0)
+
+
+#endif
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810lib.h b/xc/lib/GL/mesa/src/drv/i810/i810lib.h
deleted file mode 100644
index ad80674f0..000000000
--- a/xc/lib/GL/mesa/src/drv/i810/i810lib.h
+++ /dev/null
@@ -1,130 +0,0 @@
-/*
- * GLX Hardware Device Driver for Intel i810
- * Copyright (C) 1999 Keith Whitwell
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included
- * in all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * KEITH WHITWELL, OR ANY OTHER CONTRIBUTORS BE LIABLE FOR ANY CLAIM, 
- * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
- * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE 
- * OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- *
- *
- */
-
-#ifndef I810LIB_INC
-#define I810LIB_INC
-
-#include <stdio.h>
-
-#include "i810context.h"
-#include "mm.h"
-#include "i810log.h"
-
-
-struct i810_mem_range {
-   unsigned long Start;
-   unsigned long End;
-   unsigned long Size;
-};
-
-struct i810_batch_buffer;
-
-struct i810_ring_buffer {
-   int tail_mask;
-   struct i810_mem_range mem;
-   char *virtual_start;
-   int head;
-   int tail;
-   int space;
-   int synced;
-};
-
-typedef struct {
-   /* logging stuff */
-   GLuint logLevel;
-   FILE *logFile;
-
-   /* bookkeeping for texture swaps */
-   GLuint dma_buffer_age;
-   GLuint current_texture_age;
-        
-   /* options */
-   GLuint nullprims;		/* skip all primitive generation */
-   GLuint boxes;		/* draw performance boxes */
-   GLuint noFallback;		/* don't fall back to software */
-   GLuint skipDma;		/* don't send anything to hardware */
-
-   /* performance counters */	
-   GLuint c_setupPointers;
-   GLuint c_triangles;
-   GLuint c_points;
-   GLuint c_lines;
-   GLuint c_drawWaits;
-   GLuint c_textureSwaps;
-   GLuint c_dmaFlush;
-   GLuint c_overflows;
-
-   GLuint c_ringlost;
-   GLuint c_texlost;
-   GLuint c_ctxlost;
-	
-   GLuint hardwareWentIdle;	/* cleared each swapbuffers, set if a
-				   waitfordmacompletion ever exited
-				   without having to wait */
-
-   unsigned char *texVirtual;
-
-   
-   struct i810_batch_buffer *dma_buffer;
-   struct i810_ring_buffer LpRing;
-   unsigned char *MMIOBase;
-
-} i810Glx_t;
-
-extern i810Glx_t	i810glx;
-
-
-#define I810PACKCOLOR1555(r,g,b,a) \
-  ((((r) & 0xf8) << 7) | (((g) & 0xf8) << 2) | (((b) & 0xf8) >> 3) | \
-    ((a) ? 0x8000 : 0))
-
-#define I810PACKCOLOR565(r,g,b) \
-  ((((r) & 0xf8) << 8) | (((g) & 0xfc) << 3) | (((b) & 0xf8) >> 3))
-
-#define I810PACKCOLOR4444(r,g,b,a) \
-  ((((a) & 0xf0) << 8) | (((r) & 0xf0) << 4) | ((g) & 0xf0) | ((b) >> 4))
-
-
-#include "i810_3d_reg.h"
-
-static __inline__ GLuint i810PackColor(GLuint format, 
-				       GLubyte r, GLubyte g, 
-				       GLubyte b, GLubyte a)
-{
-  switch (format) {
-  case DV_PF_555:
-    return I810PACKCOLOR1555(r,g,b,a);
-  case DV_PF_565:
-    return I810PACKCOLOR565(r,g,b);
-  default:
-     fprintf(stderr, "unknown format %d\n", (int)format);
-    return 0;
-  }
-}
-
-
-
-
-#endif
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810pipeline.c b/xc/lib/GL/mesa/src/drv/i810/i810pipeline.c
index fb2d368fe..34171d943 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810pipeline.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810pipeline.c
@@ -1,13 +1,15 @@
 /*  #include "i810pipeline.h" */
 
 #include <stdio.h>
+
+#include "types.h"
+#include "fog.h"
+
 #include "i810vb.h"
 #include "i810dd.h"
-#include "i810lib.h"
 #include "i810tris.h"
 #include "i810pipeline.h"
 
-#include "fog.h"
 
 static struct gl_pipeline_stage i810_fast_stage = {
    "I810 fast path",
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810span.c b/xc/lib/GL/mesa/src/drv/i810/i810span.c
index 20c463b9b..aa4f22d0c 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810span.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810span.c
@@ -1,7 +1,5 @@
 #include "types.h"
 #include "i810dd.h"
-#include "i810lib.h"
-#include "i810dma.h"
 #include "i810log.h"
 #include "i810span.h"
 #include "i810ioctl.h"
@@ -127,7 +125,7 @@ do {								\
 /* 16 bit depthbuffer functions.
  */
 #define WRITE_DEPTH( _x, _y, d ) \
-   *(GLushort *)(buf + _x*2 + _y*pitch) = d;
+   *(GLushort *)(buf + _x*2 + _y*pitch)  = d;
 
 #define READ_DEPTH( d, _x, _y )	\
    d = *(GLushort *)(buf + _x*2 + _y*pitch);	 
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810state.c b/xc/lib/GL/mesa/src/drv/i810/i810state.c
index bc3391e42..80332e3b3 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810state.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810state.c
@@ -7,24 +7,40 @@
 #include "dd.h"
 
 #include "mm.h"
-#include "i810lib.h"
 #include "i810dd.h"
 #include "i810context.h"
 #include "i810state.h"
-#include "i810depth.h"
 #include "i810tex.h"
 #include "i810log.h"
 #include "i810vb.h"
 #include "i810tris.h"
+#include "i810ioctl.h"
 
+  
+
+static __inline__ GLuint i810PackColor(GLuint format, 
+				       GLubyte r, GLubyte g, 
+				       GLubyte b, GLubyte a)
+{
+   switch (format) {
+   case DV_PF_555:
+      return I810PACKCOLOR1555(r,g,b,a);
+   case DV_PF_565:
+      return I810PACKCOLOR565(r,g,b);
+   default:
+      fprintf(stderr, "unknown format %d\n", (int)format);
+      return 0;
+   }
+}
 
 
 static void i810DDAlphaFunc(GLcontext *ctx, GLenum func, GLclampf ref)
 {
    i810ContextPtr imesa = I810_CONTEXT(ctx);
-
    CARD32 a = (ZA_UPDATE_ALPHAFUNC|ZA_UPDATE_ALPHAREF);
 
+   FLUSH_BATCH(imesa);
+
    switch (ctx->Color.AlphaFunc) {
    case GL_NEVER:    a |= ZA_ALPHA_NEVER;    break;
    case GL_LESS:     a |= ZA_ALPHA_LESS;     break;
@@ -44,10 +60,6 @@ static void i810DDAlphaFunc(GLcontext *ctx, GLenum func, GLclampf ref)
    imesa->Setup[I810_CTXREG_ZA] |= a;
 }
 
-/* This shouldn't get called, as the extension is disabled.  However,
- * there are internal Mesa calls, and rogue use of the api which must be
- * caught.
- */
 static void i810DDBlendEquation(GLcontext *ctx, GLenum mode) 
 {
    if (mode != GL_FUNC_ADD_EXT) {
@@ -60,9 +72,8 @@ static void i810DDBlendEquation(GLcontext *ctx, GLenum mode)
 static void i810DDBlendFunc(GLcontext *ctx, GLenum sfactor, GLenum dfactor)
 {
    i810ContextPtr imesa = I810_CONTEXT(ctx);
-   GLuint a;
-
-   a = SDM_UPDATE_SRC_BLEND | SDM_UPDATE_DST_BLEND;
+   GLuint a = SDM_UPDATE_SRC_BLEND | SDM_UPDATE_DST_BLEND;
+   FLUSH_BATCH(imesa);
 
    switch (ctx->Color.BlendSrcRGB) {
    case GL_ZERO:                a |= SDM_SRC_ZERO; break;
@@ -73,13 +84,8 @@ static void i810DDBlendFunc(GLcontext *ctx, GLenum sfactor, GLenum dfactor)
    case GL_ONE_MINUS_SRC_ALPHA: a |= SDM_SRC_INV_SRC_ALPHA; break;
    case GL_DST_ALPHA:           a |= SDM_SRC_ONE; break;
    case GL_ONE_MINUS_DST_ALPHA: a |= SDM_SRC_ZERO; break;
-   case GL_SRC_ALPHA_SATURATE:      
-      a |= SDM_SRC_SRC_ALPHA;	/* use GFXRENDERSTATE_COLOR_FACTOR ??? */
-      break;
-   default: 
-      i810Error("unknown blend source func");
-      exit(1);
-      return;
+   case GL_SRC_ALPHA_SATURATE:  a |= SDM_SRC_SRC_ALPHA; break;
+   default: return;
    }
 
    switch (ctx->Color.BlendDstRGB) {
@@ -91,10 +97,7 @@ static void i810DDBlendFunc(GLcontext *ctx, GLenum sfactor, GLenum dfactor)
    case GL_ONE_MINUS_SRC_COLOR: a |= SDM_DST_INV_SRC_COLOR; break;
    case GL_DST_ALPHA:           a |= SDM_DST_ONE; break;
    case GL_ONE_MINUS_DST_ALPHA: a |= SDM_DST_ZERO; break;
-   default: 
-      i810Error( "unknown blend dst func");
-      exit(1);
-      return;      
+   default: return;      
    }  
 
    imesa->dirty |= I810_UPLOAD_CTX;
@@ -123,6 +126,8 @@ static void i810DDDepthFunc(GLcontext *ctx, GLenum func)
    i810ContextPtr imesa = I810_CONTEXT(ctx);
    int zmode;
 
+   FLUSH_BATCH(imesa);
+
    switch(func)  { 
    case GL_NEVER: zmode = LCS_Z_NEVER; break;
    case GL_ALWAYS: zmode = LCS_Z_ALWAYS; break;
@@ -134,9 +139,9 @@ static void i810DDDepthFunc(GLcontext *ctx, GLenum func)
    case GL_NOTEQUAL: zmode = LCS_Z_NOTEQUAL; break;
    default: return;
    }
-
+   
    imesa->Setup[I810_CTXREG_LCS] &= ~LCS_Z_MASK;
-   imesa->Setup[I810_CTXREG_LCS] |= LCS_UPDATE_ZMODE | zmode;
+   imesa->Setup[I810_CTXREG_LCS] |= LCS_UPDATE_ZMODE | zmode;   
    imesa->dirty |= I810_UPLOAD_CTX;
 }
 
@@ -144,20 +149,62 @@ static void i810DDDepthMask(GLcontext *ctx, GLboolean flag)
 {
    i810ContextPtr imesa = I810_CONTEXT(ctx);
 
+   FLUSH_BATCH(imesa);
+
    imesa->dirty |= I810_UPLOAD_CTX;
    imesa->Setup[I810_CTXREG_B2] &= ~B2_ZB_WRITE_ENABLE;
-   imesa->Setup[I810_CTXREG_B2] |= B2_UPDATE_ZB_WRITE_ENABLE;
 
    if (flag)
      imesa->Setup[I810_CTXREG_B2] |= B2_ZB_WRITE_ENABLE;
 }
 
 
+/* =============================================================
+ * Polygon stipple 
+ * 
+ * The i810 supports a 4x4 stipple natively, GL wants 32x32. 
+ * Fortunately stipple is usually a repeating pattern.  Could
+ * also consider using a multitexturing mechanism for this, but
+ * that has real issues, too.
+ */
+static void i810DDPolygonStipple( GLcontext *ctx, const GLubyte *mask )
+{
+   i810ContextPtr imesa = I810_CONTEXT(ctx);
+   const GLubyte *m = mask;
+   GLubyte p[4];
+   int i,j,k;
+   int active = (ctx->Polygon.StippleFlag && ctx->PB->primitive == GL_POLYGON);
 
+   FLUSH_BATCH(imesa);
+   ctx->Driver.TriangleCaps |= DD_TRI_STIPPLE;
 
+   if (active) {
+      imesa->dirty |= I810_UPLOAD_CTX;
+      imesa->Setup[I810_CTXREG_ST1] &= ~ST1_ENABLE;
+   }
 
-
-
+   p[0] = mask[0] & 0xf; p[0] |= p[0] << 4;
+   p[1] = mask[4] & 0xf; p[1] |= p[1] << 4;
+   p[2] = mask[8] & 0xf; p[2] |= p[2] << 4;
+   p[3] = mask[12] & 0xf; p[3] |= p[3] << 4;
+
+   for (k = 0 ; k < 8 ; k++)
+      for (j = 0 ; j < 4; j++) 
+	 for (i = 0 ; i < 4 ; i++) 
+	    if (*m++ != p[j]) {
+	       ctx->Driver.TriangleCaps &= ~DD_TRI_STIPPLE;
+	       return;
+	    }
+   
+   imesa->Setup[I810_CTXREG_ST1] &= ~0xffff;
+   imesa->Setup[I810_CTXREG_ST1] |= ( ((p[0] & 0xf) << 0) |
+				      ((p[1] & 0xf) << 4) |
+				      ((p[2] & 0xf) << 8) |
+				      ((p[3] & 0xf) << 12) );   
+
+   if (active)
+      imesa->Setup[I810_CTXREG_ST1] |= ST1_ENABLE;
+}
 
 
 
@@ -167,40 +214,26 @@ static void i810DDDepthMask(GLcontext *ctx, GLboolean flag)
 
 
 static void i810DDScissor( GLcontext *ctx, GLint x, GLint y, 
-			  GLsizei w, GLsizei h )
+			   GLsizei w, GLsizei h )
 { 
    i810ContextPtr imesa = I810_CONTEXT(ctx);
-   __DRIdrawablePrivate *dPriv = imesa->driDrawable;
-   int x1,x2,y1,y2;
 
-   x1 = ctx->Scissor.X;
-   x2 = ctx->Scissor.X + ctx->Scissor.Width - 1;
-   y1 = dPriv->h - ctx->Scissor.Y - ctx->Scissor.Height;
-   y2 = dPriv->h - ctx->Scissor.Y - 1;
-  
-   if (x1 < 0) x1 = 0;
-   if (y1 < 0) y1 = 0;
-   if (x2 >= dPriv->w) x2 = dPriv->w-1;
-   if (y2 >= dPriv->h) y2 = dPriv->h-1;
-
-   if (x1 > x2 || y1 > y2) {
-      x1 = 0; x2 = 0;
-      y2 = 0; y1 = 1;
-   }
+   FLUSH_BATCH(imesa);
+   imesa->scissor_rect.x1 = x;
+   imesa->scissor_rect.y1 = imesa->driDrawable->h - (y+h);
+   imesa->scissor_rect.x2 = x+w;
+   imesa->scissor_rect.y2 = imesa->driDrawable->h - y;
 
-   /* Need to push this into drawing rectangle.
-    */
-#if 0
-   imesa->Setup[I810_CTXREG_SCI0] = GFX_OP_SCISSOR_INFO;
-   imesa->Setup[I810_CTXREG_SCI1] = (y1<<16)|x1;
-   imesa->Setup[I810_CTXREG_SCI2] = (y2<<16)|x2;
 
+   if (I810_DEBUG&DEBUG_VERBOSE_2D) 
+      fprintf(stderr, "SET SCISSOR %d,%d-%d,%d\n", 
+	      imesa->scissor_rect.x1,
+	      imesa->scissor_rect.y1,
+	      imesa->scissor_rect.x2,
+	      imesa->scissor_rect.y2);
 
-   /* Need to intersect with cliprects???
-    */
-   imesa->dirty |= I810_UPLOAD_CTX;
-#endif
 
+   imesa->dirty |= I810_UPLOAD_CLIPRECTS;
 }
 
 
@@ -209,38 +242,56 @@ static void i810DDDither(GLcontext *ctx, GLboolean enable)
 }
 
 
-static GLboolean i810DDSetBuffer(GLcontext *ctx, GLenum mode )
+static GLboolean i810DDSetDrawBuffer(GLcontext *ctx, GLenum mode )
 {
    i810ContextPtr imesa = I810_CONTEXT(ctx);
 
+   FLUSH_BATCH(imesa);
 
-   fprintf(stderr, "i810DDSetBuffer %s\n", gl_lookup_enum_by_nr( mode ));
-
-   imesa->Fallback &= ~I810_FALLBACK_BUFFER;
+   imesa->Fallback &= ~I810_FALLBACK_DRAW_BUFFER;
 
    if (mode == GL_FRONT_LEFT) 
    {
-      imesa->drawOffset = imesa->i810Screen->fbOffset;
+      imesa->drawMap = imesa->driScreen->pFB;
       imesa->BufferSetup[I810_DESTREG_DI1] = (imesa->i810Screen->fbOffset | 
-					      imesa->i810Screen->auxPitchBits);
+					      imesa->i810Screen->backPitchBits);
       imesa->dirty |= I810_UPLOAD_BUFFERS;
       i810XMesaSetFrontClipRects( imesa );
       return GL_TRUE;
    } 
    else if (mode == GL_BACK_LEFT) 
    {
-      imesa->drawOffset = imesa->i810Screen->backOffset;
+      imesa->drawMap = imesa->i810Screen->back.map;
       imesa->BufferSetup[I810_DESTREG_DI1] = (imesa->i810Screen->backOffset | 
-					      imesa->i810Screen->auxPitchBits);
+					      imesa->i810Screen->backPitchBits);
       imesa->dirty |= I810_UPLOAD_BUFFERS;
       i810XMesaSetBackClipRects( imesa );
       return GL_TRUE;
    }
 
-   imesa->Fallback |= I810_FALLBACK_BUFFER;
+   imesa->Fallback |= I810_FALLBACK_DRAW_BUFFER;
    return GL_FALSE;
 }
 
+static void i810DDSetReadBuffer(GLcontext *ctx, GLframebuffer *colorBuffer,
+				GLenum mode )
+{
+   i810ContextPtr imesa = I810_CONTEXT(ctx);
+
+   if (mode == GL_FRONT_LEFT) 
+   {
+      imesa->readMap = imesa->driScreen->pFB;
+      imesa->Fallback &= ~I810_FALLBACK_READ_BUFFER;
+   } 
+   else if (mode == GL_BACK_LEFT) 
+   {
+      imesa->readMap = imesa->i810Screen->back.map;
+      imesa->Fallback &= ~I810_FALLBACK_READ_BUFFER;
+   }
+   else
+      imesa->Fallback |= I810_FALLBACK_READ_BUFFER;
+}
+
 
 
 static void i810DDSetColor(GLcontext *ctx, 
@@ -248,9 +299,7 @@ static void i810DDSetColor(GLcontext *ctx,
 			   GLubyte b, GLubyte a )
 {
    i810ContextPtr imesa = I810_CONTEXT(ctx);
-
-   imesa->MonoColor = i810PackColor( imesa->i810Screen->fbFormat,
-				      r, g, b, a );
+   imesa->MonoColor = i810PackColor( imesa->i810Screen->fbFormat, r, g, b, a );
 }
 
 
@@ -259,9 +308,7 @@ static void i810DDClearColor(GLcontext *ctx,
 			     GLubyte b, GLubyte a )
 {
    i810ContextPtr imesa = I810_CONTEXT(ctx);
-
-   imesa->ClearColor = i810PackColor( imesa->i810Screen->fbFormat,
-				      r, g, b, a );
+   imesa->ClearColor = i810PackColor( imesa->i810Screen->fbFormat, r, g, b, a );
 }
 
 
@@ -274,6 +321,8 @@ static void i810DDCullFaceFrontFace(GLcontext *ctx, GLenum unused)
    i810ContextPtr imesa = I810_CONTEXT(ctx);
    GLuint mode = LCS_CULL_BOTH;
    
+   FLUSH_BATCH(imesa);
+
    if (ctx->Polygon.CullFaceMode != GL_FRONT_AND_BACK) {
       mode = LCS_CULL_CW;
       if (ctx->Polygon.CullFaceMode == GL_FRONT)
@@ -288,27 +337,49 @@ static void i810DDCullFaceFrontFace(GLcontext *ctx, GLenum unused)
    {
       imesa->dirty |= I810_UPLOAD_CTX;
       imesa->Setup[I810_CTXREG_LCS] &= ~LCS_CULL_MASK;
-      imesa->Setup[I810_CTXREG_LCS] |= (LCS_UPDATE_CULL_MODE | mode);
+      imesa->Setup[I810_CTXREG_LCS] |= mode;
    }
 }
 
 
 static void i810DDReducedPrimitiveChange( GLcontext *ctx, GLenum prim )
 {
-   if (ctx->Polygon.CullFlag) {
-      i810ContextPtr imesa = I810_CONTEXT(ctx);
-      GLuint mode = imesa->LcsCullMode;
-
-      if (ctx->PB->primitive != GL_POLYGON)
-	 mode = LCS_CULL_DISABLE;
-
-      imesa->dirty |= I810_UPLOAD_CTX;
-      imesa->Setup[I810_CTXREG_LCS] &= ~LCS_CULL_MASK;
-      imesa->Setup[I810_CTXREG_LCS] |= (LCS_UPDATE_CULL_MODE | mode);
+   i810ContextPtr imesa = I810_CONTEXT(ctx);
+   FLUSH_BATCH(imesa);
 
-      LOCK_HARDWARE(imesa);
-      i810EmitHwStateLocked( imesa );	
-      UNLOCK_HARDWARE(imesa);
+   imesa->dirty |= I810_UPLOAD_CTX;
+   imesa->Setup[I810_CTXREG_LCS] &= ~LCS_CULL_MASK;
+   imesa->Setup[I810_CTXREG_ST1] &= ~ST1_ENABLE;
+   imesa->Setup[I810_CTXREG_AA] &= ~AA_ENABLE;
+
+   switch (ctx->PB->primitive) {
+   case GL_POLYGON:
+      if (ctx->Polygon.StippleFlag && 
+	  (ctx->Driver.TriangleCaps & DD_TRI_STIPPLE))
+	 imesa->Setup[I810_CTXREG_ST1] |= ST1_ENABLE;
+      if (ctx->Polygon.CullFlag) 
+	 imesa->Setup[I810_CTXREG_LCS] |= imesa->LcsCullMode;
+      else 
+	 imesa->Setup[I810_CTXREG_LCS] |= LCS_CULL_DISABLE;
+      if (ctx->Polygon.SmoothFlag)
+	 imesa->Setup[I810_CTXREG_AA] |= AA_ENABLE;
+      break;
+   case GL_LINES:
+      imesa->Setup[I810_CTXREG_LCS] &= ~LCS_LINEWIDTH_0_5;
+      if (ctx->Line.SmoothFlag) {
+	 imesa->Setup[I810_CTXREG_AA] |= AA_ENABLE;  
+	 imesa->Setup[I810_CTXREG_LCS] |= LCS_LINEWIDTH_0_5;
+      } 
+      imesa->Setup[I810_CTXREG_LCS] |= LCS_CULL_DISABLE;
+      break;
+   case GL_POINTS:
+      if (ctx->Point.SmoothFlag)
+	 imesa->Setup[I810_CTXREG_AA] |= AA_ENABLE;      
+      imesa->Setup[I810_CTXREG_LCS] |= LCS_CULL_DISABLE;
+      break;
+   default:
+      imesa->Setup[I810_CTXREG_LCS] |= LCS_CULL_DISABLE;
+      break;
    }
 }
 
@@ -318,15 +389,36 @@ static void i810DDReducedPrimitiveChange( GLcontext *ctx, GLenum prim )
  * Color masks
  */
 
-/* Mesa calls this from the wrong place.
+/* Mesa calls this from the wrong place - it is called a very large
+ * number of redundant times.
  *
- * Its a fallback...
+ * Colormask can be simulated by multipass or multitexture techniques.
  */
 static GLboolean i810DDColorMask(GLcontext *ctx, 
 				 GLboolean r, GLboolean g, 
 				 GLboolean b, GLboolean a )
 {
-   return 1;
+   i810ContextPtr imesa = I810_CONTEXT( ctx );
+   GLuint tmp = 0;
+   GLuint rv = 1;
+
+   imesa->Fallback &= ~I810_FALLBACK_COLORMASK;
+
+   if (r && g && b) {
+      tmp = imesa->Setup[I810_CTXREG_B2] | B2_FB_WRITE_ENABLE;
+   } else if (!r && !g && !b) {
+      tmp = imesa->Setup[I810_CTXREG_B2] & ~B2_FB_WRITE_ENABLE;
+   } else {
+      rv = 0;
+      imesa->Fallback |= I810_FALLBACK_COLORMASK;
+   }
+      
+   if (tmp != imesa->Setup[I810_CTXREG_B2]) {
+      FLUSH_BATCH(imesa);
+      imesa->Setup[I810_CTXREG_B2] = tmp;
+   }
+
+   return rv;
 }
 
 /* Seperate specular not fully implemented in hardware...  Needs
@@ -336,8 +428,14 @@ static GLboolean i810DDColorMask(GLcontext *ctx,
 static void i810DDLightModelfv(GLcontext *ctx, GLenum pname, 
 			      const GLfloat *param)
 {
-   if (pname == GL_LIGHT_MODEL_COLOR_CONTROL) {
-      I810_CONTEXT(ctx)->new_state |= I810_NEW_TEXTURE;
+   if (pname == GL_LIGHT_MODEL_COLOR_CONTROL) 
+   {            
+      i810ContextPtr imesa = I810_CONTEXT( ctx );
+      FLUSH_BATCH(imesa);
+
+      imesa->Fallback &= ~I810_FALLBACK_SPECULAR;
+      if (ctx->Light.Model.ColorControl == GL_SEPARATE_SPECULAR_COLOR)
+	 imesa->Fallback |= I810_FALLBACK_SPECULAR;
    }
 }
 
@@ -372,64 +470,105 @@ static void i810DDEnable(GLcontext *ctx, GLenum cap, GLboolean state)
 
    switch(cap) {
    case GL_ALPHA_TEST:
+      FLUSH_BATCH(imesa);
       imesa->dirty |= I810_UPLOAD_CTX;
       imesa->Setup[I810_CTXREG_B1] &= ~B1_ALPHA_TEST_ENABLE;
-      imesa->Setup[I810_CTXREG_B1] |= B1_UPDATE_ALPHA_TEST_ENABLE;
       if (state)
 	 imesa->Setup[I810_CTXREG_B1] |= B1_ALPHA_TEST_ENABLE;
       break;
    case GL_BLEND:
+      FLUSH_BATCH(imesa);
       imesa->dirty |= I810_UPLOAD_CTX;
       imesa->Setup[I810_CTXREG_B1] &= ~B1_BLEND_ENABLE;
-      imesa->Setup[I810_CTXREG_B1] |= B1_UPDATE_BLEND_ENABLE;
       if (state)
 	 imesa->Setup[I810_CTXREG_B1] |= B1_BLEND_ENABLE;
       break;
    case GL_DEPTH_TEST:
+      FLUSH_BATCH(imesa);
       imesa->dirty |= I810_UPLOAD_CTX;
       imesa->Setup[I810_CTXREG_B1] &= ~B1_Z_TEST_ENABLE;
-      imesa->Setup[I810_CTXREG_B1] |= B1_UPDATE_Z_TEST_ENABLE;
       if (state)
 	 imesa->Setup[I810_CTXREG_B1] |= B1_Z_TEST_ENABLE;
       break;
    case GL_SCISSOR_TEST:
-#if 0
-      imesa->dirty |= I810_UPLOAD_CTX;
-      imesa->Setup[I810_CTXREG_SC] &= ~SC_ENABLE_MASK;
-      imesa->Setup[I810_CTXREG_SC] |= SC_UPDATE_SCISSOR;
-      if (state)
-	 imesa->Setup[I810_CTXREG_SC] |= SC_ENABLE;
-#endif
+      FLUSH_BATCH(imesa);
+      imesa->scissor = state;
+      imesa->dirty |= I810_UPLOAD_CLIPRECTS;
+      break;
+   case GL_POLYGON_STIPPLE:      
+      if ((ctx->Driver.TriangleCaps & DD_TRI_STIPPLE) &&
+	  ctx->PB->primitive == GL_POLYGON) 
+      {
+	 FLUSH_BATCH(imesa);
+	 imesa->dirty |= I810_UPLOAD_CTX;
+	 imesa->Setup[I810_CTXREG_ST1] &= ~ST1_ENABLE;
+	 if (state)
+	    imesa->Setup[I810_CTXREG_ST1] |= ST1_ENABLE;
+      }
+      break;
+   case GL_LINE_SMOOTH:
+      if (ctx->PB->primitive == GL_LINE) {
+	 FLUSH_BATCH(imesa);
+	 imesa->dirty |= I810_UPLOAD_CTX;
+	 imesa->Setup[I810_CTXREG_AA] &= ~AA_ENABLE;
+	 imesa->Setup[I810_CTXREG_LCS] &= ~LCS_LINEWIDTH_0_5;
+	 if (state) {
+	    imesa->Setup[I810_CTXREG_AA] |= AA_ENABLE;
+	    imesa->Setup[I810_CTXREG_LCS] |= LCS_LINEWIDTH_0_5;
+	 }
+      }
+      break;
+   case GL_POINT_SMOOTH:
+      if (ctx->PB->primitive == GL_POINT) {
+	 FLUSH_BATCH(imesa);
+	 imesa->dirty |= I810_UPLOAD_CTX;
+	 imesa->Setup[I810_CTXREG_AA] &= ~AA_ENABLE;
+	 if (state) 
+	    imesa->Setup[I810_CTXREG_AA] |= AA_ENABLE;
+      }
+      break;
+   case GL_POLYGON_SMOOTH:
+      if (ctx->PB->primitive == GL_POLYGON) {
+	 FLUSH_BATCH(imesa);
+	 imesa->dirty |= I810_UPLOAD_CTX;
+	 imesa->Setup[I810_CTXREG_AA] &= ~AA_ENABLE;
+	 if (state) 
+	    imesa->Setup[I810_CTXREG_AA] |= AA_ENABLE;
+      }
       break;
    case GL_FOG:
+      FLUSH_BATCH(imesa);
       imesa->dirty |= I810_UPLOAD_CTX;
       imesa->Setup[I810_CTXREG_B1] &= ~B1_FOG_ENABLE;
-      imesa->Setup[I810_CTXREG_B1] |= B1_UPDATE_FOG_ENABLE;
       if (state)
 	 imesa->Setup[I810_CTXREG_B1] |= B1_FOG_ENABLE;
       break;
    case GL_CULL_FACE:
       if (ctx->PB->primitive == GL_POLYGON) {
+	 FLUSH_BATCH(imesa);
 	 imesa->dirty |= I810_UPLOAD_CTX;
 	 imesa->Setup[I810_CTXREG_LCS] &= ~LCS_CULL_MASK;
-	 imesa->Setup[I810_CTXREG_LCS] |= LCS_UPDATE_CULL_MODE;
 	 if (state)
 	    imesa->Setup[I810_CTXREG_LCS] |= imesa->LcsCullMode;
 	 else
 	    imesa->Setup[I810_CTXREG_LCS] |= LCS_CULL_DISABLE;
       }
       break;
+   case GL_TEXTURE_1D:      
+   case GL_TEXTURE_3D:      
+      FLUSH_BATCH(imesa);
+      imesa->new_state |= I810_NEW_TEXTURE;
+      break;
    case GL_TEXTURE_2D:      
+      FLUSH_BATCH(imesa);
       imesa->new_state |= I810_NEW_TEXTURE;
       imesa->dirty |= I810_UPLOAD_CTX;
       if (ctx->Texture.CurrentUnit == 0) {
 	 imesa->Setup[I810_CTXREG_MT] &= ~MT_TEXEL0_ENABLE;
-	 imesa->Setup[I810_CTXREG_MT] |= MT_UPDATE_TEXEL0_STATE;
 	 if (state)
 	    imesa->Setup[I810_CTXREG_MT] |= MT_TEXEL0_ENABLE;
       } else {
 	 imesa->Setup[I810_CTXREG_MT] &= ~MT_TEXEL1_ENABLE;
-	 imesa->Setup[I810_CTXREG_MT] |= MT_UPDATE_TEXEL1_STATE;
 	 if (state)
 	    imesa->Setup[I810_CTXREG_MT] |= MT_TEXEL1_ENABLE;
       }
@@ -444,92 +583,20 @@ static void i810DDEnable(GLcontext *ctx, GLenum cap, GLboolean state)
 /* =============================================================
  */
 
-/* Delightfully few possibilities:
- */
-void i810DDPrintState( const char *msg, GLuint state )
-{
-   fprintf(stderr, "%s (0x%x): %s\n",
-	   msg,
-	   (unsigned int) state,
-	   (state & I810_NEW_TEXTURE) ? "texture, " : "");
-}
 
 void i810DDUpdateHwState( GLcontext *ctx )
 {
    i810ContextPtr imesa = I810_CONTEXT(ctx);
 
-   if (imesa->new_state & I810_NEW_TEXTURE)
+   if (imesa->new_state & I810_NEW_TEXTURE) {
+      FLUSH_BATCH(imesa);
       i810UpdateTextureState( ctx );
-
-   imesa->new_state = 0;
-
-   if (imesa->dirty) {
-      LOCK_HARDWARE(imesa);
-      i810EmitHwStateLocked( imesa );	
-      UNLOCK_HARDWARE(imesa);
-   }
-}
-
-
-
-/*
- * i810DmaExecute
- * Add a block of data to the dma buffer
- *
- * -- In ring buffer mode, must be called with lock held, and translates to 
- *    OUT_RING rather than outbatch
- *
- * -- In dma mode, probably won't be used because state updates will have
- *    to be done via the kernel for security...
- */
-static void i810DmaExecute( i810ContextPtr imesa,
-			    GLuint *code, int dwords, const char *where ) 
-{	
-   if (I810_DEBUG & DEBUG_VERBOSE_RING)
-      fprintf(stderr, "i810DmaExecute (%s)\n", where);
-
-   if (dwords & 1) {
-      i810Error( "Misaligned buffer in i810DmaExecute\n" );
-      exit(1);
    }
 
-   {
-      int i;
-      BEGIN_BATCH( imesa, dwords);
-
-      for ( i = 0 ; i < dwords ; i++ ) {
-	 if (0) fprintf(stderr, "%d: %x\n", i, code[i]);
-	 OUT_BATCH( code[i] );
-      }
-
-      ADVANCE_BATCH();
-   }
+   imesa->new_state = 0;
 }
 
 
-
-void i810EmitScissorValues( i810ContextPtr imesa, int nr, int emit )
-{
-   int x1 = imesa->pClipRects[nr].x1 - imesa->drawX;
-   int y1 = imesa->pClipRects[nr].y1 - imesa->drawY;
-   int x2 = imesa->pClipRects[nr].x2 - imesa->drawX;
-   int y2 = imesa->pClipRects[nr].y2 - imesa->drawY;
-
-   if (I810_DEBUG&DEBUG_VERBOSE_DRI)
-      fprintf(stderr, "i810EmitScissorValues %d,%d - %d,%d\n",
-	      x1,y1,x2,y2);
-
-   imesa->ClipSetup[I810_CLIPREG_SCI1] = (x1 | (y1 << 16));
-   imesa->ClipSetup[I810_CLIPREG_SCI2] = (x2 | (y2 << 16));
-   imesa->ClipSetup[I810_CLIPREG_SC] = ( GFX_OP_SCISSOR |
-					 SC_UPDATE_SCISSOR |
-					 SC_ENABLE );
-
-   if (emit)
-      i810DmaExecute( imesa, 
-		      imesa->ClipSetup, I810_CLIP_SETUP_SIZE, "cliprect" );
-}
-
 void i810EmitDrawingRectangle( i810ContextPtr imesa )
 {
    __DRIdrawablePrivate *dPriv = imesa->driDrawable;
@@ -545,7 +612,6 @@ void i810EmitDrawingRectangle( i810ContextPtr imesa )
     */
    imesa->BufferSetup[I810_DESTREG_DR4] = ((y0<<16) | 
 					   (((unsigned)x0)&0xFFFF));
-
   
    /* Clip to screen.
     */
@@ -556,74 +622,74 @@ void i810EmitDrawingRectangle( i810ContextPtr imesa )
 
 
    /* Onscreen drawing rectangle.
-    *
-    * TODO: clip again to GL scissor values.
     */
    imesa->BufferSetup[I810_DESTREG_DR2] = ((y0<<16) | x0);
-   imesa->BufferSetup[I810_DESTREG_DR3] = ((y1<<16) | x1);
+   imesa->BufferSetup[I810_DESTREG_DR3] = (((y1+1)<<16) | (x1+1));
    imesa->dirty |= I810_UPLOAD_BUFFERS;
 }
 
 
 static void i810DDPrintDirty( const char *msg, GLuint state )
 {
-   fprintf(stderr, "%s (0x%x): %s%s%s%s%s%s%s\n",	   
+   fprintf(stderr, "%s (0x%x): %s%s%s%s%s\n",	   
 	   msg,
 	   (unsigned int) state,
-	   (state & I810_REFRESH_RING)      ? "read-lp-ring, " : "",
-	   (state & I810_REQUIRE_QUIESCENT) ? "req-quiescent, " : "",
 	   (state & I810_UPLOAD_TEX0IMAGE)  ? "upload-tex0, " : "",
 	   (state & I810_UPLOAD_TEX1IMAGE)  ? "upload-tex1, " : "",
 	   (state & I810_UPLOAD_CTX)        ? "upload-ctx, " : "",
 	   (state & I810_UPLOAD_BUFFERS)    ? "upload-bufs, " : "",
-	   (state & I810_EMIT_CLIPRECT)     ? "emit-single-cliprect, " : ""
+	   (state & I810_UPLOAD_CLIPRECTS)  ? "upload-cliprects, " : ""
 	   );
 }
 
 
-/* Spew the state onto the ringbuffer and/or texture memory.
+/* Push the state into the sarea and/or texture memory.
  */
 void i810EmitHwStateLocked( i810ContextPtr imesa )
 {
    if (I810_DEBUG & DEBUG_VERBOSE_API)
-      i810DDPrintDirty( "i810EmitHwStateLocked", imesa->dirty );
-
-   if (imesa->dirty & I810_REQUIRE_QUIESCENT) {
-      i810glx.c_drawWaits += _I810Sync( imesa );
-   }
+      i810DDPrintDirty( "\n\n\ni810EmitHwStateLocked", imesa->dirty );
 
-   /* TODO: Refine mechanism to be equivalent to UploadSubImage
-    */
-   if ((imesa->dirty & I810_UPLOAD_TEX0IMAGE) && imesa->CurrentTexObj[0])
-      i810UploadTexImages(imesa, imesa->CurrentTexObj[0]);
+   if (imesa->dirty & ~I810_UPLOAD_CLIPRECTS)
+   {
+      if ((imesa->dirty & I810_UPLOAD_TEX0IMAGE) && imesa->CurrentTexObj[0])
+	 i810UploadTexImages(imesa, imesa->CurrentTexObj[0]);
    
-   if ((imesa->dirty & I810_UPLOAD_TEX1IMAGE) && imesa->CurrentTexObj[1])
-      i810UploadTexImages(imesa, imesa->CurrentTexObj[1]);
+      if ((imesa->dirty & I810_UPLOAD_TEX1IMAGE) && imesa->CurrentTexObj[1])
+	 i810UploadTexImages(imesa, imesa->CurrentTexObj[1]);
   
-   if ((imesa->dirty & I810_UPLOAD_CTX)) {
-      i810DmaExecute( imesa, imesa->Setup, I810_CTX_SETUP_SIZE, "context" );
-
-      if (imesa->CurrentTexObj[0])
-	 i810DmaExecute( imesa, imesa->CurrentTexObj[0]->Setup,
-			 I810_TEX_SETUP_SIZE, "tex-0");
+      if (imesa->dirty & I810_UPLOAD_CTX)
+	 memcpy( imesa->sarea->ContextState, 
+		 imesa->Setup, 
+		 sizeof(imesa->Setup) );
+
+      if ((imesa->dirty & I810_UPLOAD_TEX0) && imesa->CurrentTexObj[0]) {
+	 imesa->sarea->dirty |= I810_UPLOAD_TEX0;
+	 memcpy(imesa->sarea->TexState[0],
+		imesa->CurrentTexObj[0]->Setup,
+		sizeof(imesa->sarea->TexState[0]));
+      }
 
-      if (imesa->CurrentTexObj[1]) 
-	 i810DmaExecute( imesa, imesa->CurrentTexObj[1]->Setup,
-			 I810_TEX_SETUP_SIZE, "tex-1");
+      if ((imesa->dirty & I810_UPLOAD_TEX1) && imesa->CurrentTexObj[1]) {
+	 imesa->sarea->dirty |= I810_UPLOAD_TEX1;
+	 memcpy(imesa->sarea->TexState[1],
+		imesa->CurrentTexObj[1]->Setup,
+		sizeof(imesa->sarea->TexState[1]));
+      }
+      
+      if (imesa->dirty & I810_UPLOAD_BUFFERS) 
+	 memcpy( imesa->sarea->BufferState, 
+		 imesa->BufferSetup, 
+		 sizeof(imesa->BufferSetup) );
+
+      imesa->sarea->dirty |= (imesa->dirty & 
+			      ~(I810_UPLOAD_TEX1|I810_UPLOAD_TEX0));
+      imesa->dirty &= I810_UPLOAD_CLIPRECTS;
    }
-
-   if (imesa->dirty & I810_UPLOAD_BUFFERS) 
-      i810DmaExecute( imesa, imesa->BufferSetup, 
-		      I810_DEST_SETUP_SIZE, "buffers" );
-   
-   if (imesa->dirty & I810_EMIT_CLIPRECT)
-      i810DmaExecute( imesa, imesa->ClipSetup, 
-		      I810_CLIP_SETUP_SIZE, "cliprect" );
-
-   imesa->dirty = 0;
 }
 
 
+
 void i810DDInitState( i810ContextPtr imesa )
 {
    i810ScreenPrivate *i810Screen = imesa->i810Screen;
@@ -797,7 +863,7 @@ void i810DDInitState( i810ContextPtr imesa )
 				     LCS_UPDATE_ZMODE |
 				     LCS_Z_LESS |
 				     LCS_UPDATE_LINEWIDTH |
-				     (0x2<<LCS_LINEWIDTH_SHIFT) |
+				     LCS_LINEWIDTH_1_0 |
 				     LCS_UPDATE_ALPHA_INTERP |
 				     LCS_ALPHA_INTERP |
 				     LCS_UPDATE_FOG_INTERP |
@@ -825,11 +891,9 @@ void i810DDInitState( i810ContextPtr imesa )
    imesa->Setup[I810_CTXREG_ST0] = GFX_OP_STIPPLE;
    imesa->Setup[I810_CTXREG_ST1] = 0;
 
-
-
    imesa->Setup[I810_CTXREG_AA] = ( GFX_OP_ANTIALIAS |
 				    AA_UPDATE_EDGEFLAG |
-				    0 |
+				    AA_ENABLE_EDGEFLAG | /* ? */
 				    AA_UPDATE_POLYWIDTH |
 				    AA_POLYWIDTH_05 |
 				    AA_UPDATE_LINEWIDTH |
@@ -839,38 +903,29 @@ void i810DDInitState( i810ContextPtr imesa )
 				    AA_UPDATE_AA_ENABLE |
 				    0 );
 
-
-
-   memset(imesa->ClipSetup, 0, sizeof(imesa->ClipSetup));
-   imesa->ClipSetup[I810_CLIPREG_SCI0] = GFX_OP_SCISSOR_INFO; 
-   imesa->ClipSetup[I810_CLIPREG_SCI1] = 0;
-   imesa->ClipSetup[I810_CLIPREG_SCI2] = 0;
-   imesa->ClipSetup[I810_CLIPREG_SC] = ( GFX_OP_SCISSOR | 
-					 SC_UPDATE_SCISSOR |
-					 0 );
-
-   /* This stuff is all invarient as long as we are using
-    * shared back and depth buffers.  
-    */
    memset(imesa->BufferSetup, 0, sizeof(imesa->BufferSetup));
-   imesa->drawOffset = i810Screen->backOffset;
    imesa->BufferSetup[I810_DESTREG_DI0] = CMD_OP_DESTBUFFER_INFO;
 
-   if (imesa->glCtx->Color.DriverDrawBuffer == GL_BACK_LEFT)
+   if (imesa->glCtx->Color.DriverDrawBuffer == GL_BACK_LEFT) {
+      imesa->drawMap = i810Screen->back.map;
       imesa->BufferSetup[I810_DESTREG_DI1] = (i810Screen->backOffset | 
-					      i810Screen->auxPitchBits);
-   else
+					      i810Screen->backPitchBits);
+   } else {
+      imesa->drawMap = imesa->driScreen->pFB;
       imesa->BufferSetup[I810_DESTREG_DI1] = (i810Screen->fbOffset | 
-					      i810Screen->auxPitchBits);
+					      i810Screen->backPitchBits);
+   }
 
+   if (imesa->glCtx->Color.DriverDrawBuffer == GL_BACK_LEFT) {
+      imesa->readMap = i810Screen->back.map;
+   } else {
+      imesa->readMap = imesa->driScreen->pFB;
+   }
 
    imesa->BufferSetup[I810_DESTREG_DV0] = GFX_OP_DESTBUFFER_VARS;
    imesa->BufferSetup[I810_DESTREG_DV1] = (DV_HORG_BIAS_OGL |
 					   DV_VORG_BIAS_OGL |
 					   i810Screen->fbFormat);
-   imesa->BufferSetup[I810_DESTREG_ZB0] = CMD_OP_Z_BUFFER_INFO;
-   imesa->BufferSetup[I810_DESTREG_ZB1] = (i810Screen->depthOffset | 
-					   i810Screen->auxPitchBits);
 
    imesa->BufferSetup[I810_DESTREG_DR0] = GFX_OP_DRAWRECT_INFO;
    imesa->BufferSetup[I810_DESTREG_DR1] = DR1_RECT_CLIP_ENABLE;
@@ -886,17 +941,31 @@ void i810DDUpdateState( GLcontext *ctx )
 {
    i810ContextPtr imesa = I810_CONTEXT( ctx );
 
+   /* Have to do this here to detect texture fallbacks in time:
+    */
+   if (I810_CONTEXT(ctx)->new_state & I810_NEW_TEXTURE)
+      i810DDUpdateHwState( ctx );
+
+
    if (ctx->NewState & INTERESTED) {
       i810DDChooseRenderState(ctx);  
       i810ChooseRasterSetupFunc(ctx);
    }
+
+   if (0) 
+      fprintf(stderr, "IndirectTriangles %x Fallback %x\n", 
+	      imesa->IndirectTriangles, imesa->Fallback);
    
-   /* TODO: stop mesa from clobbering these.
-    */
-   ctx->Driver.PointsFunc=imesa->PointsFunc;
-   ctx->Driver.LineFunc=imesa->LineFunc;
-   ctx->Driver.TriangleFunc=imesa->TriangleFunc;
-   ctx->Driver.QuadFunc=imesa->QuadFunc;
+   if (!imesa->Fallback)
+   {
+      ctx->IndirectTriangles &= ~DD_SW_RASTERIZE;
+      ctx->IndirectTriangles |= imesa->IndirectTriangles;
+
+      ctx->Driver.PointsFunc=imesa->PointsFunc;
+      ctx->Driver.LineFunc=imesa->LineFunc;
+      ctx->Driver.TriangleFunc=imesa->TriangleFunc;
+      ctx->Driver.QuadFunc=imesa->QuadFunc;
+   }
 }
 
 
@@ -919,8 +988,13 @@ void i810DDInitStateFuncs(GLcontext *ctx)
    ctx->Driver.ReducedPrimitiveChange = i810DDReducedPrimitiveChange;
    ctx->Driver.RenderStart = i810DDUpdateHwState; 
    ctx->Driver.RenderFinish = 0; 
+   
+   ctx->Driver.PolygonStipple = i810DDPolygonStipple;
+   ctx->Driver.LineStipple = 0;
+
+   ctx->Driver.SetReadBuffer = i810DDSetReadBuffer;
+   ctx->Driver.SetDrawBuffer = i810DDSetDrawBuffer;
 
-   ctx->Driver.SetBuffer = i810DDSetBuffer;
    ctx->Driver.Color = i810DDSetColor;
    ctx->Driver.ClearColor = i810DDClearColor;
    ctx->Driver.Dither = i810DDDither;
@@ -928,6 +1002,5 @@ void i810DDInitStateFuncs(GLcontext *ctx)
    ctx->Driver.Index = 0;
    ctx->Driver.ClearIndex = 0;
    ctx->Driver.IndexMask = 0;
-
 }
 
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810state.h b/xc/lib/GL/mesa/src/drv/i810/i810state.h
index d0551631e..b43a1a4ff 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810state.h
+++ b/xc/lib/GL/mesa/src/drv/i810/i810state.h
@@ -1,13 +1,12 @@
 #ifndef _I810_STATE_H
 #define _I810_STATE_H
 
+#include "i810context.h"
 
 extern void i810DDUpdateHwState( GLcontext *ctx );
 extern void i810DDUpdateState( GLcontext *ctx );
 extern void i810DDInitState( i810ContextPtr imesa );
 extern void i810DDInitStateFuncs( GLcontext *ctx );
 
-extern void i810DDPrintState( const char *msg, GLuint state );
-
 
 #endif
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810swap.c b/xc/lib/GL/mesa/src/drv/i810/i810swap.c
deleted file mode 100644
index 9220ef9f4..000000000
--- a/xc/lib/GL/mesa/src/drv/i810/i810swap.c
+++ /dev/null
@@ -1,259 +0,0 @@
-
-
-#include "types.h"
-#include "vbrender.h"
-#include "i810log.h"
-
-#include <stdio.h>
-#include <stdlib.h>
-
-#include "mm.h"
-#include "i810lib.h"
-#include "i810dd.h"
-#include "i810dma.h"
-#include "i810state.h"
-#include "i810swap.h"
-#include "i810dma.h"
-
-/*
- * i810BackToFront
- *
- * Blit the visible rectangles from the back buffer to the screen.
- * Respects the frontbuffer cliprects, and applies the offset
- * necessary if the backbuffer is positioned elsewhere in the screen.
- */
-static int i810BackToFront( i810ContextPtr imesa ) 
-{
-   __DRIdrawablePrivate *dPriv = imesa->driDrawable;
-   i810ScreenPrivate *i810Screen = imesa->i810Screen;
-
-
-   /* Use the frontbuffer cliprects:
-    */
-   XF86DRIClipRectPtr pbox = dPriv->pClipRects;
-   int nbox = dPriv->numClipRects;
-
-   if( nbox ) 
-   {
-      int i;
-      int pitch = i810Screen->auxPitch;
-
-      int dx = dPriv->auxX - dPriv->x;
-      int dy = dPriv->auxY - dPriv->y;
-      int ofs = ( i810Screen->backOffset + 
-		  dy * i810Screen->auxPitch +
-		  dx * i810Screen->cpp);
-
-      unsigned int BR13 = ((i810Screen->fbStride) |
-			   (0xCC << 16));
-
-
-      for (i=0; i < nbox; i++, pbox++) {
-	 int w = pbox->x2 - pbox->x1;
-	 int h = pbox->y2 - pbox->y1;
-
-	 int start = ofs + pbox->x1*i810Screen->cpp + pbox->y1*pitch;
-	 int dst = pbox->x1*i810Screen->cpp + pbox->y1*i810Screen->fbStride;
-	 
-	 if (I810_DEBUG&DEBUG_VERBOSE_2D) 
-	    fprintf(stderr, "i810BackToFront %d,%d-%dx%d\n",
-		    pbox->x1,pbox->y1,w,h);
-
-	 {
-	    BEGIN_BATCH( imesa, 6 );
-	    OUT_BATCH( BR00_BITBLT_CLIENT | BR00_OP_SRC_COPY_BLT | 0x4 );
-	    OUT_BATCH( BR13 );	/* dest pitch, rop */
-
-	    OUT_BATCH( (h << 16) | (w * i810Screen->cpp));
-	    OUT_BATCH( dst );
-
-	    OUT_BATCH( pitch );	/* src pitch */
-	    OUT_BATCH( start );
-	    ADVANCE_BATCH();
-	 }
-      }
-   }
-  
-   return Success;   
-}
-
-
-/*
- * ClearBox
- *
- * Add hardware commands to draw a filled box for the debugging
- * display.  These are drawn into the current drawbuffer, so should
- * work with both front and backbuffer rendering.  (However, it
- * is only ever called on swapbuffer - ie backbuffer rendering).
- */
-static void ClearBox( i810ContextPtr imesa,
-		      int cx, int cy, int cw, int ch, 
-		      int r, int g, int b ) 
-{
-   __DRIdrawablePrivate *dPriv = imesa->driDrawable;
-   i810ScreenPrivate *i810Screen = imesa->i810Screen;
-
-   int _nc = imesa->numClipRects;
-      
-   while (_nc--) {
-      int x1 = cx + dPriv->x;
-      int y1 = cy + dPriv->y;
-      int x2 = x1 + cw;
-      int y2 = y1 + ch;
-
-      if (imesa->needClip) {
-	 int rx1 = imesa->pClipRects[_nc].x1;
-	 int ry1 = imesa->pClipRects[_nc].y1;
-	 int rx2 = imesa->pClipRects[_nc].x2;
-	 int ry2 = imesa->pClipRects[_nc].y2;
-
-
-	 if (x2 < rx1) continue;
-	 if (y2 < ry1) continue;
-	 if (x1 > rx2) continue;
-	 if (y1 > ry2) continue;
-	 if (x1 < rx1) x1 = rx1;
-	 if (y1 < ry1) y1 = ry1;
-	 if (x2 > rx2) x2 = rx2;
-	 if (y2 > ry2) y2 = ry2;
-      }
-
-
-      {
-	 int start = (imesa->drawOffset + 
-		      y1 * i810Screen->auxPitch + 
-		      x1 * i810Screen->cpp);
-
-	 BEGIN_BATCH( imesa, 6 );
-   
-	 OUT_BATCH( BR00_BITBLT_CLIENT | BR00_OP_COLOR_BLT | 0x3 );
-	 OUT_BATCH( BR13_SOLID_PATTERN | 
-		    (0xF0 << 16) |
-		    i810Screen->auxPitch );
-	 OUT_BATCH( ((y2-y1) << 16) | ((x2-x1) * i810Screen->cpp));
-	 OUT_BATCH( start );
-	 OUT_BATCH( i810PackColor( i810Screen->fbFormat, r, g, b, 0 ) );
-	 OUT_BATCH( 0 );
-
-	 if (I810_DEBUG&DEBUG_VERBOSE_2D) 
-	    fprintf(stderr, "box %d,%x %d,%d col %x (%d,%d,%d)\n", 
-		    (int)x1,(int)y1, (int)x2,(int)y2,
-		    (int)i810PackColor( i810Screen->fbFormat, r, g, b, 0 ), 
-		    (int)r, (int)g, (int)b);
-	 
-	 ADVANCE_BATCH();
-      }
-   }
-}
-
-
-/*
- * performanceBoxes
- * Draw some small boxesin the corner of the buffer
- * based on some performance information
- */
-static void i810PerformanceBoxes( i810ContextPtr imesa ) {
-	
-
-   /* draw a box to show we are active, so if it is't seen it means
-      that it is completely software based.  Purple is traditional for
-      direct rendering */
-   ClearBox( imesa, 4, 4, 8, 8, 255, 0, 255 );
-	
-   /* draw a red box if we had to wait for drawing to complete 
-      (software render or texture swap) */
-   if ( i810glx.c_drawWaits ) {
-      ClearBox( imesa, 16, 4, 8, 8, 255, 0, 0 );
-      i810glx.c_drawWaits = 0;
-   }
-	
-   /* draw a blue box if the 3d context was lost
-    */
-   if ( i810glx.c_ctxlost ) {
-      ClearBox( imesa, 28, 4, 8, 8, 0, 0, 255 );
-      i810glx.c_ctxlost = 0;
-   }
-	
-   /* draw an orange box if texture state was stomped */
-   if ( i810glx.c_texlost ) {
-      ClearBox( imesa, 40, 4, 8, 8, 255, 128, 0 );
-      i810glx.c_texlost = 0;
-   }
-
-   /* draw a yellow box if textures were swapped */
-   if ( i810glx.c_textureSwaps ) {
-      ClearBox( imesa, 56, 4, 8, 8, 255, 255, 0 );
-      i810glx.c_textureSwaps = 0;
-   }
-		
-   /* draw a green box if we had to wait for dma to complete (full utilization) 
-      on the previous frame */
-   if ( !i810glx.hardwareWentIdle ) {
-      ClearBox( imesa, 72, 4, 8, 8, 0, 255, 0 );
-   }
-   i810glx.hardwareWentIdle = 0;
-	
-#if 0
-   /* show buffer utilization */
-   if ( i810glx.c_dmaFlush > 1 ) {
-      /* draw a solid bar if we flushed more than one buffer */
-      ClearBox( imesa, 4, 16, 252, 4, 255, 32, 32 );
-   } else {
-      /* draw bars to represent the utilization of primary and
-         secondary buffers */
-      ClearBox( imesa, 4, 16, 252, 4, 32, 32, 32 );
-      t = i810glx.dma_buffer->mem.Size;
-      w = 252 * i810glx.dma_buffer->head / t;
-      if ( w < 1 ) {
-	 w = 1;
-      }
-      ClearBox( imesa, 4, 16, w, 4, 196, 128, 128 );
-   }
-#endif
-
-   i810glx.c_dmaFlush = 0;
-}
-
-
-static void i810SendDmaFlush( i810ContextPtr imesa )
-{
-   BEGIN_BATCH( imesa, 2 );
-   OUT_BATCH( INST_PARSER_CLIENT | INST_OP_FLUSH );
-   OUT_BATCH( 0 );
-   ADVANCE_BATCH();
-}
-
-void i810SwapBuffers( i810ContextPtr imesa )
-{
-   if (I810_DEBUG & DEBUG_VERBOSE_API)
-      fprintf(stderr, "i810SwapBuffers()\n");
-
-   LOCK_HARDWARE( imesa );
-   i810SendDmaFlush( imesa );
-   if ( i810glx.boxes )  
-      i810PerformanceBoxes( imesa );
-   i810BackToFront( imesa );
-
-   /* Check if we are emitting thousands of tiny frames, and if so
-    * throttle them.  Would be better to simply go to sleep until
-    * our stuff had cleared, rather than polling.  A dma solution
-    * should provide for this.
-    */   
-   if (imesa->lastSwap >= imesa->sarea->lastWrap &&
-       imesa->lastSwap >= imesa->sarea->lastSync) {
-
-      if (I810_DEBUG & DEBUG_VERBOSE_API)
-	 fprintf(stderr, ".");
-
-      _I810Sync( imesa );
-   }
-
-   imesa->lastSwap = imesa->sarea->lastSync;
-   if (imesa->sarea->lastWrap > imesa->lastSwap) 
-      imesa->lastSwap = imesa->sarea->lastWrap;
-      
-   
-   UNLOCK_HARDWARE( imesa );
-
-/*     XSync(imesa->display, 0); */
-}
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810swap.h b/xc/lib/GL/mesa/src/drv/i810/i810swap.h
deleted file mode 100644
index 1657a717d..000000000
--- a/xc/lib/GL/mesa/src/drv/i810/i810swap.h
+++ /dev/null
@@ -1,6 +0,0 @@
-#ifndef I810_SWAP_H
-#define I810_SWAP_H
-
-extern void i810SwapBuffers( i810ContextPtr imesa );
-
-#endif
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810tex.c b/xc/lib/GL/mesa/src/drv/i810/i810tex.c
index 039c0af59..8c88f2b2b 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810tex.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810tex.c
@@ -28,9 +28,10 @@
 #include <GL/gl.h>
 
 #include "mm.h"
-#include "i810lib.h"
+#include "i810context.h"
 #include "i810tex.h"
 #include "i810log.h"
+#include "i810ioctl.h"
 #include "simple_list.h"
 #include "enums.h"
 
@@ -204,6 +205,7 @@ static i810TextureObjectPtr i810CreateTexObj(i810ContextPtr imesa,
       log_pitch++;
    
    t->dirty_images = 0;
+   t->bound = 0;
    
    for ( height = i = 0 ; i < I810_TEX_MAXLEVELS && tObj->Image[i] ; i++ ) {
       t->image[i].image = tObj->Image[i];
@@ -278,24 +280,24 @@ static i810TextureObjectPtr i810CreateTexObj(i810ContextPtr imesa,
 
 void i810DestroyTexObj(i810ContextPtr imesa, i810TextureObjectPtr t)
 {
-   int i;
    if (!t) return;
 
    /* This is sad - need to sync *in case* we upload a texture
     * to this newly free memory...
     */
    if (t->MemBlock) {
-      imesa->dirty |= I810_REQUIRE_QUIESCENT;
       mmFreeMem(t->MemBlock);
-      t->MemBlock = 0;      
+      t->MemBlock = 0;
+
+      if (t->age > imesa->dirtyAge)
+	 imesa->dirtyAge = t->age;
    }
 
    if (t->globj)
       t->globj->DriverData = 0;
 
-   for ( i = 0 ; i < 2 ; i++ ) 
-      if ( imesa->CurrentTexObj[i] == t ) 
-	 imesa->CurrentTexObj[i] = 0;
+   if (t->bound)
+      imesa->CurrentTexObj[t->bound - 1] = 0; 
 
    remove_from_list(t);
    free(t);
@@ -305,9 +307,11 @@ void i810DestroyTexObj(i810ContextPtr imesa, i810TextureObjectPtr t)
 static void i810SwapOutTexObj(i810ContextPtr imesa, i810TextureObjectPtr t)
 {
    if (t->MemBlock) {
-      imesa->dirty |= I810_REQUIRE_QUIESCENT;
       mmFreeMem(t->MemBlock);
       t->MemBlock = 0;      
+
+      if (t->age > imesa->dirtyAge)
+	 imesa->dirtyAge = t->age;
    }
 
    t->dirty_images = ~0;
@@ -413,13 +417,28 @@ static void i810UploadTexLevel( i810TextureObjectPtr t, int level )
    }
    break;
 
+   case GL_ALPHA:
+   {
+      GLushort *dst = (GLushort *)(t->BufAddr + t->image[level].offset);
+      GLubyte  *src = (GLubyte *)image->Data;
+      i810Msg(10,"  CopyA: %p %p\n", dst, src);      
+
+      for (j = 0 ; j < image->Height ; j++, dst += (t->Pitch/2)) {
+	 for (i = 0 ; i < image->Width ; i++) {
+	    dst[i] = I810PACKCOLOR4444(255,255,255,src[0]);
+	    src += 1;
+	 }
+      }
+   }
+   break;
+
    /* TODO: Translate color indices *now*:
     */
    case GL_COLOR_INDEX:
       {
 	 GLubyte *dst = (GLubyte *)(t->BufAddr + t->image[level].offset);
 	 GLubyte *src = (GLubyte *)image->Data;
-	 i810Msg(10,"  CopyIndex  %p %p\n", dst, src);      
+	 i810Msg(10,"  CopyIndex: %p %p\n", dst, src);      
 
 	 for (j = 0 ; j < image->Height ; j++, dst += t->Pitch) {
 	    for (i = 0 ; i < image->Width ; i++) {
@@ -431,8 +450,8 @@ static void i810UploadTexLevel( i810TextureObjectPtr t, int level )
    break;
       
    default:
-      i810Error("Not supported texture format %d\n",(int)image->Format);
-      exit(1);
+      i810Error("Not supported texture format %s\n",
+		gl_lookup_enum_by_nr(image->Format));
    }
 }
 
@@ -449,13 +468,19 @@ void i810PrintLocalLRU( i810ContextPtr imesa )
 		 t->MemBlock->ofs / sz,
 		 t->MemBlock->ofs,
 		 t->MemBlock->size);      
+      else
+	 fprintf(stderr, "Texture (bound %d) at %x sz %x\n", 
+		 t->bound,
+		 t->MemBlock->ofs,
+		 t->MemBlock->size);      
+
    }
 }
 
 void i810PrintGlobalLRU( i810ContextPtr imesa )
 {
    int i, j;
-   i810TexRegion *list = imesa->sarea->texList;
+   drm_i810_tex_region_t *list = imesa->sarea->texList;
 
    for (i = 0, j = I810_NR_TEX_REGIONS ; i < I810_NR_TEX_REGIONS ; i++) {
       fprintf(stderr, "list[%d] age %d next %d prev %d\n",
@@ -471,7 +496,7 @@ void i810PrintGlobalLRU( i810ContextPtr imesa )
 
 void i810ResetGlobalLRU( i810ContextPtr imesa )
 {
-   i810TexRegion *list = imesa->sarea->texList;
+   drm_i810_tex_region_t *list = imesa->sarea->texList;
    int sz = 1 << imesa->i810Screen->logTextureGranularity;
    int i;
 
@@ -481,7 +506,7 @@ void i810ResetGlobalLRU( i810ContextPtr imesa )
     * when looking up objects at a particular location in texture
     * memory.  
     */
-   for (i = 0 ; (i+1) * sz < imesa->i810Screen->textureSize ; i++) {
+   for (i = 0 ; (i+1) * sz <= imesa->i810Screen->textureSize ; i++) {
       list[i].prev = i-1;
       list[i].next = i+1;
       list[i].age = 0;
@@ -503,7 +528,7 @@ static void i810UpdateTexLRU( i810ContextPtr imesa, i810TextureObjectPtr t )
    int logsz = imesa->i810Screen->logTextureGranularity;
    int start = t->MemBlock->ofs >> logsz;
    int end = (t->MemBlock->ofs + t->MemBlock->size - 1) >> logsz;
-   i810TexRegion *list = imesa->sarea->texList;
+   drm_i810_tex_region_t *list = imesa->sarea->texList;
    
    imesa->texAge = ++imesa->sarea->texAge;
 
@@ -556,7 +581,7 @@ void i810TexturesGone( i810ContextPtr imesa,
       /* It overlaps - kick it off.  Need to hold onto the currently bound
        * objects, however.
        */
-      if (t == imesa->CurrentTexObj[0] || t == imesa->CurrentTexObj[1]) 
+      if (t->bound)
 	 i810SwapOutTexObj( imesa, t );
       else
 	 i810DestroyTexObj( imesa, t );
@@ -583,7 +608,6 @@ int i810UploadTexImages( i810ContextPtr imesa, i810TextureObjectPtr t )
 {
    int i;
    int ofs;
-   i810glx.c_textureSwaps++;
 
    /* Do we need to eject LRU texture objects?
     */
@@ -594,18 +618,24 @@ int i810UploadTexImages( i810ContextPtr imesa, i810TextureObjectPtr t )
 	 if (t->MemBlock)
 	    break;
 
+	 if (imesa->TexObjList.prev->bound) {
+  	    fprintf(stderr, "Hit bound texture in upload\n"); 
+	    i810PrintLocalLRU( imesa );
+	    return -1;
+	 }
+
 	 if (imesa->TexObjList.prev == &(imesa->TexObjList)) {
-	    fprintf(stderr, "Failed to upload texture, sz %d\n", t->totalSize);
+ 	    fprintf(stderr, "Failed to upload texture, sz %d\n", t->totalSize);
 	    mmDumpMemInfo( imesa->texHeap );
 	    return -1;
 	 }
-
+	 
 	 i810DestroyTexObj( imesa, imesa->TexObjList.prev );
       }
  
       ofs = t->MemBlock->ofs;
       t->Setup[I810_TEXREG_MI3] = imesa->i810Screen->textureOffset + ofs;
-      t->BufAddr = i810glx.texVirtual + ofs;
+      t->BufAddr = imesa->i810Screen->tex.map + ofs;
       imesa->dirty |= I810_UPLOAD_CTX;
    }
 
@@ -613,6 +643,14 @@ int i810UploadTexImages( i810ContextPtr imesa, i810TextureObjectPtr t )
     */
    i810UpdateTexLRU( imesa, t );
 
+   if (I810_DEBUG & DEBUG_VERBOSE_LRU)
+      fprintf(stderr, "dispatch age: %d age freed memory: %d\n",
+	      GET_DISPATCH_AGE(imesa), imesa->dirtyAge);
+
+   if (imesa->dirtyAge >= GET_DISPATCH_AGE(imesa)) 
+      i810WaitAgeLocked( imesa, imesa->dirtyAge );
+   
+
    if (t->dirty_images) {
       if (I810_DEBUG & DEBUG_VERBOSE_LRU)
 	 fprintf(stderr, "*");
@@ -694,21 +732,37 @@ static void i810UpdateTex0State( GLcontext *ctx )
     
    if (t->dirty_images) 
       imesa->dirty |= I810_UPLOAD_TEX0IMAGE;
-
+   
    imesa->CurrentTexObj[0] = t;
+   t->bound = 1;
+
+   if (t->MemBlock)
+      i810UpdateTexLRU( imesa, t );
   
    switch (ctx->Texture.Unit[0].EnvMode) {
-      case GL_REPLACE:
-      imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
-					MC_STAGE_0 |
-					MC_UPDATE_DEST |
-					MC_DEST_CURRENT |
-					MC_UPDATE_ARG1 |
-					MC_ARG1_TEX0_COLOR | 
-					MC_UPDATE_ARG2 |
-					MC_ARG2_ONE |
-					MC_UPDATE_OP |
-					MC_OP_ARG1 );
+   case GL_REPLACE:
+      if (t->image[0].internalFormat == GL_ALPHA) 
+	 imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
+					   MC_STAGE_0 |
+					   MC_UPDATE_DEST |
+					   MC_DEST_CURRENT |
+					   MC_UPDATE_ARG1 |
+					   MC_ARG1_TEX0_COLOR | 
+					   MC_UPDATE_ARG2 |
+					   MC_ARG2_ITERATED_COLOR |
+					   MC_UPDATE_OP |
+					   MC_OP_ARG2 );
+      else 
+	 imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
+					   MC_STAGE_0 |
+					   MC_UPDATE_DEST |
+					   MC_DEST_CURRENT |
+					   MC_UPDATE_ARG1 |
+					   MC_ARG1_TEX0_COLOR | 
+					   MC_UPDATE_ARG2 |
+					   MC_ARG2_ONE |
+					   MC_UPDATE_OP |
+					   MC_OP_ARG1 );
 
       if (t->image[0].internalFormat == GL_RGB) {
 	 ma_modulate_op = MA_OP_ARG1;
@@ -812,16 +866,28 @@ static void i810UpdateTex0State( GLcontext *ctx )
 					MA_OP_ARG1 );
       break;
    case GL_BLEND:
-      imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
-					MC_STAGE_0 |
-					MC_UPDATE_DEST |
-					MC_DEST_CURRENT |
-					MC_UPDATE_ARG1 |
-					MC_ARG1_COLOR_FACTOR | 
-					MC_UPDATE_ARG2 |
-					MC_ARG2_ITERATED_COLOR |
-					MC_UPDATE_OP |
-					MC_OP_LIN_BLEND_TEX0_COLOR );
+      if (t->image[0].internalFormat == GL_ALPHA) 
+	 imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
+					   MC_STAGE_0 |
+					   MC_UPDATE_DEST |
+					   MC_DEST_CURRENT |
+					   MC_UPDATE_ARG1 |
+					   MC_ARG1_TEX0_COLOR | 
+					   MC_UPDATE_ARG2 |
+					   MC_ARG2_ITERATED_COLOR |
+					   MC_UPDATE_OP |
+					   MC_OP_ARG2 );
+      else 
+	 imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
+					   MC_STAGE_0 |
+					   MC_UPDATE_DEST |
+					   MC_DEST_CURRENT |
+					   MC_UPDATE_ARG1 |
+					   MC_ARG1_COLOR_FACTOR | 
+					   MC_UPDATE_ARG2 |
+					   MC_ARG2_ITERATED_COLOR |
+					   MC_UPDATE_OP |
+					   MC_OP_LIN_BLEND_TEX0_COLOR );
 
       if (t->image[0].internalFormat == GL_RGB) {
 	 imesa->Setup[I810_CTXREG_MA0] = ( GFX_OP_MAP_ALPHA_STAGES |
@@ -911,7 +977,10 @@ static void i810UpdateTex1State( GLcontext *ctx )
       imesa->dirty |= I810_UPLOAD_TEX1IMAGE;
 
    imesa->CurrentTexObj[1] = t;
+   t->bound = 2;
 
+   if (t->MemBlock)
+      i810UpdateTexLRU( imesa, t );
 
    switch (ctx->Texture.Unit[1].EnvMode) {
    case GL_REPLACE:
@@ -1078,11 +1147,15 @@ static void i810UpdateTex1State( GLcontext *ctx )
 void i810UpdateTextureState( GLcontext *ctx )
 {
    i810ContextPtr imesa = I810_CONTEXT(ctx);
+   if (imesa->CurrentTexObj[0]) imesa->CurrentTexObj[0]->bound = 0;
+   if (imesa->CurrentTexObj[1]) imesa->CurrentTexObj[1]->bound = 0;
    imesa->CurrentTexObj[0] = 0;
-   imesa->CurrentTexObj[1] = 0;
+   imesa->CurrentTexObj[1] = 0;   
    i810UpdateTex0State( ctx );
    i810UpdateTex1State( ctx );
-   I810_CONTEXT( ctx )->dirty |= I810_UPLOAD_CTX;
+   I810_CONTEXT( ctx )->dirty |= (I810_UPLOAD_CTX |
+				  I810_UPLOAD_TEX0 | 
+				  I810_UPLOAD_TEX1);
 }
 
 
@@ -1091,25 +1164,30 @@ void i810UpdateTextureState( GLcontext *ctx )
  * DRIVER functions
  *****************************************/
 
-static void i810TexEnv( GLcontext *ctx, GLenum pname, const GLfloat *param )
+static void i810TexEnv( GLcontext *ctx, GLenum target, 
+			GLenum pname, const GLfloat *param )
 {
    i810ContextPtr imesa = I810_CONTEXT( ctx );
 
    if (pname == GL_TEXTURE_ENV_MODE) {
 
+      FLUSH_BATCH(imesa);	
       imesa->new_state |= I810_NEW_TEXTURE;
 
    } else if (pname == GL_TEXTURE_ENV_COLOR) {
 
       struct gl_texture_unit *texUnit = 
 	 &ctx->Texture.Unit[ctx->Texture.CurrentUnit];
-      GLubyte c[4];
+      GLfloat *fc = texUnit->EnvColor;
       GLuint col;
 
-      FLOAT_RGBA_TO_UBYTE_RGBA(texUnit->EnvColor, c);
-      col = (c[3]<<24) | (c[0]<<16) | (c[1]<<8) | (c[2]);
+      col = ((((GLubyte)fc[3])<<24) | 
+	     (((GLubyte)fc[0])<<16) | 
+	     (((GLubyte)fc[1])<<8) | 
+	     (((GLubyte)fc[2])<<0));
     
       if (imesa->Setup[I810_CTXREG_CF1] != col) {
+	 FLUSH_BATCH(imesa);	
 	 imesa->Setup[I810_CTXREG_CF1] = col;      
 	 imesa->dirty |= I810_UPLOAD_CTX;
       }
@@ -1137,6 +1215,7 @@ static void i810TexImage( GLcontext *ctx,
 
    t = (i810TextureObjectPtr) tObj->DriverData;
    if (t) {
+      if (t->bound) FLUSH_BATCH(imesa);
       /* if this is the current object, it will force an update */
       i810DestroyTexObj( imesa, t );
       tObj->DriverData = 0;
@@ -1163,10 +1242,10 @@ static void i810TexSubImage( GLcontext *ctx, GLenum target,
    
    t = (i810TextureObjectPtr) tObj->DriverData;
    if (t) {
+      if (t->bound) FLUSH_BATCH( imesa );
       i810DestroyTexObj( imesa, t );
       tObj->DriverData = 0;
       imesa->new_state |= I810_NEW_TEXTURE;
-      i810glx.c_textureSwaps++;
    }
 }
 
@@ -1183,15 +1262,18 @@ static void i810TexParameter( GLcontext *ctx, GLenum target,
    switch (pname) {
    case GL_TEXTURE_MIN_FILTER:
    case GL_TEXTURE_MAG_FILTER:
+      if (t->bound) FLUSH_BATCH( imesa );
       i810SetTexFilter(t,tObj->MinFilter,tObj->MagFilter);
       break;
 
    case GL_TEXTURE_WRAP_S:
    case GL_TEXTURE_WRAP_T:
+      if (t->bound) FLUSH_BATCH( imesa );
       i810SetTexWrapping(t,tObj->WrapS,tObj->WrapT);
       break;
   
    case GL_TEXTURE_BORDER_COLOR:
+      if (t->bound) FLUSH_BATCH( imesa );
       i810SetTexBorderColor(t,tObj->BorderColor);
       break;
 
@@ -1206,7 +1288,14 @@ static void i810BindTexture( GLcontext *ctx, GLenum target,
 			     struct gl_texture_object *tObj )
 {
    i810ContextPtr imesa = I810_CONTEXT( ctx );
-   imesa->CurrentTexObj[ctx->Texture.CurrentUnit] = 0;  
+   
+   FLUSH_BATCH(imesa);
+
+   if (imesa->CurrentTexObj[ctx->Texture.CurrentUnit]) {
+      imesa->CurrentTexObj[ctx->Texture.CurrentUnit]->bound = 0;
+      imesa->CurrentTexObj[ctx->Texture.CurrentUnit] = 0;  
+   }
+
    imesa->new_state |= I810_NEW_TEXTURE;
 }
 
@@ -1216,6 +1305,13 @@ static void i810DeleteTexture( GLcontext *ctx, struct gl_texture_object *tObj )
    i810ContextPtr imesa = I810_CONTEXT( ctx );
 
    if (t) {
+
+      if (t->bound) {
+	 FLUSH_BATCH(imesa);
+	 imesa->CurrentTexObj[t->bound-1] = 0;
+	 imesa->new_state |= I810_NEW_TEXTURE;
+      }
+
       i810DestroyTexObj(imesa,t);
       tObj->DriverData=0;
    }
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810tex.h b/xc/lib/GL/mesa/src/drv/i810/i810tex.h
index d2b153ad8..a16733ec1 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810tex.h
+++ b/xc/lib/GL/mesa/src/drv/i810/i810tex.h
@@ -61,10 +61,11 @@ struct i810_texture_object_t {
    int Height;
    int texelBytes;
    int totalSize;
+   int bound;
 
    PMemBlock MemBlock;   
    char *BufAddr;
-
+   
    GLuint min_level;
    GLuint max_level;
    GLuint dirty_images;
@@ -87,8 +88,6 @@ struct i810_texture_object_t {
 #define I810_UPDATE_PALETTE    0x2
 #define I810_FALLBACK_PALETTE  0x4
 
-typedef struct i810_texture_object_t *i810TextureObjectPtr;
-
 void i810UpdateTextureState( GLcontext *ctx );
 void i810DDInitTextureFuncs( GLcontext *ctx );
 
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810tris.c b/xc/lib/GL/mesa/src/drv/i810/i810tris.c
index 7e24da002..c601ca492 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810tris.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810tris.c
@@ -31,7 +31,6 @@
 #include "pipeline.h"
 
 #include "mm.h"
-#include "i810lib.h"
 #include "i810tris.h"
 #include "i810vb.h"
 #include "i810log.h"
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810tris.h b/xc/lib/GL/mesa/src/drv/i810/i810tris.h
index 81cdb5159..3b3843683 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810tris.h
+++ b/xc/lib/GL/mesa/src/drv/i810/i810tris.h
@@ -27,37 +27,48 @@
 #define I810TRIS_INC
 
 #include "types.h"
-#include "i810dma.h"
+#include "i810ioctl.h"
 
 extern void i810PrintRenderState( const char *msg, GLuint state );
 extern void i810DDChooseRenderState(GLcontext *ctx);
 extern void i810DDTrifuncInit( void );
 
-extern GLuint *i810AllocPrimitiveVerts( i810ContextPtr imesa, int dwords );
 
-
-
-/* Todo: 
- *    - multidraw, ...
- *    - Antialiasing (?)
- *    - line and polygon stipple
- *    - select and feedback 
- *    - stencil 
- *    - point parameters
- *    - 
- */
-#define I810_ANTIALIAS_BIT   0       /* ignored for now, no fallback */
+/* shared */
 #define I810_FLAT_BIT 	     0x1
+
+/* triangle */
 #define I810_OFFSET_BIT	     0x2	
 #define I810_TWOSIDE_BIT     0x4
-#define I810_NODRAW_BIT	     0x8
-#define I810_FALLBACK_BIT    0x10
 
-/* Not in use:
- */
-#define I810_FEEDBACK_BIT    0x20
-#define I810_SELECT_BIT      0x40
-#define I810_POINT_PARAM_BIT 0x80	/* not needed? */
+/* line */
+#define I810_WIDE_LINE_BIT    0x2 
+#define I810_STIPPLE_LINE_BIT 0x4 
+
+/* shared */
+#define I810_FALLBACK_BIT    0x8 
+
+
+
+
+
+static i810_vertex __inline__ *i810AllocTriangles( i810ContextPtr imesa, int nr)
+{
+   GLuint *start = i810AllocDwords( imesa, 30*nr, PR_TRIANGLES );
+   return (i810_vertex *)start;
+}
+
+static i810_vertex __inline__ *i810AllocLine( i810ContextPtr imesa ) 
+{
+   GLuint *start = i810AllocDwords( imesa, 20, PR_LINES );
+   return (i810_vertex *)start;
+}
+
+static i810_vertex __inline__ *i810AllocRect( i810ContextPtr imesa ) 
+{
+   GLuint *start = i810AllocDwords( imesa, 30, PR_RECTS );
+   return (i810_vertex *)start;
+}
 
 
 
@@ -66,23 +77,17 @@ static void __inline__ i810_draw_triangle( i810ContextPtr imesa,
 					   i810_vertex *v1, 
 					   i810_vertex *v2 )
 {
-   i810_vertex *wv = (i810_vertex *)i810AllocPrimitiveVerts( imesa, 30 );
+   i810_vertex *wv = i810AllocTriangles( imesa, 1 );
    wv[0] = *v0;
    wv[1] = *v1;
    wv[2] = *v2;
-   FINISH_PRIM();
 }
 
 
-
-
-/* These can go soon, but for the meantime we're using triangles for 
- * everything.
- */
 static __inline__ void i810_draw_point( i810ContextPtr imesa,
 					i810_vertex *tmp, float sz )
 {
-   i810_vertex *wv = (i810_vertex *)i810AllocPrimitiveVerts( imesa, 6*10 );
+   i810_vertex *wv = i810AllocTriangles( imesa, 2 );
 
    wv[0] = *tmp;
    wv[0].x = tmp->x - sz;
@@ -108,16 +113,25 @@ static __inline__ void i810_draw_point( i810ContextPtr imesa,
    wv[5].x = tmp->x - sz;
    wv[5].y = tmp->y - sz;
 
-   FINISH_PRIM();
 }
 
 
-static __inline__ void i810_draw_line( i810ContextPtr imesa, 
-				       i810_vertex *tmp0, 
-				       i810_vertex *tmp1,
-				       float width )
+static __inline__ void i810_draw_line_line( i810ContextPtr imesa, 
+					    i810_vertex *tmp0, 
+					    i810_vertex *tmp1 )
 {
-   i810_vertex *wv = (i810_vertex *)i810AllocPrimitiveVerts( imesa, 6 * 10 );
+   i810_vertex *wv = i810AllocLine( imesa );
+   wv[0] = *tmp0;
+   wv[1] = *tmp1;
+}
+
+static __inline__ void i810_draw_tri_line( i810ContextPtr imesa, 
+					   i810_vertex *tmp0, 
+					   i810_vertex *tmp1,
+					   float width )
+{
+   i810_vertex *wv = i810AllocTriangles( imesa, 2 );
+      
    float dx, dy, ix, iy;
 
    dx = tmp0->x - tmp1->x;
@@ -151,8 +165,15 @@ static __inline__ void i810_draw_line( i810ContextPtr imesa,
    wv[5] = *tmp1;
    wv[5].x = tmp1->x + ix;
    wv[5].y = tmp1->y + iy;
+}
 
-   FINISH_PRIM();
+
+static __inline__ void i810_draw_line( i810ContextPtr imesa, 
+				       i810_vertex *tmp0, 
+				       i810_vertex *tmp1,
+				       float width )
+{
+   i810_draw_line_line( imesa, tmp0, tmp1 );
 }
 
 #endif
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810tritmp.h b/xc/lib/GL/mesa/src/drv/i810/i810tritmp.h
index 97c101c17..9d6b43d93 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810tritmp.h
+++ b/xc/lib/GL/mesa/src/drv/i810/i810tritmp.h
@@ -63,44 +63,37 @@ static __inline__ void TAG(triangle)( GLcontext *ctx, GLuint e0,
    }
 #endif  
 
-   i810glx.c_triangles++;
-
    
-   BEGIN_CLIP_LOOP(imesa)
-      {
-	 i810_vertex *wv = (i810_vertex *)i810AllocPrimitiveVerts( imesa, 30 );
-	 wv[0] = *v0;
+   {
+      i810_vertex *wv = i810AllocTriangles( imesa, 1 );
+      wv[0] = *v0;
 #if (IND & (I810_FLAT_BIT|I810_TWOSIDE_BIT))
-	 *((int *)(&wv[0].color)) = c0;
+      *((int *)(&wv[0].color)) = c0;
 #endif
 #if (IND & I810_OFFSET_BIT)
-	 wv[0].z = v0->z + offset;
+      wv[0].z = v0->z + offset;
 #endif
 
 
-	 wv[1] = *v1;
+      wv[1] = *v1;
 #if (IND & (I810_FLAT_BIT|I810_TWOSIDE_BIT))
-	 *((int *)(&wv[1].color)) = c1;
+      *((int *)(&wv[1].color)) = c1;
 #endif
 #if (IND & I810_OFFSET_BIT)
-	 wv[1].z = v1->z + offset;
+      wv[1].z = v1->z + offset;
 #endif
 
-	 wv[2] = *v2;
+      wv[2] = *v2;
 #if (IND & (I810_FLAT_BIT|I810_TWOSIDE_BIT))
-	 *((int *)(&wv[2].color)) = c2;
+      *((int *)(&wv[2].color)) = c2;
 #endif
 #if (IND & I810_OFFSET_BIT)
-	 wv[2].z = v2->z + offset;
+      wv[2].z = v2->z + offset;
 #endif
-
-	 FINISH_PRIM();
-      }
-   END_CLIP_LOOP(imesa);
+   }
 }
 
 
-
 static void TAG(quad)( GLcontext *ctx, GLuint v0,
 		       GLuint v1, GLuint v2, GLuint v3, 
 		       GLuint pv )
@@ -109,25 +102,35 @@ static void TAG(quad)( GLcontext *ctx, GLuint v0,
    TAG(triangle)( ctx, v1, v2, v3, pv );
 }
 
-
-#if ((IND & ~I810_FLAT_BIT) == 0)
-
 static void TAG(line)( GLcontext *ctx, GLuint v0, GLuint v1, GLuint pv )
 {
    i810ContextPtr imesa = I810_CONTEXT( ctx );
    i810VertexPtr i810VB = I810_DRIVER_DATA(ctx->VB)->verts;
-   i810_vertex tmp0 = i810VB[v0].v;
-   i810_vertex tmp1 = i810VB[v1].v;
-   float width = ctx->Line.Width;
+   int tmp0, tmp1;
+   (void) tmp0; (void) tmp1;
+
 
    if (IND & I810_FLAT_BIT) {
-      *(int *)&tmp1.color = *(int *)&tmp0.color = 
-	 *(int *)&i810VB[pv].v.color;
-   }
+      tmp0 = *(int *)&i810VB[v0].v.color;
+      tmp1 = *(int *)&i810VB[v1].v.color;
+      i810VB[v0].v.color = i810VB[pv].v.color;
+      i810VB[v1].v.color = i810VB[pv].v.color;
+   }      
 
-   BEGIN_CLIP_LOOP(imesa)
-      i810_draw_line( imesa, &tmp0, &tmp1, width );
-   END_CLIP_LOOP(imesa);
+   if (IND & I810_WIDE_LINE_BIT)
+   {
+      i810_draw_tri_line( imesa, &i810VB[v0].v, &i810VB[v1].v, 
+			  ctx->Line.Width );
+   } 
+   else 
+   {
+      i810_draw_line_line( imesa, &i810VB[v0].v, &i810VB[v1].v );
+   }      
+
+   if (IND & I810_FLAT_BIT) {
+      *(int *)&i810VB[v0].v.color = tmp0;
+      *(int *)&i810VB[v1].v.color = tmp1;
+   }      
 }
 
 
@@ -143,28 +146,22 @@ static void TAG(points)( GLcontext *ctx, GLuint first, GLuint last )
     * ctx->Driver.ReducedPrimitiveChange() callback.  
     */
    
-   BEGIN_CLIP_LOOP(imesa)
-      for(i=first;i<=last;i++) {
-	 if(VB->ClipMask[i]==0) {
-	    i810_vertex *tmp = &i810VB[i].v;
-	    i810_draw_point( imesa, tmp, sz );
-	 }
+   for(i=first;i<=last;i++) {
+      if(VB->ClipMask[i]==0) {
+	 i810_vertex *tmp = &i810VB[i].v;
+	 i810_draw_point( imesa, tmp, sz );
       }
-   END_CLIP_LOOP(imesa);
+   }
 }
 
-#endif
 
 
 static void TAG(init)( void )
 {
    tri_tab[IND] = TAG(triangle);
    quad_tab[IND] = TAG(quad);
-
-#if ((IND & ~I810_FLAT_BIT) == 0) 
    line_tab[IND] = TAG(line);
    points_tab[IND] = TAG(points);
-#endif
 }
 
 
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810vb.c b/xc/lib/GL/mesa/src/drv/i810/i810vb.c
index 6ba7eb9a5..6f8d163c4 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810vb.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810vb.c
@@ -23,11 +23,9 @@
  *
  */
  
-#include "i810lib.h"
 #include "i810context.h"
 #include "i810vb.h"
 #include "i810log.h"
-#include "i810dma.h"
 
 #include "stages.h"
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/Imakefile b/xc/lib/GL/mesa/src/drv/mga/Imakefile
index e4e39b747..cbcd09821 100644
--- a/xc/lib/GL/mesa/src/drv/mga/Imakefile
+++ b/xc/lib/GL/mesa/src/drv/mga/Imakefile
@@ -1,4 +1,6 @@
 
+#include <Threads.tmpl>
+
 #define DoNormalLib NormalLibGlx
 #define DoSharedLib SharedLibGlx
 #define DoExtraLib SharedLibGlx
@@ -9,38 +11,285 @@
 ALLOC_DEFINES = -DMALLOC_0_RETURNS_NULL
 #endif
 
+#if defined(LinuxArchitecture)
+OS_SUBDIR = linux
+#endif
+#if defined(FreeBSDArchitecture)
+OS_SUBDIR = bsd
+#endif
+
+#ifdef i386Architecture
+#ifdef MesaUse3DNow
+     ASM_DEFS = -DUSE_MMX_ASM -DUSE_X86_ASM -DUSE_3DNOW_ASM
+#else
+     ASM_DEFS = -DUSE_MMX_ASM -DUSE_X86_ASM
+#endif
+#endif
+
+
 #if BuildXF86DRI
   DRI_DEFINES = GlxDefines -DDRIVERTS
  DRI_INCLUDES = -I../../../../dri -I../../../../glx \
+		-I../../../dri \
 		-I$(TOP)/include -I$(TOP)/include/GL \
 		-I$(XF86OSSRC) -I$(SERVERSRC)/GL/dri \
 		-I$(XF86DRIVERSRC)/mga \
 		-I../../../include -I../.. -I../../X -I../common \
-	        -I$(XF86OSSRC)/linux/drm/kernel
+	        -I$(XF86OSSRC)/$(OS_SUBDIR)/drm/kernel
 #endif
 
 MESA_INCLUDES = -I. -I.. -I../../include
 
 
 
-      DEFINES = $(ALLOC_DEFINES) $(DRI_DEFINES)
-     INCLUDES = -I$(XLIBSRC) -I$(EXTINCSRC) $(MESA_INCLUDES) $(DRI_INCLUDES) \
-		-I/usr/include/glide
-      DRISRCS = mgaclear.c mgacnvtex.c mgadd.c mgadepth.c	\
-		mgafastpath.c					\
+      DEFINES = $(ALLOC_DEFINES) $(DRI_DEFINES) $(ASM_DEFS)
+     INCLUDES = -I$(XLIBSRC) -I$(EXTINCSRC) $(MESA_INCLUDES) $(DRI_INCLUDES) 
+
+#if 0
+	LOSRC = ../../../../lowpc.c
+	LOOBJ = ../../../../lowpc.o
+
+	HISRC = ../../../../highpc.c
+	HIOBJ = ../../../../highpc.o
+#endif
+
+      DRISRCS = ../../../dri/dri_mesa.c \
+		../../../../dri/dri_tmm.c
+
+      DRIOBJS = ../../../dri/dri_mesa.o \
+		../../../../dri/dri_tmm.o
+
+      DRMSRCS = ../../../../dri/drm/xf86drm.c \
+		../../../../dri/drm/xf86drmHash.c \
+		../../../../dri/drm/xf86drmRandom.c \
+		../../../../dri/drm/xf86drmSL.c
+
+      DRMOBJS = ../../../../dri/drm/xf86drm.o \
+		../../../../dri/drm/xf86drmHash.o \
+		../../../../dri/drm/xf86drmRandom.o \
+		../../../../dri/drm/xf86drmSL.o
+
+
+      MGASRCS = mgaclear.c mgacnvtex.c mgadd.c \
+		mgafastpath.c mgaeltpath.c \
 		mgapipeline.c	\
 		mgaspan.c mgastate.c mgatex.c		\
-		mgatris.c mgavb.c mgaioctl.c mga_xmesa.c
+		mgatris.c mgavb.c mgaioctl.c mga_xmesa.c mgabuffers.c
 
-      DRIOBJS = mgaclear.o mgacnvtex.o mgadd.o mgadepth.o	\
-		mgafastpath.o					\
+      MGAOBJS = mgaclear.o mgacnvtex.o mgadd.o \
+		mgafastpath.o mgaeltpath.o \
 		mgapipeline.o	\
 		mgaspan.o mgastate.o mgatex.o		\
-		mgatris.o mgavb.o mgaioctl.o mga_xmesa.o
+		mgatris.o mgavb.o mgaioctl.o mga_xmesa.o mgabuffers.o
+
+     MESASRCS = ../../aatriangle.c \
+		../../accum.c \
+		../../alpha.c \
+		../../alphabuf.c \
+		../../attrib.c \
+		../../bbox.c \
+		../../bitmap.c \
+		../../blend.c \
+		../../buffers.c \
+		../../clip.c \
+		../../colortab.c \
+		../../config.c \
+		../../context.c \
+		../../copypix.c \
+		../../cva.c \
+		../../debug_xform.c \
+		../../depth.c \
+		../../dlist.c \
+		../../drawpix.c \
+		../../enable.c \
+		../../enums.c \
+		../../eval.c \
+		../../extensions.c \
+		../../feedback.c \
+		../../fog.c \
+		../../get.c \
+		../../glapi.c \
+		../../glapinoop.c \
+		../../glthread.c \
+		../../hash.c \
+		../../image.c \
+		../../imaging.c \
+		../../light.c \
+		../../lines.c \
+		../../logic.c \
+		../../masking.c \
+		../../matrix.c \
+		../../mem.c \
+		../../mmath.c \
+		../../pb.c \
+		../../pipeline.c \
+		../../pixel.c \
+		../../pixeltex.c \
+		../../points.c \
+		../../polygon.c \
+		../../quads.c \
+		../../rastpos.c \
+		../../readpix.c \
+		../../rect.c \
+		../../scissor.c \
+		../../shade.c \
+		../../span.c \
+		../../stages.c \
+		../../state.c \
+		../../stencil.c \
+		../../teximage.c \
+		../../texobj.c \
+		../../texstate.c \
+		../../texture.c \
+		../../texutil.c \
+		../../translate.c \
+		../../triangle.c \
+		../../varray.c \
+		../../vb.c \
+		../../vbcull.c \
+		../../vbfill.c \
+		../../vbindirect.c \
+		../../vbrender.c \
+		../../vbxform.c \
+		../../vector.c \
+		../../vertices.c \
+		../../winpos.c \
+		../../xform.c \
+		../../zoom.c \
+		../../X86/common_x86.c
+
+     MESAOBJS = ../../aatriangle.o \
+		../../accum.o \
+		../../alpha.o \
+		../../alphabuf.o \
+		../../attrib.o \
+		../../bbox.o \
+		../../bitmap.o \
+		../../blend.o \
+		../../buffers.o \
+		../../clip.o \
+		../../colortab.o \
+		../../config.o \
+		../../context.o \
+		../../copypix.o \
+		../../cva.o \
+		../../debug_xform.o \
+		../../depth.o \
+		../../dlist.o \
+		../../drawpix.o \
+		../../enable.o \
+		../../enums.o \
+		../../eval.o \
+		../../extensions.o \
+		../../feedback.o \
+		../../fog.o \
+		../../get.o \
+		../../hash.o \
+		../../hint.o \
+		../../image.o \
+		../../imaging.o \
+		../../light.o \
+		../../lines.o \
+		../../logic.o \
+		../../masking.o \
+		../../matrix.o \
+		../../mem.o \
+		../../mmath.o \
+		../../pb.o \
+		../../pipeline.o \
+		../../pixel.o \
+		../../pixeltex.o \
+		../../points.o \
+		../../polygon.o \
+		../../quads.o \
+		../../rastpos.o \
+		../../readpix.o \
+		../../rect.o \
+		../../scissor.o \
+		../../shade.o \
+		../../span.o \
+		../../stages.o \
+		../../state.o \
+		../../stencil.o \
+		../../teximage.o \
+		../../texobj.o \
+		../../texstate.o \
+		../../texture.o \
+		../../texutil.o \
+		../../translate.o \
+		../../triangle.o \
+		../../varray.o \
+		../../vb.o \
+		../../vbcull.o \
+		../../vbfill.o \
+		../../vbindirect.o \
+		../../vbrender.o \
+		../../vbxform.o \
+		../../vector.o \
+		../../vertices.o \
+		../../winpos.o \
+		../../xform.o \
+		../../zoom.o
+
+#ifdef i386Architecture
+     X86_SRCS = ../../X86/x86.c \
+		../../X86/x86a.S \
+		../../X86/common_x86.c \
+		../../X86/common_x86asm.S \
+		../../X86/vertex.S
+
+     X86_OBJS = ../../X86/x86.o \
+		../../X86/x86a.o \
+		../../X86/common_x86.o \
+		../../X86/common_x86asm.o \
+		../../X86/vertex.o
+
+     MMX_SRCS = ../../X86/mmx_blend.S
+
+     MMX_OBJS = ../../X86/mmx_blend.o
+
+#ifdef MesaUse3DNow
+   3DNOW_SRCS = ../../X86/3dnow.c \
+		../../X86/3dnow_norm_raw.S \
+		../../X86/3dnow_xform_masked1.S \
+		../../X86/3dnow_xform_masked2.S \
+		../../X86/3dnow_xform_masked3.S \
+		../../X86/3dnow_xform_masked4.S \
+		../../X86/3dnow_xform_raw1.S \
+		../../X86/3dnow_xform_raw2.S \
+		../../X86/3dnow_xform_raw3.S \
+		../../X86/3dnow_xform_raw4.S \
+		../../X86/vertex_3dnow.S
 
+   3DNOW_OBJS = ../../X86/3dnow.o \
+		../../X86/3dnow_norm_raw.o \
+		../../X86/3dnow_xform_masked1.o \
+		../../X86/3dnow_xform_masked2.o \
+		../../X86/3dnow_xform_masked3.o \
+		../../X86/3dnow_xform_masked4.o \
+		../../X86/3dnow_xform_raw1.o \
+		../../X86/3dnow_xform_raw2.o \
+		../../X86/3dnow_xform_raw3.o \
+		../../X86/3dnow_xform_raw4.o \
+		../../X86/vertex_3dnow.o
+
+#endif
+#endif
+
+      ASMSRCS = $(X86_SRCS) $(MMX_SRCS) $(3DNOW_SRCS)
+      ASMOBJS = $(X86_OBJS) $(MMX_OBJS) $(3DNOW_OBJS)
+
+      COMMONSRCS = ../common/mm.c ../common/hwlog.c
+      COMMONOBJS = ../common/mm.o ../common/hwlog.o
+
+         SRCS = $(LOWSRC) $(DRISRCS) $(DRMSRCS) $(MESASRCS) $(ASMSRCS) $(COMMONSRCS) $(MGASRCS) $(HISRC)
+         OBJS = $(LOWOBJ) $(DRIOBJS) $(DRMOBJS) $(MESAOBJS) $(ASMOBJS) $(COMMONOBJS) $(MGAOBJS) $(HIOBJ)
+
+REQUIREDLIBS += -lm
+#if !GlxBuiltInMga
+REQUIREDLIBS += -L../../../.. -lGL
+#endif
 
-	 SRCS = $(DRISRCS) 
-	 OBJS = $(DRIOBJS) 
 
 #if !GlxUseBuiltInDRIDriver
 #undef DoNormalLib NormalLibGlx
@@ -61,7 +310,7 @@ LIBNAME = mga_dri.so
 ALL_OBJS = $(OBJS)
 ALL_DEPS = DONE
 SharedDepModuleTarget($(LIBNAME),$(ALL_DEPS),$(ALL_OBJS))
-InstallDynamicModule($(LIBNAME),$(MODULEDIR),.)
+InstallDynamicModule($(LIBNAME),$(MODULEDIR)/dri,.)
 #endif
 
 DependTarget()
diff --git a/xc/lib/GL/mesa/src/drv/mga/mga_xmesa.c b/xc/lib/GL/mesa/src/drv/mga/mga_xmesa.c
index b7fc70b79..a0d9cdf73 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mga_xmesa.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mga_xmesa.c
@@ -78,6 +78,21 @@ static mgaContextPtr      mgaCtx = 0;
 
 mgaGlx_t	mgaglx;
 
+/* These functions are accessed externally to the driver:
+ *
+ * XMesaInitDriver
+ * XMesaResetDriver
+ * XMesaCreateVisual
+ * XMesaDestroyVisual
+ * XMesaCreateContext 
+ * XMesaDestroyContext
+ * XMesaCreateWindowBuffer
+ * XMesaCreatePixmapBuffer
+ * XMesaDestroyBuffer
+ * XMesaSwapBuffers
+ * XMesaMakeCurrent
+ *
+ */
 
 GLboolean XMesaInitDriver(__DRIscreenPrivate *sPriv)
 {
@@ -150,23 +165,21 @@ GLboolean XMesaInitDriver(__DRIscreenPrivate *sPriv)
    mgaScreen->depthPitch  =  serverInfo->depthPitch;
 
 
-   mgaScreen->agp_tex.handle = serverInfo->agp;
-   mgaScreen->agp_tex.size = serverInfo->agpSize;
+   mgaScreen->agp.handle = serverInfo->agp;
+   mgaScreen->agp.size = serverInfo->agpSize;
 
    if (drmMap(sPriv->fd, 
-	      mgaScreen->agp_tex.handle, 
-	      mgaScreen->agp_tex.size, 
-	      (drmAddress *)&mgaScreen->agp_tex.map) != 0) 
+	      mgaScreen->agp.handle, 
+	      mgaScreen->agp.size, 
+	      (drmAddress *)&mgaScreen->agp.map) != 0) 
    {
       Xfree(mgaScreen);
       sPriv->private = NULL;
       return GL_FALSE;
    }
 
-
-
    mgaScreen->textureOffset[MGA_CARD_HEAP] = serverInfo->textureOffset;
-   mgaScreen->textureOffset[MGA_AGP_HEAP] = (mgaScreen->agp_tex.handle |
+   mgaScreen->textureOffset[MGA_AGP_HEAP] = (serverInfo->agpTextureOffset |
 					     PDEA_pagpxfer_enable | 1);
 
    /*
@@ -184,12 +197,17 @@ GLboolean XMesaInitDriver(__DRIscreenPrivate *sPriv)
       serverInfo->logAgpTextureGranularity;
 
    mgaScreen->texVirtual[MGA_CARD_HEAP] = (mgaScreen->sPriv->pFB + 
-					   mgaScreen->textureOffset[0]);
-   mgaScreen->texVirtual[MGA_AGP_HEAP] = mgaScreen->agp_tex.map;
+					   serverInfo->textureOffset);
+   mgaScreen->texVirtual[MGA_AGP_HEAP] = (mgaScreen->agp.map + 
+					  serverInfo->agpTextureOffset);
 
    mgaScreen->mAccess = serverInfo->mAccess;
 
 
+   /* For calculating setupdma addresses.
+    */
+   mgaScreen->dmaOffset = serverInfo->agpBufferOffset;   
+   
    /*
    fprintf(stderr, "\n\n\nbackOffset: %x pitch %x\n", 
 	   mgaScreen->backOffset,
@@ -199,7 +217,7 @@ GLboolean XMesaInitDriver(__DRIscreenPrivate *sPriv)
    mgaScreen->Attrib = MGA_PF_565;
    mgaScreen->bufs = drmMapBufs(sPriv->fd);
    if (!mgaScreen->bufs) {
-      drmUnmap(mgaScreen->agp_tex.map, mgaScreen->agp_tex.size);
+      /*drmUnmap(mgaScreen->agp_tex.map, mgaScreen->agp_tex.size);*/
       XFree(mgaScreen);
       sPriv->private = NULL;
       return GL_FALSE;
@@ -210,6 +228,7 @@ GLboolean XMesaInitDriver(__DRIscreenPrivate *sPriv)
    memset(&mgaglx, 0, sizeof(mgaglx));
 
    mgaDDFastPathInit();
+   mgaDDEltPathInit();
    mgaDDTrifuncInit();
    mgaDDSetupInit();
 
@@ -222,7 +241,7 @@ GLboolean XMesaInitDriver(__DRIscreenPrivate *sPriv)
 void XMesaResetDriver(__DRIscreenPrivate *sPriv)
 {
    mgaScreenPrivate *mgaScreen = (mgaScreenPrivate *) sPriv->private;
-   drmUnmap(mgaScreen->agp_tex.map, mgaScreen->agp_tex.size);
+   /*drmUnmap(mgaScreen->agp_tex.map, mgaScreen->agp_tex.size);*/
    Xfree(mgaScreen);
    sPriv->private = NULL;
 }
@@ -481,4 +500,5 @@ void mgaGetLock( mgaContextPtr mmesa, GLuint flags )
 }
 
 
+
 #endif
diff --git a/xc/lib/GL/mesa/src/drv/mga/mga_xmesa.h b/xc/lib/GL/mesa/src/drv/mga/mga_xmesa.h
index d563bd429..489b71836 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mga_xmesa.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mga_xmesa.h
@@ -28,7 +28,6 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 /*
  * Authors:
  *   Keith Whitwell <keithw@precisioninsight.com>
- *   Daryll Strauss <daryll@precisioninsight.com> (Origninal tdfx driver).
  *
  */
 
@@ -42,7 +41,13 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "dri_mesaint.h"
 #include "dri_mesa.h"
 #include "types.h"
-#include "xmesaP.h"
+#include "mgaregs.h"
+
+typedef struct {
+   drmHandle handle;
+   drmSize size;
+   char *map;
+} mgaRegion, *mgaRegionPtr;
 
 typedef struct {
 
@@ -54,53 +59,69 @@ typedef struct {
    int cpp;			/* for front and back buffers */
 
    int Attrib;
+   unsigned int mAccess;
 
-   int frontOffset;
-   int frontPitch;
-   int backOffset;
-   int backPitch;
+   unsigned int frontOffset;
+   unsigned int frontPitch;
+   unsigned int backOffset;
+   unsigned int backPitch;
 
-   int depthOffset;
-   int depthPitch;
+   unsigned int depthOffset;
+   unsigned int depthPitch;
    int depthCpp;
 
-   int textureOffset;
-   int textureSize;
-   int logTextureGranularity;
+   unsigned int dmaOffset;		
+
+   unsigned int textureOffset[MGA_NR_TEX_HEAPS];
+   unsigned int textureSize[MGA_NR_TEX_HEAPS];
+   int logTextureGranularity[MGA_NR_TEX_HEAPS];
+   char *texVirtual[MGA_NR_TEX_HEAPS];
+
 
    __DRIscreenPrivate *sPriv;
    drmBufMapPtr  bufs;
 
+   /* Maps the dma buffers as well as textures ? 
+    */
+   mgaRegion agp;
+
 } mgaScreenPrivate;
 
 
 #include "mgalib.h"
 
-extern void mgaXMesaUpdateState( mgaContextPtr mmesa );
+extern void mgaGetLock( mgaContextPtr mmesa, GLuint flags );
 extern void mgaEmitHwStateLocked( mgaContextPtr mmesa );
 extern void mgaEmitScissorValues( mgaContextPtr mmesa, int box_nr, int emit );
-extern void mgaXMesaSetBackClipRects( mgaContextPtr mmesa );
-extern void mgaXMesaSetFrontClipRects( mgaContextPtr mmesa );
+
+#define GET_DISPATCH_AGE( mmesa ) mmesa->sarea->last_dispatch
+#define GET_ENQUEUE_AGE( mmesa ) mmesa->sarea->last_enqueue
 
 
 
 /* Lock the hardware and validate our state.  
  */
-#define LOCK_HARDWARE( mmesa )				\
-  do {							\
-    char __ret=0;					\
-    DRM_CAS(mmesa->driHwLock, mmesa->hHWContext,	\
-	    (DRM_LOCK_HELD|mmesa->hHWContext), __ret);	\
-    if (__ret) {					\
-        drmGetLock(mmesa->driFd, mmesa->hHWContext, 0);	\
-        mgaXMesaUpdateState( mmesa );			\
-    }							\
+#define LOCK_HARDWARE( mmesa )					\
+  do {								\
+    char __ret=0;						\
+    DRM_CAS(mmesa->driHwLock, mmesa->hHWContext,		\
+	    (DRM_LOCK_HELD|mmesa->hHWContext), __ret);		\
+    if (__ret)							\
+        mgaGetLock( mmesa, 0 );					\
   } while (0)
 
 
+/* 
+ */
+#define LOCK_HARDWARE_QUIESCENT( mmesa ) do {	                        \
+	LOCK_HARDWARE( mmesa );			                        \
+	mgaUpdateLock( mmesa, DRM_LOCK_QUIESCENT | DRM_LOCK_FLUSH );	\
+} while (0)
+
+
 /* Unlock the hardware using the global current context 
  */
-#define UNLOCK_HARDWARE(mmesa)					\
+#define UNLOCK_HARDWARE(mmesa) 				\
     DRM_UNLOCK(mmesa->driFd, mmesa->driHwLock, mmesa->hHWContext);	
 
 
@@ -109,6 +130,8 @@ extern void mgaXMesaSetFrontClipRects( mgaContextPtr mmesa );
 #define REFRESH_DRAWABLE_INFO( mmesa )		\
 do {						\
    LOCK_HARDWARE( mmesa );			\
+   mmesa->lastX = mmesa->drawX; 		\
+   mmesa->lastY = mmesa->drawY; 		\
    UNLOCK_HARDWARE( mmesa );			\
 } while (0)
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgaeltpath.c b/xc/lib/GL/mesa/src/drv/mga/mgaeltpath.c
new file mode 100644
index 000000000..4751eb091
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/mga/mgaeltpath.c
@@ -0,0 +1,457 @@
+/*
+ * GLX Hardware Device Driver for Matrox G400
+ * Copyright (C) 1999 Keith Whitwell
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * KEITH WHITWELL, OR ANY OTHER CONTRIBUTORS BE LIABLE FOR ANY CLAIM, 
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE 
+ * OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <stdio.h>
+
+#include "types.h"
+#include "enums.h"
+#include "cva.h"
+#include "vertices.h"
+#include "mmath.h"
+#include "xform.h"
+
+#include "mgalib.h"
+#include "mgapipeline.h"
+#include "mgatris.h"
+#include "mgastate.h"
+#include "mgavb.h"
+
+
+/* Always use a full-sized stride for vertices. [FIXME]
+ * Stride in the buffers must be a quadword multiple.
+ */
+#define BUFFER_STRIDE 12 
+#define CLIP_STRIDE 10 
+
+
+static void fire_elts( mgaContextPtr mmesa )
+{
+   LOCK_HARDWARE( mmesa );
+
+   /* Fire queued elements and discard that buffer if its contents
+    * won't be referenced by future elements.
+    */
+   if (mmesa->elt_buf) {
+      GLuint retain = (mmesa->elt_buf == mmesa->retained_buf);
+
+      if (mmesa->first_elt != mmesa->next_elt) {
+	 mgaFireEltsLocked( mmesa, 
+			    ((GLuint)mmesa->first_elt - 
+			     (GLuint)mmesa->elt_buf->address),
+			    ((GLuint)mmesa->next_elt - 
+			     (GLuint)mmesa->elt_buf->address),
+			    !retain );
+      } else if (!retain) 
+	 mgaReleaseBufLocked( mmesa, mmesa->elt_buf );
+      
+      mmesa->elt_buf = 0;
+   } 
+   else if (mmesa->vertex_dma_buffer) 
+   {
+      mgaFlushVerticesLocked( mmesa );
+   }
+
+   mgaGetEltBufLocked( mmesa );
+      
+   UNLOCK_HARDWARE( mmesa );
+
+   mmesa->next_vert = (GLfloat *)((GLuint)mmesa->elt_buf->address + 
+				  mmesa->elt_buf->total - 
+				  BUFFER_STRIDE * sizeof(GLfloat));
+
+   mmesa->next_vert_phys = (mmesa->mgaScreen->dmaOffset + 
+			    mmesa->elt_buf->idx * MGA_DMA_BUF_SZ +
+			    mmesa->elt_buf->total -
+			    BUFFER_STRIDE * sizeof(GLfloat));
+
+   mmesa->first_elt = (GLuint *)mmesa->elt_buf->address;
+   mmesa->next_elt = (GLuint *)mmesa->elt_buf->address;
+
+}
+
+
+static void release_bufs( mgaContextPtr mmesa )
+{
+   if (mmesa->retained_buf && mmesa->retained_buf != mmesa->elt_buf) 
+   {
+      LOCK_HARDWARE( mmesa );
+      if (mmesa->first_elt != mmesa->next_elt) {
+	 mgaFireEltsLocked( mmesa, 
+			    ((GLuint)mmesa->first_elt - 
+			     (GLuint)mmesa->elt_buf->address),
+			    ((GLuint)mmesa->next_elt - 
+			     (GLuint)mmesa->elt_buf->address),
+			    0 );
+
+	 mmesa->first_elt = mmesa->next_elt;
+      }
+
+      mgaReleaseBufLocked( mmesa, mmesa->retained_buf );
+      UNLOCK_HARDWARE( mmesa );
+   }
+
+   mmesa->retained_buf = 0;
+}
+
+
+
+
+#define NEGATIVE(f)          (f < 0)
+#define DIFFERENT_SIGNS(a,b) ((a*b) < 0)
+#define LINTERP( T, A, B )   ( (A) + (T) * ( (B) - (A) ) )
+
+
+#define INTERP_RGBA(t, out, a, b) {			\
+   int i;						\
+   for (i = 0; i < 4; i++) {				\
+      GLfloat fa = UBYTE_COLOR_TO_FLOAT_COLOR(a[i]);	\
+      GLfloat fb = UBYTE_COLOR_TO_FLOAT_COLOR(b[i]);	\
+      GLfloat fo = LINTERP(t, fa, fb);			\
+      FLOAT_COLOR_TO_UBYTE_COLOR(out[i], fo);		\
+   }							\
+}
+
+
+#define CLIP(SGN,V,PLANE)			\
+if (mask & PLANE) {				\
+   GLuint *indata = inlist[in];			\
+   GLuint *outdata = inlist[in ^= 1];		\
+   GLuint nr = n;				\
+   GLfloat *J = verts[indata[nr-1]];		\
+   GLfloat dpJ = (SGN J[V]) + J[3];		\
+						\
+   for (i = n = 0 ; i < nr ; i++) {		\
+      GLuint elt_i = indata[i];			\
+      GLfloat *I = verts[elt_i];		\
+      GLfloat dpI = (SGN I[V]) + I[3];		\
+						\
+      if (DIFFERENT_SIGNS(dpI, dpJ)) {		\
+	 GLfloat *O = verts[next_vert];	\
+	 outdata[n++] = next_vert++;		\
+						\
+	 if (NEGATIVE(dpI)) {			\
+	    GLfloat t = dpI / (dpI - dpJ);	\
+	    interp(t, O, I, J);			\
+	 }					\
+         else					\
+	 {					\
+	    GLfloat t = dpJ / (dpJ - dpI);	\
+	    interp(t, O, J, I);			\
+	 }					\
+      }						\
+						\
+      if (!NEGATIVE(dpI))			\
+	 outdata[n++] = elt_i;			\
+						\
+      J = I;					\
+      dpJ = dpI;				\
+   }						\
+						\
+   if (n < 3) return;				\
+}
+
+
+static void mga_tri_clip( mgaContextPtr mmesa,
+			  struct vertex_buffer *VB,
+			  GLuint *elt,
+			  GLubyte mask )
+{
+   struct mga_elt_tab *tab = mmesa->elt_tab;
+   mga_interp_func interp = tab->interp;
+   GLuint inlist[2][VB_MAX_CLIPPED_VERTS];
+   GLuint in = 0;
+   GLuint n = 3, next_vert = 3;
+   GLuint i;
+   GLfloat verts[VB_MAX_CLIPPED_VERTS][CLIP_STRIDE];
+
+   /* Build temporary vertices in clipspace.  This is the potential
+    * downside to this path.
+    */
+   tab->build_tri_verts( mmesa, VB, (GLfloat *)verts, elt );
+
+   inlist[0][0] = 0;
+   inlist[0][1] = 1;
+   inlist[0][2] = 2;
+
+   CLIP(-,0,CLIP_RIGHT_BIT);
+   CLIP(+,0,CLIP_LEFT_BIT);
+   CLIP(-,1,CLIP_TOP_BIT);
+   CLIP(+,1,CLIP_BOTTOM_BIT);
+   CLIP(-,2,CLIP_FAR_BIT);
+   CLIP(+,2,CLIP_NEAR_BIT); 
+
+
+   {
+      GLuint *out = inlist[in];
+      GLuint space = (GLuint)mmesa->next_vert -	(GLuint)mmesa->next_elt;
+
+      if (space < n * (BUFFER_STRIDE + 3) * sizeof(GLuint))
+         fire_elts(mmesa); 
+
+      /* Project the new vertices and emit to dma buffers.  Translate
+       * out values to physical addresses for setup dma.  
+       */
+      tab->project_and_emit_verts( mmesa, (GLfloat *)verts, out, n );
+      
+      /* Convert the planar polygon to a list of triangles and emit to
+       * elt buffers. 
+       */
+      for (i = 2 ; i < n ; i++) {
+	 mmesa->next_elt[0] = out[0];
+	 mmesa->next_elt[1] = out[i-1];
+	 mmesa->next_elt[2] = out[i];
+	 mmesa->next_elt += 3;
+      }
+   }
+}
+
+
+
+
+/* Build a table of functions to clip each primitive type.  These
+ * produce a list of elements in the appropriate 'reduced' primitive,
+ * ie (points, lines, triangles) containing all the clipped and
+ * unclipped primitives from the original list.
+ */
+
+#define INIT(x)
+
+#define TRI_THRESHOLD         (3 * sizeof(GLuint))
+
+#define UNCLIPPED_VERT(x) (mmesa->first_vert_phys - x * BUFFER_STRIDE * 4)
+
+#define TRIANGLE( e2, e1, e0 )			\
+do {						\
+   if ((GLuint)mmesa->next_vert - 		\
+       (GLuint)mmesa->next_elt < TRI_THRESHOLD)	\
+      fire_elts(mmesa);			\
+   mmesa->next_elt[0] = UNCLIPPED_VERT(e2);	\
+   mmesa->next_elt[1] = UNCLIPPED_VERT(e1);	\
+   mmesa->next_elt[2] = UNCLIPPED_VERT(e0);	\
+   mmesa->next_elt+=3;				\
+} while (0)
+
+#define CLIP_TRIANGLE( e2, e1, e0 )				\
+do {								\
+   GLubyte ormask = mask[e2] | mask[e1] | mask[e0];		\
+   if (ormask == 0) {						\
+      TRIANGLE( e2, e1, e0 );					\
+   } else if ((mask[e2] & mask[e1] & mask[e0]) == 0) {		\
+      out[0] = e2;						\
+      out[1] = e1;						\
+      out[2] = e0;						\
+      mga_tri_clip( mmesa, VB, out, ormask );			\
+   }								\
+} while (0)
+
+#define LOCAL_VARS						\
+   mgaContextPtr mmesa = MGA_CONTEXT( VB->ctx );		\
+   GLuint *elt = VB->EltPtr->data;				\
+   GLuint out[VB_MAX_CLIPPED_VERTS];				\
+   GLubyte *mask = VB->ClipMask; \
+   (void) mask; (void) out; (void) elt; (void) mmesa;
+
+
+
+#define RENDER_POINTS(start, count) 
+#define RENDER_LINE(i1, i0) 
+#define RENDER_TRI(i2, i1, i0, pv, parity)		\
+do {							\
+   GLuint e2 = elt[i2], e1 = elt[i1], e0 = elt[i0];	\
+   if (parity) e2 = elt[i1], e1 = elt[i2];		\
+   CLIP_TRIANGLE( e2, e1, e0 );				\
+} while (0)
+
+#define RENDER_QUAD(i3, i2, i1, i0, pv )		\
+  CLIP_TRIANGLE(elt[i3], elt[i2], elt[i0]);     	\
+  CLIP_TRIANGLE(elt[i2], elt[i1], elt[i0])
+
+#define TAG(x) mga_##x##_elt
+#include "render_tmp.h"
+
+
+
+#define LOCAL_VARS						\
+   mgaContextPtr mmesa = MGA_CONTEXT( VB->ctx );		\
+   GLuint *elt = VB->EltPtr->data;				\
+   (void) elt; (void) mmesa;
+
+#define RENDER_POINTS(start, count) 
+#define RENDER_LINE(i1, i0) 
+#define RENDER_TRI(i2, i1, i0, pv, parity)		\
+do {							\
+   GLuint e2 = elt[i2], e1 = elt[i1], e0 = elt[i0];	\
+   if (parity) e2 = elt[i1], e1 = elt[i2];		\
+   TRIANGLE( e2, e1, e0 );				\
+} while (0)
+
+#define RENDER_QUAD(i3, i2, i1, i0, pv )	\
+  TRIANGLE(elt[i3], elt[i2], elt[i0]);		\
+  TRIANGLE(elt[i2], elt[i1], elt[i0])
+
+#define TAG(x) mga_##x##_elt_unclipped
+#include "render_tmp.h"
+
+
+
+
+static void refresh_projection_matrix( GLcontext *ctx )
+{
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
+   GLfloat *m = mmesa->device_matrix;
+   GLmatrix *mat = &ctx->Viewport.WindowMap;
+
+   REFRESH_DRAWABLE_INFO(mmesa);
+
+   m[MAT_SX] =   mat->m[MAT_SX];
+   m[MAT_TX] =   mat->m[MAT_TX] + mmesa->drawX + .5;
+   m[MAT_SY] = (- mat->m[MAT_SY]);
+   m[MAT_TY] = (- mat->m[MAT_TY]) + mmesa->driDrawable->h + mmesa->drawY - .5;
+   m[MAT_SZ] =   mat->m[MAT_SZ] * (1.0 / 0x10000);
+   m[MAT_TZ] =   mat->m[MAT_TZ] * (1.0 / 0x10000);
+}
+
+#define CLIP_UBYTE_B 0   
+#define CLIP_UBYTE_G 1  
+#define CLIP_UBYTE_R 2
+#define CLIP_UBYTE_A 3   
+
+
+#define TYPE (0)
+#define TAG(x) x
+#include "mgaelttmp.h"
+
+#define TYPE (MGA_RGBA_BIT)
+#define TAG(x) x##_RGBA
+#include "mgaelttmp.h"
+
+#define TYPE (MGA_TEX0_BIT)
+#define TAG(x) x##_TEX0
+#include "mgaelttmp.h"
+
+#define TYPE (MGA_RGBA_BIT|MGA_TEX0_BIT)
+#define TAG(x) x##_RGBA_TEX0
+#include "mgaelttmp.h"
+
+#define TYPE (MGA_RGBA_BIT|MGA_TEX0_BIT|MGA_TEX1_BIT)
+#define TAG(x) x##_RGBA_TEX0_TEX1
+#include "mgaelttmp.h"
+
+#define TYPE (MGA_TEX0_BIT|MGA_TEX1_BIT)
+#define TAG(x) x##_TEX0_TEX1
+#include "mgaelttmp.h"
+
+
+/* Very sparsely popluated array - fix the indices.
+ */
+static struct mga_elt_tab mgaEltTab[0x80];
+
+void mgaDDEltPathInit( void )
+{
+   mga_render_init_elt();
+   mga_render_init_elt_unclipped();
+
+   mga_init_eltpath( &mgaEltTab[0] );
+   mga_init_eltpath_RGBA( &mgaEltTab[MGA_RGBA_BIT] );
+   mga_init_eltpath_TEX0( &mgaEltTab[MGA_TEX0_BIT] );
+   mga_init_eltpath_RGBA_TEX0( &mgaEltTab[MGA_RGBA_BIT|MGA_TEX0_BIT] );
+   mga_init_eltpath_TEX0_TEX1( &mgaEltTab[MGA_TEX0_BIT|MGA_TEX1_BIT] );
+   mga_init_eltpath_RGBA_TEX0_TEX1( &mgaEltTab[MGA_RGBA_BIT|MGA_TEX0_BIT|
+						  MGA_TEX1_BIT] );
+}
+
+#define VALID_SETUP (MGA_RGBA_BIT|MGA_TEX0_BIT|MGA_TEX1_BIT)
+
+
+
+/* Use a temporary array for device coordinates, so that we can easily
+ * tap into existing mesa assembly.  Otherwise consider emitting
+ * device coordinates to dma buffers directly from the project/cliptest
+ * routine.  (requires output stride, potential loss of writecombining 
+ * efficiency?)  
+ *
+ * This path is a lot closer to the standard vertex path in the
+ * initial stages than the original fastpath.  A slightly more optimal
+ * path could be constructed, but would require us to write new
+ * assembly.  
+ */
+void mgaDDEltPath( struct vertex_buffer *VB )
+{
+   GLcontext *ctx = VB->ctx;
+   GLenum prim = ctx->CVA.elt_mode;
+   mgaContextPtr mmesa = MGA_CONTEXT( ctx ); 
+   struct mga_elt_tab *tab = &mgaEltTab[mmesa->setupindex & VALID_SETUP];
+
+   VB->ClipPtr = TransformRaw(&VB->Clip, &ctx->ModelProjectMatrix, VB->ObjPtr );
+
+   refresh_projection_matrix( ctx );
+
+   VB->ClipAndMask = ~0; 
+   VB->ClipOrMask = 0;
+   VB->Projected = gl_clip_tab[VB->ClipPtr->size]( VB->ClipPtr,
+						   &VB->Win,
+						   VB->ClipMask,
+						   &VB->ClipOrMask,
+						   &VB->ClipAndMask );
+
+   if (VB->ClipAndMask) 
+      return;
+
+   if (mmesa->vertex_dma_buffer) 
+      mgaFlushVertices( mmesa );
+
+   if (mmesa->new_state)
+      mgaDDUpdateHwState( ctx );
+
+   /* Allocate a single buffer to hold unclipped vertices.  All
+    * unclipped vertices must be contiguous.  
+    */
+   if ((GLuint)mmesa->next_vert - (GLuint)mmesa->next_elt <  
+       VB->Count * BUFFER_STRIDE * sizeof(GLuint))	
+      fire_elts( mmesa );
+
+   mmesa->retained_buf = mmesa->elt_buf;
+   
+   /* Emit unclipped vertices to the buffer.
+    */
+   tab->emit_unclipped_verts( VB ); 
+	 
+   /* Emit indices and clipped vertices to one or more buffers.
+    */
+   if (VB->ClipOrMask) {
+      mmesa->elt_tab = tab;
+      mga_render_tab_elt[prim]( VB, 0, VB->EltPtr->count, 0 ); 
+   } else
+      mga_render_tab_elt_unclipped[prim]( VB, 0, VB->EltPtr->count, 0 );
+
+   /* Send to hardware and release any retained buffers.
+    */
+   release_bufs( mmesa );
+
+   /* This indicates that there is no cached data to reuse.  
+    */
+   VB->pipeline->data_valid = 0;
+   VB->pipeline->new_state = 0;
+}
+
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgaelttmp.h b/xc/lib/GL/mesa/src/drv/mga/mgaelttmp.h
new file mode 100644
index 000000000..ce758a532
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/mga/mgaelttmp.h
@@ -0,0 +1,269 @@
+/*
+ * DRI Hardware Device Driver for G200/G400
+ * Copyright (C) 1999 Keith Whitwell
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * KEITH WHITWELL, OR ANY OTHER CONTRIBUTORS BE LIABLE FOR ANY CLAIM, 
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE 
+ * OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+/* Buffers fill from high addresses down with vertices and from low
+ * addresses up with elements.  
+ */
+
+
+/* Emit the bulk of the vertices to the first dma buffer.  Leave
+ * empty slots for clipped vertices so that we can still address
+ * vertices by index.
+ */
+static void TAG(emit_unclipped_verts)( struct vertex_buffer *VB )
+{
+   GLuint i;
+   mgaContextPtr mmesa = MGA_CONTEXT(VB->ctx);
+   GLfloat *dev = VB->Projected->start;
+   GLubyte *color = VB->ColorPtr->start;
+   GLfloat *tex0_data = VB->TexCoordPtr[0]->start;
+   GLfloat *tex1_data = VB->TexCoordPtr[1]->start;
+   GLuint color_stride = VB->ColorPtr->stride;
+   GLuint tex0_stride = VB->TexCoordPtr[0]->stride;
+   GLuint tex1_stride = VB->TexCoordPtr[1]->stride;
+
+   GLfloat *f = mmesa->next_vert;
+   GLuint count = VB->Count;
+   GLubyte *clipmask = VB->ClipMask;
+
+   const GLfloat *m = mmesa->device_matrix;
+   const GLfloat sx = m[0], sy = m[5], sz = m[10];
+   const GLfloat tx = m[12], ty = m[13], tz = m[14];
+
+   mmesa->retained_buf = mmesa->elt_buf;
+   mmesa->first_vert_phys = mmesa->next_vert_phys;
+
+   for (i = 0 ; i < count ; f -= BUFFER_STRIDE, i++) 
+   {
+      if (!clipmask[i]) 
+      {
+	 f[0] = sx * dev[0] + tx;
+	 f[1] = sy * dev[1] + ty;
+	 f[2] = sz * dev[2] + tz;
+	 f[3] = dev[3]; 
+
+	 if (TYPE & MGA_RGBA_BIT) {
+#if defined(USE_X86_ASM)
+	    __asm__ (
+	    "movl (%%edx),%%eax   \n"
+	    "bswap %%eax          \n"
+	    "rorl $8,%%eax        \n"
+	    "movl %%eax,16(%%edi) \n"
+	    :
+	    : "d" (color), "D" (f)
+	    : "%eax" );
+#else
+	    GLubyte *b = (GLubyte *)&f[4];
+	    b[CLIP_UBYTE_B] = color[2];
+	    b[CLIP_UBYTE_G] = color[1];
+	    b[CLIP_UBYTE_R] = color[0];
+	    b[CLIP_UBYTE_A] = color[3];
+#endif
+	 }
+
+	 if (TYPE & MGA_TEX0_BIT) {
+/*  	    fprintf(stderr, "i %d tex0 %f, %f\n", i,  */
+/*  		    tex0_data[0], tex0_data[1]); */
+	    *(int*)&f[6] = *(int*)&tex0_data[0];
+	    *(int*)&f[7] = *(int*)&tex0_data[1];
+	 }
+
+	 if (TYPE & MGA_TEX1_BIT) {
+	    *(int*)&f[8] = *(int*)&tex1_data[0];
+	    *(int*)&f[9] = *(int*)&tex1_data[1];
+	 }
+      }
+
+      STRIDE_F(dev, 16);
+      if (TYPE & MGA_RGBA_BIT) color += color_stride;
+      if (TYPE & MGA_TEX0_BIT) STRIDE_F(tex0_data, tex0_stride);
+      if (TYPE & MGA_TEX1_BIT) STRIDE_F(tex1_data, tex1_stride);
+   }
+
+   mmesa->next_vert = f;
+   mmesa->next_vert_phys -= count * BUFFER_STRIDE * sizeof(GLuint);
+}
+
+
+/* Build three temporary clipspace vertex for clipping a triangle.
+ * Recreate from the VB data rather than trying to read back from
+ * uncached memory.  
+ */
+static void TAG(build_tri_verts)( mgaContextPtr mmesa,
+				  struct vertex_buffer *VB, 
+				  GLfloat *O, 
+				  GLuint *elt ) 
+{
+   int i;
+
+   for (i = 0 ; i < 3 ; i++, O += CLIP_STRIDE) {
+      GLfloat *clip = VB->Clip.start + elt[i]*4;
+
+      O[0] = clip[0];
+      O[1] = clip[1];
+      O[2] = clip[2]; 
+      O[3] = clip[3];
+
+      if (TYPE & MGA_RGBA_BIT) {
+	 GLubyte *col = VEC_ELT(VB->ColorPtr, GLubyte, elt[i]);
+	 GLubyte *b = (GLubyte *)&O[4];
+	 b[CLIP_UBYTE_R] = col[0];
+	 b[CLIP_UBYTE_G] = col[1];
+	 b[CLIP_UBYTE_B] = col[2];
+	 b[CLIP_UBYTE_A] = col[3];
+      }
+
+      if (0)
+	 fprintf(stderr, 
+	      "build_tri_vert elt[%d]: %d phys: %x (first_phys %x elt_buf %x\n",
+	      i, elt[i], UNCLIPPED_VERT(elt[i]), 
+	      mmesa->first_vert_phys, (GLuint)mmesa->elt_buf);
+
+      *(GLuint *)&O[5] = UNCLIPPED_VERT(elt[i]);
+
+      if (TYPE & MGA_TEX0_BIT) {
+	 GLfloat *tex0_data = VEC_ELT(VB->TexCoordPtr[0], GLfloat, elt[i]);
+	 *(int*)&O[6] = *(int*)&tex0_data[0];
+	 *(int*)&O[7] = *(int*)&tex0_data[1];
+      }
+
+      if (TYPE & MGA_TEX1_BIT) {
+	 GLfloat *tex1_data = VEC_ELT(VB->TexCoordPtr[1], GLfloat, elt[i]);
+	 *(int*)&O[8] = *(int*)&tex1_data[0];
+	 *(int*)&O[9] = *(int*)&tex1_data[1];
+      }
+   }
+}
+
+
+/* Interpolate between two of the vertices constructed above.
+ */
+static void TAG(interp)( GLfloat t, 
+			 GLfloat *O,
+			 const GLfloat *I,
+			 const GLfloat *J )
+{
+   O[0] = LINTERP(t, I[0], J[0]);
+   O[1] = LINTERP(t, I[1], J[1]);
+   O[2] = LINTERP(t, I[2], J[2]);
+   O[3] = LINTERP(t, I[3], J[3]);
+
+   if (TYPE & MGA_RGBA_BIT) {
+      INTERP_RGBA(t,
+		  ((GLubyte *)&(O[4])),
+		  ((GLubyte *)&(I[4])),
+		  ((GLubyte *)&(J[4])));
+   }
+
+   if (0) fprintf(stderr, "setting 0x%x to ~0\n", (GLuint)&O[5]);
+
+   *(GLuint *)&O[5] = ~0;	/* note that this is a new vertex */
+ 
+   if (TYPE & MGA_TEX0_BIT) {
+      O[6] = LINTERP(t, I[6], J[6]);
+      O[7] = LINTERP(t, I[7], J[7]);
+   }
+
+   if (TYPE & MGA_TEX1_BIT) {
+      O[8] = LINTERP(t, I[8], J[8]);
+      O[9] = LINTERP(t, I[9], J[9]);
+   }
+}
+
+
+
+/* When clipping is complete, scan the final vertex list and emit any
+ * new ones to dma buffers.  Update the element list to a format
+ * suitable for sending to hardware.
+ */
+static void TAG(project_and_emit_verts)( mgaContextPtr mmesa,
+					 const GLfloat *verts,
+					 GLuint *elt,
+					 int nr)
+{
+   
+   GLfloat *O = mmesa->next_vert;
+   GLuint phys = mmesa->next_vert_phys;
+
+   const GLfloat *m = mmesa->device_matrix;
+   const GLfloat sx = m[0], sy = m[5], sz = m[10];
+   const GLfloat tx = m[12], ty = m[13], tz = m[14];
+   GLuint i;
+   
+   for (i = 0 ; i < nr ; i++) {
+      const GLfloat *I = &verts[elt[i] * CLIP_STRIDE];
+      GLuint tmp = *(GLuint *)&I[5];
+
+      if (0) fprintf(stderr, "elt[%d] (tmp 0x%x %d) %d --> ", i, (GLuint)&I[5],
+	      tmp, elt[i]);
+
+
+      if ((elt[i] = tmp) == ~0) 
+      {      
+	 GLfloat oow = 1.0/I[3];
+
+	 elt[i] = phys;
+	 phys -= BUFFER_STRIDE * sizeof(GLuint);
+	 
+	 O[0] = sx * I[0] * oow + tx;
+	 O[1] = sy * I[1] * oow + ty;
+	 O[2] = sz * I[2] * oow + tz;
+	 O[3] = oow; 
+
+	 if (TYPE & MGA_RGBA_BIT) {
+	    *(int*)&O[4] = *(int*)&I[4];
+	 }
+ 
+	 if (TYPE & MGA_TEX0_BIT) {
+	    *(int*)&O[6] = *(int*)&I[6];
+	    *(int*)&O[7] = *(int*)&I[7];
+	 }
+
+	 if (TYPE & MGA_TEX1_BIT) {
+	    *(int*)&O[8] = *(int*)&I[8];
+	    *(int*)&O[9] = *(int*)&I[9];
+	 }
+
+	 O -= BUFFER_STRIDE;
+      }
+      if (0) fprintf(stderr, "0x%x\n", elt[i]);
+   }
+
+   mmesa->next_vert = O;
+   mmesa->next_vert_phys = phys;
+}
+
+
+
+static void TAG(mga_init_eltpath)( struct mga_elt_tab *tab )
+{
+   tab->emit_unclipped_verts = TAG(emit_unclipped_verts);
+   tab->build_tri_verts = TAG(build_tri_verts);
+   tab->interp = TAG(interp);
+   tab->project_and_emit_verts = TAG(project_and_emit_verts);
+}
+
+#undef TYPE
+#undef TAG
+#undef STRIDE
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgafastpath.c b/xc/lib/GL/mesa/src/drv/mga/mgafastpath.c
index cd316e162..73a939b74 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgafastpath.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgafastpath.c
@@ -488,6 +488,16 @@ void mgaDDFastPath( struct vertex_buffer *VB )
 
    gl_prepare_arrays_cva( VB );	                 /* still need this */
 
+   if (gl_reduce_prim[prim] == GL_TRIANGLES && 
+       VB->Count < (MGA_DMA_BUF_SZ / 48) &&
+       (ctx->ModelProjectMatrix.flags & (MAT_FLAG_GENERAL|
+					 MAT_FLAG_PERSPECTIVE)) &&
+       mmesa->mgaScreen->chipset == MGA_CARD_TYPE_G400) 
+   {
+      mgaDDEltPath( VB );
+      return;
+   }
+
    /* Reserve enough space for the pathological case.
     */
    if (VB->EltPtr->count * 12 > MGA_DRIVER_DATA(VB)->size) {
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgafasttmp.h b/xc/lib/GL/mesa/src/drv/mga/mgafasttmp.h
index 43caddcac..683640409 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgafasttmp.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgafasttmp.h
@@ -39,7 +39,6 @@ static void TAG(mga_setup_full)( struct vertex_buffer *VB, GLuint do_cliptest )
    const GLfloat * const m = ctx->ModelProjectMatrix.m;
    GLuint start = VB->CopyStart;
    GLuint count = VB->Count;
-   GLuint i;
 
    gl_xform_points3_v16_general(MGA_DRIVER_DATA(VB)->verts[start].f,
 				m, 
@@ -71,26 +70,63 @@ static void TAG(mga_setup_full)( struct vertex_buffer *VB, GLuint do_cliptest )
       GLuint tex1_stride = VB->TexCoordPtr[1]->stride;
 
       GLfloat *f = MGA_DRIVER_DATA(VB)->verts[start].f;
-
-      for (i = start ; i < count ; i++, f += 16) {
+      GLfloat *end = f+(16*(count-start));
+      while (f != end) {
 	 if (TYPE & MGA_RGBA_BIT) {
+#if defined(USE_X86_ASM)
+	    __asm__ (
+	    "movl (%%edx),%%eax   \n"
+	    "bswap %%eax          \n"
+	    "rorl $8,%%eax        \n"
+	    "movl %%eax,16(%%edi) \n"
+	    :
+	    : "d" (color), "D" (f)
+	    : "%eax" );
+#else
+	    GLubyte *col = color;
 	    GLubyte *b = (GLubyte *)&f[CLIP_UBYTE_COLOR];
-	    GLubyte *col = color; color += color_stride;
-	    b[CLIP_UBYTE_R] = col[0];
-	    b[CLIP_UBYTE_G] = col[1];
 	    b[CLIP_UBYTE_B] = col[2];
+	    b[CLIP_UBYTE_G] = col[1];
+	    b[CLIP_UBYTE_R] = col[0];
 	    b[CLIP_UBYTE_A] = col[3];
+#endif
 	 }
 	 if (TYPE & MGA_TEX0_BIT) {
-	    f[CLIP_S0] = tex0_data[0];
-	    f[CLIP_T0] = tex0_data[1];
-	    STRIDE_F(tex0_data, tex0_stride);
+#if defined (USE_X86_ASM)
+	    __asm__ (
+	    "movl (%%ecx), %%eax    \n"
+	    "movl %%eax, 24(%%edi)  \n"
+	    "movl 4(%%ecx), %%eax   \n"
+	    "movl %%eax, 28(%%edi)"
+	    :
+	    : "c" (tex0_data), "D" (f)
+	    : "%eax");
+#else
+	    *(unsigned int *)(f+CLIP_S0) = *(unsigned int *)tex0_data;
+	    *(unsigned int *)(f+CLIP_T0) = *(unsigned int *)(tex0_data+1);
+#endif
 	 }
 	 if (TYPE & MGA_TEX1_BIT) {
-	    f[CLIP_S1] = tex1_data[0];
-	    f[CLIP_T1] = tex1_data[1];
-	    STRIDE_F(tex1_data, tex1_stride);
+	    /* Hits a second cache line.
+	     */
+#if defined (USE_X86_ASM)
+	    __asm__ (
+	    "movl (%%esi), %%eax    \n"
+	    "movl %%eax, 32(%%edi)  \n"
+	    "movl 4(%%esi), %%eax   \n"
+	    "movl %%eax, 36(%%edi)"
+	    :
+	    : "S" (tex1_data), "D" (f)
+	    : "%eax");
+#else
+	    *(unsigned int *)(f+CLIP_S1) = *(unsigned int *)tex1_data;
+	    *(unsigned int *)(f+CLIP_T1) = *(unsigned int *)(tex1_data+1);
+#endif	      
 	 }
+	 if(TYPE & MGA_RGBA_BIT)color += color_stride;
+	 if(TYPE & MGA_TEX0_BIT)STRIDE_F(tex0_data, tex0_stride);
+	 if(TYPE & MGA_TEX1_BIT)STRIDE_F(tex1_data, tex1_stride);
+	 f += 16;
       }
    }
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgaioctl.c b/xc/lib/GL/mesa/src/drv/mga/mgaioctl.c
index c1819924d..640849e3f 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgaioctl.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgaioctl.c
@@ -14,114 +14,137 @@
 #include "mgalog.h"
 #include "mgavb.h"
 #include "mgatris.h"
+#include "mgabuffers.h"
+
 
 #include "drm.h"
 #include <sys/ioctl.h>
 
+#define DEPTH_SCALE 65535.0F
+
 static void mga_iload_dma_ioctl(mgaContextPtr mmesa,
-				int x1, int y1, int x2, int y2, 
-				unsigned long dest, unsigned int maccess)
+				unsigned long dest, 
+				int length)
 {
    int retcode;
    drm_mga_iload_t iload;
-   drmBufPtr buf = mmesa->dma_buffer;
+   drmBufPtr buf = mmesa->iload_buffer;
    
    iload.idx = buf->idx;
    iload.destOrg = dest;	
-   iload.mAccess = maccess;
-   iload.texture.x1 = x1;
-   iload.texture.y1 = y1;
-   iload.texture.y2 = x2;
-   iload.texture.x2 = y2;
+   iload.length = length;
+
+   if (MGA_DEBUG&DEBUG_VERBOSE_IOCTL) 
+      fprintf(stderr, "DRM_IOCTL_MGA_ILOAD idx %d dst %x length %d\n",
+	   iload.idx, iload.destOrg, iload.length);
+	   
    
    if ((retcode = ioctl(mmesa->driFd, DRM_IOCTL_MGA_ILOAD, &iload))) {
       printf("send iload retcode = %d\n", retcode);
       exit(1);
    }
-}
 
+   mmesa->iload_buffer = 0;
+
+   if (MGA_DEBUG&DEBUG_VERBOSE_IOCTL) 
+      fprintf(stderr, "finished iload dma put\n");
+
+}
 
-static void mga_vertex_dma_ioctl(mgaContextPtr mmesa)
+int mgaUpdateLock( mgaContextPtr mmesa, drmLockFlags flags )
 {
-   int retcode;
-   int size = MGA_DMA_BUF_SZ;
-   drmDMAReq dma;
-   drmBufPtr buf = mmesa->dma_buffer;
+   drm_lock_t lock;
+   
+   lock.flags = 0;
 
-   dma.context = mmesa->hHWContext;
-   dma.send_count = 1;
-   dma.send_list = &buf->idx;
-   dma.send_sizes = &size;
-   dma.flags = DRM_DMA_WAIT;
-   dma.request_count = 0;
-   dma.request_size = 0;
-   dma.request_list = 0;
-   dma.request_sizes = 0;
-
-   if ((retcode = drmDMA(mmesa->driFd, &dma))) {
-      printf("send iload retcode = %d\n", retcode);
-      exit(1);
+   if (mmesa->sarea->last_quiescent != mmesa->sarea->last_enqueue &&
+       flags & DRM_LOCK_QUIESCENT) {
+      if (MGA_DEBUG&DEBUG_VERBOSE_IOCTL)
+	fprintf(stderr, "mgaLockQuiescent\n");
+      lock.flags |= _DRM_LOCK_QUIESCENT;
+   }
+   
+   if (flags & DRM_LOCK_FLUSH)      lock.flags |= _DRM_LOCK_FLUSH;
+   if (flags & DRM_LOCK_FLUSH_ALL)  lock.flags |= _DRM_LOCK_FLUSH_ALL;
+ 
+   if (!lock.flags)
+      return 0;
+
+   if(ioctl(mmesa->driFd, DRM_IOCTL_MGA_FLUSH, &lock)) {
+      fprintf(stderr, "Lockupdate failed\n");
+      return -1;
    }
+   
+   if(flags & DRM_LOCK_QUIESCENT)
+     mmesa->sarea->last_quiescent = mmesa->sarea->last_enqueue;
+ 
+   return 0;
 }
 
-
-static void mga_get_buffer_ioctl( mgaContextPtr mmesa )
+static drmBufPtr mga_get_buffer_ioctl( mgaContextPtr mmesa )
 {
    int idx = 0;
    int size = 0;
    drmDMAReq dma;
    int retcode;
+   drmBufPtr buf;
    
-   fprintf(stderr,  "Getting dma buffer\n");
+   if (MGA_DEBUG&DEBUG_VERBOSE_IOCTL)
+      fprintf(stderr,  "Getting dma buffer\n");
    
    dma.context = mmesa->hHWContext;
    dma.send_count = 0;
    dma.send_list = NULL;
    dma.send_sizes = NULL;
-   dma.flags = DRM_DMA_WAIT;
+   dma.flags = 0;
    dma.request_count = 1;
    dma.request_size = MGA_DMA_BUF_SZ;
    dma.request_list = &idx;
    dma.request_sizes = &size;
-   
-   if ((retcode = drmDMA(mmesa->driFd, &dma))) {
-      fprintf(stderr, "request drmDMA retcode = %d\n", retcode);
-      exit(1);
-   }
+   dma.granted_count = 0;
 
-   mmesa->dma_buffer = &mmesa->mgaScreen->bufs->list[idx];
-}
 
-static void mga_swap_ioctl( mgaContextPtr mmesa )
-{
-   int retcode;
-   drm_mga_swap_t swap;
-      
-   if((retcode = ioctl(mmesa->driFd, DRM_IOCTL_MGA_SWAP, &swap))) {
-      printf("send swap retcode = %d\n", retcode);
-      exit(1);
-   }
-}
+   if (MGA_DEBUG&DEBUG_VERBOSE_IOCTL)
+      fprintf(stderr, "drmDMA (get) ctx %d count %d size 0x%x\n",
+	   dma.context, dma.request_count, 
+	   dma.request_size);
 
+   while (1) {
+      retcode = drmDMA(mmesa->driFd, &dma);
 
+      if (MGA_DEBUG&DEBUG_VERBOSE_IOCTL)
+	 fprintf(stderr, "retcode %d sz %d idx %d count %d\n", 
+		 retcode,
+		 dma.request_sizes[0],
+		 dma.request_list[0],
+		 dma.granted_count);
 
-static void mga_clear_ioctl( mgaContextPtr mmesa, int flags, int col, int zval )
-{
-   int retcode;
-   drm_mga_clear_t clear;
-   
-   clear.flags = flags;
-   clear.clear_color = col;
-   clear.clear_depth = zval;
-      
-   if((retcode = ioctl(mmesa->driFd, DRM_IOCTL_MGA_CLEAR, &clear))) {
-      printf("send clear retcode = %d\n", retcode);
-      exit(1);
+      if (retcode == 0 && 
+	  dma.request_sizes[0] &&
+	  dma.granted_count) 
+	 break;
+
+      if (MGA_DEBUG&DEBUG_VERBOSE_IOCTL)  
+	 fprintf(stderr, "\n\nflush");
+      mgaUpdateLock( mmesa, DRM_LOCK_FLUSH );
    }
-}
 
+   buf = &(mmesa->mgaScreen->bufs->list[idx]);
+   buf->used = 0;
 
+   if (MGA_DEBUG&DEBUG_VERBOSE_IOCTL)
+      fprintf(stderr, 
+	   "drmDMA (get) returns size[0] 0x%x idx[0] %d\n"
+	   "dma_buffer now: buf idx: %d size: %d used: %d\n",
+	   dma.request_sizes[0], dma.request_list[0],
+	   buf->idx, buf->total,
+	   buf->used);
 
+   if (MGA_DEBUG&DEBUG_VERBOSE_IOCTL)
+      fprintf(stderr, "finished getbuffer\n");
+
+   return buf;
+}
 
 
 
@@ -130,52 +153,59 @@ GLbitfield mgaClear( GLcontext *ctx, GLbitfield mask, GLboolean all,
 		     GLint cx, GLint cy, GLint cw, GLint ch ) 
 {
    mgaContextPtr mmesa = MGA_CONTEXT( ctx );
-   GLuint c = mmesa->ClearColor;
-   mgaUI32 zval = (mgaUI32) (ctx->Depth.Clear * DEPTH_SCALE);
    __DRIdrawablePrivate *dPriv = mmesa->driDrawable;
-   int flags = 0;
+   const GLuint colorMask = *((GLuint *) &ctx->Color.ColorMask);
+   drm_mga_clear_t clear;
+   int retcode;
    int i;
+   static int nrclears;
 
-   mgaMsg( 10, "mgaClear( %i, %i, %i, %i, %i )\n", 
-	   mask, x, y, width, height );
-
+   if (0) fprintf(stderr, "clear %d: %d,%d %dx%d\n", all,cx,cy,cw,ch);
 
-   mgaFlushVertices( mmesa );
+   clear.flags = 0;
+   clear.clear_color = mmesa->ClearColor;
+   clear.clear_depth = (mgaUI32) (ctx->Depth.Clear * DEPTH_SCALE);
 
+   FLUSH_BATCH( mmesa );
+	
+   if ((mask & DD_FRONT_LEFT_BIT) && colorMask == ~0) {
+      clear.flags |= MGA_FRONT;
+      mask &= ~DD_FRONT_LEFT_BIT;
+   }
 
-   if (mask & GL_COLOR_BUFFER_BIT) {
-      if (ctx->Color.DriverDrawBuffer == GL_FRONT_LEFT) {
-	 flags |= MGA_CLEAR_FRONT;
-	 mask &= ~GL_COLOR_BUFFER_BIT;
-      } else if (ctx->Color.DriverDrawBuffer == GL_BACK_LEFT) {
-	 flags |= MGA_CLEAR_BACK;
-	 mask &= ~GL_COLOR_BUFFER_BIT;
-      }
+   if ((mask & DD_BACK_LEFT_BIT) && colorMask == ~0) {
+      clear.flags |= MGA_BACK;
+      mask &= ~DD_BACK_LEFT_BIT;
    }
 
-   if ((flags & GL_DEPTH_BUFFER_BIT) && ctx->Depth.Mask) {
-      flags |= MGA_CLEAR_DEPTH;
-      mask &= ~GL_DEPTH_BUFFER_BIT;
+   if ((mask & DD_DEPTH_BIT) && ctx->Depth.Mask) {
+      clear.flags |= MGA_DEPTH;
+      mask &= ~DD_DEPTH_BIT;
    }
 
-   if (!flags)
+   if (!clear.flags)
       return mask;
 
    LOCK_HARDWARE( mmesa );
-   
+
+   if (mmesa->dirty_cliprects)
+      mgaUpdateRects( mmesa, (MGA_FRONT|MGA_BACK));
+
    /* flip top to bottom */
    cy = dPriv->h-cy-ch;
    cx += mmesa->drawX;
    cy += mmesa->drawY;
 
-   for (i = 0 ; i < dPriv->numClipRects ; ) { 	 
+   if (MGA_DEBUG&DEBUG_VERBOSE_IOCTL)
+      fprintf(stderr, "Clear, bufs %x nbox %d\n", 
+	      (int)clear.flags, (int)mmesa->numClipRects);
 
-      /* Use the cliprects for the current draw buffer
-       */
+   for (i = 0 ; i < mmesa->numClipRects ; ) 
+   { 	 
       int nr = MIN2(i + MGA_NR_SAREA_CLIPRECTS, mmesa->numClipRects);
       XF86DRIClipRectRec *box = mmesa->pClipRects;	 
-      xf86drmClipRectRec *b = mmesa->sarea->boxes;
-      mmesa->sarea->nbox = nr - i;
+      drm_clip_rect_t *b = mmesa->sarea->boxes;
+      int n = 0;
 
       if (!all) {
 	 for ( ; i < nr ; i++) {
@@ -196,16 +226,37 @@ GLbitfield mgaClear( GLcontext *ctx, GLbitfield mask, GLboolean all,
 	    b->x2 = x + w;
 	    b->y2 = y + h;
 	    b++;
+	    n++;
 	 }
       } else {
-	 for ( ; i < nr ; i++) 
-	    *b++ = *(xf86drmClipRectRec *)&box[i];
+	 for ( ; i < nr ; i++) {
+	    *b++ = *(drm_clip_rect_t *)&box[i];
+	    n++;
+	 }
       }
 
-      mga_clear_ioctl( mmesa, mask, c, zval );
+
+      if (MGA_DEBUG&DEBUG_VERBOSE_IOCTL)
+	 fprintf(stderr, 
+		 "DRM_IOCTL_MGA_CLEAR flag 0x%x color %x depth %x nbox %d\n",
+		 clear.flags, clear.clear_color, 
+		 clear.clear_depth, mmesa->sarea->nbox);
+		
+
+      mmesa->sarea->nbox = n;
+
+      retcode = ioctl(mmesa->driFd, DRM_IOCTL_MGA_CLEAR, &clear);
+      if (retcode) {
+	 printf("send clear retcode = %d\n", retcode);
+	 exit(1);
+      }
+      if (MGA_DEBUG&DEBUG_VERBOSE_IOCTL)
+	 fprintf(stderr, "finished clear %d\n", ++nrclears);
    }
 
    UNLOCK_HARDWARE( mmesa );
+   mmesa->dirty |= MGA_UPLOAD_CLIPRECTS;
+
    return mask;
 }
 
@@ -218,50 +269,61 @@ GLbitfield mgaClear( GLcontext *ctx, GLbitfield mask, GLboolean all,
 void mgaSwapBuffers( mgaContextPtr mmesa ) 
 {
    __DRIdrawablePrivate *dPriv = mmesa->driDrawable;
+   XF86DRIClipRectPtr pbox;
+   int nbox;
+   drm_mga_swap_t swap;
+   static int nrswaps;
+   int retcode;
    int i;
+   int tmp;
 
-   mgaFlushVertices( mmesa );
 
-   LOCK_HARDWARE( mmesa );
-   {
-      /* Use the frontbuffer cliprects
-       */
-      XF86DRIClipRectPtr pbox = dPriv->pClipRects;
-      int nbox = dPriv->numClipRects;
+   FLUSH_BATCH( mmesa );
 
-      for (i = 0 ; i < nbox ; )
-      {
-	 int nr = MIN2(i + MGA_NR_SAREA_CLIPRECTS, dPriv->numClipRects);
-	 XF86DRIClipRectRec *b = (XF86DRIClipRectRec *)mmesa->sarea->boxes;
-	 mmesa->sarea->nbox = nr - i;
+   LOCK_HARDWARE( mmesa );
+   
+   /* Use the frontbuffer cliprects
+    */
+   if (mmesa->dirty_cliprects & MGA_FRONT) 
+      mgaUpdateRects( mmesa, MGA_FRONT );
+   
 
-	 for ( ; i < nr ; i++) 
-	    *b++ = pbox[i];
-	 
-	 mga_swap_ioctl( mmesa );
-      }
-   }
+   pbox = dPriv->pClipRects;
+   nbox = dPriv->numClipRects;
 
+   if (0) fprintf(stderr, "swap, nbox %d\n", nbox);
 
-#if 1
-   UNLOCK_HARDWARE(mmesa);
-#else
+   for (i = 0 ; i < nbox ; )
    {
-      last_enqueue = mmesa->sarea->lastEnqueue;
-      last_dispatch = mmesa->sarea->lastDispatch;
-      UNLOCK_HARDWARE;
+      int nr = MIN2(i + MGA_NR_SAREA_CLIPRECTS, dPriv->numClipRects);
+      XF86DRIClipRectRec *b = (XF86DRIClipRectRec *)mmesa->sarea->boxes;
+
+      mmesa->sarea->nbox = nr - i;
+	 
+      for ( ; i < nr ; i++) 
+	 *b++ = pbox[i];
       
-      /* Throttle runaway apps - there should be an easier way to sleep
-       * on dma without locking out the rest of the system!
-       */
-      if (mmesa->lastSwap > last_dispatch) {
-	 drmGetLock(mmesa->driFd, mmesa->hHWContext, DRM_LOCK_QUIESCENT);	
-	 DRM_UNLOCK(mmesa->driFd, mmesa->driHwLock, mmesa->hHWContext);	
+      if (0)
+	 fprintf(stderr, "DRM_IOCTL_MGA_SWAP\n"); 
+
+      if((retcode = ioctl(mmesa->driFd, DRM_IOCTL_MGA_SWAP, &swap))) {
+	 printf("send swap retcode = %d\n", retcode);
+	 exit(1);
       }
 
-      mmesa->lastSwap = last_enqueue;
+      if (MGA_DEBUG&DEBUG_VERBOSE_IOCTL)
+	 fprintf(stderr, "finished swap %d\n", ++nrswaps);
    }
-#endif
+
+   tmp = GET_ENQUEUE_AGE(mmesa);
+
+   UNLOCK_HARDWARE( mmesa );
+
+   if (GET_DISPATCH_AGE(mmesa) < mmesa->lastSwap)
+      mgaWaitAge(mmesa, mmesa->lastSwap);
+
+   mmesa->lastSwap = tmp;
+   mmesa->dirty |= MGA_UPLOAD_CLIPRECTS;
 }
 
 
@@ -269,13 +331,165 @@ void mgaSwapBuffers( mgaContextPtr mmesa )
  */
 void mgaDDFinish( GLcontext *ctx  ) 
 {
-   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
-   drmGetLock(mmesa->driFd, mmesa->hHWContext, DRM_LOCK_QUIESCENT);	
-   DRM_UNLOCK(mmesa->driFd, mmesa->driHwLock, mmesa->hHWContext);	
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
+
+   FLUSH_BATCH( mmesa );
+
+   if (mmesa->sarea->last_quiescent != mmesa->sarea->last_enqueue) {
+      if (MGA_DEBUG&DEBUG_VERBOSE_IOCTL)
+	 fprintf(stderr, "mgaRegetLockQuiescent\n");
+
+      LOCK_HARDWARE( mmesa );
+      mgaUpdateLock( mmesa, DRM_LOCK_QUIESCENT | DRM_LOCK_FLUSH);
+      UNLOCK_HARDWARE( mmesa );
+
+      mmesa->sarea->last_quiescent = mmesa->sarea->last_enqueue;
+   }
+}
+
+void mgaWaitAgeLocked( mgaContextPtr mmesa, int age  ) 
+{
+   if (GET_DISPATCH_AGE(mmesa) < age) {
+      if (0) fprintf(stderr, "\n\n\nmgaWaitAgeLocked\n");
+      mgaUpdateLock( mmesa, DRM_LOCK_FLUSH );
+   }
+}
+
+
+void mgaWaitAge( mgaContextPtr mmesa, int age  ) 
+{
+   if (GET_DISPATCH_AGE(mmesa) < age) {
+      LOCK_HARDWARE(mmesa);
+      if (GET_DISPATCH_AGE(mmesa) < age) {
+	 if (0) fprintf(stderr, "\n\n\nmgaWaitAge\n");
+	 mgaUpdateLock( mmesa, DRM_LOCK_FLUSH );
+      }
+      UNLOCK_HARDWARE(mmesa);
+   }
+}
+
+
+static int intersect_rect( drm_clip_rect_t *out,
+			   drm_clip_rect_t *a,
+			   drm_clip_rect_t *b )
+{
+   *out = *a;
+   if (b->x1 > out->x1) out->x1 = b->x1;
+   if (b->y1 > out->y1) out->y1 = b->y1;
+   if (b->x2 < out->x2) out->x2 = b->x2;
+   if (b->y2 < out->y2) out->y2 = b->y2;
+   if (out->x1 >= out->x2) return 0;
+   if (out->y1 >= out->y2) return 0;
+   return 1;
+}
+
+
+
+
+static void age_mmesa( mgaContextPtr mmesa, int age )
+{
+   if (mmesa->CurrentTexObj[0]) mmesa->CurrentTexObj[0]->age = age;
+   if (mmesa->CurrentTexObj[1]) mmesa->CurrentTexObj[1]->age = age;
 }
 
+void mgaFlushVerticesLocked( mgaContextPtr mmesa )
+{
+   drm_clip_rect_t *pbox = (drm_clip_rect_t *)mmesa->pClipRects;
+   int nbox = mmesa->numClipRects;
+   drmBufPtr buffer = mmesa->vertex_dma_buffer;
+   drm_mga_vertex_t vertex;
+   int i;
+
+   mmesa->vertex_dma_buffer = 0;
+
+   if (!buffer)
+      return;
+
+   if (mmesa->dirty_cliprects & mmesa->draw_buffer)
+      mgaUpdateRects( mmesa, mmesa->draw_buffer );
+
+   if (mmesa->dirty & ~MGA_UPLOAD_CLIPRECTS) 
+      mgaEmitHwStateLocked( mmesa );
+
+   /* FIXME: Workaround bug in kernel module.
+    */
+   mmesa->sarea->dirty |= MGA_UPLOAD_CTX; 
+      
+   /* FIXME: dstorg bug
+    */
+   if (0)
+      if (mmesa->lastX != mmesa->drawX || mmesa->lastY != mmesa->drawY)
+	 fprintf(stderr, "****** last: %d,%d current: %d,%d\n",
+		 mmesa->lastX, mmesa->lastY,
+		 mmesa->drawX, mmesa->drawY);
+
+   vertex.idx = buffer->idx;
+   vertex.used = buffer->used;
+   vertex.discard = 0;
+
+   if (!nbox)
+      vertex.used = 0;
+
+   if (nbox >= MGA_NR_SAREA_CLIPRECTS)
+      mmesa->dirty |= MGA_UPLOAD_CLIPRECTS;
+
+   if (!vertex.used || !(mmesa->dirty & MGA_UPLOAD_CLIPRECTS)) 
+   {
+      if (nbox == 1) 
+	 mmesa->sarea->nbox = 0;
+      else
+	 mmesa->sarea->nbox = nbox;
+
+      if (MGA_DEBUG&DEBUG_VERBOSE_IOCTL)  
+	 fprintf(stderr, "Firing vertex -- case a nbox %d\n", nbox);
+
+      vertex.discard = 1;
+      ioctl(mmesa->driFd, DRM_IOCTL_MGA_VERTEX, &vertex);
+      age_mmesa(mmesa, mmesa->sarea->last_enqueue);
+   } 
+   else 
+   {      
+      for (i = 0 ; i < nbox ; )
+      {
+	 int nr = MIN2(i + MGA_NR_SAREA_CLIPRECTS, nbox);
+	 drm_clip_rect_t *b = mmesa->sarea->boxes;
+
+	 if (mmesa->scissor) {
+	    mmesa->sarea->nbox = 0;	 
+
+	    for ( ; i < nr ; i++) {	       
+	       *b = pbox[i];
+	       if (intersect_rect(b, b, &mmesa->scissor_rect)) {
+		  mmesa->sarea->nbox++;
+		  b++;
+	       } 
+	    }
+
+	    /* Culled?
+	     */
+	    if (!mmesa->sarea->nbox) {
+	       if (nr < nbox) continue;
+	       vertex.used = 0;
+	    }
+	 } else {
+	    mmesa->sarea->nbox = nr - i;
+	    for ( ; i < nr ; i++) 
+	       *b++ = pbox[i];
+	 }
+
+	 /* Finished with the buffer?
+	  */
+	 if (nr == nbox) 
+	    vertex.discard = 1;
 
+	 mmesa->sarea->dirty |= MGA_UPLOAD_CLIPRECTS;
+	 ioctl(mmesa->driFd, DRM_IOCTL_MGA_VERTEX, &vertex);
+	 age_mmesa(mmesa, mmesa->sarea->last_enqueue);
+      }
+   }
 
+   mmesa->dirty &= ~MGA_UPLOAD_CLIPRECTS;
+}
 
 void mgaFlushVertices( mgaContextPtr mmesa ) 
 {
@@ -284,38 +498,221 @@ void mgaFlushVertices( mgaContextPtr mmesa )
    UNLOCK_HARDWARE( mmesa );
 }
 
+void mgaFlushEltsLocked( mgaContextPtr mmesa ) 
+{
+   if (mmesa->first_elt != mmesa->next_elt) {
+      mgaFireEltsLocked( mmesa, 
+			 ((GLuint)mmesa->first_elt - 
+			  (GLuint)mmesa->elt_buf->address),
+			 ((GLuint)mmesa->next_elt - 
+			  (GLuint)mmesa->elt_buf->address),
+			 0 );
+      mmesa->first_elt = mmesa->next_elt;
+   }
+}
 
-void mgaFlushVerticesLocked( mgaContextPtr mmesa )
+void mgaFlushElts( mgaContextPtr mmesa ) 
+{
+   LOCK_HARDWARE( mmesa );
+   mgaFlushEltsLocked( mmesa );
+   UNLOCK_HARDWARE( mmesa );
+}
+
+
+mgaUI32 *mgaAllocVertexDwords( mgaContextPtr mmesa, int dwords )
+{
+   int bytes = dwords * 4;
+   mgaUI32 *head;
+
+   if (!mmesa->vertex_dma_buffer) {
+      LOCK_HARDWARE( mmesa );
+
+      if (mmesa->first_elt != mmesa->next_elt) 
+	 mgaFlushEltsLocked(mmesa);
+
+      mmesa->vertex_dma_buffer = mga_get_buffer_ioctl( mmesa );
+      UNLOCK_HARDWARE( mmesa );
+   } else if (mmesa->vertex_dma_buffer->used + bytes > 
+	      mmesa->vertex_dma_buffer->total) {
+      LOCK_HARDWARE( mmesa );
+      mgaFlushVerticesLocked( mmesa );
+      mmesa->vertex_dma_buffer = mga_get_buffer_ioctl( mmesa );
+      UNLOCK_HARDWARE( mmesa );
+   }
+
+   head = (mgaUI32 *)((char *)mmesa->vertex_dma_buffer->address + 
+		      mmesa->vertex_dma_buffer->used);
+
+   mmesa->vertex_dma_buffer->used += bytes;
+   return head;
+}
+
+
+void mgaFireILoadLocked( mgaContextPtr mmesa, 
+			 GLuint offset, GLuint length )
+{
+   if (!mmesa->iload_buffer) {
+      fprintf(stderr, "mgaFireILoad: no buffer\n");
+      return;
+   }
+
+   if (MGA_DEBUG&DEBUG_VERBOSE_IOCTL)
+      fprintf(stderr, "mgaFireILoad idx %d ofs 0x%x length %d\n",
+	      mmesa->iload_buffer->idx, (int)offset, (int)length );
+
+   mga_iload_dma_ioctl( mmesa, offset, length );
+}
+
+void mgaGetILoadBufferLocked( mgaContextPtr mmesa )
+{
+   if (MGA_DEBUG&DEBUG_VERBOSE_IOCTL)
+      fprintf(stderr, "mgaGetIloadBuffer (buffer now %p)\n",
+	   mmesa->iload_buffer);
+
+   mmesa->iload_buffer = mga_get_buffer_ioctl( mmesa );
+}
+
+
+ void mgaDDFlush( GLcontext *ctx )
 {
-   XF86DRIClipRectPtr pbox = mmesa->pClipRects;
+   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
+
+
+   FLUSH_BATCH( mmesa );
+
+   /* This may be called redundantly - dispatch_age may trail what
+    * has actually been sent and processed by the hardware.
+    */
+   if (GET_DISPATCH_AGE( mmesa ) < mmesa->sarea->last_enqueue) {
+      LOCK_HARDWARE( mmesa );
+      if (0) fprintf(stderr, "mgaDDFlush %d %d\n", GET_DISPATCH_AGE( mmesa ),  mmesa->sarea->last_enqueue);
+      mgaUpdateLock( mmesa, DRM_LOCK_FLUSH );
+      UNLOCK_HARDWARE( mmesa );
+   }
+}
+
+
+
+void mgaFireEltsLocked( mgaContextPtr mmesa, 
+			GLuint start, 
+			GLuint end,
+			GLuint discard )
+{
+   drm_clip_rect_t *pbox = (drm_clip_rect_t *)mmesa->pClipRects;
    int nbox = mmesa->numClipRects;
+   drmBufPtr buffer = mmesa->elt_buf;
+   drm_mga_indices_t elts;
    int i;
 
-   if (mmesa->dirty)
+
+   if (0) fprintf(stderr, "FireElts %d %d\n", start/4, end/4);
+
+   if (!buffer)
+      return;
+
+   if (mmesa->dirty_cliprects & mmesa->draw_buffer)
+      mgaUpdateRects( mmesa, mmesa->draw_buffer );
+
+   if (mmesa->dirty & ~MGA_UPLOAD_CLIPRECTS) 
       mgaEmitHwStateLocked( mmesa );
 
-   for (i = 0 ; i < nbox ; )
+   /* FIXME: Workaround bug in kernel module.
+    */
+   mmesa->sarea->dirty |= MGA_UPLOAD_CTX; 
+      
+   elts.idx = buffer->idx;
+   elts.start = start;
+   elts.end = end;
+   elts.discard = 0;
+
+   if (!nbox)
+      elts.end = start;
+
+   if (nbox >= MGA_NR_SAREA_CLIPRECTS)
+      mmesa->dirty |= MGA_UPLOAD_CLIPRECTS;
+
+   if (elts.end == start || !(mmesa->dirty & MGA_UPLOAD_CLIPRECTS)) 
    {
-      int nr = MIN2(i + MGA_NR_SAREA_CLIPRECTS, nbox);
-      XF86DRIClipRectRec *b = (XF86DRIClipRectRec *)mmesa->sarea->boxes;
-      mmesa->sarea->nbox = nr - i;
+      if (nbox == 1) 
+	 mmesa->sarea->nbox = 0;
+      else
+	 mmesa->sarea->nbox = nbox;
+
+      if (0)
+	 fprintf(stderr, "Firing elts -- case a nbox %d\n", nbox);
+
+      elts.discard = discard;
+      ioctl(mmesa->driFd, DRM_IOCTL_MGA_INDICES, &elts);
+      age_mmesa(mmesa, mmesa->sarea->last_enqueue);
+   } 
+   else 
+   {      
+      for (i = 0 ; i < nbox ; )
+      {
+	 int nr = MIN2(i + MGA_NR_SAREA_CLIPRECTS, nbox);
+	 drm_clip_rect_t *b = mmesa->sarea->boxes;
+
+	 if (mmesa->scissor) {
+	    mmesa->sarea->nbox = 0;	 
+
+	    for ( ; i < nr ; i++) {	       
+	       *b = pbox[i];
+	       if (intersect_rect(b, b, &mmesa->scissor_rect)) {
+		  mmesa->sarea->nbox++;
+		  b++;
+	       } 
+	    }
+
+	    /* Culled?
+	     */
+	    if (!mmesa->sarea->nbox) {
+	       if (nr < nbox) continue;
+	       elts.end = start;
+	    }
+	 } else {
+	    mmesa->sarea->nbox = nr - i;
+	    for ( ; i < nr ; i++) 
+	       *b++ = pbox[i];
+	 }
 
-      for ( ; i < nr ; i++) 
-	 *b++ = pbox[i];
-	 
-      mga_vertex_dma_ioctl( mmesa );
+	 /* Potentially finished with the buffer?
+	  */
+	 if (nr == nbox) 
+	    elts.discard = discard;
 
-      break;			/* fix dma multiple dispatch */
+	 if (0)
+	    fprintf(stderr, "Firing elts -- case b nbox %d\n", nbox);
+
+	 mmesa->sarea->dirty |= MGA_UPLOAD_CLIPRECTS;
+	 ioctl(mmesa->driFd, DRM_IOCTL_MGA_INDICES, &elts);
+	 age_mmesa(mmesa, mmesa->sarea->last_enqueue);
+      }
    }
 
-   mga_get_buffer_ioctl( mmesa );
+   mmesa->dirty &= ~MGA_UPLOAD_CLIPRECTS;
+}
+
+void mgaGetEltBufLocked( mgaContextPtr mmesa )
+{
+   mmesa->elt_buf = mga_get_buffer_ioctl( mmesa );
 }
 
+void mgaReleaseBufLocked( mgaContextPtr mmesa, drmBufPtr buffer )
+{
+   drm_mga_vertex_t vertex;
+
+   if (!buffer) return;
+   
+   vertex.idx = buffer->idx;
+   vertex.used = 0;
+   vertex.discard = 1;
+   ioctl(mmesa->driFd, DRM_IOCTL_MGA_VERTEX, &vertex);
+}
 
 
 void mgaDDInitIoctlFuncs( GLcontext *ctx )
 {
    ctx->Driver.Clear = mgaClear;
-   ctx->Driver.Flush = 0;
+   ctx->Driver.Flush = mgaDDFlush;
    ctx->Driver.Finish = mgaDDFinish;
 }
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgaioctl.h b/xc/lib/GL/mesa/src/drv/mga/mgaioctl.h
index 22ca26e17..900e4a4f8 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgaioctl.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgaioctl.h
@@ -9,9 +9,37 @@ GLbitfield mgaClear( GLcontext *ctx, GLbitfield mask, GLboolean all,
 
 void mgaSwapBuffers( mgaContextPtr mmesa ); 
 
+
+
+mgaUI32 *mgaAllocVertexDwords( mgaContextPtr mmesa, int dwords );
+
+
+void mgaGetILoadBufferLocked( mgaContextPtr mmesa );
+
+
+void mgaFireILoadLocked( mgaContextPtr mmesa, 
+			 GLuint offset, GLuint length );
+
+void mgaWaitAgeLocked( mgaContextPtr mmesa, int age );
+void mgaWaitAge( mgaContextPtr mmesa, int age );
+int mgaUpdateLock( mgaContextPtr mmesa, drmLockFlags flags );
+
 void mgaFlushVertices( mgaContextPtr mmesa ); 
 void mgaFlushVerticesLocked( mgaContextPtr mmesa );
 
+
+void mgaFireEltsLocked( mgaContextPtr mmesa, 
+			GLuint start, 
+			GLuint end,
+			GLuint discard );
+
+void mgaGetEltBufLocked( mgaContextPtr mmesa );
+void mgaReleaseBufLocked( mgaContextPtr mmesa, drmBufPtr buffer );
+
+void mgaFlushEltsLocked( mgaContextPtr mmesa );
+void mgaFlushElts( mgaContextPtr mmesa ) ;
+
+
 /* upload texture
  */
 
@@ -19,5 +47,12 @@ void mgaDDFinish( GLcontext *ctx );
 
 void mgaDDInitIoctlFuncs( GLcontext *ctx );
  
+#define FLUSH_BATCH(mmesa) do {						\
+        if (MGA_DEBUG&DEBUG_VERBOSE_IOCTL)  				\
+              fprintf(stderr, "FLUSH_BATCH in %s\n", __FUNCTION__);	\
+	if (mmesa->vertex_dma_buffer) mgaFlushVertices(mmesa);		\
+	else if (mmesa->next_elt != mmesa->first_elt) mgaFlushElts(mmesa);	        	\
+} while (0)
+
 
 #endif
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgalib.h b/xc/lib/GL/mesa/src/drv/mga/mgalib.h
index 3c4cdbbd8..e00525f7b 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgalib.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgalib.h
@@ -101,6 +101,21 @@ typedef void (*mga_interp_func)( GLfloat t,
 #define MGA_BLEND_ENV_COLOR 0x1
 #define MGA_BLEND_MULTITEX  0x2
 
+struct mga_elt_tab {
+   void (*emit_unclipped_verts)( struct vertex_buffer *VB );
+
+   void (*build_tri_verts)( mgaContextPtr mmesa, 
+			    struct vertex_buffer *VB, 
+			    GLfloat *O, GLuint *elt );
+
+   void (*interp)( GLfloat t, GLfloat *O, 
+		   const GLfloat *I, const GLfloat *J );
+
+   void (*project_and_emit_verts)( mgaContextPtr mmesa,
+				   const GLfloat *verts,
+				   GLuint *elts,
+				   int nr );
+};
 
 struct mga_context_t {
 
@@ -177,7 +192,7 @@ struct mga_context_t {
     */
    int dirty_cliprects;		/* which sets of cliprects are uptodate? */
    int draw_buffer;		/* which buffer are we rendering to */
-   int drawOffset;		/* draw buffer address in  space */
+   unsigned int drawOffset;		/* draw buffer address in  space */
    int read_buffer;	
    int readOffset;	
    int drawX, drawY;		/* origin of drawable in draw buffer */
@@ -192,8 +207,8 @@ struct mga_context_t {
    /* Texture aging and DMA based aging.
     */
    unsigned int texAge[MGA_NR_TEX_HEAPS];/* texture LRU age  */
-   int dirtyAge;		/* buffer age for synchronization */
-   int lastSwap;		/* throttling runaway apps */
+   unsigned int dirtyAge;		/* buffer age for synchronization */
+   unsigned int lastSwap;		/* throttling runaway apps */
 
 
 
@@ -208,6 +223,17 @@ struct mga_context_t {
    __DRIscreenPrivate *driScreen;
    mgaScreenPrivate *mgaScreen; 
    drm_mga_sarea_t *sarea;
+
+
+   /* New setupdma path
+    */
+   drmBufPtr elt_buf, retained_buf;
+   GLuint *first_elt, *next_elt;
+   GLfloat *next_vert;
+   GLuint next_vert_phys;
+   GLuint first_vert_phys;
+   struct mga_elt_tab *elt_tab;
+   GLfloat device_matrix[16];
 };
 
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgapipeline.h b/xc/lib/GL/mesa/src/drv/mga/mgapipeline.h
index d42abddfe..dc0179227 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgapipeline.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgapipeline.h
@@ -12,4 +12,7 @@ extern GLboolean mgaDDBuildPrecalcPipeline( GLcontext *ctx );
 extern void mgaDDFastPath( struct vertex_buffer *VB );
 extern void mgaDDFastPathInit( void );
 
+extern void mgaDDEltPath( struct vertex_buffer *VB );
+extern void mgaDDEltPathInit( void );
+
 #endif
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgarender.c b/xc/lib/GL/mesa/src/drv/mga/mgarender.c
deleted file mode 100644
index 6f406db2d..000000000
--- a/xc/lib/GL/mesa/src/drv/mga/mgarender.c
+++ /dev/null
@@ -1,334 +0,0 @@
-#include <stdio.h>
-#include "mgadd.h"
-#include "mgavb.h"
-#include "mgadma.h"
-#include "mgalib.h"
-#include "mgatris.h"
-#include "mgastate.h"
-#include "xsmesaP.h"
-#include "enums.h"
-
-#define POINT(x) mga_draw_point(&gWin[x], psize)
-#define LINE(x,y) mga_draw_line(&gWin[x], &gWin[y], lwidth)
-#define TRI(x,y,z) mga_draw_triangle(&gWin[x], &gWin[y], &gWin[z])
-
-
-/* Direct, and no clipping required.  I haven't written the clip funcs
- * yet, so this is only useful for the fast path, which does its own
- * clipping.
- */
-#define RENDER_POINTS( start, count )		\
-do {						\
-   GLuint e;					\
-   for(e=start;e<=count;e++)			\
-     POINT(elt[e]);				\
-} while (0)
-
-#define RENDER_LINE( i1, i )			\
-do {						\
-   GLuint e1 = elt[i1], e = elt[i];		\
-   LINE( e1, e );				\
-} while (0)
-
-
-#define RENDER_TRI( i2, i1, i, pv, parity)		\
-do {							\
-{  GLuint e2 = elt[i2], e1 = elt[i1], e = elt[i];	\
-  if (parity) {GLuint tmp = e2; e2 = e1; e1 = tmp;}	\
-  TRI(e2, e1, e);					\
-}} while (0)
-
-
-#define RENDER_QUAD( i3, i2, i1, i, pv)				\
-do {								\
-  GLuint e3 = elt[i3], e2 = elt[i2], e1 = elt[i1], e = elt[i];	\
-  TRI(e3, e2, e);						\
-  TRI(e2, e1, e);						\
-} while (0)
-
-#define LOCAL_VARS					\
-   mgaVertexPtr gWin = MGA_DRIVER_DATA(VB)->verts;	\
-   const GLuint *elt = VB->EltPtr->data;		\
-   const GLfloat lwidth = VB->ctx->Line.Width;		\
-   const GLfloat psize = VB->ctx->Point.Size;		\
-   GLcontext *ctx = VB->ctx; \
-   (void) lwidth; (void)psize; (void)ctx; (void) gWin;
-
-
-#define TAG(x) x##_mga_smooth_indirect
-#include "render_tmp.h"
-
-
-
-
-#define RENDER_POINTS( start, count )		\
-do {						\
-   GLuint e;					\
-   for(e=start;e<=count;e++)			\
-     POINT(e);					\
-} while (0)
-
-#define RENDER_LINE( i1, i )			\
-do {						\
-   LINE( i1, i );				\
-} while (0)
-
-
-#define RENDER_TRI( i2, i1, i, pv, parity)		\
-do {							\
-  GLuint e2 = i2, e1 = i1;				\
-  if (parity) {GLuint tmp = e2; e2 = e1; e1 = tmp;}	\
-  TRI(e2, e1, i);					\
-} while (0)
-
-
-#define RENDER_QUAD( i3, i2, i1, i, pv)		\
-do {						\
-  TRI(i3, i2, i);				\
-  TRI(i2, i1, i);				\
-} while (0)
-
-#define LOCAL_VARS					\
-   mgaVertexPtr gWin = MGA_DRIVER_DATA(VB)->verts;	\
-   const GLfloat lwidth = VB->ctx->Line.Width;		\
-   const GLfloat psize = VB->ctx->Point.Size;		\
-   GLcontext *ctx = VB->ctx; \
-   (void) lwidth; (void)psize; (void)ctx; (void) gWin;
-
-#define TAG(x) x##_mga_smooth_direct
-#include "render_tmp.h"
-
-
-
-
-
-/* Vertex array rendering via. the SetupDma function - no clipping required. 
- */
-#define RENDER_POINTS( start, count )		\
-do {						\
-    FatalError("Dead code in mgarender.c\n");	\
-} while (0)
-
-#define RENDER_LINE( i1, i )			\
-do {						\
-    FatalError("Dead code in mgarender.c\n");	\
-} while (0)
-
-
-#define RENDER_TRI( i2, i1, i, pv, parity)		\
-do {							\
-   GLuint e2 = elt[i2], e1 = elt[i1], e = elt[i];	\
-   if (/*parity*/0) {GLuint tmp = e2; e2 = e1; e1 = tmp;}	\
-   TRI(e2, e1, e);		\
-} while (0)
-
-#define RENDER_QUAD( i3, i2, i1, i, pv)				\
-do {								\
-  GLuint e3 = elt[i3], e2 = elt[i2], e1 = elt[i1], e = elt[i];	\
-  TRI(e3, e2, e);			\
-  TRI(e2, e1, e);			\
-} while (0)
-
-#define LOCAL_VARS					\
-   mgaVertexBufferPtr mgaVB = MGA_DRIVER_DATA(VB);	\
-   mgaUI32 phys = mgaVB->vert_phys_start;		\
-   const GLuint *elt = VB->EltPtr->data;		\
-   (void) phys; (void) elt;
-
-#define PRESERVE_VB_DEFS
-#undef TRI
-#define VERT_ADDR_10( phys, elt ) ( phys + elt * 48 )
-#define TRI( ee0, ee1, ee2 )				\
-{							\
-   mgaVB->elt_buf[0] = VERT_ADDR_10(phys, ee0);		\
-   mgaVB->elt_buf[1] = VERT_ADDR_10(phys, ee1);		\
-   mgaVB->elt_buf[2] = VERT_ADDR_10(phys, ee2);		\
-   mgaVB->elt_buf += 3;					\
-}
-#define TAG(x) x##_mga_elt_10
-#include "render_tmp.h"
-
-
-#define TAG(x) x##_mga_elt_8
-#undef TRI
-#define VERT_ADDR_8( phys, elt ) ( phys + elt * 32 )
-#define TRI( ee0, ee1, ee2 )			\
-{						\
-   mgaVB->elt_buf[0] = VERT_ADDR_8(phys, ee0);	\
-   mgaVB->elt_buf[1] = VERT_ADDR_8(phys, ee1);	\
-   mgaVB->elt_buf[2] = VERT_ADDR_8(phys, ee2);	\
-   mgaVB->elt_buf += 3;				\
-}
-#include "render_tmp.h"
-
-
-
-
-/* Currently only used on the fast path - need a set of render funcs
- * for clipped primitives.
- */
-void mgaDDRenderElementsDirect( struct vertex_buffer *VB )
-{
-   mgaVertexBufferPtr mgaVB = MGA_DRIVER_DATA(VB);
-   GLcontext *ctx = VB->ctx;
-   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
-   GLenum prim = ctx->CVA.elt_mode;
-   GLuint nr = VB->EltPtr->count;
-   render_func func = render_tab_mga_smooth_indirect[prim];
-   GLuint p = 0;
-   mgaUI32 *start = mgaVB->elt_buf;
-   
-   struct vertex_buffer *saved_vb = ctx->VB;
-
-   if (mmesa->new_state)
-      mgaDDUpdateHwState( ctx );
-
-   /* Are we using Setup DMA?
-    */
-   if (start) {
-      switch (MGA_CONTEXT(ctx)->vertsize) {
-      case 10:
-	 func = render_tab_mga_elt_10[prim];
-	 break;
-      case 8:
-	 func = render_tab_mga_elt_8[prim];
-	 break;
-      default:
-      }
-   }
-
-   ctx->VB = VB;
-
-   do {
-      func( VB, 0, nr, 0 );
-   } while (ctx->Driver.MultipassFunc &&
-	    ctx->Driver.MultipassFunc( VB, ++p ));
-
-   ctx->VB = saved_vb;
-
-
-   if (start && nr) {
-      if (0)
-	 fprintf(stderr, "%d/%d elts\n", nr, mgaVB->elt_buf-start);
-      mgaSetupDma( start, mgaVB->elt_buf - start );
-      mgaVB->elt_buf = mgaVB->vert_buf = 0;
-   }
-}
-
-
-void mgaDDRenderElementsImmediate( struct vertex_buffer *VB )
-{
-   GLcontext *ctx = VB->ctx;
-   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
-   GLuint nr = VB->EltPtr->count;
-   render_func *tab = render_tab_mga_smooth_indirect;
-   GLuint parity = VB->Parity;
-   GLuint p = 0;
-   GLuint i, next;
-
-   if (mmesa->new_state)
-      mgaDDUpdateHwState( ctx );
-
-   do {
-      for ( i= VB->CopyStart ; i < nr ; parity = 0, i = next ) 
-      {
-	 GLenum prim = VB->Primitive[i];
-	 next = VB->NextPrimitive[i];
-	 tab[prim]( VB, i, next, parity );
-      }
-   } while (ctx->Driver.MultipassFunc &&
-	    ctx->Driver.MultipassFunc( VB, ++p ));
-}
-
-
-void mgaDDRenderDirect( struct vertex_buffer *VB )
-{
-   GLcontext *ctx = VB->ctx;
-   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
-   GLuint nr = VB->Count;
-   render_func *tab = render_tab_mga_smooth_direct;
-   GLuint parity = VB->Parity;
-   GLuint p = 0;
-   GLuint i, next;
-
-   if (mmesa->new_state)
-      mgaDDUpdateHwState( ctx );
-
-   do {
-      for ( i= VB->CopyStart ; i < nr ; parity = 0, i = next ) 
-      {
-	 GLenum prim = VB->Primitive[i];
-	 next = VB->NextPrimitive[i];
-	 tab[prim]( VB, i, next, parity );
-      }
-   } while (ctx->Driver.MultipassFunc &&
-	    ctx->Driver.MultipassFunc( VB, ++p ));
-}
-
-
-static void optimized_render_vb_triangle_mga_smooth_indirect(struct vertex_buffer *VB,
-							    GLuint start,
-							    GLuint count,
-							    GLuint parity)
-{
-   mgaVertexPtr gWin = MGA_DRIVER_DATA(VB)->verts;
-   const GLuint *elt = VB->EltPtr->data;
-   GLcontext *ctx = VB->ctx;
-   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
-
-   mgaUI32 vertsize = mmesa->vertsize;
-
-   /* We know how many dwords we need so we can allocate all of it
-    * in one call */
-   mgaUI32 *wv = mgaAllocSecondaryBuffer( count * vertsize );
-
-   int vertex;
-
-   (void)ctx; (void) gWin; (void) parity;
-
-
-   for(vertex=start+2;vertex<count;vertex+=3){
-      GLuint e2=elt[vertex-2],e1=elt[vertex-1],e=elt[vertex];
-      mgaUI32 *src;
-
-      int j;
-
-      if(parity){
-         GLuint tmp=e2;
-         e2=e1;
-	 e1=tmp;
-      }
-
-      src=&gWin[e2].ui[0];
-      
-      for(j=vertsize;j;j--){
-	 *wv++=*src++;
-      }
-      
-      src=&gWin[e1].ui[0];
-      for(j=vertsize;j;j--){
-	 *wv++=*src++;
-      }
-      
-      src=&gWin[e].ui[0];
-      for(j=vertsize;j;j--){
-	 *wv++=*src++;
-      }
-   }
-   
-}
-
-
-void mgaDDRenderInit()
-{
-
-   render_init_mga_smooth_indirect();
-   render_init_mga_smooth_direct();
-   render_init_mga_elt_10();
-   render_init_mga_elt_8();
-   
-   render_tab_mga_smooth_indirect[GL_TRIANGLES] =
-      optimized_render_vb_triangle_mga_smooth_indirect; 
-}
-
-
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgarender.h b/xc/lib/GL/mesa/src/drv/mga/mgarender.h
deleted file mode 100644
index c772ac459..000000000
--- a/xc/lib/GL/mesa/src/drv/mga/mgarender.h
+++ /dev/null
@@ -1,10 +0,0 @@
-
-#ifndef _MGA_RENDER_H
-#define _MGA_RENDER_H
-
-extern void mgaDDRenderElementsDirect( struct vertex_buffer *VB );
-extern void mgaDDRenderElementsImmediate( struct vertex_buffer *VB );
-extern void mgaDDRenderDirect( struct vertex_buffer *VB );
-extern void mgaDDRenderInit();
-
-#endif
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgastate.c b/xc/lib/GL/mesa/src/drv/mga/mgastate.c
index ff9063e0b..7a3b82930 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgastate.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgastate.c
@@ -16,7 +16,7 @@
 #include "mgavb.h"
 #include "mgatris.h"
 #include "mgaregs.h"
-#include "mga_drm_public.h"
+#include "mgabuffers.h"
 
 static void mgaUpdateZMode(const GLcontext *ctx)
 {
@@ -60,17 +60,20 @@ static void mgaUpdateZMode(const GLcontext *ctx)
 
 static void mgaDDAlphaFunc(GLcontext *ctx, GLenum func, GLclampf ref)
 {
+   FLUSH_BATCH( MGA_CONTEXT(ctx) );
    MGA_CONTEXT(ctx)->new_state |= MGA_NEW_ALPHA;
 }
 
 
 static void mgaDDBlendEquation(GLcontext *ctx, GLenum mode) 
 {
+   FLUSH_BATCH( MGA_CONTEXT(ctx) );
    MGA_CONTEXT(ctx)->new_state |= MGA_NEW_ALPHA;
 }
 
 static void mgaDDBlendFunc(GLcontext *ctx, GLenum sfactor, GLenum dfactor)
 {
+   FLUSH_BATCH( MGA_CONTEXT(ctx) );
    MGA_CONTEXT(ctx)->new_state |= MGA_NEW_ALPHA;
 }
 
@@ -78,6 +81,7 @@ static void mgaDDBlendFuncSeparate( GLcontext *ctx, GLenum sfactorRGB,
 				    GLenum dfactorRGB, GLenum sfactorA,
 				    GLenum dfactorA )
 {
+   FLUSH_BATCH( MGA_CONTEXT(ctx) );
    MGA_CONTEXT(ctx)->new_state |= MGA_NEW_ALPHA;
 }
 
@@ -87,6 +91,7 @@ static void mgaDDLightModelfv(GLcontext *ctx, GLenum pname,
 			      const GLfloat *param)
 {
    if (pname == GL_LIGHT_MODEL_COLOR_CONTROL) {
+      FLUSH_BATCH( MGA_CONTEXT(ctx) );
       MGA_CONTEXT(ctx)->new_state |= MGA_NEW_TEXTURE;
    }
 }
@@ -95,6 +100,7 @@ static void mgaDDLightModelfv(GLcontext *ctx, GLenum pname,
 static void mgaDDShadeModel(GLcontext *ctx, GLenum mode)
 {
    if (1) {
+      FLUSH_BATCH( MGA_CONTEXT(ctx) );
       MGA_CONTEXT(ctx)->new_state |= MGA_NEW_TEXTURE; 
       mgaMsg(8, "mgaDDShadeModel: %x\n", mode);
    }
@@ -103,11 +109,13 @@ static void mgaDDShadeModel(GLcontext *ctx, GLenum mode)
 
 static void mgaDDDepthFunc(GLcontext *ctx, GLenum func)
 {
+   FLUSH_BATCH( MGA_CONTEXT(ctx) );
    MGA_CONTEXT(ctx)->new_state |= MGA_NEW_DEPTH;
 }
 
 static void mgaDDDepthMask(GLcontext *ctx, GLboolean flag)
 {
+   FLUSH_BATCH( MGA_CONTEXT(ctx) );
    MGA_CONTEXT(ctx)->new_state |= MGA_NEW_DEPTH;
 }
 
@@ -134,6 +142,7 @@ static void mgaUpdateFogAttrib( GLcontext *ctx )
 
 static void mgaDDFogfv(GLcontext *ctx, GLenum pname, const GLfloat *param)
 {
+   FLUSH_BATCH( MGA_CONTEXT(ctx) );
    MGA_CONTEXT(ctx)->new_state |= MGA_NEW_FOG;
 }
 
@@ -148,10 +157,11 @@ static void mgaDDFogfv(GLcontext *ctx, GLenum pname, const GLfloat *param)
 static void mgaUpdateAlphaMode(GLcontext *ctx)
 {
    mgaContextPtr mmesa = MGA_CONTEXT( ctx );
+   mgaScreenPrivate *mgaScreen = mmesa->mgaScreen;
    int a = 0;
 
    /* determine source of alpha for blending and testing */
-   if ( !ctx->Texture.Enabled || (mmesa->Fallback & MGA_FALLBACK_TEXTURE)) 
+   if ( !ctx->Texture.Enabled ) 
       a |= AC_alphasel_diffused;
    else {
       switch (ctx->Texture.Unit[0].EnvMode) {
@@ -168,7 +178,7 @@ static void mgaUpdateAlphaMode(GLcontext *ctx)
    }
 
 
-   /* alpha test control - disabled by default.
+   /* alpha test control.
     */
    if (ctx->Color.AlphaEnabled) {
       GLubyte ref = ctx->Color.AlphaRef;
@@ -219,13 +229,13 @@ static void mgaUpdateAlphaMode(GLcontext *ctx)
       case GL_ONE_MINUS_SRC_ALPHA:
 	 a |= AC_src_om_src_alpha; break;
       case GL_DST_ALPHA:
-	 if (0) /*(mgaScreen->Attrib & MGA_PF_HASALPHA)*/
+	 if (mgaScreen->Attrib & MGA_PF_HASALPHA)
 	    a |= AC_src_dst_alpha;
 	 else
 	    a |= AC_src_one;
 	 break;
       case GL_ONE_MINUS_DST_ALPHA:
-	 if (0) /*(mgaScreen->Attrib & MGA_PF_HASALPHA)*/
+	 if (mgaScreen->Attrib & MGA_PF_HASALPHA)
 	    a |= AC_src_om_dst_alpha; 
 	 else 
 	    a |= AC_src_zero;
@@ -250,13 +260,13 @@ static void mgaUpdateAlphaMode(GLcontext *ctx)
       case GL_ONE_MINUS_SRC_COLOR:
 	 a |= AC_dst_om_src_color; break;
       case GL_DST_ALPHA:
-	 if (0) /*(mgaDB->Attrib & MGA_PF_HASALPHA)*/
+	 if (mgaScreen->Attrib & MGA_PF_HASALPHA)
 	    a |= AC_dst_dst_alpha;
 	 else
 	    a |= AC_dst_one;
 	 break;
       case GL_ONE_MINUS_DST_ALPHA:
-	 if (0) /*(mgaScreen->Attrib & MGA_PF_HASALPHA)*/
+	 if (mgaScreen->Attrib & MGA_PF_HASALPHA)
 	    a |= AC_dst_om_dst_alpha;
 	 else
 	    a |= AC_dst_zero;
@@ -282,50 +292,44 @@ static void mgaUpdateAlphaMode(GLcontext *ctx)
  * Hardware clipping
  */
 
-static void mgaUpdateClipping(const GLcontext *ctx)
+void mgaUpdateClipping(const GLcontext *ctx)
 {
-#if 0
-   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
-   __DRIdrawablePrivate *dPriv = mmesa->driDrawable;	
-   int x1,x2,y1,y2;
-
-   if ( ctx->Scissor.Enabled) {
-      x1 = ctx->Scissor.X;
-      x2 = ctx->Scissor.X + ctx->Scissor.Width - 1;
-      y1 = dPriv->Height - ctx->Scissor.Y - ctx->Scissor.Height;
-      y2 = dPriv->Height - ctx->Scissor.Y - 1;
-   } else {
-      x1 = 0;
-      y1 = 0;
-      x2 = mgaDB->Width-1;
-      y2 = mgaDB->Height-1;
-   }
-  
-   if (x1 < 0) x1 = 0;
-   if (y1 < 0) y1 = 0;
-   if (x2 >= mgaDB->Width) x2 = mgaDB->Width-1;
-   if (y2 >= mgaDB->Height) y2 = mgaDB->Height-1;
-
-   if (x1 > x2 || y1 > y2) {
-      x1 = 0; x2 = 0;
-      y2 = 0; y1 = 1;
-   }
-
-
-   mmesa->Setup[MGA_CTXREG_CXBNDRY] = (MGA_FIELD(CXB_cxright,x2) | 
-				      MGA_FIELD(CXB_cxleft,x1));
-   mmesa->Setup[MGA_CTXREG_YTOP] = y1*mgaDB->Pitch;
-   mmesa->Setup[MGA_CTXREG_YBOT] = y2*mgaDB->Pitch;
-
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
 
-   mmesa->dirty |= MGA_UPLOAD_CTX;
-#endif
+   if (mmesa->driDrawable)
+   {
+      int x1 = mmesa->driDrawable->x + ctx->Scissor.X;
+      int y1 = mmesa->driDrawable->y + mmesa->driDrawable->h - (ctx->Scissor.Y+
+								ctx->Scissor.Height);
+      int x2 = mmesa->driDrawable->x + ctx->Scissor.X+ctx->Scissor.Width;
+      int y2 = mmesa->driDrawable->y + mmesa->driDrawable->h - ctx->Scissor.Y;
+
+      if (x1 < 0) x1 = 0;
+      if (y1 < 0) y1 = 0;
+      if (x2 < 0) x2 = 0;
+      if (y2 < 0) y2 = 0;
+
+      mmesa->scissor_rect.x1 = x1;
+      mmesa->scissor_rect.y1 = y1;
+      mmesa->scissor_rect.x2 = x2;
+      mmesa->scissor_rect.y2 = y2;
+
+      if (MGA_DEBUG&DEBUG_VERBOSE_2D)
+	 fprintf(stderr, "SET SCISSOR %d,%d-%d,%d\n", 
+		 mmesa->scissor_rect.x1,
+		 mmesa->scissor_rect.y1,
+		 mmesa->scissor_rect.x2,
+		 mmesa->scissor_rect.y2);
+
+      mmesa->dirty |= MGA_UPLOAD_CLIPRECTS;
+   }
 }
 
 
 static void mgaDDScissor( GLcontext *ctx, GLint x, GLint y, 
 			  GLsizei w, GLsizei h )
 {
+   FLUSH_BATCH( MGA_CONTEXT(ctx) );
    MGA_CONTEXT(ctx)->new_state |= MGA_NEW_CLIP;
 }
 
@@ -339,35 +343,6 @@ static void mgaDDDither(GLcontext *ctx, GLboolean enable)
 }
 
 
-static GLboolean mgaDDSetBuffer(GLcontext *ctx, GLenum mode )
-{
-   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
-
-   mmesa->Fallback &= ~MGA_FALLBACK_BUFFER;
-
-   if (mode == GL_FRONT_LEFT) 
-   {
-      mmesa->drawOffset = mmesa->mgaScreen->fbOffset;
-      mmesa->Setup[MGA_CTXREG_DSTORG] = mmesa->mgaScreen->fbOffset;
-      mmesa->dirty |= MGA_UPLOAD_CTX;
-      mgaXMesaSetFrontClipRects( mmesa );
-      return GL_TRUE;
-   } 
-   else if (mode == GL_BACK_LEFT) 
-   {
-      mmesa->drawOffset = mmesa->mgaScreen->backOffset;
-      mmesa->Setup[MGA_CTXREG_DSTORG] = mmesa->mgaScreen->backOffset;
-      mmesa->dirty |= MGA_UPLOAD_CTX;
-      mgaXMesaSetBackClipRects( mmesa );
-      return GL_TRUE;
-   }
-   else 
-   {
-      mmesa->Fallback |= MGA_FALLBACK_BUFFER;
-      return GL_FALSE;
-   }
-}
-
 
 
 static void mgaDDSetColor(GLcontext *ctx, 
@@ -376,7 +351,7 @@ static void mgaDDSetColor(GLcontext *ctx,
 {
    mgaContextPtr mmesa = MGA_CONTEXT(ctx);
 
-   mmesa->MonoColor = mgaPackColor( mmesa->mgaScreen->fbFormat,
+   mmesa->MonoColor = mgaPackColor( mmesa->mgaScreen->Attrib,
 				      r, g, b, a );
 }
 
@@ -387,7 +362,7 @@ static void mgaDDClearColor(GLcontext *ctx,
 {
    mgaContextPtr mmesa = MGA_CONTEXT(ctx);
 
-   mmesa->ClearColor = mgaPackColor( mmesa->mgaScreen->fbFormat,
+   mmesa->ClearColor = mgaPackColor( mmesa->mgaScreen->Attrib,
 				      r, g, b, a );
 }
 
@@ -426,6 +401,7 @@ static void mgaUpdateCull( GLcontext *ctx )
 
 static void mgaDDCullFaceFrontFace(GLcontext *ctx, GLenum mode)
 {
+   FLUSH_BATCH( MGA_CONTEXT(ctx) );
    MGA_CONTEXT(ctx)->new_state |= MGA_NEW_CULL;
 }
 
@@ -461,6 +437,7 @@ static GLboolean mgaDDColorMask(GLcontext *ctx,
    }
 
    if (mmesa->Setup[MGA_CTXREG_PLNWT] != mask) {
+      FLUSH_BATCH( MGA_CONTEXT(ctx) );
       mmesa->Setup[MGA_CTXREG_PLNWT] = mask;      
       MGA_CONTEXT(ctx)->new_state |= MGA_NEW_MASK; 
       mmesa->dirty |= MGA_UPLOAD_CTX;
@@ -470,6 +447,83 @@ static GLboolean mgaDDColorMask(GLcontext *ctx,
 }
 
 /* =============================================================
+ * Polygon stipple 
+ * 
+ * The mga supports a subset of possible 4x4 stipples natively, GL
+ * wants 32x32.  Fortunately stipple is usually a repeating pattern.
+ */
+static int mgaStipples[16] = {
+   0xffff,
+   0xa5a5,
+   0x5a5a,
+   0xa0a0,
+   0x5050,
+   0x0a0a,
+   0x0505,
+   0x8020,
+   0x0401,
+   0x1040,
+   0x0208,
+   0x0802,
+   0x4010,
+   0x0104,
+   0x2080,
+   0x0000
+};
+
+static void mgaDDPolygonStipple( GLcontext *ctx, const GLubyte *mask )
+{
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
+   const GLubyte *m = mask;
+   GLubyte p[4];
+   int i,j,k;
+   int active = (ctx->Polygon.StippleFlag && ctx->PB->primitive == GL_POLYGON);
+   GLuint stipple;
+
+   FLUSH_BATCH(mmesa);
+   ctx->Driver.TriangleCaps |= DD_TRI_STIPPLE;
+
+   if (active) {
+      mmesa->dirty |= MGA_UPLOAD_CTX;
+      mmesa->Setup[MGA_CTXREG_DWGCTL] &= ~(0xf<<20);
+   }
+
+   p[0] = mask[0] & 0xf; p[0] |= p[0] << 4;
+   p[1] = mask[4] & 0xf; p[1] |= p[1] << 4;
+   p[2] = mask[8] & 0xf; p[2] |= p[2] << 4;
+   p[3] = mask[12] & 0xf; p[3] |= p[3] << 4;
+
+   for (k = 0 ; k < 8 ; k++)
+      for (j = 0 ; j < 4; j++) 
+	 for (i = 0 ; i < 4 ; i++) 
+	    if (*m++ != p[j]) {
+	       ctx->Driver.TriangleCaps &= ~DD_TRI_STIPPLE;
+	       return;
+	    }
+
+   stipple = ( ((p[0] & 0xf) << 0) |
+	       ((p[1] & 0xf) << 4) |
+	       ((p[2] & 0xf) << 8) |
+	       ((p[3] & 0xf) << 12) );   
+
+   for (i = 0 ; i < 16 ; i++)
+      if (mgaStipples[i] == stipple) {
+	 mmesa->poly_stipple = i<<20;
+	 break;
+      }
+
+   if (i == 16) {
+      ctx->Driver.TriangleCaps &= ~DD_TRI_STIPPLE;
+      return;
+   }
+   
+   if (active) {
+      mmesa->Setup[MGA_CTXREG_DWGCTL] &= ~(0xf<<20);
+      mmesa->Setup[MGA_CTXREG_DWGCTL] |= mmesa->poly_stipple;
+   }
+}
+
+/* =============================================================
  */
 
 
@@ -477,15 +531,16 @@ static GLboolean mgaDDColorMask(GLcontext *ctx,
 
 static void mgaDDPrintDirty( const char *msg, GLuint state )
 {
-   fprintf(stderr, "%s (0x%x): %s%s%s%s%s%s\n",	   
+   fprintf(stderr, "%s (0x%x): %s%s%s%s%s%s%s\n",
 	   msg,
 	   (unsigned int) state,
-	   (state & MGA_REQUIRE_QUIESCENT) ? "req-quiescent, " : "",
+	   (state & MGA_WAIT_AGE) ? "wait-age, " : "",
 	   (state & MGA_UPLOAD_TEX0IMAGE)  ? "upload-tex0-img, " : "",
 	   (state & MGA_UPLOAD_TEX1IMAGE)  ? "upload-tex1-img, " : "",
 	   (state & MGA_UPLOAD_CTX)        ? "upload-ctx, " : "",
 	   (state & MGA_UPLOAD_TEX0)       ? "upload-tex0, " : "",
-	   (state & MGA_UPLOAD_TEX1)       ? "upload-tex1, " : ""
+	   (state & MGA_UPLOAD_TEX1)       ? "upload-tex1, " : "",
+	   (state & MGA_UPLOAD_PIPE)       ? "upload-pipe, " : ""
 	   );
 }
 
@@ -494,7 +549,7 @@ static void mgaDDPrintDirty( const char *msg, GLuint state )
  */
 void mgaEmitHwStateLocked( mgaContextPtr mmesa )
 {
-   if (MGA_DEBUG & MGA_DEBUG_VERBOSE_MSG)
+   if (MGA_DEBUG & DEBUG_VERBOSE_MSG)
       mgaDDPrintDirty( "mgaEmitHwStateLocked", mmesa->dirty );
 
    if ((mmesa->dirty & MGA_UPLOAD_TEX0IMAGE) && mmesa->CurrentTexObj[0])
@@ -503,27 +558,34 @@ void mgaEmitHwStateLocked( mgaContextPtr mmesa )
    if ((mmesa->dirty & MGA_UPLOAD_TEX1IMAGE) && mmesa->CurrentTexObj[1])
       mgaUploadTexImages(mmesa, mmesa->CurrentTexObj[1]);
   
-   if (mmesa->dirty & MGA_UPLOAD_CTX) {
+   if (mmesa->dirty & MGA_UPLOAD_CTX) 
       memcpy( mmesa->sarea->ContextState, 
 	      mmesa->Setup,
 	      sizeof(mmesa->Setup));
 
-      if (mmesa->CurrentTexObj[0])
+   if ((mmesa->dirty & MGA_UPLOAD_TEX0) && mmesa->CurrentTexObj[0])
 	 memcpy(mmesa->sarea->TexState[0],
 		mmesa->CurrentTexObj[0]->Setup,
 		sizeof(mmesa->sarea->TexState[0]));
 
-      if (mmesa->CurrentTexObj[1])
+   if ((mmesa->dirty & MGA_UPLOAD_TEX1) && mmesa->CurrentTexObj[1])
 	 memcpy(mmesa->sarea->TexState[1],
 		mmesa->CurrentTexObj[1]->Setup,
 		sizeof(mmesa->sarea->TexState[1]));
-   }
 
-   if (mmesa->dirty & MGA_UPLOAD_PIPE) 
-      mmesa->sarea->WarpPipe = mmesa->setupindex & MGA_WARP_T2GZSAF;
-      
+   mmesa->sarea->WarpPipe = ((mmesa->setupindex & MGA_WARP_T2GZSAF) | 
+			     MGA_ALPHA_BIT | MGA_SPEC_BIT | MGA_FOG_BIT);
+
+
    mmesa->sarea->dirty |= mmesa->dirty;
-   mmesa->dirty = 0;
+
+#if 0
+   mgaPrintSetupFlags("warp pipe", mmesa->sarea->WarpPipe);
+   fprintf(stderr, "in mgaEmitHwStateLocked: dirty now %x\n",
+	   mmesa->sarea->dirty);
+#endif
+
+   mmesa->dirty &= (MGA_UPLOAD_CLIPRECTS|MGA_WAIT_AGE);
 }
 
 
@@ -533,29 +595,50 @@ void mgaEmitHwStateLocked( mgaContextPtr mmesa )
 
 static void mgaDDEnable(GLcontext *ctx, GLenum cap, GLboolean state)
 {
+   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
+
    switch(cap) {
    case GL_ALPHA_TEST:
-      MGA_CONTEXT(ctx)->new_state |= MGA_NEW_ALPHA;
+      FLUSH_BATCH( mmesa );
+      mmesa->new_state |= MGA_NEW_ALPHA;
       break;
    case GL_BLEND:
-      MGA_CONTEXT(ctx)->new_state |= MGA_NEW_ALPHA;
+      FLUSH_BATCH( mmesa );
+      mmesa->new_state |= MGA_NEW_ALPHA;
       break;
    case GL_DEPTH_TEST:
-      MGA_CONTEXT(ctx)->new_state |= MGA_NEW_DEPTH;
+      FLUSH_BATCH( mmesa );
+      mmesa->new_state |= MGA_NEW_DEPTH;
       break;
    case GL_SCISSOR_TEST:
-      MGA_CONTEXT(ctx)->new_state |= MGA_NEW_CLIP;
+      FLUSH_BATCH( mmesa );
+      mmesa->scissor = state;
+      mmesa->new_state |= MGA_NEW_CLIP;
       break;
    case GL_FOG:
-      MGA_CONTEXT(ctx)->new_state |= MGA_NEW_FOG;
+      FLUSH_BATCH( mmesa );
+      mmesa->new_state |= MGA_NEW_FOG;
       break;
    case GL_CULL_FACE:
-      MGA_CONTEXT(ctx)->new_state |= MGA_NEW_CULL;
+      FLUSH_BATCH( mmesa );
+      mmesa->new_state |= MGA_NEW_CULL;
       break;
    case GL_TEXTURE_1D:
    case GL_TEXTURE_2D:
    case GL_TEXTURE_3D:
-      MGA_CONTEXT(ctx)->new_state |= MGA_NEW_TEXTURE;
+      FLUSH_BATCH( mmesa );
+      mmesa->new_state |= MGA_NEW_TEXTURE;
+      break;
+   case GL_POLYGON_STIPPLE:
+      if ((ctx->Driver.TriangleCaps & DD_TRI_STIPPLE) &&
+	  ctx->PB->primitive == GL_POLYGON) 
+      {
+	 FLUSH_BATCH(mmesa);
+	 mmesa->dirty |= MGA_UPLOAD_CTX;
+	 mmesa->Setup[MGA_CTXREG_DWGCTL] &= ~(0xf<<20);
+	 if (state)
+	    mmesa->Setup[MGA_CTXREG_DWGCTL] |= mmesa->poly_stipple;
+      }
       break;
    default:
       ; 
@@ -581,6 +664,7 @@ static void mgaWarpUpdateState( GLcontext *ctx )
    {
       mmesa->warp_pipe = index;
       mmesa->new_state |= MGA_NEW_WARP;
+      mmesa->dirty |= MGA_UPLOAD_PIPE;
    }
 }
 
@@ -611,21 +695,15 @@ void mgaDDUpdateHwState( GLcontext *ctx )
 
    if (new_state) 
    {
-      mmesa->new_state = 0;
+      FLUSH_BATCH( mmesa );
 
-      /* Emit any vertices for the current state.  This will also
-       * push the current state into the sarea.
-       */
-/*        mgaFlushVertices( mmesa ); */
+      mmesa->new_state = 0;
 
       if (MESA_VERBOSE&VERBOSE_DRIVER)
 	 mgaDDPrintState("UpdateHwState", new_state);
 
       if (new_state & MGA_NEW_DEPTH)
-      {
 	 mgaUpdateZMode(ctx);
-	 mgaDDInitDepthFuncs(ctx);
-      }
 
       if (new_state & MGA_NEW_ALPHA)
 	 mgaUpdateAlphaMode(ctx);
@@ -641,8 +719,6 @@ void mgaDDUpdateHwState( GLcontext *ctx )
 
       if (new_state & (MGA_NEW_WARP|MGA_NEW_TEXTURE))
 	 mgaUpdateTextureState(ctx);
-
-      mmesa->new_state = 0; /* tex uploads scribble newstate */
    }
 }
 
@@ -654,8 +730,17 @@ void mgaDDUpdateHwState( GLcontext *ctx )
 
 void mgaDDReducedPrimitiveChange( GLcontext *ctx, GLenum prim )
 {
-   mgaFlushVertices( MGA_CONTEXT(ctx) );
-   mgaUpdateCull(ctx);
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
+
+   FLUSH_BATCH( mmesa );
+   mgaUpdateCull(ctx);   
+
+   if (ctx->Polygon.StippleFlag && (ctx->Driver.TriangleCaps & DD_TRI_STIPPLE))
+   {
+      mmesa->Setup[MGA_CTXREG_DWGCTL] &= ~(0xf<<20);
+      if (ctx->PB->primitive == GL_POLYGON)
+	 mmesa->Setup[MGA_CTXREG_DWGCTL] |= mmesa->poly_stipple;
+   }
 }
 
 
@@ -677,10 +762,16 @@ void mgaDDUpdateState( GLcontext *ctx )
 
    /* Have to do this here to detect texture fallbacks in time:
     */
-   if (MGA_CONTEXT(ctx)->new_state & MGA_NEW_TEXTURE)
+   if (mmesa->new_state & MGA_NEW_TEXTURE)
       mgaDDUpdateHwState( ctx );
 
-   if (!mmesa->Fallback || mgaglx.noFallback) {
+   if (0) fprintf(stderr, "fallback %x indirect %x\n", mmesa->Fallback, 
+		  mmesa->IndirectTriangles);
+   
+   if (!mmesa->Fallback) {
+      ctx->IndirectTriangles &= ~DD_SW_RASTERIZE;
+      ctx->IndirectTriangles |= mmesa->IndirectTriangles;
+
       ctx->Driver.PointsFunc=mmesa->PointsFunc;
       ctx->Driver.LineFunc=mmesa->LineFunc;
       ctx->Driver.TriangleFunc=mmesa->TriangleFunc;
@@ -689,6 +780,60 @@ void mgaDDUpdateState( GLcontext *ctx )
 }
 
 
+
+void mgaInitState( mgaContextPtr mmesa )
+{
+   mgaScreenPrivate *mgaScreen = mmesa->mgaScreen;
+   GLcontext *ctx = mmesa->glCtx;
+
+   if (ctx->Color.DriverDrawBuffer == GL_BACK_LEFT) {
+      mmesa->draw_buffer = MGA_BACK;
+      mmesa->read_buffer = MGA_BACK;
+      mmesa->drawOffset = mmesa->mgaScreen->backOffset;
+      mmesa->readOffset = mmesa->mgaScreen->backOffset;
+      mmesa->Setup[MGA_CTXREG_DSTORG] = mgaScreen->backOffset;
+   } else {
+      mmesa->drawOffset = mmesa->mgaScreen->frontOffset;
+      mmesa->readOffset = mmesa->mgaScreen->frontOffset;
+      mmesa->draw_buffer = MGA_FRONT;
+      mmesa->read_buffer = MGA_FRONT;
+      mmesa->Setup[MGA_CTXREG_DSTORG] = mgaScreen->frontOffset;
+   }
+
+/*     mmesa->Setup[MGA_CTXREG_MACCESS] = mgaScreen->mAccess; */
+/*     mmesa->Setup[MGA_CTXREG_DWGCTL] = ( DC_clipdis_disable |  */
+/*  				       (0xC << DC_bop_SHIFT) | */
+/*  				       DC_shftzero_enable |  */
+/*  				       DC_zmode_nozcmp |  */
+/*  				       DC_atype_zi ); */
+
+
+   mmesa->Setup[MGA_CTXREG_MACCESS] = 0x1;
+   mmesa->Setup[MGA_CTXREG_DWGCTL] = 0xc4074;
+
+
+   mmesa->Setup[MGA_CTXREG_PLNWT] = ~0;
+   mmesa->Setup[MGA_CTXREG_ALPHACTRL] = ( AC_src_one |
+					  AC_dst_zero | 
+					  AC_amode_FCOL | 
+					  AC_astipple_disable | 
+					  AC_aten_disable | 
+					  AC_atmode_noacmp |
+					  AC_alphasel_fromtex );
+
+   mmesa->Setup[MGA_CTXREG_FOGCOLOR] = 
+      MGAPACKCOLOR888((GLubyte)(ctx->Fog.Color[0]*255.0F), 
+		      (GLubyte)(ctx->Fog.Color[1]*255.0F), 
+		      (GLubyte)(ctx->Fog.Color[2]*255.0F));
+
+   mmesa->Setup[MGA_CTXREG_WFLAG] = 0;
+   mmesa->Setup[MGA_CTXREG_TDUAL0] = 0;
+   mmesa->Setup[MGA_CTXREG_TDUAL1] = 0;
+   mmesa->Setup[MGA_CTXREG_FCOL] = 0;
+   mmesa->new_state = ~0;
+}
+
+
 void mgaDDInitStateFuncs( GLcontext *ctx )
 {
    ctx->Driver.UpdateState = mgaDDUpdateState;
@@ -710,11 +855,14 @@ void mgaDDInitStateFuncs( GLcontext *ctx )
    ctx->Driver.RenderStart = mgaDDUpdateHwState; 
    ctx->Driver.RenderFinish = 0; 
 
-   ctx->Driver.SetBuffer = mgaDDSetBuffer;
+   ctx->Driver.SetDrawBuffer = mgaDDSetDrawBuffer;
+   ctx->Driver.SetReadBuffer = mgaDDSetReadBuffer;
    ctx->Driver.Color = mgaDDSetColor;
    ctx->Driver.ClearColor = mgaDDClearColor;
    ctx->Driver.Dither = mgaDDDither;
 
+   ctx->Driver.PolygonStipple = mgaDDPolygonStipple;
+
    ctx->Driver.Index = 0;
    ctx->Driver.ClearIndex = 0;
    ctx->Driver.IndexMask = 0;
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgatex.c b/xc/lib/GL/mesa/src/drv/mga/mgatex.c
index be7ad6018..eafccbaa2 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgatex.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgatex.c
@@ -47,22 +47,25 @@
  * to it.
  */
 static void mgaDestroyTexObj( mgaContextPtr mmesa, mgaTextureObjectPtr t ) {
-	int	i;
 
  	if ( !t ) return;
   	  	
 	/* free the texture memory */
-	mmFreeMem( t->MemBlock );
+	if (t->MemBlock) {
+		mmFreeMem( t->MemBlock );
+		t->MemBlock = 0;
+
+		if (t->age > mmesa->dirtyAge)
+			mmesa->dirtyAge = t->age;
+	}
  
  	/* free mesa's link */   
-	t->tObj->DriverData = NULL;
+	if (t->tObj) 
+		t->tObj->DriverData = NULL;
 
 	/* see if it was the driver's current object */
-	for ( i = 0 ; i < 2 ; i++ ) {
-		if ( mmesa->CurrentTexObj[i] == t ) {
-			mmesa->CurrentTexObj[i] = NULL;
-		}  
-	}
+	if (t->bound)
+		mmesa->CurrentTexObj[t->bound - 1] = 0; 
 	
 	remove_from_list(t);
 	free( t );
@@ -73,6 +76,9 @@ static void mgaSwapOutTexObj(mgaContextPtr mmesa, mgaTextureObjectPtr t)
 	if (t->MemBlock) {
 		mmFreeMem(t->MemBlock);
 		t->MemBlock = 0;      
+
+		if (t->age > mmesa->dirtyAge)
+			mmesa->dirtyAge = t->age;
 	}
 
 	t->dirty_images = ~0;
@@ -80,22 +86,76 @@ static void mgaSwapOutTexObj(mgaContextPtr mmesa, mgaTextureObjectPtr t)
 }
 
 
-void mgaResetGlobalLRU( mgaContextPtr mmesa )
+static void mgaPrintLocalLRU( mgaContextPtr mmesa, int heap ) 
+{
+   mgaTextureObjectPtr t;
+   int sz = 1 << (mmesa->mgaScreen->logTextureGranularity[heap]);
+
+   fprintf(stderr, "\nLocal LRU, heap %d:\n", heap);
+
+   foreach( t, &(mmesa->TexObjList[heap]) ) {
+      if (!t->tObj)
+	 fprintf(stderr, "Placeholder %d at %x sz %x\n", 
+		 t->MemBlock->ofs / sz,
+		 t->MemBlock->ofs,
+		 t->MemBlock->size);      
+      else
+	 fprintf(stderr, "Texture (bound %d) at %x sz %x\n", 
+		 t->bound,
+		 t->MemBlock->ofs,
+		 t->MemBlock->size);      
+
+   }
+
+   fprintf(stderr, "\n\n");
+}
+
+static void mgaPrintGlobalLRU( mgaContextPtr mmesa, int heap )
+{
+   int i, j;
+   drm_mga_tex_region_t *list = mmesa->sarea->texList[heap];
+
+   fprintf(stderr, "\nGlobal LRU, heap %d list %p:\n", heap, list);
+
+   for (i = 0, j = MGA_NR_TEX_REGIONS ; i < MGA_NR_TEX_REGIONS ; i++) {
+      fprintf(stderr, "list[%d] age %d next %d prev %d\n",
+	      j, list[j].age, list[j].next, list[j].prev);
+      j = list[j].next;
+      if (j == MGA_NR_TEX_REGIONS) break;
+   }
+   
+   if (j != MGA_NR_TEX_REGIONS) {
+      fprintf(stderr, "Loop detected in global LRU\n\n\n");
+      for (i = 0 ; i < MGA_NR_TEX_REGIONS ; i++) {
+	      fprintf(stderr, "list[%d] age %d next %d prev %d\n",
+		      i, list[i].age, list[i].next, list[i].prev);
+      }
+   }
+   
+   fprintf(stderr, "\n\n");
+}
+
+
+static void mgaResetGlobalLRU( mgaContextPtr mmesa, GLuint heap )
 {
-   mgaTexRegion *list = mmesa->sarea->texList;
-   int sz = 1 << mmesa->mgaScreen->logTextureGranularity;
+   drm_mga_tex_region_t *list = mmesa->sarea->texList[heap];
+   int sz = 1 << mmesa->mgaScreen->logTextureGranularity[heap];
    int i;
 
+   mmesa->texAge[heap] = ++mmesa->sarea->texAge[heap];
+
+   if (0) fprintf(stderr, "mgaResetGlobalLRU %d\n", (int)heap);
+
    /* (Re)initialize the global circular LRU list.  The last element
     * in the array (MGA_NR_TEX_REGIONS) is the sentinal.  Keeping it
     * at the end of the array allows it to be addressed rationally
     * when looking up objects at a particular location in texture
     * memory.  
     */
-   for (i = 0 ; (i+1) * sz < mmesa->mgaScreen->textureSize ; i++) {
+   for (i = 0 ; (i+1) * sz <= mmesa->mgaScreen->textureSize[heap] ; i++) {
       list[i].prev = i-1;
       list[i].next = i+1;
-      list[i].age = 0;
+      list[i].age = mmesa->sarea->texAge[heap];   
    }
 
    i--;
@@ -104,30 +164,41 @@ void mgaResetGlobalLRU( mgaContextPtr mmesa )
    list[i].next = MGA_NR_TEX_REGIONS;
    list[MGA_NR_TEX_REGIONS].prev = i;
    list[MGA_NR_TEX_REGIONS].next = 0;
-   mmesa->sarea->texAge = 0;
+
 }
 
 
 static void mgaUpdateTexLRU( mgaContextPtr mmesa, mgaTextureObjectPtr t ) 
 {
    int i;
-   int logsz = mmesa->mgaScreen->logTextureGranularity;
+   int heap = t->heap;
+   int logsz = mmesa->mgaScreen->logTextureGranularity[heap];
    int start = t->MemBlock->ofs >> logsz;
    int end = (t->MemBlock->ofs + t->MemBlock->size - 1) >> logsz;
-   mgaTexRegion *list = mmesa->sarea->texList;
+   drm_mga_tex_region_t *list = mmesa->sarea->texList[heap];
    
-   mmesa->texAge = ++mmesa->sarea->texAge;
+   mmesa->texAge[heap] = ++mmesa->sarea->texAge[heap];
+
+   if (!t->MemBlock) {
+	   fprintf(stderr, "no memblock\n\n");
+	   return;
+   }
 
    /* Update our local LRU
     */
-   move_to_head( &(mmesa->TexObjList), t );
+   move_to_head( &(mmesa->TexObjList[heap]), t );
+
+
+   if (0)
+	   fprintf(stderr, "mgaUpdateTexLRU heap %d list %p\n", heap, list);
+
 
    /* Update the global LRU
     */
    for (i = start ; i <= end ; i++) {
 
       list[i].in_use = 1;
-      list[i].age = mmesa->texAge;
+      list[i].age = mmesa->texAge[heap];
 
       /* remove_from_list(i)
        */
@@ -141,8 +212,12 @@ static void mgaUpdateTexLRU( mgaContextPtr mmesa, mgaTextureObjectPtr t )
       list[(unsigned)list[MGA_NR_TEX_REGIONS].next].prev = i;
       list[MGA_NR_TEX_REGIONS].next = i;
    }
-}
 
+   if (0) {
+	   mgaPrintGlobalLRU(mmesa, t->heap);
+	   mgaPrintLocalLRU(mmesa, t->heap);
+   }
+}
 
 /* Called for every shared texture region which has increased in age
  * since we last held the lock.
@@ -151,23 +226,29 @@ static void mgaUpdateTexLRU( mgaContextPtr mmesa, mgaTextureObjectPtr t )
  * and pushes a placeholder texture onto the LRU list to represent 
  * the other client's textures.  
  */
-void mgaTexturesGone( mgaContextPtr mmesa,
-		       GLuint offset, 
-		       GLuint size,
-		       GLuint in_use ) 
+static void mgaTexturesGone( mgaContextPtr mmesa,
+			     GLuint heap,
+			     GLuint offset, 
+			     GLuint size,
+			     GLuint in_use ) 
 {
    mgaTextureObjectPtr t, tmp;
-   
-   foreach_s ( t, tmp, &mmesa->TexObjList ) {
+
+
+
+   foreach_s ( t, tmp, &(mmesa->TexObjList[heap]) ) {
 
       if (t->MemBlock->ofs >= offset + size ||
 	  t->MemBlock->ofs + t->MemBlock->size <= offset)
 	 continue;
 
+
+
+
       /* It overlaps - kick it off.  Need to hold onto the currently bound
        * objects, however.
        */
-      if (t == mmesa->CurrentTexObj[0] || t == mmesa->CurrentTexObj[1]) 
+      if (t->bound) 
 	 mgaSwapOutTexObj( mmesa, t );
       else
 	 mgaDestroyTexObj( mmesa, t );
@@ -178,12 +259,54 @@ void mgaTexturesGone( mgaContextPtr mmesa,
       t = (mgaTextureObjectPtr) calloc(1,sizeof(*t));
       if (!t) return;
 
-      t->MemBlock = mmAllocMem( mmesa->texHeap, size, 0, offset);      
-      insert_at_head( &mmesa->TexObjList, t );
+      t->heap = heap;
+      t->MemBlock = mmAllocMem( mmesa->texHeap[heap], size, 0, offset);      
+      if (!t->MemBlock) {
+	      fprintf(stderr, "Couldn't alloc placeholder sz %x ofs %x\n",
+		      (int)size, (int)offset);
+	      mmDumpMemInfo( mmesa->texHeap[heap]);
+	      return;
+      }
+      insert_at_head( &(mmesa->TexObjList[heap]), t );
    }
 }
 
 
+void mgaAgeTextures( mgaContextPtr mmesa, int heap )
+{
+	drm_mga_sarea_t *sarea = mmesa->sarea;
+	int sz = 1 << (mmesa->mgaScreen->logTextureGranularity[heap]);
+	int idx, nr = 0;
+
+	/* Have to go right round from the back to ensure stuff ends up
+	 * LRU in our local list...  Fix with a cursor pointer.
+	 */
+	for (idx = sarea->texList[heap][MGA_NR_TEX_REGIONS].prev ; 
+	     idx != MGA_NR_TEX_REGIONS && nr < MGA_NR_TEX_REGIONS ; 
+	     idx = sarea->texList[heap][idx].prev, nr++)
+	{
+		if (sarea->texList[heap][idx].age > mmesa->texAge[heap]) {
+			mgaTexturesGone(mmesa, heap, idx * sz, sz, 1);
+		}
+	}
+
+	if (nr == MGA_NR_TEX_REGIONS) {
+		mgaTexturesGone(mmesa, heap, 0,
+				mmesa->mgaScreen->textureSize[heap], 0);
+		mgaResetGlobalLRU( mmesa, heap );
+	}
+
+	
+	if (0) {
+		mgaPrintGlobalLRU( mmesa, heap );
+		mgaPrintLocalLRU( mmesa, heap );
+	}
+
+	mmesa->texAge[heap] = sarea->texAge[heap];
+	mmesa->dirty |= MGA_UPLOAD_TEX0IMAGE | MGA_UPLOAD_TEX1IMAGE;
+}
+
+
 /*
  * mgaSetTexWrappings
  */
@@ -281,22 +404,22 @@ static void mgaSetTexBorderColor(mgaTextureObjectPtr t, GLubyte color[4]) {
 
 
 /*
- * mgaUploadSubImage
+ * mgaUploadSubImageLocked
  *
  * Perform an iload based update of a resident buffer.  This is used for
  * both initial loading of the entire image, and texSubImage updates.
  *
  * Performed with the hardware lock held.
  */
-static void mgaUploadSubImage( mgaContextPtr mmesa,
-			       mgaTextureObjectPtr t,
-			       int level,	     
-			       int x, int y, int width, int height ) {
+static void mgaUploadSubImageLocked( mgaContextPtr mmesa,
+				     mgaTextureObjectPtr t,
+				     int level,	     
+				     int x, int y, int width, int height ) {
 	int		x2;
 	int		dwords;
-        int		dstorg;
+        int		offset;
 	struct gl_texture_image *image;
-	int		texelBytes, texelsPerDword, texelMaccess;
+	int		texelBytes, texelsPerDword, texelMaccess, length;
 
 	if ( level < 0 || level >= MGA_TEX_MAXLEVELS ) {
 		mgaMsg( 1, "mgaUploadSubImage: bad level: %i\n", level );
@@ -310,15 +433,9 @@ static void mgaUploadSubImage( mgaContextPtr mmesa,
 	}
 	
 	/* find the proper destination offset for this level */
-   	dstorg = (mmesa->mgaScreen->textureOffset + t->MemBlock->ofs + 
+   	offset = (t->MemBlock->ofs + 
 		  t->offsets[level]);
 
-    	/* turn on PCI/AGP if needed 
-    	if ( textureHeapPhysical ) {
-    		dstorg |= 1 | mgaglx.use_agp;
-    	}
-	*/
-
 	texelBytes = t->texelBytes;	
 	switch( texelBytes ) {
 	case 1:
@@ -363,7 +480,7 @@ static void mgaUploadSubImage( mgaContextPtr mmesa,
 		x = (x + (texelsPerDword-1)) & ~(texelsPerDword-1);
 		width = x2 - x;
 	}
-    
+
 	/* we may not be able to upload the entire texture in one
 	   batch due to register limits or dma buffer limits.
 	   Recursively split it up. */
@@ -374,7 +491,8 @@ static void mgaUploadSubImage( mgaContextPtr mmesa,
 		}
 		mgaMsg(10, "mgaUploadSubImage: recursively subdividing\n" );
 
-		mgaUploadSubImage( mmesa, t, level, x, y, width, height >> 1 );
+		mgaUploadSubImageLocked( mmesa, t, level, x, y, 
+					 width, height >> 1 );
 		y += ( height >> 1 );
 		height -= ( height >> 1 );
 	}
@@ -385,31 +503,50 @@ static void mgaUploadSubImage( mgaContextPtr mmesa,
 	/* bump the performance counter */
 	mgaglx.c_textureSwaps += ( dwords << 2 );
 	
-	
-#if 0
-
-	/* fill in the secondary buffer with properly converted texels
-	from the mesa buffer */
-	mgaConvertTexture( dest, texelBytes, image, x, y, width, height );
-		
-	/* send the secondary data */	
-	mgaSecondaryDma( TT_BLIT, dest, dwords );
-#endif
-
+   	length = dwords * 4;
+
+	/* Fill in the secondary buffer with properly converted texels
+	 * from the mesa buffer. */
+   	if(t->heap == MGA_CARD_HEAP) {
+	   mgaGetILoadBufferLocked( mmesa );
+	   mgaConvertTexture( (mgaUI32 *)mmesa->iload_buffer->address, 
+			     texelBytes, image, x, y, width, height ); 
+	   if(length < 64) length = 64;
+	   mgaMsg(10, "TexelBytes : %d, offset: %d, length : %d\n",
+		  texelBytes,
+		  mmesa->mgaScreen->textureOffset[t->heap] +
+		  offset + 
+		  y * width * 4/texelsPerDword,
+		  length);
+	   
+	   mgaFireILoadLocked( mmesa, 
+			      mmesa->mgaScreen->textureOffset[t->heap] +
+			      offset + 
+			      y * width * 4/texelsPerDword,
+			      length);
+	} else {
+ 	/* This works, is slower for uploads to card space and needs
+	 * additional synchronization with the dma stream.
+	 */
+	   mgaConvertTexture( (mgaUI32 *)
+			      (mmesa->mgaScreen->texVirtual[t->heap] + 
+					  offset + 
+					  y * width * 4/texelsPerDword), 
+			     texelBytes, image, x, y, width, height ); 
+	}
 }
 
 
-
 static void mgaUploadTexLevel( mgaContextPtr mmesa, 
-			       mgaTextureObjectPtr t,
-			       int l )
+				     mgaTextureObjectPtr t,
+				     int l )
 {
-	mgaUploadSubImage( mmesa,
-			   t,
-			   l,
-			   0, 0,
-			   t->tObj->Image[l]->Width,
-			   t->tObj->Image[l]->Height);
+	mgaUploadSubImageLocked( mmesa,
+				 t,
+				 l,
+				 0, 0,
+				 t->tObj->Image[l]->Width,
+				 t->tObj->Image[l]->Height);
 }
 
 
@@ -527,9 +664,11 @@ static void mgaCreateTexObj(mgaContextPtr mmesa, struct gl_texture_object *tObj)
 	/* fill in our mga texture object */
 	t->tObj = tObj;
 	t->ctx = mmesa;
+	t->age = 0;
+	t->bound = 0;
 
 	
-	insert_at_tail(&(mmesa->TexObjList), t);
+	insert_at_tail(&(mmesa->SwappedOut), t);
 	
 	t->MemBlock = 0;
 
@@ -596,34 +735,61 @@ static void mgaCreateTexObj(mgaContextPtr mmesa, struct gl_texture_object *tObj)
   	tObj->DriverData = t;
 }
 
+static void mgaMigrateTexture( mgaContextPtr mmesa, mgaTextureObjectPtr t )
+{
+	/* NOT DONE */
+}
 
 
-int mgaUploadTexImages( mgaContextPtr mmesa, mgaTextureObjectPtr t )
+static int mgaChooseTexHeap( mgaContextPtr mmesa, mgaTextureObjectPtr t )
 {
+	return 0;
+}
+
 
+int mgaUploadTexImages( mgaContextPtr mmesa, mgaTextureObjectPtr t )
+{
+	int heap;
 	int i;
 	int ofs;
 	mgaglx.c_textureSwaps++;
 
+	heap = t->heap = mgaChooseTexHeap( mmesa, t );
+
 	/* Do we need to eject LRU texture objects?
 	 */
 	if (!t->MemBlock) {
 		while (1)
 		{
-			t->MemBlock = mmAllocMem( mmesa->texHeap, t->totalSize, 12, 0 ); 
+			mgaTextureObjectPtr tmp = mmesa->TexObjList[heap].prev;
+
+			t->MemBlock = mmAllocMem( mmesa->texHeap[heap], 
+						  t->totalSize,
+						  6, 0 ); 
 			if (t->MemBlock)
 				break;
 
-			if (mmesa->TexObjList.prev == &(mmesa->TexObjList)) {
-				fprintf(stderr, "Failed to upload texture, sz %d\n", t->totalSize);
-				mmDumpMemInfo( mmesa->texHeap );
+			if (mmesa->TexObjList[heap].prev->bound) {
+				fprintf(stderr, 
+					"Hit bound texture in upload\n"); 
 				return -1;
 			}
 
-			mgaDestroyTexObj( mmesa, mmesa->TexObjList.prev );
+			if (mmesa->TexObjList[heap].prev == 
+			    &(mmesa->TexObjList[heap])) 
+			{
+				fprintf(stderr, "Failed to upload texture, "
+					"sz %d\n", t->totalSize);
+				mmDumpMemInfo( mmesa->texHeap[heap] );
+				return -1;
+			}
+	 
+			mgaDestroyTexObj( mmesa, tmp );
 		}
  
-		ofs = t->MemBlock->ofs;
+		ofs = t->MemBlock->ofs 
+			+ mmesa->mgaScreen->textureOffset[heap]
+			;
 
 		t->Setup[MGA_TEXREG_ORG]  = ofs;
 		t->Setup[MGA_TEXREG_ORG1] = ofs + t->offsets[1];
@@ -638,8 +804,16 @@ int mgaUploadTexImages( mgaContextPtr mmesa, mgaTextureObjectPtr t )
 	 */
 	mgaUpdateTexLRU( mmesa, t );
 
+	
+	if (MGA_DEBUG&DEBUG_VERBOSE_LRU)
+		fprintf(stderr, "dispatch age: %d age freed memory: %d\n",
+			GET_DISPATCH_AGE(mmesa), mmesa->dirtyAge);
+
+	if (mmesa->dirtyAge >= GET_DISPATCH_AGE(mmesa)) 
+		mgaWaitAgeLocked( mmesa, mmesa->dirtyAge );
+
 	if (t->dirty_images) {
-		if (MGA_DEBUG & MGA_DEBUG_VERBOSE_MSG)
+  		if (MGA_DEBUG&DEBUG_VERBOSE_LRU)
 			fprintf(stderr, "*");
 
 		for (i = 0 ; i <= t->lastLevel ; i++)
@@ -692,12 +866,10 @@ static void mgaUpdateTextureEnvG200( GLcontext *ctx )
 	}
 }
 
-/* I don't have the alpha values correct yet:
- */
 static void mgaUpdateTextureStage( GLcontext *ctx, int unit )
 {
 	mgaContextPtr mmesa = MGA_CONTEXT( ctx );
-	mgaUI32 *reg = &mmesa->Setup[MGA_CTXREG_TDUAL0 + unit];
+	GLuint *reg = &mmesa->Setup[MGA_CTXREG_TDUAL0 + unit];
 	GLuint source = mmesa->tmu_source[unit];
 	struct gl_texture_object *tObj = ctx->Texture.Unit[source].Current;
 
@@ -725,13 +897,13 @@ static void mgaUpdateTextureStage( GLcontext *ctx, int unit )
 			*reg = ( TD0_color_arg2_diffuse |
 				 TD0_color_sel_mul | 
 				 TD0_alpha_arg2_diffuse |
-				 TD0_alpha_sel_arg1);
+				 TD0_alpha_sel_mul);
 		else
 			*reg = ( TD0_color_arg2_prevstage |
 				 TD0_color_alpha_prevstage |
 				 TD0_color_sel_mul | 
 				 TD0_alpha_arg2_prevstage |
-				 TD0_alpha_sel_arg1);
+				 TD0_alpha_sel_mul);
 		break;
 	case GL_DECAL:
 		*reg = (TD0_color_arg2_fcol | 
@@ -751,20 +923,49 @@ static void mgaUpdateTextureStage( GLcontext *ctx, int unit )
 				 TD0_color_add_add |
 				 TD0_color_sel_add | 
 				 TD0_alpha_arg2_diffuse |
-				 TD0_alpha_sel_arg1);
+				 TD0_alpha_sel_add);
 		else
 			*reg = ( TD0_color_arg2_prevstage |
 				 TD0_color_alpha_prevstage |
 				 TD0_color_add_add |
 				 TD0_color_sel_add | 
 				 TD0_alpha_arg2_prevstage |
-				 TD0_alpha_sel_arg1);
+				 TD0_alpha_sel_add);
 		break;
 
 	case GL_BLEND:
-		/* Use a multipass mechanism to do this:
+		if (0)
+			fprintf(stderr, "GL_BLEND unit %d flags %x\n", unit,
+				mmesa->blend_flags);
+
+		if (mmesa->blend_flags) 
+			mmesa->Fallback |= MGA_FALLBACK_TEXTURE;
+			return;
+
+		/* Do singletexture GL_BLEND with 'all ones' env-color
+		 * by using both texture units.  Multitexture gl_blend
+		 * is a fallback.  
 		 */
-		mmesa->Fallback |= MGA_FALLBACK_TEXTURE;
+		if (unit == 0) {
+			/* Part 1: R1 = Rf ( 1 - Rt )
+			 *         A1 = Af At
+			 */
+			*reg = ( TD0_color_arg2_diffuse |
+				 TD0_color_arg1_inv_enable |
+				 TD0_color_sel_mul | 
+				 TD0_alpha_arg2_diffuse |
+				 TD0_alpha_sel_arg1);
+		} else {
+			/* Part 2: R2 = R1 + Rt
+			 *         A2 = A1
+			 */
+			*reg = ( TD0_color_arg2_prevstage |
+				 TD0_color_add_add | 
+				 TD0_color_sel_add | 
+				 TD0_alpha_arg2_prevstage |
+				 TD0_alpha_sel_arg2);		
+		}	
+
 		break;
 	default:
 	}
@@ -782,9 +983,8 @@ static void mgaUpdateTextureObject( GLcontext *ctx, int unit ) {
 	mgaMsg(15,"mgaUpdateTextureState %d\n", unit);
 	
 	/* disable texturing until it is known to be good */
-	mmesa->Setup[MGA_CTXREG_DWGCTL] = 
-		(( mmesa->Setup[MGA_CTXREG_DWGCTL] & DC_opcod_MASK ) | 
-		 DC_opcod_trap);
+	mmesa->Setup[MGA_CTXREG_DWGCTL] &= DC_opcod_MASK;
+	mmesa->Setup[MGA_CTXREG_DWGCTL] |= DC_opcod_trap;
 
 	enabled = (ctx->Texture.Enabled>>(source*4))&TEXTURE0_ANY;
 	if (enabled != TEXTURE0_2D) {
@@ -797,6 +997,8 @@ static void mgaUpdateTextureObject( GLcontext *ctx, int unit ) {
 
 	if ( !tObj || tObj != ctx->Texture.Unit[source].CurrentD[2] ) 
 		return;
+
+/*  	fprintf(stderr, "unit %d: %d\n", unit, tObj->Name); */
 	
 	/* if the texture object doesn't exist at all (never used or
 	   swapped out), create it now, uploading all texture images */
@@ -804,7 +1006,6 @@ static void mgaUpdateTextureObject( GLcontext *ctx, int unit ) {
 	if ( !tObj->DriverData ) {
 		/* clear the current pointer so that texture object can be
 		   swapped out if necessary to make room */
-		mmesa->CurrentTexObj[source] = NULL;
 		mgaCreateTexObj( mmesa, tObj );
 
 		if ( !tObj->DriverData ) {
@@ -815,9 +1016,8 @@ static void mgaUpdateTextureObject( GLcontext *ctx, int unit ) {
 	}
 
 	/* we definately have a valid texture now */
-	mmesa->Setup[MGA_CTXREG_DWGCTL] = 
-		(( mmesa->Setup[MGA_CTXREG_DWGCTL] & DC_opcod_MASK ) | 
-		 DC_opcod_texture_trap);
+	mmesa->Setup[MGA_CTXREG_DWGCTL] &= DC_opcod_MASK;
+	mmesa->Setup[MGA_CTXREG_DWGCTL] |= DC_opcod_texture_trap;
 
 	t = (mgaTextureObjectPtr)tObj->DriverData;
 
@@ -825,6 +1025,11 @@ static void mgaUpdateTextureObject( GLcontext *ctx, int unit ) {
 		mmesa->dirty |= (MGA_UPLOAD_TEX0IMAGE << unit);
 
 	mmesa->CurrentTexObj[unit] = t;
+	t->bound = unit+1;
+
+	if (t->MemBlock)
+		mgaUpdateTexLRU( mmesa, t );
+
 
 	t->Setup[MGA_TEXREG_CTL2] &= ~TMC_dualtex_enable;
 	if (ctx->Texture.Enabled == (TEXTURE0_2D|TEXTURE1_2D))
@@ -848,23 +1053,31 @@ void mgaUpdateTextureState( GLcontext *ctx )
 	mgaContextPtr mmesa = MGA_CONTEXT( ctx );
 	mmesa->Fallback &= ~MGA_FALLBACK_TEXTURE;
 
+	if (mmesa->CurrentTexObj[0]) mmesa->CurrentTexObj[0]->bound = 0;
+	if (mmesa->CurrentTexObj[1]) mmesa->CurrentTexObj[1]->bound = 0;
+	mmesa->CurrentTexObj[0] = 0;
+	mmesa->CurrentTexObj[1] = 0;   
+
 	if (MGA_IS_G400(mmesa)) {
 		mgaUpdateTextureObject( ctx, 0 );		
 		mgaUpdateTextureStage( ctx, 0 );
 
-		mmesa->Setup[MGA_CTXREG_TDUAL1] = mmesa->Setup[MGA_CTXREG_TDUAL0];
+		mmesa->Setup[MGA_CTXREG_TDUAL1] = 
+			mmesa->Setup[MGA_CTXREG_TDUAL0];
 
 		if (mmesa->multitex) {
 			mgaUpdateTextureObject( ctx, 1 );	
 			mgaUpdateTextureStage( ctx, 1 );
 		}
+		
+		mmesa->dirty |= MGA_UPLOAD_TEX0 | MGA_UPLOAD_TEX1;
 	} else {
 		mgaUpdateTextureObject( ctx, 0 );
-		mgaUpdateTextureEnvG200( ctx );
+		mgaUpdateTextureEnvG200( ctx );		
 	}
 
 	/* schedule the register writes */
-	mmesa->dirty |= MGA_UPLOAD_CTX;
+	mmesa->dirty |= MGA_UPLOAD_CTX | MGA_UPLOAD_TEX0;
 }
 
 
@@ -880,14 +1093,57 @@ Driver functions called directly from mesa
 /*
  * mgaTexEnv
  */      
-void mgaTexEnv( GLcontext *ctx, GLenum pname, const GLfloat *param ) {
+void mgaTexEnv( GLcontext *ctx, GLenum pname, const GLfloat *param ) 
+{
+	mgaContextPtr mmesa = MGA_CONTEXT(ctx);
 	mgaMsg( 10, "mgaTexEnv( %i )\n", pname );
 
+
 	if (pname == GL_TEXTURE_ENV_MODE) {
 		/* force the texture state to be updated */
-		MGA_CONTEXT(ctx)->CurrentTexObj[0] = 0;  
+		FLUSH_BATCH( MGA_CONTEXT(ctx) );
 		MGA_CONTEXT(ctx)->new_state |= MGA_NEW_TEXTURE;
 	}
+	else if (pname == GL_TEXTURE_ENV_COLOR)
+	{
+		struct gl_texture_unit *texUnit = 
+			&ctx->Texture.Unit[ctx->Texture.CurrentUnit];
+		GLfloat *fc = texUnit->EnvColor;
+		GLubyte c[4];
+		GLuint col;
+
+		
+		c[0] = fc[0];
+		c[1] = fc[1];
+		c[2] = fc[2];
+		c[3] = fc[3];
+
+		/* No alpha at 16bpp?
+		 */
+		col = mgaPackColor( mmesa->mgaScreen->Attrib,
+				    c[0], c[1], c[2], c[3] );
+
+  		mmesa->envcolor = (c[3]<<24) | (c[0]<<16) | (c[1]<<8) | (c[2]); 
+    
+		if (mmesa->Setup[MGA_CTXREG_FCOL] != col) {
+			FLUSH_BATCH(mmesa);	
+			mmesa->Setup[MGA_CTXREG_FCOL] = col;      
+			mmesa->dirty |= MGA_UPLOAD_CTX;
+
+			mmesa->blend_flags &= ~MGA_BLEND_ENV_COLOR;
+
+			/* Actually just require all four components to be
+			 * equal.  This permits a single-pass GL_BLEND.
+			 * 
+			 * More complex multitexture/multipass fallbacks
+			 * for blend can be done later.
+			 */
+			if (mmesa->envcolor != 0x0 &&
+			    mmesa->envcolor != 0xffffffff)
+				mmesa->blend_flags |= MGA_BLEND_ENV_COLOR;
+		}
+	}
+
 }
 
 /*
@@ -907,6 +1163,7 @@ void mgaTexImage( GLcontext *ctx, GLenum target,
 	mgaUpdateTextureState time. */  
 	t = (mgaTextureObjectPtr) tObj->DriverData;
 	if ( t ) {
+		if (t->bound) FLUSH_BATCH(mmesa);
 		/* if this is the current object, it will force an update */
  	 	mgaDestroyTexObj( mmesa, t );
 		mmesa->new_state |= MGA_NEW_TEXTURE;
@@ -936,6 +1193,7 @@ void mgaTexSubImage( GLcontext *ctx, GLenum target,
 	mgaUpdateTextureState time. */  
 	t = (mgaTextureObjectPtr) tObj->DriverData;
 	if ( t ) {
+		if (t->bound) FLUSH_BATCH(mmesa);
 		/* if this is the current object, it will force an update */
  	 	mgaDestroyTexObj( mmesa, t );
 		mmesa->new_state |= MGA_NEW_TEXTURE;
@@ -956,7 +1214,9 @@ void mgaTexSubImage( GLcontext *ctx, GLenum target,
  */
 void mgaTexParameter( GLcontext *ctx, GLenum target,
 		      struct gl_texture_object *tObj,
-		      GLenum pname, const GLfloat *params ) {
+		      GLenum pname, const GLfloat *params )
+{
+	mgaContextPtr mmesa = MGA_CONTEXT( ctx );
 	mgaTextureObjectPtr t;
 
 	mgaMsg( 10, "mgaTexParameter( %p, %i )\n", tObj, pname );
@@ -973,52 +1233,68 @@ void mgaTexParameter( GLcontext *ctx, GLenum target,
 	switch (pname) {
 	case GL_TEXTURE_MIN_FILTER:
 	case GL_TEXTURE_MAG_FILTER:
+		if (t->bound) FLUSH_BATCH(mmesa);
 		mgaSetTexFilter( t, tObj->MinFilter, tObj->MagFilter );
 		break;
 
 	case GL_TEXTURE_WRAP_S:
 	case GL_TEXTURE_WRAP_T:
+		if (t->bound) FLUSH_BATCH(mmesa);
 		mgaSetTexWrapping(t,tObj->WrapS,tObj->WrapT);
 		break;
   
 	case GL_TEXTURE_BORDER_COLOR:
+		if (t->bound) FLUSH_BATCH(mmesa);
 		mgaSetTexBorderColor(t,tObj->BorderColor);
 		break;
 
 	default:
 		return;
 	}
-	/* force the texture state to be updated */
-	MGA_CONTEXT(ctx)->CurrentTexObj[0] = NULL;  
-	MGA_CONTEXT(ctx)->new_state |= MGA_NEW_TEXTURE;
+
+	mmesa->new_state |= MGA_NEW_TEXTURE;
 }
 
 /*
  * mgaBindTexture
  */      
 void mgaBindTexture( GLcontext *ctx, GLenum target,
-		     struct gl_texture_object *tObj ) {
+		     struct gl_texture_object *tObj ) 
+{
+	mgaContextPtr mmesa = MGA_CONTEXT( ctx );
 
 	mgaMsg( 10, "mgaBindTexture( %p )\n", tObj );
-		
+
+	FLUSH_BATCH(mmesa);
+
+	if (mmesa->CurrentTexObj[ctx->Texture.CurrentUnit]) {
+		mmesa->CurrentTexObj[ctx->Texture.CurrentUnit]->bound = 0;
+		mmesa->CurrentTexObj[ctx->Texture.CurrentUnit] = 0;  
+	}
+
 	/* force the texture state to be updated 
 	 */
-	MGA_CONTEXT(ctx)->CurrentTexObj[0] = NULL;  
 	MGA_CONTEXT(ctx)->new_state |= MGA_NEW_TEXTURE;
 }
 
 /*
  * mgaDeleteTexture
  */      
-void mgaDeleteTexture( GLcontext *ctx, struct gl_texture_object *tObj ) {
+void mgaDeleteTexture( GLcontext *ctx, struct gl_texture_object *tObj ) 
+{
+	mgaContextPtr mmesa = MGA_CONTEXT( ctx );
+	mgaTextureObjectPtr t = (mgaTextureObjectPtr)tObj->DriverData;
 
 	mgaMsg( 10, "mgaDeleteTexture( %p )\n", tObj );
 	
-	/* delete our driver data */
-	if ( tObj->DriverData ) {
-		mgaContextPtr mmesa = MGA_CONTEXT( ctx );
-		mgaDestroyTexObj( mmesa,
-				  (mgaTextureObjectPtr)(tObj->DriverData) );
+	if ( t ) {
+		if (t->bound) {
+			FLUSH_BATCH(mmesa);
+			mmesa->CurrentTexObj[t->bound-1] = 0;
+			mmesa->new_state |= MGA_NEW_TEXTURE;
+		}
+
+		mgaDestroyTexObj( mmesa, t );
 		mmesa->new_state |= MGA_NEW_TEXTURE;
 	}
 }
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgatris.c b/xc/lib/GL/mesa/src/drv/mga/mgatris.c
index bd8ef083f..387a79f3e 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgatris.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgatris.c
@@ -27,6 +27,7 @@
 #include <stdio.h>
 #include <math.h>
 
+#include "types.h"
 #include "vb.h"
 #include "pipeline.h"
 
@@ -122,34 +123,14 @@ void mgaDDTrifuncInit()
    init_twoside_offset();
    init_twoside_offset_flat();
 
-   /* Hmmm...
-    */
-   for (i = 0 ; i < 0x20 ; i++) {
-      if (i & ~MGA_FLAT_BIT) {
-	 points_tab[i] = points_tab[i&MGA_FLAT_BIT];
-	 line_tab[i] = line_tab[i&MGA_FLAT_BIT];
-      }
-   }
-
    for (i = 0 ; i < 0x20 ; i++) 
-      if ((i & (MGA_NODRAW_BIT|MGA_FALLBACK_BIT)) == MGA_NODRAW_BIT ||
-	  mgaglx.nullprims) 
+      if ((i & (MGA_NODRAW_BIT|MGA_FALLBACK_BIT)) == MGA_NODRAW_BIT)
       {
 	 quad_tab[i] = mga_null_quad; 
 	 tri_tab[i] = mga_null_triangle; 
 	 line_tab[i] = mga_null_line;
 	 points_tab[i] = mga_null_points;
       }
-
-   if (mgaglx.noFallback) {
-      for (i = 0 ; i < 0x10 ; i++) {
-	 points_tab[i|MGA_FALLBACK_BIT] = points_tab[i];
-	 line_tab[i|MGA_FALLBACK_BIT] = line_tab[i];
-	 tri_tab[i|MGA_FALLBACK_BIT] = tri_tab[i];
-	 quad_tab[i|MGA_FALLBACK_BIT] = quad_tab[i];
-      }
-   }
-
 }
 
 
@@ -162,50 +143,61 @@ void mgaDDChooseRenderState( GLcontext *ctx )
    mgaContextPtr mmesa = MGA_CONTEXT( ctx );
    GLuint flags = ctx->TriangleCaps;
 
-   ctx->IndirectTriangles &= ~DD_SW_RASTERIZE;
+   if (mmesa->Fallback)
+      return;
+
+   mmesa->IndirectTriangles = 0;
 
    if (flags) {
       GLuint ind = 0;
       GLuint shared = 0;
-      GLuint fallback = MGA_FALLBACK_BIT;
-
-      if (mgaglx.noFallback) fallback = 0;
 
       if (flags & DD_Z_NEVER)                      shared |= MGA_NODRAW_BIT;
       if (flags & DD_FLATSHADE)                    shared |= MGA_FLAT_BIT;
-      if (flags & DD_MULTIDRAW)                    shared |= fallback;
-      if (flags & (DD_SELECT|DD_FEEDBACK))         shared |= MGA_FALLBACK_BIT;
+      if (flags & (DD_MULTIDRAW|
+		   DD_SELECT|
+		   DD_FEEDBACK))                   shared |= MGA_FALLBACK_BIT;
       if (flags & DD_STENCIL)                      shared |= MGA_FALLBACK_BIT;
 
       ind = shared;
+#if 0
       if (flags & DD_POINT_SMOOTH)                 ind |= MGA_ANTIALIAS_BIT;
-      if (flags & DD_POINT_ATTEN)                  ind |= fallback;
+#else
+      if (flags & DD_POINT_SMOOTH)                 ind |= MGA_FALLBACK_BIT;
+#endif
 
       mmesa->renderindex = ind;
       mmesa->PointsFunc = points_tab[ind];
       if (ind & MGA_FALLBACK_BIT) 
-	 ctx->IndirectTriangles |= DD_POINT_SW_RASTERIZE;
+	 mmesa->IndirectTriangles |= DD_POINT_SW_RASTERIZE;
 
       ind = shared;
+#if 0
       if (flags & DD_LINE_SMOOTH)                   ind |= MGA_ANTIALIAS_BIT;
-      if (flags & DD_LINE_STIPPLE)                  ind |= fallback;
+#else
+      if (flags & DD_LINE_SMOOTH)                   ind |= MGA_FALLBACK_BIT;
+#endif
+      if (flags & DD_LINE_STIPPLE)                  ind |= MGA_FALLBACK_BIT;
 
       mmesa->renderindex |= ind;
       mmesa->LineFunc = line_tab[ind];
       if (ind & MGA_FALLBACK_BIT) 
-	 ctx->IndirectTriangles |= DD_LINE_SW_RASTERIZE;
+	 mmesa->IndirectTriangles |= DD_LINE_SW_RASTERIZE;
 
       ind = shared;
       if (flags & DD_TRI_SMOOTH)                    ind |= MGA_ANTIALIAS_BIT;
       if (flags & DD_TRI_OFFSET)                    ind |= MGA_OFFSET_BIT;
       if (flags & DD_TRI_LIGHT_TWOSIDE)             ind |= MGA_TWOSIDE_BIT;
-      if (flags & (DD_TRI_UNFILLED|DD_TRI_STIPPLE)) ind |= fallback;	
+      if (flags & DD_TRI_UNFILLED)                  ind |= MGA_FALLBACK_BIT;
+      if ((flags & DD_TRI_STIPPLE) &&
+	  (ctx->IndirectTriangles & DD_TRI_STIPPLE)) ind |= MGA_FALLBACK_BIT;	
 
       mmesa->renderindex |= ind;
       mmesa->TriangleFunc = tri_tab[ind];
       mmesa->QuadFunc = quad_tab[ind];
       if (ind & MGA_FALLBACK_BIT)
-	 ctx->IndirectTriangles |= (DD_TRI_SW_RASTERIZE | DD_QUAD_SW_RASTERIZE);
+	 mmesa->IndirectTriangles |= (DD_TRI_SW_RASTERIZE | 
+				      DD_QUAD_SW_RASTERIZE);
    } 
    else if (mmesa->renderindex)
    {
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgatritmp.h b/xc/lib/GL/mesa/src/drv/mga/mgatritmp.h
index ac5a545d2..2a9d16b86 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgatritmp.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgatritmp.h
@@ -98,19 +98,31 @@ static void TAG(quad)( GLcontext *ctx, GLuint v0,
 }
 
 
-#if ((IND & ~MGA_FLAT_BIT) == 0)
-
 static void TAG(line)( GLcontext *ctx, GLuint v0, GLuint v1, GLuint pv )
 {
    mgaContextPtr mmesa = MGA_CONTEXT( ctx );
    mgaVertexPtr mgaVB = MGA_DRIVER_DATA(ctx->VB)->verts;
    float width = ctx->Line.Width;
 
-   if (IND & MGA_FLAT_BIT) {
+   if (IND & (MGA_TWOSIDE_BIT|MGA_FLAT_BIT)) {
       mgaVertex tmp0 = mgaVB[v0];
       mgaVertex tmp1 = mgaVB[v1];
-      *(int *)&tmp0.warp1.color = *(int *)&mgaVB[pv].warp1.color;
-      *(int *)&tmp1.warp1.color = *(int *)&mgaVB[pv].warp1.color;
+
+      if (IND & MGA_TWOSIDE_BIT) {
+	 GLubyte (*vbcolor)[4] = ctx->VB->ColorPtr->data;
+
+	 if (IND & MGA_FLAT_BIT) {
+ 	    MGA_COLOR((char *)&tmp0.warp1.color,vbcolor[pv]);
+	    *(int *)&tmp1.warp1.color = *(int *)&tmp0.warp1.color;
+	 } else {
+	    MGA_COLOR((char *)&tmp0.warp1.color,vbcolor[v0]);
+	    MGA_COLOR((char *)&tmp1.warp1.color,vbcolor[v1]);
+	 }
+
+      } else {
+	 *(int *)&tmp0.warp1.color = *(int *)&mgaVB[pv].warp1.color;
+	 *(int *)&tmp1.warp1.color = *(int *)&mgaVB[pv].warp1.color;
+      }
       mga_draw_line( mmesa, &tmp0, &tmp1, width );
    }
    else
@@ -125,13 +137,20 @@ static void TAG(points)( GLcontext *ctx, GLuint first, GLuint last )
    mgaVertexPtr mgaVB = MGA_DRIVER_DATA(VB)->verts;
    GLfloat sz = ctx->Point.Size * .5;
    int i;
-
+   
    for(i=first;i<=last;i++) 
-      if(VB->ClipMask[i]==0) 
-	 mga_draw_point( mmesa, &mgaVB[i], sz );
+      if(VB->ClipMask[i]==0) {
+	 if (IND & MGA_TWOSIDE_BIT) {	
+	    GLubyte (*vbcolor)[4] = VB->ColorPtr->data;
+	    mgaVertex tmp0 = mgaVB[i];
+	    MGA_COLOR((char *)&tmp0.warp1.color, vbcolor[i]);
+	    mga_draw_point( mmesa, &tmp0, sz );
+	 } else
+	    mga_draw_point( mmesa, &mgaVB[i], sz );
+      }
 }
 
-#endif
+
 
 
 static void TAG(init)( void )
@@ -139,10 +158,9 @@ static void TAG(init)( void )
    tri_tab[IND] = TAG(triangle);
    quad_tab[IND] = TAG(quad);
 
-#if ((IND & ~MGA_FLAT_BIT) == 0) 
    line_tab[IND] = TAG(line);
    points_tab[IND] = TAG(points);
-#endif
+
 }
 
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgavb.c b/xc/lib/GL/mesa/src/drv/mga/mgavb.c
index 4772e8a34..3f246d9a7 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgavb.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgavb.c
@@ -104,7 +104,7 @@ static void name(struct vertex_buffer *VB, GLuint start, GLuint end)	\
    GLfloat yoffset = mmesa->driDrawable->h + mmesa->drawY + SUBPIXEL_Y;	\
    int i;								\
    (void) xoffset; (void) yoffset;					\
-									\
+   if (0) fprintf(stderr, "V");						\
    gl_import_client_data( VB, VB->ctx->RenderFlags,			\
 			  (VB->ClipOrMask				\
 			   ? VEC_WRITABLE|VEC_GOOD_STRIDE		\
diff --git a/xc/lib/GL/mesa/src/drv/r128/Imakefile b/xc/lib/GL/mesa/src/drv/r128/Imakefile
new file mode 100644
index 000000000..4cd07b8ff
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/Imakefile
@@ -0,0 +1,325 @@
+XCOMM $XFree86$
+
+#include <Threads.tmpl>
+
+#define DoNormalLib NormalLibGlx
+#define DoSharedLib SharedLibGlx
+#define DoExtraLib SharedLibGlx
+#define DoDebugLib DebugLibGlx
+#define DoProfileLib ProfileLibGlx
+
+#if Malloc0ReturnsNull
+ALLOC_DEFINES = -DMALLOC_0_RETURNS_NULL
+#endif
+
+#if BuildXF86DRI
+  DRI_DEFINES = GlxDefines -DDRIVERTS
+ DRI_INCLUDES = -I../../../../dri \
+		-I../../../../glx \
+		-I../../../dri \
+		-I$(TOP)/include \
+		-I$(TOP)/include/GL \
+		-I$(XF86OSSRC) \
+		-I$(XF86COMSRC) \
+		-I$(SERVERSRC)/GL/dri \
+		-I$(XF86DRIVERSRC)/r128 \
+		-I../../../include \
+		-I../.. \
+		-I../../X \
+		-I../common
+#endif
+
+LinkSourceFile(mm.c, ../common)
+LinkSourceFile(mm.h, ../common)
+LinkSourceFile(hwlog.c, ../common)
+LinkSourceFile(hwlog.h, ../common)
+
+MESA_INCLUDES = -I. -I.. -I../../include
+
+      DEFINES = $(ALLOC_DEFINES) $(DRI_DEFINES)
+     INCLUDES = -I$(XLIBSRC) -I$(EXTINCSRC) $(MESA_INCLUDES) $(DRI_INCLUDES)
+
+     R128SRCS = r128_cce.c \
+		r128_clear.c \
+		r128_context.c \
+		r128_dd.c \
+		r128_fastpath.c \
+		r128_pipeline.c \
+                r128_screen.c \
+		r128_span.c \
+		r128_state.c \
+		r128_swap.c \
+		r128_tex.c \
+		r128_tris.c \
+		r128_vb.c \
+		r128_xmesa.c
+
+     R128OBJS = r128_cce.o \
+		r128_clear.o \
+		r128_context.o \
+		r128_dd.o \
+		r128_fastpath.o \
+		r128_pipeline.o \
+                r128_screen.o \
+		r128_span.o \
+		r128_state.o \
+		r128_swap.o \
+		r128_tex.o \
+		r128_tris.o \
+		r128_vb.o \
+		r128_xmesa.o
+
+#if !GlxUseBuiltInDRIDriver
+      DRISRCS = ../../../dri/dri_mesa.c \
+		../../../../dri/dri_tmm.c
+
+      DRIOBJS = ../../../dri/dri_mesa.o \
+		../../../../dri/dri_tmm.o
+
+      DRMSRCS = ../../../../dri/drm/xf86drm.c \
+		../../../../dri/drm/xf86drmHash.c \
+		../../../../dri/drm/xf86drmRandom.c \
+		../../../../dri/drm/xf86drmSL.c \
+		../../../../dri/drm/xf86drmR128.c
+
+      DRMOBJS = ../../../../dri/drm/xf86drm.o \
+		../../../../dri/drm/xf86drmHash.o \
+		../../../../dri/drm/xf86drmRandom.o \
+		../../../../dri/drm/xf86drmSL.o \
+		../../../../dri/drm/xf86drmR128.o
+
+     MESASRCS = ../../aatriangle.c \
+		../../accum.c \
+		../../alpha.c \
+		../../alphabuf.c \
+		../../attrib.c \
+		../../bbox.c \
+		../../bitmap.c \
+		../../blend.c \
+		../../buffers.c \
+		../../clip.c \
+		../../colortab.c \
+		../../config.c \
+		../../context.c \
+		../../copypix.c \
+		../../cva.c \
+		../../debug_xform.c \
+		../../depth.c \
+		../../dlist.c \
+		../../drawpix.c \
+		../../enable.c \
+		../../enums.c \
+		../../eval.c \
+		../../extensions.c \
+		../../feedback.c \
+		../../fog.c \
+		../../get.c \
+		../../glapi.c \
+		../../glapinoop.c \
+		../../glthread.c \
+		../../hash.c \
+		../../image.c \
+		../../imaging.c \
+		../../light.c \
+		../../lines.c \
+		../../logic.c \
+		../../masking.c \
+		../../matrix.c \
+		../../mem.c \
+		../../mmath.c \
+		../../pb.c \
+		../../pipeline.c \
+		../../pixel.c \
+		../../pixeltex.c \
+		../../points.c \
+		../../polygon.c \
+		../../quads.c \
+		../../rastpos.c \
+		../../readpix.c \
+		../../rect.c \
+		../../scissor.c \
+		../../shade.c \
+		../../span.c \
+		../../stages.c \
+		../../state.c \
+		../../stencil.c \
+		../../teximage.c \
+		../../texobj.c \
+		../../texstate.c \
+		../../texture.c \
+		../../texutil.c \
+		../../translate.c \
+		../../triangle.c \
+		../../varray.c \
+		../../vb.c \
+		../../vbcull.c \
+		../../vbfill.c \
+		../../vbindirect.c \
+		../../vbrender.c \
+		../../vbxform.c \
+		../../vector.c \
+		../../vertices.c \
+		../../winpos.c \
+		../../xform.c \
+		../../zoom.c \
+		../../X86/common_x86.c
+
+     MESAOBJS = ../../aatriangle.o \
+		../../accum.o \
+		../../alpha.o \
+		../../alphabuf.o \
+		../../attrib.o \
+		../../bbox.o \
+		../../bitmap.o \
+		../../blend.o \
+		../../buffers.o \
+		../../clip.o \
+		../../colortab.o \
+		../../config.o \
+		../../context.o \
+		../../copypix.o \
+		../../cva.o \
+		../../debug_xform.o \
+		../../depth.o \
+		../../dlist.o \
+		../../drawpix.o \
+		../../enable.o \
+		../../enums.o \
+		../../eval.o \
+		../../extensions.o \
+		../../feedback.o \
+		../../fog.o \
+		../../get.o \
+		../../hash.o \
+		../../hint.o \
+		../../image.o \
+		../../imaging.o \
+		../../light.o \
+		../../lines.o \
+		../../logic.o \
+		../../masking.o \
+		../../matrix.o \
+		../../mem.o \
+		../../mmath.o \
+		../../pb.o \
+		../../pipeline.o \
+		../../pixel.o \
+		../../pixeltex.o \
+		../../points.o \
+		../../polygon.o \
+		../../quads.o \
+		../../rastpos.o \
+		../../readpix.o \
+		../../rect.o \
+		../../scissor.o \
+		../../shade.o \
+		../../span.o \
+		../../stages.o \
+		../../state.o \
+		../../stencil.o \
+		../../teximage.o \
+		../../texobj.o \
+		../../texstate.o \
+		../../texture.o \
+		../../texutil.o \
+		../../translate.o \
+		../../triangle.o \
+		../../varray.o \
+		../../vb.o \
+		../../vbcull.o \
+		../../vbfill.o \
+		../../vbindirect.o \
+		../../vbrender.o \
+		../../vbxform.o \
+		../../vector.o \
+		../../vertices.o \
+		../../winpos.o \
+		../../xform.o \
+		../../zoom.o
+
+#ifdef i386Architecture
+     X86_SRCS = ../../X86/x86.c \
+		../../X86/x86a.S \
+		../../X86/common_x86.c \
+		../../X86/common_x86asm.S \
+		../../X86/vertex.S
+
+     X86_OBJS = ../../X86/x86.o \
+		../../X86/x86a.o \
+		../../X86/common_x86.o \
+		../../X86/common_x86asm.o \
+		../../X86/vertex.o
+
+     MMX_SRCS = ../../X86/mmx_blend.S
+
+     MMX_OBJS = ../../X86/mmx_blend.o
+
+XCOMM Disabling 3Dnow code for the time being.
+#if 0
+   3DNOW_SRCS = ../../X86/3dnow.c \
+		../../X86/3dnow_norm_raw.S \
+		../../X86/3dnow_xform_masked1.S \
+		../../X86/3dnow_xform_masked2.S \
+		../../X86/3dnow_xform_masked3.S \
+		../../X86/3dnow_xform_masked4.S \
+		../../X86/3dnow_xform_raw1.S \
+		../../X86/3dnow_xform_raw2.S \
+		../../X86/3dnow_xform_raw3.S \
+		../../X86/3dnow_xform_raw4.S \
+		../../X86/vertex_3dnow.S
+
+   3DNOW_OBJS = ../../X86/3dnow.o \
+		../../X86/3dnow_norm_raw.o \
+		../../X86/3dnow_xform_masked1.o \
+		../../X86/3dnow_xform_masked2.o \
+		../../X86/3dnow_xform_masked3.o \
+		../../X86/3dnow_xform_masked4.o \
+		../../X86/3dnow_xform_raw1.o \
+		../../X86/3dnow_xform_raw2.o \
+		../../X86/3dnow_xform_raw3.o \
+		../../X86/3dnow_xform_raw4.o \
+		../../X86/vertex_3dnow.o
+#endif
+
+#endif
+#endif
+
+      ASMSRCS = $(X86_SRCS) $(MMX_SRCS) $(3DNOW_SRCS)
+      ASMOBJS = $(X86_OBJS) $(MMX_OBJS) $(3DNOW_OBJS)
+
+   COMMONSRCS = mm.c hwlog.c
+   COMMONOBJS = mm.o hwlog.o
+
+         SRCS = $(DRISRCS) $(DRMSRCS) $(MESASRCS) $(ASMSRCS) \
+		$(COMMONSRCS) $(R128SRCS)
+         OBJS = $(DRIOBJS) $(DRMOBJS) $(MESAOBJS) $(ASMOBJS) \
+		$(COMMONOBJS) $(R128OBJS)
+
+REQUIREDLIBS += -lm
+#if !GlxBuiltInMga
+REQUIREDLIBS += -L../../../.. -lGL
+#endif
+
+#if !GlxUseBuiltInDRIDriver
+#undef DoNormalLib NormalLibGlx
+#undef DoExtraLib SharedLibGlx
+#undef DoDebugLib DebugLibGlx
+#undef DoProfileLib ProfileLibGlx
+#endif
+
+#include <Library.tmpl>
+
+LibraryObjectRule()
+
+SubdirLibraryRule($(OBJS))
+NormalLintTarget($(SRCS))
+
+#if !GlxUseBuiltInDRIDriver
+LIBNAME = r128_dri.so
+ALL_OBJS = $(OBJS)
+ALL_DEPS = DONE
+SharedDepModuleTarget($(LIBNAME),$(ALL_DEPS),$(ALL_OBJS))
+InstallDynamicModule($(LIBNAME),$(MODULEDIR)/dri,.)
+#endif
+
+DependTarget()
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_cce.c b/xc/lib/GL/mesa/src/drv/r128/r128_cce.c
new file mode 100644
index 000000000..e949dd37c
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_cce.c
@@ -0,0 +1,389 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas.
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#include "r128_init.h"
+#include "r128_mesa.h"
+#include "r128_xmesa.h"
+#include "r128_context.h"
+#include "r128_lock.h"
+#include "r128_reg.h"
+#include "r128_cce.h"
+
+#if USE_USER_SPACE_RING
+
+#define USE_IOCTLS          0
+#define DO_SECURITY_CHECK   1
+#define R128_TIMEOUT        2000000
+
+#if DEBUG
+static void r128RingStatus(r128ScreenPtr pScrn)
+{
+    unsigned char *R128MMIO = pScrn->mmio;
+
+    R128_DEBUG(("GUI_STAT           = 0x%08x\n",
+		(unsigned int)INREG(R128_GUI_STAT)));
+    R128_DEBUG(("PM4_STAT           = 0x%08x\n",
+		(unsigned int)INREG(R128_PM4_STAT)));
+    R128_DEBUG(("PM4_BUFFER_DL_WPTR = 0x%08x\n",
+		(unsigned int)INREG(R128_PM4_BUFFER_DL_WPTR)));
+    R128_DEBUG(("PM4_BUFFER_DL_RPTR = 0x%08x\n",
+		(unsigned int)INREG(R128_PM4_BUFFER_DL_RPTR)));
+    R128_DEBUG(("ringWrite          = 0x%08x\n", pScrn->SAREA->ringWrite));
+    R128_DEBUG(("*ringReadPtr       = 0x%08x\n", *pScrn->ringReadPtr));
+}
+#endif
+
+/* FIXME: Requires access to secure registers */
+static int INPLL(r128ScreenPtr pScrn, int addr)
+{
+    unsigned char *R128MMIO = pScrn->mmio;
+
+    OUTREG8(R128_CLOCK_CNTL_INDEX, addr & 0x1f);
+    return INREG(R128_CLOCK_CNTL_DATA);
+}
+
+/* FIXME: Requires access to secure registers */
+static void r128EngineFlush(r128ScreenPtr pScrn)
+{
+    unsigned char *R128MMIO = pScrn->mmio;
+    int           i;
+
+    OUTREGP(R128_PC_NGUI_CTLSTAT, R128_PC_FLUSH_ALL, ~R128_PC_FLUSH_ALL);
+    for (i = 0; i < R128_TIMEOUT; i++) {
+	if (!(INREG(R128_PC_NGUI_CTLSTAT) & R128_PC_BUSY)) break;
+    }
+}
+
+/* FIXME: Requires access to secure registers */
+static void r128EngineReset(r128ScreenPtr pScrn)
+{
+#if USE_IOCTLS
+#if DEBUG
+    r128RingStatus(pScrn);
+#endif
+
+    (void)drmR128EngineReset(pScrn->driScreen->fd);
+#else
+    unsigned char *R128MMIO = pScrn->mmio;
+    CARD32        clock_cntl_index;
+    CARD32        mclk_cntl;
+    CARD32        gen_reset_cntl;
+	
+#if DEBUG
+    r128RingStatus(pScrn);
+#endif
+
+    r128EngineFlush(pScrn);
+
+    clock_cntl_index = INREG(R128_CLOCK_CNTL_INDEX);
+    mclk_cntl        = INPLL(pScrn, R128_MCLK_CNTL);
+
+    OUTPLL(R128_MCLK_CNTL, mclk_cntl | R128_FORCE_GCP | R128_FORCE_PIPE3D_CPP);
+
+    gen_reset_cntl   = INREG(R128_GEN_RESET_CNTL);
+
+    OUTREG(R128_GEN_RESET_CNTL, gen_reset_cntl | R128_SOFT_RESET_GUI);
+    INREG(R128_GEN_RESET_CNTL);
+    OUTREG(R128_GEN_RESET_CNTL, gen_reset_cntl & ~R128_SOFT_RESET_GUI);
+    INREG(R128_GEN_RESET_CNTL);
+
+    OUTPLL(R128_MCLK_CNTL,        mclk_cntl);
+    OUTREG(R128_CLOCK_CNTL_INDEX, clock_cntl_index);
+    OUTREG(R128_GEN_RESET_CNTL,   gen_reset_cntl);
+
+    /* Reset the ring buffer status */
+    if (R128CCE_USE_RING_BUFFER(pScrn->CCEMode)) {
+	OUTREG(R128_PM4_BUFFER_DL_WPTR, 0);
+	OUTREG(R128_PM4_BUFFER_DL_RPTR, 0);
+	pScrn->SAREA->ringWrite = 0;
+	*pScrn->ringReadPtr = 0;
+    }
+#endif
+    fprintf(stderr, "Error: Rage 128 timed out... exiting\n");
+    exit(-1);
+}
+
+#define r128CCEWaitForFifo(pScrn, entries)                                   \
+do {                                                                         \
+    if (pScrn->CCEFifoSlots < entries)                                       \
+	r128CCEWaitForFifoFunction(pScrn, entries);                          \
+    pScrn->CCEFifoSlots -= entries;                                          \
+} while (0)
+
+/* Wait for at least `entries' slots are free.  The actual number of
+   slots available is stored in info->CCEFifoSize. */
+static void r128CCEWaitForFifoFunction(r128ScreenPtr pScrn, int entries)
+{
+    unsigned char *R128MMIO = pScrn->mmio;
+    int           i;
+
+    for (;;) {
+	for (i = 0; i < R128_TIMEOUT; i++) {
+	    pScrn->CCEFifoSlots = INREG(R128_PM4_STAT) & R128_PM4_FIFOCNT_MASK;
+	    if (pScrn->CCEFifoSlots >= entries) return;
+	}
+	R128_DEBUG(("Reseting Engine: PM4_STAT = 0x%08x\n",
+		    (unsigned int)INREG(R128_PM4_STAT)));
+	r128EngineReset(pScrn);
+    }
+}
+
+/* Wait until the CCE is completely idle: the FIFO has drained and the
+   CCE is idle. */
+void r128CCEWaitForIdle(r128ScreenPtr pScrn)
+{
+#if USE_IOCTLS
+    if (drmR128WaitForIdle(pScrn->driScreen->fd) < 0) {
+	fprintf(stderr, "Error: Rage 128 timed out... exiting\n");
+	exit(-1);
+    }
+#else
+    unsigned char *R128MMIO = pScrn->mmio;
+    int           i;
+
+    if (R128CCE_USE_RING_BUFFER(pScrn->CCEMode)) {
+	OUTREGP(R128_PM4_BUFFER_DL_WPTR,
+		R128_PM4_BUFFER_DL_DONE, ~R128_PM4_BUFFER_DL_DONE);
+
+	for (;;) {
+	    for (i = 0; i < R128_TIMEOUT; i++) {
+		if (*pScrn->ringReadPtr == pScrn->SAREA->ringWrite) {
+		    int pm4stat = INREG(R128_PM4_STAT);
+		    if ((pm4stat & R128_PM4_FIFOCNT_MASK) == pScrn->CCEFifoSize
+			&& !(pm4stat & (R128_PM4_BUSY | R128_PM4_GUI_ACTIVE)))
+			return;
+		}
+	    }
+	    R128_DEBUG(("Reseting Engine: PM4_STAT = 0x%08x\n",
+			(unsigned int)INREG(R128_PM4_STAT)));
+	    r128EngineReset(pScrn);
+	}
+    } else {
+	r128CCEWaitForFifoFunction(pScrn, pScrn->CCEFifoSize);
+
+	for (;;) {
+	    for (i = 0; i < R128_TIMEOUT; i++) {
+		if (!(INREG(R128_PM4_STAT)
+		      & (R128_PM4_BUSY | R128_PM4_GUI_ACTIVE))) {
+		    r128EngineFlush(pScrn);
+		    return;
+		}
+	    }
+	    R128_DEBUG(("Reseting Engine: PM4_STAT = 0x%08x\n",
+			(unsigned int)INREG(R128_PM4_STAT)));
+	    r128EngineReset(pScrn);
+	}
+    }
+#endif
+}
+
+/* Send a packet to the CCE via the PIO registers */
+static void r128CCESubmitPacketPIO(r128ContextPtr r128ctx,
+				   CARD32 *buf, int count)
+{
+    unsigned char *R128MMIO     = r128ctx->r128Screen->mmio;
+#if DO_SECURITY_CHECK
+    CARD32         tmp          = 0;
+    int            psize        = 0;
+    int            writing      = 1;
+    int            addr         = R128_PM4_FIFO_DATA_EVEN;
+#endif
+
+#if DO_SECURITY_CHECK
+    while (count > 0) {
+	tmp = *buf++;
+	if (!psize) {
+	    writing = 1;
+
+	    if ((tmp & R128_CCE_PACKET_MASK) == R128_CCE_PACKET0) {
+		if ((tmp & R128_CCE_PACKET0_REG_MASK) <= (0x1004 >> 2)) {
+		    if ((tmp & R128_CCE_PACKET0_REG_MASK) !=
+			(R128_PM4_VC_FPU_SETUP >> 2)) {
+			writing = 0;
+		    }
+		}
+		psize = ((tmp & R128_CCE_PACKET_COUNT_MASK) >> 16) + 2;
+	    } else if ((tmp & R128_CCE_PACKET_MASK) == R128_CCE_PACKET1) {
+		if ((tmp & R128_CCE_PACKET1_REG0_MASK) <= (0x1004 >> 2)) {
+		    if ((tmp & R128_CCE_PACKET1_REG0_MASK) !=
+			(R128_PM4_VC_FPU_SETUP >> 2)) {
+			writing = 0;
+		    }
+		} else if ((tmp & R128_CCE_PACKET1_REG1_MASK) <=
+			   (0x1004 << 9)) {
+		    if ((tmp & R128_CCE_PACKET1_REG1_MASK) !=
+			(R128_PM4_VC_FPU_SETUP << 9)) {
+			writing = 0;
+		    }
+		}
+		psize = 3;
+	    } else {
+		psize = ((tmp & R128_CCE_PACKET_COUNT_MASK) >> 16) + 2;
+	    }
+	}
+	psize--;
+
+	if (writing) {
+	    r128CCEWaitForFifo(r128ctx->r128Screen, 1);
+	    OUTREG(addr, tmp);
+	    addr ^= 0x0004;
+	}
+
+	count--;
+    }
+
+    if (addr == R128_PM4_FIFO_DATA_ODD) {
+	    r128CCEWaitForFifo(r128ctx->r128Screen, 1);
+	    OUTREG(addr, R128_CCE_PACKET2);
+    }
+#else
+    while (count > 1) {
+	r128CCEWaitForFifo(r128ctx->r128Screen, 2);
+	OUTREG(R128_PM4_FIFO_DATA_EVEN, *buf++);
+	OUTREG(R128_PM4_FIFO_DATA_ODD,  *buf++);
+	count -= 2;
+    }
+
+    if (count) {
+	r128CCEWaitForFifo(r128ctx->r128Screen, 2);
+	OUTREG(R128_PM4_FIFO_DATA_EVEN, *buf);
+	OUTREG(R128_PM4_FIFO_DATA_ODD,  R128_CCE_PACKET2);
+    }
+#endif
+}
+
+/* Send a packet to the CCE via the ring */
+static void r128CCESubmitPacketRing(r128ContextPtr r128ctx,
+				    CARD32 *buf, int count)
+{
+    r128ScreenPtr     r128scrn  = r128ctx->r128Screen;
+    unsigned char    *R128MMIO  = r128ctx->r128Screen->mmio;
+    int            ringWrite    = r128scrn->SAREA->ringWrite;
+    int           *ringWritePtr = r128scrn->ringStartPtr + ringWrite;
+    int            timeout;
+#if DO_SECURITY_CHECK
+    CARD32         tmp          = 0;
+    int            psize        = 0;
+    int            writing      = 1;
+#endif
+
+    if (count > r128scrn->ringEntries) {
+	/* FIXME */
+	return;
+    }
+
+    while (count > 0) {
+#if DO_SECURITY_CHECK
+	tmp = *buf++;
+	if (!psize) {
+	    writing = 1;
+
+	    if ((tmp & R128_CCE_PACKET_MASK) == R128_CCE_PACKET0) {
+		if ((tmp & R128_CCE_PACKET0_REG_MASK) <= (0x1004 >> 2)) {
+		    if ((tmp & R128_CCE_PACKET0_REG_MASK) !=
+			(R128_PM4_VC_FPU_SETUP >> 2)) {
+			writing = 0;
+		    }
+		}
+		psize = ((tmp & R128_CCE_PACKET_COUNT_MASK) >> 16) + 2;
+	    } else if ((tmp & R128_CCE_PACKET_MASK) == R128_CCE_PACKET1) {
+		if ((tmp & R128_CCE_PACKET1_REG0_MASK) <= (0x1004 >> 2)) {
+		    if ((tmp & R128_CCE_PACKET1_REG0_MASK) !=
+			(R128_PM4_VC_FPU_SETUP >> 2)) {
+			writing = 0;
+		    }
+		} else if ((tmp & R128_CCE_PACKET1_REG1_MASK) <=
+			   (0x1004 << 9)) {
+		    if ((tmp & R128_CCE_PACKET1_REG1_MASK) !=
+			(R128_PM4_VC_FPU_SETUP << 9)) {
+			writing = 0;
+		    }
+		}
+		psize = 3;
+	    } else {
+		psize = ((tmp & R128_CCE_PACKET_COUNT_MASK) >> 16) + 2;
+	    }
+	}
+	psize--;
+
+	if (writing) {
+	    ringWrite++;
+	    *ringWritePtr++ = tmp;
+	}
+#else
+	ringWrite++;
+	*ringWritePtr++ = *buf++;
+#endif
+	if (ringWrite >= r128scrn->ringEntries) {
+	    ringWrite    = 0;
+	    ringWritePtr = r128scrn->ringStartPtr;
+	}
+	timeout = 0;
+	while (ringWrite == *r128scrn->ringReadPtr) {
+	    (void)INREG(R128_PM4_BUFFER_DL_RPTR);
+	    if (timeout++ >= R128_TIMEOUT) r128EngineReset(r128scrn);
+	}
+
+	count--;
+    }
+
+    if (ringWrite < 32) {
+	memcpy(r128scrn->ringEndPtr,
+	       r128scrn->ringStartPtr,
+	       ringWrite * sizeof(int));
+    }
+
+    r128scrn->SAREA->ringWrite = ringWrite;
+    OUTREG(R128_PM4_BUFFER_DL_WPTR, ringWrite);
+}
+
+void r128CCESubmitPackets(r128ContextPtr r128ctx, CARD32 *buf, int count)
+{
+#if USE_IOCTLS
+    if (drmR128SubmitPackets(r128ctx->r128Screen->driScreen->fd,
+			     r128ctx->CCEbuf, r128ctx->CCEcount,
+			     0) < 0) {
+	r128EngineReset(r128ctx->r128Screen);
+	fprintf(stderr, "Error: Illegal CCE packet... exiting\n");
+	exit(-1);
+    }
+#else
+    if (R128CCE_USE_RING_BUFFER(r128ctx->r128Screen->CCEMode))
+	r128CCESubmitPacketRing(r128ctx, buf, count);
+    else
+	r128CCESubmitPacketPIO(r128ctx, buf, count);
+#endif
+}
+
+#endif
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_cce.h b/xc/lib/GL/mesa/src/drv/r128/r128_cce.h
new file mode 100644
index 000000000..162686cb7
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_cce.h
@@ -0,0 +1,142 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas.
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#ifndef _R128_CCE_H_
+#define _R128_CCE_H_
+
+#ifdef GLX_DIRECT_RENDERING
+
+#include "r128_dri.h"
+#include "r128_reg.h"
+
+#include "xf86drmR128.h"
+
+typedef union {
+    float f;
+    int   i;
+} floatTOint;
+
+#define R128_DEFAULT_TOTAL_CCE_TIMEOUT 1000000 /* usecs */
+
+/* Insert an integer value into the CCE ring buffer. */
+#define R128CCE(v)                                                           \
+    do {                                                                     \
+	r128ctx->CCEbuf[r128ctx->CCEcount] = (v);                            \
+	r128ctx->CCEcount++;                                                 \
+    } while (0)
+
+/* Insert an floating point value into the CCE ring buffer. */
+#define R128CCEF(v)                                                          \
+    do {                                                                     \
+	floatTOint fTi;                                                      \
+	fTi.f = (v);                                                         \
+	r128ctx->CCEbuf[r128ctx->CCEcount] = fTi.i;                          \
+	r128ctx->CCEcount++;                                                 \
+    } while (0)
+
+#if USE_USER_SPACE_RING
+
+#define R128CCE_SUBMIT_PACKETS()                                             \
+    do {                                                                     \
+	r128CCESubmitPackets(r128ctx, r128ctx->CCEbuf, r128ctx->CCEcount);   \
+	r128ctx->CCEcount = 0;                                               \
+    } while (0)
+
+#define R128CCE_WAIT_FOR_IDLE(r128ctx)                                       \
+    r128CCEWaitForIdle(r128ctx->r128Screen)
+
+#else /* if !USE_USER_SPACE_RING */
+
+#define R128CCE_SUBMIT_PACKETS()                                             \
+    do {                                                                     \
+	CARD32 *_buf;                                                        \
+	int     _c   = r128ctx->CCEcount;                                    \
+	int     _fd  = r128ctx->r128Screen->driScreen->fd;                   \
+	int     _to  = 0;                                                    \
+	int     _ret;                                                        \
+                                                                             \
+	do {                                                                 \
+	    _buf = r128ctx->CCEbuf + (r128ctx->CCEcount - _c);               \
+	    _ret = drmR128SubmitPackets(_fd, _buf, &_c, 0);                  \
+	} while (_ret < 0 && _ret == -EBUSY && _to++ < r128ctx->CCEtimeout); \
+	if (_ret < 0) {                                                      \
+	    (void)drmR128EngineReset(_fd);                                   \
+	    fprintf(stderr, "Error: Could not submit packet... exiting\n");  \
+	    exit(-1);                                                        \
+	}                                                                    \
+	r128ctx->CCEcount = 0;                                               \
+    } while (0)
+
+#define R128CCE_WAIT_FOR_IDLE(r128ctx)                                       \
+    do {                                                                     \
+	int    _fd  = r128ctx->r128Screen->driScreen->fd;                    \
+	int    _to  = 0;                                                     \
+	int    _ret;                                                         \
+                                                                             \
+	(void)drmR128EngineFlush(_fd);                                       \
+	do {                                                                 \
+	    _ret = drmR128CCEWaitForIdle(_fd);                               \
+	} while (_ret < 0 && _ret == -EBUSY && _to++ < r128ctx->CCEtimeout); \
+	if (_ret < 0) {                                                      \
+	    (void)drmR128EngineReset(_fd);                                   \
+	    fprintf(stderr, "Error: Rage 128 timed out... exiting\n");       \
+	    exit(-1);                                                        \
+	}                                                                    \
+    } while (0)
+
+#endif
+
+#define R128CCE_WAIT_FOR_IDLE_LOCK(r128ctx)                                  \
+    do {                                                                     \
+	LOCK_HARDWARE(r128ctx);                                              \
+	R128CCE_WAIT_FOR_IDLE(r128ctx);                                      \
+	UNLOCK_HARDWARE(r128ctx);                                            \
+    } while (0)
+
+
+/* Insert a type-[0123] packet header into the ring buffer */
+#define R128CCE0(p,r,n)   R128CCE((p) | ((n) << 16) | ((r) >> 2))
+#define R128CCE1(p,r1,r2) R128CCE((p) | (((r2) >> 2) << 11) | ((r1) >> 2))
+#define R128CCE2(p)       R128CCE((p))
+#define R128CCE3(p,n)     R128CCE((p) | ((n) << 16))
+
+
+#if USE_USER_SPACE_RING
+extern void r128CCEWaitForIdle(r128ScreenPtr pScrn);
+extern void r128CCESubmitPackets(r128ContextPtr r128ctx,
+				 CARD32 *buf, int count);
+#endif
+
+#endif
+#endif /* _R128_CCE_H_ */
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_ccevb.h b/xc/lib/GL/mesa/src/drv/r128/r128_ccevb.h
new file mode 100644
index 000000000..2084b8672
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_ccevb.h
@@ -0,0 +1,224 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#ifndef _R128_CCEVB_H_
+#define _R128_CCEVB_H_
+
+#ifdef GLX_DIRECT_RENDERING
+
+#include "xf86drmR128.h"
+
+typedef struct {
+    CARD32          *buf;   /* Pointer to current VB */
+    int              index; /* Index of current VB */
+    int              start; /* in bytes from the beginning of the VB map */
+    int              size;  /* in vertices */
+    int              count; /* in vertices */
+    drmR128PrimType  prim;  /* Primitive type for vertices in VB */
+    int              done;  /* VB has been sent to ring */
+} r128CCEVertBuf, *r128CCEVertBufPtr;
+
+
+
+/* Flush the CPU's write-combining cache */
+/* FIXME: This code is both processor and compiler specific */
+#define R128_FLUSH_WC_MEMORY()                                               \
+    do {                                                                     \
+	int xchangeDummy;                                                    \
+                                                                             \
+	__asm__ volatile("push %%eax ;"                                      \
+			 "xchg %%eax, %0 ;"                                  \
+			 "pop %%eax" : : "m" (xchangeDummy));                \
+	__asm__ volatile("push %%eax ;"                                      \
+			 "push %%ebx ;"                                      \
+			 "push %%ecx ;"                                      \
+			 "push %%edx ;"                                      \
+			 "movl $0,%%eax ;"                                   \
+			 "cpuid ;"                                           \
+			 "pop %%edx ;"                                       \
+			 "pop %%ecx ;"                                       \
+			 "pop %%ebx ;"                                       \
+			 "pop %%eax" : /* no outputs */ : /* no inputs */);  \
+    } while (0)
+
+
+/* Get a new VB from the pool of vertex buffers in AGP space */
+/* NOTE: By default the primitive type for a VB is set to individual tris */
+#define R128CCE_GET_NEW_VB(r128ctx)                                          \
+    do {                                                                     \
+	r128CCEVertBufPtr  vb    = &r128ctx->vb;                             \
+	int                to    = 0;                                        \
+	int                fd    = r128ctx->r128Screen->driScreen->fd;       \
+	int                num;                                              \
+	int                index;                                            \
+	int                size;                                             \
+                                                                             \
+	vb->buf = NULL;                                                      \
+                                                                             \
+	LOCK_HARDWARE(r128ctx);                                              \
+	while (!vb->buf && to++ < r128ctx->CCEtimeout) {                     \
+	    /* Ask the kernel a new vertex buffer */                         \
+	    num = drmR128GetVertexBuffers(fd, 1, &index, &size);             \
+                                                                             \
+	    if (num > 0) {                                                   \
+		vb->buf   = (CARD32 *)(r128ctx->r128Screen->                 \
+				       vbBufs->list[index].address);         \
+		vb->index = index;                                           \
+		vb->start = index*r128ctx->r128Screen->vbBufSize;            \
+		vb->size  = size/sizeof(r128_vertex);                        \
+		vb->count = 0;                                               \
+		vb->prim  = DRM_R128_PRIM_TRI_LIST;                          \
+		vb->done  = GL_FALSE;                                        \
+	    }                                                                \
+	}                                                                    \
+	UNLOCK_HARDWARE(r128ctx);                                            \
+                                                                             \
+	if (!vb->buf) {                                                      \
+	    (void)drmR128EngineReset(fd);                                    \
+	    fprintf(stderr, "Error: Could not get new VB... exiting\n");     \
+	    exit(-1);                                                        \
+	}                                                                    \
+    } while (0)
+
+/* Submit a VB to the hardware for processing */
+/* NOTE: This macro is only called while holding the hardware lock */
+#define R128CCE_SUBMIT_VB(r128ctx)                                           \
+    do {                                                                     \
+	int       index = r128ctx->vb.index;                                 \
+	int       size  = r128ctx->vb.count;                                 \
+	int       fd    = r128ctx->r128Screen->driScreen->fd;                \
+	int       to    = 0;                                                 \
+	int       ret;                                                       \
+	CARD32    prim;                                                      \
+                                                                             \
+	switch (r128ctx->vb.prim) {                                          \
+	case DRM_R128_PRIM_NONE:                                             \
+	    prim = R128_CCE_VC_CNTL_PRIM_TYPE_NONE;                          \
+	    break;                                                           \
+	case DRM_R128_PRIM_POINT:                                            \
+	    prim = R128_CCE_VC_CNTL_PRIM_TYPE_POINT;                         \
+	    break;                                                           \
+	case DRM_R128_PRIM_LINE:                                             \
+	    prim = R128_CCE_VC_CNTL_PRIM_TYPE_LINE;                          \
+	    break;                                                           \
+	case DRM_R128_PRIM_POLY_LINE:                                        \
+	    prim = R128_CCE_VC_CNTL_PRIM_TYPE_POLY_LINE;                     \
+	    break;                                                           \
+	case DRM_R128_PRIM_TRI_LIST:                                         \
+	    prim = R128_CCE_VC_CNTL_PRIM_TYPE_TRI_LIST;                      \
+	    break;                                                           \
+	case DRM_R128_PRIM_TRI_FAN:                                          \
+	    prim = R128_CCE_VC_CNTL_PRIM_TYPE_TRI_FAN;                       \
+	    break;                                                           \
+	case DRM_R128_PRIM_TRI_STRIP:                                        \
+	    prim = R128_CCE_VC_CNTL_PRIM_TYPE_TRI_STRIP;                     \
+	    break;                                                           \
+	case DRM_R128_PRIM_TRI_TYPE2:                                        \
+	    prim = R128_CCE_VC_CNTL_PRIM_TYPE_TRI_TYPE2;                     \
+	    break;                                                           \
+	default:                                                             \
+	    prim = R128_CCE_VC_CNTL_PRIM_TYPE_NONE;                          \
+	    break;                                                           \
+	}                                                                    \
+                                                                             \
+	/* Send the vertex buffer to the hardware */                         \
+	BEGIN_CLIP_LOOP(r128ctx);                                            \
+                                                                             \
+	R128CCE3(R128_CCE_PACKET3_3D_RNDR_GEN_INDX_PRIM, 3);                 \
+	R128CCE(r128ctx->r128Screen->vbOffset + r128ctx->vb.start);          \
+	R128CCE(r128ctx->vb.count);                                          \
+	R128CCE(R128_FULL_VERTEX_FORMAT);                                    \
+	R128CCE(prim | R128_CCE_VC_CNTL_PRIM_WALK_LIST |                     \
+		(r128ctx->vb.count << R128_CCE_VC_CNTL_NUM_SHIFT));          \
+	R128CCE_SUBMIT_PACKETS();                                            \
+                                                                             \
+	END_CLIP_LOOP(r128ctx);                                              \
+                                                                             \
+	/* Tell the kernel to release the vertex buffer */                   \
+	do {                                                                 \
+	    ret = drmR128FlushVertexBuffers(fd, 1, &index, &size,            \
+					    r128ctx->vb.prim);               \
+	} while (ret < 0 && ret == -EBUSY && to++ < r128ctx->CCEtimeout);    \
+	if (ret < 0) {                                                       \
+	    (void)drmR128EngineReset(fd);                                    \
+	    fprintf(stderr, "Error: Could not flush VB... exiting\n");       \
+	    exit(-1);                                                        \
+	}                                                                    \
+                                                                             \
+	r128ctx->vb.done = GL_TRUE;                                          \
+    } while (0)
+
+/* Flush the vertex buffer (assuming we already hold the lock) */
+#define R128CCE_FLUSH_VB(r128ctx)                                            \
+    do {                                                                     \
+	if (r128ctx->vb.count && !r128ctx->vb.done) {                        \
+	    /* FIXME: Is this _really_ needed? */                            \
+	    if (!R128CCE_USE_RING_BUFFER(r128ctx->r128Screen->CCEMode))      \
+		R128CCE_WAIT_FOR_IDLE(r128ctx);                              \
+                                                                             \
+	    R128CCE_SUBMIT_VB(r128ctx);                                      \
+	}                                                                    \
+    } while (0)
+
+/* Flush the vertex buffer */
+#define R128CCE_FLUSH_VB_LOCK(r128ctx)                                       \
+    do {                                                                     \
+	LOCK_HARDWARE(r128ctx);                                              \
+	R128CCE_FLUSH_VB(r128ctx);                                           \
+	UNLOCK_HARDWARE(r128ctx);                                            \
+    } while (0)
+
+/* Reserve space in a VB to store `n' vertices, and set the pointer `p'
+   to point to the next vertex in the current VB.  If adding these
+   vertices to the VB would exceed the size of the VB, then flush the
+   current VB, get a new one from the pool of VBs, reserve space in the
+   new VB for the `n' vertices, and set the pointer `p' to point to the
+   first vertex in the new VB. */
+#define R128CCE_ALLOC_VB_SPACE(r128ctx, p, n)                                \
+    do {                                                                     \
+	r128CCEVertBufPtr  vb = &r128ctx->vb;                                \
+                                                                             \
+	if (vb->done) R128CCE_GET_NEW_VB(r128ctx);                           \
+                                                                             \
+	if (vb->count + (n) > vb->size) {                                    \
+	    R128CCE_FLUSH_VB_LOCK(r128ctx);                                  \
+	    R128CCE_GET_NEW_VB(r128ctx);                                     \
+	}                                                                    \
+                                                                             \
+	(p)        = (r128_vertex *)vb->buf + vb->count;                     \
+	vb->count += (n);                                                    \
+    } while (0)
+
+#endif
+#endif /* _R128_CCEVB_H_ */
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_clear.c b/xc/lib/GL/mesa/src/drv/r128/r128_clear.c
new file mode 100644
index 000000000..9ab0581b0
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_clear.c
@@ -0,0 +1,248 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#include "r128_init.h"
+#include "r128_mesa.h"
+#include "r128_xmesa.h"
+#include "r128_context.h"
+#include "r128_lock.h"
+#include "r128_reg.h"
+#include "r128_cce.h"
+#include "r128_state.h"
+#include "r128_vb.h"
+#include "r128_clear.h"
+
+/* Clear the depth buffer */
+void r128ClearDepthBuffer(r128ContextPtr r128ctx, GLboolean all,
+			  GLint cx, GLint cy, GLint cw, GLint ch)
+{
+    __DRIdrawablePrivate *dPriv    = r128ctx->driDrawable;
+    int                   nc;
+    XF86DRIClipRectPtr    c;
+    int                   dst_bpp;
+    CARD32                write_mask;
+
+    if (!(r128ctx->regs.tex_cntl_c & R128_Z_WRITE_ENABLE)) return;
+
+    switch (r128ctx->regs.z_sten_cntl_c & R128_Z_PIX_WIDTH_MASK) {
+    case R128_Z_PIX_WIDTH_16:
+	write_mask = 0x0000ffff;
+	dst_bpp    = R128_GMC_DST_16BPP;
+	break;
+    case R128_Z_PIX_WIDTH_24:
+	write_mask = 0x00ffffff;
+	dst_bpp    = R128_GMC_DST_32BPP;
+	break;
+    case R128_Z_PIX_WIDTH_32:
+	write_mask = 0xffffffff;
+	dst_bpp    = R128_GMC_DST_32BPP;
+	break;
+    default: return;
+    }
+
+    cx += dPriv->x;
+    cy  = dPriv->y + dPriv->h - cy - ch;
+
+    LOCK_HARDWARE(r128ctx);
+
+    /* Flush any outstanding vertex buffers */
+    R128CCE_FLUSH_VB(r128ctx);
+
+    /* Init the clip rects here in case they changed during the
+       LOCK_HARDWARE macro */
+    c  = dPriv->pClipRects;
+    nc = dPriv->numClipRects;
+
+    /* Set the write mask so that we _only_ clear the Z buffer */
+    R128CCE0(R128_CCE_PACKET0, R128_DP_WRITE_MASK, 0);
+    R128CCE(write_mask);
+
+    /* Temporarily disable Z and stencil buffer and texture mapping modes */
+    R128CCE0(R128_CCE_PACKET0, R128_TEX_CNTL_C, 0);
+    R128CCE(r128ctx->regs.tex_cntl_c & ~(R128_Z_ENABLE |
+					 R128_STENCIL_ENABLE |
+					 R128_TEXMAP_ENABLE));
+
+    /* Cycle through the clip rects */
+    while (nc--) {
+	int x = c[nc].x1;
+	int y = c[nc].y1;
+	int w = c[nc].x2 - x;
+	int h = c[nc].y2 - y;
+
+	if (!all) {
+	    if (x < cx) w -= cx - x, x = cx;
+	    if (y < cy) h -= cy - y, y = cy;
+
+	    if (x + w > cx + cw) w = cx + cw - x;
+	    if (y + h > cy + ch) h = cy + ch - y;
+
+	    if (w <= 0 || h <= 0) continue;
+	}
+
+	x += r128ctx->r128Screen->depthX;
+	y += r128ctx->r128Screen->depthY;
+
+	R128CCE3(R128_CCE_PACKET3_CNTL_PAINT_MULTI, 3);
+	R128CCE(R128_GMC_BRUSH_SOLID_COLOR
+		| dst_bpp
+		| R128_GMC_SRC_DATATYPE_COLOR
+		| R128_ROP3_P
+		| R128_GMC_3D_FCN_EN            /* FIXME?? */
+		| R128_GMC_CLR_CMP_CNTL_DIS     /* FIXME?? */
+		| R128_AUX_CLIP_DIS             /* FIXME?? */
+		| R128_GMC_WR_MSK_DIS);         /* FIXME?? */
+	R128CCE(r128ctx->ClearDepth);
+	R128CCE((x << 16) | y);
+	R128CCE((w << 16) | h);
+    }
+
+#if 0
+    /* Set the write mask so that we _only_ clear the Z buffer */
+    R128CCE0(R128_CCE_PACKET0, R128_DP_WRITE_MASK, 0);
+    R128CCE(0xffffffff);
+
+    /* Restore Z and stencil buffer and texture mapping modes */
+    R128CCE0(R128_CCE_PACKET0, R128_TEX_CNTL_C, 0);
+    R128CCE(r128ctx->regs.tex_cntl_c);
+#else
+    /* FIXME: We should be able to optimize this by restoring only the
+       registers that change (above) */
+    /* NOTE: The restore of TEX_CNTL_C and R128_DP_WRITE_MASK is handled by
+    vvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvv             */
+    r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+    r128ctx->dirty_context |= R128_CTX_ALL_DIRTY;
+#endif
+
+    R128CCE_SUBMIT_PACKETS();
+
+    UNLOCK_HARDWARE(r128ctx);
+}
+
+/* Clear a color buffer */
+void r128ClearColorBuffer(r128ContextPtr r128ctx, GLboolean all,
+			  GLint cx, GLint cy, GLint cw, GLint ch,
+			  GLint drawX, GLint drawY)
+{
+    __DRIdrawablePrivate *dPriv    = r128ctx->driDrawable;
+    int                   nc;
+    XF86DRIClipRectPtr    c;
+    int                   dst_bpp;
+
+    switch (r128ctx->r128Screen->bpp) {
+    case 8:
+	dst_bpp = R128_GMC_DST_8BPP_CI;
+	break;
+    case 16:
+	if (r128ctx->r128Screen->depth == 15) dst_bpp = R128_GMC_DST_15BPP;
+	else                                  dst_bpp = R128_GMC_DST_16BPP;
+	break;
+    case 24:
+	dst_bpp = R128_GMC_DST_24BPP;
+	break;
+    case 32:
+    default:
+	dst_bpp = R128_GMC_DST_32BPP;
+	break;
+    }
+
+    cx += dPriv->x;
+    cy  = dPriv->y + dPriv->h - cy - ch;
+
+    LOCK_HARDWARE(r128ctx);
+
+    /* Flush any outstanding vertex buffers */
+    R128CCE_FLUSH_VB(r128ctx);
+
+    /* Init the clip rects here in case they changed during the
+       LOCK_HARDWARE macro */
+    c  = dPriv->pClipRects;
+    nc = dPriv->numClipRects;
+
+    /* Temporarily disable Z and stencil buffer and texture mapping modes */
+    R128CCE0(R128_CCE_PACKET0, R128_TEX_CNTL_C, 0);
+    R128CCE(r128ctx->regs.tex_cntl_c & ~(R128_Z_ENABLE |
+					 R128_STENCIL_ENABLE |
+					 R128_TEXMAP_ENABLE));
+
+    /* Cycle through the clip rects */
+    while (nc--) {
+	int x = c[nc].x1;
+	int y = c[nc].y1;
+	int w = c[nc].x2 - x;
+	int h = c[nc].y2 - y;
+
+	if (!all) {
+	    if (x < cx) w -= cx - x, x = cx;
+	    if (y < cy) h -= cy - y, y = cy;
+
+	    if (x + w > cx + cw) w = cx + cw - x;
+	    if (y + h > cy + ch) h = cy + ch - h;
+
+	    if (w <= 0 || h <= 0) continue;
+	}
+
+	x += drawX;
+	y += drawY;
+
+	R128CCE3(R128_CCE_PACKET3_CNTL_PAINT_MULTI, 3);
+	R128CCE(R128_GMC_BRUSH_SOLID_COLOR
+		| dst_bpp
+		| R128_GMC_SRC_DATATYPE_COLOR
+		| R128_ROP3_P
+		| R128_GMC_3D_FCN_EN            /* FIXME?? */
+		| R128_GMC_CLR_CMP_CNTL_DIS     /* FIXME?? */
+		| R128_AUX_CLIP_DIS);           /* FIXME?? */
+	R128CCE(r128ctx->ClearColor);
+	R128CCE((x << 16) | y);
+	R128CCE((w << 16) | h);
+    }
+
+#if 0
+    /* Restore Z and stencil buffer and texture mapping modes */
+    R128CCE0(R128_CCE_PACKET0, R128_TEX_CNTL_C, 0);
+    R128CCE(r128ctx->regs.tex_cntl_c);
+#else
+    /* FIXME: We should be able to optimize this by restoring only the
+       registers that change (above) */
+    /* NOTE: The restore of TEX_CNTL_C is handled by
+    vvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvv             */
+    r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+    r128ctx->dirty_context |= R128_CTX_ALL_DIRTY;
+#endif
+
+    R128CCE_SUBMIT_PACKETS();
+
+    UNLOCK_HARDWARE(r128ctx);
+}
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_clear.h b/xc/lib/GL/mesa/src/drv/r128/r128_clear.h
new file mode 100644
index 000000000..7018b05b4
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_clear.h
@@ -0,0 +1,47 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#ifndef _R128_CLEAR_H_
+#define _R128_CLEAR_H_
+
+#ifdef GLX_DIRECT_RENDERING
+
+extern void r128ClearDepthBuffer(r128ContextPtr r128ctx, GLboolean all,
+				 GLint x, GLint y, GLint width, GLint height);
+extern void r128ClearColorBuffer(r128ContextPtr r128ctx, GLboolean all,
+				 GLint x, GLint y, GLint width, GLint height,
+				 GLint drawX, GLint drawY);
+
+#endif
+#endif /* _R128_CLEAR_H_ */
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_context.c b/xc/lib/GL/mesa/src/drv/r128/r128_context.c
new file mode 100644
index 000000000..01573f559
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_context.c
@@ -0,0 +1,203 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas.
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#include <stdlib.h>
+
+#include "r128_init.h"
+#include "r128_mesa.h"
+#include "r128_xmesa.h"
+#include "r128_context.h"
+#include "r128_cce.h"
+#include "r128_dd.h"
+#include "r128_state.h"
+#include "r128_span.h"
+#include "r128_tex.h"
+#include "r128_vb.h"
+#include "r128_pipeline.h"
+
+#include "context.h"
+#include "simple_list.h"
+
+/* Create the device specific context */
+GLboolean r128CreateContext(Display *dpy, GLvisual *glVisual,
+			    __DRIcontextPrivate *driContextPriv)
+{
+    r128ContextPtr  r128ctx;
+    r128ScreenPtr   r128scrn;
+    GLcontext      *glCtx    = driContextPriv->mesaContext;
+    int             i;
+    char           *v;
+
+    r128ctx = (r128ContextPtr)Xmalloc(sizeof(*r128ctx));
+    if (!r128ctx) return GL_FALSE;
+
+    /* Initialize r128Context */
+    r128ctx->glCtx         = glCtx;
+    r128ctx->display       = dpy;
+    r128ctx->driContext    = driContextPriv;
+    r128ctx->driDrawable   = NULL; /* Set by XMesaMakeCurrent */
+
+    if (getenv("LIBGL_SOFTWARE_RENDERING"))
+	r128ctx->SWonly    = GL_TRUE;
+    else
+	r128ctx->SWonly    = GL_FALSE;
+    if (getenv("LIBGL_NO_SOFTWARE_FALLBACKS"))
+	r128ctx->SWfallbackDisable = GL_TRUE;
+    else
+	r128ctx->SWfallbackDisable = GL_FALSE;
+    r128scrn = r128ctx->r128Screen =
+	(r128ScreenPtr)(driContextPriv->driScreenPriv->private);
+
+    r128ctx->CurrentTexObj[0] = NULL;
+    r128ctx->CurrentTexObj[1] = NULL;
+    make_empty_list(&r128ctx->SwappedOut);
+    for (i = 0; i < r128scrn->NRTexHeaps; i++) {
+	make_empty_list(&r128ctx->TexObjList[i]);
+	r128ctx->texHeap[i] = mmInit(0, r128scrn->texSize[i]);
+	r128ctx->lastTexAge[i] = -1;
+    }
+
+    r128ctx->lastSwapAge      = 0;
+
+    r128ctx->useFastPath      = GL_FALSE;
+
+    r128ctx->vb.start         = 0;
+    r128ctx->vb.count         = 0;
+    r128ctx->vb.size          = 0;
+    r128ctx->vb.index         = 0;
+    r128ctx->vb.buf           = NULL;
+    r128ctx->vb.done          = GL_TRUE;
+
+    if (r128scrn->IsPCI || getenv("LIBGL_DISABLE_VERTEX_BUFFERS"))
+	r128ctx->disableVB    = GL_TRUE;
+    else
+	r128ctx->disableVB    = GL_FALSE;
+
+    r128ctx->CCEbuf           = (CARD32 *)malloc(sizeof(*r128ctx->CCEbuf) *
+						 r128scrn->ringEntries);
+    r128ctx->CCEcount         = 0;
+
+    if ((v = getenv("LIBGL_CCE_TIMEOUT")))
+	r128ctx->CCEtimeout   = strtoul(v, NULL, 10);
+    else
+	r128ctx->CCEtimeout   = (R128_DEFAULT_TOTAL_CCE_TIMEOUT /
+				 R128_DEFAULT_CCE_TIMEOUT);
+    if (r128ctx->CCEtimeout <= 0) r128ctx->CCEtimeout = 1;
+
+    /* Initialize GLcontext */
+    glCtx->DriverCtx = (void *)r128ctx;
+
+    r128DDInitExtensions(glCtx);
+
+    r128DDInitDriverFuncs(glCtx);
+    r128DDInitStateFuncs(glCtx);
+    r128DDInitSpanFuncs(glCtx);
+    r128DDInitTextureFuncs(glCtx);
+
+    glCtx->Driver.TriangleCaps = (DD_TRI_CULL
+				  | DD_TRI_LIGHT_TWOSIDE
+				  | DD_TRI_OFFSET);
+#if 0
+    /* FIXME */
+    glCtx->TriangleCaps |= DD_CLIP_FOG_COORD;
+#endif
+
+    /* Reset Mesa's current 2D texture pointers to the driver's textures */
+    glCtx->Shared->DefaultD[2][0].DriverData = NULL;
+    glCtx->Shared->DefaultD[2][1].DriverData = NULL;
+
+    /* If Mesa has current a vertex buffer, make sure the driver's VB
+       data is up to date */
+    if (glCtx->VB) r128DDRegisterVB(glCtx->VB);
+
+    /* Register the fast path */
+    if (glCtx->NrPipelineStages)
+	glCtx->NrPipelineStages =
+	    r128RegisterPipelineStages(glCtx->PipelineStage,
+				       glCtx->PipelineStage,
+				       glCtx->NrPipelineStages);
+
+    r128DDInitState(r128ctx);
+
+    driContextPriv->driverPrivate = (void *)r128ctx;
+
+    return GL_TRUE;
+}
+
+/* Destroy the device specific context */
+void r128DestroyContext(r128ContextPtr r128ctx)
+{
+    if (r128ctx) {
+	r128TexObjPtr t, next_t;
+	int           i;
+
+	free(r128ctx->CCEbuf);
+
+	for (i = 0; i < r128ctx->r128Screen->NRTexHeaps; i++) {
+	    foreach_s (t, next_t, &r128ctx->TexObjList[i])
+		r128DestroyTexObj(r128ctx, t);
+	}
+
+	foreach_s (t, next_t, &r128ctx->SwappedOut)
+	    r128DestroyTexObj(r128ctx, t);
+
+	gl_destroy_context(r128ctx->glCtx);
+	Xfree(r128ctx);
+    }
+}
+
+/* Load the device specific context into the hardware.  The actual
+   setting of the hardware state is done in the r128UpdateHWState(). */
+r128ContextPtr r128MakeCurrent(r128ContextPtr oldCtx,
+			       r128ContextPtr newCtx,
+			       __DRIdrawablePrivate *dPriv)
+{
+    if (oldCtx) {
+	if (!R128CCE_USE_RING_BUFFER(newCtx->r128Screen->CCEMode))
+	    newCtx->dirty         |= R128_REQUIRE_QUIESCENCE;
+	if (oldCtx != newCtx) {
+	    newCtx->dirty         |= R128_UPDATE_CONTEXT;
+	    newCtx->dirty_context |= R128_CTX_ALL_DIRTY;
+	}
+	if (oldCtx->driDrawable != dPriv)
+	    newCtx->dirty         |= R128_UPDATE_WINPOS;
+    } else {
+	newCtx->dirty             |= R128_UPDATE_CONTEXT;
+	newCtx->dirty_context     |= R128_CTX_ALL_DIRTY;
+    }
+
+    newCtx->driDrawable = dPriv;
+
+    return newCtx;
+}
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_context.h b/xc/lib/GL/mesa/src/drv/r128/r128_context.h
new file mode 100644
index 000000000..189a22cbe
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_context.h
@@ -0,0 +1,219 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas.
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#ifndef _R128_CONTEXT_H_
+#define _R128_CONTEXT_H_
+
+#ifdef GLX_DIRECT_RENDERING
+
+#include "r128_texobj.h"
+#include "r128_fastpath.h"
+#include "r128_ccevb.h"
+
+/* Flags for what needs to be updated before a new primitive is rendered */
+#define R128_CLEAN                  0x0000
+#define R128_REQUIRE_QUIESCENCE     0x0001
+#define R128_UPDATE_CONTEXT         0x0002
+#define R128_UPDATE_WINPOS          0x0004
+#define R128_UPDATE_TEX0IMAGES      0x0008
+#define R128_UPDATE_TEX1IMAGES      0x0010
+#define R128_UPDATE_TEXSTATE        0x0020
+#define R128_ALL_DIRTY              0xffff
+
+/* Flags for what context state needs to be updated */
+#define R128_CTX_CLEAN              0x0000
+#define R128_CTX_MISC               0x0001
+#define R128_CTX_ENGINESTATE        0x0002
+#define R128_CTX_TEX0STATE          0x0004
+#define R128_CTX_TEX1STATE          0x0008
+#define R128_CTX_TEXENVSTATE        0x0010
+#define R128_CTX_FOGSTATE           0x0020
+#define R128_CTX_FOGTABLE           0x0040
+#define R128_CTX_ZSTENSTATE         0x0080
+#define R128_CTX_SCISSORS           0x0100
+#define R128_CTX_ALPHASTATE         0x0200
+#define R128_CTX_SETUPSTATE         0x0400
+#define R128_CTX_WIN_Z_POS          0x0800
+#define R128_CTX_FLUSH_PIX_CACHE    0x1000
+#define R128_CTX_ALL_DIRTY          0xffff
+
+/* Flags for software fallback cases */
+#define R128_FALLBACK_TEXTURE       0x0001
+#define R128_FALLBACK_DRAW_BUFFER   0x0002
+#define R128_FALLBACK_READ_BUFFER   0x0004
+#define R128_FALLBACK_COLORMASK     0x0008
+#define R128_FALLBACK_STIPPLE       0x0010
+
+/* NOTE: The groups below need to be kept together so that a single
+   memcpy can be used to transfer data to the ring buffer */
+typedef struct {
+    CARD32  scale_3d_cntl;                        /* 0x1a00 */
+
+    CARD32  aux_sc_cntl;                          /* 0x1660 */
+    CARD32  aux1_sc_left;
+    CARD32  aux1_sc_right;
+    CARD32  aux1_sc_top;
+    CARD32  aux1_sc_bottom;
+    CARD32  aux2_sc_left;
+    CARD32  aux2_sc_right;
+    CARD32  aux2_sc_top;
+    CARD32  aux2_sc_bottom;
+    CARD32  aux3_sc_left;
+    CARD32  aux3_sc_right;
+    CARD32  aux3_sc_top;
+    CARD32  aux3_sc_bottom;                       /* 0x1690 */
+
+    CARD32  dst_pitch_offset_c;                   /* 0x1c80 */
+    CARD32  dp_gui_master_cntl;
+    CARD32  sc_top_left_c;
+    CARD32  sc_bottom_right_c;
+    CARD32  z_offset_c;
+    CARD32  z_pitch_c;
+    CARD32  z_sten_cntl_c;
+    CARD32  tex_cntl_c;
+    CARD32  misc_3d_state_cntl_reg;
+    CARD32  texture_clr_cmp_clr_c;
+    CARD32  texture_clr_cmp_msk_c;
+    CARD32  fog_color_c;
+    CARD32  prim_tex_cntl_c;
+    CARD32  prim_texture_combine_cntl_c;
+    CARD32  tex_size_pitch_c;
+    CARD32  prim_tex_offset[R128_TEX_MAXLEVELS];  /* 0x1ce4 */
+
+    CARD32  sec_tex_cntl_c;                       /* 0x1d00 */
+    CARD32  sec_tex_combine_cntl_c;
+    CARD32  sec_tex_offset[R128_TEX_MAXLEVELS];
+    CARD32  constant_color_c;
+    CARD32  prim_texture_border_color_c;
+    CARD32  sec_texture_border_color_c;
+    CARD32  sten_ref_mask_c;
+    CARD32  plane_3d_mask_c;                      /* 0x1d44 */
+
+    CARD32  setup_cntl;                           /* 0x1bc4 */
+
+    CARD32  pm4_vc_fpu_setup;                     /* 0x071c */
+
+    CARD32  fog_3d_table_start;                   /* 0x1810 */
+    CARD32  fog_3d_table_end;
+    CARD32  fog_3d_table_density;                 /* 0x181c */
+
+    CARD32  window_xy_offset;                     /* 0x1bcc */
+
+    CARD32  dp_write_mask;                        /* 0x16cc */
+
+    CARD32  pc_gui_ctlstat;                       /* 0x1748 */
+} r128ContextRegs;
+
+typedef struct {
+    GLcontext            *glCtx;	/* Mesa context */
+    int                   dirty;	/* Hardware state to be updated */
+    int                   dirty_context; /* Context state to be updated */
+
+    int                   SWonly;	/* Force software-only rendering */
+    int                   SWfallbackDisable; /* Disable software fallbacks */
+
+    r128TexObjPtr         CurrentTexObj[2]; /* Ptr to current texture
+					       object associated with
+					       each texture unit */
+					/* List of tex swapped in per heap*/
+    r128TexObj            TexObjList[R128_NR_TEX_HEAPS];
+    r128TexObj            SwappedOut;	/* List of textures swapped out */
+    memHeap_t            *texHeap[R128_NR_TEX_HEAPS]; /* Global tex heaps */
+					/* Last known global tex heap ages */
+    int                   lastTexAge[R128_NR_TEX_HEAPS];
+
+    CARD32                lastSwapAge;  /* Last known swap age */
+
+    GLenum                FogMode;      /* Current fog equation */
+
+    int                   Scissor;
+    XF86DRIClipRectRec    ScissorRect;  /* Current software scissor */
+
+    int                   useFastPath;	/* Currently using Fast Path code */
+    int                   SetupIndex;	/* Raster setup function index */
+    int                   SetupDone;	/* Partial raster setup done? */
+    int                   RenderIndex;	/* Render state function index */
+    r128InterpFunc        interp;	/* Current vert interp function */
+
+    r128CCEVertBuf        vb;		/* VB currently being filled */
+    int                   disableVB;    /* Disable the use of vertex buffers */
+
+    points_func           PointsFunc;	/* Current Points, Line, Triangle */
+    line_func             LineFunc;	/* and Quad rendering functions */
+    triangle_func         TriangleFunc;
+    quad_func             QuadFunc;
+
+    CARD32                IndirectTriangles; /* Flags for point, line,
+						tri and quad software
+						fallbacks */
+    CARD32                Fallback;     /* Need software fallback */
+
+    r128ContextRegs       regs;		/* Hardware state */
+    CARD32                Color;	/* Current draw color */
+    CARD32                ClearColor;	/* Color used to clear color buffer */
+    CARD32                ClearDepth;	/* Value used to clear depth buffer */
+
+    int                   drawX;	/* x-offset to current draw buffer */
+    int                   drawY;	/* y-offset to current draw buffer */
+
+    int                   readX;	/* x-offset to current read buffer */
+    int                   readY;	/* y-offset to current read buffer */
+
+    CARD32               *CCEbuf;	/* buffer to submit to CCE */
+    int                   CCEcount;	/* number of dwords in CCEbuf */
+
+    int                   CCEtimeout;   /* number of times to loop
+					   before exiting */
+
+    Display              *display;	/* X server display */
+
+    __DRIcontextPrivate  *driContext;	/* DRI context */
+    __DRIdrawablePrivate *driDrawable;	/* DRI drawable bound to this ctx */
+
+    r128ScreenPtr         r128Screen;	/* Screen private DRI data */
+} r128ContextRec, *r128ContextPtr;
+
+#define R128_MESACTX(r128ctx)     ((r128ctx)->glCtx)
+#define R128_DRIDRAWABLE(r128ctx) ((r128ctx)->driDrawable)
+#define R128_DRISCREEN(r128ctx)   ((r128ctx)->r128Screen->driScreen)
+
+extern GLboolean r128CreateContext(Display *dpy, GLvisual *glVisual,
+				   __DRIcontextPrivate *driContextPriv);
+extern void           r128DestroyContext(r128ContextPtr r128ctx);
+extern r128ContextPtr r128MakeCurrent(r128ContextPtr oldCtx,
+				      r128ContextPtr newCtx,
+				      __DRIdrawablePrivate *dPriv);
+
+#endif
+#endif /* _R128_CONTEXT_H_ */
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_dd.c b/xc/lib/GL/mesa/src/drv/r128/r128_dd.c
new file mode 100644
index 000000000..0ed7ae471
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_dd.c
@@ -0,0 +1,182 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#include "r128_init.h"
+#include "r128_mesa.h"
+#include "r128_xmesa.h"
+#include "r128_context.h"
+#include "r128_lock.h"
+#include "r128_cce.h"
+#include "r128_clear.h"
+#include "r128_state.h"
+#include "r128_vb.h"
+#include "r128_pipeline.h"
+#include "r128_dd.h"
+
+/* Driver entry point for clearing color and ancillary buffers */
+static GLbitfield r128DDClear(GLcontext *ctx, GLbitfield mask, GLboolean all,
+			      GLint x, GLint y, GLint width, GLint height)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+
+    if (r128ctx->SWonly) {
+	/* FIXME: Provide software fallback for this case?? */
+    }
+
+    if (mask & DD_FRONT_LEFT_BIT) {
+	r128ClearColorBuffer(r128ctx, all, x, y, width, height,
+			     r128ctx->r128Screen->fbX,
+			     r128ctx->r128Screen->fbY);
+	mask &= ~DD_FRONT_LEFT_BIT;
+    }
+
+    if (mask & DD_BACK_LEFT_BIT) {
+	r128ClearColorBuffer(r128ctx, all, x, y, width, height,
+			     r128ctx->r128Screen->backX,
+			     r128ctx->r128Screen->backY);
+	mask &= ~DD_BACK_LEFT_BIT;
+    }
+
+    if (mask & DD_DEPTH_BIT) {
+	r128ClearDepthBuffer(r128ctx, all, x, y, width, height);
+	mask &= ~DD_DEPTH_BIT;
+    }
+
+#if 0
+    /* FIXME: Add stencil support */
+    if (mask & DD_STENCIL_BIT) {
+	r128ClearStencilBuffer(r128ctx, all, x, y, width, height);
+	mask &= ~DD_STENCIL_BIT;
+    }
+#endif
+
+    return mask;
+}
+
+/* Return the current color buffer size */
+static void r128DDGetBufferSize(GLcontext *ctx, GLuint *width, GLuint *height)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+
+    *width  = r128ctx->driDrawable->w;
+    *height = r128ctx->driDrawable->h;
+}
+
+/* Return various strings for glGetString() */
+static const GLubyte *r128DDGetString(GLcontext *ctx, GLenum name)
+{
+    switch (name) {
+    case GL_VENDOR:
+	return (GLubyte *)"Precision Insight, Inc.";
+    case GL_RENDERER:
+	return (GLubyte *)"Mesa DRI Rage128 20000320";
+    default:
+	return NULL;
+    }
+}
+
+/* Send all commands to the hardware.  If vertex buffers or indirect
+   buffers are in use, then we need to make sure they are sent to the
+   hardware.  All commands that are normally sent to the ring are
+   already considered `flushed'. */
+static void r128DDFlush(GLcontext *ctx)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+
+    R128CCE_FLUSH_VB_LOCK(r128ctx);
+}
+
+/* Make sure all commands have been sent to the hardware and have
+   completed processing. */
+static void r128DDFinish(GLcontext *ctx)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+
+    r128DDFlush(ctx);
+    R128CCE_WAIT_FOR_IDLE_LOCK(r128ctx);
+}
+
+/* Return various parameters requested by Mesa */
+static GLint r128DDGetParameteri(const GLcontext *ctx, GLint param)
+{
+    switch (param) {
+#if 0
+	/* FIXME: Support for these needs to be added to Mesa */
+    case DD_MAX_TEXTURE_SIZE:  return 1024;
+    case DD_MAX_TEXTURES:      return 2;
+#endif
+#if 0
+    case DD_HAVE_HARDWARE_FOG: return 1; /* FIXME: Add HW fog support */
+#endif
+    default:                   return 0;
+    }
+}
+
+/* Initialize the extensions supported by this driver */
+void r128DDInitExtensions(GLcontext *ctx)
+{
+    /* FIXME: Are there other extensions to enable/disable??? */
+    gl_extensions_disable(ctx, "GL_EXT_shared_texture_palette");
+    gl_extensions_disable(ctx, "GL_EXT_paletted_texture");
+    gl_extensions_disable(ctx, "GL_EXT_point_parameters");
+    gl_extensions_disable(ctx, "ARB_imaging");
+    gl_extensions_disable(ctx, "GL_EXT_blend_minmax");
+    gl_extensions_disable(ctx, "GL_EXT_blend_logic_op");
+    gl_extensions_disable(ctx, "GL_EXT_blend_subtract");
+    gl_extensions_disable(ctx, "GL_INGR_blend_func_separate");
+
+    if (getenv("LIBGL_NO_MULTITEXTURE"))
+	gl_extensions_disable(ctx, "GL_ARB_multitexture");
+}
+
+/* Initialize the driver's misc functions */
+void r128DDInitDriverFuncs(GLcontext *ctx)
+{
+    ctx->Driver.Clear                = r128DDClear;
+
+    ctx->Driver.GetBufferSize        = r128DDGetBufferSize;
+    ctx->Driver.GetString            = r128DDGetString;
+    ctx->Driver.Finish               = r128DDFinish;
+    ctx->Driver.Flush                = r128DDFlush;
+
+    ctx->Driver.Error                = NULL;
+    ctx->Driver.GetParameteri        = r128DDGetParameteri;
+
+    ctx->Driver.DrawPixels           = NULL;
+    ctx->Driver.Bitmap               = NULL;
+
+    ctx->Driver.RegisterVB           = r128DDRegisterVB;
+    ctx->Driver.UnregisterVB         = r128DDUnregisterVB;
+    ctx->Driver.BuildPrecalcPipeline = r128DDBuildPrecalcPipeline;
+}
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_dd.h b/xc/lib/GL/mesa/src/drv/r128/r128_dd.h
new file mode 100644
index 000000000..91aa02586
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_dd.h
@@ -0,0 +1,44 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#ifndef _R128_DD_H_
+#define _R128_DD_H_
+
+#ifdef GLX_DIRECT_RENDERING
+
+extern void r128DDInitExtensions(GLcontext *ctx);
+extern void r128DDInitDriverFuncs(GLcontext *ctx);
+
+#endif
+#endif /* _R128_DD_H_ */
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_fastpath.c b/xc/lib/GL/mesa/src/drv/r128/r128_fastpath.c
new file mode 100644
index 000000000..330e5a23d
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_fastpath.c
@@ -0,0 +1,551 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#include "r128_init.h"
+#include "r128_mesa.h"
+#include "r128_state.h"
+#include "r128_vb.h"
+#include "r128_tris.h"
+#include "r128_fastpath.h"
+
+#include "mmath.h"
+#include "cva.h"
+#include "vertices.h"
+
+/* FIXME: These routines were copied from the i810 driver, and were only
+   slightly modified for the Rage 128.  They still need to be optmizied
+   and cleaned up.  Also, support for USE_RHW2 needs to be added. */
+
+typedef struct r128_fast_table {
+    r128BuildVerticesFunc build_vertices;
+    r128InterpFunc        interp;
+} r128FastPathTable;
+
+#define POINT(x)   r128DrawPointVB(r128ctx, &vert[x].v, psize)
+#define LINE(x,y)  r128DrawLineVB(r128ctx, &vert[x].v, &vert[y].v, lwidth)
+#define TRI(x,y,z) r128DrawTriangleVB(r128ctx, &vert[x].v, &vert[y].v, &vert[z].v)
+
+/* Direct, and no clipping required.  The clip funcs have not been
+  written yet, so this is only useful for the fast path. */
+#define RENDER_POINTS(start, count)                                       \
+do {                                                                      \
+    GLuint e;                                                             \
+    for (e = start; e <= count; e++)                                      \
+	POINT(elt[e]);                                                    \
+} while (0)
+
+#define RENDER_LINE(i1, i)                                                \
+do {                                                                      \
+    GLuint e1 = elt[i1], e = elt[i];                                      \
+    LINE(e1, e);                                                          \
+} while (0)
+
+#define RENDER_TRI(i2, i1, i, pv, parity)                                 \
+do {                                                                      \
+    GLuint e2 = elt[i2], e1 = elt[i1], e = elt[i];                        \
+    if (parity) {                                                         \
+	GLuint tmp = e2;                                                  \
+	e2 = e1;                                                          \
+	e1 = tmp;                                                         \
+    }                                                                     \
+    TRI(e2, e1, e);                                                       \
+} while (0)
+
+#define RENDER_QUAD(i3, i2, i1, i, pv)                                    \
+do {                                                                      \
+    GLuint e3 = elt[i3], e2 = elt[i2], e1 = elt[i1], e = elt[i];          \
+    TRI(e3, e2, e);                                                       \
+    TRI(e2, e1, e);                                                       \
+} while (0)
+
+#define LOCAL_VARS                                                        \
+    r128VertexPtr   vert    = R128_DRIVER_DATA(VB)->verts;                \
+    const GLuint   *elt     = VB->EltPtr->data;                           \
+    GLcontext      *ctx     = VB->ctx;                                    \
+    r128ContextPtr  r128ctx = R128_CONTEXT(ctx);                          \
+    const GLfloat   lwidth  = ctx->Line.Width;                            \
+    const GLfloat   psize   = ctx->Point.Size;                            \
+    (void) lwidth; (void)psize; (void) vert;
+
+#define TAG(x) x##_r128_smooth_indirect
+#include "render_tmp.h"
+
+
+
+#define NEGATIVE(f)          (f < 0)
+#define DIFFERENT_SIGNS(a,b) ((a*b) < 0)
+#define LINTERP(T, A, B)     ((A) + (T) * ((B) - (A)))
+
+
+#define INTERP_RGBA(t, out, a, b)                                         \
+do {                                                                      \
+    int i;                                                                \
+    for (i = 0; i < 4; i++) {                                             \
+	GLfloat fa = UBYTE_COLOR_TO_FLOAT_COLOR(a[i]);                    \
+	GLfloat fb = UBYTE_COLOR_TO_FLOAT_COLOR(b[i]);                    \
+	GLfloat fo = LINTERP(t, fa, fb);                                  \
+	FLOAT_COLOR_TO_UBYTE_COLOR(out[i], fo);                           \
+   }                                                                      \
+} while (0)
+
+
+#define CLIP(SGN, V, PLANE)                                               \
+do {                                                                      \
+    if (mask & PLANE) {                                                   \
+	GLuint   *indata  = inlist[in];                                   \
+	GLuint   *outdata = inlist[in ^= 1];                              \
+	GLuint   nr       = n;                                            \
+	GLfloat *J        = verts[indata[nr-1]].f;                        \
+	GLfloat  dpJ      = (SGN J[V]) + J[3];                            \
+                                                                          \
+	inlist[0] = vlist1;                                               \
+	for (i = n = 0 ; i < nr ; i++) {                                  \
+	    GLuint   elt_i = indata[i];                                   \
+	    GLfloat *I     = verts[elt_i].f;                              \
+	    GLfloat  dpI   = (SGN I[V]) + I[3];                           \
+                                                                          \
+	    if (DIFFERENT_SIGNS(dpI, dpJ)) {                              \
+		GLfloat *O = verts[next_vert].f;                          \
+		GLfloat t, *in, *out;                                     \
+                                                                          \
+		if (NEGATIVE(dpI)) {                                      \
+		    t = dpI / (dpI - dpJ);                                \
+		    in = I;                                               \
+		    out = J;                                              \
+		} else {                                                  \
+		    t = dpJ / (dpJ - dpI);                                \
+		    in = J;                                               \
+		    out = I;                                              \
+		}                                                         \
+                                                                          \
+		interp(t, O, in, out);                                    \
+                                                                          \
+		clipmask[next_vert] = 0;                                  \
+		outdata[n++] = next_vert++;                               \
+	    }                                                             \
+                                                                          \
+	    clipmask[elt_i] |= PLANE;      /* don't set up */             \
+                                                                          \
+	    if (!NEGATIVE(dpI)) {                                         \
+		outdata[n++] = elt_i;                                     \
+		clipmask[elt_i] &= ~PLANE; /* set up after all */         \
+	    }                                                             \
+                                                                          \
+	    J = I;                                                        \
+	    dpJ = dpI;                                                    \
+	}                                                                 \
+                                                                          \
+	if (n < 3) return;                                                \
+    }                                                                     \
+} while (0)
+
+#define LINE_CLIP(x,y,z,w,PLANE)                                          \
+do {                                                                      \
+    if (mask & PLANE) {                                                   \
+	GLfloat dpI = DOT4V(I,x,y,z,w);                                   \
+	GLfloat dpJ = DOT4V(J,x,y,z,w);                                   \
+                                                                          \
+	if (DIFFERENT_SIGNS(dpI, dpJ)) {                                  \
+	    GLfloat *O = verts[next_vert].f;                              \
+	    GLfloat  t = dpI / (dpI - dpJ);                               \
+                                                                          \
+	    interp(t, O, I, J);                                           \
+                                                                          \
+	    clipmask[next_vert] = 0;                                      \
+                                                                          \
+	    if (NEGATIVE(dpI)) {                                          \
+		clipmask[elts[0]] |= PLANE;                               \
+		I = O;                                                    \
+		elts[0] = next_vert++;                                    \
+	    } else {                                                      \
+		clipmask[elts[1]] |= PLANE;                               \
+		J = O;                                                    \
+		elts[1] = next_vert++;                                    \
+	    }                                                             \
+	} else if (NEGATIVE(dpI)) return;                                 \
+    }                                                                     \
+} while (0)
+
+
+static void r128TriClip(GLuint **p_elts,
+			r128Vertex *verts,
+			GLubyte *clipmask,
+			GLuint *p_next_vert,
+			GLubyte mask,
+			r128InterpFunc interp)
+{
+    GLuint *elts      = *p_elts;
+    GLuint  next_vert = *p_next_vert;
+    GLuint  in        = 0;
+    GLuint  n         = 3;
+    GLuint  vlist1[VB_MAX_CLIPPED_VERTS];
+    GLuint  vlist2[VB_MAX_CLIPPED_VERTS];
+    GLuint *inlist[2];
+    GLuint *out;
+    GLuint  i;
+
+    inlist[0] = elts;
+    inlist[1] = vlist2;
+
+    CLIP(-,0,CLIP_RIGHT_BIT);
+    CLIP(+,0,CLIP_LEFT_BIT);
+    CLIP(-,1,CLIP_TOP_BIT);
+    CLIP(+,1,CLIP_BOTTOM_BIT);
+    CLIP(-,2,CLIP_FAR_BIT);
+    CLIP(+,2,CLIP_NEAR_BIT);
+
+    /* Convert the planar polygon to a list of triangles */
+    out = inlist[in];
+   
+    for (i = 2 ; i < n ; i++) {
+	elts[0] = out[0];
+	elts[1] = out[i-1];
+	elts[2] = out[i];
+	elts += 3;
+    }
+
+    *p_next_vert = next_vert;
+    *p_elts      = elts;
+}
+
+
+static void r128LineClip(GLuint **p_elts,
+			 r128Vertex *verts,
+			 GLubyte *clipmask,
+			 GLuint *p_next_vert,
+			 GLubyte mask,
+			 r128InterpFunc interp)
+{
+    GLuint  *elts      = *p_elts;
+    GLfloat *I         = verts[elts[0]].f;
+    GLfloat *J         = verts[elts[1]].f;
+    GLuint   next_vert = *p_next_vert;
+
+    LINE_CLIP(1,0,0,-1,CLIP_LEFT_BIT);
+    LINE_CLIP(-1,0,0,1,CLIP_RIGHT_BIT);
+    LINE_CLIP(0,1,0,-1,CLIP_TOP_BIT);
+    LINE_CLIP(0,-1,0,1,CLIP_BOTTOM_BIT);
+    LINE_CLIP(0,0,1,-1,CLIP_FAR_BIT);
+    LINE_CLIP(0,0,-1,1,CLIP_NEAR_BIT);
+
+    *p_next_vert  = next_vert;
+    *p_elts      += 2;
+}
+
+
+
+#define CLIP_POINT(e)                                                     \
+do {                                                                      \
+   if (mask[e]) *out++ = e;                                               \
+} while (0)
+
+#define CLIP_LINE(e1, e0)                                                 \
+do {                                                                      \
+    GLubyte ormask = mask[e0] | mask[e1];                                 \
+    out[0] = e1;                                                          \
+    out[1] = e0;                                                          \
+    out += 2;                                                             \
+    if (ormask) {                                                         \
+	out-=2;                                                           \
+	if (!(mask[e0] & mask[e1])) {                                     \
+	    r128LineClip(&out, verts, mask, &next_vert, ormask, interp);  \
+	}                                                                 \
+    }                                                                     \
+} while (0)
+
+#define CLIP_TRIANGLE(e2, e1, e0)                                         \
+do {                                                                      \
+    GLubyte ormask;                                                       \
+    out[0] = e2;                                                          \
+    out[1] = e1;                                                          \
+    out[2] = e0;                                                          \
+    out += 3;                                                             \
+    ormask = mask[e2] | mask[e1] | mask[e0];                              \
+    if (ormask) {                                                         \
+	out -= 3;                                                         \
+	if (!(mask[e2] & mask[e1] & mask[e0])) {                          \
+	    r128TriClip(&out, verts, mask, &next_vert, ormask, interp);   \
+	}                                                                 \
+    }                                                                     \
+} while (0)
+
+
+
+/* Build a table of functions to clip each primitive type.  These
+ * produce a list of elements in the appropriate 'reduced' primitive,
+ * ie (points, lines, triangles) containing all the clipped and
+ * unclipped primitives from the original list.
+ */
+#define LOCAL_VARS                                                        \
+    r128ContextPtr       r128ctx   = R128_CONTEXT(VB->ctx);               \
+    r128VertexBufferPtr  r128VB    = R128_DRIVER_DATA(VB);                \
+    GLuint              *elt       = VB->EltPtr->data;                    \
+    r128Vertex          *verts     = r128VB->verts;                       \
+    GLuint               next_vert = r128VB->last_vert;                   \
+    GLuint              *out       = r128VB->clipped_elements.data;       \
+    GLubyte             *mask      = VB->ClipMask;                        \
+    r128InterpFunc       interp    = r128ctx->interp;                     \
+    (void) interp; (void) verts;
+
+#define POSTFIX                                                           \
+    r128VB->clipped_elements.count = out - r128VB->clipped_elements.data; \
+    r128VB->last_vert = next_vert;
+
+
+#define INIT(x)
+
+#define RENDER_POINTS(start, count)                                       \
+do {                                                                      \
+    GLuint i;                                                             \
+    for (i = start; i < count; i++)                                       \
+	CLIP_POINT(elt[i]);                                               \
+} while (0)
+
+#define RENDER_LINE(i1, i0)                                               \
+do {                                                                      \
+    CLIP_LINE(elt[i1], elt[i0]);                                          \
+} while (0)
+
+#define RENDER_TRI(i2, i1, i0, pv, parity)                                \
+do {                                                                      \
+    GLuint e2 = elt[i2], e1 = elt[i1], e0 = elt[i0];                      \
+    if (parity) e2 = elt[i1], e1 = elt[i2];                               \
+    CLIP_TRIANGLE(e2, e1, e0);                                            \
+} while (0)
+
+#define RENDER_QUAD(i3, i2, i1, i0, pv)                                   \
+do {                                                                      \
+  CLIP_TRIANGLE(elt[i3], elt[i2], elt[i0]);                               \
+  CLIP_TRIANGLE(elt[i2], elt[i1], elt[i0]);                               \
+} while (0)
+
+#define TAG(x) r128_clip_##x##_elt
+#include "render_tmp.h"
+
+
+/* Pack rgba and/or texture into the remaining half of a 32 byte vertex.
+ */
+#define CLIP_UBYTE_COLOR  4
+#define CLIP_UBYTE_B 0   
+#define CLIP_UBYTE_G 1  
+#define CLIP_UBYTE_R 2
+#define CLIP_UBYTE_A 3   
+#define CLIP_S0 6
+#define CLIP_T0 7
+#define CLIP_S1 8
+#define CLIP_T1 9
+
+#define TYPE (0)
+#define TAG(x) x
+#include "r128_fasttmp.h"
+
+#define TYPE (R128_RGBA_BIT)
+#define TAG(x) x##_RGBA
+#include "r128_fasttmp.h"
+
+#define TYPE (R128_TEX0_BIT)
+#define TAG(x) x##_TEX0
+#include "r128_fasttmp.h"
+
+#define TYPE (R128_RGBA_BIT | R128_TEX0_BIT)
+#define TAG(x) x##_RGBA_TEX0
+#include "r128_fasttmp.h"
+
+#define TYPE (R128_RGBA_BIT | R128_TEX0_BIT | R128_TEX1_BIT)
+#define TAG(x) x##_RGBA_TEX0_TEX1
+#include "r128_fasttmp.h"
+
+/* This one *could* get away with sneaking TEX1 into the color and
+ * specular slots, thus fitting inside a cache line.  Would be even
+ * better to switch to a smaller vertex.
+ */
+#define TYPE (R128_TEX0_BIT | R128_TEX1_BIT)
+#define TAG(x) x##_TEX0_TEX1
+#include "r128_fasttmp.h"
+
+
+
+/* Render elements directly from original list of vertices. */
+static void r128RenderElementsDirect(struct vertex_buffer *VB)
+{
+    GLcontext           *ctx     = VB->ctx;
+    r128ContextPtr       r128ctx = R128_CONTEXT(ctx);
+    GLenum               prim    = ctx->CVA.elt_mode;
+    GLuint               nr      = VB->EltPtr->count;
+    render_func          func    = render_tab_r128_smooth_indirect[prim];
+    GLuint               p       = 0;
+   
+    if (r128ctx->dirty) r128UpdateHWState(r128ctx);
+
+    do {
+	func(VB, 0, nr, 0);
+    } while (ctx->Driver.MultipassFunc &&
+	     ctx->Driver.MultipassFunc(VB, ++p));
+}
+
+/* Project vertices from clip to device space */
+static void r128ProjectVertices(struct vertex_buffer *VB)
+{
+    GLcontext           *ctx     = VB->ctx;
+    GLmatrix            *mat     = &ctx->Viewport.WindowMap;
+    r128ContextPtr       r128ctx = R128_CONTEXT(ctx);
+    r128VertexBufferPtr  r128VB  = R128_DRIVER_DATA(VB);
+    GLfloat              m[16];
+
+    m[MAT_SX] =  mat->m[MAT_SX];
+    m[MAT_TX] =  mat->m[MAT_TX];
+    m[MAT_SY] = -mat->m[MAT_SY];
+    m[MAT_TY] = -mat->m[MAT_TY] + r128ctx->driDrawable->h;
+    m[MAT_SZ] =  mat->m[MAT_SZ];
+    m[MAT_TZ] =  mat->m[MAT_TZ];
+
+    switch (ctx->Visual->DepthBits) {
+    case 16: m[MAT_SZ] /= 65536.0;      m[MAT_TZ] /= 65536.0;      break;
+    case 24: m[MAT_SZ] /= 16777216.0;   m[MAT_TZ] /= 16777216.0;   break;
+    case 32: m[MAT_SZ] /= 4294967296.0; m[MAT_TZ] /= 4294967296.0; break;
+    default: m[MAT_SZ] /= 65536.0;      m[MAT_TZ] /= 65536.0;      break;
+    }
+
+#if USE_RHW2
+    /* FIXME: Handle RHW2?? */
+    gl_project_v16(r128VB->verts[VB->CopyStart].f,
+		   r128VB->verts[r128VB->last_vert].f,
+		   m,
+		   16 * 4);
+#else
+    gl_project_v16(r128VB->verts[VB->CopyStart].f,
+		   r128VB->verts[r128VB->last_vert].f,
+		   m,
+		   16 * 4);
+#endif
+}
+
+/* Project clipped vertices from clip to device space */
+static void r128ProjectClippedVertices(struct vertex_buffer *VB)
+{
+    GLcontext           *ctx     = VB->ctx;
+    GLmatrix            *mat     = &ctx->Viewport.WindowMap;
+    r128ContextPtr       r128ctx = R128_CONTEXT(ctx);
+    r128VertexBufferPtr  r128VB  = R128_DRIVER_DATA(VB);
+    GLfloat              m[16];
+
+    m[MAT_SX] =  mat->m[MAT_SX];
+    m[MAT_TX] =  mat->m[MAT_TX];
+    m[MAT_SY] = -mat->m[MAT_SY];
+    m[MAT_TY] = -mat->m[MAT_TY] + r128ctx->driDrawable->h;
+    m[MAT_SZ] =  mat->m[MAT_SZ];
+    m[MAT_TZ] =  mat->m[MAT_TZ];
+
+    switch (ctx->Visual->DepthBits) {
+    case 16: m[MAT_SZ] /= 65536.0;      m[MAT_TZ] /= 65536.0;      break;
+    case 24: m[MAT_SZ] /= 16777216.0;   m[MAT_TZ] /= 16777216.0;   break;
+    case 32: m[MAT_SZ] /= 4294967296.0; m[MAT_TZ] /= 4294967296.0; break;
+    default: m[MAT_SZ] /= 65536.0;      m[MAT_TZ] /= 65536.0;      break;
+    }
+
+#if USE_RHW2
+    /* FIXME: Handle RHW2?? */
+    gl_project_clipped_v16(r128VB->verts[VB->CopyStart].f,
+			   r128VB->verts[r128VB->last_vert].f,
+			   m,
+			   16 * 4,
+			   VB->ClipMask + VB->CopyStart);
+#else
+    gl_project_clipped_v16(r128VB->verts[VB->CopyStart].f,
+			   r128VB->verts[r128VB->last_vert].f,
+			   m,
+			   16 * 4,
+			   VB->ClipMask + VB->CopyStart);
+#endif
+}
+
+static r128FastPathTable r128FastTab[0x80];
+
+/* Initialize the table of fast path support functions */
+void r128FastPathInit(void)
+{
+   r128_clip_render_init_elt();
+   render_init_r128_smooth_indirect();
+
+   r128_init_fastpath(&r128FastTab[0]);
+   r128_init_fastpath_RGBA(&r128FastTab[R128_RGBA_BIT]);
+   r128_init_fastpath_TEX0(&r128FastTab[R128_TEX0_BIT]);
+   r128_init_fastpath_RGBA_TEX0(&r128FastTab[R128_RGBA_BIT|R128_TEX0_BIT]);
+   r128_init_fastpath_TEX0_TEX1(&r128FastTab[R128_TEX0_BIT|R128_TEX1_BIT]);
+   r128_init_fastpath_RGBA_TEX0_TEX1(&r128FastTab[R128_RGBA_BIT|R128_TEX0_BIT|
+						  R128_TEX1_BIT]);
+}
+
+#define VALID_SETUP (R128_RGBA_BIT | R128_TEX0_BIT | R128_TEX1_BIT)
+
+void r128FastPath(struct vertex_buffer *VB)
+{
+    GLcontext         *ctx  = VB->ctx;
+    GLenum             prim = ctx->CVA.elt_mode;
+    r128ContextPtr     r128ctx = R128_CONTEXT(ctx);
+    r128FastPathTable *tab  = &r128FastTab[r128ctx->SetupIndex & VALID_SETUP];
+    GLuint             do_cliptest = 1;
+
+    gl_prepare_arrays_cva(VB);   /* still need this */
+
+    /* Reserve enough space for the pathological case */
+    if (VB->EltPtr->count * 12 > R128_DRIVER_DATA(VB)->size) {
+	r128ResizeVB(VB, VB->EltPtr->count * 12);
+	do_cliptest = 1;
+    }
+
+    tab->build_vertices(VB, do_cliptest);         /* object->clip space */
+
+    if (r128ctx->dirty) r128UpdateHWState(r128ctx);
+
+    if (VB->ClipOrMask) {
+	if (!VB->ClipAndMask) {
+	    render_func *clip = r128_clip_render_tab_elt;
+
+	    r128ctx->interp = tab->interp;
+	    clip[prim](VB, 0, VB->EltPtr->count, 0); /* build new elts */
+	    ctx->CVA.elt_mode = gl_reduce_prim[prim];
+	    VB->EltPtr = &(R128_DRIVER_DATA(VB)->clipped_elements);
+	    r128ProjectClippedVertices(VB);    /* clip->device space */
+	    r128RenderElementsDirect(VB);      /* render using new list */
+	}
+    } else {
+	r128ProjectVertices(VB);               /* clip->device space  */
+	r128RenderElementsDirect(VB);          /* render using orig list */
+    }
+
+    /* This indicates that there is no cached data to reuse */
+    VB->pipeline->data_valid = 0;
+    VB->pipeline->new_state = 0;
+}
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_fastpath.h b/xc/lib/GL/mesa/src/drv/r128/r128_fastpath.h
new file mode 100644
index 000000000..47bb92049
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_fastpath.h
@@ -0,0 +1,48 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#ifndef _R128_FASTPATH_H_
+#define _R128_FASTPATH_H_
+
+typedef void (*r128BuildVerticesFunc)(struct vertex_buffer *VB,
+				      GLuint do_cliptest);
+typedef void (*r128InterpFunc)(GLfloat t,
+			       GLfloat *result,
+			       const GLfloat *in,
+			       const GLfloat *out);
+
+extern void r128FastPathInit(void);
+extern void r128FastPath(struct vertex_buffer *VB);
+
+#endif /* _R128_FASTPATH_H_ */
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_fasttmp.h b/xc/lib/GL/mesa/src/drv/r128/r128_fasttmp.h
new file mode 100644
index 000000000..e76dd52b4
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_fasttmp.h
@@ -0,0 +1,155 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+/* FIXME: These routines were copied from the i810 driver, and were only
+   slightly modified for the Rage 128.  They still need to be optmizied
+   and cleaned up.  Also, support for USE_RHW2 needs to be added. */
+
+/* The first part of setup is applied to all vertices, clipped or
+ * unclipped.  This data w!ill be used for clipping, and then all
+ * vertices with a zero clipmask will be projected to device space.
+ *
+ * This could be split into several loops, but - it seems that the
+ * large stride of the fxVertices makes cache issues the big
+ * performance factor, and that multiple loops mean multiple cache
+ * misses....
+ */
+static void TAG(r128_setup_full)(struct vertex_buffer *VB, GLuint do_cliptest)
+{
+    GLcontext           *ctx    = VB->ctx;
+    r128VertexBufferPtr  r128VB = R128_DRIVER_DATA(VB);
+    const GLfloat       *m      = ctx->ModelProjectMatrix.m;
+    GLuint               start  = VB->CopyStart;
+    GLuint               count  = VB->Count;
+    GLuint               i;
+
+    gl_xform_points3_v16_general(r128VB->verts[start].f,
+				 m, 
+				 VB->ObjPtr->start,
+				 VB->ObjPtr->stride,
+				 count - start);
+
+    if (do_cliptest) {
+	VB->ClipAndMask = ~0; 
+	VB->ClipOrMask = 0;
+	gl_cliptest_points4_v16(r128VB->verts[start].f,
+				r128VB->verts[count].f,
+				&(VB->ClipOrMask),
+				&(VB->ClipAndMask),
+				VB->ClipMask + start);
+    }
+
+    /* These branches are all resolved at compile time.  Hopefully all
+     * the pointers are valid addresses even when not enabled.
+     */
+    if (TYPE) {
+	GLubyte *color        = VB->ColorPtr->start;
+	GLfloat *tex0_data    = VB->TexCoordPtr[0]->start;
+	GLfloat *tex1_data    = VB->TexCoordPtr[1]->start;
+
+	GLuint   color_stride = VB->ColorPtr->stride;
+	GLuint   tex0_stride  = VB->TexCoordPtr[0]->stride;
+	GLuint   tex1_stride  = VB->TexCoordPtr[1]->stride;
+
+	GLfloat *f            = r128VB->verts[start].f;
+
+	for (i = start ; i < count ; i++, f += 16) {
+	    if (TYPE & R128_RGBA_BIT) {
+		GLubyte *b = (GLubyte *)&f[CLIP_UBYTE_COLOR];
+		GLubyte *col = color; color += color_stride;
+		b[CLIP_UBYTE_R] = col[0];
+		b[CLIP_UBYTE_G] = col[1];
+		b[CLIP_UBYTE_B] = col[2];
+		b[CLIP_UBYTE_A] = col[3];
+	    }
+	    if (TYPE & R128_TEX0_BIT) {
+		f[CLIP_S0] = tex0_data[0];
+		f[CLIP_T0] = tex0_data[1];
+		STRIDE_F(tex0_data, tex0_stride);
+	    }
+	    if (TYPE & R128_TEX1_BIT) {
+		f[CLIP_S1] = tex1_data[0];
+		f[CLIP_T1] = tex1_data[1];
+		STRIDE_F(tex1_data, tex1_stride);
+	    }
+	}
+    }
+
+    r128VB->clipped_elements.count = start;
+    r128VB->last_vert = count;
+}
+
+
+/* Changed to just put the interp func instead of the whole clip
+ * routine into the header.  Less code and better chance of doing some
+ * of this stuff in assembly.
+ */
+static void TAG(r128_interp_vert)(GLfloat t, 
+				  GLfloat *O,
+				  const GLfloat *I,
+				  const GLfloat *J)
+{
+    O[0] = LINTERP(t, I[0], J[0]);
+    O[1] = LINTERP(t, I[1], J[1]);
+    O[2] = LINTERP(t, I[2], J[2]);
+    O[3] = LINTERP(t, I[3], J[3]);
+
+    if (TYPE & R128_RGBA_BIT) {
+	INTERP_RGBA(t,
+		    ((GLubyte *)&(O[4])),
+		    ((GLubyte *)&(I[4])),
+		    ((GLubyte *)&(J[4])));
+    }
+
+    if (TYPE & R128_TEX0_BIT) {
+	O[6] = LINTERP(t, I[6], J[6]);
+	O[7] = LINTERP(t, I[7], J[7]);
+    }
+
+    if (TYPE & R128_TEX1_BIT) {
+	O[8] = LINTERP(t, I[8], J[8]);
+	O[9] = LINTERP(t, I[9], J[9]);
+    }
+}
+
+
+static void TAG(r128_init_fastpath)(r128FastPathTable *tab)
+{
+   tab->build_vertices = TAG(r128_setup_full);
+   tab->interp         = TAG(r128_interp_vert);
+}
+
+#undef TYPE
+#undef TAG
+#undef SIZE
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_init.h b/xc/lib/GL/mesa/src/drv/r128/r128_init.h
new file mode 100644
index 000000000..fcaf70a51
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_init.h
@@ -0,0 +1,92 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas.
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#ifndef _R128_INIT_H_
+#define _R128_INIT_H_
+
+#ifdef GLX_DIRECT_RENDERING
+
+#include <X11/Xlibint.h>
+
+#include "types.h"
+#include "xf86drm.h"
+#include "dri_tmm.h"
+#include "dri_mesaint.h"
+
+#include "r128_screen.h"
+#include "r128_context.h"
+
+/* NOTE: The vertex buffer code is currently unstable because of the
+   switches between CCE and MMIO mode in the X server.  Fixing the X
+   server to use the CCE should fix this problem.  So, for now, it is
+   recommended that you do not use it. */
+
+#define DEBUG                       1
+#define DEBUG_LOCKING               1
+#define USE_FAST_PATH               1
+#define USE_USER_SPACE_RING         0
+
+
+#if DEBUG
+
+#include <stdio.h>
+#define R128_DEBUG(p)                                                       \
+    do {                                                                    \
+	printf p ;                                                          \
+	fflush(stdout);                                                     \
+    } while (0)
+
+extern int R128_DEBUG_FLAGS;
+
+#define DEBUG_VERBOSE_2D            0x0001
+#define DEBUG_VERBOSE_CCE           0x0008
+#define DEBUG_VERBOSE_OUTREG        0x0010
+#define DEBUG_ALWAYS_SYNC           0x0040
+#define DEBUG_VERBOSE_MSG           0x0080
+#define DEBUG_NO_OUTRING            0x0100
+#define DEBUG_NO_OUTREG             0x0200
+#define DEBUG_VERBOSE_API           0x0400
+#define DEBUG_VALIDATE_RING         0x0800
+#define DEBUG_VERBOSE_LRU           0x1000
+#define DEBUG_VERBOSE_DRI           0x2000
+#define DEBUG_VERBOSE_IOCTL         0x4000
+
+#else
+
+#define R128_DEBUG_FLAGS            0
+
+#endif
+
+#endif
+#endif /* _R128_INIT_H_ */
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_lock.h b/xc/lib/GL/mesa/src/drv/r128/r128_lock.h
new file mode 100644
index 000000000..7fd81f807
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_lock.h
@@ -0,0 +1,134 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#ifndef _R128_LOCK_H_
+#define _R128_LOCK_H_
+
+#ifdef GLX_DIRECT_RENDERING
+
+/* Turn DEBUG_LOCKING on to find locking conflicts (see r128_init.h) */
+#if DEBUG_LOCKING
+extern char *prevLockFile;
+extern int   prevLockLine;
+
+#define DEBUG_LOCK()                                                    \
+    do {                                                                \
+	prevLockFile = (__FILE__);                                      \
+	prevLockLine = (__LINE__);                                      \
+    } while (0)
+
+#define DEBUG_RESET()                                                   \
+    do {                                                                \
+	prevLockFile = 0;                                               \
+	prevLockLine = 0;                                               \
+    } while (0)
+
+#define DEBUG_CHECK_LOCK()                                              \
+    do {                                                                \
+	if (prevLockFile) {                                             \
+	    fprintf(stderr,                                             \
+		    "LOCK SET!\n\tPrevious %s:%d\n\tCurrent: %s:%d\n",  \
+		    prevLockFile, prevLockLine, __FILE__, __LINE__);    \
+	    exit(1);                                                    \
+	}                                                               \
+    } while (0)
+
+#else
+
+#define DEBUG_LOCK()
+#define DEBUG_RESET()
+#define DEBUG_CHECK_LOCK()
+
+#endif
+
+/*
+ * !!! We may want to separate locks from locks with validation.  This
+ * could be used to improve performance for those things commands that
+ * do not do any drawing !!!
+ */
+
+/* Lock the hardware using the current context */
+#define LOCK_HARDWARE(CC)                                               \
+    do {                                                                \
+	char                  __ret = 0;                                \
+	__DRIcontextPrivate  *cPriv = CC->driContext;                   \
+	__DRIscreenPrivate   *sPriv = CC->r128Screen->driScreen;        \
+                                                                        \
+	DEBUG_CHECK_LOCK();                                             \
+	DRM_CAS(&sPriv->pSAREA->lock, cPriv->hHWContext,                \
+		DRM_LOCK_HELD|cPriv->hHWContext, __ret);                \
+	if (__ret) {                                                    \
+	    /* We lost the context, so we need to request the lock from \
+               the kernel and update our state. */                      \
+	    drmGetLock(sPriv->fd, cPriv->hHWContext, 0);                \
+	    XMesaUpdateState(cPriv);                                    \
+	}                                                               \
+	DEBUG_LOCK();                                                   \
+    } while (0)
+
+/* Unlock the hardware using the current context */
+#define UNLOCK_HARDWARE(CC)                                             \
+    do {                                                                \
+	__DRIcontextPrivate  *cPriv = CC->driContext;                   \
+	__DRIscreenPrivate   *sPriv = CC->r128Screen->driScreen;        \
+                                                                        \
+	DRM_UNLOCK(sPriv->fd, &sPriv->pSAREA->lock, cPriv->hHWContext); \
+	DEBUG_RESET();                                                  \
+    } while (0)
+
+/*
+ * This pair of macros makes a loop over the drawing operations, so it
+ * is not self contained and does not have the nice single statement
+ * semantics of most macros.
+ */
+#define BEGIN_CLIP_LOOP(CC)                                             \
+    do {                                                                \
+	__DRIdrawablePrivate *_dPriv = CC->driDrawable;                 \
+	XF86DRIClipRectPtr    _pc = _dPriv->pClipRects;                 \
+	int                   _nc, _sc;                                 \
+                                                                        \
+	for (_nc = _dPriv->numClipRects; _nc > 0; _nc -= 3, _pc += 3) { \
+	    _sc = (_nc <= 3) ? _nc : 3;                                 \
+	    r128SetClipRects(CC, _pc, _sc)
+
+/* FIXME: This should be a function call to turn off aux clipping */
+#define END_CLIP_LOOP(CC)                                               \
+	    R128CCE0(R128_CCE_PACKET0, R128_AUX_SC_CNTL, 0);            \
+	    R128CCE(0x00000000);                                        \
+	    R128CCE_SUBMIT_PACKETS();                                   \
+	}                                                               \
+    } while (0)
+
+#endif
+#endif /* _R128_LOCK_H_ */
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_mesa.h b/xc/lib/GL/mesa/src/drv/r128/r128_mesa.h
new file mode 100644
index 000000000..3a697b466
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_mesa.h
@@ -0,0 +1,43 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#ifndef _R128_MESA_H_
+#define _R128_MESA_H_
+
+#ifdef GLX_DIRECT_RENDERING
+
+#define R128_CONTEXT(ctx) ((r128ContextPtr)(ctx->DriverCtx))
+
+#endif
+#endif /* _R128_MESA_H_ */
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_pipeline.c b/xc/lib/GL/mesa/src/drv/r128/r128_pipeline.c
new file mode 100644
index 000000000..55487ffcd
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_pipeline.c
@@ -0,0 +1,126 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#include "r128_init.h"
+#include "r128_mesa.h"
+#include "r128_vb.h"
+#include "r128_fastpath.h"
+#include "r128_pipeline.h"
+
+#include "types.h"
+
+static struct gl_pipeline_stage r128FastStage = {
+    "R128 Fast Path",
+    (PIPE_OP_VERT_XFORM   |
+     PIPE_OP_RAST_SETUP_0 |
+     PIPE_OP_RAST_SETUP_1 |
+     PIPE_OP_RENDER),
+    PIPE_PRECALC,
+    0, 0, 0, 0, 0, 0, 0, 0, 0,   
+    r128FastPath
+};
+
+/* Build the PRECALC pipeline with our stage, if possible.  Otherwise,
+   return GL_FALSE */
+GLboolean r128DDBuildPrecalcPipeline(GLcontext *ctx)
+{
+#if USE_FAST_PATH
+    r128ContextPtr      r128ctx = R128_CONTEXT(ctx);
+    struct gl_pipeline *pipe    = &ctx->CVA.pre;
+
+    if (r128ctx->RenderIndex == 0 &&
+	!(ctx->Enabled & (TEXTURE0_3D     |
+			  TEXTURE1_3D     |
+			  ENABLE_TEXMAT0  |
+			  ENABLE_TEXMAT1  |
+			  ENABLE_TEXGEN0  |
+			  ENABLE_TEXGEN1  |
+			  ENABLE_USERCLIP |
+			  ENABLE_LIGHT    |
+			  ENABLE_FOG)) &&
+	(ctx->Array.Flags & (VERT_OBJ_234 |
+			     VERT_TEX0_4  |
+			     VERT_TEX1_4  |
+			     VERT_ELT)) == (VERT_OBJ_23 |
+					    VERT_ELT)) {
+	pipe->stages[0]  = &r128FastStage;
+	pipe->stages[1]  = 0;
+	pipe->new_inputs = ctx->RenderFlags & VERT_DATA;
+	pipe->ops        = pipe->stages[0]->ops;
+
+	r128ctx->useFastPath = GL_TRUE;
+	return GL_TRUE;
+    }
+
+    if (r128ctx->useFastPath) {
+	r128ctx->useFastPath = GL_FALSE;
+
+	ctx->CVA.VB->ClipOrMask   = 0;
+	ctx->CVA.VB->ClipAndMask  = CLIP_ALL_BITS;
+	ctx->Array.NewArrayState |= ctx->Array.Summary;
+    }
+#endif
+
+    return GL_FALSE;
+}
+
+/* Register the pipeline with our stages included */
+GLuint r128RegisterPipelineStages(struct gl_pipeline_stage *out,
+				  const struct gl_pipeline_stage *in,
+				  GLuint nr)
+{
+#if USE_FAST_PATH
+    int i;
+
+    for (i = 0; i < nr; i++) {
+	out[i] = in[i];
+	switch (in[i].ops) {
+	case PIPE_OP_RAST_SETUP_0:
+	    out[i].cva_state_change = (NEW_LIGHTING  |
+				       NEW_TEXTURING |
+				       NEW_RASTER_OPS);
+	    out[i].state_change     = ~0;
+	    out[i].check            = r128CheckPartialRasterSetup;
+	    out[i].run              = r128PartialRasterSetup;
+	    break;
+
+	case PIPE_OP_RAST_SETUP_0|PIPE_OP_RAST_SETUP_1:
+	    out[i].run              = r128DoRasterSetup;
+	    break;
+	}
+    }
+#endif
+
+    return nr;
+}
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_pipeline.h b/xc/lib/GL/mesa/src/drv/r128/r128_pipeline.h
new file mode 100644
index 000000000..280e6b943
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_pipeline.h
@@ -0,0 +1,43 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#ifndef _R128_PIPELINE_H_
+#define _R128_PIPELINE_H_
+
+extern GLboolean r128DDBuildPrecalcPipeline(GLcontext *ctx);
+extern GLuint    r128RegisterPipelineStages(struct gl_pipeline_stage *out,
+					    const struct gl_pipeline_stage *in,
+					    GLuint nr);
+
+#endif /* _R128_PIPELINE_H_ */
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_screen.c b/xc/lib/GL/mesa/src/drv/r128/r128_screen.c
new file mode 100644
index 000000000..b385271f4
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_screen.c
@@ -0,0 +1,258 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#include "r128_dri.h"
+#include "r128_reg.h"
+
+#include "r128_init.h"
+#include "r128_context.h"
+#include "r128_xmesa.h"
+#include "r128_cce.h"
+#include "r128_tris.h"
+#include "r128_vb.h"
+#include "r128_fastpath.h"
+
+#include <sys/mman.h>
+
+/* Create the device specific screen private data struct */
+r128ScreenPtr r128CreateScreen(__DRIscreenPrivate *sPriv)
+{
+    r128ScreenPtr    r128Screen;
+    R128DRIPtr       r128DRIPriv = (R128DRIPtr)sPriv->pDevPriv;
+
+    /* Allocate the private area */
+    r128Screen = (r128ScreenPtr)Xmalloc(sizeof(*r128Screen));
+    if (!r128Screen) return NULL;
+
+    /* This is first since which regions we map depends on whether or
+       not we are using a PCI card */
+    r128Screen->IsPCI          = r128DRIPriv->IsPCI;
+
+    r128Screen->mmioRgn.handle = r128DRIPriv->registerHandle;
+    r128Screen->mmioRgn.size   = r128DRIPriv->registerSize;
+    if (drmMap(sPriv->fd,
+	       r128Screen->mmioRgn.handle,
+	       r128Screen->mmioRgn.size,
+	       (drmAddressPtr)&r128Screen->mmio)) {
+	Xfree(r128Screen);
+	return NULL;
+    }
+
+    if (!r128Screen->IsPCI) {
+	r128Screen->ringRgn.handle = r128DRIPriv->ringHandle;
+	r128Screen->ringRgn.size   = r128DRIPriv->ringMapSize;
+	if (drmMap(sPriv->fd,
+		   r128Screen->ringRgn.handle,
+		   r128Screen->ringRgn.size,
+		   (drmAddressPtr)&r128Screen->ring)) {
+	    drmUnmap((drmAddress)r128Screen->mmio, r128Screen->mmioRgn.size);
+	    Xfree(r128Screen);
+	    return NULL;
+	}
+
+	r128Screen->ringReadRgn.handle = r128DRIPriv->ringReadPtrHandle;
+	r128Screen->ringReadRgn.size   = r128DRIPriv->ringReadMapSize;
+	if (drmMap(sPriv->fd,
+		   r128Screen->ringReadRgn.handle,
+		   r128Screen->ringReadRgn.size,
+		   (drmAddressPtr)&r128Screen->ringReadPtr)) {
+	    drmUnmap((drmAddress)r128Screen->ring, r128Screen->ringRgn.size);
+	    drmUnmap((drmAddress)r128Screen->mmio, r128Screen->mmioRgn.size);
+	    Xfree(r128Screen);
+	    return NULL;
+	}
+
+	r128Screen->vbRgn.handle = r128DRIPriv->vbHandle;
+	r128Screen->vbRgn.size   = r128DRIPriv->vbMapSize;
+	r128Screen->vbOffset     = r128DRIPriv->vbOffset;
+	if (drmMap(sPriv->fd,
+		   r128Screen->vbRgn.handle,
+		   r128Screen->vbRgn.size,
+		   (drmAddressPtr)&r128Screen->vb)) {
+	    drmUnmap((drmAddress)r128Screen->ringReadPtr,
+		     r128Screen->ringReadRgn.size);
+	    drmUnmap((drmAddress)r128Screen->ring, r128Screen->ringRgn.size);
+	    drmUnmap((drmAddress)r128Screen->mmio, r128Screen->mmioRgn.size);
+	    Xfree(r128Screen);
+	    return NULL;
+	}
+	r128Screen->vbOffset     = r128DRIPriv->vbOffset;
+	r128Screen->vbBufSize    = r128DRIPriv->vbBufSize;
+
+	r128Screen->indRgn.handle = r128DRIPriv->indHandle;
+	r128Screen->indRgn.size   = r128DRIPriv->indMapSize;
+	if (drmMap(sPriv->fd,
+		   r128Screen->indRgn.handle,
+		   r128Screen->indRgn.size,
+		   (drmAddressPtr)&r128Screen->ind)) {
+	    drmUnmap((drmAddress)r128Screen->vb, r128Screen->vbRgn.size);
+	    drmUnmap((drmAddress)r128Screen->ringReadPtr,
+		     r128Screen->ringReadRgn.size);
+	    drmUnmap((drmAddress)r128Screen->ring, r128Screen->ringRgn.size);
+	    drmUnmap((drmAddress)r128Screen->mmio, r128Screen->mmioRgn.size);
+	    Xfree(r128Screen);
+	    return NULL;
+	}
+
+	r128Screen->agpTexRgn.handle = r128DRIPriv->agpTexHandle;
+	r128Screen->agpTexRgn.size   = r128DRIPriv->agpTexMapSize;
+	if (drmMap(sPriv->fd,
+		   r128Screen->agpTexRgn.handle,
+		   r128Screen->agpTexRgn.size,
+		   (drmAddressPtr)&r128Screen->agpTex)) {
+	    drmUnmap((drmAddress)r128Screen->ind,  r128Screen->indRgn.size);
+	    drmUnmap((drmAddress)r128Screen->vb,   r128Screen->vbRgn.size);
+	    drmUnmap((drmAddress)r128Screen->ringReadPtr,
+		     r128Screen->ringReadRgn.size);
+	    drmUnmap((drmAddress)r128Screen->ring, r128Screen->ringRgn.size);
+	    drmUnmap((drmAddress)r128Screen->mmio, r128Screen->mmioRgn.size);
+	    Xfree(r128Screen);
+	    return NULL;
+	}
+	r128Screen->agpTexOffset   = r128DRIPriv->agpTexOffset;
+
+	if (!(r128Screen->vbBufs = drmMapBufs(sPriv->fd))) {
+	    drmUnmap((drmAddress)r128Screen->agpTex,
+		     r128Screen->agpTexRgn.size);
+	    drmUnmap((drmAddress)r128Screen->ind,  r128Screen->indRgn.size);
+	    drmUnmap((drmAddress)r128Screen->vb,   r128Screen->vbRgn.size);
+	    drmUnmap((drmAddress)r128Screen->ringReadPtr,
+		     r128Screen->ringReadRgn.size);
+	    drmUnmap((drmAddress)r128Screen->ring, r128Screen->ringRgn.size);
+	    drmUnmap((drmAddress)r128Screen->mmio, r128Screen->mmioRgn.size);
+	    Xfree(r128Screen);
+	    return NULL;
+	}
+    }
+
+    r128Screen->deviceID         = r128DRIPriv->deviceID;
+
+    r128Screen->width            = r128DRIPriv->width;
+    r128Screen->height           = r128DRIPriv->height;
+    r128Screen->depth            = r128DRIPriv->depth;
+    r128Screen->bpp              = r128DRIPriv->bpp;
+    r128Screen->pixel_code       = (r128Screen->bpp != 16 ?
+				    r128Screen->bpp :
+				    r128Screen->depth);
+
+    r128Screen->fb               = sPriv->pFB;
+    r128Screen->fbOffset         = sPriv->fbOrigin;
+    r128Screen->fbStride         = sPriv->fbStride;
+    r128Screen->fbSize           = sPriv->fbSize;
+
+    r128Screen->fbX              = r128DRIPriv->fbX;
+    r128Screen->fbY              = r128DRIPriv->fbY;
+    r128Screen->backX            = r128DRIPriv->backX;
+    r128Screen->backY            = r128DRIPriv->backY;
+    r128Screen->depthX           = r128DRIPriv->depthX;
+    r128Screen->depthY           = r128DRIPriv->depthY;
+
+    r128Screen->texOffset[R128_LOCAL_TEX_HEAP]     = (r128DRIPriv->textureY *
+						      r128Screen->fbStride +
+						      r128DRIPriv->textureX *
+						      (r128Screen->bpp/8));
+    r128Screen->texSize[R128_LOCAL_TEX_HEAP]       = r128DRIPriv->textureSize;
+    r128Screen->log2TexGran[R128_LOCAL_TEX_HEAP]   = r128DRIPriv->log2TexGran;
+
+    if (r128Screen->IsPCI) {
+	r128Screen->texOffset[R128_AGP_TEX_HEAP]   = 0;
+	r128Screen->texSize[R128_AGP_TEX_HEAP]     = 0;
+	r128Screen->log2TexGran[R128_AGP_TEX_HEAP] = 0;
+	r128Screen->NRTexHeaps                     = R128_NR_TEX_HEAPS-1;
+    } else {
+	r128Screen->texOffset[R128_AGP_TEX_HEAP]   = 0;
+	r128Screen->texSize[R128_AGP_TEX_HEAP]     =
+	    r128DRIPriv->agpTexMapSize;
+	r128Screen->log2TexGran[R128_AGP_TEX_HEAP] =
+	    r128DRIPriv->log2AGPTexGran;
+	r128Screen->NRTexHeaps                     = R128_NR_TEX_HEAPS;
+    }
+
+#if 1
+    /* FIXME: For testing only */
+    if (getenv("LIBGL_SHOW_BUFFERS")) {
+	r128Screen->backX        = 0;
+	r128Screen->backY        = r128DRIPriv->height/2;
+	r128Screen->depthX       = r128DRIPriv->width/2;
+	r128Screen->depthY       = r128DRIPriv->height/2;
+    }
+#endif
+
+    r128Screen->CCEMode          = r128DRIPriv->CCEMode;
+    r128Screen->CCEFifoSize      = r128DRIPriv->CCEFifoSize;
+
+    r128Screen->ringEntries      = r128DRIPriv->ringSize/sizeof(CARD32);
+    if (!r128Screen->IsPCI) {
+	r128Screen->ringStartPtr = (int *)r128Screen->ring;
+	r128Screen->ringEndPtr   = (int *)(r128Screen->ring
+					   + r128DRIPriv->ringSize);
+    }
+
+    r128Screen->MMIOFifoSlots    = 0;
+    r128Screen->CCEFifoSlots     = 0;
+
+    r128Screen->CCEFifoAddr      = R128_PM4_FIFO_DATA_EVEN;
+
+    r128Screen->SAREA            = (R128SAREAPrivPtr)((char *)sPriv->pSAREA +
+						      sizeof(XF86DRISAREARec));
+
+    r128Screen->driScreen        = sPriv;
+
+    r128FastPathInit();
+    r128TriangleFuncsInit();
+    r128SetupInit();
+
+    return r128Screen;
+}
+
+/* Destroy the device specific screen private data struct */
+void r128DestroyScreen(__DRIscreenPrivate *sPriv)
+{
+    r128ScreenPtr r128Screen = (r128ScreenPtr)sPriv->private;
+
+    if (!r128Screen->IsPCI) {
+	drmUnmapBufs(r128Screen->vbBufs);
+
+	drmUnmap((drmAddress)r128Screen->agpTex, r128Screen->agpTexRgn.size);
+	drmUnmap((drmAddress)r128Screen->ind,    r128Screen->indRgn.size);
+	drmUnmap((drmAddress)r128Screen->vb,     r128Screen->vbRgn.size);
+	drmUnmap((drmAddress)r128Screen->ringReadPtr,
+		 r128Screen->ringReadRgn.size);
+	drmUnmap((drmAddress)r128Screen->ring,   r128Screen->ringRgn.size);
+    }
+    drmUnmap((drmAddress)r128Screen->mmio,   r128Screen->mmioRgn.size);
+
+    Xfree(r128Screen);
+    sPriv->private = NULL;
+}
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_screen.h b/xc/lib/GL/mesa/src/drv/r128/r128_screen.h
new file mode 100644
index 000000000..02deb64e9
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_screen.h
@@ -0,0 +1,129 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#ifndef _R128_SCREEN_H_
+#define _R128_SCREEN_H_
+
+#ifdef GLX_DIRECT_RENDERING
+
+#include "r128_sarea.h"
+
+typedef struct {
+    drmHandle  handle;			/* Handle to the DRM region */
+    drmSize    size;			/* Size of the DRM region */
+} r128RegionRec, *r128RegionPtr;
+
+typedef struct {
+    /* MMIO register data */
+    r128RegionRec       mmioRgn;
+    unsigned char      *mmio;
+
+    /* CCE ring buffer data */
+    r128RegionRec       ringRgn;
+    unsigned char      *ring;
+
+    /* CCE ring read pointer data */
+    r128RegionRec       ringReadRgn;
+
+    /* CCE vertex buffer data */
+    r128RegionRec       vbRgn;
+    unsigned char      *vb;
+    int                 vbOffset;
+    int                 vbBufSize;
+    drmBufMapPtr        vbBufs;
+
+    /* CCE indirect buffer data */
+    r128RegionRec       indRgn;
+    unsigned char      *ind;
+
+    /* CCE AGP Texture data */
+    r128RegionRec       agpTexRgn;
+    unsigned char      *agpTex;
+    int                 agpTexOffset;
+
+    /* Frame buffer data */
+    unsigned char      *fb;
+    unsigned long       fbOffset;
+    int                 fbStride;
+    int                 fbSize;
+
+    int                 IsPCI;		/* Current card is a PCI card */
+
+    int                 CCEMode;	/* CCE mode that server/clients use */
+    int                 CCEFifoSize;	/* Size of the CCE command FIFO */
+
+    /* CCE ring buffer data */
+    int                 ringEntries;
+
+    volatile int       *ringReadPtr;	/* Pointer to current read addr */
+    int                *ringStartPtr;   /* Pointer to end of ring buffer */
+    int                *ringEndPtr;     /* Pointer to end of ring buffer */
+
+    /* DRI screen private data */
+    int                 deviceID;	/* PCI device ID */
+    int                 width;		/* Width in pixels of display */
+    int                 height;		/* Height in scanlines of display */
+    int                 depth;		/* Depth of display (8, 15, 16, 24) */
+    int                 bpp;		/* Bit depth of disp (8, 16, 24, 32) */
+    int                 pixel_code;	/* 8, 15, 16, 24, 32 */
+
+    int                 fbX;		/* Start of frame buffer */
+    int                 fbY;
+    int                 backX;		/* Start of shared back buffer */
+    int                 backY;
+    int                 depthX;		/* Start of shared depth buffer */
+    int                 depthY;
+
+    /* Shared texture data */
+    int                 NRTexHeaps;
+    int                 texOffset[R128_NR_TEX_HEAPS];
+    int                 texSize[R128_NR_TEX_HEAPS];
+    int                 log2TexGran[R128_NR_TEX_HEAPS];
+
+    int                 MMIOFifoSlots;	/* Free slots in the FIFO (64 max) */
+    int                 CCEFifoSlots;	/* Free slots in the CCE FIFO */
+
+    int                 CCEFifoAddr;    /* MMIO offset to write next CCE
+					   value (only used when CCE is
+					   in PIO mode). */
+    R128SAREAPrivPtr    SAREA;		/* Pointer to SAREA private data */
+
+    __DRIscreenPrivate *driScreen;
+} r128ScreenRec, *r128ScreenPtr;
+
+r128ScreenPtr r128CreateScreen(__DRIscreenPrivate *sPriv);
+void          r128DestroyScreen(__DRIscreenPrivate *sPriv);
+
+#endif
+#endif /* _R128_SCREEN_H_ */
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_span.c b/xc/lib/GL/mesa/src/drv/r128/r128_span.c
new file mode 100644
index 000000000..b5d1774fd
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_span.c
@@ -0,0 +1,329 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *   Keith Whitwell <keith@precisioninsight.com>
+ *   Gareth Hughes <gareth@precisioninsight.com>
+ *
+ */
+
+#include "r128_init.h"
+#include "r128_mesa.h"
+#include "r128_xmesa.h"
+#include "r128_context.h"
+#include "r128_lock.h"
+#include "r128_state.h"
+#include "r128_reg.h"
+#include "r128_cce.h"
+#include "r128_span.h"
+
+#define DBG 0
+
+#define LOCAL_VARS                                                            \
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);                               \
+    r128ScreenPtr r128scrn = r128ctx->r128Screen;                             \
+    __DRIdrawablePrivate *dPriv = r128ctx->driDrawable;                       \
+    GLuint pitch = r128scrn->fbStride;                                        \
+    GLuint height = dPriv->h;                                                 \
+    char *buf = (char *)(r128scrn->fb +                                       \
+			 (r128ctx->drawX + dPriv->x) * (r128scrn->bpp/8) +    \
+			 (r128ctx->drawY + dPriv->y) * pitch);                \
+    char *read_buf = (char *)(r128scrn->fb +                                  \
+			      (r128ctx->readX + dPriv->x) * (r128scrn->bpp/8)+\
+			      (r128ctx->readY + dPriv->y) * pitch);           \
+    GLushort p;                                                               \
+    (void) read_buf; (void) buf; (void) p
+
+#define LOCAL_DEPTH_VARS                                                      \
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);                               \
+    r128ScreenPtr r128scrn = r128ctx->r128Screen;                             \
+    __DRIdrawablePrivate *dPriv = r128ctx->driDrawable;                       \
+    GLuint pitch = r128scrn->fbStride;                                        \
+    GLuint height = dPriv->h;                                                 \
+    char *buf = (char *)(r128scrn->fb +                                       \
+			 (r128scrn->depthX + dPriv->x) * (r128scrn->bpp/8) +  \
+			 (r128scrn->depthY + dPriv->y) * pitch);              \
+    (void) buf
+
+#define INIT_MONO_PIXEL(p)                                                    \
+    p = R128_CONTEXT(ctx)->Color
+
+#define CLIPPIXEL(_x, _y)                                                     \
+    ((_x >= minx) && (_x < maxx) && (_y >= miny) && (_y < maxy))
+
+
+#define CLIPSPAN(_x, _y, _n, _x1, _n1, _i)                                    \
+    if (( _y < miny) || (_y >= maxy)) {                                       \
+	_n1 = 0, _x1 = x;                                                     \
+    } else {                                                                  \
+	_n1 = _n;                                                             \
+	_x1 = _x;                                                             \
+	if (_x1 < minx) _i += (minx - _x1), _x1 = minx;                       \
+	if (_x1 + _n1 >= maxx) n1 -= (_x1 + n1 - maxx) + 1;                   \
+    }
+
+#define Y_FLIP(_y) (height - _y - 1)
+
+
+#define HW_LOCK()                                                             \
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);                               \
+    LOCK_HARDWARE(r128ctx);                                                   \
+    R128CCE_WAIT_FOR_IDLE(r128ctx);
+
+#define HW_CLIPLOOP()                                                         \
+    do {                                                                      \
+	__DRIdrawablePrivate *dPriv = r128ctx->driDrawable;                   \
+	int _nc = dPriv->numClipRects;                                        \
+                                                                              \
+	while (_nc--) {                                                       \
+	    int minx = dPriv->pClipRects[_nc].x1 - dPriv->x;                  \
+	    int miny = dPriv->pClipRects[_nc].y1 - dPriv->y;                  \
+	    int maxx = dPriv->pClipRects[_nc].x2 - dPriv->x;                  \
+	    int maxy = dPriv->pClipRects[_nc].y2 - dPriv->y;
+
+#define HW_ENDCLIPLOOP()                                                      \
+	}                                                                     \
+    } while (0)
+
+#define HW_UNLOCK()                                                           \
+    UNLOCK_HARDWARE(r128ctx)
+
+
+
+/* 16 bit, RGB565 color spanline and pixel functions */
+#define WRITE_RGBA(_x, _y, r, g, b, a)                                        \
+    *(GLushort *)(buf + _x*2 + _y*pitch) = ((((int)r & 0xf8) << 8) |          \
+					    (((int)g & 0xfc) << 3) |          \
+					    (((int)b & 0xf8) >> 3) )
+
+#define WRITE_PIXEL(_x, _y, p)                                                \
+    *(GLushort *)(buf + _x*2 + _y*pitch) = p
+
+#define READ_RGBA(rgba, _x, _y)                                               \
+    do {                                                                      \
+	GLushort p = *(GLushort *)(read_buf + _x*2 + _y*pitch);               \
+	rgba[0] = (p >> 8) & 0xf8;                                            \
+	rgba[1] = (p >> 3) & 0xfc;                                            \
+	rgba[2] = (p << 3) & 0xf8;                                            \
+	rgba[3] = 0xff;                                                       \
+    } while (0)
+
+#define TAG(x) r128##x##_RGB565
+#include "spantmp.h"
+
+
+
+/* 15 bit, ARGB1555 color spanline and pixel functions */
+#define WRITE_RGBA(_x, _y, r, g, b, a)                                        \
+    *(GLushort *)(buf + _x*2 + _y*pitch) = (((r & 0xf8) << 7) |               \
+					    ((g & 0xf8) << 2) |               \
+					    ((b & 0xf8) >> 3) |               \
+					    ((a) ? 0x8000 : 0))
+
+#define WRITE_PIXEL(_x, _y, p)                                                \
+    *(GLushort *)(buf + _x*2 + _y*pitch) = p
+
+#define READ_RGBA(rgba, _x, _y)                                               \
+    do {                                                                      \
+	GLushort p = *(GLushort *)(read_buf + _x*2 + _y*pitch);               \
+	rgba[0] = (p >> 7) & 0xf8;                                            \
+	rgba[1] = (p >> 2) & 0xf8;                                            \
+	rgba[2] = (p << 3) & 0xf8;                                            \
+	rgba[3] = (p & 0x8000) ? 0xff : 0;                                    \
+    } while (0)
+
+#define TAG(x) r128##x##_ARGB1555
+#include "spantmp.h"
+
+
+
+/* 16 bit depthbuffer functions */
+#define WRITE_DEPTH(_x, _y, d)                                                \
+    *(GLdepth *)(buf + _x*2 + _y*pitch) = d
+
+#define READ_DEPTH(d, _x, _y)                                                 \
+    d = *(GLdepth *)(buf + _x*2 + _y*pitch)
+
+#define TAG(x) r128##x##_16
+#include "depthtmp.h"
+
+
+
+/* 24 bit, RGB888 color spanline and pixel functions */
+#define WRITE_RGBA(_x, _y, r, g, b, a)                                        \
+    *(GLuint *)(buf + _x*3 + _y*pitch) = ((r << 16) |                         \
+					  (g << 8)  |                         \
+					  (b << 0))
+
+#define WRITE_PIXEL(_x, _y, p)                                                \
+    *(GLuint *)(buf + _x*3 + _y*pitch) = p
+
+#define READ_RGBA(rgba, _x, _y)                                               \
+    do {                                                                      \
+	GLuint p = *(GLuint *)(read_buf + _x*3 + _y*pitch);                   \
+	rgba[0] = (p >> 16) & 0xff;                                           \
+	rgba[1] = (p >> 8)  & 0xff;                                           \
+	rgba[2] = (p >> 0)  & 0xff;                                           \
+	rgba[3] = 0xff;                                                       \
+    } while (0)
+
+#define TAG(x) r128##x##_RGB888
+#include "spantmp.h"
+
+
+
+/* 24 bit depthbuffer functions */
+#define WRITE_DEPTH(_x, _y, d)                                                \
+    *(GLdepth *)(buf + _x*3 + _y*pitch) = d
+
+#define READ_DEPTH(d, _x, _y)                                                 \
+    d = *(GLdepth *)(buf + _x*3 + _y*pitch)
+
+#define TAG(x) r128##x##_24
+#include "depthtmp.h"
+
+
+
+/* 32 bit, ARGB8888 color spanline and pixel functions */
+#define WRITE_RGBA(_x, _y, r, g, b, a)                                        \
+    *(GLuint *)(buf + _x*4 + _y*pitch) = ((r << 16) |                         \
+					  (g << 8)  |                         \
+					  (b << 0)  |                         \
+					  (a << 24) )
+
+#define WRITE_PIXEL(_x, _y, p)                                                \
+    *(GLuint *)(buf + _x*4 + _y*pitch) = p
+
+#define READ_RGBA(rgba, _x, _y)                                               \
+    do {                                                                      \
+	GLuint p = *(GLuint *)(read_buf + _x*4 + _y*pitch);                   \
+	rgba[0] = (p >> 16) & 0xff;                                           \
+	rgba[1] = (p >> 8)  & 0xff;                                           \
+	rgba[2] = (p >> 0)  & 0xff;                                           \
+	rgba[3] = (p >> 24) & 0xff;                                           \
+    } while (0)
+
+#define TAG(x) r128##x##_ARGB8888
+#include "spantmp.h"
+
+
+
+/* 32 bit depthbuffer functions */
+#define WRITE_DEPTH(_x, _y, d)                                                \
+    *(GLdepth *)(buf + _x*4 + _y*pitch) = d
+
+#define READ_DEPTH(d, _x, _y)                                                 \
+    d = *(GLdepth *)(buf + _x*4 + _y*pitch)
+
+#define TAG(x) r128##x##_32
+#include "depthtmp.h"
+
+
+
+void r128DDInitSpanFuncs(GLcontext *ctx)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+
+    switch (r128ctx->r128Screen->pixel_code) {
+    case 8: /* Color Index mode not supported */
+	break;
+
+    case 15:
+	ctx->Driver.WriteRGBASpan       = r128WriteRGBASpan_ARGB1555;
+	ctx->Driver.WriteRGBSpan        = r128WriteRGBSpan_ARGB1555;
+	ctx->Driver.WriteMonoRGBASpan   = r128WriteMonoRGBASpan_ARGB1555;
+	ctx->Driver.WriteRGBAPixels     = r128WriteRGBAPixels_ARGB1555;
+	ctx->Driver.WriteMonoRGBAPixels = r128WriteMonoRGBAPixels_ARGB1555;
+	ctx->Driver.ReadRGBASpan        = r128ReadRGBASpan_ARGB1555;
+	ctx->Driver.ReadRGBAPixels      = r128ReadRGBAPixels_ARGB1555;
+
+	ctx->Driver.ReadDepthSpan       = r128ReadDepthSpan_16;
+	ctx->Driver.WriteDepthSpan      = r128WriteDepthSpan_16;
+	ctx->Driver.ReadDepthPixels     = r128ReadDepthPixels_16;
+	ctx->Driver.WriteDepthPixels    = r128WriteDepthPixels_16;
+	break;
+
+    case 16:
+	ctx->Driver.WriteRGBASpan       = r128WriteRGBASpan_RGB565;
+	ctx->Driver.WriteRGBSpan        = r128WriteRGBSpan_RGB565;
+	ctx->Driver.WriteMonoRGBASpan   = r128WriteMonoRGBASpan_RGB565;
+	ctx->Driver.WriteRGBAPixels     = r128WriteRGBAPixels_RGB565;
+	ctx->Driver.WriteMonoRGBAPixels = r128WriteMonoRGBAPixels_RGB565;
+	ctx->Driver.ReadRGBASpan        = r128ReadRGBASpan_RGB565;
+	ctx->Driver.ReadRGBAPixels      = r128ReadRGBAPixels_RGB565;
+
+	ctx->Driver.ReadDepthSpan       = r128ReadDepthSpan_16;
+	ctx->Driver.WriteDepthSpan      = r128WriteDepthSpan_16;
+	ctx->Driver.ReadDepthPixels     = r128ReadDepthPixels_16;
+	ctx->Driver.WriteDepthPixels    = r128WriteDepthPixels_16;
+	break;
+
+    case 24:
+	ctx->Driver.WriteRGBASpan       = r128WriteRGBASpan_RGB888;
+	ctx->Driver.WriteRGBSpan        = r128WriteRGBSpan_RGB888;
+	ctx->Driver.WriteMonoRGBASpan   = r128WriteMonoRGBASpan_RGB888;
+	ctx->Driver.WriteRGBAPixels     = r128WriteRGBAPixels_RGB888;
+	ctx->Driver.WriteMonoRGBAPixels = r128WriteMonoRGBAPixels_RGB888;
+	ctx->Driver.ReadRGBASpan        = r128ReadRGBASpan_RGB888;
+	ctx->Driver.ReadRGBAPixels      = r128ReadRGBAPixels_RGB888;
+
+	ctx->Driver.ReadDepthSpan       = r128ReadDepthSpan_24;
+	ctx->Driver.WriteDepthSpan      = r128WriteDepthSpan_24;
+	ctx->Driver.ReadDepthPixels     = r128ReadDepthPixels_24;
+	ctx->Driver.WriteDepthPixels    = r128WriteDepthPixels_24;
+	break;
+
+    case 32:
+	ctx->Driver.WriteRGBASpan       = r128WriteRGBASpan_ARGB8888;
+	ctx->Driver.WriteRGBSpan        = r128WriteRGBSpan_ARGB8888;
+	ctx->Driver.WriteMonoRGBASpan   = r128WriteMonoRGBASpan_ARGB8888;
+	ctx->Driver.WriteRGBAPixels     = r128WriteRGBAPixels_ARGB8888;
+	ctx->Driver.WriteMonoRGBAPixels = r128WriteMonoRGBAPixels_ARGB8888;
+	ctx->Driver.ReadRGBASpan        = r128ReadRGBASpan_ARGB8888;
+	ctx->Driver.ReadRGBAPixels      = r128ReadRGBAPixels_ARGB8888;
+
+	ctx->Driver.ReadDepthSpan       = r128ReadDepthSpan_32;
+	ctx->Driver.WriteDepthSpan      = r128WriteDepthSpan_32;
+	ctx->Driver.ReadDepthPixels     = r128ReadDepthPixels_32;
+	ctx->Driver.WriteDepthPixels    = r128WriteDepthPixels_32;
+	break;
+
+    default:
+	break;
+    }
+
+    ctx->Driver.WriteCI8Span      = NULL;
+    ctx->Driver.WriteCI32Span     = NULL;
+    ctx->Driver.WriteMonoCISpan   = NULL;
+    ctx->Driver.WriteCI32Pixels   = NULL;
+    ctx->Driver.WriteMonoCIPixels = NULL;
+    ctx->Driver.ReadCI32Span      = NULL;
+    ctx->Driver.ReadCI32Pixels    = NULL;
+}
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_span.h b/xc/lib/GL/mesa/src/drv/r128/r128_span.h
new file mode 100644
index 000000000..f83a4da9f
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_span.h
@@ -0,0 +1,44 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas.
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *   Gareth Hughes <gareth@precisioninsight.com>
+ *
+ */
+
+#ifndef _R128_SPAN_H_
+#define _R128_SPAN_H_
+
+#ifdef GLX_DIRECT_RENDERING
+
+extern void r128DDInitSpanFuncs(GLcontext *ctx);
+
+#endif
+#endif /* _R128_SPAN_H_ */
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_state.c b/xc/lib/GL/mesa/src/drv/r128/r128_state.c
new file mode 100644
index 000000000..3d1ceaa24
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_state.c
@@ -0,0 +1,1309 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas.
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#include "r128_init.h"
+#include "r128_mesa.h"
+#include "r128_xmesa.h"
+#include "r128_context.h"
+#include "r128_lock.h"
+#include "r128_state.h"
+#include "r128_reg.h"
+#include "r128_cce.h"
+#include "r128_tris.h"
+#include "r128_vb.h"
+#include "r128_tex.h"
+
+#include "mmath.h"
+
+#define INTERESTED (~(NEW_MODELVIEW      | \
+		      NEW_PROJECTION     | \
+		      NEW_TEXTURE_MATRIX | \
+		      NEW_USER_CLIP      | \
+		      NEW_CLIENT_STATE   | \
+		      NEW_TEXTURE_ENABLE))
+
+static void r128DDUpdateState(GLcontext *ctx)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+
+    if (R128_DEBUG_FLAGS & DEBUG_VERBOSE_API) {
+	fprintf(stderr, "r128DDUpdateState(%p)\n", ctx);
+    }
+
+    if (ctx->NewState & INTERESTED) {
+	r128ChooseRenderState(ctx);
+	r128ChooseRasterSetupFunc(ctx);
+    }
+
+    if (!r128ctx->Fallback) {
+	ctx->IndirectTriangles   &= ~DD_SW_RASTERIZE;
+	ctx->IndirectTriangles   |= r128ctx->IndirectTriangles;
+
+	ctx->Driver.PointsFunc    = r128ctx->PointsFunc;
+	ctx->Driver.LineFunc      = r128ctx->LineFunc;
+	ctx->Driver.TriangleFunc  = r128ctx->TriangleFunc;
+	ctx->Driver.QuadFunc      = r128ctx->QuadFunc;
+	ctx->Driver.RectFunc      = NULL;
+    }
+}
+
+static void r128DDUpdateHWState(GLcontext *ctx)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+
+    /* FIXME: state is being updated too often */
+    if (r128ctx->dirty)
+	r128UpdateHWState(r128ctx);
+}
+
+static void r128DDReducedPrimitiveChange(GLcontext *ctx, GLenum prim)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+
+    /* FIXME: Also need to flush between tris and tristrips/fans when we
+       support them directly */
+    R128CCE_FLUSH_VB_LOCK(r128ctx);
+}
+
+static void r128DDClearColor(GLcontext *ctx, 
+                             GLubyte r, GLubyte g, GLubyte b, GLubyte a)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+
+    r128ctx->ClearColor = r128PackColor(r128ctx->r128Screen->depth,
+					r, g, b, a);
+}
+
+static void r128DDColor(GLcontext *ctx,
+			GLubyte r, GLubyte g, GLubyte b, GLubyte a)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+
+    r128ctx->Color = r128PackColor(r128ctx->r128Screen->depth, r, g, b, a);
+}
+
+static GLboolean r128DDSetDrawBuffer(GLcontext *ctx, GLenum mode)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+    int            x       = r128ctx->driDrawable->x;
+    int            y       = r128ctx->driDrawable->y;
+    int            found;
+
+    r128ctx->Fallback &= ~R128_FALLBACK_DRAW_BUFFER;
+
+    switch (mode) {
+    case GL_FRONT_LEFT:
+	r128ctx->drawX = r128ctx->r128Screen->fbX;
+	r128ctx->drawY = r128ctx->r128Screen->fbY;
+	found = GL_TRUE;
+	break;
+    case GL_BACK_LEFT:
+	r128ctx->drawX = r128ctx->r128Screen->backX;
+	r128ctx->drawY = r128ctx->r128Screen->backY;
+	found = GL_TRUE;
+	break;
+    default:
+	r128ctx->Fallback |= R128_FALLBACK_DRAW_BUFFER;
+	found = GL_FALSE;
+	break;
+    }
+
+    x += r128ctx->drawX;
+    y += r128ctx->drawY;
+
+    R128CCE_FLUSH_VB_LOCK(r128ctx);
+    r128ctx->regs.window_xy_offset = ((y << R128_WINDOW_Y_SHIFT) |
+				      (x << R128_WINDOW_X_SHIFT));
+
+    /* Recalculate the Z buffer offset since we might be drawing to the
+       back buffer and window_xy_offset affects both color buffer and
+       depth drawing */
+    r128ctx->regs.z_offset_c        = ((r128ctx->r128Screen->depthX -
+					r128ctx->drawX) *
+				       (r128ctx->r128Screen->bpp/8) +
+				       (r128ctx->r128Screen->depthY -
+					r128ctx->drawY) *
+				       r128ctx->r128Screen->fbStride);
+
+    r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+    r128ctx->dirty_context |= R128_CTX_WIN_Z_POS;
+    return found;
+}
+
+static void r128DDSetReadBuffer(GLcontext *ctx,
+				GLframebuffer *colorBuffer,
+				GLenum mode)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+
+    r128ctx->Fallback &= ~R128_FALLBACK_READ_BUFFER;
+
+    switch (mode) {
+    case GL_FRONT_LEFT:
+	r128ctx->readX = r128ctx->r128Screen->fbX;
+	r128ctx->readY = r128ctx->r128Screen->fbY;
+	break;
+    case GL_BACK_LEFT:
+	r128ctx->readX = r128ctx->r128Screen->backX;
+	r128ctx->readY = r128ctx->r128Screen->backY;
+	break;
+    default:
+	r128ctx->Fallback |= R128_FALLBACK_READ_BUFFER;
+	break;
+    }
+}
+
+static GLboolean r128DDColorMask(GLcontext *ctx,
+				 GLboolean r, GLboolean g,
+				 GLboolean b, GLboolean a)
+{
+    r128ContextPtr r128ctx  = R128_CONTEXT(ctx);
+
+    GLuint mask = r128PackColor(r128ctx->r128Screen->pixel_code,
+				ctx->Color.ColorMask[RCOMP],
+				ctx->Color.ColorMask[GCOMP],
+				ctx->Color.ColorMask[BCOMP],
+				ctx->Color.ColorMask[ACOMP]);
+
+    if (r128ctx->regs.plane_3d_mask_c != mask) {
+	R128CCE_FLUSH_VB_LOCK(r128ctx);
+	r128ctx->regs.plane_3d_mask_c = mask;
+
+	/* FIXME: Load into hardware now??? */
+	r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+	r128ctx->dirty_context |= R128_CTX_MISC;
+    }
+
+    return GL_TRUE;
+}
+
+static void r128DDDither(GLcontext *ctx, GLboolean enable)
+{
+}
+
+static void r128DDAlphaFunc(GLcontext *ctx, GLenum func, GLclampf ref)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+    CARD32         a       = r128ctx->regs.misc_3d_state_cntl_reg;
+
+    a &= ~(R128_ALPHA_TEST_MASK | R128_REF_ALPHA_MASK);
+    a |= ctx->Color.AlphaRef & R128_REF_ALPHA_MASK;
+
+    switch (func) {
+    case GL_NEVER:    a |= R128_ALPHA_TEST_NEVER;        break;
+    case GL_LESS:     a |= R128_ALPHA_TEST_LESS;         break;
+    case GL_LEQUAL:   a |= R128_ALPHA_TEST_LESSEQUAL;    break;
+    case GL_EQUAL:    a |= R128_ALPHA_TEST_EQUAL;        break;
+    case GL_GEQUAL:   a |= R128_ALPHA_TEST_GREATEREQUAL; break;
+    case GL_GREATER:  a |= R128_ALPHA_TEST_GREATER;      break;
+    case GL_NOTEQUAL: a |= R128_ALPHA_TEST_NEQUAL;       break;
+    case GL_ALWAYS:   a |= R128_ALPHA_TEST_ALWAYS;
+	break;
+    default:
+	/* ERROR!!! */
+	return;
+    }
+
+    if (r128ctx->regs.misc_3d_state_cntl_reg != a) {
+	R128CCE_FLUSH_VB_LOCK(r128ctx);
+	r128ctx->regs.misc_3d_state_cntl_reg = a;
+
+	/* FIXME: Load into hardware now??? */
+	r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+	r128ctx->dirty_context |= R128_CTX_ALPHASTATE;
+    }
+}
+
+static void r128DDBlendFunc(GLcontext *ctx, GLenum sfactor, GLenum dfactor)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+    CARD32         b       = r128ctx->regs.misc_3d_state_cntl_reg;
+
+    b &= ~(R128_ALPHA_BLEND_SRC_MASK | R128_ALPHA_BLEND_DST_MASK);
+
+    switch (sfactor) {
+    case GL_ZERO:                     b |= R128_ALPHA_BLEND_SRC_ZERO;
+	break;
+    case GL_ONE:                      b |= R128_ALPHA_BLEND_SRC_ONE;
+	break;
+    case GL_DST_COLOR:                b |= R128_ALPHA_BLEND_SRC_DESTCOLOR;
+	break;
+    case GL_ONE_MINUS_DST_COLOR:      b |= R128_ALPHA_BLEND_SRC_INVDESTCOLOR;
+	break;
+    case GL_SRC_ALPHA:                b |= R128_ALPHA_BLEND_SRC_SRCALPHA;
+	break;
+    case GL_ONE_MINUS_SRC_ALPHA:      b |= R128_ALPHA_BLEND_SRC_INVSRCALPHA;
+	break;
+    case GL_DST_ALPHA:                b |= R128_ALPHA_BLEND_SRC_DESTALPHA;
+	break;
+    case GL_ONE_MINUS_DST_ALPHA:      b |= R128_ALPHA_BLEND_SRC_INVDESTALPHA;
+	break;
+    case GL_SRC_ALPHA_SATURATE:       b |= R128_ALPHA_BLEND_SRC_SRCALPHASAT;
+	break;
+#if 0
+	/* FIXME: These are not supported directly by the Rage 128.
+           They could be emulated using something like the TexEnv
+           modes. */
+    case GL_CONSTANT_COLOR:           b |= 0;
+	break;
+    case GL_ONE_MINUS_CONSTANT_COLOR: b |= 0;
+	break;
+    case GL_CONSTANT_ALPHA:           b |= 0;
+	break;
+    case GL_ONE_MINUS_CONSTANT_ALPHA: b |= 0;
+	break;
+#endif
+    default:
+	/* ERROR!!! */
+	return;
+    }
+
+    switch (dfactor) {
+    case GL_ZERO:                     b |= R128_ALPHA_BLEND_DST_ZERO;
+	break;
+    case GL_ONE:                      b |= R128_ALPHA_BLEND_DST_ONE;
+	break;
+    case GL_SRC_COLOR:                b |= R128_ALPHA_BLEND_DST_SRCCOLOR;
+	break;
+    case GL_ONE_MINUS_SRC_COLOR:      b |= R128_ALPHA_BLEND_DST_INVSRCCOLOR;
+	break;
+    case GL_SRC_ALPHA:                b |= R128_ALPHA_BLEND_DST_SRCALPHA;
+	break;
+    case GL_ONE_MINUS_SRC_ALPHA:      b |= R128_ALPHA_BLEND_DST_INVSRCALPHA;
+	break;
+    case GL_DST_ALPHA:                b |= R128_ALPHA_BLEND_DST_DESTALPHA;
+	break;
+    case GL_ONE_MINUS_DST_ALPHA:      b |= R128_ALPHA_BLEND_DST_INVDESTALPHA;
+	break;
+#if 0
+	/* FIXME: These are not supported directly by the Rage 128.
+           They could be emulated using something like the TexEnv
+           modes. */
+    case GL_CONSTANT_COLOR:           b |= 0;
+	break;
+    case GL_ONE_MINUS_CONSTANT_COLOR: b |= 0;
+	break;
+    case GL_CONSTANT_ALPHA:           b |= 0;
+	break;
+    case GL_ONE_MINUS_CONSTANT_ALPHA: b |= 0;
+	break;
+#endif
+    default:
+	/* ERROR!!! */
+	return;
+    }
+
+    if (r128ctx->regs.misc_3d_state_cntl_reg != b) {
+	R128CCE_FLUSH_VB_LOCK(r128ctx);
+	r128ctx->regs.misc_3d_state_cntl_reg = b;
+
+	/* FIXME: Load into hardware now??? */
+	r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+	r128ctx->dirty_context |= R128_CTX_ALPHASTATE;
+    }
+}
+
+static void r128DDBlendFuncSeparate(GLcontext *ctx, GLenum sfactorRGB,
+				    GLenum dfactorRGB, GLenum sfactorA,
+				    GLenum dfactorA)
+{
+    if (sfactorRGB != sfactorA || dfactorRGB != dfactorA) {
+	/* ERROR!!! */
+	return;
+    }
+
+    r128DDBlendFunc(ctx, sfactorRGB, dfactorRGB);
+}
+
+static void r128DDClearDepth(GLcontext *ctx, GLclampd d)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+
+    switch (r128ctx->regs.z_sten_cntl_c & R128_Z_PIX_WIDTH_MASK) {
+    case R128_Z_PIX_WIDTH_16: r128ctx->ClearDepth = d * 0x0000ffff; break;
+    case R128_Z_PIX_WIDTH_24: r128ctx->ClearDepth = d * 0x00ffffff; break;
+    case R128_Z_PIX_WIDTH_32: r128ctx->ClearDepth = d * 0xffffffff; break;
+    default: return;
+    }
+}
+
+static void r128DDCullFace(GLcontext *ctx, GLenum mode)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+    CARD32         f       = r128ctx->regs.pm4_vc_fpu_setup;
+
+    if (!ctx->Polygon.CullFlag) return;
+
+    f &= ~(R128_BACKFACE_MASK | R128_FRONTFACE_MASK);
+
+    switch (mode) {
+    case GL_FRONT:          f |= R128_BACKFACE_SOLID;  break;
+    case GL_BACK:           f |= R128_FRONTFACE_SOLID; break;
+    case GL_FRONT_AND_BACK:                            break;
+    default: return;
+    }
+
+    if (r128ctx->regs.pm4_vc_fpu_setup != f) {
+	R128CCE_FLUSH_VB_LOCK(r128ctx);
+	r128ctx->regs.pm4_vc_fpu_setup = f;
+
+	/* FIXME: Load into hardware now??? */
+	r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+	r128ctx->dirty_context |= R128_CTX_SETUPSTATE;
+    }
+}
+
+static void r128DDFrontFace(GLcontext *ctx, GLenum mode)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+    CARD32         f       = r128ctx->regs.pm4_vc_fpu_setup;
+
+    f &= ~R128_FRONT_DIR_MASK;
+
+    switch (mode) {
+    case GL_CW:  f |= R128_FRONT_DIR_CW;  break;
+    case GL_CCW: f |= R128_FRONT_DIR_CCW; break;
+    default: return;
+    }
+
+    if (r128ctx->regs.pm4_vc_fpu_setup != f) {
+	R128CCE_FLUSH_VB_LOCK(r128ctx);
+	r128ctx->regs.pm4_vc_fpu_setup = f;
+
+	/* FIXME: Load into hardware now??? */
+	r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+	r128ctx->dirty_context |= R128_CTX_SETUPSTATE;
+    }
+}
+
+static void r128DDDepthFunc(GLcontext *ctx, GLenum func)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+    CARD32         z       = r128ctx->regs.z_sten_cntl_c;
+
+    z &= ~R128_Z_TEST_MASK;
+
+    switch (func) {
+    case GL_NEVER:    z |= R128_Z_TEST_NEVER;        break;
+    case GL_LESS:     z |= R128_Z_TEST_LESS;         break;
+    case GL_LEQUAL:   z |= R128_Z_TEST_LESSEQUAL;    break;
+    case GL_EQUAL:    z |= R128_Z_TEST_EQUAL;        break;
+    case GL_GEQUAL:   z |= R128_Z_TEST_GREATEREQUAL; break;
+    case GL_GREATER:  z |= R128_Z_TEST_GREATER;      break;
+    case GL_NOTEQUAL: z |= R128_Z_TEST_NEQUAL;       break;
+    case GL_ALWAYS:   z |= R128_Z_TEST_ALWAYS;       break;
+    default: return;
+    }
+
+    if (r128ctx->regs.z_sten_cntl_c != z) {
+	R128CCE_FLUSH_VB_LOCK(r128ctx);
+	r128ctx->regs.z_sten_cntl_c = z;
+
+	/* FIXME: Load into hardware now??? */
+	r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+	r128ctx->dirty_context |= R128_CTX_ZSTENSTATE;
+    }
+}
+
+static void r128DDDepthMask(GLcontext *ctx, GLboolean flag)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+    CARD32         t       = r128ctx->regs.tex_cntl_c;
+
+    if (flag) t |=  R128_Z_WRITE_ENABLE;
+    else      t &= ~R128_Z_WRITE_ENABLE;
+
+    if (r128ctx->regs.tex_cntl_c != t) {
+	R128CCE_FLUSH_VB_LOCK(r128ctx);
+	r128ctx->regs.tex_cntl_c = t;
+
+	/* FIXME: Load into hardware now??? */
+	r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+	r128ctx->dirty_context |= R128_CTX_ENGINESTATE;
+    }
+}
+
+static void r128DDEnable(GLcontext *ctx, GLenum cap, GLboolean state)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+    CARD32         t       = r128ctx->regs.tex_cntl_c;
+    CARD32         f       = r128ctx->regs.pm4_vc_fpu_setup;
+
+    if (R128_DEBUG_FLAGS & DEBUG_VERBOSE_API) {
+	fprintf(stderr, "r128DDEnable( %p, 0x%x = %s )\n",
+		ctx, cap, state ? "GL_TRUE" : "GL_FALSE");
+    }
+
+    switch (cap) {
+    case GL_ALPHA_TEST:
+	if (state) t |=  R128_ALPHA_TEST_ENABLE;
+	else       t &= ~R128_ALPHA_TEST_ENABLE;
+	break;
+
+    case GL_AUTO_NORMAL: return;
+
+    case GL_BLEND:
+	if (state) t |=  R128_ALPHA_ENABLE;
+	else       t &= ~R128_ALPHA_ENABLE;
+	break;
+
+    case GL_CLIP_PLANE0:
+    case GL_CLIP_PLANE1:
+    case GL_CLIP_PLANE2:
+    case GL_CLIP_PLANE3:
+    case GL_CLIP_PLANE4:
+    case GL_CLIP_PLANE5:
+    case GL_COLOR_MATERIAL: return;
+
+    case GL_CULL_FACE:
+	f &= ~(R128_BACKFACE_MASK | R128_FRONTFACE_MASK);
+	if (state) {
+	    switch (ctx->Polygon.CullFaceMode) {
+	    case GL_FRONT:          f |= R128_BACKFACE_SOLID;  break;
+	    case GL_BACK:           f |= R128_FRONTFACE_SOLID; break;
+	    case GL_FRONT_AND_BACK:                            break;
+	    default: return;
+	    }
+	} else {
+	    f |= R128_BACKFACE_SOLID | R128_FRONTFACE_SOLID;
+	}
+	break;
+
+    case GL_DEPTH_TEST:
+	if (state) t |=  R128_Z_ENABLE;
+	else       t &= ~R128_Z_ENABLE;
+	break;
+
+    case GL_DITHER:
+	if (state) t |=  R128_DITHER_ENABLE;
+	else       t &= ~R128_DITHER_ENABLE;
+	break;
+
+    case GL_FOG:
+	if (state) t |=  R128_FOG_ENABLE;
+	else       t &= ~R128_FOG_ENABLE;
+	break;
+
+    case GL_LIGHT0:
+    case GL_LIGHT1:
+    case GL_LIGHT2:
+    case GL_LIGHT3:
+    case GL_LIGHT4:
+    case GL_LIGHT5:
+    case GL_LIGHT6:
+    case GL_LIGHT7:
+    case GL_LIGHTING:
+    case GL_LINE_SMOOTH:
+    case GL_LINE_STIPPLE:
+    case GL_INDEX_LOGIC_OP:
+    case GL_COLOR_LOGIC_OP:
+    case GL_MAP1_COLOR_4:
+    case GL_MAP1_INDEX:
+    case GL_MAP1_NORMAL:
+    case GL_MAP1_TEXTURE_COORD_1:
+    case GL_MAP1_TEXTURE_COORD_2:
+    case GL_MAP1_TEXTURE_COORD_3:
+    case GL_MAP1_TEXTURE_COORD_4:
+    case GL_MAP1_VERTEX_3:
+    case GL_MAP1_VERTEX_4:
+    case GL_MAP2_COLOR_4:
+    case GL_MAP2_INDEX:
+    case GL_MAP2_NORMAL:
+    case GL_MAP2_TEXTURE_COORD_1:
+    case GL_MAP2_TEXTURE_COORD_2:
+    case GL_MAP2_TEXTURE_COORD_3:
+    case GL_MAP2_TEXTURE_COORD_4:
+    case GL_MAP2_VERTEX_3:
+    case GL_MAP2_VERTEX_4:
+    case GL_NORMALIZE:
+    case GL_POINT_SMOOTH:
+    case GL_POLYGON_SMOOTH:
+    case GL_POLYGON_STIPPLE:
+    case GL_POLYGON_OFFSET_POINT:
+    case GL_POLYGON_OFFSET_LINE:
+    case GL_POLYGON_OFFSET_FILL:
+    case GL_RESCALE_NORMAL_EXT: return;
+
+    case GL_SCISSOR_TEST:
+	/* FIXME: Hook up the software scissor */
+#if 0
+	r128ctx->Scissor = state;
+#endif
+	break;
+
+    case GL_SHARED_TEXTURE_PALETTE_EXT:
+    case GL_STENCIL_TEST: return;
+
+    case GL_TEXTURE_1D:
+    case GL_TEXTURE_2D:
+	/* This is handled in r128UpdateTex[01]State() */
+	r128ctx->dirty |= R128_UPDATE_TEXSTATE;
+	break;
+
+    case GL_TEXTURE_3D:
+    case GL_TEXTURE_GEN_Q:
+    case GL_TEXTURE_GEN_R:
+    case GL_TEXTURE_GEN_S:
+    case GL_TEXTURE_GEN_T: return;
+
+	/* Client state */
+    case GL_VERTEX_ARRAY:
+    case GL_NORMAL_ARRAY:
+    case GL_COLOR_ARRAY:
+    case GL_INDEX_ARRAY:
+    case GL_TEXTURE_COORD_ARRAY:
+    case GL_EDGE_FLAG_ARRAY: return;
+
+    default: return;
+    }
+
+    if (r128ctx->regs.tex_cntl_c != t) {
+	R128CCE_FLUSH_VB_LOCK(r128ctx);
+	r128ctx->regs.tex_cntl_c = t;
+
+	/* FIXME: Load into hardware now??? */
+	r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+	r128ctx->dirty_context |= R128_CTX_ENGINESTATE;
+    }
+    if (r128ctx->regs.pm4_vc_fpu_setup != f) {
+	R128CCE_FLUSH_VB_LOCK(r128ctx);
+	r128ctx->regs.pm4_vc_fpu_setup = f;
+
+	/* FIXME: Load into hardware now??? */
+	r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+	r128ctx->dirty_context |= R128_CTX_SETUPSTATE;
+    }
+
+}
+
+static void r128DDFogfv(GLcontext *ctx, GLenum pname, const GLfloat *param)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+    GLubyte        c[4];
+    CARD32         col;
+    floatTOint     fog;
+    GLenum         mode;
+
+    if (R128_DEBUG_FLAGS & DEBUG_VERBOSE_API) {
+	fprintf(stderr, "r128DDFogfv(%p, 0x%x)\n", ctx, pname);
+    }
+
+    switch (pname) {
+    case GL_FOG_MODE:
+	mode = (GLenum)(GLint)*param;
+	if (r128ctx->FogMode != mode) {
+	    R128CCE_FLUSH_VB_LOCK(r128ctx);
+	    r128ctx->FogMode = mode;
+
+	    /* FIXME: Load into hardware now??? */
+	    r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+	    r128ctx->dirty_context |= R128_CTX_FOGTABLE;
+	}
+	break;
+
+    case GL_FOG_DENSITY:
+	fog.f = *param;
+	if (r128ctx->regs.fog_3d_table_density != fog.i) {
+	    R128CCE_FLUSH_VB_LOCK(r128ctx);
+	    r128ctx->regs.fog_3d_table_density = fog.i;
+
+	    /* FIXME: Load into hardware now??? */
+	    r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+	    r128ctx->dirty_context |= R128_CTX_FOGSTATE;
+	}
+	break;
+
+    case GL_FOG_START:
+	fog.f = *param;
+	if (r128ctx->regs.fog_3d_table_start != fog.i) {
+	    R128CCE_FLUSH_VB_LOCK(r128ctx);
+	    r128ctx->regs.fog_3d_table_start = fog.i;
+
+	    /* FIXME: Load into hardware now??? */
+	    r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+	    r128ctx->dirty_context |= R128_CTX_FOGSTATE;
+	}
+	break;
+
+    case GL_FOG_END:
+	fog.f = *param;
+	if (r128ctx->regs.fog_3d_table_end != fog.i) {
+	    R128CCE_FLUSH_VB_LOCK(r128ctx);
+	    r128ctx->regs.fog_3d_table_end = fog.i;
+
+	    /* FIXME: Load into hardware now??? */
+	    r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+	    r128ctx->dirty_context |= R128_CTX_FOGSTATE;
+	}
+	break;
+
+    case GL_FOG_COLOR:
+	FLOAT_RGBA_TO_UBYTE_RGBA(c, ctx->Fog.Color);
+	col = r128PackColor(32, c[0], c[1], c[2], c[3]);
+	if (r128ctx->regs.fog_color_c != col) {
+	    R128CCE_FLUSH_VB_LOCK(r128ctx);
+	    r128ctx->regs.fog_color_c = col;
+
+	    /* FIXME: Load into hardware now??? */
+	    r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+	    r128ctx->dirty_context |= R128_CTX_FOGSTATE;
+	}
+	break;
+
+    default:
+	return;
+    }
+}
+
+static void r128DDScissor(GLcontext *ctx,
+			  GLint x, GLint y, GLsizei w, GLsizei h)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+
+    r128ctx->ScissorRect.x1 = x;
+    r128ctx->ScissorRect.y1 = r128ctx->driDrawable->h - (y + h);
+    r128ctx->ScissorRect.x2 = x + w;
+    r128ctx->ScissorRect.y2 = r128ctx->driDrawable->h - y;
+}
+
+/* Initialize the driver's state functions */
+void r128DDInitStateFuncs(GLcontext *ctx)
+{
+    ctx->Driver.UpdateState            = r128DDUpdateState;
+
+    ctx->Driver.ClearIndex             = NULL;
+    ctx->Driver.ClearColor             = r128DDClearColor;
+    ctx->Driver.Index                  = NULL;
+    ctx->Driver.Color                  = r128DDColor;
+    ctx->Driver.SetDrawBuffer          = r128DDSetDrawBuffer;
+    ctx->Driver.SetReadBuffer          = r128DDSetReadBuffer;
+
+    ctx->Driver.IndexMask              = NULL;
+    ctx->Driver.ColorMask              = r128DDColorMask;
+    ctx->Driver.LogicOp                = NULL;
+    ctx->Driver.Dither                 = r128DDDither;
+
+    ctx->Driver.NearFar                = NULL;
+
+    ctx->Driver.RenderStart            = r128DDUpdateHWState;
+    ctx->Driver.RenderFinish           = NULL;
+    ctx->Driver.RasterSetup            = NULL;
+
+    ctx->Driver.RenderVBClippedTab     = NULL;
+    ctx->Driver.RenderVBCulledTab      = NULL;
+    ctx->Driver.RenderVBRawTab         = NULL;
+
+    ctx->Driver.ReducedPrimitiveChange = r128DDReducedPrimitiveChange;
+    ctx->Driver.MultipassFunc          = NULL;
+
+    ctx->Driver.AlphaFunc              = r128DDAlphaFunc;
+    ctx->Driver.BlendEquation          = NULL;
+    ctx->Driver.BlendFunc              = r128DDBlendFunc;
+    ctx->Driver.BlendFuncSeparate      = r128DDBlendFuncSeparate;
+    ctx->Driver.ClearDepth             = r128DDClearDepth;
+    ctx->Driver.CullFace               = r128DDCullFace;
+    ctx->Driver.FrontFace              = r128DDFrontFace;
+    ctx->Driver.DepthFunc              = r128DDDepthFunc;
+    ctx->Driver.DepthMask              = r128DDDepthMask;
+    ctx->Driver.DepthRange             = NULL;
+    ctx->Driver.Enable                 = r128DDEnable;
+    ctx->Driver.Fogfv                  = r128DDFogfv;
+    ctx->Driver.Hint                   = NULL;
+    ctx->Driver.Lightfv                = NULL;
+    ctx->Driver.LightModelfv           = NULL;
+    ctx->Driver.PolygonMode            = NULL;
+    ctx->Driver.Scissor                = r128DDScissor;
+    ctx->Driver.ShadeModel             = NULL;
+    ctx->Driver.ClearStencil           = NULL;
+    ctx->Driver.StencilFunc            = NULL;
+    ctx->Driver.StencilMask            = NULL;
+    ctx->Driver.StencilOp              = NULL;
+    ctx->Driver.Viewport               = NULL;
+}
+
+/* Initialize the context's hardware state */
+void r128DDInitState(r128ContextPtr r128ctx)
+{
+    int    dst_bpp, depth_bpp, pitch, i;
+    CARD32 depthClear;
+
+    pitch = r128ctx->r128Screen->fbStride / r128ctx->r128Screen->bpp;
+
+    switch (r128ctx->r128Screen->pixel_code) {
+    case  8: dst_bpp = R128_GMC_DST_8BPP_CI; break;
+    case 15: dst_bpp = R128_GMC_DST_15BPP;   break;
+    case 16: dst_bpp = R128_GMC_DST_16BPP;   break;
+    case 24: dst_bpp = R128_GMC_DST_24BPP;   break;
+    case 32: dst_bpp = R128_GMC_DST_32BPP;   break;
+    default:
+	fprintf(stderr, "Error: Unsupported pixel depth %d... exiting\n",
+		r128ctx->r128Screen->pixel_code);
+	exit(-1);
+    }
+
+    /* FIXME: Figure out how to use 16bpp depth buffer in 32bpp mode */
+    switch (r128ctx->glCtx->Visual->DepthBits) {
+    case 16: depthClear = 0x0000ffff; depth_bpp  = R128_Z_PIX_WIDTH_16; break;
+    case 24: depthClear = 0x00ffffff; depth_bpp  = R128_Z_PIX_WIDTH_24; break;
+    case 32: depthClear = 0xffffffff; depth_bpp  = R128_Z_PIX_WIDTH_32; break;
+    default:
+	fprintf(stderr, "Error: Unsupported depth %d... exiting\n",
+		r128ctx->r128Screen->bpp);
+	exit(-1);
+	break;
+    }
+
+    r128ctx->dirty             = R128_ALL_DIRTY;
+    r128ctx->dirty_context     = R128_CTX_ALL_DIRTY;
+
+    r128ctx->RenderIndex       = R128_FALLBACK_BIT;
+    r128ctx->PointsFunc        = NULL;
+    r128ctx->LineFunc          = NULL;
+    r128ctx->TriangleFunc      = NULL;
+    r128ctx->QuadFunc          = NULL;
+
+    r128ctx->IndirectTriangles = 0;
+    r128ctx->Fallback          = 0;
+
+    if (r128ctx->glCtx->Visual->DBflag) {
+	r128ctx->drawX         = r128ctx->r128Screen->backX;
+	r128ctx->drawY         = r128ctx->r128Screen->backY;
+	r128ctx->readX         = r128ctx->r128Screen->backX;
+	r128ctx->readY         = r128ctx->r128Screen->backY;
+    } else {
+	r128ctx->drawX         = r128ctx->r128Screen->fbX;
+	r128ctx->drawY         = r128ctx->r128Screen->fbY;
+	r128ctx->readX         = r128ctx->r128Screen->fbX;
+	r128ctx->readY         = r128ctx->r128Screen->fbY;
+    }
+
+    r128ctx->ClearColor = 0x00000000;
+    r128ctx->ClearDepth = depthClear;
+
+    r128ctx->regs.scale_3d_cntl =
+	R128_SCALE_DITHER_TABLE         |
+	R128_TEX_CACHE_SIZE_FULL        |
+	R128_DITHER_INIT_RESET          |
+	R128_SCALE_3D_TEXMAP_SHADE      |
+	R128_SCALE_PIX_REPLICATE        |
+     /* R128_TEX_CACHE_SPLIT            | */
+	R128_ALPHA_COMB_ADD_CLAMP       |
+	R128_FOG_TABLE                  |
+	R128_ALPHA_BLEND_SRC_ONE        |
+	R128_ALPHA_BLEND_DST_ZERO       |
+	R128_ALPHA_TEST_ALWAYS          |
+	R128_COMPOSITE_SHADOW_CMP_EQUAL |
+	R128_TEX_MAP_ALPHA_IN_TEXTURE   |
+	R128_TEX_CACHE_LINE_SIZE_8QW;
+
+    r128ctx->regs.dst_pitch_offset_c = pitch << R128_PITCH_SHIFT;
+
+    r128ctx->regs.dp_gui_master_cntl =
+	R128_GMC_DST_PITCH_OFFSET_CNTL |
+	R128_GMC_DST_CLIPPING          |
+	R128_GMC_BRUSH_SOLID_COLOR     |
+	dst_bpp                        |
+	R128_GMC_SRC_DATATYPE_COLOR    |
+	R128_GMC_BYTE_MSB_TO_LSB       |
+	R128_GMC_CONVERSION_TEMP_6500  |
+	R128_ROP3_S                    |
+	R128_DP_SRC_SOURCE_MEMORY      |
+	R128_GMC_3D_FCN_EN             |
+	R128_GMC_CLR_CMP_CNTL_DIS      |
+	R128_AUX_CLIP_DIS              |
+	R128_GMC_WR_MSK_DIS;
+
+    r128ctx->regs.sc_top_left_c     = 0x00000000;
+    r128ctx->regs.sc_bottom_right_c = 0x1fff1fff;
+
+    r128ctx->regs.aux_sc_cntl       = 0x00000000;
+
+    r128ctx->regs.aux1_sc_left      = 0x00000000;
+    r128ctx->regs.aux1_sc_right     = 0x00001fff;
+    r128ctx->regs.aux1_sc_top       = 0x00000000;
+    r128ctx->regs.aux1_sc_bottom    = 0x00001fff;
+
+    r128ctx->regs.aux2_sc_left      = 0x00000000;
+    r128ctx->regs.aux2_sc_right     = 0x00001fff;
+    r128ctx->regs.aux2_sc_top       = 0x00000000;
+    r128ctx->regs.aux2_sc_bottom    = 0x00001fff;
+
+    r128ctx->regs.aux3_sc_left      = 0x00000000;
+    r128ctx->regs.aux3_sc_right     = 0x00001fff;
+    r128ctx->regs.aux3_sc_top       = 0x00000000;
+    r128ctx->regs.aux3_sc_bottom    = 0x00001fff;
+
+    r128ctx->regs.z_offset_c        = (r128ctx->r128Screen->depthX *
+				       (r128ctx->r128Screen->bpp/8) +
+				       r128ctx->r128Screen->depthY *
+				       r128ctx->r128Screen->fbStride);
+    r128ctx->regs.z_pitch_c         = pitch;
+
+    r128ctx->regs.z_sten_cntl_c =
+	depth_bpp                |
+	R128_Z_TEST_LESS         |
+	R128_STENCIL_TEST_ALWAYS |
+	R128_STENCIL_S_FAIL_KEEP |
+	R128_STENCIL_ZPASS_KEEP  |
+	R128_STENCIL_ZFAIL_KEEP;
+
+    r128ctx->regs.tex_cntl_c =
+	R128_Z_WRITE_ENABLE          |
+	R128_SHADE_ENABLE            |
+	R128_DITHER_ENABLE           |
+	R128_ALPHA_IN_TEX_COMPLETE_A |
+	R128_LIGHT_DIS               |
+	R128_ALPHA_LIGHT_DIS         |
+	R128_TEX_CACHE_FLUSH         |
+	(0x0f << R128_LOD_BIAS_SHIFT);
+
+    r128ctx->regs.misc_3d_state_cntl_reg =
+	R128_MISC_SCALE_3D_TEXMAP_SHADE |
+	R128_MISC_SCALE_PIX_REPLICATE   |
+	R128_ALPHA_COMB_ADD_CLAMP       |
+	R128_FOG_TABLE                  |
+	R128_ALPHA_BLEND_SRC_ONE        |
+	R128_ALPHA_BLEND_DST_ZERO       |
+	R128_ALPHA_TEST_ALWAYS;
+
+    r128ctx->regs.texture_clr_cmp_clr_c = 0x00000000;
+    r128ctx->regs.texture_clr_cmp_msk_c = 0xffffffff;
+
+    r128ctx->regs.prim_tex_cntl_c =
+	R128_MIN_BLEND_NEAREST |
+	R128_MAG_BLEND_NEAREST |
+	R128_MIP_MAP_DISABLE   |
+	R128_TEX_CLAMP_S_WRAP  |
+	R128_TEX_CLAMP_T_WRAP;
+
+    r128ctx->regs.prim_texture_combine_cntl_c =
+	R128_COMB_MODULATE          |
+	R128_COLOR_FACTOR_TEX       |
+	R128_INPUT_FACTOR_INT_COLOR |
+	R128_COMB_ALPHA_COPY        |
+	R128_ALPHA_FACTOR_TEX_ALPHA |
+	R128_INP_FACTOR_A_INT_ALPHA;
+
+    r128ctx->regs.tex_size_pitch_c     =
+	(0 << R128_TEX_PITCH_SHIFT)      |
+	(0 << R128_TEX_SIZE_SHIFT)       |
+	(0 << R128_TEX_HEIGHT_SHIFT)     |
+	(0 << R128_TEX_MIN_SIZE_SHIFT)   |
+	(0 << R128_SEC_TEX_PITCH_SHIFT)  |
+	(0 << R128_SEC_TEX_SIZE_SHIFT)   |
+	(0 << R128_SEC_TEX_HEIGHT_SHIFT) |
+	(0 << R128_SEC_TEX_MIN_SIZE_SHIFT);
+
+    for (i = 0; i < R128_TEX_MAXLEVELS; i++)
+	r128ctx->regs.prim_tex_offset[i]  = 0x00000000;
+
+    r128ctx->regs.sec_tex_cntl_c =
+	R128_SEC_SELECT_PRIM_ST;
+
+    r128ctx->regs.sec_tex_combine_cntl_c =
+	R128_COMB_DIS                |
+	R128_COLOR_FACTOR_TEX        |
+	R128_INPUT_FACTOR_PREV_COLOR |
+	R128_COMB_ALPHA_DIS          |
+	R128_ALPHA_FACTOR_TEX_ALPHA  |
+	R128_INP_FACTOR_A_PREV_ALPHA;
+
+    for (i = 0; i < R128_TEX_MAXLEVELS; i++)
+	r128ctx->regs.sec_tex_offset[i] = 0x00000000;
+
+    r128ctx->regs.constant_color_c            = 0x00ffffff;
+    r128ctx->regs.prim_texture_border_color_c = 0x00ffffff;
+    r128ctx->regs.sec_texture_border_color_c  = 0x00ffffff;
+    r128ctx->regs.sten_ref_mask_c             = 0xffff0000;
+    r128ctx->regs.plane_3d_mask_c             = 0xffffffff;
+
+    r128ctx->regs.setup_cntl =
+	R128_COLOR_GOURAUD         |
+	R128_PRIM_TYPE_TRI         |
+#if 1
+	/* FIXME: Let r128 multiply? */
+	R128_TEXTURE_ST_MULT_W     |
+#else
+	/* FIXME: Or, pre multiply? */
+	R128_TEXTURE_ST_DIRECT     |
+#endif
+	R128_STARTING_VERTEX_1     |
+	R128_ENDING_VERTEX_3       |
+	R128_SU_POLY_LINE_NOT_LAST |
+	R128_SUB_PIX_4BITS;
+
+    r128ctx->regs.pm4_vc_fpu_setup =
+	R128_FRONT_DIR_CCW         |
+	R128_BACKFACE_SOLID        |
+	R128_FRONTFACE_SOLID       |
+	R128_FPU_COLOR_GOURAUD     |
+	R128_FPU_SUB_PIX_4BITS     |
+	R128_FPU_MODE_3D           |
+	R128_TRAP_BITS_DISABLE     |
+	R128_XFACTOR_2             |
+	R128_YFACTOR_2             |
+	R128_FLAT_SHADE_VERTEX_OGL |
+	R128_FPU_ROUND_TRUNCATE    |
+	R128_WM_SEL_8DW;
+
+    r128ctx->FogMode                   = GL_EXP;
+    r128ctx->regs.fog_color_c          = 0x00808080;
+    r128ctx->regs.fog_3d_table_start   = 0x00000000;
+    r128ctx->regs.fog_3d_table_end     = 0xffffffff;
+    r128ctx->regs.fog_3d_table_density = 0x00000000;
+
+    r128ctx->regs.window_xy_offset     = 0x00000000;
+
+    r128ctx->regs.dp_write_mask        = 0xffffffff;
+
+    r128ctx->regs.pc_gui_ctlstat       = R128_PC_FLUSH_GUI;
+
+    r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+    r128ctx->dirty_context |= R128_CTX_ALL_DIRTY;
+}
+
+/* Upload the fog table for the current fog mode */
+static void r128UploadFogTable(r128ContextPtr r128ctx)
+{
+    int i;
+
+    R128CCE0(R128_CCE_PACKET0, R128_FOG_TABLE_INDEX, 0);
+    R128CCE(0x00000000);
+
+    R128CCE0(R128_CCE_PACKET0_ONE_REG_WR, R128_FOG_TABLE_DATA, 255);
+
+    switch (r128ctx->FogMode) {
+    case GL_LINEAR:
+	for (i = 0; i < 256; i++) {
+	    R128CCE(255 - i);
+	}
+	break;
+    case GL_EXP:
+	for (i = 0; i < 256; i++) {
+	    R128CCE(255 - FLOAT_TO_UBYTE(exp(i - 255)));
+	}
+	break;
+    case GL_EXP2:
+	for (i = 0; i < 256; i++) {
+	    R128CCE(255 - FLOAT_TO_UBYTE(exp((i - 255) * (i - 255))));
+	}
+	break;
+    }
+}
+
+/* Load the current context's state into the hardware */
+/* NOTE: This function is only called while holding the hardware lock */
+static void r128LoadContext(r128ContextPtr r128ctx)
+{
+    int i;
+    int tex_size_pitch_done = GL_FALSE;
+
+    if (R128_DEBUG_FLAGS & DEBUG_VERBOSE_API) {
+	fprintf(stderr, "r128LoadContext(%p)\n", r128ctx->glCtx);
+    }
+
+#if 0
+    r128ctx->dirty_context = R128_CTX_ALL_DIRTY;
+#endif
+
+#if 1
+    /* FIXME: Why do these need to be updated even when they don't change? */
+    r128ctx->dirty_context |= (R128_CTX_MISC        |
+			       R128_CTX_ENGINESTATE |
+			       R128_CTX_ALPHASTATE);
+#endif
+
+#if 1
+    /* FIXME: Is this _really_ needed? */
+    if (r128ctx->dirty_context)
+	if (!R128CCE_USE_RING_BUFFER(r128ctx->r128Screen->CCEMode))
+	    R128CCE_WAIT_FOR_IDLE(r128ctx);
+#endif
+
+    if (r128ctx->dirty_context & R128_CTX_MISC) {
+	R128CCE1(R128_CCE_PACKET1, R128_SCALE_3D_CNTL, R128_DP_WRITE_MASK);
+	R128CCE(r128ctx->regs.scale_3d_cntl);
+	R128CCE(r128ctx->regs.dp_write_mask);
+
+	R128CCE0(R128_CCE_PACKET0, R128_DST_PITCH_OFFSET_C, 1);
+	R128CCE(r128ctx->regs.dst_pitch_offset_c);
+	R128CCE(r128ctx->regs.dp_gui_master_cntl);
+
+	R128CCE0(R128_CCE_PACKET0, R128_TEXTURE_CLR_CMP_CLR_C, 1);
+	R128CCE(r128ctx->regs.texture_clr_cmp_clr_c);
+	R128CCE(r128ctx->regs.texture_clr_cmp_msk_c);
+
+	R128CCE0(R128_CCE_PACKET0, R128_STEN_REF_MASK_C, 1);
+	R128CCE(r128ctx->regs.sten_ref_mask_c);
+	R128CCE(r128ctx->regs.plane_3d_mask_c);
+    }
+
+    if (r128ctx->dirty_context & R128_CTX_ENGINESTATE) {
+	R128CCE0(R128_CCE_PACKET0, R128_TEX_CNTL_C, 0);
+	R128CCE(r128ctx->regs.tex_cntl_c);
+    }
+
+    if (r128ctx->dirty_context & R128_CTX_TEX0STATE) {
+	R128CCE0(R128_CCE_PACKET0, R128_PRIM_TEX_CNTL_C, 2+R128_TEX_MAXLEVELS);
+	R128CCE(r128ctx->regs.prim_tex_cntl_c);
+	R128CCE(r128ctx->regs.prim_texture_combine_cntl_c);
+	R128CCE(r128ctx->regs.tex_size_pitch_c);
+	for (i = 0; i < R128_TEX_MAXLEVELS; i++)
+	    R128CCE(r128ctx->regs.prim_tex_offset[i]);
+
+	R128CCE0(R128_CCE_PACKET0, R128_PRIM_TEXTURE_BORDER_COLOR_C, 0);
+	R128CCE(r128ctx->regs.prim_texture_border_color_c);
+
+	tex_size_pitch_done = GL_TRUE;
+    }
+
+    if (r128ctx->dirty_context & R128_CTX_TEX1STATE) {
+	if (!tex_size_pitch_done) {
+	    R128CCE0(R128_CCE_PACKET0, R128_TEX_SIZE_PITCH_C, 0);
+	    R128CCE(r128ctx->regs.tex_size_pitch_c);
+	}
+
+	R128CCE0(R128_CCE_PACKET0, R128_SEC_TEX_CNTL_C, 1+R128_TEX_MAXLEVELS);
+	R128CCE(r128ctx->regs.sec_tex_cntl_c);
+	R128CCE(r128ctx->regs.sec_tex_combine_cntl_c);
+	for (i = 0; i < R128_TEX_MAXLEVELS; i++)
+	    R128CCE(r128ctx->regs.sec_tex_offset[i]);
+
+	R128CCE0(R128_CCE_PACKET0, R128_SEC_TEXTURE_BORDER_COLOR_C, 0);
+	R128CCE(r128ctx->regs.sec_texture_border_color_c);
+    }
+
+    if (r128ctx->dirty_context & R128_CTX_TEXENVSTATE) {
+	R128CCE0(R128_CCE_PACKET0, R128_CONSTANT_COLOR_C, 0);
+	R128CCE(r128ctx->regs.constant_color_c);
+    }
+
+    if (r128ctx->dirty_context & R128_CTX_FOGSTATE) {
+	R128CCE0(R128_CCE_PACKET0, R128_FOG_3D_TABLE_START, 1);
+	R128CCE(r128ctx->regs.fog_3d_table_start);
+	R128CCE(r128ctx->regs.fog_3d_table_end);
+
+	R128CCE1(R128_CCE_PACKET1,
+		 R128_FOG_COLOR_C, R128_FOG_3D_TABLE_DENSITY);
+	R128CCE(r128ctx->regs.fog_color_c);
+	R128CCE(r128ctx->regs.fog_3d_table_density);
+    }
+
+    if (r128ctx->dirty_context & R128_CTX_FOGTABLE) {
+	r128UploadFogTable(r128ctx);
+    }
+
+    if (r128ctx->dirty_context & R128_CTX_ZSTENSTATE) {
+	R128CCE0(R128_CCE_PACKET0, R128_Z_STEN_CNTL_C, 0);
+	R128CCE(r128ctx->regs.z_sten_cntl_c);
+    }
+
+    if (r128ctx->dirty_context & R128_CTX_SCISSORS) {
+	R128CCE0(R128_CCE_PACKET0, R128_SC_TOP_LEFT_C, 1);
+	R128CCE(r128ctx->regs.sc_top_left_c);
+	R128CCE(r128ctx->regs.sc_bottom_right_c);
+    }
+
+    if (r128ctx->dirty_context & (R128_CTX_ALPHASTATE |
+				  R128_CTX_FOGSTATE)) {
+	R128CCE0(R128_CCE_PACKET0, R128_MISC_3D_STATE_CNTL_REG, 0);
+	R128CCE(r128ctx->regs.misc_3d_state_cntl_reg);
+    }
+
+    if (r128ctx->dirty_context & R128_CTX_SETUPSTATE) {
+	R128CCE1(R128_CCE_PACKET1, R128_SETUP_CNTL, R128_PM4_VC_FPU_SETUP);
+	R128CCE(r128ctx->regs.setup_cntl);
+	R128CCE(r128ctx->regs.pm4_vc_fpu_setup);
+    }
+
+    if (r128ctx->dirty_context & R128_CTX_WIN_Z_POS) {
+	R128CCE0(R128_CCE_PACKET0, R128_WINDOW_XY_OFFSET, 0);
+	R128CCE(r128ctx->regs.window_xy_offset);
+
+	R128CCE0(R128_CCE_PACKET0, R128_Z_OFFSET_C, 1);
+	R128CCE(r128ctx->regs.z_offset_c);
+	R128CCE(r128ctx->regs.z_pitch_c);
+    }
+
+#if 0
+    if (r128ctx->dirty_context & R128_CTX_FLUSH_PIX_CACHE) {
+	R128CCE0(R128_CCE_PACKET0, R128_PC_GUI_CTLSTAT, 0);
+	R128CCE(r128ctx->regs.pc_gui_ctlstat);
+    }
+#endif
+
+    R128CCE_SUBMIT_PACKETS();
+
+    /* Turn off the texture cache flushing */
+    r128ctx->regs.tex_cntl_c     &= ~R128_TEX_CACHE_FLUSH;
+
+    /* Turn off the pixel cache flushing */
+    r128ctx->regs.pc_gui_ctlstat &= ~R128_PC_FLUSH_ALL;
+
+    r128ctx->dirty_context = R128_CTX_CLEAN;
+}
+
+/* Set the hardware clip rects for drawing to the current color buffer */
+/* NOTE: This function is only called while holding the hardware lock */
+void r128SetClipRects(r128ContextPtr r128ctx,
+		      XF86DRIClipRectPtr pc, int nc)
+{
+    if (!pc) return;
+
+    /* Clear any previous auxiliary scissors */
+    r128ctx->regs.aux_sc_cntl = 0x00000000;
+
+    switch (nc) {
+    case 3:
+	R128CCE0(R128_CCE_PACKET0, R128_AUX3_SC_LEFT, 3);
+	R128CCE(pc[2].x1   + r128ctx->drawX);
+	R128CCE(pc[2].x2-1 + r128ctx->drawX);
+	R128CCE(pc[2].y1   + r128ctx->drawY);
+	R128CCE(pc[2].y2-1 + r128ctx->drawY);
+
+	r128ctx->regs.aux_sc_cntl |= R128_AUX3_SC_EN | R128_AUX3_SC_MODE_OR;
+
+    case 2:
+	R128CCE0(R128_CCE_PACKET0, R128_AUX2_SC_LEFT, 3);
+	R128CCE(pc[1].x1   + r128ctx->drawX);
+	R128CCE(pc[1].x2-1 + r128ctx->drawX);
+	R128CCE(pc[1].y1   + r128ctx->drawY);
+	R128CCE(pc[1].y2-1 + r128ctx->drawY);
+
+	r128ctx->regs.aux_sc_cntl |= R128_AUX2_SC_EN | R128_AUX2_SC_MODE_OR;
+
+    case 1:
+	R128CCE0(R128_CCE_PACKET0, R128_AUX1_SC_LEFT, 3);
+	R128CCE(pc[0].x1   + r128ctx->drawX);
+	R128CCE(pc[0].x2-1 + r128ctx->drawX);
+	R128CCE(pc[0].y1   + r128ctx->drawY);
+	R128CCE(pc[0].y2-1 + r128ctx->drawY);
+
+	r128ctx->regs.aux_sc_cntl |= R128_AUX1_SC_EN | R128_AUX1_SC_MODE_OR;
+	break;
+
+    default:
+	return;
+    }
+
+    R128CCE0(R128_CCE_PACKET0, R128_AUX_SC_CNTL, 0);
+    R128CCE(r128ctx->regs.aux_sc_cntl);
+
+    R128CCE_SUBMIT_PACKETS();
+}
+
+/* Update the driver's notion of the window position */
+/* NOTE: This function is only called while holding the hardware lock */
+static void r128UpdateWindowPosition(r128ContextPtr r128ctx)
+{
+    int x = r128ctx->driDrawable->x + r128ctx->drawX;
+    int y = r128ctx->driDrawable->y + r128ctx->drawY;
+
+#if 0
+    /* FIXME: Is this _really_ needed? */
+    R128CCE_FLUSH_VB(r128ctx);
+#endif
+    r128ctx->regs.window_xy_offset = ((y << R128_WINDOW_Y_SHIFT) |
+				      (x << R128_WINDOW_X_SHIFT));
+
+    /* Recalculate the Z buffer offset since we might be drawing to the
+       back buffer and window_xy_offset affects both color buffer and
+       depth drawing */
+    r128ctx->regs.z_offset_c        = ((r128ctx->r128Screen->depthX -
+					r128ctx->drawX) *
+				       (r128ctx->r128Screen->bpp/8) +
+				       (r128ctx->r128Screen->depthY -
+					r128ctx->drawY) *
+				       r128ctx->r128Screen->fbStride);
+
+    r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+    r128ctx->dirty_context |= R128_CTX_WIN_Z_POS;
+}
+
+/* Update the hardware state */
+/* NOTE: This function is only called while holding the hardware lock */
+static void r128UpdateHWStateLocked(r128ContextPtr r128ctx)
+{
+    if (r128ctx->dirty & R128_REQUIRE_QUIESCENCE)
+	R128CCE_WAIT_FOR_IDLE(r128ctx);
+
+    /* Update any state that might have changed recently */
+
+    /* Update the clip rects */
+    if (r128ctx->dirty & R128_UPDATE_WINPOS)
+	r128UpdateWindowPosition(r128ctx);
+
+    /* Update texture state and then upload the images */
+    /* Note: Texture images can only be updated after the state has been set */
+    if (r128ctx->dirty & R128_UPDATE_TEXSTATE)
+	r128UpdateTextureState(r128ctx);
+    if (r128ctx->dirty & R128_UPDATE_TEX0IMAGES)
+        r128UploadTexImages(r128ctx, r128ctx->CurrentTexObj[0]);
+    if (r128ctx->dirty & R128_UPDATE_TEX1IMAGES)
+        r128UploadTexImages(r128ctx, r128ctx->CurrentTexObj[1]);
+
+    /* Load the state into the hardware */
+    /* Note: This must be done after all other state has been set */
+    if (r128ctx->dirty & R128_UPDATE_CONTEXT)
+	r128LoadContext(r128ctx);
+
+    r128ctx->dirty = R128_CLEAN;
+}
+
+/* Update the hardware state */
+void r128UpdateHWState(r128ContextPtr r128ctx)
+{
+    LOCK_HARDWARE(r128ctx);
+    r128UpdateHWStateLocked(r128ctx);
+    UNLOCK_HARDWARE(r128ctx);
+}
+
+/* Update the driver's state */
+/* NOTE: This function is only called while holding the hardware lock */
+void r128UpdateState(r128ContextPtr r128ctx, int winMoved)
+{
+    R128SAREAPrivPtr sarea = r128ctx->r128Screen->SAREA;
+    int              i;
+
+    if (sarea->ctxOwner != r128ctx->driContext->hHWContext) {
+	sarea->ctxOwner = r128ctx->driContext->hHWContext;
+	r128ctx->dirty_context = R128_CTX_ALL_DIRTY;
+	r128LoadContext(r128ctx);
+    }
+
+    for (i = 0; i < r128ctx->r128Screen->NRTexHeaps; i++)
+	r128AgeTextures(r128ctx, i);
+
+    if (winMoved) r128ctx->dirty |= R128_UPDATE_WINPOS;
+}
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_state.h b/xc/lib/GL/mesa/src/drv/r128/r128_state.h
new file mode 100644
index 000000000..f847452f0
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_state.h
@@ -0,0 +1,50 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas.
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#ifndef _R128_STATE_H_
+#define _R128_STATE_H_
+
+#ifdef GLX_DIRECT_RENDERING
+
+extern void   r128DDInitState(r128ContextPtr r128ctx);
+extern void   r128DDInitStateFuncs(GLcontext *ctx);
+
+extern void   r128UpdateState(r128ContextPtr r128ctx, int winMoved);
+extern void   r128UpdateHWState(r128ContextPtr r128ctx);
+
+extern void   r128SetClipRects(r128ContextPtr r128ctx,
+			       XF86DRIClipRectPtr pc, int nc);
+
+#endif
+#endif /* _R128_STATE_H_ */
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_swap.c b/xc/lib/GL/mesa/src/drv/r128/r128_swap.c
new file mode 100644
index 000000000..6b347023d
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_swap.c
@@ -0,0 +1,125 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#include "r128_init.h"
+#include "r128_mesa.h"
+#include "r128_xmesa.h"
+#include "r128_context.h"
+#include "r128_lock.h"
+#include "r128_reg.h"
+#include "r128_cce.h"
+#include "r128_state.h"
+#include "r128_vb.h"
+#include "r128_swap.h"
+
+/* Copy the back color buffer to the front color buffer */
+void r128SwapBuffers(r128ContextPtr r128ctx)
+{
+    unsigned char *R128MMIO        = r128ctx->r128Screen->mmio;
+    __DRIdrawablePrivate *dPriv    = r128ctx->driDrawable;
+    int                   nc;
+    XF86DRIClipRectPtr    c;
+    int                   dst_bpp;
+    CARD32                swapAge;
+
+    if (r128ctx->SWonly) {
+	/* FIXME: Provide software fallback for this case?? */
+    }
+
+    switch (r128ctx->r128Screen->bpp) {
+    case 8:
+	dst_bpp = R128_GMC_DST_8BPP_CI;
+	break;
+    case 16:
+	if (r128ctx->r128Screen->depth == 15) dst_bpp = R128_GMC_DST_15BPP;
+	else                                  dst_bpp = R128_GMC_DST_16BPP;
+	break;
+    case 24:
+	dst_bpp = R128_GMC_DST_24BPP;
+	break;
+    case 32:
+    default:
+	dst_bpp = R128_GMC_DST_32BPP;
+	break;
+    }
+
+    LOCK_HARDWARE(r128ctx);
+
+    /* Flush any outstanding vertex buffers */
+    R128CCE_FLUSH_VB(r128ctx);
+
+    /* Throttle the frame rate -- only allow one pending swap buffers
+       request at a time */
+    while (r128ctx->lastSwapAge > (swapAge = INREG(R128_SWAP_AGE_REG)));
+
+    /* Init the clip rects here in case they changed during the
+       LOCK_HARDWARE macro */
+    c  = dPriv->pClipRects;
+    nc = dPriv->numClipRects;
+
+    /* Cycle through the clip rects */
+    while (nc--) {
+	int fx = c[nc].x1;
+	int fy = c[nc].y1;
+	int fw = c[nc].x2 - fx;
+	int fh = c[nc].y2 - fy;
+	int bx = fx + r128ctx->r128Screen->backX;
+	int by = fy + r128ctx->r128Screen->backY;
+
+	fx += r128ctx->r128Screen->fbX;
+	fy += r128ctx->r128Screen->fbY;
+
+	R128CCE3(R128_CCE_PACKET3_CNTL_BITBLT_MULTI, 3);
+	R128CCE(R128_GMC_BRUSH_NONE
+		| R128_GMC_SRC_DATATYPE_COLOR
+		| R128_DP_SRC_SOURCE_MEMORY
+		| dst_bpp
+		| R128_ROP3_S);
+	R128CCE((bx << 16) | by);
+	R128CCE((fx << 16) | fy);
+	R128CCE((fw << 16) | fh);
+    }
+
+    ++swapAge;
+    R128CCE0(R128_CCE_PACKET0, R128_SWAP_AGE_REG, 0);
+    R128CCE(swapAge);
+    r128ctx->lastSwapAge = swapAge;
+
+    r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+    r128ctx->dirty_context |= R128_CTX_ALL_DIRTY;
+
+    R128CCE_SUBMIT_PACKETS();
+
+    UNLOCK_HARDWARE(r128ctx);
+}
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_swap.h b/xc/lib/GL/mesa/src/drv/r128/r128_swap.h
new file mode 100644
index 000000000..32851041c
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_swap.h
@@ -0,0 +1,43 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#ifndef _R128_SWAP_H_
+#define _R128_SWAP_H_
+
+#ifdef GLX_DIRECT_RENDERING
+
+extern void r128SwapBuffers(r128ContextPtr r128ctx);
+
+#endif
+#endif /* _R128_SWAP_H_ */
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_tex.c b/xc/lib/GL/mesa/src/drv/r128/r128_tex.c
new file mode 100644
index 000000000..d75345404
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_tex.c
@@ -0,0 +1,1851 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas.
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *   Gareth Hughes <gareth@precisioninsight.com>
+ *
+ */
+
+#include "r128_init.h"
+#include "r128_mesa.h"
+#include "r128_xmesa.h"
+#include "r128_context.h"
+#include "r128_lock.h"
+#include "r128_state.h"
+#include "r128_reg.h"
+#include "r128_cce.h"
+#include "r128_vb.h"
+#include "r128_tex.h"
+
+#include "mmath.h"
+#include "simple_list.h"
+
+static void r128SetTexWrap(r128TexObjPtr t, GLenum srwap, GLenum twrap);
+static void r128SetTexFilter(r128TexObjPtr t, GLenum minf, GLenum magf);
+static void r128SetTexBorderColor(r128TexObjPtr t, GLubyte c[4]);
+
+/* Allocate and initialize hardware state associated with texture `t' */
+/* NOTE: This function is only called while holding the hardware lock */
+static r128TexObjPtr r128CreateTexObj(r128ContextPtr r128ctx,
+				      struct gl_texture_object *tObj)
+{
+    r128TexObjPtr            t;
+    struct gl_texture_image *image;
+    int                      log2Pitch, log2Height, log2Size, log2MinSize;
+    int                      totalSize;
+    int                      i;
+
+    image = tObj->Image[0];
+    if (!image) return NULL; /* ERROR!!! */
+
+    t = (r128TexObjPtr)calloc(1,sizeof(*t));
+    if (!t) return NULL;     /* ERROR!!! */
+
+    if (R128_DEBUG_FLAGS & DEBUG_VERBOSE_API)
+	fprintf(stderr, "r128CreateTexObj(%p)\n", tObj);
+
+    switch (image->Format) {
+    case GL_RGBA:
+    case GL_ALPHA:
+    case GL_LUMINANCE_ALPHA:
+    case GL_INTENSITY:
+	if (r128ctx->r128Screen->bpp == 32) {
+	    t->texelBytes    = 4;
+	    t->textureFormat = R128_DATATYPE_ARGB8888;
+	} else {
+	    t->texelBytes    = 2;
+	    t->textureFormat = R128_DATATYPE_ARGB4444;
+	}
+	break;
+
+    case GL_RGB:
+	if (r128ctx->r128Screen->bpp == 32) {
+	    t->texelBytes    = 4;
+	    t->textureFormat = R128_DATATYPE_ARGB8888;
+	} else {
+	    t->texelBytes    = 2;
+	    t->textureFormat = R128_DATATYPE_RGB565;
+	}
+	break;
+
+    case GL_LUMINANCE:
+	if (r128ctx->r128Screen->bpp == 32) {
+	    t->texelBytes    = 4;
+	    t->textureFormat = R128_DATATYPE_ARGB8888;
+	} else {
+	    t->texelBytes    = 2;
+					/* Use this to get true greys */
+	    t->textureFormat = R128_DATATYPE_ARGB1555;
+	}
+	break;
+
+    case GL_COLOR_INDEX:
+	t->texelBytes    = 1;
+	t->textureFormat = R128_DATATYPE_CI8;
+	break;
+
+    default:
+	/* ERROR!!! */
+	fprintf(stderr, "r128CreateTexObj: bad image->Format\n");
+	free(t);
+	return NULL;
+    }
+
+    /* Calculate dimensions in log domain */
+    for (i = 1, log2Height = 0; i < image->Height; i *= 2) log2Height++;
+    for (i = 1, log2Pitch = 0;  i < image->Width;  i *= 2) log2Pitch++;
+    if (image->Width > image->Height) log2Size = log2Pitch;
+    else                              log2Size = log2Height;
+
+    t->dirty_images = 0;
+
+    /* Calculate mipmap offsets and dimensions */
+    totalSize = 0;
+    for (i = 0; i <= log2Size && tObj->Image[i]; i++) {
+	t->image[i].offset  = totalSize;
+	t->image[i].width   = tObj->Image[i]->Width;
+	t->image[i].height  = tObj->Image[i]->Height;
+	t->dirty_images    |= 1 << i;
+	totalSize          += (tObj->Image[i]->Height *
+			       tObj->Image[i]->Width *
+			       t->texelBytes);
+
+	/* Offsets must be 32-byte aligned for host data blits */
+	totalSize           = (totalSize + 31) & ~31;
+    }
+    log2MinSize = log2Size - i + 1;
+
+    t->totalSize    = totalSize;
+    t->internFormat = image->IntFormat;
+
+    t->bound        = 0;
+    t->heap         = 0; /* This is set in r128UploadTexImages */
+    t->tObj         = tObj;
+
+    t->memBlock     = NULL;
+    t->bufAddr      = NULL;
+
+    t->regs.tex_cntl         = t->textureFormat;
+    t->regs.size_pitch       = ((log2Pitch   << R128_TEX_PITCH_SHIFT)  |
+				(log2Size    << R128_TEX_SIZE_SHIFT)   |
+				(log2Height  << R128_TEX_HEIGHT_SHIFT) |
+				(log2MinSize << R128_TEX_MIN_SIZE_SHIFT));
+    t->regs.border_color     = 0x00000000;
+
+    if (log2MinSize == log2Size ||
+	log2MinSize != 0)
+	t->regs.tex_cntl |= R128_MIP_MAP_DISABLE;
+
+    r128SetTexWrap(t, tObj->WrapS, tObj->WrapT);
+    r128SetTexFilter(t, tObj->MinFilter, tObj->MagFilter);
+    r128SetTexBorderColor(t, tObj->BorderColor);
+
+    tObj->DriverData = t;
+
+    make_empty_list(t);
+
+    return t;
+}
+
+/* Destroy hardware state associated with texture `t' */
+/* NOTE: This function can be called while holding the hardware lock and
+   while not holding the lock*/
+void r128DestroyTexObj(r128ContextPtr r128ctx, r128TexObjPtr t)
+{
+    if (!t) return;
+
+    if (t->memBlock) {
+	mmFreeMem(t->memBlock);
+	t->memBlock = NULL;
+    }
+
+    if (t->tObj)  t->tObj->DriverData                = NULL;
+    if (t->bound) r128ctx->CurrentTexObj[t->bound-1] = NULL;
+
+    remove_from_list(t);
+    free(t);
+}
+
+/* Keep track of swapped out texture objects */
+/* NOTE: This function is only called while holding the hardware lock */
+static void r128SwapOutTexObj(r128ContextPtr r128ctx, r128TexObjPtr t)
+{
+    if (t->memBlock) {
+	mmFreeMem(t->memBlock);
+	t->memBlock = NULL;
+    }
+
+    t->dirty_images = ~0;
+    move_to_tail(&r128ctx->SwappedOut, t);
+}
+
+/* Print out debugging information about texture LRU */
+void r128PrintLocalLRU(r128ContextPtr r128ctx, int heap)
+{
+    r128TexObjPtr t;
+    int           sz = 1 << (r128ctx->r128Screen->log2TexGran[heap]);
+
+    foreach(t, &r128ctx->TexObjList[heap]) {
+	if (!t->tObj) {
+	    fprintf(stderr, "Placeholder %d at 0x%x sz 0x%x\n",
+		    t->memBlock->ofs / sz,
+		    t->memBlock->ofs,
+		    t->memBlock->size);
+	} else {
+	    fprintf(stderr, "Texture (bound %d) at 0x%x sz 0x%x\n",
+		    t->bound,
+		    t->memBlock->ofs,
+		    t->memBlock->size);
+	}
+    }
+}
+
+void r128PrintGlobalLRU(r128ContextPtr r128ctx, int heap)
+{
+    R128TexRegion *list = r128ctx->r128Screen->SAREA->texList[heap];
+    int            i, j;
+
+    for (i = 0, j = R128_NR_TEX_REGIONS ; i < R128_NR_TEX_REGIONS ; i++) {
+	fprintf(stderr, "list[%d] age %d next %d prev %d\n",
+		j, list[j].age, list[j].next, list[j].prev);
+	j = list[j].next;
+	if (j == R128_NR_TEX_REGIONS) break;
+    }
+
+    if (j != R128_NR_TEX_REGIONS) {
+	fprintf(stderr, "Loop detected in global LRU\n");
+    }
+}
+
+/* Reset the global texture LRU */
+/* NOTE: This function is only called while holding the hardware lock */
+static void r128ResetGlobalLRU(r128ContextPtr r128ctx, int heap)
+{
+    R128TexRegion *list   = r128ctx->r128Screen->SAREA->texList[heap];
+    int            log2sz = 1 << r128ctx->r128Screen->log2TexGran[heap];
+    int            i;
+
+    /*
+     * (Re)initialize the global circular LRU list.  The last element in
+     * the array (R128_NR_TEX_REGIONS) is the sentinal.  Keeping it at
+     * the end of the array allows it to be addressed rationally when
+     * looking up objects at a particular location in texture memory.
+     */
+
+    for (i = 0; (i+1) * log2sz <= r128ctx->r128Screen->texSize[heap]; i++) {
+	list[i].prev = i-1;
+	list[i].next = i+1;
+	list[i].age = 0;
+    }
+
+    i--;
+    list[0].prev = R128_NR_TEX_REGIONS;
+    list[i].prev = i-1;
+    list[i].next = R128_NR_TEX_REGIONS;
+    list[R128_NR_TEX_REGIONS].prev = i;
+    list[R128_NR_TEX_REGIONS].next = 0;
+    r128ctx->r128Screen->SAREA->texAge[heap] = 0;
+}
+
+/* Update the local and glock texture LRUs */
+/* NOTE: This function is only called while holding the hardware lock */
+static void r128UpdateTexLRU(r128ContextPtr r128ctx, r128TexObjPtr t)
+{
+    int            heap   = t->heap;
+    R128TexRegion *list = r128ctx->r128Screen->SAREA->texList[heap];
+    int            log2sz = r128ctx->r128Screen->log2TexGran[heap];
+
+    int            start  = t->memBlock->ofs >> log2sz;
+    int            end    = (t->memBlock->ofs + t->memBlock->size-1) >> log2sz;
+    int            i;
+
+    r128ctx->lastTexAge[heap] = ++r128ctx->r128Screen->SAREA->texAge[heap];
+
+    /* Update our local LRU */
+    move_to_head(&r128ctx->TexObjList[heap], t);
+
+    /* Update the global LRU */
+    for (i = start ; i <= end ; i++) {
+	list[i].in_use = 1;
+	list[i].age = r128ctx->lastTexAge[heap];
+
+	/* remove_from_list(i) */
+	list[(CARD32)list[i].next].prev = list[i].prev;
+	list[(CARD32)list[i].prev].next = list[i].next;
+
+	/* insert_at_head(list, i) */
+	list[i].prev = R128_NR_TEX_REGIONS;
+	list[i].next = list[R128_NR_TEX_REGIONS].next;
+	list[(CARD32)list[R128_NR_TEX_REGIONS].next].prev = i;
+	list[R128_NR_TEX_REGIONS].next = i;
+    }
+}
+
+/* Update our notion of what textures have been changed since we last
+   held the lock.  This pertains to both our local textures and the
+   textures belonging to other clients.  Keep track of other client's
+   textures by pushing a placeholder texture onto the LRU list -- these
+   are denoted by (tObj == NULL). */
+/* NOTE: This function is only called while holding the hardware lock */
+static void r128TexturesGone(r128ContextPtr r128ctx, int heap,
+			     int offset, int size, int in_use)
+{
+    r128TexObjPtr t, tmp;
+   
+    foreach_s (t, tmp, &r128ctx->TexObjList[heap]) {
+	if (t->memBlock->ofs >= offset + size ||
+	    t->memBlock->ofs + t->memBlock->size <= offset)
+	    continue;
+
+	/* It overlaps - kick it out.  Need to hold onto the currently
+	   bound objects, however. */
+	if (t->bound) r128SwapOutTexObj(r128ctx, t);
+	else          r128DestroyTexObj(r128ctx, t);
+    }
+
+    if (in_use) {
+	t = (r128TexObjPtr) calloc(1,sizeof(*t));
+	if (!t) return;
+
+	t->memBlock = mmAllocMem(r128ctx->texHeap[heap], size, 0, offset);
+	insert_at_head(&r128ctx->TexObjList[heap], t);
+    }
+}
+
+/* Update our client's shared texture state.  If another client has
+   modified a region in which we have textures, then we need to figure
+   out which of our textures has been removed, and update our global
+   LRU. */
+void r128AgeTextures(r128ContextPtr r128ctx, int heap)
+{
+    R128SAREAPrivPtr sarea = r128ctx->r128Screen->SAREA;
+
+    if (sarea->texAge[heap] != r128ctx->lastTexAge[heap]) {
+	int log2sz = 1 << r128ctx->r128Screen->log2TexGran[heap];
+	int nr     = 0;
+	int idx;
+
+	for (idx = sarea->texList[heap][R128_NR_TEX_REGIONS].prev;
+	     idx != R128_NR_TEX_REGIONS && nr < R128_NR_TEX_REGIONS;
+	     idx = sarea->texList[heap][idx].prev, nr++) {
+
+	    /* If switching texturing schemes, then the SAREA might not
+               have been properly cleared, so we need to reset the
+               global texture LRU. */
+	    if (idx * log2sz > r128ctx->r128Screen->texSize[heap]) {
+		nr = R128_NR_TEX_REGIONS;
+		break;
+	    }
+
+	    if (sarea->texList[heap][idx].age > r128ctx->lastTexAge[heap])
+		r128TexturesGone(r128ctx, heap, idx * log2sz, log2sz,
+				 sarea->texList[heap][idx].in_use);
+	}
+
+	if (nr == R128_NR_TEX_REGIONS) {
+	    r128TexturesGone(r128ctx, heap,
+			     0, r128ctx->r128Screen->texSize[heap], 0);
+	    r128ResetGlobalLRU(r128ctx, heap);
+	}
+
+	r128ctx->dirty |= R128_UPDATE_TEX0IMAGES;
+	r128ctx->dirty |= R128_UPDATE_TEX1IMAGES;
+	r128ctx->lastTexAge[heap] = sarea->texAge[heap];
+    }
+}
+
+/* Convert a block of Mesa-formatted texture to an 8bpp hardware format */
+static void r128ConvertTexture8bpp(r128ContextPtr r128ctx,
+				   struct gl_texture_image *image,
+				   int x, int y, int width, int height,
+				   int pitch)
+{
+    CARD8  *src;
+    CARD32  pix;
+    int     i, j;
+
+    switch (image->Format) {
+    case GL_RGB:
+	for (i = 0; i < height; i++) {
+	    src = (CARD8 *)image->Data + ((y + i) * pitch + x) * 3;
+	    for (j = width >> 2; j; j--) {
+		pix = ((R128PACKCOLOR332( src[0], src[1],  src[2])       ) |
+		       (R128PACKCOLOR332( src[3], src[4],  src[5])  <<  8) |
+		       (R128PACKCOLOR332( src[6], src[7],  src[8])  << 16) |
+		       (R128PACKCOLOR332( src[9], src[10], src[11]) << 24));
+		R128CCE(pix);
+		src += 12;
+	    }
+	}
+	break;
+
+    case GL_ALPHA:
+    case GL_LUMINANCE:
+    case GL_INTENSITY:
+    case GL_COLOR_INDEX:
+	for (i = 0; i < height; i++) {
+	    src = (CARD8 *)image->Data + ((y + i) * pitch + x);
+	    for (j = width >> 2; j; j--) {
+		pix = src[0] | (src[1] << 8) | (src[2] << 16) | (src[3] << 24);
+		R128CCE(pix);
+		src += 4;
+	    }
+	}
+	break;
+
+    default:
+	fprintf(stderr, "r128ConvertTexture8bpp: unsupported format 0x%x\n",
+		image->Format);
+    }
+}
+
+/* Convert a block of Mesa-formatted texture to a 16bpp hardware format */
+static void r128ConvertTexture16bpp(r128ContextPtr r128ctx,
+				    struct gl_texture_image *image,
+				    int x, int y, int width, int height,
+				    int pitch)
+{
+    CARD8  *src;
+    CARD32  pix;
+    int     i, j;
+
+    switch (image->Format) {
+    case GL_RGB:
+	for (i = 0; i < height; i++) {
+	    src = (CARD8 *)image->Data + ((y + i) * pitch + x) * 3;
+	    for (j = width >> 1; j; j--) {
+		pix = ((R128PACKCOLOR565(src[0], src[1], src[2])      ) |
+		       (R128PACKCOLOR565(src[3], src[4], src[5]) << 16));
+		R128CCE(pix);
+		src += 6;
+	    }
+	}
+	break;
+
+    case GL_RGBA:
+	for (i = 0; i < height; i++) {
+	    src = (CARD8 *)image->Data + ((y + i) * pitch + x) * 4;
+	    for (j = width >> 1; j; j--) {
+		pix =
+		    ((R128PACKCOLOR4444(src[0], src[1], src[2], src[3])    ) |
+		     (R128PACKCOLOR4444(src[4], src[5], src[6], src[7])<<16));
+		R128CCE(pix);
+		src += 8;
+	    }
+	}
+	break;
+
+    case GL_ALPHA:
+	for (i = 0; i < height; i++) {
+	    src = (CARD8 *)image->Data + ((y + i) * pitch + x);
+	    for (j = width >> 1; j; j--) {
+		pix = ((R128PACKCOLOR4444(0xff, 0xff, 0xff, src[0])      ) |
+		       (R128PACKCOLOR4444(0xff, 0xff, 0xff, src[1]) << 16));
+		R128CCE(pix);
+		src += 2;
+	    }
+	}
+	break;
+
+    case GL_LUMINANCE:
+	for (i = 0; i < height; i++) {
+	    src = (CARD8 *)image->Data + ((y + i) * pitch + x);
+	    for (j = width >> 1; j; j--) {
+		pix = ((R128PACKCOLOR1555(src[0], src[0], src[0], 0xff)    ) |
+		       (R128PACKCOLOR1555(src[1], src[1], src[1], 0xff)<<16));
+		R128CCE(pix);
+		src += 2;
+	    }
+	}
+	break;
+
+    case GL_LUMINANCE_ALPHA:
+	for (i = 0; i < height; i++) {
+	    src = (CARD8 *)image->Data + ((y + i) * pitch + x) * 2;
+	    for (j = width >> 1; j; j--) {
+		pix =
+		    ((R128PACKCOLOR4444(src[0], src[0], src[0], src[1])    ) |
+		     (R128PACKCOLOR4444(src[2], src[2], src[2], src[3])<<16));
+		R128CCE(pix);
+		src += 4;
+	    }
+	}
+	break;
+
+    case GL_INTENSITY:
+	for (i = 0; i < height; i++) {
+	    src = (CARD8 *)image->Data + ((y + i) * pitch + x);
+	    for (j = width >> 1; j; j--) {
+		pix =
+		    ((R128PACKCOLOR4444(src[0], src[0], src[0], src[0])    ) |
+		     (R128PACKCOLOR4444(src[1], src[1], src[1], src[1])<<16));
+		R128CCE(pix);
+		src += 2;
+	    }
+	}
+	break;
+
+    default:
+	fprintf(stderr, "r128ConvertTexture16bpp: unsupported format 0x%x\n",
+		image->Format);
+    }
+}
+
+/* Convert a block of Mesa-formatted texture to a 32bpp hardware format */
+static void r128ConvertTexture32bpp(r128ContextPtr r128ctx,
+				    struct gl_texture_image *image,
+				    int x, int y, int width, int height,
+				    int pitch)
+{
+    CARD8  *src;
+    CARD32  pix;
+    int     i, j;
+
+    switch (image->Format) {
+    case GL_RGB:
+	for (i = 0; i < height; i++) {
+	    src = (CARD8 *)image->Data + ((y + i) * pitch + x) * 3;
+	    for (j = width; j; j--) {
+		pix = R128PACKCOLOR8888(src[0], src[1], src[2], 0xff);
+		R128CCE(pix);
+		src += 3;
+	    }
+	}
+	break;
+
+    case GL_RGBA:
+	for (i = 0; i < height; i++) {
+	    src = (CARD8 *)image->Data + ((y + i) * pitch + x) * 4;
+	    for (j = width; j; j--) {
+		pix = R128PACKCOLOR8888(src[0], src[1], src[2], src[3]);
+		R128CCE(pix);
+		src += 4;
+	    }
+	}
+	break;
+
+    case GL_ALPHA:
+	for (i = 0; i < height; i++) {
+	    src = (CARD8 *)image->Data + ((y + i) * pitch + x);
+	    for (j = width; j; j--) {
+		pix = R128PACKCOLOR8888(0xff, 0xff, 0xff, src[0]);
+		R128CCE(pix);
+		src += 1;
+	    }
+	}
+	break;
+
+    case GL_LUMINANCE:
+	for (i = 0 ; i < height ; i++) {
+	    src = (CARD8 *)image->Data + ((y + i) * pitch + x);
+	    for (j = width; j; j--) {
+		pix = R128PACKCOLOR8888(src[0], src[0], src[0], 0xff);
+		R128CCE(pix);
+		src += 1;
+	    }
+	}
+	break;
+
+    case GL_LUMINANCE_ALPHA:
+	for (i = 0; i < height; i++) {
+	    src = (CARD8 *)image->Data + ((y + i) * pitch + x) * 2;
+	    for (j = width; j; j-- ) {
+		pix = R128PACKCOLOR8888(src[0], src[0], src[0], src[1]);
+		R128CCE(pix);
+		src += 2;
+	    }
+	}
+	break;
+
+    case GL_INTENSITY:
+	for (i = 0; i < height; i++) {
+	    src = (CARD8 *)image->Data + ((y + i) * pitch + x);
+	    for (j = width; j; j--) {
+		pix = R128PACKCOLOR8888(src[0], src[0], src[0], src[0]);
+		R128CCE(pix);
+		src += 1;
+	    }
+	}
+	break;
+
+    default:
+	fprintf(stderr, "r128ConvertTexture32bpp: unsupported format 0x%x\n",
+		image->Format);
+    }
+}
+
+/* Upload the texture image associated with texture `t' at level `level'
+   at the address relative to `start'. */
+/* NOTE: This function is only called while holding the hardware lock */
+static void r128UploadSubImage(r128ContextPtr r128ctx,
+			       r128TexObjPtr t, int level,
+			       int x, int y, int width, int height)
+{
+    struct gl_texture_image *image;
+    int                      texelsPerDword = 0;
+    int                      imageWidth, imageHeight;
+    int                      remaining, rows;
+    int                      format, pitch, dwords;
+    CARD32                   offset;
+
+    /* Ensure we have a valid texture to upload */
+    if (level < 0 || level > R128_TEX_MAXLEVELS) return;
+    if (!(image = t->tObj->Image[level]))        return;
+
+    /* FIXME: support RGB888 (i.e., 24bpp) textures? */
+    switch (t->texelBytes) {
+    case 1: texelsPerDword = 4; break;
+    case 2: texelsPerDword = 2; break;
+    case 4: texelsPerDword = 1; break;
+    }
+
+    imageWidth  = image->Width;
+    imageHeight = image->Height;
+
+    format = t->textureFormat >> 16;
+
+    /* The texel upload routines have a minimum width, so force the size
+       if needed */
+    if (imageWidth < texelsPerDword) {
+	int factor;
+
+	factor        = texelsPerDword / imageWidth;
+	imageWidth    = texelsPerDword;
+	imageHeight /= factor;
+	if (imageHeight == 0) {
+	    /* In this case, the texel converter will actually walk a
+	       texel or two off the end of the image, but normal malloc
+	       alignment should prevent it from ever causing a fault. */
+	    imageHeight = 1;
+	}
+    }
+
+    /* We can't upload to a pitch less than 8 texels so we will need to
+       linearly upload all modified rows for textures smaller than this.
+       This makes the x/y/width/height different for the blitter and the
+       texture walker. */
+    if (imageWidth >= 8) {
+	/* The texture walker and the blitter look identical */
+	pitch      = imageWidth >> 3;
+    } else {
+	int factor;
+	int y2;
+	int start, end;
+
+	start = (y * imageWidth) & ~7;
+	end   = (y + height) * imageWidth;
+
+	if (end - start < 8) {
+	    /* Handle the case where the total number of texels uploaded
+               is < 8 */
+	    x       = 0;
+	    y       = start / 8;
+	    width   = end - start;
+	    height  = 1;
+	} else {
+	    /* Upload some number of full 8 texel blit rows */
+	    factor  = 8 / imageWidth;
+
+	    y2      = y + height - 1;
+	    y      /= factor;
+	    y2     /= factor;
+
+	    x       = 0;
+	    width   = 8;
+	    height  = y2 - y + 1;
+	}
+
+	/* Fixed pitch of 8 */
+	pitch = 1;
+    }
+
+    dwords = width * height / texelsPerDword;
+    offset = (CARD32)(t->bufAddr + t->image[level].offset);
+
+    /* Fix offset for AGP textures */
+    if (t->heap == R128_AGP_TEX_HEAP)
+	offset += R128_AGP_TEX_OFFSET + r128ctx->r128Screen->agpTexOffset;
+
+    if (R128_DEBUG_FLAGS & DEBUG_VERBOSE_API) {
+	fprintf(stderr, "r128UploadSubImage: %d,%d of %d,%d at %d,%d\n",
+		width, height, image->Width, image->Height, x, y);
+	fprintf(stderr, "          blit ofs: 0x%08x pitch: 0x%x dwords: %d "
+				  "level: %d format: %x\n",
+		(int)offset, pitch, dwords, level, format);
+    }
+
+    /* Subdivide the texture if required */
+    if (dwords < R128_CCE_PACKET_MAX_DWORDS) {
+	rows = height;
+    } else {
+	rows = (R128_CCE_PACKET_MAX_DWORDS * texelsPerDword) / (2 * width);
+    }
+
+    /* Flush the pixel cache, and mark the contents as Read Invalid */
+    R128CCE0(R128_CCE_PACKET0, R128_PC_GUI_CTLSTAT, 0);
+    R128CCE(r128ctx->regs.pc_gui_ctlstat |
+	    R128_PC_RI_GUI               |
+	    R128_PC_FLUSH_GUI);
+    R128CCE_SUBMIT_PACKETS();
+
+    /* Build the CCE host data blit header */
+    R128CCE3(R128_CCE_PACKET3_CNTL_HOSTDATA_BLT, 0);
+
+    /* DP_GUI_MASTER_CNTL */
+    R128CCE(R128_GMC_DST_PITCH_OFFSET_CNTL |
+	    R128_GMC_BRUSH_NONE            |
+	    (format << 8)                  |
+	    R128_GMC_SRC_DATATYPE_COLOR    |
+	    R128_ROP3_S                    |
+	    R128_DP_SRC_SOURCE_HOST_DATA   |
+	    R128_GMC_CLR_CMP_CNTL_DIS      |
+	    R128_AUX_CLIP_DIS              |
+	    R128_GMC_WR_MSK_DIS           );
+
+    /* DST_OFFSET_PITCH - fixed at the moment until we get better ring
+       control */
+    R128CCE((pitch << 21) | (offset>>5));
+
+    /* FRGD_COLOR, BKGD_COLOR */
+    R128CCE(0xffffffff);
+    R128CCE(0xffffffff);
+
+    for (remaining = height; remaining > 0; remaining -= rows, y += rows) {
+	height = (remaining >= rows) ? rows : remaining;
+	dwords = width * height / texelsPerDword;
+
+	if (R128_DEBUG_FLAGS & DEBUG_VERBOSE_API) {
+	    fprintf(stderr, "          blitting: %d,%d at %d,%d - %d dwords\n",
+		    width, height, x, y, dwords);
+	}
+
+	r128ctx->CCEbuf[0] &= ~R128_CCE_PACKET_COUNT_MASK;
+	r128ctx->CCEbuf[0] |= (dwords + 6) << 16;
+
+	/* Blit coords, size */
+	R128CCE((y      << 16) | x);
+	R128CCE((height << 16) | width);
+	R128CCE(dwords);
+
+	/* Actually do the texture conversion */
+	switch (t->texelBytes) {
+	case 1:
+	    r128ConvertTexture8bpp(r128ctx, image,
+				   x, y, width, height, width);
+	    break;
+	case 2:
+	    r128ConvertTexture16bpp(r128ctx, image,
+				    x, y, width, height, width);
+	    break;
+	case 4:
+	    r128ConvertTexture32bpp(r128ctx,
+				    image, x, y, width, height, width);
+	    break;
+	}
+
+	/* Flush the pixel cache */
+	R128CCE0(R128_CCE_PACKET0, R128_PC_GUI_CTLSTAT, 0 );
+	R128CCE(r128ctx->regs.pc_gui_ctlstat | R128_PC_FLUSH_GUI);
+
+	/* Save the partial blit header */
+	R128CCE_SUBMIT_PACKETS();
+	r128ctx->CCEcount = 5;
+    }
+    
+    /* Clean up CCE ring buffer */
+    r128ctx->CCEcount = 0;
+}
+
+/* Upload the texture images associated with texture `t'.  This might
+   require removing our own and/or other client's texture objects to
+   make room for these images. */
+/* NOTE: This function is only called while holding the hardware lock */
+int r128UploadTexImages(r128ContextPtr r128ctx, r128TexObjPtr t)
+{
+    int i;
+    int minLevel;
+    int maxLevel;
+    int heap;
+
+    if (!t) return 0;
+
+    /* Choose the heap appropriately */
+    heap = t->heap = R128_LOCAL_TEX_HEAP;
+    if (!r128ctx->r128Screen->IsPCI &&
+	t->totalSize > r128ctx->r128Screen->texSize[heap])
+	heap = t->heap = R128_AGP_TEX_HEAP;
+
+    /* Do we need to eject LRU texture objects? */
+    if (!t->memBlock) {
+	/* Allocate a memory block on a 4k boundary (1<<12 == 4096) */
+	t->memBlock = mmAllocMem(r128ctx->texHeap[heap], t->totalSize, 12, 0);
+
+	/* Try AGP before kicking anything out of local mem */
+	if (!t->memBlock && heap == R128_LOCAL_TEX_HEAP) {
+	    t->memBlock = mmAllocMem(r128ctx->texHeap[R128_AGP_TEX_HEAP],
+				     t->totalSize, 12, 0);
+
+	    if (t->memBlock) heap = t->heap = R128_AGP_TEX_HEAP;
+	}
+
+	/* Kick out textures until the requested texture fits */
+	while (!t->memBlock) {
+	    if (r128ctx->TexObjList[heap].prev->bound) {
+		fprintf(stderr,
+			"r128UploadTexImages: ran into bound texture\n");
+		return -1;
+	    }
+	    if (r128ctx->TexObjList[heap].prev ==
+		&(r128ctx->TexObjList[heap])) {
+		if (r128ctx->r128Screen->IsPCI) {
+		    fprintf(stderr, "r128UploadTexImages: upload texture "
+			    "failure on local texture heaps, sz=%d\n",
+			    t->totalSize);
+		    return -1;
+		} else if (heap == R128_LOCAL_TEX_HEAP) {
+		    heap = t->heap = R128_AGP_TEX_HEAP;
+		    continue;
+		} else {
+		    fprintf(stderr, "r128UploadTexImages: upload texture "
+			    "failure on both local and AGP texture heaps, "
+			    "sz=%d\n",
+			    t->totalSize);
+		    return -1;
+		}
+	    }
+
+	    r128DestroyTexObj(r128ctx, r128ctx->TexObjList[heap].prev);
+
+	    t->memBlock = mmAllocMem(r128ctx->texHeap[heap],
+				     t->totalSize, 12, 0);
+	}
+
+	/* Set the base offset of the texture image */
+	t->bufAddr = (unsigned char *)r128ctx->r128Screen->texOffset[heap];
+	t->bufAddr += t->memBlock->ofs;
+
+	maxLevel = ((t->regs.size_pitch & R128_TEX_SIZE_MASK) >>
+		    R128_TEX_SIZE_SHIFT);
+	minLevel = ((t->regs.size_pitch & R128_TEX_MIN_SIZE_MASK) >>
+		    R128_TEX_MIN_SIZE_SHIFT);
+
+	/* Update the hardware's texture image addresses */
+	switch (t->bound) {
+	case 1:
+	    /* Set texture offsets */
+	    if (t->regs.tex_cntl & R128_MIP_MAP_DISABLE) {
+		for (i = 0; i < R128_TEX_MAXLEVELS; i++)
+		    r128ctx->regs.prim_tex_offset[i] = (CARD32)t->bufAddr;
+	    } else {
+		for (i = maxLevel; i >= minLevel; i--)
+		    r128ctx->regs.prim_tex_offset[i] =
+			t->image[maxLevel-i].offset + (CARD32)t->bufAddr;
+	    }
+	    /* Fix AGP texture offsets */
+	    if (heap == R128_AGP_TEX_HEAP)
+		for (i = 0; i < R128_TEX_MAXLEVELS; i++)
+		    r128ctx->regs.prim_tex_offset[i] +=
+			R128_AGP_TEX_OFFSET +
+			r128ctx->r128Screen->agpTexOffset;
+	    /* Force loading the new state into the hardware */
+	    r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+	    r128ctx->dirty_context |= R128_CTX_TEX0STATE;
+	    break;
+
+	case 2:
+	    /* Set texture offsets */
+	    if (t->regs.tex_cntl & R128_MIP_MAP_DISABLE) {
+		for (i = 0; i < R128_TEX_MAXLEVELS; i++)
+		    r128ctx->regs.sec_tex_offset[i] = (CARD32)t->bufAddr;
+	    } else {
+		for (i = maxLevel; i >= minLevel; i--)
+		    r128ctx->regs.sec_tex_offset[i] = 
+			t->image[maxLevel-i].offset + (CARD32)t->bufAddr;
+	    }
+	    /* Fix AGP texture offsets */
+	    if (heap == R128_AGP_TEX_HEAP)
+		for (i = 0; i < R128_TEX_MAXLEVELS; i++)
+		    r128ctx->regs.sec_tex_offset[i] +=
+			R128_AGP_TEX_OFFSET +
+			r128ctx->r128Screen->agpTexOffset;
+	    /* Force loading the new state into the hardware */
+	    r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+	    r128ctx->dirty_context |= R128_CTX_TEX1STATE;
+	    break;
+
+	default:
+	    return -1;
+	}
+    }
+
+    /* Let the world know we've used this memory recently */
+    r128UpdateTexLRU(r128ctx, t);
+
+    /* Upload any images that are new */
+    if (t->dirty_images) {
+	int num_levels = (((t->regs.size_pitch & R128_TEX_SIZE_MASK) >>
+			   R128_TEX_SIZE_SHIFT) -
+			  ((t->regs.size_pitch & R128_TEX_MIN_SIZE_MASK) >>
+			   R128_TEX_MIN_SIZE_SHIFT));
+
+	for (i = 0; i <= num_levels; i++) {
+	    if (t->dirty_images & (1<<i)) {
+		r128UploadSubImage(r128ctx, t, i, 0, 0,
+				   t->image[i].width, t->image[i].height);
+	    }
+	}
+
+	r128ctx->regs.tex_cntl_c |= R128_TEX_CACHE_FLUSH;
+
+	r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+	r128ctx->dirty_context |= R128_CTX_ENGINESTATE;
+    }
+
+    t->dirty_images = 0;
+    return 0;
+}
+
+/* Update the hardware state for texture unit 0 */
+/* NOTE: This function is only called while holding the hardware lock */
+static void r128UpdateTex0State(r128ContextPtr r128ctx)
+{
+    GLcontext                *ctx = r128ctx->glCtx;
+    r128TexObjPtr             t;
+    struct gl_texture_object *tObj;
+    int                       i;
+    CARD32                    tex_size_pitch, tex_combine_cntl;
+
+    /* Only update the hardware texture state if the texture is current,
+       complete and enabled. */
+    if (!(tObj = ctx->Texture.Unit[0].Current))       return;
+    if ((tObj != ctx->Texture.Unit[0].CurrentD[2]) &&
+	(tObj != ctx->Texture.Unit[0].CurrentD[1]))   return;
+    if (!tObj->Complete)                              return;
+
+    /* If neither tex0 nor tex1 is enabled, then disable tex0.  However,
+       if tex1 is enabled but tex0 is disabled, then we need to enable
+       tex0 and have it to copy the input (see how tex_combine_cntl is
+       setup below). */
+    if (!(ctx->Texture.Enabled & (ENABLE_TEX0 | ENABLE_TEX1))) {
+	r128ctx->regs.tex_cntl_c &= ~R128_TEXMAP_ENABLE;
+	return;
+    }
+
+    /* If this is the first time the texture has been used, then create
+       a new texture object for it. */
+    t = tObj->DriverData;
+    if (!t) t = r128CreateTexObj(r128ctx, tObj);
+    if (!t) return;
+
+    if (R128_DEBUG_FLAGS & DEBUG_VERBOSE_API)
+	fprintf(stderr, "r128UpdateTex0State(%p, 0x%08x)\n",
+		tObj, (int)t->dirty_images);
+
+    /* Force any texture images to be loaded into the hardware */
+    if (t->dirty_images) r128ctx->dirty |= R128_UPDATE_TEX0IMAGES;
+
+    /* Bind texture to texture 0 unit */
+    r128ctx->CurrentTexObj[0] = t;
+    t->bound = 1;
+
+    if (t->memBlock) r128UpdateTexLRU(r128ctx, t);
+
+    /* Set the texture environment state */
+    switch (ctx->Texture.Unit[0].EnvMode) {
+    case GL_REPLACE:
+	switch (tObj->Image[0]->Format) {
+	case GL_RGBA:
+	case GL_LUMINANCE_ALPHA:
+	case GL_INTENSITY:
+	    tex_combine_cntl = (R128_COMB_DIS                |
+				R128_COLOR_FACTOR_TEX        |
+				R128_INPUT_FACTOR_INT_COLOR  |
+				R128_COMB_ALPHA_DIS          |
+				R128_ALPHA_FACTOR_TEX_ALPHA  |
+				R128_INP_FACTOR_A_INT_ALPHA);
+	    break;
+	case GL_RGB:
+	case GL_LUMINANCE:
+	    tex_combine_cntl = (R128_COMB_DIS                |
+				R128_COLOR_FACTOR_TEX        |
+				R128_INPUT_FACTOR_INT_COLOR  |
+				R128_COMB_ALPHA_COPY_INP     |
+				R128_ALPHA_FACTOR_TEX_ALPHA  |
+				R128_INP_FACTOR_A_INT_ALPHA);
+	    break;
+	case GL_ALPHA:
+	    tex_combine_cntl = (R128_COMB_COPY_INP           |
+				R128_COLOR_FACTOR_TEX        |
+				R128_INPUT_FACTOR_INT_COLOR  |
+				R128_COMB_ALPHA_DIS          |
+				R128_ALPHA_FACTOR_TEX_ALPHA  |
+				R128_INP_FACTOR_A_INT_ALPHA);
+	    break;
+	case GL_COLOR_INDEX:
+	default:
+	    return;
+	}
+	break;
+
+    case GL_BLEND:
+	switch (tObj->Image[0]->Format) {
+	case GL_RGBA:
+	case GL_LUMINANCE_ALPHA:
+	    switch (r128ctx->regs.constant_color_c &
+		    R128_CONSTANT_COLOR_MASK) {
+	    case R128_CONSTANT_COLOR_ZERO:
+		tex_combine_cntl = (R128_COMB_MODULATE           |
+				    R128_COLOR_FACTOR_NTEX       |
+				    R128_INPUT_FACTOR_INT_COLOR  |
+				    R128_COMB_ALPHA_MODULATE     |
+				    R128_ALPHA_FACTOR_TEX_ALPHA  |
+				    R128_INP_FACTOR_A_INT_ALPHA);
+		break;
+	    case R128_CONSTANT_COLOR_ONE:
+	    default:
+		tex_combine_cntl = (R128_COMB_MODULATE           |
+				    R128_COLOR_FACTOR_TEX        |
+				    R128_INPUT_FACTOR_INT_COLOR  |
+				    R128_COMB_ALPHA_MODULATE     |
+				    R128_ALPHA_FACTOR_TEX_ALPHA  |
+				    R128_INP_FACTOR_A_INT_ALPHA);
+		r128ctx->Fallback |= R128_FALLBACK_TEXTURE;
+		break;
+	    }
+	    break;
+	case GL_RGB:
+	case GL_LUMINANCE:
+	    switch (r128ctx->regs.constant_color_c &
+		    R128_CONSTANT_COLOR_MASK) {
+	    case R128_CONSTANT_COLOR_ZERO:
+		tex_combine_cntl = (R128_COMB_MODULATE           |
+				    R128_COLOR_FACTOR_NTEX       |
+				    R128_INPUT_FACTOR_INT_COLOR  |
+				    R128_COMB_ALPHA_COPY_INP     |
+				    R128_ALPHA_FACTOR_TEX_ALPHA  |
+				    R128_INP_FACTOR_A_INT_ALPHA);
+		break;
+	    case R128_CONSTANT_COLOR_ONE:
+	    default:
+		tex_combine_cntl = (R128_COMB_MODULATE           |
+				    R128_COLOR_FACTOR_TEX        |
+				    R128_INPUT_FACTOR_INT_COLOR  |
+				    R128_COMB_ALPHA_COPY_INP     |
+				    R128_ALPHA_FACTOR_TEX_ALPHA  |
+				    R128_INP_FACTOR_A_INT_ALPHA);
+		r128ctx->Fallback |= R128_FALLBACK_TEXTURE;
+		break;
+	    }
+	    break;
+	case GL_ALPHA:
+	    tex_combine_cntl = (R128_COMB_COPY_INP           |
+				R128_COLOR_FACTOR_TEX        |
+				R128_INPUT_FACTOR_INT_COLOR  |
+				R128_COMB_ALPHA_MODULATE     |
+				R128_ALPHA_FACTOR_TEX_ALPHA  |
+				R128_INP_FACTOR_A_INT_ALPHA);
+	    break;
+	case GL_INTENSITY:
+	    switch (r128ctx->regs.constant_color_c &
+		    R128_CONSTANT_COLOR_MASK) {
+	    case R128_CONSTANT_COLOR_ZERO:
+		tex_combine_cntl = (R128_COMB_MODULATE           |
+				    R128_COLOR_FACTOR_NTEX       |
+				    R128_INPUT_FACTOR_INT_COLOR  |
+				    R128_COMB_ALPHA_MODULATE     |
+				    R128_ALPHA_FACTOR_NTEX_ALPHA |
+				    R128_INP_FACTOR_A_INT_ALPHA);
+		break;
+	    case R128_CONSTANT_COLOR_ONE:
+	    default:
+		tex_combine_cntl = (R128_COMB_MODULATE           |
+				    R128_COLOR_FACTOR_TEX        |
+				    R128_INPUT_FACTOR_INT_COLOR  |
+				    R128_COMB_ALPHA_MODULATE     |
+				    R128_ALPHA_FACTOR_TEX_ALPHA  |
+				    R128_INP_FACTOR_A_INT_ALPHA);
+		r128ctx->Fallback |= R128_FALLBACK_TEXTURE;
+		break;
+	    }
+	    break;
+	case GL_COLOR_INDEX:
+	default:
+	    return;
+	}
+	break;
+
+    case GL_MODULATE:
+	switch (tObj->Image[0]->Format) {
+	case GL_RGBA:
+	case GL_LUMINANCE_ALPHA:
+	case GL_INTENSITY:
+	    tex_combine_cntl = (R128_COMB_MODULATE           |
+				R128_COLOR_FACTOR_TEX        |
+				R128_INPUT_FACTOR_INT_COLOR  |
+				R128_COMB_ALPHA_MODULATE     |
+				R128_ALPHA_FACTOR_TEX_ALPHA  |
+				R128_INP_FACTOR_A_INT_ALPHA);
+	    break;
+	case GL_RGB:
+	case GL_LUMINANCE:
+	    tex_combine_cntl = (R128_COMB_MODULATE           |
+				R128_COLOR_FACTOR_TEX        |
+				R128_INPUT_FACTOR_INT_COLOR  |
+				R128_COMB_ALPHA_COPY_INP     |
+				R128_ALPHA_FACTOR_TEX_ALPHA  |
+				R128_INP_FACTOR_A_INT_ALPHA);
+	    break;
+	case GL_ALPHA:
+	    tex_combine_cntl = (R128_COMB_COPY_INP           |
+				R128_COLOR_FACTOR_TEX        |
+				R128_INPUT_FACTOR_INT_COLOR  |
+				R128_COMB_ALPHA_MODULATE     |
+				R128_ALPHA_FACTOR_TEX_ALPHA  |
+				R128_INP_FACTOR_A_INT_ALPHA);
+	    break;
+	case GL_COLOR_INDEX:
+	default:
+	    return;
+	}
+	break;
+
+    case GL_DECAL:
+	switch (tObj->Image[0]->Format) {
+	case GL_RGBA:
+	    tex_combine_cntl = (R128_COMB_BLEND_TEXTURE      |
+				R128_COLOR_FACTOR_TEX        |
+				R128_INPUT_FACTOR_INT_COLOR  |
+				R128_COMB_ALPHA_COPY_INP     |
+				R128_ALPHA_FACTOR_TEX_ALPHA  |
+				R128_INP_FACTOR_A_INT_ALPHA);
+	    break;
+	case GL_RGB:
+	    tex_combine_cntl = (R128_COMB_DIS                |
+				R128_COLOR_FACTOR_TEX        |
+				R128_INPUT_FACTOR_INT_COLOR  |
+				R128_COMB_ALPHA_COPY_INP     |
+				R128_ALPHA_FACTOR_TEX_ALPHA  |
+				R128_INP_FACTOR_A_INT_ALPHA);
+	    break;
+	case GL_ALPHA:
+	case GL_LUMINANCE:
+	case GL_LUMINANCE_ALPHA:
+	case GL_INTENSITY:
+	    /* Undefined behaviour - just copy the input */
+	    tex_combine_cntl = (R128_COMB_COPY_INP           |
+				R128_COLOR_FACTOR_TEX        |
+				R128_INPUT_FACTOR_INT_COLOR  |
+				R128_COMB_ALPHA_COPY_INP     |
+				R128_ALPHA_FACTOR_TEX_ALPHA  |
+				R128_INP_FACTOR_A_INT_ALPHA);
+	    break;
+	case GL_COLOR_INDEX:
+	default:
+	    return;
+	}
+	break;
+
+    case GL_ADD:
+	switch (tObj->Image[0]->Format) {
+	case GL_RGBA:
+	case GL_LUMINANCE_ALPHA:
+	case GL_INTENSITY:
+	    tex_combine_cntl = (R128_COMB_ADD                |
+				R128_COLOR_FACTOR_TEX        |
+				R128_INPUT_FACTOR_INT_COLOR  |
+				R128_COMB_ALPHA_MODULATE     |
+				R128_ALPHA_FACTOR_TEX_ALPHA  |
+				R128_INP_FACTOR_A_INT_ALPHA);
+	    break;
+	case GL_RGB:
+	case GL_LUMINANCE:
+	    tex_combine_cntl = (R128_COMB_ADD                |
+				R128_COLOR_FACTOR_TEX        |
+				R128_INPUT_FACTOR_INT_COLOR  |
+				R128_COMB_ALPHA_COPY_INP     |
+				R128_ALPHA_FACTOR_TEX_ALPHA  |
+				R128_INP_FACTOR_A_INT_ALPHA);
+	    break;
+	case GL_ALPHA:
+	    tex_combine_cntl = (R128_COMB_COPY_INP           |
+				R128_COLOR_FACTOR_TEX        |
+				R128_INPUT_FACTOR_INT_COLOR  |
+				R128_COMB_ALPHA_MODULATE     |
+				R128_ALPHA_FACTOR_TEX_ALPHA  |
+				R128_INP_FACTOR_A_INT_ALPHA);
+	    break;
+	case GL_COLOR_INDEX:
+	default:
+	    return;
+	}
+	break;
+
+    default:
+	return;
+    }
+
+    /* If tex0 is disabled, then make sure it just copies the input */
+    if (!(ctx->Texture.Enabled & ENABLE_TEX0))
+	tex_combine_cntl = (R128_COMB_COPY_INP           |
+			    R128_COLOR_FACTOR_TEX        |
+			    R128_INPUT_FACTOR_INT_COLOR  |
+			    R128_COMB_ALPHA_COPY_INP     |
+			    R128_ALPHA_FACTOR_TEX_ALPHA  |
+			    R128_INP_FACTOR_A_INT_ALPHA);
+
+    /* Enable tex0 */
+    r128ctx->regs.tex_cntl_c |= R128_TEXMAP_ENABLE;
+
+    tex_size_pitch  = r128ctx->regs.tex_size_pitch_c;
+    tex_size_pitch &= ~R128_TEX_SIZE_PITCH_MASK;
+    tex_size_pitch |= t->regs.size_pitch << R128_TEX_SIZE_PITCH_SHIFT;
+
+    /* Set the primary texture state in r128ctx->regs */
+    r128ctx->regs.prim_tex_cntl_c             = t->regs.tex_cntl;
+    r128ctx->regs.prim_texture_combine_cntl_c = tex_combine_cntl;
+    r128ctx->regs.tex_size_pitch_c            = tex_size_pitch;
+    r128ctx->regs.prim_texture_border_color_c = t->regs.border_color;
+
+    /* Set texture offsets */
+    if (t->regs.tex_cntl & R128_MIP_MAP_DISABLE) {
+	for (i = 0; i < R128_TEX_MAXLEVELS; i++)
+	    r128ctx->regs.prim_tex_offset[i] = (CARD32)t->bufAddr;
+    } else {
+	int maxLevel = ((t->regs.size_pitch & R128_TEX_SIZE_MASK) >>
+			R128_TEX_SIZE_SHIFT);
+	int minLevel = ((t->regs.size_pitch & R128_TEX_MIN_SIZE_MASK) >>
+			R128_TEX_MIN_SIZE_SHIFT);
+	for (i = maxLevel; i >= minLevel; i--)
+	    r128ctx->regs.prim_tex_offset[i] = 
+		t->image[maxLevel-i].offset + (CARD32)t->bufAddr;
+    }
+    /* Fix AGP texture offsets */
+    if (t->heap == R128_AGP_TEX_HEAP)
+	for (i = 0; i < R128_TEX_MAXLEVELS; i++)
+	    r128ctx->regs.prim_tex_offset[i] +=
+		R128_AGP_TEX_OFFSET + r128ctx->r128Screen->agpTexOffset;
+
+    /* Force loading the new state into the hardware */
+    r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+    r128ctx->dirty_context |= R128_CTX_TEX0STATE | R128_CTX_ENGINESTATE;
+}
+
+/* Update the hardware state for texture unit 1 */
+/* NOTE: This function is only called while holding the hardware lock */
+static void r128UpdateTex1State(r128ContextPtr r128ctx)
+{
+    GLcontext                *ctx = r128ctx->glCtx;
+    r128TexObjPtr             t;
+    struct gl_texture_object *tObj;
+    int                       i;
+    CARD32                    tex_size_pitch, tex_combine_cntl, tex_cntl;
+
+    /* Only update the hardware texture state if the texture is current,
+       complete and enabled. */
+    if (!(tObj = ctx->Texture.Unit[1].Current))       return;
+    if ((tObj != ctx->Texture.Unit[1].CurrentD[2]) &&
+	(tObj != ctx->Texture.Unit[1].CurrentD[1]))   return;
+    if (!tObj->Complete)                              return;
+
+    /* If tex1 is not enabled, then disable it */
+    if (!(ctx->Texture.Enabled & ENABLE_TEX1)) {
+	r128ctx->regs.tex_cntl_c &= ~R128_SEC_TEXMAP_ENABLE;
+	return;
+    }
+
+    /* If this is the first time the texture has been used, then create
+       a new texture object for it. */
+    t = tObj->DriverData;
+    if (!t) t = r128CreateTexObj(r128ctx, tObj);
+    if (!t) return;
+
+    if (R128_DEBUG_FLAGS & DEBUG_VERBOSE_API)
+	fprintf(stderr, "r128UpdateTex1State(%p, 0x%08x)\n",
+		tObj, (int)t->dirty_images);
+
+    /* Force any texture images to be loaded into the hardware */
+    if (t->dirty_images) r128ctx->dirty |= R128_UPDATE_TEX1IMAGES;
+
+    /* Bind texture to texture 1 unit */
+    r128ctx->CurrentTexObj[1] = t;
+    t->bound = 2;
+
+    if (t->memBlock) r128UpdateTexLRU(r128ctx, t);
+
+    /* Set the texture environment state */
+    switch (ctx->Texture.Unit[1].EnvMode) {
+    case GL_REPLACE:
+	switch (tObj->Image[0]->Format) {
+	case GL_RGBA:
+	case GL_LUMINANCE_ALPHA:
+	case GL_INTENSITY:
+	    tex_combine_cntl = (R128_COMB_DIS                 |
+				R128_COLOR_FACTOR_TEX         |
+				R128_INPUT_FACTOR_PREV_COLOR  |
+				R128_COMB_ALPHA_DIS           |
+				R128_ALPHA_FACTOR_TEX_ALPHA   |
+				R128_INP_FACTOR_A_PREV_ALPHA);
+	    break;
+	case GL_RGB:
+	case GL_LUMINANCE:
+	    tex_combine_cntl = (R128_COMB_DIS                 |
+				R128_COLOR_FACTOR_TEX         |
+				R128_INPUT_FACTOR_PREV_COLOR  |
+				R128_COMB_ALPHA_COPY_INP      |
+				R128_ALPHA_FACTOR_TEX_ALPHA   |
+				R128_INP_FACTOR_A_PREV_ALPHA);
+	    break;
+	case GL_ALPHA:
+	    tex_combine_cntl = (R128_COMB_COPY_INP            |
+				R128_COLOR_FACTOR_TEX         |
+				R128_INPUT_FACTOR_PREV_COLOR  |
+				R128_COMB_ALPHA_DIS           |
+				R128_ALPHA_FACTOR_TEX_ALPHA   |
+				R128_INP_FACTOR_A_PREV_ALPHA);
+	    break;
+	case GL_COLOR_INDEX:
+	default:
+	    return;
+	}
+	break;
+
+    case GL_BLEND:
+	switch (tObj->Image[0]->Format) {
+	case GL_RGBA:
+	case GL_LUMINANCE_ALPHA:
+	    switch ( r128ctx->regs.constant_color_c & R128_CONSTANT_COLOR_MASK ) {
+	    case R128_CONSTANT_COLOR_ZERO:
+		tex_combine_cntl = (R128_COMB_MODULATE            |
+				    R128_COLOR_FACTOR_NTEX        |
+				    R128_INPUT_FACTOR_PREV_COLOR  |
+				    R128_COMB_ALPHA_MODULATE      |
+				    R128_ALPHA_FACTOR_TEX_ALPHA   |
+				    R128_INP_FACTOR_A_PREV_ALPHA);
+		break;
+	    case R128_CONSTANT_COLOR_ONE:
+	    default:
+		tex_combine_cntl = (R128_COMB_MODULATE            |
+				    R128_COLOR_FACTOR_TEX         |
+				    R128_INPUT_FACTOR_PREV_COLOR  |
+				    R128_COMB_ALPHA_MODULATE      |
+				    R128_ALPHA_FACTOR_TEX_ALPHA   |
+				    R128_INP_FACTOR_A_PREV_ALPHA);
+		r128ctx->Fallback |= R128_FALLBACK_TEXTURE;
+		break;
+	    }
+	    break;
+	case GL_RGB:
+	case GL_LUMINANCE:
+	    switch (r128ctx->regs.constant_color_c &
+		    R128_CONSTANT_COLOR_MASK) {
+	    case R128_CONSTANT_COLOR_ZERO:
+		tex_combine_cntl = (R128_COMB_MODULATE            |
+				    R128_COLOR_FACTOR_NTEX        |
+				    R128_INPUT_FACTOR_PREV_COLOR  |
+				    R128_COMB_ALPHA_COPY_INP      |
+				    R128_ALPHA_FACTOR_TEX_ALPHA   |
+				    R128_INP_FACTOR_A_PREV_ALPHA );
+		break;
+	    case R128_CONSTANT_COLOR_ONE:
+	    default:
+		tex_combine_cntl = (R128_COMB_MODULATE            |
+				    R128_COLOR_FACTOR_TEX         |
+				    R128_INPUT_FACTOR_PREV_COLOR  |
+				    R128_COMB_ALPHA_COPY_INP      |
+				    R128_ALPHA_FACTOR_TEX_ALPHA   |
+				    R128_INP_FACTOR_A_PREV_ALPHA );
+		r128ctx->Fallback |= R128_FALLBACK_TEXTURE;
+		break;
+	    }
+	    break;
+	case GL_ALPHA:
+	    tex_combine_cntl = (R128_COMB_COPY_INP            |
+				R128_COLOR_FACTOR_TEX         |
+				R128_INPUT_FACTOR_PREV_COLOR  |
+				R128_COMB_ALPHA_MODULATE      |
+				R128_ALPHA_FACTOR_TEX_ALPHA   |
+				R128_INP_FACTOR_A_PREV_ALPHA);
+	    break;
+	case GL_INTENSITY:
+	    switch (r128ctx->regs.constant_color_c &
+		    R128_CONSTANT_COLOR_MASK) {
+	    case R128_CONSTANT_COLOR_ZERO:
+		tex_combine_cntl = (R128_COMB_MODULATE            |
+				    R128_COLOR_FACTOR_NTEX        |
+				    R128_INPUT_FACTOR_PREV_COLOR  |
+				    R128_COMB_ALPHA_MODULATE      |
+				    R128_ALPHA_FACTOR_NTEX_ALPHA  |
+				    R128_INP_FACTOR_A_PREV_ALPHA);
+		break;
+	    case R128_CONSTANT_COLOR_ONE:
+	    default:
+		tex_combine_cntl = (R128_COMB_MODULATE            |
+				    R128_COLOR_FACTOR_TEX         |
+				    R128_INPUT_FACTOR_PREV_COLOR  |
+				    R128_COMB_ALPHA_MODULATE      |
+				    R128_ALPHA_FACTOR_TEX_ALPHA   |
+				    R128_INP_FACTOR_A_PREV_ALPHA);
+		r128ctx->Fallback |= R128_FALLBACK_TEXTURE;
+		break;
+	    }
+	    break;
+	case GL_COLOR_INDEX:
+	default:
+	    return;
+	}
+	break;
+
+    case GL_MODULATE:
+	switch (tObj->Image[0]->Format) {
+	case GL_RGBA:
+	case GL_LUMINANCE_ALPHA:
+	case GL_INTENSITY:
+	    tex_combine_cntl = (R128_COMB_MODULATE            |
+				R128_COLOR_FACTOR_TEX         |
+				R128_INPUT_FACTOR_PREV_COLOR  |
+				R128_COMB_ALPHA_MODULATE      |
+				R128_ALPHA_FACTOR_TEX_ALPHA   |
+				R128_INP_FACTOR_A_PREV_ALPHA);
+	    break;
+	case GL_RGB:
+	case GL_LUMINANCE:
+	    tex_combine_cntl = (R128_COMB_MODULATE            |
+				R128_COLOR_FACTOR_TEX         |
+				R128_INPUT_FACTOR_PREV_COLOR  |
+				R128_COMB_ALPHA_COPY_INP      |
+				R128_ALPHA_FACTOR_TEX_ALPHA   |
+				R128_INP_FACTOR_A_PREV_ALPHA);
+	    break;
+	case GL_ALPHA:
+	    tex_combine_cntl = (R128_COMB_COPY_INP            |
+				R128_COLOR_FACTOR_TEX         |
+				R128_INPUT_FACTOR_PREV_COLOR  |
+				R128_COMB_ALPHA_MODULATE      |
+				 R128_ALPHA_FACTOR_TEX_ALPHA   |
+				R128_INP_FACTOR_A_PREV_ALPHA);
+	    break;
+	case GL_COLOR_INDEX:
+	default:
+	    return;
+	}
+	break;
+
+    case GL_DECAL:
+	switch (tObj->Image[0]->Format) {
+	case GL_RGBA:
+	    tex_combine_cntl = (R128_COMB_BLEND_TEXTURE       |
+				R128_COLOR_FACTOR_TEX         |
+				R128_INPUT_FACTOR_PREV_COLOR  |
+				R128_COMB_ALPHA_COPY_INP      |
+				R128_ALPHA_FACTOR_TEX_ALPHA   |
+				R128_INP_FACTOR_A_PREV_ALPHA);
+	    break;
+	case GL_RGB:
+	    tex_combine_cntl = (R128_COMB_DIS                 |
+				R128_COLOR_FACTOR_TEX         |
+				R128_INPUT_FACTOR_PREV_COLOR  |
+				R128_COMB_ALPHA_COPY_INP      |
+				R128_ALPHA_FACTOR_TEX_ALPHA   |
+				R128_INP_FACTOR_A_PREV_ALPHA);
+	    break;
+	case GL_ALPHA:
+	case GL_LUMINANCE:
+	case GL_LUMINANCE_ALPHA:
+	case GL_INTENSITY:
+	    /* Undefined behaviour - just copy the input */
+	    tex_combine_cntl = (R128_COMB_COPY_INP            |
+				R128_COLOR_FACTOR_TEX         |
+				R128_INPUT_FACTOR_PREV_COLOR  |
+				R128_COMB_ALPHA_COPY_INP      |
+				R128_ALPHA_FACTOR_TEX_ALPHA   |
+				R128_INP_FACTOR_A_PREV_ALPHA);
+	    break;
+	case GL_COLOR_INDEX:
+	default:
+	    return;
+	}
+	break;
+
+    case GL_ADD:
+	switch (tObj->Image[0]->Format) {
+	case GL_RGBA:
+	case GL_LUMINANCE_ALPHA:
+	case GL_INTENSITY:
+	    tex_combine_cntl = (R128_COMB_ADD                 |
+				R128_COLOR_FACTOR_TEX         |
+				R128_INPUT_FACTOR_PREV_COLOR  |
+				R128_COMB_ALPHA_MODULATE      |
+				R128_ALPHA_FACTOR_TEX_ALPHA   |
+				R128_INP_FACTOR_A_PREV_ALPHA);
+	    break;
+	case GL_RGB:
+	case GL_LUMINANCE:
+	    tex_combine_cntl = (R128_COMB_ADD                 |
+				R128_COLOR_FACTOR_TEX         |
+				R128_INPUT_FACTOR_PREV_COLOR  |
+				R128_COMB_ALPHA_COPY_INP      |
+				R128_ALPHA_FACTOR_TEX_ALPHA   |
+				R128_INP_FACTOR_A_PREV_ALPHA);
+	    break;
+	case GL_ALPHA:
+	    tex_combine_cntl = (R128_COMB_COPY_INP            |
+				R128_COLOR_FACTOR_TEX         |
+				R128_INPUT_FACTOR_PREV_COLOR  |
+				R128_COMB_ALPHA_MODULATE      |
+				R128_ALPHA_FACTOR_TEX_ALPHA   |
+				R128_INP_FACTOR_A_PREV_ALPHA);
+	    break;
+	case GL_COLOR_INDEX:
+	default:
+	    return;
+	}
+	break;
+
+    default:
+	return;
+    }
+
+    /* Enable tex1 */
+    r128ctx->regs.tex_cntl_c |= R128_SEC_TEXMAP_ENABLE;
+
+    tex_size_pitch  = r128ctx->regs.tex_size_pitch_c;
+    tex_size_pitch &= ~R128_SEC_TEX_SIZE_PITCH_MASK;
+    tex_size_pitch |= t->regs.size_pitch << R128_SEC_TEX_SIZE_PITCH_SHIFT;
+
+    tex_cntl        = t->regs.tex_cntl | R128_SEC_SELECT_SEC_ST;
+
+    /* Set the secondary texture state in r128ctx->regs */
+    r128ctx->regs.sec_tex_cntl_c              = tex_cntl;
+    r128ctx->regs.sec_tex_combine_cntl_c      = tex_combine_cntl;
+    r128ctx->regs.tex_size_pitch_c            = tex_size_pitch;
+    r128ctx->regs.sec_texture_border_color_c  = t->regs.border_color;
+
+    /* Set texture offsets */
+    if (t->regs.tex_cntl & R128_MIP_MAP_DISABLE) {
+	for (i = 0; i < R128_TEX_MAXLEVELS; i++)
+	    r128ctx->regs.sec_tex_offset[i] = (CARD32)t->bufAddr;
+    } else {
+	int maxLevel = ((t->regs.size_pitch & R128_TEX_SIZE_MASK) >>
+			R128_TEX_SIZE_SHIFT);
+	int minLevel = ((t->regs.size_pitch & R128_TEX_MIN_SIZE_MASK) >>
+			R128_TEX_MIN_SIZE_SHIFT);
+	for (i = maxLevel; i >= minLevel; i--)
+	    r128ctx->regs.sec_tex_offset[i] = 
+		t->image[maxLevel-i].offset + (CARD32)t->bufAddr;
+    }
+    /* Fix AGP texture offsets */
+    if (t->heap == R128_AGP_TEX_HEAP)
+	for (i = 0; i < R128_TEX_MAXLEVELS; i++)
+	    r128ctx->regs.sec_tex_offset[i] +=
+		R128_AGP_TEX_OFFSET + r128ctx->r128Screen->agpTexOffset;
+
+    /* Force loading the new state into the hardware */
+    r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+    r128ctx->dirty_context |= R128_CTX_TEX1STATE | R128_CTX_ENGINESTATE;
+}
+
+/* Update the hardware texture state */
+/* NOTE: This function is only called while holding the hardware lock */
+void r128UpdateTextureState(r128ContextPtr r128ctx)
+{
+    /* Clear the GL_BLEND texturing fallback */
+    r128ctx->Fallback &= ~R128_FALLBACK_TEXTURE;
+
+    /* Unbind any currently bound textures */
+    if (r128ctx->CurrentTexObj[0]) r128ctx->CurrentTexObj[0]->bound = 0;
+    if (r128ctx->CurrentTexObj[1]) r128ctx->CurrentTexObj[1]->bound = 0;
+    r128ctx->CurrentTexObj[0] = NULL;
+    r128ctx->CurrentTexObj[1] = NULL;
+
+    /* Update the texture unit 0/1 state */
+    r128UpdateTex0State(r128ctx);
+    r128UpdateTex1State(r128ctx);
+}
+
+/* Set the texture wrap mode */
+static void r128SetTexWrap(r128TexObjPtr t, GLenum swrap, GLenum twrap)
+{
+    t->regs.tex_cntl &= ~(R128_TEX_CLAMP_S_MASK | R128_TEX_CLAMP_T_MASK);
+
+    switch (swrap) {
+    case GL_CLAMP:  t->regs.tex_cntl |= R128_TEX_CLAMP_S_CLAMP; break;
+    case GL_REPEAT: t->regs.tex_cntl |= R128_TEX_CLAMP_S_WRAP;  break;
+    default: /* ERROR!! */ return;
+    }
+
+    switch (twrap) {
+    case GL_CLAMP:  t->regs.tex_cntl |= R128_TEX_CLAMP_T_CLAMP; break;
+    case GL_REPEAT: t->regs.tex_cntl |= R128_TEX_CLAMP_T_WRAP;  break;
+    default: /* ERROR!! */ return;
+    }
+}
+
+/* Set the texture filter mode */
+static void r128SetTexFilter(r128TexObjPtr t, GLenum minf, GLenum magf)
+{
+    t->regs.tex_cntl &= ~(R128_MIN_BLEND_MASK | R128_MAG_BLEND_MASK);
+
+    switch (minf) {
+    case GL_NEAREST:
+	t->regs.tex_cntl |= R128_MIN_BLEND_NEAREST;
+	break;
+    case GL_LINEAR:
+	t->regs.tex_cntl |= R128_MIN_BLEND_LINEAR;
+	break;
+    case GL_NEAREST_MIPMAP_NEAREST:
+	t->regs.tex_cntl |= R128_MIN_BLEND_MIPNEAREST;
+	break;
+    case GL_LINEAR_MIPMAP_NEAREST:
+	t->regs.tex_cntl |= R128_MIN_BLEND_LINEARMIPNEAREST;
+	break;
+    case GL_NEAREST_MIPMAP_LINEAR:
+	t->regs.tex_cntl |= R128_MIN_BLEND_MIPLINEAR;
+	break;
+    case GL_LINEAR_MIPMAP_LINEAR:
+	t->regs.tex_cntl |= R128_MIN_BLEND_LINEARMIPLINEAR;
+	break;
+    default: /* ERROR!! */ return;
+    }
+
+    switch (magf) {
+    case GL_NEAREST:
+	t->regs.tex_cntl |= R128_MAG_BLEND_NEAREST;
+	break;
+    case GL_LINEAR:
+	t->regs.tex_cntl |= R128_MAG_BLEND_LINEAR;
+	break;
+    }
+}
+
+/* Set the texture border color */
+static void r128SetTexBorderColor(r128TexObjPtr t, GLubyte c[4])
+{
+    t->regs.border_color = r128PackColor(32, c[0], c[1], c[2], c[3]);
+}
+
+/* Set the texture environment state */
+static void r128DDTexEnv(GLcontext *ctx, GLenum target, GLenum pname,
+			 const GLfloat *param)
+{
+    r128ContextPtr          r128ctx = R128_CONTEXT(ctx);
+    struct gl_texture_unit *texUnit;
+    GLubyte                 c[4];
+    CARD32                  col;
+    
+    /* FIXME: Add texture LOD bias extension */
+
+    switch (pname) {
+    case GL_TEXTURE_ENV_MODE:
+	/* TexEnv modes are handled in UpdateTextureState */
+	R128CCE_FLUSH_VB_LOCK(r128ctx);
+	r128ctx->dirty |= R128_UPDATE_TEXSTATE;
+	break;
+    case GL_TEXTURE_ENV_COLOR:
+	texUnit = &ctx->Texture.Unit[ctx->Texture.CurrentUnit];
+	FLOAT_RGBA_TO_UBYTE_RGBA(texUnit->EnvColor, c);
+	col = r128PackColor(32, c[0], c[1], c[2], c[3]);
+	if (r128ctx->regs.constant_color_c != col) {
+	    R128CCE_FLUSH_VB_LOCK(r128ctx);
+	    r128ctx->regs.constant_color_c = col;
+
+	    /* FIXME: Load into hardware now??? */
+	    r128ctx->dirty         |= R128_UPDATE_CONTEXT;
+	    r128ctx->dirty_context |= R128_CTX_TEXENVSTATE;
+	}
+	break;
+    default:
+	return;
+    }
+}
+
+/* Upload a new texture image */
+static void r128DDTexImage(GLcontext *ctx, GLenum target,
+			   struct gl_texture_object *tObj, GLint level,
+			   GLint internalFormat,
+			   const struct gl_texture_image *image)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+    r128TexObjPtr  t;
+
+    if (R128_DEBUG_FLAGS & DEBUG_VERBOSE_API)
+	fprintf(stderr, "r128DDTexImage(%p, level %d)\n", tObj, level);
+
+    if ((target != GL_TEXTURE_2D) && (target != GL_TEXTURE_1D)) return;
+    if (level >= R128_TEX_MAXLEVELS) return;
+
+    t = (r128TexObjPtr)tObj->DriverData;
+    if (t) {
+	if (t->bound) R128CCE_FLUSH_VB_LOCK(r128ctx);
+
+	/* Destroy the old texture, and upload a new one.  The actual
+           uploading of the texture image occurs in the UploadSubImage
+           function. */
+	r128DestroyTexObj(r128ctx, t);
+	r128ctx->dirty |= R128_UPDATE_TEXSTATE;
+    }
+}
+
+/* Upload a new texture sub-image */
+static void r128DDTexSubImage(GLcontext *ctx, GLenum target,
+			      struct gl_texture_object *tObj, GLint level,
+			      GLint xoffset, GLint yoffset,
+			      GLsizei width, GLsizei height,
+			      GLint internalFormat,
+			      const struct gl_texture_image *image)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+    r128TexObjPtr  t;
+
+    if (R128_DEBUG_FLAGS & DEBUG_VERBOSE_API)
+	fprintf(stderr, "r128DDTexSubImage(%p, level %d) "
+		"size: %d,%d of %d,%d\n",
+		tObj, level, width, height, image->Width, image->Height);
+
+    if ((target != GL_TEXTURE_2D) && (target != GL_TEXTURE_1D)) return;
+    if (level >= R128_TEX_MAXLEVELS) return;
+
+    t = (r128TexObjPtr)tObj->DriverData;
+    if (t) {
+	if (t->bound) R128CCE_FLUSH_VB_LOCK(r128ctx);
+
+	LOCK_HARDWARE(r128ctx);
+	r128UploadSubImage(r128ctx, t, level,
+			   xoffset, yoffset, width, height);
+	UNLOCK_HARDWARE(r128ctx);
+
+	/* Update the context state */
+	r128ctx->regs.tex_cntl_c |= R128_TEX_CACHE_FLUSH;
+
+	r128ctx->dirty           |= (R128_UPDATE_CONTEXT |
+				     R128_UPDATE_TEXSTATE);
+	r128ctx->dirty_context   |= (R128_CTX_ENGINESTATE |
+				     R128_CTX_TEX0STATE   |
+				     R128_CTX_TEX1STATE);
+    }
+}
+
+/* Set the texture parameter state */
+static void r128DDTexParameter(GLcontext *ctx, GLenum target,
+			       struct gl_texture_object *tObj,
+			       GLenum pname, const GLfloat *params)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+    r128TexObjPtr  t       = (r128TexObjPtr)tObj->DriverData;
+
+    if (!t) return;
+    if ((target != GL_TEXTURE_2D) && (target != GL_TEXTURE_1D)) return;
+
+    switch (pname) {
+    case GL_TEXTURE_MIN_FILTER:
+    case GL_TEXTURE_MAG_FILTER:
+	if (t->bound) R128CCE_FLUSH_VB_LOCK(r128ctx);
+	r128SetTexFilter(t, tObj->MinFilter, tObj->MagFilter);
+	break;
+
+    case GL_TEXTURE_WRAP_S:
+    case GL_TEXTURE_WRAP_T:
+	if (t->bound) R128CCE_FLUSH_VB_LOCK(r128ctx);
+	r128SetTexWrap(t, tObj->WrapS, tObj->WrapT);
+	break;
+  
+    case GL_TEXTURE_BORDER_COLOR:
+	if (t->bound) R128CCE_FLUSH_VB_LOCK(r128ctx);
+	r128SetTexBorderColor(t, tObj->BorderColor);
+	break;
+
+    default:
+	return;
+    }
+
+    r128ctx->dirty |= R128_UPDATE_TEXSTATE;
+}
+
+/* Bind a texture to the currently active texture unit */
+static void r128DDBindTexture(GLcontext *ctx, GLenum target,
+			      struct gl_texture_object *tObj)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+
+    R128CCE_FLUSH_VB_LOCK(r128ctx);
+
+    /* Unbind the old texture */
+    if (r128ctx->CurrentTexObj[ctx->Texture.CurrentUnit]) {
+	r128ctx->CurrentTexObj[ctx->Texture.CurrentUnit]->bound = 0;
+	r128ctx->CurrentTexObj[ctx->Texture.CurrentUnit] = NULL;
+    }
+
+    /* The actualy binding occurs in the Tex[01]UpdateState function */
+    r128ctx->dirty |= R128_UPDATE_TEXSTATE;
+}
+
+/* Remove texture from AGP/local texture memory */
+static void r128DDDeleteTexture(GLcontext *ctx,
+				struct gl_texture_object *tObj)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+    r128TexObjPtr  t       = (r128TexObjPtr)tObj->DriverData;
+
+    if (t) {
+	if (t->bound) {
+	    R128CCE_FLUSH_VB_LOCK(r128ctx);
+
+	    r128ctx->CurrentTexObj[t->bound-1] = 0;
+	    r128ctx->dirty |= R128_UPDATE_TEXSTATE;
+	}
+
+	r128DestroyTexObj(r128ctx, t);
+	tObj->DriverData = NULL;
+    }
+}
+
+/* Determine if a texture is currently residing in either AGP/local
+   texture memory */
+static GLboolean r128DDIsTextureResident(GLcontext *ctx, 
+					 struct gl_texture_object *tObj)
+{
+    r128TexObjPtr t = (r128TexObjPtr)tObj->DriverData;
+
+    return t && t->memBlock;
+}
+
+/* Initialize the driver's texture functions */
+void r128DDInitTextureFuncs(GLcontext *ctx)
+{
+    ctx->Driver.TexEnv                  = r128DDTexEnv;
+    ctx->Driver.TexImage                = r128DDTexImage;
+    ctx->Driver.TexSubImage             = r128DDTexSubImage;
+    ctx->Driver.TexParameter            = r128DDTexParameter;
+    ctx->Driver.BindTexture             = r128DDBindTexture;
+    ctx->Driver.DeleteTexture           = r128DDDeleteTexture;
+    ctx->Driver.UpdateTexturePalette    = NULL;
+    ctx->Driver.ActiveTexture           = NULL;
+    ctx->Driver.IsTextureResident       = r128DDIsTextureResident;
+    ctx->Driver.PrioritizeTexture       = NULL;
+}
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_tex.h b/xc/lib/GL/mesa/src/drv/r128/r128_tex.h
new file mode 100644
index 000000000..20893110c
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_tex.h
@@ -0,0 +1,82 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *   Gareth Hughes <gareth@precisioninsight.com>
+ *
+ */
+
+#ifndef _R128_TEX_H_
+#define _R128_TEX_H_
+
+#ifdef GLX_DIRECT_RENDERING
+
+extern void r128AgeTextures(r128ContextPtr r128ctx, int heap);
+extern int  r128UploadTexImages(r128ContextPtr r128ctx, r128TexObjPtr t);
+extern void r128UpdateTextureState(r128ContextPtr r128ctx);
+extern void r128DestroyTexObj(r128ContextPtr r128ctx, r128TexObjPtr t);
+
+extern void r128DDInitTextureFuncs(GLcontext *ctx);
+
+#define R128PACKCOLOR332(r, g, b)                                             \
+    (((r) & 0xe0) | (((g) & 0xe0) >> 3) | (((b) & 0xc0) >> 6))
+
+#define R128PACKCOLOR1555(r, g, b, a)                                         \
+    ((((r) & 0xf8) << 7) | (((g) & 0xf8) << 2) | (((b) & 0xf8) >> 3) |        \
+     ((a) ? 0x8000 : 0))
+
+#define R128PACKCOLOR565(r, g, b)                                             \
+    ((((r) & 0xf8) << 8) | (((g) & 0xfc) << 3) | (((b) & 0xf8) >> 3))
+
+#define R128PACKCOLOR888(r, g, b)                                             \
+    (((r) << 16) | ((g) << 8) | (b))
+
+#define R128PACKCOLOR8888(r, g, b, a)                                         \
+    (((a) << 24) | ((r) << 16) | ((g) << 8) | (b))
+
+#define R128PACKCOLOR4444(r, g, b, a)                                         \
+    ((((a) & 0xf0) << 8) | (((r) & 0xf0) << 4) | ((g) & 0xf0) | ((b) >> 4))
+
+static __inline__ CARD32 r128PackColor(GLuint depth,
+				       GLubyte r, GLubyte g,
+				       GLubyte b, GLubyte a)
+{
+    switch (depth) {
+    case  8: return R128PACKCOLOR332(r, g, b);
+    case 15: return R128PACKCOLOR1555(r, g, b, a);
+    case 16: return R128PACKCOLOR565(r, g, b);
+    case 24: return R128PACKCOLOR888(r, g, b);
+    case 32: return R128PACKCOLOR8888(r, g, b, a);
+    default: return 0;
+    }
+}
+
+#endif
+#endif /* _R128_TEX_H_ */
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_texobj.h b/xc/lib/GL/mesa/src/drv/r128/r128_texobj.h
new file mode 100644
index 000000000..588b768ed
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_texobj.h
@@ -0,0 +1,87 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas.
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *   Gareth Hughes <gareth@precisioninsight.com>
+ *
+ */
+
+#ifndef _R128_TEXOBJ_H_
+#define _R128_TEXOBJ_H_
+
+#include "mm.h"
+
+#define R128_TEX_MAXLEVELS 11
+
+/* Setup registers for each texture */
+typedef struct {
+    CARD32  tex_cntl;
+    CARD32  size_pitch;
+    CARD32  border_color;
+} r128TextureRegs;
+
+/* Individual texture image information */
+typedef struct {
+    int offset;		/* Offset into locally shared texture space (i.e.,
+			   relative to bufAddr (below) */
+    int width;		/* Width of texture image */
+    int height;		/* Height of texture image */
+} r128TexImage;
+
+typedef struct r128_tex_obj r128TexObj, *r128TexObjPtr;
+
+/* Texture object in locally shared texture space */
+struct r128_tex_obj {
+    r128TexObjPtr    next, prev;
+
+    struct gl_texture_object *tObj;	/* Mesa texture object */
+
+    PMemBlock        memBlock;		/* Memory block containing texture */
+    unsigned char   *bufAddr;		/* Offset to start of locally
+					   shared texture block */
+
+    CARD32           dirty_images;	/* Flags for whether or not
+					   images need to be uploaded to
+					   local or AGP texture space */
+    int              bound;		/* Texture unit currently bound to */
+    int              heap;              /* Texture heap currently stored in */
+    r128TexImage     image[R128_TEX_MAXLEVELS]; /* Image data for all
+						   mipmap levels */
+    int              totalSize;		/* Total size of the texture
+					   including all mipmap levels */
+    int              internFormat;	/* Internal GL format used to store
+					   texture on card */
+    int              textureFormat;	/* Actual hardware format */
+    int              texelBytes;	/* Number of bytes per texel */
+
+    r128TextureRegs  regs;		/* Setup regs for texture */
+};
+
+#endif /* _R128_TEXOBJ_H_ */
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_tris.c b/xc/lib/GL/mesa/src/drv/r128/r128_tris.c
new file mode 100644
index 000000000..1c1b45daf
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_tris.c
@@ -0,0 +1,548 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas.
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#include "r128_init.h"
+#include "r128_mesa.h"
+#include "r128_xmesa.h"
+#include "r128_context.h"
+#include "r128_lock.h"
+#include "r128_reg.h"
+#include "r128_cce.h"
+#include "r128_vb.h"
+#include "r128_tris.h"
+#include "r128_state.h"
+
+static triangle_func tri_tab[0x40];    /* only 0x20 actually used */
+static quad_func     quad_tab[0x40];   /* only 0x20 actually used */
+static line_func     line_tab[0x40];   /* less than 0x20 used */
+static points_func   points_tab[0x40]; /* less than 0x20 used */
+
+/* Draw a triangle from the vertices in the vertex buffer */
+void r128DrawTriangleVB(r128ContextPtr r128ctx,
+			r128_vertex *v0,
+			r128_vertex *v1,
+			r128_vertex *v2)
+{
+    r128_vertex *vbptr;
+
+    if (r128ctx->disableVB) {
+	r128DrawTriangle(r128ctx, v0, v1, v2);
+	return;
+    }
+
+    R128CCE_ALLOC_VB_SPACE(r128ctx, vbptr, 3);
+
+    *vbptr++ = *v0;
+    *vbptr++ = *v1;
+    *vbptr++ = *v2;
+}
+
+/* Draw a triangle */
+void r128DrawTriangle(r128ContextPtr r128ctx,
+		      r128_vertex *v0,
+		      r128_vertex *v1,
+		      r128_vertex *v2)
+{
+    LOCK_HARDWARE(r128ctx);
+    BEGIN_CLIP_LOOP(r128ctx);
+
+#if USE_RHW2
+    R128CCE3(R128_CCE_PACKET3_3D_RNDR_GEN_PRIM, 34);
+#else
+    R128CCE3(R128_CCE_PACKET3_3D_RNDR_GEN_PRIM, 31);
+#endif
+    R128CCE(R128_FULL_VERTEX_FORMAT);
+    R128CCE(R128_CCE_VC_CNTL_PRIM_TYPE_TRI_LIST |
+	    R128_CCE_VC_CNTL_PRIM_WALK_RING     |
+	    (3 << R128_CCE_VC_CNTL_NUM_SHIFT));
+
+    R128CCE_SEND_VERTEX(v0);
+    R128CCE_SEND_VERTEX(v1);
+    R128CCE_SEND_VERTEX(v2);
+
+    R128CCE_SUBMIT_PACKETS();
+
+    END_CLIP_LOOP(r128ctx);
+    UNLOCK_HARDWARE(r128ctx);
+}
+
+/* Draw a line from the vertices in the vertex buffer */
+void r128DrawLineVB(r128ContextPtr r128ctx,
+		    r128_vertex *v0, r128_vertex *v1,
+		    float width)
+{
+    float        dx, dy, ix, iy;
+    r128_vertex *vbptr;
+
+    /* FIXME: Use r128's line primitive for width 1 lines */
+    if (r128ctx->disableVB) {
+	r128DrawLine(r128ctx, v0, v1, width);
+	return;
+    }
+
+    R128CCE_ALLOC_VB_SPACE(r128ctx, vbptr, 3);
+
+    dx = v0->x - v1->x;
+    dy = v0->y - v1->y;
+
+    ix = width * 0.5; iy = 0;
+    if (dx*dx > dy*dy) {
+	iy = ix; ix = 0;
+    }
+
+    *vbptr   = *v0; vbptr->x = v0->x - ix; vbptr->y = v0->y - iy;
+    *++vbptr = *v1; vbptr->x = v1->x + ix; vbptr->y = v1->y + iy;
+    *++vbptr = *v0; vbptr->x = v0->x + ix; vbptr->y = v0->y + iy;
+    *++vbptr = *v0; vbptr->x = v0->x - ix; vbptr->y = v0->y - iy;
+    *++vbptr = *v1; vbptr->x = v1->x - ix; vbptr->y = v1->y - iy;
+    *++vbptr = *v1; vbptr->x = v1->x + ix; vbptr->y = v1->y + iy;
+}
+
+/* Draw a line */
+void r128DrawLine(r128ContextPtr r128ctx,
+		  r128_vertex *v0, r128_vertex *v1,
+		  float width)
+{
+    LOCK_HARDWARE(r128ctx);
+    if (width != 1) {
+	float dx, dy, ix, iy;
+
+	dx = v0->x - v1->x;
+	dy = v0->y - v1->y;
+
+	ix = width * 0.5; iy = 0;
+	if (dx*dx > dy*dy) {
+	    iy = ix; ix = 0;
+	}
+
+	BEGIN_CLIP_LOOP(r128ctx);
+
+#if USE_RHW2
+	R128CCE3(R128_CCE_PACKET3_3D_RNDR_GEN_PRIM, 67);
+#else
+	R128CCE3(R128_CCE_PACKET3_3D_RNDR_GEN_PRIM, 61);
+#endif
+	R128CCE(R128_FULL_VERTEX_FORMAT);
+	R128CCE(R128_CCE_VC_CNTL_PRIM_TYPE_TRI_LIST |
+		R128_CCE_VC_CNTL_PRIM_WALK_RING     |
+		(6 << R128_CCE_VC_CNTL_NUM_SHIFT));
+
+	R128CCEF(v0->x - ix);
+	R128CCEF(v0->y - iy);
+	R128CCEF(v0->z);
+	R128CCEF(v0->rhw);
+	R128CCE(*(int *)&v0->dif_argb);
+	R128CCE(*(int *)&v0->spec_frgb);
+	R128CCEF(v0->tu0);
+	R128CCEF(v0->tv0);
+	R128CCEF(v0->tu1);
+	R128CCEF(v0->tv1);
+#if USE_RHW2
+	R128CCEF(v0->rhw2);
+#endif
+
+	R128CCEF(v1->x + ix);
+	R128CCEF(v1->y + iy);
+	R128CCEF(v1->z);
+	R128CCEF(v1->rhw);
+	R128CCE(*(int *)&v1->dif_argb);
+	R128CCE(*(int *)&v1->spec_frgb);
+	R128CCEF(v1->tu0);
+	R128CCEF(v1->tv0);
+	R128CCEF(v1->tu1);
+	R128CCEF(v1->tv1);
+#if USE_RHW2
+	R128CCEF(v1->rhw2);
+#endif
+
+	R128CCEF(v0->x + ix);
+	R128CCEF(v0->y + iy);
+	R128CCEF(v0->z);
+	R128CCEF(v0->rhw);
+	R128CCE(*(int *)&v0->dif_argb);
+	R128CCE(*(int *)&v0->spec_frgb);
+	R128CCEF(v0->tu0);
+	R128CCEF(v0->tv0);
+	R128CCEF(v0->tu1);
+	R128CCEF(v0->tv1);
+#if USE_RHW2
+	R128CCEF(v0->rhw2);
+#endif
+
+	R128CCEF(v0->x - ix);
+	R128CCEF(v0->y - iy);
+	R128CCEF(v0->z);
+	R128CCEF(v0->rhw);
+	R128CCE(*(int *)&v0->dif_argb);
+	R128CCE(*(int *)&v0->spec_frgb);
+	R128CCEF(v0->tu0);
+	R128CCEF(v0->tv0);
+	R128CCEF(v0->tu1);
+	R128CCEF(v0->tv1);
+#if USE_RHW2
+	R128CCEF(v0->rhw2);
+#endif
+
+	R128CCEF(v1->x - ix);
+	R128CCEF(v1->y - iy);
+	R128CCEF(v1->z);
+	R128CCEF(v1->rhw);
+	R128CCE(*(int *)&v1->dif_argb);
+	R128CCE(*(int *)&v1->spec_frgb);
+	R128CCEF(v1->tu0);
+	R128CCEF(v1->tv0);
+	R128CCEF(v1->tu1);
+	R128CCEF(v1->tv1);
+#if USE_RHW2
+	R128CCEF(v1->rhw2);
+#endif
+
+	R128CCEF(v1->x + ix);
+	R128CCEF(v1->y + iy);
+	R128CCEF(v1->z);
+	R128CCEF(v1->rhw);
+	R128CCE(*(int *)&v1->dif_argb);
+	R128CCE(*(int *)&v1->spec_frgb);
+	R128CCEF(v1->tu0);
+	R128CCEF(v1->tv0);
+	R128CCEF(v1->tu1);
+	R128CCEF(v1->tv1);
+#if USE_RHW2
+	R128CCEF(v1->rhw2);
+#endif
+
+	R128CCE_SUBMIT_PACKETS();
+
+	END_CLIP_LOOP(r128ctx);
+    } else {
+	BEGIN_CLIP_LOOP(r128ctx);
+
+#if USE_RHW2
+	R128CCE3(R128_CCE_PACKET3_3D_RNDR_GEN_PRIM, 23);
+#else
+	R128CCE3(R128_CCE_PACKET3_3D_RNDR_GEN_PRIM, 21);
+#endif
+	R128CCE(R128_FULL_VERTEX_FORMAT);
+	R128CCE(R128_CCE_VC_CNTL_PRIM_TYPE_LINE     |
+		R128_CCE_VC_CNTL_PRIM_WALK_RING     |
+		(2 << R128_CCE_VC_CNTL_NUM_SHIFT));
+
+	R128CCE_SEND_VERTEX(v0);
+	R128CCE_SEND_VERTEX(v1);
+
+	R128CCE_SUBMIT_PACKETS();
+
+	END_CLIP_LOOP(r128ctx);
+    }
+    UNLOCK_HARDWARE(r128ctx);
+}
+
+/* Draw a point from the vertices in the vertex buffer */
+void r128DrawPointVB(r128ContextPtr r128ctx,
+		     r128_vertex *v0, float size)
+{
+    r128_vertex *vbptr;
+
+    /* FIXME: Use r128's point primitive for diameter 1 points */
+    if (r128ctx->disableVB) {
+	r128DrawPoint(r128ctx, v0, size);
+	return;
+    }
+
+    R128CCE_ALLOC_VB_SPACE(r128ctx, vbptr, 3);
+
+    *vbptr   = *v0; vbptr->x = v0->x - size; vbptr->y = v0->y - size;
+    *++vbptr = *v0; vbptr->x = v0->x + size; vbptr->y = v0->y - size;
+    *++vbptr = *v0; vbptr->x = v0->x + size; vbptr->y = v0->y + size;
+    *++vbptr = *v0; vbptr->x = v0->x + size; vbptr->y = v0->y + size;
+    *++vbptr = *v0; vbptr->x = v0->x - size; vbptr->y = v0->y + size;
+    *++vbptr = *v0; vbptr->x = v0->x - size; vbptr->y = v0->y - size;
+}
+
+/* Draw a point */
+void r128DrawPoint(r128ContextPtr r128ctx,
+		   r128_vertex *v0, float size)
+{
+    LOCK_HARDWARE(r128ctx);
+    if (size != 0.5) {
+	BEGIN_CLIP_LOOP(r128ctx);
+
+#if USE_RHW2
+	R128CCE3(R128_CCE_PACKET3_3D_RNDR_GEN_PRIM, 67);
+#else
+	R128CCE3(R128_CCE_PACKET3_3D_RNDR_GEN_PRIM, 61);
+#endif
+	R128CCE(R128_FULL_VERTEX_FORMAT);
+	R128CCE(R128_CCE_VC_CNTL_PRIM_TYPE_TRI_LIST |
+		R128_CCE_VC_CNTL_PRIM_WALK_RING     |
+		(6 << R128_CCE_VC_CNTL_NUM_SHIFT));
+
+	R128CCEF(v0->x - size);
+	R128CCEF(v0->y - size);
+	R128CCEF(v0->z);
+	R128CCEF(v0->rhw);
+	R128CCE(*(int *)&v0->dif_argb);
+	R128CCE(*(int *)&v0->spec_frgb);
+	R128CCEF(v0->tu0);
+	R128CCEF(v0->tv0);
+	R128CCEF(v0->tu1);
+	R128CCEF(v0->tv1);
+#if USE_RHW2
+	R128CCEF(v0->rhw2);
+#endif
+
+	R128CCEF(v0->x + size);
+	R128CCEF(v0->y - size);
+	R128CCEF(v0->z);
+	R128CCEF(v0->rhw);
+	R128CCE(*(int *)&v0->dif_argb);
+	R128CCE(*(int *)&v0->spec_frgb);
+	R128CCEF(v0->tu0);
+	R128CCEF(v0->tv0);
+	R128CCEF(v0->tu1);
+	R128CCEF(v0->tv1);
+#if USE_RHW2
+	R128CCEF(v0->rhw2);
+#endif
+
+	R128CCEF(v0->x + size);
+	R128CCEF(v0->y + size);
+	R128CCEF(v0->z);
+	R128CCEF(v0->rhw);
+	R128CCE(*(int *)&v0->dif_argb);
+	R128CCE(*(int *)&v0->spec_frgb);
+	R128CCEF(v0->tu0);
+	R128CCEF(v0->tv0);
+	R128CCEF(v0->tu1);
+	R128CCEF(v0->tv1);
+#if USE_RHW2
+	R128CCEF(v0->rhw2);
+#endif
+
+	R128CCEF(v0->x + size);
+	R128CCEF(v0->y + size);
+	R128CCEF(v0->z);
+	R128CCEF(v0->rhw);
+	R128CCE(*(int *)&v0->dif_argb);
+	R128CCE(*(int *)&v0->spec_frgb);
+	R128CCEF(v0->tu0);
+	R128CCEF(v0->tv0);
+	R128CCEF(v0->tu1);
+	R128CCEF(v0->tv1);
+#if USE_RHW2
+	R128CCEF(v0->rhw2);
+#endif
+
+	R128CCEF(v0->x - size);
+	R128CCEF(v0->y + size);
+	R128CCEF(v0->z);
+	R128CCEF(v0->rhw);
+	R128CCE(*(int *)&v0->dif_argb);
+	R128CCE(*(int *)&v0->spec_frgb);
+	R128CCEF(v0->tu0);
+	R128CCEF(v0->tv0);
+	R128CCEF(v0->tu1);
+	R128CCEF(v0->tv1);
+#if USE_RHW2
+	R128CCEF(v0->rhw2);
+#endif
+
+	R128CCEF(v0->x - size);
+	R128CCEF(v0->y - size);
+	R128CCEF(v0->z);
+	R128CCEF(v0->rhw);
+	R128CCE(*(int *)&v0->dif_argb);
+	R128CCE(*(int *)&v0->spec_frgb);
+	R128CCEF(v0->tu0);
+	R128CCEF(v0->tv0);
+	R128CCEF(v0->tu1);
+	R128CCEF(v0->tv1);
+#if USE_RHW2
+	R128CCEF(v0->rhw2);
+#endif
+
+	R128CCE_SUBMIT_PACKETS();
+
+	END_CLIP_LOOP(r128ctx);
+    } else {
+	BEGIN_CLIP_LOOP(r128ctx);
+
+#if USE_RHW2
+	R128CCE3(R128_CCE_PACKET3_3D_RNDR_GEN_PRIM, 12);
+#else
+	R128CCE3(R128_CCE_PACKET3_3D_RNDR_GEN_PRIM, 11);
+#endif
+	R128CCE(R128_FULL_VERTEX_FORMAT);
+	R128CCE(R128_CCE_VC_CNTL_PRIM_TYPE_POINT    |
+		R128_CCE_VC_CNTL_PRIM_WALK_RING     |
+		(1 << R128_CCE_VC_CNTL_NUM_SHIFT));
+
+	R128CCE_SEND_VERTEX(v0);
+
+	R128CCE_SUBMIT_PACKETS();
+
+	END_CLIP_LOOP(r128ctx);
+    }
+    UNLOCK_HARDWARE(r128ctx);
+}
+
+#define R128_COLOR(to, from)                                                \
+do {                                                                        \
+    (to)[0] = (from)[2];                                                    \
+    (to)[1] = (from)[1];                                                    \
+    (to)[2] = (from)[0];                                                    \
+    (to)[3] = (from)[3];                                                    \
+} while (0)
+
+#define IND (0)
+#define TAG(x) x
+#include "r128_tritmp.h"
+
+#define IND (R128_FLAT_BIT)
+#define TAG(x) x##_flat
+#include "r128_tritmp.h"
+
+#define IND (R128_OFFSET_BIT)
+#define TAG(x) x##_offset
+#include "r128_tritmp.h"
+
+#define IND (R128_OFFSET_BIT | R128_FLAT_BIT)
+#define TAG(x) x##_offset_flat
+#include "r128_tritmp.h"
+
+#define IND (R128_TWOSIDE_BIT)
+#define TAG(x) x##_twoside
+#include "r128_tritmp.h"
+
+#define IND (R128_TWOSIDE_BIT | R128_FLAT_BIT)
+#define TAG(x) x##_twoside_flat
+#include "r128_tritmp.h"
+
+#define IND (R128_TWOSIDE_BIT | R128_OFFSET_BIT)
+#define TAG(x) x##_twoside_offset
+#include "r128_tritmp.h"
+
+#define IND (R128_TWOSIDE_BIT | R128_OFFSET_BIT | R128_FLAT_BIT)
+#define TAG(x) x##_twoside_offset_flat
+#include "r128_tritmp.h"
+
+/* Initialize the table of points, line and triangle drawing functions */
+void r128TriangleFuncsInit(void)
+{
+    init();
+    init_flat();
+    init_offset();
+    init_offset_flat();
+    init_twoside();
+    init_twoside_flat();
+    init_twoside_offset();
+    init_twoside_offset_flat();
+}
+
+/* Setup the Point, Line, Triangle and Quad functions based on the
+   current rendering state.  Wherever possible, use the hardware to
+   render the primitive.  Otherwise, fallback to software rendering. */
+void r128ChooseRenderState(GLcontext *ctx)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+    GLuint         flags   = ctx->TriangleCaps;
+
+    if (r128ctx->Fallback) return;
+
+    if (r128ctx->SWonly) {
+	r128ctx->IndirectTriangles = DD_SW_RASTERIZE;
+	r128ctx->PointsFunc   = NULL;
+	r128ctx->LineFunc     = NULL;
+	r128ctx->TriangleFunc = NULL;
+	r128ctx->QuadFunc     = NULL;
+	return;
+    } else {
+	r128ctx->IndirectTriangles = 0;
+    }
+
+    if (flags) {
+	CARD32 index    = 0;
+	CARD32 shared   = 0;
+	CARD32 fallback = R128_FALLBACK_BIT;
+
+	if (r128ctx->SWfallbackDisable) fallback = 0; /* No fallbacks */
+	if (r128ctx->SWonly) shared = fallback; /* Everything's a fallback */
+
+	if (flags & DD_FLATSHADE)               shared |= R128_FLAT_BIT;
+	if (flags & DD_MULTIDRAW)               shared |= fallback;
+	if (flags & DD_SELECT)                  shared |= R128_FALLBACK_BIT;
+	if (flags & DD_FEEDBACK)                shared |= R128_FALLBACK_BIT;
+
+	/* Setup PointFunc */
+	index = shared;
+	if (flags & DD_POINT_SMOOTH)            index  |= fallback;
+	if (flags & DD_POINT_ATTEN)             index  |= fallback;
+
+	r128ctx->RenderIndex = index;
+	r128ctx->PointsFunc  = points_tab[index];
+	if (index & R128_FALLBACK_BIT)
+	    r128ctx->IndirectTriangles |= DD_POINT_SW_RASTERIZE;
+
+	/* Setup LineFunc */
+	index = shared;
+	if (flags & DD_LINE_SMOOTH)             index  |= fallback;
+	if (flags & DD_LINE_STIPPLE)            index  |= fallback;
+
+	r128ctx->RenderIndex |= index;
+	r128ctx->LineFunc     = line_tab[index];
+	if (index & R128_FALLBACK_BIT)
+	    r128ctx->IndirectTriangles |= DD_LINE_SW_RASTERIZE;
+
+	/* Setup TriangleFunc and QuadFunc */
+	index = shared;
+	if (flags & DD_TRI_SMOOTH)              index  |= fallback;
+	if (flags & DD_TRI_OFFSET)              index  |= R128_OFFSET_BIT;
+	if (flags & DD_TRI_LIGHT_TWOSIDE)       index  |= R128_TWOSIDE_BIT;
+	if (flags & DD_TRI_UNFILLED)            index  |= fallback;
+	if (flags & DD_TRI_STIPPLE)             index  |= fallback;
+
+	r128ctx->RenderIndex  |= index;
+	r128ctx->TriangleFunc  = tri_tab[index];
+	r128ctx->QuadFunc      = quad_tab[index];
+	if (index & R128_FALLBACK_BIT)
+	    r128ctx->IndirectTriangles |= (DD_TRI_SW_RASTERIZE |
+					   DD_QUAD_SW_RASTERIZE);
+    } else if (r128ctx->RenderIndex) {
+	r128ctx->RenderIndex  = 0;
+	r128ctx->PointsFunc   = points_tab[0];
+	r128ctx->LineFunc     = line_tab[0];
+	r128ctx->TriangleFunc = tri_tab[0];
+	r128ctx->QuadFunc     = quad_tab[0];
+    }
+}
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_tris.h b/xc/lib/GL/mesa/src/drv/r128/r128_tris.h
new file mode 100644
index 000000000..0980f2791
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_tris.h
@@ -0,0 +1,76 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#ifndef _R128_TRIS_H_
+#define _R128_TRIS_H_
+
+#ifdef GLX_DIRECT_RENDERING
+
+#include "r128_vb.h"
+
+#define R128_ANTIALIAS_BIT   0x00   /* ignored for now, no fallback */
+#define R128_FLAT_BIT        0x01
+#define R128_OFFSET_BIT      0x02
+#define R128_TWOSIDE_BIT     0x04
+#define R128_NODRAW_BIT      0x08
+#define R128_FALLBACK_BIT    0x10
+#define R128_FEEDBACK_BIT    0x20
+#define R128_SELECT_BIT      0x40
+#define R128_POINT_PARAM_BIT 0x80   /* not needed? */
+
+extern void r128DrawTriangleVB(r128ContextPtr r128ctx,
+			       r128_vertex *v0,
+			       r128_vertex *v1,
+			       r128_vertex *v2);
+extern void r128DrawLineVB(r128ContextPtr r128ctx,
+			   r128_vertex *tmp0, r128_vertex *tmp1,
+			   float width);
+extern void r128DrawPointVB(r128ContextPtr r128ctx,
+			    r128_vertex *tmp, float size);
+
+extern void r128DrawTriangle(r128ContextPtr r128ctx,
+			     r128_vertex *v0,
+			     r128_vertex *v1,
+			     r128_vertex *v2);
+extern void r128DrawLine(r128ContextPtr r128ctx,
+			 r128_vertex *tmp0, r128_vertex *tmp1,
+			 float width);
+extern void r128DrawPoint(r128ContextPtr r128ctx,
+			  r128_vertex *tmp, float size);
+
+extern void r128ChooseRenderState(GLcontext *ctx);
+extern void r128TriangleFuncsInit(void);
+
+#endif
+#endif /* _R128_TRIS_H_ */
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_tritmp.h b/xc/lib/GL/mesa/src/drv/r128/r128_tritmp.h
new file mode 100644
index 000000000..833bd832a
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_tritmp.h
@@ -0,0 +1,327 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#if !defined(TAG) || !defined(IND)
+    this is an error
+#endif
+
+/* Draw a single triangle.  Note that the device-dependent vertex data
+   might need to be changed based on the render state. */
+static void TAG(triangle)(GLcontext *ctx,
+			  GLuint e0, GLuint e1, GLuint e2,
+			  GLuint pv)
+{
+    r128ContextPtr        r128ctx   = R128_CONTEXT(ctx);
+    struct vertex_buffer *VB        = ctx->VB;
+    r128VertexPtr         r128verts = R128_DRIVER_DATA(VB)->verts;
+    const r128_vertex    *v0        = &r128verts[e0].v;  
+    const r128_vertex    *v1        = &r128verts[e1].v;  
+    const r128_vertex    *v2        = &r128verts[e2].v;  
+
+#if (IND & R128_OFFSET_BIT)
+    GLfloat offset;
+#endif
+
+#if (IND & (R128_FLAT_BIT | R128_TWOSIDE_BIT))
+    int c0 = *(int *)&r128verts[pv].v.dif_argb; 
+    int c1 = c0;
+    int c2 = c0;
+#endif
+
+#if (IND & R128_OFFSET_BIT)
+    switch (ctx->Visual->DepthBits) {
+    case 16: offset = ctx->Polygon.OffsetUnits / 65536.0;      break;
+    case 24: offset = ctx->Polygon.OffsetUnits / 16777216.0;   break;
+    case 32: offset = ctx->Polygon.OffsetUnits / 4294967296.0; break;
+    default: offset = ctx->Polygon.OffsetUnits / 65536.0;      break;
+    }
+#endif
+
+#if (IND & (R128_TWOSIDE_BIT | R128_OFFSET_BIT)) 
+    {
+	GLfloat ex = v0->x - v2->x;
+	GLfloat ey = v0->y - v2->y;
+	GLfloat fx = v1->x - v2->x;
+	GLfloat fy = v1->y - v2->y;
+	GLfloat c  = ex*fy - ey*fx;
+         
+#if (IND & R128_TWOSIDE_BIT) 
+	{
+	    GLuint  facing        = (c > 0.0) ^ ctx->Polygon.FrontBit;
+	    GLubyte (*vbcolor)[4] = VB->Color[facing]->data;
+	    if (IND & R128_FLAT_BIT) {
+		R128_COLOR((char *)&c0, vbcolor[pv]);
+		c2 = c1 = c0;
+	    } else {
+		R128_COLOR((char *)&c0, vbcolor[e0]);
+		R128_COLOR((char *)&c1, vbcolor[e1]);
+		R128_COLOR((char *)&c2, vbcolor[e2]);
+	    }
+	}
+#endif
+      
+#if (IND & R128_OFFSET_BIT) 
+	{
+	    if (c * c > 1e-16) {
+		GLfloat factor = ctx->Polygon.OffsetFactor;
+		GLfloat ez     = v0->z - v2->z;
+		GLfloat fz     = v1->z - v2->z;
+		GLfloat a      = ey*fz - ez*fy;
+		GLfloat b      = ez*fx - ex*fz;
+		GLfloat ic     = 1.0 / c;
+		GLfloat ac     = a * ic;
+		GLfloat bc     = b * ic;
+		if (ac < 0.0f) ac = -ac;
+		if (bc < 0.0f) bc = -bc;
+		offset += MAX2(ac, bc) * factor;
+	    }
+	}
+#endif
+    }
+#endif  
+
+    LOCK_HARDWARE(r128ctx);
+    if (r128ctx->regs.tex_cntl_c & (R128_TEXMAP_ENABLE |
+				    R128_SEC_TEXMAP_ENABLE)) {
+	BEGIN_CLIP_LOOP(r128ctx);
+
+#if USE_RHW2
+	R128CCE3(R128_CCE_PACKET3_3D_RNDR_GEN_PRIM, 34);
+#else
+	R128CCE3(R128_CCE_PACKET3_3D_RNDR_GEN_PRIM, 31);
+#endif
+	R128CCE(R128_FULL_VERTEX_FORMAT);
+	R128CCE(R128_CCE_VC_CNTL_PRIM_TYPE_TRI_LIST |
+		R128_CCE_VC_CNTL_PRIM_WALK_RING     |
+		(3 << R128_CCE_VC_CNTL_NUM_SHIFT));
+
+	R128CCEF(v0->x);
+	R128CCEF(v0->y);
+#if (IND & R128_OFFSET_BIT)
+	R128CCEF(v0->z + offset);
+#else
+	R128CCEF(v0->z);
+#endif
+	R128CCEF(v0->rhw);
+
+#if (IND & (R128_FLAT_BIT | R128_TWOSIDE_BIT))
+	R128CCE(c0);
+#else
+	R128CCE(*(int *)&v0->dif_argb);
+#endif
+	R128CCE(*(int *)&v0->spec_frgb);
+
+	R128CCEF(v0->tu0);
+	R128CCEF(v0->tv0);
+	R128CCEF(v0->tu1);
+	R128CCEF(v0->tv1);
+#if USE_RHW2
+	R128CCEF(v0->rhw2);
+#endif
+
+	R128CCEF(v1->x);
+	R128CCEF(v1->y);
+#if (IND & R128_OFFSET_BIT)
+	R128CCEF(v1->z + offset);
+#else
+	R128CCEF(v1->z);
+#endif
+	R128CCEF(v1->rhw);
+
+#if (IND & (R128_FLAT_BIT | R128_TWOSIDE_BIT))
+	R128CCE(c1);
+#else
+	R128CCE(*(int *)&v1->dif_argb);
+#endif
+	R128CCE(*(int *)&v1->spec_frgb);
+
+	R128CCEF(v1->tu0);
+	R128CCEF(v1->tv0);
+	R128CCEF(v1->tu1);
+	R128CCEF(v1->tv1);
+#if USE_RHW2
+	R128CCEF(v1->rhw2);
+#endif
+
+	R128CCEF(v2->x);
+	R128CCEF(v2->y);
+#if (IND & R128_OFFSET_BIT)
+	R128CCEF(v2->z + offset);
+#else
+	R128CCEF(v2->z);
+#endif
+	R128CCEF(v2->rhw);
+
+#if (IND & (R128_FLAT_BIT | R128_TWOSIDE_BIT))
+	R128CCE(c2);
+#else
+	R128CCE(*(int *)&v2->dif_argb);
+#endif
+	R128CCE(*(int *)&v2->spec_frgb);
+
+	R128CCEF(v2->tu0);
+	R128CCEF(v2->tv0);
+	R128CCEF(v2->tu1);
+	R128CCEF(v2->tv1);
+#if USE_RHW2
+	R128CCEF(v2->rhw2);
+#endif
+
+	R128CCE_SUBMIT_PACKETS();
+
+	END_CLIP_LOOP(r128ctx);
+    } else {
+	BEGIN_CLIP_LOOP(r128ctx);
+
+	R128CCE3(R128_CCE_PACKET3_3D_RNDR_GEN_PRIM, 13);
+	R128CCE(R128_CCE_VC_FRMT_DIFFUSE_ARGB);
+	R128CCE(R128_CCE_VC_CNTL_PRIM_TYPE_TRI_LIST |
+		R128_CCE_VC_CNTL_PRIM_WALK_RING     |
+		(3 << R128_CCE_VC_CNTL_NUM_SHIFT));
+
+	R128CCEF(v0->x);
+	R128CCEF(v0->y);
+#if (IND & R128_OFFSET_BIT)
+	R128CCEF(v0->z + offset);
+#else
+	R128CCEF(v0->z);
+#endif
+
+#if (IND & (R128_FLAT_BIT | R128_TWOSIDE_BIT))
+	R128CCE(c0);
+#else
+	R128CCE(*(int *)&v0->dif_argb);
+#endif
+
+	R128CCEF(v1->x);
+	R128CCEF(v1->y);
+#if (IND & R128_OFFSET_BIT)
+	R128CCEF(v1->z + offset);
+#else
+	R128CCEF(v1->z);
+#endif
+
+#if (IND & (R128_FLAT_BIT | R128_TWOSIDE_BIT))
+	R128CCE(c1);
+#else
+	R128CCE(*(int *)&v1->dif_argb);
+#endif
+
+	R128CCEF(v2->x);
+	R128CCEF(v2->y);
+#if (IND & R128_OFFSET_BIT)
+	R128CCEF(v2->z + offset);
+#else
+	R128CCEF(v2->z);
+#endif
+
+#if (IND & (R128_FLAT_BIT | R128_TWOSIDE_BIT))
+	R128CCE(c2);
+#else
+	R128CCE(*(int *)&v2->dif_argb);
+#endif
+
+	R128CCE_SUBMIT_PACKETS();
+
+	END_CLIP_LOOP(r128ctx);
+    }
+    UNLOCK_HARDWARE(r128ctx);
+}
+
+static void TAG(quad)(GLcontext *ctx,
+		      GLuint v0, GLuint v1, GLuint v2, GLuint v3,
+		      GLuint pv)
+{
+    TAG(triangle)(ctx, v0, v1, v3, pv);
+    TAG(triangle)(ctx, v1, v2, v3, pv);
+}
+
+#if ((IND & ~R128_FLAT_BIT) == 0)
+
+/* Draw a single line.  Note that the device-dependent vertex data might
+   need to be changed based on the render state. */
+static void TAG(line)(GLcontext *ctx,
+		      GLuint v0, GLuint v1,
+		      GLuint pv)
+{
+    r128ContextPtr  r128ctx   = R128_CONTEXT(ctx);
+    r128VertexPtr   r128verts = R128_DRIVER_DATA(ctx->VB)->verts;
+    r128_vertex     tmp0      = r128verts[v0].v;
+    r128_vertex     tmp1      = r128verts[v1].v;
+    float           width     = ctx->Line.Width;
+
+    if (IND & R128_FLAT_BIT) {
+	*(int *)&tmp1.dif_argb =
+	    *(int *)&tmp0.dif_argb = 
+	    *(int *)&r128verts[pv].v.dif_argb;
+    }
+
+    r128DrawLine(r128ctx, &tmp0, &tmp1, width);
+}
+
+/* Draw a set of points.  Note that the device-dependent vertex data
+   might need to be changed based on the render state. */
+static void TAG(points)(GLcontext *ctx,
+			GLuint first, GLuint last)
+{
+    r128ContextPtr        r128ctx   = R128_CONTEXT(ctx);
+    struct vertex_buffer *VB        = ctx->VB;
+    r128VertexPtr         r128verts = R128_DRIVER_DATA(VB)->verts;
+    GLfloat               size      = ctx->Point.Size * 0.5;
+    int                   i;
+
+    for(i = first; i <= last; i++) {
+	if(VB->ClipMask[i] == 0) {
+            r128_vertex *tmp = &r128verts[i].v;
+            r128DrawPoint(r128ctx, tmp, size);
+	}
+    }
+}
+
+#endif
+
+/* Initialize the table of primitives to render. */
+static void TAG(init)(void)
+{
+    tri_tab[IND]    = TAG(triangle);
+    quad_tab[IND]   = TAG(quad);
+
+#if ((IND & ~R128_FLAT_BIT) == 0)
+    line_tab[IND]   = TAG(line);
+    points_tab[IND] = TAG(points);
+#endif
+}
+
+#undef IND
+#undef TAG
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_vb.c b/xc/lib/GL/mesa/src/drv/r128/r128_vb.c
new file mode 100644
index 000000000..f14ec21bd
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_vb.c
@@ -0,0 +1,470 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#include "r128_init.h"
+#include "r128_mesa.h"
+#include "r128_xmesa.h"
+#include "r128_context.h"
+#include "r128_lock.h"
+#include "r128_reg.h"
+#include "r128_cce.h"
+#include "r128_state.h"
+#include "r128_vb.h"
+
+#include "stages.h"
+
+#define TEX0                                                            \
+do {                                                                    \
+    v->v.tu0 = tc0[i][0];                                               \
+    v->v.tv0 = tc0[i][1];                                               \
+} while (0)
+
+#define TEX1                                                            \
+do {                                                                    \
+    v->v.tu1 = tc1[i][0];                                               \
+    v->v.tv1 = tc1[i][1];                                               \
+} while (0)
+
+#define SPC                                                             \
+do {                                                                    \
+    GLubyte *spec = &(VB->Spec[0][i][0]);                               \
+    v->v.spec_frgb.r = spec[0];                                         \
+    v->v.spec_frgb.g = spec[1];                                         \
+    v->v.spec_frgb.b = spec[2];                                         \
+} while (0)
+
+#define FOG                                                             \
+do {                                                                    \
+    GLubyte *spec = &(VB->Spec[0][i][0]);                               \
+    v->v.spec_frgb.a = spec[3];                                         \
+} while (0)
+
+#define COL                                                             \
+do {                                                                    \
+    GLubyte *col = &(VB->Color[0]->data[i][0]);                         \
+    v->v.dif_argb.a = col[3];                                           \
+    v->v.dif_argb.r = col[0];                                           \
+    v->v.dif_argb.g = col[1];                                           \
+    v->v.dif_argb.b = col[2];                                           \
+} while (0)
+
+#if 1
+/* FIXME: These are handled by the Rage 128 */
+#define TEX0_4
+#define TEX1_4
+#else
+#define TEX0_4                                                          \
+do {                                                                    \
+    if (VB->TexCoordPtr[0]->size == 4) {                                \
+	GLfloat (*tc)[4] = VB->TexCoordPtr[0]->data;                    \
+	v = &(R128_DRIVER_DATA(VB)->verts[start]);                      \
+	for (i = start; i < end; i++, v++)   {                          \
+	    float oow = 1.0 / tc[i][3];                                 \
+	    v->v.rhw *= tc[i][3];                                       \
+	    v->v.tu0 *= oow;                                            \
+	    v->v.tv0 *= oow;                                            \
+	}                                                               \
+    }                                                                   \
+} while (0)
+
+#if USE_RHW2
+#define TEX1_4                                                          \
+do {                                                                    \
+    if (VB->TexCoordPtr[1]->size == 4) {                                \
+	GLfloat (*tc)[4] = VB->TexCoordPtr[1]->data;                    \
+	v = &(R128_DRIVER_DATA(VB)->verts[start]);                      \
+	for (i = start; i < end; i++, v++)   {                          \
+	    float oow = 1.0 / tc[i][3];                                 \
+	    v->v.rhw2 *= tc[i][3];                                      \
+	    v->v.tu1  *= oow;                                           \
+	    v->v.tv1  *= oow;                                           \
+	}                                                               \
+    }                                                                   \
+} while (0)
+#else
+#define TEX1_4
+#endif
+#endif
+
+#if USE_RHW2
+#define COORD                                                           \
+do {                                                                    \
+    GLfloat *win = VB->Win.data[i];                                     \
+    v->v.x =                win[0];                                     \
+    v->v.y =   r128height - win[1];                                     \
+    v->v.z =        scale * win[2];                                     \
+    v->v.rhw = v->v.rhw2 =  win[3];                                     \
+} while (0)
+#else
+#define COORD                                                           \
+do {                                                                    \
+    GLfloat *win = VB->Win.data[i];                                     \
+    v->v.x =                win[0];                                     \
+    v->v.y =   r128height - win[1];                                     \
+    v->v.z =        scale * win[2];                                     \
+    v->v.rhw =              win[3];                                     \
+} while (0)
+#endif
+
+#define NOP
+
+/* Setup the r128 vertex buffer entries */
+#define SETUPFUNC(name,win,col,tex0,tex1,tex0_4,tex1_4,spec,fog)        \
+static void name(struct vertex_buffer *VB, GLuint start, GLuint end)    \
+{                                                                       \
+    r128ContextPtr        r128ctx    = R128_CONTEXT(VB->ctx);           \
+    __DRIdrawablePrivate *dPriv      = r128ctx->driDrawable;            \
+    r128VertexPtr         v;                                            \
+    GLfloat               (*tc0)[4];                                    \
+    GLfloat               (*tc1)[4];                                    \
+    GLfloat               r128height = dPriv->h;                        \
+    GLfloat               scale;                                        \
+    int                   i;                                            \
+                                                                        \
+    (void) r128height; (void) r128ctx;                                  \
+                                                                        \
+    gl_import_client_data(VB, VB->ctx->RenderFlags,                     \
+			  (VB->ClipOrMask                               \
+			   ? VEC_WRITABLE | VEC_GOOD_STRIDE             \
+			   : VEC_GOOD_STRIDE));                         \
+                                                                        \
+    switch (VB->ctx->Visual->DepthBits) {                               \
+    case 16: scale = 1.0 / 65536.0;      break;                         \
+    case 24: scale = 1.0 / 16777216.0;   break;                         \
+    case 32: scale = 1.0 / 4294967296.0; break;                         \
+    default: scale = 1.0 / 65536.0;      break;                         \
+    }                                                                   \
+                                                                        \
+    tc0 = VB->TexCoordPtr[0]->data;                                     \
+    tc1 = VB->TexCoordPtr[1]->data;                                     \
+                                                                        \
+    v = &(R128_DRIVER_DATA(VB)->verts[start]);                          \
+                                                                        \
+    if (VB->ClipOrMask == 0)                                            \
+	for (i = start; i < end; i++, v++) {                            \
+	    win;                                                        \
+	    col;                                                        \
+	    spec;                                                       \
+	    fog;                                                        \
+	    tex0;                                                       \
+	    tex1;                                                       \
+	}                                                               \
+    else                                                                \
+	for (i = start; i < end; i++, v++) {                            \
+	    if (VB->ClipMask[i] == 0) {                                 \
+		win;                                                    \
+		spec;                                                   \
+		fog;                                                    \
+		tex0;                                                   \
+		tex1;                                                   \
+	    }                                                           \
+	    col;                                                        \
+	}                                                               \
+    tex0_4;                                                             \
+    tex1_4;                                                             \
+}
+
+
+SETUPFUNC(rs_wt0,       COORD, NOP, TEX0, NOP,  TEX0_4, NOP,    NOP, NOP)
+SETUPFUNC(rs_wt1,       COORD, NOP, NOP,  TEX1, NOP,    TEX1_4, NOP, NOP)
+SETUPFUNC(rs_wt0t1,     COORD, NOP, TEX0, TEX1, TEX0_4, TEX1_4, NOP, NOP)
+SETUPFUNC(rs_wft0,      COORD, NOP, TEX0, NOP,  TEX0_4, NOP,    NOP, FOG)
+SETUPFUNC(rs_wft1,      COORD, NOP, NOP,  TEX1, NOP,    TEX1_4, NOP, FOG)
+SETUPFUNC(rs_wft0t1,    COORD, NOP, TEX0, TEX1, TEX0_4, TEX1_4, NOP, FOG)
+SETUPFUNC(rs_wg,        COORD, COL, NOP,  NOP,  NOP,    NOP,    NOP, NOP)
+SETUPFUNC(rs_wgs,       COORD, COL, NOP,  NOP,  NOP,    NOP,    SPC, NOP)
+SETUPFUNC(rs_wgt0,      COORD, COL, TEX0, NOP,  TEX0_4, NOP,    NOP, NOP)
+SETUPFUNC(rs_wgt1,      COORD, COL, NOP,  TEX1, NOP,    TEX1_4, NOP, NOP)
+SETUPFUNC(rs_wgt0t1,    COORD, COL, TEX0, TEX1, TEX0_4, TEX1_4, NOP, NOP)
+SETUPFUNC(rs_wgst0,     COORD, COL, TEX0, NOP,  TEX0_4, NOP,    SPC, NOP)
+SETUPFUNC(rs_wgst1,     COORD, COL, NOP,  TEX1, NOP,    TEX1_4, SPC, NOP)
+SETUPFUNC(rs_wgst0t1,   COORD, COL, TEX0, TEX1, TEX0_4, TEX1_4, SPC, NOP)
+SETUPFUNC(rs_wgf,       COORD, COL, NOP,  NOP,  NOP,    NOP,    NOP, FOG)
+SETUPFUNC(rs_wgfs,      COORD, COL, NOP,  NOP,  NOP,    NOP,    SPC, FOG)
+SETUPFUNC(rs_wgft0,     COORD, COL, TEX0, NOP,  TEX0_4, NOP,    NOP, FOG)
+SETUPFUNC(rs_wgft1,     COORD, COL, NOP,  TEX1, NOP,    TEX1_4, NOP, FOG)
+SETUPFUNC(rs_wgft0t1,   COORD, COL, TEX0, TEX1, TEX0_4, TEX1_4, NOP, FOG)
+SETUPFUNC(rs_wgfst0,    COORD, COL, TEX0, NOP,  TEX0_4, NOP,    SPC, FOG)
+SETUPFUNC(rs_wgfst1  ,  COORD, COL, NOP,  TEX1, NOP,    TEX1_4, SPC, FOG)
+SETUPFUNC(rs_wgfst0t1,  COORD, COL, TEX0, TEX1, TEX0_4, TEX1_4, SPC, FOG)
+
+SETUPFUNC(rs_t0,        NOP,   NOP, TEX0, NOP,  TEX0_4, NOP,    NOP, NOP)
+SETUPFUNC(rs_t1,        NOP,   NOP, NOP,  TEX1, NOP,    TEX1_4, NOP, NOP)
+SETUPFUNC(rs_t0t1,      NOP,   NOP, TEX0, TEX1, TEX0_4, TEX1_4, NOP, NOP)
+SETUPFUNC(rs_f,         NOP,   NOP, NOP,  NOP,  NOP,    NOP,    NOP, FOG)
+SETUPFUNC(rs_ft0,       NOP,   NOP, TEX0, NOP,  TEX0_4, NOP,    NOP, FOG)
+SETUPFUNC(rs_ft1,       NOP,   NOP, NOP,  TEX1, NOP,    TEX1_4, NOP, FOG)
+SETUPFUNC(rs_ft0t1,     NOP,   NOP, TEX0, TEX1, TEX0_4, TEX1_4, NOP, FOG)
+SETUPFUNC(rs_g,         NOP,   COL, NOP,  NOP,  NOP,    NOP,    NOP, NOP)
+SETUPFUNC(rs_gs,        NOP,   COL, NOP,  NOP,  NOP,    NOP,    SPC, NOP)
+SETUPFUNC(rs_gt0,       NOP,   COL, TEX0, NOP,  TEX0_4, NOP,    NOP, NOP)
+SETUPFUNC(rs_gt1,       NOP,   COL, NOP,  TEX1, NOP,    TEX1_4, NOP, NOP)
+SETUPFUNC(rs_gt0t1,     NOP,   COL, TEX0, TEX1, TEX0_4, TEX1_4, NOP, NOP)
+SETUPFUNC(rs_gst0,      NOP,   COL, TEX0, NOP,  TEX0_4, NOP,    SPC, NOP)
+SETUPFUNC(rs_gst1,      NOP,   COL, NOP,  TEX1, NOP,    TEX1_4, SPC, NOP)
+SETUPFUNC(rs_gst0t1,    NOP,   COL, TEX0, TEX1, TEX0_4, TEX1_4, SPC, NOP)
+SETUPFUNC(rs_gf,        NOP,   COL, NOP,  NOP,  NOP,    NOP,    NOP, FOG)
+SETUPFUNC(rs_gfs,       NOP,   COL, NOP,  NOP,  NOP,    NOP,    SPC, FOG)
+SETUPFUNC(rs_gft0,      NOP,   COL, TEX0, NOP,  TEX0_4, NOP,    NOP, FOG)
+SETUPFUNC(rs_gft1,      NOP,   COL, NOP,  TEX1, NOP,    TEX1_4, NOP, FOG)
+SETUPFUNC(rs_gft0t1,    NOP,   COL, TEX0, TEX1, TEX0_4, TEX1_4, NOP, FOG)
+SETUPFUNC(rs_gfst0,     NOP,   COL, TEX0, NOP,  TEX0_4, NOP,    SPC, FOG)
+SETUPFUNC(rs_gfst1,     NOP,   COL, NOP,  TEX1, NOP,    TEX1_4, SPC, FOG)
+SETUPFUNC(rs_gfst0t1,   NOP,   COL, TEX0, TEX1, TEX0_4, TEX1_4, SPC, FOG)
+
+static void rs_invalid(struct vertex_buffer *VB, GLuint start, GLuint end)
+{
+    fprintf(stderr, "r128RasterSetup(): invalid setup function\n");
+}
+
+typedef void (*setupFunc)(struct vertex_buffer *, GLuint, GLuint);
+static setupFunc setup_func[0x80];
+
+/* Initialize the table of vertex buffer setup functions */
+void r128SetupInit(void)
+{
+    int i;
+
+    for (i = 0; i < 0x80; i++) setup_func[i] = rs_invalid;
+
+    /* Funcs to build vertices from scratch */
+    setup_func[R128_WIN_BIT|R128_TEX0_BIT] = rs_wt0;
+    setup_func[R128_WIN_BIT|R128_TEX1_BIT] = rs_wt1;
+    setup_func[R128_WIN_BIT|R128_TEX0_BIT|R128_TEX1_BIT] = rs_wt0t1;
+    setup_func[R128_WIN_BIT|R128_FOG_BIT|R128_TEX0_BIT] = rs_wft0;
+    setup_func[R128_WIN_BIT|R128_FOG_BIT|R128_TEX1_BIT] = rs_wft1;
+    setup_func[R128_WIN_BIT|R128_FOG_BIT|R128_TEX0_BIT|R128_TEX1_BIT] = rs_wft0t1;
+    setup_func[R128_WIN_BIT|R128_RGBA_BIT] = rs_wg;
+    setup_func[R128_WIN_BIT|R128_RGBA_BIT|R128_SPEC_BIT] = rs_wgs;
+    setup_func[R128_WIN_BIT|R128_RGBA_BIT|R128_TEX0_BIT] = rs_wgt0;
+    setup_func[R128_WIN_BIT|R128_RGBA_BIT|R128_TEX1_BIT] = rs_wgt1;
+    setup_func[R128_WIN_BIT|R128_RGBA_BIT|R128_TEX0_BIT|R128_TEX1_BIT] = rs_wgt0t1;
+    setup_func[R128_WIN_BIT|R128_RGBA_BIT|R128_SPEC_BIT|R128_TEX0_BIT] = rs_wgst0;
+    setup_func[R128_WIN_BIT|R128_RGBA_BIT|R128_SPEC_BIT|R128_TEX1_BIT] = rs_wgst1;
+    setup_func[R128_WIN_BIT|R128_RGBA_BIT|R128_SPEC_BIT|R128_TEX0_BIT|R128_TEX1_BIT] = rs_wgst0t1;
+    setup_func[R128_WIN_BIT|R128_RGBA_BIT|R128_FOG_BIT] = rs_wgf;
+    setup_func[R128_WIN_BIT|R128_RGBA_BIT|R128_FOG_BIT|R128_SPEC_BIT] = rs_wgfs;
+    setup_func[R128_WIN_BIT|R128_RGBA_BIT|R128_FOG_BIT|R128_TEX0_BIT] = rs_wgft0;
+    setup_func[R128_WIN_BIT|R128_RGBA_BIT|R128_FOG_BIT|R128_TEX1_BIT] = rs_wgft1;
+    setup_func[R128_WIN_BIT|R128_RGBA_BIT|R128_FOG_BIT|R128_TEX0_BIT|R128_TEX1_BIT] = rs_wgft0t1;
+    setup_func[R128_WIN_BIT|R128_RGBA_BIT|R128_FOG_BIT|R128_SPEC_BIT|R128_TEX0_BIT] = rs_wgfst0;
+    setup_func[R128_WIN_BIT|R128_RGBA_BIT|R128_FOG_BIT|R128_SPEC_BIT|R128_TEX1_BIT] = rs_wgfst1;
+    setup_func[R128_WIN_BIT|R128_RGBA_BIT|R128_FOG_BIT|R128_SPEC_BIT|R128_TEX0_BIT|R128_TEX1_BIT] = rs_wgfst0t1;
+
+    /* Funcs to repair vertices */
+    setup_func[R128_TEX0_BIT] = rs_t0;
+    setup_func[R128_TEX1_BIT] = rs_t1;
+    setup_func[R128_TEX0_BIT|R128_TEX1_BIT] = rs_t0t1;
+    setup_func[R128_FOG_BIT] = rs_f;
+    setup_func[R128_FOG_BIT|R128_TEX0_BIT] = rs_ft0;
+    setup_func[R128_FOG_BIT|R128_TEX1_BIT] = rs_ft1;
+    setup_func[R128_FOG_BIT|R128_TEX0_BIT|R128_TEX1_BIT] = rs_ft0t1;
+    setup_func[R128_RGBA_BIT] = rs_g;
+    setup_func[R128_RGBA_BIT|R128_SPEC_BIT] = rs_gs;
+    setup_func[R128_RGBA_BIT|R128_TEX0_BIT] = rs_gt0;
+    setup_func[R128_RGBA_BIT|R128_TEX1_BIT] = rs_gt1;
+    setup_func[R128_RGBA_BIT|R128_TEX0_BIT|R128_TEX1_BIT] = rs_gt0t1;
+    setup_func[R128_RGBA_BIT|R128_SPEC_BIT|R128_TEX0_BIT] = rs_gst0;
+    setup_func[R128_RGBA_BIT|R128_SPEC_BIT|R128_TEX1_BIT] = rs_gst1;
+    setup_func[R128_RGBA_BIT|R128_SPEC_BIT|R128_TEX0_BIT|R128_TEX1_BIT] = rs_gst0t1;
+    setup_func[R128_RGBA_BIT|R128_FOG_BIT] = rs_gf;
+    setup_func[R128_RGBA_BIT|R128_FOG_BIT|R128_SPEC_BIT] = rs_gfs;
+    setup_func[R128_RGBA_BIT|R128_FOG_BIT|R128_TEX0_BIT] = rs_gft0;
+    setup_func[R128_RGBA_BIT|R128_FOG_BIT|R128_TEX1_BIT] = rs_gft1;
+    setup_func[R128_RGBA_BIT|R128_FOG_BIT|R128_TEX0_BIT|R128_TEX1_BIT] = rs_gft0t1;
+    setup_func[R128_RGBA_BIT|R128_FOG_BIT|R128_SPEC_BIT|R128_TEX0_BIT] = rs_gfst0;
+    setup_func[R128_RGBA_BIT|R128_FOG_BIT|R128_SPEC_BIT|R128_TEX1_BIT] = rs_gfst1;
+    setup_func[R128_RGBA_BIT|R128_FOG_BIT|R128_SPEC_BIT|R128_TEX0_BIT|R128_TEX1_BIT] = rs_gfst0t1;
+}
+
+/* Initialize the vertex buffer setup functions based on the current
+   rendering state */
+void r128ChooseRasterSetupFunc(GLcontext *ctx)
+{
+    r128ContextPtr r128ctx   = R128_CONTEXT(ctx);
+    int            funcIndex = R128_WIN_BIT | R128_RGBA_BIT;
+
+    if (ctx->Texture.Enabled & 0xf) {
+	if (ctx->Texture.Unit[0].EnvMode == GL_REPLACE)
+	    funcIndex &= ~R128_RGBA_BIT;
+	funcIndex |= R128_TEX0_BIT;
+    }
+
+    if (ctx->Texture.Enabled & 0xf0)
+	funcIndex |= R128_TEX1_BIT;
+
+    /* FIXME: Verify this works properly */
+    if (ctx->Light.Model.ColorControl == GL_SEPARATE_SPECULAR_COLOR)
+	funcIndex |= R128_SPEC_BIT;
+
+    if (ctx->FogMode == FOG_FRAGMENT)
+	funcIndex |= R128_FOG_BIT;
+
+    r128ctx->SetupIndex = funcIndex;
+    ctx->Driver.RasterSetup = setup_func[funcIndex];
+}
+
+/* Check to see if any updates of the vertex buffer entries are needed */
+void r128CheckPartialRasterSetup(GLcontext *ctx,
+				 struct gl_pipeline_stage *s)
+{
+    r128ContextPtr r128ctx = R128_CONTEXT(ctx);
+    int            tmp     = r128ctx->SetupDone;
+
+    s->type = 0;
+    r128ctx->SetupDone = GL_FALSE;
+
+    if ((ctx->Array.Summary & VERT_OBJ_ANY) == 0) return;
+    if (ctx->IndirectTriangles) return;
+
+    r128ctx->SetupDone = tmp;
+}
+
+/* Update the vertex buffer entries, if necessary */
+void r128PartialRasterSetup(struct vertex_buffer *VB)
+{
+    r128ContextPtr r128ctx   = R128_CONTEXT(VB->ctx);
+    int            new       = VB->pipeline->new_outputs;
+    int            available = VB->pipeline->outputs;
+    int            index     = 0;
+
+    if (new & VERT_WIN) {
+	new = available;
+	index |= R128_WIN_BIT | R128_FOG_BIT;
+    }
+
+    if (new & VERT_RGBA)      index |= R128_RGBA_BIT | R128_SPEC_BIT;
+    if (new & VERT_TEX0_ANY)  index |= R128_TEX0_BIT;
+    if (new & VERT_TEX1_ANY)  index |= R128_TEX1_BIT;
+#if 0
+    /* FIXME */
+    if (new & VERT_FOG_COORD) index |= R128_FOG_BIT;
+#endif
+
+    r128ctx->SetupDone &= ~index;
+    index              &= r128ctx->SetupIndex;
+    r128ctx->SetupDone |= index;
+
+    if (index) setup_func[index & ~R128_ALPHA_BIT](VB, VB->Start, VB->Count);
+}
+
+/* Perform the raster setup for the fast path, if using CVA */
+void r128DoRasterSetup(struct vertex_buffer *VB)
+{
+    GLcontext *ctx = VB->ctx;
+
+    if (VB->Type == VB_CVA_PRECALC)   r128PartialRasterSetup(VB);
+    else if (ctx->Driver.RasterSetup) ctx->Driver.RasterSetup(VB,
+							      VB->CopyStart,
+							      VB->Count);
+}
+
+/* Resize an existing vertex buffer */
+void r128ResizeVB(struct vertex_buffer *VB, GLuint size)
+{
+    r128VertexBufferPtr r128vb = R128_DRIVER_DATA(VB);
+
+    while (r128vb->size < size)
+	r128vb->size *= 2;
+
+    free(r128vb->vert_store);
+    r128vb->vert_store = malloc(sizeof(r128Vertex) * r128vb->size + 31);
+    if (!r128vb->vert_store) {
+	fprintf(stderr, "Cannot allocate vertex store!  Exiting...\n");
+	exit(1);
+    }
+
+    r128vb->verts = (r128VertexPtr)(((CARD32)r128vb->vert_store + 31) & ~31);
+
+    gl_vector1ui_free(&r128vb->clipped_elements);
+    gl_vector1ui_alloc(&r128vb->clipped_elements,
+		       VEC_WRITABLE, r128vb->size, 32);
+    if (!r128vb->clipped_elements.start) {
+	fprintf(stderr, "Cannot allocate clipped elements!  Exiting...\n");
+	exit(1);
+    }
+
+    free(VB->ClipMask);
+    VB->ClipMask = (GLubyte *)malloc(sizeof(GLubyte) * r128vb->size);
+    if (!VB->ClipMask) {
+	fprintf(stderr, "Cannot allocate clipmask!  Exiting...\n");
+	exit(1);
+    }
+}
+
+/* Create a new device-dependent vertex buffer */
+void r128DDRegisterVB(struct vertex_buffer *VB)
+{
+    r128VertexBufferPtr r128vb;
+
+    r128vb = (r128VertexBufferPtr)calloc(1, sizeof(*r128vb));
+
+    r128vb->size = VB->Size * 2;
+    r128vb->vert_store = malloc(sizeof(r128Vertex) * r128vb->size + 31);
+    if (!r128vb->vert_store) {
+	fprintf(stderr, "Cannot allocate vertex store!  Exiting...\n");
+	exit(1);
+    }
+
+    r128vb->verts = (r128VertexPtr)(((CARD32)r128vb->vert_store + 31) & ~31);
+
+    gl_vector1ui_alloc(&r128vb->clipped_elements,
+		       VEC_WRITABLE, r128vb->size, 32);
+    if (!r128vb->clipped_elements.start) {
+	fprintf(stderr, "Cannot allocate clipped elements!  Exiting...\n");
+	exit(1);
+    }
+
+    free(VB->ClipMask);
+    VB->ClipMask = (GLubyte *)malloc(sizeof(GLubyte) * r128vb->size);
+    if (!VB->ClipMask) {
+	fprintf(stderr, "Cannot allocate clipmask!  Exiting...\n");
+	exit(1);
+    }
+
+    VB->driver_data = r128vb;
+}
+
+/* Destroy a device-dependent vertex buffer */
+void r128DDUnregisterVB(struct vertex_buffer *VB)
+{
+    r128VertexBufferPtr r128vb = R128_DRIVER_DATA(VB);
+
+    if (r128vb) {
+	if (r128vb->vert_store)     free(r128vb->vert_store);
+	gl_vector1ui_free(&r128vb->clipped_elements);
+	free(r128vb);
+	VB->driver_data = 0;
+    }
+}
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_vb.h b/xc/lib/GL/mesa/src/drv/r128/r128_vb.h
new file mode 100644
index 000000000..4f7c11d7f
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_vb.h
@@ -0,0 +1,150 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#ifndef _R128_VB_H_
+#define _R128_VB_H_
+
+#ifdef GLX_DIRECT_RENDERING
+
+#define USE_RHW2 0
+
+/* FIXME: This is endian-specific */
+typedef struct {
+    GLubyte b;
+    GLubyte g;
+    GLubyte r;
+    GLubyte a;
+} r128Color;
+
+typedef struct {
+    GLfloat   x, y, z;		/* Coordinates in screen space */
+    GLfloat   rhw;		/* Reciprocal homogeneous w */
+    r128Color dif_argb;		/* Diffuse color */
+    r128Color spec_frgb;	/* Specular color (alpha is fog) */
+    GLfloat   tu0, tv0;		/* Texture 0 coordinates */
+    GLfloat   tu1, tv1;		/* Texture 1 coordinates */
+#if USE_RHW2
+    GLfloat   rhw2;		/* Reciprocal homogeneous w */
+#endif
+} r128_vertex;
+
+#if USE_RHW2
+/* Format of vertices in r128_vertex struct */
+#define R128_FULL_VERTEX_FORMAT                                            \
+    R128_CCE_VC_FRMT_RHW          |                                        \
+    R128_CCE_VC_FRMT_DIFFUSE_ARGB |                                        \
+    R128_CCE_VC_FRMT_SPEC_FRGB    |                                        \
+    R128_CCE_VC_FRMT_S_T          |                                        \
+    R128_CCE_VC_FRMT_S2_T2        |                                        \
+    R128_CCE_VC_FRMT_RHW2
+
+/* Send a single vertex to the ring buffer */
+#define R128CCE_SEND_VERTEX(v)                                              \
+    do {                                                                    \
+	R128CCEF((v)->x);                                                   \
+	R128CCEF((v)->y);                                                   \
+	R128CCEF((v)->z);                                                   \
+	R128CCEF((v)->rhw);                                                 \
+	R128CCE(*(int *)&(v)->dif_argb);                                    \
+	R128CCE(*(int *)&(v)->spec_frgb);                                   \
+	R128CCEF((v)->tu0);                                                 \
+	R128CCEF((v)->tv0);                                                 \
+	R128CCEF((v)->tu1);                                                 \
+	R128CCEF((v)->tv1);                                                 \
+	R128CCEF((v)->rhw2);                                                \
+    } while (0)
+
+#else /* !USE_RHW2 */
+
+/* Format of vertices in r128_vertex struct */
+#define R128_FULL_VERTEX_FORMAT                                            \
+    R128_CCE_VC_FRMT_RHW          |                                        \
+    R128_CCE_VC_FRMT_DIFFUSE_ARGB |                                        \
+    R128_CCE_VC_FRMT_SPEC_FRGB    |                                        \
+    R128_CCE_VC_FRMT_S_T          |                                        \
+    R128_CCE_VC_FRMT_S2_T2
+
+/* Send a single vertex to the ring buffer */
+#define R128CCE_SEND_VERTEX(v)                                              \
+    do {                                                                    \
+	R128CCEF((v)->x);                                                   \
+	R128CCEF((v)->y);                                                   \
+	R128CCEF((v)->z);                                                   \
+	R128CCEF((v)->rhw);                                                 \
+	R128CCE(*(int *)&(v)->dif_argb);                                    \
+	R128CCE(*(int *)&(v)->spec_frgb);                                   \
+	R128CCEF((v)->tu0);                                                 \
+	R128CCEF((v)->tv0);                                                 \
+	R128CCEF((v)->tu1);                                                 \
+	R128CCEF((v)->tv1);                                                 \
+    } while (0)
+#endif
+
+/* FIXME: We currently only have assembly for 16-stride vertices */
+typedef union {
+    r128_vertex v;
+    float       f[16];
+} r128Vertex, *r128VertexPtr;
+
+/* Vertex buffer for use when on the fast path */
+typedef struct {
+    GLuint         size;		/* Number of vertices in store */
+    void          *vert_store;		/* Storage for vertex buffer */
+    r128VertexPtr  verts;		/* Aligned start of verts in storage */
+    int            last_vert;		/* Index of last vertex used */
+    GLvector1ui    clipped_elements;	/* List of clipped elements */
+} *r128VertexBufferPtr;
+
+#define R128_DRIVER_DATA(vb) ((r128VertexBufferPtr)((vb)->driver_data))
+
+#define R128_SPEC_BIT       0x01
+#define R128_FOG_BIT        0x02
+#define R128_ALPHA_BIT      0x04 /* GL_BLEND, not used */
+#define R128_TEX1_BIT       0x08
+#define R128_TEX0_BIT       0x10        
+#define R128_RGBA_BIT       0x20
+#define R128_WIN_BIT        0x40
+
+extern void         r128SetupInit(void);
+extern void         r128ChooseRasterSetupFunc(GLcontext *ctx);
+extern void         r128CheckPartialRasterSetup(GLcontext *ctx,
+						struct gl_pipeline_stage *s);
+extern void         r128PartialRasterSetup(struct vertex_buffer *VB);
+extern void         r128DoRasterSetup(struct vertex_buffer *VB);
+extern void         r128ResizeVB(struct vertex_buffer *VB, GLuint size);
+extern void         r128DDRegisterVB(struct vertex_buffer *VB);
+extern void         r128DDUnregisterVB(struct vertex_buffer *VB);
+
+#endif
+#endif /* _R128_VB_H_ */
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_xmesa.c b/xc/lib/GL/mesa/src/drv/r128/r128_xmesa.c
new file mode 100644
index 000000000..60d528813
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_xmesa.c
@@ -0,0 +1,257 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas.
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#ifdef GLX_DIRECT_RENDERING
+
+/* r128 Mesa driver includes */
+#include "r128_init.h"
+#include "r128_context.h"
+#include "r128_xmesa.h"
+#include "r128_state.h"
+#include "r128_tex.h"
+#include "r128_swap.h"
+
+/* Mesa src includes */
+#include "context.h"
+#include "simple_list.h"
+#include "mmath.h"
+
+#ifndef R128_DEBUG_FLAGS
+int R128_DEBUG_FLAGS = (0
+#if 0
+			| DEBUG_ALWAYS_SYNC
+			| DEBUG_VERBOSE_CCE
+			| DEBUG_VERBOSE_OUTREG
+			| DEBUG_VERBOSE_MSG
+			| DEBUG_NO_OUTRING
+			| DEBUG_NO_OUTREG
+			| DEBUG_VERBOSE_API
+			| DEBUG_VERBOSE_2D
+			| DEBUG_VERBOSE_DRI
+			| DEBUG_VALIDATE_RING
+			| DEBUG_VERBOSE_IOCTL
+#endif
+    );
+#endif
+
+#if DEBUG_LOCKING
+char *prevLockFile = NULL;
+int   prevLockLine = 0;
+#endif
+
+static r128ContextPtr r128Context = NULL;
+
+
+/* Initialize the driver specific screen private data */
+GLboolean XMesaInitDriver(__DRIscreenPrivate *sPriv)
+{
+    sPriv->private = (void *)r128CreateScreen(sPriv);
+    if (!sPriv->private) {
+	r128DestroyScreen(sPriv);
+	return GL_FALSE;
+    }
+
+    return GL_TRUE;
+}
+
+/* Reset the driver specific screen private data */
+void XMesaResetDriver(__DRIscreenPrivate *sPriv)
+{
+    r128DestroyScreen(sPriv);
+}
+
+/* Create and initialize the Mesa and driver specific visual data */
+GLvisual *XMesaCreateVisual(Display *dpy,
+			    __DRIscreenPrivate *driScrnPriv,
+			    const XVisualInfo *visinfo,
+			    const __GLXvisualConfig *config)
+{
+    /* Drivers may change the args to _mesa_create_visual() in order to
+     * setup special visuals.
+     */
+    return _mesa_create_visual(config->rgba,
+			       config->doubleBuffer,
+			       config->stereo,
+			       _mesa_bitcount(visinfo->red_mask),
+			       _mesa_bitcount(visinfo->green_mask),
+			       _mesa_bitcount(visinfo->blue_mask),
+			       config->alphaSize,
+			       0, /* index bits */
+			       config->depthSize,
+			       config->stencilSize,
+			       config->accumRedSize,
+			       config->accumGreenSize,
+			       config->accumBlueSize,
+			       config->accumAlphaSize,
+			       0 /* num samples */);
+}
+
+/* Create and initialize the Mesa and driver specific context data */
+GLboolean XMesaCreateContext(Display *dpy, GLvisual *mesaVis,
+			     __DRIcontextPrivate *driContextPriv)
+{
+    return r128CreateContext(dpy, mesaVis, driContextPriv);
+}
+
+/* Destroy the Mesa and driver specific context data */
+void XMesaDestroyContext(__DRIcontextPrivate *driContextPriv)
+{
+    r128ContextPtr r128ctx = (r128ContextPtr)driContextPriv->driverPrivate;
+
+    if (r128ctx == (void *)r128Context) r128Context = NULL;
+    r128DestroyContext(r128ctx);
+}
+
+/* Create and initialize the Mesa and driver specific pixmap buffer data */
+GLframebuffer *XMesaCreateWindowBuffer(Display *dpy,
+				       __DRIscreenPrivate *driScrnPriv,
+				       __DRIdrawablePrivate *driDrawPriv,
+				       GLvisual *mesaVis)
+{
+   return gl_create_framebuffer(mesaVis,
+                                GL_FALSE,  /* software depth buffer? */
+                                mesaVis->StencilBits > 0,
+                                mesaVis->AccumRedBits > 0,
+                                mesaVis->AlphaBits > 0
+                                );
+}
+
+/* Create and initialize the Mesa and driver specific pixmap buffer data */
+GLframebuffer *XMesaCreatePixmapBuffer( Display *dpy,
+                                        __DRIscreenPrivate *driScrnPriv,
+                                        __DRIdrawablePrivate *driDrawPriv,
+                                        GLvisual *mesaVis)
+{
+#if 0
+    /* Different drivers may have different combinations of hardware and
+     * software ancillary buffers.
+     */
+    return gl_create_framebuffer(mesaVis,
+				 GL_FALSE,  /* software depth buffer? */
+				 mesaVis->StencilBits > 0,
+				 mesaVis->AccumRedBits > 0,
+				 mesaVis->AlphaBits > 0);
+#else
+    return NULL;  /* not implemented yet */
+#endif
+}
+
+/* Copy the back color buffer to the front color buffer */
+void XMesaSwapBuffers(__DRIdrawablePrivate *driDrawPriv)
+{
+    /* FIXME: This assumes buffer is currently bound to a context.  This
+       needs to be able to swap buffers when not currently bound.  Also,
+       this needs to swap according to buffer, and NOT according to
+       context! */
+    if (r128Context == NULL) return;
+
+    /* Only swap buffers when a back buffer exists */
+    if (R128_MESACTX(r128Context)->Visual->DBflag) {
+	FLUSH_VB(R128_MESACTX(r128Context), "swap buffers");
+	r128SwapBuffers(r128Context);
+    }
+}
+
+/* Force the context `c' to be the current context and associate with it
+   buffer `b' */
+GLboolean XMesaMakeCurrent(__DRIcontextPrivate *driContextPriv,
+			   __DRIdrawablePrivate *driDrawPriv,
+			   __DRIdrawablePrivate *driReadPriv)
+{
+    if (driContextPriv) {
+	r128ContextPtr r128ctx = (r128ContextPtr)driContextPriv->driverPrivate;
+
+	if (r128Context &&
+	    r128ctx == (void *)r128Context &&
+	    driDrawPriv == R128_DRIDRAWABLE(r128Context))
+	    return GL_TRUE;
+
+	r128Context = r128MakeCurrent(r128Context, r128ctx, driDrawPriv);
+
+	gl_make_current2(R128_MESACTX(r128Context),
+			 driDrawPriv->mesaBuffer, driReadPriv->mesaBuffer);
+
+	if (!R128_MESACTX(r128Context)->Viewport.Width) {
+	    gl_Viewport(R128_MESACTX(r128Context), 0, 0,
+			driDrawPriv->w, driDrawPriv->h);
+	}
+    } else {
+	gl_make_current(0,0);
+	r128Context = NULL;
+    }
+
+    return GL_TRUE;
+}
+
+/* Force the context `c' to be unbound from its buffer */
+GLboolean XMesaUnbindContext(__DRIcontextPrivate *driContextPriv)
+{
+    return GL_TRUE;
+}
+
+/* Update the hardware state.  This is called if another context has
+   grabbed the hardware lock, which includes the X server.  This
+   function also updates the driver's window state after the X server
+   moves, resizes or restacks a window -- the change will be reflected
+   in the drawable position and clip rects.  Since the X server grabs
+   the hardware lock when it changes the window state, this routine will
+   automatically be called after such a change. */
+/* NOTE: This routine is only called while holding the hardware lock. */
+void XMesaUpdateState(__DRIcontextPrivate *driContextPriv)
+{
+    r128ContextPtr        r128ctx = driContextPriv->driverPrivate;
+    __DRIscreenPrivate   *sPriv   = R128_DRISCREEN(r128ctx);
+    __DRIdrawablePrivate *dPriv   = R128_DRIDRAWABLE(r128ctx);
+    int                   stamp   = dPriv->lastStamp;
+
+    /* The window might have moved, so we might need to get new clip
+       rects.
+
+       NOTE: This releases and regrabs the hw lock to allow the X server
+       to respond to the DRI protocol request for new drawable info.
+       Since the hardware state depends on having the latest drawable
+       clip rects, all state checking must be done _after_ this call. */
+    XMESA_VALIDATE_DRAWABLE_INFO(r128ctx->display, sPriv, dPriv);
+
+    r128UpdateState(r128ctx, (stamp != dPriv->lastStamp));
+}
+
+/* This function is called by libGL.so as soon as libGL.so is loaded.
+ * This is where we'd register new extension functions with the dispatcher.
+ */
+void __driRegisterExtensions(void)
+{
+}
+
+#endif
diff --git a/xc/lib/GL/mesa/src/drv/r128/r128_xmesa.h b/xc/lib/GL/mesa/src/drv/r128/r128_xmesa.h
new file mode 100644
index 000000000..ca6fc127b
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/r128/r128_xmesa.h
@@ -0,0 +1,44 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#ifndef _R128_XMESA_H_
+#define _R128_XMESA_H_
+
+#ifdef GLX_DIRECT_RENDERING
+
+extern void XMesaUpdateState(__DRIcontextPrivate *driContextPriv);
+extern void __driRegisterExtensions(void);
+
+#endif
+#endif /* _R128_XMESA_H_ */
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/Imakefile b/xc/lib/GL/mesa/src/drv/tdfx/Imakefile
index c159a8e76..4a6a8e0ff 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/Imakefile
+++ b/xc/lib/GL/mesa/src/drv/tdfx/Imakefile
@@ -79,7 +79,8 @@ LinkSourceFile(fxvs_tmp.h, ../../../../../../extras/Mesa/src/FX)
 		fxglidew.o fxpipeline.o fxrender.o fxsanity.o fxsetup.o \
 		fxtexman.o fxtrifuncs.o fxvsetup.o
 
-     MESASRCS = ../../accum.c \
+     MESASRCS = ../../aatriangle.c \
+		../../accum.c \
 		../../alpha.c \
 		../../alphabuf.c \
 		../../attrib.c \
@@ -109,7 +110,7 @@ LinkSourceFile(fxvs_tmp.h, ../../../../../../extras/Mesa/src/FX)
 		../../glthread.c \
 		../../hash.c \
 		../../image.c \
-		../../imaging.o \
+		../../imaging.c \
 		../../light.c \
 		../../lines.c \
 		../../logic.c \
@@ -120,6 +121,7 @@ LinkSourceFile(fxvs_tmp.h, ../../../../../../extras/Mesa/src/FX)
 		../../pb.c \
 		../../pipeline.c \
 		../../pixel.c \
+		../../pixeltex.c \
 		../../points.c \
 		../../polygon.c \
 		../../quads.c \
@@ -136,6 +138,7 @@ LinkSourceFile(fxvs_tmp.h, ../../../../../../extras/Mesa/src/FX)
 		../../texobj.c \
 		../../texstate.c \
 		../../texture.c \
+		../../texutil.c \
 		../../translate.c \
 		../../triangle.c \
 		../../varray.c \
@@ -152,7 +155,8 @@ LinkSourceFile(fxvs_tmp.h, ../../../../../../extras/Mesa/src/FX)
 		../../zoom.c \
 		../../X86/common_x86.c
 
-     MESAOBJS = ../../accum.o \
+     MESAOBJS = ../../aatriangle.o \
+		../../accum.o \
 		../../alpha.o \
 		../../alphabuf.o \
 		../../attrib.o \
@@ -191,6 +195,7 @@ LinkSourceFile(fxvs_tmp.h, ../../../../../../extras/Mesa/src/FX)
 		../../pb.o \
 		../../pipeline.o \
 		../../pixel.o \
+		../../pixeltex.o \
 		../../points.o \
 		../../polygon.o \
 		../../quads.o \
@@ -207,6 +212,7 @@ LinkSourceFile(fxvs_tmp.h, ../../../../../../extras/Mesa/src/FX)
 		../../texobj.o \
 		../../texstate.o \
 		../../texture.o \
+		../../texutil.o \
 		../../translate.o \
 		../../triangle.o \
 		../../varray.o \
@@ -239,8 +245,7 @@ LinkSourceFile(fxvs_tmp.h, ../../../../../../extras/Mesa/src/FX)
 
      MMX_OBJS = ../../X86/mmx_blend.o
 
-XCOMM Disabling 3Dnow code for the time being.
-#if 0
+#if MesaUse3DNow
    3DNOW_SRCS = ../../X86/3dnow.c \
 		../../X86/3dnow_norm_raw.S \
 		../../X86/3dnow_xform_masked1.S \
@@ -271,10 +276,14 @@ XCOMM Disabling 3Dnow code for the time being.
       ASMSRCS = $(X86_SRCS) $(MMX_SRCS) $(3DNOW_SRCS)
       ASMOBJS = $(X86_OBJS) $(MMX_OBJS) $(3DNOW_OBJS)
 
-	 SRCS = $(DRISRCS) $(DRMSRCS) $(TDFXSRCS) $(MESASRCS) $(ASMSRCS)
-	 OBJS = $(DRIOBJS) $(DRMOBJS) $(TDFXOBJS) $(MESAOBJS) $(ASMOBJS)
+         SRCS = $(DRISRCS) $(DRMSRCS) $(TDFXSRCS) $(MESASRCS) $(ASMSRCS)
+         OBJS = $(DRIOBJS) $(DRMOBJS) $(TDFXOBJS) $(MESAOBJS) $(ASMOBJS)
+
+REQUIREDLIBS = -lglide3x -lm
+#if !GlxBuiltInTdfx
+REQUIREDLIBS += -L../../../.. -lGL
+#endif
 
-REQUIREDLIBS += -lglide3x
 
 #if !GlxUseBuiltInDRIDriver
 #undef DoNormalLib NormalLibGlx
diff --git a/xc/programs/Xserver/GL/dri/dri.c b/xc/programs/Xserver/GL/dri/dri.c
index 0bff80a6c..c48c156bd 100644
--- a/xc/programs/Xserver/GL/dri/dri.c
+++ b/xc/programs/Xserver/GL/dri/dri.c
@@ -112,7 +112,7 @@ DRIScreenInit(ScreenPtr pScreen, DRIInfoPtr pDRIInfo, int *pDRMFD)
     fd = drmOpen(pDRIInfo->drmDriverName, NULL);
     if (fd < 0) {
         /* failed to open DRM */
-        pScreen->devPrivates[DRIScreenPrivIndex].ptr = 0;
+        pScreen->devPrivates[DRIScreenPrivIndex].ptr = NULL;
         return FALSE;
     }
 
@@ -125,7 +125,7 @@ DRIScreenInit(ScreenPtr pScreen, DRIInfoPtr pDRIInfo, int *pDRMFD)
 
     pDRIPriv = (DRIScreenPrivPtr) xalloc(sizeof(DRIScreenPrivRec));
     if (!pDRIPriv) {
-        pScreen->devPrivates[DRIScreenPrivIndex].ptr = 0;
+        pScreen->devPrivates[DRIScreenPrivIndex].ptr = NULL;
         return FALSE;
     }
 
@@ -137,6 +137,7 @@ DRIScreenInit(ScreenPtr pScreen, DRIInfoPtr pDRIInfo, int *pDRMFD)
 
     if (drmSetBusid(pDRIPriv->drmFD, pDRIPriv->pDriverInfo->busIdString) < 0) {
 	pDRIPriv->directRenderingSupport = FALSE;
+	pScreen->devPrivates[DRIScreenPrivIndex].ptr = NULL;
 	drmClose(pDRIPriv->drmFD);
 	return FALSE;
     }
@@ -155,6 +156,7 @@ DRIScreenInit(ScreenPtr pScreen, DRIInfoPtr pDRIInfo, int *pDRMFD)
 		   &pDRIPriv->hSAREA) < 0)
     {
 	pDRIPriv->directRenderingSupport = FALSE;
+	pScreen->devPrivates[DRIScreenPrivIndex].ptr = NULL;
 	drmClose(pDRIPriv->drmFD);
 	return FALSE;
     }
@@ -168,6 +170,7 @@ DRIScreenInit(ScreenPtr pScreen, DRIInfoPtr pDRIInfo, int *pDRMFD)
 		(drmAddressPtr)(&pDRIPriv->pSAREA)) < 0)
     {
 	pDRIPriv->directRenderingSupport = FALSE;
+	pScreen->devPrivates[DRIScreenPrivIndex].ptr = NULL;
 	drmClose(pDRIPriv->drmFD);
 	return FALSE;
     }
@@ -182,6 +185,7 @@ DRIScreenInit(ScreenPtr pScreen, DRIInfoPtr pDRIInfo, int *pDRMFD)
 		   &pDRIPriv->hFrameBuffer) < 0)
     {
 	pDRIPriv->directRenderingSupport = FALSE;
+	pScreen->devPrivates[DRIScreenPrivIndex].ptr = NULL;
 	drmUnmap(pDRIPriv->pSAREA, pDRIPriv->pDriverInfo->SAREASize);
 	drmClose(pDRIPriv->drmFD);
 	return FALSE;
diff --git a/xc/programs/Xserver/Imakefile b/xc/programs/Xserver/Imakefile
index c1e91e6c5..cf9770537 100644
--- a/xc/programs/Xserver/Imakefile
+++ b/xc/programs/Xserver/Imakefile
@@ -619,7 +619,7 @@ XF86IDRVOBJS = $(XF86SRC)/input/drvConf.o
 XF86IDRVLIBS = $(XF86IDRIVERLIB)
 XF86SCANLIB = $(XF86SRC)/scanpci/LibraryTargetName(scanpci)
 XF86LIBS  = $(XF86INIT) $(XF86COMLIB) $(XF86RACLIB) $(XF86PARSLIB) \
-	    $(XF86OSLIB) $(XF86INT10LIB)
+	    $(XF86OSLIB) $(XF86INT10LIB) $(XF86DDCLIB)
 #else
 XF86LIBS  = $(XF86INIT) $(XF86COMLIB) $(XF86PARSLIB) $(XF86OSLIB) 
 #endif
diff --git a/xc/programs/Xserver/hw/xfree86/doc/sgml/DRI.sgml b/xc/programs/Xserver/hw/xfree86/doc/sgml/DRI.sgml
index 895ce09a7..e95ff113a 100644
--- a/xc/programs/Xserver/hw/xfree86/doc/sgml/DRI.sgml
+++ b/xc/programs/Xserver/hw/xfree86/doc/sgml/DRI.sgml
@@ -3,17 +3,17 @@
 ]>
 
 <!-- Created: Mon Feb 28 13:00:00 2000 by brian@precisioninsight.com -->
-<!-- Revised: Mon Mar  6 20:10:02 2000 by kevin@precisioninsight.com -->
+<!-- Revised: Fri May 19 09:41:48 2000 by martin@valinux.com -->
 
   <article>
 
-      <title>DRI Users Guide
+      <title>DRI User Guide
       <author><htmlurl url="http://www.precisioninsight.com/"
                        name="Precision Insight, Inc.">
-      <date>6 March 2000
+      <date>18 May 2000
 
     <ident>
-    $XFree86: xc/programs/Xserver/hw/xfree86/doc/sgml/DRI.sgml,v 1.3 2000/03/08 05:38:41 dawes Exp $
+    $XFree86: xc/programs/Xserver/hw/xfree86/doc/sgml/DRI.sgml,v 1.3 2000/04/05 05:38:41 brianp Exp $
     </ident>
 
     <toc>
@@ -25,7 +25,7 @@
           <bf>Copyright &copy; 2000 by Precision Insight, Inc.,
           Cedar Park, Texas.
           All Rights Reserved.</bf>
-        <p>
+<p>
           <bf>Permission is granted to make and distribute verbatim copies 
           of this document provided the copyright notice and this permission
           notice are preserved on all copies.</bf>
@@ -43,6 +43,8 @@
           trademarks of 3Dlabs Inc. Ltd.
           3dfx, Voodoo3, Voodoo4, and Voodoo5 are registered trademarks of
           3dfx Interactive, Incorporated.
+	  Matrox is a registered trademark of Matrox Electronic Systems Ltd.
+	  ATI Rage is a registered trademark of ATI Technologies, Inc.
           All other trademarks mentioned are the property of their
           respective owners.
 
@@ -60,9 +62,13 @@
 	software downloads.
       <p>
         This document does not cover compilation or installation of
-        XFree86 4.0;
-        it is assumed that you've already installed a Linux distribution which
-        includes XFree86 4.0.  
+        XFree86 4.0.
+        It is assumed that you've already installed a Linux distribution which
+        includes XFree86 4.0 or that you're an experienced Linux developer
+        who has compiled the DRI for himself.
+        DRI download, compilation and installation instructions can be found
+        at <htmlurl url="http://dri.sourceforge.net/DRIcompile.html"
+           name="http://dri.sourceforge.net/DRIcompile.html">
 
     <sect>Supported Hardware
 <p>
@@ -71,8 +77,8 @@
         Support for Alpha, and perhaps other CPUs, should be available in the
         future.
       <p>
-        XFree86 4.0 includes 3D acceleration for the following
-        graphics hardware:
+        XFree86 4.0 (or later versions) includes 3D acceleration for the
+        following graphics hardware:
 
         <itemize>
           <item>3dfx:
@@ -87,6 +93,19 @@
             </itemize>
             There are many configurations of 3dfx cards on the market.
             Not all have been tested.
+	  <item>Matrox G400
+	  <item>ATI Rage 128
+            <itemize>
+              <item>Rage Fury AGP
+              <item>Rage Magnum AGP
+              <item>XPERT 2000 AGP
+              <item>XPERT 128 AGP
+	      <item>XPERT 99 AGP
+	      <item>All-in-Wonder 128 AGP
+            </itemize>
+	    The PCI versions of these cards also have minimal support.
+	    Note that there are also Rage 128 Pro based boards on the
+	    market, and these are not yet supported.
           <item>3Dlabs Oxygen GMX 2000 (MX/Gamma based)
         </itemize>
 
@@ -94,8 +113,6 @@
         Support for the following hardware is underway:
 	<itemize>
           <item>Intel i810
-	  <item>Matrox G400
-	  <item>ATI Rage 128
 	  <item>3dfx Voodoo4 and Voodoo5 series
 	</itemize>
 
@@ -104,8 +121,10 @@
 <p>
         <itemize>
           <item>XFree86 4.0
-          <item>Linux kernel 2.2.x (later kernels will be supported in
+          <item>For the 3dfx Voodoo3 driver, Linux kernel 2.2.x (later
+                kernels will be supported in
                 the near future, and may be required for some chipsets)
+          <item>For the Matrox G400, Linux kernel 2.3.51, with AGP support
         </itemize>
       <p>
         Mesa 3.3 (beta) is included with XFree86 4.0; there is no need to
@@ -119,20 +138,17 @@
 
       <sect1>Kernel module
 <p>
-          Before starting the X server you must install the correct kernel
-          module for your hardware.
-        <p>
-          This can be done by executing the following as root:
-        <verb>
-        insmod XXX/drivername.o
-        </verb>
-        <p>
-          For example, on 3dfx hardware, the kernel module is called tdfx.o
-          so you you would type insmod XXX/tdfx.o
-
-        <p>
-          Verify that the kernel module was installed by checking that
-          /proc/dri/0 exists. 
+          XFree86 4.0.1 added automatic kernel module loading to the X server.
+          On Linux, the X server uses modprobe to load kernel modules.
+          The DRM kernel modules should be in /lib/modules/KERNEL-VERSION/misc/
+          for automatic loading to work.
+<p>
+          Optionally, DRM kernel modules can be loaded manually with insmod
+          prior to starting the X server.
+<p>
+          You can verify that the kernel module was installed with lsmod,
+          checking the X server startup log, and checking that /proc/dri/0
+          exists. 
 
       <sect1>XF86Config file
 <p>
@@ -199,6 +215,7 @@
         <p>
           Next, the Device section of the XF86Config file must describe your
           particular hardware.
+        <p>
           For example, here's the Device section for a 3dfx Voodoo3 card:
 
         <verb>
@@ -210,6 +227,27 @@
         </verb>
 
         <p>
+          Here's the Device section for a Matrox G400 card:
+        <verb>
+	Section "Device"
+	    Identifier  "G400"
+	    VendorName  "Matrox"
+	    Driver      "mga"
+	    VideoRam    32768
+	EndSection
+        </verb>
+
+        <p>
+          Here's the Device section for an ATI Rage 128 card:
+        <verb>
+	Section "Device"
+	    Identifier  "Rage128"
+	    VendorName  "ATI"
+	    Driver      "r128"
+	EndSection
+        </verb>
+
+        <p>
           Finally, the Screen section of the XF86Config file may have to be
           specially configured as well.
           For example, Voodoo3 hardware acceleration is only available
@@ -227,7 +265,10 @@
 	    EndSubsection
         EndSection
         </verb>
-
+        <p>
+          Replace <em>Voodoo3</em> with <em>G400</em> for Matrox G400.
+        <p>
+          Replace <em>Voodoo3</em> with <em>Rage128</em> for ATI Rage 128.
         <p>
           If there are errors in the XF86Config file, the X server will
           log errors to the file /var/log/XFree86.0.log
@@ -301,6 +342,25 @@
           if means that the libGL.so.1 file is not the right location.
           Proceed to the trouble shooting section.          
 
+      <sect1>libOSMesa.so
+<p>
+          OSMesa (Off-Screen Mesa) is an interface and driver for rendering
+          3D images into a user-allocated block of memory rather than an
+          on-screen window.
+<p>
+          libOSMesa.so implements the OSMesa interface and must be linked
+          with your application if you want to use the OSMesa functions.
+          You must also link with libGL.so.  For example:
+        <verb>
+        gcc osdemo.c -L/usr/X11R6/lib -lOSMesa -lGLU -lGL -o osdemo
+        </verb>
+<p>
+          In stand-alone Mesa this interface was compiled into the monolithic
+          libGL.so (formerly libMesaGL.so) library.
+          In XFree86 4.0.1 and later this interface is implemented in a
+          separate library.
+<p>
+
       <sect1>glxinfo
 <p>
           glxinfo is a useful program for checking which version of
@@ -346,20 +406,25 @@
           <enum>
             <item>
               <tt/LIBGL_DEBUG/, if defined will cause libGL.so to print error
-              and diagnostic messages.  This can help to solve problems.
+              and diagnostic messages.
+              This can help to solve problems.
+              Setting <tt/LIBGL_DEBUG/ to <tt/verbose/ may provide additional
+              information.
 	    <item>
 	      <tt/LIBGL_ALWAYS_INDIRECT/, if defined this will force libGL.so
 	      to always use indirect rendering instead of hardware
-               acceleration.
+              acceleration.
 	      This can be useful to isolate rendering errors.
 	    <item>
-	      <tt/LIBGL_DRIVERS_DIR/ can be used to override the default
-	      directory which is searched for 3D drivers.
+	      <tt/LIBGL_DRIVERS_PATH/ can be used to override the default
+	      directories which are searched for 3D drivers.
+              The value is one or more paths separated by colons.
 	      In a typical XFree86 installation, the 3D drivers should be in
-	      /usr/X11R6/lib/modules/dri/.
-	      This environment variable can be used to specify a different
-	      directory.
+	      /usr/X11R6/lib/modules/dri/ and <tt/LIBGL_DRIVERS_PATH/ need
+              not be defined.
 	      Note that this feature is disabled for set-uid programs.
+              This variable replaces the <tt/LIBGL_DRIVERS_DIR/ env var used
+              in XFree86 4.0.
 	  </enum>
         <p>
           Mesa-based drivers (this includes most of the drivers listed
@@ -531,6 +596,40 @@
 
       <sect1>3dfx Voodoo3
 <p>
+        <sect2>Configuration
+<p>
+            Your XF86Config file's device section must specify the
+            <tt>tdfx</tt> device:
+            <verb>
+        Section "Device"
+            Identifier  "Voodoo3"
+            VendorName  "3dfx"
+            Driver      "tdfx"
+        EndSection
+            </verb>
+            The Screen section should then reference the Voodoo3 device:
+            <verb>
+	Section "Screen"
+	    Identifier  "Screen 1"
+	    Device      "Voodoo3"
+	    Monitor     "High Res Monitor"
+	    DefaultDepth 16
+	    Subsection "Display"
+		Depth       16
+		Modes       "1280x1024" "1024x768" "800x600" "640x480"
+		ViewPort    0 0
+	    EndSubsection
+        EndSection
+            </verb>
+          <p>
+            The kernel module for the Voodoo3 is named <tt>tdfx.o</tt> and
+	    should be installed in /lib/modules/KERNEL-VERSION/misc/.
+            It will be automatically loaded by the Xserver if needed.
+          <p>
+	    The DRI 3D driver for the Voodoo3 should be in
+	    <tt>/usr/X11R6/lib/modules/dri/tdfx_dri.so</tt>.
+            This will be automatically loaded by libGL.so.
+
         <sect2>Troubleshooting
 <p>
           <itemize>
@@ -539,6 +638,9 @@
               bit/pixel screen mode.
               Use <tt/xdpyinfo/ to verify that all your visuals are depth 16.
               Edit your XF86Config file if needed.
+           <item>
+              The <tt>/dev/3dfx</tt> device is not used for DRI; it's only for
+              Glide on older 3dfx hardware.
           </itemize>
 
         <sect2>Performance
@@ -565,6 +667,9 @@
 <p>
           <itemize>
             <item>
+              Glide cannot be used directly; only OpenGL-based programs are
+              supported on the Voodoo3.
+            <item>
               SSystem has problems because of poorly set near and far
               clipping planes.
               The office.unc Performer model also suffers from this problem.
@@ -573,10 +678,120 @@
 
       <sect1>Intel i810
 <p>
+          Information to be added in the future.
+
       <sect1>Matrox G400
 <p>
+        <sect2>Configuration
+<p>
+            Your XF86Config file's device section must specify the
+            <tt>mga</tt> device:
+            <verb>
+        Section "Device"
+            Identifier  "G400"
+            VendorName  "Matrox"
+            Driver      "mga"
+        EndSection
+            </verb>
+            The Screen section should then reference the G400 mga device:
+            <verb>
+	Section "Screen"
+	    Identifier  "Screen 1"
+	    Device      "G400"
+	    Monitor     "High Res Monitor"
+	    DefaultDepth 16
+	    Subsection "Display"
+		Depth       16
+		Modes       "1280x1024" "1024x768" "800x600" "640x480"
+		ViewPort    0 0
+	    EndSubsection
+        EndSection
+            </verb>
+          <p>
+            The kernel module for the G400 is named <tt>mga.o</tt> and
+	    should be installed in /lib/modules/KERNEL-VERSION/misc/.
+            It will be automatically loaded by the Xserver if needed.
+          <p>
+	    The DRI 3D driver for the G400 should be in
+	    <tt>/usr/X11R6/lib/modules/dri/mga_dri.so</tt>.
+            This will be automatically loaded by libGL.so.
+
+        <sect2>Troubleshooting
+<p>
+          <itemize>
+            <item>
+              3D acceleration for the G400 is only supported in the 16
+              bit/pixel screen mode at this time.  32bpp will be supported
+              in the future.
+              Use <tt/xdpyinfo/ to verify that all your visuals are depth 16.
+              Edit your XF86Config file if needed.
+          </itemize>
+
+        <sect2>Performance
+<p>
+          No data at this time.
+
+        <sect2>Known Problems
+<p>
+          No data at this time.
+
+
       <sect1>ATI Rage 128
 <p>
+        <sect2>Configuration
+<p>
+            Your XF86Config file's device section must specify the
+            <tt>r128</tt> device:
+            <verb>
+        Section "Device"
+            Identifier  "Rage128"
+            VendorName  "ATI"
+            Driver      "r128"
+        EndSection
+            </verb>
+            The Screen section should then reference the Rage 128 device:
+            <verb>
+	Section "Screen"
+	    Identifier  "Screen 1"
+	    Device      "Rage128"
+	    Monitor     "High Res Monitor"
+	    DefaultDepth 16
+	    Subsection "Display"
+		Depth       16
+		Modes       "1280x1024" "1024x768" "800x600" "640x480"
+		ViewPort    0 0
+	    EndSubsection
+	    Subsection "Display"
+		Depth       32
+		Modes       "1280x1024" "1024x768" "800x600" "640x480"
+		ViewPort    0 0
+	    EndSubsection
+        EndSection
+            </verb>
+          <p>
+            The kernel module for the Rage 128 is named <tt>r128.o</tt> and
+	    should be installed in /lib/modules/KERNEL-VERSION/misc/.
+            It will be automatically loaded by the Xserver if needed.
+          <p>
+	    The DRI 3D driver for the Rage 128 should be in
+	    <tt>/usr/X11R6/lib/modules/dri/r128_dri.so</tt>.
+            This will be automatically loaded by libGL.so.
+          <p>
+            You may also set your screen depth to 32 for 32bpp mode.
+          <p>
+
+        <sect2>Performance
+<p>
+          While PCI Rage 128 based cards are supported, they do not yet
+          support PCI GART, so they will not perform as well as their
+          AGP counterparts.
+
+        <sect2>Known Problems
+<p>
+          DGA is not yet supported in the ATI Rage 128 X server.  This
+          feature will be added in a future release.
+
+
       <sect1>3DLabs Oxygen GMX 2000
 <p>
           The driver for this hardware was experimental and is no longer being
@@ -621,6 +836,50 @@
           which will be addressed in the future.
 
 
+      <sect1>libGL.so and dlopen()
+<p>
+          A number of popular OpenGL applications on Linux (such as Quake3,
+          HereticII, Heavy Gear 2, etc) dynamically open the libGL.so
+          library at runtime with dlopen(), rather than linking with -lGL
+          at compile/link time.
+<p>
+          If dynamic loading of libGL.so is not implemented carefully, there
+          can be a number of serious problems.
+          Here are the things to be careful of in your application:
+          <itemize>
+          <item>Specify the RTLD_GLOBAL flag to dlopen().
+            If you don't do this then you'll likely see a runtime error message
+            complaining that _glapi_Context is undefined when libGL.so
+            tries to open a hardware-specific driver.
+            Without this flag, <em>nested</em> opening of dynamic libraries
+            does not work.
+          <item>Do not close the library with dlclose() until after
+            XCloseDisplay() has been called.
+            When libGL.so initializes itself it registers several callbacks
+            functions with Xlib.
+            When XCloseDisplay() is called those callback functions are
+            called.
+            If libGL.so has already been unloaded with dlclose() this will
+            cause a segmentation fault.
+          <item>
+            Your application should link with -lpthread.
+            On Linux, libGL.so uses the pthreads library in order to provide
+            thread safety.
+            There is apparently a bug in the dlopen()/dlclose() code which
+            causes crashes if the library uses pthreads but the parent
+            application doesn't.
+            The only known work-around is to link the application with
+            -lpthread.
+          </itemize>
+
+          Some applications don't yet incorporate these procedures and
+          may fail.
+          For example, changing the graphics settings in some video games
+          will expose this problem.
+          The DRI developers are working with game vendors to prevent this
+          problem in the future.
+
+
       <sect1>Bug Database
 <p>
           The DRI bug database which includes bugs related to specific
@@ -654,6 +913,9 @@
           <item>Visit the <htmlurl url="http://dri.sourceforge.net"
             name="DRI project on SourceForge.net"> for the latest development
             news about the DRI and 3D drivers.
+          <item>The <htmlurl url="http://dri.sourceforge.net/DRIcompile.html"
+            name="DRI Compilation Guide"> explains how to download, compile
+            and install the DRI for yourself.
         </itemize>
 
       <sect1>Support
@@ -664,54 +926,13 @@
             <htmlurl url="http://sourceforge.net/mail/?group_id=387"
             name="SourceForge"> is a forum for people to discuss DRI problems.
           <item>
-            XXX IHV support?
-          <item>
-            XXX Linux distro support?
+            In the future there may be IHV and Linux vendor support resources
+            for the DRI.
         </itemize>
 
   </article>
 
 
-<!--
-1. Introduction
-2. Supported Hardware
-3. Prerequisite Software
-4. Start-Up
-5. Using 3D Acceleration
-   5.1 libGL.so
-   5.2 glxinfo
-   5.3 Environment variables
-6. General Trouble Shooting
-7. Hardware-Specific Information
-   7.1 3dfx
-   7.2 Intel i810
-   7.3 Matrox G400
-   7.4 ATI Rage 128
-   7.5 3DLabs Gamma
-8. Limitation and Known Bugs
-   8.1 OpenGL
-   8.2 GLX
-   8.3 Signal Handling
-   8.4 Scheduling
-   8.5 Bug Database
-9. Resources
-   9.1 Software Resources
-       GLU
-       GLUT
-       Glide3
-       Utilities and demos
-       Sample XF86Config files
-   9.2 Documentation
-       www.XFree86.org
-       www.opengl.org
-       www.linux.com
-       www.precisioninsight.com
-       sourceforge.net
-   9.3 Support
-       IHVs
-       RedHat
- -->
-
   <!-- Local Variables: -->
   <!-- fill-column: 72  -->
   <!-- End:             -->
diff --git a/xc/programs/Xserver/hw/xfree86/doc/sgml/DRIcomp.sgml b/xc/programs/Xserver/hw/xfree86/doc/sgml/DRIcomp.sgml
new file mode 100644
index 000000000..5d0888342
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/doc/sgml/DRIcomp.sgml
@@ -0,0 +1,501 @@
+<!doctype linuxdoc system>
+<!-- Created: Sun Mar 12 13:00:00 2000 by brian@precisioninsight.com -->
+<!-- Revised: Fri May 19 09:36:02 2000 by martin@valinux.com -->
+<!-- $Id: DRIcomp.sgml,v 1.6.14.1 2000/06/07 16:06:45 brianp Exp $ -->
+
+  <article>
+
+      <title>DRI Compilation Guide
+      <author>
+          <htmlurl url="http://www.precisioninsight.com/"
+            name="Precision Insight, Inc.">
+      <date>18 May 2000
+        
+    <ident>
+    $XFree86: xc/programs/Xserver/hw/xfree86/doc/sgml/DRI.sgml,v 1.3 2000/04/05 05:38:41 brianp Exp $
+    </ident>
+
+    <toc>
+
+    <sect>Preamble
+
+      <sect1>Copyright
+        <p>
+          <bf>Copyright &copy; 2000 by Precision Insight, Inc.,
+          Cedar Park, Texas.
+          All Rights Reserved.</bf>
+        <p>
+          <bf>Permission is granted to make and distribute verbatim copies 
+          of this document provided the copyright notice and this permission
+          notice are preserved on all copies.</bf>
+          
+      <sect1>Trademarks
+        <p>
+          OpenGL is a registered trademark and SGI is a trademark of
+          Silicon Graphics, Inc.
+          Unix is a registered trademark of The Open Group.
+          The `X' device and X Window System are trademarks of The Open Group.
+          XFree86 is a trademark of The XFree86 Project.
+          Linux is a registered trademark of Linus Torvalds.
+          Intel is a registered trademark of Intel Corporation.
+          3Dlabs, GLINT, and Oxygen are either registered trademarks or
+          trademarks of 3Dlabs Inc. Ltd.
+          3dfx, Voodoo3, Voodoo4, and Voodoo5 are registered trademarks of
+          3dfx Interactive, Incorporated.
+	  Matrox is a registered trademark of Matrox Electronic Systems Ltd.
+	  ATI Rage is a registered trademark of ATI Technologies, Inc.
+          All other trademarks mentioned are the property of their
+          respective owners.
+
+
+    <sect>Introduction
+      <p>
+        This document describes how to download, compile and install the
+        DRI project.
+        This information is intended for experienced Linux developers.
+        Beginners are probably better off installing precompiled packages.
+      <p>
+        Edits, corrections and updates to this document may be mailed
+        to brian@precisioninsight.com.
+
+
+    <sect>Prerequisites
+      <p>
+        You'll need the following:
+        <itemize>
+          <item>At least 400MB of free disk space.  More is needed if you want
+            to build with debugging information or keep several build trees.
+          <item>A fast system.  Using a PIII-550 it takes about 1/2 hour to
+            build everthing.
+          <item>GCC compiler and related tools.
+          <item>ssh (secure shell) for registed developer downloading of the
+            DRI source tree
+        </itemize>
+
+      <p>
+        For 3dfx Voodoo3 hardware, you'll also need:
+        <itemize>
+          <item>Glide3x headers and runtime library if you want to use the
+            3dfx driver.
+            These can be obtained from
+            <htmlurl url="http://linux.3dfx.com/open_source/glide_kit.htm"
+            name="linux.3dfx.com">.
+          <item>Linux kernel 2.2.x.
+            The DRI developers have been using stock RedHat 6.1 systems
+            (kernel 2.2.12-20).
+            Later kernel versions will be supported in the future.
+        </itemize>
+
+      <p>
+        For Matrox G400 hardware, you'll also need:
+        <itemize>
+          <item>Linux kernel 2.3.51.  Other kernel versions may work but
+            this one is known to work.
+        </itemize>
+
+      <p>
+        For ATI Rage hardware, you'll also need:
+        <itemize>
+          <item>Linux kernel 2.3.51.  Other kernel versions may work but
+            this one is known to work.
+        </itemize>
+
+
+    <sect>Kernel preparation
+      <p>
+        You may have to upgrade your Linux kernel in order to use the DRI.
+        This is because you need a kernel version which supports AGP.
+        Building a new Linux kernel can be difficult for beginners but
+        there are resouces on the Internet to help.
+        This document assumes experience with configuring, building and
+        installing Linux kernels.
+      <p>
+        Linux kernels can be downloaded from
+        <htmlurl url="http://www.kernel.org/pub/linux/kernel/"
+        name="www.kernel.org">
+      <p>
+        Download the needed kernel and put it in /usr/src.
+        Create a directory for the source and unpack it.
+        For example:
+            <verb>
+            cd /usr/src
+	    rm -f linux
+            mkdir linux-2.3.51
+            ln -s linux-2.3.51 linux
+            bzcat linux-2.3.51.tar.bz2 | tar xf -
+            </verb>
+      <p>
+        Now configure your kernel.
+        You might, for example, use <tt>make menuconfig</tt> and do the
+        following:
+
+        <itemize>
+        <item>Go to <em>Code maturity level options</em>
+	<item>Enable <em>Prompt for development and/or incomplete
+          code/drivers</em>
+	<item>hit ESC
+	<item>Go to <em>Character devices</em>
+	<item>Disable <em>Direct Rendering Manager (XFree86 DRI support)</em>
+          since we'll use the DRI module from the XFree86/DRI tree.
+        <item>Go to <em>/dev/agpgart (AGP Support) (EXPERIMENTAL) (NEW)</em>
+        <item>Hit SPACE twice
+	<item>Enable all chipsets' support for AGP
+	</itemize>
+
+      <p>It's recommended that you turn on MTRRs under <em>Processor type
+        and Features</em>, but not required.
+
+      <p>
+        Configure the rest of the kernel as required for your system
+        (i.e. ethernet, scsi, etc)
+      <p>
+        Exit, saving your kernel configuration.
+      <p>
+        Edit your /etc/lilo.conf file.
+	Make sure you have an image entry as follows (or similar):
+            <verb>
+            image=/boot/vmlinuz
+                  label=linux.2.3.51
+                  read-only
+                  root=/dev/hda1
+            </verb>
+      <p>
+	The important part is that you have /boot/vmlinuz without a
+	trailing version number.
+	If this is the first entry in your /etc/lilo.conf AND you
+	haven't set a default, then this will be your default kernel.
+
+      <p>
+        Now compile the new kernel:
+
+            <verb>
+            cd /usr/src/linux-2.3.51
+            make dep ; make bzImage
+            make modules ; make modules_install
+            make install
+            </verb>
+	Note that the final part, make install, will automatically run
+	lilo for you.
+      <p>
+        Now reboot to use this new kernel.
+
+
+    <sect>Downloading the XFree86/DRI CVS Sources
+      <p>
+        The DRI project is hosted by VA Linux Systems'
+        <htmlurl url="http://sourceforge.net/project/?group_id=387"
+        name="SourceForge">.
+        The DRI source code, which is a subset of the XFree86 source tree,
+        is kept in a CVS repository there.
+      <p>
+        The DRI CVS sources may be accessed either anonymously or as a
+        registered SourceForge user.
+        It's recommended that you become a registered SourceForge user so
+        that you may submit non-annonymous bug reports and can participate
+        in the mailing lists.
+
+      <sect1>Anonymous CVS download:
+      <p>
+        <enum>
+          <item>Create a directory to store the CVS files:
+            <p>
+            <verb>
+            cd ~
+            mkdir DRI-CVS
+            </verb>
+            You could put your CVS directory in a different place but we'll
+            use <tt>~/DRI-CVS/</tt> here.
+            <p>
+          <item>Check out the CVS sources:
+            <p>
+            <verb>
+            cd ~/DRI-CVS
+            cvs -d:pserver:anonymous@cvs.dri.sourceforge.net:/cvsroot/dri login
+              (hit ENTER when prompted for a password)
+            cvs -z3 -d:pserver:anonymous@cvs.dri.sourceforge.net:/cvsroot/dri co xc
+            </verb>
+            <p>
+            The -z3 flag causes compression to be used in order to reduce
+            the download time.
+        </enum>
+
+
+      <sect1>Registered CVS download:
+      <p>
+        <enum>
+          <item>Create a directory to store the CVS files:
+            <p>
+            <verb>
+            cd ~
+            mkdir DRI-CVS
+            </verb>
+            You could put your CVS directory in a different place but we'll
+            use <tt>~/DRI-CVS/</tt> here.
+            <p>
+          <item>Set the CVS_RSH environment variable:
+            <p>
+            <verb>
+            setenv CVS_RSH ssh      // if using csh or tcsh
+            export CVS_RSH=ssh      // if using sh or bash
+            </verb>
+          <item>Check out the CVS sources:
+            <p>
+            <verb>
+            cd ~/DRI-CVS
+            cvs -z3 -dYOURID@cvs.dri.sourceforge.net:/cvsroot/dri co xc
+            </verb>
+            Replace <tt>YOURID</tt> with your CVS login name.
+            You'll be prompted to enter your sourceforge password.
+            <p>
+            The -z3 flag causes compression to be used in order to reduce
+            the download time.
+        </enum>
+
+
+    <sect>Updating your CVS sources
+      <p>
+        In the future you'll want to occasionally update your local copy of
+        the DRI source code to get the latest changes.
+        This can be done with:
+        <verb>
+            cd ~/DRI-CVS
+            cvs -z3 update -dA xc
+        </verb>
+        The -d flag causes any new subdirectories to be created and -A causes
+        most recent trunc sources to be fetched, not branch sources.
+
+
+    <sect>Compiling the XFree86/DRI tree
+      <sect1>Make a build tree
+        <p>
+          Rather than placing object files and library files right in the
+          source tree, they're instead put into a parallel <em>build</em> tree.
+          The build tree is made with the <tt>lndir</tt> command:
+          <verb>
+            cd ~/DRI-CVS
+            ln -s xc XFree40
+            mkdir build
+            cd build
+            lndir -silent -ignorelinks ../XFree40
+          </verb>
+        <p>
+          The build tree will be populated with symbolic links which point
+          back into the CVS source tree.
+        <p>
+          Advanced users may have several build trees for compiling and
+          testing with different options.
+
+      <sect1>Edit the host.def file
+        <p>
+          The <tt>~/DRI-CVS/build/xc/config/cf/host.def</tt> file is used
+          to configure the XFree86 build process.
+          You can change it to customize your build options or make adjustments
+          for your particular system configuration
+        <p>
+          The default <tt>host.def</tt> file will look something like this:
+          <verb>
+            #define DefaultCCOptions -Wall
+            #define DefaultGcc2i386Opt -O2
+            #define LibraryCDebugFlags -O2
+            #define BuildServersOnly YES
+            #define XF86CardDrivers vga tdfx mga r128
+            #define LinuxDistribution LinuxRedHat
+            #define DefaultCCOptions -ansi GccWarningOptions -pipe
+            #define BuildXF86DRI YES
+            #define HasGlide3 YES
+            /* Optionally turn these on for debugging */
+            /* #define GlxBuiltInTdfx YES */
+            /* #define GlxBuiltInMga YES */
+            /* #define GlxBuiltInR128 YES */
+            /* #define DoLoadableServer NO */
+            #define SharedLibFont NO
+          </verb>
+          The <tt>ProjectRoot</tt> variable specifies where the XFree86 files
+          will be installed.
+          You probably don't want to use <tt>/usr/X11R6/</tt> because that
+          would overwrite your default X files.
+          The following is recommended:
+          <verb>
+            #define ProjectRoot /usr/XF86-main
+          </verb>
+        <p>
+          Especially note the <em>XF86CardDrivers</em> line to be sure your
+          driver is listed.
+        <p>
+          If you have 3dfx hardware be sure that the Glide 3x headers are
+          installed in <tt>/usr/include/glide3/</tt> and that the Glide 3x
+          library is installed at <tt>/usr/lib/libglide3x.so</tt>.
+        <p>
+          If you do not have 3dfx hardware comment out the <tt>HasGlide3</tt>
+          line in <tt>host.def</tt>.
+        <p>
+
+      <sect1>Compile
+        <p>
+          To compile the complete DRI tree:
+          <verb>
+            cd ~/DRI-CVS/build/xc/
+            make World >& World.LOG
+          </verb>
+          Or if you want to watch the compilation progress:
+          <verb>
+            cd ~/DRI-CVS/build/xc/
+            make World >& World.LOG &
+            tail -f World.LOG
+          </verb>
+          With the default compilation flags it's normal to get a lot of
+          warnings during compilation.
+        <p>
+          Building will take some time so you may want to go check your
+          email or visit slashdot.
+
+      <sect1>Check for compilation errors
+        <p>
+          Using your text editor, examine <tt>World.LOG</tt> for errors
+          by searching for the pattern <tt>***</tt>.
+        <p>
+          Verify that the DRI kernel module(s) for your system were built:
+          <verb>
+          cd ~/DRI-CVS/build/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel
+          ls
+          </verb>
+        <p>
+          For the 3dfx Voodoo, you should see <em>tdfx.o</em>.
+          For the Matrox G400, you should see <em>mga.o</em>.
+          For the ATI Rage 128, you should see <em>r128.o</em>.
+        <p>
+          If the DRI kernel module(s) failed to build you should verify
+          that you're using the right version of the Linux kernel.
+          The most recent kernels are not always supported.
+        <p>
+          If your build machine is running a different version of the kernel
+          than your target machine (i.e. 2.2.12-20 vs. 2.3.99-pre6), make will
+          select the wrong kernel headers. This can be fixed by explicitly
+          setting the value of <tt>TREE</tt>.
+          If the path to your kernel source is 
+          <tt>/bigdisk/linux-2.3.99-pre6</tt>,
+          <verb>
+          cd ~/DRI-CVS/build/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel
+          make TREE=/bigdisk/linux-2.3.99-pre6
+          </verb>
+          or alternatively, edit Makefile to include this change.
+        <p>
+          After fixing the errors, do <tt>make World</tt> again.
+          Later, you might just compile parts of the source tree but it's
+          important that the whole tree will build first.
+
+
+    <sect>Installing
+      <p>
+        After the DRI tree has been compiled you can install the XFree86
+        headers, libraries, programs, etc for testing.
+
+      <sect1>X Installation
+	<p>
+	  As mentioned above, the install directory is specified by the
+	  <tt>ProjectRoot</tt> variable in the <tt>host.def</tt> file.
+	  Create that directory now if it doesn't already exist:
+	  <verb>
+	      mkdir /usr/XF86-main
+	  </verb>
+	<p>
+	  You'll have to change to root since the install process puts
+	  several files in <tt>/etc/X11/</tt> and sets the setuid flag on
+	  the X server executable.
+	  <verb>
+	      cd ~/DRI-CVS/build/xc
+	      su
+	      make install
+	  </verb>
+	  Edit your <tt>/etc/ld.so.conf</tt> file and put
+	  <tt>/usr/XF86-main/lib</tt> as the first line.
+	  Continue with:
+	  <verb>
+	      ldconfig
+	      exit
+	  </verb>
+	<p>
+	  Look in <tt>/usr/XF86-main</tt> to be sure the files installed
+          there.
+	<P>
+	  Strictly speaking, installing the DRI tree isn't required.
+	  It's possible to run and test the X server directly from the
+	  build directory but it's a bit error prone.
+
+      <sect1>Update Locale Information
+        <p>
+          To update your X locale information do the following:
+          <verb>
+             cd ~/DRI-CVS/build/xc/nls
+             xmkmf -a
+             make
+             make install
+          </verb>
+          This will prevent a locale error message from being printed
+          when you run Xlib programs.
+
+    <sect>X Server Configuration
+      <p>
+        If your X server is currently running you'll have to stop it
+        and return to a virtual terminal.
+      <p>
+        First, setup your XF86Config file.
+        It should load the GLX and DRI modules and specify the driver to
+        use for your hardware.
+        See the <htmlurl url="http://dri.sourceforge.net/DRIuserguide.html"
+        name="DRI User Guide"> for detailed information.
+      <p>
+        You may want to make a backup copy of your existing
+        <tt>/etc/X11/XF86Config</tt> file first.
+      <p>
+        It's very important that you set the <tt>ModulePath</tt> option to
+        point to your installation directory:
+        <verb>
+            ModulePath "/usr/XF86-main/lib/modules"
+        </verb>
+      <p>
+        Double check with this:
+        <verb>
+            grep ModulePath /etc/X11/XF86Config
+        </verb>
+      <p>
+        Next, your <tt>~/.xinitrc</tt> file controls which clients will be
+        launched when your X server starts.
+        You might put the following in yours:
+        <verb>
+            xset b off
+            xsetroot -solid "#004070"
+            xmodmap -e "clear mod4"
+            xrdb -merge ~/.Xdefaults
+            xterm -geometry +0+0 &
+            xterm -geometry +512+0 &
+            fvwm
+        </verb>
+      <p>
+
+    <sect>X Server Start-up
+      <p>
+        The X server can be started with:
+      <p>
+        <verb>
+            xinit -- /usr/XF86-main/bin/XFree86
+        </verb>
+      <p>
+        Automatic loading of DRM kernel modules was added to the X server
+        in XFree86 4.0.1.
+        This feature, and manual loading of kernel modules, is documented
+        in the DRI user guide.
+      <p>
+        At this point your X server should be up and running with
+        hardware-accelerated direct rendering.
+        Please read the
+        <htmlurl url="http://dri.sourceforge.net/DRIuserguide.html"
+        name="DRI User Guide"> for trouble shooting information.
+
+
+  </article>
+
+
+  <!-- Local Variables: -->
+  <!-- fill-column: 72  -->
+  <!-- End:             -->
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/glint/Imakefile b/xc/programs/Xserver/hw/xfree86/drivers/glint/Imakefile
index c8624e19f..d72589c49 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/glint/Imakefile
+++ b/xc/programs/Xserver/hw/xfree86/drivers/glint/Imakefile
@@ -14,10 +14,12 @@ DRIINCLUDES=-I$(SERVERSRC)/GL/dri -I$(LIBSRC)/GL/dri
 
 SRCS = glint_driver.c pm2_dac.c pm2ramdac.c pm2_accel.c pm_dac.c IBMramdac.c \
        pm_accel.c tx_dac.c tx_accel.c pm2v_dac.c pm2vramdac.c pm2_video.c \
-       TIramdac.c dualmx_dac.c dualmx_accel.c glint_shadow.c $(DRISRC)
+       TIramdac.c dualmx_dac.c dualmx_accel.c glint_shadow.c $(DRISRC) \
+       glint_dga.c
 OBJS = glint_driver.o pm2_dac.o pm2ramdac.o pm2_accel.o pm_dac.o IBMramdac.o \
        pm_accel.o tx_dac.o tx_accel.o pm2v_dac.o pm2vramdac.o pm2_video.o \
-       TIramdac.o dualmx_dac.o dualmx_accel.o glint_shadow.o $(DRIOBJ)
+       TIramdac.o dualmx_dac.o dualmx_accel.o glint_shadow.o $(DRIOBJ) \
+       glint_dga.o
 
 #if defined(XF86DriverSDK)
 INCLUDES = -I. -I../../include
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/glint/dualmx_accel.c b/xc/programs/Xserver/hw/xfree86/drivers/glint/dualmx_accel.c
index a6d34a4bf..974ee3b4f 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/glint/dualmx_accel.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/glint/dualmx_accel.c
@@ -105,7 +105,6 @@ DualMXInitializeEngine(ScrnInfoPtr pScrn)
     GLINTPtr pGlint = GLINTPTR(pScrn);
 
     pGlint->rasterizerMode = RMMultiGLINT;
-    pGlint->pprod |= FBRM_ScanlineInt2;
 
     /* Initialize the Accelerator Engine to defaults */
 
@@ -120,8 +119,10 @@ DualMXInitializeEngine(ScrnInfoPtr pScrn)
     /* Make sure the rest of the register writes go to both MX's */
     GLINT_SLOW_WRITE_REG(3, BroadcastMask);
 
+    GLINT_SLOW_WRITE_REG(pGlint->rasterizerMode, RasterizerMode);
     GLINT_SLOW_WRITE_REG(UNIT_DISABLE, ScissorMode);
-    GLINT_SLOW_WRITE_REG(pGlint->pprod,	LBReadMode);
+    GLINT_SLOW_WRITE_REG(pGlint->pprod | LBRM_ScanlineInt2,	LBReadMode);
+    pGlint->pprod |= FBRM_ScanlineInt2;
     GLINT_SLOW_WRITE_REG(pGlint->pprod,	FBReadMode);
     GLINT_SLOW_WRITE_REG(0, dXSub);
     GLINT_SLOW_WRITE_REG(UNIT_DISABLE,	LBWriteMode);
@@ -149,7 +150,6 @@ DualMXInitializeEngine(ScrnInfoPtr pScrn)
     GLINT_SLOW_WRITE_REG(0x400,		FilterMode);
     GLINT_SLOW_WRITE_REG(0xffffffff,	FBHardwareWriteMask);
     GLINT_SLOW_WRITE_REG(0xffffffff,	FBSoftwareWriteMask);
-    GLINT_SLOW_WRITE_REG(pGlint->rasterizerMode, RasterizerMode);
     GLINT_SLOW_WRITE_REG(UNIT_DISABLE,	GLINTDepth);
     GLINT_SLOW_WRITE_REG(UNIT_DISABLE,	FBSourceOffset);
     GLINT_SLOW_WRITE_REG(UNIT_DISABLE,	FBPixelOffset);
@@ -172,6 +172,7 @@ DualMXInitializeEngine(ScrnInfoPtr pScrn)
 	    GLINT_SLOW_WRITE_REG(0x0,	PixelSize);
 	    break;
     }
+
     pGlint->ROP = 0xFF;
     pGlint->ClippingOn = FALSE;
     pGlint->startxsub = 0;
@@ -212,32 +213,34 @@ DualMXAccelInit(ScreenPtr pScreen)
     infoPtr->SetClippingRectangle = DualMXSetClippingRectangle;
     infoPtr->DisableClipping = DualMXDisableClipping;
     infoPtr->ClippingFlags = HARDWARE_CLIP_MONO_8x8_FILL |
+#if 0
 			     HARDWARE_CLIP_SCREEN_TO_SCREEN_COPY |
+#endif
 			     HARDWARE_CLIP_SOLID_FILL;
 
     infoPtr->SolidFillFlags = 0;
     infoPtr->SetupForSolidFill = DualMXSetupForFillRectSolid;
     infoPtr->SubsequentSolidFillRect = DualMXSubsequentFillRectSolid;
 
-    /*
-     * The following optimized routines are copied from tx_accel.c, 
-     * but haven't been ported to a dual MX board. 
-     */
-#ifdef NOT_DONE 
     infoPtr->SolidLineFlags = 0;
     infoPtr->PolySegmentThinSolidFlags = 0;
     infoPtr->PolylinesThinSolidFlags = 0;
     infoPtr->SetupForSolidLine = DualMXSetupForSolidLine;
     infoPtr->SubsequentSolidHorVertLine = DualMXSubsequentHorVertLine;
-    infoPtr->SubsequentSolidBresenhamLine = 
+    if (!(pScrn->overlayFlags & OVERLAY_8_32_PLANAR))
+    {
+        infoPtr->SubsequentSolidBresenhamLine = 
 					DualMXSubsequentSolidBresenhamLine;
+    }
     infoPtr->PolySegmentThinSolid = DualMXPolySegmentThinSolidWrapper;
     infoPtr->PolylinesThinSolid = DualMXPolylinesThinSolidWrapper;
 
+#if 0
     infoPtr->ScreenToScreenCopyFlags = NO_TRANSPARENCY |
 				       ONLY_LEFT_TO_RIGHT_BITBLT;
     infoPtr->SetupForScreenToScreenCopy = DualMXSetupForScreenToScreenCopy;
     infoPtr->SubsequentScreenToScreenCopy = DualMXSubsequentScreenToScreenCopy;
+#endif
 
     infoPtr->Mono8x8PatternFillFlags = HARDWARE_PATTERN_PROGRAMMED_ORIGIN |
 				       HARDWARE_PATTERN_SCREEN_ORIGIN |
@@ -246,6 +249,7 @@ DualMXAccelInit(ScreenPtr pScreen)
     infoPtr->SubsequentMono8x8PatternFillRect = 
 				    DualMXSubsequentMono8x8PatternFillRect;
 
+#if 0
     if (!pGlint->UsePCIRetry) {
         infoPtr->ScanlineCPUToScreenColorExpandFillFlags = 
 					       TRANSPARENCY_ONLY |
@@ -285,6 +289,7 @@ DualMXAccelInit(ScreenPtr pScreen)
         infoPtr->SubsequentCPUToScreenColorExpandFill = 
 				DualMXSubsequentCPUToScreenColorExpandFill;
     } 
+#endif
 
     infoPtr->ColorExpandRange = MAX_FIFO_ENTRIES;
 
@@ -299,7 +304,6 @@ DualMXAccelInit(ScreenPtr pScreen)
 					  pScrn->bitsPerPixel / 8);
 
     xf86InitFBManager(pScreen, &AvailFBArea);
-#endif /* NOT_DONE */
 
     return (XAAInit(pScreen, infoPtr));
 }
@@ -312,11 +316,6 @@ static void DualMXLoadCoord(
 ){
     GLINTPtr pGlint = GLINTPTR(pScrn);
     
-    /*
-     * Optimization from tx_accel.c where these values are cached on the
-     * host doesn't appear to work.  We definitely need to reload at least
-     * the StartXDom value.  I'll play it safe and reload them all.
-     */
     GLINT_WRITE_REG(w<<16, StartXSub);
     GLINT_WRITE_REG(x<<16,StartXDom);
     GLINT_WRITE_REG(y<<16,StartY);
@@ -686,14 +685,15 @@ DualMXSubsequentMono8x8PatternFillRect(
     if (pGlint->FrameBufferReadMode != -1) {
 	if (pGlint->ROP == GXcopy) {
 	  GLINT_WRITE_REG(pGlint->BackGroundColor, FBBlockColor);
-	  GLINT_WRITE_REG(PrimitiveTrapezoid | FastFillEnable,Render);
+	  span = 0;
 	} else {
   	  GLINT_WRITE_REG(pGlint->BackGroundColor, PatternRamData0);
-	  GLINT_WRITE_REG(2<<1|2<<4|patternx<<7|patterny<<12|ASM_InvertPattern |
+	  span = SpanOperation;
+	}
+	GLINT_WRITE_REG(2<<1|2<<4|patternx<<7|patterny<<12|ASM_InvertPattern |
 					UNIT_ENABLE, AreaStippleMode);
-	  GLINT_WRITE_REG(AreaStippleEnable | SpanOperation | FastFillEnable | 
+	GLINT_WRITE_REG(AreaStippleEnable | span | FastFillEnable | 
 						PrimitiveTrapezoid, Render);
-	}
     }
 
     if (pGlint->ROP == GXcopy) {
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/glint/dualmx_dac.c b/xc/programs/Xserver/hw/xfree86/drivers/glint/dualmx_dac.c
index a17ab1c8b..0a7728ded 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/glint/dualmx_dac.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/glint/dualmx_dac.c
@@ -454,8 +454,7 @@ DualMXInit(ScrnInfoPtr pScrn, DisplayModePtr mode)
     pReg->glintRegs[VTGVSyncStart >> 3] = temp2;
     pReg->glintRegs[VTGVBlankEnd >> 3] = mode->CrtcVTotal - mode->CrtcVDisplay;
 
-    pReg->glintRegs[VTGPolarity >> 3] = (((mode->Flags & V_PHSYNC) ? 0:2)<<2) |
-			     ((mode->Flags & V_PVSYNC) ? 0 : 2) | (0xb0);
+    pReg->glintRegs[VTGPolarity >> 3] = 0xba;
 
     pReg->glintRegs[VClkCtl >> 3] = 0; 
     pReg->glintRegs[VTGVGateStart >> 3] = pReg->glintRegs[VTGVBlankEnd>>3] - 1; 
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/glint/glint.h b/xc/programs/Xserver/hw/xfree86/drivers/glint/glint.h
index a84f7219a..9ef168854 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/glint/glint.h
+++ b/xc/programs/Xserver/hw/xfree86/drivers/glint/glint.h
@@ -105,6 +105,10 @@ typedef struct {
     unsigned char *     IOBaseVGA;
     unsigned char *	FbBase;
     long		FbMapSize;
+    DGAModePtr		DGAModes;
+    int			numDGAModes;
+    Bool		DGAactive;
+    int			DGAViewportStatus;
     Bool		DoubleBuffer;
     Bool		NoAccel;
     Bool		FBDev;
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_dga.c b/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_dga.c
new file mode 100644
index 000000000..f3aa7ebe5
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_dga.c
@@ -0,0 +1,294 @@
+/*
+ * Copyright 2000 by Alan Hourihane, Sychdyn, North Wales, UK.
+ *
+ * Permission to use, copy, modify, distribute, and sell this software and its
+ * documentation for any purpose is hereby granted without fee, provided that
+ * the above copyright notice appear in all copies and that both that
+ * copyright notice and this permission notice appear in supporting
+ * documentation, and that the name of Alan Hourihane not be used in
+ * advertising or publicity pertaining to distribution of the software without
+ * specific, written prior permission.  Alan Hourihane makes no representations
+ * about the suitability of this software for any purpose.  It is provided
+ * "as is" without express or implied warranty.
+ *
+ * ALAN HOURIHANE DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE,
+ * INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS, IN NO
+ * EVENT SHALL ALAN HOURIHANE BE LIABLE FOR ANY SPECIAL, INDIRECT OR
+ * CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE,
+ * DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER
+ * TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+ * PERFORMANCE OF THIS SOFTWARE.
+ *
+ * Authors:  Alan Hourihane, <alanh@fairlite.demon.co.uk>
+ */
+/* $XFree86: $ */
+
+#include "xf86.h"
+#include "xf86_OSproc.h"
+#include "xf86_ansic.h"
+#include "xf86Pci.h"
+#include "xf86PciInfo.h"
+#include "xaa.h"
+#include "xaalocal.h"
+#include "glint.h"
+#include "glint_regs.h"
+#include "dgaproc.h"
+
+static Bool GLINT_OpenFramebuffer(ScrnInfoPtr, char **, unsigned char **, 
+					int *, int *, int *);
+static Bool GLINT_SetMode(ScrnInfoPtr, DGAModePtr);
+static void GLINT_Sync(ScrnInfoPtr);
+static int  GLINT_GetViewport(ScrnInfoPtr);
+static void GLINT_SetViewport(ScrnInfoPtr, int, int, int);
+static void GLINT_FillRect(ScrnInfoPtr, int, int, int, int, unsigned long);
+static void GLINT_BlitRect(ScrnInfoPtr, int, int, int, int, int, int);
+static void GLINT_BlitTransRect(ScrnInfoPtr, int, int, int, int, int, int, 
+					unsigned long);
+
+static
+DGAFunctionRec GLINTDGAFuncs = {
+   GLINT_OpenFramebuffer,
+   NULL,
+   GLINT_SetMode,
+   GLINT_SetViewport,
+   GLINT_GetViewport,
+   GLINT_Sync,
+   GLINT_FillRect,
+   NULL,
+#if 0
+   GLINT_BlitRect,
+   GLINT_BlitTransRect
+#else
+   NULL
+#endif
+};
+
+Bool
+GLINTDGAInit(ScreenPtr pScreen)
+{   
+   ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
+   GLINTPtr pGlint = GLINTPTR(pScrn);
+   DGAModePtr modes = NULL, newmodes = NULL, currentMode;
+   DisplayModePtr pMode, firstMode;
+   int Bpp = pScrn->bitsPerPixel >> 3;
+   int num = 0;
+   Bool oneMore;
+
+   pMode = firstMode = pScrn->modes;
+
+   while(pMode) {
+
+	if(0 /*pScrn->displayWidth != pMode->HDisplay*/) {
+	    newmodes = xrealloc(modes, (num + 2) * sizeof(DGAModeRec));
+	    oneMore = TRUE;
+	} else {
+	    newmodes = xrealloc(modes, (num + 1) * sizeof(DGAModeRec));
+	    oneMore = FALSE;
+	}
+
+	if(!newmodes) {
+	   xfree(modes);
+	   return FALSE;
+	}
+	modes = newmodes;
+
+SECOND_PASS:
+
+	currentMode = modes + num;
+	num++;
+
+	currentMode->mode = pMode;
+	currentMode->flags = DGA_CONCURRENT_ACCESS /*| DGA_PIXMAP_AVAILABLE*/;
+	if(!pGlint->NoAccel)
+	   currentMode->flags |= DGA_FILL_RECT /*| DGA_BLIT_RECT*/;
+	if(pMode->Flags & V_DBLSCAN)
+	   currentMode->flags |= DGA_DOUBLESCAN;
+	if(pMode->Flags & V_INTERLACE)
+	   currentMode->flags |= DGA_INTERLACED;
+	currentMode->byteOrder = pScrn->imageByteOrder;
+	currentMode->depth = pScrn->depth;
+	currentMode->bitsPerPixel = pScrn->bitsPerPixel;
+	currentMode->red_mask = pScrn->mask.red;
+	currentMode->green_mask = pScrn->mask.green;
+	currentMode->blue_mask = pScrn->mask.blue;
+	currentMode->visualClass = (Bpp == 1) ? PseudoColor : TrueColor;
+	currentMode->viewportWidth = pMode->HDisplay;
+	currentMode->viewportHeight = pMode->VDisplay;
+	currentMode->xViewportStep = 1;
+	currentMode->yViewportStep = 1;
+	currentMode->viewportFlags = DGA_FLIP_RETRACE;
+	currentMode->offset = 0;
+	currentMode->address = pGlint->FbBase;
+
+	if(oneMore) { /* first one is narrow width */
+	    currentMode->bytesPerScanline = ((pMode->HDisplay * Bpp) + 3) & ~3L;
+	    currentMode->imageWidth = pMode->HDisplay;
+	    currentMode->imageHeight =  pMode->VDisplay;
+	    currentMode->pixmapWidth = currentMode->imageWidth;
+	    currentMode->pixmapHeight = currentMode->imageHeight;
+	    currentMode->maxViewportX = currentMode->imageWidth - 
+					currentMode->viewportWidth;
+	    /* this might need to get clamped to some maximum */
+	    currentMode->maxViewportY = currentMode->imageHeight -
+					currentMode->viewportHeight;
+	    oneMore = FALSE;
+	    goto SECOND_PASS;
+	} else {
+	    currentMode->bytesPerScanline = 
+			((pScrn->displayWidth * Bpp) + 3) & ~3L;
+	    currentMode->imageWidth = pScrn->displayWidth;
+	    currentMode->imageHeight =  pMode->VDisplay;
+	    currentMode->pixmapWidth = currentMode->imageWidth;
+	    currentMode->pixmapHeight = currentMode->imageHeight;
+	    currentMode->maxViewportX = currentMode->imageWidth - 
+					currentMode->viewportWidth;
+	    /* this might need to get clamped to some maximum */
+	    currentMode->maxViewportY = currentMode->imageHeight -
+					currentMode->viewportHeight;
+	}	    
+
+	pMode = pMode->next;
+	if(pMode == firstMode)
+	   break;
+   }
+
+   pGlint->numDGAModes = num;
+   pGlint->DGAModes = modes;
+
+   return DGAInit(pScreen, &GLINTDGAFuncs, modes, num);  
+}
+
+
+static Bool
+GLINT_SetMode(
+   ScrnInfoPtr pScrn,
+   DGAModePtr pMode
+){
+   static int OldDisplayWidth[MAXSCREENS];
+   int index = pScrn->pScreen->myNum;
+   GLINTPtr pGlint = GLINTPTR(pScrn);
+
+   if(!pMode) { /* restore the original mode */
+	/* put the ScreenParameters back */
+	
+	pScrn->displayWidth = OldDisplayWidth[index];
+	
+        GLINTSwitchMode(index, pScrn->currentMode, 0);
+	pGlint->DGAactive = FALSE;
+   } else {
+	if(!pGlint->DGAactive) {  /* save the old parameters */
+	    OldDisplayWidth[index] = pScrn->displayWidth;
+
+	    pGlint->DGAactive = TRUE;
+	}
+
+	pScrn->displayWidth = pMode->bytesPerScanline / 
+			      (pMode->bitsPerPixel >> 3);
+
+        GLINTSwitchMode(index, pMode->mode, 0);
+   }
+   
+   return TRUE;
+}
+
+static int  
+GLINT_GetViewport(
+  ScrnInfoPtr pScrn
+){
+    GLINTPtr pGlint = GLINTPTR(pScrn);
+
+    return pGlint->DGAViewportStatus;
+}
+
+static void 
+GLINT_SetViewport(
+   ScrnInfoPtr pScrn, 
+   int x, int y, 
+   int flags
+){
+   GLINTPtr pGlint = GLINTPTR(pScrn);
+
+   GLINTAdjustFrame(pScrn->pScreen->myNum, x, y, flags);
+   pGlint->DGAViewportStatus = 0;  /* GLINTAdjustFrame loops until finished */
+}
+
+static void 
+GLINT_FillRect (
+   ScrnInfoPtr pScrn, 
+   int x, int y, int w, int h, 
+   unsigned long color
+){
+    GLINTPtr pGlint = GLINTPTR(pScrn);
+
+    if(pGlint->AccelInfoRec) {
+	(*pGlint->AccelInfoRec->SetupForSolidFill)(pScrn, color, GXcopy, ~0);
+	(*pGlint->AccelInfoRec->SubsequentSolidFillRect)(pScrn, x, y, w, h);
+	SET_SYNC_FLAG(pGlint->AccelInfoRec);
+    }
+}
+
+static void 
+GLINT_Sync(
+   ScrnInfoPtr pScrn
+){
+    GLINTPtr pGlint = GLINTPTR(pScrn);
+
+    if(pGlint->AccelInfoRec) {
+	(*pGlint->AccelInfoRec->Sync)(pScrn);
+    }
+}
+
+static void 
+GLINT_BlitRect(
+   ScrnInfoPtr pScrn, 
+   int srcx, int srcy, 
+   int w, int h, 
+   int dstx, int dsty
+){
+    GLINTPtr pGlint = GLINTPTR(pScrn);
+
+    if(pGlint->AccelInfoRec) {
+	int xdir = ((srcx < dstx) && (srcy == dsty)) ? -1 : 1;
+	int ydir = (srcy < dsty) ? -1 : 1;
+
+	(*pGlint->AccelInfoRec->SetupForScreenToScreenCopy)(
+		pScrn, xdir, ydir, GXcopy, ~0, -1);
+	(*pGlint->AccelInfoRec->SubsequentScreenToScreenCopy)(
+		pScrn, srcx, srcy, dstx, dsty, w, h);
+	SET_SYNC_FLAG(pGlint->AccelInfoRec);
+    }
+}
+
+
+static void 
+GLINT_BlitTransRect(
+   ScrnInfoPtr pScrn, 
+   int srcx, int srcy, 
+   int w, int h, 
+   int dstx, int dsty,
+   unsigned long color
+){
+  /* this one should be separate since the XAA function would
+     prohibit usage of ~0 as the key */
+}
+
+
+static Bool 
+GLINT_OpenFramebuffer(
+   ScrnInfoPtr pScrn, 
+   char **name,
+   unsigned char **mem,
+   int *size,
+   int *offset,
+   int *flags
+){
+    GLINTPtr pGlint = GLINTPTR(pScrn);
+
+    *name = NULL; 		/* no special device */
+    *mem = (unsigned char*)pGlint->FbAddress;
+    *size = pGlint->FbMapSize;
+    *offset = 0;
+    *flags = DGA_NEED_ROOT;
+
+    return TRUE;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_dri.c b/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_dri.c
index 0913cada0..9aa55de42 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_dri.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_dri.c
@@ -40,11 +40,7 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 #include "xf86PciInfo.h"
 #include "xf86Pci.h"
-#define PSZ 8
-#include "cfb.h"
-#undef PSZ
-#include "cfb16.h"
-#include "cfb32.h"
+#include "fb.h"
 
 #include "miline.h"
 
@@ -61,18 +57,7 @@ static void GLINTDestroyContext(ScreenPtr pScreen, drmContext hwContext,
                                 DRIContextType contextStore);
 
 static int 
-GLINTDRIControlInitSingleMX(int drmSubFD, int irq)
-{
-    
-    int retcode;
-
-				/* Perhaps add flag to for single mx here? */
-    if ((retcode = drmCtlInstHandler(drmSubFD, irq))) return 1;
-    return 0;
-}
-
-static int 
-GLINTDRIControlInitDualMX(int drmSubFD, int irq)
+GLINTDRIControlInit(int drmSubFD, int irq)
 {
     int retcode;
 
@@ -105,7 +90,7 @@ GLINTInitVisualConfigs(ScreenPtr pScreen)
     case 32:
 	/* if(pGlint->Overlay): differentiate overlays when we support
 	   either alpha buffer or 3D rendering in Overlay */
-	numConfigs = 5;
+	numConfigs = 4;
 
 	if (!(pConfigs = (__GLXvisualConfig *)xnfcalloc(
 					    sizeof(__GLXvisualConfig),
@@ -260,6 +245,7 @@ GLINTInitVisualConfigs(ScreenPtr pScreen)
 	pConfigs[3].transparentIndex = 0;
 	pGlintConfigs[3].index = 3;
 
+#if 0
 	/* config 4: db=TRUE, depth=16, stencil=8, conformant (a lie) */ 
 	pConfigs[4].vid              = -1;
 	pConfigs[4].class            = -1;
@@ -291,6 +277,7 @@ GLINTInitVisualConfigs(ScreenPtr pScreen)
 	pConfigs[4].transparentAlpha = 0;
 	pConfigs[4].transparentIndex = 0;
 	pGlintConfigs[4].index = 4;
+#endif
 
 	break;
     }
@@ -424,6 +411,7 @@ GLINTDRIScreenInit(ScreenPtr pScreen)
 	DRIDestroyInfoRec(pGlint->pDRIInfo);
 	return FALSE;
     }
+
     pDRIInfo->devPrivate     = pGlintDRI;
     pDRIInfo->devPrivateSize = sizeof(GLINTDRIRec);
     pDRIInfo->contextSize    = sizeof(GLINTDRIContextRec);
@@ -463,6 +451,11 @@ GLINTDRIScreenInit(ScreenPtr pScreen)
         }
     }
 
+    /* Tell the client driver how many MX's we have */
+    pGlintDRI->numMXDevices = pGlint->numMXDevices;
+    /* Tell the client about our screen size setup */
+    pGlintDRI->pprod = pGlint->pprod;
+   
     /* setup device specific direct rendering memory maps */
 
     /* pci region 0: control regs, first 4k page, priveledged writes */
@@ -543,8 +536,8 @@ GLINTDRIScreenInit(ScreenPtr pScreen)
 	    if ((bufs = drmAddBufs(pGlint->drmSubFD,
 				   xf86ConfigDRI.bufs[i].count,
 				   xf86ConfigDRI.bufs[i].size,
-				   0, /* flags */
-				   0  /* offset */)) <= 0) {
+				   0,
+				   0 /* flags */)) <= 0) {
 		xf86DrvMsg(pScrn->scrnIndex, X_ERROR, 
 			   "[drm] failure adding %d %d byte DMA buffers\n",
 			   xf86ConfigDRI.bufs[i].count,
@@ -564,8 +557,8 @@ GLINTDRIScreenInit(ScreenPtr pScreen)
 	if ((bufs = drmAddBufs(pGlint->drmSubFD,
 			       GLINT_DRI_BUF_COUNT,
 			       GLINT_DRI_BUF_SIZE,
-			       0, /* flags */
-			       0  /* offset */)) <= 0) {
+			       0,
+			       0 /* flags */)) <= 0) {
 	    xf86DrvMsg(pScrn->scrnIndex, X_INFO,
 		       "[drm] failure adding %d %d byte DMA buffers\n",
 		       GLINT_DRI_BUF_COUNT,
@@ -602,26 +595,13 @@ GLINTDRIScreenInit(ScreenPtr pScreen)
 						->thisCard)->funcnum);
     }
     
-    if (pGlint->numMXDevices == 2) {
-    	if ( (pGlint->irq <= 0) || 
-	      GLINTDRIControlInitDualMX(pGlint->drmSubFD, pGlint->irq) ) {
-	    xf86DrvMsg(pScrn->scrnIndex, X_INFO,
-		   "[drm] cannot initialize dma with IRQ %d\n",
-		   pGlint->irq);
-	    DRICloseScreen(pScreen);
-	    return FALSE;
-	}
-    }
-
-    if (pGlint->numMXDevices == 1) {
-    	if ( (pGlint->irq <= 0) || 
-	      GLINTDRIControlInitSingleMX(pGlint->drmSubFD, pGlint->irq) ) {
-	    xf86DrvMsg(pScrn->scrnIndex, X_INFO,
-		   "[drm] cannot initialize dma with IRQ %d\n",
-		   pGlint->irq);
-	    DRICloseScreen(pScreen);
-	    return FALSE;
-	}
+   if ( (pGlint->irq <= 0) || 
+	      GLINTDRIControlInit(pGlint->drmSubFD, pGlint->irq) ) {
+	xf86DrvMsg(pScrn->scrnIndex, X_INFO,
+	   "[drm] cannot initialize dma with IRQ %d\n",
+	   pGlint->irq);
+	DRICloseScreen(pScreen);
+	return FALSE;
     }
 
     xf86DrvMsg(pScrn->scrnIndex, X_INFO,
@@ -670,7 +650,7 @@ GLINTCreateContext(ScreenPtr pScreen,
                    VisualPtr visual,
                    drmContext hwContext,
                    void *pVisualConfigPriv,
-                   DRIContextType context)
+		   DRIContextType contextStore)
 {
     ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
     GLINTPtr pGlint = GLINTPTR(pScrn);
@@ -1161,7 +1141,7 @@ GLINTDRISwapContext(
 
 	    /* send gamma the context dump command */
 	    GLINT_WAIT(3);
-            if (pGlint->numMXDevices > 1)
+            if (pGlint->numMXDevices == 2)
 	    	GLINT_WRITE_REG(1, BroadcastMask);
 	    GLINT_WRITE_REG(3<<14, FilterMode);  /* context bits on gamma */
 	    GLINT_WRITE_REG(GLINT_GAMMA_CONTEXT_MASK, ContextDump);
@@ -1206,7 +1186,7 @@ dumpIndex,readValue);
 
 	    /* send context restore command */
 	    GLINT_WAIT(1);
-            if (pGlint->numMXDevices > 1)
+            if (pGlint->numMXDevices == 2)
 	    	GLINT_WRITE_REG(1, BroadcastMask);
 
 	    GLINT_WAIT(3);
@@ -1235,7 +1215,7 @@ dumpIndex,pWC->Gamma[dumpIndex]);
 	    pGlint->AccelInfoRec->NeedToSync = TRUE;
 	    
 	    /* finally the MX portions */
-            if (pGlint->numMXDevices > 1)
+            if (pGlint->numMXDevices == 2)
 	    	GLINT_SLOW_WRITE_REG(1, BroadcastMask);
 	    GLINT_SLOW_WRITE_REG(pWC->MX1.CSStart,                  SStart);
 	    GLINT_SLOW_WRITE_REG(pWC->MX1.CdSdx,                    dSdx);
@@ -1488,7 +1468,7 @@ dumpIndex,pWC->Gamma[dumpIndex]);
 	/* restore the 2D portion of the new context */
 
 	/* Restore MX1's registers */
-        if (pGlint->numMXDevices > 1)
+        if (pGlint->numMXDevices == 2)
 	    GLINT_SLOW_WRITE_REG(1, BroadcastMask);
 	GLINT_SLOW_WRITE_REG(pWC->MX1.CStartXDom,               StartXDom);
 	GLINT_SLOW_WRITE_REG(pWC->MX1.CdXDom,                   dXDom);
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_dri.h b/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_dri.h
index 03f4d9fed..9e11edda8 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_dri.h
+++ b/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_dri.h
@@ -50,6 +50,8 @@ typedef struct {
     drmMapFlags 	flagsControlRegs1;
     drmMapFlags 	flagsControlRegs2;
     drmMapFlags		flagsControlRegs3;
+    int			numMXDevices;
+    int			pprod;
 } GLINTDRIRec, *GLINTDRIPtr;
 
 #define GLINT_DRI_BUF_COUNT 20
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_dripriv.h b/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_dripriv.h
index c92d90f98..e026f8b18 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_dripriv.h
+++ b/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_dripriv.h
@@ -42,15 +42,15 @@ extern void GlxSetVisualConfigs(
 extern Bool GLINTCreateContext(ScreenPtr pScreen,
                                VisualPtr visual,
                                drmContext hwContext,
-                               void *pVisualConfigPriv,
-                               DRIContextType context);
+                               void* pVisualConfigPriv,
+			       DRIContextType contextStore);
 
 extern void GLINTDRISwapContext( ScreenPtr pScreen,
 				 DRISyncType syncType,
 				 DRIContextType readContextType,
-				 void *readContextStore,
+				 void* readContextStore,
 				 DRIContextType writeContextType,
-				 void *writeContextStore);
+				 void* writeContextStore);
 
 /* Macros to Setup Generic Kernel Device Driver to Handle DMA for gamma */
 
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_driver.c b/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_driver.c
index edd418ae3..f0b316729 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_driver.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_driver.c
@@ -2658,6 +2658,9 @@ GLINTScreenInit(int scrnIndex, ScreenPtr pScreen, int argc, char **argv)
 
     xf86SetBlackWhitePixels(pScreen);
 
+    if (!pGlint->ShadowFB)
+	GLINTDGAInit(pScreen);
+
     if (pScrn->bitsPerPixel > 8) {
         /* Fixup RGB ordering */
         visual = pScreen->visuals + pScreen->numVisuals;
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_regs.h b/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_regs.h
index f33e42c80..489bf8eed 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_regs.h
+++ b/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_regs.h
@@ -35,6 +35,9 @@
 #define PCI_CHIP_3DLABS_PERMEDIA				0x04
 #define PCI_CHIP_3DLABS_MX					0x06
 #define PCI_CHIP_3DLABS_PERMEDIA2				0x07
+#define PCI_CHIP_3DLABS_GAMMA					0x08
+#define PCI_CHIP_3DLABS_PERMEDIA2V				0x09
+#define PCI_CHIP_3DLABS_PERMEDIA3				0x0A
 #define PCI_CHIP_TI_PERMEDIA 	  				0x3d04
 
 #define CFGRevisionId						0x08
@@ -90,6 +93,10 @@
 #define   SCLK_SEL_MCLK_HALF	(3 << 10)
 #define ByDMAControl							0x00D8
 
+/* GLINT R3 & Permedia3 Region 0 Bypass Controls */
+#define PM3ByAperture1Mode						0x0300
+#define PM3ByAperture2Mode						0x0328
+
 /* GLINT 500TX LocalBuffer Registers */
 #define LBMemoryCtl							0x1000
 #define   LBNumBanksMask	0x00000001
@@ -192,6 +199,17 @@
 #define PMFramebufferWriteMask					        0x1140
 #define PMCount								0x1180
 
+/* Permedia 3 & GLINT R3 Memory Control */
+#define PM3MemCounter							0x1000
+#define PM3MemBypassWriteMask						0x1008
+#define PM3MemScratch							0x1010
+#define PM3LocalMemCaps							0x1018
+#define PM3LocalMemTimings						0x1020
+#define PM3LocalMemControl						0x1028
+#define PM3LocalMemRefresh						0x1030
+#define PM3LocalMemPowerDown						0x1038
+#define PM3RemoteMemControl						0x1100
+
 /* Framebuffer Registers */
 #define FBMemoryCtl							0x1800
 #define FBModeSel							0x1808
@@ -1146,14 +1164,15 @@
 #define GLINT_READ_REG(r)					\
 	GLINT_VERB_READ_REG(pGlint,r,__FILE__,__LINE__)
 #else
-#define GLINT_WRITE_REG(v,r)					\
-do{								\
-	*(volatile CARD32 *)((char *)pGlint->IOBase+(r))=v;	\
-	mem_barrier();						\
-}while(0)
-#define GLINT_READ_REG(r)					\
-	(*(volatile CARD32 *)((char *)pGlint->IOBase+(r)))
-#endif                                        
+
+#define GLINT_WRITE_REG(v,r) MMIO_OUT32(pGlint->IOBase,(unsigned long)r, v)
+#define GLINT_READ_REG(r) MMIO_IN32(pGlint->IOBase,(unsigned long)r)
+#define GLINT_SECONDARY_WRITE_REG(v,r) \
+			MMIO_OUT32(pGlint->IOBase,(unsigned long)r+0x10000, v)
+#define GLINT_SECONDARY_READ_REG(r) \
+			MMIO_IN32(pGlint->IOBase,(unsigned long)r+0x10000)
+
+#endif /* DEBUG */
 
 #define GLINT_WAIT(n)						\
 do{								\
@@ -1179,23 +1198,12 @@ do{								\
 	GLINTDACDelay(5);					\
 }while(0)
 
-#define GLINT_SECONDARY_WRITE_REG(v,r)                          	\
-do{									\
-	*(volatile CARD32 *)((char *)pGlint->SecondaryBase+(r))=v;	\
-}while(0)
-
-#define GLINT_SECONDARY_READ_REG(r)                             	\
-	(*(volatile CARD32 *)((char *)pGlint->SecondaryBase+(r)))
-
 #define GLINT_SECONDARY_SLOW_WRITE_REG(v,r)				\
 do{									\
 	while(GLINT_READ_REG(InFIFOSpace)<1);				\
         GLINT_SECONDARY_WRITE_REG(v,r);					\
 }while(0)
 
-#define GLINT_LOAD_PAR(v,r) \
-        *(unsigned long *)((char*)pGlint->IOBase+r) = *((unsigned long *) &v)
-
 #define REPLICATE(r)						\
 {								\
 	if (pScrn->bitsPerPixel == 16) {			\
@@ -1208,14 +1216,21 @@ do{									\
 	}							\
 }
 
+#ifndef XF86DRI
 #define LOADROP(rop)						\
 {								\
-	if (pGlint->ROP != rop) {				\
+	if (pGlint->ROP != rop)	{				\
 		GLINT_WRITE_REG(rop<<1|UNIT_ENABLE, LogicalOpMode);	\
 		pGlint->ROP = rop;				\
 	}							\
 }
-
+#else
+#define LOADROP(rop) \
+	{				\
+		GLINT_WRITE_REG(rop<<1|UNIT_ENABLE, LogicalOpMode);	\
+		pGlint->ROP = rop;				\
+	}
+#endif
 	
 #define CHECKCLIPPING						\
 {								\
@@ -1226,6 +1241,7 @@ do{									\
 	}							\
 }
 
+#ifndef XF86DRI
 #define DO_PLANEMASK(planemask)					\
 { 								\
 	if (planemask != pGlint->planemask) {			\
@@ -1234,4 +1250,12 @@ do{									\
 		GLINT_WRITE_REG(planemask, FBHardwareWriteMask);\
 	}							\
 } 
+#else
+#define DO_PLANEMASK(planemask)					\
+	{							\
+		pGlint->planemask = planemask;			\
+		REPLICATE(planemask); 				\
+		GLINT_WRITE_REG(planemask, FBHardwareWriteMask);\
+	}
+#endif
 #endif
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/i810/Imakefile b/xc/programs/Xserver/hw/xfree86/drivers/i810/Imakefile
index 331396f38..9e167d6c7 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/i810/Imakefile
+++ b/xc/programs/Xserver/hw/xfree86/drivers/i810/Imakefile
@@ -6,13 +6,9 @@ XCOMM
 #define IHaveModules
 #include <Server.tmpl>
 
-#
-# Uncomment these to build with DRI support when available
-#
-#undef BuildXF86DRI
 #if BuildXF86DRI
-DRISRCS = i810_dri.c i810_drm.c
-DRIOBJS = i810_dri.o i810_drm.o
+DRISRCS = i810_dri.c
+DRIOBJS = i810_dri.o
 DRIINCLUDES = -I$(SERVERSRC)/GL/dri -I$(LIBSRC)/GL/dri \
 	-I$(XF86OSSRC)/linux/drm/kernel
 DRIDEFINES = $(GLX_DEFINES)
@@ -63,7 +59,6 @@ InstallDriverSDKNonExecFile(i810_cursor.c,$(DRIVERSDKDIR)/drivers/i810)
 InstallDriverSDKNonExecFile(i810_driver.c,$(DRIVERSDKDIR)/drivers/i810)
 InstallDriverSDKNonExecFile(i810_dri.c,$(DRIVERSDKDIR)/drivers/i810)
 InstallDriverSDKNonExecFile(i810_dri.h,$(DRIVERSDKDIR)/drivers/i810)
-InstallDriverSDKNonExecFile(i810_dripriv.h,$(DRIVERSDKDIR)/drivers/i810)
 InstallDriverSDKNonExecFile(i810_io.c,$(DRIVERSDKDIR)/drivers/i810)
 InstallDriverSDKNonExecFile(i810_memory.c,$(DRIVERSDKDIR)/drivers/i810)
 InstallDriverSDKNonExecFile(i810_wmark.c,$(DRIVERSDKDIR)/drivers/i810)
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/i810/i810.h b/xc/programs/Xserver/hw/xfree86/drivers/i810/i810.h
index a37ddca72..030b66a91 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/i810/i810.h
+++ b/xc/programs/Xserver/hw/xfree86/drivers/i810/i810.h
@@ -43,9 +43,6 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "xaa.h"
 #include "xf86Cursor.h"
 
-#undef XF86DRI
-
-
 #ifdef XF86DRI
 #include "xf86drm.h"
 #include "sarea.h"
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_accel.c b/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_accel.c
index 0f67180d2..1a410bcd1 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_accel.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_accel.c
@@ -271,6 +271,7 @@ I810WaitLpRing( ScrnInfoPtr pScrn, int n, int timeout_millis )
 	    DRICloseScreen(screenInfo.screens[pScrn->scrnIndex]);
 	 }
 #endif
+	 pI810->AccelInfoRec = NULL; /* Stops recursive behavior */
 	 FatalError("lockup\n"); 
       }
 
@@ -301,13 +302,9 @@ I810Sync( ScrnInfoPtr pScrn )
 #ifdef XF86DRI
    /* VT switching tries to do this.  
     */
-   if (!pI810->LockHeld) {
+   if (!pI810->LockHeld && pI810->directRenderingEnabled) {
       return;   
    }
-
-
-/*     if (pI810->directRenderingEnabled)  */
-/*        DRIUnlockLockQueiscent( pScrn->pScreen ); */
 #endif
 
    /* Send a flush instruction and then wait till the ring is empty.
@@ -635,19 +632,19 @@ I810RefreshRing(ScrnInfoPtr pScrn)
 }
 
 
-
+/* Emit on gaining VT?
+ */
 void 
 I810EmitInvarientState(ScrnInfoPtr pScrn)
 {
    I810Ptr pI810 = I810PTR(pScrn);
-   BEGIN_LP_RING( 8 );
+   BEGIN_LP_RING( 10 );
 
    OUT_RING( INST_PARSER_CLIENT | INST_OP_FLUSH | INST_FLUSH_MAP_CACHE );
    OUT_RING( GFX_CMD_CONTEXT_SEL | CS_UPDATE_USE | CS_USE_CTX0 );
    OUT_RING( INST_PARSER_CLIENT | INST_OP_FLUSH | INST_FLUSH_MAP_CACHE);
    OUT_RING( 0 );
 
-
    OUT_RING( GFX_OP_COLOR_CHROMA_KEY );
    OUT_RING( CC1_UPDATE_KILL_WRITE | 
 	     CC1_DISABLE_KILL_WRITE | 
@@ -658,5 +655,8 @@ I810EmitInvarientState(ScrnInfoPtr pScrn)
    OUT_RING( 0 );
    OUT_RING( 0 );
 
+/*     OUT_RING( CMD_OP_Z_BUFFER_INFO ); */
+/*     OUT_RING( pI810->DepthBuffer.Start | pI810->auxPitchBits); */
+
    ADVANCE_LP_RING();      
 }
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_dri.c b/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_dri.c
index 7de2d1d7a..e2c48f68f 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_dri.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_dri.c
@@ -73,13 +73,26 @@ Bool I810InitDma(ScrnInfoPtr pScrn)
 {
    I810Ptr pI810 = I810PTR(pScrn);
    I810RingBuffer *ring = &(pI810->LpRing);
+   drmI810Init info;
    Bool ret_val;
    
-   ret_val = drmI810InitDma(pI810->drmSubFD, ring->mem.Start,
-			    ring->mem.End, ring->mem.Size,
-			    6, 5, sizeof(XF86DRISAREARec));
-   if(ret_val == FALSE)
-      xf86DrvMsg(pScrn->scrnIndex, X_ERROR, "I810 Dma Initialization Failed\n");
+   info.start = ring->mem.Start;
+   info.end = ring->mem.End; 
+   info.size = ring->mem.Size;
+   info.ring_map_idx = 6; 
+   info.buffer_map_idx = 5; 
+   info.sarea_off = sizeof(XF86DRISAREARec);
+
+   info.front_offset = 0;
+   info.back_offset = pI810->BackBuffer.Start;
+   info.depth_offset = pI810->DepthBuffer.Start;
+   info.w = pScrn->virtualX;
+   info.h = pScrn->virtualY;
+   info.pitch = pI810->auxPitch;
+   info.pitch_bits = pI810->auxPitchBits;
+
+   ret_val = drmI810InitDma(pI810->drmSubFD, &info);
+   if(ret_val == FALSE) ErrorF("I810 Dma Initialization Failed\n");
    return ret_val;
 }
 
@@ -214,7 +227,7 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
    
 #if XFree86LOADER
    /* Check that the GLX, DRI, and DRM modules have been loaded by testing
-    * for canonical symbols in each module. */
+    * for known symbols in each module. */
    if (!LoaderSymbol("GlxSetVisualConfigs")) return FALSE;
    if (!LoaderSymbol("DRIScreenInit"))       return FALSE;
    if (!LoaderSymbol("drmAvailable"))        return FALSE;
@@ -248,6 +261,7 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
    pDRIInfo->drmDriverName = I810KernelDriverName;
    pDRIInfo->clientDriverName = I810ClientDriverName;
    pDRIInfo->busIdString = xalloc(64);
+
    sprintf(pDRIInfo->busIdString, "PCI:%d:%d:%d",
 	   ((pciConfigPtr)pI810->PciInfo->thisCard)->busnum,
 	   ((pciConfigPtr)pI810->PciInfo->thisCard)->devnum,
@@ -370,7 +384,6 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
 
    drmAgpAlloc(pI810->drmSubFD, 4096 * 1024, 1, NULL, &dcacheHandle);
    pI810->dcacheHandle = dcacheHandle;
-
    xf86DrvMsg(pScreen->myNum, X_INFO, "dcacheHandle : %p\n", dcacheHandle);
    
 #define Elements(x) sizeof(x)/sizeof(*x)
@@ -386,10 +399,7 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
    }
    else {
       back_size = i810_pitches[pitch_idx] * (pScrn->virtualY + 4);
-      xf86DrvMsg(pScreen->myNum, X_INFO, "back_size: %d\n", back_size);
-      /* Round to 4k */
       back_size = ((back_size + 4096 - 1) / 4096) * 4096;
-      xf86DrvMsg(pScreen->myNum, X_INFO, "back_size 2 : %d\n", back_size);
    }
    
    sysmem_size = pScrn->videoRam * 1024;
@@ -407,15 +417,18 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
    else {
       sysmem_size = sysmem_size - 2*back_size;
    }
-   xf86DrvMsg(pScreen->myNum, X_INFO, "sysmem_size : %d\n", sysmem_size);
-   
-   sysmem_size -= 4096;
-   if (sysmem_size > ((48*1024*1024) - 1) ) {
-      sysmem_size = (48*1024*1024) - (2*4096);
-      xf86DrvMsg(pScrn->scrnIndex, X_INFO, "User requested more memory then fits in the agp aperture\n"
+
+   if(sysmem_size > 48*1024*1024) {
+      sysmem_size = 48*1024*1024;
+
+      xf86DrvMsg(pScrn->scrnIndex, X_INFO, 
+		 "User requested more memory then fits in the agp aperture\n"
 		 "Truncating to %d bytes of memory\n",
 		 sysmem_size);
    }
+
+   sysmem_size -= 4096;		/* remove 4k for the hw cursor */
+
    pI810->SysMem.Start = 0;
    pI810->SysMem.Size = sysmem_size;
    pI810->SysMem.End = sysmem_size;
@@ -425,11 +438,7 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
 
    if (dcacheHandle != 0) {
       /* The Z buffer is always aligned to the 48 mb mark in the aperture */
-      xf86DrvMsg(pScreen->myNum, X_INFO,
-                 "dcacheHandle : %p, aligned to : %lx\n",
-                 dcacheHandle, 48*1024*1024);
-
-      if (drmAgpBind(pI810->drmSubFD, dcacheHandle, 48*1024*1024) == 0) {
+      if(drmAgpBind(pI810->drmSubFD, dcacheHandle, 48*1024*1024) == 0) {
 	 xf86memset (&pI810->DcacheMem, 0, sizeof(I810MemRange));
 	 xf86DrvMsg(pScrn->scrnIndex, X_INFO, 
 		    "GART: Found 4096K Z buffer memory\n");
@@ -459,8 +468,6 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
    
    if (agpHandle != 0) {
       /* The backbuffer is always aligned to the 56 mb mark in the aperture */
-      xf86DrvMsg(pScreen->myNum, X_INFO, "agpHandle : %p, aligned to : %lx\n",
-                 agpHandle, 56*1024*1024);
       if(drmAgpBind(pI810->drmSubFD, agpHandle, 56*1024*1024) == 0) {
 	 xf86DrvMsg(pScrn->scrnIndex, X_INFO,
 		    "Bound backbuffer memory\n");
@@ -488,10 +495,6 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
       pI810->zHandle = agpHandle;
 
       if(agpHandle != 0) {
-	 xf86DrvMsg(pScreen->myNum, X_INFO,
-                    "agpHandle : %p, aligned to : %lx\n", dcacheHandle, 
-                    48*1024*1024);
-
 	 if(drmAgpBind(pI810->drmSubFD, agpHandle, 48*1024*1024) == 0) {
 	    xf86DrvMsg(pScrn->scrnIndex, X_INFO, "Bound depthbuffer memory\n");
 	    pI810->DepthBuffer.Start = 48*1024*1024;
@@ -515,7 +518,6 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
    /* Now allocate and bind the agp space.  This memory will include the
     * regular framebuffer as well as texture memory.
     */
-   
    drmAgpAlloc(pI810->drmSubFD, sysmem_size, 0, NULL, &agpHandle);
    if (agpHandle == 0) {
       xf86DrvMsg(pScreen->myNum, X_ERROR, "drmAgpAlloc failed\n");
@@ -542,7 +544,6 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
 		    "GART: Allocated 4K for mouse cursor image\n");
 	 pI810->CursorStart = tom;	 
 	 tom += 4096;
-	 xf86DrvMsg(pScreen->myNum, X_INFO, "tom : %lx\n", tom);
       }
       else {
 	 xf86DrvMsg(pScrn->scrnIndex, X_INFO, "GART: cursor bind failed\n");
@@ -554,28 +555,21 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
       pI810->CursorPhysical = 0;
    }
 
-   ErrorF("i : %d pI810->DepthBuffer %x\n", pitch_idx, pI810->DepthBuffer);
-         
    I810SetTiledMemory(pScrn, 1,
 		      pI810->DepthBuffer.Start,
 		      i810_pitches[pitch_idx],
 		      8*1024*1024);
-   ErrorF("i : %d\n", pitch_idx);
    
    I810SetTiledMemory(pScrn, 2,
 		      pI810->BackBuffer.Start,
 		      i810_pitches[pitch_idx],
 		      8*1024*1024);
-   ErrorF("i : %d\n", pitch_idx);
    
    pI810->auxPitch = i810_pitches[pitch_idx];
    pI810->auxPitchBits = i810_pitch_flags[pitch_idx];
-   
    pI810->SavedDcacheMem = pI810->DcacheMem;
-   
    pI810DRI->backbufferSize = pI810->BackBuffer.Size;
-   xf86DrvMsg(pScreen->myNum, X_INFO, "Backbuffer map : %lx\n",
-              pI810->BackBuffer.Start);
+
    if (drmAddMap(pI810->drmSubFD, (drmHandle)pI810->BackBuffer.Start,
 		 pI810->BackBuffer.Size, DRM_AGP, 0, 
 		 &pI810DRI->backbuffer) < 0) {
@@ -584,8 +578,6 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
       return FALSE;
    }
    
-   xf86DrvMsg(pScreen->myNum, X_INFO, "Depth map : %lx\n",
-              pI810->DepthBuffer.Start);
    pI810DRI->depthbufferSize = pI810->DepthBuffer.Size;
    if (drmAddMap(pI810->drmSubFD, (drmHandle)pI810->DepthBuffer.Start,
 		 pI810->DepthBuffer.Size, DRM_AGP, 0, 
@@ -624,7 +616,6 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
    pI810DRI->agp_buffers = pI810->buffer_map;
    pI810DRI->agp_buf_size = pI810->BufferMem.Size;
 
-   ErrorF("Ring map : %lx\n", pI810->LpRing.mem.Start);
    if (drmAddMap(pI810->drmSubFD, (drmHandle)pI810->LpRing.mem.Start,
 		 pI810->LpRing.mem.Size, DRM_AGP, 0,
 		 &pI810->ring_map) < 0) {
@@ -636,8 +627,6 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
    /* Use the rest of memory for textures. */
    pI810DRI->textureSize = pI810->SysMem.Size;
 
-   ErrorF("INITIAL texsize: %x\n", pI810DRI->textureSize);
-
    i = mylog2(pI810DRI->textureSize / I810_NR_TEX_REGIONS);
 
    if (i < I810_LOG_MIN_TEX_REGION_SIZE)
@@ -646,18 +635,15 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
    pI810DRI->logTextureGranularity = i;
    pI810DRI->textureSize = (pI810DRI->textureSize >> i) << i; /* truncate */
 
-   xf86DrvMsg(pScreen->myNum, X_INFO,
-              "REDUCED texsize: %x (i: %d)\n", pI810DRI->textureSize, i);
-
-   if (pI810DRI->textureSize < 512*1024) {
-      xf86DrvMsg(pScreen->myNum, X_ERROR, "Less then 512k for textures\n");
+   if(pI810DRI->textureSize < 512*1024) {
+      ErrorF("Less then 512k for textures\n");
       DRICloseScreen(pScreen);
+      return FALSE;
    }
    
    I810AllocLow( &(pI810->TexMem), &(pI810->SysMem),
 		 pI810DRI->textureSize);
    
-   ErrorF("Texture map : %lx\n", pI810->TexMem.Start);
    if (drmAddMap(pI810->drmSubFD, (drmHandle)pI810->TexMem.Start,
 		 pI810->TexMem.Size, DRM_AGP, 0,
 		 &pI810DRI->textures) < 0) {
@@ -666,7 +652,6 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
       return FALSE;
    }
    
-   ErrorF("pI810->BufferMem.Start : %lx\n", pI810->BufferMem.Start);
    if((bufs = drmAddBufs(pI810->drmSubFD,
 			 I810_DMA_BUF_NR,
 			 I810_DMA_BUF_SZ,
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_driver.c b/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_driver.c
index 3db1809ae..6d4c8aafb 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_driver.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_driver.c
@@ -1028,7 +1028,6 @@ DoRestore(ScrnInfoPtr pScrn, vgaRegPtr vgaReg, I810RegPtr i810Reg,
    pI810 = I810PTR(pScrn);
    hwp = VGAHWPTR(pScrn);
 
-
    if (I810_DEBUG&DEBUG_VERBOSE_VGA) {
       ErrorF("Setting mode in I810Restore:\n");
       i810PrintMode( vgaReg, i810Reg );
@@ -1094,7 +1093,7 @@ DoRestore(ScrnInfoPtr pScrn, vgaRegPtr vgaReg, I810RegPtr i810Reg,
    temp &= ~INTERLACE_ENABLE;
    temp |= i810Reg->InterlaceControl;
    hwp->writeCrtc(hwp, INTERLACE_CNTL, temp);
-
+   
    temp=pI810->readControl(pI810, GRX, ADDRESS_MAPPING);
    temp &= 0xE0; /* Save reserved bits 7:5 */
    temp |= i810Reg->AddressMapping;
@@ -1532,14 +1531,6 @@ I810AllocateFront(ScrnInfoPtr pScrn) {
 		   pI810->FbMemBox.y2 * 
 		   pI810->cpp) + 4095) & ~4095);
    
-   fprintf(stderr, "Framebuffer %x, size %x\n", 
-	   pI810->FrontBuffer.Start,
-	   pI810->FrontBuffer.Size);
-   
-   fprintf(stderr, "DisplayWidth %x, virtualX %x\n", 
-	   pScrn->displayWidth,
-	   pScrn->virtualX);
-   
    xf86memset( &(pI810->LpRing), 0, sizeof( I810RingBuffer ) );
    if(I810AllocLow( &(pI810->LpRing.mem), &(pI810->SysMem), 16*4096 )) {
 	 if (I810_DEBUG & DEBUG_VERBOSE_MEMORY)
@@ -1728,13 +1719,6 @@ I810ScreenInit(int scrnIndex, ScreenPtr pScreen, int argc, char **argv) {
    }
 #endif
    
-   ErrorF("InitFB box: %d,%d-%d,%d virt: %d,%d\n",
-	  pI810->FbMemBox.x1,
-	  pI810->FbMemBox.y1,
-	  pI810->FbMemBox.x2,
-	  pI810->FbMemBox.y2,
-	  pScrn->virtualX,
-	  pScrn->virtualY);
 
    if (!xf86InitFBManager(pScreen, &(pI810->FbMemBox))) {
       xf86DrvMsg(pScrn->scrnIndex, X_ERROR, 
@@ -1890,9 +1874,10 @@ I810LeaveVT(int scrnIndex, int flags) {
       pI810->LockHeld = 1;
    }
 #endif
-
-   I810RefreshRing( pScrn );
-   I810Sync( pScrn );
+   if(pI810->AccelInfoRec != NULL) {
+      I810RefreshRing( pScrn );
+      I810Sync( pScrn );
+   }
    I810Restore(pScrn);
    vgaHWLock(hwp);
 }
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_memory.c b/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_memory.c
index 2cedd9682..f499ffa2d 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_memory.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_memory.c
@@ -46,7 +46,9 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "i810.h"
 #include "i810_reg.h"
 
+#ifdef linux
 #include <linux/agpgart.h>
+#endif
 
 int I810AllocLow( I810MemRange *result, I810MemRange *pool, int size )
 {
@@ -83,6 +85,7 @@ extern int xf86errno;
 
 int I810AllocateGARTMemory( ScrnInfoPtr pScrn ) 
 {
+#ifdef linux
    struct _agp_info agpinf;
    struct _agp_bind bind;
    struct _agp_allocate alloc;
@@ -220,18 +223,23 @@ int I810AllocateGARTMemory( ScrnInfoPtr pScrn )
 
 
    return TRUE;
+#else
+   return FALSE;
+#endif
 }
 
 
 
 void I810FreeGARTMemory( ScrnInfoPtr pScrn ) 
 {
+#ifdef linux
    I810Ptr pI810 = I810PTR(pScrn);
 
    if (pI810->gartfd != -1) {
       xf86close( pI810->gartfd );
       pI810->gartfd = -1;
    }
+#endif
 }
 
 
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dri.c b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dri.c
index ffa073561..7f91b6630 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dri.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dri.c
@@ -502,10 +502,8 @@ Bool MGADRIScreenInit(ScreenPtr pScreen)
       pMGADRI->chipset = MGA_CARD_TYPE_G400;
       break;
    case PCI_CHIP_MGAG200:
-#if 0
       pMGADRI->chipset = MGA_CARD_TYPE_G200;
       break;
-#endif
    case PCI_CHIP_MGAG200_PCI:
    default:
       xf86DrvMsg(pScreen->myNum, X_ERROR,
@@ -561,29 +559,37 @@ Bool MGADRIScreenInit(ScreenPtr pScreen)
    pMGADRI->textureSize = (pMGADRI->textureSize >> i) << i; /* truncate */
 
 
-   /* Calculate similar constants for AGP texture space
-    */
-   i = mylog2(pMGADRI->agpSize / MGA_NR_TEX_REGIONS);
-   if (i < MGA_LOG_MIN_TEX_REGION_SIZE)
-      i = MGA_LOG_MIN_TEX_REGION_SIZE;
-  
-   pMGADRI->logAgpTextureGranularity = i;
-   pMGADRI->agpTextureSize = (pMGADRI->agpSize >> i) << i; 
-
-
    /* Here is where we need to do initialization of the dma engine */
-   if ((bufs = drmAddBufs(pMGA->drmSubFD,
+   if((bufs = drmAddBufs(pMGA->drmSubFD,
                          MGA_DMA_BUF_NR,
                          MGA_DMA_BUF_SZ,
                          DRM_AGP_BUFFER,
-			init_offset)) <= 0) {
-      xf86DrvMsg(pScrn->scrnIndex, X_INFO,
-                 "[drm] failure adding %d %d byte DMA buffers\n",
-                 MGA_DMA_BUF_NR,
-                 MGA_DMA_BUF_SZ);
-      DRICloseScreen(pScreen);
-      return FALSE;
+                         init_offset)) <= 0) {
+     xf86DrvMsg(pScrn->scrnIndex, X_INFO,
+                "[drm] failure adding %d %d byte DMA buffers\n",
+                MGA_DMA_BUF_NR,
+                MGA_DMA_BUF_SZ);
+     DRICloseScreen(pScreen);
+     return FALSE;
    }
+
+   pMGADRI->agpBufferOffset = init_offset + pMGADRIServer->agp_private;
+
+   /* Calculate texture constants for AGP texture space
+    */
+   {
+      CARD32 agpTextureOffset = MGA_DMA_BUF_SZ * MGA_DMA_BUF_NR;
+      CARD32 agpTextureSize = pMGADRI->agpSize - agpTextureOffset;
+
+      i = mylog2(agpTextureSize / MGA_NR_TEX_REGIONS);
+      if (i < MGA_LOG_MIN_TEX_REGION_SIZE)
+         i = MGA_LOG_MIN_TEX_REGION_SIZE;
+
+      pMGADRI->logAgpTextureGranularity = i;
+      pMGADRI->agpTextureSize = (agpTextureSize >> i) << i; 
+      pMGADRI->agpTextureOffset = agpTextureOffset;
+   }
+
    
    xf86DrvMsg(pScrn->scrnIndex, X_INFO,
               "[drm] added %d %d byte DMA buffers\n",
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dri.h b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dri.h
index 7196bc954..b8518ee89 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dri.h
+++ b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dri.h
@@ -4,9 +4,11 @@
 #define _MGA_DRI_
 
 #include <xf86drm.h>
-#include "mga_drm_public.h"
+#include <xf86drmMga.h>
 
-typedef struct _mga_dri_server_private {
+#define MGA_MAX_DRAWABLES 256
+
+typedef struct {
    int reserved_map_agpstart;
    int reserved_map_idx;
    int buffer_map_idx;
@@ -17,6 +19,7 @@ typedef struct _mga_dri_server_private {
    int sgram;
    unsigned long agpMode;
    unsigned long agpHandle;
+   Bool agpAcquired;
    drmHandle agp_private;
    drmSize agpSizep;
    drmAddress agpBase;
@@ -24,27 +27,115 @@ typedef struct _mga_dri_server_private {
    drmHandle regs;
    drmSize regsSize;
    drmAddress regsMap;
-   mgaWarpIndex WarpIndex[MGA_MAX_WARP_PIPES];
+   drmMgaWarpIndex WarpIndex[MGA_MAX_WARP_PIPES];
    drmBufMapPtr drmBufs;
    CARD8 *agp_map;
-} MGADRIServerPrivate, *MGADRIServerPrivatePtr;
+} MGADRIServerPrivateRec, *MGADRIServerPrivatePtr;
 
 typedef struct {
-  int deviceID;
-  int width;
-  int height;
-  int mem;
-  int cpp;
-  int stride;
-  int fbOffset;
-  int backOffset;
-  int depthOffset;
-  int textureOffset;
-  int textureSize;
-  drmHandle agp;
-  drmSize agpSize;
+   int chipset;
+   int width;
+   int height;
+   int mem;
+   int cpp;
+   unsigned int frontOffset;
+   unsigned int frontPitch;
+
+   unsigned int backOffset;
+   unsigned int backPitch;
+
+   unsigned int depthOffset;
+   unsigned int depthPitch;
+
+   unsigned int textureOffset;
+   unsigned int textureSize;
+   int logTextureGranularity;
+
+   /* Allow calculation of setup dma addresses.
+    */
+   unsigned int agpBufferOffset;
+
+   unsigned int agpTextureOffset;
+   unsigned int agpTextureSize;
+   int logAgpTextureGranularity;
+
+   /* Redundant?
+    */
+   unsigned int frontOrg;
+   unsigned int backOrg;
+   unsigned int depthOrg;
+
+   unsigned int mAccess;
+
+   drmHandle agp;
+   drmSize agpSize;
 } MGADRIRec, *MGADRIPtr;
 
 
+/* WARNING: Do not change the SAREA structure without changing the kernel
+ * as well */
+typedef struct {
+   unsigned char next, prev;
+   unsigned char in_use;
+   unsigned int age;
+} MGATexRegionRec, *MGATexRegionPtr;
+
+typedef struct {
+   /* The channel for communication of state information to the kernel
+    * on firing a vertex dma buffer.
+    */
+   unsigned int ContextState[MGA_CTX_SETUP_SIZE];
+   unsigned int ServerState[MGA_2D_SETUP_SIZE];
+   unsigned int TexState[2][MGA_TEX_SETUP_SIZE];
+   unsigned int WarpPipe;
+   unsigned int dirty;
+   
+   unsigned int nbox;
+   XF86DRIClipRectRec boxes[MGA_NR_SAREA_CLIPRECTS];
+   
+   /* Information about the most recently used 3d drawable.  The
+    * client fills in the req_* fields, the server fills in the 
+    * exported_ fields and puts the cliprects into boxes, above.
+    *
+    * The client clears the exported_drawable field before
+    * clobbering the boxes data.
+    */
+   unsigned int req_drawable;       /* the X drawable id */
+   unsigned int req_draw_buffer;    /* MGA_FRONT or MGA_BACK */
+   
+   unsigned int exported_drawable;
+   unsigned int exported_index;
+   unsigned int exported_stamp;
+   unsigned int exported_buffers;
+   unsigned int exported_nfront;
+   unsigned int exported_nback;
+   int exported_back_x, exported_front_x, exported_w;
+   int exported_back_y, exported_front_y, exported_h;
+   XF86DRIClipRectRec exported_boxes[MGA_NR_SAREA_CLIPRECTS];
+   
+   /* Counters for aging textures and for client-side throttling.
+    */
+   unsigned int last_enqueue;       /* last time a buffer was enqueued */
+   unsigned int last_dispatch;      /* age of the most recently dispatched buffer */
+   unsigned int last_quiescent;     /*  */
+   
+   /* LRU lists for texture memory in agp space and on the card */
+   
+   MGATexRegionRec texList[MGA_NR_TEX_HEAPS][MGA_NR_TEX_REGIONS+1];
+   unsigned int texAge[MGA_NR_TEX_HEAPS];
+   /* Mechanism to validate card state.
+    */
+   int ctxOwner;
+} MGASAREARec, *MGASAREAPtr;
+
+typedef struct {
+  /* Nothing here yet */
+  int dummy;
+} MGAConfigPrivRec, *MGAConfigPrivPtr;
+
+typedef struct {
+  /* Nothing here yet */
+  int dummy;
+} MGADRIContextRec, *MGADRIContextPtr;
 
 #endif
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/r128/Imakefile b/xc/programs/Xserver/hw/xfree86/drivers/r128/Imakefile
index e3f7178bf..8f0f33a35 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/r128/Imakefile
+++ b/xc/programs/Xserver/hw/xfree86/drivers/r128/Imakefile
@@ -6,9 +6,16 @@ XCOMM
 #define IHaveModules
 #include <Server.tmpl>
 
-SRCS = r128_driver.c r128_cursor.c r128_accel.c # r128_i2c.c
+#if BuildXF86DRI
+DRISRCS     = r128_dri.c
+DRIOBJS     = r128_dri.o
+DRIINCLUDES = -I$(SERVERSRC)/GL/dri -I$(LIBSRC)/GL/dri
+DRIDEFINES  = $(GLX_DEFINES)
+#endif
+
+SRCS = r128_driver.c r128_cursor.c r128_accel.c $(DRISRCS) # r128_i2c.c
 
-OBJS = r128_driver.o r128_cursor.o r128_accel.o # r128_i2c.o
+OBJS = r128_driver.o r128_cursor.o r128_accel.o $(DRIOBJS) # r128_i2c.o
 
 #if defined(XF86DriverSDK)
 INCLUDES = -I. -I../../include
@@ -22,9 +29,12 @@ INCLUDES = -I. -I$(XF86COMSRC) -I$(XF86OSSRC) \
 	   -I$(XF86SRC)/ddc -I$(XF86SRC)/i2c -I$(XF86OSSRC)/vbe \
 	   -I$(XF86SRC)/int10 -I$(SERVERSRC)/Xext \
            -I$(FONTINCSRC) -I$(SERVERSRC)/include -I$(XINCLUDESRC) \
-	   -I$(EXTINCSRC) -I$(XF86SRC)/xf24_32bpp
+	   -I$(EXTINCSRC) -I$(XF86SRC)/xf24_32bpp \
+	   $(DRIINCLUDES)
 #endif
 
+DEFINES = $(DRIDEFINES)
+
 #if MakeHasPosixVariableSubstitutions
 SubdirLibraryRule($(OBJS))
 #endif
@@ -49,5 +59,10 @@ InstallDriverSDKNonExecFile(r128_cursor.c,$(DRIVERSDKDIR)/drivers/r128)
 InstallDriverSDKNonExecFile(r128_driver.c,$(DRIVERSDKDIR)/drivers/r128)
 InstallDriverSDKNonExecFile(r128_reg.h,$(DRIVERSDKDIR)/drivers/r128)
 
+InstallDriverSDKNonExecFile(r128_dri.c,$(DRIVERSDKDIR)/drivers/r128)
+InstallDriverSDKNonExecFile(r128_dri.h,$(DRIVERSDKDIR)/drivers/r128)
+InstallDriverSDKNonExecFile(r128_dripriv.h,$(DRIVERSDKDIR)/drivers/r128)
+InstallDriverSDKNonExecFile(r128_sarea.h,$(DRIVERSDKDIR)/drivers/r128)
+
 InstallDriverSDKObjectModule(r128,$(DRIVERSDKMODULEDIR),drivers)
 
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/r128/r128.h b/xc/programs/Xserver/hw/xfree86/drivers/r128/r128.h
index d5d120153..acb04f96e 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/r128/r128.h
+++ b/xc/programs/Xserver/hw/xfree86/drivers/r128/r128.h
@@ -1,8 +1,8 @@
 /* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/r128/r128.h,v 1.8 2000/02/23 04:47:18 martin Exp $ */
 /**************************************************************************
 
-Copyright 1999 ATI Technologies Inc. and Precision Insight, Inc.,
-                                         Cedar Park, Texas. 
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas.
 All Rights Reserved.
 
 Permission is hereby granted, free of charge, to any person obtaining a
@@ -40,6 +40,9 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define R128_TIMEOUT  2000000	/* Fall out of wait loops after this count */
 #define R128_MMIOSIZE 0x80000
 
+				/* R128_NAME is used for the server-side
+                                   ddx driver, the client-side DRI driver,
+                                   and the kernel-level DRM driver. */
 #define R128_NAME "r128"
 #define R128_VERSION_MAJOR 3
 #define R128_VERSION_MINOR 0
@@ -157,7 +160,7 @@ typedef struct {
     unsigned char     *FB;	  /* Map of frame buffer                     */
 
     CARD32            MemCntl;
-    CARD32	      BusCntl;
+    CARD32            BusCntl;
     unsigned long     FbMapSize;  /* Size of frame buffer, in bytes          */
     int               Flags;	  /* Saved copy of mode flags                */
     
@@ -168,6 +171,8 @@ typedef struct {
     R128SaveRec       ModeReg;	  /* Current mode                            */
     Bool              (*CloseScreen)(int, ScreenPtr);
 
+    Bool              PaletteSavedOnVT; /* Palette saved on last VT switch   */
+
     I2CBusPtr         i2c;
     XAAInfoRecPtr     accel;
     xf86CursorInfoPtr cursor;
@@ -199,8 +204,98 @@ typedef struct {
     int               scanline_words;
     int               scanline_direct;
     int               scanline_bpp; /* Only used for ImageWrite */
+
+#ifdef XF86DRI
+    Bool              directRenderingEnabled;
+    DRIInfoPtr        pDRIInfo;
+    int               drmFD;
+    int               numVisualConfigs;
+    __GLXvisualConfig *pVisualConfigs;
+    R128ConfigPrivPtr pVisualConfigsPriv;
+    
+    drmHandle         fbHandle;
+    
+    drmSize           registerSize;
+    drmHandle         registerHandle;
+    
+    Bool              IsPCI;		/* Current card is a PCI card */
+
+    drmSize           agpSize;
+    drmHandle         agpMemHandle;     /* Handle from drmAgpAlloc */
+    unsigned long     agpOffset;
+    unsigned char     *AGP;	        /* Map */
+    int               agpMode;
+
+    Bool              CCEInUse;		/* CCE is currently active */
+    int               CCEMode;		/* CCE mode that server/clients use */
+    int               CCEFifoSize;	/* Size of the CCE command FIFO */
+    Bool              CCESecure;	/* CCE security enabled */
+    int               CCEusecTimeout;	/* CCE timeout in usecs */
+    Bool              CCE2D;            /* CCE is used for X server 2D prims */
+
+				/* CCE ring buffer data */
+    unsigned long     ringStart;	/* Offset into AGP space */
+    drmHandle         ringHandle;	/* Handle from drmAddMap */
+    drmSize           ringMapSize;	/* Size of map */
+    int               ringSize;		/* Size of ring (in MB) */
+    unsigned char     *ring;		/* Map */
+    int               ringSizeLog2QW;
+
+    unsigned long     ringReadOffset;	/* Offset into AGP space */
+    drmHandle         ringReadPtrHandle; /* Handle from drmAddMap */
+    drmSize           ringReadMapSize;	/* Size of map */
+    unsigned char     *ringReadPtr;	/* Map */
+
+				/* CCE vertex buffer data */
+    unsigned long     vbStart;		/* Offset into AGP space */
+    drmHandle         vbHandle;		/* Handle from drmAddMap */
+    drmSize           vbMapSize;	/* Size of map */
+    int               vbSize;		/* Size of vert bufs (in MB) */
+    unsigned char     *vb;		/* Map */
+    int               vbBufSize;	/* Size of individual vert buf */
+    int               vbNumBufs;	/* Number of vert bufs */
+    drmBufMapPtr      vbBufs;		/* Buffer map */
+
+				/* CCE indirect buffer data */
+    unsigned long     indStart;		/* Offset into AGP space */
+    drmHandle         indHandle;	/* Handle from drmAddMap */
+    drmSize           indMapSize;	/* Size of map */
+    int               indSize;		/* Size of indirect bufs (in MB) */
+    unsigned char     *ind;		/* Map */
+
+				/* CCE AGP Texture data */
+    unsigned long     agpTexStart;	/* Offset into AGP space */
+    drmHandle         agpTexHandle;	/* Handle from drmAddMap */
+    drmSize           agpTexMapSize;	/* Size of map */
+    int               agpTexSize;	/* Size of AGP tex space (in MB) */
+    unsigned char     *agpTex;		/* Map */
+    int               log2AGPTexGran;
+
+				/* DRI screen private data */
+    int               fbX;
+    int               fbY;
+    int               backX;
+    int               backY;
+    int               depthX;
+    int               depthY;
+    int               textureX;
+    int               textureY;
+    int               textureSize;
+    int               log2TexGran;
+#endif
 } R128InfoRec, *R128InfoPtr;
 
+#define R128WaitForFifo(pScrn, entries)                                      \
+do {                                                                         \
+    if (info->fifo_slots < entries) R128WaitForFifoFunction(pScrn, entries); \
+    info->fifo_slots -= entries;                                             \
+} while (0)
+
+extern void        R128WaitForFifoFunction(ScrnInfoPtr pScrn, int entries);
+extern void        R128WaitForIdle(ScrnInfoPtr pScrn);
+extern void        R128EngineReset(ScrnInfoPtr pScrn);
+extern void        R128EngineFlush(ScrnInfoPtr pScrn);
+
 extern int         INPLL(ScrnInfoPtr pScrn, int addr);
 extern void        R128WaitForVerticalSync(ScrnInfoPtr pScrn);
 
@@ -208,4 +303,16 @@ extern Bool        R128AccelInit(ScreenPtr pScreen);
 extern void        R128EngineInit(ScrnInfoPtr pScrn);
 extern Bool        R128CursorInit(ScreenPtr pScreen);
 
+extern int         R128MinBits(int val);
+
+#ifdef XF86DRI
+extern Bool        R128DRIScreenInit(ScreenPtr pScreen);
+extern void        R128DRICloseScreen(ScreenPtr pScreen);
+extern Bool        R128DRIFinishScreenInit(ScreenPtr pScreen);
+extern void        R128CCEStart(ScrnInfoPtr pScrn);
+extern void        R128CCEStop(ScrnInfoPtr pScrn);
+extern void        R128CCEResetRing(ScrnInfoPtr pScrn);
+extern void        R128CCEWaitForIdle(ScrnInfoPtr pScrn);
+#endif
+
 #endif
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_accel.c b/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_accel.c
index 5b0223661..f42569767 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_accel.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_accel.c
@@ -1,8 +1,8 @@
 /* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/r128/r128_accel.c,v 1.7 2000/02/23 04:47:18 martin Exp $ */
 /**************************************************************************
 
-Copyright 1999 ATI Technologies Inc. and Precision Insight, Inc.,
-                                         Cedar Park, Texas. 
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
 All Rights Reserved.
 
 Permission is hereby granted, free of charge, to any person obtaining a
@@ -95,6 +95,19 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 				/* DDC support */
 #include "xf86DDC.h"
 
+				/* DRI support */
+#ifdef XF86DRI
+#include "GL/glxint.h"
+#include "xf86drm.h"
+#include "sarea.h"
+#define _XF86DRI_SERVER_
+#include "xf86dri.h"
+#include "dri.h"
+#include "r128_dri.h"
+#include "r128_dripriv.h"
+#include "r128_sarea.h"
+#endif
+
 				/* Driver data structures */
 #include "r128.h"
 #include "r128_reg.h"
@@ -122,7 +135,7 @@ static struct {
 };
 
 /* Flush all dirty data in the Pixel Cache to memory. */
-static void R128EngineFlush(ScrnInfoPtr pScrn)
+void R128EngineFlush(ScrnInfoPtr pScrn)
 {
     int i;
     R128MMIO_VARS();
@@ -134,11 +147,12 @@ static void R128EngineFlush(ScrnInfoPtr pScrn)
 }
 
 /* Reset graphics card to known state. */
-static void R128EngineReset(ScrnInfoPtr pScrn)
+void R128EngineReset(ScrnInfoPtr pScrn)
 {
-    CARD32 clock_cntl_index;
-    CARD32 mclk_cntl;
-    CARD32 gen_reset_cntl;
+    R128InfoPtr info = R128PTR(pScrn);
+    CARD32      clock_cntl_index;
+    CARD32      mclk_cntl;
+    CARD32      gen_reset_cntl;
     R128MMIO_VARS();
 	
     R128EngineFlush(pScrn);
@@ -158,17 +172,15 @@ static void R128EngineReset(ScrnInfoPtr pScrn)
     OUTPLL(R128_MCLK_CNTL,        mclk_cntl);
     OUTREG(R128_CLOCK_CNTL_INDEX, clock_cntl_index);
     OUTREG(R128_GEN_RESET_CNTL,   gen_reset_cntl);
-}
 
-#define R128WaitForFifo(pScrn, entries)                                      \
-do {                                                                         \
-    if (info->fifo_slots < entries) R128WaitForFifoFunction(pScrn, entries); \
-    info->fifo_slots -= entries;                                             \
-} while (0)
+#ifdef XF86DRI
+    if (R128CCE_USE_RING_BUFFER(info->CCEMode)) R128CCEResetRing(pScrn);
+#endif
+}
 
 /* The FIFO has 64 slots.  This routines waits until at least `entries' of
    these slots are empty. */
-static void R128WaitForFifoFunction(ScrnInfoPtr pScrn, int entries)
+void R128WaitForFifoFunction(ScrnInfoPtr pScrn, int entries)
 {
     R128InfoPtr info = R128PTR(pScrn);
     int         i;
@@ -186,15 +198,19 @@ static void R128WaitForFifoFunction(ScrnInfoPtr pScrn, int entries)
 	xf86DrvMsg(pScrn->scrnIndex, X_ERROR,
 		   "FIFO timed out, resetting engine...\n");
 	R128EngineReset(pScrn);
+#if XF86DRI
+	if (info->CCE2D) R128CCEStart(pScrn);
+#endif
     }
 }
 
 /* Wait for the graphics engine to be completely idle: the FIFO has
    drained, the Pixel Cache is flushed, and the engine is idle.  This is a
    standard "sync" function that will make the hardware "quiescent". */
-static void R128WaitForIdle(ScrnInfoPtr pScrn)
+void R128WaitForIdle(ScrnInfoPtr pScrn)
 {
-    int i;
+    R128InfoPtr info = R128PTR(pScrn);
+    int         i;
     R128MMIO_VARS();
 
     R128WaitForFifoFunction(pScrn, 64);
@@ -213,6 +229,9 @@ static void R128WaitForIdle(ScrnInfoPtr pScrn)
 	xf86DrvMsg(pScrn->scrnIndex, X_ERROR,
 		   "Idle timed out, resetting engine...\n");
 	R128EngineReset(pScrn);
+#if XF86DRI
+	if (info->CCE2D) R128CCEStart(pScrn);
+#endif
     }
 }
 
@@ -966,19 +985,33 @@ void R128EngineInit(ScrnInfoPtr pScrn)
     OUTREG(R128_DP_SRC_BKGD_CLR,   0x00000000);
     OUTREG(R128_DP_WRITE_MASK,     0xffffffff);
 
+#if X_BYTE_ORDER == X_BIG_ENDIAN
+    OUTREGP(R128_DP_DATATYPE,
+	    R128_HOST_BIG_ENDIAN_EN, ~R128_HOST_BIG_ENDIAN_EN);
+#else
+    OUTREGP(R128_DP_DATATYPE, 0, ~R128_HOST_BIG_ENDIAN_EN);
+#endif
+
     R128WaitForIdle(pScrn);
 }
 
-/* Initialize XAA for supported acceleration and also initialize the
-   graphics hardware for acceleration. */
-Bool R128AccelInit(ScreenPtr pScreen)
+#ifdef XF86DRI
+    /* FIXME: When direct rendering is enabled, we should use the CCE to
+       draw 2D commands */
+static void R128CCEAccelInit(ScrnInfoPtr pScrn, XAAInfoRecPtr a)
 {
-    ScrnInfoPtr   pScrn = xf86Screens[pScreen->myNum];
-    R128InfoPtr   info  = R128PTR(pScrn);
-    XAAInfoRecPtr a;
+    a->Flags                            = 0;
+
+				/* Sync */
+    a->Sync                             = R128CCEWaitForIdle;
+
+}
+#endif
+
+static void R128MMIOAccelInit(ScrnInfoPtr pScrn, XAAInfoRecPtr a)
+{
+    R128InfoPtr info = R128PTR(pScrn);
 
-    if (!(a = info->accel = XAACreateInfoRec())) return FALSE;
-    
     a->Flags                            = (PIXMAP_CACHE
 					   | OFFSCREEN_PIXMAPS
 					   | LINEAR_FRAMEBUFFER);
@@ -1065,7 +1098,26 @@ Bool R128AccelInit(ScreenPtr pScreen)
 					  | SCANLINE_PAD_DWORD
 					  | SYNC_AFTER_IMAGE_WRITE);
 #endif
-    
+}
+
+/* Initialize XAA for supported acceleration and also initialize the
+   graphics hardware for acceleration. */
+Bool R128AccelInit(ScreenPtr pScreen)
+{
+    ScrnInfoPtr   pScrn = xf86Screens[pScreen->myNum];
+    R128InfoPtr   info  = R128PTR(pScrn);
+    XAAInfoRecPtr a;
+
+    if (!(a = info->accel = XAACreateInfoRec())) return FALSE;
+
+#ifdef XF86DRI
+    /* FIXME: When direct rendering is enabled, we should use the CCE to
+       draw 2D commands */
+    if (info->CCE2D) R128CCEAccelInit(pScrn, a);
+    else
+#endif
+	R128MMIOAccelInit(pScrn, a);
+
     R128EngineInit(pScrn);
     return XAAInit(pScreen, a);
 }
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_cursor.c b/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_cursor.c
index a96872702..2667443d9 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_cursor.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_cursor.c
@@ -1,8 +1,8 @@
 /* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/r128/r128_cursor.c,v 1.6 2000/03/06 22:59:26 dawes Exp $ */
 /**************************************************************************
 
-Copyright 1999 ATI Technologies Inc. and Precision Insight, Inc.,
-                                         Cedar Park, Texas. 
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
 All Rights Reserved.
 
 Permission is hereby granted, free of charge, to any person obtaining a
@@ -60,6 +60,19 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 				/* DDC support */
 #include "xf86DDC.h"
 
+				/* DRI support */
+#ifdef XF86DRI
+#include "GL/glxint.h"
+#include "xf86drm.h"
+#include "sarea.h"
+#define _XF86DRI_SERVER_
+#include "xf86dri.h"
+#include "dri.h"
+#include "r128_dri.h"
+#include "r128_dripriv.h"
+#include "r128_sarea.h"
+#endif
+
 				/* Driver data structures */
 #include "r128.h"
 #include "r128_reg.h"
@@ -92,16 +105,19 @@ static void R128SetCursorColors(ScrnInfoPtr pScrn, int bg, int fg)
    (xorigin,yorigin). */
 static void R128SetCursorPosition(ScrnInfoPtr pScrn, int x, int y)
 {
-    R128InfoPtr   info    = R128PTR(pScrn);
-    int           xorigin = 0;
-    int           yorigin = 0;
-    int           total_y = pScrn->frameY1 - pScrn->frameY0;
+    R128InfoPtr           info    = R128PTR(pScrn);
+    xf86CursorInfoPtr     cursor  = info->cursor;
+    int                   xorigin = 0;
+    int                   yorigin = 0;
+    int                   total_y = pScrn->frameY1 - pScrn->frameY0;
     R128MMIO_VARS();
     
-    if (x < 0)                   xorigin = -x;
-    if (y < 0)                   yorigin = -y;
-    if (y > total_y)             y       = total_y;
-    if (info->Flags & V_DBLSCAN) y       *= 2;
+    if (x < 0)                        xorigin = -x;
+    if (y < 0)                        yorigin = -y;
+    if (y > total_y)                  y       = total_y;
+    if (info->Flags & V_DBLSCAN)      y       *= 2;
+    if (xorigin >= cursor->MaxWidth)  xorigin = cursor->MaxWidth - 1;
+    if (yorigin >= cursor->MaxHeight) yorigin = cursor->MaxHeight - 1;
 
     OUTREG(R128_CUR_HORZ_VERT_OFF,  R128_CUR_LOCK | (xorigin << 16) | yorigin);
     OUTREG(R128_CUR_HORZ_VERT_POSN, (R128_CUR_LOCK
@@ -126,47 +142,58 @@ static void R128LoadCursorImage(ScrnInfoPtr pScrn, unsigned char *image)
  
 #if X_BYTE_ORDER == X_BIG_ENDIAN
     switch(info->pixel_bytes) {
-       case 4:
-       case 3:
-          for (y = 0; y < 64; y++) {
-             P_SWAP32(d,s);
-             d++; s++;
-             P_SWAP32(d,s);
-             d++; s++;
-             P_SWAP32(d,s);
-             d++; s++;
-             P_SWAP32(d,s);
-             d++; s++;
-          }
-          break;
-       case 2:
-          for (y = 0; y < 64; y++) {
-             P_SWAP16(d,s);
-             d++; s++;
-             P_SWAP16(d,s);
-             d++; s++;
-             P_SWAP16(d,s);
-             d++; s++;
-             P_SWAP16(d,s);
-             d++; s++;
-          }
-          break;
-      default:
-         for (y = 0; y < 64; y++) {
+    case 4:
+    case 3:
+	for (y = 0; y < 64; y++) {
+	    P_SWAP32(d,s);
+	    d++; s++;
+	    P_SWAP32(d,s);
+	    d++; s++;
+	    P_SWAP32(d,s);
+	    d++; s++;
+	    P_SWAP32(d,s);
+	    d++; s++;
+	}
+	break;
+    case 2:
+	for (y = 0; y < 64; y++) {
+	    P_SWAP16(d,s);
+	    d++; s++;
+	    P_SWAP16(d,s);
+	    d++; s++;
+	    P_SWAP16(d,s);
+	    d++; s++;
+	    P_SWAP16(d,s);
+	    d++; s++;
+	}
+	break;
+    default:
+	for (y = 0; y < 64; y++) {
             *d++ = *s++;
             *d++ = *s++;
             *d++ = *s++;
             *d++ = *s++;
-         }
+	}
     }
 #else
-   for (y = 0; y < 64; y++) {
+    for (y = 0; y < 64; y++) {
 	*d++ = *s++;
 	*d++ = *s++;
 	*d++ = *s++;
 	*d++ = *s++;
     }
 #endif
+
+    /* Set the area after the cursor to be all transparent so that we
+       won't display corrupted cursors on the screen */
+    for (y = 0; y < 64; y++) {
+	*d++ = 0xffffffff; /* The AND bits */
+	*d++ = 0xffffffff;
+	*d++ = 0x00000000; /* The XOR bits */
+	*d++ = 0x00000000;
+    }
+
+   
     OUTREG(R128_CRTC_GEN_CNTL, save);
 }
 
@@ -204,6 +231,7 @@ Bool R128CursorInit(ScreenPtr pScreen)
     FBAreaPtr             fbarea;
     int                   width;
     int                   height;
+    int                   size;
 
 
     if (!(cursor = info->cursor = xf86CreateCursorInfoRec())) return FALSE;
@@ -227,8 +255,9 @@ Bool R128CursorInit(ScreenPtr pScreen)
     cursor->ShowCursor        = R128ShowCursor;
     cursor->UseHWCursor       = R128UseHWCursor;
 
+    size                      = (cursor->MaxWidth/4) * cursor->MaxHeight;
     width                     = pScrn->displayWidth;
-    height                    = (1024 + 1023) / pScrn->displayWidth;
+    height                    = (size*2 + 1023) / pScrn->displayWidth;
     fbarea                    = xf86AllocateOffscreenArea(pScreen,
 							  width,
 							  height,
@@ -246,7 +275,7 @@ Bool R128CursorInit(ScreenPtr pScreen)
 	info->cursor_start    = R128_ALIGN((fbarea->box.x1
 					    + width * fbarea->box.y1)
 					   * info->pixel_bytes, 16);
-	info->cursor_end      = info->cursor_start + 1024;
+	info->cursor_end      = info->cursor_start + size;
     }
 
     R128TRACE(("R128CursorInit (0x%08x-0x%08x)\n",
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_dri.c b/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_dri.c
new file mode 100644
index 000000000..cb647d6b6
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_dri.c
@@ -0,0 +1,1242 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *   Rickard E. Faith <faith@precisioninsight.com>
+ *   Daryll Strauss <daryll@precisioninsight.com>
+ *
+ */
+
+
+				/* X and server generic header files */
+#include "xf86.h"
+#include "xf86_ansic.h"
+#include "xf86_OSproc.h"
+#include "xf86Resources.h"
+#include "xf86RAC.h"
+#include "xf86cmap.h"
+#include "xf86fbman.h"
+
+				/* Backing store, software cursor, and
+                                   colormap initialization */
+#include "mibstore.h"
+#include "mipointer.h"
+#include "micmap.h"
+
+				/* CFB support */
+#define PSZ 8
+#include "cfb.h"
+#undef PSZ
+#include "cfb16.h"
+#include "cfb24.h"
+#include "cfb32.h"
+#include "cfb24_32.h"
+
+				/* XAA and Cursor Support */
+#include "xaa.h"
+#include "xf86Cursor.h"
+
+				/* PCI support */
+#include "xf86PciInfo.h"
+#include "xf86Pci.h"
+
+				/* DDC support */
+#include "xf86DDC.h"
+
+				/* DRI support */
+#include "GL/glxint.h"
+#include "GL/glxtokens.h"
+#include "xf86drm.h"
+#include "xf86drmR128.h"
+#include "sarea.h"
+#define _XF86DRI_SERVER_
+#include "xf86dri.h"
+#include "dri.h"
+#include "r128_dri.h"
+#include "r128_sarea.h"
+#include "r128_dripriv.h"
+
+				/* Driver data structures */
+#include "r128.h"
+#include "r128_reg.h"
+
+#define R128_WATERMARK_L      16
+#define R128_WATERMARK_M       8
+#define R128_WATERMARK_N       8
+#define R128_WATERMARK_K     128
+
+static int CCEFifoSlots = 0;
+
+#define R128CCEWaitForFifo(pScrn, entries)                                   \
+do {                                                                         \
+    if (CCEFifoSlots < entries) R128WaitForFifoFunction(pScrn, entries);     \
+    CCEFifoSlots -= entries;                                                 \
+} while (0)
+
+/* Wait for at least `entries' slots are free.  The actual number of
+   slots available is stored in info->CCEFifoSize. */
+static void R128CCEWaitForFifoFunction(ScrnInfoPtr pScrn, int entries)
+{
+    R128InfoPtr    info     = R128PTR(pScrn);
+    unsigned char *R128MMIO = info->MMIO;
+    int            i;
+
+    for (;;) {
+	for (i = 0; i < R128_TIMEOUT; i++) {
+	    CCEFifoSlots = INREG(R128_PM4_STAT) & R128_PM4_FIFOCNT_MASK;
+	    if (CCEFifoSlots >= entries) return;
+	}
+	R128EngineReset(pScrn);
+	if (info->CCE2D) R128CCEStart(pScrn);
+    }
+}
+
+/* Wait until the CCE is completely idle: the FIFO has drained and the
+   CCE is idle. */
+void R128CCEWaitForIdle(ScrnInfoPtr pScrn)
+{
+    R128InfoPtr    info     = R128PTR(pScrn);
+    unsigned char *R128MMIO = info->MMIO;
+    int            i;
+
+    if (!info->CCEInUse || info->CCEMode == R128_PM4_NONPM4) return;
+
+    if (R128CCE_USE_RING_BUFFER(info->CCEMode)) {
+	volatile CARD32 *r128RingReadPtr =
+	    (volatile CARD32 *)(info->ringReadPtr);
+	R128SAREAPrivPtr pSAREAPriv;
+
+	OUTREGP(R128_PM4_BUFFER_DL_WPTR,
+		R128_PM4_BUFFER_DL_DONE, ~R128_PM4_BUFFER_DL_DONE);
+
+	pSAREAPriv = (R128SAREAPrivPtr)DRIGetSAREAPrivate(pScrn->pScreen);
+
+	for (;;) {
+	    for (i = 0; i < R128_TIMEOUT; i++) {
+		if (*r128RingReadPtr == pSAREAPriv->ringWrite) {
+		    int pm4stat = INREG(R128_PM4_STAT);
+		    if ((pm4stat & R128_PM4_FIFOCNT_MASK) >= info->CCEFifoSize
+			&& !(pm4stat & (R128_PM4_BUSY | R128_PM4_GUI_ACTIVE)))
+			return;
+		}
+	    }
+	    R128EngineReset(pScrn);
+	    if (info->CCE2D) R128CCEStart(pScrn);
+	}
+    } else {
+	R128CCEWaitForFifoFunction(pScrn, info->CCEFifoSize);
+
+	for (;;) {
+	    for (i = 0; i < R128_TIMEOUT; i++) {
+		if (!(INREG(R128_PM4_STAT)
+		      & (R128_PM4_BUSY | R128_PM4_GUI_ACTIVE))) {
+		    R128EngineFlush(pScrn);
+		    return;
+		}
+	    }
+	    R128EngineReset(pScrn);
+	    if (info->CCE2D) R128CCEStart(pScrn);
+	}
+    }
+}
+
+/* Reset the ring buffer status, if the engine was reset */
+void R128CCEResetRing(ScrnInfoPtr pScrn)
+{
+    R128InfoPtr       info            = R128PTR(pScrn);
+    unsigned char    *R128MMIO        = info->MMIO;
+    R128SAREAPrivPtr  pSAREAPriv;
+    volatile CARD32  *r128RingReadPtr;
+
+    if (!info->CCEInUse || info->CCEMode == R128_PM4_NONPM4) return;
+
+    r128RingReadPtr = (volatile CARD32 *)(info->ringReadPtr);
+    pSAREAPriv      = (R128SAREAPrivPtr)DRIGetSAREAPrivate(pScrn->pScreen);
+
+    OUTREG(R128_PM4_BUFFER_DL_WPTR, 0);
+    OUTREG(R128_PM4_BUFFER_DL_RPTR, 0);
+    pSAREAPriv->ringWrite = 0;
+    *r128RingReadPtr = 0;
+
+    /* Resetting the ring turns off the CCE */
+    info->CCEInUse = FALSE;
+}
+
+/* Start the CCE, but only if it is not already in use and the requested
+   mode is a CCE mode.  The mode is stored in info->CCEMode. */
+void R128CCEStart(ScrnInfoPtr pScrn)
+{
+    R128InfoPtr    info     = R128PTR(pScrn);
+    unsigned char *R128MMIO = info->MMIO;
+
+    if (info->CCEInUse || info->CCEMode == R128_PM4_NONPM4) return;
+
+    R128WaitForIdle(pScrn);
+    OUTREG(R128_PM4_BUFFER_CNTL, info->CCEMode | info->ringSizeLog2QW);
+    (void)INREG(R128_PM4_BUFFER_ADDR); /* as per the sample code */
+    OUTREG(R128_PM4_MICRO_CNTL, R128_PM4_MICRO_FREERUN);
+    info->CCEInUse = TRUE;
+}
+
+/* Stop the CCE, but only if it is in use and the requested mode is not
+   the non-CCE mode.  This function also flushes any outstanding
+   requests before switching modes.*/
+void R128CCEStop(ScrnInfoPtr pScrn)
+{
+    R128InfoPtr    info     = R128PTR(pScrn);
+    unsigned char *R128MMIO = info->MMIO;
+
+    if (!info->CCEInUse || info->CCEMode == R128_PM4_NONPM4) return;
+
+    R128CCEWaitForIdle(pScrn);
+    OUTREG(R128_PM4_MICRO_CNTL, 0);
+    OUTREG(R128_PM4_BUFFER_CNTL, R128_PM4_NONPM4);
+    R128EngineReset(pScrn);
+    info->CCEInUse = FALSE;
+}
+
+/* Initialize the visual configs that are supported by the hardware.
+   These are combined with the visual configs that the indirect
+   rendering core supports, and the intersection is exported to the
+   client. */
+static Bool R128InitVisualConfigs(ScreenPtr pScreen)
+{
+    ScrnInfoPtr       pScrn            = xf86Screens[pScreen->myNum];
+    R128InfoPtr       pR128            = R128PTR(pScrn);
+    int               numConfigs       = 0;
+    __GLXvisualConfig *pConfigs        = 0;
+    R128ConfigPrivPtr pR128Configs     = 0;
+    R128ConfigPrivPtr *pR128ConfigPtrs = 0;
+    int               i, accum, stencil;
+    
+    switch (pR128->pixel_code) {
+    case 8:  /* 8bpp mode is not support */
+    case 15: /* FIXME */
+    case 24: /* FIXME */
+	return FALSE;
+
+#define R128_USE_ACCUM   1
+#define R128_USE_STENCIL 0 /* Only in 24 depth mode */
+
+    case 16:
+	numConfigs = 1;
+	if (R128_USE_ACCUM) numConfigs *= 2;
+	
+	if (!(pConfigs
+	      = (__GLXvisualConfig*)xnfcalloc(sizeof(__GLXvisualConfig),
+					      numConfigs))) {
+	    return FALSE;
+	}
+	if (!(pR128Configs
+	      = (R128ConfigPrivPtr)xnfcalloc(sizeof(R128ConfigPrivRec),
+					     numConfigs))) {
+	    xfree(pConfigs);
+	    return FALSE;
+	}
+	if (!(pR128ConfigPtrs
+	      = (R128ConfigPrivPtr*)xnfcalloc(sizeof(R128ConfigPrivPtr),
+					      numConfigs))) {
+	    xfree(pConfigs);
+	    xfree(pR128Configs);
+	    return FALSE;
+	}
+	
+	i = 0;
+	for (accum = 0; accum <= R128_USE_ACCUM; accum++) {
+	    pR128ConfigPtrs[i] = &pR128Configs[i];
+
+	    pConfigs[i].vid                = -1;
+	    pConfigs[i].class              = -1;
+	    pConfigs[i].rgba               = TRUE;
+	    pConfigs[i].redSize            = 5;
+	    pConfigs[i].greenSize          = 6;
+	    pConfigs[i].blueSize           = 5;
+	    pConfigs[i].redMask            = 0x0000F800;
+	    pConfigs[i].greenMask          = 0x000007E0;
+	    pConfigs[i].blueMask           = 0x0000001F;
+
+	    pConfigs[i].alphaMask          = 0;
+	    if (accum) { /* Simulated in software */
+		pConfigs[i].accumRedSize   = 16;
+		pConfigs[i].accumGreenSize = 16;
+		pConfigs[i].accumBlueSize  = 16;
+		pConfigs[i].accumAlphaSize = 0;
+	    } else {
+		pConfigs[i].accumRedSize   = 0;
+		pConfigs[i].accumGreenSize = 0;
+		pConfigs[i].accumBlueSize  = 0;
+		pConfigs[i].accumAlphaSize = 0;
+	    }
+	    pConfigs[i].doubleBuffer       = TRUE;
+	    pConfigs[i].stereo             = FALSE;
+	    pConfigs[i].bufferSize         = 16;
+	    pConfigs[i].depthSize          = 16;
+	    pConfigs[i].stencilSize        = 0;
+	    pConfigs[i].auxBuffers         = 0;
+	    pConfigs[i].level              = 0;
+	    if (accum)
+		pConfigs[i].visualRating   = GLX_SLOW_VISUAL_EXT;
+	    else
+		pConfigs[i].visualRating   = GLX_NONE_EXT;
+	    pConfigs[i].transparentPixel   = 0;
+	    pConfigs[i].transparentRed     = 0;
+	    pConfigs[i].transparentGreen   = 0;
+	    pConfigs[i].transparentBlue    = 0;
+	    pConfigs[i].transparentAlpha   = 0;
+	    pConfigs[i].transparentIndex   = 0;
+	    i++;
+	}
+	break;
+    case 32:
+	numConfigs = 1;
+	if (R128_USE_ACCUM)   numConfigs *= 2;
+	if (R128_USE_STENCIL) numConfigs *= 2;
+	
+	if (!(pConfigs
+	      = (__GLXvisualConfig*)xnfcalloc(sizeof(__GLXvisualConfig),
+					      numConfigs))) {
+	    return FALSE;
+	}
+	if (!(pR128Configs
+	      = (R128ConfigPrivPtr)xnfcalloc(sizeof(R128ConfigPrivRec),
+					     numConfigs))) {
+	    xfree(pConfigs);
+	    return FALSE;
+	}
+	if (!(pR128ConfigPtrs
+	      = (R128ConfigPrivPtr*)xnfcalloc(sizeof(R128ConfigPrivPtr),
+					      numConfigs))) {
+	    xfree(pConfigs);
+	    xfree(pR128Configs);
+	    return FALSE;
+	}
+	
+	i = 0;
+	for (accum = 0; accum <= R128_USE_ACCUM; accum++) {
+	    for (stencil = 0; stencil <= R128_USE_STENCIL; stencil++) {
+		pR128ConfigPtrs[i] = &pR128Configs[i];
+
+		pConfigs[i].vid                = -1;
+		pConfigs[i].class              = -1;
+		pConfigs[i].rgba               = TRUE;
+		pConfigs[i].redSize            = 8;
+		pConfigs[i].greenSize          = 8;
+		pConfigs[i].blueSize           = 8;
+		pConfigs[i].redMask            = 0x00FF0000;
+		pConfigs[i].greenMask          = 0x0000FF00;
+		pConfigs[i].blueMask           = 0x000000FF;
+
+		pConfigs[i].alphaMask          = 0;
+		if (accum) { /* Simulated in software */
+		    pConfigs[i].accumRedSize   = 16;
+		    pConfigs[i].accumGreenSize = 16;
+		    pConfigs[i].accumBlueSize  = 16;
+		    pConfigs[i].accumAlphaSize = 0;
+		} else {
+		    pConfigs[i].accumRedSize   = 0;
+		    pConfigs[i].accumGreenSize = 0;
+		    pConfigs[i].accumBlueSize  = 0;
+		    pConfigs[i].accumAlphaSize = 0;
+		}
+		pConfigs[i].doubleBuffer       = TRUE;
+		pConfigs[i].stereo             = FALSE;
+		pConfigs[i].bufferSize         = 24;
+		if (stencil) {
+		    pConfigs[i].depthSize      = 24;
+		    pConfigs[i].stencilSize    = 8;
+		} else {
+		    pConfigs[i].depthSize      = 32;
+		    pConfigs[i].stencilSize    = 0;
+		}
+		pConfigs[i].auxBuffers         = 0;
+		pConfigs[i].level              = 0;
+		if (accum)
+		    pConfigs[i].visualRating   = GLX_SLOW_VISUAL_EXT;
+		else
+		    pConfigs[i].visualRating   = GLX_NONE_EXT;
+		pConfigs[i].transparentPixel   = 0;
+		pConfigs[i].transparentRed     = 0;
+		pConfigs[i].transparentGreen   = 0;
+		pConfigs[i].transparentBlue    = 0;
+		pConfigs[i].transparentAlpha   = 0;
+		pConfigs[i].transparentIndex   = 0;
+		i++;
+	    }
+	}
+	break;
+    }
+    
+    pR128->numVisualConfigs   = numConfigs;
+    pR128->pVisualConfigs     = pConfigs;
+    pR128->pVisualConfigsPriv = pR128Configs;
+    GlxSetVisualConfigs(numConfigs, pConfigs, (void**)pR128ConfigPtrs);
+    return TRUE;
+}
+
+/* Create the Rage 128-specific context information */
+static Bool R128CreateContext(ScreenPtr pScreen, VisualPtr visual, 
+			      drmContext hwContext, void *pVisualConfigPriv,
+			      DRIContextType contextStore)
+{
+    /* Nothing yet */
+    return TRUE;
+}
+
+/* Destroy the Rage 128-specific context information */
+static void R128DestroyContext(ScreenPtr pScreen, drmContext hwContext, 
+			       DRIContextType contextStore)
+{
+    /* Nothing yet */
+}
+
+/* Called when the X server is woken up to allow the last client's
+   context to be saved and the X server's context to be loaded.  This is
+   not necessary for the Rage 128 since the client detects when it's
+   context is not currently loaded and then load's it itself.  Since the
+   registers to start and stop the CCE are privileged, only the X server
+   can start/stop the engine. */
+static void R128EnterServer(ScreenPtr pScreen)
+{
+    ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
+    R128InfoPtr pR128 = R128PTR(pScrn);
+
+    if (pR128->accel) pR128->accel->NeedToSync = TRUE;
+
+#if 1
+    if (!pR128->CCE2D) R128CCEStop(pScrn);
+#else
+    if (pR128->CCE2D) R128CCEWaitForIdle(pScrn);
+    else              R128CCEStop(pScrn);
+#endif
+}
+
+/* Called when the X server goes to sleep to allow the X server's
+   context to be saved and the last client's context to be loaded.  This
+   is not necessary for the Rage 128 since the client detects when it's
+   context is not currently loaded and then load's it itself.  Since the
+   registers to start and stop the CCE are privileged, only the X server
+   can start/stop the engine. */
+static void R128LeaveServer(ScreenPtr pScreen)
+{
+    ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
+    R128InfoPtr pR128 = R128PTR(pScrn);
+
+#if 1
+    if (!pR128->CCE2D) R128CCEStart(pScrn);
+#else
+    if (pR128->CCE2D) R128CCEWaitForIdle(pScrn);
+    else              R128CCEStart(pScrn);
+#endif
+}
+
+/* Contexts can be swapped by the X server if necessary.  This callback
+   is currently only used to perform any functions necessary when
+   entering or leaving the X server, and in the future might not be
+   necessary. */
+static void R128DRISwapContext(ScreenPtr pScreen, DRISyncType syncType, 
+			       DRIContextType oldContextType, void *oldContext,
+			       DRIContextType newContextType, void *newContext)
+{
+    if ((syncType==DRI_3D_SYNC) && (oldContextType==DRI_2D_CONTEXT) &&
+	(newContextType==DRI_2D_CONTEXT)) { /* Entering from Wakeup */
+	R128EnterServer(pScreen);
+    }
+    if ((syncType==DRI_2D_SYNC) && (oldContextType==DRI_NO_CONTEXT) &&
+	(newContextType==DRI_2D_CONTEXT)) { /* Exiting from Block Handler */
+	R128LeaveServer(pScreen);
+    }
+}
+
+/* Initialize the state of the back and depth buffers. */
+static void R128DRIInitBuffers(WindowPtr pWin, RegionPtr prgn, CARD32 index)
+{
+    /* FIXME: This routine needs to have acceleration turned on */
+    ScreenPtr   pScreen = pWin->drawable.pScreen;
+    ScrnInfoPtr pScrn   = xf86Screens[pScreen->myNum];
+    R128InfoPtr pR128   = R128PTR(pScrn);
+    BoxPtr      pbox;
+    int         nbox;
+    int         depth;
+
+    /* FIXME: Use accel when CCE 2D code is written */
+    if (pR128->CCE2D) return;
+
+    /* FIXME: This should be based on the __GLXvisualConfig info */
+    switch (pScrn->bitsPerPixel) {
+    case  8: depth = 0x000000ff; break;
+    case 16: depth = 0x0000ffff; break;
+    case 24: depth = 0x00ffffff; break;
+    case 32: depth = 0xffffffff; break;
+    default: depth = 0x00000000; break;
+    }
+
+    /* FIXME: Copy XAAPaintWindow() and use REGION_TRANSLATE() */
+    /* FIXME: Only initialize the back and depth buffers for contexts
+       that request them */
+
+    pbox = REGION_RECTS(prgn);
+    nbox = REGION_NUM_RECTS(prgn);
+
+    (*pR128->accel->SetupForSolidFill)(pScrn, 0, GXcopy, -1);
+    for (; nbox; nbox--, pbox++) {
+	(*pR128->accel->SubsequentSolidFillRect)(pScrn,
+						 pbox->x1 + pR128->fbX,
+						 pbox->y1 + pR128->fbY,
+						 pbox->x2 - pbox->x1,
+						 pbox->y2 - pbox->y1);
+	(*pR128->accel->SubsequentSolidFillRect)(pScrn,
+						 pbox->x1 + pR128->backX,
+						 pbox->y1 + pR128->backY,
+						 pbox->x2 - pbox->x1,
+						 pbox->y2 - pbox->y1);
+    }
+
+    (*pR128->accel->SetupForSolidFill)(pScrn, depth, GXcopy, -1);
+    for (; nbox; nbox--, pbox++)
+	(*pR128->accel->SubsequentSolidFillRect)(pScrn,
+						 pbox->x1 + pR128->depthX,
+						 pbox->y1 + pR128->depthY,
+						 pbox->x2 - pbox->x1,
+						 pbox->y2 - pbox->y1);
+
+    pR128->accel->NeedToSync = TRUE;
+}
+
+/* Copy the back and depth buffers when the X server moves a window. */
+static void R128DRIMoveBuffers(WindowPtr pWin, DDXPointRec ptOldOrg, 
+			       RegionPtr prgnSrc, CARD32 index)
+{
+    ScreenPtr   pScreen = pWin->drawable.pScreen;
+    ScrnInfoPtr pScrn   = xf86Screens[pScreen->myNum];
+    R128InfoPtr pR128   = R128PTR(pScrn);
+
+    /* FIXME: This routine needs to have acceleration turned on */
+    /* FIXME: Copy XAACopyWindow() and use REGION_TRANSLATE() */
+    /* FIXME: Only initialize the back and depth buffers for contexts
+       that request them */
+
+    /* FIXME: Use accel when CCE 2D code is written */
+    if (pR128->CCE2D) return;
+}
+
+/* Initialize the AGP state.  Request memory for use in AGP space, and
+   initialize the Rage 128 registers to point to that memory. */
+static Bool R128DRIAgpInit(R128InfoPtr pR128, ScreenPtr pScreen)
+{
+    unsigned char *R128MMIO = pR128->MMIO;
+    unsigned long mode;
+    unsigned int  vendor, device;
+    int           ret;
+    unsigned long cntl;
+    int           s, l;
+    int           flags;
+    
+    if (drmAgpAcquire(pR128->drmFD) < 0) {
+	xf86DrvMsg(pScreen->myNum, X_ERROR, "[agp] AGP not available\n");
+	return FALSE;
+    }
+
+				/* Modify the mode if the default mode is
+                                   not appropriate for this particular
+                                   combination of graphics card and AGP
+                                   chipset. */
+
+    mode   = drmAgpGetMode(pR128->drmFD);	/* Default mode */
+    vendor = drmAgpVendorId(pR128->drmFD);
+    device = drmAgpDeviceId(pR128->drmFD);
+
+    mode &= ~R128_AGP_MODE_MASK;
+    switch (pR128->agpMode) {
+    case 2:          mode |= R128_AGP_2X_MODE;
+    case 1: default: mode |= R128_AGP_1X_MODE;
+    }
+
+    xf86DrvMsg(pScreen->myNum, X_INFO,
+	       "[agp] Mode 0x%08lx [AGP 0x%04x/0x%04x; Card 0x%04x/0x%04x]\n",
+	       mode, vendor, device,
+	       pR128->PciInfo->vendor,
+	       pR128->PciInfo->chipType);
+
+    if (drmAgpEnable(pR128->drmFD, mode) < 0) {
+	xf86DrvMsg(pScreen->myNum, X_ERROR, "[agp] AGP not enabled\n");
+	drmAgpRelease(pR128->drmFD);
+	return FALSE;
+    }
+
+    pR128->agpOffset = 0;
+
+    if ((ret = drmAgpAlloc(pR128->drmFD, pR128->agpSize*1024*1024, 0, NULL,
+			   &pR128->agpMemHandle)) < 0) {
+	xf86DrvMsg(pScreen->myNum, X_ERROR, "[agp] Out of memory (%d)\n", ret);
+	drmAgpRelease(pR128->drmFD);
+	return FALSE;
+    }
+    xf86DrvMsg(pScreen->myNum, X_INFO,
+	       "[agp] %d kB allocated with handle 0x%08x\n",
+	       pR128->agpSize*1024, pR128->agpMemHandle);
+
+    if (drmAgpBind(pR128->drmFD, pR128->agpMemHandle, pR128->agpOffset) < 0) {
+	xf86DrvMsg(pScreen->myNum, X_ERROR, "[agp] Could not bind\n");
+	drmAgpFree(pR128->drmFD, pR128->agpMemHandle);
+	drmAgpRelease(pR128->drmFD);
+	return FALSE;
+    }
+
+				/* Initialize the CCE ring buffer data */
+    pR128->ringStart       = pR128->agpOffset;
+    pR128->ringMapSize     = pR128->ringSize*1024*1024 + 4096;
+    pR128->ringSizeLog2QW  = R128MinBits(pR128->ringSize*1024*1024/8) - 1;
+
+    pR128->ringReadOffset  = pR128->ringStart + pR128->ringMapSize;
+    pR128->ringReadMapSize = 4096;
+
+				/* Reserve space for the vertex buffer */
+    pR128->vbStart         = pR128->ringReadOffset + pR128->ringReadMapSize;
+    pR128->vbMapSize       = pR128->vbSize*1024*1024;
+
+				/* Reserve space for the indirect buffer */
+    pR128->indStart        = pR128->vbStart + pR128->vbMapSize;
+    pR128->indMapSize      = pR128->indSize*1024*1024;
+
+				/* Reserve the rest for AGP textures */
+    pR128->agpTexStart     = pR128->indStart + pR128->indMapSize;
+    s = (pR128->agpSize*1024*1024 - pR128->agpTexStart);
+    l = R128MinBits((s-1) / R128_NR_TEX_REGIONS);
+    if (l < R128_LOG_TEX_GRANULARITY) l = R128_LOG_TEX_GRANULARITY;
+    pR128->agpTexMapSize   = (s >> l) << l;
+    pR128->log2AGPTexGran  = l;
+
+    if (pR128->CCESecure) flags = DRM_READ_ONLY;
+    else                  flags = 0;
+
+    if (drmAddMap(pR128->drmFD, pR128->ringStart, pR128->ringMapSize,
+		  DRM_AGP, flags, &pR128->ringHandle) < 0) {
+	xf86DrvMsg(pScreen->myNum, X_ERROR,
+		   "[agp] Could not add ring mapping\n");
+	return FALSE;
+    }
+    xf86DrvMsg(pScreen->myNum, X_INFO,
+	       "[agp] ring handle = 0x%08lx\n", pR128->ringHandle);
+
+    if (drmMap(pR128->drmFD, pR128->ringHandle, pR128->ringMapSize,
+	       (drmAddressPtr)&pR128->ring) < 0) {
+	xf86DrvMsg(pScreen->myNum, X_ERROR, "[agp] Could not map ring\n");
+	return FALSE;
+    }
+    xf86DrvMsg(pScreen->myNum, X_INFO,
+	       "[agp] Ring mapped at 0x%08lx\n",
+	       (unsigned long)pR128->ring);
+
+    if (drmAddMap(pR128->drmFD, pR128->ringReadOffset, pR128->ringReadMapSize,
+		  DRM_AGP, flags, &pR128->ringReadPtrHandle) < 0) {
+	xf86DrvMsg(pScreen->myNum, X_ERROR,
+		   "[agp] Could not add ring read ptr mapping\n");
+	return FALSE;
+    }
+    xf86DrvMsg(pScreen->myNum, X_INFO,
+	       "[agp] ring read ptr handle = 0x%08lx\n",
+	       pR128->ringReadPtrHandle);
+
+    if (drmMap(pR128->drmFD, pR128->ringReadPtrHandle, pR128->ringReadMapSize,
+	       (drmAddressPtr)&pR128->ringReadPtr) < 0) {
+	xf86DrvMsg(pScreen->myNum, X_ERROR,
+		   "[agp] Could not map ring read ptr\n");
+	return FALSE;
+    }
+    xf86DrvMsg(pScreen->myNum, X_INFO,
+	       "[agp] Ring read ptr mapped at 0x%08lx\n",
+	       (unsigned long)pR128->ringReadPtr);
+    
+    if (drmAddMap(pR128->drmFD, pR128->vbStart, pR128->vbMapSize,
+		  DRM_AGP, 0, &pR128->vbHandle) < 0) {
+	xf86DrvMsg(pScreen->myNum, X_ERROR,
+		   "[agp] Could not add vertex buffers mapping\n");
+	return FALSE;
+    }
+    xf86DrvMsg(pScreen->myNum, X_INFO,
+	       "[agp] vertex buffers handle = 0x%08lx\n", pR128->vbHandle);
+
+    if (drmMap(pR128->drmFD, pR128->vbHandle, pR128->vbMapSize,
+	       (drmAddressPtr)&pR128->vb) < 0) {
+	xf86DrvMsg(pScreen->myNum, X_ERROR,
+		   "[agp] Could not map vertex buffers\n");
+	return FALSE;
+    }
+    xf86DrvMsg(pScreen->myNum, X_INFO,
+	       "[agp] Vertex buffers mapped at 0x%08lx\n",
+	       (unsigned long)pR128->vb);
+    
+    if (drmAddMap(pR128->drmFD, pR128->indStart, pR128->indMapSize,
+		  DRM_AGP, flags, &pR128->indHandle) < 0) {
+	xf86DrvMsg(pScreen->myNum, X_ERROR,
+		   "[agp] Could not add indirect buffers mapping\n");
+	return FALSE;
+    }
+    xf86DrvMsg(pScreen->myNum, X_INFO,
+	       "[agp] indirect buffers handle = 0x%08lx\n", pR128->indHandle);
+
+    if (drmMap(pR128->drmFD, pR128->indHandle, pR128->indMapSize,
+	       (drmAddressPtr)&pR128->ind) < 0) {
+	xf86DrvMsg(pScreen->myNum, X_ERROR,
+		   "[agp] Could not map indirect buffers\n");
+	return FALSE;
+    }
+    xf86DrvMsg(pScreen->myNum, X_INFO,
+	       "[agp] Indirect buffers mapped at 0x%08lx\n",
+	       (unsigned long)pR128->ind);
+    
+    if (drmAddMap(pR128->drmFD, pR128->agpTexStart, pR128->agpTexMapSize,
+		  DRM_AGP, 0, &pR128->agpTexHandle) < 0) {
+	xf86DrvMsg(pScreen->myNum, X_ERROR,
+		   "[agp] Could not add AGP texture map mapping\n");
+	return FALSE;
+    }
+    xf86DrvMsg(pScreen->myNum, X_INFO,
+	       "[agp] AGP texture map handle = 0x%08lx\n",
+	       pR128->agpTexHandle);
+
+    if (drmMap(pR128->drmFD, pR128->agpTexHandle, pR128->agpTexMapSize,
+	       (drmAddressPtr)&pR128->agpTex) < 0) {
+	xf86DrvMsg(pScreen->myNum, X_ERROR,
+		   "[agp] Could not map AGP texture map\n");
+	return FALSE;
+    }
+    xf86DrvMsg(pScreen->myNum, X_INFO,
+	       "[agp] AGP Texture map mapped at 0x%08lx\n",
+	       (unsigned long)pR128->agpTex);
+
+				/* Initialize Rage 128's AGP registers */
+    cntl  = INREG(R128_AGP_CNTL);
+    cntl &= ~R128_AGP_APER_SIZE_MASK;
+    switch (pR128->agpSize) {
+    case 256: cntl |= R128_AGP_APER_SIZE_256MB; break;
+    case 128: cntl |= R128_AGP_APER_SIZE_128MB; break;
+    case  64: cntl |= R128_AGP_APER_SIZE_64MB;  break;
+    case  32: cntl |= R128_AGP_APER_SIZE_32MB;  break;
+    case  16: cntl |= R128_AGP_APER_SIZE_16MB;  break;
+    case   8: cntl |= R128_AGP_APER_SIZE_8MB;   break;
+    case   4: cntl |= R128_AGP_APER_SIZE_4MB;   break;
+    default:
+	xf86DrvMsg(pScreen->myNum, X_ERROR,
+		   "[agp] Illegal aperture size %d kB\n",
+		   pR128->agpSize*1024);
+	return FALSE;
+    }
+    OUTREG(R128_AGP_BASE, pR128->ringHandle); /* Ring buf is at AGP offset 0 */
+    OUTREG(R128_AGP_CNTL, cntl);
+
+    return TRUE;
+}
+
+/* Add a map for the MMIO registers that will be accessed by any
+   DRI-based clients. */
+static Bool R128DRIMapInit(R128InfoPtr pR128, ScreenPtr pScreen)
+{
+    int flags;
+
+    if (pR128->CCESecure) flags = DRM_READ_ONLY;
+    else                  flags = 0;
+
+				/* Map registers */
+    pR128->registerSize = R128_MMIOSIZE;
+    if (drmAddMap(pR128->drmFD, pR128->MMIOAddr, pR128->registerSize,
+		  DRM_REGISTERS, flags, &pR128->registerHandle) < 0) {
+	return FALSE;	
+    }
+    xf86DrvMsg(pScreen->myNum, X_INFO,
+	       "[drm] register handle = 0x%08lx\n", pR128->registerHandle);
+    
+    return TRUE;
+}
+
+/* Initialize the ring buffer state for use in the X server and any
+   DRI-based clients. */
+static void R128DRICCEInitRingBuffer(ScrnInfoPtr pScrn)
+{
+    R128InfoPtr    info     = R128PTR(pScrn);
+    unsigned char *R128MMIO = info->MMIO;
+    unsigned long  addr;
+
+    /* FIXME: When we use the CCE for the X server, we should move this
+       function (and the support functions above) to r128_accel.c */
+
+				/* The manual (p. 2) says this address is
+                                   in "VM space".  This means it's an
+                                   offset from the start of AGP space. */
+    OUTREG(R128_PM4_BUFFER_OFFSET, info->ringStart | 0x02000000);
+
+    OUTREG(R128_PM4_BUFFER_DL_WPTR, 0);
+    OUTREG(R128_PM4_BUFFER_DL_RPTR, 0);
+
+				/* DL_RPTR_ADDR is a physical address.
+                                   This should be in the SAREA. */
+    *(volatile long unsigned *)(info->ringReadPtr) = 0;
+    OUTREG(R128_PM4_BUFFER_DL_RPTR_ADDR, (info->ringReadPtrHandle));
+
+				/* Set watermark control */
+    OUTREG(R128_PM4_BUFFER_WM_CNTL,
+	   ((R128_WATERMARK_L/4) << R128_WMA_SHIFT)
+	   | ((R128_WATERMARK_M/4) << R128_WMB_SHIFT)
+	   | ((R128_WATERMARK_N/4) << R128_WMC_SHIFT)
+	   | ((R128_WATERMARK_K/64) << R128_WB_WM_SHIFT));
+
+    addr = INREG(R128_PM4_BUFFER_ADDR);	/* Force read.  Why?  Because it's
+                                           in the examples... */
+    
+#if 0
+    R128CCEWaitForIdle(pScrn);
+#endif
+
+				/* Turn on bus mastering */
+    info->BusCntl &= ~R128_BUS_MASTER_DIS;
+    OUTREGP(R128_BUS_CNTL, 0, ~R128_BUS_MASTER_DIS);
+}
+
+/* Initialize the kernel data structures. */
+static int R128DRIKernelInit(R128InfoPtr pR128, ScreenPtr pScreen)
+{
+    drmR128Init      drmInfo;
+
+    drmInfo.sarea_priv_offset   = sizeof(XF86DRISAREARec);
+    drmInfo.is_pci              = pR128->IsPCI;
+    drmInfo.cce_mode            = pR128->CCEMode;
+    drmInfo.cce_fifo_size       = pR128->CCEFifoSize;
+    drmInfo.cce_secure          = pR128->CCESecure;
+    drmInfo.ring_size           = pR128->ringSize*1024*1024;
+    drmInfo.usec_timeout        = pR128->CCEusecTimeout;
+
+    drmInfo.fb_offset           = pR128->LinearAddr;
+    drmInfo.agp_ring_offset     = pR128->ringHandle;
+    drmInfo.agp_read_ptr_offset = pR128->ringReadPtrHandle;
+    drmInfo.agp_vertbufs_offset = pR128->vbHandle;
+    drmInfo.agp_indbufs_offset  = pR128->indHandle;
+    drmInfo.agp_textures_offset = pR128->agpTexHandle;
+    drmInfo.mmio_offset         = pR128->registerHandle;
+
+    if (drmR128InitCCE(pR128->drmFD, &drmInfo) < 0) return FALSE;
+
+    return TRUE;
+}
+
+/* Add a map for the vertex buffers that will be accessed by any
+   DRI-based clients. */
+static Bool R128DRIBufInit(R128InfoPtr pR128, ScreenPtr pScreen)
+{
+				/* Initialize vertex buffers */
+    if ((pR128->vbNumBufs = drmAddBufs(pR128->drmFD,
+				       pR128->vbMapSize / pR128->vbBufSize,
+				       pR128->vbBufSize,
+				       DRM_AGP_BUFFER,
+				       pR128->vbStart)) <= 0) {
+	xf86DrvMsg(pScreen->myNum, X_ERROR,
+		   "[drm] Could not create vertex buffers list\n");
+	return FALSE;
+    }
+    xf86DrvMsg(pScreen->myNum, X_INFO,
+	       "[drm] Added %d %d byte vertex buffers\n",
+	       pR128->vbNumBufs, pR128->vbBufSize);
+
+    if (drmMarkBufs(pR128->drmFD, 0.133333, 0.266666)) {
+	xf86DrvMsg(pScreen->myNum, X_ERROR,
+		   "[drm] Failed to mark vertex buffers list\n");
+	return FALSE;
+    }
+
+    if (!(pR128->vbBufs = drmMapBufs(pR128->drmFD))) {
+	xf86DrvMsg(pScreen->myNum, X_ERROR,
+		   "[drm] Failed to map vertex buffers list\n");
+	return FALSE;
+    }
+    xf86DrvMsg(pScreen->myNum, X_INFO,
+	       "[drm] Mapped %d vertex buffers\n",
+	       pR128->vbBufs->count);
+
+    return TRUE;
+}
+
+/* Load the microcode for the CCE */
+static void R128DRILoadMicrocode(ScrnInfoPtr pScrn)
+{
+    unsigned char *R128MMIO = R128PTR(pScrn)->MMIO;
+    int            i;
+    unsigned long  R128Microcode[] = {
+	/* CCE microcode (from ATI) */
+    0, 276838400, 0, 268449792, 2, 142, 2, 145, 0, 1076765731, 0, 1617039951,
+    0, 774592877, 0, 1987540286, 0, 2307490946U, 0, 599558925, 0, 589505315, 0,
+    596487092, 0, 589505315, 1, 11544576, 1, 206848, 1, 311296, 1, 198656, 2,
+    912273422, 11, 262144, 0, 0, 1, 33559837, 1, 7438, 1, 14809, 1, 6615, 12,
+    28, 1, 6614, 12, 28, 2, 23, 11, 18874368, 0, 16790922, 1, 409600, 9, 30, 1,
+    147854772, 16, 420483072, 3, 8192, 0, 10240, 1, 198656, 1, 15630, 1, 51200,
+    10, 34858, 9, 42, 1, 33559823, 2, 10276, 1, 15717, 1, 15718, 2, 43, 1,
+    15936948, 1, 570480831, 1, 14715071, 12, 322123831, 1, 33953125, 12, 55, 1,
+    33559908, 1, 15718, 2, 46, 4, 2099258, 1, 526336, 1, 442623, 4, 4194365, 1,
+    509952, 1, 459007, 3, 0, 12, 92, 2, 46, 12, 176, 1, 15734, 1, 206848, 1,
+    18432, 1, 133120, 1, 100670734, 1, 149504, 1, 165888, 1, 15975928, 1,
+    1048576, 6, 3145806, 1, 15715, 16, 2150645232U, 2, 268449859, 2, 10307, 12,
+    176, 1, 15734, 1, 15735, 1, 15630, 1, 15631, 1, 5253120, 6, 3145810, 16,
+    2150645232U, 1, 15864, 2, 82, 1, 343310, 1, 1064207, 2, 3145813, 1, 15728,
+    1, 7817, 1, 15729, 3, 15730, 12, 92, 2, 98, 1, 16168, 1, 16167, 1, 16002,
+    1, 16008, 1, 15974, 1, 15975, 1, 15990, 1, 15976, 1, 15977, 1, 15980, 0,
+    15981, 1, 10240, 1, 5253120, 1, 15720, 1, 198656, 6, 110, 1, 180224, 1,
+    103824738, 2, 112, 2, 3145839, 0, 536885440, 1, 114880, 14, 125, 12,
+    206975, 1, 33559995, 12, 198784, 0, 33570236, 1, 15803, 0, 15804, 3,
+    294912, 1, 294912, 3, 442370, 1, 11544576, 0, 811612160, 1, 12593152, 1,
+    11536384, 1, 14024704, 7, 310382726, 0, 10240, 1, 14796, 1, 14797, 1,
+    14793, 1, 14794, 0, 14795, 1, 268679168, 1, 9437184, 1, 268449792, 1,
+    198656, 1, 9452827, 1, 1075854602, 1, 1075854603, 1, 557056, 1, 114880, 14,
+    159, 12, 198784, 1, 1109409213, 12, 198783, 1, 1107312059, 12, 198784, 1,
+    1109409212, 2, 162, 1, 1075854781, 1, 1073757627, 1, 1075854780, 1, 540672,
+    1, 10485760, 6, 3145894, 16, 274741248, 9, 168, 3, 4194304, 3, 4209949, 0,
+    0, 0, 256, 14, 174, 1, 114857, 1, 33560007, 12, 176, 0, 10240, 1, 114858,
+    1, 33560018, 1, 114857, 3, 33560007, 1, 16008, 1, 114874, 1, 33560360, 1,
+    114875, 1, 33560154, 0, 15963, 0, 256, 0, 4096, 1, 409611, 9, 188, 0,
+    10240, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+    0, 0, 0, 0, 0, 0, 0
+    };
+
+    R128WaitForIdle(pScrn);
+
+    OUTREG(R128_PM4_MICROCODE_ADDR, 0);
+    for (i = 0; i < 256; i += 1) {
+        OUTREG(R128_PM4_MICROCODE_DATAH, R128Microcode[i*2]);
+        OUTREG(R128_PM4_MICROCODE_DATAL, R128Microcode[i*2 + 1]);
+    }
+}
+
+/* Initialize the CCE state, and start the CCE (if used by the X server) */
+static void R128DRICCEInit(ScrnInfoPtr pScrn)
+{
+    R128InfoPtr    info     = R128PTR(pScrn);
+    unsigned char *R128MMIO = info->MMIO;
+
+				/* CCEMode is initialized in r128_driver.c */
+    switch (info->CCEMode) {
+    case R128_PM4_NONPM4:                 info->CCEFifoSize = 0;   break;
+    case R128_PM4_192PIO:                 info->CCEFifoSize = 192; break;
+    case R128_PM4_192BM:                  info->CCEFifoSize = 192; break;
+    case R128_PM4_128PIO_64INDBM:         info->CCEFifoSize = 128; break;
+    case R128_PM4_128BM_64INDBM:          info->CCEFifoSize = 128; break;
+    case R128_PM4_64PIO_128INDBM:         info->CCEFifoSize = 64;  break;
+    case R128_PM4_64BM_128INDBM:          info->CCEFifoSize = 64;  break;
+    case R128_PM4_64PIO_64VCBM_64INDBM:   info->CCEFifoSize = 64;  break;
+    case R128_PM4_64BM_64VCBM_64INDBM:    info->CCEFifoSize = 64;  break;
+    case R128_PM4_64PIO_64VCPIO_64INDPIO: info->CCEFifoSize = 64;  break;
+    }
+
+    if (info->CCE2D) {
+				/* Make sure the CCE is on for the X server */
+	R128CCEStart(pScrn);
+    } else {
+				/* Make sure the CCE is off for the X server */
+	OUTREG(R128_PM4_MICRO_CNTL, 0);
+	OUTREG(R128_PM4_BUFFER_CNTL, R128_PM4_NONPM4);
+    }
+}
+
+/* Initialize the screen-specific data structures for the DRI and the
+   Rage 128.  This is the main entry point to the device-specific
+   initialization code.  It calls device-independent DRI functions to
+   create the DRI data structures and initialize the DRI state. */
+Bool R128DRIScreenInit(ScreenPtr pScreen)
+{
+    ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
+    R128InfoPtr pR128 = R128PTR(pScrn);
+    DRIInfoPtr  pDRIInfo;
+    R128DRIPtr  pR128DRI;
+    
+    switch (pR128->pixel_code) {
+    case 8:
+	/* These modes are not supported (yet). */
+    case 15:
+    case 24:
+	return FALSE;
+
+	/* Only 16 and 32 color depths are supports currently. */
+    case 16:
+    case 32:
+	break;
+    }
+
+    /* Create the DRI data structure, and fill it in before calling the
+       DRIScreenInit(). */
+    if (!(pDRIInfo = DRICreateInfoRec())) return FALSE;
+
+    pR128->pDRIInfo                      = pDRIInfo;
+    pDRIInfo->drmDriverName              = R128_NAME;
+    pDRIInfo->clientDriverName           = R128_NAME;
+    pDRIInfo->busIdString                = xalloc(64);
+    sprintf(pDRIInfo->busIdString,
+	    "PCI:%d:%d:%d",
+	    pR128->PciInfo->bus,
+	    pR128->PciInfo->device,
+	    pR128->PciInfo->func);
+    pDRIInfo->ddxDriverMajorVersion      = R128_VERSION_MAJOR;
+    pDRIInfo->ddxDriverMinorVersion      = R128_VERSION_MINOR;
+    pDRIInfo->ddxDriverPatchVersion      = R128_VERSION_PATCH;
+    pDRIInfo->frameBufferPhysicalAddress = pR128->LinearAddr;
+    pDRIInfo->frameBufferSize            = pR128->FbMapSize;
+    pDRIInfo->frameBufferStride          = (pScrn->displayWidth
+					    * pR128->pixel_bytes);
+    pDRIInfo->ddxDrawableTableEntry      = R128_MAX_DRAWABLES;
+    pDRIInfo->maxDrawableTableEntry      = (SAREA_MAX_DRAWABLES
+					    < R128_MAX_DRAWABLES
+					    ? SAREA_MAX_DRAWABLES
+					    : R128_MAX_DRAWABLES);
+    
+#ifdef NOT_DONE
+    /* FIXME: Need to extend DRI protocol to pass this size back to
+     * client for SAREA mapping that includes a device private record
+     */
+    pDRIInfo->SAREASize = 
+	((sizeof(XF86DRISAREARec) + 0xfff) & 0x1000); /* round to page */
+    /* + shared memory device private rec */
+#else
+    /* For now the mapping works by using a fixed size defined
+     * in the SAREA header
+     */
+    if (sizeof(XF86DRISAREARec)+sizeof(R128SAREAPriv)>SAREA_MAX) {
+	ErrorF("Data does not fit in SAREA\n");
+	return FALSE;
+    }
+    pDRIInfo->SAREASize = SAREA_MAX;
+#endif
+    
+    if (!(pR128DRI = (R128DRIPtr)xnfcalloc(sizeof(R128DRIRec),1))) {
+	DRIDestroyInfoRec(pR128->pDRIInfo);
+	pR128->pDRIInfo = NULL;
+	return FALSE;
+    }
+    pDRIInfo->devPrivate     = pR128DRI;
+    pDRIInfo->devPrivateSize = sizeof(R128DRIRec);
+    pDRIInfo->contextSize    = sizeof(R128DRIContextRec);
+    
+    pDRIInfo->CreateContext  = R128CreateContext;
+    pDRIInfo->DestroyContext = R128DestroyContext;
+    pDRIInfo->SwapContext    = R128DRISwapContext;
+    pDRIInfo->InitBuffers    = R128DRIInitBuffers;
+    pDRIInfo->MoveBuffers    = R128DRIMoveBuffers;
+    pDRIInfo->bufferRequests = DRI_ALL_WINDOWS;
+
+    if (!DRIScreenInit(pScreen, pDRIInfo, &pR128->drmFD)) {
+	xfree(pDRIInfo->devPrivate);
+	pDRIInfo->devPrivate = NULL;
+	DRIDestroyInfoRec(pDRIInfo);
+	pDRIInfo = NULL;
+	return FALSE;
+    }
+
+				/* Initialize AGP */
+    if (!pR128->IsPCI && !R128DRIAgpInit(pR128, pScreen)) {
+	R128DRICloseScreen(pScreen);
+	return FALSE;
+    }
+
+				/* DRIScreenInit doesn't add all the
+                                   common mappings.  Add additional
+                                   mappings here. */
+    if (!R128DRIMapInit(pR128, pScreen)) {
+	R128DRICloseScreen(pScreen);
+	return FALSE;
+    }
+
+				/* Initialize the ring buffer */
+    if (!pR128->IsPCI) R128DRICCEInitRingBuffer(pScrn);
+
+				/* Initialize the kernel data structures */
+    if (!R128DRIKernelInit(pR128, pScreen)) {
+	R128DRICloseScreen(pScreen);
+	return FALSE;
+    }
+
+				/* Initialize vertex buffers list */
+    if (!pR128->IsPCI && !R128DRIBufInit(pR128, pScreen)) {
+	R128DRICloseScreen(pScreen);
+	return FALSE;
+    }
+
+				/* FIXME: When are these mappings unmapped? */
+
+    if (!R128InitVisualConfigs(pScreen)) {
+	R128DRICloseScreen(pScreen);
+	return FALSE;
+    }
+    xf86DrvMsg(pScrn->scrnIndex, X_INFO, "Visual configs initialized\n");
+
+				/* Load the CCE Microcode */
+    R128DRILoadMicrocode(pScrn);
+
+				/* Reset the Graphics Engine */
+    R128EngineReset(pScrn);
+
+    return TRUE;
+}
+
+/* Finish initializing the device-dependent DRI state, and call
+   DRIFinishScreenInit() to complete the device-independent DRI
+   initialization. */
+Bool R128DRIFinishScreenInit(ScreenPtr pScreen)
+{
+    ScrnInfoPtr      pScrn      = xf86Screens[pScreen->myNum];
+    R128InfoPtr      pR128      = R128PTR(pScrn);
+    R128SAREAPrivPtr pSAREAPriv;
+    R128DRIPtr       pR128DRI;
+    
+    /* Init and start the CCE */
+    R128DRICCEInit(pScrn);
+
+    pSAREAPriv = (R128SAREAPrivPtr)DRIGetSAREAPrivate(pScreen);
+    memset(pSAREAPriv, 0, sizeof(*pSAREAPriv));
+
+    pR128->pDRIInfo->driverSwapMethod = DRI_HIDE_X_CONTEXT;
+    /* pR128->pDRIInfo->driverSwapMethod = DRI_SERVER_SWAP; */
+    
+    pR128DRI                 = (R128DRIPtr)pR128->pDRIInfo->devPrivate;
+    pR128DRI->registerHandle = pR128->registerHandle;
+    pR128DRI->registerSize   = pR128->registerSize;
+
+    pR128DRI->ringHandle     = pR128->ringHandle;
+    pR128DRI->ringMapSize    = pR128->ringMapSize;
+    pR128DRI->ringSize       = pR128->ringSize*1024*1024;
+
+    pR128DRI->ringReadPtrHandle = pR128->ringReadPtrHandle;
+    pR128DRI->ringReadMapSize   = pR128->ringReadMapSize;
+
+    pR128DRI->vbHandle       = pR128->vbHandle;
+    pR128DRI->vbMapSize      = pR128->vbMapSize;
+    pR128DRI->vbOffset       = pR128->vbStart;
+    pR128DRI->vbBufSize      = pR128->vbBufSize;
+
+    pR128DRI->indHandle      = pR128->indHandle;
+    pR128DRI->indMapSize     = pR128->indMapSize;
+
+    pR128DRI->agpTexHandle   = pR128->agpTexHandle;
+    pR128DRI->agpTexMapSize  = pR128->agpTexMapSize;
+    pR128DRI->log2AGPTexGran = pR128->log2AGPTexGran;
+    pR128DRI->agpTexOffset   = pR128->agpTexStart;
+
+    pR128DRI->deviceID       = pR128->Chipset;
+    pR128DRI->width          = pScrn->virtualX;
+    pR128DRI->height         = pScrn->virtualY;
+    pR128DRI->depth          = pScrn->depth;
+    pR128DRI->bpp            = pScrn->bitsPerPixel;
+
+    pR128DRI->fbX            = pR128->fbX;
+    pR128DRI->fbY            = pR128->fbY;
+    pR128DRI->backX          = pR128->backX;
+    pR128DRI->backY          = pR128->backY;
+    pR128DRI->depthX         = pR128->depthX;
+    pR128DRI->depthY         = pR128->depthY;
+    pR128DRI->textureX       = pR128->textureX;
+    pR128DRI->textureY       = pR128->textureY;
+    pR128DRI->textureSize    = pR128->textureSize;
+    pR128DRI->log2TexGran    = pR128->log2TexGran;
+
+    pR128DRI->IsPCI          = pR128->IsPCI;
+
+    pR128DRI->CCEMode        = pR128->CCEMode;
+    pR128DRI->CCEFifoSize    = pR128->CCEFifoSize;
+
+    xf86DrvMsg(pScrn->scrnIndex, X_INFO, "0x%08lx %d\n",
+	       pR128DRI->registerHandle, pR128DRI->registerSize);
+    return DRIFinishScreenInit(pScreen);
+}
+
+/* The screen is being closed, so clean up any state and free any
+   resources used by the DRI. */
+void R128DRICloseScreen(ScreenPtr pScreen)
+{
+    ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
+    R128InfoPtr pR128 = R128PTR(pScrn);
+
+				/* Stop the CCE if it is still in use */
+    if (pR128->CCE2D) R128CCEStop(pScrn);
+
+				/* De-allocate vertex buffers */
+    if (pR128->vbBufs) {
+	drmUnmapBufs(pR128->vbBufs);
+	pR128->vbBufs = NULL;
+    }
+
+				/* De-allocate all kernel resources */
+    drmR128CleanupCCE(pR128->drmFD);
+
+				/* De-allocate all AGP resources */
+    if (pR128->agpTex) {
+	drmUnmap(pR128->agpTex, pR128->agpTexMapSize);
+	pR128->agpTex = NULL;
+    }
+    if (pR128->ind) {
+	drmUnmap(pR128->ind, pR128->indMapSize);
+	pR128->ind = NULL;
+    }
+    if (pR128->vb) {
+	drmUnmap(pR128->vb, pR128->vbMapSize);
+	pR128->vb = NULL;
+    }
+    if (pR128->ringReadPtr) {
+	drmUnmap(pR128->ringReadPtr, pR128->ringReadMapSize);
+	pR128->ringReadPtr = NULL;
+    }
+    if (pR128->ring) {
+	drmUnmap(pR128->ring, pR128->ringMapSize);
+	pR128->ring = NULL;
+    }
+    if (pR128->agpMemHandle) {
+	drmAgpUnbind(pR128->drmFD, pR128->agpMemHandle);
+	drmAgpFree(pR128->drmFD, pR128->agpMemHandle);
+	pR128->agpMemHandle = 0;
+	drmAgpRelease(pR128->drmFD);
+    }
+    
+				/* De-allocate all DRI resources */
+    DRICloseScreen(pScreen);
+    
+				/* De-allocate all DRI data structures */
+    if (pR128->pDRIInfo) {
+	if (pR128->pDRIInfo->devPrivate) {
+	    xfree(pR128->pDRIInfo->devPrivate);
+	    pR128->pDRIInfo->devPrivate = NULL;
+	}
+	DRIDestroyInfoRec(pR128->pDRIInfo);
+	pR128->pDRIInfo = NULL;
+    }
+    if (pR128->pVisualConfigs) {
+	xfree(pR128->pVisualConfigs);
+	pR128->pVisualConfigs = NULL;
+    }
+    if (pR128->pVisualConfigsPriv) {
+	xfree(pR128->pVisualConfigsPriv);
+	pR128->pVisualConfigsPriv = NULL;
+    }
+}
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_dri.h b/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_dri.h
new file mode 100644
index 000000000..533aadb2f
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_dri.h
@@ -0,0 +1,116 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *   Rickard E. Faith <faith@precisioninsight.com>
+ *
+ */
+
+#ifndef _R128_DRI_
+#define _R128_DRI_
+
+#include <xf86drm.h>
+
+/* DRI Driver defaults */
+#define R128_DEFAULT_CCE_PIO_MODE R128_PM4_64PIO_64VCBM_64INDBM
+#define R128_DEFAULT_CCE_BM_MODE  R128_PM4_64BM_64VCBM_64INDBM
+#define R128_DEFAULT_AGP_MODE     2
+#define R128_DEFAULT_AGP_SIZE     8 /* MB (must be a power of 2 and > 4MB) */
+#define R128_DEFAULT_RING_SIZE    1 /* MB (must be page aligned) */
+#define R128_DEFAULT_VB_SIZE      1 /* MB (must be page aligned) */
+#define R128_DEFAULT_IND_SIZE     1 /* MB (must be page aligned) */
+#define R128_DEFAULT_AGP_TEX_SIZE 1 /* MB (must be page aligned) */
+
+#define R128_DEFAULT_VB_BUF_SIZE  16384  /* bytes */
+#define R128_DEFAULT_CCE_TIMEOUT  10000  /* usecs */
+
+#define R128_AGP_MAX_MODE         2
+
+#define R128CCE_USE_RING_BUFFER(m)                                        \
+(((m) == R128_PM4_192BM) ||                                               \
+ ((m) == R128_PM4_128BM_64INDBM) ||                                       \
+ ((m) == R128_PM4_64BM_128INDBM) ||                                       \
+ ((m) == R128_PM4_64BM_64VCBM_64INDBM))
+
+typedef struct {
+    /* MMIO register data */
+    drmHandle     registerHandle;
+    drmSize       registerSize;
+
+    /* CCE ring buffer data */
+    drmHandle     ringHandle;
+    drmSize       ringMapSize;
+    int           ringSize;
+
+    /* CCE ring read pointer data */
+    drmHandle     ringReadPtrHandle;
+    drmSize       ringReadMapSize;
+
+    /* CCE vertex buffer data */
+    drmHandle     vbHandle;
+    drmSize       vbMapSize;
+    int           vbOffset;
+    int           vbBufSize;
+
+    /* CCE indirect buffer data */
+    drmHandle     indHandle;
+    drmSize       indMapSize;
+
+    /* CCE AGP Texture data */
+    drmHandle     agpTexHandle;
+    drmSize       agpTexMapSize;
+    int           log2AGPTexGran;
+    int           agpTexOffset;
+
+    /* DRI screen private data */
+    int           deviceID;	/* PCI device ID */
+    int           width;	/* Width in pixels of display */
+    int           height;	/* Height in scanlines of display */
+    int           depth;	/* Depth of display (8, 15, 16, 24) */
+    int           bpp;		/* Bit depth of display (8, 16, 24, 32) */
+
+    int           fbX;		/* Start of frame buffer */
+    int           fbY;
+    int           backX;	/* Start of shared back buffer */
+    int           backY;
+    int           depthX;	/* Start of shared depth buffer */
+    int           depthY;
+    int           textureX;	/* Start of texture data in frame buffer */
+    int           textureY;
+    int           textureSize;
+    int           log2TexGran;
+
+    int           IsPCI;	/* Current card is a PCI card */
+
+    int           CCEMode;	/* CCE mode that server/clients use */
+    int           CCEFifoSize;	/* Size of the CCE command FIFO */
+} R128DRIRec, *R128DRIPtr;
+
+#endif
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_dripriv.h b/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_dripriv.h
new file mode 100644
index 000000000..acec5e269
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_dripriv.h
@@ -0,0 +1,54 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Rickard E. Faith <faith@precisioninsight.com>
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#ifndef _R128_DRIPRIV_H_
+#define _R128_DRIPRIV_H_
+
+#define R128_MAX_DRAWABLES 256
+
+extern void GlxSetVisualConfigs(int nconfigs, __GLXvisualConfig *configs,
+				void **configprivs);
+
+typedef struct {
+    /* Nothing here yet */
+    int dummy;
+} R128ConfigPrivRec, *R128ConfigPrivPtr;
+
+typedef struct {
+    /* Nothing here yet */
+    int dummy;
+} R128DRIContextRec, *R128DRIContextPtr;
+
+#endif
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_driver.c b/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_driver.c
index b5da4cb6e..f30d86aa8 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_driver.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_driver.c
@@ -1,8 +1,8 @@
 /* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/r128/r128_driver.c,v 1.26 2000/03/06 23:17:44 martin Exp $ */
 /**************************************************************************
 
-Copyright 1999 ATI Technologies Inc. and Precision Insight, Inc.,
-                                         Cedar Park, Texas. 
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
 All Rights Reserved.
 
 Permission is hereby granted, free of charge, to any person obtaining a
@@ -107,6 +107,19 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 				/* VESA support */
 #include "vbe.h"
 
+				/* DRI support */
+#ifdef XF86DRI
+#include "GL/glxint.h"
+#include "xf86drm.h"
+#include "sarea.h"
+#define _XF86DRI_SERVER_
+#include "xf86dri.h"
+#include "dri.h"
+#include "r128_dri.h"
+#include "r128_dripriv.h"
+#include "r128_sarea.h"
+#endif
+
 				/* Driver data structures */
 #include "r128.h"
 #include "r128_reg.h"
@@ -176,17 +189,41 @@ typedef enum {
   OPTION_HW_CURSOR,
   OPTION_DAC_6BIT,
   OPTION_DAC_8BIT,
+#ifdef XF86DRI
+  OPTION_IS_PCI,
+  OPTION_CCE_PIO,
+  OPTION_NO_SECURITY,
+  OPTION_USEC_TIMEOUT,
+  OPTION_AGP_MODE,
+  OPTION_AGP_SIZE,
+  OPTION_RING_SIZE,
+  OPTION_VERT_SIZE,
+  OPTION_VBUF_SIZE,
+  OPTION_USE_CCE_2D,
+#endif
   OPTION_FBDEV
 } R128Opts;
 
 static OptionInfoRec R128Options[] = {
-  { OPTION_NOACCEL,   "NoAccel",  OPTV_BOOLEAN, {0}, FALSE },
-  { OPTION_SW_CURSOR, "SWcursor", OPTV_BOOLEAN, {0}, FALSE },
-  { OPTION_HW_CURSOR, "HWcursor", OPTV_BOOLEAN, {0}, TRUE  },
-  { OPTION_DAC_6BIT,  "Dac6Bit",  OPTV_BOOLEAN, {0}, FALSE },
-  { OPTION_DAC_8BIT,  "Dac8Bit",  OPTV_BOOLEAN, {0}, TRUE  },
-  { OPTION_FBDEV,     "UseFBDev", OPTV_BOOLEAN, {0}, FALSE },
-  { -1,               NULL,       OPTV_NONE,    {0}, FALSE }
+  { OPTION_NOACCEL,      "NoAccel",          OPTV_BOOLEAN, {0}, FALSE },
+  { OPTION_SW_CURSOR,    "SWcursor",         OPTV_BOOLEAN, {0}, FALSE },
+  { OPTION_HW_CURSOR,    "HWcursor",         OPTV_BOOLEAN, {0}, TRUE  },
+  { OPTION_DAC_6BIT,     "Dac6Bit",          OPTV_BOOLEAN, {0}, FALSE },
+  { OPTION_DAC_8BIT,     "Dac8Bit",          OPTV_BOOLEAN, {0}, TRUE  },
+#ifdef XF86DRI
+  { OPTION_IS_PCI,       "ForcePCIMode",     OPTV_BOOLEAN, {0}, FALSE },
+  { OPTION_CCE_PIO,      "CCEPIOMode",       OPTV_BOOLEAN, {0}, FALSE },
+  { OPTION_NO_SECURITY,  "CCENoSecurity",    OPTV_BOOLEAN, {0}, FALSE },
+  { OPTION_USEC_TIMEOUT, "CCEusecTimeout",   OPTV_INTEGER, {0}, FALSE },
+  { OPTION_AGP_MODE,     "AGPMode",          OPTV_INTEGER, {0}, FALSE },
+  { OPTION_AGP_SIZE,     "AGPSize",          OPTV_INTEGER, {0}, FALSE },
+  { OPTION_RING_SIZE,    "RingSize",         OPTV_INTEGER, {0}, FALSE },
+  { OPTION_VERT_SIZE,    "VBListSize",       OPTV_INTEGER, {0}, FALSE },
+  { OPTION_VBUF_SIZE,    "VBSize",           OPTV_INTEGER, {0}, FALSE },
+  { OPTION_USE_CCE_2D,   "UseCCEfor2D",      OPTV_BOOLEAN, {0}, FALSE },
+#endif
+  { OPTION_FBDEV,        "UseFBDev",         OPTV_BOOLEAN, {0}, FALSE },
+  { -1,                  NULL,               OPTV_NONE,    {0}, FALSE }
 };
 
 R128RAMRec R128RAM[] = {	/* Memory Specifications
@@ -299,6 +336,38 @@ static const char *ramdacSymbols[] = {
     NULL
 };
 
+#ifdef XF86DRI
+static const char *drmSymbols[] = {
+    "drmAddBufs",
+    "drmAddMap",
+    "drmAvailable",
+    "drmCtlAddCommand",
+    "drmCtlInstHandler",
+    "drmGetInterruptFromBusID",
+    "drmMapBufs",
+    "drmMarkBufs",
+    "drmUnmapBufs",
+    NULL
+};
+
+static const char *driSymbols[] = {
+    "DRIGetDrawableIndex",
+    "DRIFinishScreenInit",
+    "DRIDestroyInfoRec",
+    "DRICloseScreen",
+    "DRIDestroyInfoRec",
+    "DRIScreenInit",
+    "DRIDestroyInfoRec",
+    "DRICreateInfoRec",
+    "DRILock",
+    "DRIUnlock",
+    "DRIGetSAREAPrivate",
+    "DRIGetContext",
+    "GlxSetVisualConfigs",
+    NULL
+};
+#endif
+
 static MODULESETUPPROTO(R128Setup);
 
 static XF86ModuleVersionInfo R128VersRec =
@@ -347,6 +416,10 @@ static pointer R128Setup(pointer module, pointer opts, int *errmaj,
 			  xaaSymbols, 
                           xf8_32bppSymbols,
 			  ramdacSymbols,
+#ifdef XF86DRI
+			  drmSymbols,
+			  driSymbols,
+#endif
 			  fbdevHWSymbols,
 			  vbeSymbols,
                           0 /* ddcsymbols */,
@@ -518,7 +591,7 @@ static void R128Unblank(ScrnInfoPtr pScrn)
 }
 
 /* Compute log base 2 of val. */
-static int R128MinBits(int val)
+int R128MinBits(int val)
 {
     int bits;
 
@@ -779,6 +852,7 @@ static Bool R128PreInitConfig(ScrnInfoPtr pScrn)
     int           offset = 0;	/* RAM Type */
     MessageType   from;
     unsigned char *R128MMIO;
+
 				/* Chipset */
     from = X_PROBED;
     if (dev->chipset && *dev->chipset) {
@@ -876,7 +950,7 @@ static Bool R128PreInitConfig(ScrnInfoPtr pScrn)
     switch (info->MemCntl & 0x3) {
     case 0:			/* SDR SGRAM 1:1 */
 	switch (info->Chipset) {
-	case PCI_CHIP_RAGE128RE: 
+	case PCI_CHIP_RAGE128RE:
 	case PCI_CHIP_RAGE128RF: offset = 0; break; /* 128-bit SDR SGRAM 1:1 */
 	case PCI_CHIP_RAGE128RK:
 	case PCI_CHIP_RAGE128RL:
@@ -895,12 +969,28 @@ static Bool R128PreInitConfig(ScrnInfoPtr pScrn)
 		   pScrn->videoRam);
 	from             = X_CONFIG;
 	pScrn->videoRam  = dev->videoRam;
-    } 
+    }
     pScrn->videoRam  &= ~1023;
     info->FbMapSize  = pScrn->videoRam * 1024;
     xf86DrvMsg(pScrn->scrnIndex, from,
 	       "VideoRAM: %d kByte (%s)\n", pScrn->videoRam, info->ram->name);
 
+#ifdef XF86DRI
+				/* AGP/PCI */
+    if (xf86ReturnOptValBool(R128Options, OPTION_IS_PCI, FALSE)) {
+	info->IsPCI = TRUE;
+	xf86DrvMsg(pScrn->scrnIndex, X_CONFIG, "Forced into PCI-only mode\n");
+    } else {
+	switch (info->Chipset) {
+	case PCI_CHIP_RAGE128RE:
+	case PCI_CHIP_RAGE128RK: info->IsPCI = TRUE;  break;
+	case PCI_CHIP_RAGE128RF:
+	case PCI_CHIP_RAGE128RL:
+	case PCI_CHIP_RAGE128PF:
+	default:                 info->IsPCI = FALSE; break;
+	}
+    }
+#endif
     return TRUE;
 }
 
@@ -1053,6 +1143,143 @@ static Bool R128PreInitInt10(ScrnInfoPtr pScrn)
     return TRUE;
 }
 
+#ifdef XF86DRI
+static Bool R128PreInitDRI(ScrnInfoPtr pScrn)
+{
+    R128InfoPtr   info = R128PTR(pScrn);    
+
+    if (info->IsPCI) {
+	xf86DrvMsg(pScrn->scrnIndex, X_CONFIG, "CCE in PIO mode\n");
+	info->CCEMode = R128_DEFAULT_CCE_PIO_MODE;
+    } else if (xf86ReturnOptValBool(R128Options, OPTION_CCE_PIO, FALSE)) {
+	xf86DrvMsg(pScrn->scrnIndex, X_CONFIG, "CCE in PIO mode\n");
+	info->CCEMode = R128_DEFAULT_CCE_PIO_MODE;
+    } else {
+	xf86DrvMsg(pScrn->scrnIndex, X_CONFIG, "CCE in BM mode\n");
+	info->CCEMode = R128_DEFAULT_CCE_BM_MODE;
+    }
+
+    if (xf86ReturnOptValBool(R128Options, OPTION_USE_CCE_2D, FALSE)) {
+	xf86DrvMsg(pScrn->scrnIndex, X_CONFIG, "Using CCE for 2D\n");
+	info->CCE2D = TRUE;
+    } else {
+	info->CCE2D = FALSE;
+    }
+
+    if (xf86ReturnOptValBool(R128Options, OPTION_NO_SECURITY, FALSE)) {
+	xf86DrvMsg(pScrn->scrnIndex, X_CONFIG,
+		   "WARNING!!!  CCE Security checks disabled!!! **********\n");
+	info->CCESecure = FALSE;
+    } else {
+	info->CCESecure = TRUE;
+    }
+
+    info->agpMode        = R128_DEFAULT_AGP_MODE;
+    info->agpSize        = R128_DEFAULT_AGP_SIZE;
+    info->ringSize       = R128_DEFAULT_RING_SIZE;
+    info->vbSize         = R128_DEFAULT_VB_SIZE;
+    info->indSize        = R128_DEFAULT_IND_SIZE;
+    info->agpTexSize     = R128_DEFAULT_AGP_TEX_SIZE;
+
+    info->vbBufSize      = R128_DEFAULT_VB_BUF_SIZE;
+
+    info->CCEusecTimeout = R128_DEFAULT_CCE_TIMEOUT;
+
+    if (!info->IsPCI) {
+	if (xf86GetOptValInteger(R128Options,
+				 OPTION_AGP_MODE, &(info->agpMode))) {
+	    if (info->agpMode < 1 || info->agpMode > R128_AGP_MAX_MODE) {
+		xf86DrvMsg(pScrn->scrnIndex, X_ERROR,
+			   "Illegal AGP Mode: %d\n", info->agpMode);
+		return FALSE;
+	    }
+	    xf86DrvMsg(pScrn->scrnIndex, X_CONFIG,
+		       "Using AGP %dx mode\n", info->agpMode);
+	}
+
+	if (xf86GetOptValInteger(R128Options,
+				 OPTION_AGP_SIZE, (int *)&(info->agpSize))) {
+	    switch (info->agpSize) {
+	    case 4:
+	    case 8:
+	    case 16:
+	    case 32:
+	    case 64:
+	    case 128:
+	    case 256:
+		break;
+	    default:
+		xf86DrvMsg(pScrn->scrnIndex, X_ERROR,
+			   "Illegal AGP size: %d MB\n", info->agpSize);
+		return FALSE;
+	    }
+	}
+
+	if (xf86GetOptValInteger(R128Options,
+				 OPTION_RING_SIZE, &(info->ringSize))) {
+	    if (info->ringSize < 1 || info->ringSize >= info->agpSize) {
+		xf86DrvMsg(pScrn->scrnIndex, X_ERROR,
+			   "Illegal ring buffer size: %d MB\n",
+			   info->ringSize);
+		return FALSE;
+	    }
+	}
+
+	if (xf86GetOptValInteger(R128Options,
+				 OPTION_VERT_SIZE, &(info->vbSize))) {
+	    if (info->vbSize < 1 || info->vbSize >= info->agpSize) {
+		xf86DrvMsg(pScrn->scrnIndex, X_ERROR,
+			   "Illegal vertex buffers list size: %d MB\n",
+			   info->vbSize);
+		return FALSE;
+	    }
+	}
+
+	if (xf86GetOptValInteger(R128Options,
+				 OPTION_VBUF_SIZE, &(info->vbBufSize))) {
+	    int numBufs = info->vbSize*1024*1024/info->vbBufSize;
+	    if (numBufs < 2 || numBufs > 512) { /* FIXME: 512 is arbitrary */
+		xf86DrvMsg(pScrn->scrnIndex, X_ERROR,
+			   "Illegal individual vertex buffer size: %d bytes\n",
+			   info->vbBufSize);
+		return FALSE;
+	    }
+	}
+
+	if (info->ringSize + info->vbSize + info->indSize + info->agpTexSize >
+	    info->agpSize) {
+	    xf86DrvMsg(pScrn->scrnIndex, X_ERROR,
+		       "Buffers are too big for requested AGP space\n");
+	    return FALSE;
+	}
+
+	info->agpTexSize = info->agpSize - (info->ringSize +
+					    info->vbSize +
+					    info->indSize);
+
+	xf86DrvMsg(pScrn->scrnIndex, X_CONFIG,
+		   "Using %d MB AGP aperture\n", info->agpSize);
+	xf86DrvMsg(pScrn->scrnIndex, X_CONFIG,
+		   "Using %d MB for the ring buffer\n", info->ringSize);
+	xf86DrvMsg(pScrn->scrnIndex, X_CONFIG,
+		   "Using %d MB for vertex buffers\n", info->vbSize);
+	xf86DrvMsg(pScrn->scrnIndex, X_CONFIG,
+		   "Using %d MB for indirect buffers\n", info->indSize);
+	xf86DrvMsg(pScrn->scrnIndex, X_CONFIG,
+		   "Using %d MB for AGP textures\n", info->agpTexSize);
+	xf86DrvMsg(pScrn->scrnIndex, X_CONFIG,
+		   "Using %d byte vertex buffers\n", info->vbBufSize);
+    }
+
+    if (xf86GetOptValInteger(R128Options, OPTION_USEC_TIMEOUT,
+			     &(info->CCEusecTimeout))) {
+	/* This option checked by the R128 DRM kernel module */
+    }
+
+    return TRUE;
+}
+#endif
+
 extern xf86MonPtr ConfiguredMonitor;
 
 static void
@@ -1152,6 +1379,10 @@ static Bool R128PreInit(ScrnInfoPtr pScrn, int flags)
     
     if (!R128PreInitAccel(pScrn))     goto fail;
 
+#ifdef XF86DRI
+    if (!R128PreInitDRI(pScrn))       goto fail;
+#endif
+
     return TRUE;
 
   fail:
@@ -1231,9 +1462,20 @@ static Bool R128ScreenInit(int scrnIndex, ScreenPtr pScreen,
 
     R128TRACE(("R128ScreenInit %x %d\n", pScrn->memPhysBase, pScrn->fbOffset));
     
+#ifdef XF86DRI
+				/* Turn off the CCE for now. */
+    info->CCEInUse     = FALSE;
+#endif
+
     if (!R128MapMem(pScrn)) return FALSE;
     pScrn->fbOffset    = 0;
-    
+#ifdef XF86DRI
+    info->fbX          = 0;
+    info->fbY          = 0;
+#endif
+
+    info->PaletteSavedOnVT = FALSE;
+
     R128Save(pScrn);
     if (info->FBDev) {
 	if (!fbdevHWModeInit(pScrn, pScrn->currentMode)) return FALSE;
@@ -1251,6 +1493,18 @@ static Bool R128ScreenInit(int scrnIndex, ScreenPtr pScreen,
 			  pScrn->rgbBits,
 			  pScrn->defaultVisual)) return FALSE;
 
+#ifdef XF86DRI
+				/* Setup DRI after visuals have been
+                                   established, but before cfbScreenInit is
+                                   called.  cfbScreenInit will eventually
+                                   call the driver's InitGLXVisuals call
+                                   back. */
+    if (!xf86ReturnOptValBool(R128Options, OPTION_NOACCEL, FALSE))
+	info->directRenderingEnabled = R128DRIScreenInit(pScreen);
+    else
+	info->directRenderingEnabled = FALSE;
+#endif
+
 #ifdef USE_FB
     if (!fbScreenInit (pScreen, info->FB,
 		       pScrn->virtualX, pScrn->virtualY,
@@ -1358,6 +1612,101 @@ static Bool R128ScreenInit(int scrnIndex, ScreenPtr pScreen,
 	}
     }
 
+#ifdef XF86DRI
+				/* Allocate frame buffer space for the
+                                   shared back and depth buffers as well
+                                   as for local textures. */
+    if (info->directRenderingEnabled) {
+	FBAreaPtr fbarea;
+	int       width_bytes = pScrn->displayWidth * info->pixel_bytes;
+	int       maxy        = info->FbMapSize / width_bytes;
+	int       l;
+	
+				/* Allocate the shared back buffer */
+	if ((fbarea = xf86AllocateOffscreenArea(pScreen,
+						pScrn->virtualX,
+						pScrn->virtualY,
+						32, NULL, NULL, NULL))) {
+	    xf86DrvMsg(scrnIndex, X_INFO,
+		       "Reserved back buffer from (%d,%d) to (%d,%d)\n",
+		       fbarea->box.x1, fbarea->box.y1,
+		       fbarea->box.x2, fbarea->box.y2);
+
+	    info->backX = fbarea->box.x1;
+	    info->backY = fbarea->box.y1;
+	} else {
+	    xf86DrvMsg(scrnIndex, X_ERROR, "Unable to reserve back buffer\n");
+	    info->backX = -1;
+	    info->backY = -1;
+	}
+
+				/* Allocate the shared depth buffer */
+	if ((fbarea = xf86AllocateOffscreenArea(pScreen,
+						pScrn->virtualX,
+						pScrn->virtualY,
+						32, NULL, NULL, NULL))) {
+	    xf86DrvMsg(scrnIndex, X_INFO,
+		       "Reserved depth buffer from (%d,%d) to (%d,%d)\n",
+		       fbarea->box.x1, fbarea->box.y1,
+		       fbarea->box.x2, fbarea->box.y2);
+
+	    info->depthX = fbarea->box.x1;
+	    info->depthY = fbarea->box.y1;
+	} else {
+	    xf86DrvMsg(scrnIndex, X_ERROR, "Unable to reserve depth buffer\n");
+	    info->depthX = -1;
+	    info->depthY = -1;
+	}
+
+				/* Allocate local texture space */
+	if (((maxy - MemBox.y2 - 1) * width_bytes) >
+	    (pScrn->virtualX * pScrn->virtualY * 2 * info->pixel_bytes)) {
+	    info->textureX = 0;
+	    info->textureY = MemBox.y2 + 1;
+	    info->textureSize = (maxy - MemBox.y2 - 1) * width_bytes;
+
+	    l = R128MinBits((info->textureSize-1) / R128_NR_TEX_REGIONS);
+	    if (l < R128_LOG_TEX_GRANULARITY) l = R128_LOG_TEX_GRANULARITY;
+
+	    info->log2TexGran = l;
+	    info->textureSize = (info->textureSize >> l) << l;
+
+	    xf86DrvMsg(scrnIndex, X_INFO,
+		       "Reserved %d kb for textures: (%d,%d)-(%d,%d)\n",
+		       info->textureSize/1024,
+		       info->textureX,      info->textureY,
+		       pScrn->displayWidth, maxy);
+	} else if ((fbarea = xf86AllocateOffscreenArea(pScreen,
+						       pScrn->virtualX,
+						       pScrn->virtualY * 2,
+						       32,
+						       NULL, NULL, NULL))) {
+	    info->textureX = fbarea->box.x1;
+	    info->textureY = fbarea->box.y1;
+	    info->textureSize = ((fbarea->box.y2 - fbarea->box.y1) *
+				 (fbarea->box.x2 - fbarea->box.x1) *
+				 info->pixel_bytes);
+
+	    l = R128MinBits((info->textureSize-1) / R128_NR_TEX_REGIONS);
+	    if (l < R128_LOG_TEX_GRANULARITY) l = R128_LOG_TEX_GRANULARITY;
+
+	    info->log2TexGran = l;
+	    info->textureSize = (info->textureSize >> l) << l;
+
+	    xf86DrvMsg(scrnIndex, X_INFO,
+		       "Reserved %d kb for textures: (%d,%d)-(%d,%d)\n",
+		       info->textureSize/1024,
+		       fbarea->box.x1, fbarea->box.y1,
+		       fbarea->box.x2, fbarea->box.y2);
+	} else {
+	    xf86DrvMsg(scrnIndex, X_ERROR,
+		       "Unable to reserve texture space in frame buffer\n");
+	    info->textureX = -1;
+	    info->textureY = -1;
+	}
+    }
+#endif
+
 				/* Backing store setup */
     miInitializeBackingStore(pScreen);
     xf86SetBackingStore(pScreen);
@@ -1453,6 +1802,21 @@ static Bool R128ScreenInit(int scrnIndex, ScreenPtr pScreen,
     if (serverGeneration == 1)
 	xf86ShowUnusedOptions(pScrn->scrnIndex, pScrn->options);
 
+#ifdef XF86DRI
+				/* DRI finalization */
+    if (info->directRenderingEnabled) {
+				/* Now that mi, cfb, drm and others have
+                                   done their thing, complete the DRI
+                                   setup. */
+        info->directRenderingEnabled = R128DRIFinishScreenInit(pScreen);
+    }
+    if (info->directRenderingEnabled) {
+        xf86DrvMsg(pScrn->scrnIndex, X_INFO, "Direct rendering enabled\n");
+    } else {
+        xf86DrvMsg(pScrn->scrnIndex, X_INFO, "Direct rendering disabled\n");
+    }
+#endif
+
     return TRUE;
 }
 
@@ -1473,7 +1837,7 @@ static void R128RestoreCommonRegisters(ScrnInfoPtr pScrn, R128SavePtr restore)
     OUTREG(R128_GEN_INT_CNTL,         restore->gen_int_cntl);
     OUTREG(R128_CAP0_TRIG_CNTL,       restore->cap0_trig_cntl);
     OUTREG(R128_CAP1_TRIG_CNTL,       restore->cap1_trig_cntl);
-    OUTREG(R128_BUS_CNTL,	      restore->bus_cntl);
+    OUTREG(R128_BUS_CNTL,             restore->bus_cntl);
 }
 
 /* Write CRTC registers. */
@@ -1603,7 +1967,7 @@ static void R128SaveCommonRegisters(ScrnInfoPtr pScrn, R128SavePtr save)
     save->gen_int_cntl       = INREG(R128_GEN_INT_CNTL);
     save->cap0_trig_cntl     = INREG(R128_CAP0_TRIG_CNTL);
     save->cap1_trig_cntl     = INREG(R128_CAP1_TRIG_CNTL);
-    save->bus_cntl	     = INREG(R128_BUS_CNTL);
+    save->bus_cntl           = INREG(R128_BUS_CNTL);
 }
 
 /* Read CRTC registers. */
@@ -2025,7 +2389,7 @@ static Bool R128Init(ScrnInfoPtr pScrn, DisplayModePtr mode, R128SavePtr save)
     R128InitPLLRegisters(pScrn, save, mode, &info->pll, dot_clock);
     if (!R128InitDDARegisters(pScrn, save, mode, &info->pll, info))
 	return FALSE;
-    R128InitPalette(save, info);
+    if (!info->PaletteSavedOnVT) R128InitPalette(save, info);
     
     R128TRACE(("R128Init returns %p\n", save));
     return TRUE;
@@ -2037,6 +2401,7 @@ static Bool R128ModeInit(ScrnInfoPtr pScrn, DisplayModePtr mode)
     R128InfoPtr info      = R128PTR(pScrn);
     
     if (!R128Init(pScrn, mode, &info->ModeReg)) return FALSE;
+				/* FIXME?  DRILock/DRIUnlock here? */
     R128Blank(pScrn);
     R128RestoreMode(pScrn, &info->ModeReg);
     R128Unblank(pScrn);
@@ -2096,14 +2461,43 @@ static void R128AdjustFrame(int scrnIndex, int x, int y, int flags)
    mode. */
 static Bool R128EnterVT(int scrnIndex, int flags)
 {
-    ScrnInfoPtr pScrn = xf86Screens[scrnIndex];
+    ScrnInfoPtr pScrn   = xf86Screens[scrnIndex];
+    R128InfoPtr info  = R128PTR(pScrn);
 
     R128TRACE(("R128EnterVT\n"));
+#ifdef XF86DRI
+    if (R128PTR(pScrn)->directRenderingEnabled) {
+	R128CCEStart(pScrn);
+	DRIUnlock(pScrn->pScreen);
+    }
+#endif
     if (!R128ModeInit(pScrn, pScrn->currentMode)) return FALSE;
+    info->PaletteSavedOnVT = FALSE;
     R128AdjustFrame(scrnIndex, pScrn->frameX0, pScrn->frameY0, 0);
+
     return TRUE;
 }
 
+/* Called when VT switching away from the X server.  Restore the original
+   text mode. */
+static void R128LeaveVT(int scrnIndex, int flags)
+{
+    ScrnInfoPtr pScrn = xf86Screens[scrnIndex];
+    R128InfoPtr info  = R128PTR(pScrn);
+    R128SavePtr save  = &info->ModeReg;
+
+    R128TRACE(("R128LeaveVT\n"));
+#ifdef XF86DRI
+    if (R128PTR(pScrn)->directRenderingEnabled) {
+	DRILock(pScrn->pScreen, 0);
+	R128CCEStop(pScrn);
+    }
+#endif
+    R128SavePalette(pScrn, save);
+    info->PaletteSavedOnVT = TRUE;
+    R128Restore(pScrn);
+}
+
 static Bool
 R128EnterVTFBDev(int scrnIndex, int flags)
 {
@@ -2116,8 +2510,6 @@ R128EnterVTFBDev(int scrnIndex, int flags)
     return TRUE;
 }
 
-/* Called when VT switching away from the X server.  Restore the original
-   text mode. */
 static void R128LeaveVTFBDev(int scrnIndex, int flags)
 {
     ScrnInfoPtr pScrn = xf86Screens[scrnIndex];
@@ -2127,14 +2519,6 @@ static void R128LeaveVTFBDev(int scrnIndex, int flags)
     fbdevHWLeaveVT(scrnIndex,flags);
 }
 
-static void R128LeaveVT(int scrnIndex, int flags)
-{
-    ScrnInfoPtr pScrn = xf86Screens[scrnIndex];
-
-    R128TRACE(("R128LeaveVT\n"));
-    R128Restore(pScrn);
-}
-
 /* Called at the end of each server generation.  Restore the original text
    mode, unmap video memory, and unwrap and call the saved CloseScreen
    function.  */
@@ -2144,6 +2528,15 @@ static Bool R128CloseScreen(int scrnIndex, ScreenPtr pScreen)
     R128InfoPtr info  = R128PTR(pScrn);
 
     R128TRACE(("R128CloseScreen\n"));
+
+#ifdef XF86DRI
+				/* Disable direct rendering */
+    if (info->directRenderingEnabled) {
+	R128DRICloseScreen(pScreen);
+	info->directRenderingEnabled = FALSE;
+    }
+#endif
+    
     if (pScrn->vtSema) {
 	R128Restore(pScrn);
 	R128UnmapMem(pScrn);
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_reg.h b/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_reg.h
index 86652affa..144fd9efe 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_reg.h
+++ b/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_reg.h
@@ -1,8 +1,8 @@
 /* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/r128/r128_reg.h,v 1.6 2000/02/23 04:47:19 martin Exp $ */
 /**************************************************************************
 
-Copyright 1999 ATI Technologies Inc. and Precision Insight, Inc.,
-                                         Cedar Park, Texas. 
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas.
 All Rights Reserved.
 
 Permission is hereby granted, free of charge, to any person obtaining a
@@ -159,9 +159,20 @@ static inline unsigned short regr16(volatile unsigned long base_addr, unsigned l
 #define R128_AGP_APER_OFFSET              0x0178
 #define R128_AGP_BASE                     0x0170
 #define R128_AGP_CNTL                     0x0174
+#       define R128_AGP_APER_SIZE_256MB   (0x00 << 0)
+#       define R128_AGP_APER_SIZE_128MB   (0x20 << 0)
+#       define R128_AGP_APER_SIZE_64MB    (0x30 << 0)
+#       define R128_AGP_APER_SIZE_32MB    (0x38 << 0)
+#       define R128_AGP_APER_SIZE_16MB    (0x3c << 0)
+#       define R128_AGP_APER_SIZE_8MB     (0x3e << 0)
+#       define R128_AGP_APER_SIZE_4MB     (0x3f << 0)
+#       define R128_AGP_APER_SIZE_MASK    (0x3f << 0)
 #define R128_AGP_COMMAND                  0x0f58 /* PCI */
 #define R128_AGP_PLL_CNTL                 0x0010 /* PLL */
 #define R128_AGP_STATUS                   0x0f54 /* PCI */
+#       define R128_AGP_1X_MODE           0x01
+#       define R128_AGP_2X_MODE           0x02
+#       define R128_AGP_MODE_MASK         0x03
 #define R128_AMCGPIO_A_REG                0x01a0
 #define R128_AMCGPIO_EN_REG               0x01a8
 #define R128_AMCGPIO_MASK                 0x0194
@@ -169,6 +180,15 @@ static inline unsigned short regr16(volatile unsigned long base_addr, unsigned l
 #define R128_ATTRDR                       0x03c1 /* VGA */
 #define R128_ATTRDW                       0x03c0 /* VGA */
 #define R128_ATTRX                        0x03c0 /* VGA */
+#       define R128_AUX1_SC_EN            (1 << 0)
+#       define R128_AUX1_SC_MODE_OR       (0 << 1)
+#       define R128_AUX1_SC_MODE_NAND     (1 << 1)
+#       define R128_AUX2_SC_EN            (1 << 2)
+#       define R128_AUX2_SC_MODE_OR       (0 << 3)
+#       define R128_AUX2_SC_MODE_NAND     (1 << 3)
+#       define R128_AUX3_SC_EN            (1 << 4)
+#       define R128_AUX3_SC_MODE_OR       (0 << 5)
+#       define R128_AUX3_SC_MODE_NAND     (1 << 5)
 #define R128_AUX_SC_CNTL                  0x1660
 #define R128_AUX1_SC_BOTTOM               0x1670
 #define R128_AUX1_SC_LEFT                 0x1664
@@ -257,11 +277,12 @@ static inline unsigned short regr16(volatile unsigned long base_addr, unsigned l
 #define R128_BRUSH_SCALE                  0x1470
 #define R128_BRUSH_Y_X                    0x1474
 #define R128_BUS_CNTL                     0x0030
-#	define R128_BUS_RD_DISCARD_EN	  (1 << 24)
-#	define R128_BUS_RD_ABORT_EN	  (1 << 25)
-#	define R128_BUS_MSTR_DISCONNECT_EN (1 << 28)
-#	define R128_BUS_WRT_BURST         (1 << 29)
-#	define R128_BUS_READ_BURST	  (1 << 30)
+#       define R128_BUS_MASTER_DIS         (1 << 6)
+#       define R128_BUS_RD_DISCARD_EN      (1 << 24)
+#       define R128_BUS_RD_ABORT_EN        (1 << 25)
+#       define R128_BUS_MSTR_DISCONNECT_EN (1 << 28)
+#       define R128_BUS_WRT_BURST          (1 << 29)
+#       define R128_BUS_READ_BURST         (1 << 30)
 #define R128_BUS_CNTL1                    0x0034
 
 #define R128_CACHE_CNTL                   0x1724
@@ -297,6 +318,9 @@ static inline unsigned short regr16(volatile unsigned long base_addr, unsigned l
 #define R128_CONFIG_REG_APER_SIZE         0x0110
 #define R128_CONFIG_XSTRAP                0x00e4
 #define R128_CONSTANT_COLOR_C             0x1d34
+#       define R128_CONSTANT_COLOR_MASK   0x00ffffff
+#       define R128_CONSTANT_COLOR_ONE    0x00ffffff
+#       define R128_CONSTANT_COLOR_ZERO   0x00000000
 #define R128_CRC_CMDFIFO_ADDR             0x0740
 #define R128_CRC_CMDFIFO_DOUT             0x0744
 #define R128_CRTC_CRNT_FRAME              0x0214
@@ -360,8 +384,13 @@ static inline unsigned short regr16(volatile unsigned long base_addr, unsigned l
 #define R128_DEFAULT_SC_BOTTOM_RIGHT      0x16e8
 #       define R128_DEFAULT_SC_RIGHT_MAX  (0x1fff <<  0)
 #       define R128_DEFAULT_SC_BOTTOM_MAX (0x1fff << 16)
-#define R128_DESTINATION_3D_CLR_CMP_MSK   0x1824
+#define R128_FOG_3D_TABLE_START           0x1810
+#define R128_FOG_3D_TABLE_END             0x1814
+#define R128_FOG_3D_TABLE_DENSITY         0x181c
+#define R128_FOG_TABLE_INDEX              0x1a14
+#define R128_FOG_TABLE_DATA               0x1a18
 #define R128_DESTINATION_3D_CLR_CMP_VAL   0x1820
+#define R128_DESTINATION_3D_CLR_CMP_MSK   0x1824
 #define R128_DEVICE_ID                    0x0f02 /* PCI */
 #define R128_DP_BRUSH_BKGD_CLR            0x1478
 #define R128_DP_BRUSH_FRGD_CLR            0x147c
@@ -373,6 +402,7 @@ static inline unsigned short regr16(volatile unsigned long base_addr, unsigned l
 #       define R128_DST_Y_DIR_TOP_TO_BOTTOM (1 << 15)
 #       define R128_DST_X_DIR_LEFT_TO_RIGHT (1 << 31)
 #define R128_DP_DATATYPE                  0x16c4
+#       define R128_HOST_BIG_ENDIAN_EN    (1 << 29)
 #define R128_DP_GUI_MASTER_CNTL           0x146c
 #       define R128_GMC_SRC_PITCH_OFFSET_CNTL (1    <<  0)
 #       define R128_GMC_DST_PITCH_OFFSET_CNTL (1    <<  1)
@@ -388,6 +418,19 @@ static inline unsigned short regr16(volatile unsigned long base_addr, unsigned l
 #       define R128_GMC_BRUSH_8x8_COLOR       (10   <<  4)
 #       define R128_GMC_BRUSH_1X8_COLOR       (12   <<  4)
 #       define R128_GMC_BRUSH_SOLID_COLOR     (13   <<  4)
+#       define R128_GMC_BRUSH_NONE            (15   <<  4)
+#       define R128_GMC_DST_8BPP_CI           (2    <<  8)
+#       define R128_GMC_DST_15BPP             (3    <<  8)
+#       define R128_GMC_DST_16BPP             (4    <<  8)
+#       define R128_GMC_DST_24BPP             (5    <<  8)
+#       define R128_GMC_DST_32BPP             (6    <<  8)
+#       define R128_GMC_DST_8BPP_RGB          (7    <<  8)
+#       define R128_GMC_DST_Y8                (8    <<  8)
+#       define R128_GMC_DST_RGB8              (9    <<  8)
+#       define R128_GMC_DST_VYUY              (11   <<  8)
+#       define R128_GMC_DST_YVYU              (12   <<  8)
+#       define R128_GMC_DST_AYUV444           (14   <<  8)
+#       define R128_GMC_DST_ARGB4444          (15   <<  8)
 #       define R128_GMC_DST_DATATYPE_MASK     (0x0f <<  8)
 #       define R128_GMC_DST_DATATYPE_SHIFT    8
 #       define R128_GMC_SRC_DATATYPE_MASK       (3    << 12)
@@ -395,8 +438,11 @@ static inline unsigned short regr16(volatile unsigned long base_addr, unsigned l
 #       define R128_GMC_SRC_DATATYPE_MONO_FG_LA (1    << 12)
 #       define R128_GMC_SRC_DATATYPE_COLOR      (3    << 12)
 #       define R128_GMC_BYTE_PIX_ORDER        (1    << 14)
+#       define R128_GMC_BYTE_MSB_TO_LSB       (0    << 14)
 #       define R128_GMC_BYTE_LSB_TO_MSB       (1    << 14)
 #       define R128_GMC_CONVERSION_TEMP       (1    << 15)
+#       define R128_GMC_CONVERSION_TEMP_6500  (0    << 15)
+#       define R128_GMC_CONVERSION_TEMP_9300  (1    << 15)
 #       define R128_GMC_ROP3_MASK             (0xff << 16)
 #       define R128_DP_SRC_SOURCE_MASK        (7    << 24)
 #       define R128_DP_SRC_SOURCE_MEMORY      (2    << 24)
@@ -404,7 +450,7 @@ static inline unsigned short regr16(volatile unsigned long base_addr, unsigned l
 #       define R128_GMC_3D_FCN_EN             (1    << 27)
 #       define R128_GMC_CLR_CMP_CNTL_DIS      (1    << 28)
 #       define R128_AUX_CLIP_DIS              (1    << 29)
-#       define R128_GMC_WR_MSK_DS             (1    << 30)
+#       define R128_GMC_WR_MSK_DIS            (1    << 30)
 #       define R128_GMC_LD_BRUSH_Y_X          (1    << 31)
 #       define R128_ROP3_ZERO             0x00000000
 #       define R128_ROP3_DSa              0x00880000
@@ -456,6 +502,7 @@ static inline unsigned short regr16(volatile unsigned long base_addr, unsigned l
 #define R128_DST_PITCH                    0x1408
 #define R128_DST_PITCH_OFFSET             0x142c
 #define R128_DST_PITCH_OFFSET_C           0x1c80
+#       define R128_PITCH_SHIFT               21
 #define R128_DST_WIDTH                    0x140c
 #define R128_DST_WIDTH_HEIGHT             0x1598
 #define R128_DST_WIDTH_X                  0x1588
@@ -573,7 +620,6 @@ static inline unsigned short regr16(volatile unsigned long base_addr, unsigned l
 #define R128_MEM_VGA_RP_SEL               0x003c
 #define R128_MEM_VGA_WP_SEL               0x0038
 #define R128_MIN_GRANT                    0x0f3e /* PCI */
-#define R128_MISC_3D_STATE_CNTL_REG       0x1CA0
 #define R128_MM_DATA                      0x0004
 #define R128_MM_INDEX                     0x0000
 #define R128_MPLL_CNTL                    0x000e /* PLL */
@@ -592,7 +638,10 @@ static inline unsigned short regr16(volatile unsigned long base_addr, unsigned l
 #define R128_PC_DEBUG_MODE                0x1760
 #define R128_PC_GUI_CTLSTAT               0x1748
 #define R128_PC_GUI_MODE                  0x1744
+#       define R128_PC_IGNORE_UNIFY       (1 << 5)
 #define R128_PC_NGUI_CTLSTAT              0x0184
+#       define R128_PC_FLUSH_GUI          (3 << 0)
+#       define R128_PC_RI_GUI             (1 << 2)
 #       define R128_PC_FLUSH_ALL          0x00ff
 #       define R128_PC_BUSY               (1 << 31)
 #define R128_PC_NGUI_MODE                 0x0180
@@ -689,6 +738,538 @@ static inline unsigned short regr16(volatile unsigned long base_addr, unsigned l
 #define R128_XDLL_CNTL                    0x000c /* PLL */
 #define R128_XPLL_CNTL                    0x000b /* PLL */
 
+				/* Registers for CCE and Microcode Engine */
+#define R128_PM4_MICROCODE_ADDR           0x07d4
+#define R128_PM4_MICROCODE_RADDR          0x07d8
+#define R128_PM4_MICROCODE_DATAH          0x07dc
+#define R128_PM4_MICROCODE_DATAL          0x07e0
+
+#define R128_PM4_BUFFER_OFFSET            0x0700
+#define R128_PM4_BUFFER_CNTL              0x0704
+#       define R128_PM4_NONPM4                 (0  << 28)
+#       define R128_PM4_192PIO                 (1  << 28)
+#       define R128_PM4_192BM                  (2  << 28)
+#       define R128_PM4_128PIO_64INDBM         (3  << 28)
+#       define R128_PM4_128BM_64INDBM          (4  << 28)
+#       define R128_PM4_64PIO_128INDBM         (5  << 28)
+#       define R128_PM4_64BM_128INDBM          (6  << 28)
+#       define R128_PM4_64PIO_64VCBM_64INDBM   (7  << 28)
+#       define R128_PM4_64BM_64VCBM_64INDBM    (8  << 28)
+#       define R128_PM4_64PIO_64VCPIO_64INDPIO (15 << 28)
+#define R128_PM4_BUFFER_WM_CNTL           0x0708
+#       define R128_WMA_SHIFT                  0
+#       define R128_WMB_SHIFT                  8
+#       define R128_WMC_SHIFT                 16
+#       define R128_WB_WM_SHIFT               24
+#define R128_PM4_BUFFER_DL_RPTR_ADDR      0x070c
+#define R128_PM4_BUFFER_DL_RPTR           0x0710
+#define R128_PM4_BUFFER_DL_WPTR           0x0714
+#       define R128_PM4_BUFFER_DL_DONE    (1 << 31)
+#define R128_PM4_VC_FPU_SETUP             0x071c
+#       define R128_FRONT_DIR_CW          (0 <<  0)
+#       define R128_FRONT_DIR_CCW         (1 <<  0)
+#       define R128_FRONT_DIR_MASK        (1 <<  0)
+#       define R128_BACKFACE_CULL         (0 <<  1)
+#       define R128_BACKFACE_POINTS       (1 <<  1)
+#       define R128_BACKFACE_LINES        (2 <<  1)
+#       define R128_BACKFACE_SOLID        (3 <<  1)
+#       define R128_BACKFACE_MASK         (3 <<  1)
+#       define R128_FRONTFACE_CULL        (0 <<  3)
+#       define R128_FRONTFACE_POINTS      (1 <<  3)
+#       define R128_FRONTFACE_LINES       (2 <<  3)
+#       define R128_FRONTFACE_SOLID       (3 <<  3)
+#       define R128_FRONTFACE_MASK        (3 <<  3)
+#       define R128_FPU_COLOR_SOLID       (0 <<  5)
+#       define R128_FPU_COLOR_FLAT        (1 <<  5)
+#       define R128_FPU_COLOR_GOURAUD     (2 <<  5)
+#       define R128_FPU_COLOR_GOURAUD2    (3 <<  5)
+#       define R128_FPU_SUB_PIX_2BITS     (0 <<  7)
+#       define R128_FPU_SUB_PIX_4BITS     (1 <<  7)
+#       define R128_FPU_MODE_2D           (0 <<  8)
+#       define R128_FPU_MODE_3D           (1 <<  8)
+#       define R128_TRAP_BITS_DISABLE     (1 <<  9)
+#       define R128_EDGE_ANTIALIAS        (1 << 10)
+#       define R128_SUPERSAMPLE           (1 << 11)
+#       define R128_XFACTOR_2             (0 << 12)
+#       define R128_XFACTOR_4             (1 << 12)
+#       define R128_YFACTOR_2             (0 << 13)
+#       define R128_YFACTOR_4             (1 << 13)
+#       define R128_FLAT_SHADE_VERTEX_D3D (0 << 14)
+#       define R128_FLAT_SHADE_VERTEX_OGL (1 << 14)
+#       define R128_FPU_ROUND_TRUNCATE    (0 << 15)
+#       define R128_FPU_ROUND_NEAREST     (1 << 15)
+#       define R128_WM_SEL_8DW            (0 << 16)
+#       define R128_WM_SEL_16DW           (1 << 16)
+#       define R128_WM_SEL_32DW           (2 << 16)
+#define R128_PM4_STAT                     0x07b8
+#       define R128_PM4_FIFOCNT_MASK      0x0fff
+#       define R128_PM4_BUSY              (1 << 16)
+#       define R128_PM4_GUI_ACTIVE        (1 << 31)
+#define R128_PM4_BUFFER_ADDR              0x07f0
+#define R128_PM4_MICRO_CNTL               0x07fc
+#       define R128_PM4_MICRO_FREERUN     (1 << 30)
+#define R128_PM4_FIFO_DATA_EVEN           0x1000
+#define R128_PM4_FIFO_DATA_ODD            0x1004
+
 #define R128_SCALE_3D_CNTL                0x1a00
+#       define R128_SCALE_DITHER_ERR_DIFF         (0  <<  1)
+#       define R128_SCALE_DITHER_TABLE            (1  <<  1)
+#       define R128_TEX_CACHE_SIZE_FULL           (0  <<  2)
+#       define R128_TEX_CACHE_SIZE_HALF           (1  <<  2)
+#       define R128_DITHER_INIT_CURR              (0  <<  3)
+#       define R128_DITHER_INIT_RESET             (1  <<  3)
+#       define R128_ROUND_24BIT                   (1  <<  4)
+#       define R128_TEX_CACHE_DISABLE             (1  <<  5)
+#       define R128_SCALE_3D_NOOP                 (0  <<  6)
+#       define R128_SCALE_3D_SCALE                (1  <<  6)
+#       define R128_SCALE_3D_TEXMAP_SHADE         (2  <<  6)
+#       define R128_SCALE_PIX_BLEND               (0  <<  8)
+#       define R128_SCALE_PIX_REPLICATE           (1  <<  8)
+#       define R128_TEX_CACHE_SPLIT               (1  <<  9)
+#       define R128_APPLE_YUV_MODE                (1  << 10)
+#       define R128_TEX_CACHE_PALLETE_MODE        (1  << 11)
+#       define R128_ALPHA_COMB_ADD_CLAMP          (0  << 12)
+#       define R128_ALPHA_COMB_ADD_NCLAMP         (1  << 12)
+#       define R128_ALPHA_COMB_SUB_DST_SRC_CLAMP  (2  << 12)
+#       define R128_ALPHA_COMB_SUB_DST_SRC_NCLAMP (3  << 12)
+#       define R128_FOG_TABLE                     (1  << 14)
+#       define R128_SIGNED_DST_CLAMP              (1  << 15)
+#       define R128_ALPHA_BLEND_SRC_ZERO          (0  << 16)
+#       define R128_ALPHA_BLEND_SRC_ONE           (1  << 16)
+#       define R128_ALPHA_BLEND_SRC_SRCCOLOR      (2  << 16)
+#       define R128_ALPHA_BLEND_SRC_INVSRCCOLOR   (3  << 16)
+#       define R128_ALPHA_BLEND_SRC_SRCALPHA      (4  << 16)
+#       define R128_ALPHA_BLEND_SRC_INVSRCALPHA   (5  << 16)
+#       define R128_ALPHA_BLEND_SRC_DSTALPHA      (6  << 16)
+#       define R128_ALPHA_BLEND_SRC_INVDSTALPHA   (7  << 16)
+#       define R128_ALPHA_BLEND_SRC_DSTCOLOR      (8  << 16)
+#       define R128_ALPHA_BLEND_SRC_INVDSTCOLOR   (9  << 16)
+#       define R128_ALPHA_BLEND_SRC_SAT           (10 << 16)
+#       define R128_ALPHA_BLEND_SRC_BLEND         (11 << 16)
+#       define R128_ALPHA_BLEND_SRC_INVBLEND      (12 << 16)
+#       define R128_ALPHA_BLEND_DST_ZERO          (0  << 20)
+#       define R128_ALPHA_BLEND_DST_ONE           (1  << 20)
+#       define R128_ALPHA_BLEND_DST_SRCCOLOR      (2  << 20)
+#       define R128_ALPHA_BLEND_DST_INVSRCCOLOR   (3  << 20)
+#       define R128_ALPHA_BLEND_DST_SRCALPHA      (4  << 20)
+#       define R128_ALPHA_BLEND_DST_INVSRCALPHA   (5  << 20)
+#       define R128_ALPHA_BLEND_DST_DSTALPHA      (6  << 20)
+#       define R128_ALPHA_BLEND_DST_INVDSTALPHA   (7  << 20)
+#       define R128_ALPHA_BLEND_DST_DSTCOLOR      (8  << 20)
+#       define R128_ALPHA_BLEND_DST_INVDSTCOLOR   (9  << 20)
+#       define R128_ALPHA_TEST_NEVER              (0  << 24)
+#       define R128_ALPHA_TEST_LESS               (1  << 24)
+#       define R128_ALPHA_TEST_LESSEQUAL          (2  << 24)
+#       define R128_ALPHA_TEST_EQUAL              (3  << 24)
+#       define R128_ALPHA_TEST_GREATEREQUAL       (4  << 24)
+#       define R128_ALPHA_TEST_GREATER            (5  << 24)
+#       define R128_ALPHA_TEST_NEQUAL             (6  << 24)
+#       define R128_ALPHA_TEST_ALWAYS             (7  << 24)
+#       define R128_COMPOSITE_SHADOW_CMP_EQUAL    (0  << 28)
+#       define R128_COMPOSITE_SHADOW_CMP_NEQUAL   (1  << 28)
+#       define R128_COMPOSITE_SHADOW              (1  << 29)
+#       define R128_TEX_MAP_ALPHA_IN_TEXTURE      (1  << 30)
+#       define R128_TEX_CACHE_LINE_SIZE_8QW       (0  << 31)
+#       define R128_TEX_CACHE_LINE_SIZE_4QW       (1  << 31)
+
+#define R128_SETUP_CNTL                   0x1bc4
+#       define R128_DONT_START_TRIANGLE   (1 <<  0)
+#       define R128_Z_BIAS                (0 <<  1)
+#       define R128_DONT_START_ANY_ON     (1 <<  2)
+#       define R128_COLOR_SOLID_COLOR     (0 <<  3)
+#       define R128_COLOR_FLAT_VERT_1     (1 <<  3)
+#       define R128_COLOR_FLAT_VERT_2     (2 <<  3)
+#       define R128_COLOR_FLAT_VERT_3     (3 <<  3)
+#       define R128_COLOR_GOURAUD         (4 <<  3)
+#       define R128_PRIM_TYPE_TRI         (0 <<  7)
+#       define R128_PRIM_TYPE_LINE        (1 <<  7)
+#       define R128_PRIM_TYPE_POINT       (2 <<  7)
+#       define R128_PRIM_TYPE_POLY_EDGE   (3 <<  7)
+#       define R128_TEXTURE_ST_MULT_W     (0 <<  9)
+#       define R128_TEXTURE_ST_DIRECT     (1 <<  9)
+#       define R128_STARTING_VERTEX_1     (1 << 14)
+#       define R128_STARTING_VERTEX_2     (2 << 14)
+#       define R128_STARTING_VERTEX_3     (3 << 14)
+#       define R128_ENDING_VERTEX_1       (1 << 16)
+#       define R128_ENDING_VERTEX_2       (2 << 16)
+#       define R128_ENDING_VERTEX_3       (3 << 16)
+#       define R128_SU_POLY_LINE_LAST     (0 << 18)
+#       define R128_SU_POLY_LINE_NOT_LAST (1 << 18)
+#       define R128_SUB_PIX_2BITS         (0 << 19)
+#       define R128_SUB_PIX_4BITS         (1 << 19)
+#       define R128_SET_UP_CONTINUE       (1 << 31)
+
+#define R128_WINDOW_XY_OFFSET             0x1bcc
+#       define R128_WINDOW_Y_SHIFT        4
+#       define R128_WINDOW_X_SHIFT        20
+
+#define R128_Z_OFFSET_C                   0x1c90
+#define R128_Z_PITCH_C                    0x1c94
+#define R128_Z_STEN_CNTL_C                0x1c98
+#       define R128_Z_PIX_WIDTH_16            (0 <<  1)
+#       define R128_Z_PIX_WIDTH_24            (1 <<  1)
+#       define R128_Z_PIX_WIDTH_32            (2 <<  1)
+#       define R128_Z_PIX_WIDTH_MASK          (3 <<  1)
+#       define R128_Z_TEST_NEVER              (0 <<  4)
+#       define R128_Z_TEST_LESS               (1 <<  4)
+#       define R128_Z_TEST_LESSEQUAL          (2 <<  4)
+#       define R128_Z_TEST_EQUAL              (3 <<  4)
+#       define R128_Z_TEST_GREATEREQUAL       (4 <<  4)
+#       define R128_Z_TEST_GREATER            (5 <<  4)
+#       define R128_Z_TEST_NEQUAL             (6 <<  4)
+#       define R128_Z_TEST_ALWAYS             (7 <<  4)
+#       define R128_Z_TEST_MASK               (7 <<  4)
+#       define R128_STENCIL_TEST_NEVER        (0 << 12)
+#       define R128_STENCIL_TEST_LESS         (1 << 12)
+#       define R128_STENCIL_TEST_LESSEQUAL    (2 << 12)
+#       define R128_STENCIL_TEST_EQUAL        (3 << 12)
+#       define R128_STENCIL_TEST_GREATEREQUAL (4 << 12)
+#       define R128_STENCIL_TEST_GREATER      (5 << 12)
+#       define R128_STENCIL_TEST_NEQUAL       (6 << 12)
+#       define R128_STENCIL_TEST_ALWAYS       (7 << 12)
+#       define R128_STENCIL_S_FAIL_KEEP       (0 << 16)
+#       define R128_STENCIL_S_FAIL_ZERO       (1 << 16)
+#       define R128_STENCIL_S_FAIL_REPLACE    (2 << 16)
+#       define R128_STENCIL_S_FAIL_INC        (3 << 16)
+#       define R128_STENCIL_S_FAIL_DEC        (4 << 16)
+#       define R128_STENCIL_S_FAIL_INV        (5 << 16)
+#       define R128_STENCIL_ZPASS_KEEP        (0 << 20)
+#       define R128_STENCIL_ZPASS_ZERO        (1 << 20)
+#       define R128_STENCIL_ZPASS_REPLACE     (2 << 20)
+#       define R128_STENCIL_ZPASS_INC         (3 << 20)
+#       define R128_STENCIL_ZPASS_DEC         (4 << 20)
+#       define R128_STENCIL_ZPASS_INV         (5 << 20)
+#       define R128_STENCIL_ZFAIL_KEEP        (0 << 24)
+#       define R128_STENCIL_ZFAIL_ZERO        (1 << 24)
+#       define R128_STENCIL_ZFAIL_REPLACE     (2 << 24)
+#       define R128_STENCIL_ZFAIL_INC         (3 << 24)
+#       define R128_STENCIL_ZFAIL_DEC         (4 << 24)
+#       define R128_STENCIL_ZFAIL_INV         (5 << 24)
+#define R128_TEX_CNTL_C                   0x1c9c
+#       define R128_Z_ENABLE                   (1 <<  0)
+#       define R128_Z_WRITE_ENABLE             (1 <<  1)
+#       define R128_STENCIL_ENABLE             (1 <<  3)
+#       define R128_SHADE_ENABLE               (0 <<  4)
+#       define R128_TEXMAP_ENABLE              (1 <<  4)
+#       define R128_SEC_TEXMAP_ENABLE          (1 <<  5)
+#       define R128_FOG_ENABLE                 (1 <<  7)
+#       define R128_DITHER_ENABLE              (1 <<  8)
+#       define R128_ALPHA_ENABLE               (1 <<  9)
+#       define R128_ALPHA_TEST_ENABLE          (1 << 10)
+#       define R128_SPEC_LIGHT_ENABLE          (1 << 11)
+#       define R128_TEX_CHROMA_KEY_ENABLE      (1 << 12)
+#       define R128_ALPHA_IN_TEX_COMPLETE_A    (0 << 13)
+#       define R128_ALPHA_IN_TEX_LSB_A         (1 << 13)
+#       define R128_LIGHT_DIS                  (0 << 14)
+#       define R128_LIGHT_COPY                 (1 << 14)
+#       define R128_LIGHT_MODULATE             (2 << 14)
+#       define R128_LIGHT_ADD                  (3 << 14)
+#       define R128_LIGHT_BLEND_CONSTANT       (4 << 14)
+#       define R128_LIGHT_BLEND_TEXTURE        (5 << 14)
+#       define R128_LIGHT_BLEND_VERTEX         (6 << 14)
+#       define R128_LIGHT_BLEND_CONST_COLOR    (7 << 14)
+#       define R128_ALPHA_LIGHT_DIS            (0 << 18)
+#       define R128_ALPHA_LIGHT_COPY           (1 << 18)
+#       define R128_ALPHA_LIGHT_MODULATE       (2 << 18)
+#       define R128_ALPHA_LIGHT_ADD            (3 << 18)
+#       define R128_ANTI_ALIAS                 (1 << 21)
+#       define R128_TEX_CACHE_FLUSH            (1 << 23)
+#       define R128_LOD_BIAS_SHIFT             24
+#define R128_MISC_3D_STATE_CNTL_REG       0x1ca0
+#       define R128_REF_ALPHA_MASK                  0xff
+#       define R128_MISC_SCALE_3D_NOOP              (0  <<  8)
+#       define R128_MISC_SCALE_3D_SCALE             (1  <<  8)
+#       define R128_MISC_SCALE_3D_TEXMAP_SHADE      (2  <<  8)
+#       define R128_MISC_SCALE_PIX_BLEND            (0  << 10)
+#       define R128_MISC_SCALE_PIX_REPLICATE        (1  << 10)
+#       define R128_ALPHA_COMB_ADD_CLAMP            (0  << 12)
+#       define R128_ALPHA_COMB_ADD_NO_CLAMP         (1  << 12)
+#       define R128_ALPHA_COMB_SUB_SRC_DST_CLAMP    (2  << 12)
+#       define R128_ALPHA_COMB_SUB_SRC_DST_NO_CLAMP (3  << 12)
+#       define R128_FOG_VERTEX                      (0  << 14)
+#       define R128_FOG_TABLE                       (1  << 14)
+#       define R128_ALPHA_BLEND_SRC_ZERO            (0  << 16)
+#       define R128_ALPHA_BLEND_SRC_ONE             (1  << 16)
+#       define R128_ALPHA_BLEND_SRC_SRCCOLOR        (2  << 16)
+#       define R128_ALPHA_BLEND_SRC_INVSRCCOLOR     (3  << 16)
+#       define R128_ALPHA_BLEND_SRC_SRCALPHA        (4  << 16)
+#       define R128_ALPHA_BLEND_SRC_INVSRCALPHA     (5  << 16)
+#       define R128_ALPHA_BLEND_SRC_DESTALPHA       (6  << 16)
+#       define R128_ALPHA_BLEND_SRC_INVDESTALPHA    (7  << 16)
+#       define R128_ALPHA_BLEND_SRC_DESTCOLOR       (8  << 16)
+#       define R128_ALPHA_BLEND_SRC_INVDESTCOLOR    (9  << 16)
+#       define R128_ALPHA_BLEND_SRC_SRCALPHASAT     (10 << 16)
+#       define R128_ALPHA_BLEND_SRC_BOTHSRCALPHA    (11 << 16)
+#       define R128_ALPHA_BLEND_SRC_BOTHINVSRCALPHA (12 << 16)
+#       define R128_ALPHA_BLEND_SRC_MASK            (15 << 16)
+#       define R128_ALPHA_BLEND_DST_ZERO            (0  << 20)
+#       define R128_ALPHA_BLEND_DST_ONE             (1  << 20)
+#       define R128_ALPHA_BLEND_DST_SRCCOLOR        (2  << 20)
+#       define R128_ALPHA_BLEND_DST_INVSRCCOLOR     (3  << 20)
+#       define R128_ALPHA_BLEND_DST_SRCALPHA        (4  << 20)
+#       define R128_ALPHA_BLEND_DST_INVSRCALPHA     (5  << 20)
+#       define R128_ALPHA_BLEND_DST_DESTALPHA       (6  << 20)
+#       define R128_ALPHA_BLEND_DST_INVDESTALPHA    (7  << 20)
+#       define R128_ALPHA_BLEND_DST_DESTCOLOR       (8  << 20)
+#       define R128_ALPHA_BLEND_DST_INVDESTCOLOR    (9  << 20)
+#       define R128_ALPHA_BLEND_DST_SRCALPHASAT     (10 << 20)
+#       define R128_ALPHA_BLEND_DST_MASK            (15 << 20)
+#       define R128_ALPHA_TEST_NEVER                (0  << 24)
+#       define R128_ALPHA_TEST_LESS                 (1  << 24)
+#       define R128_ALPHA_TEST_LESSEQUAL            (2  << 24)
+#       define R128_ALPHA_TEST_EQUAL                (3  << 24)
+#       define R128_ALPHA_TEST_GREATEREQUAL         (4  << 24)
+#       define R128_ALPHA_TEST_GREATER              (5  << 24)
+#       define R128_ALPHA_TEST_NEQUAL               (6  << 24)
+#       define R128_ALPHA_TEST_ALWAYS               (7  << 24)
+#       define R128_ALPHA_TEST_MASK                 (7  << 24)
+#define R128_TEXTURE_CLR_CMP_CLR_C        0x1ca4
+#define R128_TEXTURE_CLR_CMP_MSK_C        0x1ca8
+#define R128_FOG_COLOR_C                  0x1cac
+#       define R128_FOG_BLUE_SHIFT             0
+#       define R128_FOG_GREEN_SHIFT            8
+#       define R128_FOG_RED_SHIFT             16
+#define R128_PRIM_TEX_CNTL_C              0x1cb0
+#       define R128_MIN_BLEND_NEAREST          (0  <<  1)
+#       define R128_MIN_BLEND_LINEAR           (1  <<  1)
+#       define R128_MIN_BLEND_MIPNEAREST       (2  <<  1)
+#       define R128_MIN_BLEND_MIPLINEAR        (3  <<  1)
+#       define R128_MIN_BLEND_LINEARMIPNEAREST (4  <<  1)
+#       define R128_MIN_BLEND_LINEARMIPLINEAR  (5  <<  1)
+#       define R128_MIN_BLEND_MASK             (7  <<  1)
+#       define R128_MAG_BLEND_NEAREST          (0  <<  4)
+#       define R128_MAG_BLEND_LINEAR           (1  <<  4)
+#       define R128_MAG_BLEND_MASK             (7  <<  4)
+#       define R128_MIP_MAP_DISABLE            (1  <<  7)
+#       define R128_TEX_CLAMP_S_WRAP           (0  <<  8)
+#       define R128_TEX_CLAMP_S_MIRROR         (1  <<  8)
+#       define R128_TEX_CLAMP_S_CLAMP          (2  <<  8)
+#       define R128_TEX_CLAMP_S_BORDER_COLOR   (3  <<  8)
+#       define R128_TEX_CLAMP_S_MASK           (3  <<  8)
+#       define R128_TEX_WRAP_S                 (1  << 10)
+#       define R128_TEX_CLAMP_T_WRAP           (0  << 11)
+#       define R128_TEX_CLAMP_T_MIRROR         (1  << 11)
+#       define R128_TEX_CLAMP_T_CLAMP          (2  << 11)
+#       define R128_TEX_CLAMP_T_BORDER_COLOR   (3  << 11)
+#       define R128_TEX_CLAMP_T_MASK           (3  << 11)
+#       define R128_TEX_WRAP_T                 (1  << 13)
+#       define R128_TEX_PERSPECTIVE_DISABLE    (1  << 14)
+#       define R128_DATATYPE_VQ                (0  << 16)
+#       define R128_DATATYPE_CI4               (1  << 16)
+#       define R128_DATATYPE_CI8               (2  << 16)
+#       define R128_DATATYPE_ARGB1555          (3  << 16)
+#       define R128_DATATYPE_RGB565            (4  << 16)
+#       define R128_DATATYPE_RGB888            (5  << 16)
+#       define R128_DATATYPE_ARGB8888          (6  << 16)
+#       define R128_DATATYPE_RGB332            (7  << 16)
+#       define R128_DATATYPE_Y8                (8  << 16)
+#       define R128_DATATYPE_RGB8              (9  << 16)
+#       define R128_DATATYPE_CI16              (10 << 16)
+#       define R128_DATATYPE_YUV422            (11 << 16)
+#       define R128_DATATYPE_YUV422_2          (12 << 16)
+#       define R128_DATATYPE_AYUV444           (14 << 16)
+#       define R128_DATATYPE_ARGB4444          (15 << 16)
+#       define R128_PALLETE_EITHER             (0  << 20)
+#       define R128_PALLETE_1                  (1  << 20)
+#       define R128_PALLETE_2                  (2  << 20)
+#       define R128_PSEUDOCOLOR_DT_RGB565      (0  << 24)
+#       define R128_PSEUDOCOLOR_DT_ARGB1555    (1  << 24)
+#       define R128_PSEUDOCOLOR_DT_ARGB4444    (2  << 24)
+#define R128_PRIM_TEXTURE_COMBINE_CNTL_C  0x1cb4
+#       define R128_COMB_DIS                 (0  <<  0)
+#       define R128_COMB_COPY                (1  <<  0)
+#       define R128_COMB_COPY_INP            (2  <<  0)
+#       define R128_COMB_MODULATE            (3  <<  0)
+#       define R128_COMB_MODULATE2X          (4  <<  0)
+#       define R128_COMB_MODULATE4X          (5  <<  0)
+#       define R128_COMB_ADD                 (6  <<  0)
+#       define R128_COMB_ADD_SIGNED          (7  <<  0)
+#       define R128_COMB_BLEND_VERTEX        (8  <<  0)
+#       define R128_COMB_BLEND_TEXTURE       (9  <<  0)
+#       define R128_COMB_BLEND_CONST         (10 <<  0)
+#       define R128_COMB_BLEND_PREMULT       (11 <<  0)
+#       define R128_COMB_BLEND_PREV          (12 <<  0)
+#       define R128_COMB_BLEND_PREMULT_INV   (13 <<  0)
+#       define R128_COMB_ADD_SIGNED2X        (14 <<  0)
+#       define R128_COMB_BLEND_CONST_COLOR   (15 <<  0)
+#       define R128_COMB_MASK                (15 <<  0)
+#       define R128_COLOR_FACTOR_TEX         (4  <<  4)
+#       define R128_COLOR_FACTOR_NTEX        (5  <<  4)
+#       define R128_COLOR_FACTOR_ALPHA       (6  <<  4)
+#       define R128_COLOR_FACTOR_NALPHA      (7  <<  4)
+#       define R128_COLOR_FACTOR_MASK        (15 <<  4)
+#       define R128_INPUT_FACTOR_CONST_COLOR (2  << 10)
+#       define R128_INPUT_FACTOR_CONST_ALPHA (3  << 10)
+#       define R128_INPUT_FACTOR_INT_COLOR   (4  << 10)
+#       define R128_INPUT_FACTOR_INT_ALPHA   (5  << 10)
+#       define R128_INPUT_FACTOR_MASK        (15 << 10)
+#       define R128_COMB_ALPHA_DIS           (0  << 14)
+#       define R128_COMB_ALPHA_COPY          (1  << 14)
+#       define R128_COMB_ALPHA_COPY_INP      (2  << 14)
+#       define R128_COMB_ALPHA_MODULATE      (3  << 14)
+#       define R128_COMB_ALPHA_MODULATE2X    (4  << 14)
+#       define R128_COMB_ALPHA_MODULATE4X    (5  << 14)
+#       define R128_COMB_ALPHA_ADD           (6  << 14)
+#       define R128_COMB_ALPHA_ADD_SIGNED    (7  << 14)
+#       define R128_COMB_ALPHA_ADD_SIGNED2X  (14 << 14)
+#       define R128_COMB_ALPHA_MASK          (15 << 14)
+#       define R128_ALPHA_FACTOR_TEX_ALPHA   (6  << 18)
+#       define R128_ALPHA_FACTOR_NTEX_ALPHA  (7  << 18)
+#       define R128_ALPHA_FACTOR_MASK        (15 << 18)
+#       define R128_INP_FACTOR_A_CONST_ALPHA (1  << 25)
+#       define R128_INP_FACTOR_A_INT_ALPHA   (2  << 25)
+#       define R128_INP_FACTOR_A_MASK        (7  << 25)
+#define R128_TEX_SIZE_PITCH_C             0x1cb8
+#       define R128_TEX_PITCH_SHIFT           0
+#       define R128_TEX_SIZE_SHIFT            4
+#       define R128_TEX_HEIGHT_SHIFT          8
+#       define R128_TEX_MIN_SIZE_SHIFT       12
+#       define R128_SEC_TEX_PITCH_SHIFT      16
+#       define R128_SEC_TEX_SIZE_SHIFT       20
+#       define R128_SEC_TEX_HEIGHT_SHIFT     24
+#       define R128_SEC_TEX_MIN_SIZE_SHIFT   28
+#       define R128_TEX_PITCH_MASK           (0x0f <<  0)
+#       define R128_TEX_SIZE_MASK            (0x0f <<  4)
+#       define R128_TEX_HEIGHT_MASK          (0x0f <<  8)
+#       define R128_TEX_MIN_SIZE_MASK        (0x0f << 12)
+#       define R128_SEC_TEX_PITCH_MASK       (0x0f << 16)
+#       define R128_SEC_TEX_SIZE_MASK        (0x0f << 20)
+#       define R128_SEC_TEX_HEIGHT_MASK      (0x0f << 24)
+#       define R128_SEC_TEX_MIN_SIZE_MASK    (0x0f << 28)
+#       define R128_TEX_SIZE_PITCH_SHIFT      0
+#       define R128_SEC_TEX_SIZE_PITCH_SHIFT 16
+#       define R128_TEX_SIZE_PITCH_MASK      (0xffff <<  0)
+#       define R128_SEC_TEX_SIZE_PITCH_MASK  (0xffff << 16)
+#define R128_PRIM_TEX_0_OFFSET_C          0x1cbc
+#define R128_PRIM_TEX_1_OFFSET_C          0x1cc0
+#define R128_PRIM_TEX_2_OFFSET_C          0x1cc4
+#define R128_PRIM_TEX_3_OFFSET_C          0x1cc8
+#define R128_PRIM_TEX_4_OFFSET_C          0x1ccc
+#define R128_PRIM_TEX_5_OFFSET_C          0x1cd0
+#define R128_PRIM_TEX_6_OFFSET_C          0x1cd4
+#define R128_PRIM_TEX_7_OFFSET_C          0x1cd8
+#define R128_PRIM_TEX_8_OFFSET_C          0x1cdc
+#define R128_PRIM_TEX_9_OFFSET_C          0x1ce0
+#define R128_PRIM_TEX_10_OFFSET_C         0x1ce4
+#       define R128_TEX_NO_TILE           (0 << 30)
+#       define R128_TEX_TILED_BY_HOST     (1 << 30)
+#       define R128_TEX_TILED_BY_STORAGE  (2 << 30)
+#       define R128_TEX_TILED_BY_STORAGE2 (3 << 30)
+
+#define R128_SEC_TEX_CNTL_C               0x1d00
+#       define R128_SEC_SELECT_PRIM_ST    (0  <<  0)
+#       define R128_SEC_SELECT_SEC_ST     (1  <<  0)
+#define R128_SEC_TEX_COMBINE_CNTL_C       0x1d04
+#       define R128_INPUT_FACTOR_PREV_COLOR (8  << 10)
+#       define R128_INPUT_FACTOR_PREV_ALPHA (9  << 10)
+#       define R128_INP_FACTOR_A_PREV_ALPHA (4  << 25)
+#define R128_SEC_TEX_0_OFFSET_C           0x1d08
+#define R128_SEC_TEX_1_OFFSET_C           0x1d0c
+#define R128_SEC_TEX_2_OFFSET_C           0x1d10
+#define R128_SEC_TEX_3_OFFSET_C           0x1d14
+#define R128_SEC_TEX_4_OFFSET_C           0x1d18
+#define R128_SEC_TEX_5_OFFSET_C           0x1d1c
+#define R128_SEC_TEX_6_OFFSET_C           0x1d20
+#define R128_SEC_TEX_7_OFFSET_C           0x1d24
+#define R128_SEC_TEX_8_OFFSET_C           0x1d28
+#define R128_SEC_TEX_9_OFFSET_C           0x1d2c
+#define R128_SEC_TEX_10_OFFSET_C          0x1d30
+#define R128_CONSTANT_COLOR_C             0x1d34
+#       define R128_CONSTANT_BLUE_SHIFT        0
+#       define R128_CONSTANT_GREEN_SHIFT       8
+#       define R128_CONSTANT_RED_SHIFT        16
+#       define R128_CONSTANT_ALPHA_SHIFT      24
+#define R128_PRIM_TEXTURE_BORDER_COLOR_C  0x1d38
+#       define R128_PRIM_TEX_BORDER_BLUE_SHIFT   0
+#       define R128_PRIM_TEX_BORDER_GREEN_SHIFT  8
+#       define R128_PRIM_TEX_BORDER_RED_SHIFT   16
+#       define R128_PRIM_TEX_BORDER_ALPHA_SHIFT 24
+#define R128_SEC_TEXTURE_BORDER_COLOR_C   0x1d3c
+#       define R128_SEC_TEX_BORDER_BLUE_SHIFT   0
+#       define R128_SEC_TEX_BORDER_GREEN_SHIFT  8
+#       define R128_SEC_TEX_BORDER_RED_SHIFT   16
+#       define R128_SEC_TEX_BORDER_ALPHA_SHIFT 24
+#define R128_STEN_REF_MASK_C              0x1d40
+#       define R128_STEN_REFERENCE_SHIFT       0
+#       define R128_STEN_MASK_SHIFT           16
+#       define R128_STEN_WRITE_MASK_SHIFT     24
+#define R128_PLANE_3D_MASK_C              0x1d44
+
+
+				/* Constants */
+#define R128_AGP_TEX_OFFSET               0x02000000
+
+#define R128_VB_AGE_REG                   R128_GUI_SCRATCH_REG0
+#define R128_SWAP_AGE_REG                 R128_GUI_SCRATCH_REG1
+
+				/* CCE packet types */
+#define R128_CCE_PACKET0                         0x00000000
+#define R128_CCE_PACKET0_ONE_REG_WR              0x00008000
+#define R128_CCE_PACKET1                         0x40000000
+#define R128_CCE_PACKET2                         0x80000000
+#define R128_CCE_PACKET3_NOP                     0xC0001000
+#define R128_CCE_PACKET3_PAINT                   0xC0001100
+#define R128_CCE_PACKET3_BITBLT                  0xC0001200
+#define R128_CCE_PACKET3_SMALLTEXT               0xC0001300
+#define R128_CCE_PACKET3_HOSTDATA_BLT            0xC0001400
+#define R128_CCE_PACKET3_POLYLINE                0xC0001500
+#define R128_CCE_PACKET3_SCALING                 0xC0001600
+#define R128_CCE_PACKET3_TRANS_SCALING           0xC0001700
+#define R128_CCE_PACKET3_POLYSCANLINES           0xC0001800
+#define R128_CCE_PACKET3_NEXT_CHAR               0xC0001900
+#define R128_CCE_PACKET3_PAINT_MULTI             0xC0001A00
+#define R128_CCE_PACKET3_BITBLT_MULTI            0xC0001B00
+#define R128_CCE_PACKET3_PLY_NEXTSCAN            0xC0001D00
+#define R128_CCE_PACKET3_SET_SCISSORS            0xC0001E00
+#define R128_CCE_PACKET3_SET_MODE24BPP           0xC0001F00
+#define R128_CCE_PACKET3_CNTL_PAINT              0xC0009100
+#define R128_CCE_PACKET3_CNTL_BITBLT             0xC0009200
+#define R128_CCE_PACKET3_CNTL_SMALLTEXT          0xC0009300
+#define R128_CCE_PACKET3_CNTL_HOSTDATA_BLT       0xC0009400
+#define R128_CCE_PACKET3_CNTL_POLYLINE           0xC0009500
+#define R128_CCE_PACKET3_CNTL_SCALING            0xC0009600
+#define R128_CCE_PACKET3_CNTL_TRANS_SCALING      0xC0009700
+#define R128_CCE_PACKET3_CNTL_POLYSCANLINES      0xC0009800
+#define R128_CCE_PACKET3_CNTL_NEXT_CHAR          0xC0009900
+#define R128_CCE_PACKET3_CNTL_PAINT_MULTI        0xC0009A00
+#define R128_CCE_PACKET3_CNTL_BITBLT_MULTI       0xC0009B00
+#define R128_CCE_PACKET3_CNTL_TRANS_BITBLT       0xC0009C00
+#define R128_CCE_PACKET3_3D_SAVE_CONTEXT         0xC0002000
+#define R128_CCE_PACKET3_3D_PLAY_CONTEXT         0xC0002100
+#define R128_CCE_PACKET3_3D_RNDR_GEN_INDX_PRIM   0xC0002300
+#define R128_CCE_PACKET3_3D_RNDR_GEN_PRIM        0xC0002500
+#define R128_CCE_PACKET3_LOAD_PALETTE            0xC0002C00
+#define R128_CCE_PACKET3_PURGE                   0xC0002D00
+#define R128_CCE_PACKET3_NEXT_VERTEX_BUNDLE      0xC0002E00
+#       define R128_CCE_PACKET_MASK              0xC0000000
+#       define R128_CCE_PACKET_COUNT_MASK        0x3fff0000
+#	define R128_CCE_PACKET_MAX_DWORDS        (1 << 14)
+#       define R128_CCE_PACKET0_REG_MASK         0x000007ff
+#       define R128_CCE_PACKET1_REG0_MASK        0x000007ff
+#       define R128_CCE_PACKET1_REG1_MASK        0x003ff800
+
+#define R128_CCE_VC_FRMT_RHW                     0x00000001
+#define R128_CCE_VC_FRMT_DIFFUSE_BGR             0x00000002
+#define R128_CCE_VC_FRMT_DIFFUSE_A               0x00000004
+#define R128_CCE_VC_FRMT_DIFFUSE_ARGB            0x00000008
+#define R128_CCE_VC_FRMT_SPEC_BGR                0x00000010
+#define R128_CCE_VC_FRMT_SPEC_F                  0x00000020
+#define R128_CCE_VC_FRMT_SPEC_FRGB               0x00000040
+#define R128_CCE_VC_FRMT_S_T                     0x00000080
+#define R128_CCE_VC_FRMT_S2_T2                   0x00000100
+#define R128_CCE_VC_FRMT_RHW2                    0x00000200
+
+#define R128_CCE_VC_CNTL_PRIM_TYPE_NONE          0x00000000
+#define R128_CCE_VC_CNTL_PRIM_TYPE_POINT         0x00000001
+#define R128_CCE_VC_CNTL_PRIM_TYPE_LINE          0x00000002
+#define R128_CCE_VC_CNTL_PRIM_TYPE_POLY_LINE     0x00000003
+#define R128_CCE_VC_CNTL_PRIM_TYPE_TRI_LIST      0x00000004
+#define R128_CCE_VC_CNTL_PRIM_TYPE_TRI_FAN       0x00000005
+#define R128_CCE_VC_CNTL_PRIM_TYPE_TRI_STRIP     0x00000006
+#define R128_CCE_VC_CNTL_PRIM_TYPE_TRI_TYPE2     0x00000007
+#define R128_CCE_VC_CNTL_PRIM_WALK_IND           0x00000010
+#define R128_CCE_VC_CNTL_PRIM_WALK_LIST          0x00000020
+#define R128_CCE_VC_CNTL_PRIM_WALK_RING          0x00000030
+#define R128_CCE_VC_CNTL_NUM_SHIFT               16
 
 #endif
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_sarea.h b/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_sarea.h
new file mode 100644
index 000000000..c5d99df34
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/drivers/r128/r128_sarea.h
@@ -0,0 +1,77 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
+                                               Cedar Park, Texas. 
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ */
+
+#ifndef _R128_SAREA_H_
+#define _R128_SAREA_H_
+
+/* There are 2 heaps (local/AGP).  Each region within a heap is a
+   minimum of 64k, and there are at most 64 of them per heap. */
+#define R128_LOCAL_TEX_HEAP       0
+#define R128_AGP_TEX_HEAP         1
+#define R128_NR_TEX_HEAPS         2
+#define R128_NR_TEX_REGIONS      64
+#define R128_LOG_TEX_GRANULARITY 16
+
+typedef struct {
+    unsigned char next, prev; /* indices to form a circular LRU  */
+    unsigned char in_use;     /* owned by a client, or free? */
+    int           age;        /* tracked by clients to update local LRU's */
+} R128TexRegion;
+
+typedef struct {
+    /* Maintain an LRU of contiguous regions of texture space.  If you
+     * think you own a region of texture memory, and it has an age
+     * different to the one you set, then you are mistaken and it has
+     * been stolen by another client.  If global texAge hasn't changed,
+     * there is no need to walk the list.
+     *
+     * These regions can be used as a proxy for the fine-grained texture
+     * information of other clients - by maintaining them in the same
+     * lru which is used to age their own textures, clients have an
+     * approximate lru for the whole of global texture space, and can
+     * make informed decisions as to which areas to kick out.  There is
+     * no need to choose whether to kick out your own texture or someone
+     * else's - simply eject them all in LRU order.
+     */
+				/* Last elt is sentinal */
+    R128TexRegion texList[R128_NR_TEX_HEAPS][R128_NR_TEX_REGIONS+1];
+				/* last time texture was uploaded */
+    int           texAge[R128_NR_TEX_HEAPS];
+
+    int           ctxOwner;     /* last context to upload state */
+
+    int           ringWrite;    /* current ring buffer write index */
+} R128SAREAPriv, *R128SAREAPrivPtr;
+
+#endif
diff --git a/xc/programs/Xserver/hw/xfree86/loader/xf86sym.c b/xc/programs/Xserver/hw/xfree86/loader/xf86sym.c
index 3d21e1c87..9844d6d90 100644
--- a/xc/programs/Xserver/hw/xfree86/loader/xf86sym.c
+++ b/xc/programs/Xserver/hw/xfree86/loader/xf86sym.c
@@ -54,6 +54,12 @@
 #include "xf86Priv.h"
 #include "vbe.h"
 
+#ifdef __FreeBSD__
+/* XXX used in drmOpen(). This should change to use a less os-specific
+ * method. */
+int sysctlbyname(const char*, void *, size_t *, void *, size_t);
+#endif
+
 extern xf86MonPtr ConfiguredMonitor;
 
 /* XXX Should get all of these from elsewhere */
@@ -894,6 +900,10 @@ LOOKUP xfree86LookupTab[] = {
 #endif
 #endif
 
+#ifdef __FreeBSD__
+   SYMFUNC(sysctlbyname)
+#endif
+
 /*
  * and now some variables
  */
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/Imakefile b/xc/programs/Xserver/hw/xfree86/os-support/Imakefile
index 7041d435a..c42f5b2b7 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/Imakefile
+++ b/xc/programs/Xserver/hw/xfree86/os-support/Imakefile
@@ -100,6 +100,7 @@ OS_SUBDIR = sco
 #if BuildXF86DRI && !DoLoadableServer
 DRM_SRC = $(OS_SUBDIR)/drm/?*.c
 DRM_OBJ = $(OS_SUBDIR)/drm/?*.o
+DRM_DONES = $(OS_SUBDIR)/drm/DONE
 #endif
 
 SUBDIRS = $(OS_SUBDIR) $(BUS_SUBDIR) misc vbe
@@ -107,7 +108,7 @@ SUBDIRS = $(OS_SUBDIR) $(BUS_SUBDIR) misc vbe
 SRCS = $(OS_SUBDIR)/?*.c $(BUS_SUBDIR)/?*.c misc/?*.c vbe/?*.c $(DRM_SRC)
 OBJS = $(OS_SUBDIR)/?*.o $(BUS_SUBDIR)/?*.o misc/?*.o vbe/?*.o $(DRM_OBJ)
 
-DONES = $(OS_SUBDIR)/DONE $(BUS_SUBDIR)/DONE misc/DONE vbe/DONE
+DONES = $(OS_SUBDIR)/DONE $(BUS_SUBDIR)/DONE misc/DONE vbe/DONE $(DRM_DONES)
 
 #if HasParallelMake
 MakeMutex($(SUBDIRS) $(OBJS) $(DONES))
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/Imakefile b/xc/programs/Xserver/hw/xfree86/os-support/bsd/Imakefile
index b0ec00a5e..81f28bea8 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/bsd/Imakefile
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/Imakefile
@@ -43,6 +43,11 @@ IOPERM_OBJ = ioperm_noop.o
 # endif
 #endif
 
+#if BuildXF86DRI
+DRI_SRC = sigio.c
+DRI_OBJ = sigio.o
+#endif
+
 MOUSESRC = bsd_mouse.c
 MOUSEOBJ = bsd_mouse.o
 
@@ -84,12 +89,12 @@ AXP_OBJ=bsd_ev56.o
 SRCS = bsd_init.c bsd_video.c bsd_io.c bsd_VTsw.c \
        libc_wrapper.c $(IOPERM_SRC) std_kbdEv.c posix_tty.c $(MOUSESRC) \
        $(RES_SRC) stdPci.c vidmem.c $(JOYSTICK_SRC) sigio.c $(APMSRC) \
-       $(AXP_SRC)
+       $(AXP_SRC) kmod_noop.c
 
 OBJS = bsd_init.o bsd_video.o bsd_io.o bsd_VTsw.o \
        libc_wrapper.o $(IOPERM_OBJ) std_kbdEv.o posix_tty.o $(MOUSEOBJ) \
        $(RES_OBJ) stdPci.o vidmem.o $(JOYSTICK_OBJ) sigio.o $(APMOBJ) \
-       $(AXP_OBJ)
+       $(AXP_OBJ) kmod_noop.o
 
 INCLUDES = -I$(XF86COMSRC) -I$(XF86OSSRC) -I. -I$(SERVERSRC)/include \
            -I$(XINCLUDESRC) -I$(EXTINCSRC) -I$(SERVERSRC)/mi -I$(APINCLUDES) \
@@ -114,6 +119,14 @@ SpecialObjectRule(bsd_ev56.o, bsd_ev56.c, -mcpu=ev56)
 SubdirLibraryRule($(OBJS))
 NormalLibraryObjectRule()
 
+#if BuildXF86DRI
+#define IHaveSubdirs
+SUBDIRS = drm
+
+MakeSubdirs($(SUBDIRS))
+DependSubdirs($(SUBDIRS))
+#endif
+
 #if BuildXInputExt
 # if DoLoadableServer
 #  if JoystickSupport
@@ -136,6 +149,7 @@ LinkSourceFile(stdResource.c,../shared)
 LinkSourceFile(stdPci.c,../shared)
 LinkSourceFile(vidmem.c,../shared)
 LinkSourceFile(sigio.c,../shared)
+LinkSourceFile(kmod_noop.c,../shared)
 
 DependTarget()
 
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/Imakefile b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/Imakefile
new file mode 100644
index 000000000..4639b530d
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/Imakefile
@@ -0,0 +1,41 @@
+XCOMM $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/Imakefile,v 1.4 1999/09/25 14:37:49 dawes Exp $
+XCOMM $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/Imakefile,v 1.7 1999/09/14 19:55:15 faith Exp $
+
+#define IHaveModules
+#include <Server.tmpl>
+
+#if DoLoadableServer
+MSRC = drmmodule.c
+MOBJ = drmmodule.o
+#endif
+
+#if BuildXF86DRI
+#if HasMTRRSupport
+MTRR_DEFINES = -DHAS_MTRR_SUPPORT
+#endif
+
+SRCS = xf86drm.c xf86drmHash.c xf86drmRandom.c xf86drmSL.c xf86drmI810.c xf86drmMga.c xf86drmR128.c $(MSRC)
+OBJS = xf86drm.o xf86drmHash.o xf86drmRandom.o xf86drmSL.o xf86drmI810.o xf86drmMga.o xf86drmR128.o $(MOBJ)
+
+
+INCLUDES = -I$(XF86COMSRC) -I$(XF86OSSRC) -I. -I$(SERVERSRC)/include \
+           -I$(XINCLUDESRC) -I$(EXTINCSRC) -I../.. -Ikernel
+
+DEFINES = $(MTRR_DEFINES) $(GLX_DEFINES)
+
+ModuleObjectRule()
+LibraryModuleTarget(drm,$(OBJS))
+NormalLintTarget($(SRCS))
+
+InstallLibraryModule(drm,$(MODULEDIR),freebsd)
+
+#define IHaveSubdirs
+SUBDIRS = kernel
+
+MakeSubdirs($(SUBDIRS))
+DependSubdirs($(SUBDIRS))
+#endif
+
+
+
+InstallDriverSDKLibraryModule(drm,$(DRIVERSDKMODULEDIR),freebsd)
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/drmmodule.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/drmmodule.c
new file mode 100644
index 000000000..1ddf3f4c1
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/drmmodule.c
@@ -0,0 +1,56 @@
+/* drmmodule.c -- Module initialization
+ * Created: Fri Jun  4 09:05:48 1999 by faith@precisioninsight.com
+ * Revised: Fri Jun  4 09:09:22 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/drmmodule.c,v 1.1 1999/06/07 13:01:42 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/drmmodule.c,v 1.1 1999/06/14 07:32:01 dawes Exp $
+ * 
+ */
+
+#include "xf86Module.h"
+
+static MODULESETUPPROTO(drmSetup);
+
+static XF86ModuleVersionInfo VersRec =
+{
+        "drm",
+        MODULEVENDORSTRING,
+        MODINFOSTRING1,
+        MODINFOSTRING2,
+        XF86_VERSION_CURRENT,
+        1, 0, 0,
+        ABI_CLASS_EXTENSION,
+        ABI_EXTENSION_VERSION,
+        MOD_CLASS_NONE,
+        {0,0,0,0}
+};
+
+XF86ModuleData drmModuleData = { &VersRec, drmSetup, NULL };
+
+static pointer
+drmSetup(pointer module, pointer opts, int *errmaj, int *errmin)
+{
+    return (void *)1;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/Imakefile b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/Imakefile
new file mode 100644
index 000000000..cf042d218
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/Imakefile
@@ -0,0 +1,31 @@
+XCOMM $XFree86$
+XCOMM $PI$
+
+#include <Server.tmpl>
+
+LinkSourceFile(xf86drm.c,..)
+LinkSourceFile(xf86drmHash.c,..)
+LinkSourceFile(xf86drmRandom.c,..)
+LinkSourceFile(xf86drmSL.c,..)
+LinkSourceFile(xf86drm.h,$(XF86OSSRC))
+LinkSourceFile(xf86_OSproc.h,$(XF86OSSRC))
+LinkSourceFile(sigio.c,$(XF86OSSRC)/shared)
+
+XCOMM This is a kludge until we determine how best to build the
+XCOMM kernel-specific device driver.  This allows us to continue
+XCOMM to maintain the single Makefile.bsd with kernel-specific
+XCOMM support.  Later, we can move to a different Imakefile.
+
+#if BuildXF86DRI && BuildXF86DRM
+all::
+	$(MAKE) -f Makefile.bsd
+
+install::
+	$(MAKE) -f Makefile.bsd install
+#else
+all::
+	echo 'Use "make -f Makefile.bsd" to manually build drm.o'
+#endif
+
+clean::
+	$(MAKE) -f Makefile.bsd clean
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/Makefile.bsd b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/Makefile.bsd
new file mode 100644
index 000000000..ff26c7628
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/Makefile.bsd
@@ -0,0 +1,5 @@
+# $FreeBSD$
+
+SUBDIR = drm tdfx gamma
+
+.include <bsd.subdir.mk>
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm.h b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm.h
new file mode 100644
index 000000000..9e0ade357
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm.h
@@ -0,0 +1,359 @@
+/* drm.h -- Header for Direct Rendering Manager -*- c -*-
+ * Created: Mon Jan  4 10:05:05 1999 by faith@precisioninsight.com
+ * Revised: Fri Aug 20 13:08:18 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm.h,v 1.46 1999/08/20 20:00:53 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm.h,v 1.1 1999/09/25 14:37:58 dawes Exp $
+ * 
+ */
+
+#ifndef _DRM_H_
+#define _DRM_H_
+
+#include <sys/ioccom.h>		/* For _IO* macros */
+
+#define DRM_DEV_DRM	 "/dev/drm"
+#define DRM_DEV_MODE	 (S_IRUSR|S_IWUSR|S_IRGRP|S_IWGRP)
+#define DRM_DEV_UID	 0
+#define DRM_DEV_GID	 0
+
+
+#define DRM_NAME	"drm"	  /* Name in kernel, /dev		    */
+#define DRM_MIN_ORDER	5	  /* At least 2^5 bytes = 32 bytes	    */
+#define DRM_MAX_ORDER	22	  /* Up to 2^22 bytes = 4MB		    */
+#define DRM_RAM_PERCENT 10	  /* How much system ram can we lock?	    */
+
+#define _DRM_LOCK_HELD	0x80000000 /* Hardware lock is held		    */
+#define _DRM_LOCK_CONT	0x40000000 /* Hardware lock is contended	    */
+#define _DRM_LOCK_IS_HELD(lock)	   ((lock) & _DRM_LOCK_HELD)
+#define _DRM_LOCK_IS_CONT(lock)	   ((lock) & _DRM_LOCK_CONT)
+#define _DRM_LOCKING_CONTEXT(lock) ((lock) & ~(_DRM_LOCK_HELD|_DRM_LOCK_CONT))
+
+typedef unsigned long drm_handle_t;
+typedef unsigned int  drm_context_t;
+typedef unsigned int  drm_drawable_t;
+typedef unsigned int  drm_magic_t;
+
+/* Warning: If you change this structure, make sure you change 
+ * XF86DRIClipRectRec in the server as well */
+
+typedef struct drm_clip_rect {
+           unsigned short x1;
+           unsigned short y1;
+           unsigned short x2;
+           unsigned short y2;
+} drm_clip_rect_t;
+
+/* Seperate include files for the i810/mga specific structures */
+#include "mga_drm.h"
+#include "i810_drm.h"
+#include "r128_drm.h"
+
+typedef struct drm_version {
+	int    version_major;	  /* Major version			    */
+	int    version_minor;	  /* Minor version			    */
+	int    version_patchlevel;/* Patch level			    */
+	size_t name_len;	  /* Length of name buffer		    */
+	char   *name;		  /* Name of driver			    */
+	size_t date_len;	  /* Length of date buffer		    */
+	char   *date;		  /* User-space buffer to hold date	    */
+	size_t desc_len;	  /* Length of desc buffer		    */
+	char   *desc;		  /* User-space buffer to hold desc	    */
+} drm_version_t;
+
+typedef struct drm_unique {
+	size_t unique_len;	  /* Length of unique			    */
+	char   *unique;		  /* Unique name for driver instantiation   */
+} drm_unique_t;
+
+typedef struct drm_list {
+	int		 count;	  /* Length of user-space structures	    */
+	drm_version_t	 *version;
+} drm_list_t;
+
+typedef struct drm_block {
+	int		 unused;
+} drm_block_t;
+
+typedef struct drm_control {
+	enum {
+		DRM_ADD_COMMAND,
+		DRM_RM_COMMAND,
+		DRM_INST_HANDLER,
+		DRM_UNINST_HANDLER
+	}		 func;
+	int		 irq;
+} drm_control_t;
+
+typedef enum drm_map_type {
+	_DRM_FRAME_BUFFER = 0,	  /* WC (no caching), no core dump	    */
+	_DRM_REGISTERS	  = 1,	  /* no caching, no core dump		    */
+	_DRM_SHM	  = 2,	  /* shared, cached			    */
+	_DRM_AGP	  = 3	  /* AGP/GART				    */
+} drm_map_type_t;
+
+typedef enum drm_map_flags {
+	_DRM_RESTRICTED	     = 0x01, /* Cannot be mapped to user-virtual    */
+	_DRM_READ_ONLY	     = 0x02,
+	_DRM_LOCKED	     = 0x04, /* shared, cached, locked		    */
+	_DRM_KERNEL	     = 0x08, /* kernel requires access		    */
+	_DRM_WRITE_COMBINING = 0x10, /* use write-combining if available    */
+	_DRM_CONTAINS_LOCK   = 0x20  /* SHM page that contains lock	    */
+} drm_map_flags_t;
+
+typedef struct drm_map {
+	unsigned long	offset;	 /* Requested physical address (0 for SAREA)*/
+	unsigned long	size;	 /* Requested physical size (bytes)	    */
+	drm_map_type_t	type;	 /* Type of memory to map		    */
+	drm_map_flags_t flags;	 /* Flags				    */
+	void		*handle; /* User-space: "Handle" to pass to mmap    */
+				 /* Kernel-space: kernel-virtual address    */
+	int		mtrr;	 /* MTRR slot used			    */
+				 /* Private data			    */
+} drm_map_t;
+
+typedef enum drm_lock_flags {
+	_DRM_LOCK_READY	     = 0x01, /* Wait until hardware is ready for DMA */
+	_DRM_LOCK_QUIESCENT  = 0x02, /* Wait until hardware quiescent	     */
+	_DRM_LOCK_FLUSH	     = 0x04, /* Flush this context's DMA queue first */
+	_DRM_LOCK_FLUSH_ALL  = 0x08, /* Flush all DMA queues first	     */
+				/* These *HALT* flags aren't supported yet
+				   -- they will be used to support the
+				   full-screen DGA-like mode. */
+	_DRM_HALT_ALL_QUEUES = 0x10, /* Halt all current and future queues   */
+	_DRM_HALT_CUR_QUEUES = 0x20  /* Halt all current queues		     */
+} drm_lock_flags_t;
+
+typedef struct drm_lock {
+	int		 context;
+	drm_lock_flags_t flags;
+} drm_lock_t;
+
+typedef enum drm_dma_flags {	      /* These values *MUST* match xf86drm.h */
+				      /* Flags for DMA buffer dispatch	     */
+	_DRM_DMA_BLOCK	      = 0x01, /* Block until buffer dispatched.
+					 Note, the buffer may not yet have
+					 been processed by the hardware --
+					 getting a hardware lock with the
+					 hardware quiescent will ensure
+					 that the buffer has been
+					 processed.			     */
+	_DRM_DMA_WHILE_LOCKED = 0x02, /* Dispatch while lock held	     */
+	_DRM_DMA_PRIORITY     = 0x04, /* High priority dispatch		     */
+
+				      /* Flags for DMA buffer request	     */
+	_DRM_DMA_WAIT	      = 0x10, /* Wait for free buffers		     */
+	_DRM_DMA_SMALLER_OK   = 0x20, /* Smaller-than-requested buffers ok   */
+	_DRM_DMA_LARGER_OK    = 0x40  /* Larger-than-requested buffers ok    */
+} drm_dma_flags_t;
+
+typedef struct drm_buf_desc {
+	int	      count;	 /* Number of buffers of this size	     */
+	int	      size;	 /* Size in bytes			     */
+	int	      low_mark;	 /* Low water mark			     */
+	int	      high_mark; /* High water mark			     */
+	enum {
+		_DRM_PAGE_ALIGN = 0x01, /* Align on page boundaries for DMA  */
+		_DRM_AGP_BUFFER = 0x02  /* Buffer is in agp space            */
+	}	      flags;
+	unsigned long agp_start; /* Start address of where the agp buffers
+				  * are in the agp aperture */
+} drm_buf_desc_t;
+
+typedef struct drm_buf_info {
+	int	       count;	/* Entries in list			     */
+	drm_buf_desc_t *list;
+} drm_buf_info_t;
+
+typedef struct drm_buf_free {
+	int	       count;
+	int	       *list;
+} drm_buf_free_t;
+
+typedef struct drm_buf_pub {
+	int		  idx;	       /* Index into master buflist	     */
+	int		  total;       /* Buffer size			     */
+	int		  used;	       /* Amount of buffer in use (for DMA)  */
+	void		  *address;    /* Address of buffer		     */
+} drm_buf_pub_t;
+
+typedef struct drm_buf_map {
+	int	      count;	/* Length of buflist			    */
+	void	      *virtual;	/* Mmaped area in user-virtual		    */
+	drm_buf_pub_t *list;	/* Buffer information			    */
+} drm_buf_map_t;
+
+typedef struct drm_dma {
+				/* Indices here refer to the offset into
+				   buflist in drm_buf_get_t.  */
+	int		context;	  /* Context handle		    */
+	int		send_count;	  /* Number of buffers to send	    */
+	int		*send_indices;	  /* List of handles to buffers	    */
+	int		*send_sizes;	  /* Lengths of data to send	    */
+	drm_dma_flags_t flags;		  /* Flags			    */
+	int		request_count;	  /* Number of buffers requested    */
+	int		request_size;	  /* Desired size for buffers	    */
+	int		*request_indices; /* Buffer information		    */
+	int		*request_sizes;
+	int		granted_count;	  /* Number of buffers granted	    */
+} drm_dma_t;
+
+typedef enum {
+	_DRM_CONTEXT_PRESERVED = 0x01,
+	_DRM_CONTEXT_2DONLY    = 0x02
+} drm_ctx_flags_t;
+
+typedef struct drm_ctx {
+	drm_context_t	handle;
+	drm_ctx_flags_t flags;
+} drm_ctx_t;
+
+typedef struct drm_ctx_res {
+	int		count;
+	drm_ctx_t	*contexts;
+} drm_ctx_res_t;
+
+typedef struct drm_draw {
+	drm_drawable_t	handle;
+} drm_draw_t;
+
+typedef struct drm_auth {
+	drm_magic_t	magic;
+} drm_auth_t;
+
+typedef struct drm_irq_busid {
+	int irq;
+	int busnum;
+	int devnum;
+	int funcnum;
+} drm_irq_busid_t;
+
+typedef struct drm_agp_mode {
+	unsigned long mode;
+} drm_agp_mode_t;
+
+				/* For drm_agp_alloc -- allocated a buffer */
+typedef struct drm_agp_buffer {
+	unsigned long size;	/* In bytes -- will round to page boundary */
+	unsigned long handle;	/* Used for BIND/UNBIND ioctls */
+	unsigned long type;     /* Type of memory to allocate  */
+        unsigned long physical; /* Physical used by i810       */
+} drm_agp_buffer_t;
+
+				/* For drm_agp_bind */
+typedef struct drm_agp_binding {
+	unsigned long handle;   /* From drm_agp_buffer */
+	unsigned long offset;	/* In bytes -- will round to page boundary */
+} drm_agp_binding_t;
+
+typedef struct drm_agp_info {
+	int            agp_version_major;
+	int            agp_version_minor;
+	unsigned long  mode;
+	unsigned long  aperture_base;  /* physical address */
+	unsigned long  aperture_size;  /* bytes */
+	unsigned long  memory_allowed; /* bytes */
+	unsigned long  memory_used;
+
+				/* PCI information */
+	unsigned short id_vendor;
+	unsigned short id_device;
+} drm_agp_info_t;
+
+#define DRM_IOCTL_BASE	     'd'
+#define DRM_IOCTL_NR(n)	     ((n) & 0xff)
+#define DRM_IO(nr)	     _IO(DRM_IOCTL_BASE,nr)
+#define DRM_IOR(nr,size)     _IOR(DRM_IOCTL_BASE,nr,size)
+#define DRM_IOW(nr,size)     _IOW(DRM_IOCTL_BASE,nr,size)
+#define DRM_IOWR(nr,size)    _IOWR(DRM_IOCTL_BASE,nr,size)
+
+
+#define DRM_IOCTL_VERSION    DRM_IOWR(0x00, drm_version_t)
+#define DRM_IOCTL_GET_UNIQUE DRM_IOWR(0x01, drm_unique_t)
+#define DRM_IOCTL_GET_MAGIC  DRM_IOR( 0x02, drm_auth_t)
+#define DRM_IOCTL_IRQ_BUSID  DRM_IOWR(0x03, drm_irq_busid_t)
+
+#define DRM_IOCTL_SET_UNIQUE DRM_IOW( 0x10, drm_unique_t)
+#define DRM_IOCTL_AUTH_MAGIC DRM_IOW( 0x11, drm_auth_t)
+#define DRM_IOCTL_BLOCK	     DRM_IOWR(0x12, drm_block_t)
+#define DRM_IOCTL_UNBLOCK    DRM_IOWR(0x13, drm_block_t)
+#define DRM_IOCTL_CONTROL    DRM_IOW( 0x14, drm_control_t)
+#define DRM_IOCTL_ADD_MAP    DRM_IOWR(0x15, drm_map_t)
+#define DRM_IOCTL_ADD_BUFS   DRM_IOWR(0x16, drm_buf_desc_t)
+#define DRM_IOCTL_MARK_BUFS  DRM_IOW( 0x17, drm_buf_desc_t)
+#define DRM_IOCTL_INFO_BUFS  DRM_IOWR(0x18, drm_buf_info_t)
+#define DRM_IOCTL_MAP_BUFS   DRM_IOWR(0x19, drm_buf_map_t)
+#define DRM_IOCTL_FREE_BUFS  DRM_IOW( 0x1a, drm_buf_free_t)
+
+#define DRM_IOCTL_ADD_CTX    DRM_IOWR(0x20, drm_ctx_t)
+#define DRM_IOCTL_RM_CTX     DRM_IOWR(0x21, drm_ctx_t)
+#define DRM_IOCTL_MOD_CTX    DRM_IOW( 0x22, drm_ctx_t)
+#define DRM_IOCTL_GET_CTX    DRM_IOWR(0x23, drm_ctx_t)
+#define DRM_IOCTL_SWITCH_CTX DRM_IOW( 0x24, drm_ctx_t)
+#define DRM_IOCTL_NEW_CTX    DRM_IOW( 0x25, drm_ctx_t)
+#define DRM_IOCTL_RES_CTX    DRM_IOWR(0x26, drm_ctx_res_t)
+#define DRM_IOCTL_ADD_DRAW   DRM_IOWR(0x27, drm_draw_t)
+#define DRM_IOCTL_RM_DRAW    DRM_IOWR(0x28, drm_draw_t)
+#define DRM_IOCTL_DMA	     DRM_IOWR(0x29, drm_dma_t)
+#define DRM_IOCTL_LOCK	     DRM_IOW( 0x2a, drm_lock_t)
+#define DRM_IOCTL_UNLOCK     DRM_IOW( 0x2b, drm_lock_t)
+#define DRM_IOCTL_FINISH     DRM_IOW( 0x2c, drm_lock_t)
+
+#define DRM_IOCTL_AGP_ACQUIRE DRM_IO(  0x30)
+#define DRM_IOCTL_AGP_RELEASE DRM_IO(  0x31)
+#define DRM_IOCTL_AGP_ENABLE  DRM_IOW( 0x32, drm_agp_mode_t)
+#define DRM_IOCTL_AGP_INFO    DRM_IOR( 0x33, drm_agp_info_t)
+#define DRM_IOCTL_AGP_ALLOC   DRM_IOWR(0x34, drm_agp_buffer_t)
+#define DRM_IOCTL_AGP_FREE    DRM_IOW( 0x35, drm_agp_buffer_t)
+#define DRM_IOCTL_AGP_BIND    DRM_IOW( 0x36, drm_agp_binding_t)
+#define DRM_IOCTL_AGP_UNBIND  DRM_IOW( 0x37, drm_agp_binding_t)
+
+/* Mga specific ioctls */
+#define DRM_IOCTL_MGA_INIT    DRM_IOW( 0x40, drm_mga_init_t)
+#define DRM_IOCTL_MGA_SWAP    DRM_IOW( 0x41, drm_mga_swap_t)
+#define DRM_IOCTL_MGA_CLEAR   DRM_IOW( 0x42, drm_mga_clear_t)
+#define DRM_IOCTL_MGA_ILOAD   DRM_IOW( 0x43, drm_mga_iload_t)
+#define DRM_IOCTL_MGA_VERTEX  DRM_IOW( 0x44, drm_mga_vertex_t)
+#define DRM_IOCTL_MGA_FLUSH   DRM_IOW( 0x45, drm_lock_t )
+#define DRM_IOCTL_MGA_INDICES DRM_IOW( 0x46, drm_mga_indices_t)
+
+/* I810 specific ioctls */
+#define DRM_IOCTL_I810_INIT    DRM_IOW( 0x40, drm_i810_init_t)
+#define DRM_IOCTL_I810_VERTEX  DRM_IOW( 0x41, drm_i810_vertex_t)
+#define DRM_IOCTL_I810_CLEAR   DRM_IOW( 0x42, drm_i810_clear_t)
+#define DRM_IOCTL_I810_FLUSH   DRM_IO ( 0x43)
+#define DRM_IOCTL_I810_GETAGE  DRM_IO ( 0x44)
+#define DRM_IOCTL_I810_GETBUF  DRM_IOWR(0x45, drm_i810_dma_t)
+#define DRM_IOCTL_I810_SWAP    DRM_IO ( 0x46)
+
+/* Rage 128 specific ioctls */
+#define DRM_IOCTL_R128_INIT	DRM_IOW( 0x40, drm_r128_init_t)
+#define DRM_IOCTL_R128_RESET	DRM_IO(  0x41)
+#define DRM_IOCTL_R128_FLUSH	DRM_IO(  0x42)
+#define DRM_IOCTL_R128_CCEIDL	DRM_IO(  0x43)
+#define DRM_IOCTL_R128_PACKET	DRM_IOW( 0x44, drm_r128_packet_t)
+#define DRM_IOCTL_R128_VERTEX	DRM_IOW( 0x45, drm_r128_vertex_t)
+
+#endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/Makefile b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/Makefile
new file mode 100644
index 000000000..6a70a5b0f
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/Makefile
@@ -0,0 +1,17 @@
+# $FreeBSD$
+
+KMOD	= drm
+SRCS	= init.c memory.c auth.c context.c drawable.c bufs.c \
+	  lists.c lock.c ioctl.c fops.c vm.c dma.c sysctl.c  \
+	  agpsupport.c
+SRCS   += device_if.h bus_if.h pci_if.h
+CFLAGS += ${DEBUG_FLAGS} -I.. # -DDRM_AGP
+KERN    = /usr/src/sys
+
+@:
+	ln -sf /sys @
+
+machine:
+	ln -sf /sys/i386/include machine
+
+.include "/usr/src/sys/conf/kmod.mk"
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/agpsupport.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/agpsupport.c
new file mode 100644
index 000000000..53444c907
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/agpsupport.c
@@ -0,0 +1,271 @@
+/* agpsupport.c -- DRM support for AGP/GART backend
+ * Created: Mon Dec 13 09:56:45 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * Author: Rickard E. (Rik) Faith <faith@precisioninsight.com>
+ *
+ * $XFree86$
+ *
+ */
+
+#define __NO_VERSION__
+#include "drmP.h"
+
+#ifdef DRM_AGP
+
+#include <pci/agpvar.h>
+
+MODULE_DEPEND(drm, agp, 1, 1, 1);
+
+int
+drm_agp_info(dev_t kdev, u_long cmd, caddr_t data,
+	     int flags, struct proc *p)
+{
+	drm_device_t	*dev	= kdev->si_drv1;
+	struct agp_info *kern;
+	drm_agp_info_t   info;
+
+	if (!dev->agp->acquired) return EINVAL;
+
+	kern                   = &dev->agp->info;
+	agp_get_info(dev->agp->agpdev, kern);
+	info.agp_version_major = 1;
+	info.agp_version_minor = 0;
+	info.mode              = kern->ai_mode;
+	info.aperture_base     = kern->ai_aperture_base;
+	info.aperture_size     = kern->ai_aperture_size;
+	info.memory_allowed    = kern->ai_memory_allowed;
+	info.memory_used       = kern->ai_memory_used;
+	info.id_vendor         = kern->ai_devid & 0xffff;
+	info.id_device         = kern->ai_devid >> 16;
+
+	*(drm_agp_info_t *) data = info;
+	return 0;
+}
+
+int
+drm_agp_acquire(dev_t kdev, u_long cmd, caddr_t data,
+		int flags, struct proc *p)
+{
+	drm_device_t *dev = kdev->si_drv1;
+	int          retcode;
+
+	if (dev->agp->acquired) return EINVAL;
+	retcode = agp_acquire(dev->agp->agpdev);
+	if (retcode) return retcode;
+	dev->agp->acquired = 1;
+	return 0;
+}
+
+int
+drm_agp_release(dev_t kdev, u_long cmd, caddr_t data,
+		int flags, struct proc *p)
+{
+	drm_device_t *dev = kdev->si_drv1;
+
+	if (!dev->agp->acquired) return EINVAL;
+	agp_release(dev->agp->agpdev);
+	dev->agp->acquired = 0;
+	return 0;
+	
+}
+
+int
+drm_agp_enable(dev_t kdev, u_long cmd, caddr_t data,
+	       int flags, struct proc *p)
+{
+	drm_device_t   *dev = kdev->si_drv1;
+	drm_agp_mode_t mode;
+
+	if (!dev->agp->acquired) return EINVAL;
+
+	mode = *(drm_agp_mode_t *) data;
+	
+	dev->agp->mode    = mode.mode;
+	agp_enable(dev->agp->agpdev, mode.mode);
+	dev->agp->base    = dev->agp->info.ai_aperture_base;
+	dev->agp->enabled = 1;
+	return 0;
+}
+
+int drm_agp_alloc(dev_t kdev, u_long cmd, caddr_t data,
+		  int flags, struct proc *p)
+{
+	drm_device_t     *dev = kdev->si_drv1;
+	drm_agp_buffer_t request;
+	drm_agp_mem_t    *entry;
+	void	         *handle;
+	unsigned long    pages;
+	u_int32_t	 type;
+	struct agp_memory_info info;
+
+	if (!dev->agp->acquired) return EINVAL;
+
+	request = *(drm_agp_buffer_t *) data;
+
+	if (!(entry = drm_alloc(sizeof(*entry), DRM_MEM_AGPLISTS)))
+		return ENOMEM;
+   
+   	memset(entry, 0, sizeof(*entry));
+
+	pages = (request.size + PAGE_SIZE - 1) / PAGE_SIZE;
+	type = (u_int32_t) request.type;
+
+	if (!(handle = drm_alloc_agp(pages, type))) {
+		drm_free(entry, sizeof(*entry), DRM_MEM_AGPLISTS);
+		return ENOMEM;
+	}
+	
+	entry->handle    = handle;
+	entry->bound     = 0;
+	entry->pages     = pages;
+	entry->prev      = NULL;
+	entry->next      = dev->agp->memory;
+	if (dev->agp->memory) dev->agp->memory->prev = entry;
+	dev->agp->memory = entry;
+
+	agp_memory_info(dev->agp->agpdev, entry->handle, &info);
+
+	request.handle   = (unsigned long) entry->handle;
+        request.physical = info.ami_physical;
+
+	*(drm_agp_buffer_t *) data = request;
+
+	return 0;
+}
+
+static drm_agp_mem_t *
+drm_agp_lookup_entry(drm_device_t *dev, void *handle)
+{
+	drm_agp_mem_t *entry;
+
+	for (entry = dev->agp->memory; entry; entry = entry->next) {
+		if (entry->handle == handle) return entry;
+	}
+	return NULL;
+}
+
+int
+drm_agp_unbind(dev_t kdev, u_long cmd, caddr_t data,
+	       int flags, struct proc *p)
+{
+	drm_device_t      *dev = kdev->si_drv1;
+	drm_agp_binding_t request;
+	drm_agp_mem_t     *entry;
+
+	if (!dev->agp->acquired) return EINVAL;
+	request = *(drm_agp_binding_t *) data;
+	if (!(entry = drm_agp_lookup_entry(dev, (void *) request.handle)))
+		return EINVAL;
+	if (!entry->bound) return EINVAL;
+	return drm_unbind_agp(entry->handle);
+}
+
+int drm_agp_bind(dev_t kdev, u_long cmd, caddr_t data,
+		 int flags, struct proc *p)
+{
+	drm_device_t      *dev = kdev->si_drv1;
+	drm_agp_binding_t request;
+	drm_agp_mem_t     *entry;
+	int               retcode;
+	int               page;
+	
+	if (!dev->agp->acquired) return EINVAL;
+	request = *(drm_agp_binding_t *) data;
+	if (!(entry = drm_agp_lookup_entry(dev, (void *) request.handle)))
+		return EINVAL;
+	if (entry->bound) return EINVAL;
+	page = (request.offset + PAGE_SIZE - 1) / PAGE_SIZE;
+	if ((retcode = drm_bind_agp(entry->handle, page))) return retcode;
+	entry->bound = dev->agp->base + (page << PAGE_SHIFT);
+	return 0;
+}
+
+int drm_agp_free(dev_t kdev, u_long cmd, caddr_t data,
+		 int flags, struct proc *p)
+{
+	drm_device_t     *dev = kdev->si_drv1;
+	drm_agp_buffer_t request;
+	drm_agp_mem_t    *entry;
+	
+	if (!dev->agp->acquired) return EINVAL;
+	request = *(drm_agp_buffer_t *) data;
+	if (!(entry = drm_agp_lookup_entry(dev, (void*) request.handle)))
+		return EINVAL;
+	if (entry->bound) drm_unbind_agp(entry->handle);
+   
+	if (entry->prev) entry->prev->next = entry->next;
+	else             dev->agp->memory  = entry->next;
+	if (entry->next) entry->next->prev = entry->prev;
+	drm_free_agp(entry->handle, entry->pages);
+	drm_free(entry, sizeof(*entry), DRM_MEM_AGPLISTS);
+	return 0;
+}
+
+drm_agp_head_t *drm_agp_init(void)
+{
+	device_t agpdev;
+	drm_agp_head_t *head   = NULL;
+	int      agp_available = 1;
+   
+	agpdev = agp_find_device();
+	if (!agpdev)
+		agp_available = 0;
+
+	DRM_DEBUG("agp_available = %d\n", agp_available);
+
+	if (agp_available) {
+		if (!(head = drm_alloc(sizeof(*head), DRM_MEM_AGPLISTS)))
+			return NULL;
+		head->agpdev = agpdev;
+		memset((void *)head, 0, sizeof(*head));
+		agp_get_info(agpdev, &head->info);
+		head->memory = NULL;
+#if 0				/* bogus */
+		switch (head->agp_info.chipset) {
+		case INTEL_GENERIC:  head->chipset = "Intel";          break;
+		case INTEL_LX:       head->chipset = "Intel 440LX";    break;
+		case INTEL_BX:       head->chipset = "Intel 440BX";    break;
+		case INTEL_GX:       head->chipset = "Intel 440GX";    break;
+		case INTEL_I810:     head->chipset = "Intel i810";     break;
+		case VIA_GENERIC:    head->chipset = "VIA";            break;
+		case VIA_VP3:        head->chipset = "VIA VP3";        break;
+		case VIA_MVP3:       head->chipset = "VIA MVP3";       break;
+		case VIA_APOLLO_PRO: head->chipset = "VIA Apollo Pro"; break;
+		case SIS_GENERIC:    head->chipset = "SiS";            break;
+		case AMD_GENERIC:    head->chipset = "AMD";            break;
+		case AMD_IRONGATE:   head->chipset = "AMD Irongate";   break;
+		case ALI_GENERIC:    head->chipset = "ALi";            break;
+		case ALI_M1541:      head->chipset = "ALi M1541";      break;
+		default:
+		}
+#endif
+		DRM_INFO("AGP at 0x%08x %dMB\n",
+			 head->info.ai_aperture_base,
+			 head->info.ai_aperture_size >> 20);
+	}
+	return head;
+}
+
+#endif /* DRM_AGP */
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/auth.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/auth.c
new file mode 100644
index 000000000..f7b3bc493
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/auth.c
@@ -0,0 +1,168 @@
+/* auth.c -- IOCTLs for authentication -*- c -*-
+ * Created: Tue Feb  2 08:37:54 1999 by faith@precisioninsight.com
+ * Revised: Fri Aug 20 11:31:48 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/auth.c,v 1.4 1999/08/30 13:05:00 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/auth.c,v 1.1 1999/09/25 14:37:57 dawes Exp $
+ *
+ */
+
+#define __NO_VERSION__
+#include "drmP.h"
+
+static int drm_hash_magic(drm_magic_t magic)
+{
+	return magic & (DRM_HASH_SIZE-1);
+}
+
+static drm_file_t *drm_find_file(drm_device_t *dev, drm_magic_t magic)
+{
+	drm_file_t	  *retval = NULL;
+	drm_magic_entry_t *pt;
+	int		  hash	  = drm_hash_magic(magic);
+
+	lockmgr(&dev->dev_lock, LK_EXCLUSIVE, 0, curproc);
+	for (pt = dev->magiclist[hash].head; pt; pt = pt->next) {
+		if (pt->priv->authenticated) continue;
+		if (pt->magic == magic) {
+			retval = pt->priv;
+			break;
+		}
+	}
+	lockmgr(&dev->dev_lock, LK_RELEASE, 0, curproc);
+
+	return retval;
+}
+
+int drm_add_magic(drm_device_t *dev, drm_file_t *priv, drm_magic_t magic)
+{
+	int		  hash;
+	drm_magic_entry_t *entry;
+	
+	DRM_DEBUG("%d\n", magic);
+	
+	hash	     = drm_hash_magic(magic);
+	entry	     = drm_alloc(sizeof(*entry), DRM_MEM_MAGIC);
+	if (!entry) return ENOMEM;
+	entry->magic = magic;
+	entry->priv  = priv;
+	entry->next  = NULL;
+
+	lockmgr(&dev->dev_lock, LK_EXCLUSIVE, 0, curproc);
+	if (dev->magiclist[hash].tail) {
+		dev->magiclist[hash].tail->next = entry;
+		dev->magiclist[hash].tail	= entry;
+	} else {
+		dev->magiclist[hash].head	= entry;
+		dev->magiclist[hash].tail	= entry;
+	}
+	lockmgr(&dev->dev_lock, LK_RELEASE, 0, curproc);
+	
+	return 0;
+}
+
+int drm_remove_magic(drm_device_t *dev, drm_magic_t magic)
+{
+	drm_magic_entry_t *prev = NULL;
+	drm_magic_entry_t *pt;
+	int		  hash;
+	
+	DRM_DEBUG("%d\n", magic);
+	hash = drm_hash_magic(magic);
+	
+	lockmgr(&dev->dev_lock, LK_EXCLUSIVE, 0, curproc);
+	for (pt = dev->magiclist[hash].head; pt; prev = pt, pt = pt->next) {
+		if (pt->magic == magic) {
+			if (dev->magiclist[hash].head == pt) {
+				dev->magiclist[hash].head = pt->next;
+			}
+			if (dev->magiclist[hash].tail == pt) {
+				dev->magiclist[hash].tail = prev;
+			}
+			if (prev) {
+				prev->next = pt->next;
+			}
+			lockmgr(&dev->dev_lock, LK_RELEASE, 0, curproc);
+			return 0;
+		}
+	}
+	lockmgr(&dev->dev_lock, LK_RELEASE, 0, curproc);
+
+	drm_free(pt, sizeof(*pt), DRM_MEM_MAGIC);
+	
+	return EINVAL;
+}
+
+int drm_getmagic(dev_t kdev, u_long cmd, caddr_t data,
+		 int flags, struct proc *p)
+{
+	static drm_magic_t sequence = 0;
+#if 0
+	static struct simplelock lock; /* XXX */
+#endif
+	drm_device_t	   *dev = kdev->si_drv1;
+	drm_file_t	   *priv;
+	drm_auth_t	   auth;
+
+				/* Find unique magic */
+	priv = drm_find_file_by_proc(dev, p);
+	if (!priv) {
+	    DRM_DEBUG("can't find file structure\n");
+	    return EINVAL;
+	}
+	if (priv->magic) {
+		auth.magic = priv->magic;
+	} else {
+		simple_lock(&lock);
+		do {
+			if (!sequence) ++sequence; /* reserve 0 */
+			auth.magic = sequence++;
+		} while (drm_find_file(dev, auth.magic));
+		simple_unlock(&lock);
+		priv->magic = auth.magic;
+		drm_add_magic(dev, priv, auth.magic);
+	}
+	
+	DRM_DEBUG("%u\n", auth.magic);
+	*(drm_auth_t *) data = auth;
+	return 0;
+}
+
+int drm_authmagic(dev_t kdev, u_long cmd, caddr_t data,
+		  int flags, struct proc *p)
+{
+	drm_device_t	   *dev	    = kdev->si_drv1;
+	drm_auth_t	   auth;
+	drm_file_t	   *file;
+
+	auth = *(drm_auth_t *) data;
+	DRM_DEBUG("%u\n", auth.magic);
+	if ((file = drm_find_file(dev, auth.magic))) {
+		file->authenticated = 1;
+		drm_remove_magic(dev, auth.magic);
+		return 0;
+	}
+	return EINVAL;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/bufs.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/bufs.c
new file mode 100644
index 000000000..fc08b69d2
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/bufs.c
@@ -0,0 +1,500 @@
+/* bufs.c -- IOCTLs to manage buffers -*- c -*-
+ * Created: Tue Feb  2 08:37:54 1999 by faith@precisioninsight.com
+ * Revised: Fri Aug 20 22:48:10 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/bufs.c,v 1.8 1999/08/30 13:05:00 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/bufs.c,v 1.1 1999/09/25 14:37:57 dawes Exp $
+ *
+ */
+
+#define __NO_VERSION__
+#include "drmP.h"
+#include <sys/mman.h>
+#include <vm/vm.h>
+#include <vm/pmap.h>
+#include <vm/vm_extern.h>
+#include <vm/vm_map.h>
+
+				/* Compute order.  Can be made faster. */
+int drm_order(unsigned long size)
+{
+	int	      order;
+	unsigned long tmp;
+
+	for (order = 0, tmp = size; tmp >>= 1; ++order);
+	if (size & ~(1 << order)) ++order;
+	return order;
+}
+
+int drm_addmap(dev_t kdev, u_long cmd, caddr_t data,
+	       int flags, struct proc *p)
+{
+	drm_device_t	*dev	= kdev->si_drv1;
+	drm_map_t	*map;
+	
+	if (!(dev->flags & (FREAD|FWRITE)))
+		return EACCES; /* Require read/write */
+
+	map	     = drm_alloc(sizeof(*map), DRM_MEM_MAPS);
+	if (!map) return ENOMEM;
+	*map = *(drm_map_t *) data;
+
+	DRM_DEBUG("offset = 0x%08lx, size = 0x%08lx, type = %d\n",
+		  map->offset, map->size, map->type);
+	if ((map->offset & (PAGE_SIZE-1)) || (map->size & (PAGE_SIZE-1))) {
+		drm_free(map, sizeof(*map), DRM_MEM_MAPS);
+		DRM_DEBUG("offset or size not page aligned\n");
+		return EINVAL;
+	}
+	map->mtrr   = -1;
+	map->handle = 0;
+
+	switch (map->type) {
+	case _DRM_REGISTERS:
+	case _DRM_FRAME_BUFFER:	
+		if (map->offset + map->size < map->offset
+		    /* || map->offset < virt_to_phys(high_memory) */) {
+			drm_free(map, sizeof(*map), DRM_MEM_MAPS);
+			DRM_DEBUG("bad frame buffer size\n");
+			return EINVAL;
+		}
+#ifdef CONFIG_MTRR
+		if (map->type == _DRM_FRAME_BUFFER
+		    || (map->flags & _DRM_WRITE_COMBINING)) {
+			map->mtrr = mtrr_add(map->offset, map->size,
+					     MTRR_TYPE_WRCOMB, 1);
+		}
+#endif
+		map->handle = drm_ioremap(map->offset, map->size);
+		break;
+			
+
+	case _DRM_SHM:
+		DRM_DEBUG("%ld %d\n", map->size, drm_order(map->size));
+		map->handle = (void *)drm_alloc_pages(drm_order(map->size)
+						      - PAGE_SHIFT,
+						      DRM_MEM_SAREA);
+		if (!map->handle) {
+			drm_free(map, sizeof(*map), DRM_MEM_MAPS);
+			return ENOMEM;
+		}
+		map->offset = (unsigned long)map->handle;
+		if (map->flags & _DRM_CONTAINS_LOCK) {
+			dev->lock.hw_lock = map->handle; /* Pointer to lock */
+		}
+		break;
+#ifdef DRM_AGP
+	case _DRM_AGP:
+		map->offset = map->offset + dev->agp->base;
+		break;
+#endif
+	default:
+		drm_free(map, sizeof(*map), DRM_MEM_MAPS);
+		DRM_DEBUG("bad type\n");
+		return EINVAL;
+	}
+
+	lockmgr(&dev->dev_lock, LK_EXCLUSIVE, 0, curproc);
+	if (dev->maplist) {
+		++dev->map_count;
+		dev->maplist = drm_realloc(dev->maplist,
+					   (dev->map_count-1)
+					   * sizeof(*dev->maplist),
+					   dev->map_count
+					   * sizeof(*dev->maplist),
+					   DRM_MEM_MAPS);
+	} else {
+		dev->map_count = 1;
+		dev->maplist = drm_alloc(dev->map_count*sizeof(*dev->maplist),
+					 DRM_MEM_MAPS);
+	}
+	dev->maplist[dev->map_count-1] = map;
+	lockmgr(&dev->dev_lock, LK_RELEASE, 0, curproc);
+
+	*(drm_map_t *) data = *map;
+	if (map->type != _DRM_SHM)
+		((drm_map_t *)data)->handle = (void *) map->offset;
+
+	return 0;
+}
+
+int drm_addbufs(dev_t kdev, u_long cmd, caddr_t data,
+		int flags, struct proc *p)
+{
+	drm_device_t	 *dev	 = kdev->si_drv1;
+	drm_device_dma_t *dma	 = dev->dma;
+	drm_buf_desc_t	 request;
+	int		 count;
+	int		 order;
+	int		 size;
+	int		 total;
+	int		 page_order;
+	drm_buf_entry_t	 *entry;
+	unsigned long	 page;
+	drm_buf_t	 *buf;
+	int		 alignment;
+	unsigned long	 offset;
+	int		 i;
+	int		 byte_count;
+	int		 page_count;
+
+	if (!dma) return EINVAL;
+
+	request = *(drm_buf_desc_t *) data;
+
+	count	   = request.count;
+	order	   = drm_order(request.size);
+	size	   = 1 << order;
+	
+	DRM_DEBUG("count = %d, size = %d (%d), order = %d, queue_count = %d\n",
+		  request.count, request.size, size, order, dev->queue_count);
+
+	if (order < DRM_MIN_ORDER || order > DRM_MAX_ORDER) return EINVAL;
+	if (dev->queue_count) return EBUSY; /* Not while in use */
+
+	alignment  = (request.flags & _DRM_PAGE_ALIGN) ? round_page(size) :size;
+	page_order = order - PAGE_SHIFT > 0 ? order - PAGE_SHIFT : 0;
+	total	   = PAGE_SIZE << page_order;
+
+	simple_lock(&dev->count_lock);
+	if (dev->buf_use) {
+		simple_unlock(&dev->count_lock);
+		return EBUSY;
+	}
+	atomic_inc(&dev->buf_alloc);
+	simple_unlock(&dev->count_lock);
+	
+	lockmgr(&dev->dev_lock, LK_EXCLUSIVE, 0, curproc);
+	entry = &dma->bufs[order];
+	if (entry->buf_count) {
+		lockmgr(&dev->dev_lock, LK_RELEASE, 0, curproc);
+		atomic_dec(&dev->buf_alloc);
+		return ENOMEM;	/* May only call once for each order */
+	}
+	
+	entry->buflist = drm_alloc(count * sizeof(*entry->buflist),
+				   DRM_MEM_BUFS);
+	if (!entry->buflist) {
+		lockmgr(&dev->dev_lock, LK_RELEASE, 0, curproc);
+		atomic_dec(&dev->buf_alloc);
+		return ENOMEM;
+	}
+	memset(entry->buflist, 0, count * sizeof(*entry->buflist));
+
+	entry->seglist = drm_alloc(count * sizeof(*entry->seglist),
+				   DRM_MEM_SEGS);
+	if (!entry->seglist) {
+		drm_free(entry->buflist,
+			 count * sizeof(*entry->buflist),
+			 DRM_MEM_BUFS);
+		lockmgr(&dev->dev_lock, LK_RELEASE, 0, curproc);
+		atomic_dec(&dev->buf_alloc);
+		return ENOMEM;
+	}
+	memset(entry->seglist, 0, count * sizeof(*entry->seglist));
+
+	dma->pagelist = drm_realloc(dma->pagelist,
+				    dma->page_count * sizeof(*dma->pagelist),
+				    (dma->page_count + (count << page_order))
+				    * sizeof(*dma->pagelist),
+				    DRM_MEM_PAGES);
+	DRM_DEBUG("pagelist: %d entries\n",
+		  dma->page_count + (count << page_order));
+
+
+	entry->buf_size	  = size;
+	entry->page_order = page_order;
+	byte_count	  = 0;
+	page_count	  = 0;
+	while (entry->buf_count < count) {
+		if (!(page = drm_alloc_pages(page_order, DRM_MEM_DMA))) break;
+		entry->seglist[entry->seg_count++] = page;
+		for (i = 0; i < (1 << page_order); i++) {
+			DRM_DEBUG("page %d @ 0x%08lx\n",
+				  dma->page_count + page_count,
+				  page + PAGE_SIZE * i);
+			dma->pagelist[dma->page_count + page_count++]
+				= page + PAGE_SIZE * i;
+		}
+		for (offset = 0;
+		     offset + size <= total && entry->buf_count < count;
+		     offset += alignment, ++entry->buf_count) {
+			buf	     = &entry->buflist[entry->buf_count];
+			buf->idx     = dma->buf_count + entry->buf_count;
+			buf->total   = alignment;
+			buf->order   = order;
+			buf->used    = 0;
+			buf->offset  = (dma->byte_count + byte_count + offset);
+			buf->address = (void *)(page + offset);
+			buf->next    = NULL;
+			buf->waiting = 0;
+			buf->pending = 0;
+			buf->dma_wait = 0;
+			buf->pid     = 0;
+#if DRM_DMA_HISTOGRAM
+			timespecclear(&buf->time_queued);
+			timespecclear(&buf->time_dispatched);
+			timespecclear(&buf->time_completed);
+			timespecclear(&buf->time_freed);
+#endif
+			DRM_DEBUG("buffer %d @ %p\n",
+				  entry->buf_count, buf->address);
+		}
+		byte_count += PAGE_SIZE << page_order;
+	}
+
+	dma->buflist = drm_realloc(dma->buflist,
+				   dma->buf_count * sizeof(*dma->buflist),
+				   (dma->buf_count + entry->buf_count)
+				   * sizeof(*dma->buflist),
+				   DRM_MEM_BUFS);
+	for (i = dma->buf_count; i < dma->buf_count + entry->buf_count; i++)
+		dma->buflist[i] = &entry->buflist[i - dma->buf_count];
+
+	dma->buf_count	+= entry->buf_count;
+	dma->seg_count	+= entry->seg_count;
+	dma->page_count += entry->seg_count << page_order;
+	dma->byte_count += PAGE_SIZE * (entry->seg_count << page_order);
+	
+	drm_freelist_create(&entry->freelist, entry->buf_count);
+	for (i = 0; i < entry->buf_count; i++) {
+		drm_freelist_put(dev, &entry->freelist, &entry->buflist[i]);
+	}
+	
+	lockmgr(&dev->dev_lock, LK_RELEASE, 0, curproc);
+
+	request.count = entry->buf_count;
+	request.size  = size;
+
+	*(drm_buf_desc_t *) data = request;
+	
+	atomic_dec(&dev->buf_alloc);
+	return 0;
+}
+
+int drm_infobufs(dev_t kdev, u_long cmd, caddr_t data,
+		 int flags, struct proc *p)
+{
+	drm_device_t	 *dev	 = kdev->si_drv1;
+	drm_device_dma_t *dma	 = dev->dma;
+	drm_buf_info_t	 request;
+	int		 i;
+	int		 count;
+
+	if (!dma) return EINVAL;
+
+	simple_lock(&dev->count_lock);
+	if (atomic_read(&dev->buf_alloc)) {
+		simple_unlock(&dev->count_lock);
+		return EBUSY;
+	}
+	++dev->buf_use;		/* Can't allocate more after this call */
+	simple_unlock(&dev->count_lock);
+
+	request = *(drm_buf_info_t *) data;
+
+	for (i = 0, count = 0; i < DRM_MAX_ORDER+1; i++) {
+		if (dma->bufs[i].buf_count) ++count;
+	}
+	
+	DRM_DEBUG("count = %d\n", count);
+	
+	if (request.count >= count) {
+		for (i = 0, count = 0; i < DRM_MAX_ORDER+1; i++) {
+			if (dma->bufs[i].buf_count) {
+				int error;
+				error = copyout(&dma->bufs[i].buf_count,
+						&request.list[count].count,
+						sizeof(dma->bufs[0]
+						       .buf_count));
+				if (error) return error;
+				error = copyout(&dma->bufs[i].buf_size,
+						&request.list[count].size,
+						sizeof(dma->bufs[0].buf_size));
+				if (error) return error;
+				error = copyout(&dma->bufs[i]
+						.freelist.low_mark,
+						&request.list[count].low_mark,
+						sizeof(dma->bufs[0]
+						       .freelist.low_mark));
+				if (error) return error;
+				error = copyout(&dma->bufs[i]
+						.freelist.high_mark,
+						&request.list[count].high_mark,
+						sizeof(dma->bufs[0]
+						       .freelist.high_mark));
+				if (error) return error;
+				DRM_DEBUG("%d %d %d %d %d\n",
+					  i,
+					  dma->bufs[i].buf_count,
+					  dma->bufs[i].buf_size,
+					  dma->bufs[i].freelist.low_mark,
+					  dma->bufs[i].freelist.high_mark);
+				++count;
+			}
+		}
+	}
+	request.count = count;
+
+	*(drm_buf_info_t *) data = request;
+	
+	return 0;
+}
+
+int drm_markbufs(dev_t kdev, u_long cmd, caddr_t data,
+		 int flags, struct proc *p)
+{
+	drm_device_t	 *dev	 = kdev->si_drv1;
+	drm_device_dma_t *dma	 = dev->dma;
+	drm_buf_desc_t	 request;
+	int		 order;
+	drm_buf_entry_t	 *entry;
+
+	if (!dma) return EINVAL;
+
+	request = *(drm_buf_desc_t *) data;
+
+	DRM_DEBUG("%d, %d, %d\n",
+		  request.size, request.low_mark, request.high_mark);
+	order = drm_order(request.size);
+	if (order < DRM_MIN_ORDER || order > DRM_MAX_ORDER) return EINVAL;
+	entry = &dma->bufs[order];
+
+	if (request.low_mark < 0 || request.low_mark > entry->buf_count)
+		return EINVAL;
+	if (request.high_mark < 0 || request.high_mark > entry->buf_count)
+		return EINVAL;
+
+	entry->freelist.low_mark  = request.low_mark;
+	entry->freelist.high_mark = request.high_mark;
+	
+	return 0;
+}
+
+int drm_freebufs(dev_t kdev, u_long cmd, caddr_t data,
+		 int flags, struct proc *p)
+{
+	drm_device_t	 *dev	 = kdev->si_drv1;
+	drm_device_dma_t *dma	 = dev->dma;
+	drm_buf_free_t	 request;
+	int		 i;
+	int		 idx;
+	int		 error;
+	drm_buf_t	 *buf;
+
+	if (!dma) return EINVAL;
+
+	request = *(drm_buf_free_t *) data;
+
+	DRM_DEBUG("%d\n", request.count);
+	for (i = 0; i < request.count; i++) {
+		error = copyin(&request.list[i], &idx, sizeof(idx));
+		if (error)
+			return error;
+		if (idx < 0 || idx >= dma->buf_count) {
+			DRM_ERROR("Index %d (of %d max)\n",
+				  idx, dma->buf_count - 1);
+			return EINVAL;
+		}
+		buf = dma->buflist[idx];
+		if (buf->pid != p->p_pid) {
+			DRM_ERROR("Process %d freeing buffer owned by %d\n",
+				  p->p_pid, buf->pid);
+			return EINVAL;
+		}
+		drm_free_buffer(dev, buf);
+	}
+	
+	return 0;
+}
+
+int drm_mapbufs(dev_t kdev, u_long cmd, caddr_t data,
+		int flags, struct proc *p)
+{
+	drm_device_t	 *dev	 = kdev->si_drv1;
+	drm_device_dma_t *dma	 = dev->dma;
+	int		 retcode = 0;
+	const int	 zero	 = 0;
+	vm_offset_t	 virtual;
+	vm_offset_t	 address;
+	drm_buf_map_t	 request;
+	int		 i;
+
+	if (!dma) return EINVAL;
+	
+	DRM_DEBUG("\n");
+
+	simple_lock(&dev->count_lock);
+	if (atomic_read(&dev->buf_alloc)) {
+		simple_unlock(&dev->count_lock);
+		return EBUSY;
+	}
+	++dev->buf_use;		/* Can't allocate more after this call */
+	simple_unlock(&dev->count_lock);
+
+	request = *(drm_buf_map_t *) data;
+
+	if (request.count >= dma->buf_count) {
+		virtual = 0;
+		retcode = vm_mmap(&p->p_vmspace->vm_map,
+				  &virtual,
+				  round_page(dma->byte_count),
+				  PROT_READ|PROT_WRITE, VM_PROT_ALL,
+				  MAP_SHARED,
+				  SLIST_FIRST(&kdev->si_hlist),
+				  0);
+		if (retcode)
+			goto done;
+
+		request.virtual = (void *)virtual;
+
+		for (i = 0; i < dma->buf_count; i++) {
+			retcode = copyout(&dma->buflist[i]->idx,
+					  &request.list[i].idx,
+					  sizeof(request.list[0].idx));
+			if (retcode) goto done;
+			retcode = copyout(&dma->buflist[i]->total,
+					  &request.list[i].total,
+					  sizeof(request.list[0].total));
+			if (retcode) goto done;
+			retcode = copyout(&zero,
+					  &request.list[i].used,
+					  sizeof(request.list[0].used));
+			if (retcode) goto done;
+			address = virtual + dma->buflist[i]->offset;
+			retcode = copyout(&address,
+					  &request.list[i].address,
+					  sizeof(address));
+			if (retcode) goto done;
+		}
+	}
+done:
+	request.count = dma->buf_count;
+	DRM_DEBUG("%d buffers, retcode = %d\n", request.count, retcode);
+
+	*(drm_buf_map_t *) data = request;
+
+	return retcode;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/context.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/context.c
new file mode 100644
index 000000000..d79990f86
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/context.c
@@ -0,0 +1,297 @@
+/* context.c -- IOCTLs for contexts and DMA queues -*- c -*-
+ * Created: Tue Feb  2 08:37:54 1999 by faith@precisioninsight.com
+ * Revised: Fri Aug 20 11:32:09 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/context.c,v 1.5 1999/08/30 13:05:00 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/context.c,v 1.1 1999/09/25 14:37:58 dawes Exp $
+ *
+ */
+
+#define __NO_VERSION__
+#include "drmP.h"
+
+static int drm_init_queue(drm_device_t *dev, drm_queue_t *q, drm_ctx_t *ctx)
+{
+	DRM_DEBUG("\n");
+	
+	if (atomic_read(&q->use_count) != 1
+	    || atomic_read(&q->finalization)
+	    || atomic_read(&q->block_count)) {
+		DRM_ERROR("New queue is already in use: u%d f%d b%d\n",
+			  atomic_read(&q->use_count),
+			  atomic_read(&q->finalization),
+			  atomic_read(&q->block_count));
+	}
+		  
+	atomic_set(&q->finalization,  0);
+	atomic_set(&q->block_count,   0);
+	atomic_set(&q->block_read,    0);
+	atomic_set(&q->block_write,   0);
+	atomic_set(&q->total_queued,  0);
+	atomic_set(&q->total_flushed, 0);
+	atomic_set(&q->total_locks,   0);
+
+	q->write_queue = 0;
+	q->read_queue = 0;
+	q->flush_queue = 0;
+
+	q->flags = ctx->flags;
+
+	drm_waitlist_create(&q->waitlist, dev->dma->buf_count);
+
+	return 0;
+}
+
+
+/* drm_alloc_queue:
+PRE: 1) dev->queuelist[0..dev->queue_count] is allocated and will not
+	disappear (so all deallocation must be done after IOCTLs are off)
+     2) dev->queue_count < dev->queue_slots
+     3) dev->queuelist[i].use_count == 0 and
+	dev->queuelist[i].finalization == 0 if i not in use 
+POST: 1) dev->queuelist[i].use_count == 1
+      2) dev->queue_count < dev->queue_slots */
+		
+static int drm_alloc_queue(drm_device_t *dev)
+{
+	int	    i;
+	drm_queue_t *queue;
+	int	    oldslots;
+	int	    newslots;
+				/* Check for a free queue */
+	for (i = 0; i < dev->queue_count; i++) {
+		atomic_inc(&dev->queuelist[i]->use_count);
+		if (atomic_read(&dev->queuelist[i]->use_count) == 1
+		    && !atomic_read(&dev->queuelist[i]->finalization)) {
+			DRM_DEBUG("%d (free)\n", i);
+			return i;
+		}
+		atomic_dec(&dev->queuelist[i]->use_count);
+	}
+				/* Allocate a new queue */
+	lockmgr(&dev->dev_lock, LK_EXCLUSIVE, 0, curproc);
+	
+	queue = drm_alloc(sizeof(*queue), DRM_MEM_QUEUES);
+	memset(queue, 0, sizeof(*queue));
+	atomic_set(&queue->use_count, 1);
+	
+	++dev->queue_count;
+	if (dev->queue_count >= dev->queue_slots) {
+		oldslots = dev->queue_slots * sizeof(*dev->queuelist);
+		if (!dev->queue_slots) dev->queue_slots = 1;
+		dev->queue_slots *= 2;
+		newslots = dev->queue_slots * sizeof(*dev->queuelist);
+
+		dev->queuelist = drm_realloc(dev->queuelist,
+					     oldslots,
+					     newslots,
+					     DRM_MEM_QUEUES);
+		if (!dev->queuelist) {
+			lockmgr(&dev->dev_lock, LK_RELEASE, 0, curproc);
+			DRM_DEBUG("out of memory\n");
+			return -ENOMEM;
+		}
+	}
+	dev->queuelist[dev->queue_count-1] = queue;
+	
+	lockmgr(&dev->dev_lock, LK_RELEASE, 0, curproc);
+	DRM_DEBUG("%d (new)\n", dev->queue_count - 1);
+	return dev->queue_count - 1;
+}
+
+int drm_resctx(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	drm_ctx_res_t	res;
+	drm_ctx_t	ctx;
+	int		i;
+	int		error;
+
+	DRM_DEBUG("%d\n", DRM_RESERVED_CONTEXTS);
+	res = *(drm_ctx_res_t *) data;
+	if (res.count >= DRM_RESERVED_CONTEXTS) {
+		memset(&ctx, 0, sizeof(ctx));
+		for (i = 0; i < DRM_RESERVED_CONTEXTS; i++) {
+			ctx.handle = i;
+			error = copyout(&i, &res.contexts[i],
+					sizeof(i));
+			if (error) return error;
+		}
+	}
+	res.count = DRM_RESERVED_CONTEXTS;
+	*(drm_ctx_res_t *) data = res;
+	return 0;
+}
+
+
+int drm_addctx(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	drm_device_t	*dev	= kdev->si_drv1;
+	drm_ctx_t	ctx;
+
+	ctx = *(drm_ctx_t *) data;
+	if ((ctx.handle = drm_alloc_queue(dev)) == DRM_KERNEL_CONTEXT) {
+				/* Init kernel's context and get a new one. */
+		drm_init_queue(dev, dev->queuelist[ctx.handle], &ctx);
+		ctx.handle = drm_alloc_queue(dev);
+	}
+	drm_init_queue(dev, dev->queuelist[ctx.handle], &ctx);
+	DRM_DEBUG("%d\n", ctx.handle);
+	*(drm_ctx_t *) data = ctx;
+	return 0;
+}
+
+int drm_modctx(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	drm_device_t	*dev	= kdev->si_drv1;
+	drm_ctx_t	ctx;
+	drm_queue_t	*q;
+		
+	ctx = *(drm_ctx_t *) data;
+	
+	DRM_DEBUG("%d\n", ctx.handle);
+	
+	if (ctx.handle < 0 || ctx.handle >= dev->queue_count) return -EINVAL;
+	q = dev->queuelist[ctx.handle];
+	
+	atomic_inc(&q->use_count);
+	if (atomic_read(&q->use_count) == 1) {
+				/* No longer in use */
+		atomic_dec(&q->use_count);
+		return -EINVAL;
+	}
+
+	if (DRM_BUFCOUNT(&q->waitlist)) {
+		atomic_dec(&q->use_count);
+		return -EBUSY;
+	}
+	
+	q->flags = ctx.flags;
+	
+	atomic_dec(&q->use_count);
+	return 0;
+}
+
+int drm_getctx(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	drm_device_t	*dev	= kdev->si_drv1;
+	drm_ctx_t	ctx;
+	drm_queue_t	*q;
+		
+	ctx = *(drm_ctx_t *) data;
+	
+	DRM_DEBUG("%d\n", ctx.handle);
+	
+	if (ctx.handle >= dev->queue_count) return -EINVAL;
+	q = dev->queuelist[ctx.handle];
+	
+	atomic_inc(&q->use_count);
+	if (atomic_read(&q->use_count) == 1) {
+				/* No longer in use */
+		atomic_dec(&q->use_count);
+		return -EINVAL;
+	}
+	
+	ctx.flags = q->flags;
+	atomic_dec(&q->use_count);
+	
+	*(drm_ctx_t *) data = ctx;
+	
+	return 0;
+}
+
+int drm_switchctx(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	drm_device_t	*dev	= kdev->si_drv1;
+	drm_ctx_t	ctx;
+
+	ctx = *(drm_ctx_t *) data;
+	DRM_DEBUG("%d\n", ctx.handle);
+	return drm_context_switch(dev, dev->last_context, ctx.handle);
+}
+
+int drm_newctx(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	drm_device_t	*dev	= kdev->si_drv1;
+	drm_ctx_t	ctx;
+
+	ctx = *(drm_ctx_t *) data;
+	DRM_DEBUG("%d\n", ctx.handle);
+	drm_context_switch_complete(dev, ctx.handle);
+
+	return 0;
+}
+
+int drm_rmctx(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	drm_device_t	*dev	= kdev->si_drv1;
+	drm_ctx_t	ctx;
+	drm_queue_t	*q;
+	drm_buf_t	*buf;
+
+	ctx = *(drm_ctx_t *) data;
+	DRM_DEBUG("%d\n", ctx.handle);
+	
+	if (ctx.handle >= dev->queue_count) return -EINVAL;
+	q = dev->queuelist[ctx.handle];
+	
+	atomic_inc(&q->use_count);
+	if (atomic_read(&q->use_count) == 1) {
+				/* No longer in use */
+		atomic_dec(&q->use_count);
+		return -EINVAL;
+	}
+	
+	atomic_inc(&q->finalization); /* Mark queue in finalization state */
+	atomic_sub(2, &q->use_count); /* Mark queue as unused (pending
+					 finalization) */
+
+	/* Wait while interrupt servicing is in progress */
+	while (test_and_set_bit(0, &dev->interrupt_flag)) {
+		int never;
+		int error = tsleep(&never, PZERO|PCATCH, "drmrc", 1);
+		if (error) {
+			clear_bit(0, &dev->interrupt_flag);
+			return error;
+		}
+	}
+				/* Remove queued buffers */
+	while ((buf = drm_waitlist_get(&q->waitlist))) {
+		drm_free_buffer(dev, buf);
+	}
+	clear_bit(0, &dev->interrupt_flag);
+	
+				/* Wakeup blocked processes */
+	wakeup(&q->read_queue);
+	wakeup(&q->write_queue);
+	wakeup(&q->flush_queue);
+	
+				/* Finalization over.  Queue is made
+				   available when both use_count and
+				   finalization become 0, which won't
+				   happen until all the waiting processes
+				   stop waiting. */
+	atomic_dec(&q->finalization);
+	return 0;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/dma.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/dma.c
new file mode 100644
index 000000000..149f45934
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/dma.c
@@ -0,0 +1,534 @@
+/* dma.c -- DMA IOCTL and function support -*- c -*-
+ * Created: Fri Mar 19 14:30:16 1999 by faith@precisioninsight.com
+ * Revised: Thu Sep 16 12:55:39 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/dma.c,v 1.7 1999/09/16 16:56:18 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/dma.c,v 1.1 1999/09/25 14:37:58 dawes Exp $
+ *
+ */
+
+#define __NO_VERSION__
+#include "drmP.h"
+
+void drm_dma_setup(drm_device_t *dev)
+{
+	int i;
+	
+	dev->dma = drm_alloc(sizeof(*dev->dma), DRM_MEM_DRIVER);
+	memset(dev->dma, 0, sizeof(*dev->dma));
+	for (i = 0; i <= DRM_MAX_ORDER; i++)
+		memset(&dev->dma->bufs[i], 0, sizeof(dev->dma->bufs[0]));
+}
+
+void drm_dma_takedown(drm_device_t *dev)
+{
+	drm_device_dma_t  *dma = dev->dma;
+	int		  i, j;
+
+	if (!dma) return;
+	
+				/* Clear dma buffers */
+	for (i = 0; i <= DRM_MAX_ORDER; i++) {
+		if (dma->bufs[i].seg_count) {
+			DRM_DEBUG("order %d: buf_count = %d,"
+				  " seg_count = %d\n",
+				  i,
+				  dma->bufs[i].buf_count,
+				  dma->bufs[i].seg_count);
+			for (j = 0; j < dma->bufs[i].seg_count; j++) {
+				drm_free_pages(dma->bufs[i].seglist[j],
+					       dma->bufs[i].page_order,
+					       DRM_MEM_DMA);
+			}
+			drm_free(dma->bufs[i].buflist,
+				 dma->buf_count
+				 * sizeof(*dma->bufs[0].buflist),
+				 DRM_MEM_BUFS);
+			drm_free(dma->bufs[i].seglist,
+				 dma->buf_count
+				 * sizeof(*dma->bufs[0].seglist),
+				 DRM_MEM_SEGS);
+			drm_freelist_destroy(&dma->bufs[i].freelist);
+		}
+	}
+	
+	if (dma->buflist) {
+		drm_free(dma->buflist,
+			 dma->buf_count * sizeof(*dma->buflist),
+			 DRM_MEM_BUFS);
+	}
+
+	if (dma->pagelist) {
+		drm_free(dma->pagelist,
+			 dma->page_count * sizeof(*dma->pagelist),
+			 DRM_MEM_PAGES);
+	}
+	drm_free(dev->dma, sizeof(*dev->dma), DRM_MEM_DRIVER);
+	dev->dma = NULL;
+}
+
+#if DRM_DMA_HISTOGRAM
+/* This is slow, but is useful for debugging. */
+int drm_histogram_slot(struct timespec *ts)
+{
+	long count = ts->tv_sec * 1000 + ts->tv_nsec / 1000000;
+	int value = DRM_DMA_HISTOGRAM_INITIAL;
+	int slot;
+
+	for (slot = 0;
+	     slot < DRM_DMA_HISTOGRAM_SLOTS;
+	     ++slot, value = DRM_DMA_HISTOGRAM_NEXT(value)) {
+		if (count < value) return slot;
+	}
+	return DRM_DMA_HISTOGRAM_SLOTS - 1;
+}
+
+void drm_histogram_compute(drm_device_t *dev, drm_buf_t *buf)
+{
+	struct timespec queued_to_dispatched;
+	struct timespec dispatched_to_completed;
+	struct timespec completed_to_freed;
+	int	 q2d, d2c, c2f, q2c, q2f;
+	
+	if (timespecisset(&buf->time_queued)) {
+		queued_to_dispatched = buf->time_dispatched;
+		timespecsub(&queued_to_dispatched, &buf->time_queued);
+		dispatched_to_completed = buf->time_completed;
+		timespecsub(&dispatched_to_completed, &buf->time_dispatched);
+		completed_to_freed = buf->time_freed;
+		timespecsub(&completed_to_freed, &buf->time_completed);
+
+		q2d = drm_histogram_slot(&queued_to_dispatched);
+		d2c = drm_histogram_slot(&dispatched_to_completed);
+		c2f = drm_histogram_slot(&completed_to_freed);
+
+		timespecadd(&queued_to_dispatched, &dispatched_to_completed);
+		q2c = drm_histogram_slot(&queued_to_dispatched);
+		timespecadd(&queued_to_dispatched, &completed_to_freed);
+		q2f = drm_histogram_slot(&queued_to_dispatched);
+		
+		atomic_inc(&dev->histo.total);
+		atomic_inc(&dev->histo.queued_to_dispatched[q2d]);
+		atomic_inc(&dev->histo.dispatched_to_completed[d2c]);
+		atomic_inc(&dev->histo.completed_to_freed[c2f]);
+		
+		atomic_inc(&dev->histo.queued_to_completed[q2c]);
+		atomic_inc(&dev->histo.queued_to_freed[q2f]);
+
+	}
+	timespecclear(&buf->time_queued);
+	timespecclear(&buf->time_dispatched);
+	timespecclear(&buf->time_completed);
+	timespecclear(&buf->time_freed);
+}
+#endif
+
+void drm_free_buffer(drm_device_t *dev, drm_buf_t *buf)
+{
+	drm_device_dma_t *dma = dev->dma;
+
+	if (!buf) return;
+	
+	buf->waiting  = 0;
+	buf->pending  = 0;
+	buf->pid      = 0;
+	buf->used     = 0;
+#if DRM_DMA_HISTOGRAMxx
+	buf->time_completed = get_cycles();
+#endif
+	if (buf->dma_wait) {
+		buf->dma_wait = 0;
+		wakeup(&buf->dma_wait);
+	} else {
+				/* If processes are waiting, the last one
+				   to wake will put the buffer on the free
+				   list.  If no processes are waiting, we
+				   put the buffer on the freelist here. */
+		drm_freelist_put(dev, &dma->bufs[buf->order].freelist, buf);
+	}
+}
+
+void drm_reclaim_buffers(drm_device_t *dev, pid_t pid)
+{
+	drm_device_dma_t *dma = dev->dma;
+	int		 i;
+
+	if (!dma) return;
+	for (i = 0; i < dma->buf_count; i++) {
+		if (dma->buflist[i]->pid == pid) {
+			switch (dma->buflist[i]->list) {
+			case DRM_LIST_NONE:
+				drm_free_buffer(dev, dma->buflist[i]);
+				break;
+			case DRM_LIST_WAIT:
+				dma->buflist[i]->list = DRM_LIST_RECLAIM;
+				break;
+			default:
+				/* Buffer already on hardware. */
+				break;
+			}
+		}
+	}
+}
+
+int drm_context_switch(drm_device_t *dev, int old, int new)
+{
+	char	    buf[64];
+	drm_queue_t *q;
+
+	atomic_inc(&dev->total_ctx);
+
+	if (test_and_set_bit(0, &dev->context_flag)) {
+		DRM_ERROR("Reentering -- FIXME\n");
+		return EBUSY;
+	}
+
+#if DRM_DMA_HISTOGRAM
+	getnanotime(&dev->ctx_start);
+#endif
+	
+	DRM_DEBUG("Context switch from %d to %d\n", old, new);
+
+	if (new >= dev->queue_count) {
+		clear_bit(0, &dev->context_flag);
+		return EINVAL;
+	}
+
+	if (new == dev->last_context) {
+		clear_bit(0, &dev->context_flag);
+		return 0;
+	}
+	
+	q = dev->queuelist[new];
+	atomic_inc(&q->use_count);
+	if (atomic_read(&q->use_count) == 1) {
+		atomic_dec(&q->use_count);
+		clear_bit(0, &dev->context_flag);
+		return EINVAL;
+	}
+
+	if (drm_flags & DRM_FLAG_NOCTX) {
+		drm_context_switch_complete(dev, new);
+	} else {
+		sprintf(buf, "C %d %d\n", old, new);
+		drm_write_string(dev, buf);
+	}
+	
+	atomic_dec(&q->use_count);
+	
+	return 0;
+}
+
+int drm_context_switch_complete(drm_device_t *dev, int new)
+{
+	drm_device_dma_t *dma = dev->dma;
+	
+	dev->last_context = new;  /* PRE/POST: This is the _only_ writer. */
+	dev->last_switch = ticks;
+	
+	if (!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("Lock isn't held after context switch\n");
+	}
+
+	if (!dma || !(dma->next_buffer && dma->next_buffer->while_locked)) {
+		if (drm_lock_free(dev, &dev->lock.hw_lock->lock,
+				  DRM_KERNEL_CONTEXT)) {
+			DRM_ERROR("Cannot free lock\n");
+		}
+	}
+	
+#if DRM_DMA_HISTOGRAM
+	{
+		struct timespec ts;
+		getnanotime(&ts);
+		timespecsub(&ts, &dev->ctx_start);
+		atomic_inc(&dev->histo.ctx[drm_histogram_slot(&ts)]);
+	}
+#endif
+	clear_bit(0, &dev->context_flag);
+	wakeup(&dev->context_wait);
+	
+	return 0;
+}
+
+void drm_clear_next_buffer(drm_device_t *dev)
+{
+	drm_device_dma_t *dma = dev->dma;
+	
+	dma->next_buffer = NULL;
+	if (dma->next_queue && !DRM_BUFCOUNT(&dma->next_queue->waitlist)) {
+		wakeup(&dma->next_queue->flush_queue);
+	}
+	dma->next_queue	 = NULL;
+}
+
+
+int drm_select_queue(drm_device_t *dev, void (*wrapper)(void *))
+{
+	int	   i;
+	int	   candidate = -1;
+	int	   j	     = ticks;
+
+	if (!dev) {
+		DRM_ERROR("No device\n");
+		return -1;
+	}
+	if (!dev->queuelist || !dev->queuelist[DRM_KERNEL_CONTEXT]) {
+				/* This only happens between the time the
+				   interrupt is initialized and the time
+				   the queues are initialized. */
+		return -1;
+	}
+
+				/* Doing "while locked" DMA? */
+	if (DRM_WAITCOUNT(dev, DRM_KERNEL_CONTEXT)) {
+		return DRM_KERNEL_CONTEXT;
+	}
+
+				/* If there are buffers on the last_context
+				   queue, and we have not been executing
+				   this context very long, continue to
+				   execute this context. */
+	if (dev->last_switch <= j
+	    && dev->last_switch + DRM_TIME_SLICE > j
+	    && DRM_WAITCOUNT(dev, dev->last_context)) {
+		return dev->last_context;
+	}
+
+				/* Otherwise, find a candidate */
+	for (i = dev->last_checked + 1; i < dev->queue_count; i++) {
+		if (DRM_WAITCOUNT(dev, i)) {
+			candidate = dev->last_checked = i;
+			break;
+		}
+	}
+
+	if (candidate < 0) {
+		for (i = 0; i < dev->queue_count; i++) {
+			if (DRM_WAITCOUNT(dev, i)) {
+				candidate = dev->last_checked = i;
+				break;
+			}
+		}
+	}
+
+	if (wrapper
+	    && candidate >= 0
+	    && candidate != dev->last_context
+	    && dev->last_switch <= j
+	    && dev->last_switch + DRM_TIME_SLICE > j) {
+		int s = splclock();
+		if (dev->timer.c_time != dev->last_switch + DRM_TIME_SLICE) {
+			callout_reset(&dev->timer,
+				      dev->last_switch + DRM_TIME_SLICE - j,
+				      wrapper,
+				      dev);
+		}
+		splx(s);
+		return -1;
+	}
+
+	return candidate;
+}
+
+
+int drm_dma_enqueue(drm_device_t *dev, drm_dma_t *d)
+{
+	int		  i;
+	drm_queue_t	  *q;
+	drm_buf_t	  *buf;
+	int		  idx;
+	int		  while_locked = 0;
+	drm_device_dma_t  *dma = dev->dma;
+	int		  error;
+
+	DRM_DEBUG("%d\n", d->send_count);
+
+	if (d->flags & _DRM_DMA_WHILE_LOCKED) {
+		int context = dev->lock.hw_lock->lock;
+		
+		if (!_DRM_LOCK_IS_HELD(context)) {
+			DRM_ERROR("No lock held during \"while locked\""
+				  " request\n");
+			return EINVAL;
+		}
+		if (d->context != _DRM_LOCKING_CONTEXT(context)
+		    && _DRM_LOCKING_CONTEXT(context) != DRM_KERNEL_CONTEXT) {
+			DRM_ERROR("Lock held by %d while %d makes"
+				  " \"while locked\" request\n",
+				  _DRM_LOCKING_CONTEXT(context),
+				  d->context);
+			return EINVAL;
+		}
+		q = dev->queuelist[DRM_KERNEL_CONTEXT];
+		while_locked = 1;
+	} else {
+		q = dev->queuelist[d->context];
+	}
+
+
+	atomic_inc(&q->use_count);
+	if (atomic_read(&q->block_write)) {
+		atomic_inc(&q->block_count);
+		for (;;) {
+			if (!atomic_read(&q->block_write)) break;
+			error = tsleep(&q->block_write, PZERO|PCATCH,
+				       "dmawr", 0);
+			if (error) {
+				atomic_dec(&q->use_count);
+				return error;
+			}
+		}
+		atomic_dec(&q->block_count);
+	}
+	
+	for (i = 0; i < d->send_count; i++) {
+		idx = d->send_indices[i];
+		if (idx < 0 || idx >= dma->buf_count) {
+			atomic_dec(&q->use_count);
+			DRM_ERROR("Index %d (of %d max)\n",
+				  d->send_indices[i], dma->buf_count - 1);
+			return EINVAL;
+		}
+		buf = dma->buflist[ idx ];
+		if (buf->pid != curproc->p_pid) {
+			atomic_dec(&q->use_count);
+			DRM_ERROR("Process %d using buffer owned by %d\n",
+				  curproc->p_pid, buf->pid);
+			return EINVAL;
+		}
+		if (buf->list != DRM_LIST_NONE) {
+			atomic_dec(&q->use_count);
+			DRM_ERROR("Process %d using buffer %d on list %d\n",
+				  curproc->p_pid, buf->idx, buf->list);
+		}
+		buf->used	  = d->send_sizes[i];
+		buf->while_locked = while_locked;
+		buf->context	  = d->context;
+		if (!buf->used) {
+			DRM_ERROR("Queueing 0 length buffer\n");
+		}
+		if (buf->pending) {
+			atomic_dec(&q->use_count);
+			DRM_ERROR("Queueing pending buffer:"
+				  " buffer %d, offset %d\n",
+				  d->send_indices[i], i);
+			return EINVAL;
+		}
+		if (buf->waiting) {
+			atomic_dec(&q->use_count);
+			DRM_ERROR("Queueing waiting buffer:"
+				  " buffer %d, offset %d\n",
+				  d->send_indices[i], i);
+			return EINVAL;
+		}
+		buf->waiting = 1;
+		if (atomic_read(&q->use_count) == 1
+		    || atomic_read(&q->finalization)) {
+			drm_free_buffer(dev, buf);
+		} else {
+			drm_waitlist_put(&q->waitlist, buf);
+			atomic_inc(&q->total_queued);
+		}
+	}
+	atomic_dec(&q->use_count);
+	
+	return 0;
+}
+
+static int drm_dma_get_buffers_of_order(drm_device_t *dev, drm_dma_t *d,
+					int order)
+{
+	int		  i;
+	int		  error;
+	drm_buf_t	  *buf;
+	drm_device_dma_t  *dma = dev->dma;
+	
+	for (i = d->granted_count; i < d->request_count; i++) {
+		buf = drm_freelist_get(&dma->bufs[order].freelist,
+				       d->flags & _DRM_DMA_WAIT);
+		if (!buf) break;
+		if (buf->pending || buf->waiting) {
+			DRM_ERROR("Free buffer %d in use by %d (w%d, p%d)\n",
+				  buf->idx,
+				  buf->pid,
+				  buf->waiting,
+				  buf->pending);
+		}
+		buf->pid     = curproc->p_pid;
+		error = copyout(&buf->idx,
+				&d->request_indices[i],
+				sizeof(buf->idx));
+		if (error)
+			return error;
+		error = copyout(&buf->total,
+				&d->request_sizes[i],
+				sizeof(buf->total));
+		if (error)
+			return error;
+		++d->granted_count;
+	}
+	return 0;
+}
+
+
+int drm_dma_get_buffers(drm_device_t *dev, drm_dma_t *dma)
+{
+	int		  order;
+	int		  retcode = 0;
+	int		  tmp_order;
+	
+	order = drm_order(dma->request_size);
+
+	dma->granted_count = 0;
+	retcode		   = drm_dma_get_buffers_of_order(dev, dma, order);
+
+	if (dma->granted_count < dma->request_count
+	    && (dma->flags & _DRM_DMA_SMALLER_OK)) {
+		for (tmp_order = order - 1;
+		     !retcode
+			     && dma->granted_count < dma->request_count
+			     && tmp_order >= DRM_MIN_ORDER;
+		     --tmp_order) {
+			
+			retcode = drm_dma_get_buffers_of_order(dev, dma,
+							       tmp_order);
+		}
+	}
+
+	if (dma->granted_count < dma->request_count
+	    && (dma->flags & _DRM_DMA_LARGER_OK)) {
+		for (tmp_order = order + 1;
+		     !retcode
+			     && dma->granted_count < dma->request_count
+			     && tmp_order <= DRM_MAX_ORDER;
+		     ++tmp_order) {
+			
+			retcode = drm_dma_get_buffers_of_order(dev, dma,
+							       tmp_order);
+		}
+	}
+	return 0;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/drawable.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/drawable.c
new file mode 100644
index 000000000..d8005af6a
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/drawable.c
@@ -0,0 +1,50 @@
+/* drawable.c -- IOCTLs for drawables -*- c -*-
+ * Created: Tue Feb  2 08:37:54 1999 by faith@precisioninsight.com
+ * Revised: Fri Aug 20 09:27:03 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drawable.c,v 1.3 1999/08/30 13:05:00 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drawable.c,v 1.1 1999/09/25 14:37:58 dawes Exp $
+ *
+ */
+
+#define __NO_VERSION__
+#include "drmP.h"
+
+int drm_adddraw(dev_t kdev, u_long cmd, caddr_t data,
+		int flags, struct proc *p)
+{
+	drm_draw_t draw;
+
+	draw.handle = 0;	/* NOOP */
+	DRM_DEBUG("%d\n", draw.handle);
+	*(drm_draw_t *) data = draw;
+	return 0;
+}
+
+int drm_rmdraw(dev_t kdev, u_long cmd, caddr_t data,
+	       int flags, struct proc *p)
+{
+	return 0;		/* NOOP */
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/drmstat.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/drmstat.c
new file mode 100644
index 000000000..0ce76b01a
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/drmstat.c
@@ -0,0 +1,418 @@
+/* drmstat.c -- DRM device status and testing program
+ * Created: Tue Jan  5 08:19:24 1999 by faith@precisioninsight.com
+ * Revised: Sun Aug  1 11:02:00 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drmstat.c,v 1.28 1999/08/04 18:12:11 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drmstat.c,v 1.1 1999/09/25 14:37:59 dawes Exp $
+ * 
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <sys/types.h>
+#include <sys/time.h>
+#include <sys/mman.h>
+#include <strings.h>
+#include <errno.h>
+#include <signal.h>
+#include <fcntl.h>
+#include "xf86drm.h"
+
+int sigio_fd;
+
+static double usec(struct timeval *end, struct timeval *start)
+{
+    double e = end->tv_sec   * 1000000 + end->tv_usec;
+    double s = start->tv_sec * 1000000 + start->tv_usec;
+
+    return e - s;
+}
+
+static void getversion(int fd)
+{
+    drmVersionPtr version;
+    
+    version = drmGetVersion(fd);
+    if (version) {
+	printf( "Name: %s\n", version->name ? version->name : "?" );
+	printf( "    Version: %d.%d.%d\n",
+		version->version_major,
+		version->version_minor,
+		version->version_patchlevel );
+	printf( "    Date: %s\n", version->date ? version->date : "?" );
+	printf( "    Desc: %s\n", version->desc ? version->desc : "?" );
+	drmFreeVersion(version);
+    } else {
+	printf( "No driver available\n" );
+    }
+}
+	
+void handler(int fd, void *oldctx, void *newctx)
+{
+    printf("Got fd %d\n", fd);
+}
+
+void process_sigio(char *device)
+{
+    int              fd;
+
+printf("%s\n", device);
+    if ((fd = open(device, 0)) < 0) {
+printf("%d\n", errno);
+	drmError(-errno, __FUNCTION__);
+	exit(1);
+    }
+
+    sigio_fd = fd;
+    drmInstallSIGIOHandler(fd, handler);
+    for (;;) sleep(60);
+}
+
+int main(int argc, char **argv)
+{
+    int            c;
+    int            r  = 0;
+    int            fd = -1;
+    drmHandle      handle;
+    void           *address;
+    char           *pt;
+    unsigned long  count;
+    unsigned long  offset;
+    unsigned long  size;
+    drmContext     context;
+    int            loops;
+    char           buf[1024];
+    int            i;
+    drmBufInfoPtr  info;
+    drmBufMapPtr   bufs;
+    drmLockPtr     lock;
+    int            secs;
+
+    while ((c = getopt(argc, argv,
+		       "lc:vo:O:f:s:w:W:b:r:R:P:L:C:XS:B:F:")) != EOF)
+	switch (c) {
+	case 'F':
+	    count  = strtoul(optarg, NULL, 0);
+	    if (!fork()) {
+		dup(fd);
+		sleep(count);
+	    }
+	    close(fd);
+	    break;
+	case 'v': getversion(fd);                                        break;
+	case 'X':
+	    if ((r = drmCreateContext(fd, &context))) {
+		drmError(r, argv[0]);
+		return 1;
+	    }
+	    printf( "Got %d\n", context);
+	    break;
+	case 'S':
+	    process_sigio(optarg);
+	    break;
+	case 'C':
+	    if ((r = drmSwitchToContext(fd, strtoul(optarg, NULL, 0)))) {
+		drmError(r, argv[0]);
+		return 1;
+	    }
+	    break;
+	case 'c':
+	    if ((r = drmSetBusid(fd,optarg))) {
+		drmError(r, argv[0]);
+		return 1;
+	    }
+	    break;
+	case 'o':
+	    if ((fd = drmOpen(optarg, NULL)) < 0) {
+		drmError(fd, argv[0]);
+		return 1;
+	    }
+	    break;
+	case 'O':
+	    if ((fd = drmOpen(NULL, optarg)) < 0) {
+		drmError(fd, argv[0]);
+		return 1;
+	    }
+	    break;
+	case 'B':		/* Test buffer allocation */
+	    count  = strtoul(optarg, &pt, 0);
+	    size   = strtoul(pt+1, &pt, 0);
+	    secs   = strtoul(pt+1, NULL, 0);
+	    {
+		drmDMAReq      dma;
+		int            *indices, *sizes;
+
+		indices = alloca(sizeof(*indices) * count);
+		sizes   = alloca(sizeof(*sizes)   * count);
+		dma.context         = context;
+		dma.send_count      = 0;
+		dma.request_count   = count;
+		dma.request_size    = size;
+		dma.request_list    = indices;
+		dma.request_sizes   = sizes;
+		dma.flags           = DRM_DMA_WAIT;
+		if ((r = drmDMA(fd, &dma))) {
+		    drmError(r, argv[0]);
+		    return 1;
+		}
+		for (i = 0; i < dma.granted_count; i++) {
+		    printf("%5d: index = %d, size = %d\n",
+			   i, dma.request_list[i], dma.request_sizes[i]);
+		}
+		sleep(secs);
+		drmFreeBufs(fd, dma.granted_count, indices);
+	    }
+	    break;
+	case 'b':
+	    count   = strtoul(optarg, &pt, 0);
+	    size    = strtoul(pt+1, NULL, 0);
+	    if ((r = drmAddBufs(fd, count, size, 0, 0)) < 0) {
+		drmError(r, argv[0]);
+		return 1;
+	    }
+	    if (!(info = drmGetBufInfo(fd))) {
+		drmError(0, argv[0]);
+		return 1;
+	    }
+	    for (i = 0; i < info->count; i++) {
+		printf("%5d buffers of size %6d (low = %d, high = %d)\n",
+		       info->list[i].count,
+		       info->list[i].size,
+		       info->list[i].low_mark,
+		       info->list[i].high_mark);
+	    }
+	    if ((r = drmMarkBufs(fd, 0.50, 0.80))) {
+		drmError(r, argv[0]);
+		return 1;
+	    }
+	    if (!(info = drmGetBufInfo(fd))) {
+		drmError(0, argv[0]);
+		return 1;
+	    }
+	    for (i = 0; i < info->count; i++) {
+		printf("%5d buffers of size %6d (low = %d, high = %d)\n",
+		       info->list[i].count,
+		       info->list[i].size,
+		       info->list[i].low_mark,
+		       info->list[i].high_mark);
+	    }
+	    printf("===== /proc/drm/1/meminfo =====\n");
+	    sprintf(buf, "cat /proc/drm/1/meminfo");
+	    system(buf);
+#if 1
+	    if (!(bufs = drmMapBufs(fd))) {
+		drmError(0, argv[0]);
+		return 1;
+	    }
+	    printf("===============================\n");
+	    printf( "%d bufs\n", bufs->count);
+	    for (i = 0; i < bufs->count; i++) {
+		printf( "  %4d: %8d bytes at %p\n",
+			i,
+			bufs->list[i].total,
+			bufs->list[i].address);
+	    }
+	    printf("===== /proc/drm/1/vmainfo =====\n");
+	    sprintf(buf, "cat /proc/drm/1/vmainfo");
+	    system(buf);
+#endif
+	    break;
+	case 'f':
+	    offset  = strtoul(optarg, &pt, 0);
+	    size    = strtoul(pt+1, NULL, 0);
+	    handle  = 0;
+	    if ((r = drmAddMap(fd, offset, size,
+			       DRM_FRAME_BUFFER, 0, &handle))) {
+		drmError(r, argv[0]);
+		return 1;
+	    }
+	    printf("0x%08lx:0x%04lx added\n", offset, size);
+	    printf("===== /proc/drm/1/meminfo =====\n");
+	    sprintf(buf, "cat /proc/drm/1/meminfo");
+	    system(buf);
+	    break;
+	case 'r':
+	case 'R':
+	    offset  = strtoul(optarg, &pt, 0);
+	    size    = strtoul(pt+1, NULL, 0);
+	    handle  = 0;
+	    if ((r = drmAddMap(fd, offset, size,
+			       DRM_REGISTERS,
+			       c == 'R' ? DRM_READ_ONLY : 0,
+			       &handle))) {
+		drmError(r, argv[0]);
+		return 1;
+	    }
+	    printf("0x%08lx:0x%04lx added\n", offset, size);
+	    printf("===== /proc/drm/1/meminfo =====\n");
+	    sprintf(buf, "cat /proc/drm/1/meminfo");
+	    system(buf);
+	    break;
+	case 's':
+	    size = strtoul(optarg, &pt, 0);
+	    handle = 0;
+	    if ((r = drmAddMap(fd, 0, size,
+			       DRM_SHM, DRM_CONTAINS_LOCK,
+			       &handle))) {
+		drmError(r, argv[0]);
+		return 1;
+	    }
+	    printf("0x%04lx byte shm added at 0x%08lx\n", size, handle);
+	    sprintf(buf, "sysctl hw.graphics.0.vm");
+	    system(buf);
+	    break;
+	case 'P':
+	    offset  = strtoul(optarg, &pt, 0);
+	    size    = strtoul(pt+1, NULL, 0);
+	    address = NULL;
+	    if ((r = drmMap(fd, offset, size, &address))) {
+		drmError(r, argv[0]);
+		return 1;
+	    }
+	    printf("0x%08lx:0x%04lx mapped at %p for pid %d\n",
+		   offset, size, address, getpid());
+	    printf("===== hw.graphics.0.vma =====\n");
+	    sprintf(buf, "sysctl hw.graphics.0.vma");
+	    system(buf);
+	    mprotect((void *)offset, size, PROT_READ);
+	    printf("===== hw.graphics.0.vma =====\n");
+	    sprintf(buf, "sysctl hw.graphics.0.vma");
+	    system(buf);
+	    break;
+	case 'w':
+	case 'W':
+	    offset  = strtoul(optarg, &pt, 0);
+	    size    = strtoul(pt+1, NULL, 0);
+	    address = NULL;
+	    if ((r = drmMap(fd, offset, size, &address))) {
+		drmError(r, argv[0]);
+		return 1;
+	    }
+	    printf("0x%08lx:0x%04lx mapped at %p for pid %d\n",
+		   offset, size, address, getpid());
+	    printf("===== /proc/%d/maps =====\n", getpid());
+	    sprintf(buf, "cat /proc/%d/maps", getpid());
+	    system(buf);
+	    printf("===== /proc/drm/1/meminfo =====\n");
+	    sprintf(buf, "cat /proc/drm/1/meminfo");
+	    system(buf);
+	    printf("===== /proc/drm/1/vmainfo =====\n");
+	    sprintf(buf, "cat /proc/drm/1/vmainfo");
+	    system(buf);
+	    printf("===== READING =====\n");
+	    for (i = 0; i < 0x10; i++)
+		printf("%02x ", (unsigned int)((unsigned char *)address)[i]);
+	    printf("\n");
+	    if (c == 'w') {
+		printf("===== WRITING =====\n");
+		for (i = 0; i < size; i+=2) {
+		    ((char *)address)[i]   = i & 0xff;
+		    ((char *)address)[i+1] = i & 0xff;
+		}
+	    }
+	    printf("===== READING =====\n");
+	    for (i = 0; i < 0x10; i++)
+		printf("%02x ", (unsigned int)((unsigned char *)address)[i]);
+	    printf("\n");
+	    printf("===== /proc/drm/1/vmainfo =====\n");
+	    sprintf(buf, "cat /proc/drm/1/vmainfo");
+	    system(buf);
+	    break;
+	case 'L':
+	    context = strtoul(optarg, &pt, 0);
+	    offset  = strtoul(pt+1, &pt, 0);
+	    size    = strtoul(pt+1, &pt, 0);
+	    loops   = strtoul(pt+1, NULL, 0);
+	    address = NULL;
+	    if ((r = drmMap(fd, offset, size, &address))) {
+		drmError(r, argv[0]);
+		return 1;
+	    }
+	    lock       = address;
+#if 1
+	    {
+		int            counter = 0;
+		struct timeval loop_start, loop_end;
+		struct timeval lock_start, lock_end;
+		double         wt;
+#define HISTOSIZE 9
+		int            histo[HISTOSIZE];
+		int            output = 0;
+		int            fast   = 0;
+
+		if (loops < 0) {
+		    loops = -loops;
+		    ++output;
+		}
+
+		for (i = 0; i < HISTOSIZE; i++) histo[i] = 0;
+
+		gettimeofday(&loop_start, NULL);
+		for (i = 0; i < loops; i++) {
+		    gettimeofday(&lock_start, NULL);
+		    DRM_LIGHT_LOCK_COUNT(fd,lock,context,fast);
+		    gettimeofday(&lock_end, NULL);
+		    DRM_UNLOCK(fd,lock,context);
+		    ++counter;
+		    wt = usec(&lock_end, &lock_start);
+		    if      (wt <=      2.5) ++histo[8];
+		    if      (wt <       5.0) ++histo[0];
+		    else if (wt <      50.0) ++histo[1];
+		    else if (wt <     500.0) ++histo[2];
+		    else if (wt <    5000.0) ++histo[3];
+		    else if (wt <   50000.0) ++histo[4];
+		    else if (wt <  500000.0) ++histo[5];
+		    else if (wt < 5000000.0) ++histo[6];
+		    else                     ++histo[7];
+		    if (output) printf( "%.2f uSec, %d fast\n", wt, fast);
+		}
+		gettimeofday(&loop_end, NULL);
+		printf( "Average wait time = %.2f usec, %d fast\n",
+			usec(&loop_end, &loop_start) /  counter, fast);
+		printf( "%9d <=     2.5 uS\n", histo[8]);
+		printf( "%9d <        5 uS\n", histo[0]);
+		printf( "%9d <       50 uS\n", histo[1]);
+		printf( "%9d <      500 uS\n", histo[2]);
+		printf( "%9d <     5000 uS\n", histo[3]);
+		printf( "%9d <    50000 uS\n", histo[4]);
+		printf( "%9d <   500000 uS\n", histo[5]);
+		printf( "%9d <  5000000 uS\n", histo[6]);
+		printf( "%9d >= 5000000 uS\n", histo[7]);
+	    }
+#else
+	    printf( "before lock: 0x%08x\n", lock->lock);
+	    printf( "lock: 0x%08x\n", lock->lock);
+	    sleep(5);
+	    printf( "unlock: 0x%08x\n", lock->lock);
+#endif
+	    break;
+	default:
+	    fprintf( stderr, "Usage: drmstat [options]\n" );
+	    return 1;
+	}
+
+    return r; 
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/fops.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/fops.c
new file mode 100644
index 000000000..837fc7db0
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/fops.c
@@ -0,0 +1,260 @@
+/* fops.c -- File operations for DRM -*- c -*-
+ * Created: Mon Jan  4 08:58:31 1999 by faith@precisioninsight.com
+ * Revised: Tue Oct 12 08:48:59 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/fops.c,v 1.3 1999/08/20 15:36:45 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/fops.c,v 1.1 1999/09/25 14:37:59 dawes Exp $
+ *
+ */
+
+#define __NO_VERSION__
+#include "drmP.h"
+#include <sys/signalvar.h>
+#include <sys/poll.h>
+
+drm_file_t *drm_find_file_by_proc(drm_device_t *dev, struct proc *p)
+{
+	uid_t uid = p->p_cred->p_svuid;
+	pid_t pid = p->p_pid;
+	drm_file_t *priv;
+
+	TAILQ_FOREACH(priv, &dev->files, link)
+		if (priv->pid == pid && priv->uid == uid)
+			return priv;
+	return NULL;
+}
+
+
+/* drm_open is called whenever a process opens /dev/drm. */
+
+int drm_open_helper(dev_t kdev, int flags, int fmt, struct proc *p,
+		    drm_device_t *dev)
+{
+	int	     m = minor(kdev);
+	drm_file_t   *priv;
+
+	if (flags & O_EXCL)
+		return EBUSY; /* No exclusive opens */
+
+	dev->flags = flags;
+
+	DRM_DEBUG("pid = %d, device = %p, minor = %d\n",
+		  p->p_pid, dev->device, m);
+
+	priv = drm_find_file_by_proc(dev, p);
+	if (priv) {
+		priv->refs++;
+	} else {
+		priv		    = drm_alloc(sizeof(*priv), DRM_MEM_FILES);
+		memset(priv, 0, sizeof(*priv));
+		priv->uid	    = p->p_cred->p_svuid;
+		priv->pid	    = p->p_pid;
+		priv->refs          = 1;
+		priv->minor	    = m;
+		priv->devXX	    = dev;
+		priv->ioctl_count   = 0;
+		priv->authenticated = !suser(p);
+		lockmgr(&dev->dev_lock, LK_EXCLUSIVE, 0, p);
+		TAILQ_INSERT_TAIL(&dev->files, priv, link);
+		lockmgr(&dev->dev_lock, LK_RELEASE, 0, p);
+	}
+
+	kdev->si_drv1 = dev;
+	
+	return 0;
+}
+
+int drm_write(dev_t kdev, struct uio *uio, int ioflag)
+{
+	struct proc   *p      = curproc;
+	drm_device_t  *dev    = kdev->si_drv1;
+
+	DRM_DEBUG("pid = %d, device = %p, open_count = %d\n",
+		  p->p_pid, dev->device, dev->open_count);
+	return 0;
+}
+
+/* drm_release is called whenever a process closes /dev/drm*. */
+
+int drm_close(dev_t kdev, int fflag, int devtype, struct proc *p)
+{
+	drm_device_t  *dev  = kdev->si_drv1;
+	drm_file_t    *priv;
+
+	DRM_DEBUG("pid = %d, device = %p, open_count = %d\n",
+		  p->p_pid, dev->device, dev->open_count);
+
+	priv = drm_find_file_by_proc(dev, p);
+	if (!priv) {
+		DRM_DEBUG("can't find authenticator\n");
+		return EINVAL;
+	}
+
+	if (dev->lock.hw_lock && _DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)
+	    && dev->lock.pid == p->p_pid) {
+		DRM_ERROR("Process %d dead, freeing lock for context %d\n",
+			  p->p_pid,
+			  _DRM_LOCKING_CONTEXT(dev->lock.hw_lock->lock));
+		drm_lock_free(dev,
+			      &dev->lock.hw_lock->lock,
+			      _DRM_LOCKING_CONTEXT(dev->lock.hw_lock->lock));
+		
+				/* FIXME: may require heavy-handed reset of
+                                   hardware at this point, possibly
+                                   processed via a callback to the X
+                                   server. */
+	}
+	drm_reclaim_buffers(dev, priv->pid);
+
+	funsetown(dev->buf_sigio);
+
+	lockmgr(&dev->dev_lock, LK_EXCLUSIVE, 0, p);
+	priv = drm_find_file_by_proc(dev, p);
+	if (priv) {
+		priv->refs--;
+		if (!priv->refs) {
+			TAILQ_REMOVE(&dev->files, priv, link);
+			drm_free(priv, sizeof(*priv), DRM_MEM_FILES);
+		}
+	}
+	lockmgr(&dev->dev_lock, LK_RELEASE, 0, p);
+	
+	return 0;
+}
+
+/* The drm_read and drm_write_string code (especially that which manages
+   the circular buffer), is based on Alessandro Rubini's LINUX DEVICE
+   DRIVERS (Cambridge: O'Reilly, 1998), pages 111-113. */
+
+ssize_t drm_read(dev_t kdev, struct uio *uio, int ioflag)
+{
+	drm_device_t  *dev    = kdev->si_drv1;
+	int	      left;
+	int	      avail;
+	int	      send;
+	int	      cur;
+	int	      error = 0;
+
+	DRM_DEBUG("%p, %p\n", dev->buf_rp, dev->buf_wp);
+	
+	while (dev->buf_rp == dev->buf_wp) {
+		DRM_DEBUG("  sleeping\n");
+		if (dev->flags & FASYNC) {
+			return EWOULDBLOCK;
+		}
+		error = tsleep(&dev->buf_rp, PZERO|PCATCH, "drmrd", 0);
+		if (error) {
+			DRM_DEBUG("  interrupted\n");
+			return error;
+		}
+		DRM_DEBUG("  awake\n");
+	}
+
+	left  = (dev->buf_rp + DRM_BSZ - dev->buf_wp) % DRM_BSZ;
+	avail = DRM_BSZ - left;
+	send  = DRM_MIN(avail, uio->uio_resid);
+
+	while (send) {
+		if (dev->buf_wp > dev->buf_rp) {
+			cur = DRM_MIN(send, dev->buf_wp - dev->buf_rp);
+		} else {
+			cur = DRM_MIN(send, dev->buf_end - dev->buf_rp);
+		}
+		error = uiomove(dev->buf_rp, cur, uio);
+		if (error)
+			break;
+		dev->buf_rp += cur;
+		if (dev->buf_rp == dev->buf_end) dev->buf_rp = dev->buf;
+		send -= cur;
+	}
+	
+	wakeup(&dev->buf_wp);
+
+	return error;
+}
+
+int drm_write_string(drm_device_t *dev, const char *s)
+{
+	int left   = (dev->buf_rp + DRM_BSZ - dev->buf_wp) % DRM_BSZ;
+	int send   = strlen(s);
+	int count;
+
+	DRM_DEBUG("%d left, %d to send (%p, %p)\n",
+		  left, send, dev->buf_rp, dev->buf_wp);
+	
+	if (left == 1 || dev->buf_wp != dev->buf_rp) {
+		DRM_ERROR("Buffer not empty (%d left, wp = %p, rp = %p)\n",
+			  left,
+			  dev->buf_wp,
+			  dev->buf_rp);
+	}
+
+	while (send) {
+		if (dev->buf_wp >= dev->buf_rp) {
+			count = DRM_MIN(send, dev->buf_end - dev->buf_wp);
+			if (count == left) --count; /* Leave a hole */
+		} else {
+			count = DRM_MIN(send, dev->buf_rp - dev->buf_wp - 1);
+		}
+		strncpy(dev->buf_wp, s, count);
+		dev->buf_wp += count;
+		if (dev->buf_wp == dev->buf_end) dev->buf_wp = dev->buf;
+		send -= count;
+	}
+
+	if (dev->buf_selecting) {
+		dev->buf_selecting = 0;
+		selwakeup(&dev->buf_sel);
+	}
+		
+	DRM_DEBUG("dev->buf_sigio=%p\n", dev->buf_sigio);
+	if (dev->buf_sigio) {
+		DRM_DEBUG("dev->buf_sigio->sio_pgid=%d\n", dev->buf_sigio->sio_pgid);
+		pgsigio(dev->buf_sigio, SIGIO, 0);
+	}
+
+	DRM_DEBUG("waking\n");
+	wakeup(&dev->buf_rp);
+	return 0;
+}
+
+int drm_poll(dev_t kdev, int events, struct proc *p)
+{
+	drm_device_t  *dev    = kdev->si_drv1;
+	int           s;
+	int	      revents = 0;
+
+	s = spldrm();
+	if (events & (POLLIN | POLLRDNORM)) {
+		int left  = (dev->buf_rp + DRM_BSZ - dev->buf_wp) % DRM_BSZ;
+		if (left > 0)
+			revents |= events & (POLLIN | POLLRDNORM);
+		else
+			selrecord(p, &dev->buf_sel);
+	}
+	splx(s);
+
+	return revents;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/init.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/init.c
new file mode 100644
index 000000000..44e9be995
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/init.c
@@ -0,0 +1,101 @@
+/* init.c -- Setup/Cleanup for DRM -*- c -*-
+ * Created: Mon Jan  4 08:58:31 1999 by faith@precisioninsight.com
+ * Revised: Fri Aug 20 09:27:02 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/init.c,v 1.3 1999/08/20 15:07:01 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/init.c,v 1.1 1999/09/25 14:38:01 dawes Exp $
+ *
+ */
+
+#define __NO_VERSION__
+#include "drmP.h"
+
+MODULE_VERSION(drm, 1);
+
+int			      drm_flags		= 0;
+
+/* drm_parse_option parses a single option.  See description for
+   drm_parse_drm for details. */
+
+static void drm_parse_option(char *s)
+{
+	char *c, *r;
+	
+	DRM_DEBUG("\"%s\"\n", s);
+	if (!s || !*s) return;
+	for (c = s; *c && *c != ':'; c++); /* find : or \0 */
+	if (*c) r = c + 1; else r = NULL;  /* remember remainder */
+	*c = '\0';			   /* terminate */
+	if (!strcmp(s, "noctx")) {
+		drm_flags |= DRM_FLAG_NOCTX;
+		DRM_INFO("Server-mediated context switching OFF\n");
+		return;
+	}
+	if (!strcmp(s, "debug")) {
+		drm_flags |= DRM_FLAG_DEBUG;
+		DRM_INFO("Debug messages ON\n");
+		return;
+	}
+	DRM_ERROR("\"%s\" is not a valid option\n", s);
+	return;
+}
+
+/* drm_parse_options parse the insmod "drm=" options, or the command-line
+ * options passed to the kernel via LILO.  The grammar of the format is as
+ * follows:
+ *
+ * drm		::= 'drm=' option_list
+ * option_list	::= option [ ';' option_list ]
+ * option	::= 'device:' major
+ *		|   'debug' 
+ *		|   'noctx'
+ * major	::= INTEGER
+ *
+ * Note that 's' contains option_list without the 'drm=' part.
+ *
+ * device=major,minor specifies the device number used for /dev/drm
+ *	  if major == 0 then the misc device is used
+ *	  if major == 0 and minor == 0 then dynamic misc allocation is used
+ * debug=on specifies that debugging messages will be printk'd
+ * debug=trace specifies that each function call will be logged via printk
+ * debug=off turns off all debugging options
+ *
+ */
+
+void drm_parse_options(char *s)
+{
+	char *h, *t, *n;
+	
+	DRM_DEBUG("\"%s\"\n", s ?: "");
+	if (!s || !*s) return;
+
+	for (h = t = n = s; h && *h; h = n) {
+		for (; *t && *t != ';'; t++);	       /* find ; or \0 */
+		if (*t) n = t + 1; else n = NULL;      /* remember next */
+		*t = '\0';			       /* terminate */
+		drm_parse_option(h);		       /* parse */
+	}
+}
+
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/ioctl.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/ioctl.c
new file mode 100644
index 000000000..55bdeedac
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/ioctl.c
@@ -0,0 +1,120 @@
+/* ioctl.c -- IOCTL processing for DRM -*- c -*-
+ * Created: Fri Jan  8 09:01:26 1999 by faith@precisioninsight.com
+ * Revised: Fri Aug 20 09:27:02 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/ioctl.c,v 1.3 1999/08/30 13:05:00 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/ioctl.c,v 1.1 1999/09/25 14:38:01 dawes Exp $
+ *
+ */
+
+#define __NO_VERSION__
+#include "drmP.h"
+#include <sys/bus.h>
+#include <pci/pcivar.h>
+
+int
+drm_irq_busid(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	drm_irq_busid_t id;
+	devclass_t pci;
+	device_t bus, dev;
+	device_t *kids;
+	int error, i, num_kids;
+
+	id = *(drm_irq_busid_t *) data;
+	pci = devclass_find("pci");
+	if (!pci)
+		return ENOENT;
+	bus = devclass_get_device(pci, id.busnum);
+	if (!bus)
+		return ENOENT;
+	error = device_get_children(bus, &kids, &num_kids);
+	if (error)
+		return error;
+
+	dev = 0;
+	for (i = 0; i < num_kids; i++) {
+		dev = kids[i];
+		if (pci_get_slot(dev) == id.devnum
+		    && pci_get_function(dev) == id.funcnum)
+			break;
+	}
+
+	free(kids, M_TEMP);
+
+	if (i != num_kids)
+		id.irq = pci_get_irq(dev);
+	else
+		id.irq = 0;
+
+	DRM_DEBUG("%d:%d:%d => IRQ %d\n",
+		  id.busnum, id.devnum, id.funcnum, id.irq);
+	*(drm_irq_busid_t *) data = id;
+
+	return 0;
+}
+
+int
+drm_getunique(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	drm_device_t	 *dev	 = kdev->si_drv1;
+	drm_unique_t	 u;
+	int		 error;
+
+	u = *(drm_unique_t *) data;
+	if (u.unique_len >= dev->unique_len) {
+		error = copyout(dev->unique, u.unique, dev->unique_len);
+		if (error)
+			return error;
+	}
+	u.unique_len = dev->unique_len;
+	*(drm_unique_t *) data = u;
+	return 0;
+}
+
+int
+drm_setunique(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	drm_device_t	 *dev	 = kdev->si_drv1;
+	drm_unique_t	 u;
+	int		 error;
+
+	if (dev->unique_len || dev->unique) return EBUSY;
+
+	u = *(drm_unique_t *) data;
+	
+	dev->unique_len = u.unique_len;
+	dev->unique	= drm_alloc(u.unique_len + 1, DRM_MEM_DRIVER);
+	error = copyin(u.unique, dev->unique, dev->unique_len);
+	if (error)
+		return error;
+	dev->unique[dev->unique_len] = '\0';
+
+	dev->devname = drm_alloc(strlen(dev->name) + strlen(dev->unique) + 2,
+				 DRM_MEM_DRIVER);
+	sprintf(dev->devname, "%s@%s", dev->name, dev->unique);
+
+	return 0;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/lists.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/lists.c
new file mode 100644
index 000000000..9f9b5f7ad
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/lists.c
@@ -0,0 +1,258 @@
+/* lists.c -- Buffer list handling routines -*- c -*-
+ * Created: Mon Apr 19 20:54:22 1999 by faith@precisioninsight.com
+ * Revised: Fri Aug 20 09:27:01 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/lists.c,v 1.3 1999/08/20 15:07:02 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/lists.c,v 1.1 1999/09/25 14:38:01 dawes Exp $
+ *
+ */
+
+#define __NO_VERSION__
+#include "drmP.h"
+
+int drm_waitlist_create(drm_waitlist_t *bl, int count)
+{
+	DRM_DEBUG("%d\n", count);
+	if (bl->count) return EINVAL;
+	
+	bl->count      = count;
+	bl->bufs       = drm_alloc((bl->count + 2) * sizeof(*bl->bufs),
+				   DRM_MEM_BUFLISTS);
+	bl->rp	       = bl->bufs;
+	bl->wp	       = bl->bufs;
+	bl->end	       = &bl->bufs[bl->count+1];
+	simple_lock_init(&bl->write_lock);
+	simple_lock_init(&bl->read_lock);
+	return 0;
+}
+
+int drm_waitlist_destroy(drm_waitlist_t *bl)
+{
+	DRM_DEBUG("\n");
+	if (bl->rp != bl->wp) return EINVAL;
+	if (bl->bufs) drm_free(bl->bufs,
+			       (bl->count + 2) * sizeof(*bl->bufs),
+			       DRM_MEM_BUFLISTS);
+	bl->count = 0;
+	bl->bufs  = NULL;
+	bl->rp	  = NULL;
+	bl->wp	  = NULL;
+	bl->end	  = NULL;
+	return 0;
+}
+
+int drm_waitlist_put(drm_waitlist_t *bl, drm_buf_t *buf)
+{							
+	int	      left;
+	int	      s;
+
+	left = DRM_LEFTCOUNT(bl);
+	DRM_DEBUG("put %d (%d left, rp = %p, wp = %p)\n",
+		  buf->idx, left, bl->rp, bl->wp);
+	if (!left) {
+		DRM_ERROR("Overflow while adding buffer %d from pid %d\n",
+			  buf->idx, buf->pid);
+		return EINVAL;
+	}
+#if DRM_DMA_HISTOGRAM
+	getnanotime(&buf->time_queued);
+#endif
+	buf->list	 = DRM_LIST_WAIT;
+	
+	simple_lock(&bl->write_lock);
+	s = spldrm();
+	*bl->wp = buf;
+	if (++bl->wp >= bl->end) bl->wp = bl->bufs;
+	splx(s);
+	simple_unlock(&bl->write_lock);
+	
+	return 0;
+}
+
+drm_buf_t *drm_waitlist_get(drm_waitlist_t *bl)
+{
+	drm_buf_t     *buf;
+	int	       s;
+
+	simple_lock(&bl->read_lock);
+	s = spldrm();
+	buf = *bl->rp;
+	if (bl->rp == bl->wp) {
+		splx(s);
+		simple_unlock(&bl->read_lock);
+		return NULL;
+	}				     
+	if (++bl->rp >= bl->end) bl->rp = bl->bufs;
+	splx(s);
+	simple_unlock(&bl->read_lock);
+	
+	DRM_DEBUG("get %d\n", buf->idx);
+	return buf;
+}
+
+int drm_freelist_create(drm_freelist_t *bl, int count)
+{
+	DRM_DEBUG("\n");
+	atomic_set(&bl->count, 0);
+	bl->next      = NULL;
+	bl->waiting   = 0;
+	bl->low_mark  = 0;
+	bl->high_mark = 0;
+	atomic_set(&bl->wfh,   0);
+	++bl->initialized;
+	return 0;
+}
+
+int drm_freelist_destroy(drm_freelist_t *bl)
+{
+	DRM_DEBUG("\n");
+	atomic_set(&bl->count, 0);
+	bl->next = NULL;
+	return 0;
+}
+
+int drm_freelist_put(drm_device_t *dev, drm_freelist_t *bl, drm_buf_t *buf)
+{
+	unsigned int	 old;
+	unsigned int	 new;
+	char		 failed;
+	int		 count = 0;
+	drm_device_dma_t *dma = dev->dma;
+
+	if (!dma) {
+		DRM_ERROR("No DMA support\n");
+		return 1;
+	}
+
+	if (buf->waiting || buf->pending || buf->list == DRM_LIST_FREE) {
+		DRM_ERROR("Freed buffer %d: w%d, p%d, l%d\n",
+			  buf->idx, buf->waiting, buf->pending, buf->list);
+	}
+	DRM_DEBUG("%d, count = %d, wfh = %d, w%d, p%d\n",
+		  buf->idx, atomic_read(&bl->count), atomic_read(&bl->wfh),
+		  buf->waiting, buf->pending);
+	if (!bl) return 1;
+#if DRM_DMA_HISTOGRAM
+	getnanotime(&buf->time_freed);
+	drm_histogram_compute(dev, buf);
+#endif
+	buf->list	= DRM_LIST_FREE;
+	do {
+		old	  = (unsigned long)bl->next;
+		buf->next = (void *)old;
+		new	  = (unsigned long)buf;
+		_DRM_CAS(&bl->next, old, new, failed);
+		if (++count > DRM_LOOPING_LIMIT) {
+			DRM_ERROR("Looping\n");
+			return 1;
+		}
+	} while (failed);
+	atomic_inc(&bl->count);
+	if (atomic_read(&bl->count) > dma->buf_count) {
+		DRM_ERROR("%d of %d buffers free after addition of %d\n",
+			  atomic_read(&bl->count), dma->buf_count, buf->idx);
+		return 1;
+	}
+				/* Check for high water mark */
+	if (atomic_read(&bl->wfh) && atomic_read(&bl->count)>=bl->high_mark) {
+		atomic_set(&bl->wfh, 0);
+		if (bl->waiting)
+			wakeup(&bl->waiting);
+	}
+	return 0;
+}
+
+static drm_buf_t *drm_freelist_try(drm_freelist_t *bl)
+{
+	unsigned int	  old;
+	unsigned int	  new;
+	char		  failed;
+	drm_buf_t	  *buf;
+	int		  count = 0;
+
+	if (!bl) return NULL;
+	
+				/* Get buffer */
+	do {
+		old = (unsigned int)bl->next;
+		if (!old) {
+			return NULL;
+		}
+		new = (unsigned long)bl->next->next;
+		_DRM_CAS(&bl->next, old, new, failed);
+		if (++count > DRM_LOOPING_LIMIT) {
+			DRM_ERROR("Looping\n");
+			return NULL;
+		}
+	} while (failed);
+	atomic_dec(&bl->count);
+	
+	buf	  = (drm_buf_t *)old;
+	buf->next = NULL;
+	buf->list = DRM_LIST_NONE;
+	DRM_DEBUG("%d, count = %d, wfh = %d, w%d, p%d\n",
+		  buf->idx, atomic_read(&bl->count), atomic_read(&bl->wfh),
+		  buf->waiting, buf->pending);
+	if (buf->waiting || buf->pending) {
+		DRM_ERROR("Free buffer %d: w%d, p%d, l%d\n",
+			  buf->idx, buf->waiting, buf->pending, buf->list);
+	}
+	
+	return buf;
+}
+
+drm_buf_t *drm_freelist_get(drm_freelist_t *bl, int block)
+{
+	drm_buf_t	  *buf	= NULL;
+	int		   error;
+
+	if (!bl || !bl->initialized) return NULL;
+	
+				/* Check for low water mark */
+	if (atomic_read(&bl->count) <= bl->low_mark) /* Became low */
+		atomic_set(&bl->wfh, 1);
+	if (atomic_read(&bl->wfh)) {
+		DRM_DEBUG("Block = %d, count = %d, wfh = %d\n",
+			  block, atomic_read(&bl->count),
+			  atomic_read(&bl->wfh));
+		if (block) {
+			atomic_inc(&bl->waiting);
+			for (;;) {
+				if (!atomic_read(&bl->wfh)
+				    && (buf = drm_freelist_try(bl))) break;
+				error = tsleep(&bl->waiting, PZERO|PCATCH,
+					       "drmfg", 0);
+				if (error)
+					break;
+			}
+			atomic_dec(&bl->waiting);
+		}
+		return buf;
+	}
+		
+	DRM_DEBUG("Count = %d, wfh = %d\n",
+		  atomic_read(&bl->count), atomic_read(&bl->wfh));
+	return drm_freelist_try(bl);
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/lock.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/lock.c
new file mode 100644
index 000000000..cd14a8825
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/lock.c
@@ -0,0 +1,220 @@
+/* lock.c -- IOCTLs for locking -*- c -*-
+ * Created: Tue Feb  2 08:37:54 1999 by faith@precisioninsight.com
+ * Revised: Tue Oct 12 08:51:06 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/lock.c,v 1.5 1999/08/30 13:05:00 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/lock.c,v 1.1 1999/09/25 14:38:01 dawes Exp $
+ *
+ */
+
+#define __NO_VERSION__
+#include "drmP.h"
+
+int
+drm_block(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	DRM_DEBUG("\n");
+	return 0;
+}
+
+int
+drm_unblock(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	DRM_DEBUG("\n");
+	return 0;
+}
+
+int drm_lock_take(__volatile__ unsigned int *lock, unsigned int context)
+{
+	unsigned int old;
+	unsigned int new;
+	char	     failed;
+
+	DRM_DEBUG("%d attempts\n", context);
+	do {
+		old = *lock;
+		if (old & _DRM_LOCK_HELD) new = old | _DRM_LOCK_CONT;
+		else			  new = context | _DRM_LOCK_HELD;
+		_DRM_CAS(lock, old, new, failed);
+	} while (failed);
+	if (_DRM_LOCKING_CONTEXT(old) == context) {
+		if (old & _DRM_LOCK_HELD) {
+			if (context != DRM_KERNEL_CONTEXT) {
+				DRM_ERROR("%d holds heavyweight lock\n",
+					  context);
+			}
+			return 0;
+		}
+	}
+	if (new == (context | _DRM_LOCK_HELD)) {
+				/* Have lock */
+		DRM_DEBUG("%d\n", context);
+		return 1;
+	}
+	DRM_DEBUG("%d unable to get lock held by %d\n",
+		  context, _DRM_LOCKING_CONTEXT(old));
+	return 0;
+}
+
+/* This takes a lock forcibly and hands it to context.	Should ONLY be used
+   inside *_unlock to give lock to kernel before calling *_dma_schedule. */
+int drm_lock_transfer(drm_device_t *dev,
+		      __volatile__ unsigned int *lock, unsigned int context)
+{
+	unsigned int old;
+	unsigned int new;
+	char	     failed;
+
+	dev->lock.pid = 0;
+	do {
+		old = *lock;
+		new = context | _DRM_LOCK_HELD;
+		_DRM_CAS(lock, old, new, failed);
+	} while (failed);
+	DRM_DEBUG("%d => %d\n", _DRM_LOCKING_CONTEXT(old), context);
+	return 1;
+}
+
+int drm_lock_free(drm_device_t *dev,
+		  __volatile__ unsigned int *lock, unsigned int context)
+{
+	unsigned int old;
+	unsigned int new;
+	char	     failed;
+	pid_t        pid = dev->lock.pid;
+
+	DRM_DEBUG("%d\n", context);
+	dev->lock.pid = 0;
+	do {
+		old = *lock;
+		new = 0;
+		_DRM_CAS(lock, old, new, failed);
+	} while (failed);
+	if (_DRM_LOCK_IS_HELD(old) && _DRM_LOCKING_CONTEXT(old) != context) {
+		DRM_ERROR("%d freed heavyweight lock held by %d (pid %d)\n",
+			  context,
+			  _DRM_LOCKING_CONTEXT(old),
+			  pid);
+		return 1;
+	}
+	wakeup(&dev->lock.lock_queue);
+	return 0;
+}
+
+static int drm_flush_queue(drm_device_t *dev, int context)
+{
+	int		  ret	= 0;
+	int		  error;
+	drm_queue_t	  *q	= dev->queuelist[context];
+	
+	DRM_DEBUG("\n");
+	
+	atomic_inc(&q->use_count);
+	if (atomic_read(&q->use_count) > 1) {
+		atomic_inc(&q->block_write);
+		atomic_inc(&q->block_count);
+		error = tsleep(&q->flush_queue, PCATCH|PZERO, "drmfq", 0);
+		if (error)
+			return error;
+		atomic_dec(&q->block_count);
+	}
+	atomic_dec(&q->use_count);
+	atomic_inc(&q->total_flushed);
+		
+				/* NOTE: block_write is still incremented!
+				   Use drm_flush_unlock_queue to decrement. */
+	return ret;
+}
+
+static int drm_flush_unblock_queue(drm_device_t *dev, int context)
+{
+	drm_queue_t	  *q	= dev->queuelist[context];
+	
+	DRM_DEBUG("\n");
+	
+	atomic_inc(&q->use_count);
+	if (atomic_read(&q->use_count) > 1) {
+		if (atomic_read(&q->block_write)) {
+			atomic_dec(&q->block_write);
+			wakeup(&q->write_queue);
+		}
+	}
+	atomic_dec(&q->use_count);
+	return 0;
+}
+
+int drm_flush_block_and_flush(drm_device_t *dev, int context,
+			      drm_lock_flags_t flags)
+{
+	int ret = 0;
+	int i;
+	
+	DRM_DEBUG("\n");
+	
+	if (flags & _DRM_LOCK_FLUSH) {
+		ret = drm_flush_queue(dev, DRM_KERNEL_CONTEXT);
+		if (!ret) ret = drm_flush_queue(dev, context);
+	}
+	if (flags & _DRM_LOCK_FLUSH_ALL) {
+		for (i = 0; !ret && i < dev->queue_count; i++) {
+			ret = drm_flush_queue(dev, i);
+		}
+	}
+	return ret;
+}
+
+int drm_flush_unblock(drm_device_t *dev, int context, drm_lock_flags_t flags)
+{
+	int ret = 0;
+	int i;
+	
+	DRM_DEBUG("\n");
+	
+	if (flags & _DRM_LOCK_FLUSH) {
+		ret = drm_flush_unblock_queue(dev, DRM_KERNEL_CONTEXT);
+		if (!ret) ret = drm_flush_unblock_queue(dev, context);
+	}
+	if (flags & _DRM_LOCK_FLUSH_ALL) {
+		for (i = 0; !ret && i < dev->queue_count; i++) {
+			ret = drm_flush_unblock_queue(dev, i);
+		}
+	}
+		
+	return ret;
+}
+
+int drm_finish(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	drm_device_t	  *dev	  = kdev->si_drv1;
+	int		  ret	  = 0;
+	drm_lock_t	  lock;
+
+	DRM_DEBUG("\n");
+
+	lock = *(drm_lock_t *) data;
+	ret = drm_flush_block_and_flush(dev, lock.context, lock.flags);
+	drm_flush_unblock(dev, lock.context, lock.flags);
+	return ret;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/memory.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/memory.c
new file mode 100644
index 000000000..a8a936df0
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/memory.c
@@ -0,0 +1,458 @@
+/* memory.c -- Memory management wrappers for DRM -*- c -*-
+ * Created: Thu Feb  4 14:00:34 1999 by faith@precisioninsight.com
+ * Revised: Fri Aug 20 13:04:33 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/memory.c,v 1.4 1999/08/20 20:00:53 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/memory.c,v 1.1 1999/09/25 14:38:02 dawes Exp $
+ *
+ */
+
+#define __NO_VERSION__
+#include "drmP.h"
+
+#include <vm/vm.h>
+#include <vm/pmap.h>
+#ifdef DRM_AGP
+#include <sys/agpio.h>
+#endif
+
+MALLOC_DEFINE(M_DRM, "drm", "DRM Data Structures");
+
+typedef struct drm_mem_stats {
+	const char	  *name;
+	int		  succeed_count;
+	int		  free_count;
+	int		  fail_count;
+	unsigned long	  bytes_allocated;
+	unsigned long	  bytes_freed;
+} drm_mem_stats_t;
+
+#ifdef SMP
+static struct simplelock  drm_mem_lock;
+#endif
+static unsigned long	  drm_ram_available = 0;
+static unsigned long	  drm_ram_used	    = 0;
+static drm_mem_stats_t	  drm_mem_stats[]   = {
+	[DRM_MEM_DMA]	   = { "dmabufs"  },
+	[DRM_MEM_SAREA]	   = { "sareas"	  },
+	[DRM_MEM_DRIVER]   = { "driver"	  },
+	[DRM_MEM_MAGIC]	   = { "magic"	  },
+	[DRM_MEM_IOCTLS]   = { "ioctltab" },
+	[DRM_MEM_MAPS]	   = { "maplist"  },
+	[DRM_MEM_VMAS]	   = { "vmalist"  },
+	[DRM_MEM_BUFS]	   = { "buflist"  },
+	[DRM_MEM_SEGS]	   = { "seglist"  },
+	[DRM_MEM_PAGES]	   = { "pagelist" },
+	[DRM_MEM_FILES]	   = { "files"	  },
+	[DRM_MEM_QUEUES]   = { "queues"	  },
+	[DRM_MEM_CMDS]	   = { "commands" },
+	[DRM_MEM_MAPPINGS] = { "mappings" },
+	[DRM_MEM_BUFLISTS] = { "buflists" },
+	[DRM_MEM_AGPLISTS]  = { "agplist"  },
+	[DRM_MEM_TOTALAGP]  = { "totalagp" },
+	[DRM_MEM_BOUNDAGP]  = { "boundagp" },
+	[DRM_MEM_CTXBITMAP] = { "ctxbitmap"},
+	{ NULL, 0, }		/* Last entry must be null */
+};
+
+void drm_mem_init(void)
+{
+	drm_mem_stats_t *mem;
+	
+	for (mem = drm_mem_stats; mem->name; ++mem) {
+		mem->succeed_count   = 0;
+		mem->free_count	     = 0;
+		mem->fail_count	     = 0;
+		mem->bytes_allocated = 0;
+		mem->bytes_freed     = 0;
+	}
+	
+	drm_ram_available = 0; /* si.totalram; */
+	drm_ram_used	  = 0;
+}
+
+/* drm_mem_info is called whenever a process reads /dev/drm/mem. */
+
+static int _drm_mem_info SYSCTL_HANDLER_ARGS
+{
+	drm_mem_stats_t *pt;
+	char buf[128];
+	int error;
+
+	DRM_SYSCTL_PRINT("		  total counts			"
+		       " |    outstanding  \n");
+	DRM_SYSCTL_PRINT("type	   alloc freed fail	bytes	   freed"
+		       " | allocs      bytes\n\n");
+	DRM_SYSCTL_PRINT("%-9.9s %5d %5d %4d %10lu	    |\n",
+		       "system", 0, 0, 0, drm_ram_available);
+	DRM_SYSCTL_PRINT("%-9.9s %5d %5d %4d %10lu	    |\n",
+		       "locked", 0, 0, 0, drm_ram_used);
+	DRM_SYSCTL_PRINT("\n");
+	for (pt = drm_mem_stats; pt->name; pt++) {
+		DRM_SYSCTL_PRINT("%-9.9s %5d %5d %4d %10lu %10lu | %6d %10ld\n",
+			       pt->name,
+			       pt->succeed_count,
+			       pt->free_count,
+			       pt->fail_count,
+			       pt->bytes_allocated,
+			       pt->bytes_freed,
+			       pt->succeed_count - pt->free_count,
+			       (long)pt->bytes_allocated
+			       - (long)pt->bytes_freed);
+	}
+	SYSCTL_OUT(req, "", 1);
+	
+	return 0;
+}
+
+int drm_mem_info SYSCTL_HANDLER_ARGS
+{
+	int ret;
+	
+	simple_lock(&drm_mem_lock);
+	ret = _drm_mem_info(oidp, arg1, arg2, req);
+	simple_unlock(&drm_mem_lock);
+	return ret;
+}
+
+void *drm_alloc(size_t size, int area)
+{
+	void *pt;
+	
+	if (!size) {
+		DRM_MEM_ERROR(area, "Allocating 0 bytes\n");
+		return NULL;
+	}
+	
+	if (!(pt = malloc(size, M_DRM, M_NOWAIT))) {
+		simple_lock(&drm_mem_lock);
+		++drm_mem_stats[area].fail_count;
+		simple_unlock(&drm_mem_lock);
+		return NULL;
+	}
+	simple_lock(&drm_mem_lock);
+	++drm_mem_stats[area].succeed_count;
+	drm_mem_stats[area].bytes_allocated += size;
+	simple_unlock(&drm_mem_lock);
+	return pt;
+}
+
+void *drm_realloc(void *oldpt, size_t oldsize, size_t size, int area)
+{
+	void *pt;
+	
+	if (!(pt = drm_alloc(size, area))) return NULL;
+	if (oldpt && oldsize) {
+		memcpy(pt, oldpt, oldsize);
+		drm_free(oldpt, oldsize, area);
+	}
+	return pt;
+}
+
+char *drm_strdup(const char *s, int area)
+{
+	char *pt;
+	int	 length = s ? strlen(s) : 0;
+	
+	if (!(pt = drm_alloc(length+1, area))) return NULL;
+	strcpy(pt, s);
+	return pt;
+}
+
+void drm_strfree(char *s, int area)
+{
+	unsigned int size;
+	
+	if (!s) return;
+	
+	size = 1 + (s ? strlen(s) : 0);
+	drm_free((void *)s, size, area);
+}
+
+void drm_free(void *pt, size_t size, int area)
+{
+	int alloc_count;
+	int free_count;
+	
+	if (!pt) DRM_MEM_ERROR(area, "Attempt to free NULL pointer\n");
+	else	 free(pt, M_DRM);
+	simple_lock(&drm_mem_lock);
+	drm_mem_stats[area].bytes_freed += size;
+	free_count  = ++drm_mem_stats[area].free_count;
+	alloc_count =	drm_mem_stats[area].succeed_count;
+	simple_unlock(&drm_mem_lock);
+	if (free_count > alloc_count) {
+		DRM_MEM_ERROR(area, "Excess frees: %d frees, %d allocs\n",
+			      free_count, alloc_count);
+	}
+}
+
+unsigned long drm_alloc_pages(int order, int area)
+{
+	vm_offset_t address;
+	unsigned long bytes	  = PAGE_SIZE << order;
+	unsigned long addr;
+	unsigned int  sz;
+	
+	simple_lock(&drm_mem_lock);
+	if (drm_ram_used > +(DRM_RAM_PERCENT * drm_ram_available) / 100) {
+		simple_unlock(&drm_mem_lock);
+		return 0;
+	}
+	simple_unlock(&drm_mem_lock);
+	
+	address = (vm_offset_t) contigmalloc(1<<order, M_DRM, M_WAITOK, 0, ~0, 1, 0);
+	if (!address) {
+		simple_lock(&drm_mem_lock);
+		++drm_mem_stats[area].fail_count;
+		simple_unlock(&drm_mem_lock);
+		return 0;
+	}
+	simple_lock(&drm_mem_lock);
+	++drm_mem_stats[area].succeed_count;
+	drm_mem_stats[area].bytes_allocated += bytes;
+	drm_ram_used		            += bytes;
+	simple_unlock(&drm_mem_lock);
+	
+	
+				/* Zero outside the lock */
+	memset((void *)address, 0, bytes);
+	
+				/* Reserve */
+	for (addr = address, sz = bytes;
+	     sz > 0;
+	     addr += PAGE_SIZE, sz -= PAGE_SIZE) {
+		/* mem_map_reserve(MAP_NR(addr));*/
+	}
+	
+	return address;
+}
+
+void drm_free_pages(unsigned long address, int order, int area)
+{
+	unsigned long bytes = PAGE_SIZE << order;
+	int		  alloc_count;
+	int		  free_count;
+	unsigned long addr;
+	unsigned int  sz;
+	
+	if (!address) {
+		DRM_MEM_ERROR(area, "Attempt to free address 0\n");
+	} else {
+				/* Unreserve */
+		for (addr = address, sz = bytes;
+		     sz > 0;
+		     addr += PAGE_SIZE, sz -= PAGE_SIZE) {
+			/* mem_map_unreserve(MAP_NR(addr));*/
+		}
+		contigfree((void *) address, bytes, M_DRM);
+	}
+	
+	simple_lock(&drm_mem_lock);
+	free_count  = ++drm_mem_stats[area].free_count;
+	alloc_count =	drm_mem_stats[area].succeed_count;
+	drm_mem_stats[area].bytes_freed += bytes;
+	drm_ram_used			-= bytes;
+	simple_unlock(&drm_mem_lock);
+	if (free_count > alloc_count) {
+		DRM_MEM_ERROR(area,
+			      "Excess frees: %d frees, %d allocs\n",
+			      free_count, alloc_count);
+	}
+}
+
+void *drm_ioremap(unsigned long offset, unsigned long size)
+{
+	void *pt;
+	
+	if (!size) {
+		DRM_MEM_ERROR(DRM_MEM_MAPPINGS,
+			      "Mapping 0 bytes at 0x%08lx\n", offset);
+		return NULL;
+	}
+	
+	if (!(pt = pmap_mapdev(offset, size))) {
+		simple_lock(&drm_mem_lock);
+		++drm_mem_stats[DRM_MEM_MAPPINGS].fail_count;
+		simple_unlock(&drm_mem_lock);
+		return NULL;
+	}
+	simple_lock(&drm_mem_lock);
+	++drm_mem_stats[DRM_MEM_MAPPINGS].succeed_count;
+	drm_mem_stats[DRM_MEM_MAPPINGS].bytes_allocated += size;
+	simple_unlock(&drm_mem_lock);
+	return pt;
+}
+
+void drm_ioremapfree(void *pt, unsigned long size)
+{
+	int alloc_count;
+	int free_count;
+	
+	if (!pt)
+		DRM_MEM_ERROR(DRM_MEM_MAPPINGS,
+			      "Attempt to free NULL pointer\n");
+	else
+		pmap_unmapdev((vm_offset_t) pt, size);
+	
+	simple_lock(&drm_mem_lock);
+	drm_mem_stats[DRM_MEM_MAPPINGS].bytes_freed += size;
+	free_count  = ++drm_mem_stats[DRM_MEM_MAPPINGS].free_count;
+	alloc_count =	drm_mem_stats[DRM_MEM_MAPPINGS].succeed_count;
+	simple_unlock(&drm_mem_lock);
+	if (free_count > alloc_count) {
+		DRM_MEM_ERROR(DRM_MEM_MAPPINGS,
+			      "Excess frees: %d frees, %d allocs\n",
+			      free_count, alloc_count);
+	}
+}
+
+#ifdef DRM_AGP
+void *drm_alloc_agp(int pages, u_int32_t type)
+{
+	device_t dev = agp_find_device();
+	void *handle;
+
+	if (!dev)
+		return NULL;
+
+	if (!pages) {
+		DRM_MEM_ERROR(DRM_MEM_TOTALAGP, "Allocating 0 pages\n");
+		return NULL;
+	}
+	
+	if ((handle = agp_alloc_memory(dev, type, pages << AGP_PAGE_SHIFT))) {
+		simple_lock(&drm_mem_lock);
+		++drm_mem_stats[DRM_MEM_TOTALAGP].succeed_count;
+		drm_mem_stats[DRM_MEM_TOTALAGP].bytes_allocated
+			+= pages << PAGE_SHIFT;
+		simple_unlock(&drm_mem_lock);
+		return handle;
+	}
+	simple_lock(&drm_mem_lock);
+	++drm_mem_stats[DRM_MEM_TOTALAGP].fail_count;
+	simple_unlock(&drm_mem_lock);
+	return NULL;
+}
+
+int drm_free_agp(void *handle, int pages)
+{
+	device_t dev = agp_find_device();
+	int           alloc_count;
+	int           free_count;
+	int           retval = EINVAL;
+
+	if (!dev)
+		return EINVAL;
+
+	if (!handle) {
+		DRM_MEM_ERROR(DRM_MEM_TOTALAGP,
+			      "Attempt to free NULL AGP handle\n");
+		return retval;
+	}
+	
+	agp_free_memory(dev, handle);
+	simple_lock(&drm_mem_lock);
+	free_count  = ++drm_mem_stats[DRM_MEM_TOTALAGP].free_count;
+	alloc_count =   drm_mem_stats[DRM_MEM_TOTALAGP].succeed_count;
+	drm_mem_stats[DRM_MEM_TOTALAGP].bytes_freed
+		+= pages << PAGE_SHIFT;
+	simple_unlock(&drm_mem_lock);
+	if (free_count > alloc_count) {
+		DRM_MEM_ERROR(DRM_MEM_TOTALAGP,
+			      "Excess frees: %d frees, %d allocs\n",
+			      free_count, alloc_count);
+	}
+	return 0;
+}
+
+int drm_bind_agp(void *handle, unsigned int start)
+{
+	device_t dev = agp_find_device();
+	int retcode  = EINVAL;
+	struct agp_memory_info info;
+
+	DRM_DEBUG("drm_bind_agp called\n");
+
+	if (!dev)
+		return EINVAL;
+
+	if (!handle) {
+		DRM_MEM_ERROR(DRM_MEM_BOUNDAGP,
+			      "Attempt to bind NULL AGP handle\n");
+		return retcode;
+	}
+
+	if (!(retcode = agp_bind_memory(dev, handle,
+					start << AGP_PAGE_SHIFT))) {
+		simple_lock(&drm_mem_lock);
+		++drm_mem_stats[DRM_MEM_BOUNDAGP].succeed_count;
+		agp_memory_info(dev, handle, &info);
+		drm_mem_stats[DRM_MEM_BOUNDAGP].bytes_allocated
+			+= info.ami_size;
+		simple_unlock(&drm_mem_lock);
+		DRM_DEBUG("drm_agp.bind_memory: retcode %d\n", retcode);
+		return retcode;
+	}
+	simple_lock(&drm_mem_lock);
+	++drm_mem_stats[DRM_MEM_BOUNDAGP].fail_count;
+	simple_unlock(&drm_mem_lock);
+	return retcode;
+}
+
+int drm_unbind_agp(void *handle)
+{
+	device_t dev = agp_find_device();
+	int alloc_count;
+	int free_count;
+	int retcode = EINVAL;
+	struct agp_memory_info info;
+	
+	if (!dev)
+		return EINVAL;
+
+	if (!handle) {
+		DRM_MEM_ERROR(DRM_MEM_BOUNDAGP,
+			      "Attempt to unbind NULL AGP handle\n");
+		return retcode;
+	}
+
+	
+	agp_memory_info(dev, handle, &info);
+	if ((retcode = agp_unbind_memory(dev, handle)))
+		return retcode;
+	simple_lock(&drm_mem_lock);
+	free_count  = ++drm_mem_stats[DRM_MEM_BOUNDAGP].free_count;
+	alloc_count = drm_mem_stats[DRM_MEM_BOUNDAGP].succeed_count;
+	drm_mem_stats[DRM_MEM_BOUNDAGP].bytes_freed += info.ami_size;
+	simple_unlock(&drm_mem_lock);
+	if (free_count > alloc_count) {
+		DRM_MEM_ERROR(DRM_MEM_BOUNDAGP,
+			      "Excess frees: %d frees, %d allocs\n",
+			      free_count, alloc_count);
+	}
+	return retcode;
+}
+#endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/proc.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/proc.c
new file mode 100644
index 000000000..12168aa3d
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/proc.c
@@ -0,0 +1,568 @@
+/* proc.c -- /proc support for DRM -*- c -*-
+ * Created: Mon Jan 11 09:48:47 1999 by faith@precisioninsight.com
+ * Revised: Fri Aug 20 11:31:48 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/proc.c,v 1.4 1999/08/20 15:36:46 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/proc.c,v 1.1 1999/09/25 14:38:02 dawes Exp $
+ *
+ */
+
+#define __NO_VERSION__
+#include "drmP.h"
+
+static struct proc_dir_entry *drm_root	   = NULL;
+static struct proc_dir_entry *drm_dev_root = NULL;
+static char		     drm_slot_name[64];
+
+static int	   drm_name_info(char *buf, char **start, off_t offset,
+				 int len, int *eof, void *data);
+static int	   drm_vm_info(char *buf, char **start, off_t offset,
+			       int len, int *eof, void *data);
+static int	   drm_clients_info(char *buf, char **start, off_t offset,
+				    int len, int *eof, void *data);
+static int	   drm_queues_info(char *buf, char **start, off_t offset,
+				   int len, int *eof, void *data);
+static int	   drm_bufs_info(char *buf, char **start, off_t offset,
+				 int len, int *eof, void *data);
+#if DRM_DEBUG_CODE
+static int	   drm_vma_info(char *buf, char **start, off_t offset,
+				int len, int *eof, void *data);
+#endif
+#if DRM_DMA_HISTOGRAM
+static int	   drm_histo_info(char *buf, char **start, off_t offset,
+				  int len, int *eof, void *data);
+#endif
+
+struct drm_proc_list {
+	const char *name;
+	int	   (*f)(char *, char **, off_t, int, int *, void *);
+} drm_proc_list[] = {
+	{ "name",    drm_name_info    },
+	{ "mem",     drm_mem_info     },
+	{ "vm",	     drm_vm_info      },
+	{ "clients", drm_clients_info },
+	{ "queues",  drm_queues_info  },
+	{ "bufs",    drm_bufs_info    },
+#if DRM_DEBUG_CODE
+	{ "vma",     drm_vma_info     },
+#endif
+#if DRM_DMA_HISTOGRAM
+	{ "histo",   drm_histo_info   },
+#endif
+};
+#define DRM_PROC_ENTRIES (sizeof(drm_proc_list)/sizeof(drm_proc_list[0]))
+
+int drm_proc_init(drm_device_t *dev)
+{
+	struct proc_dir_entry *ent;
+	int		      i, j;
+
+	drm_root = create_proc_entry("graphics", S_IFDIR, NULL);
+	if (!drm_root) {
+		DRM_ERROR("Cannot create /proc/graphics\n");
+		return -1;
+	}
+
+				/* Instead of doing this search, we should
+				   add some global support for /proc/graphics. */
+	for (i = 0; i < 8; i++) {
+		sprintf(drm_slot_name, "graphics/%d", i);
+		drm_dev_root = create_proc_entry(drm_slot_name, S_IFDIR, NULL);
+		if (!drm_dev_root) {
+			DRM_ERROR("Cannot create /proc/%s\n", drm_slot_name);
+			remove_proc_entry("graphics", NULL);
+		}
+		if (drm_dev_root->nlink == 2) break;
+		drm_dev_root = NULL;
+	}
+	if (!drm_dev_root) {
+		DRM_ERROR("Cannot find slot in /proc/graphics\n");
+		return -1;
+	}
+
+	for (i = 0; i < DRM_PROC_ENTRIES; i++) {
+		ent = create_proc_entry(drm_proc_list[i].name,
+					S_IFREG|S_IRUGO, drm_dev_root);
+		if (!ent) {
+			DRM_ERROR("Cannot create /proc/%s/%s\n",
+				  drm_slot_name, drm_proc_list[i].name);
+			for (j = 0; j < i; j++)
+				remove_proc_entry(drm_proc_list[i].name,
+						  drm_dev_root);
+			remove_proc_entry(drm_slot_name, NULL);
+			remove_proc_entry("graphics", NULL);
+			return -1;
+		}
+		ent->read_proc = drm_proc_list[i].f;
+		ent->data      = dev;
+	}
+
+	return 0;
+}
+
+
+int drm_proc_cleanup(void)
+{
+	int i;
+	
+	if (drm_root) {
+		if (drm_dev_root) {
+			for (i = 0; i < DRM_PROC_ENTRIES; i++) {
+				remove_proc_entry(drm_proc_list[i].name,
+						  drm_dev_root);
+			}
+			remove_proc_entry(drm_slot_name, NULL);
+		}
+		remove_proc_entry("graphics", NULL);
+		remove_proc_entry(DRM_NAME, NULL);
+	}
+	drm_root = drm_dev_root = NULL;
+	return 0;
+}
+
+static int drm_name_info(char *buf, char **start, off_t offset, int len,
+			 int *eof, void *data)
+{
+	drm_device_t *dev = (drm_device_t *)data;
+
+	if (offset > 0) return 0; /* no partial requests */
+	len  = 0;
+	*eof = 1;
+
+	if (dev->unique) {
+		DRM_PROC_PRINT("%s 0x%x %s\n",
+			       dev->name, dev->device, dev->unique);
+	} else {
+		DRM_PROC_PRINT("%s 0x%x\n", dev->name, dev->device);
+	}
+	return len;
+}
+
+static int _drm_vm_info(char *buf, char **start, off_t offset, int len,
+			int *eof, void *data)
+{
+	drm_device_t *dev = (drm_device_t *)data;
+	drm_map_t    *map;
+	const char   *types[] = { "FB", "REG", "SHM" };
+	const char   *type;
+	int	     i;
+
+	if (offset > 0) return 0; /* no partial requests */
+	len  = 0;
+	*eof = 1;
+	DRM_PROC_PRINT("slot	 offset	      size type flags	 "
+		       "address mtrr\n\n");
+	for (i = 0; i < dev->map_count; i++) {
+		map = dev->maplist[i];
+		if (map->type < 0 || map->type > 2) type = "??";
+		else				    type = types[map->type];
+		DRM_PROC_PRINT("%4d 0x%08lx 0x%08lx %4.4s  0x%02x 0x%08lx ",
+			       i,
+			       map->offset,
+			       map->size,
+			       type,
+			       map->flags,
+			       (unsigned long)map->handle);
+		if (map->mtrr < 0) {
+			DRM_PROC_PRINT("none\n");
+		} else {
+			DRM_PROC_PRINT("%4d\n", map->mtrr);
+		}
+	}
+
+	return len;
+}
+
+static int drm_vm_info(char *buf, char **start, off_t offset, int len,
+		       int *eof, void *data)
+{
+	drm_device_t *dev = (drm_device_t *)data;
+	int	     ret;
+
+	down(&dev->struct_sem);
+	ret = _drm_vm_info(buf, start, offset, len, eof, data);
+	up(&dev->struct_sem);
+	return ret;
+}
+
+
+static int _drm_queues_info(char *buf, char **start, off_t offset, int len,
+			    int *eof, void *data)
+{
+	drm_device_t *dev = (drm_device_t *)data;
+	int	     i;
+	drm_queue_t  *q;
+
+	if (offset > 0) return 0; /* no partial requests */
+	len  = 0;
+	*eof = 1;
+	DRM_PROC_PRINT("  ctx/flags   use   fin"
+		       "   blk/rw/rwf  wait    flushed	   queued"
+		       "      locks\n\n");
+	for (i = 0; i < dev->queue_count; i++) {
+		q = dev->queuelist[i];
+		atomic_inc(&q->use_count);
+		DRM_PROC_PRINT_RET(atomic_dec(&q->use_count),
+				   "%5d/0x%03x %5d %5d"
+				   " %5d/%c%c/%c%c%c %5d %10d %10d %10d\n",
+				   i,
+				   q->flags,
+				   atomic_read(&q->use_count),
+				   atomic_read(&q->finalization),
+				   atomic_read(&q->block_count),
+				   atomic_read(&q->block_read) ? 'r' : '-',
+				   atomic_read(&q->block_write) ? 'w' : '-',
+				   waitqueue_active(&q->read_queue) ? 'r':'-',
+				   waitqueue_active(&q->write_queue) ? 'w':'-',
+				   waitqueue_active(&q->flush_queue) ? 'f':'-',
+				   DRM_BUFCOUNT(&q->waitlist),
+				   atomic_read(&q->total_flushed),
+				   atomic_read(&q->total_queued),
+				   atomic_read(&q->total_locks));
+		atomic_dec(&q->use_count);
+	}
+	
+	return len;
+}
+
+static int drm_queues_info(char *buf, char **start, off_t offset, int len,
+			   int *eof, void *data)
+{
+	drm_device_t *dev = (drm_device_t *)data;
+	int	     ret;
+
+	down(&dev->struct_sem);
+	ret = _drm_queues_info(buf, start, offset, len, eof, data);
+	up(&dev->struct_sem);
+	return ret;
+}
+
+/* drm_bufs_info is called whenever a process reads
+   /dev/drm/<dev>/bufs. */
+
+static int _drm_bufs_info(char *buf, char **start, off_t offset, int len,
+			  int *eof, void *data)
+{
+	drm_device_t	 *dev = (drm_device_t *)data;
+	drm_device_dma_t *dma = dev->dma;
+	int		 i;
+
+	if (!dma)	return 0;
+	if (offset > 0) return 0; /* no partial requests */
+	len  = 0;
+	*eof = 1;
+	DRM_PROC_PRINT(" o     size count  free	 segs pages    kB\n\n");
+	for (i = 0; i <= DRM_MAX_ORDER; i++) {
+		if (dma->bufs[i].buf_count)
+			DRM_PROC_PRINT("%2d %8d %5d %5d %5d %5d %5ld\n",
+				       i,
+				       dma->bufs[i].buf_size,
+				       dma->bufs[i].buf_count,
+				       atomic_read(&dma->bufs[i]
+						   .freelist.count),
+				       dma->bufs[i].seg_count,
+				       dma->bufs[i].seg_count
+				       *(1 << dma->bufs[i].page_order),
+				       (dma->bufs[i].seg_count
+					* (1 << dma->bufs[i].page_order))
+				       * PAGE_SIZE / 1024);
+	}
+	DRM_PROC_PRINT("\n");
+	for (i = 0; i < dma->buf_count; i++) {
+		if (i && !(i%32)) DRM_PROC_PRINT("\n");
+		DRM_PROC_PRINT(" %d", dma->buflist[i]->list);
+	}
+	DRM_PROC_PRINT("\n");
+
+	return len;
+}
+
+static int drm_bufs_info(char *buf, char **start, off_t offset, int len,
+			 int *eof, void *data)
+{
+	drm_device_t *dev = (drm_device_t *)data;
+	int	     ret;
+
+	down(&dev->struct_sem);
+	ret = _drm_bufs_info(buf, start, offset, len, eof, data);
+	up(&dev->struct_sem);
+	return ret;
+}
+
+
+static int _drm_clients_info(char *buf, char **start, off_t offset, int len,
+			     int *eof, void *data)
+{
+	drm_device_t *dev = (drm_device_t *)data;
+	drm_file_t   *priv;
+
+	if (offset > 0) return 0; /* no partial requests */
+	len  = 0;
+	*eof = 1;
+	DRM_PROC_PRINT("a dev	pid    uid	magic	  ioctls\n\n");
+	for (priv = dev->file_first; priv; priv = priv->next) {
+		DRM_PROC_PRINT("%c %3d %5d %5d %10u %10lu\n",
+			       priv->authenticated ? 'y' : 'n',
+			       priv->minor,
+			       priv->pid,
+			       priv->uid,
+			       priv->magic,
+			       priv->ioctl_count);
+	}
+
+	return len;
+}
+
+static int drm_clients_info(char *buf, char **start, off_t offset, int len,
+			    int *eof, void *data)
+{
+	drm_device_t *dev = (drm_device_t *)data;
+	int	     ret;
+
+	down(&dev->struct_sem);
+	ret = _drm_clients_info(buf, start, offset, len, eof, data);
+	up(&dev->struct_sem);
+	return ret;
+}
+
+#if DRM_DEBUG_CODE
+
+static int _drm_vma_info(char *buf, char **start, off_t offset, int len,
+			 int *eof, void *data)
+{
+	drm_device_t	      *dev = (drm_device_t *)data;
+	drm_vma_entry_t	      *pt;
+	pgd_t		      *pgd;
+	pmd_t		      *pmd;
+	pte_t		      *pte;
+	unsigned long	      i;
+	struct vm_area_struct *vma;
+	unsigned long	      address;
+#if defined(__i386__)
+	unsigned int	      pgprot;
+#endif
+
+	if (offset > 0) return 0; /* no partial requests */
+	len  = 0;
+	*eof = 1;
+	DRM_PROC_PRINT("vma use count: %d, high_memory = %p, 0x%08lx\n",
+		       atomic_read(&dev->vma_count),
+		       high_memory, virt_to_phys(high_memory));
+	for (pt = dev->vmalist; pt; pt = pt->next) {
+		if (!(vma = pt->vma)) continue;
+		DRM_PROC_PRINT("\n%5d 0x%08lx-0x%08lx %c%c%c%c%c%c 0x%08lx",
+			       pt->pid,
+			       vma->vm_start,
+			       vma->vm_end,
+			       vma->vm_flags & VM_READ	   ? 'r' : '-',
+			       vma->vm_flags & VM_WRITE	   ? 'w' : '-',
+			       vma->vm_flags & VM_EXEC	   ? 'x' : '-',
+			       vma->vm_flags & VM_MAYSHARE ? 's' : 'p',
+			       vma->vm_flags & VM_LOCKED   ? 'l' : '-',
+			       vma->vm_flags & VM_IO	   ? 'i' : '-',
+			       vma->vm_offset );
+#if defined(__i386__)
+		pgprot = pgprot_val(vma->vm_page_prot);
+		DRM_PROC_PRINT(" %c%c%c%c%c%c%c%c%c",
+			       pgprot & _PAGE_PRESENT  ? 'p' : '-',
+			       pgprot & _PAGE_RW       ? 'w' : 'r',
+			       pgprot & _PAGE_USER     ? 'u' : 's',
+			       pgprot & _PAGE_PWT      ? 't' : 'b',
+			       pgprot & _PAGE_PCD      ? 'u' : 'c',
+			       pgprot & _PAGE_ACCESSED ? 'a' : '-',
+			       pgprot & _PAGE_DIRTY    ? 'd' : '-',
+			       pgprot & _PAGE_4M       ? 'm' : 'k',
+			       pgprot & _PAGE_GLOBAL   ? 'g' : 'l' );
+#endif		
+		DRM_PROC_PRINT("\n");
+		for (i = vma->vm_start; i < vma->vm_end; i += PAGE_SIZE) {
+			pgd = pgd_offset(vma->vm_mm, i);
+			pmd = pmd_offset(pgd, i);
+			pte = pte_offset(pmd, i);
+			if (pte_present(*pte)) {
+				address = __pa(pte_page(*pte))
+					+ (i & (PAGE_SIZE-1));
+				DRM_PROC_PRINT("      0x%08lx -> 0x%08lx"
+					       " %c%c%c%c%c\n",
+					       i,
+					       address,
+					       pte_read(*pte)  ? 'r' : '-',
+					       pte_write(*pte) ? 'w' : '-',
+					       pte_exec(*pte)  ? 'x' : '-',
+					       pte_dirty(*pte) ? 'd' : '-',
+					       pte_young(*pte) ? 'a' : '-' );
+			} else {
+				DRM_PROC_PRINT("      0x%08lx\n", i);
+			}
+		}
+	}
+	
+	return len;
+}
+
+static int drm_vma_info(char *buf, char **start, off_t offset, int len,
+			int *eof, void *data)
+{
+	drm_device_t *dev = (drm_device_t *)data;
+	int	     ret;
+
+	down(&dev->struct_sem);
+	ret = _drm_vma_info(buf, start, offset, len, eof, data);
+	up(&dev->struct_sem);
+	return ret;
+}
+#endif
+
+
+#if DRM_DMA_HISTOGRAM
+static int _drm_histo_info(char *buf, char **start, off_t offset, int len,
+			   int *eof, void *data)
+{
+	drm_device_t	 *dev = (drm_device_t *)data;
+	drm_device_dma_t *dma = dev->dma;
+	int		 i;
+	unsigned long	 slot_value = DRM_DMA_HISTOGRAM_INITIAL;
+	unsigned long	 prev_value = 0;
+	drm_buf_t	 *buffer;
+
+	if (offset > 0) return 0; /* no partial requests */
+	len  = 0;
+	*eof = 1;
+
+	DRM_PROC_PRINT("general statistics:\n");
+	DRM_PROC_PRINT("total	 %10u\n", atomic_read(&dev->histo.total));
+	DRM_PROC_PRINT("open	 %10u\n", atomic_read(&dev->total_open));
+	DRM_PROC_PRINT("close	 %10u\n", atomic_read(&dev->total_close));
+	DRM_PROC_PRINT("ioctl	 %10u\n", atomic_read(&dev->total_ioctl));
+	DRM_PROC_PRINT("irq	 %10u\n", atomic_read(&dev->total_irq));
+	DRM_PROC_PRINT("ctx	 %10u\n", atomic_read(&dev->total_ctx));
+	
+	DRM_PROC_PRINT("\nlock statistics:\n");
+	DRM_PROC_PRINT("locks	 %10u\n", atomic_read(&dev->total_locks));
+	DRM_PROC_PRINT("unlocks	 %10u\n", atomic_read(&dev->total_unlocks));
+	DRM_PROC_PRINT("contends %10u\n", atomic_read(&dev->total_contends));
+	DRM_PROC_PRINT("sleeps	 %10u\n", atomic_read(&dev->total_sleeps));
+
+
+	if (dma) {
+		DRM_PROC_PRINT("\ndma statistics:\n");
+		DRM_PROC_PRINT("prio	 %10u\n",
+			       atomic_read(&dma->total_prio));
+		DRM_PROC_PRINT("bytes	 %10u\n",
+			       atomic_read(&dma->total_bytes));
+		DRM_PROC_PRINT("dmas	 %10u\n",
+			       atomic_read(&dma->total_dmas));
+		DRM_PROC_PRINT("missed:\n");
+		DRM_PROC_PRINT("  dma	 %10u\n",
+			       atomic_read(&dma->total_missed_dma));
+		DRM_PROC_PRINT("  lock	 %10u\n",
+			       atomic_read(&dma->total_missed_lock));
+		DRM_PROC_PRINT("  free	 %10u\n",
+			       atomic_read(&dma->total_missed_free));
+		DRM_PROC_PRINT("  sched	 %10u\n",
+			       atomic_read(&dma->total_missed_sched));
+		DRM_PROC_PRINT("tried	 %10u\n",
+			       atomic_read(&dma->total_tried));
+		DRM_PROC_PRINT("hit	 %10u\n",
+			       atomic_read(&dma->total_hit));
+		DRM_PROC_PRINT("lost	 %10u\n",
+			       atomic_read(&dma->total_lost));
+		
+		buffer = dma->next_buffer;
+		if (buffer) {
+			DRM_PROC_PRINT("next_buffer %7d\n", buffer->idx);
+		} else {
+			DRM_PROC_PRINT("next_buffer    none\n");
+		}
+		buffer = dma->this_buffer;
+		if (buffer) {
+			DRM_PROC_PRINT("this_buffer %7d\n", buffer->idx);
+		} else {
+			DRM_PROC_PRINT("this_buffer    none\n");
+		}
+	}
+	
+
+	DRM_PROC_PRINT("\nvalues:\n");
+	if (dev->lock.hw_lock) {
+		DRM_PROC_PRINT("lock	       0x%08x\n",
+			       dev->lock.hw_lock->lock);
+	} else {
+		DRM_PROC_PRINT("lock		     none\n");
+	}
+	DRM_PROC_PRINT("context_flag   0x%08x\n", dev->context_flag);
+	DRM_PROC_PRINT("interrupt_flag 0x%08x\n", dev->interrupt_flag);
+	DRM_PROC_PRINT("dma_flag       0x%08x\n", dev->dma_flag);
+
+	DRM_PROC_PRINT("queue_count    %10d\n",	 dev->queue_count);
+	DRM_PROC_PRINT("last_context   %10d\n",	 dev->last_context);
+	DRM_PROC_PRINT("last_switch    %10lu\n", dev->last_switch);
+	DRM_PROC_PRINT("last_checked   %10d\n",	 dev->last_checked);
+		
+	
+	DRM_PROC_PRINT("\n		       q2d	  d2c	     c2f"
+		       "	q2c	   q2f	      dma	 sch"
+		       "	ctx	  lacq	     lhld\n\n");
+	for (i = 0; i < DRM_DMA_HISTOGRAM_SLOTS; i++) {
+		DRM_PROC_PRINT("%s %10lu %10u %10u %10u %10u %10u"
+			       " %10u %10u %10u %10u %10u\n",
+			       i == DRM_DMA_HISTOGRAM_SLOTS - 1 ? ">=" : "< ",
+			       i == DRM_DMA_HISTOGRAM_SLOTS - 1
+			       ? prev_value : slot_value ,
+			       
+			       atomic_read(&dev->histo
+					   .queued_to_dispatched[i]),
+			       atomic_read(&dev->histo
+					   .dispatched_to_completed[i]),
+			       atomic_read(&dev->histo
+					   .completed_to_freed[i]),
+			       
+			       atomic_read(&dev->histo
+					   .queued_to_completed[i]),
+			       atomic_read(&dev->histo
+					   .queued_to_freed[i]),
+			       atomic_read(&dev->histo.dma[i]),
+			       atomic_read(&dev->histo.schedule[i]),
+			       atomic_read(&dev->histo.ctx[i]),
+			       atomic_read(&dev->histo.lacq[i]),
+			       atomic_read(&dev->histo.lhld[i]));
+		prev_value = slot_value;
+		slot_value = DRM_DMA_HISTOGRAM_NEXT(slot_value);
+	}
+	return len;
+}
+
+static int drm_histo_info(char *buf, char **start, off_t offset, int len,
+			  int *eof, void *data)
+{
+	drm_device_t *dev = (drm_device_t *)data;
+	int	     ret;
+
+	down(&dev->struct_sem);
+	ret = _drm_histo_info(buf, start, offset, len, eof, data);
+	up(&dev->struct_sem);
+	return ret;
+}
+#endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/sysctl.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/sysctl.c
new file mode 100644
index 000000000..7c736abff
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/sysctl.c
@@ -0,0 +1,554 @@
+/* proc.c -- /proc support for DRM -*- c -*-
+ * Created: Mon Jan 11 09:48:47 1999 by faith@precisioninsight.com
+ * Revised: Fri Aug 20 11:31:48 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI$
+ * $XFree86$
+ *
+ */
+
+#define __NO_VERSION__
+#include "drmP.h"
+#include <sys/sysctl.h>
+
+SYSCTL_NODE(_hw, OID_AUTO, dri, CTLFLAG_RW, 0, "DRI Graphics");
+
+static int	   drm_name_info SYSCTL_HANDLER_ARGS;
+static int	   drm_vm_info SYSCTL_HANDLER_ARGS;
+static int	   drm_clients_info SYSCTL_HANDLER_ARGS;
+static int	   drm_queues_info SYSCTL_HANDLER_ARGS;
+static int	   drm_bufs_info SYSCTL_HANDLER_ARGS;
+#if DRM_DEBUG_CODExx
+static int	   drm_vma_info SYSCTL_HANDLER_ARGS;
+#endif
+#if DRM_DMA_HISTOGRAM
+static int	   drm_histo_info SYSCTL_HANDLER_ARGS;
+#endif
+
+struct drm_sysctl_list {
+	const char *name;
+	int	   (*f) SYSCTL_HANDLER_ARGS;
+} drm_sysctl_list[] = {
+	{ "name",    drm_name_info    },
+	{ "mem",     drm_mem_info     },
+	{ "vm",	     drm_vm_info      },
+	{ "clients", drm_clients_info },
+	{ "queues",  drm_queues_info  },
+	{ "bufs",    drm_bufs_info    },
+#if DRM_DEBUG_CODExx
+	{ "vma",     drm_vma_info     },
+#endif
+#if DRM_DMA_HISTOGRAM
+	{ "histo",   drm_histo_info   },
+#endif
+};
+#define DRM_SYSCTL_ENTRIES (sizeof(drm_sysctl_list)/sizeof(drm_sysctl_list[0]))
+
+struct drm_sysctl_info {
+	struct sysctl_oid      oids[DRM_SYSCTL_ENTRIES + 1];
+	struct sysctl_oid_list list;
+	char		       name[2];
+};
+
+int drm_sysctl_init(drm_device_t *dev)
+{
+	struct drm_sysctl_info *info;
+	struct sysctl_oid *oid;
+	struct sysctl_oid *top;
+	int		  i;
+
+	/* Find the next free slot under hw.graphics */
+	i = 0;
+	SLIST_FOREACH(oid, &sysctl__hw_dri_children, oid_link) {
+		if (i <= oid->oid_arg2)
+			i = oid->oid_arg2 + 1;
+	}
+	
+	info = drm_alloc(sizeof *info, DRM_MEM_DRIVER);
+	dev->sysctl = info;
+
+	/* Construct the node under hw.graphics */
+	info->name[0] = '0' + i;
+	info->name[1] = 0;
+	oid = &info->oids[DRM_SYSCTL_ENTRIES];
+	bzero(oid, sizeof(*oid));
+	oid->oid_parent = &sysctl__hw_dri_children;
+	oid->oid_number = OID_AUTO;
+	oid->oid_kind = CTLTYPE_NODE | CTLFLAG_RW;
+	oid->oid_arg1 = &info->list;
+	oid->oid_arg2 = i;
+	oid->oid_name = info->name;
+	oid->oid_handler = 0;
+	oid->oid_fmt = "N";
+	SLIST_INIT(&info->list);
+	sysctl_register_oid(oid);
+	top = oid;
+
+	for (i = 0; i < DRM_SYSCTL_ENTRIES; i++) {
+		oid = &info->oids[i];
+		bzero(oid, sizeof(*oid));
+		oid->oid_parent = top->oid_arg1;
+		oid->oid_number = OID_AUTO;
+		oid->oid_kind = CTLTYPE_INT | CTLFLAG_RD;
+		oid->oid_arg1 = dev;
+		oid->oid_arg2 = 0;
+		oid->oid_name = drm_sysctl_list[i].name;
+		oid->oid_handler = drm_sysctl_list[i].f;
+		oid->oid_fmt = "A";
+		sysctl_register_oid(oid);
+	}
+
+	return 0;
+}
+
+int drm_sysctl_cleanup(drm_device_t *dev)
+{
+	int i;
+
+	DRM_DEBUG("dev->sysctl=%p\n", dev->sysctl);
+	for (i = 0; i < DRM_SYSCTL_ENTRIES + 1; i++)
+		sysctl_unregister_oid(&dev->sysctl->oids[i]);
+
+	drm_free(dev->sysctl, sizeof *dev->sysctl, DRM_MEM_DRIVER);
+	dev->sysctl = NULL;
+
+	return 0;
+}
+
+static int drm_name_info SYSCTL_HANDLER_ARGS
+{
+	drm_device_t *dev = arg1;
+	char buf[128];
+	int error;
+
+	if (dev->unique) {
+		DRM_SYSCTL_PRINT("%s 0x%x %s\n",
+			       dev->name, dev2udev(dev->devnode), dev->unique);
+	} else {
+		DRM_SYSCTL_PRINT("%s 0x%x\n", dev->name, dev2udev(dev->devnode));
+	}
+
+	SYSCTL_OUT(req, "", 1);
+
+	return 0;
+}
+
+static int _drm_vm_info SYSCTL_HANDLER_ARGS
+{
+	drm_device_t *dev = arg1;
+	drm_map_t    *map;
+	const char   *types[] = { "FB", "REG", "SHM" };
+	const char   *type;
+	int	     i;
+	char         buf[128];
+	int          error;
+
+	DRM_SYSCTL_PRINT("slot	 offset	      size type flags	 "
+			 "address mtrr\n\n");
+	error = SYSCTL_OUT(req, buf, strlen(buf));
+	if (error) return error;
+
+	for (i = 0; i < dev->map_count; i++) {
+		map = dev->maplist[i];
+		if (map->type < 0 || map->type > 2) type = "??";
+		else				    type = types[map->type];
+		DRM_SYSCTL_PRINT("%4d 0x%08lx 0x%08lx %4.4s  0x%02x 0x%08lx ",
+				 i,
+				 map->offset,
+				 map->size,
+				 type,
+				 map->flags,
+				 (unsigned long)map->handle);
+		if (map->mtrr < 0) {
+			DRM_SYSCTL_PRINT("none\n");
+		} else {
+			DRM_SYSCTL_PRINT("%4d\n", map->mtrr);
+		}
+	}
+	SYSCTL_OUT(req, "", 1);
+
+	return 0;
+}
+
+static int drm_vm_info SYSCTL_HANDLER_ARGS
+{
+	drm_device_t *dev = arg1;
+	int	     ret;
+
+	lockmgr(&dev->dev_lock, LK_EXCLUSIVE, 0, curproc);
+	ret = _drm_vm_info(oidp, arg1, arg2, req);
+	lockmgr(&dev->dev_lock, LK_RELEASE, 0, curproc);
+
+	return ret;
+}
+
+
+static int _drm_queues_info SYSCTL_HANDLER_ARGS
+{
+	drm_device_t *dev = arg1;
+	int	     i;
+	drm_queue_t  *q;
+	char         buf[128];
+	int          error;
+
+	DRM_SYSCTL_PRINT("  ctx/flags   use   fin"
+			 "   blk/rw/rwf  wait    flushed	   queued"
+			 "      locks\n\n");
+	for (i = 0; i < dev->queue_count; i++) {
+		q = dev->queuelist[i];
+		atomic_inc(&q->use_count);
+		DRM_SYSCTL_PRINT_RET(atomic_dec(&q->use_count),
+				     "%5d/0x%03x %5d %5d"
+				     " %5d/%c%c/%c%c%c %5d %10d %10d %10d\n",
+				     i,
+				     q->flags,
+				     atomic_read(&q->use_count),
+				     atomic_read(&q->finalization),
+				     atomic_read(&q->block_count),
+				     atomic_read(&q->block_read) ? 'r' : '-',
+				     atomic_read(&q->block_write) ? 'w' : '-',
+				     q->read_queue ? 'r':'-',
+				     q->write_queue ? 'w':'-',
+				     q->flush_queue ? 'f':'-',
+				     DRM_BUFCOUNT(&q->waitlist),
+				     atomic_read(&q->total_flushed),
+				     atomic_read(&q->total_queued),
+				     atomic_read(&q->total_locks));
+		atomic_dec(&q->use_count);
+	}
+
+	SYSCTL_OUT(req, "", 1);
+	return 0;
+}
+
+static int drm_queues_info SYSCTL_HANDLER_ARGS
+{
+	drm_device_t *dev = arg1;
+	int	     ret;
+
+	lockmgr(&dev->dev_lock, LK_EXCLUSIVE, 0, curproc);
+	ret = _drm_queues_info(oidp, arg1, arg2, req);
+	lockmgr(&dev->dev_lock, LK_RELEASE, 0, curproc);
+	return ret;
+}
+
+/* drm_bufs_info is called whenever a process reads
+   hw.dri.0.bufs. */
+
+static int _drm_bufs_info SYSCTL_HANDLER_ARGS
+{
+	drm_device_t	 *dev = arg1;
+	drm_device_dma_t *dma = dev->dma;
+	int		 i;
+	char             buf[128];
+	int              error;
+
+	if (!dma)	return 0;
+	DRM_SYSCTL_PRINT(" o     size count  free	 segs pages    kB\n\n");
+	for (i = 0; i <= DRM_MAX_ORDER; i++) {
+		if (dma->bufs[i].buf_count)
+			DRM_SYSCTL_PRINT("%2d %8d %5d %5d %5d %5d %5d\n",
+				       i,
+				       dma->bufs[i].buf_size,
+				       dma->bufs[i].buf_count,
+				       atomic_read(&dma->bufs[i]
+						   .freelist.count),
+				       dma->bufs[i].seg_count,
+				       dma->bufs[i].seg_count
+				       *(1 << dma->bufs[i].page_order),
+				       (dma->bufs[i].seg_count
+					* (1 << dma->bufs[i].page_order))
+				       * PAGE_SIZE / 1024);
+	}
+	DRM_SYSCTL_PRINT("\n");
+	for (i = 0; i < dma->buf_count; i++) {
+		if (i && !(i%32)) DRM_SYSCTL_PRINT("\n");
+		DRM_SYSCTL_PRINT(" %d", dma->buflist[i]->list);
+	}
+	DRM_SYSCTL_PRINT("\n");
+
+	SYSCTL_OUT(req, "", 1);
+	return 0;
+}
+
+static int drm_bufs_info SYSCTL_HANDLER_ARGS
+{
+	drm_device_t *dev = arg1;
+	int	     ret;
+
+	lockmgr(&dev->dev_lock, LK_EXCLUSIVE, 0, curproc);
+	ret = _drm_bufs_info(oidp, arg1, arg2, req);
+	lockmgr(&dev->dev_lock, LK_RELEASE, 0, curproc);
+	return ret;
+}
+
+
+static int _drm_clients_info SYSCTL_HANDLER_ARGS
+{
+	drm_device_t *dev = arg1;
+	drm_file_t   *priv;
+	char         buf[128];
+	int          error;
+
+	DRM_SYSCTL_PRINT("a dev	pid    uid	magic	  ioctls\n\n");
+	TAILQ_FOREACH(priv, &dev->files, link) {
+		DRM_SYSCTL_PRINT("%c %3d %5d %5d %10u %10lu\n",
+			       priv->authenticated ? 'y' : 'n',
+			       priv->minor,
+			       priv->pid,
+			       priv->uid,
+			       priv->magic,
+			       priv->ioctl_count);
+	}
+
+	SYSCTL_OUT(req, "", 1);
+	return 0;
+}
+
+static int drm_clients_info SYSCTL_HANDLER_ARGS
+{
+	drm_device_t *dev = arg1;
+	int	     ret;
+
+	lockmgr(&dev->dev_lock, LK_EXCLUSIVE, 0, curproc);
+	ret = _drm_clients_info(oidp, arg1, arg2, req);
+	lockmgr(&dev->dev_lock, LK_RELEASE, 0, curproc);
+	return ret;
+}
+
+#if DRM_DEBUG_CODExx
+
+static int _drm_vma_info SYSCTL_HANDLER_ARGS
+{
+	drm_device_t	      *dev = arg1;
+	drm_vma_entry_t	      *pt;
+	pgd_t		      *pgd;
+	pmd_t		      *pmd;
+	pte_t		      *pte;
+	unsigned long	      i;
+	struct vm_area_struct *vma;
+	unsigned long	      address;
+#if defined(__i386__)
+	unsigned int	      pgprot;
+#endif
+	char		      buf[128];
+	int		      error;
+
+	DRM_SYSCTL_PRINT("vma use count: %d, high_memory = %p, 0x%08lx\n",
+		       atomic_read(&dev->vma_count),
+		       high_memory, virt_to_phys(high_memory));
+	for (pt = dev->vmalist; pt; pt = pt->next) {
+		if (!(vma = pt->vma)) continue;
+		DRM_SYSCTL_PRINT("\n%5d 0x%08lx-0x%08lx %c%c%c%c%c%c 0x%08lx",
+			       pt->pid,
+			       vma->vm_start,
+			       vma->vm_end,
+			       vma->vm_flags & VM_READ	   ? 'r' : '-',
+			       vma->vm_flags & VM_WRITE	   ? 'w' : '-',
+			       vma->vm_flags & VM_EXEC	   ? 'x' : '-',
+			       vma->vm_flags & VM_MAYSHARE ? 's' : 'p',
+			       vma->vm_flags & VM_LOCKED   ? 'l' : '-',
+			       vma->vm_flags & VM_IO	   ? 'i' : '-',
+			       vma->vm_offset );
+#if defined(__i386__)
+		pgprot = pgprot_val(vma->vm_page_prot);
+		DRM_SYSCTL_PRINT(" %c%c%c%c%c%c%c%c%c",
+			       pgprot & _PAGE_PRESENT  ? 'p' : '-',
+			       pgprot & _PAGE_RW       ? 'w' : 'r',
+			       pgprot & _PAGE_USER     ? 'u' : 's',
+			       pgprot & _PAGE_PWT      ? 't' : 'b',
+			       pgprot & _PAGE_PCD      ? 'u' : 'c',
+			       pgprot & _PAGE_ACCESSED ? 'a' : '-',
+			       pgprot & _PAGE_DIRTY    ? 'd' : '-',
+			       pgprot & _PAGE_4M       ? 'm' : 'k',
+			       pgprot & _PAGE_GLOBAL   ? 'g' : 'l' );
+#endif		
+		DRM_SYSCTL_PRINT("\n");
+		for (i = vma->vm_start; i < vma->vm_end; i += PAGE_SIZE) {
+			pgd = pgd_offset(vma->vm_mm, i);
+			pmd = pmd_offset(pgd, i);
+			pte = pte_offset(pmd, i);
+			if (pte_present(*pte)) {
+				address = __pa(pte_page(*pte))
+					+ (i & (PAGE_SIZE-1));
+				DRM_SYSCTL_PRINT("      0x%08lx -> 0x%08lx"
+					       " %c%c%c%c%c\n",
+					       i,
+					       address,
+					       pte_read(*pte)  ? 'r' : '-',
+					       pte_write(*pte) ? 'w' : '-',
+					       pte_exec(*pte)  ? 'x' : '-',
+					       pte_dirty(*pte) ? 'd' : '-',
+					       pte_young(*pte) ? 'a' : '-' );
+			} else {
+				DRM_SYSCTL_PRINT("      0x%08lx\n", i);
+			}
+		}
+	}
+	
+	SYSCTL_OUT(req, "", 1);
+	return 0;
+}
+
+static int drm_vma_info SYSCTL_HANDLER_ARGS
+{
+	drm_device_t *dev = arg1;
+	int	     ret;
+
+	lockmgr(&dev->dev_lock, LK_EXCLUSIVE, 0, curproc);
+	ret = _drm_vma_info(oidp, arg1, arg2, req);
+	lockmgr(&dev->dev_lock, LK_RELEASE, 0, curproc);
+	return ret;
+}
+#endif
+
+
+#if DRM_DMA_HISTOGRAM
+static int _drm_histo_info SYSCTL_HANDLER_ARGS
+{
+	drm_device_t	 *dev = arg1;
+	drm_device_dma_t *dma = dev->dma;
+	int		 i;
+	unsigned long	 slot_value = DRM_DMA_HISTOGRAM_INITIAL;
+	unsigned long	 prev_value = 0;
+	drm_buf_t	 *buffer;
+	char		 buf[128];
+	int              error;
+
+	DRM_SYSCTL_PRINT("general statistics:\n");
+	DRM_SYSCTL_PRINT("total	 %10u\n", atomic_read(&dev->histo.total));
+	DRM_SYSCTL_PRINT("open	 %10u\n", atomic_read(&dev->total_open));
+	DRM_SYSCTL_PRINT("close	 %10u\n", atomic_read(&dev->total_close));
+	DRM_SYSCTL_PRINT("ioctl	 %10u\n", atomic_read(&dev->total_ioctl));
+	DRM_SYSCTL_PRINT("irq	 %10u\n", atomic_read(&dev->total_irq));
+	DRM_SYSCTL_PRINT("ctx	 %10u\n", atomic_read(&dev->total_ctx));
+	
+	DRM_SYSCTL_PRINT("\nlock statistics:\n");
+	DRM_SYSCTL_PRINT("locks	 %10u\n", atomic_read(&dev->total_locks));
+	DRM_SYSCTL_PRINT("unlocks	 %10u\n", atomic_read(&dev->total_unlocks));
+	DRM_SYSCTL_PRINT("contends %10u\n", atomic_read(&dev->total_contends));
+	DRM_SYSCTL_PRINT("sleeps	 %10u\n", atomic_read(&dev->total_sleeps));
+
+
+	if (dma) {
+		DRM_SYSCTL_PRINT("\ndma statistics:\n");
+		DRM_SYSCTL_PRINT("prio	 %10u\n",
+			       atomic_read(&dma->total_prio));
+		DRM_SYSCTL_PRINT("bytes	 %10u\n",
+			       atomic_read(&dma->total_bytes));
+		DRM_SYSCTL_PRINT("dmas	 %10u\n",
+			       atomic_read(&dma->total_dmas));
+		DRM_SYSCTL_PRINT("missed:\n");
+		DRM_SYSCTL_PRINT("  dma	 %10u\n",
+			       atomic_read(&dma->total_missed_dma));
+		DRM_SYSCTL_PRINT("  lock	 %10u\n",
+			       atomic_read(&dma->total_missed_lock));
+		DRM_SYSCTL_PRINT("  free	 %10u\n",
+			       atomic_read(&dma->total_missed_free));
+		DRM_SYSCTL_PRINT("  sched	 %10u\n",
+			       atomic_read(&dma->total_missed_sched));
+		DRM_SYSCTL_PRINT("tried	 %10u\n",
+			       atomic_read(&dma->total_tried));
+		DRM_SYSCTL_PRINT("hit	 %10u\n",
+			       atomic_read(&dma->total_hit));
+		DRM_SYSCTL_PRINT("lost	 %10u\n",
+			       atomic_read(&dma->total_lost));
+		
+		buffer = dma->next_buffer;
+		if (buffer) {
+			DRM_SYSCTL_PRINT("next_buffer %7d\n", buffer->idx);
+		} else {
+			DRM_SYSCTL_PRINT("next_buffer    none\n");
+		}
+		buffer = dma->this_buffer;
+		if (buffer) {
+			DRM_SYSCTL_PRINT("this_buffer %7d\n", buffer->idx);
+		} else {
+			DRM_SYSCTL_PRINT("this_buffer    none\n");
+		}
+	}
+	
+
+	DRM_SYSCTL_PRINT("\nvalues:\n");
+	if (dev->lock.hw_lock) {
+		DRM_SYSCTL_PRINT("lock	       0x%08x\n",
+			       dev->lock.hw_lock->lock);
+	} else {
+		DRM_SYSCTL_PRINT("lock		     none\n");
+	}
+	DRM_SYSCTL_PRINT("context_flag   0x%08x\n", dev->context_flag);
+	DRM_SYSCTL_PRINT("interrupt_flag 0x%08x\n", dev->interrupt_flag);
+	DRM_SYSCTL_PRINT("dma_flag       0x%08x\n", dev->dma_flag);
+
+	DRM_SYSCTL_PRINT("queue_count    %10d\n",	 dev->queue_count);
+	DRM_SYSCTL_PRINT("last_context   %10d\n",	 dev->last_context);
+	DRM_SYSCTL_PRINT("last_switch    %10u\n",	 dev->last_switch);
+	DRM_SYSCTL_PRINT("last_checked   %10d\n",	 dev->last_checked);
+		
+	
+	DRM_SYSCTL_PRINT("\n		       q2d	  d2c	     c2f"
+		       "	q2c	   q2f	      dma	 sch"
+		       "	ctx	  lacq	     lhld\n\n");
+	for (i = 0; i < DRM_DMA_HISTOGRAM_SLOTS; i++) {
+		DRM_SYSCTL_PRINT("%s %10lu %10u %10u %10u %10u %10u"
+			       " %10u %10u %10u %10u %10u\n",
+			       i == DRM_DMA_HISTOGRAM_SLOTS - 1 ? ">=" : "< ",
+			       i == DRM_DMA_HISTOGRAM_SLOTS - 1
+			       ? prev_value : slot_value ,
+			       
+			       atomic_read(&dev->histo
+					   .queued_to_dispatched[i]),
+			       atomic_read(&dev->histo
+					   .dispatched_to_completed[i]),
+			       atomic_read(&dev->histo
+					   .completed_to_freed[i]),
+			       
+			       atomic_read(&dev->histo
+					   .queued_to_completed[i]),
+			       atomic_read(&dev->histo
+					   .queued_to_freed[i]),
+			       atomic_read(&dev->histo.dma[i]),
+			       atomic_read(&dev->histo.schedule[i]),
+			       atomic_read(&dev->histo.ctx[i]),
+			       atomic_read(&dev->histo.lacq[i]),
+			       atomic_read(&dev->histo.lhld[i]));
+		prev_value = slot_value;
+		slot_value = DRM_DMA_HISTOGRAM_NEXT(slot_value);
+	}
+	SYSCTL_OUT(req, "", 1);
+	return 0;
+}
+
+static int drm_histo_info SYSCTL_HANDLER_ARGS
+{
+	drm_device_t *dev = arg1;
+	int	     ret;
+
+	lockmgr(&dev->dev_lock, LK_EXCLUSIVE, 0, curproc);
+	ret = _drm_histo_info(oidp, arg1, arg2, req);
+	lockmgr(&dev->dev_lock, LK_RELEASE, 0, curproc);
+	return ret;
+}
+#endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/vm.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/vm.c
new file mode 100644
index 000000000..9c457fca3
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drm/vm.c
@@ -0,0 +1,104 @@
+/* vm.c -- Memory mapping for DRM -*- c -*-
+ * Created: Mon Jan  4 08:58:31 1999 by faith@precisioninsight.com
+ * Revised: Fri Aug 20 22:48:11 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/vm.c,v 1.7 1999/08/21 02:48:34 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/vm.c,v 1.1 1999/09/25 14:38:02 dawes Exp $
+ *
+ */
+
+#define __NO_VERSION__
+#include "drmP.h"
+
+#include <vm/vm.h>
+#include <vm/pmap.h>
+
+static int drm_dma_mmap(dev_t kdev, vm_offset_t offset, int prot)
+{
+	drm_device_t	 *dev	 = kdev->si_drv1;
+	drm_device_dma_t *dma	 = dev->dma;
+	unsigned long	 physical;
+	unsigned long	 page;
+
+	if (!dma)		   return -1; /* Error */
+	if (!dma->pagelist)	   return -1; /* Nothing allocated */
+
+	page	 = offset >> PAGE_SHIFT;
+	physical = dma->pagelist[page];
+
+	DRM_DEBUG("0x%08x (page %lu) => 0x%08lx\n", offset, page, physical);
+	return atop(physical);
+}
+
+int drm_mmap(dev_t kdev, vm_offset_t offset, int prot)
+{
+	drm_device_t	*dev	= kdev->si_drv1;
+	drm_map_t	*map	= NULL;
+	int		i;
+	
+	/* DRM_DEBUG("offset = 0x%x\n", offset); */
+
+	if (dev->dma
+	    && offset >= 0
+	    && offset < ptoa(dev->dma->page_count))
+		return drm_dma_mmap(kdev, offset, prot);
+
+				/* A sequential search of a linked list is
+				   fine here because: 1) there will only be
+				   about 5-10 entries in the list and, 2) a
+				   DRI client only has to do this mapping
+				   once, so it doesn't have to be optimized
+				   for performance, even if the list was a
+				   bit longer. */
+	for (i = 0; i < dev->map_count; i++) {
+		map = dev->maplist[i];
+		/* DRM_DEBUG("considering 0x%x..0x%x\n", map->offset, map->offset + map->size - 1); */
+		if (offset >= map->offset
+		    && offset < map->offset + map->size) break;
+	}
+	
+	if (i >= dev->map_count) {
+		DRM_DEBUG("can't find map\n");
+		return -1;
+	}
+	if (!map || ((map->flags&_DRM_RESTRICTED) && suser(curproc))) {
+		DRM_DEBUG("restricted map\n");
+		return -1;
+	}
+
+	switch (map->type) {
+	case _DRM_FRAME_BUFFER:
+	case _DRM_REGISTERS:
+	case _DRM_AGP:
+		return atop(offset);
+	case _DRM_SHM:
+		return atop(vtophys(offset));
+	default:
+		return -1;	/* This should never happen. */
+	}
+	DRM_DEBUG("bailing out\n");
+	
+	return -1;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drmP.h b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drmP.h
new file mode 100644
index 000000000..863836a66
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/drmP.h
@@ -0,0 +1,708 @@
+/* drmP.h -- Private header for Direct Rendering Manager -*- c -*-
+ * Created: Mon Jan  4 10:05:05 1999 by faith@precisioninsight.com
+ * Revised: Tue Oct 12 08:51:07 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drmP.h,v 1.58 1999/08/30 13:05:00 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drmP.h,v 1.1 1999/09/25 14:37:59 dawes Exp $
+ * 
+ */
+
+#ifndef _DRM_P_H_
+#define _DRM_P_H_
+
+#ifdef _KERNEL
+#include <sys/param.h>
+#include <sys/queue.h>
+#include <sys/malloc.h>
+#include <sys/kernel.h>
+#include <sys/module.h>
+#include <sys/systm.h>
+#include <sys/conf.h>
+#include <sys/stat.h>
+#include <sys/proc.h>
+#include <sys/lock.h>
+#include <sys/fcntl.h>
+#include <sys/uio.h>
+#include <sys/filio.h>
+#include <sys/sysctl.h>
+#include <sys/select.h>
+#include <sys/bus.h>
+#include <sys/taskqueue.h>
+
+#ifdef DRM_AGP
+#include <pci/agpvar.h>
+#endif
+
+#include "drm.h"
+
+typedef u_int32_t atomic_t;
+typedef u_int32_t cycles_t;
+typedef u_int32_t spinlock_t;
+#define atomic_set(p, v)	(*(p) = (v))
+#define atomic_read(p)		(*(p))
+#define atomic_inc(p)		atomic_add_int(p, 1)
+#define atomic_dec(p)		atomic_subtract_int(p, 1)
+#define atomic_add(n, p)	atomic_add_int(p, n)
+#define atomic_sub(n, p)	atomic_subtract_int(p, n)
+
+/* Fake this */
+static __inline u_int32_t
+test_and_set_bit(int b, volatile u_int32_t *p)
+{
+	u_int32_t m = 1<<b;
+	u_int32_t r = *p & m;
+	*p |= m;
+	return r;
+}
+
+static __inline void
+clear_bit(int b, volatile u_int32_t *p)
+{
+    atomic_clear_int(p + (b >> 5), 1 << (b & 0x1f));
+}
+
+static __inline void
+set_bit(int b, volatile u_int32_t *p)
+{
+    atomic_set_int(p + (b >> 5), 1 << (b & 0x1f));
+}
+
+static __inline int
+test_bit(int b, volatile u_int32_t *p)
+{
+    return p[b >> 5] & (1 << (b & 0x1f));
+}
+
+static __inline int
+find_first_zero_bit(volatile u_int32_t *p, int max)
+{
+    int b;
+
+    for (b = 0; b < max; b += 32) {
+	if (p[b >> 5]) {
+	    for (;;) {
+		if (p[b >> 5] & (1 << (b & 0x1f)))
+		    return b;
+		b++;
+	    }
+	}
+    }
+    return max;
+}
+
+#define spldrm()		spltty()
+
+#define memset(p, v, s)		bzero(p, s)
+
+/*
+ * Software interrupts for DMA pipe feeding. The FreeBSD kernel apis
+ * are severely lacking here.
+ */
+#define SWI_DRI		(SWI_VM+2)
+
+#define DRM_DEBUG_CODE 2	  /* Include debugging code (if > 1, then
+				     also include looping detection. */
+#define DRM_DMA_HISTOGRAM 1	  /* Make histogram of DMA latency. */
+
+#define DRM_HASH_SIZE	      16 /* Size of key hash table		  */
+#define DRM_KERNEL_CONTEXT    0	 /* Change drm_resctx if changed	  */
+#define DRM_RESERVED_CONTEXTS 1	 /* Change drm_resctx if changed	  */
+#define DRM_LOOPING_LIMIT     5000000
+#define DRM_BSZ		      1024 /* Buffer size for /dev/drm? output	  */
+#define DRM_TIME_SLICE	      (hz/20) /* Time slice for GLXContexts       */
+#define DRM_LOCK_SLICE	      1	/* Time slice for lock, in jiffies	  */
+
+#define DRM_FLAG_DEBUG	  0x01
+#define DRM_FLAG_NOCTX	  0x02
+
+#define DRM_MEM_DMA	  0
+#define DRM_MEM_SAREA	  1
+#define DRM_MEM_DRIVER	  2
+#define DRM_MEM_MAGIC	  3
+#define DRM_MEM_IOCTLS	  4
+#define DRM_MEM_MAPS	  5
+#define DRM_MEM_VMAS	  6
+#define DRM_MEM_BUFS	  7
+#define DRM_MEM_SEGS	  8
+#define DRM_MEM_PAGES	  9
+#define DRM_MEM_FILES	 10
+#define DRM_MEM_QUEUES	 11
+#define DRM_MEM_CMDS	 12
+#define DRM_MEM_MAPPINGS 13
+#define DRM_MEM_BUFLISTS 14
+#define DRM_MEM_AGPLISTS  15
+#define DRM_MEM_TOTALAGP  16
+#define DRM_MEM_BOUNDAGP  17
+#define DRM_MEM_CTXBITMAP 18
+
+#define DRM_MAX_CTXBITMAP (PAGE_SIZE * 8)
+
+				/* Backward compatibility section */
+#ifndef _PAGE_PWT
+				/* The name of _PAGE_WT was changed to
+				   _PAGE_PWT in Linux 2.2.6 */
+#define _PAGE_PWT _PAGE_WT
+#endif
+
+#define __drm_dummy_lock(lock) (*(__volatile__ unsigned int *)lock)
+#define _DRM_CAS(lock,old,new,__ret)				       \
+	do {							       \
+		int __dummy;	/* Can't mark eax as clobbered */      \
+		__asm__ __volatile__(				       \
+			"lock ; cmpxchg %4,%1\n\t"		       \
+			"setnz %0"				       \
+			: "=d" (__ret),				       \
+			  "=m" (__drm_dummy_lock(lock)),	       \
+			  "=a" (__dummy)			       \
+			: "2" (old),				       \
+			  "r" (new));				       \
+	} while (0)
+
+
+
+				/* Macros to make printk easier */
+#define DRM_ERROR(fmt, arg...) \
+	printf("error: " "[" DRM_NAME ":" __FUNCTION__ "] *ERROR* " fmt , ##arg)
+#define DRM_MEM_ERROR(area, fmt, arg...) \
+	printf("error: " "[" DRM_NAME ":" __FUNCTION__ ":%s] *ERROR* " fmt , \
+	       drm_mem_stats[area].name , ##arg)
+#define DRM_INFO(fmt, arg...)  printf("info: " "[" DRM_NAME "] " fmt , ##arg)
+
+#if DRM_DEBUG_CODE
+#define DRM_DEBUG(fmt, arg...)						  \
+	do {								  \
+		if (drm_flags&DRM_FLAG_DEBUG)				  \
+			printf("[" DRM_NAME ":" __FUNCTION__ "] " fmt ,	  \
+			       ##arg);					  \
+	} while (0)
+#else
+#define DRM_DEBUG(fmt, arg...)		 do { } while (0)
+#endif
+
+#define DRM_PROC_LIMIT (PAGE_SIZE-80)
+
+#define DRM_SYSCTL_PRINT(fmt, arg...)		\
+  snprintf(buf, sizeof(buf), fmt, ##arg);	\
+  error = SYSCTL_OUT(req, buf, strlen(buf));	\
+  if (error) return error;
+
+#define DRM_SYSCTL_PRINT_RET(ret, fmt, arg...)	\
+  snprintf(buf, sizeof(buf), fmt, ##arg);	\
+  error = SYSCTL_OUT(req, buf, strlen(buf));	\
+  if (error) { ret; return error; }
+
+				/* Internal types and structures */
+#define DRM_ARRAY_SIZE(x) (sizeof(x)/sizeof(x[0]))
+#define DRM_MIN(a,b) ((a)<(b)?(a):(b))
+#define DRM_MAX(a,b) ((a)>(b)?(a):(b))
+
+#define DRM_LEFTCOUNT(x) (((x)->rp + (x)->count - (x)->wp) % ((x)->count + 1))
+#define DRM_BUFCOUNT(x) ((x)->count - DRM_LEFTCOUNT(x))
+#define DRM_WAITCOUNT(dev,idx) DRM_BUFCOUNT(&dev->queuelist[idx]->waitlist)
+
+typedef struct drm_ioctl_desc {
+	d_ioctl_t	     *func;
+	int		     auth_needed;
+	int		     root_only;
+} drm_ioctl_desc_t;
+
+typedef struct drm_devstate {
+	pid_t		  owner;	/* X server pid holding x_lock */
+	
+} drm_devstate_t;
+
+typedef struct drm_magic_entry {
+	drm_magic_t	       magic;
+	struct drm_file	       *priv;
+	struct drm_magic_entry *next;
+} drm_magic_entry_t;
+
+typedef struct drm_magic_head {
+       struct drm_magic_entry *head;
+       struct drm_magic_entry *tail;
+} drm_magic_head_t;
+
+typedef struct drm_vma_entry {
+	struct vm_area_struct *vma;
+	struct drm_vma_entry  *next;
+	pid_t		      pid;
+} drm_vma_entry_t;
+
+typedef struct drm_buf {
+	int		  idx;	       /* Index into master buflist	     */
+	int		  total;       /* Buffer size			     */
+	int		  order;       /* log-base-2(total)		     */
+	int		  used;	       /* Amount of buffer in use (for DMA)  */
+	unsigned long	  offset;      /* Byte offset (used internally)	     */
+	void		  *address;    /* Address of buffer		     */
+	unsigned long	  bus_address; /* Bus address of buffer		     */
+	struct drm_buf	  *next;       /* Kernel-only: used for free list    */
+	__volatile__ int  waiting;     /* On kernel DMA queue		     */
+	__volatile__ int  pending;     /* On hardware DMA queue		     */
+	int		  dma_wait;    /* Processes waiting		     */
+	pid_t		  pid;	       /* PID of holding process	     */
+	int		  context;     /* Kernel queue for this buffer	     */
+	int		  while_locked;/* Dispatch this buffer while locked  */
+	enum {
+		DRM_LIST_NONE	 = 0,
+		DRM_LIST_FREE	 = 1,
+		DRM_LIST_WAIT	 = 2,
+		DRM_LIST_PEND	 = 3,
+		DRM_LIST_PRIO	 = 4,
+		DRM_LIST_RECLAIM = 5
+	}		  list;	       /* Which list we're on		     */
+
+	void *dev_private;
+	int dev_priv_size;
+
+#if DRM_DMA_HISTOGRAM
+	struct timespec	  time_queued;	   /* Queued to kernel DMA queue     */
+	struct timespec	  time_dispatched; /* Dispatched to hardware	     */
+	struct timespec	  time_completed;  /* Completed by hardware	     */
+	struct timespec	  time_freed;	   /* Back on freelist		     */
+#endif
+} drm_buf_t;
+
+#if DRM_DMA_HISTOGRAM
+#define DRM_DMA_HISTOGRAM_SLOTS		  9
+#define DRM_DMA_HISTOGRAM_INITIAL	 10
+#define DRM_DMA_HISTOGRAM_NEXT(current)	 ((current)*10)
+typedef struct drm_histogram {
+	atomic_t	  total;
+	
+	atomic_t	  queued_to_dispatched[DRM_DMA_HISTOGRAM_SLOTS];
+	atomic_t	  dispatched_to_completed[DRM_DMA_HISTOGRAM_SLOTS];
+	atomic_t	  completed_to_freed[DRM_DMA_HISTOGRAM_SLOTS];
+	
+	atomic_t	  queued_to_completed[DRM_DMA_HISTOGRAM_SLOTS];
+	atomic_t	  queued_to_freed[DRM_DMA_HISTOGRAM_SLOTS];
+	
+	atomic_t	  dma[DRM_DMA_HISTOGRAM_SLOTS];
+	atomic_t	  schedule[DRM_DMA_HISTOGRAM_SLOTS];
+	atomic_t	  ctx[DRM_DMA_HISTOGRAM_SLOTS];
+	atomic_t	  lacq[DRM_DMA_HISTOGRAM_SLOTS];
+	atomic_t	  lhld[DRM_DMA_HISTOGRAM_SLOTS];
+} drm_histogram_t;
+#endif
+
+				/* bufs is one longer than it has to be */
+typedef struct drm_waitlist {
+	int		  count;	/* Number of possible buffers	   */
+	drm_buf_t	  **bufs;	/* List of pointers to buffers	   */
+	drm_buf_t	  **rp;		/* Read pointer			   */
+	drm_buf_t	  **wp;		/* Write pointer		   */
+	drm_buf_t	  **end;	/* End pointer			   */
+	spinlock_t	  read_lock;
+	spinlock_t	  write_lock;
+} drm_waitlist_t;
+
+typedef struct drm_freelist {
+	int		  initialized; /* Freelist in use		   */
+	atomic_t	  count;       /* Number of free buffers	   */
+	drm_buf_t	  *next;       /* End pointer			   */
+	
+	int		  waiting;     /* Processes waiting on free bufs   */
+	int		  low_mark;    /* Low water mark		   */
+	int		  high_mark;   /* High water mark		   */
+	atomic_t	  wfh;	       /* If waiting for high mark	   */
+} drm_freelist_t;
+
+typedef struct drm_buf_entry {
+	int		  buf_size;
+	int		  buf_count;
+	drm_buf_t	  *buflist;
+	int		  seg_count;
+	int		  page_order;
+	unsigned long	  *seglist;
+
+	drm_freelist_t	  freelist;
+} drm_buf_entry_t;
+
+typedef struct drm_hw_lock {
+	__volatile__ unsigned int lock;
+	char			  padding[60]; /* Pad to cache line */
+} drm_hw_lock_t;
+
+typedef TAILQ_HEAD(drm_file_list, drm_file) drm_file_list_t;
+typedef struct drm_file {
+	TAILQ_ENTRY(drm_file) link;
+	int		  authenticated;
+	int		  minor;
+	pid_t		  pid;
+	uid_t		  uid;
+	int		  refs;
+	drm_magic_t	  magic;
+	unsigned long	  ioctl_count;
+	struct drm_device *devXX;
+} drm_file_t;
+
+
+typedef struct drm_queue {
+	atomic_t	  use_count;	/* Outstanding uses (+1)	    */
+	atomic_t	  finalization;	/* Finalization in progress	    */
+	atomic_t	  block_count;	/* Count of processes waiting	    */
+	atomic_t	  block_read;	/* Queue blocked for reads	    */
+	int		  read_queue;	/* Processes waiting on block_read  */
+	atomic_t	  block_write;	/* Queue blocked for writes	    */
+	int		  write_queue;	/* Processes waiting on block_write */
+	atomic_t	  total_queued;	/* Total queued statistic	    */
+	atomic_t	  total_flushed;/* Total flushes statistic	    */
+	atomic_t	  total_locks;	/* Total locks statistics	    */
+	drm_ctx_flags_t	  flags;	/* Context preserving and 2D-only   */
+	drm_waitlist_t	  waitlist;	/* Pending buffers		    */
+	int		  flush_queue;	/* Processes waiting until flush    */
+} drm_queue_t;
+
+typedef struct drm_lock_data {
+	drm_hw_lock_t	  *hw_lock;	/* Hardware lock		   */
+	pid_t		  pid;		/* PID of lock holder (0=kernel)   */
+	int		  lock_queue;	/* Queue of blocked processes	   */
+	unsigned long	  lock_time;	/* Time of last lock in jiffies	   */
+} drm_lock_data_t;
+
+typedef struct drm_device_dma {
+				/* Performance Counters */
+	atomic_t	  total_prio;	/* Total DRM_DMA_PRIORITY	   */
+	atomic_t	  total_bytes;	/* Total bytes DMA'd		   */
+	atomic_t	  total_dmas;	/* Total DMA buffers dispatched	   */
+	
+	atomic_t	  total_missed_dma;  /* Missed drm_do_dma	    */
+	atomic_t	  total_missed_lock; /* Missed lock in drm_do_dma   */
+	atomic_t	  total_missed_free; /* Missed drm_free_this_buffer */
+	atomic_t	  total_missed_sched;/* Missed drm_dma_schedule	    */
+
+	atomic_t	  total_tried;	/* Tried next_buffer		    */
+	atomic_t	  total_hit;	/* Sent next_buffer		    */
+	atomic_t	  total_lost;	/* Lost interrupt		    */
+
+	drm_buf_entry_t	  bufs[DRM_MAX_ORDER+1];
+	int		  buf_count;
+	drm_buf_t	  **buflist;	/* Vector of pointers info bufs	   */
+	int		  seg_count; 
+	int		  page_count;
+	vm_offset_t	  *pagelist;
+	unsigned long	  byte_count;
+	enum {
+	   _DRM_DMA_USE_AGP = 0x01
+	} flags;
+
+				/* DMA support */
+	drm_buf_t	  *this_buffer;	/* Buffer being sent		   */
+	drm_buf_t	  *next_buffer; /* Selected buffer to send	   */
+	drm_queue_t	  *next_queue;	/* Queue from which buffer selected*/
+	int		   waiting;	/* Processes waiting on free bufs  */
+} drm_device_dma_t;
+
+#ifdef DRM_AGP
+
+typedef struct drm_agp_mem {
+	void               *handle;
+	unsigned long      bound; /* address */
+	int                pages;
+	struct drm_agp_mem *prev;
+	struct drm_agp_mem *next;
+} drm_agp_mem_t;
+
+typedef struct drm_agp_head {
+	device_t	   agpdev;
+	struct agp_info    info;
+	const char         *chipset;
+	drm_agp_mem_t      *memory;
+	unsigned long      mode;
+	int                enabled;
+	int                acquired;
+	unsigned long      base;
+   	int 		   agp_mtrr;
+} drm_agp_head_t;
+
+#endif
+
+typedef struct drm_device {
+	const char	  *name;	/* Simple driver name		   */
+	char		  *unique;	/* Unique identifier: e.g., busid  */
+	int		  unique_len;	/* Length of unique field	   */
+	device_t	  device;	/* Device instance from newbus     */
+	dev_t		  devnode;	/* Device number for mknod	   */
+	char		  *devname;	/* For /proc/interrupts		   */
+	
+	int		  blocked;	/* Blocked due to VC switch?	   */
+	int		  flags;	/* Flags to open(2)		   */
+	int		  writable;	/* Opened with FWRITE		   */
+	struct proc_dir_entry *root;	/* Root for this device's entries  */
+
+				/* Locks */
+	struct simplelock count_lock;	/* For inuse, open_count, buf_use  */
+	struct lock	  dev_lock;	/* For others			   */
+
+				/* Usage Counters */
+	int		  open_count;	/* Outstanding files open	   */
+	atomic_t	  ioctl_count;	/* Outstanding IOCTLs pending	   */
+	atomic_t	  vma_count;	/* Outstanding vma areas open	   */
+	int		  buf_use;	/* Buffers in use -- cannot alloc  */
+	atomic_t	  buf_alloc;	/* Buffer allocation in progress   */
+
+				/* Performance Counters */
+	atomic_t	  total_open;
+	atomic_t	  total_close;
+	atomic_t	  total_ioctl;
+	atomic_t	  total_irq;	/* Total interruptions		   */
+	atomic_t	  total_ctx;	/* Total context switches	   */
+	
+	atomic_t	  total_locks;
+	atomic_t	  total_unlocks;
+	atomic_t	  total_contends;
+	atomic_t	  total_sleeps;
+
+				/* Authentication */
+	drm_file_list_t   files;
+	drm_magic_head_t  magiclist[DRM_HASH_SIZE];
+
+				/* Memory management */
+	drm_map_t	  **maplist;	/* Vector of pointers to regions   */
+	int		  map_count;	/* Number of mappable regions	   */
+
+	drm_vma_entry_t	  *vmalist;	/* List of vmas (for debugging)	   */
+	drm_lock_data_t	  lock;		/* Information on hardware lock	   */
+
+				/* DMA queues (contexts) */
+	int		  queue_count;	/* Number of active DMA queues	   */
+	int		  queue_reserved; /* Number of reserved DMA queues */
+	int		  queue_slots;	/* Actual length of queuelist	   */
+	drm_queue_t	  **queuelist;	/* Vector of pointers to DMA queues */
+	drm_device_dma_t  *dma;		/* Optional pointer for DMA support */
+
+				/* Context support */
+	struct resource   *irq;		/* Interrupt used by board	   */
+	void		  *irqh;	/* Handle from bus_setup_intr      */
+	__volatile__ int  context_flag;	 /* Context swapping flag	   */
+	__volatile__ int  interrupt_flag;/* Interruption handler flag	   */
+	__volatile__ int  dma_flag;	 /* DMA dispatch flag		   */
+	struct callout	  timer;	/* Timer for delaying ctx switch   */
+	int		  context_wait; /* Processes waiting on ctx switch */
+	int		  last_checked;	/* Last context checked for DMA	   */
+	int		  last_context;	/* Last current context		   */
+	int		  last_switch;	/* Time at last context switch  */
+	struct task	  task;
+	struct timespec	  ctx_start;
+	struct timespec	  lck_start;
+#if DRM_DMA_HISTOGRAM
+	drm_histogram_t	  histo;
+#endif
+	
+				/* Callback to X server for context switch
+				   and for heavy-handed reset. */
+	char		  buf[DRM_BSZ]; /* Output buffer		   */
+	char		  *buf_rp;	/* Read pointer			   */
+	char		  *buf_wp;	/* Write pointer		   */
+	char		  *buf_end;	/* End pointer			   */
+	struct sigio	  *buf_sigio;   /* Processes waiting for SIGIO	   */
+	struct selinfo    buf_sel;	/* Workspace for select/poll       */
+	int		  buf_readers;	/* Processes waiting to read	   */
+	int		  buf_writers;	/* Processes waiting to ctx switch */
+	int		  buf_selecting; /* True if poll sleeper           */
+
+				/* Sysctl support */
+	struct drm_sysctl_info *sysctl;
+
+#ifdef DRM_AGP
+	drm_agp_head_t    *agp;
+#endif
+	u_int32_t	  *ctx_bitmap;
+	void		  *dev_private;
+} drm_device_t;
+
+
+				/* Internal function definitions */
+
+				/* Misc. support (init.c) */
+extern int	     drm_flags;
+extern void	     drm_parse_options(char *s);
+
+
+				/* Device support (fops.c) */
+extern drm_file_t *drm_find_file_by_proc(drm_device_t *dev, struct proc *p);
+extern int drm_open_helper(dev_t kdev, int flags, int fmt, struct proc *p,
+			   drm_device_t *dev);
+extern d_close_t     drm_close;
+extern d_read_t	     drm_read;
+extern d_write_t     drm_write;
+extern d_poll_t	     drm_poll;
+extern int	     drm_fsetown(dev_t kdev, u_long cmd, caddr_t data,
+				 int flags, struct proc *p);
+extern int	     drm_fgetown(dev_t kdev, u_long cmd, caddr_t data,
+				 int flags, struct proc *p);
+extern int	     drm_write_string(drm_device_t *dev, const char *s);
+
+#if 0
+				/* Mapping support (vm.c) */
+extern unsigned long drm_vm_nopage(struct vm_area_struct *vma,
+				   unsigned long address,
+				   int write_access);
+extern unsigned long drm_vm_shm_nopage(struct vm_area_struct *vma,
+				       unsigned long address,
+				       int write_access);
+extern unsigned long drm_vm_dma_nopage(struct vm_area_struct *vma,
+				       unsigned long address,
+				       int write_access);
+extern void	     drm_vm_open(struct vm_area_struct *vma);
+extern void	     drm_vm_close(struct vm_area_struct *vma);
+extern int	     drm_mmap_dma(struct file *filp,
+				  struct vm_area_struct *vma);
+#endif
+extern d_mmap_t	     drm_mmap;
+
+				/* Proc support (proc.c) */
+extern int	     drm_sysctl_init(drm_device_t *dev);
+extern int	     drm_sysctl_cleanup(drm_device_t *dev);
+
+				/* Memory management support (memory.c) */
+extern void	     drm_mem_init(void);
+extern int	     drm_mem_info SYSCTL_HANDLER_ARGS;
+extern void	     *drm_alloc(size_t size, int area);
+extern void	     *drm_realloc(void *oldpt, size_t oldsize, size_t size,
+				  int area);
+extern char	     *drm_strdup(const char *s, int area);
+extern void	     drm_strfree(char *s, int area);
+extern void	     drm_free(void *pt, size_t size, int area);
+extern unsigned long drm_alloc_pages(int order, int area);
+extern void	     drm_free_pages(unsigned long address, int order,
+				    int area);
+extern void	     *drm_ioremap(unsigned long offset, unsigned long size);
+extern void	     drm_ioremapfree(void *pt, unsigned long size);
+
+#ifdef DRM_AGP
+extern void	    *drm_alloc_agp(int pages, u_int32_t type);
+extern int           drm_free_agp(void *handle, int pages);
+extern int           drm_bind_agp(void *handle, unsigned int start);
+extern int           drm_unbind_agp(void *handle);
+#endif
+
+				/* Buffer management support (bufs.c) */
+extern int	     drm_order(unsigned long size);
+extern d_ioctl_t     drm_addmap;
+extern d_ioctl_t     drm_addbufs;
+extern d_ioctl_t     drm_infobufs;
+extern d_ioctl_t     drm_markbufs;
+extern d_ioctl_t     drm_freebufs;
+extern d_ioctl_t     drm_mapbufs;
+
+
+				/* Buffer list management support (lists.c) */
+extern int	     drm_waitlist_create(drm_waitlist_t *bl, int count);
+extern int	     drm_waitlist_destroy(drm_waitlist_t *bl);
+extern int	     drm_waitlist_put(drm_waitlist_t *bl, drm_buf_t *buf);
+extern drm_buf_t     *drm_waitlist_get(drm_waitlist_t *bl);
+
+extern int	     drm_freelist_create(drm_freelist_t *bl, int count);
+extern int	     drm_freelist_destroy(drm_freelist_t *bl);
+extern int	     drm_freelist_put(drm_device_t *dev, drm_freelist_t *bl,
+				      drm_buf_t *buf);
+extern drm_buf_t     *drm_freelist_get(drm_freelist_t *bl, int block);
+
+				/* DMA support (gen_dma.c) */
+extern void	     drm_dma_setup(drm_device_t *dev);
+extern void	     drm_dma_takedown(drm_device_t *dev);
+extern void	     drm_free_buffer(drm_device_t *dev, drm_buf_t *buf);
+extern void	     drm_reclaim_buffers(drm_device_t *dev, pid_t pid);
+extern int	     drm_context_switch(drm_device_t *dev, int old, int new);
+extern int	     drm_context_switch_complete(drm_device_t *dev, int new);
+extern void	     drm_wakeup(drm_device_t *dev, drm_buf_t *buf);
+extern void	     drm_clear_next_buffer(drm_device_t *dev);
+extern int	     drm_select_queue(drm_device_t *dev,
+				      void (*wrapper)(void *));
+extern int	     drm_dma_enqueue(drm_device_t *dev, drm_dma_t *dma);
+extern int	     drm_dma_get_buffers(drm_device_t *dev, drm_dma_t *dma);
+#if DRM_DMA_HISTOGRAM
+extern int	     drm_histogram_slot(struct timespec *ts);
+extern void	     drm_histogram_compute(drm_device_t *dev, drm_buf_t *buf);
+#endif
+
+
+				/* Misc. IOCTL support (ioctl.c) */
+extern d_ioctl_t     drm_irq_busid;
+extern d_ioctl_t     drm_getunique;
+extern d_ioctl_t     drm_setunique;
+
+
+				/* Context IOCTL support (context.c) */
+extern d_ioctl_t     drm_resctx;
+extern d_ioctl_t     drm_addctx;
+extern d_ioctl_t     drm_modctx;
+extern d_ioctl_t     drm_getctx;
+extern d_ioctl_t     drm_switchctx;
+extern d_ioctl_t     drm_newctx;
+extern d_ioctl_t     drm_rmctx;
+
+
+				/* Drawable IOCTL support (drawable.c) */
+extern d_ioctl_t     drm_adddraw;
+extern d_ioctl_t     drm_rmdraw;
+
+
+				/* Authentication IOCTL support (auth.c) */
+extern int	     drm_add_magic(drm_device_t *dev, drm_file_t *priv,
+				   drm_magic_t magic);
+extern int	     drm_remove_magic(drm_device_t *dev, drm_magic_t magic);
+extern d_ioctl_t     drm_getmagic;
+extern d_ioctl_t     drm_authmagic;
+
+
+				/* Locking IOCTL support (lock.c) */
+extern d_ioctl_t     drm_block;
+extern d_ioctl_t     drm_unblock;
+extern int	     drm_lock_take(__volatile__ unsigned int *lock,
+				   unsigned int context);
+extern int	     drm_lock_transfer(drm_device_t *dev,
+				       __volatile__ unsigned int *lock,
+				       unsigned int context);
+extern int	     drm_lock_free(drm_device_t *dev,
+				   __volatile__ unsigned int *lock,
+				   unsigned int context);
+extern d_ioctl_t     drm_finish;
+extern int	     drm_flush_unblock(drm_device_t *dev, int context,
+				       drm_lock_flags_t flags);
+extern int	     drm_flush_block_and_flush(drm_device_t *dev, int context,
+					       drm_lock_flags_t flags);
+
+				/* Context Bitmap support (ctxbitmap.c) */
+extern int	     drm_ctxbitmap_init(drm_device_t *dev);
+extern void	     drm_ctxbitmap_cleanup(drm_device_t *dev);
+extern int	     drm_ctxbitmap_next(drm_device_t *dev);
+extern void	     drm_ctxbitmap_free(drm_device_t *dev, int ctx_handle);
+
+#ifdef DRM_AGP
+				/* AGP/GART support (agpsupport.c) */
+extern drm_agp_head_t *drm_agp_init(void);
+extern d_ioctl_t     drm_agp_acquire;
+extern d_ioctl_t     drm_agp_release;
+extern d_ioctl_t     drm_agp_enable;
+extern d_ioctl_t     drm_agp_info;
+extern d_ioctl_t     drm_agp_alloc;
+extern d_ioctl_t     drm_agp_free;
+extern d_ioctl_t     drm_agp_unbind;
+extern d_ioctl_t     drm_agp_bind;
+#endif
+#endif
+#endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/gamma/Makefile b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/gamma/Makefile
new file mode 100644
index 000000000..dd6110382
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/gamma/Makefile
@@ -0,0 +1,15 @@
+# $FreeBSD$
+
+KMOD	= gamma
+SRCS    = gamma_drv.c gamma_dma.c
+SRCS   += device_if.h bus_if.h pci_if.h
+CFLAGS += ${DEBUG_FLAGS} -I..
+KERN    = /usr/src/sys
+
+@:
+	ln -sf /sys @
+
+machine:
+	ln -sf /sys/i386/include machine
+
+.include "/usr/src/sys/conf/kmod.mk"
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/gamma/gamma_dma.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/gamma/gamma_dma.c
new file mode 100644
index 000000000..177440db8
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/gamma/gamma_dma.c
@@ -0,0 +1,802 @@
+/* gamma_dma.c -- DMA support for GMX 2000 -*- c -*-
+ * Created: Fri Mar 19 14:30:16 1999 by faith@precisioninsight.com
+ * Revised: Thu Sep 16 12:55:37 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_dma.c,v 1.9 1999/09/16 16:56:18 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_dma.c,v 1.1 1999/09/25 14:38:00 dawes Exp $
+ *
+ */
+
+#define __NO_VERSION__
+#include "drmP.h"
+#include "gamma_drv.h"
+
+#include <machine/bus.h>
+#include <machine/resource.h>
+#include <sys/rman.h>
+#include <vm/vm.h>
+#include <vm/pmap.h>
+
+/* WARNING!!! MAGIC NUMBER!!!  The number of regions already added to the
+   kernel must be specified here.  Currently, the number is 2.	This must
+   match the order the X server uses for instantiating register regions ,
+   or must be passed in a new ioctl. */
+#define GAMMA_REG(reg)						   \
+	(2							   \
+	 + ((reg < 0x1000)					   \
+	    ? 0							   \
+	    : ((reg < 0x10000) ? 1 : ((reg < 0x11000) ? 2 : 3))))
+
+#define GAMMA_OFF(reg)						   \
+	((reg < 0x1000)						   \
+	 ? reg							   \
+	 : ((reg < 0x10000)					   \
+	    ? (reg - 0x1000)					   \
+	    : ((reg < 0x11000)					   \
+	       ? (reg - 0x10000)				   \
+	       : (reg - 0x11000))))
+
+#define GAMMA_BASE(reg)	 ((unsigned long)dev->maplist[GAMMA_REG(reg)]->handle)
+#define GAMMA_ADDR(reg)	 (GAMMA_BASE(reg) + GAMMA_OFF(reg))
+#define GAMMA_DEREF(reg) *(__volatile__ int *)GAMMA_ADDR(reg)
+#define GAMMA_READ(reg)	 GAMMA_DEREF(reg)
+#define GAMMA_WRITE(reg,val) do { GAMMA_DEREF(reg) = val; } while (0)
+
+#define GAMMA_BROADCASTMASK    0x9378
+#define GAMMA_COMMANDINTENABLE 0x0c48
+#define GAMMA_DMAADDRESS       0x0028
+#define GAMMA_DMACOUNT	       0x0030
+#define GAMMA_FILTERMODE       0x8c00
+#define GAMMA_GCOMMANDINTFLAGS 0x0c50
+#define GAMMA_GCOMMANDMODE     0x0c40
+#define GAMMA_GCOMMANDSTATUS   0x0c60
+#define GAMMA_GDELAYTIMER      0x0c38
+#define GAMMA_GDMACONTROL      0x0060
+#define GAMMA_GINTENABLE       0x0808
+#define GAMMA_GINTFLAGS	       0x0810
+#define GAMMA_INFIFOSPACE      0x0018
+#define GAMMA_OUTFIFOWORDS     0x0020
+#define GAMMA_OUTPUTFIFO       0x2000
+#define GAMMA_SYNC	       0x8c40
+#define GAMMA_SYNC_TAG	       0x0188
+
+static __inline void gamma_dma_dispatch(drm_device_t *dev,
+					vm_offset_t address,
+					vm_size_t length)
+{
+	GAMMA_WRITE(GAMMA_DMAADDRESS, vtophys(address));
+	while (GAMMA_READ(GAMMA_GCOMMANDSTATUS) != 4)
+		;
+	GAMMA_WRITE(GAMMA_DMACOUNT, length / 4);
+}
+
+static __inline void gamma_dma_quiescent(drm_device_t *dev)
+{
+	while (GAMMA_READ(GAMMA_DMACOUNT))
+		;
+	while (GAMMA_READ(GAMMA_INFIFOSPACE) < 3)
+		;
+	GAMMA_WRITE(GAMMA_BROADCASTMASK, 3);
+	GAMMA_WRITE(GAMMA_FILTERMODE, 1 << 10);
+	GAMMA_WRITE(GAMMA_SYNC, 0);
+	
+				/* Read from first MX */
+	do {
+		while (!GAMMA_READ(GAMMA_OUTFIFOWORDS))
+			;
+	} while (GAMMA_READ(GAMMA_OUTPUTFIFO) != GAMMA_SYNC_TAG);
+	
+
+				/* Read from second MX */
+	do {
+		while (!GAMMA_READ(GAMMA_OUTFIFOWORDS + 0x10000))
+			;
+	} while (GAMMA_READ(GAMMA_OUTPUTFIFO + 0x10000) != GAMMA_SYNC_TAG);
+}
+
+static __inline void gamma_dma_ready(drm_device_t *dev)
+{
+	while (GAMMA_READ(GAMMA_DMACOUNT))
+		;
+}
+
+static __inline int gamma_dma_is_ready(drm_device_t *dev)
+{
+	return !GAMMA_READ(GAMMA_DMACOUNT);
+}
+
+static void gamma_dma_service(void *arg)
+{
+	drm_device_t	 *dev = (drm_device_t *)arg;
+	drm_device_dma_t *dma = dev->dma;
+	
+	atomic_inc(&dev->total_irq);
+	GAMMA_WRITE(GAMMA_GDELAYTIMER, 0xc350/2); /* 0x05S */
+	GAMMA_WRITE(GAMMA_GCOMMANDINTFLAGS, 8);
+	GAMMA_WRITE(GAMMA_GINTFLAGS, 0x2001);
+	if (gamma_dma_is_ready(dev)) {
+				/* Free previous buffer */
+		if (test_and_set_bit(0, &dev->dma_flag)) {
+			atomic_inc(&dma->total_missed_free);
+			return;
+		}
+		if (dma->this_buffer) {
+			drm_free_buffer(dev, dma->this_buffer);
+			dma->this_buffer = NULL;
+		}
+		clear_bit(0, &dev->dma_flag);
+
+#if 0
+				/* Dispatch new buffer */
+		queue_task(&dev->tq, &tq_immediate);
+		mark_bh(IMMEDIATE_BH);
+#endif
+	}
+}
+
+/* Only called by gamma_dma_schedule. */
+static int gamma_do_dma(drm_device_t *dev, int locked)
+{
+	unsigned long	 address;
+	unsigned long	 length;
+	drm_buf_t	 *buf;
+	int		 retcode = 0;
+	drm_device_dma_t *dma = dev->dma;
+#if DRM_DMA_HISTOGRAM
+	struct timespec	 dma_start, dma_stop;
+#endif
+
+	if (test_and_set_bit(0, &dev->dma_flag)) {
+		atomic_inc(&dma->total_missed_dma);
+		return EBUSY;
+	}
+	
+#if DRM_DMA_HISTOGRAM
+	getnanotime(&dma_start);
+#endif
+
+	if (!dma->next_buffer) {
+		DRM_ERROR("No next_buffer\n");
+		clear_bit(0, &dev->dma_flag);
+		return EINVAL;
+	}
+
+	buf	= dma->next_buffer;
+	address = (unsigned long)buf->address;
+	length	= buf->used;
+
+	DRM_DEBUG("context %d, buffer %d (%ld bytes)\n",
+		  buf->context, buf->idx, length);
+
+	if (buf->list == DRM_LIST_RECLAIM) {
+		drm_clear_next_buffer(dev);
+		drm_free_buffer(dev, buf);
+		clear_bit(0, &dev->dma_flag);
+		return EINVAL;
+	}
+
+	if (!length) {
+		DRM_ERROR("0 length buffer\n");
+		drm_clear_next_buffer(dev);
+		drm_free_buffer(dev, buf);
+		clear_bit(0, &dev->dma_flag);
+		return 0;
+	}
+	
+	if (!gamma_dma_is_ready(dev)) {
+		clear_bit(0, &dev->dma_flag);
+		return EBUSY;
+	}
+
+	if (buf->while_locked) {
+		if (!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+			DRM_ERROR("Dispatching buffer %d from pid %d"
+				  " \"while locked\", but no lock held\n",
+				  buf->idx, buf->pid);
+		}
+	} else {
+		if (!locked && !drm_lock_take(&dev->lock.hw_lock->lock,
+					      DRM_KERNEL_CONTEXT)) {
+			atomic_inc(&dma->total_missed_lock);
+			clear_bit(0, &dev->dma_flag);
+			return EBUSY;
+		}
+	}
+
+	if (dev->last_context != buf->context
+	    && !(dev->queuelist[buf->context]->flags
+		 & _DRM_CONTEXT_PRESERVED)) {
+				/* PRE: dev->last_context != buf->context */
+		if (drm_context_switch(dev, dev->last_context, buf->context)) {
+			drm_clear_next_buffer(dev);
+			drm_free_buffer(dev, buf);
+		}
+		retcode = EBUSY;
+		goto cleanup;
+			
+				/* POST: we will wait for the context
+				   switch and will dispatch on a later call
+				   when dev->last_context == buf->context.
+				   NOTE WE HOLD THE LOCK THROUGHOUT THIS
+				   TIME! */
+	}
+
+	drm_clear_next_buffer(dev);
+	buf->pending	 = 1;
+	buf->waiting	 = 0;
+	buf->list	 = DRM_LIST_PEND;
+#if DRM_DMA_HISTOGRAM
+	getnanotime(&buf->time_dispatched);
+#endif
+
+	gamma_dma_dispatch(dev, address, length);
+	drm_free_buffer(dev, dma->this_buffer);
+	dma->this_buffer = buf;
+
+	atomic_add(length, &dma->total_bytes);
+	atomic_inc(&dma->total_dmas);
+
+	if (!buf->while_locked && !dev->context_flag && !locked) {
+		if (drm_lock_free(dev, &dev->lock.hw_lock->lock,
+				  DRM_KERNEL_CONTEXT)) {
+			DRM_ERROR("\n");
+		}
+	}
+cleanup:
+
+	clear_bit(0, &dev->dma_flag);
+
+#if DRM_DMA_HISTOGRAM
+	getnanotime(&dma_stop);
+	timespecsub(&dma_stop, &dma_start);
+	atomic_inc(&dev->histo.ctx[drm_histogram_slot(&dma_stop)]);
+#endif
+
+	return retcode;
+}
+
+static void gamma_dma_schedule_wrapper(void *dev)
+{
+	gamma_dma_schedule(dev, 0);
+}
+
+int gamma_dma_schedule(drm_device_t *dev, int locked)
+{
+	int		 next;
+	drm_queue_t	 *q;
+	drm_buf_t	 *buf;
+	int		 retcode   = 0;
+	int		 processed = 0;
+	int		 missed;
+	int		 expire	   = 20;
+	drm_device_dma_t *dma	   = dev->dma;
+#if DRM_DMA_HISTOGRAM
+	struct timespec	 schedule_start;
+#endif
+
+	if (test_and_set_bit(0, &dev->interrupt_flag)) {
+				/* Not reentrant */
+		atomic_inc(&dma->total_missed_sched);
+		return EBUSY;
+	}
+	missed = atomic_read(&dma->total_missed_sched);
+
+#if DRM_DMA_HISTOGRAM
+	getnanotime(&schedule_start);
+#endif
+
+again:
+	if (dev->context_flag) {
+		clear_bit(0, &dev->interrupt_flag);
+		return EBUSY;
+	}
+	if (dma->next_buffer) {
+				/* Unsent buffer that was previously
+				   selected, but that couldn't be sent
+				   because the lock could not be obtained
+				   or the DMA engine wasn't ready.  Try
+				   again. */
+		atomic_inc(&dma->total_tried);
+		if (!(retcode = gamma_do_dma(dev, locked))) {
+			atomic_inc(&dma->total_hit);
+			++processed;
+		}
+	} else {
+		do {
+			next = drm_select_queue(dev,
+						gamma_dma_schedule_wrapper);
+			if (next >= 0) {
+				q   = dev->queuelist[next];
+				buf = drm_waitlist_get(&q->waitlist);
+				dma->next_buffer = buf;
+				dma->next_queue	 = q;
+				if (buf && buf->list == DRM_LIST_RECLAIM) {
+					drm_clear_next_buffer(dev);
+					drm_free_buffer(dev, buf);
+				}
+			}
+		} while (next >= 0 && !dma->next_buffer);
+		if (dma->next_buffer) {
+			if (!(retcode = gamma_do_dma(dev, locked))) {
+				++processed;
+			}
+		}
+	}
+
+	if (--expire) {
+		if (missed != atomic_read(&dma->total_missed_sched)) {
+			atomic_inc(&dma->total_lost);
+			if (gamma_dma_is_ready(dev)) goto again;
+		}
+		if (processed && gamma_dma_is_ready(dev)) {
+			atomic_inc(&dma->total_lost);
+			processed = 0;
+			goto again;
+		}
+	}
+	
+	clear_bit(0, &dev->interrupt_flag);
+	
+#if DRM_DMA_HISTOGRAM
+	{
+		struct timespec ts;
+		getnanotime(&ts);
+		timespecsub(&ts, &schedule_start);
+		atomic_inc(&dev->histo.schedule[drm_histogram_slot(&ts)]);
+	}
+#endif
+	return retcode;
+}
+
+static int gamma_dma_priority(drm_device_t *dev, drm_dma_t *d)
+{
+	struct proc	  *p = curproc;
+	unsigned long	  address;
+	unsigned long	  length;
+	int		  must_free = 0;
+	int		  retcode   = 0;
+	int		  i;
+	int		  idx;
+	drm_buf_t	  *buf;
+	drm_buf_t	  *last_buf = NULL;
+	drm_device_dma_t  *dma	    = dev->dma;
+	static int	  never;
+
+				/* Turn off interrupt handling */
+	while (test_and_set_bit(0, &dev->interrupt_flag)) {
+		retcode = tsleep(&never, PZERO|PCATCH, "gamp1", 1);
+		if (retcode)
+			return retcode;
+	}
+	if (!(d->flags & _DRM_DMA_WHILE_LOCKED)) {
+		while (!drm_lock_take(&dev->lock.hw_lock->lock,
+				      DRM_KERNEL_CONTEXT)) {
+			retcode = tsleep(&never, PZERO|PCATCH, "gamp2", 1);
+			if (retcode)
+				return retcode;
+		}
+		++must_free;
+	}
+	atomic_inc(&dma->total_prio);
+
+	for (i = 0; i < d->send_count; i++) {
+		idx = d->send_indices[i];
+		if (idx < 0 || idx >= dma->buf_count) {
+			DRM_ERROR("Index %d (of %d max)\n",
+				  d->send_indices[i], dma->buf_count - 1);
+			continue;
+		}
+		buf = dma->buflist[ idx ];
+		if (buf->pid != p->p_pid) {
+			DRM_ERROR("Process %d using buffer owned by %d\n",
+				  p->p_pid, buf->pid);
+			retcode = EINVAL;
+			goto cleanup;
+		}
+		if (buf->list != DRM_LIST_NONE) {
+			DRM_ERROR("Process %d using %d's buffer on list %d\n",
+				  p->p_pid, buf->pid, buf->list);
+			retcode = EINVAL;
+			goto cleanup;
+		}
+				/* This isn't a race condition on
+				   buf->list, since our concern is the
+				   buffer reclaim during the time the
+				   process closes the /dev/drm? handle, so
+				   it can't also be doing DMA. */
+		buf->list	  = DRM_LIST_PRIO;
+		buf->used	  = d->send_sizes[i];
+		buf->context	  = d->context;
+		buf->while_locked = d->flags & _DRM_DMA_WHILE_LOCKED;
+		address		  = (unsigned long)buf->address;
+		length		  = buf->used;
+		if (!length) {
+			DRM_ERROR("0 length buffer\n");
+		}
+		if (buf->pending) {
+			DRM_ERROR("Sending pending buffer:"
+				  " buffer %d, offset %d\n",
+				  d->send_indices[i], i);
+			retcode = EINVAL;
+			goto cleanup;
+		}
+		if (buf->waiting) {
+			DRM_ERROR("Sending waiting buffer:"
+				  " buffer %d, offset %d\n",
+				  d->send_indices[i], i);
+			retcode = EINVAL;
+			goto cleanup;
+		}
+		buf->pending = 1;
+		
+		if (dev->last_context != buf->context
+		    && !(dev->queuelist[buf->context]->flags
+			 & _DRM_CONTEXT_PRESERVED)) {
+			atomic_inc(&dev->context_wait);
+				/* PRE: dev->last_context != buf->context */
+			drm_context_switch(dev, dev->last_context,
+					   buf->context);
+				/* POST: we will wait for the context
+				   switch and will dispatch on a later call
+				   when dev->last_context == buf->context.
+				   NOTE WE HOLD THE LOCK THROUGHOUT THIS
+				   TIME! */
+			retcode = tsleep(&dev->context_wait,  PZERO|PCATCH,
+				       "gamctx", 0);
+			atomic_dec(&dev->context_wait);
+			if (retcode)
+				goto cleanup;
+			if (dev->last_context != buf->context) {
+				DRM_ERROR("Context mismatch: %d %d\n",
+					  dev->last_context,
+					  buf->context);
+			}
+		}
+
+#if DRM_DMA_HISTOGRAM
+		getnanotime(&buf->time_queued);
+		buf->time_dispatched = buf->time_queued;
+#endif
+		gamma_dma_dispatch(dev, address, length);
+		atomic_add(length, &dma->total_bytes);
+		atomic_inc(&dma->total_dmas);
+		
+		if (last_buf) {
+			drm_free_buffer(dev, last_buf);
+		}
+		last_buf = buf;
+	}
+
+
+cleanup:
+	if (last_buf) {
+		gamma_dma_ready(dev);
+		drm_free_buffer(dev, last_buf);
+	}
+	
+	if (must_free && !dev->context_flag) {
+		if (drm_lock_free(dev, &dev->lock.hw_lock->lock,
+				  DRM_KERNEL_CONTEXT)) {
+			DRM_ERROR("\n");
+		}
+	}
+	clear_bit(0, &dev->interrupt_flag);
+	return retcode;
+}
+
+static int gamma_dma_send_buffers(drm_device_t *dev, drm_dma_t *d)
+{
+	struct proc	  *p = curproc;
+	drm_buf_t	  *last_buf = NULL;
+	int		  retcode   = 0;
+	drm_device_dma_t  *dma	    = dev->dma;
+
+	
+	if ((retcode = drm_dma_enqueue(dev, d))) {
+		return retcode;
+	}
+	
+	gamma_dma_schedule(dev, 0);
+	
+	if (d->flags & _DRM_DMA_BLOCK) {
+		last_buf = dma->buflist[d->send_indices[d->send_count-1]];
+		atomic_inc(&last_buf->dma_wait);
+	}
+
+	if (d->flags & _DRM_DMA_BLOCK) {
+		DRM_DEBUG("%d waiting\n", p->p_pid);
+		for (;;) {
+			retcode = tsleep(&last_buf->dma_wait, PZERO|PCATCH,
+					 "gamdw", 0);
+			if (!last_buf->waiting
+			    && !last_buf->pending)
+				break; /* finished */
+			if (retcode)
+				break;
+		}
+			
+		DRM_DEBUG("%d running\n", p->p_pid);
+		atomic_dec(&last_buf->dma_wait);
+		if (!retcode
+		    || (last_buf->list==DRM_LIST_PEND && !last_buf->pending)) {
+			if (!last_buf->dma_wait) {
+				drm_free_buffer(dev, last_buf);
+			}
+		}
+		if (retcode) {
+			DRM_ERROR("ctx%d w%d p%d c%d i%d l%d %d/%d\n",
+				  d->context,
+				  last_buf->waiting,
+				  last_buf->pending,
+				  DRM_WAITCOUNT(dev, d->context),
+				  last_buf->idx,
+				  last_buf->list,
+				  last_buf->pid,
+				  p->p_pid);
+		}
+	}
+	return retcode;
+}
+
+int gamma_dma(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	drm_device_t	  *dev	    = kdev->si_drv1;
+	drm_device_dma_t  *dma	    = dev->dma;
+	int		  retcode   = 0;
+	drm_dma_t	  d;
+
+	d = *(drm_dma_t *) data;
+	DRM_DEBUG("%d %d: %d send, %d req\n",
+		  p->p_pid, d.context, d.send_count, d.request_count);
+
+	if (d.context == DRM_KERNEL_CONTEXT || d.context >= dev->queue_slots) {
+		DRM_ERROR("Process %d using context %d\n",
+			  p->p_pid, d.context);
+		return EINVAL;
+	}
+	if (d.send_count < 0 || d.send_count > dma->buf_count) {
+		DRM_ERROR("Process %d trying to send %d buffers (of %d max)\n",
+			  p->p_pid, d.send_count, dma->buf_count);
+		return EINVAL;
+	}
+	if (d.request_count < 0 || d.request_count > dma->buf_count) {
+		DRM_ERROR("Process %d trying to get %d buffers (of %d max)\n",
+			  p->p_pid, d.request_count, dma->buf_count);
+		return EINVAL;
+	}
+
+	if (d.send_count) {
+		if (d.flags & _DRM_DMA_PRIORITY)
+			retcode = gamma_dma_priority(dev, &d);
+		else 
+			retcode = gamma_dma_send_buffers(dev, &d);
+	}
+
+	d.granted_count = 0;
+
+	if (!retcode && d.request_count) {
+		retcode = drm_dma_get_buffers(dev, &d);
+	}
+
+	DRM_DEBUG("%d returning, granted = %d\n",
+		  p->p_pid, d.granted_count);
+	*(drm_dma_t *) data = d;
+
+	return retcode;
+}
+
+int gamma_irq_install(drm_device_t *dev, int irq)
+{
+	int rid;
+	int retcode;
+
+	if (!irq)     return EINVAL;
+	
+	lockmgr(&dev->dev_lock, LK_EXCLUSIVE, 0, curproc);
+	if (dev->irq) {
+		lockmgr(&dev->dev_lock, LK_RELEASE, 0, curproc);
+		return EBUSY;
+	}
+	lockmgr(&dev->dev_lock, LK_RELEASE, 0, curproc);
+	
+	DRM_DEBUG("%d\n", irq);
+
+	dev->context_flag     = 0;
+	dev->interrupt_flag   = 0;
+	dev->dma_flag	      = 0;
+	
+	dev->dma->next_buffer = NULL;
+	dev->dma->next_queue  = NULL;
+	dev->dma->this_buffer = NULL;
+
+#if 0
+	dev->tq.next	      = NULL;
+	dev->tq.sync	      = 0;
+	dev->tq.routine	      = gamma_dma_schedule_tq_wrapper;
+	dev->tq.data	      = dev;
+#endif
+				/* Before installing handler */
+	GAMMA_WRITE(GAMMA_GCOMMANDMODE, 0);
+	GAMMA_WRITE(GAMMA_GDMACONTROL, 0);
+	
+				/* Install handler */
+	rid = 0;
+	dev->irq = bus_alloc_resource(dev->device, SYS_RES_IRQ, &rid,
+				      0, ~0, 1, RF_SHAREABLE);
+	if (!dev->irq)
+		return ENOENT;
+
+	retcode = bus_setup_intr(dev->device, dev->irq, INTR_TYPE_TTY,
+				 gamma_dma_service, dev, &dev->irqh);
+	if (retcode) {
+		bus_release_resource(dev->device, SYS_RES_IRQ, 0, dev->irq);
+		dev->irq = 0;
+		return retcode;
+	}
+
+				/* After installing handler */
+	GAMMA_WRITE(GAMMA_GINTENABLE,	    0x2001);
+	GAMMA_WRITE(GAMMA_COMMANDINTENABLE, 0x0008);
+	GAMMA_WRITE(GAMMA_GDELAYTIMER,	   0x39090);
+	
+	return 0;
+}
+
+int gamma_irq_uninstall(drm_device_t *dev)
+{
+	if (!dev->irq)
+		return EINVAL;
+
+	DRM_DEBUG("%ld\n", rman_get_start(dev->irq));
+
+	GAMMA_WRITE(GAMMA_GDELAYTIMER,	    0);
+	GAMMA_WRITE(GAMMA_COMMANDINTENABLE, 0);
+	GAMMA_WRITE(GAMMA_GINTENABLE,	    0);
+
+	bus_teardown_intr(dev->device, dev->irq, dev->irqh);
+	bus_release_resource(dev->device, SYS_RES_IRQ, 0, dev->irq);
+	dev->irq = 0;
+	
+	return 0;
+}
+
+
+int gamma_control(dev_t kdev, u_long cmd, caddr_t data,
+		  int flags, struct proc *p)
+{
+	drm_device_t	*dev	= kdev->si_drv1;
+	drm_control_t	ctl;
+	int		retcode;
+	
+	ctl = *(drm_control_t *) data;
+	
+	switch (ctl.func) {
+	case DRM_INST_HANDLER:
+		if ((retcode = gamma_irq_install(dev, ctl.irq)))
+			return retcode;
+		break;
+	case DRM_UNINST_HANDLER:
+		if ((retcode = gamma_irq_uninstall(dev)))
+			return retcode;
+		break;
+	default:
+		return EINVAL;
+	}
+	return 0;
+}
+
+int gamma_lock(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	drm_device_t	  *dev	  = kdev->si_drv1;
+	int		  ret	= 0;
+	drm_lock_t	  lock;
+	drm_queue_t	  *q;
+#if DRM_DMA_HISTOGRAM
+	struct timespec	  start;
+
+	getnanotime(&start);
+	dev->lck_start = start;
+#endif
+
+	lock = *(drm_lock_t *) data;
+
+	if (lock.context == DRM_KERNEL_CONTEXT) {
+		DRM_ERROR("Process %d using kernel context %d\n",
+			  p->p_pid, lock.context);
+		return EINVAL;
+	}
+
+	DRM_DEBUG("%d (pid %d) requests lock (0x%08x), flags = 0x%08x\n",
+		  lock.context, p->p_pid, dev->lock.hw_lock->lock,
+		  lock.flags);
+
+	if (lock.context < 0 || lock.context >= dev->queue_count)
+		return EINVAL;
+	q = dev->queuelist[lock.context];
+	
+	ret = drm_flush_block_and_flush(dev, lock.context, lock.flags);
+
+	if (!ret) {
+		if (_DRM_LOCKING_CONTEXT(dev->lock.hw_lock->lock)
+		    != lock.context) {
+			long j = ticks - dev->lock.lock_time;
+
+			if (j > 0 && j <= DRM_LOCK_SLICE) {
+				/* Can't take lock if we just had it and
+				   there is contention. */
+				static int never;
+				ret = tsleep(&never, PZERO|PCATCH,
+					     "gaml1", j);
+				if (ret)
+					return ret;
+			}
+		}
+		atomic_inc(&dev->lock.lock_queue);
+		for (;;) {
+			if (!dev->lock.hw_lock) {
+				/* Device has been unregistered */
+				ret = EINTR;
+				break;
+			}
+			if (drm_lock_take(&dev->lock.hw_lock->lock,
+					  lock.context)) {
+				dev->lock.pid	    = p->p_pid;
+				dev->lock.lock_time = ticks;
+				atomic_inc(&dev->total_locks);
+				atomic_inc(&q->total_locks);
+				break;	/* Got lock */
+			}
+			
+				/* Contention */
+			atomic_inc(&dev->total_sleeps);
+			ret = tsleep(&dev->lock.lock_queue, PZERO|PCATCH,
+					 "gaml2", 0);
+			if (ret)
+				break;
+		}
+		atomic_dec(&dev->lock.lock_queue);
+	}
+
+	drm_flush_unblock(dev, lock.context, lock.flags); /* cleanup phase */
+	
+	if (!ret) {
+		if (lock.flags & _DRM_LOCK_READY)
+			gamma_dma_ready(dev);
+		if (lock.flags & _DRM_LOCK_QUIESCENT)
+			gamma_dma_quiescent(dev);
+	}
+	DRM_DEBUG("%d %s\n", lock.context, ret ? "interrupted" : "has lock");
+
+#if DRM_DMA_HISTOGRAM
+	{
+		struct timespec ts;
+		getnanotime(&ts);
+		timespecsub(&ts, &start);
+		atomic_inc(&dev->histo.lacq[drm_histogram_slot(&ts)]);
+	}
+#endif
+	
+	return ret;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/gamma/gamma_drv.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/gamma/gamma_drv.c
new file mode 100644
index 000000000..fe71737a5
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/gamma/gamma_drv.c
@@ -0,0 +1,574 @@
+/* gamma.c -- 3dlabs GMX 2000 driver -*- c -*-
+ * Created: Mon Jan  4 08:58:31 1999 by faith@precisioninsight.com
+ * Revised: Tue Oct 12 08:51:36 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drv.c,v 1.17 1999/08/30 13:05:00 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drv.c,v 1.1 1999/09/25 14:38:00 dawes Exp $
+ *
+ */
+
+#include "drmP.h"
+#include "gamma_drv.h"
+
+#include <pci/pcivar.h>
+
+MODULE_DEPEND(gamma, drm, 1, 1, 1);
+
+static int gamma_init(device_t nbdev);
+static void gamma_cleanup(device_t nbdev);
+
+static int gamma_probe(device_t dev)
+{
+	const char *s = 0;
+
+	switch (pci_get_devid(dev)) {
+	case 0x00083d3d:
+		s = "3D Labs Gamma graphics accelerator";
+		break;
+	}
+
+	if (s) {
+		device_set_desc(dev, s);
+		return 0;
+	}
+
+	return ENXIO;
+}
+
+static int gamma_attach(device_t dev)
+{
+	gamma_init(dev);
+	return 0;
+}
+
+static int gamma_detach(device_t dev)
+{
+	gamma_cleanup(dev);
+	return 0;
+}
+
+static device_method_t gamma_methods[] = {
+	/* Device interface */
+	DEVMETHOD(device_probe,		gamma_probe),
+	DEVMETHOD(device_attach,	gamma_attach),
+	DEVMETHOD(device_detach,	gamma_detach),
+
+	{ 0, 0 }
+};
+
+static driver_t gamma_driver = {
+	"drm",
+	gamma_methods,
+	sizeof(drm_device_t),
+};
+
+static devclass_t gamma_devclass;
+#define GAMMA_SOFTC(unit) \
+	((drm_device_t *) devclass_get_softc(gamma_devclass, unit))
+
+DRIVER_MODULE(if_gamma, pci, gamma_driver, gamma_devclass, 0, 0);
+
+#define GAMMA_NAME	 "gamma"
+#define GAMMA_DESC	 "3dlabs GMX 2000"
+#define GAMMA_DATE	 "19990830"
+#define GAMMA_MAJOR	 0
+#define GAMMA_MINOR	 0
+#define GAMMA_PATCHLEVEL 5
+
+#define CDEV_MAJOR	200
+
+static struct cdevsw gamma_cdevsw = {
+	/* open */	gamma_open,
+	/* close */	gamma_close,
+	/* read */	drm_read,
+	/* write */	drm_write,
+	/* ioctl */	gamma_ioctl,
+	/* poll */	nopoll,
+	/* mmap */	drm_mmap,
+	/* strategy */	nostrategy,
+	/* name */	"gamma",
+	/* maj */	CDEV_MAJOR,
+	/* dump */	nodump,
+	/* psize */	nopsize,
+	/* flags */	D_TTY | D_TRACKCLOSE,
+	/* bmaj */	-1
+};
+
+static drm_ioctl_desc_t	      gamma_ioctls[] = {
+	[DRM_IOCTL_NR(DRM_IOCTL_VERSION)]    = { gamma_version,	  0, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_GET_UNIQUE)] = { drm_getunique,	  0, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_GET_MAGIC)]  = { drm_getmagic,	  0, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_IRQ_BUSID)]  = { drm_irq_busid,	  0, 1 },
+
+	[DRM_IOCTL_NR(DRM_IOCTL_SET_UNIQUE)] = { drm_setunique,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_BLOCK)]	     = { drm_block,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_UNBLOCK)]    = { drm_unblock,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_CONTROL)]    = { gamma_control,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_AUTH_MAGIC)] = { drm_authmagic,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_ADD_MAP)]    = { drm_addmap,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_ADD_BUFS)]   = { drm_addbufs,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_MARK_BUFS)]  = { drm_markbufs,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_INFO_BUFS)]  = { drm_infobufs,	  1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_MAP_BUFS)]   = { drm_mapbufs,	  1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_FREE_BUFS)]  = { drm_freebufs,	  1, 0 },
+	
+	[DRM_IOCTL_NR(DRM_IOCTL_ADD_CTX)]    = { drm_addctx,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_RM_CTX)]     = { drm_rmctx,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_MOD_CTX)]    = { drm_modctx,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_GET_CTX)]    = { drm_getctx,	  1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_SWITCH_CTX)] = { drm_switchctx,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_NEW_CTX)]    = { drm_newctx,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_RES_CTX)]    = { drm_resctx,	  1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_ADD_DRAW)]   = { drm_adddraw,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_RM_DRAW)]    = { drm_rmdraw,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_DMA)]	     = { gamma_dma,	  1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_LOCK)]	     = { gamma_lock,	  1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_UNLOCK)]     = { gamma_unlock,	  1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_FINISH)]     = { drm_finish,	  1, 0 },
+};
+#define GAMMA_IOCTL_COUNT DRM_ARRAY_SIZE(gamma_ioctls)
+
+static int gamma_setup(drm_device_t *dev)
+{
+	int i;
+	
+	device_busy(dev->device);
+
+	atomic_set(&dev->ioctl_count, 0);
+	atomic_set(&dev->vma_count, 0);
+	dev->buf_use	  = 0;
+	atomic_set(&dev->buf_alloc, 0);
+
+	drm_dma_setup(dev);
+
+	atomic_set(&dev->total_open, 0);
+	atomic_set(&dev->total_close, 0);
+	atomic_set(&dev->total_ioctl, 0);
+	atomic_set(&dev->total_irq, 0);
+	atomic_set(&dev->total_ctx, 0);
+	atomic_set(&dev->total_locks, 0);
+	atomic_set(&dev->total_unlocks, 0);
+	atomic_set(&dev->total_contends, 0);
+	atomic_set(&dev->total_sleeps, 0);
+
+	for (i = 0; i < DRM_HASH_SIZE; i++) {
+		dev->magiclist[i].head = NULL;
+		dev->magiclist[i].tail = NULL;
+	}
+	dev->maplist	    = NULL;
+	dev->map_count	    = 0;
+	dev->vmalist	    = NULL;
+	dev->lock.hw_lock   = NULL;
+	dev->lock.lock_queue = 0;
+	dev->queue_count    = 0;
+	dev->queue_reserved = 0;
+	dev->queue_slots    = 0;
+	dev->queuelist	    = NULL;
+	dev->irq	    = 0;
+	dev->context_flag   = 0;
+	dev->interrupt_flag = 0;
+	dev->dma_flag	    = 0;
+	dev->last_context   = 0;
+	dev->last_switch    = 0;
+	dev->last_checked   = 0;
+	callout_init(&dev->timer);
+	dev->context_wait   = 0;
+#if DRM_DMA_HISTO
+	memset(&dev->histo, 0, sizeof(dev->histo));
+#endif
+	timespecclear(&dev->ctx_start);
+	timespecclear(&dev->lck_start);
+	
+	dev->buf_rp	  = dev->buf;
+	dev->buf_wp	  = dev->buf;
+	dev->buf_end	  = dev->buf + DRM_BSZ;
+	dev->buf_sigio	  = NULL;
+			
+	DRM_DEBUG("\n");
+			
+	/* The kernel's context could be created here, but is now created
+	   in drm_dma_enqueue.	This is more resource-efficient for
+	   hardware that does not do DMA, but may mean that
+	   drm_select_queue fails between the time the interrupt is
+	   initialized and the time the queues are initialized. */
+			
+	return 0;
+}
+
+
+static int
+gamma_takedown(drm_device_t *dev)
+{
+	int		  i;
+	drm_magic_entry_t *pt, *next;
+	drm_map_t	  *map;
+	drm_vma_entry_t	  *vma, *vma_next;
+
+	DRM_DEBUG("\n");
+
+	if (dev->irq) gamma_irq_uninstall(dev);
+	
+	lockmgr(&dev->dev_lock, LK_EXCLUSIVE, 0, curproc);
+	callout_stop(&dev->timer);
+
+	if (dev->devname) {
+		drm_free(dev->devname, strlen(dev->devname)+1, DRM_MEM_DRIVER);
+		dev->devname = NULL;
+	}
+	
+	if (dev->unique) {
+		drm_free(dev->unique, strlen(dev->unique)+1, DRM_MEM_DRIVER);
+		dev->unique = NULL;
+		dev->unique_len = 0;
+	}
+				/* Clear pid list */
+	for (i = 0; i < DRM_HASH_SIZE; i++) {
+		for (pt = dev->magiclist[i].head; pt; pt = next) {
+			next = pt->next;
+			drm_free(pt, sizeof(*pt), DRM_MEM_MAGIC);
+		}
+		dev->magiclist[i].head = dev->magiclist[i].tail = NULL;
+	}
+	
+				/* Clear vma list (only built for debugging) */
+	if (dev->vmalist) {
+		for (vma = dev->vmalist; vma; vma = vma_next) {
+			vma_next = vma->next;
+			drm_free(vma, sizeof(*vma), DRM_MEM_VMAS);
+		}
+		dev->vmalist = NULL;
+	}
+	
+				/* Clear map area and mtrr information */
+	if (dev->maplist) {
+		for (i = 0; i < dev->map_count; i++) {
+			map = dev->maplist[i];
+			switch (map->type) {
+			case _DRM_REGISTERS:
+			case _DRM_FRAME_BUFFER:
+#ifdef CONFIG_MTRR
+				if (map->mtrr >= 0) {
+					int retcode;
+					retcode = mtrr_del(map->mtrr,
+							   map->offset,
+							   map->size);
+					DRM_DEBUG("mtrr_del = %d\n", retcode);
+				}
+#endif
+				drm_ioremapfree(map->handle, map->size);
+				break;
+			case _DRM_SHM:
+				drm_free_pages((unsigned long)map->handle,
+					       drm_order(map->size)
+					       - PAGE_SHIFT,
+					       DRM_MEM_SAREA);
+				break;
+			}
+			drm_free(map, sizeof(*map), DRM_MEM_MAPS);
+		}
+		drm_free(dev->maplist,
+			 dev->map_count * sizeof(*dev->maplist),
+			 DRM_MEM_MAPS);
+		dev->maplist   = NULL;
+		dev->map_count = 0;
+	}
+	
+	if (dev->queuelist) {
+		for (i = 0; i < dev->queue_count; i++) {
+			drm_waitlist_destroy(&dev->queuelist[i]->waitlist);
+			if (dev->queuelist[i]) {
+				drm_free(dev->queuelist[i],
+					 sizeof(*dev->queuelist[0]),
+					 DRM_MEM_QUEUES);
+				dev->queuelist[i] = NULL;
+			}
+		}
+		drm_free(dev->queuelist,
+			 dev->queue_slots * sizeof(*dev->queuelist),
+			 DRM_MEM_QUEUES);
+		dev->queuelist	 = NULL;
+	}
+
+	drm_dma_takedown(dev);
+
+	dev->queue_count     = 0;
+	if (dev->lock.hw_lock) {
+		dev->lock.hw_lock    = NULL; /* SHM removed */
+		dev->lock.pid	     = 0;
+		wakeup(&dev->lock.lock_queue);
+	}
+	lockmgr(&dev->dev_lock, LK_RELEASE, 0, curproc);
+	
+	device_unbusy(dev->device);
+
+	return 0;
+}
+
+/* gamma_init is called via gamma_attach at module load time */
+
+static int
+gamma_init(device_t nbdev)
+{
+	drm_device_t	      *dev = device_get_softc(nbdev);
+
+	DRM_DEBUG("\n");
+
+	memset((void *)dev, 0, sizeof(*dev));
+	simple_lock_init(&dev->count_lock);
+	lockinit(&dev->dev_lock, PZERO, "drmlk", 0, 0);
+	
+#if 0				/* XXX use getenv I guess */
+	drm_parse_options(gamma);
+#endif
+
+#if 0
+	if ((retcode = misc_register(&gamma_misc))) {
+		DRM_ERROR("Cannot register \"%s\"\n", GAMMA_NAME);
+		return retcode;
+	}
+#endif
+	dev->device = nbdev;
+	dev->devnode = make_dev(&gamma_cdevsw,
+				device_get_unit(nbdev),
+				DRM_DEV_UID,
+				DRM_DEV_GID,
+				DRM_DEV_MODE,
+				GAMMA_NAME);
+	dev->name   = GAMMA_NAME;
+
+	drm_mem_init();
+	drm_sysctl_init(dev);
+
+	DRM_INFO("Initialized %s %d.%d.%d %s on minor %d\n",
+		 GAMMA_NAME,
+		 GAMMA_MAJOR,
+		 GAMMA_MINOR,
+		 GAMMA_PATCHLEVEL,
+		 GAMMA_DATE,
+		 device_get_unit(nbdev));
+
+	return 0;
+}
+
+/* gamma_cleanup is called via gamma_detach at module unload time. */
+
+static void
+gamma_cleanup(device_t nbdev)
+{
+	drm_device_t	      *dev = device_get_softc(nbdev);
+
+	DRM_DEBUG("\n");
+	
+	drm_sysctl_cleanup(dev);
+#if 0
+	if (misc_deregister(&gamma_misc)) {
+		DRM_ERROR("Cannot unload module\n");
+	} else {
+		DRM_INFO("Module unloaded\n");
+	}
+#endif
+	device_busy(dev->device);
+	gamma_takedown(dev);
+}
+
+SYSUNINIT(gamma_cleanup, SI_SUB_DRIVERS, SI_ORDER_ANY, gamma_cleanup, 0);
+
+#if 0
+int gamma_version(struct inode *inode, struct file *filp, unsigned int cmd,
+		  unsigned long arg)
+{
+	drm_version_t version;
+	int	      len;
+
+	copy_from_user_ret(&version,
+			   (drm_version_t *)arg,
+			   sizeof(version),
+			   -EFAULT);
+
+#define DRM_COPY(name,value)				     \
+	len = strlen(value);				     \
+	if (len > name##_len) len = name##_len;		     \
+	name##_len = strlen(value);			     \
+	if (len && name) {				     \
+		copy_to_user_ret(name, value, len, -EFAULT); \
+	}
+
+	version.version_major	   = GAMMA_MAJOR;
+	version.version_minor	   = GAMMA_MINOR;
+	version.version_patchlevel = GAMMA_PATCHLEVEL;
+
+	DRM_COPY(version.name, GAMMA_NAME);
+	DRM_COPY(version.date, GAMMA_DATE);
+	DRM_COPY(version.desc, GAMMA_DESC);
+
+	copy_to_user_ret((drm_version_t *)arg,
+			 &version,
+			 sizeof(version),
+			 -EFAULT);
+	return 0;
+}
+#endif
+
+int
+gamma_open(dev_t kdev, int flags, int fmt, struct proc *p)
+{
+	drm_device_t  *dev    = GAMMA_SOFTC(minor(kdev));
+	int	      retcode = 0;
+	
+	DRM_DEBUG("open_count = %d\n", dev->open_count);
+
+	device_busy(dev->device);
+	if (!(retcode = drm_open_helper(kdev, flags, fmt, p, dev))) {
+		atomic_inc(&dev->total_open);
+		simple_lock(&dev->count_lock);
+		if (!dev->open_count++) {
+			simple_unlock(&dev->count_lock);
+			retcode = gamma_setup(dev);
+		}
+		simple_unlock(&dev->count_lock);
+	}
+	device_unbusy(dev->device);
+
+	return retcode;
+}
+
+int
+gamma_close(dev_t kdev, int flags, int fmt, struct proc *p)
+{
+	drm_device_t	 *dev	 = kdev->si_drv1;
+	int	      retcode = 0;
+
+	DRM_DEBUG("open_count = %d\n", dev->open_count);
+	if (!(retcode = drm_close(kdev, flags, fmt, p))) {
+		atomic_inc(&dev->total_close);
+		simple_lock(&dev->count_lock);
+		if (!--dev->open_count) {
+			if (atomic_read(&dev->ioctl_count) || dev->blocked) {
+				DRM_ERROR("Device busy: %d %d\n",
+					  atomic_read(&dev->ioctl_count),
+					  dev->blocked);
+				simple_unlock(&dev->count_lock);
+				return EBUSY;
+			}
+			simple_unlock(&dev->count_lock);
+			return gamma_takedown(dev);
+		}
+		simple_unlock(&dev->count_lock);
+	}
+	return retcode;
+}
+
+/* drm_ioctl is called whenever a process performs an ioctl on /dev/drm. */
+
+int
+gamma_ioctl(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	int		 nr	 = DRM_IOCTL_NR(cmd);
+	drm_device_t	 *dev	 = kdev->si_drv1;
+	drm_file_t	 *priv;
+	int		 retcode = 0;
+	drm_ioctl_desc_t *ioctl;
+	d_ioctl_t	 *func;
+
+	priv = drm_find_file_by_proc(dev, p);
+	if (!priv) {
+		DRM_DEBUG("can't find authenticator\n");
+		return EINVAL;
+	}
+
+	atomic_inc(&dev->ioctl_count);
+	atomic_inc(&dev->total_ioctl);
+	++priv->ioctl_count;
+	
+	DRM_DEBUG("pid = %d, cmd = 0x%02lx, nr = 0x%02x, auth = %d\n",
+		  p->p_pid, cmd, nr, priv->authenticated);
+
+	switch (cmd) {
+	case FIOSETOWN:
+		return fsetown(*(int *)data, &dev->buf_sigio);
+
+	case FIOGETOWN:
+		*(int *) data = fgetown(dev->buf_sigio);
+		return 0;
+	}
+
+	if (nr >= GAMMA_IOCTL_COUNT) {
+		retcode = EINVAL;
+	} else {
+		ioctl	  = &gamma_ioctls[nr];
+		func	  = ioctl->func;
+
+		if (!func) {
+			DRM_DEBUG("no function\n");
+			retcode = EINVAL;
+		} else if ((ioctl->root_only && suser(p))
+			    || (ioctl->auth_needed && !priv->authenticated)) {
+			retcode = EACCES;
+		} else {
+			retcode = (func)(kdev, cmd, data, flags, p);
+		}
+	}
+	
+	atomic_dec(&dev->ioctl_count);
+	return retcode;
+}
+
+int gamma_unlock(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	drm_device_t	  *dev	  = kdev->si_drv1;
+	drm_lock_t	  *lockp  = (drm_lock_t *) data;
+
+	if (lockp->context == DRM_KERNEL_CONTEXT) {
+		DRM_ERROR("Process %d using kernel context %d\n",
+			  p->p_pid, lockp->context);
+		return -EINVAL;
+	}
+
+	DRM_DEBUG("%d frees lock (%d holds)\n",
+		  lockp->context,
+		  _DRM_LOCKING_CONTEXT(dev->lock.hw_lock->lock));
+	atomic_inc(&dev->total_unlocks);
+	if (_DRM_LOCK_IS_CONT(dev->lock.hw_lock->lock))
+		atomic_inc(&dev->total_contends);
+	drm_lock_transfer(dev, &dev->lock.hw_lock->lock, DRM_KERNEL_CONTEXT);
+	gamma_dma_schedule(dev, 1);
+	if (!dev->context_flag) {
+		if (drm_lock_free(dev, &dev->lock.hw_lock->lock,
+				  DRM_KERNEL_CONTEXT)) {
+			DRM_ERROR("\n");
+		}
+	}
+
+#if DRM_DMA_HISTOGRAM
+	{
+	    struct timespec ts;
+	    getnanotime(&ts);
+	    timespecsub(&ts, &dev->lck_start);
+	    atomic_inc(&dev->histo.lhld[drm_histogram_slot(&ts)]);
+	}
+#endif
+	
+	return 0;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/gamma/gamma_drv.h b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/gamma/gamma_drv.h
new file mode 100644
index 000000000..e48882207
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/gamma/gamma_drv.h
@@ -0,0 +1,50 @@
+/* gamma_drv.h -- Private header for 3dlabs GMX 2000 driver -*- c -*-
+ * Created: Mon Jan  4 10:05:05 1999 by faith@precisioninsight.com
+ * Revised: Fri Aug 20 09:24:27 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drv.h,v 1.4 1999/08/30 13:05:00 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drv.h,v 1.1 1999/09/25 14:38:00 dawes Exp $
+ * 
+ */
+
+#ifndef _GAMMA_DRV_H_
+#define _GAMMA_DRV_H_
+
+				/* gamma_drv.c */
+extern d_open_t gamma_open;
+extern d_close_t gamma_close;
+extern d_ioctl_t gamma_ioctl;
+extern d_ioctl_t gamma_version;
+extern d_ioctl_t gamma_dma;
+extern d_ioctl_t gamma_lock;
+extern d_ioctl_t gamma_unlock;
+extern d_ioctl_t gamma_control;
+
+				/* gamma_dma.c */
+extern int  gamma_dma_schedule(drm_device_t *dev, int locked);
+extern int  gamma_irq_install(drm_device_t *dev, int irq);
+extern int  gamma_irq_uninstall(drm_device_t *dev);
+
+#endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/i810_drm.h b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/i810_drm.h
new file mode 100644
index 000000000..4c8e09f6a
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/i810_drm.h
@@ -0,0 +1,188 @@
+#ifndef _I810_DRM_H_
+#define _I810_DRM_H_
+
+/* WARNING: These defines must be the same as what the Xserver uses.
+ * if you change them, you must change the defines in the Xserver.
+ */
+
+#ifndef _I810_DEFINES_
+#define _I810_DEFINES_
+
+#define I810_DMA_BUF_ORDER		12
+#define I810_DMA_BUF_SZ 		(1<<I810_DMA_BUF_ORDER)
+#define I810_DMA_BUF_NR 		256
+#define I810_NR_SAREA_CLIPRECTS 	8
+
+/* Each region is a minimum of 64k, and there are at most 64 of them.
+ */
+#define I810_NR_TEX_REGIONS 64
+#define I810_LOG_MIN_TEX_REGION_SIZE 16
+#endif
+
+#define I810_UPLOAD_TEX0IMAGE  0x1 /* handled clientside */
+#define I810_UPLOAD_TEX1IMAGE  0x2 /* handled clientside */
+#define I810_UPLOAD_CTX        0x4
+#define I810_UPLOAD_BUFFERS    0x8
+#define I810_UPLOAD_TEX0       0x10
+#define I810_UPLOAD_TEX1       0x20
+#define I810_UPLOAD_CLIPRECTS  0x40
+
+
+/* Indices into buf.Setup where various bits of state are mirrored per
+ * context and per buffer.  These can be fired at the card as a unit,
+ * or in a piecewise fashion as required.
+ */
+
+/* Destbuffer state 
+ *    - backbuffer linear offset and pitch -- invarient in the current dri
+ *    - zbuffer linear offset and pitch -- also invarient
+ *    - drawing origin in back and depth buffers.
+ *
+ * Keep the depth/back buffer state here to acommodate private buffers
+ * in the future.
+ */
+#define I810_DESTREG_DI0  0	/* CMD_OP_DESTBUFFER_INFO (2 dwords) */
+#define I810_DESTREG_DI1  1
+#define I810_DESTREG_DV0  2	/* GFX_OP_DESTBUFFER_VARS (2 dwords) */
+#define I810_DESTREG_DV1  3
+#define I810_DESTREG_DR0  4	/* GFX_OP_DRAWRECT_INFO (4 dwords) */
+#define I810_DESTREG_DR1  5
+#define I810_DESTREG_DR2  6
+#define I810_DESTREG_DR3  7
+#define I810_DESTREG_DR4  8
+#define I810_DEST_SETUP_SIZE 10
+
+/* Context state
+ */
+#define I810_CTXREG_CF0   0	/* GFX_OP_COLOR_FACTOR */
+#define I810_CTXREG_CF1   1	
+#define I810_CTXREG_ST0   2     /* GFX_OP_STIPPLE */
+#define I810_CTXREG_ST1   3
+#define I810_CTXREG_VF    4	/* GFX_OP_VERTEX_FMT */
+#define I810_CTXREG_MT    5	/* GFX_OP_MAP_TEXELS */
+#define I810_CTXREG_MC0   6	/* GFX_OP_MAP_COLOR_STAGES - stage 0 */
+#define I810_CTXREG_MC1   7     /* GFX_OP_MAP_COLOR_STAGES - stage 1 */
+#define I810_CTXREG_MC2   8	/* GFX_OP_MAP_COLOR_STAGES - stage 2 */
+#define I810_CTXREG_MA0   9	/* GFX_OP_MAP_ALPHA_STAGES - stage 0 */
+#define I810_CTXREG_MA1   10	/* GFX_OP_MAP_ALPHA_STAGES - stage 1 */
+#define I810_CTXREG_MA2   11	/* GFX_OP_MAP_ALPHA_STAGES - stage 2 */
+#define I810_CTXREG_SDM   12	/* GFX_OP_SRC_DEST_MONO */
+#define I810_CTXREG_FOG   13	/* GFX_OP_FOG_COLOR */
+#define I810_CTXREG_B1    14	/* GFX_OP_BOOL_1 */
+#define I810_CTXREG_B2    15	/* GFX_OP_BOOL_2 */
+#define I810_CTXREG_LCS   16	/* GFX_OP_LINEWIDTH_CULL_SHADE_MODE */
+#define I810_CTXREG_PV    17	/* GFX_OP_PV_RULE -- Invarient! */
+#define I810_CTXREG_ZA    18	/* GFX_OP_ZBIAS_ALPHAFUNC */
+#define I810_CTXREG_AA    19	/* GFX_OP_ANTIALIAS */
+#define I810_CTX_SETUP_SIZE 20 
+
+/* Texture state (per tex unit)
+ */
+#define I810_TEXREG_MI0  0	/* GFX_OP_MAP_INFO (4 dwords) */
+#define I810_TEXREG_MI1  1	
+#define I810_TEXREG_MI2  2	
+#define I810_TEXREG_MI3  3	
+#define I810_TEXREG_MF   4	/* GFX_OP_MAP_FILTER */
+#define I810_TEXREG_MLC  5	/* GFX_OP_MAP_LOD_CTL */
+#define I810_TEXREG_MLL  6	/* GFX_OP_MAP_LOD_LIMITS */
+#define I810_TEXREG_MCS  7	/* GFX_OP_MAP_COORD_SETS ??? */
+#define I810_TEX_SETUP_SIZE 8
+
+#define I810_FRONT   0x1
+#define I810_BACK    0x2
+#define I810_DEPTH   0x4
+
+
+typedef struct _drm_i810_init {
+	enum {
+		I810_INIT_DMA = 0x01,
+		I810_CLEANUP_DMA = 0x02
+	} func;
+	int ring_map_idx;
+	int buffer_map_idx;
+	int sarea_priv_offset;
+	unsigned int ring_start;
+	unsigned int ring_end;
+	unsigned int ring_size;
+	unsigned int front_offset;
+	unsigned int back_offset;
+	unsigned int depth_offset;
+	unsigned int w;
+	unsigned int h;
+	unsigned int pitch;
+	unsigned int pitch_bits; 
+} drm_i810_init_t;
+
+/* Warning: If you change the SAREA structure you must change the Xserver
+ * structure as well */
+
+typedef struct _drm_i810_tex_region {
+	unsigned char next, prev; /* indices to form a circular LRU  */
+	unsigned char in_use;	/* owned by a client, or free? */
+	int age;		/* tracked by clients to update local LRU's */
+} drm_i810_tex_region_t;
+
+typedef struct _drm_i810_sarea {
+   	unsigned int ContextState[I810_CTX_SETUP_SIZE];
+   	unsigned int BufferState[I810_DEST_SETUP_SIZE];
+   	unsigned int TexState[2][I810_TEX_SETUP_SIZE];
+   	unsigned int dirty;
+
+	unsigned int nbox;
+	drm_clip_rect_t boxes[I810_NR_SAREA_CLIPRECTS];
+
+	/* Maintain an LRU of contiguous regions of texture space.  If
+	 * you think you own a region of texture memory, and it has an
+	 * age different to the one you set, then you are mistaken and
+	 * it has been stolen by another client.  If global texAge
+	 * hasn't changed, there is no need to walk the list.
+	 *
+	 * These regions can be used as a proxy for the fine-grained
+	 * texture information of other clients - by maintaining them
+	 * in the same lru which is used to age their own textures,
+	 * clients have an approximate lru for the whole of global
+	 * texture space, and can make informed decisions as to which
+	 * areas to kick out.  There is no need to choose whether to
+	 * kick out your own texture or someone else's - simply eject
+	 * them all in LRU order.  
+	 */
+   
+	drm_i810_tex_region_t texList[I810_NR_TEX_REGIONS+1]; 
+				/* Last elt is sentinal */
+        int texAge;		/* last time texture was uploaded */
+        int last_enqueue;	/* last time a buffer was enqueued */
+	int last_dispatch;	/* age of the most recently dispatched buffer */
+	int last_quiescent;     /*  */
+	int ctxOwner;		/* last context to upload state */
+
+	int vertex_prim;
+
+} drm_i810_sarea_t;
+
+typedef struct _drm_i810_clear {
+	int clear_color;
+	int clear_depth;
+	int flags;
+} drm_i810_clear_t;
+
+
+
+/* These may be placeholders if we have more cliprects than
+ * I810_NR_SAREA_CLIPRECTS.  In that case, the client sets discard to
+ * false, indicating that the buffer will be dispatched again with a
+ * new set of cliprects.
+ */
+typedef struct _drm_i810_vertex {
+   	int idx;		/* buffer index */
+	int used;		/* nr bytes in use */
+	int discard;		/* client is finished with the buffer? */
+} drm_i810_vertex_t;
+
+typedef struct drm_i810_dma {
+	void *virtual;
+	int request_idx;
+	int request_size;
+	int granted;
+} drm_i810_dma_t;
+
+#endif /* _I810_DRM_H_ */
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/mga_drm.h b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/mga_drm.h
new file mode 100644
index 000000000..8bfa2b974
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/mga_drm.h
@@ -0,0 +1,269 @@
+/* mga_drm.h -- Public header for the Matrox g200/g400 driver
+ * Created: Tue Jan 25 01:50:01 1999 by jhartmann@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Jeff Hartmann <jhartmann@precisioninsight.com>
+ *          Keith Whitwell <keithw@precisioninsight.com>
+ *
+ * $XFree86$
+ */
+
+#ifndef _MGA_DRM_H_
+#define _MGA_DRM_H_
+
+/* WARNING: If you change any of these defines, make sure to change the
+ * defines in the Xserver file (xf86drmMga.h)
+ */
+#ifndef _MGA_DEFINES_
+#define _MGA_DEFINES_
+
+#define MGA_F  0x1		/* fog */
+#define MGA_A  0x2		/* alpha */
+#define MGA_S  0x4		/* specular */
+#define MGA_T2 0x8		/* multitexture */
+
+#define MGA_WARP_TGZ            0
+#define MGA_WARP_TGZF           (MGA_F)
+#define MGA_WARP_TGZA           (MGA_A)
+#define MGA_WARP_TGZAF          (MGA_F|MGA_A)
+#define MGA_WARP_TGZS           (MGA_S)
+#define MGA_WARP_TGZSF          (MGA_S|MGA_F)
+#define MGA_WARP_TGZSA          (MGA_S|MGA_A)
+#define MGA_WARP_TGZSAF         (MGA_S|MGA_F|MGA_A)
+#define MGA_WARP_T2GZ           (MGA_T2)
+#define MGA_WARP_T2GZF          (MGA_T2|MGA_F)
+#define MGA_WARP_T2GZA          (MGA_T2|MGA_A)
+#define MGA_WARP_T2GZAF         (MGA_T2|MGA_A|MGA_F)
+#define MGA_WARP_T2GZS          (MGA_T2|MGA_S)
+#define MGA_WARP_T2GZSF         (MGA_T2|MGA_S|MGA_F)
+#define MGA_WARP_T2GZSA         (MGA_T2|MGA_S|MGA_A)
+#define MGA_WARP_T2GZSAF        (MGA_T2|MGA_S|MGA_F|MGA_A)
+
+#define MGA_MAX_G400_PIPES 16
+#define MGA_MAX_G200_PIPES  8	/* no multitex */
+#define MGA_MAX_WARP_PIPES MGA_MAX_G400_PIPES
+
+#define MGA_CARD_TYPE_G200 1
+#define MGA_CARD_TYPE_G400 2
+
+#define MGA_FRONT   0x1
+#define MGA_BACK    0x2
+#define MGA_DEPTH   0x4
+
+/* 3d state excluding texture units:
+ */
+#define MGA_CTXREG_DSTORG   0	/* validated */
+#define MGA_CTXREG_MACCESS  1	
+#define MGA_CTXREG_PLNWT    2	
+#define MGA_CTXREG_DWGCTL    3	
+#define MGA_CTXREG_ALPHACTRL 4
+#define MGA_CTXREG_FOGCOLOR  5
+#define MGA_CTXREG_WFLAG     6
+#define MGA_CTXREG_TDUAL0    7
+#define MGA_CTXREG_TDUAL1    8
+#define MGA_CTXREG_FCOL      9
+#define MGA_CTX_SETUP_SIZE   10
+
+/* 2d state
+ */
+#define MGA_2DREG_PITCH 	0
+#define MGA_2D_SETUP_SIZE 	1
+
+/* Each texture unit has a state:
+ */
+#define MGA_TEXREG_CTL        0
+#define MGA_TEXREG_CTL2       1
+#define MGA_TEXREG_FILTER     2
+#define MGA_TEXREG_BORDERCOL  3
+#define MGA_TEXREG_ORG        4 /* validated */
+#define MGA_TEXREG_ORG1       5
+#define MGA_TEXREG_ORG2       6
+#define MGA_TEXREG_ORG3       7
+#define MGA_TEXREG_ORG4       8
+#define MGA_TEXREG_WIDTH      9
+#define MGA_TEXREG_HEIGHT     10
+#define MGA_TEX_SETUP_SIZE    11
+
+/* What needs to be changed for the current vertex dma buffer?
+ */
+#define MGA_UPLOAD_CTX        0x1
+#define MGA_UPLOAD_TEX0       0x2
+#define MGA_UPLOAD_TEX1       0x4
+#define MGA_UPLOAD_PIPE       0x8
+#define MGA_UPLOAD_TEX0IMAGE  0x10 /* handled client-side */
+#define MGA_UPLOAD_TEX1IMAGE  0x20 /* handled client-side */
+#define MGA_UPLOAD_2D 	      0x40
+#define MGA_WAIT_AGE          0x80 /* handled client-side */
+#define MGA_UPLOAD_CLIPRECTS  0x100 /* handled client-side */
+#define MGA_DMA_FLUSH	      0x200 /* set when someone gets the lock
+                                       quiescent */
+
+/* 32 buffers of 64k each, total 2 meg.
+ */
+#define MGA_DMA_BUF_ORDER     16
+#define MGA_DMA_BUF_SZ        (1<<MGA_DMA_BUF_ORDER)
+#define MGA_DMA_BUF_NR        31
+
+/* Keep these small for testing.
+ */
+#define MGA_NR_SAREA_CLIPRECTS 8
+
+/* 2 heaps (1 for card, 1 for agp), each divided into upto 128
+ * regions, subject to a minimum region size of (1<<16) == 64k. 
+ *
+ * Clients may subdivide regions internally, but when sharing between
+ * clients, the region size is the minimum granularity. 
+ */
+
+#define MGA_CARD_HEAP 0
+#define MGA_AGP_HEAP  1
+#define MGA_NR_TEX_HEAPS 2
+#define MGA_NR_TEX_REGIONS 16
+#define MGA_LOG_MIN_TEX_REGION_SIZE 16
+#endif
+
+typedef struct _drm_mga_warp_index {
+   	int installed;
+   	unsigned long phys_addr;
+   	int size;
+} drm_mga_warp_index_t;
+
+typedef struct drm_mga_init {
+   	enum { 
+	   	MGA_INIT_DMA = 0x01,
+	       	MGA_CLEANUP_DMA = 0x02
+	} func;
+   	int reserved_map_agpstart;
+   	int reserved_map_idx;
+   	int buffer_map_idx;
+   	int sarea_priv_offset;
+   	int primary_size;
+   	int warp_ucode_size;
+   	unsigned int frontOffset;
+   	unsigned int backOffset;
+   	unsigned int depthOffset;
+   	unsigned int textureOffset;
+   	unsigned int textureSize;
+        unsigned int agpTextureOffset;
+        unsigned int agpTextureSize;
+   	unsigned int cpp;
+   	unsigned int stride;
+   	int sgram;
+	int chipset;
+   	drm_mga_warp_index_t WarpIndex[MGA_MAX_WARP_PIPES];
+	unsigned int mAccess;
+} drm_mga_init_t;
+
+/* Warning: if you change the sarea structure, you must change the Xserver
+ * structures as well */
+
+typedef struct _drm_mga_tex_region {
+	unsigned char next, prev;	
+	unsigned char in_use;	
+	unsigned int age;			
+} drm_mga_tex_region_t;
+
+typedef struct _drm_mga_sarea {
+	/* The channel for communication of state information to the kernel
+	 * on firing a vertex dma buffer.
+	 */
+   	unsigned int ContextState[MGA_CTX_SETUP_SIZE];
+   	unsigned int ServerState[MGA_2D_SETUP_SIZE];
+   	unsigned int TexState[2][MGA_TEX_SETUP_SIZE];
+   	unsigned int WarpPipe;
+   	unsigned int dirty;
+
+   	unsigned int nbox;
+   	drm_clip_rect_t boxes[MGA_NR_SAREA_CLIPRECTS];
+
+
+	/* Information about the most recently used 3d drawable.  The
+	 * client fills in the req_* fields, the server fills in the 
+	 * exported_ fields and puts the cliprects into boxes, above.
+	 *
+	 * The client clears the exported_drawable field before
+	 * clobbering the boxes data.
+	 */
+        unsigned int req_drawable;	 /* the X drawable id */
+	unsigned int req_draw_buffer;	 /* MGA_FRONT or MGA_BACK */
+
+        unsigned int exported_drawable;	 
+	unsigned int exported_index; 
+        unsigned int exported_stamp;	
+        unsigned int exported_buffers;	 
+        unsigned int exported_nfront;
+        unsigned int exported_nback;
+	int exported_back_x, exported_front_x, exported_w;	
+	int exported_back_y, exported_front_y, exported_h;
+   	drm_clip_rect_t exported_boxes[MGA_NR_SAREA_CLIPRECTS];
+   
+	/* Counters for aging textures and for client-side throttling.
+	 */
+        unsigned int last_enqueue;	/* last time a buffer was enqueued */
+	unsigned int last_dispatch;	/* age of the most recently dispatched buffer */
+	unsigned int last_quiescent;     /*  */
+
+
+	/* LRU lists for texture memory in agp space and on the card
+	 */
+	drm_mga_tex_region_t texList[MGA_NR_TEX_HEAPS][MGA_NR_TEX_REGIONS+1];
+	unsigned int texAge[MGA_NR_TEX_HEAPS];
+	
+	/* Mechanism to validate card state.
+	 */
+   	int ctxOwner;
+} drm_mga_sarea_t;	
+
+/* Device specific ioctls:
+ */
+typedef struct _drm_mga_clear {
+	unsigned int clear_color;
+	unsigned int clear_depth;
+	unsigned int flags;
+} drm_mga_clear_t;
+
+typedef struct _drm_mga_swap {
+   	int dummy;
+} drm_mga_swap_t;
+
+typedef struct _drm_mga_iload {
+	int idx;
+	int length;
+	unsigned int destOrg;
+} drm_mga_iload_t;
+
+typedef struct _drm_mga_vertex {
+   	int idx;		/* buffer to queue */
+	int used;		/* bytes in use */
+	int discard;		/* client finished with buffer?  */
+} drm_mga_vertex_t;
+
+typedef struct _drm_mga_indices {
+   	int idx;		/* buffer to queue */
+	unsigned int start;		
+	unsigned int end;		
+	int discard;		/* client finished with buffer?  */
+} drm_mga_indices_t;
+
+#endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/r128_drm.h b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/r128_drm.h
new file mode 100644
index 000000000..0379a5fae
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/r128_drm.h
@@ -0,0 +1,111 @@
+/* r128_drm.h -- Public header for the r128 driver
+ * Created: Wed Apr  5 19:24:19 2000 by kevin@precisioninsight.com
+ *
+ * Copyright 2000 Precision Insight, Inc., Cedar Park, Texas.
+ * All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ * $XFree86$
+ */
+
+#ifndef _R128_DRM_H_
+#define _R128_DRM_H_
+
+/* WARNING: If you change any of these defines, make sure to change the
+ * defines in the Xserver file (xf86drmR128.h)
+ */
+typedef struct drm_r128_init {
+	enum { 
+		R128_INIT_CCE    = 0x01,
+		R128_CLEANUP_CCE = 0x02
+	} func;
+	int sarea_priv_offset;
+	int is_pci;
+	int cce_mode;
+	int cce_fifo_size;
+	int cce_secure;
+	int ring_size;
+	int usec_timeout;
+
+	int fb_offset;
+	int agp_ring_offset;
+	int agp_read_ptr_offset;
+	int agp_vertbufs_offset;
+	int agp_indbufs_offset;
+	int agp_textures_offset;
+	int mmio_offset;
+} drm_r128_init_t;
+
+typedef struct drm_r128_packet {
+	unsigned long *buffer;
+	int            count;
+	int            flags;
+} drm_r128_packet_t;
+
+typedef enum drm_r128_prim {
+	_DRM_R128_PRIM_NONE		= 0x0001,
+	_DRM_R128_PRIM_POINT		= 0x0002,
+	_DRM_R128_PRIM_LINE		= 0x0004,
+	_DRM_R128_PRIM_POLY_LINE	= 0x0008,
+	_DRM_R128_PRIM_TRI_LIST		= 0x0010,
+	_DRM_R128_PRIM_TRI_FAN		= 0x0020,
+	_DRM_R128_PRIM_TRI_STRIP	= 0x0040,
+	_DRM_R128_PRIM_TRI_TYPE2	= 0x0080
+} drm_r128_prim_t;
+
+typedef struct drm_r128_vertex {
+				/* Indices here refer to the offset into
+				   buflist in drm_buf_get_t.  */
+	int		send_count;	  /* Number of buffers to send	    */
+	int		*send_indices;	  /* List of handles to buffers	    */
+	int		*send_sizes;	  /* Lengths of data to send	    */
+	drm_r128_prim_t	prim;		  /* Primitive type		    */
+	int		request_count;	  /* Number of buffers requested    */
+	int		*request_indices; /* Buffer information		    */
+	int		*request_sizes;
+	int		granted_count;	  /* Number of buffers granted	    */
+} drm_r128_vertex_t;
+
+/* WARNING: If you change any of these defines, make sure to change the
+ * defines in the Xserver file (r128_sarea.h)
+ */
+#define R128_LOCAL_TEX_HEAP       0
+#define R128_AGP_TEX_HEAP         1
+#define R128_NR_TEX_HEAPS         2
+#define R128_NR_TEX_REGIONS      64
+#define R128_LOG_TEX_GRANULARITY 16
+
+typedef struct drm_tex_region {
+	unsigned char next, prev;       
+	unsigned char in_use;   
+	int age;                        
+} drm_tex_region_t;
+
+typedef struct drm_r128_sarea {
+	drm_tex_region_t tex_list[R128_NR_TEX_HEAPS][R128_NR_TEX_REGIONS+1];
+	int              tex_age[R128_NR_TEX_HEAPS];
+	int              ctx_owner;
+	int              ring_write;
+} drm_r128_sarea_t;
+
+#endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/tdfx/Makefile b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/tdfx/Makefile
new file mode 100644
index 000000000..e0ff8ffa4
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/tdfx/Makefile
@@ -0,0 +1,15 @@
+# $FreeBSD$
+
+KMOD	= tdfx
+SRCS    = tdfx_drv.c tdfx_context.c
+SRCS   += device_if.h bus_if.h pci_if.h
+CFLAGS += ${DEBUG_FLAGS} -I..
+KERN    = /usr/src/sys
+
+@:
+	ln -sf /sys @
+
+machine:
+	ln -sf /sys/i386/include machine
+
+.include "/usr/src/sys/conf/kmod.mk"
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/tdfx/tdfx_context.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/tdfx/tdfx_context.c
new file mode 100644
index 000000000..0aecf762a
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/tdfx/tdfx_context.c
@@ -0,0 +1,201 @@
+/* tdfx_context.c -- IOCTLs for tdfx contexts -*- c -*-
+ * Created: Thu Oct  7 10:50:22 1999 by faith@precisioninsight.com
+ * Revised: Sat Oct  9 23:39:56 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI$
+ * $XFree86$
+ *
+ */
+
+#include "drmP.h"
+#include "tdfx_drv.h"
+
+extern drm_ctx_t tdfx_res_ctx;
+
+static int tdfx_alloc_queue(drm_device_t *dev)
+{
+	static int context = 0;
+
+	return ++context;	/* Should this reuse contexts in the future? */
+}
+
+int tdfx_context_switch(drm_device_t *dev, int old, int new)
+{
+        char        buf[64];
+
+        atomic_inc(&dev->total_ctx);
+
+        if (test_and_set_bit(0, &dev->context_flag)) {
+                DRM_ERROR("Reentering -- FIXME\n");
+                return -EBUSY;
+        }
+
+#if DRM_DMA_HISTOGRAM
+        getnanotime(&dev->ctx_start);
+#endif
+        
+        DRM_DEBUG("Context switch from %d to %d\n", old, new);
+
+        if (new == dev->last_context) {
+                clear_bit(0, &dev->context_flag);
+                return 0;
+        }
+        
+        if (drm_flags & DRM_FLAG_NOCTX) {
+                tdfx_context_switch_complete(dev, new);
+        } else {
+                sprintf(buf, "C %d %d\n", old, new);
+                drm_write_string(dev, buf);
+        }
+        
+        return 0;
+}
+
+int tdfx_context_switch_complete(drm_device_t *dev, int new)
+{
+        dev->last_context = new;  /* PRE/POST: This is the _only_ writer. */
+        dev->last_switch  = ticks;
+        
+        if (!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+                DRM_ERROR("Lock isn't held after context switch\n");
+        }
+
+				/* If a context switch is ever initiated
+                                   when the kernel holds the lock, release
+                                   that lock here. */
+#if DRM_DMA_HISTOGRAM
+	{
+	    struct timespec ts;
+	    getnanotime(&ts);
+	    timespecsub(&ts, &dev->lck_start);
+	    atomic_inc(&dev->histo.ctx[drm_histogram_slot(&ts)]);
+	}
+#endif
+        clear_bit(0, &dev->context_flag);
+        wakeup(&dev->context_wait);
+        
+        return 0;
+}
+
+
+int
+tdfx_resctx(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	drm_ctx_res_t	res;
+	drm_ctx_t	ctx;
+	int		i, error;
+
+	DRM_DEBUG("%d\n", DRM_RESERVED_CONTEXTS);
+	res = *(drm_ctx_res_t *) data;
+	if (res.count >= DRM_RESERVED_CONTEXTS) {
+		memset(&ctx, 0, sizeof(ctx));
+		for (i = 0; i < DRM_RESERVED_CONTEXTS; i++) {
+			ctx.handle = i;
+			error = copyout(&i, &res.contexts[i], sizeof(i));
+			if (error) return error;
+		}
+	}
+	res.count = DRM_RESERVED_CONTEXTS;
+	*(drm_ctx_res_t *) data = res;
+	return 0;
+}
+
+
+int
+tdfx_addctx(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	drm_device_t	*dev	= kdev->si_drv1;
+	drm_ctx_t	ctx;
+
+	ctx = *(drm_ctx_t *) data;
+	if ((ctx.handle = tdfx_alloc_queue(dev)) == DRM_KERNEL_CONTEXT) {
+				/* Skip kernel's context and get a new one. */
+		ctx.handle = tdfx_alloc_queue(dev);
+	}
+	DRM_DEBUG("%d\n", ctx.handle);
+	*(drm_ctx_t *) data = ctx;
+	return 0;
+}
+
+int
+tdfx_modctx(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	drm_ctx_t ctx;
+
+	ctx = *(drm_ctx_t *) data;
+	if (ctx.flags==_DRM_CONTEXT_PRESERVED)
+		tdfx_res_ctx.handle=ctx.handle;
+	return 0;
+}
+
+int
+tdfx_getctx(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	drm_ctx_t ctx;
+
+	ctx = *(drm_ctx_t *) data;
+	/* This is 0, because we don't hanlde any context flags */
+	ctx.flags = 0;
+	*(drm_ctx_t *) data = ctx;
+	return 0;
+}
+
+int
+tdfx_switchctx(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	drm_device_t	*dev	= kdev->si_drv1;
+	drm_ctx_t	ctx;
+
+	ctx = *(drm_ctx_t *) data;
+	DRM_DEBUG("%d\n", ctx.handle);
+	return tdfx_context_switch(dev, dev->last_context, ctx.handle);
+}
+
+int
+tdfx_newctx(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	drm_device_t	*dev	= kdev->si_drv1;
+	drm_ctx_t	ctx;
+
+	ctx = *(drm_ctx_t *) data;
+	DRM_DEBUG("%d\n", ctx.handle);
+	tdfx_context_switch_complete(dev, ctx.handle);
+
+	return 0;
+}
+
+int
+tdfx_rmctx(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	drm_ctx_t	ctx;
+
+	ctx = *(drm_ctx_t *) data;
+	DRM_DEBUG("%d\n", ctx.handle);
+				/* This is currently a noop because we
+				   don't reuse context values.  Perhaps we
+				   should? */
+	
+	return 0;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/tdfx/tdfx_drv.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/tdfx/tdfx_drv.c
new file mode 100644
index 000000000..573cfcf14
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/tdfx/tdfx_drv.c
@@ -0,0 +1,694 @@
+/* tdfx.c -- tdfx driver -*- c -*-
+ * Created: Thu Oct  7 10:38:32 1999 by faith@precisioninsight.com
+ * Revised: Tue Oct 12 08:51:35 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI$
+ * $XFree86$
+ *
+ */
+
+#include "drmP.h"
+#include "tdfx_drv.h"
+
+#include <pci/pcivar.h>
+
+MODULE_DEPEND(tdfx, drm, 1, 1, 1);
+
+#define TDFX_NAME	 "tdfx"
+#define TDFX_DESC	 "tdfx"
+#define TDFX_DATE	 "19991009"
+#define TDFX_MAJOR	 0
+#define TDFX_MINOR	 0
+#define TDFX_PATCHLEVEL  1
+
+static int tdfx_init(device_t nbdev);
+static void tdfx_cleanup(device_t nbdev);
+
+drm_ctx_t	              tdfx_res_ctx;
+
+static int tdfx_probe(device_t dev)
+{
+	const char *s = 0;
+
+	switch (pci_get_devid(dev)) {
+	case 0x0003121a:
+		s = "3Dfx Voodoo Banshee graphics accelerator";
+		break;
+		
+	case 0x0005121a:
+		s = "3Dfx Voodoo 3 graphics accelerator";
+		break;
+	}
+
+	if (s) {
+		device_set_desc(dev, s);
+		return 0;
+	}
+
+	return ENXIO;
+}
+
+static int tdfx_attach(device_t dev)
+{
+	tdfx_init(dev);
+	return 0;
+}
+
+static int tdfx_detach(device_t dev)
+{
+	tdfx_cleanup(dev);
+	return 0;
+}
+
+static device_method_t tdfx_methods[] = {
+	/* Device interface */
+	DEVMETHOD(device_probe,		tdfx_probe),
+	DEVMETHOD(device_attach,	tdfx_attach),
+	DEVMETHOD(device_detach,	tdfx_detach),
+
+	{ 0, 0 }
+};
+
+static driver_t tdfx_driver = {
+	"drm",
+	tdfx_methods,
+	sizeof(drm_device_t),
+};
+
+static devclass_t tdfx_devclass;
+#define TDFX_SOFTC(unit) \
+	((drm_device_t *) devclass_get_softc(tdfx_devclass, unit))
+
+DRIVER_MODULE(if_tdfx, pci, tdfx_driver, tdfx_devclass, 0, 0);
+
+#define CDEV_MAJOR	145
+				/* tdfx_drv.c */
+static d_open_t tdfx_open;
+static d_close_t tdfx_close;
+static d_ioctl_t tdfx_version;
+static d_ioctl_t tdfx_ioctl;
+static d_ioctl_t tdfx_lock;
+static d_ioctl_t tdfx_unlock;
+
+static struct cdevsw tdfx_cdevsw = {
+	/* open */	tdfx_open,
+	/* close */	tdfx_close,
+	/* read */	drm_read,
+	/* write */	drm_write,
+	/* ioctl */	tdfx_ioctl,
+	/* poll */	drm_poll,
+	/* mmap */	drm_mmap,
+	/* strategy */	nostrategy,
+	/* name */	"tdfx",
+	/* maj */	CDEV_MAJOR,
+	/* dump */	nodump,
+	/* psize */	nopsize,
+	/* flags */	D_TTY | D_TRACKCLOSE,
+	/* bmaj */	-1
+};
+
+static drm_ioctl_desc_t	      tdfx_ioctls[] = {
+	[DRM_IOCTL_NR(DRM_IOCTL_VERSION)]    = { tdfx_version,	  0, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_GET_UNIQUE)] = { drm_getunique,	  0, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_GET_MAGIC)]  = { drm_getmagic,	  0, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_IRQ_BUSID)]  = { drm_irq_busid,	  0, 1 },
+
+	[DRM_IOCTL_NR(DRM_IOCTL_SET_UNIQUE)] = { drm_setunique,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_BLOCK)]	     = { drm_block,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_UNBLOCK)]    = { drm_unblock,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_AUTH_MAGIC)] = { drm_authmagic,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_ADD_MAP)]    = { drm_addmap,	  1, 1 },
+	
+	[DRM_IOCTL_NR(DRM_IOCTL_ADD_CTX)]    = { tdfx_addctx,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_RM_CTX)]     = { tdfx_rmctx,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_MOD_CTX)]    = { tdfx_modctx,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_GET_CTX)]    = { tdfx_getctx,	  1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_SWITCH_CTX)] = { tdfx_switchctx,  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_NEW_CTX)]    = { tdfx_newctx,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_RES_CTX)]    = { tdfx_resctx,	  1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_ADD_DRAW)]   = { drm_adddraw,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_RM_DRAW)]    = { drm_rmdraw,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_LOCK)]	     = { tdfx_lock,	  1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_UNLOCK)]     = { tdfx_unlock,	  1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_FINISH)]     = { drm_finish,	  1, 0 },
+};
+#define TDFX_IOCTL_COUNT DRM_ARRAY_SIZE(tdfx_ioctls)
+
+static int
+tdfx_setup(drm_device_t *dev)
+{
+	int i;
+	
+	device_busy(dev->device);
+
+	atomic_set(&dev->ioctl_count, 0);
+	atomic_set(&dev->vma_count, 0);
+	dev->buf_use	  = 0;
+	atomic_set(&dev->buf_alloc, 0);
+
+	atomic_set(&dev->total_open, 0);
+	atomic_set(&dev->total_close, 0);
+	atomic_set(&dev->total_ioctl, 0);
+	atomic_set(&dev->total_irq, 0);
+	atomic_set(&dev->total_ctx, 0);
+	atomic_set(&dev->total_locks, 0);
+	atomic_set(&dev->total_unlocks, 0);
+	atomic_set(&dev->total_contends, 0);
+	atomic_set(&dev->total_sleeps, 0);
+
+	for (i = 0; i < DRM_HASH_SIZE; i++) {
+		dev->magiclist[i].head = NULL;
+		dev->magiclist[i].tail = NULL;
+	}
+	dev->maplist	    = NULL;
+	dev->map_count	    = 0;
+	dev->vmalist	    = NULL;
+	dev->lock.hw_lock   = NULL;
+	dev->lock.lock_queue = 0;
+	dev->queue_count    = 0;
+	dev->queue_reserved = 0;
+	dev->queue_slots    = 0;
+	dev->queuelist	    = NULL;
+	dev->irq	    = 0;
+	dev->context_flag   = 0;
+	dev->interrupt_flag = 0;
+	dev->dma            = 0;
+	dev->dma_flag	    = 0;
+	dev->last_context   = 0;
+	dev->last_switch    = 0;
+	dev->last_checked   = 0;
+	callout_init(&dev->timer);
+	dev->context_wait   = 0;
+
+	timespecclear(&dev->ctx_start);
+	timespecclear(&dev->lck_start);
+	
+	dev->buf_rp	  = dev->buf;
+	dev->buf_wp	  = dev->buf;
+	dev->buf_end	  = dev->buf + DRM_BSZ;
+	bzero(&dev->buf_sel, sizeof dev->buf_sel);
+	dev->buf_sigio	  = NULL;
+	dev->buf_readers  = 0;
+	dev->buf_writers  = 0;
+	dev->buf_selecting = 0;
+
+	tdfx_res_ctx.handle=-1;
+			
+	DRM_DEBUG("\n");
+			
+	/* The kernel's context could be created here, but is now created
+	   in drm_dma_enqueue.	This is more resource-efficient for
+	   hardware that does not do DMA, but may mean that
+	   drm_select_queue fails between the time the interrupt is
+	   initialized and the time the queues are initialized. */
+			
+	return 0;
+}
+
+
+static int
+tdfx_takedown(drm_device_t *dev)
+{
+	int		  i;
+	drm_magic_entry_t *pt, *next;
+	drm_map_t	  *map;
+	drm_vma_entry_t	  *vma, *vma_next;
+
+	DRM_DEBUG("\n");
+
+	lockmgr(&dev->dev_lock, LK_EXCLUSIVE, 0, curproc);
+	callout_stop(&dev->timer);
+	
+	if (dev->devname) {
+		drm_free(dev->devname, strlen(dev->devname)+1, DRM_MEM_DRIVER);
+		dev->devname = NULL;
+	}
+	
+	if (dev->unique) {
+		drm_free(dev->unique, strlen(dev->unique)+1, DRM_MEM_DRIVER);
+		dev->unique = NULL;
+		dev->unique_len = 0;
+	}
+				/* Clear pid list */
+	for (i = 0; i < DRM_HASH_SIZE; i++) {
+		for (pt = dev->magiclist[i].head; pt; pt = next) {
+			next = pt->next;
+			drm_free(pt, sizeof(*pt), DRM_MEM_MAGIC);
+		}
+		dev->magiclist[i].head = dev->magiclist[i].tail = NULL;
+	}
+	
+				/* Clear vma list (only built for debugging) */
+	if (dev->vmalist) {
+		for (vma = dev->vmalist; vma; vma = vma_next) {
+			vma_next = vma->next;
+			drm_free(vma, sizeof(*vma), DRM_MEM_VMAS);
+		}
+		dev->vmalist = NULL;
+	}
+	
+				/* Clear map area and mtrr information */
+	if (dev->maplist) {
+		for (i = 0; i < dev->map_count; i++) {
+			map = dev->maplist[i];
+			switch (map->type) {
+			case _DRM_REGISTERS:
+			case _DRM_FRAME_BUFFER:
+#ifdef CONFIG_MTRR
+				if (map->mtrr >= 0) {
+					int retcode;
+					retcode = mtrr_del(map->mtrr,
+							   map->offset,
+							   map->size);
+					DRM_DEBUG("mtrr_del = %d\n", retcode);
+				}
+#endif
+				drm_ioremapfree(map->handle, map->size);
+				break;
+			case _DRM_SHM:
+				drm_free_pages((unsigned long)map->handle,
+					       drm_order(map->size)
+					       - PAGE_SHIFT,
+					       DRM_MEM_SAREA);
+				break;
+			case _DRM_AGP:
+				break; /* XXX */
+			}
+			drm_free(map, sizeof(*map), DRM_MEM_MAPS);
+		}
+		drm_free(dev->maplist,
+			 dev->map_count * sizeof(*dev->maplist),
+			 DRM_MEM_MAPS);
+		dev->maplist   = NULL;
+		dev->map_count = 0;
+	}
+	
+	if (dev->lock.hw_lock) {
+		dev->lock.hw_lock    = NULL; /* SHM removed */
+		dev->lock.pid	     = 0;
+		wakeup(&dev->lock.lock_queue);
+	}
+	lockmgr(&dev->dev_lock, LK_RELEASE, 0, curproc);
+
+	device_unbusy(dev->device);
+	
+	return 0;
+}
+
+/* tdfx_init is called via tdfx_attach at module load time, */
+
+static int
+tdfx_init(device_t nbdev)
+{
+	drm_device_t *dev = device_get_softc(nbdev);
+
+	DRM_DEBUG("\n");
+
+	memset((void *)dev, 0, sizeof(*dev));
+	simple_lock_init(&dev->count_lock);
+	lockinit(&dev->dev_lock, PZERO, "drmlk", 0, 0);
+	
+#if 0
+	drm_parse_options(tdfx);
+#endif
+
+	dev->device = nbdev;
+	dev->devnode = make_dev(&tdfx_cdevsw,
+				device_get_unit(nbdev),
+				DRM_DEV_UID,
+				DRM_DEV_GID,
+				DRM_DEV_MODE,
+				TDFX_NAME);
+	dev->name   = TDFX_NAME;
+
+	drm_mem_init();
+	drm_sysctl_init(dev);
+	TAILQ_INIT(&dev->files);
+
+	DRM_INFO("Initialized %s %d.%d.%d %s on minor %d\n",
+		 TDFX_NAME,
+		 TDFX_MAJOR,
+		 TDFX_MINOR,
+		 TDFX_PATCHLEVEL,
+		 TDFX_DATE,
+		 device_get_unit(nbdev));
+	
+	return 0;
+}
+
+/* tdfx_cleanup is called via tdfx_detach at module unload time. */
+
+static void
+tdfx_cleanup(device_t nbdev)
+{
+	drm_device_t *dev = device_get_softc(nbdev);
+
+	DRM_DEBUG("\n");
+	
+	drm_sysctl_cleanup(dev);
+	destroy_dev(dev->devnode);
+
+	DRM_INFO("Module unloaded\n");
+
+	device_busy(dev->device);
+	tdfx_takedown(dev);
+}
+
+static int
+tdfx_version(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	drm_version_t version;
+	int	      len;
+
+	version = *(drm_version_t *) data;
+
+#define DRM_COPY(name,value)				\
+	len = strlen(value);				\
+	if (len > name##_len) len = name##_len;		\
+	name##_len = strlen(value);			\
+	if (len && name) {				\
+		int error = copyout(value, name, len);	\
+		if (error) return error;		\
+	}
+
+	version.version_major	   = TDFX_MAJOR;
+	version.version_minor	   = TDFX_MINOR;
+	version.version_patchlevel = TDFX_PATCHLEVEL;
+
+	DRM_COPY(version.name, TDFX_NAME);
+	DRM_COPY(version.date, TDFX_DATE);
+	DRM_COPY(version.desc, TDFX_DESC);
+
+	*(drm_version_t *) data = version;
+	return 0;
+}
+
+static int
+tdfx_open(dev_t kdev, int flags, int fmt, struct proc *p)
+{
+	drm_device_t  *dev    = TDFX_SOFTC(minor(kdev));
+	int	      retcode = 0;
+	
+	DRM_DEBUG("open_count = %d\n", dev->open_count);
+
+	device_busy(dev->device);
+	if (!(retcode = drm_open_helper(kdev, flags, fmt, p, dev))) {
+		atomic_inc(&dev->total_open);
+		simple_lock(&dev->count_lock);
+		if (!dev->open_count++) {
+			simple_unlock(&dev->count_lock);
+			retcode = tdfx_setup(dev);
+		}
+		simple_unlock(&dev->count_lock);
+	}
+	device_unbusy(dev->device);
+
+	return retcode;
+}
+
+static int
+tdfx_close(dev_t kdev, int flags, int fmt, struct proc *p)
+{
+	drm_device_t  *dev    = kdev->si_drv1;
+	int	      retcode = 0;
+
+	DRM_DEBUG("open_count = %d\n", dev->open_count);
+	if (!(retcode = drm_close(kdev, flags, fmt, p))) {
+		atomic_inc(&dev->total_close);
+		simple_lock(&dev->count_lock);
+		if (!--dev->open_count) {
+			if (atomic_read(&dev->ioctl_count) || dev->blocked) {
+				DRM_ERROR("Device busy: %d %d\n",
+					  atomic_read(&dev->ioctl_count),
+					  dev->blocked);
+				simple_unlock(&dev->count_lock);
+				return EBUSY;
+			}
+			simple_unlock(&dev->count_lock);
+			return tdfx_takedown(dev);
+		}
+		simple_unlock(&dev->count_lock);
+	}
+
+	return retcode;
+}
+
+/* tdfx_ioctl is called whenever a process performs an ioctl on /dev/drm. */
+
+static int
+tdfx_ioctl(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	int		 nr	 = DRM_IOCTL_NR(cmd);
+	drm_device_t	 *dev	 = kdev->si_drv1;
+	drm_file_t       *priv;
+	int		 retcode = 0;
+	drm_ioctl_desc_t *ioctl;
+	d_ioctl_t	 *func;
+
+	DRM_DEBUG("dev=%p\n", dev);
+	priv = drm_find_file_by_proc(dev, p);
+	if (!priv) {
+		DRM_DEBUG("can't find authenticator\n");
+		return EINVAL;
+	}
+
+	atomic_inc(&dev->ioctl_count);
+	atomic_inc(&dev->total_ioctl);
+	++priv->ioctl_count;
+	
+	DRM_DEBUG("pid = %d, cmd = 0x%02lx, nr = 0x%02x, auth = %d\n",
+		  p->p_pid, cmd, nr, priv->authenticated);
+
+	switch (cmd) {
+	case FIONBIO:
+		atomic_dec(&dev->ioctl_count);
+		return 0;
+
+	case FIOASYNC:
+		atomic_dec(&dev->ioctl_count);
+		dev->flags |= FASYNC;
+		return 0;
+
+	case FIOSETOWN:
+		atomic_dec(&dev->ioctl_count);
+		return fsetown(*(int *)data, &dev->buf_sigio);
+
+	case FIOGETOWN:
+		atomic_dec(&dev->ioctl_count);
+		*(int *) data = fgetown(dev->buf_sigio);
+		return 0;
+	}
+
+	if (nr >= TDFX_IOCTL_COUNT) {
+		retcode = EINVAL;
+	} else {
+		ioctl	  = &tdfx_ioctls[nr];
+		func	  = ioctl->func;
+
+		if (!func) {
+			DRM_DEBUG("no function\n");
+			retcode = EINVAL;
+		} else if ((ioctl->root_only && suser(p))
+			    || (ioctl->auth_needed && !priv->authenticated)) {
+			retcode = EACCES;
+		} else {
+			retcode = (func)(kdev, cmd, data, flags, p);
+		}
+	}
+	
+	atomic_dec(&dev->ioctl_count);
+	return retcode;
+}
+
+static int
+tdfx_lock(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+        drm_device_t      *dev    = kdev->si_drv1;
+        int               ret   = 0;
+        drm_lock_t        lock;
+#if DRM_DMA_HISTOGRAM
+
+        getnanotime(&dev->lck_start);
+#endif
+
+	lock = *(drm_lock_t *) data;
+
+        if (lock.context == DRM_KERNEL_CONTEXT) {
+                DRM_ERROR("Process %d using kernel context %d\n",
+                          p->p_pid, lock.context);
+                return EINVAL;
+        }
+
+        DRM_DEBUG("%d (pid %d) requests lock (0x%08x), flags = 0x%08x\n",
+                  lock.context, p->p_pid, dev->lock.hw_lock->lock,
+                  lock.flags);
+
+#if 0
+				/* dev->queue_count == 0 right now for
+                                   tdfx.  FIXME? */
+        if (lock.context < 0 || lock.context >= dev->queue_count)
+                return EINVAL;
+#endif
+        
+        if (!ret) {
+#if 0
+                if (_DRM_LOCKING_CONTEXT(dev->lock.hw_lock->lock)
+                    != lock.context) {
+                        long j = ticks - dev->lock.lock_time;
+
+                        if (lock.context == tdfx_res_ctx.handle &&
+				j >= 0 && j < DRM_LOCK_SLICE) {
+                                /* Can't take lock if we just had it and
+                                   there is contention. */
+                                DRM_DEBUG("%d (pid %d) delayed j=%d dev=%d ticks=%d\n",
+					lock.context, p->p_pid, j, 
+					dev->lock.lock_time, ticks);
+				ret = tsleep(&never, PZERO|PCATCH, "drmlk1",
+					       DRM_LOCK_SLICE - j);
+				if (ret)
+					return ret;
+				DRM_DEBUG("ticks=%d\n", ticks);
+                        }
+                }
+#endif
+                for (;;) {
+                        if (!dev->lock.hw_lock) {
+                                /* Device has been unregistered */
+                                ret = EINTR;
+                                break;
+                        }
+                        if (drm_lock_take(&dev->lock.hw_lock->lock,
+                                          lock.context)) {
+                                dev->lock.pid       = p->p_pid;
+                                dev->lock.lock_time = ticks;
+                                atomic_inc(&dev->total_locks);
+                                break;  /* Got lock */
+                        }
+                        
+                                /* Contention */
+                        atomic_inc(&dev->total_sleeps);
+			ret = tsleep(&dev->lock.lock_queue,
+				       PZERO|PCATCH,
+				       "drmlk2",
+				       0);
+			if (ret)
+				break;
+                }
+        }
+
+#if 0
+	if (!ret && dev->last_context != lock.context &&
+		lock.context != tdfx_res_ctx.handle &&
+		dev->last_context != tdfx_res_ctx.handle) {
+		add_wait_queue(&dev->context_wait, &entry);
+	        current->state = TASK_INTERRUPTIBLE;
+                /* PRE: dev->last_context != lock.context */
+	        tdfx_context_switch(dev, dev->last_context, lock.context);
+		/* POST: we will wait for the context
+                   switch and will dispatch on a later call
+                   when dev->last_context == lock.context
+                   NOTE WE HOLD THE LOCK THROUGHOUT THIS
+                   TIME! */
+		current->policy |= SCHED_YIELD;
+	        schedule();
+	        current->state = TASK_RUNNING;
+	        remove_wait_queue(&dev->context_wait, &entry);
+	        if (signal_pending(current)) {
+	                ret = EINTR;
+	        } else if (dev->last_context != lock.context) {
+			DRM_ERROR("Context mismatch: %d %d\n",
+                        	dev->last_context, lock.context);
+	        }
+	}
+#endif
+
+        if (!ret) {
+                if (lock.flags & _DRM_LOCK_READY) {
+				/* Wait for space in DMA/FIFO */
+		}
+                if (lock.flags & _DRM_LOCK_QUIESCENT) {
+				/* Make hardware quiescent */
+#if 0
+                        tdfx_quiescent(dev);
+#endif
+		}
+        }
+
+#if 0
+	DRM_ERROR("pid = %5d, old counter = %5ld\n", 
+		p->p_pid, current->counter);
+#endif
+#if 0
+	while (current->counter > 25)
+		current->counter >>= 1; /* decrease time slice */
+	DRM_ERROR("pid = %5d, new counter = %5ld\n",
+		 p->p_pid, current->counter);
+#endif
+        DRM_DEBUG("%d %s\n", lock.context, ret ? "interrupted" : "has lock");
+
+#if DRM_DMA_HISTOGRAM
+	{
+	    struct timespec ts;
+	    getnanotime(&ts);
+	    timespecsub(&ts, &dev->lck_start);
+	    atomic_inc(&dev->histo.lhld[drm_histogram_slot(&ts)]);
+	}
+#endif
+        
+        return ret;
+}
+
+
+static int
+tdfx_unlock(dev_t kdev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	drm_device_t	  *dev	  = kdev->si_drv1;
+	drm_lock_t	  lock;
+
+	lock = *(drm_lock_t *) data;
+	
+	if (lock.context == DRM_KERNEL_CONTEXT) {
+		DRM_ERROR("Process %d using kernel context %d\n",
+			  p->p_pid, lock.context);
+		return EINVAL;
+	}
+
+	DRM_DEBUG("%d frees lock (%d holds)\n",
+		  lock.context,
+		  _DRM_LOCKING_CONTEXT(dev->lock.hw_lock->lock));
+	atomic_inc(&dev->total_unlocks);
+	if (_DRM_LOCK_IS_CONT(dev->lock.hw_lock->lock))
+		atomic_inc(&dev->total_contends);
+	drm_lock_transfer(dev, &dev->lock.hw_lock->lock, DRM_KERNEL_CONTEXT);
+				/* FIXME: Try to send data to card here */
+	if (!dev->context_flag) {
+		if (drm_lock_free(dev, &dev->lock.hw_lock->lock,
+				  DRM_KERNEL_CONTEXT)) {
+			DRM_ERROR("\n");
+		}
+	}
+
+	return 0;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/tdfx/tdfx_drv.h b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/tdfx/tdfx_drv.h
new file mode 100644
index 000000000..213f8ef1e
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/tdfx/tdfx_drv.h
@@ -0,0 +1,47 @@
+/* tdfx_drv.h -- Private header for tdfx driver -*- c -*-
+ * Created: Thu Oct  7 10:40:04 1999 by faith@precisioninsight.com
+ * Revised: Sat Oct  9 23:38:19 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI$
+ * $XFree86$
+ * 
+ */
+
+#ifndef _TDFX_DRV_H_
+#define _TDFX_DRV_H_
+
+				/* tdfx_context.c */
+
+extern d_ioctl_t tdfx_resctx;
+extern d_ioctl_t tdfx_addctx;
+extern d_ioctl_t tdfx_modctx;
+extern d_ioctl_t tdfx_getctx;
+extern d_ioctl_t tdfx_switchctx;
+extern d_ioctl_t tdfx_newctx;
+extern d_ioctl_t tdfx_rmctx;
+
+extern int  tdfx_context_switch(drm_device_t *dev, int old, int new);
+extern int  tdfx_context_switch_complete(drm_device_t *dev, int new);
+#endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/xf86drm.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/xf86drm.c
new file mode 100644
index 000000000..985c48f9d
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/xf86drm.c
@@ -0,0 +1,1108 @@
+/* xf86drm.c -- User-level interface to DRM device
+ * Created: Tue Jan  5 08:16:21 1999 by faith@precisioninsight.com
+ * Revised: Wed Aug  4 07:54:23 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drm.c,v 1.43 1999/08/04 18:14:43 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drm.c,v 1.4 1999/09/25 14:37:49 dawes Exp $
+ * 
+ */
+
+#ifdef XFree86Server
+# include "xf86.h"
+# include "xf86_OSproc.h"
+# include "xf86_ansic.h"
+# include "xf86Priv.h"
+# define _DRM_MALLOC xalloc
+# define _DRM_FREE   xfree
+# ifndef XFree86LOADER
+#  include <sys/stat.h>
+#  include <sys/mman.h>
+# endif
+#else
+# include <stdio.h>
+# include <stdlib.h>
+# include <unistd.h>
+# include <string.h>
+# include <ctype.h>
+# include <fcntl.h>
+# include <errno.h>
+# include <signal.h>
+# include <sys/types.h>
+# include <sys/stat.h>
+# include <sys/ioctl.h>
+# include <sys/mman.h>
+# include <sys/time.h>
+# include <sys/types.h>
+# ifdef DRM_USE_MALLOC
+#  define _DRM_MALLOC malloc
+#  define _DRM_FREE   free
+extern int xf86InstallSIGIOHandler(int fd, void (*f)(int, void*), void*);
+extern int xf86RemoveSIGIOHandler(int fd);
+# else
+#  include <Xlibint.h>
+#  define _DRM_MALLOC Xmalloc
+#  define _DRM_FREE   Xfree
+# endif
+#endif
+
+int sysctlbyname(const char *, void *, size_t *, void *, size_t);
+
+/* Not all systems have MAP_FAILED defined */
+#ifndef MAP_FAILED
+#define MAP_FAILED ((void *)-1)
+#endif
+
+#include "xf86drm.h"
+#include "drm.h"
+
+#define makedev(x,y)    ((dev_t)(((x) << 8) | (y)))     /* create dev_t */
+
+static int drmOpenByBusid(const char *busid);
+static int drmOpenDevice(const char *path, long dev,
+			 mode_t mode, uid_t user, gid_t group);
+
+static void *drmHashTable = NULL; /* Context switch callbacks */
+
+typedef struct drmHashEntry {
+    int      fd;
+    void     (*f)(int, void *, void *);
+    void     *tagTable;
+} drmHashEntry;
+
+void *drmMalloc(int size)
+{
+    void *pt;
+    if ((pt = _DRM_MALLOC(size))) memset(pt, 0, size);
+    return pt;
+}
+
+void drmFree(void *pt)
+{
+    if (pt) _DRM_FREE(pt);
+}
+
+static char *drmStrdup(const char *s)
+{
+    return s ? strdup(s) : NULL;
+}
+
+
+static unsigned long drmGetKeyFromFd(int fd)
+{
+#ifdef XFree86LOADER
+    struct xf86stat st;
+#else
+    struct stat     st;
+#endif
+
+    st.st_rdev = 0;
+    fstat(fd, &st);
+    return st.st_rdev;
+}
+
+static drmHashEntry *drmGetEntry(int fd)
+{
+    unsigned long key = drmGetKeyFromFd(fd);
+    void          *value;
+    drmHashEntry  *entry;
+
+    if (!drmHashTable) drmHashTable = drmHashCreate();
+
+    if (drmHashLookup(drmHashTable, key, &value)) {
+	entry           = drmMalloc(sizeof(*entry));
+	entry->fd       = fd;
+	entry->f        = NULL;
+	entry->tagTable = drmHashCreate();
+	drmHashInsert(drmHashTable, key, entry);
+    } else {
+	entry = value;
+    }
+    return entry;
+}
+
+/* drm_open is used to open the /dev/drm device */
+
+static int drm_open(const char *file)
+{
+    int fd = open(file, O_RDWR, 0);
+
+    if (fd >= 0) return fd;
+    return -errno;
+}
+
+/* drmAvailable looks for a usable device node in /dev/dri. */
+
+int drmAvailable(void)
+{
+    int i, fd;
+    char   dev_name[64];
+    mode_t mode   = DRM_DEV_MODE;
+    mode_t dirmode;
+    gid_t  group  = DRM_DEV_GID;
+    uid_t  user   = DRM_DEV_UID;
+    long   dev     = 0;
+
+    if (!geteuid()) {
+	dirmode = mode;
+	if (dirmode & S_IRUSR) dirmode |= S_IXUSR;
+	if (dirmode & S_IRGRP) dirmode |= S_IXGRP;
+	if (dirmode & S_IROTH) dirmode |= S_IXOTH;
+	dirmode &= ~(S_IWGRP | S_IWOTH);
+	mkdir("/dev/dri", 0);
+	chown("/dev/dri", user, group);
+	chmod("/dev/dri", dirmode);
+    }
+
+    for (i = 0; i < 8; i++) {
+	sprintf(dev_name, "/dev/dri/card%d", i);
+	dev = makedev(145, i);
+	fd = drmOpenDevice(dev_name, dev, mode, user, group);
+	if (fd < 0) {
+	    remove(dev_name);
+	    continue;
+	}
+	close(fd);
+	return 1;
+    }
+    return 0;
+}
+
+static int drmOpenDevice(const char *path, long dev,
+			 mode_t mode, uid_t user, gid_t group)
+{
+#ifdef XFree86LOADER
+    struct xf86stat st;
+#else
+    struct stat     st;
+#endif
+
+    if (!stat(path, &st) && st.st_rdev == dev) return drm_open(path);
+
+    if (geteuid()) return DRM_ERR_NOT_ROOT;
+    remove(path);
+    if (mknod(path, S_IFCHR, dev)) {
+	remove(path);
+	return DRM_ERR_NOT_ROOT;
+    }
+    chown(path, user, group);
+    chmod(path, mode);
+    return drm_open(path);
+}
+
+static int drmOpenByName(const char *name)
+{
+    int    i;
+    char   dev_name[64];
+    mode_t mode   = DRM_DEV_MODE;
+    mode_t dirmode;
+    gid_t  group  = DRM_DEV_GID;
+    uid_t  user   = DRM_DEV_UID;
+    long   dev     = 0;
+    drmVersionPtr version;
+    int    fd;
+
+#if defined(XFree86Server)
+    mode  = xf86ConfigDRI.mode ? xf86ConfigDRI.mode : DRM_DEV_MODE;
+    group = xf86ConfigDRI.group ? xf86ConfigDRI.group : DRM_DEV_GID;
+#endif
+
+#if defined(XFree86Server)
+    if (!drmAvailable()) {
+        /* try to load the kernel module now */
+        if (!xf86LoadKernelModule(name)) {
+            ErrorF("[drm] failed to load kernel module \"%s\"\n",
+		   name);
+            return -1;
+        }
+    }
+#else
+    if (!drmAvailable())
+       return -1;
+#endif
+
+    if (!geteuid()) {
+	dirmode = mode;
+	if (dirmode & S_IRUSR) dirmode |= S_IXUSR;
+	if (dirmode & S_IRGRP) dirmode |= S_IXGRP;
+	if (dirmode & S_IROTH) dirmode |= S_IXOTH;
+	dirmode &= ~(S_IWGRP | S_IWOTH);
+	mkdir("/dev/dri", 0);
+	chown("/dev/dri", user, group);
+	chmod("/dev/dri", dirmode);
+    }
+
+    for (i = 0; i < 8; i++) {
+	sprintf(dev_name, "/dev/dri/card%d", i);
+	dev = makedev(145, i);
+	fd = drmOpenDevice(dev_name, dev, mode, user, group);
+	if (fd < 0) {
+	    remove(dev_name);
+	    continue;
+	}
+	version = drmGetVersion(fd);
+	if (!version) {
+	    continue;
+	}
+	if (!strcmp(version->name, name)) {
+	    drmFreeVersion(version);
+	    return fd;
+	}
+	drmFreeVersion(version);
+    }
+
+#if 0
+    for (i = 0; i < 8; i++) {
+	sprintf(sysctl_name, "hw.dri.%d.name", i);
+	sprintf(dev_name, "/dev/dri/card%d", i);
+	len = sizeof(buf);
+	if (sysctlbyname(sysctl_name, buf, &len, 0, 0) == 0) {
+	    buf[len-1] = '\0';
+	    for (driver = pt = buf; *pt && *pt != ' '; ++pt)
+		;
+	    if (*pt) {	/* Device is next */
+		*pt = '\0';
+		if (!strcmp(driver, name)) { /* Match */
+		    for (devstring = ++pt; *pt && *pt != ' '; ++pt)
+			;
+		    if (*pt) { /* Found busid */
+			return drmOpenByBusid(++pt);
+		    } else {	/* No busid */
+			dev = strtol(devstring, NULL, 0);
+			return drmOpenDevice(dev_name, dev,
+					     mode, user, group);
+		    }
+		}
+	    }
+	} else remove(dev_name);
+    }
+#endif
+
+    return -1;
+}
+
+static int drmOpenByBusid(const char *busid)
+{
+    int    i;
+    char   dev_name[64];
+    char   *buf;
+    int    fd;
+
+    for (i = 0; i < 8; i++) {
+	sprintf(dev_name, "/dev/dri/card%d", i);
+	if ((fd = drm_open(dev_name)) >= 0) {
+	    buf = drmGetBusid(fd);
+	    if (buf && !strcmp(buf, busid)) {
+	      drmFreeBusid(buf);
+	      return fd;
+	    }
+	    if (buf) drmFreeBusid(buf);
+	    close(fd);
+	}
+    }
+    return -1;
+}
+
+
+/* drmOpen looks up the specified name and busid, and opens the device
+   found.  The entry in /dev/dri is created if necessary (and if root).
+   A file descriptor is returned.  On error, the return value is
+   negative. */
+
+int drmOpen(const char *name, const char *busid)
+{
+    if (busid) return drmOpenByBusid(busid);
+    return drmOpenByName(name);
+}
+
+void drmFreeVersion(drmVersionPtr v)
+{
+    if (!v) return;
+    if (v->name) drmFree(v->name);
+    if (v->date) drmFree(v->date);
+    if (v->desc) drmFree(v->desc);
+    drmFree(v);
+}
+
+static void drmFreeKernelVersion(drm_version_t *v)
+{
+    if (!v) return;
+    if (v->name) drmFree(v->name);
+    if (v->date) drmFree(v->date);
+    if (v->desc) drmFree(v->desc);
+    drmFree(v);
+}
+
+static void drmCopyVersion(drmVersionPtr d, drm_version_t *s)
+{
+    d->version_major      = s->version_major;
+    d->version_minor      = s->version_minor;
+    d->version_patchlevel = s->version_patchlevel;
+    d->name_len           = s->name_len;
+    d->name               = drmStrdup(s->name);
+    d->date_len           = s->date_len;
+    d->date               = drmStrdup(s->date);
+    d->desc_len           = s->desc_len;
+    d->desc               = drmStrdup(s->desc);
+}
+
+/* drmVersion obtains the version information via an ioctl.  Similar
+ * information is available via /proc/drm. */
+
+drmVersionPtr drmGetVersion(int fd)
+{
+    drmVersionPtr retval;
+    drm_version_t *version = drmMalloc(sizeof(*version));
+
+				/* First, get the lengths */
+    version->name_len    = 0;
+    version->name        = NULL;
+    version->date_len    = 0;
+    version->date        = NULL;
+    version->desc_len    = 0;
+    version->desc        = NULL;
+    
+    if (ioctl(fd, DRM_IOCTL_VERSION, version)) {
+	drmFreeKernelVersion(version);
+	return NULL;
+    }
+
+				/* Now, allocate space and get the data */
+    if (version->name_len)
+	version->name    = drmMalloc(version->name_len + 1);
+    if (version->date_len)
+	version->date    = drmMalloc(version->date_len + 1);
+    if (version->desc_len)
+	version->desc    = drmMalloc(version->desc_len + 1);
+    
+    if (ioctl(fd, DRM_IOCTL_VERSION, version)) {
+	drmFreeKernelVersion(version);
+	return NULL;
+    }
+
+				/* The results might not be null-terminated
+                                   strings, so terminate them. */
+
+    if (version->name_len) version->name[version->name_len] = '\0';
+    if (version->date_len) version->date[version->date_len] = '\0';
+    if (version->desc_len) version->desc[version->desc_len] = '\0';
+
+				/* Now, copy it all back into the
+                                   client-visible data structure... */
+    retval = drmMalloc(sizeof(*retval));
+    drmCopyVersion(retval, version);
+    drmFreeKernelVersion(version);
+    return retval;
+}
+
+void drmFreeBusid(const char *busid)
+{
+    drmFree((void *)busid);
+}
+
+char *drmGetBusid(int fd)
+{
+    drm_unique_t u;
+
+    u.unique_len = 0;
+    u.unique     = NULL;
+
+    if (ioctl(fd, DRM_IOCTL_GET_UNIQUE, &u)) return NULL;
+    u.unique = drmMalloc(u.unique_len + 1);
+    if (ioctl(fd, DRM_IOCTL_GET_UNIQUE, &u)) return NULL;
+    u.unique[u.unique_len] = '\0';
+    return u.unique;
+}
+
+int drmSetBusid(int fd, const char *busid)
+{
+    drm_unique_t u;
+
+    u.unique     = (char *)busid;
+    u.unique_len = strlen(busid);
+
+    if (ioctl(fd, DRM_IOCTL_SET_UNIQUE, &u)) return -errno;
+    return 0;
+}
+
+int drmGetMagic(int fd, drmMagicPtr magic)
+{
+    drm_auth_t auth;
+
+    *magic = 0;
+    if (ioctl(fd, DRM_IOCTL_GET_MAGIC, &auth)) return -errno;
+    *magic = auth.magic;
+    return 0;
+}
+
+int drmAuthMagic(int fd, drmMagic magic)
+{
+    drm_auth_t auth;
+
+    auth.magic = magic;
+    if (ioctl(fd, DRM_IOCTL_AUTH_MAGIC, &auth)) return -errno;
+    return 0;
+}
+
+int drmAddMap(int fd,
+	      drmHandle offset,
+	      drmSize size,
+	      drmMapType type,
+	      drmMapFlags flags,
+	      drmHandlePtr handle)
+{
+    drm_map_t map;
+
+    map.offset  = offset;
+    map.size    = size;
+    map.handle  = 0;
+    map.type    = type;
+    map.flags   = flags;
+    if (ioctl(fd, DRM_IOCTL_ADD_MAP, &map)) return -errno;
+    if (handle) *handle = (drmHandle)map.handle;
+    return 0;
+}
+
+int drmAddBufs(int fd, int count, int size, drmBufDescFlags flags,
+	       int agp_offset)
+{
+    drm_buf_desc_t request;
+    
+    request.count     = count;
+    request.size      = size;
+    request.low_mark  = 0;
+    request.high_mark = 0;
+    request.flags     = flags;
+    request.agp_start = agp_offset;
+   
+    if (ioctl(fd, DRM_IOCTL_ADD_BUFS, &request)) return -errno;
+    return request.count;
+}
+
+int drmMarkBufs(int fd, double low, double high)
+{
+    drm_buf_info_t info;
+    int            i;
+
+    info.count = 0;
+    info.list  = NULL;
+
+    if (ioctl(fd, DRM_IOCTL_INFO_BUFS, &info)) return -EINVAL;
+
+    if (!info.count) return -EINVAL;
+    
+    if (!(info.list = drmMalloc(info.count * sizeof(*info.list))))
+	return -ENOMEM;
+	
+    if (ioctl(fd, DRM_IOCTL_INFO_BUFS, &info)) {
+	int retval = -errno;
+	drmFree(info.list);
+	return retval;
+    }
+    
+    for (i = 0; i < info.count; i++) {
+	info.list[i].low_mark  = low  * info.list[i].count;
+	info.list[i].high_mark = high * info.list[i].count;
+	if (ioctl(fd, DRM_IOCTL_MARK_BUFS, &info.list[i])) {
+	    int retval = -errno;
+	    drmFree(info.list);
+	    return retval;
+	}
+    }
+    drmFree(info.list);
+    
+    return 0;
+}
+
+int drmFreeBufs(int fd, int count, int *list)
+{
+    drm_buf_free_t request;
+
+    request.count = count;
+    request.list  = list;
+    if (ioctl(fd, DRM_IOCTL_FREE_BUFS, &request)) return -errno;
+    return 0;
+}
+
+int drmClose(int fd)
+{
+    unsigned long key    = drmGetKeyFromFd(fd);
+    drmHashEntry  *entry = drmGetEntry(fd);
+
+    drmHashDestroy(entry->tagTable);
+    entry->fd       = 0;
+    entry->f        = NULL;
+    entry->tagTable = NULL;
+
+    drmHashDelete(drmHashTable, key);
+    drmFree(entry);
+
+    return close(fd);
+}
+
+int drmMap(int fd,
+	   drmHandle handle,
+	   drmSize size,
+	   drmAddressPtr address)
+{
+    if (fd < 0) return -EINVAL;
+    *address = mmap(0, size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, handle);
+    if (*address == MAP_FAILED) return -errno;
+    return 0;
+}
+
+int drmUnmap(drmAddress address, drmSize size)
+{
+    return munmap(address, size);
+}
+
+drmBufInfoPtr drmGetBufInfo(int fd)
+{
+    drm_buf_info_t info;
+    drmBufInfoPtr  retval;
+    int            i;
+
+    info.count = 0;
+    info.list  = NULL;
+
+    if (ioctl(fd, DRM_IOCTL_INFO_BUFS, &info)) return NULL;
+
+    if (info.count) {
+	if (!(info.list = drmMalloc(info.count * sizeof(*info.list))))
+	    return NULL;
+	
+	if (ioctl(fd, DRM_IOCTL_INFO_BUFS, &info)) {
+	    drmFree(info.list);
+	    return NULL;
+	}
+				/* Now, copy it all back into the
+                                   client-visible data structure... */
+	retval = drmMalloc(sizeof(*retval));
+	retval->count = info.count;
+	retval->list  = drmMalloc(info.count * sizeof(*retval->list));
+	for (i = 0; i < info.count; i++) {
+	    retval->list[i].count     = info.list[i].count;
+	    retval->list[i].size      = info.list[i].size;
+	    retval->list[i].low_mark  = info.list[i].low_mark;
+	    retval->list[i].high_mark = info.list[i].high_mark;
+	}
+	drmFree(info.list);
+	return retval;
+    }
+    return NULL;
+}
+
+drmBufMapPtr drmMapBufs(int fd)
+{
+    drm_buf_map_t bufs;
+    drmBufMapPtr  retval;
+    int           i;
+    
+    bufs.count = 0;
+    bufs.list  = NULL;
+    if (ioctl(fd, DRM_IOCTL_MAP_BUFS, &bufs)) return NULL;
+
+    if (bufs.count) {
+	if (!(bufs.list = drmMalloc(bufs.count * sizeof(*bufs.list))))
+	    return NULL;
+
+	if (ioctl(fd, DRM_IOCTL_MAP_BUFS, &bufs)) {
+	    drmFree(bufs.list);
+	    return NULL;
+	}
+				/* Now, copy it all back into the
+                                   client-visible data structure... */
+	retval = drmMalloc(sizeof(*retval));
+	retval->count = bufs.count;
+	retval->list  = drmMalloc(bufs.count * sizeof(*retval->list));
+	for (i = 0; i < bufs.count; i++) {
+	    retval->list[i].idx     = bufs.list[i].idx;
+	    retval->list[i].total   = bufs.list[i].total;
+	    retval->list[i].used    = 0;
+	    retval->list[i].address = bufs.list[i].address;
+	}
+	return retval;
+    }
+    return NULL;
+}
+
+int drmUnmapBufs(drmBufMapPtr bufs)
+{
+    int i;
+    
+    for (i = 0; i < bufs->count; i++) {
+	munmap(bufs->list[i].address, bufs->list[i].total);
+    }
+    return 0;
+}
+
+int drmDMA(int fd, drmDMAReqPtr request)
+{
+    drm_dma_t dma;
+
+				/* Copy to hidden structure */
+    dma.context         = request->context;
+    dma.send_count      = request->send_count;
+    dma.send_indices    = request->send_list;
+    dma.send_sizes      = request->send_sizes;
+    dma.flags           = request->flags;
+    dma.request_count   = request->request_count;
+    dma.request_size    = request->request_size;
+    dma.request_indices = request->request_list;
+    dma.request_sizes   = request->request_sizes;
+    if (ioctl(fd, DRM_IOCTL_DMA, &dma)) return -errno;
+    request->granted_count = dma.granted_count;
+    
+    return 0;
+}
+
+int drmGetLock(int fd, drmContext context, drmLockFlags flags)
+{
+    drm_lock_t lock;
+
+    lock.context = context;
+    lock.flags   = 0;
+    if (flags & DRM_LOCK_READY)      lock.flags |= _DRM_LOCK_READY;
+    if (flags & DRM_LOCK_QUIESCENT)  lock.flags |= _DRM_LOCK_QUIESCENT;
+    if (flags & DRM_LOCK_FLUSH)      lock.flags |= _DRM_LOCK_FLUSH;
+    if (flags & DRM_LOCK_FLUSH_ALL)  lock.flags |= _DRM_LOCK_FLUSH_ALL;
+    if (flags & DRM_HALT_ALL_QUEUES) lock.flags |= _DRM_HALT_ALL_QUEUES;
+    if (flags & DRM_HALT_CUR_QUEUES) lock.flags |= _DRM_HALT_CUR_QUEUES;
+    
+    while (ioctl(fd, DRM_IOCTL_LOCK, &lock))
+	;
+    return 0;
+}
+
+int drmUnlock(int fd, drmContext context)
+{
+    drm_lock_t lock;
+
+    lock.context = context;
+    lock.flags   = 0;
+    return ioctl(fd, DRM_IOCTL_UNLOCK, &lock);
+}
+
+drmContextPtr drmGetReservedContextList(int fd, int *count)
+{
+    drm_ctx_res_t res;
+    drm_ctx_t     *list;
+    drmContextPtr retval;
+    int           i;
+
+    res.count    = 0;
+    res.contexts = NULL;
+    if (ioctl(fd, DRM_IOCTL_RES_CTX, &res)) return NULL;
+
+    if (!res.count) return NULL;
+
+    if (!(list   = drmMalloc(res.count * sizeof(*list)))) return NULL;
+    if (!(retval = drmMalloc(res.count * sizeof(*retval)))) {
+	drmFree(list);
+	return NULL;
+    }
+
+    res.contexts = list;
+    if (ioctl(fd, DRM_IOCTL_RES_CTX, &res)) return NULL;
+
+    for (i = 0; i < res.count; i++) retval[i] = list[i].handle;
+    drmFree(list);
+
+    *count = res.count;
+    return retval;
+}
+
+void drmFreeReservedContextList(drmContextPtr pt)
+{
+    drmFree(pt);
+}
+
+int drmCreateContext(int fd, drmContextPtr handle)
+{
+    drm_ctx_t ctx;
+
+    ctx.flags = 0;	/* Modified with functions below */
+    if (ioctl(fd, DRM_IOCTL_ADD_CTX, &ctx)) return -errno;
+    *handle = ctx.handle;
+    return 0;
+}
+
+int drmSwitchToContext(int fd, drmContext context)
+{
+    drm_ctx_t ctx;
+
+    ctx.handle = context;
+    if (ioctl(fd, DRM_IOCTL_SWITCH_CTX, &ctx)) return -errno;
+    return 0;
+}
+
+int drmSetContextFlags(int fd, drmContext context, drmContextFlags flags)
+{
+    drm_ctx_t ctx;
+
+				/* Context preserving means that no context
+                                   switched are done between DMA buffers
+                                   from one context and the next.  This is
+                                   suitable for use in the X server (which
+                                   promises to maintain hardware context,
+                                   or in the client-side library when
+                                   buffers are swapped on behalf of two
+                                   threads. */
+    ctx.handle = context;
+    ctx.flags  = 0;
+    if (flags & DRM_CONTEXT_PRESERVED) ctx.flags |= _DRM_CONTEXT_PRESERVED;
+    if (flags & DRM_CONTEXT_2DONLY)    ctx.flags |= _DRM_CONTEXT_2DONLY;
+    if (ioctl(fd, DRM_IOCTL_MOD_CTX, &ctx)) return -errno;
+    return 0;
+}
+
+int drmGetContextFlags(int fd, drmContext context, drmContextFlagsPtr flags)
+{
+    drm_ctx_t ctx;
+
+    ctx.handle = context;
+    if (ioctl(fd, DRM_IOCTL_GET_CTX, &ctx)) return -errno;
+    *flags = 0;
+    if (ctx.flags & _DRM_CONTEXT_PRESERVED) *flags |= DRM_CONTEXT_PRESERVED;
+    if (ctx.flags & _DRM_CONTEXT_2DONLY)    *flags |= DRM_CONTEXT_2DONLY;
+    return 0;
+}
+    
+int drmDestroyContext(int fd, drmContext handle)
+{
+    drm_ctx_t ctx;
+    ctx.handle = handle;
+    if (ioctl(fd, DRM_IOCTL_RM_CTX, &ctx)) return -errno;
+    return 0;
+}
+
+int drmCreateDrawable(int fd, drmDrawablePtr handle)
+{
+    drm_draw_t draw;
+    if (ioctl(fd, DRM_IOCTL_ADD_DRAW, &draw)) return -errno;
+    *handle = draw.handle;
+    return 0;
+}
+
+int drmDestroyDrawable(int fd, drmDrawable handle)
+{
+    drm_draw_t draw;
+    draw.handle = handle;
+    if (ioctl(fd, DRM_IOCTL_RM_DRAW, &draw)) return -errno;
+    return 0;
+}
+
+int drmAgpAcquire(int fd)
+{
+    if (ioctl(fd, DRM_IOCTL_AGP_ACQUIRE, NULL)) return -errno;
+    return 0;
+}
+
+int drmAgpRelease(int fd)
+{
+    if (ioctl(fd, DRM_IOCTL_AGP_RELEASE, NULL)) return -errno;
+    return 0;
+}
+
+int drmAgpEnable(int fd, unsigned long mode)
+{
+    drm_agp_mode_t m;
+
+    m.mode = mode;
+    if (ioctl(fd, DRM_IOCTL_AGP_ENABLE, &m)) return -errno;
+    return 0;
+}
+
+int drmAgpAlloc(int fd, unsigned long size, unsigned long type,
+		unsigned long *address, unsigned long *handle)
+{
+    drm_agp_buffer_t b;
+    *handle = 0;
+    b.size   = size;
+    b.handle = 0;
+    b.type   = type;
+    if (ioctl(fd, DRM_IOCTL_AGP_ALLOC, &b)) return -errno;
+    if (address != 0UL) *address = b.physical;
+    *handle = b.handle;
+    return 0;
+}
+
+int drmAgpFree(int fd, unsigned long handle)
+{
+    drm_agp_buffer_t b;
+
+    b.size   = 0;
+    b.handle = handle;
+    if (ioctl(fd, DRM_IOCTL_AGP_FREE, &b)) return -errno;
+    return 0;
+}
+
+int drmAgpBind(int fd, unsigned long handle, unsigned long offset)
+{
+    drm_agp_binding_t b;
+
+    b.handle = handle;
+    b.offset = offset;
+    if (ioctl(fd, DRM_IOCTL_AGP_BIND, &b)) return -errno;
+    return 0;
+}
+
+int drmAgpUnbind(int fd, unsigned long handle)
+{
+    drm_agp_binding_t b;
+
+    b.handle = handle;
+    b.offset = 0;
+    if (ioctl(fd, DRM_IOCTL_AGP_UNBIND, &b)) return -errno;
+    return 0;
+}
+
+int drmAgpVersionMajor(int fd)
+{
+    drm_agp_info_t i;
+
+    if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) return -errno;
+    return i.agp_version_major;
+}
+
+int drmAgpVersionMinor(int fd)
+{
+    drm_agp_info_t i;
+
+    if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) return -errno;
+    return i.agp_version_minor;
+}
+
+unsigned long drmAgpGetMode(int fd)
+{
+    drm_agp_info_t i;
+
+    if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) return 0;
+    return i.mode;
+}
+
+unsigned long drmAgpBase(int fd)
+{
+    drm_agp_info_t i;
+
+    if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) return 0;
+    return i.aperture_base;
+}
+
+unsigned long drmAgpSize(int fd)
+{
+    drm_agp_info_t i;
+
+    if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) return 0;
+    return i.aperture_size;
+}
+
+unsigned long drmAgpMemoryUsed(int fd)
+{
+    drm_agp_info_t i;
+
+    if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) return 0;
+    return i.memory_used;
+}
+
+unsigned long drmAgpMemoryAvail(int fd)
+{
+    drm_agp_info_t i;
+
+    if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) return 0;
+    return i.memory_allowed;
+}
+
+unsigned int drmAgpVendorId(int fd)
+{
+    drm_agp_info_t i;
+
+    if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) return 0;
+    return i.id_vendor;
+}
+
+unsigned int drmAgpDeviceId(int fd)
+{
+    drm_agp_info_t i;
+
+    if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) return 0;
+    return i.id_device;
+}
+
+int drmError(int err, const char *label)
+{
+    switch (err) {
+    case DRM_ERR_NO_DEVICE: fprintf(stderr, "%s: no device\n", label);   break;
+    case DRM_ERR_NO_ACCESS: fprintf(stderr, "%s: no access\n", label);   break;
+    case DRM_ERR_NOT_ROOT:  fprintf(stderr, "%s: not root\n", label);    break;
+    case DRM_ERR_INVALID:   fprintf(stderr, "%s: invalid args\n", label);break;
+    default:
+	if (err < 0) err = -err;
+	fprintf( stderr, "%s: error %d (%s)\n", label, err, strerror(err) );
+	break;
+    }
+
+    return 1;
+}
+
+int drmCtlInstHandler(int fd, int irq)
+{
+    drm_control_t ctl;
+
+    ctl.func  = DRM_INST_HANDLER;
+    ctl.irq   = irq;
+    if (ioctl(fd, DRM_IOCTL_CONTROL, &ctl)) return -errno;
+    return 0;
+}
+
+int drmCtlUninstHandler(int fd)
+{
+    drm_control_t ctl;
+
+    ctl.func  = DRM_UNINST_HANDLER;
+    ctl.irq   = 0;
+    if (ioctl(fd, DRM_IOCTL_CONTROL, &ctl)) return -errno;
+    return 0;
+}
+
+int drmFinish(int fd, int context, drmLockFlags flags)
+{
+    drm_lock_t lock;
+
+    lock.context = context;
+    lock.flags   = 0;
+    if (flags & DRM_LOCK_READY)      lock.flags |= _DRM_LOCK_READY;
+    if (flags & DRM_LOCK_QUIESCENT)  lock.flags |= _DRM_LOCK_QUIESCENT;
+    if (flags & DRM_LOCK_FLUSH)      lock.flags |= _DRM_LOCK_FLUSH;
+    if (flags & DRM_LOCK_FLUSH_ALL)  lock.flags |= _DRM_LOCK_FLUSH_ALL;
+    if (flags & DRM_HALT_ALL_QUEUES) lock.flags |= _DRM_HALT_ALL_QUEUES;
+    if (flags & DRM_HALT_CUR_QUEUES) lock.flags |= _DRM_HALT_CUR_QUEUES;
+    if (ioctl(fd, DRM_IOCTL_FINISH, &lock)) return -errno;
+    return 0;
+}
+
+int drmGetInterruptFromBusID(int fd, int busnum, int devnum, int funcnum)
+{
+    drm_irq_busid_t p;
+
+    p.busnum  = busnum;
+    p.devnum  = devnum;
+    p.funcnum = funcnum;
+    if (ioctl(fd, DRM_IOCTL_IRQ_BUSID, &p)) return -errno;
+    return p.irq;
+}
+
+int drmAddContextTag(int fd, drmContext context, void *tag)
+{
+    drmHashEntry  *entry = drmGetEntry(fd);
+
+    if (drmHashInsert(entry->tagTable, context, tag)) {
+	drmHashDelete(entry->tagTable, context);
+	drmHashInsert(entry->tagTable, context, tag);
+    }
+    return 0;
+}
+
+int drmDelContextTag(int fd, drmContext context)
+{
+    drmHashEntry  *entry = drmGetEntry(fd);
+
+    return drmHashDelete(entry->tagTable, context);
+}
+
+void *drmGetContextTag(int fd, drmContext context)
+{
+    drmHashEntry  *entry = drmGetEntry(fd);
+    void          *value;
+    
+    if (drmHashLookup(entry->tagTable, context, &value)) return NULL;
+
+    return value;
+}
+
+#if defined(XFree86Server) || defined(DRM_USE_MALLOC)
+static void drmSIGIOHandler(int interrupt, void *arg)
+{
+    unsigned long key;
+    void          *value;
+    ssize_t       count;
+    drm_ctx_t     ctx;
+    typedef void  (*_drmCallback)(int, void *, void *);
+    char          buf[256];
+    drmContext    old;
+    drmContext    new;
+    void          *oldctx;
+    void          *newctx;
+    char          *pt;
+    drmHashEntry  *entry;
+
+    if (!drmHashTable) return;
+    if (drmHashFirst(drmHashTable, &key, &value)) {
+	entry = value;
+	do {
+#if 1
+	    fprintf(stderr, "Trying %d\n", entry->fd);
+#endif
+	    if ((count = read(entry->fd, buf, sizeof(buf)))) {
+		buf[count] = '\0';
+#if 1
+		fprintf(stderr, "Got %s\n", buf);
+#endif
+		
+		for (pt = buf; *pt != ' '; ++pt); /* Find first space */
+		++pt;
+		old    = strtol(pt, &pt, 0);
+		new    = strtol(pt, NULL, 0);
+		oldctx = drmGetContextTag(entry->fd, old);
+		newctx = drmGetContextTag(entry->fd, new);
+#if 1
+		fprintf(stderr, "%d %d %p %p\n", old, new, oldctx, newctx);
+#endif
+		((_drmCallback)entry->f)(entry->fd, oldctx, newctx);
+		ctx.handle = new;
+		ioctl(entry->fd, DRM_IOCTL_NEW_CTX, &ctx);
+	    }
+	} while (drmHashNext(drmHashTable, &key, &value));
+    }
+}
+
+int drmInstallSIGIOHandler(int fd, void (*f)(int, void *, void *))
+{
+    drmHashEntry     *entry;
+
+    entry     = drmGetEntry(fd);
+    entry->f  = f;
+
+    return xf86InstallSIGIOHandler(fd, drmSIGIOHandler, 0);
+}
+
+int drmRemoveSIGIOHandler(int fd)
+{
+    drmHashEntry     *entry = drmGetEntry(fd);
+
+    entry->f = NULL;
+    
+    return xf86RemoveSIGIOHandler(fd);
+}
+#endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/xf86drmHash.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/xf86drmHash.c
new file mode 100644
index 000000000..a3c9481b5
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/xf86drmHash.c
@@ -0,0 +1,435 @@
+/* xf86drmHash.c -- Small hash table support for integer -> integer mapping
+ * Created: Sun Apr 18 09:35:45 1999 by faith@precisioninsight.com
+ * Revised: Thu Jun  3 16:11:06 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmHash.c,v 1.3 1999/06/07 13:01:42 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmHash.c,v 1.1 1999/06/14 07:32:02 dawes Exp $
+ *
+ * DESCRIPTION
+ *
+ * This file contains a straightforward implementation of a fixed-sized
+ * hash table using self-organizing linked lists [Knuth73, pp. 398-399] for
+ * collision resolution.  There are two potentially interesting things
+ * about this implementation:
+ *
+ * 1) The table is power-of-two sized.  Prime sized tables are more
+ * traditional, but do not have a significant advantage over power-of-two
+ * sized table, especially when double hashing is not used for collision
+ * resolution.
+ *
+ * 2) The hash computation uses a table of random integers [Hanson97,
+ * pp. 39-41].
+ *
+ * FUTURE ENHANCEMENTS
+ *
+ * With a table size of 512, the current implementation is sufficient for a
+ * few hundred keys.  Since this is well above the expected size of the
+ * tables for which this implementation was designed, the implementation of
+ * dynamic hash tables was postponed until the need arises.  A common (and
+ * naive) approach to dynamic hash table implementation simply creates a
+ * new hash table when necessary, rehashes all the data into the new table,
+ * and destroys the old table.  The approach in [Larson88] is superior in
+ * two ways: 1) only a portion of the table is expanded when needed,
+ * distributing the expansion cost over several insertions, and 2) portions
+ * of the table can be locked, enabling a scalable thread-safe
+ * implementation.
+ *
+ * REFERENCES
+ *
+ * [Hanson97] David R. Hanson.  C Interfaces and Implementations:
+ * Techniques for Creating Reusable Software.  Reading, Massachusetts:
+ * Addison-Wesley, 1997.
+ *
+ * [Knuth73] Donald E. Knuth. The Art of Computer Programming.  Volume 3:
+ * Sorting and Searching.  Reading, Massachusetts: Addison-Wesley, 1973.
+ *
+ * [Larson88] Per-Ake Larson. "Dynamic Hash Tables".  CACM 31(4), April
+ * 1988, pp. 446-457.
+ *
+ */
+
+#define HASH_MAIN 0
+
+#if HASH_MAIN
+# include <stdio.h>
+# include <stdlib.h>
+#else
+# include "xf86drm.h"
+# ifdef XFree86LOADER
+#  include "xf86.h"
+#  include "xf86_ansic.h"
+# else
+#  include <stdio.h>
+#  include <stdlib.h>
+# endif
+#endif
+
+#define N(x)  drm##x
+
+#define HASH_MAGIC 0xdeadbeef
+#define HASH_DEBUG 0
+#define HASH_SIZE  512		/* Good for about 100 entries */
+				/* If you change this value, you probably
+                                   have to change the HashHash hashing
+                                   function! */
+
+#if HASH_MAIN
+#define HASH_ALLOC malloc
+#define HASH_FREE  free
+#define HASH_RANDOM_DECL
+#define HASH_RANDOM_INIT(seed)  srandom(seed)
+#define HASH_RANDOM             random()
+#else
+#define HASH_ALLOC drmMalloc
+#define HASH_FREE  drmFree
+#define HASH_RANDOM_DECL        void *state
+#define HASH_RANDOM_INIT(seed)  state = drmRandomCreate(seed)
+#define HASH_RANDOM             drmRandom(state)
+
+#endif
+
+typedef struct HashBucket {
+    unsigned long     key;
+    void              *value;
+    struct HashBucket *next;
+} HashBucket, *HashBucketPtr;
+
+typedef struct HashTable {
+    unsigned long    magic;
+    unsigned long    entries;
+    unsigned long    hits;	/* At top of linked list */
+    unsigned long    partials;	/* Not at top of linked list */
+    unsigned long    misses;	/* Not in table */
+    HashBucketPtr    buckets[HASH_SIZE];
+    int              p0;
+    HashBucketPtr    p1;
+} HashTable, *HashTablePtr;
+
+#if HASH_MAIN
+extern void *N(HashCreate)(void);
+extern int  N(HashDestroy)(void *t);
+extern int  N(HashLookup)(void *t, unsigned long key, unsigned long *value);
+extern int  N(HashInsert)(void *t, unsigned long key, unsigned long value);
+extern int  N(HashDelete)(void *t, unsigned long key);
+#endif
+
+static unsigned long HashHash(unsigned long key)
+{
+    unsigned long        hash = 0;
+    unsigned long        tmp  = key;
+    static int           init = 0;
+    static unsigned long scatter[256];
+    int                  i;
+
+    if (!init) {
+	HASH_RANDOM_DECL;
+	HASH_RANDOM_INIT(37);
+	for (i = 0; i < 256; i++) scatter[i] = HASH_RANDOM;
+	++init;
+    }
+
+    while (tmp) {
+	hash = (hash << 1) + scatter[tmp & 0xff];
+	tmp >>= 8;
+    }
+
+    hash %= HASH_SIZE;
+#if HASH_DEBUG
+    printf( "Hash(%d) = %d\n", key, hash);
+#endif
+    return hash;
+}
+
+void *N(HashCreate)(void)
+{
+    HashTablePtr table;
+    int          i;
+
+    table           = HASH_ALLOC(sizeof(*table));
+    if (!table) return NULL;
+    table->magic    = HASH_MAGIC;
+    table->entries  = 0;
+    table->hits     = 0;
+    table->partials = 0;
+    table->misses   = 0;
+
+    for (i = 0; i < HASH_SIZE; i++) table->buckets[i] = NULL;
+    return table;
+}
+
+int N(HashDestroy)(void *t)
+{
+    HashTablePtr  table = (HashTablePtr)t;
+    HashBucketPtr bucket;
+    HashBucketPtr next;
+    int           i;
+
+    if (table->magic != HASH_MAGIC) return -1; /* Bad magic */
+    
+    for (i = 0; i < HASH_SIZE; i++) {
+	for (bucket = table->buckets[i]; bucket;) {
+	    next = bucket->next;
+	    HASH_FREE(bucket);
+	    bucket = next;
+	}
+    }
+    HASH_FREE(table);
+    return 0;
+}
+
+/* Find the bucket and organize the list so that this bucket is at the
+   top. */
+
+static HashBucketPtr HashFind(HashTablePtr table,
+			      unsigned long key, unsigned long *h)
+{
+    unsigned long hash = HashHash(key);
+    HashBucketPtr prev = NULL;
+    HashBucketPtr bucket;
+
+    if (h) *h = hash;
+
+    for (bucket = table->buckets[hash]; bucket; bucket = bucket->next) {
+	if (bucket->key == key) {
+	    if (prev) {
+				/* Organize */
+		prev->next           = bucket->next;
+		bucket->next         = table->buckets[hash];
+		table->buckets[hash] = bucket;
+		++table->partials;
+	    } else {
+		++table->hits;
+	    }
+	    return bucket;
+	}
+	prev = bucket;
+    }
+    ++table->misses;
+    return NULL;
+}
+
+int N(HashLookup)(void *t, unsigned long key, void **value)
+{
+    HashTablePtr  table = (HashTablePtr)t;
+    HashBucketPtr bucket;
+
+    if (table->magic != HASH_MAGIC) return -1; /* Bad magic */
+    
+    bucket = HashFind(table, key, NULL);
+    if (!bucket) return 1;	/* Not found */
+    *value = bucket->value;
+    return 0;			/* Found */
+}
+
+int N(HashInsert)(void *t, unsigned long key, void *value)
+{
+    HashTablePtr  table = (HashTablePtr)t;
+    HashBucketPtr bucket;
+    unsigned long hash;
+
+    if (table->magic != HASH_MAGIC) return -1; /* Bad magic */
+    
+    if (HashFind(table, key, &hash)) return 1; /* Already in table */
+
+    bucket               = HASH_ALLOC(sizeof(*bucket));
+    if (!bucket) return -1;	/* Error */
+    bucket->key          = key;
+    bucket->value        = value;
+    bucket->next         = table->buckets[hash];
+    table->buckets[hash] = bucket;
+#if HASH_DEBUG
+    printf("Inserted %d at %d/%p\n", key, hash, bucket);
+#endif
+    return 0;			/* Added to table */
+}
+
+int N(HashDelete)(void *t, unsigned long key)
+{
+    HashTablePtr  table = (HashTablePtr)t;
+    unsigned long hash;
+    HashBucketPtr bucket;
+
+    if (table->magic != HASH_MAGIC) return -1; /* Bad magic */
+    
+    bucket = HashFind(table, key, &hash);
+
+    if (!bucket) return 1;	/* Not found */
+
+    table->buckets[hash] = bucket->next;
+    HASH_FREE(bucket);
+    return 0;
+}
+
+int N(HashNext)(void *t, unsigned long *key, void **value)
+{
+    HashTablePtr  table = (HashTablePtr)t;
+    
+    for (; table->p0 < HASH_SIZE;
+	 ++table->p0, table->p1 = table->buckets[table->p0]) {
+	if (table->p1) {
+	    *key       = table->p1->key;
+	    *value     = table->p1->value;
+	    table->p1  = table->p1->next;
+	    return 1;
+	}
+    }
+    return 0;
+}
+
+int N(HashFirst)(void *t, unsigned long *key, void **value)
+{
+    HashTablePtr  table = (HashTablePtr)t;
+    
+    if (table->magic != HASH_MAGIC) return -1; /* Bad magic */
+
+    table->p0 = 0;
+    table->p1 = table->buckets[0];
+    return N(HashNext)(table, key, value);
+}
+
+#if HASH_MAIN
+#define DIST_LIMIT 10
+static int dist[DIST_LIMIT];
+
+static void clear_dist(void) {
+    int i;
+
+    for (i = 0; i < DIST_LIMIT; i++) dist[i] = 0;
+}
+
+static int count_entries(HashBucketPtr bucket)
+{
+    int count = 0;
+
+    for (; bucket; bucket = bucket->next) ++count;
+    return count;
+}
+
+static void update_dist(int count)
+{
+    if (count >= DIST_LIMIT) ++dist[DIST_LIMIT-1];
+    else                     ++dist[count];
+}
+
+static void compute_dist(HashTablePtr table)
+{
+    int           i;
+    HashBucketPtr bucket;
+    
+    printf("Entries = %ld, hits = %ld, partials = %ld, misses = %ld\n",
+	   table->entries, table->hits, table->partials, table->misses);
+    clear_dist();
+    for (i = 0; i < HASH_SIZE; i++) {
+	bucket = table->buckets[i];
+	update_dist(count_entries(bucket));
+    }
+    for (i = 0; i < DIST_LIMIT; i++) {
+	if (i != DIST_LIMIT-1) printf("%5d %10d\n", i, dist[i]);
+	else                   printf("other %10d\n", dist[i]);
+    }
+}
+
+static void check_table(HashTablePtr table,
+			unsigned long key, unsigned long value)
+{
+    unsigned long retval  = 0;
+    int           retcode = N(HashLookup)(table, key, &retval);
+    
+    switch (retcode) {
+    case -1:
+	printf("Bad magic = 0x%08lx:"
+	       " key = %lu, expected = %lu, returned = %lu\n",
+	       table->magic, key, value, retval);
+	break;
+    case 1:
+	printf("Not found: key = %lu, expected = %lu returned = %lu\n",
+	       key, value, retval);
+	break;
+    case 0:
+	if (value != retval)
+	    printf("Bad value: key = %lu, expected = %lu, returned = %lu\n",
+		   key, value, retval);
+	break;
+    default:
+	printf("Bad retcode = %d: key = %lu, expected = %lu, returned = %lu\n",
+	       retcode, key, value, retval);
+	break;
+    }
+}
+
+int main(void)
+{
+    HashTablePtr table;
+    int          i;
+
+    printf("\n***** 256 consecutive integers ****\n");
+    table = N(HashCreate)();
+    for (i = 0; i < 256; i++) N(HashInsert)(table, i, i);
+    for (i = 0; i < 256; i++) check_table(table, i, i);
+    for (i = 256; i >= 0; i--) check_table(table, i, i);
+    compute_dist(table);
+    N(HashDestroy)(table);
+    
+    printf("\n***** 1024 consecutive integers ****\n");
+    table = N(HashCreate)();
+    for (i = 0; i < 1024; i++) N(HashInsert)(table, i, i);
+    for (i = 0; i < 1024; i++) check_table(table, i, i);
+    for (i = 1024; i >= 0; i--) check_table(table, i, i);
+    compute_dist(table);
+    N(HashDestroy)(table);
+    
+    printf("\n***** 1024 consecutive page addresses (4k pages) ****\n");
+    table = N(HashCreate)();
+    for (i = 0; i < 1024; i++) N(HashInsert)(table, i*4096, i);
+    for (i = 0; i < 1024; i++) check_table(table, i*4096, i);
+    for (i = 1024; i >= 0; i--) check_table(table, i*4096, i);
+    compute_dist(table);
+    N(HashDestroy)(table);
+    
+    printf("\n***** 1024 random integers ****\n");
+    table = N(HashCreate)();
+    srandom(0xbeefbeef);
+    for (i = 0; i < 1024; i++) N(HashInsert)(table, random(), i);
+    srandom(0xbeefbeef);
+    for (i = 0; i < 1024; i++) check_table(table, random(), i);
+    srandom(0xbeefbeef);
+    for (i = 0; i < 1024; i++) check_table(table, random(), i);
+    compute_dist(table);
+    N(HashDestroy)(table);
+
+    printf("\n***** 5000 random integers ****\n");
+    table = N(HashCreate)();
+    srandom(0xbeefbeef);
+    for (i = 0; i < 5000; i++) N(HashInsert)(table, random(), i);
+    srandom(0xbeefbeef);
+    for (i = 0; i < 5000; i++) check_table(table, random(), i);
+    srandom(0xbeefbeef);
+    for (i = 0; i < 5000; i++) check_table(table, random(), i);
+    compute_dist(table);
+    N(HashDestroy)(table);
+    
+    return 0;
+}
+#endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/xf86drmI810.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/xf86drmI810.c
new file mode 100644
index 000000000..067c6376e
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/xf86drmI810.c
@@ -0,0 +1,86 @@
+#ifdef XFree86Server
+# include "xf86.h"
+# include "xf86_OSproc.h"
+# include "xf86_ansic.h"
+# include "xf86Priv.h"
+# define _DRM_MALLOC xalloc
+# define _DRM_FREE   xfree
+# ifndef XFree86LOADER
+#  include <sys/stat.h>
+#  include <sys/mman.h>
+# endif
+#else
+# include <stdio.h>
+# include <stdlib.h>
+# include <unistd.h>
+# include <string.h>
+# include <ctype.h>
+# include <fcntl.h>
+# include <errno.h>
+# include <signal.h>
+# include <sys/types.h>
+# include <sys/stat.h>
+# include <sys/ioctl.h>
+# include <sys/mman.h>
+# include <sys/time.h>
+# ifdef DRM_USE_MALLOC
+#  define _DRM_MALLOC malloc
+#  define _DRM_FREE   free
+extern int xf86InstallSIGIOHandler(int fd, void (*f)(int, void *), void *);
+extern int xf86RemoveSIGIOHandler(int fd);
+# else
+#  include <Xlibint.h>
+#  define _DRM_MALLOC Xmalloc
+#  define _DRM_FREE   Xfree
+# endif
+#endif
+
+/* Not all systems have MAP_FAILED defined */
+#ifndef MAP_FAILED
+#define MAP_FAILED ((void *)-1)
+#endif
+
+#include "xf86drm.h"
+#include "xf86drmI810.h"
+#include "drm.h"
+
+Bool drmI810CleanupDma(int driSubFD)
+{
+   drm_i810_init_t init;
+   
+   memset(&init, 0, sizeof(drm_i810_init_t));
+   init.func = I810_CLEANUP_DMA;
+   
+   if(ioctl(driSubFD, DRM_IOCTL_I810_INIT, &init)) {
+      return FALSE;
+   }
+   
+   return TRUE;
+}
+
+Bool drmI810InitDma(int driSubFD, drmI810Init *info)
+{
+   drm_i810_init_t init;
+   
+   memset(&init, 0, sizeof(drm_i810_init_t));
+
+   init.func = I810_INIT_DMA;
+   init.ring_map_idx = info->ring_map_idx;
+   init.buffer_map_idx = info->buffer_map_idx;
+   init.ring_start = info->start;
+   init.ring_end = info->end;
+   init.ring_size = info->size;
+   init.sarea_priv_offset = info->sarea_off;
+   init.front_offset = info->front_offset;
+   init.back_offset = info->back_offset;
+   init.depth_offset = info->depth_offset;
+   init.w = info->w;
+   init.h = info->h;
+   init.pitch = info->pitch;
+   init.pitch_bits = info->pitch_bits;
+
+   if(ioctl(driSubFD, DRM_IOCTL_I810_INIT, &init)) {
+      return FALSE;
+   }
+   return TRUE;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/xf86drmMga.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/xf86drmMga.c
new file mode 100644
index 000000000..aa3a6c745
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/xf86drmMga.c
@@ -0,0 +1,116 @@
+#ifdef XFree86Server
+# include "xf86.h"
+# include "xf86_OSproc.h"
+# include "xf86_ansic.h"
+# include "xf86Priv.h"
+# define _DRM_MALLOC xalloc
+# define _DRM_FREE   xfree
+# ifndef XFree86LOADER
+#  include <sys/stat.h>
+#  include <sys/mman.h>
+# endif
+#else
+# include <stdio.h>
+# include <stdlib.h>
+# include <unistd.h>
+# include <string.h>
+# include <ctype.h>
+# include <fcntl.h>
+# include <errno.h>
+# include <signal.h>
+# include <sys/types.h>
+# include <sys/stat.h>
+# include <sys/ioctl.h>
+# include <sys/mman.h>
+# include <sys/time.h>
+# ifdef DRM_USE_MALLOC
+#  define _DRM_MALLOC malloc
+#  define _DRM_FREE   free
+extern int xf86InstallSIGIOHandler(int fd, void (*f)(int, void *), void *);
+extern int xf86RemoveSIGIOHandler(int fd);
+# else
+#  include <Xlibint.h>
+#  define _DRM_MALLOC Xmalloc
+#  define _DRM_FREE   Xfree
+# endif
+#endif
+
+/* Not all systems have MAP_FAILED defined */
+#ifndef MAP_FAILED
+#define MAP_FAILED ((void *)-1)
+#endif
+
+#include "xf86drm.h"
+#include "xf86drmMga.h"
+#include "drm.h"
+
+Bool drmMgaCleanupDma(int driSubFD)
+{
+   drm_mga_init_t init;
+   memset(&init, 0, sizeof(drm_mga_init_t));
+   init.func = MGA_CLEANUP_DMA;
+   if(ioctl(driSubFD, DRM_IOCTL_MGA_INIT, &init)) {
+      return FALSE;
+   }
+   
+   return TRUE;
+}
+
+Bool drmMgaLockUpdate(int driSubFD, drmLockFlags flags)
+{
+   drm_lock_t lock;
+   
+   memset(&lock, 0, sizeof(drm_lock_t));
+   
+   if (flags & DRM_LOCK_QUIESCENT)  lock.flags |= _DRM_LOCK_QUIESCENT;
+   if (flags & DRM_LOCK_FLUSH)      lock.flags |= _DRM_LOCK_FLUSH;
+   if (flags & DRM_LOCK_FLUSH_ALL)  lock.flags |= _DRM_LOCK_FLUSH_ALL;
+   
+   if(ioctl(driSubFD, DRM_IOCTL_MGA_FLUSH, &lock)) {
+      return FALSE;
+   }
+   
+   return TRUE;
+}
+
+Bool drmMgaInitDma(int driSubFD, drmMgaInit *info)
+{
+      drm_mga_init_t init;
+      int i;
+   
+      memset(&init, 0, sizeof(drm_mga_init_t));
+      init.func = MGA_INIT_DMA;
+      init.reserved_map_agpstart = info->reserved_map_agpstart;
+      init.reserved_map_idx = info->reserved_map_idx;
+      init.buffer_map_idx = info->buffer_map_idx;
+      init.sarea_priv_offset = info->sarea_priv_offset;
+      init.primary_size = info->primary_size;
+      init.warp_ucode_size = info->warp_ucode_size;
+      init.frontOffset = info->frontOffset;
+      init.backOffset = info->backOffset;
+      init.depthOffset = info->depthOffset;
+      init.textureOffset = info->textureOffset;
+      init.textureSize = info->textureSize;
+      init.agpTextureSize = info->agpTextureSize;
+      init.agpTextureOffset = info->agpTextureOffset;
+      init.cpp = info->cpp;
+      init.stride = info->stride;
+      init.sgram = info->sgram;
+      init.chipset = info->chipset;
+   
+      for(i = 0; i < MGA_MAX_WARP_PIPES; i++) {
+	       init.WarpIndex[i].installed = info->WarpIndex[i].installed;
+	       init.WarpIndex[i].phys_addr = info->WarpIndex[i].phys_addr;
+	       init.WarpIndex[i].size = info->WarpIndex[i].size;
+      }
+   
+      init.mAccess = info->mAccess;
+   
+   
+   
+      if(ioctl(driSubFD, DRM_IOCTL_MGA_INIT, &init)) {
+	       return FALSE;
+      }
+      return TRUE;
+}
+
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/xf86drmR128.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/xf86drmR128.c
new file mode 100644
index 000000000..4195c90ba
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/xf86drmR128.c
@@ -0,0 +1,198 @@
+/* xf86drmR128.c -- User-level interface to Rage 128 DRM device
+ * Created: Sun Apr  9 18:13:54 2000 by kevin@precisioninsight.com
+ *
+ * Copyright 1999, 2000 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * Author: Kevin E. Martin <kevin@precisioninsight.com>
+ * 
+ * $XFree86$
+ * 
+ */
+
+#ifdef XFree86Server
+# include "xf86.h"
+# include "xf86_OSproc.h"
+# include "xf86_ansic.h"
+# include "xf86Priv.h"
+# define _DRM_MALLOC xalloc
+# define _DRM_FREE   xfree
+# ifndef XFree86LOADER
+#  include <sys/stat.h>
+#  include <sys/mman.h>
+# endif
+#else
+# include <stdio.h>
+# include <stdlib.h>
+# include <unistd.h>
+# include <string.h>
+# include <ctype.h>
+# include <fcntl.h>
+# include <errno.h>
+# include <signal.h>
+# include <sys/types.h>
+# include <sys/stat.h>
+# include <sys/ioctl.h>
+# include <sys/mman.h>
+# include <sys/time.h>
+# ifdef DRM_USE_MALLOC
+#  define _DRM_MALLOC malloc
+#  define _DRM_FREE   free
+extern int xf86InstallSIGIOHandler(int fd, void (*f)(int, void *), void *);
+extern int xf86RemoveSIGIOHandler(int fd);
+# else
+#  include <Xlibint.h>
+#  define _DRM_MALLOC Xmalloc
+#  define _DRM_FREE   Xfree
+# endif
+#endif
+
+/* Not all systems have MAP_FAILED defined */
+#ifndef MAP_FAILED
+#define MAP_FAILED ((void *)-1)
+#endif
+
+#include "xf86drm.h"
+#include "xf86drmR128.h"
+#include "drm.h"
+
+int drmR128InitCCE(int fd, drmR128Init *info)
+{
+    drm_r128_init_t init;
+   
+    memset(&init, 0, sizeof(drm_r128_init_t));
+
+    init.func                = R128_INIT_CCE;
+    init.sarea_priv_offset   = info->sarea_priv_offset;
+    init.is_pci              = info->is_pci;
+    init.cce_mode            = info->cce_mode;
+    init.cce_fifo_size       = info->cce_fifo_size;
+    init.cce_secure          = info->cce_secure;
+    init.ring_size           = info->ring_size;
+    init.usec_timeout        = info->usec_timeout;
+
+    init.fb_offset           = info->fb_offset;
+    init.agp_ring_offset     = info->agp_ring_offset;
+    init.agp_read_ptr_offset = info->agp_read_ptr_offset;
+    init.agp_vertbufs_offset = info->agp_vertbufs_offset;
+    init.agp_indbufs_offset  = info->agp_indbufs_offset;
+    init.agp_textures_offset = info->agp_textures_offset;
+    init.mmio_offset         = info->mmio_offset;
+
+    if (ioctl(fd, DRM_IOCTL_R128_INIT, &init)) return -errno;
+
+    return 0;
+}
+
+int drmR128CleanupCCE(int fd)
+{
+    drm_r128_init_t init;
+
+    memset(&init, 0, sizeof(drm_r128_init_t));
+
+    init.func = R128_CLEANUP_CCE;
+
+    if (ioctl(fd, DRM_IOCTL_R128_INIT, &init)) return -errno;
+
+    return 0;
+}
+
+int drmR128EngineReset(int fd)
+{
+    if (ioctl(fd, DRM_IOCTL_R128_RESET, NULL)) return -errno;
+
+    return 0;
+}
+
+int drmR128EngineFlush(int fd)
+{
+    if (ioctl(fd, DRM_IOCTL_R128_FLUSH, NULL)) return -errno;
+
+    return 0;
+}
+
+int drmR128CCEWaitForIdle(int fd)
+{
+    if (ioctl(fd, DRM_IOCTL_R128_CCEIDL, NULL)) return -errno;
+
+    return 0;
+}
+
+int drmR128SubmitPackets(int fd, CARD32 *buffer, int *count, int flags)
+{
+    drm_r128_packet_t packet;
+    int               ret;
+
+    memset(&packet, 0, sizeof(drm_r128_packet_t));
+
+    packet.count = *count;
+    packet.flags = flags;
+
+    while (packet.count > 0) {
+	packet.buffer = buffer + (*count - packet.count);
+	ret = ioctl(fd, DRM_IOCTL_R128_PACKET, &packet);
+	if (ret < 0 && ret != -EAGAIN) {
+	    *count = packet.count;
+	    return -errno;
+	}
+    }
+
+    *count = 0;
+    return 0;
+}
+
+int drmR128GetVertexBuffers(int fd, int count, int *indices, int *sizes)
+{
+    drm_r128_vertex_t v;
+
+    v.send_count      = 0;
+    v.send_indices    = NULL;
+    v.send_sizes      = NULL;
+    v.prim            = DRM_R128_PRIM_NONE;
+    v.request_count   = count;
+    v.request_indices = indices;
+    v.request_sizes   = sizes;
+    v.granted_count   = 0;
+
+    if (ioctl(fd, DRM_IOCTL_R128_VERTEX, &v)) return -errno;
+
+    return v.granted_count;
+}
+
+int drmR128FlushVertexBuffers(int fd, int count, int *indices,
+			      int *sizes, drmR128PrimType prim)
+{
+    drm_r128_vertex_t v;
+
+    v.send_count      = count;
+    v.send_indices    = indices;
+    v.send_sizes      = sizes;
+    v.prim            = prim;
+    v.request_count   = 0;
+    v.request_indices = NULL;
+    v.request_sizes   = NULL;
+    v.granted_count   = 0;
+
+    if (ioctl(fd, DRM_IOCTL_R128_VERTEX, &v) < 0) return -errno;
+
+    return 0;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/xf86drmRandom.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/xf86drmRandom.c
new file mode 100644
index 000000000..dad836d02
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/xf86drmRandom.c
@@ -0,0 +1,219 @@
+/* xf86drmRandom.c -- "Minimal Standard" PRNG Implementation
+ * Created: Mon Apr 19 08:28:13 1999 by faith@precisioninsight.com
+ * Revised: Thu Jun 24 14:53:45 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmRandom.c,v 1.4 1999/06/24 18:54:55 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmRandom.c,v 1.2 1999/06/27 14:08:20 dawes Exp $
+ *
+ * DESCRIPTION
+ *
+ * This file contains a simple, straightforward implementation of the Park
+ * & Miller "Minimal Standard" PRNG [PM88, PMS93], which is a Lehmer
+ * multiplicative linear congruential generator (MLCG) with a period of
+ * 2^31-1.
+ *
+ * This implementation is intended to provide a reliable, portable PRNG
+ * that is suitable for testing a hash table implementation and for
+ * implementing skip lists.
+ *
+ * FUTURE ENHANCEMENTS
+ *
+ * If initial seeds are not selected randomly, two instances of the PRNG
+ * can be correlated.  [Knuth81, pp. 32-33] describes a shuffling technique
+ * that can eliminate this problem.
+ *
+ * If PRNGs are used for simulation, the period of the current
+ * implementation may be too short.  [LE88] discusses methods of combining
+ * MLCGs to produce much longer periods, and suggests some alternative
+ * values for A and M.  [LE90 and Sch92] also provide information on
+ * long-period PRNGs.
+ *
+ * REFERENCES
+ *
+ * [Knuth81] Donald E. Knuth. The Art of Computer Programming.  Volume 2:
+ * Seminumerical Algorithms.  Reading, Massachusetts: Addison-Wesley, 1981.
+ *
+ * [LE88] Pierre L'Ecuyer. "Efficient and Portable Combined Random Number
+ * Generators".  CACM 31(6), June 1988, pp. 742-774.
+ *
+ * [LE90] Pierre L'Ecuyer. "Random Numbers for Simulation". CACM 33(10,
+ * October 1990, pp. 85-97.
+ *
+ * [PM88] Stephen K. Park and Keith W. Miller. "Random Number Generators:
+ * Good Ones are Hard to Find". CACM 31(10), October 1988, pp. 1192-1201.
+ *
+ * [Sch92] Bruce Schneier. "Pseudo-Ransom Sequence Generator for 32-Bit
+ * CPUs".  Dr. Dobb's Journal 17(2), February 1992, pp. 34, 37-38, 40.
+ *
+ * [PMS93] Stephen K. Park, Keith W. Miller, and Paul K. Stockmeyer.  In
+ * "Technical Correspondence: Remarks on Choosing and Implementing Random
+ * Number Generators". CACM 36(7), July 1993, pp. 105-110.
+ *
+ */
+
+#define RANDOM_MAIN 0
+
+#if RANDOM_MAIN
+# include <stdio.h>
+# include <stdlib.h>
+#else
+# include "xf86drm.h"
+# ifdef XFree86LOADER
+#  include "xf86.h"
+#  include "xf86_ansic.h"
+# else
+#  include <stdio.h>
+#  include <stdlib.h>
+# endif
+#endif
+
+#define N(x)  drm##x
+
+#define RANDOM_MAGIC 0xfeedbeef
+#define RANDOM_DEBUG 0
+
+#if RANDOM_MAIN
+#define RANDOM_ALLOC malloc
+#define RANDOM_FREE  free
+#else
+#define RANDOM_ALLOC drmMalloc
+#define RANDOM_FREE  drmFree
+#endif
+
+typedef struct RandomState {
+    unsigned long magic;
+    unsigned long a;
+    unsigned long m;
+    unsigned long q;		/* m div a */
+    unsigned long r;		/* m mod a */
+    unsigned long check;
+    long          seed;
+} RandomState;
+
+#if RANDOM_MAIN
+extern void          *N(RandomCreate)(unsigned long seed);
+extern int           N(RandomDestroy)(void *state);
+extern unsigned long N(Random)(void *state);
+extern double        N(RandomDouble)(void *state);
+#endif
+
+void *N(RandomCreate)(unsigned long seed)
+{
+    RandomState  *state;
+
+    state           = RANDOM_ALLOC(sizeof(*state));
+    if (!state) return NULL;
+    state->magic    = RANDOM_MAGIC;
+#if 0
+				/* Park & Miller, October 1988 */
+    state->a        = 16807;
+    state->m        = 2147483647;
+    state->check    = 1043618065; /* After 10000 iterations */
+#else
+				/* Park, Miller, and Stockmeyer, July 1993 */
+    state->a        = 48271;
+    state->m        = 2147483647;
+    state->check    = 399268537; /* After 10000 iterations */
+#endif
+    state->q        = state->m / state->a;
+    state->r        = state->m % state->a;
+
+    state->seed     = seed;
+				/* Check for illegal boundary conditions,
+                                   and choose closest legal value. */
+    if (state->seed <= 0)        state->seed = 1;
+    if (state->seed >= state->m) state->seed = state->m - 1;
+
+    return state;
+}
+
+int N(RandomDestroy)(void *state)
+{
+    RANDOM_FREE(state);
+    return 0;
+}
+
+unsigned long N(Random)(void *state)
+{
+    RandomState   *s = (RandomState *)state;
+    long          hi;
+    long          lo;
+
+    hi      = s->seed / s->q;
+    lo      = s->seed % s->q;
+    s->seed = s->a * lo - s->r * hi;
+    if (s->seed <= 0) s->seed += s->m;
+
+    return s->seed;
+}
+
+double N(RandomDouble)(void *state)
+{
+    RandomState *s = (RandomState *)state;
+    
+    return (double)N(Random)(state)/(double)s->m;
+}
+
+#if RANDOM_MAIN
+static void check_period(long seed)
+{
+    unsigned long count = 0;
+    unsigned long initial;
+    void          *state;
+    
+    state = N(RandomCreate)(seed);
+    initial = N(Random)(state);
+    ++count;
+    while (initial != N(Random)(state)) {
+	if (!++count) break;
+    }
+    printf("With seed of %10ld, period = %10lu (0x%08lx)\n",
+	   seed, count, count);
+    N(RandomDestroy)(state);
+}
+
+int main(void)
+{
+    RandomState   *state;
+    int           i;
+    unsigned long rand;
+
+    state = N(RandomCreate)(1);
+    for (i = 0; i < 10000; i++) {
+	rand = N(Random)(state);
+    }
+    printf("After 10000 iterations: %lu (%lu expected): %s\n",
+	   rand, state->check,
+	   rand - state->check ? "*INCORRECT*" : "CORRECT");
+    N(RandomDestroy)(state);
+
+    printf("Checking periods...\n");
+    check_period(1);
+    check_period(2);
+    check_period(31415926);
+    
+    return 0;
+}
+#endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/xf86drmSL.c b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/xf86drmSL.c
new file mode 100644
index 000000000..d5dd63da8
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/bsd/drm/xf86drmSL.c
@@ -0,0 +1,490 @@
+/* xf86drmSL.c -- Skip list support
+ * Created: Mon May 10 09:28:13 1999 by faith@precisioninsight.com
+ * Revised: Thu Jun  3 16:13:01 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * $PI: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmSL.c,v 1.2 1999/06/07 13:01:42 faith Exp $
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmSL.c,v 1.1 1999/06/14 07:32:02 dawes Exp $
+ *
+ * DESCRIPTION
+ *
+ * This file contains a straightforward skip list implementation.n
+ *
+ * FUTURE ENHANCEMENTS
+ *
+ * REFERENCES
+ *
+ * [Pugh90] William Pugh.  Skip Lists: A Probabilistic Alternative to
+ * Balanced Trees. CACM 33(6), June 1990, pp. 668-676.
+ *
+ */
+
+#define SL_MAIN 0
+
+#if SL_MAIN
+# include <stdio.h>
+# include <stdlib.h>
+#  include <sys/time.h>
+#else
+# include "xf86drm.h"
+# ifdef XFree86LOADER
+#  include "xf86.h"
+#  include "xf86_ansic.h"
+# else
+#  include <stdio.h>
+#  include <stdlib.h>
+# endif
+#endif
+
+#define N(x)  drm##x
+
+#define SL_LIST_MAGIC  0xfacade00LU
+#define SL_ENTRY_MAGIC 0x00fab1edLU
+#define SL_FREED_MAGIC 0xdecea5edLU
+#define SL_MAX_LEVEL   16
+#define SL_DEBUG       0
+#define SL_RANDOM_SEED 0xc01055a1LU
+
+#if SL_MAIN
+#define SL_ALLOC malloc
+#define SL_FREE  free
+#define SL_RANDOM_DECL        static int state = 0;
+#define SL_RANDOM_INIT(seed)  if (!state) { srandom(seed); ++state; }
+#define SL_RANDOM             random()
+#else
+#define SL_ALLOC drmMalloc
+#define SL_FREE  drmFree
+#define SL_RANDOM_DECL        static void *state = NULL
+#define SL_RANDOM_INIT(seed)  if (!state) state = drmRandomCreate(seed)
+#define SL_RANDOM             drmRandom(state)
+
+#endif
+
+typedef struct SLEntry {
+    unsigned long     magic;	   /* SL_ENTRY_MAGIC */
+    unsigned long     key;
+    void              *value;
+    int               levels;
+    struct SLEntry    *forward[1]; /* variable sized array */
+} SLEntry, *SLEntryPtr;
+
+typedef struct SkipList {
+    unsigned long    magic;	/* SL_LIST_MAGIC */
+    int              level;
+    int              count;
+    SLEntryPtr       head;
+    SLEntryPtr       p0;	/* Position for iteration */
+} SkipList, *SkipListPtr;
+
+#if SL_MAIN
+extern void *N(SLCreate)(void);
+extern int  N(SLDestroy)(void *l);
+extern int  N(SLLookup)(void *l, unsigned long key, void **value);
+extern int  N(SLInsert)(void *l, unsigned long key, void *value);
+extern int  N(SLDelete)(void *l, unsigned long key);
+extern int  N(SLNext)(void *l, unsigned long *key, void **value);
+extern int  N(SLFirst)(void *l, unsigned long *key, void **value);
+extern void N(SLDump)(void *l);
+extern int  N(SLLookupNeighbors)(void *l, unsigned long key,
+				 unsigned long *prev_key, void **prev_value,
+				 unsigned long *next_key, void **next_value);
+#endif
+
+static SLEntryPtr SLCreateEntry(int max_level, unsigned long key, void *value)
+{
+    SLEntryPtr entry;
+    
+    if (max_level < 0 || max_level > SL_MAX_LEVEL) max_level = SL_MAX_LEVEL;
+
+    entry         = SL_ALLOC(sizeof(*entry)
+			     + (max_level + 1) * sizeof(entry->forward[0]));
+    if (!entry) return NULL;
+    entry->magic  = SL_ENTRY_MAGIC;
+    entry->key    = key;
+    entry->value  = value;
+    entry->levels = max_level + 1;
+
+    return entry;
+}
+
+static int SLRandomLevel(void)
+{
+    int level = 1;
+    SL_RANDOM_DECL;
+
+    SL_RANDOM_INIT(SL_RANDOM_SEED);
+    
+    while ((SL_RANDOM & 0x01) && level < SL_MAX_LEVEL) ++level;
+    return level;
+}
+
+void *N(SLCreate)(void)
+{
+    SkipListPtr  list;
+    int          i;
+
+    list           = SL_ALLOC(sizeof(*list));
+    if (!list) return NULL;
+    list->magic    = SL_LIST_MAGIC;
+    list->level    = 0;
+    list->head     = SLCreateEntry(SL_MAX_LEVEL, 0, NULL);
+    list->count    = 0;
+
+    for (i = 0; i <= SL_MAX_LEVEL; i++) list->head->forward[i] = NULL;
+    
+    return list;
+}
+
+int N(SLDestroy)(void *l)
+{
+    SkipListPtr   list  = (SkipListPtr)l;
+    SLEntryPtr    entry;
+    SLEntryPtr    next;
+
+    if (list->magic != SL_LIST_MAGIC) return -1; /* Bad magic */
+
+    for (entry = list->head; entry; entry = next) {
+	if (entry->magic != SL_ENTRY_MAGIC) return -1; /* Bad magic */
+	next         = entry->forward[0];
+	entry->magic = SL_FREED_MAGIC;
+	SL_FREE(entry);
+    }
+
+    list->magic = SL_FREED_MAGIC;
+    SL_FREE(list);
+    return 0;
+}
+
+static SLEntryPtr SLLocate(void *l, unsigned long key, SLEntryPtr *update)
+{
+    SkipListPtr   list  = (SkipListPtr)l;
+    SLEntryPtr    entry;
+    int           i;
+
+    if (list->magic != SL_LIST_MAGIC) return NULL;
+
+    for (i = list->level, entry = list->head; i >= 0; i--) {
+	while (entry->forward[i] && entry->forward[i]->key < key)
+	    entry = entry->forward[i];
+	update[i] = entry;
+    }
+
+    return entry->forward[0];
+}
+
+int N(SLInsert)(void *l, unsigned long key, void *value)
+{
+    SkipListPtr   list  = (SkipListPtr)l;
+    SLEntryPtr    entry;
+    SLEntryPtr    update[SL_MAX_LEVEL + 1];
+    int           level;
+    int           i;
+
+    if (list->magic != SL_LIST_MAGIC) return -1; /* Bad magic */
+
+    entry = SLLocate(list, key, update);
+
+    if (entry && entry->key == key) return 1; /* Already in list */
+
+
+    level = SLRandomLevel();
+    if (level > list->level) {
+	level = ++list->level;
+	update[level] = list->head;
+    }
+
+    entry = SLCreateEntry(level, key, value);
+
+				/* Fix up forward pointers */
+    for (i = 0; i <= level; i++) {
+	entry->forward[i]     = update[i]->forward[i];
+	update[i]->forward[i] = entry;
+    }
+
+    ++list->count;
+    return 0;			/* Added to table */
+}
+
+int N(SLDelete)(void *l, unsigned long key)
+{
+    SkipListPtr   list = (SkipListPtr)l;
+    SLEntryPtr    update[SL_MAX_LEVEL + 1];
+    SLEntryPtr    entry;
+    int           i;
+
+    if (list->magic != SL_LIST_MAGIC) return -1; /* Bad magic */
+
+    entry = SLLocate(list, key, update);
+
+    if (!entry || entry->key != key) return 1; /* Not found */
+
+				/* Fix up forward pointers */
+    for (i = 0; i <= list->level; i++) {
+	if (update[i]->forward[i] == entry)
+	    update[i]->forward[i] = entry->forward[i];
+    }
+
+    entry->magic = SL_FREED_MAGIC;
+    SL_FREE(entry);
+
+    while (list->level && !list->head->forward[list->level]) --list->level;
+    --list->count;
+    return 0;
+}
+
+int N(SLLookup)(void *l, unsigned long key, void **value)
+{
+    SkipListPtr   list = (SkipListPtr)l;
+    SLEntryPtr    update[SL_MAX_LEVEL + 1];
+    SLEntryPtr    entry;
+
+    entry = SLLocate(list, key, update);
+
+    if (entry && entry->key == key) {
+	*value = entry;
+	return 0;
+    }
+    *value = NULL;
+    return -1;
+}
+
+int N(SLLookupNeighbors)(void *l, unsigned long key,
+			 unsigned long *prev_key, void **prev_value,
+			 unsigned long *next_key, void **next_value)
+{
+    SkipListPtr   list = (SkipListPtr)l;
+    SLEntryPtr    update[SL_MAX_LEVEL + 1];
+    SLEntryPtr    entry;
+    int           retcode = 0;
+
+    entry = SLLocate(list, key, update);
+
+    *prev_key   = *next_key   = key;
+    *prev_value = *next_value = NULL;
+	
+    if (update[0]) {
+	*prev_key   = update[0]->key;
+	*prev_value = update[0]->value;
+	++retcode;
+	if (update[0]->forward[0]) {
+	    *next_key   = update[0]->forward[0]->key;
+	    *next_value = update[0]->forward[0]->value;
+	    ++retcode;
+	}
+    }
+    return retcode;
+}
+
+int N(SLNext)(void *l, unsigned long *key, void **value)
+{
+    SkipListPtr   list = (SkipListPtr)l;
+    SLEntryPtr    entry;
+    
+    if (list->magic != SL_LIST_MAGIC) return -1; /* Bad magic */
+
+    entry    = list->p0;
+
+    if (entry) {
+	list->p0 = entry->forward[0];
+	*key     = entry->key;
+	*value   = entry->value;
+	return 1;
+    }
+    list->p0 = NULL;
+    return 0;
+}
+
+int N(SLFirst)(void *l, unsigned long *key, void **value)
+{
+    SkipListPtr   list = (SkipListPtr)l;
+    
+    if (list->magic != SL_LIST_MAGIC) return -1; /* Bad magic */
+    
+    list->p0 = list->head->forward[0];
+    return N(SLNext)(list, key, value);
+}
+
+/* Dump internal data structures for debugging. */
+void N(SLDump)(void *l)
+{
+    SkipListPtr   list = (SkipListPtr)l;
+    SLEntryPtr    entry;
+    int           i;
+    
+    if (list->magic != SL_LIST_MAGIC) {
+	printf("Bad magic: 0x%08lx (expected 0x%08lx)\n",
+	       list->magic, SL_LIST_MAGIC);
+	return;
+    }
+
+    printf("Level = %d, count = %d\n", list->level, list->count);
+    for (entry = list->head; entry; entry = entry->forward[0]) {
+	if (entry->magic != SL_ENTRY_MAGIC) {
+	    printf("Bad magic: 0x%08lx (expected 0x%08lx)\n",
+		   list->magic, SL_ENTRY_MAGIC);
+	}
+	printf("\nEntry %p <0x%08lx, %p> has %2d levels\n",
+	       entry, entry->key, entry->value, entry->levels);
+	for (i = 0; i < entry->levels; i++) {
+	    if (entry->forward[i]) {
+		printf("   %2d: %p <0x%08lx, %p>\n",
+		       i,
+		       entry->forward[i],
+		       entry->forward[i]->key,
+		       entry->forward[i]->value);
+	    } else {
+		printf("   %2d: %p\n", i, entry->forward[i]);
+	    }
+	}
+    }
+}
+
+#if SL_MAIN
+static void print(SkipListPtr list)
+{
+    unsigned long key;
+    void          *value;
+    
+    if (N(SLFirst)(list, &key, &value)) {
+	do {
+	    printf("key = %5lu, value = %p\n", key, value);
+	} while (N(SLNext)(list, &key, &value));
+    }
+}
+
+static double do_time(int size, int iter)
+{
+    SkipListPtr    list;
+    int            i, j;
+    unsigned long  keys[1000000];
+    unsigned long  previous;
+    unsigned long  key;
+    void           *value;
+    struct timeval start, stop;
+    double         usec;
+    SL_RANDOM_DECL;
+
+    SL_RANDOM_INIT(12345);
+    
+    list = N(SLCreate)();
+
+    for (i = 0; i < size; i++) {
+	keys[i] = SL_RANDOM;
+	N(SLInsert)(list, keys[i], NULL);
+    }
+
+    previous = 0;
+    if (N(SLFirst)(list, &key, &value)) {
+	do {
+	    if (key <= previous) {
+		printf( "%lu !< %lu\n", previous, key);
+	    }
+	    previous = key;
+	} while (N(SLNext)(list, &key, &value));
+    }
+    
+    gettimeofday(&start, NULL);
+    for (j = 0; j < iter; j++) {
+	for (i = 0; i < size; i++) {
+	    if (N(SLLookup)(list, keys[i], &value))
+		printf("Error %lu %d\n", keys[i], i);
+	}
+    }
+    gettimeofday(&stop, NULL);
+    
+    usec = (double)(stop.tv_sec * 1000000 + stop.tv_usec
+		    - start.tv_sec * 1000000 - start.tv_usec) / (size * iter);
+    
+    printf("%0.2f microseconds for list length %d\n", usec, size);
+
+    N(SLDestroy)(list);
+    
+    return usec;
+}
+
+static void print_neighbors(void *list, unsigned long key)
+{
+    unsigned long prev_key = 0;
+    unsigned long next_key = 0;
+    void          *prev_value;
+    void          *next_value;
+    int           retval;
+
+    retval = drmSLLookupNeighbors(list, key,
+				  &prev_key, &prev_value,
+				  &next_key, &next_value);
+    printf("Neighbors of %5lu: %d %5lu %5lu\n",
+	   key, retval, prev_key, next_key);
+}
+
+int main(void)
+{
+    SkipListPtr    list;
+    double         usec, usec2, usec3, usec4;
+
+    list = N(SLCreate)();
+    printf( "list at %p\n", list);
+
+    print(list);
+    printf("\n==============================\n\n");
+
+    N(SLInsert)(list, 123, NULL);
+    N(SLInsert)(list, 213, NULL);
+    N(SLInsert)(list, 50, NULL);
+    print(list);
+    printf("\n==============================\n\n");
+    
+    print_neighbors(list, 0);
+    print_neighbors(list, 50);
+    print_neighbors(list, 51);
+    print_neighbors(list, 123);
+    print_neighbors(list, 200);
+    print_neighbors(list, 213);
+    print_neighbors(list, 256);
+    printf("\n==============================\n\n");    
+    
+    N(SLDelete)(list, 50);
+    print(list);
+    printf("\n==============================\n\n");
+
+    N(SLDump)(list);
+    N(SLDestroy)(list);
+    printf("\n==============================\n\n");
+
+    usec  = do_time(100, 10000);
+    usec2 = do_time(1000, 500);
+    printf("Table size increased by %0.2f, search time increased by %0.2f\n",
+	   1000.0/100.0, usec2 / usec);
+    
+    usec3 = do_time(10000, 50);
+    printf("Table size increased by %0.2f, search time increased by %0.2f\n",
+	   10000.0/100.0, usec3 / usec);
+    
+    usec4 = do_time(100000, 4);
+    printf("Table size increased by %0.2f, search time increased by %0.2f\n",
+	   100000.0/100.0, usec4 / usec);
+
+    return 0;
+}
+#endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/Imakefile b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/Imakefile
index 96388450b..fdaecb972 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/Imakefile
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/Imakefile
@@ -13,15 +13,16 @@ MOBJ = drmmodule.o
 MTRR_DEFINES = -DHAS_MTRR_SUPPORT
 #endif
 
-SRCS = xf86drm.c xf86drmHash.c xf86drmRandom.c xf86drmSL.c $(MSRC)
-OBJS = xf86drm.o xf86drmHash.o xf86drmRandom.o xf86drmSL.o $(MOBJ)
-
+SRCS = xf86drm.c xf86drmHash.c xf86drmRandom.c xf86drmSL.c xf86drmI810.c xf86drmMga.c xf86drmR128.c $(MSRC)
+OBJS = xf86drm.o xf86drmHash.o xf86drmRandom.o xf86drmSL.o xf86drmI810.o xf86drmMga.o xf86drmR128.o $(MOBJ)
 
 INCLUDES = -I$(XF86COMSRC) -I$(XF86OSSRC) -I. -I$(SERVERSRC)/include \
            -I$(XINCLUDESRC) -I$(EXTINCSRC) -I../.. -Ikernel
 
 DEFINES = $(MTRR_DEFINES) $(GLX_DEFINES)
 
+#if DoLoadableServer
+
 ModuleObjectRule()
 LibraryModuleTarget(drm,$(OBJS))
 NormalLintTarget($(SRCS))
@@ -30,6 +31,14 @@ NormalLintTarget($(SRCS))
 InstallLibraryModule(drm,$(MODULEDIR),linux)
 #endif
 
+#else
+
+SubdirLibraryRule($(OBJS))
+NormalLibraryObjectRule()
+
+#endif
+
+
 #define IHaveSubdirs
 SUBDIRS = kernel
 
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/Makefile.kernel b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/Makefile.kernel
index 44d75c487..a169473af 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/Makefile.kernel
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/Makefile.kernel
@@ -14,7 +14,8 @@
 L_TARGET        := libdrm.a
 
 L_OBJS		:= init.o memory.o proc.o auth.o context.o drawable.o bufs.o \
-			lists.o lock.o ioctl.o fops.o vm.o dma.o
+			lists.o lock.o ioctl.o fops.o vm.o dma.o ctxbitmap.o \
+			agpsupport.o
 
 M_OBJS		:= 
 
@@ -26,6 +27,14 @@ ifdef CONFIG_DRM_TDFX
 M_OBJS          += tdfx.o
 endif
 
+ifdef CONFIG_DRM_MGA
+M_OBJS		+= mga.o
+endif
+
+ifdef CONFIG_DRM_R128
+M_OBJS		+= r128.o
+endif
+
 include $(TOPDIR)/Rules.make
 
 gamma.o: gamma_drv.o gamma_dma.o $(L_TARGET)
@@ -33,3 +42,12 @@ gamma.o: gamma_drv.o gamma_dma.o $(L_TARGET)
 
 tdfx.o: tdfx_drv.o tdfx_context.o $(L_TARGET)
 	$(LD) $(LD_RFLAG) -r -o $@ tdfx_drv.o tdfx_context.o -L. -ldrm
+
+i810.o: i810_drv.o i810_context.o $(L_TARGET)
+	$(LD) $(LD_RFLAG) -r -o $@ i810_drv.o i810_bufs.o i810_dma.o i810_context.o -L. -ldrm
+
+mga.o: mga_drv.o mga_context.o mga_dma.o mga_bufs.o $(L_TARGET)
+	$(LD) $(LD_RFLAG) -r -o $@ mga_drv.o mga_bufs.o mga_dma.o mga_context.o mga_state.o -L. -ldrm
+
+r128.o: r128_drv.o r128_context.o $(L_TARGET)
+	$(LD) $(LD_RFLAG) -r -o $@ r128_drv.o r128_context.o -L. -ldrm
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/Makefile.linux b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/Makefile.linux
index 9810fe3d5..ecc196bdd 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/Makefile.linux
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/Makefile.linux
@@ -47,7 +47,7 @@
 
 # **** End of SMP/MODVERSIONS detection
 
-MODS=           gamma.o tdfx.o
+MODS=           gamma.o tdfx.o r128.o
 LIBS=           libdrm.a
 PROGS=		drmstat
 
@@ -61,6 +61,9 @@ GAMMAHEADERS=   gamma_drv.h $(DRMHEADERS)
 TDFXOBJS=       tdfx_drv.o tdfx_context.o
 TDFXHEADERS=    tdfx_drv.h $(DRMHEADERS)
 
+R128OBJS=	r128_drv.o r128_dma.o r128_bufs.o r128_context.o
+R128HEADERS=	r128_drv.h r128_drm.h $(DRMHEADERS)
+
 PROGOBJS=       drmstat.po xf86drm.po xf86drmHash.po xf86drmRandom.po sigio.po
 PROGHEADERS=    xf86drm.h $(DRMHEADERS)
 
@@ -167,6 +170,9 @@ gamma.o: $(GAMMAOBJS) $(LIBS)
 tdfx.o: $(TDFXOBJS) $(LIBS)
 	$(LD) -r $^ -o $@
 
+r128.o: $(R128OBJS) $(LIBS)
+	$(LD) -r $^ -o $@
+
 ifeq ($(AGP),1)
 mga.o: $(MGAOBJS) $(LIBS)
 	$(LD) -r $^ -o $@
@@ -196,6 +202,7 @@ ChangeLog:
 $(DRMOBJS): $(DRMHEADERS)
 $(GAMMAOBJS): $(GAMMAHEADERS)
 $(TDFXOBJS): $(TDFXHEADERS)
+$(R128OBJS): $(R128HEADERS)
 ifeq ($(AGP),1)
 $(MGAOBJS): $(MGAHEADERS)
 $(I810OBJS): $(I810HEADERS)
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/agpsupport.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/agpsupport.c
index 262d63adf..c89c3e258 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/agpsupport.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/agpsupport.c
@@ -238,6 +238,8 @@ int drm_agp_bind(struct inode *inode, struct file *filp, unsigned int cmd,
 	page = (request.offset + PAGE_SIZE - 1) / PAGE_SIZE;
 	if ((retcode = drm_bind_agp(entry->memory, page))) return retcode;
 	entry->bound = dev->agp->base + (page << PAGE_SHIFT);
+	DRM_DEBUG("base = 0x%lx entry->bound = 0x%lx\n", 
+		  dev->agp->base, entry->bound);
 	return 0;
 }
 
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm.h
index 15491aeeb..c8c5581da 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drm.h
@@ -72,9 +72,10 @@ typedef struct drm_clip_rect {
            unsigned short y2;
 } drm_clip_rect_t;
 
-/* Seperate include files for the i810/mga specific structures */
+/* Seperate include files for the i810/mga/r128 specific structures */
 #include "mga_drm.h"
 #include "i810_drm.h"
+#include "r128_drm.h"
 
 typedef struct drm_version {
 	int    version_major;	  /* Major version			    */
@@ -297,7 +298,7 @@ typedef struct drm_agp_info {
 
 #define DRM_IOCTL_VERSION    DRM_IOWR(0x00, drm_version_t)
 #define DRM_IOCTL_GET_UNIQUE DRM_IOWR(0x01, drm_unique_t)
-#define DRM_IOCTL_GET_MAGIC  DRM_IOW( 0x02, drm_auth_t)
+#define DRM_IOCTL_GET_MAGIC  DRM_IOR( 0x02, drm_auth_t)
 #define DRM_IOCTL_IRQ_BUSID  DRM_IOWR(0x03, drm_irq_busid_t)
 
 #define DRM_IOCTL_SET_UNIQUE DRM_IOW( 0x10, drm_unique_t)
@@ -328,11 +329,11 @@ typedef struct drm_agp_info {
 
 #define DRM_IOCTL_AGP_ACQUIRE DRM_IO(  0x30)
 #define DRM_IOCTL_AGP_RELEASE DRM_IO(  0x31)
-#define DRM_IOCTL_AGP_ENABLE  DRM_IOR( 0x32, drm_agp_mode_t)
-#define DRM_IOCTL_AGP_INFO    DRM_IOW( 0x33, drm_agp_info_t)
+#define DRM_IOCTL_AGP_ENABLE  DRM_IOW( 0x32, drm_agp_mode_t)
+#define DRM_IOCTL_AGP_INFO    DRM_IOR( 0x33, drm_agp_info_t)
 #define DRM_IOCTL_AGP_ALLOC   DRM_IOWR(0x34, drm_agp_buffer_t)
 #define DRM_IOCTL_AGP_FREE    DRM_IOW( 0x35, drm_agp_buffer_t)
-#define DRM_IOCTL_AGP_BIND    DRM_IOWR(0x36, drm_agp_binding_t)
+#define DRM_IOCTL_AGP_BIND    DRM_IOW( 0x36, drm_agp_binding_t)
 #define DRM_IOCTL_AGP_UNBIND  DRM_IOW( 0x37, drm_agp_binding_t)
 
 /* Mga specific ioctls */
@@ -342,12 +343,23 @@ typedef struct drm_agp_info {
 #define DRM_IOCTL_MGA_ILOAD   DRM_IOW( 0x43, drm_mga_iload_t)
 #define DRM_IOCTL_MGA_VERTEX  DRM_IOW( 0x44, drm_mga_vertex_t)
 #define DRM_IOCTL_MGA_FLUSH   DRM_IOW( 0x45, drm_lock_t )
+#define DRM_IOCTL_MGA_INDICES DRM_IOW( 0x46, drm_mga_indices_t)
 
 /* I810 specific ioctls */
 #define DRM_IOCTL_I810_INIT    DRM_IOW( 0x40, drm_i810_init_t)
 #define DRM_IOCTL_I810_VERTEX  DRM_IOW( 0x41, drm_i810_vertex_t)
-#define DRM_IOCTL_I810_DMA     DRM_IOW( 0x42, drm_i810_general_t)
+#define DRM_IOCTL_I810_CLEAR   DRM_IOW( 0x42, drm_i810_clear_t)
 #define DRM_IOCTL_I810_FLUSH   DRM_IO ( 0x43)
 #define DRM_IOCTL_I810_GETAGE  DRM_IO ( 0x44)
+#define DRM_IOCTL_I810_GETBUF  DRM_IOW( 0x45, drm_i810_dma_t)
+#define DRM_IOCTL_I810_SWAP    DRM_IO ( 0x46)
+
+/* Rage 128 specific ioctls */
+#define DRM_IOCTL_R128_INIT	DRM_IOW( 0x40, drm_r128_init_t)
+#define DRM_IOCTL_R128_RESET	DRM_IO(  0x41)
+#define DRM_IOCTL_R128_FLUSH	DRM_IO(  0x42)
+#define DRM_IOCTL_R128_CCEIDL	DRM_IO(  0x43)
+#define DRM_IOCTL_R128_PACKET	DRM_IOW( 0x44, drm_r128_packet_t)
+#define DRM_IOCTL_R128_VERTEX	DRM_IOW( 0x45, drm_r128_vertex_t)
 
 #endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drmP.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drmP.h
index ecf50436d..350d1ef9b 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drmP.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drmP.h
@@ -229,8 +229,8 @@ typedef struct drm_magic_entry {
 } drm_magic_entry_t;
 
 typedef struct drm_magic_head {
-       struct drm_magic_entry *head;
-       struct drm_magic_entry *tail;
+	struct drm_magic_entry *head;
+	struct drm_magic_entry *tail;
 } drm_magic_head_t;
 
 typedef struct drm_vma_entry {
@@ -263,16 +263,15 @@ typedef struct drm_buf {
 		DRM_LIST_RECLAIM = 5
 	}		  list;	       /* Which list we're on		     */
 
-
-	void *dev_private;
-	int dev_priv_size;
-
 #if DRM_DMA_HISTOGRAM
 	cycles_t	  time_queued;	   /* Queued to kernel DMA queue     */
 	cycles_t	  time_dispatched; /* Dispatched to hardware	     */
 	cycles_t	  time_completed;  /* Completed by hardware	     */
 	cycles_t	  time_freed;	   /* Back on freelist		     */
 #endif
+
+	int		  dev_priv_size; /* Size of buffer private stoarge   */
+	void		  *dev_private;  /* Per-buffer private storage       */
 } drm_buf_t;
 
 #if DRM_DMA_HISTOGRAM
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_dma.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_dma.c
index 3372f51ea..eb78c0374 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_dma.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_dma.c
@@ -88,13 +88,31 @@ static inline void gamma_dma_dispatch(drm_device_t *dev, unsigned long address,
 	GAMMA_WRITE(GAMMA_DMACOUNT, length / 4);
 }
 
-static inline void gamma_dma_quiescent(drm_device_t *dev)
+static inline void gamma_dma_quiescent_single(drm_device_t *dev)
 {
 	while (GAMMA_READ(GAMMA_DMACOUNT))
 		;
 	while (GAMMA_READ(GAMMA_INFIFOSPACE) < 3)
 		;
+
+	GAMMA_WRITE(GAMMA_FILTERMODE, 1 << 10);
+	GAMMA_WRITE(GAMMA_SYNC, 0);
+	
+	do {
+		while (!GAMMA_READ(GAMMA_OUTFIFOWORDS))
+			;
+	} while (GAMMA_READ(GAMMA_OUTPUTFIFO) != GAMMA_SYNC_TAG);
+}
+
+static inline void gamma_dma_quiescent_dual(drm_device_t *dev)
+{
+	while (GAMMA_READ(GAMMA_DMACOUNT))
+		;
+	while (GAMMA_READ(GAMMA_INFIFOSPACE) < 3)
+		;
+
 	GAMMA_WRITE(GAMMA_BROADCASTMASK, 3);
+
 	GAMMA_WRITE(GAMMA_FILTERMODE, 1 << 10);
 	GAMMA_WRITE(GAMMA_SYNC, 0);
 	
@@ -104,7 +122,6 @@ static inline void gamma_dma_quiescent(drm_device_t *dev)
 			;
 	} while (GAMMA_READ(GAMMA_OUTPUTFIFO) != GAMMA_SYNC_TAG);
 	
-
 				/* Read from second MX */
 	do {
 		while (!GAMMA_READ(GAMMA_OUTFIFOWORDS + 0x10000))
@@ -789,8 +806,13 @@ int gamma_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 	if (!ret) {
 		if (lock.flags & _DRM_LOCK_READY)
 			gamma_dma_ready(dev);
-		if (lock.flags & _DRM_LOCK_QUIESCENT)
-			gamma_dma_quiescent(dev);
+		if (lock.flags & _DRM_LOCK_QUIESCENT) {
+			if (gamma_found() == 1) {
+				gamma_dma_quiescent_single(dev);
+			} else {
+				gamma_dma_quiescent_dual(dev);
+			}
+		}
 	}
 	DRM_DEBUG("%d %s\n", lock.context, ret ? "interrupted" : "has lock");
 
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drv.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drv.c
index c6838cb74..d42cf4aea 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drv.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drv.c
@@ -32,6 +32,7 @@
 #define EXPORT_SYMTAB
 #include "drmP.h"
 #include "gamma_drv.h"
+#include <linux/pci.h>
 EXPORT_SYMBOL(gamma_init);
 EXPORT_SYMBOL(gamma_cleanup);
 
@@ -99,10 +100,13 @@ static drm_ioctl_desc_t	      gamma_ioctls[] = {
 int			      init_module(void);
 void			      cleanup_module(void);
 static char		      *gamma = NULL;
+static int 		      devices = 0;
 
 MODULE_AUTHOR("Precision Insight, Inc., Cedar Park, Texas.");
 MODULE_DESCRIPTION("3dlabs GMX 2000");
 MODULE_PARM(gamma, "s");
+MODULE_PARM(devices, "i");
+MODULE_PARM_DESC(devices, "devices=x, where x is the number of MX chips on your card\n");
 
 /* init_module is called when insmod is used to load the module */
 
@@ -317,6 +321,34 @@ static int gamma_takedown(drm_device_t *dev)
 	return 0;
 }
 
+int gamma_found(void)
+{
+	return devices;
+}
+
+int gamma_find_devices(void)
+{
+	struct pci_dev *d = NULL, *one = NULL, *two = NULL;
+
+	d = pci_find_device(PCI_VENDOR_ID_3DLABS,PCI_DEVICE_ID_3DLABS_GAMMA,d);
+	if (!d) return 0;
+
+	one = pci_find_device(PCI_VENDOR_ID_3DLABS,PCI_DEVICE_ID_3DLABS_MX,d);
+	if (!one) return 0;
+
+	/* Make sure it's on the same card, if not - no MX's found */
+	if (PCI_SLOT(d->devfn) != PCI_SLOT(one->devfn)) return 0;
+
+	two = pci_find_device(PCI_VENDOR_ID_3DLABS,PCI_DEVICE_ID_3DLABS_MX,one);
+	if (!two) return 1;
+
+	/* Make sure it's on the same card, if not - only 1 MX found */
+	if (PCI_SLOT(d->devfn) != PCI_SLOT(two->devfn)) return 1;
+
+	/* Two MX's found - we don't currently support more than 2 */
+	return 2;
+}
+
 /* gamma_init is called via init_module at module load time, or via
  * linux/init/main.c (this is not currently supported). */
 
@@ -334,6 +366,8 @@ int gamma_init(void)
 #ifdef MODULE
 	drm_parse_options(gamma);
 #endif
+	devices = gamma_find_devices();
+	if (devices == 0) return -1;
 
 	if ((retcode = misc_register(&gamma_misc))) {
 		DRM_ERROR("Cannot register \"%s\"\n", GAMMA_NAME);
@@ -345,13 +379,14 @@ int gamma_init(void)
 	drm_mem_init();
 	drm_proc_init(dev);
 
-	DRM_INFO("Initialized %s %d.%d.%d %s on minor %d\n",
+	DRM_INFO("Initialized %s %d.%d.%d %s on minor %d with %d MX devices\n",
 		 GAMMA_NAME,
 		 GAMMA_MAJOR,
 		 GAMMA_MINOR,
 		 GAMMA_PATCHLEVEL,
 		 GAMMA_DATE,
-		 gamma_misc.minor);
+		 gamma_misc.minor,
+		 devices);
 	
 	return 0;
 }
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drv.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drv.h
index 622cedfe3..55dc26bec 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drv.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/gamma_drv.h
@@ -51,5 +51,7 @@ extern int  gamma_irq_install(drm_device_t *dev, int irq);
 extern int  gamma_irq_uninstall(drm_device_t *dev);
 extern int  gamma_control(struct inode *inode, struct file *filp,
 			  unsigned int cmd, unsigned long arg);
+extern int  gamma_find_devices(void);
+extern int  gamma_found(void);
 
 #endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_bufs.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_bufs.c
index 315f34378..fa1f84dcd 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_bufs.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_bufs.c
@@ -119,6 +119,7 @@ int i810_addbufs_agp(struct inode *inode, struct file *filp, unsigned int cmd,
 		buf->dev_private = drm_alloc(sizeof(drm_i810_buf_priv_t), 
 					     DRM_MEM_BUFS);
 		buf->dev_priv_size = sizeof(drm_i810_buf_priv_t);
+	   	memset(buf->dev_private, 0, sizeof(drm_i810_buf_priv_t));
 
 #if DRM_DMA_HISTOGRAM
 		buf->time_queued = 0;
@@ -331,113 +332,3 @@ int i810_freebufs(struct inode *inode, struct file *filp, unsigned int cmd,
 	return 0;
 }
 
-int i810_mapbufs(struct inode *inode, struct file *filp, unsigned int cmd,
-		 unsigned long arg)
-{
-	drm_file_t	 *priv	 = filp->private_data;
-	drm_device_t	 *dev	 = priv->dev;
-	drm_device_dma_t *dma	 = dev->dma;
-	int		 retcode = 0;
-	const int	 zero	 = 0;
-	unsigned long	 virtual;
-	unsigned long	 address;
-	drm_buf_map_t	 request;
-	int		 i;
-
-	if (!dma) return -EINVAL;
-	
-	DRM_DEBUG("\n");
-
-	spin_lock(&dev->count_lock);
-	if (atomic_read(&dev->buf_alloc)) {
-		spin_unlock(&dev->count_lock);
-		DRM_DEBUG("Busy\n");
-		return -EBUSY;
-	}
-	++dev->buf_use;		/* Can't allocate more after this call */
-	spin_unlock(&dev->count_lock);
-
-	copy_from_user_ret(&request,
-			   (drm_buf_map_t *)arg,
-			   sizeof(request),
-			   -EFAULT);
-   	DRM_DEBUG("dma->flags : %lx\n", dma->flags);
-   	if (request.count >= dma->buf_count) {
-		if(dma->flags & _DRM_DMA_USE_AGP) {
-			drm_i810_private_t *dev_priv = 
-				(drm_i810_private_t *)dev->dev_private;
-			drm_map_t *map = NULL;
-	      
-			map = dev->maplist[dev_priv->buffer_map_idx];
-			if (!map) {
-				DRM_DEBUG("map is null\n");
-				retcode = -EINVAL;
-				goto done;
-			}
-			DRM_DEBUG("map->offset : %lx\n", map->offset);
-			DRM_DEBUG("map->size : %lx\n", map->size);
-			DRM_DEBUG("map->type : %d\n", map->type);
-			DRM_DEBUG("map->flags : %x\n", map->flags);
-			DRM_DEBUG("map->handle : %lx\n", map->handle);
-			DRM_DEBUG("map->mtrr : %d\n", map->mtrr);
-			down(&current->mm->mmap_sem);
-			virtual = do_mmap(filp, 0, map->size, 
-					  PROT_READ|PROT_WRITE,
-					  MAP_SHARED, 
-					  (unsigned long)map->offset);
-
-			up(&current->mm->mmap_sem);
-		} else {
-			down(&current->mm->mmap_sem);
-			virtual = do_mmap(filp, 0, dma->byte_count,
-					  PROT_READ|PROT_WRITE, MAP_SHARED, 0);
-			up(&current->mm->mmap_sem);
-		}
-		if (virtual > -1024UL) {
-			/* Real error */
-			DRM_DEBUG("mmap error\n");
-			retcode = (signed long)virtual;
-			goto done;
-		}
-		request.virtual = (void *)virtual;
-	   
-		for (i = 0; i < dma->buf_count; i++) {
-			if (copy_to_user(&request.list[i].idx,
-					 &dma->buflist[i]->idx,
-					 sizeof(request.list[0].idx))) {
-				retcode = -EFAULT;
-				goto done;
-			}
-			if (copy_to_user(&request.list[i].total,
-					 &dma->buflist[i]->total,
-					 sizeof(request.list[0].total))) {
-				retcode = -EFAULT;
-				goto done;
-			}
-			if (copy_to_user(&request.list[i].used,
-					 &zero,
-					 sizeof(zero))) {
-				retcode = -EFAULT;
-				goto done;
-			}
-			address = virtual + dma->buflist[i]->offset;
-			if (copy_to_user(&request.list[i].address,
-					 &address,
-					 sizeof(address))) {
-				retcode = -EFAULT;
-				goto done;
-			}
-		}
-	}
- done:
-	request.count = dma->buf_count;
-	DRM_DEBUG("%d buffers, retcode = %d\n", request.count, retcode);
-   
-	copy_to_user_ret((drm_buf_map_t *)arg,
-			 &request,
-			 sizeof(request),
-			 -EFAULT);
-
-	DRM_DEBUG("retcode : %d\n", retcode);
-	return retcode;
-}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_dma.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_dma.c
index d82ed049e..94f35b61c 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_dma.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_dma.c
@@ -36,8 +36,17 @@
 
 #include <linux/interrupt.h>	/* For task queue support */
 
-#define I810_BUF_FREE		1
-#define I810_BUF_USED		0
+/* in case we don't have a 2.3.99-pre6 kernel or later: */
+#ifndef VM_DONTCOPY
+#define VM_DONTCOPY 0
+#endif
+
+#define I810_BUF_FREE		2
+#define I810_BUF_CLIENT		1
+#define I810_BUF_HARDWARE      	0
+
+#define I810_BUF_UNMAPPED 0
+#define I810_BUF_MAPPED   1
 
 #define I810_REG(reg)		2
 #define I810_BASE(reg)		((unsigned long) \
@@ -90,7 +99,7 @@ static inline void i810_print_status_page(drm_device_t *dev)
       	DRM_DEBUG(  "hw_status: Reserved : %x\n", temp[3]);
    	DRM_DEBUG(  "hw_status: Driver Counter : %d\n", temp[5]);
    	for(i = 6; i < dma->buf_count + 6; i++) {
-	   	DRM_DEBUG(  "buffer status idx : %d used: %d\n", i - 6, temp[i]);
+	   	DRM_DEBUG( "buffer status idx : %d used: %d\n", i - 6, temp[i]);
 	}
 }
 
@@ -107,7 +116,7 @@ static drm_buf_t *i810_freelist_get(drm_device_t *dev)
 	   	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
 		/* In use is already a pointer */
 	   	used = cmpxchg(buf_priv->in_use, I810_BUF_FREE, 
-			       I810_BUF_USED);
+			       I810_BUF_CLIENT);
 	   	if(used == I810_BUF_FREE) {
 			return buf;
 		}
@@ -125,8 +134,8 @@ static int i810_freelist_put(drm_device_t *dev, drm_buf_t *buf)
    	int used;
    
    	/* In use is already a pointer */
-   	used = cmpxchg(buf_priv->in_use, I810_BUF_USED, I810_BUF_FREE);
-   	if(used != I810_BUF_USED) {
+   	used = cmpxchg(buf_priv->in_use, I810_BUF_CLIENT, I810_BUF_FREE);
+   	if(used != I810_BUF_CLIENT) {
 	   	DRM_ERROR("Freeing buffer thats not in use : %d\n", buf->idx);
 	   	return -EINVAL;
 	}
@@ -134,26 +143,114 @@ static int i810_freelist_put(drm_device_t *dev, drm_buf_t *buf)
    	return 0;
 }
 
-static int i810_dma_get_buffers(drm_device_t *dev, drm_dma_t *d)
+static struct file_operations i810_buffer_fops = {
+	open:	 i810_open,
+	flush:	 drm_flush,
+	release: i810_release,
+	ioctl:	 i810_ioctl,
+	mmap:	 i810_mmap_buffers,
+	read:	 drm_read,
+	fasync:	 drm_fasync,
+      	poll:	 drm_poll,
+};
+
+int i810_mmap_buffers(struct file *filp, struct vm_area_struct *vma)
+{
+	drm_file_t	  *priv	  = filp->private_data;
+	drm_device_t	  *dev	  = priv->dev;
+	drm_i810_private_t *dev_priv = dev->dev_private;
+	drm_buf_t *buf = dev_priv->mmap_buffer;
+	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+   
+	vma->vm_flags |= (VM_IO | VM_DONTCOPY);
+	vma->vm_file = filp;
+   
+   	buf_priv->currently_mapped = I810_BUF_MAPPED;
+
+	if (remap_page_range(vma->vm_start,
+			     VM_OFFSET(vma),
+			     vma->vm_end - vma->vm_start,
+			     vma->vm_page_prot)) return -EAGAIN;
+	return 0;
+}
+
+static int i810_map_buffer(drm_buf_t *buf, struct file *filp)
+{
+	drm_file_t	  *priv	  = filp->private_data;
+	drm_device_t	  *dev	  = priv->dev;
+	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+      	drm_i810_private_t *dev_priv = dev->dev_private;
+   	struct file_operations *old_fops;
+	int retcode = 0;
+
+	if(buf_priv->currently_mapped == I810_BUF_MAPPED) return -EINVAL;
+	down(&current->mm->mmap_sem);
+   	old_fops = filp->f_op;
+	filp->f_op = &i810_buffer_fops;
+	dev_priv->mmap_buffer = buf;
+	buf_priv->virtual = (void *)do_mmap(filp, 0, buf->total, 
+					    PROT_READ|PROT_WRITE,
+					    MAP_SHARED, 
+					    buf->bus_address);
+	dev_priv->mmap_buffer = NULL;
+   	filp->f_op = old_fops;
+	if ((unsigned long)buf_priv->virtual > -1024UL) {
+		/* Real error */
+		DRM_DEBUG("mmap error\n");
+		retcode = (signed int)buf_priv->virtual;
+		buf_priv->virtual = 0;
+	}
+   	up(&current->mm->mmap_sem);
+	return retcode;
+}
+
+static int i810_unmap_buffer(drm_buf_t *buf)
+{
+	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+	int retcode = 0;
+
+	if(buf_priv->currently_mapped != I810_BUF_MAPPED) return -EINVAL;
+	down(&current->mm->mmap_sem);
+        retcode = do_munmap((unsigned long)buf_priv->virtual, 
+			    (size_t) buf->total);
+   	buf_priv->currently_mapped = I810_BUF_UNMAPPED;
+   	buf_priv->virtual = 0;
+   	up(&current->mm->mmap_sem);
+
+	return retcode;
+}
+
+static int i810_dma_get_buffer(drm_device_t *dev, drm_i810_dma_t *d, 
+			       struct file *filp)
 {
-	int		  i;
+	drm_file_t	  *priv	  = filp->private_data;
 	drm_buf_t	  *buf;
+	drm_i810_buf_priv_t *buf_priv;
+	int retcode = 0;
 
-	for (i = d->granted_count; i < d->request_count; i++) {
-		buf = i810_freelist_get(dev);
-		if (!buf) break;
-		buf->pid     = current->pid;
-		copy_to_user_ret(&d->request_indices[i],
-				 &buf->idx,
-				 sizeof(buf->idx),
-				 -EFAULT);
-		copy_to_user_ret(&d->request_sizes[i],
-				 &buf->total,
-				 sizeof(buf->total),
-				 -EFAULT);
-		++d->granted_count;
+	buf = i810_freelist_get(dev);
+	if (!buf) {
+		retcode = -ENOMEM;
+	   	DRM_DEBUG("%s retcode %d\n", __FUNCTION__, retcode);
+		goto out_get_buf;
 	}
-	return 0;
+   
+	retcode = i810_map_buffer(buf, filp);
+	if(retcode) {
+		i810_freelist_put(dev, buf);
+	   	DRM_DEBUG("mapbuf failed in %s retcode %d\n", 
+			  __FUNCTION__, retcode);
+	   	goto out_get_buf;
+	}
+	buf->pid     = priv->pid;
+	buf_priv = buf->dev_private;	
+	d->granted = 1;
+   	d->request_idx = buf->idx;
+   	d->request_size = buf->total;
+   	d->virtual = buf_priv->virtual;
+
+out_get_buf:
+	return retcode;
 }
 
 static unsigned long i810_alloc_page(drm_device_t *dev)
@@ -184,7 +281,10 @@ static void i810_free_page(drm_device_t *dev, unsigned long page)
 
 static int i810_dma_cleanup(drm_device_t *dev)
 {
+	drm_device_dma_t *dma = dev->dma;
+
 	if(dev->dev_private) {
+		int i;
 	   	drm_i810_private_t *dev_priv = 
 	     		(drm_i810_private_t *) dev->dev_private;
 	   
@@ -200,6 +300,12 @@ static int i810_dma_cleanup(drm_device_t *dev)
 	   	drm_free(dev->dev_private, sizeof(drm_i810_private_t), 
 			 DRM_MEM_DRIVER);
 	   	dev->dev_private = NULL;
+
+		for (i = 0; i < dma->buf_count; i++) {
+			drm_buf_t *buf = dma->buflist[ i ];
+			drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+			drm_ioremapfree(buf_priv->kernel_virtual, buf->total);
+		}
 	}
    	return 0;
 }
@@ -210,6 +316,7 @@ static int i810_wait_ring(drm_device_t *dev, int n)
    	drm_i810_ring_buffer_t *ring = &(dev_priv->ring);
    	int iters = 0;
    	unsigned long end;
+	unsigned int last_head = I810_READ(LP_RING + RING_HEAD) & HEAD_ADDR;
 
 	end = jiffies + (HZ*3);
    	while (ring->space < n) {
@@ -217,9 +324,11 @@ static int i810_wait_ring(drm_device_t *dev, int n)
 	
 	   	ring->head = I810_READ(LP_RING + RING_HEAD) & HEAD_ADDR;
 	   	ring->space = ring->head - (ring->tail+8);
+		if (ring->space < 0) ring->space += ring->Size;
 	   
-	   	if (ring->space < 0) ring->space += ring->Size;
-	   
+		if (ring->head != last_head)
+		   end = jiffies + (HZ*3);
+	  
 	   	iters++;
 		if((signed)(end - jiffies) <= 0) {
 		   	DRM_ERROR("space: %d wanted %d\n", ring->space, n);
@@ -249,9 +358,9 @@ static int i810_freelist_init(drm_device_t *dev)
 {
       	drm_device_dma_t *dma = dev->dma;
    	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
-   	u8 *hw_status = (u8 *)dev_priv->hw_status_page;
-   	int i;
    	int my_idx = 24;
+   	u32 *hw_status = (u32 *)(dev_priv->hw_status_page + my_idx);
+   	int i;
    
    	if(dma->buf_count > 1019) {
 	   	/* Not enough space in the status page for the freelist */
@@ -262,11 +371,14 @@ static int i810_freelist_init(drm_device_t *dev)
 	   	drm_buf_t *buf = dma->buflist[ i ];
 	   	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
 	   
-	   	buf_priv->in_use = hw_status + my_idx;
-	   	DRM_DEBUG("buf_priv->in_use : %p\n", buf_priv->in_use);
-	   	*buf_priv->in_use = I810_BUF_FREE;
+	   	buf_priv->in_use = hw_status++;
 	   	buf_priv->my_use_idx = my_idx;
 	   	my_idx += 4;
+
+	   	*buf_priv->in_use = I810_BUF_FREE;
+
+		buf_priv->kernel_virtual = drm_ioremap(buf->bus_address, 
+						       buf->total);
 	}
 	return 0;
 }
@@ -300,9 +412,11 @@ static int i810_dma_initialize(drm_device_t *dev,
    	dev_priv->ring.Start = init->ring_start;
    	dev_priv->ring.End = init->ring_end;
    	dev_priv->ring.Size = init->ring_size;
+
    	dev_priv->ring.virtual_start = drm_ioremap(dev->agp->base + 
 						   init->ring_start, 
 						   init->ring_size);
+
    	dev_priv->ring.tail_mask = dev_priv->ring.Size - 1;
    
    	if (dev_priv->ring.virtual_start == NULL) {
@@ -311,6 +425,17 @@ static int i810_dma_initialize(drm_device_t *dev,
 			  " ring buffer\n");
 	   	return -ENOMEM;
 	}
+
+	dev_priv->w = init->w;
+	dev_priv->h = init->h;
+	dev_priv->pitch = init->pitch;
+	dev_priv->back_offset = init->back_offset;
+	dev_priv->depth_offset = init->depth_offset;
+
+	dev_priv->front_di1 = init->front_offset | init->pitch_bits;
+	dev_priv->back_di1 = init->back_offset | init->pitch_bits;
+	dev_priv->zi1 = init->depth_offset | init->pitch_bits;
+	
    
    	/* Program Hardware Status Page */
    	dev_priv->hw_status_page = i810_alloc_page(dev);
@@ -365,37 +490,270 @@ int i810_dma_init(struct inode *inode, struct file *filp,
    	return retcode;
 }
 
-static void i810_dma_dispatch_general(drm_device_t *dev, drm_buf_t *buf,
-				      int used )
+
+
+/* Most efficient way to verify state for the i810 is as it is
+ * emitted.  Non-conformant state is silently dropped.
+ *
+ * Use 'volatile' & local var tmp to force the emitted values to be
+ * identical to the verified ones.
+ */
+static void i810EmitContextVerified( drm_device_t *dev, 
+				     volatile unsigned int *code ) 
+{	
+   	drm_i810_private_t *dev_priv = dev->dev_private;
+	int i, j = 0;
+	unsigned int tmp;
+	RING_LOCALS;
+
+	BEGIN_LP_RING( I810_CTX_SETUP_SIZE );
+
+	OUT_RING( GFX_OP_COLOR_FACTOR );
+	OUT_RING( code[I810_CTXREG_CF1] );
+
+	OUT_RING( GFX_OP_STIPPLE );
+	OUT_RING( code[I810_CTXREG_ST1] );
+
+	for ( i = 4 ; i < I810_CTX_SETUP_SIZE ; i++ ) {
+		tmp = code[i];
+
+		if ((tmp & (7<<29)) == (3<<29) &&
+		    (tmp & (0x1f<<24)) < (0x1d<<24)) 
+		{
+			OUT_RING( tmp ); 
+			j++;
+		} 
+	}
+
+	if (j & 1) 
+		OUT_RING( 0 ); 
+
+	ADVANCE_LP_RING();
+}
+
+static void i810EmitTexVerified( drm_device_t *dev, 
+				 volatile unsigned int *code ) 
+{	
+   	drm_i810_private_t *dev_priv = dev->dev_private;
+	int i, j = 0;
+	unsigned int tmp;
+	RING_LOCALS;
+
+	BEGIN_LP_RING( I810_TEX_SETUP_SIZE );
+
+	OUT_RING( GFX_OP_MAP_INFO );
+	OUT_RING( code[I810_TEXREG_MI1] );
+	OUT_RING( code[I810_TEXREG_MI2] );
+	OUT_RING( code[I810_TEXREG_MI3] );
+
+	for ( i = 4 ; i < I810_TEX_SETUP_SIZE ; i++ ) {
+		tmp = code[i];
+
+		if ((tmp & (7<<29)) == (3<<29) &&
+		    (tmp & (0x1f<<24)) < (0x1d<<24)) 
+		{
+			OUT_RING( tmp ); 
+			j++;
+		}
+	} 
+		
+	if (j & 1) 
+		OUT_RING( 0 ); 
+
+	ADVANCE_LP_RING();
+}
+
+
+/* Need to do some additional checking when setting the dest buffer.
+ */
+static void i810EmitDestVerified( drm_device_t *dev, 
+				  volatile unsigned int *code ) 
+{	
+   	drm_i810_private_t *dev_priv = dev->dev_private;
+	unsigned int tmp;
+	RING_LOCALS;
+
+	BEGIN_LP_RING( I810_DEST_SETUP_SIZE + 2 );
+
+	tmp = code[I810_DESTREG_DI1];
+	if (tmp == dev_priv->front_di1 || tmp == dev_priv->back_di1) {
+		OUT_RING( CMD_OP_DESTBUFFER_INFO );
+		OUT_RING( tmp );
+	} else
+	   DRM_DEBUG("bad di1 %x (allow %x or %x)\n",
+		     tmp, dev_priv->front_di1, dev_priv->back_di1);
+
+	/* invarient:
+	 */
+	OUT_RING( CMD_OP_Z_BUFFER_INFO );
+	OUT_RING( dev_priv->zi1 );
+
+	OUT_RING( GFX_OP_DESTBUFFER_VARS );
+	OUT_RING( code[I810_DESTREG_DV1] );
+
+	OUT_RING( GFX_OP_DRAWRECT_INFO );
+	OUT_RING( code[I810_DESTREG_DR1] );
+	OUT_RING( code[I810_DESTREG_DR2] );
+	OUT_RING( code[I810_DESTREG_DR3] );
+	OUT_RING( code[I810_DESTREG_DR4] );
+	OUT_RING( 0 );
+
+	ADVANCE_LP_RING();
+}
+
+
+
+static void i810EmitState( drm_device_t *dev )
 {
-      	drm_i810_private_t *dev_priv = dev->dev_private;
-   	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
-	unsigned long address = (unsigned long)buf->bus_address;
-	unsigned long start = address - dev->agp->base;
-   	RING_LOCALS;
+   	drm_i810_private_t *dev_priv = dev->dev_private;
+      	drm_i810_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	unsigned int dirty = sarea_priv->dirty;
 
-   	dev_priv->counter++;
-   	DRM_DEBUG(  "dispatch counter : %ld\n", dev_priv->counter);
-   	DRM_DEBUG(  "i810_dma_dispatch\n");
-   	DRM_DEBUG(  "start : 0x%lx\n", start);
-	DRM_DEBUG(  "used : 0x%x\n", used);
-   	DRM_DEBUG(  "start + used - 4 : 0x%lx\n", start + used - 4);
-   	i810_kernel_lost_context(dev);
+	if (dirty & I810_UPLOAD_BUFFERS) {
+		i810EmitDestVerified( dev, sarea_priv->BufferState );
+		sarea_priv->dirty &= ~I810_UPLOAD_BUFFERS;
+	}
 
-   	BEGIN_LP_RING(10);
-   	OUT_RING( CMD_OP_BATCH_BUFFER );
-   	OUT_RING( start | BB1_PROTECTED );
-   	OUT_RING( start + used - 4 );
-      	OUT_RING( CMD_STORE_DWORD_IDX );
-   	OUT_RING( 20 );
-   	OUT_RING( dev_priv->counter );
-   	OUT_RING( CMD_STORE_DWORD_IDX );
-   	OUT_RING( buf_priv->my_use_idx );
-   	OUT_RING( I810_BUF_FREE );   
-      	OUT_RING( CMD_REPORT_HEAD );
-      	ADVANCE_LP_RING();
+	if (dirty & I810_UPLOAD_CTX) {
+		i810EmitContextVerified( dev, sarea_priv->ContextState );
+		sarea_priv->dirty &= ~I810_UPLOAD_CTX;
+	}
+
+	if (dirty & I810_UPLOAD_TEX0) {
+		i810EmitTexVerified( dev, sarea_priv->TexState[0] );
+		sarea_priv->dirty &= ~I810_UPLOAD_TEX0;
+	}
+
+	if (dirty & I810_UPLOAD_TEX1) {
+		i810EmitTexVerified( dev, sarea_priv->TexState[1] );
+		sarea_priv->dirty &= ~I810_UPLOAD_TEX1;
+	}
 }
 
+
+
+/* need to verify 
+ */
+static void i810_dma_dispatch_clear( drm_device_t *dev, int flags, 
+				     unsigned int clear_color,
+				     unsigned int clear_zval )
+{
+   	drm_i810_private_t *dev_priv = dev->dev_private;
+      	drm_i810_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	int nbox = sarea_priv->nbox;
+	drm_clip_rect_t *pbox = sarea_priv->boxes;
+	int pitch = dev_priv->pitch;
+	int cpp = 2;
+	int i;
+	RING_LOCALS;
+
+  	i810_kernel_lost_context(dev);
+
+      	if (nbox > I810_NR_SAREA_CLIPRECTS)
+     		nbox = I810_NR_SAREA_CLIPRECTS;
+
+	for (i = 0 ; i < nbox ; i++, pbox++) {
+		unsigned int x = pbox->x1;
+		unsigned int y = pbox->y1;
+		unsigned int width = (pbox->x2 - x) * cpp;
+		unsigned int height = pbox->y2 - y;
+		unsigned int start = y * pitch + x * cpp;
+
+		if (pbox->x1 > pbox->x2 ||
+		    pbox->y1 > pbox->y2 ||
+		    pbox->x2 > dev_priv->w ||
+		    pbox->y2 > dev_priv->h)
+			continue;
+
+	   	if ( flags & I810_FRONT ) {	    
+		   	DRM_DEBUG("clear front\n");
+			BEGIN_LP_RING( 6 );	    
+			OUT_RING( BR00_BITBLT_CLIENT | 
+				  BR00_OP_COLOR_BLT | 0x3 );
+			OUT_RING( BR13_SOLID_PATTERN | (0xF0 << 16) | pitch );
+			OUT_RING( (height << 16) | width );
+			OUT_RING( start );
+			OUT_RING( clear_color );
+			OUT_RING( 0 );
+			ADVANCE_LP_RING();
+		}
+
+		if ( flags & I810_BACK ) {
+			DRM_DEBUG("clear back\n");
+			BEGIN_LP_RING( 6 );	    
+			OUT_RING( BR00_BITBLT_CLIENT | 
+				  BR00_OP_COLOR_BLT | 0x3 );
+			OUT_RING( BR13_SOLID_PATTERN | (0xF0 << 16) | pitch );
+			OUT_RING( (height << 16) | width );
+			OUT_RING( dev_priv->back_offset + start );
+			OUT_RING( clear_color );
+			OUT_RING( 0 );
+			ADVANCE_LP_RING();
+		}
+
+		if ( flags & I810_DEPTH ) {
+			DRM_DEBUG("clear depth\n");
+			BEGIN_LP_RING( 6 );	    
+			OUT_RING( BR00_BITBLT_CLIENT | 
+				  BR00_OP_COLOR_BLT | 0x3 );
+			OUT_RING( BR13_SOLID_PATTERN | (0xF0 << 16) | pitch );
+			OUT_RING( (height << 16) | width );
+			OUT_RING( dev_priv->depth_offset + start );
+			OUT_RING( clear_zval );
+			OUT_RING( 0 );
+			ADVANCE_LP_RING();
+		}
+	}
+}
+
+static void i810_dma_dispatch_swap( drm_device_t *dev )
+{
+   	drm_i810_private_t *dev_priv = dev->dev_private;
+      	drm_i810_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	int nbox = sarea_priv->nbox;
+	drm_clip_rect_t *pbox = sarea_priv->boxes;
+	int pitch = dev_priv->pitch;
+	int cpp = 2;
+	int ofs = dev_priv->back_offset;
+	int i;
+	RING_LOCALS;
+
+	DRM_DEBUG("swapbuffers\n");
+
+  	i810_kernel_lost_context(dev);
+
+      	if (nbox > I810_NR_SAREA_CLIPRECTS)
+     		nbox = I810_NR_SAREA_CLIPRECTS;
+
+	for (i = 0 ; i < nbox; i++, pbox++) 
+	{
+		unsigned int w = pbox->x2 - pbox->x1;
+		unsigned int h = pbox->y2 - pbox->y1;
+		unsigned int dst = pbox->x1*cpp + pbox->y1*pitch;
+		unsigned int start = ofs + dst;
+
+		if (pbox->x1 > pbox->x2 ||
+		    pbox->y1 > pbox->y2 ||
+		    pbox->x2 > dev_priv->w ||
+		    pbox->y2 > dev_priv->h)
+			continue;
+ 
+	   	DRM_DEBUG("dispatch swap %d,%d-%d,%d!\n",
+			  pbox[i].x1, pbox[i].y1,
+			  pbox[i].x2, pbox[i].y2);
+
+		BEGIN_LP_RING( 6 );
+		OUT_RING( BR00_BITBLT_CLIENT | BR00_OP_SRC_COPY_BLT | 0x4 );
+		OUT_RING( pitch | (0xCC << 16));
+		OUT_RING( (h << 16) | (w * cpp));
+		OUT_RING( dst );
+		OUT_RING( pitch );	
+		OUT_RING( start );
+		ADVANCE_LP_RING();
+	}
+}
+
+
 static void i810_dma_dispatch_vertex(drm_device_t *dev, 
 				     drm_buf_t *buf,
 				     int discard,
@@ -408,14 +766,29 @@ static void i810_dma_dispatch_vertex(drm_device_t *dev,
    	int nbox = sarea_priv->nbox;
 	unsigned long address = (unsigned long)buf->bus_address;
 	unsigned long start = address - dev->agp->base;     
-	int i = 0;
+	int i = 0, u;
    	RING_LOCALS;
 
-   
+   	i810_kernel_lost_context(dev);
+
    	if (nbox > I810_NR_SAREA_CLIPRECTS) 
 		nbox = I810_NR_SAREA_CLIPRECTS;
-   
-   	DRM_DEBUG("dispatch vertex addr 0x%lx, used 0x%x nbox %d\n", 
+
+	if (discard) {
+		u = cmpxchg(buf_priv->in_use, I810_BUF_CLIENT, 
+			    I810_BUF_HARDWARE);
+		if(u != I810_BUF_CLIENT) {
+			DRM_DEBUG("xxxx 2\n");
+		}
+	}
+
+	if (used > 4*1024) 
+		used = 0;
+
+	if (sarea_priv->dirty)
+	   i810EmitState( dev );
+
+  	DRM_DEBUG("dispatch vertex addr 0x%lx, used 0x%x nbox %d\n", 
 		  address, used, nbox);
 
    	dev_priv->counter++;
@@ -424,8 +797,20 @@ static void i810_dma_dispatch_vertex(drm_device_t *dev,
    	DRM_DEBUG(  "start : %lx\n", start);
 	DRM_DEBUG(  "used : %d\n", used);
    	DRM_DEBUG(  "start + used - 4 : %ld\n", start + used - 4);
-   	i810_kernel_lost_context(dev);
 
+	if (buf_priv->currently_mapped == I810_BUF_MAPPED) {
+		*(u32 *)buf_priv->virtual = (GFX_OP_PRIMITIVE |
+					     sarea_priv->vertex_prim |
+					     ((used/4)-2));
+		
+		if (used & 4) {
+			*(u32 *)((u32)buf_priv->virtual + used) = 0;
+			used += 4;
+		}
+
+		i810_unmap_buffer(buf);
+	}
+		   
 	if (used) {
 		do {
 			if (i < nbox) {
@@ -433,7 +818,7 @@ static void i810_dma_dispatch_vertex(drm_device_t *dev,
 				OUT_RING( GFX_OP_SCISSOR | SC_UPDATE_SCISSOR | 
 					  SC_ENABLE );
 				OUT_RING( GFX_OP_SCISSOR_INFO );
-				OUT_RING( box[i].x1 | (box[i].y1 << 16) );
+				OUT_RING( box[i].x1 | (box[i].y1<<16) );
 				OUT_RING( (box[i].x2-1) | ((box[i].y2-1)<<16) );
 				ADVANCE_LP_RING();
 			}
@@ -478,7 +863,9 @@ static void i810_dma_service(int irq, void *device, struct pt_regs *regs)
    	temp = temp & ~(0x6000);
    	if(temp != 0) I810_WRITE16(I810REG_INT_IDENTITY_R, 
 				   temp); /* Clear all interrupts */
-   
+	else
+	   return;
+ 
    	queue_task(&dev->tq, &tq_immediate);
    	mark_bh(IMMEDIATE_BH);
 }
@@ -538,7 +925,7 @@ int i810_irq_install(drm_device_t *dev, int irq)
 				/* Install handler */
 	if ((retcode = request_irq(dev->irq,
 				   i810_dma_service,
-				   0,
+				   SA_SHIRQ,
 				   dev->devname,
 				   dev))) {
 		down(&dev->struct_sem);
@@ -559,6 +946,9 @@ int i810_irq_uninstall(drm_device_t *dev)
 	int irq;
    	u16 temp;
 
+
+/*  	return 0; */
+
 	down(&dev->struct_sem);
 	irq	 = dev->irq;
 	dev->irq = 0;
@@ -617,10 +1007,15 @@ static inline void i810_dma_emit_flush(drm_device_t *dev)
    	RING_LOCALS;
 
    	i810_kernel_lost_context(dev);
+
    	BEGIN_LP_RING(2);
       	OUT_RING( CMD_REPORT_HEAD );
-   	OUT_RING( GFX_OP_USER_INTERRUPT );
+      	OUT_RING( GFX_OP_USER_INTERRUPT );
       	ADVANCE_LP_RING();
+
+/*  	i810_wait_ring( dev, dev_priv->ring.Size - 8 ); */
+/*     	atomic_set(&dev_priv->flush_done, 1); */
+/*     	wake_up_interruptible(&dev_priv->flush_queue); */
 }
 
 static inline void i810_dma_quiescent_emit(drm_device_t *dev)
@@ -629,13 +1024,17 @@ static inline void i810_dma_quiescent_emit(drm_device_t *dev)
    	RING_LOCALS;
 
   	i810_kernel_lost_context(dev);
-   	BEGIN_LP_RING(4);
 
+   	BEGIN_LP_RING(4);
    	OUT_RING( INST_PARSER_CLIENT | INST_OP_FLUSH | INST_FLUSH_MAP_CACHE );
    	OUT_RING( CMD_REPORT_HEAD );
-   	OUT_RING( GFX_OP_USER_INTERRUPT );
-   	OUT_RING( 0 );
+      	OUT_RING( 0 );
+      	OUT_RING( GFX_OP_USER_INTERRUPT );
    	ADVANCE_LP_RING();
+
+/*  	i810_wait_ring( dev, dev_priv->ring.Size - 8 ); */
+/*     	atomic_set(&dev_priv->flush_done, 1); */
+/*     	wake_up_interruptible(&dev_priv->flush_queue); */
 }
 
 static void i810_dma_quiescent(drm_device_t *dev)
@@ -675,8 +1074,9 @@ static int i810_flush_queue(drm_device_t *dev)
 {
    	DECLARE_WAITQUEUE(entry, current);
   	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
+	drm_device_dma_t *dma = dev->dma;
 	unsigned long end;
-   	int ret = 0;      
+   	int i, ret = 0;      
 
    	if(dev_priv == NULL) {
 	   	return 0;
@@ -701,7 +1101,21 @@ static int i810_flush_queue(drm_device_t *dev)
    
    	current->state = TASK_RUNNING;
    	remove_wait_queue(&dev_priv->flush_queue, &entry);
-   
+
+
+   	for (i = 0; i < dma->buf_count; i++) {
+	   	drm_buf_t *buf = dma->buflist[ i ];
+	   	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+	   
+		int used = cmpxchg(buf_priv->in_use, I810_BUF_HARDWARE, 
+				   I810_BUF_FREE);
+
+		if (used == I810_BUF_HARDWARE)
+			DRM_DEBUG("reclaimed from HARDWARE\n");
+		if (used == I810_BUF_CLIENT)
+			DRM_DEBUG("still on client HARDWARE\n");
+	}
+
    	return ret;
 }
 
@@ -712,20 +1126,23 @@ void i810_reclaim_buffers(drm_device_t *dev, pid_t pid)
 	int		 i;
 
 	if (!dma) return;
-      	if(dev->dev_private == NULL) return;
-	if(dma->buflist == NULL) return;
+      	if (!dev->dev_private) return;
+	if (!dma->buflist) return;
+
         i810_flush_queue(dev);
 
 	for (i = 0; i < dma->buf_count; i++) {
 	   	drm_buf_t *buf = dma->buflist[ i ];
 	   	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
 	   
-		/* Only buffers that need to get reclaimed ever 
-		 * get set to free 
-		 */
 		if (buf->pid == pid && buf_priv) {
-		   	cmpxchg(buf_priv->in_use, 
-				I810_BUF_USED, I810_BUF_FREE);
+			int used = cmpxchg(buf_priv->in_use, I810_BUF_CLIENT, 
+					   I810_BUF_FREE);
+
+			if (used == I810_BUF_CLIENT)
+				DRM_DEBUG("reclaimed from client\n");
+		   	if(buf_priv->currently_mapped == I810_BUF_MAPPED)
+		     		buf_priv->currently_mapped = I810_BUF_UNMAPPED;
 		}
 	}
 }
@@ -759,19 +1176,6 @@ int i810_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 	 */
 
 	if (!ret) {
-#if 0
-	   	if (_DRM_LOCKING_CONTEXT(dev->lock.hw_lock->lock)
-		    != lock.context) {
-			long j = jiffies - dev->lock.lock_time;
-
-			if (j > 0 && j <= DRM_LOCK_SLICE) {
-				/* Can't take lock if we just had it and
-				   there is contention. */
-				current->state = TASK_INTERRUPTIBLE;
-				schedule_timeout(j);
-			}
-		}
-#endif
 		add_wait_queue(&dev->lock.lock_queue, &entry);
 		for (;;) {
 			if (!dev->lock.hw_lock) {
@@ -828,90 +1232,80 @@ int i810_flush_ioctl(struct inode *inode, struct file *filp,
    	return 0;
 }
 
-static int i810DmaGeneral(drm_device_t *dev, drm_i810_general_t *args)
-{
-	drm_device_dma_t *dma = dev->dma;
-   	drm_buf_t *buf = dma->buflist[ args->idx ];
-   	
-	if (!args->used) {
-	   	i810_freelist_put(dev, buf);
-	} else {  
-		i810_dma_dispatch_general( dev, buf, args->used );
-		atomic_add(args->used, &dma->total_bytes);
-		atomic_inc(&dma->total_dmas);
-	}
-	return 0; 
-}
 
-static int i810DmaVertex(drm_device_t *dev, drm_i810_vertex_t *args)
-{
-	drm_device_dma_t *dma = dev->dma;
-	drm_buf_t *buf = dma->buflist[ args->idx ];
-   	i810_dma_dispatch_vertex( dev, buf, args->discard, args->used );
-   	atomic_add(args->used, &dma->total_bytes);
-	atomic_inc(&dma->total_dmas);
-   	return 0;
-}
-
-int i810_dma_general(struct inode *inode, struct file *filp,
-		     unsigned int cmd, unsigned long arg)
+int i810_dma_vertex(struct inode *inode, struct file *filp,
+	       unsigned int cmd, unsigned long arg)
 {
 	drm_file_t *priv = filp->private_data;
 	drm_device_t *dev = priv->dev;
-	drm_i810_general_t general;
-      	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
+	drm_device_dma_t *dma = dev->dma;
+   	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
       	u32 *hw_status = (u32 *)dev_priv->hw_status_page;
    	drm_i810_sarea_t *sarea_priv = (drm_i810_sarea_t *) 
      					dev_priv->sarea_priv; 
+	drm_i810_vertex_t vertex;
 
-	int retcode = 0;
-	
-	copy_from_user_ret(&general, (drm_i810_general_t *)arg, sizeof(general),
+	copy_from_user_ret(&vertex, (drm_i810_vertex_t *)arg, sizeof(vertex),
 			   -EFAULT);
 
-	DRM_DEBUG("i810 dma general idx %d used %d\n",
-		  general.idx, general.used);
-   
    	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
-		DRM_ERROR("i810_dma_general called without lock held\n");
+		DRM_ERROR("i810_dma_vertex called without lock held\n");
 		return -EINVAL;
 	}
 
-	retcode = i810DmaGeneral(dev, &general);
+	DRM_DEBUG("i810 dma vertex, idx %d used %d discard %d\n",
+		  vertex.idx, vertex.used, vertex.discard);
+
+	i810_dma_dispatch_vertex( dev, 
+				  dma->buflist[ vertex.idx ], 
+				  vertex.discard, vertex.used );
+
+   	atomic_add(vertex.used, &dma->total_bytes);
+	atomic_inc(&dma->total_dmas);
 	sarea_priv->last_enqueue = dev_priv->counter-1;
    	sarea_priv->last_dispatch = (int) hw_status[5];
    
-	return retcode;
+	return 0;
 }
 
-int i810_dma_vertex(struct inode *inode, struct file *filp,
-	       unsigned int cmd, unsigned long arg)
+
+
+int i810_clear_bufs(struct inode *inode, struct file *filp,
+		   unsigned int cmd, unsigned long arg)
 {
 	drm_file_t *priv = filp->private_data;
 	drm_device_t *dev = priv->dev;
-   	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
-      	u32 *hw_status = (u32 *)dev_priv->hw_status_page;
-   	drm_i810_sarea_t *sarea_priv = (drm_i810_sarea_t *) 
-     					dev_priv->sarea_priv; 
-	drm_i810_vertex_t vertex;
-	int retcode = 0;
+	drm_i810_clear_t clear;
 
-	copy_from_user_ret(&vertex, (drm_i810_vertex_t *)arg, sizeof(vertex),
+   	copy_from_user_ret(&clear, (drm_i810_clear_t *)arg, sizeof(clear), 
 			   -EFAULT);
+   
    	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
-		DRM_ERROR("i810_dma_vertex called without lock held\n");
+		DRM_ERROR("i810_clear_bufs called without lock held\n");
 		return -EINVAL;
 	}
 
-	DRM_DEBUG("i810 dma vertex, idx %d used %d discard %d\n",
-		  vertex.idx, vertex.used, vertex.discard);
+	i810_dma_dispatch_clear( dev, clear.flags, 
+				 clear.clear_color, 
+				 clear.clear_depth );
+   	return 0;
+}
 
-	retcode = i810DmaVertex(dev, &vertex);
-	sarea_priv->last_enqueue = dev_priv->counter-1;
-   	sarea_priv->last_dispatch = (int) hw_status[5];
+int i810_swap_bufs(struct inode *inode, struct file *filp,
+		  unsigned int cmd, unsigned long arg)
+{
+	drm_file_t *priv = filp->private_data;
+	drm_device_t *dev = priv->dev;
    
-	return retcode;
+	DRM_DEBUG("i810_swap_bufs\n");
 
+   	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("i810_swap_buf called without lock held\n");
+		return -EINVAL;
+	}
+
+	i810_dma_dispatch_swap( dev );
+   	return 0;
 }
 
 int i810_getage(struct inode *inode, struct file *filp, unsigned int cmd,
@@ -928,53 +1322,32 @@ int i810_getage(struct inode *inode, struct file *filp, unsigned int cmd,
 	return 0;
 }
 
-int i810_dma(struct inode *inode, struct file *filp, unsigned int cmd,
-	    unsigned long arg)
+int i810_getbuf(struct inode *inode, struct file *filp, unsigned int cmd,
+		unsigned long arg)
 {
 	drm_file_t	  *priv	    = filp->private_data;
 	drm_device_t	  *dev	    = priv->dev;
-	drm_device_dma_t  *dma	    = dev->dma;
 	int		  retcode   = 0;
-	drm_dma_t	  d;
+	drm_i810_dma_t	  d;
    	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
    	u32 *hw_status = (u32 *)dev_priv->hw_status_page;
    	drm_i810_sarea_t *sarea_priv = (drm_i810_sarea_t *) 
      					dev_priv->sarea_priv; 
 
-
-   	copy_from_user_ret(&d, (drm_dma_t *)arg, sizeof(d), -EFAULT);
-	DRM_DEBUG("%d %d: %d send, %d req\n",
-		  current->pid, d.context, d.send_count, d.request_count);
+	DRM_DEBUG("getbuf\n");
+   	copy_from_user_ret(&d, (drm_i810_dma_t *)arg, sizeof(d), -EFAULT);
    
 	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
 		DRM_ERROR("i810_dma called without lock held\n");
 		return -EINVAL;
 	}
-
-	/* Please don't send us buffers.
-	 */
-	if (d.send_count != 0) {
-		DRM_ERROR("Process %d trying to send %d buffers via drmDMA\n",
-			  current->pid, d.send_count);
-		return -EINVAL;
-	}
 	
-	/* We'll send you buffers.
-	 */
-	if (d.request_count < 0 || d.request_count > dma->buf_count) {
-		DRM_ERROR("Process %d trying to get %d buffers (of %d max)\n",
-			  current->pid, d.request_count, dma->buf_count);
-		return -EINVAL;
-	}
-	
-	d.granted_count = 0;
+	d.granted = 0;
 
-	if (!retcode && d.request_count) {
-		retcode = i810_dma_get_buffers(dev, &d);
-	}
+	retcode = i810_dma_get_buffer(dev, &d, filp);
 
-	DRM_DEBUG("i810_dma: %d returning, granted = %d\n",
-		  current->pid, d.granted_count);
+	DRM_DEBUG("i810_dma: %d returning %d, granted = %d\n",
+		  current->pid, retcode, d.granted);
 
 	copy_to_user_ret((drm_dma_t *)arg, &d, sizeof(d), -EFAULT);   
    	sarea_priv->last_dispatch = (int) hw_status[5];
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_drm.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_drm.h
new file mode 100644
index 000000000..4c8e09f6a
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_drm.h
@@ -0,0 +1,188 @@
+#ifndef _I810_DRM_H_
+#define _I810_DRM_H_
+
+/* WARNING: These defines must be the same as what the Xserver uses.
+ * if you change them, you must change the defines in the Xserver.
+ */
+
+#ifndef _I810_DEFINES_
+#define _I810_DEFINES_
+
+#define I810_DMA_BUF_ORDER		12
+#define I810_DMA_BUF_SZ 		(1<<I810_DMA_BUF_ORDER)
+#define I810_DMA_BUF_NR 		256
+#define I810_NR_SAREA_CLIPRECTS 	8
+
+/* Each region is a minimum of 64k, and there are at most 64 of them.
+ */
+#define I810_NR_TEX_REGIONS 64
+#define I810_LOG_MIN_TEX_REGION_SIZE 16
+#endif
+
+#define I810_UPLOAD_TEX0IMAGE  0x1 /* handled clientside */
+#define I810_UPLOAD_TEX1IMAGE  0x2 /* handled clientside */
+#define I810_UPLOAD_CTX        0x4
+#define I810_UPLOAD_BUFFERS    0x8
+#define I810_UPLOAD_TEX0       0x10
+#define I810_UPLOAD_TEX1       0x20
+#define I810_UPLOAD_CLIPRECTS  0x40
+
+
+/* Indices into buf.Setup where various bits of state are mirrored per
+ * context and per buffer.  These can be fired at the card as a unit,
+ * or in a piecewise fashion as required.
+ */
+
+/* Destbuffer state 
+ *    - backbuffer linear offset and pitch -- invarient in the current dri
+ *    - zbuffer linear offset and pitch -- also invarient
+ *    - drawing origin in back and depth buffers.
+ *
+ * Keep the depth/back buffer state here to acommodate private buffers
+ * in the future.
+ */
+#define I810_DESTREG_DI0  0	/* CMD_OP_DESTBUFFER_INFO (2 dwords) */
+#define I810_DESTREG_DI1  1
+#define I810_DESTREG_DV0  2	/* GFX_OP_DESTBUFFER_VARS (2 dwords) */
+#define I810_DESTREG_DV1  3
+#define I810_DESTREG_DR0  4	/* GFX_OP_DRAWRECT_INFO (4 dwords) */
+#define I810_DESTREG_DR1  5
+#define I810_DESTREG_DR2  6
+#define I810_DESTREG_DR3  7
+#define I810_DESTREG_DR4  8
+#define I810_DEST_SETUP_SIZE 10
+
+/* Context state
+ */
+#define I810_CTXREG_CF0   0	/* GFX_OP_COLOR_FACTOR */
+#define I810_CTXREG_CF1   1	
+#define I810_CTXREG_ST0   2     /* GFX_OP_STIPPLE */
+#define I810_CTXREG_ST1   3
+#define I810_CTXREG_VF    4	/* GFX_OP_VERTEX_FMT */
+#define I810_CTXREG_MT    5	/* GFX_OP_MAP_TEXELS */
+#define I810_CTXREG_MC0   6	/* GFX_OP_MAP_COLOR_STAGES - stage 0 */
+#define I810_CTXREG_MC1   7     /* GFX_OP_MAP_COLOR_STAGES - stage 1 */
+#define I810_CTXREG_MC2   8	/* GFX_OP_MAP_COLOR_STAGES - stage 2 */
+#define I810_CTXREG_MA0   9	/* GFX_OP_MAP_ALPHA_STAGES - stage 0 */
+#define I810_CTXREG_MA1   10	/* GFX_OP_MAP_ALPHA_STAGES - stage 1 */
+#define I810_CTXREG_MA2   11	/* GFX_OP_MAP_ALPHA_STAGES - stage 2 */
+#define I810_CTXREG_SDM   12	/* GFX_OP_SRC_DEST_MONO */
+#define I810_CTXREG_FOG   13	/* GFX_OP_FOG_COLOR */
+#define I810_CTXREG_B1    14	/* GFX_OP_BOOL_1 */
+#define I810_CTXREG_B2    15	/* GFX_OP_BOOL_2 */
+#define I810_CTXREG_LCS   16	/* GFX_OP_LINEWIDTH_CULL_SHADE_MODE */
+#define I810_CTXREG_PV    17	/* GFX_OP_PV_RULE -- Invarient! */
+#define I810_CTXREG_ZA    18	/* GFX_OP_ZBIAS_ALPHAFUNC */
+#define I810_CTXREG_AA    19	/* GFX_OP_ANTIALIAS */
+#define I810_CTX_SETUP_SIZE 20 
+
+/* Texture state (per tex unit)
+ */
+#define I810_TEXREG_MI0  0	/* GFX_OP_MAP_INFO (4 dwords) */
+#define I810_TEXREG_MI1  1	
+#define I810_TEXREG_MI2  2	
+#define I810_TEXREG_MI3  3	
+#define I810_TEXREG_MF   4	/* GFX_OP_MAP_FILTER */
+#define I810_TEXREG_MLC  5	/* GFX_OP_MAP_LOD_CTL */
+#define I810_TEXREG_MLL  6	/* GFX_OP_MAP_LOD_LIMITS */
+#define I810_TEXREG_MCS  7	/* GFX_OP_MAP_COORD_SETS ??? */
+#define I810_TEX_SETUP_SIZE 8
+
+#define I810_FRONT   0x1
+#define I810_BACK    0x2
+#define I810_DEPTH   0x4
+
+
+typedef struct _drm_i810_init {
+	enum {
+		I810_INIT_DMA = 0x01,
+		I810_CLEANUP_DMA = 0x02
+	} func;
+	int ring_map_idx;
+	int buffer_map_idx;
+	int sarea_priv_offset;
+	unsigned int ring_start;
+	unsigned int ring_end;
+	unsigned int ring_size;
+	unsigned int front_offset;
+	unsigned int back_offset;
+	unsigned int depth_offset;
+	unsigned int w;
+	unsigned int h;
+	unsigned int pitch;
+	unsigned int pitch_bits; 
+} drm_i810_init_t;
+
+/* Warning: If you change the SAREA structure you must change the Xserver
+ * structure as well */
+
+typedef struct _drm_i810_tex_region {
+	unsigned char next, prev; /* indices to form a circular LRU  */
+	unsigned char in_use;	/* owned by a client, or free? */
+	int age;		/* tracked by clients to update local LRU's */
+} drm_i810_tex_region_t;
+
+typedef struct _drm_i810_sarea {
+   	unsigned int ContextState[I810_CTX_SETUP_SIZE];
+   	unsigned int BufferState[I810_DEST_SETUP_SIZE];
+   	unsigned int TexState[2][I810_TEX_SETUP_SIZE];
+   	unsigned int dirty;
+
+	unsigned int nbox;
+	drm_clip_rect_t boxes[I810_NR_SAREA_CLIPRECTS];
+
+	/* Maintain an LRU of contiguous regions of texture space.  If
+	 * you think you own a region of texture memory, and it has an
+	 * age different to the one you set, then you are mistaken and
+	 * it has been stolen by another client.  If global texAge
+	 * hasn't changed, there is no need to walk the list.
+	 *
+	 * These regions can be used as a proxy for the fine-grained
+	 * texture information of other clients - by maintaining them
+	 * in the same lru which is used to age their own textures,
+	 * clients have an approximate lru for the whole of global
+	 * texture space, and can make informed decisions as to which
+	 * areas to kick out.  There is no need to choose whether to
+	 * kick out your own texture or someone else's - simply eject
+	 * them all in LRU order.  
+	 */
+   
+	drm_i810_tex_region_t texList[I810_NR_TEX_REGIONS+1]; 
+				/* Last elt is sentinal */
+        int texAge;		/* last time texture was uploaded */
+        int last_enqueue;	/* last time a buffer was enqueued */
+	int last_dispatch;	/* age of the most recently dispatched buffer */
+	int last_quiescent;     /*  */
+	int ctxOwner;		/* last context to upload state */
+
+	int vertex_prim;
+
+} drm_i810_sarea_t;
+
+typedef struct _drm_i810_clear {
+	int clear_color;
+	int clear_depth;
+	int flags;
+} drm_i810_clear_t;
+
+
+
+/* These may be placeholders if we have more cliprects than
+ * I810_NR_SAREA_CLIPRECTS.  In that case, the client sets discard to
+ * false, indicating that the buffer will be dispatched again with a
+ * new set of cliprects.
+ */
+typedef struct _drm_i810_vertex {
+   	int idx;		/* buffer index */
+	int used;		/* nr bytes in use */
+	int discard;		/* client is finished with the buffer? */
+} drm_i810_vertex_t;
+
+typedef struct drm_i810_dma {
+	void *virtual;
+	int request_idx;
+	int request_size;
+	int granted;
+} drm_i810_dma_t;
+
+#endif /* _I810_DRM_H_ */
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_drv.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_drv.c
index eec9eb0f3..b523db902 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_drv.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_drv.c
@@ -79,7 +79,6 @@ static drm_ioctl_desc_t	      i810_ioctls[] = {
 	[DRM_IOCTL_NR(DRM_IOCTL_ADD_BUFS)]    = { i810_addbufs,	  1, 1 },
 	[DRM_IOCTL_NR(DRM_IOCTL_MARK_BUFS)]   = { i810_markbufs,  1, 1 },
 	[DRM_IOCTL_NR(DRM_IOCTL_INFO_BUFS)]   = { i810_infobufs,  1, 0 },
-	[DRM_IOCTL_NR(DRM_IOCTL_MAP_BUFS)]    = { i810_mapbufs,	  1, 0 },
 	[DRM_IOCTL_NR(DRM_IOCTL_FREE_BUFS)]   = { i810_freebufs,  1, 0 },
 
 	[DRM_IOCTL_NR(DRM_IOCTL_ADD_CTX)]     = { i810_addctx,	  1, 1 },
@@ -92,8 +91,6 @@ static drm_ioctl_desc_t	      i810_ioctls[] = {
 	[DRM_IOCTL_NR(DRM_IOCTL_ADD_DRAW)]    = { drm_adddraw,	  1, 1 },
 	[DRM_IOCTL_NR(DRM_IOCTL_RM_DRAW)]     = { drm_rmdraw,	  1, 1 },
 
-	[DRM_IOCTL_NR(DRM_IOCTL_DMA)]	      = { i810_dma,	  1, 0 },
-
 	[DRM_IOCTL_NR(DRM_IOCTL_LOCK)]	      = { i810_lock,	  1, 0 },
 	[DRM_IOCTL_NR(DRM_IOCTL_UNLOCK)]      = { i810_unlock,	  1, 0 },
 	[DRM_IOCTL_NR(DRM_IOCTL_FINISH)]      = { drm_finish,	  1, 0 },
@@ -106,11 +103,14 @@ static drm_ioctl_desc_t	      i810_ioctls[] = {
 	[DRM_IOCTL_NR(DRM_IOCTL_AGP_FREE)]    = { drm_agp_free,    1, 1 },
 	[DRM_IOCTL_NR(DRM_IOCTL_AGP_BIND)]    = { drm_agp_bind,    1, 1 },
 	[DRM_IOCTL_NR(DRM_IOCTL_AGP_UNBIND)]  = { drm_agp_unbind,  1, 1 },
+
    	[DRM_IOCTL_NR(DRM_IOCTL_I810_INIT)]   = { i810_dma_init,   1, 1 },
    	[DRM_IOCTL_NR(DRM_IOCTL_I810_VERTEX)] = { i810_dma_vertex, 1, 0 },
-   	[DRM_IOCTL_NR(DRM_IOCTL_I810_DMA)]    = { i810_dma_general,1, 0 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_I810_CLEAR)]  = { i810_clear_bufs, 1, 0 },
       	[DRM_IOCTL_NR(DRM_IOCTL_I810_FLUSH)]  = { i810_flush_ioctl,1, 0 },
    	[DRM_IOCTL_NR(DRM_IOCTL_I810_GETAGE)] = { i810_getage,     1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_I810_GETBUF)] = { i810_getbuf,     1, 0 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_I810_SWAP)]   = { i810_swap_bufs,  1, 0 },
 };
 
 #define I810_IOCTL_COUNT DRM_ARRAY_SIZE(i810_ioctls)
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_drv.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_drv.h
index 690710c83..c387bf727 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_drv.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_drv.h
@@ -32,6 +32,16 @@
 #ifndef _I810_DRV_H_
 #define _I810_DRV_H_
 
+typedef struct drm_i810_buf_priv {
+   	u32 *in_use;
+   	int my_use_idx;
+	int currently_mapped;
+	void *virtual;
+	void *kernel_virtual;
+	int map_count;
+   	struct vm_area_struct *vma;
+} drm_i810_buf_priv_t;
+
 typedef struct _drm_i810_ring_buffer{
 	int tail_mask;
 	unsigned long Start;
@@ -55,6 +65,15 @@ typedef struct drm_i810_private {
 
    	atomic_t flush_done;
    	wait_queue_head_t flush_queue;	/* Processes waiting until flush    */
+	drm_buf_t *mmap_buffer;
+
+	
+	u32 front_di1, back_di1, zi1;
+	
+	int back_offset;
+	int depth_offset;
+	int w, h;
+	int pitch;
 } drm_i810_private_t;
 
 				/* i810_drv.c */
@@ -71,8 +90,8 @@ extern int  i810_unlock(struct inode *inode, struct file *filp,
 
 				/* i810_dma.c */
 extern int  i810_dma_schedule(drm_device_t *dev, int locked);
-extern int  i810_dma(struct inode *inode, struct file *filp,
-		      unsigned int cmd, unsigned long arg);
+extern int  i810_getbuf(struct inode *inode, struct file *filp,
+			unsigned int cmd, unsigned long arg);
 extern int  i810_irq_install(drm_device_t *dev, int irq);
 extern int  i810_irq_uninstall(drm_device_t *dev);
 extern int  i810_control(struct inode *inode, struct file *filp,
@@ -86,6 +105,7 @@ extern int  i810_flush_ioctl(struct inode *inode, struct file *filp,
 extern void i810_reclaim_buffers(drm_device_t *dev, pid_t pid);
 extern int  i810_getage(struct inode *inode, struct file *filp, unsigned int cmd,
 			unsigned long arg);
+extern int i810_mmap_buffers(struct file *filp, struct vm_area_struct *vma);
 
 
 				/* i810_bufs.c */
@@ -97,8 +117,6 @@ extern int  i810_markbufs(struct inode *inode, struct file *filp,
 			 unsigned int cmd, unsigned long arg);
 extern int  i810_freebufs(struct inode *inode, struct file *filp,
 			 unsigned int cmd, unsigned long arg);
-extern int  i810_mapbufs(struct inode *inode, struct file *filp,
-			unsigned int cmd, unsigned long arg);
 extern int  i810_addmap(struct inode *inode, struct file *filp,
 		       unsigned int cmd, unsigned long arg);
 
@@ -121,32 +139,17 @@ extern int  i810_rmctx(struct inode *inode, struct file *filp,
 extern int  i810_context_switch(drm_device_t *dev, int old, int new);
 extern int  i810_context_switch_complete(drm_device_t *dev, int new);
 
-
-
-
-/* Copy the outstanding cliprects for every I810_DMA_VERTEX buffer.
- * This can be fixed by emitting directly to the ringbuffer in the
- * 'vertex_dma' ioctl.  
-*/
-typedef struct {
-   	u32 *in_use;
-   	int my_use_idx;
-} drm_i810_buf_priv_t;
-
-
-#define I810_DMA_GENERAL 0
-#define I810_DMA_VERTEX  1
-#define I810_DMA_DISCARD 2	/* not used */
-
 #define I810_VERBOSE 0
 
 
 int i810_dma_vertex(struct inode *inode, struct file *filp,
 		    unsigned int cmd, unsigned long arg);
 
-int i810_dma_general(struct inode *inode, struct file *filp,
-		     unsigned int cmd, unsigned long arg);
+int i810_swap_bufs(struct inode *inode, struct file *filp,
+		   unsigned int cmd, unsigned long arg);
 
+int i810_clear_bufs(struct inode *inode, struct file *filp,
+		    unsigned int cmd, unsigned long arg);
 
 #define GFX_OP_USER_INTERRUPT 		((0<<29)|(2<<23))
 #define GFX_OP_BREAKPOINT_INTERRUPT	((0<<29)|(1<<23))
@@ -200,5 +203,22 @@ int i810_dma_general(struct inode *inode, struct file *filp,
 #define SCI_YMAX_MASK      (0xffff<<16)
 #define SCI_XMAX_MASK      (0xffff<<0)
 
+#define GFX_OP_COLOR_FACTOR      ((0x3<<29)|(0x1d<<24)|(0x1<<16)|0x0)
+#define GFX_OP_STIPPLE           ((0x3<<29)|(0x1d<<24)|(0x83<<16))
+#define GFX_OP_MAP_INFO          ((0x3<<29)|(0x1d<<24)|0x2)
+#define GFX_OP_DESTBUFFER_VARS   ((0x3<<29)|(0x1d<<24)|(0x85<<16)|0x0)
+#define GFX_OP_DRAWRECT_INFO     ((0x3<<29)|(0x1d<<24)|(0x80<<16)|(0x3))
+#define GFX_OP_PRIMITIVE         ((0x3<<29)|(0x1f<<24))
+
+#define CMD_OP_Z_BUFFER_INFO     ((0x0<<29)|(0x16<<23))
+#define CMD_OP_DESTBUFFER_INFO   ((0x0<<29)|(0x15<<23))
+
+#define BR00_BITBLT_CLIENT   0x40000000
+#define BR00_OP_COLOR_BLT    0x10000000
+#define BR00_OP_SRC_COPY_BLT 0x10C00000
+#define BR13_SOLID_PATTERN   0x80000000
+
+
+
 #endif
 
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_dma.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_dma.c
index b03544bc2..25e3622ca 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_dma.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_dma.c
@@ -52,6 +52,7 @@ static unsigned long mga_alloc_page(drm_device_t *dev)
 {
 	unsigned long address;
    
+	DRM_DEBUG("%s\n", __FUNCTION__);
 	address = __get_free_page(GFP_KERNEL);
 	if(address == 0UL) {
 		return 0;
@@ -64,6 +65,8 @@ static unsigned long mga_alloc_page(drm_device_t *dev)
 
 static void mga_free_page(drm_device_t *dev, unsigned long page)
 {
+	DRM_DEBUG("%s\n", __FUNCTION__);
+
 	if(page == 0UL) {
 		return;
 	}
@@ -79,9 +82,11 @@ static void mga_delay(void)
    	return;
 }
 
-static void mga_flush_write_combine(void)
+void mga_flush_write_combine(void)
 {
    	int xchangeDummy;
+	DRM_DEBUG("%s\n", __FUNCTION__);
+
    	__asm__ volatile(" push %%eax ; xchg %%eax, %0 ; pop %%eax" : : "m" (xchangeDummy));
    	__asm__ volatile(" push %%eax ; push %%ebx ; push %%ecx ; push %%edx ;"
 			 " movl $0,%%eax ; cpuid ; pop %%edx ; pop %%ecx ; pop %%ebx ;"
@@ -93,18 +98,6 @@ static void mga_flush_write_combine(void)
 #define MGA_BUF_USED 	0xffffffff
 #define MGA_BUF_FREE	0
 
-static void mga_freelist_debug(drm_mga_freelist_t *item)
-{
-   	if(item->buf != NULL) {
-     		DRM_DEBUG("buf index : %d\n", item->buf->idx);
-	} else {
-	   	DRM_DEBUG("Freelist head\n");
-	}
-   	DRM_DEBUG("item->age : %x\n", item->age);
-   	DRM_DEBUG("item->next : %p\n", item->next);
-   	DRM_DEBUG("item->prev : %p\n", item->prev);
-}
-
 static int mga_freelist_init(drm_device_t *dev)
 {
       	drm_device_dma_t *dma = dev->dma;
@@ -113,7 +106,9 @@ static int mga_freelist_init(drm_device_t *dev)
       	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
    	drm_mga_freelist_t *item;
    	int i;
-   
+
+	DRM_DEBUG("%s\n", __FUNCTION__);
+
    	dev_priv->head = drm_alloc(sizeof(drm_mga_freelist_t), DRM_MEM_DRIVER);
 	if(dev_priv->head == NULL) return -ENOMEM;
    	memset(dev_priv->head, 0, sizeof(drm_mga_freelist_t));
@@ -135,19 +130,10 @@ static int mga_freelist_init(drm_device_t *dev)
 	   	item->buf = buf;
 	   	buf_priv->my_freelist = item;
 		buf_priv->discard = 0;
+		buf_priv->dispatched = 0;
 	   	dev_priv->head->next = item;
 	}
    
-   	item = dev_priv->head;
-   	while(item) {
-		mga_freelist_debug(item);
-		item = item->next;
-	}
-   	DRM_DEBUG("Head\n");
-   	mga_freelist_debug(dev_priv->head);
-   	DRM_DEBUG("Tail\n");
-   	mga_freelist_debug(dev_priv->tail);
-   
    	return 0;
 }
 
@@ -157,6 +143,8 @@ static void mga_freelist_cleanup(drm_device_t *dev)
    	drm_mga_freelist_t *item;
    	drm_mga_freelist_t *prev;
 
+	DRM_DEBUG("%s\n", __FUNCTION__);
+
    	item = dev_priv->head;
    	while(item) {
 	   	prev = item;
@@ -173,14 +161,14 @@ static inline void mga_dma_quiescent(drm_device_t *dev)
 	drm_device_dma_t  *dma      = dev->dma;
 	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
 	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
-	__volatile__ unsigned int *status =
-		(__volatile__ unsigned int *)dev_priv->status_page;
    	unsigned long end;
 	int i;
 
+	DRM_DEBUG("%s\n", __FUNCTION__);
 	end = jiffies + (HZ*3);
     	while(1) {
-		if(!test_and_set_bit(0, &dev_priv->dispatch_lock)) {
+		if(!test_and_set_bit(MGA_IN_DISPATCH, 
+				     &dev_priv->dispatch_status)) {
 			break;
 		}
 	   	if((signed)(end - jiffies) <= 0) {
@@ -204,51 +192,25 @@ static inline void mga_dma_quiescent(drm_device_t *dev)
 		}
 		for (i = 0 ; i < 2000 ; i++) mga_delay();	  
 	}
-    	DRM_DEBUG("status[1] : %x last_sync_tag : %x\n", status[1],
-		  dev_priv->last_sync_tag);
     	sarea_priv->dirty |= MGA_DMA_FLUSH;
 
 out_status:
-    	clear_bit(0, &dev_priv->dispatch_lock);
+    	clear_bit(MGA_IN_DISPATCH, &dev_priv->dispatch_status);
 out_nolock:
 }
 
-#define FREELIST_INITIAL (MGA_DMA_BUF_NR * 2)
-#define FREELIST_COMPARE(age) ((age >> 2))
-
-unsigned int mga_create_sync_tag(drm_device_t *dev)
+static void mga_reset_freelist(drm_device_t *dev)
 {
-      	drm_mga_private_t *dev_priv = 
-     		(drm_mga_private_t *) dev->dev_private;
-   	unsigned int temp;
-      	drm_buf_t *buf;
+   	drm_device_dma_t  *dma      = dev->dma;
+   	drm_buf_t *buf;
    	drm_mga_buf_priv_t *buf_priv;
-      	drm_device_dma_t *dma = dev->dma;
-   	int i;
-   
-   	dev_priv->sync_tag++;
-   
-   	if(dev_priv->sync_tag < FREELIST_INITIAL) {
-	   	dev_priv->sync_tag = FREELIST_INITIAL;
-	}
-   	if(dev_priv->sync_tag > 0x3fffffff) {
-		mga_flush_queue(dev);
-	   	mga_dma_quiescent(dev);
-	    
-	   	for (i = 0; i < dma->buf_count; i++) {
-			buf = dma->buflist[ i ];
-			buf_priv = buf->dev_private;
-			buf_priv->my_freelist->age = MGA_BUF_FREE;
-		}
-	   	
-	   	dev_priv->sync_tag = FREELIST_INITIAL;
-	}
-   	temp = dev_priv->sync_tag << 2;
-
-	dev_priv->sarea_priv->last_enqueue = temp;
+	int i;
 
-   	DRM_DEBUG("sync_tag : %x\n", temp);
-   	return temp;
+   	for (i = 0; i < dma->buf_count; i++) {
+	   	buf = dma->buflist[ i ];
+	        buf_priv = buf->dev_private;
+		buf_priv->my_freelist->age = MGA_BUF_FREE;
+	}
 }
 
 /* Least recently used :
@@ -257,23 +219,52 @@ unsigned int mga_create_sync_tag(drm_device_t *dev)
 
 drm_buf_t *mga_freelist_get(drm_device_t *dev)
 {
+   	DECLARE_WAITQUEUE(entry, current);
    	drm_mga_private_t *dev_priv = 
      		(drm_mga_private_t *) dev->dev_private;
-   	__volatile__ unsigned int *status = 
-     		(__volatile__ unsigned int *)dev_priv->status_page;
 	drm_mga_freelist_t *prev;
    	drm_mga_freelist_t *next;
+	static int failed = 0;
+
+	DRM_DEBUG("%s : tail->age : %d last_prim_age : %d\n", __FUNCTION__,
+	       dev_priv->tail->age, dev_priv->last_prim_age);
    
-   	if((dev_priv->tail->age >> 2) <= FREELIST_COMPARE(status[1])) {
+	if(failed >= 1000 && dev_priv->tail->age >= dev_priv->last_prim_age) {
+		DRM_DEBUG("I'm waiting on the freelist!!! %d\n", 
+		       dev_priv->last_prim_age);
+	   	set_bit(MGA_IN_GETBUF, &dev_priv->dispatch_status);
+		current->state = TASK_INTERRUPTIBLE;
+	   	add_wait_queue(&dev_priv->buf_queue, &entry);
+	   	for (;;) {
+		   	mga_dma_schedule(dev, 0);
+		   	if(!test_bit(MGA_IN_GETBUF, 
+				     &dev_priv->dispatch_status)) 
+				break;
+		   	atomic_inc(&dev->total_sleeps);
+		   	schedule();
+		   	if (signal_pending(current)) {
+				clear_bit(MGA_IN_GETBUF,
+					  &dev_priv->dispatch_status);
+			   	goto failed_getbuf;
+			}
+		}
+	   	current->state = TASK_RUNNING;
+	   	remove_wait_queue(&dev_priv->buf_queue, &entry);
+	}
+   
+   	if(dev_priv->tail->age < dev_priv->last_prim_age) {
 		prev = dev_priv->tail->prev;
 	   	next = dev_priv->tail;
 	   	prev->next = NULL;
 	   	next->prev = next->next = NULL;
 	   	dev_priv->tail = prev;
 	   	next->age = MGA_BUF_USED;
+		failed = 0;
 	   	return next->buf;
-	} 
+	}
 
+failed_getbuf:
+	failed++;
    	return NULL;
 }
 
@@ -286,6 +277,8 @@ int mga_freelist_put(drm_device_t *dev, drm_buf_t *buf)
    	drm_mga_freelist_t *head;
    	drm_mga_freelist_t *next;
 
+	DRM_DEBUG("%s\n", __FUNCTION__);
+
    	if(buf_priv->my_freelist->age == MGA_BUF_USED) {
 		/* Discarded buffer, put it on the tail */
 		next = buf_priv->my_freelist;
@@ -312,32 +305,6 @@ int mga_freelist_put(drm_device_t *dev, drm_buf_t *buf)
    	return 0;
 }
 
-static void mga_print_all_primary(drm_device_t *dev)
-{
-      	drm_mga_private_t *dev_priv = dev->dev_private;
-   	drm_mga_prim_buf_t *prim;
-   	int i;
-
-   	DRM_DEBUG("Full list of primarys\n");
-     	for(i = 0; i < MGA_NUM_PRIM_BUFS; i++) {
-		prim = dev_priv->prim_bufs[i];
-		DRM_DEBUG("index : %d num_dwords : %d "
-			  "max_dwords : %d phy_head : %x\n",
-			  prim->idx, prim->num_dwords, 
-			  prim->max_dwords, prim->phys_head);
-		DRM_DEBUG("sec_used : %d swap_pending : %x "
-			  "in_use : %x force_fire : %d\n",
-			  prim->sec_used, prim->swap_pending, 
-			  prim->in_use, atomic_read(&prim->force_fire));
-		DRM_DEBUG("needs_overflow : %d\n", 
-			  atomic_read(&prim->needs_overflow));
-	}
-   
-   	DRM_DEBUG("current_idx : %d, next_idx : %d, last_idx : %d\n",
-		  dev_priv->next_prim->idx, dev_priv->last_prim->idx,
-		  dev_priv->current_prim->idx);
-}
-
 static int mga_init_primary_bufs(drm_device_t *dev, drm_mga_init_t *init)
 {
    	drm_mga_private_t *dev_priv = dev->dev_private;
@@ -345,10 +312,9 @@ static int mga_init_primary_bufs(drm_device_t *dev, drm_mga_init_t *init)
    	int i, temp, size_of_buf;
    	int offset = init->reserved_map_agpstart;
 
-   	DRM_DEBUG("mga_init_primary_bufs\n");
+	DRM_DEBUG("%s\n", __FUNCTION__);
    	dev_priv->primary_size = ((init->primary_size + PAGE_SIZE - 1) / 
 				  PAGE_SIZE) * PAGE_SIZE;
-   	DRM_DEBUG("primary_size\n");
    	size_of_buf = dev_priv->primary_size / MGA_NUM_PRIM_BUFS;
 	dev_priv->warp_ucode_size = init->warp_ucode_size;
    	dev_priv->prim_bufs = drm_alloc(sizeof(drm_mga_prim_buf_t *) * 
@@ -358,18 +324,12 @@ static int mga_init_primary_bufs(drm_device_t *dev, drm_mga_init_t *init)
 		DRM_ERROR("Unable to allocate memory for prim_buf\n");
 		return -ENOMEM;
 	}
-   	DRM_DEBUG("memset\n");
    	memset(dev_priv->prim_bufs, 
 	       0, sizeof(drm_mga_prim_buf_t *) * (MGA_NUM_PRIM_BUFS + 1));
    
    	temp = init->warp_ucode_size + dev_priv->primary_size;
 	temp = ((temp + PAGE_SIZE - 1) / PAGE_SIZE) * PAGE_SIZE;
 	   
-   	DRM_DEBUG("temp : %x\n", temp);
-   	DRM_DEBUG("dev->agp->base: %lx\n", dev->agp->base);
-   	DRM_DEBUG("init->reserved_map_agpstart: %x\n", 
-		  init->reserved_map_agpstart);
-   	DRM_DEBUG("ioremap\n");
 	dev_priv->ioremap = drm_ioremap(dev->agp->base + offset, 
 					temp);
 	if(dev_priv->ioremap == NULL) {
@@ -379,11 +339,9 @@ static int mga_init_primary_bufs(drm_device_t *dev, drm_mga_init_t *init)
    	init_waitqueue_head(&dev_priv->wait_queue);
    
    	for(i = 0; i < MGA_NUM_PRIM_BUFS; i++) {
-	   	DRM_DEBUG("For loop\n");
 	   	prim_buffer = drm_alloc(sizeof(drm_mga_prim_buf_t), 
 					DRM_MEM_DRIVER);
 	   	if(prim_buffer == NULL) return -ENOMEM;
-	   	DRM_DEBUG("memset\n");
 	   	memset(prim_buffer, 0, sizeof(drm_mga_prim_buf_t));
 	   	prim_buffer->phys_head = offset + dev->agp->base;
 	   	prim_buffer->current_dma_ptr = 
@@ -396,17 +354,18 @@ static int mga_init_primary_bufs(drm_device_t *dev, drm_mga_init_t *init)
 	   	prim_buffer->max_dwords -= 5; /* Leave room for the softrap */
 	   	prim_buffer->sec_used = 0;
 	   	prim_buffer->idx = i;
+		prim_buffer->prim_age = i + 1;
 	   	offset = offset + size_of_buf;
 	   	dev_priv->prim_bufs[i] = prim_buffer;
-	   	DRM_DEBUG("Looping\n");
 	}
 	dev_priv->current_prim_idx = 0;
         dev_priv->next_prim = 
 		dev_priv->last_prim = 
 		dev_priv->current_prim =
         	dev_priv->prim_bufs[0];
-   	set_bit(0, &dev_priv->current_prim->in_use);
-	DRM_DEBUG("init done\n");
+	dev_priv->next_prim_age = 2;	
+	dev_priv->last_prim_age = 1;
+   	set_bit(MGA_BUF_IN_USE, &dev_priv->current_prim->buffer_status);
    	return 0;
 }
 
@@ -420,9 +379,8 @@ void mga_fire_primary(drm_device_t *dev, drm_mga_prim_buf_t *prim)
    	int i;
    	int next_idx;
        	PRIMLOCALS;
-   
-    	DRM_DEBUG("mga_fire_primary\n");
-    	dev_priv->last_sync_tag = mga_create_sync_tag(dev); 
+
+   	DRM_DEBUG("%s\n", __FUNCTION__);
    	dev_priv->last_prim = prim;
    
  	/* We never check for overflow, b/c there is always room */
@@ -433,7 +391,7 @@ void mga_fire_primary(drm_device_t *dev, drm_mga_prim_buf_t *prim)
 	}
  	PRIMOUTREG( MGAREG_DMAPAD, 0);
  	PRIMOUTREG( MGAREG_DMAPAD, 0);
-       	PRIMOUTREG( MGAREG_DWGSYNC, dev_priv->last_sync_tag);
+       	PRIMOUTREG( MGAREG_DMAPAD, 0);
    	PRIMOUTREG( MGAREG_SOFTRAP, 0);
     	PRIMFINISH(prim);
 
@@ -468,13 +426,13 @@ void mga_fire_primary(drm_device_t *dev, drm_mga_prim_buf_t *prim)
 			for (i = 0 ; i < 4096 ; i++) mga_delay();
 		}
 	}
-	
+
    	mga_flush_write_combine();
     	atomic_inc(&dev_priv->pending_bufs);
-	atomic_inc(&dma->total_lost);
        	MGA_WRITE(MGAREG_PRIMADDRESS, phys_head | TT_GENERAL);
  	MGA_WRITE(MGAREG_PRIMEND, (phys_head + num_dwords * 4) | use_agp);
    	prim->num_dwords = 0;
+	sarea_priv->last_enqueue = prim->prim_age;
     
    	next_idx = prim->idx + 1;
     	if(next_idx >= MGA_NUM_PRIM_BUFS) 
@@ -486,11 +444,10 @@ void mga_fire_primary(drm_device_t *dev, drm_mga_prim_buf_t *prim)
  out_prim_wait:
 	prim->num_dwords = 0;
 	prim->sec_used = 0;
-	clear_bit(0, &prim->in_use);
+	clear_bit(MGA_BUF_IN_USE, &prim->buffer_status);
    	wake_up_interruptible(&dev_priv->wait_queue);
-	clear_bit(0, &prim->swap_pending);
-	clear_bit(0, &dev_priv->dispatch_lock);
-	atomic_dec(&dev_priv->pending_bufs);
+	clear_bit(MGA_BUF_SWAP_PENDING, &prim->buffer_status);
+	clear_bit(MGA_IN_DISPATCH, &dev_priv->dispatch_status);
 }
 
 int mga_advance_primary(drm_device_t *dev)
@@ -505,27 +462,28 @@ int mga_advance_primary(drm_device_t *dev)
    	/* This needs to reset the primary buffer if available,
 	 * we should collect stats on how many times it bites
 	 * it's tail */
+	DRM_DEBUG("%s\n", __FUNCTION__);
    
    	next_prim_idx = dev_priv->current_prim_idx + 1;
    	if(next_prim_idx >= MGA_NUM_PRIM_BUFS)
      		next_prim_idx = 0;
    	prim_buffer = dev_priv->prim_bufs[next_prim_idx];
-   	atomic_set(&dev_priv->in_wait, 1);
+	set_bit(MGA_IN_WAIT, &dev_priv->dispatch_status);
    
       	/* In use is cleared in interrupt handler */
    
-   	if(test_and_set_bit(0, &prim_buffer->in_use)) {
+   	if(test_and_set_bit(MGA_BUF_IN_USE, &prim_buffer->buffer_status)) {
 	   	add_wait_queue(&dev_priv->wait_queue, &entry);
+		current->state = TASK_INTERRUPTIBLE;
+
 	   	for (;;) {
-		   	current->state = TASK_INTERRUPTIBLE;
 		   	mga_dma_schedule(dev, 0);
-		   	if(!test_and_set_bit(0, &prim_buffer->in_use)) break;
+		   	if(!test_and_set_bit(MGA_BUF_IN_USE, 
+					     &prim_buffer->buffer_status)) 
+				break;
 		   	atomic_inc(&dev->total_sleeps);
 		   	atomic_inc(&dma->total_missed_sched);
-		   	mga_print_all_primary(dev);
-		   	DRM_DEBUG("Schedule in advance\n");
-		   	/* Three second delay */
-		   	schedule_timeout(HZ*3);
+		   	schedule();
 		   	if (signal_pending(current)) {
 			   	ret = -ERESTARTSYS;
 			   	break;
@@ -535,16 +493,27 @@ int mga_advance_primary(drm_device_t *dev)
 	   	remove_wait_queue(&dev_priv->wait_queue, &entry);
 	   	if(ret) return ret;
 	}
-   	atomic_set(&dev_priv->in_wait, 0);
+	clear_bit(MGA_IN_WAIT, &dev_priv->dispatch_status);
+
    	/* This primary buffer is now free to use */
    	prim_buffer->current_dma_ptr = prim_buffer->head;
    	prim_buffer->num_dwords = 0;
    	prim_buffer->sec_used = 0;
-   	atomic_set(&prim_buffer->needs_overflow, 0);
+	prim_buffer->prim_age = dev_priv->next_prim_age++;
+	if(prim_buffer->prim_age == 0 || prim_buffer->prim_age == 0xffffffff) {
+	   mga_flush_queue(dev);
+	   mga_dma_quiescent(dev);
+	   mga_reset_freelist(dev);
+	   prim_buffer->prim_age = (dev_priv->next_prim_age += 2);
+	}
+
+	/* Reset all buffer status stuff */
+	clear_bit(MGA_BUF_NEEDS_OVERFLOW, &prim_buffer->buffer_status);
+	clear_bit(MGA_BUF_FORCE_FIRE, &prim_buffer->buffer_status);
+	clear_bit(MGA_BUF_SWAP_PENDING, &prim_buffer->buffer_status);
+
    	dev_priv->current_prim = prim_buffer;
    	dev_priv->current_prim_idx = next_prim_idx;
-   	DRM_DEBUG("Primarys at advance\n");
-   	mga_print_all_primary(dev);
    	return 0;
 }
 
@@ -553,21 +522,29 @@ static inline int mga_decide_to_fire(drm_device_t *dev)
 {
    	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
       	drm_device_dma_t  *dma	    = dev->dma;
-	
-   	if(atomic_read(&dev_priv->next_prim->force_fire))
-	{
+
+   	DRM_DEBUG("%s\n", __FUNCTION__);
+
+   	if(test_bit(MGA_BUF_FORCE_FIRE, &dev_priv->next_prim->buffer_status)) {
 	   	atomic_inc(&dma->total_prio);
 	   	return 1;
 	}
 
-	if (atomic_read(&dev_priv->in_flush) && dev_priv->next_prim->num_dwords)
-	{
+	if (test_bit(MGA_IN_GETBUF, &dev_priv->dispatch_status) &&
+	    dev_priv->next_prim->num_dwords) {
+	   	atomic_inc(&dma->total_prio);
+	   	return 1;
+	}
+
+	if (test_bit(MGA_IN_FLUSH, &dev_priv->dispatch_status) &&
+	    dev_priv->next_prim->num_dwords) {
 	   	atomic_inc(&dma->total_prio);
 	   	return 1;
 	}
    
    	if(atomic_read(&dev_priv->pending_bufs) <= MGA_NUM_PRIM_BUFS - 1) {
-		if(test_bit(0, &dev_priv->next_prim->swap_pending)) {
+		if(test_bit(MGA_BUF_SWAP_PENDING, 
+			    &dev_priv->next_prim->buffer_status)) {
 			atomic_inc(&dma->total_dmas);
 			return 1;
 		}
@@ -601,10 +578,11 @@ int mga_dma_schedule(drm_device_t *dev, int locked)
 		return -EBUSY;
 	}
    
-       	DRM_DEBUG("mga_dma_schedule\n");
+	DRM_DEBUG("%s\n", __FUNCTION__);
 
-   	if(atomic_read(&dev_priv->in_flush) || 
-	   atomic_read(&dev_priv->in_wait)) {
+   	if(test_bit(MGA_IN_FLUSH, &dev_priv->dispatch_status) || 
+	   test_bit(MGA_IN_WAIT, &dev_priv->dispatch_status) ||
+	   test_bit(MGA_IN_GETBUF, &dev_priv->dispatch_status)) {
 		locked = 1;
 	}
    
@@ -612,27 +590,25 @@ int mga_dma_schedule(drm_device_t *dev, int locked)
 	    !drm_lock_take(&dev->lock.hw_lock->lock, DRM_KERNEL_CONTEXT)) {
 	   	atomic_inc(&dma->total_missed_lock);
 	   	clear_bit(0, &dev->dma_flag);
+		DRM_DEBUG("Not locked\n");
 	   	return -EBUSY;
 	}
    	DRM_DEBUG("I'm locked\n");
 
-   
-   	if(!test_and_set_bit(0, &dev_priv->dispatch_lock)) {
+   	if(!test_and_set_bit(MGA_IN_DISPATCH, &dev_priv->dispatch_status)) {
 	   	/* Fire dma buffer */
 	   	if(mga_decide_to_fire(dev)) {
-		   	DRM_DEBUG("mga_fire_primary\n");
 		   	DRM_DEBUG("idx :%d\n", dev_priv->next_prim->idx);
-		   	atomic_set(&dev_priv->next_prim->force_fire, 0);
-		   	if(dev_priv->current_prim == dev_priv->next_prim &&
-			   dev_priv->next_prim->num_dwords != 0) {
+			clear_bit(MGA_BUF_FORCE_FIRE, 
+				  &dev_priv->next_prim->buffer_status);
+		   	if(dev_priv->current_prim == dev_priv->next_prim) {
 				/* Schedule overflow for a later time */
-				atomic_set(
-					&dev_priv->current_prim->needs_overflow,
-					1);
+				set_bit(MGA_BUF_NEEDS_OVERFLOW,
+					&dev_priv->next_prim->buffer_status);
 			}
 		   	mga_fire_primary(dev, dev_priv->next_prim);
 		} else {
-			clear_bit(0, &dev_priv->dispatch_lock);
+			clear_bit(MGA_IN_DISPATCH, &dev_priv->dispatch_status);
 		}
 	} else {
 		DRM_DEBUG("I can't get the dispatch lock\n");
@@ -645,17 +621,26 @@ int mga_dma_schedule(drm_device_t *dev, int locked)
 		}
 	}
 
-   	clear_bit(0, &dev->dma_flag);
-   
-      	if(atomic_read(&dev_priv->in_flush) == 1 &&
-	   dev_priv->next_prim->num_dwords == 0) {
-	   	/* Everything is on the hardware */
-	   	DRM_DEBUG("Primarys at Flush\n");
-	   	mga_print_all_primary(dev);
-	   	atomic_set(&dev_priv->in_flush, 0);
+      	if(test_bit(MGA_IN_FLUSH, &dev_priv->dispatch_status) &&
+	   dev_priv->next_prim->num_dwords == 0 &&
+	   atomic_read(&dev_priv->pending_bufs) == 0) {
+	   	/* Everything has been processed by the hardware */
+		clear_bit(MGA_IN_FLUSH, &dev_priv->dispatch_status);
 	   	wake_up_interruptible(&dev_priv->flush_queue);
 	}
 
+	if(test_bit(MGA_IN_GETBUF, &dev_priv->dispatch_status) &&
+	   dev_priv->tail->age < dev_priv->last_prim_age) {
+		clear_bit(MGA_IN_GETBUF, &dev_priv->dispatch_status);
+		DRM_DEBUG("Waking up buf queue\n");
+		wake_up_interruptible(&dev_priv->buf_queue);
+	} else if (test_bit(MGA_IN_GETBUF, &dev_priv->dispatch_status)) {
+	   	DRM_DEBUG("Not waking buf_queue on %d %d\n", 
+			  atomic_read(&dev->total_irq), 
+			  dev_priv->last_prim_age);
+	}
+
+   	clear_bit(0, &dev->dma_flag);
 	return 0;
 }
 
@@ -664,33 +649,35 @@ static void mga_dma_service(int irq, void *device, struct pt_regs *regs)
     	drm_device_t	 *dev = (drm_device_t *)device;
     	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
     	drm_mga_prim_buf_t *last_prim_buffer;
-    	__volatile__ unsigned int *status = 
-      		(__volatile__ unsigned int *)dev_priv->status_page;
-    
+
+	DRM_DEBUG("%s\n", __FUNCTION__);
     	atomic_inc(&dev->total_irq);
+	if((MGA_READ(MGAREG_STATUS) & 0x00000001) != 0x00000001) return;
       	MGA_WRITE(MGAREG_ICLEAR, 0x00000001);
    	last_prim_buffer = dev_priv->last_prim;
     	last_prim_buffer->num_dwords = 0;
     	last_prim_buffer->sec_used = 0;
-      	clear_bit(0, &last_prim_buffer->in_use);
+	dev_priv->sarea_priv->last_dispatch = 
+		dev_priv->last_prim_age = last_prim_buffer->prim_age;
+      	clear_bit(MGA_BUF_IN_USE, &last_prim_buffer->buffer_status);
    	wake_up_interruptible(&dev_priv->wait_queue);
-      	clear_bit(0, &last_prim_buffer->swap_pending);
-      	clear_bit(0, &dev_priv->dispatch_lock);
+      	clear_bit(MGA_BUF_SWAP_PENDING, &last_prim_buffer->buffer_status);
+      	clear_bit(MGA_IN_DISPATCH, &dev_priv->dispatch_status);
       	atomic_dec(&dev_priv->pending_bufs);
-   	dev_priv->sarea_priv->last_dispatch = status[1];
    	queue_task(&dev->tq, &tq_immediate);
    	mark_bh(IMMEDIATE_BH);
 }
 
 static void mga_dma_task_queue(void *device)
 {
-   	drm_device_t *dev = (drm_device_t *) device;
-
-	mga_dma_schedule(dev, 0);
+	DRM_DEBUG("%s\n", __FUNCTION__);
+	mga_dma_schedule((drm_device_t *)device, 0);
 }
 
 int mga_dma_cleanup(drm_device_t *dev)
 {
+	DRM_DEBUG("%s\n", __FUNCTION__);
+
 	if(dev->dev_private) {
 		drm_mga_private_t *dev_priv = 
 			(drm_mga_private_t *) dev->dev_private;
@@ -739,14 +726,13 @@ static int mga_dma_initialize(drm_device_t *dev, drm_mga_init_t *init) {
 	drm_map_t *sarea_map = NULL;
 	int i;
 
+	DRM_DEBUG("%s\n", __FUNCTION__);
+
 	dev_priv = drm_alloc(sizeof(drm_mga_private_t), DRM_MEM_DRIVER);
 	if(dev_priv == NULL) return -ENOMEM;
 	dev->dev_private = (void *) dev_priv;
 
-	DRM_DEBUG("dev_private\n");
-
 	memset(dev_priv, 0, sizeof(drm_mga_private_t));
-      	atomic_set(&dev_priv->in_flush, 0);
 
 	if((init->reserved_map_idx >= dev->map_count) ||
 	   (init->buffer_map_idx >= dev->map_count)) {
@@ -761,7 +747,6 @@ static int mga_dma_initialize(drm_device_t *dev, drm_mga_init_t *init) {
 	dev_priv->sarea_priv = (drm_mga_sarea_t *) 
 		((u8 *)sarea_map->handle + 
 		 init->sarea_priv_offset);
-	DRM_DEBUG("sarea_priv\n");
 
 	/* Scale primary size to the next page */
 	dev_priv->chipset = init->chipset;
@@ -776,6 +761,7 @@ static int mga_dma_initialize(drm_device_t *dev, drm_mga_init_t *init) {
 
 	dev_priv->mAccess = init->mAccess;
    	init_waitqueue_head(&dev_priv->flush_queue);
+	init_waitqueue_head(&dev_priv->buf_queue);
 	dev_priv->WarpPipe = -1;
 
    	DRM_DEBUG("chipset: %d ucode_size: %d backOffset: %x depthOffset: %x\n",
@@ -795,20 +781,17 @@ static int mga_dma_initialize(drm_device_t *dev, drm_mga_init_t *init) {
 			  dev_priv->WarpIndex[i].phys_addr,
 			  dev_priv->WarpIndex[i].size);
 
-   	DRM_DEBUG("Doing init prim buffers\n");
    	if(mga_init_primary_bufs(dev, init) != 0) {
 		DRM_ERROR("Can not initialize primary buffers\n");
 		mga_dma_cleanup(dev);
 		return -ENOMEM;
 	}
-   	DRM_DEBUG("Done with init prim buffers\n");
    	dev_priv->real_status_page = mga_alloc_page(dev);
       	if(dev_priv->real_status_page == 0UL) {
 		mga_dma_cleanup(dev);
 		DRM_ERROR("Can not allocate status page\n");
 		return -ENOMEM;
 	}
-   	DRM_DEBUG("Status page at %lx\n", dev_priv->real_status_page);
 
    	dev_priv->status_page = 
 		ioremap_nocache(virt_to_bus((void *)dev_priv->real_status_page),
@@ -820,38 +803,23 @@ static int mga_dma_initialize(drm_device_t *dev, drm_mga_init_t *init) {
 		return -ENOMEM;
 	}
 
-   	DRM_DEBUG("Status page remapped to %p\n", dev_priv->status_page);
    	/* Write status page when secend or softrap occurs */
    	MGA_WRITE(MGAREG_PRIMPTR, 
 		  virt_to_bus((void *)dev_priv->real_status_page) | 0x00000003);
-   
-   	dev_priv->device = pci_find_device(0x102b, 0x0525, NULL);
-   	if(dev_priv->device == NULL) {
-		DRM_ERROR("Could not find pci device for card\n");
-		mga_dma_cleanup(dev);
-		return -EINVAL;
-	}
-   
-   	DRM_DEBUG("dma initialization\n");
+      
 
 	/* Private is now filled in, initialize the hardware */
 	{
-	   	__volatile__ unsigned int *status = 
-			(unsigned int *)dev_priv->status_page;
 		PRIMLOCALS;
 		PRIMGETPTR( dev_priv );
-	   
-	   	dev_priv->last_sync_tag = mga_create_sync_tag(dev);
-	   
+	   	   
 		PRIMOUTREG(MGAREG_DMAPAD, 0);
 		PRIMOUTREG(MGAREG_DMAPAD, 0);
-		PRIMOUTREG(MGAREG_DWGSYNC, dev_priv->last_sync_tag);
+		PRIMOUTREG(MGAREG_DWGSYNC, 0x0100);
 		PRIMOUTREG(MGAREG_SOFTRAP, 0);
 		/* Poll for the first buffer to insure that
 		 * the status register will be correct
 		 */
-	   	DRM_DEBUG("phys_head : %lx\n", (unsigned long)phys_head);
-   		status[1] = 0;
 	   
 		mga_flush_write_combine();
 	   	MGA_WRITE(MGAREG_PRIMADDRESS, phys_head | TT_GENERAL);
@@ -859,9 +827,7 @@ static int mga_dma_initialize(drm_device_t *dev, drm_mga_init_t *init) {
 		MGA_WRITE(MGAREG_PRIMEND, ((phys_head + num_dwords * 4) | 
 					   PDEA_pagpxfer_enable));
 	   
-	   	while(MGA_READ(MGAREG_DWGSYNC) != dev_priv->last_sync_tag) ;
-	   	DRM_DEBUG("status[0] after initialization : %x\n", status[0]);
-	   	DRM_DEBUG("status[1] after initialization : %x\n", status[1]);
+	   	while(MGA_READ(MGAREG_DWGSYNC) != 0x0100) ;
 	}
 
 	if(mga_freelist_init(dev) != 0) {
@@ -869,7 +835,6 @@ static int mga_dma_initialize(drm_device_t *dev, drm_mga_init_t *init) {
 	   	mga_dma_cleanup(dev);
 	   	return -ENOMEM;
 	}
-   	DRM_DEBUG("dma init was successful\n");
 	return 0;
 }
 
@@ -880,6 +845,8 @@ int mga_dma_init(struct inode *inode, struct file *filp,
 	drm_device_t *dev = priv->dev;
 	drm_mga_init_t init;
    
+   	DRM_DEBUG("%s\n", __FUNCTION__);
+
 	copy_from_user_ret(&init, (drm_mga_init_t *)arg, sizeof(init), -EFAULT);
    
 	switch(init.func) {
@@ -924,7 +891,7 @@ int mga_irq_install(drm_device_t *dev, int irq)
    				/* Install handler */
 	if ((retcode = request_irq(dev->irq,
 				   mga_dma_service,
-				   0,
+				   SA_SHIRQ,
 				   dev->devname,
 				   dev))) {
 		down(&dev->struct_sem);
@@ -963,7 +930,9 @@ int mga_control(struct inode *inode, struct file *filp, unsigned int cmd,
 	drm_control_t	ctl;
    
 	copy_from_user_ret(&ctl, (drm_control_t *)arg, sizeof(ctl), -EFAULT);
-	
+
+   	DRM_DEBUG("%s\n", __FUNCTION__);
+
 	switch (ctl.func) {
 	case DRM_INST_HANDLER:
 		return mga_irq_install(dev, ctl.irq);
@@ -980,30 +949,33 @@ static int mga_flush_queue(drm_device_t *dev)
   	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
    	int ret = 0;
 
+   	DRM_DEBUG("%s\n", __FUNCTION__);
+
    	if(dev_priv == NULL) {
 	   	return 0;
 	}
    
    	if(dev_priv->next_prim->num_dwords != 0) {
-	   	atomic_set(&dev_priv->in_flush, 1);
+		set_bit(MGA_IN_FLUSH, &dev_priv->dispatch_status);
    		current->state = TASK_INTERRUPTIBLE;
    		add_wait_queue(&dev_priv->flush_queue, &entry);
    		for (;;) {
 		   	mga_dma_schedule(dev, 0);
-	   		if (atomic_read(&dev_priv->in_flush) == 0) 
+	   		if (!test_bit(MGA_IN_FLUSH, 
+				      &dev_priv->dispatch_status)) 
 				break;
 		   	atomic_inc(&dev->total_sleeps);
-		   	DRM_DEBUG("Schedule in flush_queue\n");
-	      		schedule_timeout(HZ*3);
+	      		schedule();
 	      		if (signal_pending(current)) {
 		   		ret = -EINTR; /* Can't restart */
+				clear_bit(MGA_IN_FLUSH, 
+					  &dev_priv->dispatch_status);
 		   		break;
 			}
 		}
    		current->state = TASK_RUNNING;
    		remove_wait_queue(&dev_priv->flush_queue, &entry);
 	}
-   	atomic_set(&dev_priv->in_flush, 0);
    	return ret;
 }
 
@@ -1017,6 +989,7 @@ void mga_reclaim_buffers(drm_device_t *dev, pid_t pid)
       	if(dev->dev_private == NULL) return;
 	if(dma->buflist == NULL) return;
 
+	DRM_DEBUG("%s\n", __FUNCTION__);
         mga_flush_queue(dev);
 
 	for (i = 0; i < dma->buf_count; i++) {
@@ -1042,6 +1015,7 @@ int mga_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 	int		  ret	= 0;
 	drm_lock_t	  lock;
 
+	DRM_DEBUG("%s\n", __FUNCTION__);
 	copy_from_user_ret(&lock, (drm_lock_t *)arg, sizeof(lock), -EFAULT);
 
 	if (lock.context == DRM_KERNEL_CONTEXT) {
@@ -1080,7 +1054,6 @@ int mga_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 				/* Contention */
 			atomic_inc(&dev->total_sleeps);
 			current->state = TASK_INTERRUPTIBLE;
-		   	DRM_DEBUG("Calling lock schedule\n");
 			schedule();
 			if (signal_pending(current)) {
 				ret = -ERESTARTSYS;
@@ -1110,11 +1083,8 @@ int mga_flush_ioctl(struct inode *inode, struct file *filp,
     	drm_device_t	  *dev	  = priv->dev;
 	drm_lock_t	  lock;
       	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
-   	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
-      	__volatile__ unsigned int *status = 
-     		(__volatile__ unsigned int *)dev_priv->status_page;
-	int		 i;
- 
+
+   	DRM_DEBUG("%s\n", __FUNCTION__);
 	copy_from_user_ret(&lock, (drm_lock_t *)arg, sizeof(lock), -EFAULT);
 
 	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
@@ -1123,28 +1093,17 @@ int mga_flush_ioctl(struct inode *inode, struct file *filp,
 	}
 
    	if(lock.flags & _DRM_LOCK_FLUSH || lock.flags & _DRM_LOCK_FLUSH_ALL) {
-		mga_flush_queue(dev);
-
-		if((MGA_READ(MGAREG_STATUS) & 0x00030001) == 0x00020000 &&
-		   status[1] != dev_priv->last_sync_tag) 
-		{
-			DRM_DEBUG("Reseting hardware status\n");
-			MGA_WRITE(MGAREG_DWGSYNC, dev_priv->last_sync_tag);
-
-			while(MGA_READ(MGAREG_DWGSYNC) != 
-			      dev_priv->last_sync_tag) 
-			{
-				for(i = 0; i < 4096; i++) mga_delay();
-			}
-
-			status[1] = 
-				sarea_priv->last_dispatch = 
-				dev_priv->last_sync_tag;
-		} else {
-			sarea_priv->last_dispatch = status[1];
-		}
+		drm_mga_prim_buf_t *temp_buf =
+			dev_priv->prim_bufs[dev_priv->current_prim_idx];
+
+		if(temp_buf && temp_buf->num_dwords) {
+			set_bit(MGA_BUF_FORCE_FIRE, &temp_buf->buffer_status);
+			mga_advance_primary(dev);
+			mga_dma_schedule(dev, 1);
+ 		}
 	}
    	if(lock.flags & _DRM_LOCK_QUIESCENT) {
+		mga_flush_queue(dev);
 		mga_dma_quiescent(dev);
 	}
 
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drm.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drm.h
index 53d3590b1..e75e91a4f 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drm.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drm.h
@@ -37,6 +37,7 @@
  */
 #ifndef _MGA_DEFINES_
 #define _MGA_DEFINES_
+
 #define MGA_F  0x1		/* fog */
 #define MGA_A  0x2		/* alpha */
 #define MGA_S  0x4		/* specular */
@@ -61,11 +62,11 @@
 
 #define MGA_MAX_G400_PIPES 16
 #define MGA_MAX_G200_PIPES  8	/* no multitex */
-
 #define MGA_MAX_WARP_PIPES MGA_MAX_G400_PIPES
 
 #define MGA_CARD_TYPE_G200 1
 #define MGA_CARD_TYPE_G400 2
+
 #define MGA_FRONT   0x1
 #define MGA_BACK    0x2
 #define MGA_DEPTH   0x4
@@ -110,19 +111,19 @@
 #define MGA_UPLOAD_TEX0       0x2
 #define MGA_UPLOAD_TEX1       0x4
 #define MGA_UPLOAD_PIPE       0x8
-#define MGA_UPLOAD_TEX0IMAGE  0x10
-#define MGA_UPLOAD_TEX1IMAGE  0x20
+#define MGA_UPLOAD_TEX0IMAGE  0x10 /* handled client-side */
+#define MGA_UPLOAD_TEX1IMAGE  0x20 /* handled client-side */
 #define MGA_UPLOAD_2D 	      0x40
 #define MGA_WAIT_AGE          0x80 /* handled client-side */
 #define MGA_UPLOAD_CLIPRECTS  0x100 /* handled client-side */
 #define MGA_DMA_FLUSH	      0x200 /* set when someone gets the lock
                                        quiescent */
 
-/* 64 buffers of 16k each, total 1 meg.
+/* 32 buffers of 64k each, total 2 meg.
  */
-#define MGA_DMA_BUF_ORDER     14
+#define MGA_DMA_BUF_ORDER     16
 #define MGA_DMA_BUF_SZ        (1<<MGA_DMA_BUF_ORDER)
-#define MGA_DMA_BUF_NR        63
+#define MGA_DMA_BUF_NR        31
 
 /* Keep these small for testing.
  */
@@ -159,19 +160,19 @@ typedef struct drm_mga_init {
    	int sarea_priv_offset;
    	int primary_size;
    	int warp_ucode_size;
-   	int frontOffset;
-   	int backOffset;
-   	int depthOffset;
-   	int textureOffset;
-   	int textureSize;
-        int agpTextureOffset;
-        int agpTextureSize;
-   	int cpp;
-   	int stride;
+   	unsigned int frontOffset;
+   	unsigned int backOffset;
+   	unsigned int depthOffset;
+   	unsigned int textureOffset;
+   	unsigned int textureSize;
+        unsigned int agpTextureOffset;
+        unsigned int agpTextureSize;
+   	unsigned int cpp;
+   	unsigned int stride;
    	int sgram;
 	int chipset;
    	drm_mga_warp_index_t WarpIndex[MGA_MAX_WARP_PIPES];
-	int mAccess;
+	unsigned int mAccess;
 } drm_mga_init_t;
 
 /* Warning: if you change the sarea structure, you must change the Xserver
@@ -180,7 +181,7 @@ typedef struct drm_mga_init {
 typedef struct _drm_mga_tex_region {
 	unsigned char next, prev;	
 	unsigned char in_use;	
-	int age;			
+	unsigned int age;			
 } drm_mga_tex_region_t;
 
 typedef struct _drm_mga_sarea {
@@ -219,9 +220,9 @@ typedef struct _drm_mga_sarea {
    
 	/* Counters for aging textures and for client-side throttling.
 	 */
-        int last_enqueue;	/* last time a buffer was enqueued */
-	int last_dispatch;	/* age of the most recently dispatched buffer */
-	int last_quiescent;     /*  */
+        unsigned int last_enqueue;	/* last time a buffer was enqueued */
+	unsigned int last_dispatch;	/* age of the most recently dispatched buffer */
+	unsigned int last_quiescent;     /*  */
 
 
 	/* LRU lists for texture memory in agp space and on the card
@@ -237,9 +238,9 @@ typedef struct _drm_mga_sarea {
 /* Device specific ioctls:
  */
 typedef struct _drm_mga_clear {
-	int clear_color;
-	int clear_depth;
-	int flags;
+	unsigned int clear_color;
+	unsigned int clear_depth;
+	unsigned int flags;
 } drm_mga_clear_t;
 
 typedef struct _drm_mga_swap {
@@ -258,4 +259,11 @@ typedef struct _drm_mga_vertex {
 	int discard;		/* client finished with buffer?  */
 } drm_mga_vertex_t;
 
+typedef struct _drm_mga_indices {
+   	int idx;		/* buffer to queue */
+	unsigned int start;		
+	unsigned int end;		
+	int discard;		/* client finished with buffer?  */
+} drm_mga_indices_t;
+
 #endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drv.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drv.c
index b066fe9d9..4b2c835fa 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drv.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drv.c
@@ -111,6 +111,7 @@ static drm_ioctl_desc_t	      mga_ioctls[] = {
    	[DRM_IOCTL_NR(DRM_IOCTL_MGA_ILOAD)]   = { mga_iload,       1, 0 },
    	[DRM_IOCTL_NR(DRM_IOCTL_MGA_VERTEX)]  = { mga_vertex,      1, 0 },
    	[DRM_IOCTL_NR(DRM_IOCTL_MGA_FLUSH)]   = { mga_flush_ioctl, 1, 0 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_MGA_INDICES)] = { mga_indices,     1, 0 },
 };
 
 #define MGA_IOCTL_COUNT DRM_ARRAY_SIZE(mga_ioctls)
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drv.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drv.h
index fe6e4ef53..f217acb97 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drv.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drv.h
@@ -32,18 +32,21 @@
 #ifndef _MGA_DRV_H_
 #define _MGA_DRV_H_
 
+#define MGA_BUF_IN_USE         0
+#define MGA_BUF_SWAP_PENDING   1
+#define MGA_BUF_FORCE_FIRE     2
+#define MGA_BUF_NEEDS_OVERFLOW 3
+
 typedef struct {
+	u32 buffer_status;
    	unsigned int num_dwords;
    	unsigned int max_dwords;
    	u32 *current_dma_ptr;
    	u32 *head;
    	u32 phys_head;
+	unsigned int prim_age;
    	int sec_used;
    	int idx;
-   	int swap_pending;
-   	u32 in_use;
-   	atomic_t force_fire;
-   	atomic_t needs_overflow;
 } drm_mga_prim_buf_t;
 
 typedef struct _drm_mga_freelist {
@@ -53,31 +56,33 @@ typedef struct _drm_mga_freelist {
    	struct _drm_mga_freelist *prev;
 } drm_mga_freelist_t;
 
+#define MGA_IN_DISPATCH   0
+#define MGA_IN_FLUSH      1
+#define MGA_IN_WAIT       2
+#define MGA_IN_GETBUF	  3
+
 typedef struct _drm_mga_private {
+   	u32 dispatch_status;
+	unsigned int next_prim_age;
+	__volatile__ unsigned int last_prim_age;
    	int reserved_map_idx;
    	int buffer_map_idx;
    	drm_mga_sarea_t *sarea_priv;
    	int primary_size;
    	int warp_ucode_size;
    	int chipset;
-   	int frontOffset;
-   	int backOffset;
-   	int depthOffset;
-   	int textureOffset;
-   	int textureSize;
+   	unsigned int frontOffset;
+   	unsigned int backOffset;
+   	unsigned int depthOffset;
+   	unsigned int textureOffset;
+   	unsigned int textureSize;
    	int cpp;
-   	int stride;
+   	unsigned int stride;
    	int sgram;
 	int use_agp;
    	drm_mga_warp_index_t WarpIndex[MGA_MAX_G400_PIPES];
 	unsigned int WarpPipe;
-   	__volatile__ unsigned long softrap_age;
-   	u32 dispatch_lock;
-   	atomic_t in_flush;
-   	atomic_t in_wait;
    	atomic_t pending_bufs;
-   	unsigned int last_sync_tag;
-   	unsigned int sync_tag;
    	void *status_page;
    	unsigned long real_status_page;
    	u8 *ioremap;
@@ -86,12 +91,11 @@ typedef struct _drm_mga_private {
    	drm_mga_prim_buf_t *last_prim;
    	drm_mga_prim_buf_t *current_prim;
    	int current_prim_idx;
-   	struct pci_dev *device;
    	drm_mga_freelist_t *head;
    	drm_mga_freelist_t *tail;
    	wait_queue_head_t flush_queue;	/* Processes waiting until flush    */
       	wait_queue_head_t wait_queue;	/* Processes waiting until interrupt */
-
+	wait_queue_head_t buf_queue;    /* Processes waiting for a free buf */
 	/* Some validated register values:
 	 */	
 	u32 mAccess;
@@ -126,7 +130,7 @@ extern int mga_dma_init(struct inode *inode, struct file *filp,
 extern int mga_dma_cleanup(drm_device_t *dev);
 extern int mga_flush_ioctl(struct inode *inode, struct file *filp,
 			   unsigned int cmd, unsigned long arg);
-
+extern void mga_flush_write_combine(void);
 extern unsigned int mga_create_sync_tag(drm_device_t *dev);
 extern drm_buf_t *mga_freelist_get(drm_device_t *dev);
 extern int mga_freelist_put(drm_device_t *dev, drm_buf_t *buf);
@@ -156,6 +160,8 @@ extern int  mga_iload(struct inode *inode, struct file *filp,
 		      unsigned int cmd, unsigned long arg);
 extern int  mga_vertex(struct inode *inode, struct file *filp,
 		      unsigned int cmd, unsigned long arg);
+extern int  mga_indices(struct inode *inode, struct file *filp,
+			unsigned int cmd, unsigned long arg);
 				/* mga_context.c */
 extern int  mga_resctx(struct inode *inode, struct file *filp,
 		       unsigned int cmd, unsigned long arg);
@@ -186,6 +192,7 @@ typedef enum {
 typedef struct {
    	drm_mga_freelist_t *my_freelist;
 	int discard;
+	int dispatched;
 } drm_mga_buf_priv_t;
 
 #define DWGREG0 	0x1c00
@@ -206,15 +213,16 @@ typedef struct {
 
 #define PRIM_OVERFLOW(dev, dev_priv, length) do {			\
 	drm_mga_prim_buf_t *tmp_buf =					\
-		dev_priv->prim_bufs[dev_priv->current_prim_idx];	\
-	if( tmp_buf->max_dwords - tmp_buf->num_dwords < length ||	\
-	    tmp_buf->sec_used > MGA_DMA_BUF_NR/2) {			\
-		atomic_set(&tmp_buf->force_fire, 1);			\
-		mga_advance_primary(dev);				\
-		mga_dma_schedule(dev, 1);				\
-	} else if( atomic_read(&tmp_buf->needs_overflow)) {		\
-		mga_advance_primary(dev);				\
-		mga_dma_schedule(dev, 1);				\
+ 		dev_priv->prim_bufs[dev_priv->current_prim_idx];	\
+	if( test_bit(MGA_BUF_NEEDS_OVERFLOW,				\
+		  &tmp_buf->buffer_status)) {				\
+ 		mga_advance_primary(dev);				\
+ 		mga_dma_schedule(dev, 1);				\
+ 	} else if( tmp_buf->max_dwords - tmp_buf->num_dwords < length ||\
+ 	    tmp_buf->sec_used > MGA_DMA_BUF_NR/2) {			\
+		set_bit(MGA_BUF_FORCE_FIRE, &tmp_buf->buffer_status);	\
+ 		mga_advance_primary(dev);				\
+ 		mga_dma_schedule(dev, 1);				\
 	}								\
 } while(0)
 
@@ -266,6 +274,13 @@ drm_mga_prim_buf_t *tmp_buf = 					\
 	tmp_buf->sec_used++;						\
 } while (0)
 
+#define AGEBUF(dev_priv, buf_priv)	do {				\
+	drm_mga_prim_buf_t *tmp_buf =					\
+		dev_priv->prim_bufs[dev_priv->current_prim_idx];	\
+	buf_priv->my_freelist->age = tmp_buf->prim_age;			\
+} while (0)
+
+
 #define PRIMOUTREG(reg, val) do {					\
 	tempIndex[outcount]=ADRINDEX(reg);				\
 	dma_ptr[1+outcount] = val;					\
@@ -356,28 +371,61 @@ drm_mga_prim_buf_t *tmp_buf = 					\
 #define MGAREG_YTOP 				0x1c98
 #define MGAREG_ZORG 				0x1c0c
 
-#define DC_atype_rstr				0x10
-#define DC_atype_blk				0x40
 #define PDEA_pagpxfer_enable			0x2
+
 #define WIA_wmode_suspend			0x0
 #define WIA_wmode_start 			0x3
 #define WIA_wagp_agp				0x4
-#define DC_opcod_trap				0x4
-#define DC_arzero_enable			0x1000
-#define DC_sgnzero_enable			0x2000
-#define DC_shftzero_enable			0x4000
-#define DC_bop_SHIFT				16
-#define DC_clipdis_enable			0x80000000
-#define DC_solid_enable				0x800
-#define DC_transc_enable			0x40000000
+
+#define DC_opcod_line_open 			0x0
+#define DC_opcod_autoline_open 			0x1
+#define DC_opcod_line_close 			0x2
+#define DC_opcod_autoline_close 		0x3
+#define DC_opcod_trap 				0x4
+#define DC_opcod_texture_trap 			0x6
 #define DC_opcod_bitblt 			0x8
-#define DC_atype_rpl				0x0
-#define DC_linear_xy				0x0
-#define DC_solid_disable			0x0
-#define DC_arzero_disable			0x0
-#define DC_bltmod_bfcol				0x4000000
-#define DC_pattern_disable			0x0
-#define DC_transc_disable			0x0
+#define DC_opcod_iload 				0x9
+#define DC_atype_rpl 				0x0
+#define DC_atype_rstr 				0x10
+#define DC_atype_zi 				0x30
+#define DC_atype_blk 				0x40
+#define DC_atype_i 				0x70
+#define DC_linear_xy 				0x0 		
+#define DC_linear_linear 			0x80 		
+#define DC_zmode_nozcmp 			0x0
+#define DC_zmode_ze 				0x200
+#define DC_zmode_zne 				0x300
+#define DC_zmode_zlt 				0x400
+#define DC_zmode_zlte 				0x500
+#define DC_zmode_zgt 				0x600
+#define DC_zmode_zgte 				0x700
+#define DC_solid_disable 			0x0 		
+#define DC_solid_enable 			0x800 		
+#define DC_arzero_disable 			0x0 		
+#define DC_arzero_enable 			0x1000 		
+#define DC_sgnzero_disable 			0x0 		
+#define DC_sgnzero_enable 			0x2000 		
+#define DC_shftzero_disable 			0x0 		
+#define DC_shftzero_enable 			0x4000 		
+#define DC_bop_SHIFT 				16 		
+#define DC_trans_SHIFT 				20 		
+#define DC_bltmod_bmonolef 			0x0
+#define DC_bltmod_bmonowf 			0x8000000
+#define DC_bltmod_bplan 			0x2000000
+#define DC_bltmod_bfcol 			0x4000000
+#define DC_bltmod_bu32bgr 			0x6000000
+#define DC_bltmod_bu32rgb 			0xe000000
+#define DC_bltmod_bu24bgr 			0x16000000
+#define DC_bltmod_bu24rgb 			0x1e000000
+#define DC_pattern_disable 			0x0 		
+#define DC_pattern_enable 			0x20000000 	
+#define DC_transc_disable 			0x0 		
+#define DC_transc_enable 			0x40000000 	
+#define DC_clipdis_disable 			0x0 		
+#define DC_clipdis_enable 			0x80000000 	
+
+#define SETADD_mode_vertlist                   	0x0 	
+
 
 #define MGA_CLEAR_CMD (DC_opcod_trap | DC_arzero_enable | 		\
 		       DC_sgnzero_enable | DC_shftzero_enable | 	\
@@ -392,4 +440,8 @@ drm_mga_prim_buf_t *tmp_buf = 					\
 		      DC_pattern_disable | DC_transc_disable | 		\
 		      DC_clipdis_enable)				\
 
+#define MGA_FLUSH_CMD (DC_opcod_texture_trap | (0xF << DC_trans_SHIFT) |\
+		       DC_arzero_enable | DC_sgnzero_enable |		\
+		       DC_atype_i)
+
 #endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_state.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_state.c
index 0a50aa4f5..3134b7853 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_state.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_state.c
@@ -40,15 +40,23 @@ static void mgaEmitClipRect( drm_mga_private_t *dev_priv,
    	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
 	unsigned int *regs = sarea_priv->ContextState;
 	PRIMLOCALS;
+	DRM_DEBUG("%s\n", __FUNCTION__);
 
    	/* This takes 10 dwords */
 	PRIMGETPTR( dev_priv );
    
    	/* Force reset of dwgctl (eliminates clip disable) */
+#if 1
    	PRIMOUTREG( MGAREG_DMAPAD, 0 );
-	PRIMOUTREG( MGAREG_DWGSYNC, dev_priv->last_sync_tag - 1 );
-	PRIMOUTREG( MGAREG_DWGSYNC, dev_priv->last_sync_tag - 1 );
+	PRIMOUTREG( MGAREG_DWGSYNC, 0 );
+	PRIMOUTREG( MGAREG_DWGSYNC, 0 );
    	PRIMOUTREG( MGAREG_DWGCTL, regs[MGA_CTXREG_DWGCTL] );
+#else
+   	PRIMOUTREG( MGAREG_DWGCTL, regs[MGA_CTXREG_DWGCTL] );
+	PRIMOUTREG( MGAREG_LEN + MGAREG_MGA_EXEC, 0x80000000 );
+   	PRIMOUTREG( MGAREG_DWGCTL, regs[MGA_CTXREG_DWGCTL] );
+	PRIMOUTREG( MGAREG_LEN + MGAREG_MGA_EXEC, 0x80000000 );
+#endif
 
    	PRIMOUTREG( MGAREG_DMAPAD, 0 );
 	PRIMOUTREG( MGAREG_CXBNDRY, ((box->x2)<<16)|(box->x1) );
@@ -63,6 +71,7 @@ static void mgaEmitContext(drm_mga_private_t *dev_priv )
       	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
 	unsigned int *regs = sarea_priv->ContextState;
 	PRIMLOCALS;
+	DRM_DEBUG("%s\n", __FUNCTION__);
 	
    	/* This takes a max of 15 dwords */
 	PRIMGETPTR( dev_priv );
@@ -80,13 +89,13 @@ static void mgaEmitContext(drm_mga_private_t *dev_priv )
  	if (dev_priv->chipset == MGA_CARD_TYPE_G400) { 
 		PRIMOUTREG( MGAREG_WFLAG1, regs[MGA_CTXREG_WFLAG] );
 		PRIMOUTREG( MGAREG_TDUALSTAGE0, regs[MGA_CTXREG_TDUAL0] );
-		PRIMOUTREG( MGAREG_TDUALSTAGE1, regs[MGA_CTXREG_TDUAL1] );      
+		PRIMOUTREG( MGAREG_TDUALSTAGE1, regs[MGA_CTXREG_TDUAL1] );
 		PRIMOUTREG( MGAREG_FCOL, regs[MGA_CTXREG_FCOL] );
 	} else {
 	   	PRIMOUTREG( MGAREG_FCOL, regs[MGA_CTXREG_FCOL] );
-	   	PRIMOUTREG( MGAREG_DMAPAD, 0);
-	   	PRIMOUTREG( MGAREG_DMAPAD, 0);
-	   	PRIMOUTREG( MGAREG_DMAPAD, 0);
+	   	PRIMOUTREG( MGAREG_DMAPAD, 0 );
+	   	PRIMOUTREG( MGAREG_DMAPAD, 0 );
+	   	PRIMOUTREG( MGAREG_DMAPAD, 0 );
 	}
    
 	PRIMADVANCE( dev_priv );
@@ -97,27 +106,28 @@ static void mgaG200EmitTex( drm_mga_private_t *dev_priv )
       	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
 	unsigned int *regs = sarea_priv->TexState[0];
 	PRIMLOCALS;
+	DRM_DEBUG("%s\n", __FUNCTION__);
 
 	PRIMGETPTR( dev_priv );
    
    	/* This takes 20 dwords */
 
-	PRIMOUTREG(MGAREG_TEXCTL2, regs[MGA_TEXREG_CTL2] );
-	PRIMOUTREG(MGAREG_TEXCTL, regs[MGA_TEXREG_CTL] );
-	PRIMOUTREG(MGAREG_TEXFILTER, regs[MGA_TEXREG_FILTER] );
-	PRIMOUTREG(MGAREG_TEXBORDERCOL, regs[MGA_TEXREG_BORDERCOL] );
+	PRIMOUTREG( MGAREG_TEXCTL2, regs[MGA_TEXREG_CTL2] );
+	PRIMOUTREG( MGAREG_TEXCTL, regs[MGA_TEXREG_CTL] );
+	PRIMOUTREG( MGAREG_TEXFILTER, regs[MGA_TEXREG_FILTER] );
+	PRIMOUTREG( MGAREG_TEXBORDERCOL, regs[MGA_TEXREG_BORDERCOL] );
 
-	PRIMOUTREG(MGAREG_TEXORG, regs[MGA_TEXREG_ORG] );
-	PRIMOUTREG(MGAREG_TEXORG1, regs[MGA_TEXREG_ORG1] );
-	PRIMOUTREG(MGAREG_TEXORG2, regs[MGA_TEXREG_ORG2] );
-	PRIMOUTREG(MGAREG_TEXORG3, regs[MGA_TEXREG_ORG3] );
+	PRIMOUTREG( MGAREG_TEXORG, regs[MGA_TEXREG_ORG] );
+	PRIMOUTREG( MGAREG_TEXORG1, regs[MGA_TEXREG_ORG1] );
+	PRIMOUTREG( MGAREG_TEXORG2, regs[MGA_TEXREG_ORG2] );
+	PRIMOUTREG( MGAREG_TEXORG3, regs[MGA_TEXREG_ORG3] );
 
-	PRIMOUTREG(MGAREG_TEXORG4, regs[MGA_TEXREG_ORG4] );		
-	PRIMOUTREG(MGAREG_TEXWIDTH, regs[MGA_TEXREG_WIDTH] );
-	PRIMOUTREG(MGAREG_TEXHEIGHT, regs[MGA_TEXREG_HEIGHT] );   
-	PRIMOUTREG(0x2d00 + 24*4, regs[MGA_TEXREG_WIDTH] );
+	PRIMOUTREG( MGAREG_TEXORG4, regs[MGA_TEXREG_ORG4] );		
+	PRIMOUTREG( MGAREG_TEXWIDTH, regs[MGA_TEXREG_WIDTH] );
+	PRIMOUTREG( MGAREG_TEXHEIGHT, regs[MGA_TEXREG_HEIGHT] );   
+	PRIMOUTREG( 0x2d00 + 24*4, regs[MGA_TEXREG_WIDTH] );
 
-	PRIMOUTREG(0x2d00 + 34*4, regs[MGA_TEXREG_HEIGHT] );
+	PRIMOUTREG( 0x2d00 + 34*4, regs[MGA_TEXREG_HEIGHT] );
 	PRIMOUTREG( MGAREG_TEXTRANS, 0xffff );
 	PRIMOUTREG( MGAREG_TEXTRANSHIGH, 0xffff );
 	PRIMOUTREG( MGAREG_DMAPAD, 0 );
@@ -131,34 +141,35 @@ static void mgaG400EmitTex0( drm_mga_private_t *dev_priv )
 	unsigned int *regs = sarea_priv->TexState[0];
 	int multitex = sarea_priv->WarpPipe & MGA_T2;
 	PRIMLOCALS;
+	DRM_DEBUG("%s\n", __FUNCTION__);
    
 	PRIMGETPTR( dev_priv );
    
    	/* This takes a max of 30 dwords */
 
-	PRIMOUTREG(MGAREG_TEXCTL2, regs[MGA_TEXREG_CTL2] );
-	PRIMOUTREG(MGAREG_TEXCTL, regs[MGA_TEXREG_CTL] );
-	PRIMOUTREG(MGAREG_TEXFILTER, regs[MGA_TEXREG_FILTER] );
-	PRIMOUTREG(MGAREG_TEXBORDERCOL, regs[MGA_TEXREG_BORDERCOL] );
+	PRIMOUTREG( MGAREG_TEXCTL2, regs[MGA_TEXREG_CTL2] );
+	PRIMOUTREG( MGAREG_TEXCTL, regs[MGA_TEXREG_CTL] );
+	PRIMOUTREG( MGAREG_TEXFILTER, regs[MGA_TEXREG_FILTER] );
+	PRIMOUTREG( MGAREG_TEXBORDERCOL, regs[MGA_TEXREG_BORDERCOL] );
 
-	PRIMOUTREG(MGAREG_TEXORG, regs[MGA_TEXREG_ORG] );
-	PRIMOUTREG(MGAREG_TEXORG1, regs[MGA_TEXREG_ORG1] );
-	PRIMOUTREG(MGAREG_TEXORG2, regs[MGA_TEXREG_ORG2] );
-	PRIMOUTREG(MGAREG_TEXORG3, regs[MGA_TEXREG_ORG3] );
+	PRIMOUTREG( MGAREG_TEXORG, regs[MGA_TEXREG_ORG] );
+	PRIMOUTREG( MGAREG_TEXORG1, regs[MGA_TEXREG_ORG1] );
+	PRIMOUTREG( MGAREG_TEXORG2, regs[MGA_TEXREG_ORG2] );
+	PRIMOUTREG( MGAREG_TEXORG3, regs[MGA_TEXREG_ORG3] );
 
-	PRIMOUTREG(MGAREG_TEXORG4, regs[MGA_TEXREG_ORG4] );		
-	PRIMOUTREG(MGAREG_TEXWIDTH, regs[MGA_TEXREG_WIDTH] );
-	PRIMOUTREG(MGAREG_TEXHEIGHT, regs[MGA_TEXREG_HEIGHT] );   
-	PRIMOUTREG(0x2d00 + 49*4, 0);
+	PRIMOUTREG( MGAREG_TEXORG4, regs[MGA_TEXREG_ORG4] );		
+	PRIMOUTREG( MGAREG_TEXWIDTH, regs[MGA_TEXREG_WIDTH] );
+	PRIMOUTREG( MGAREG_TEXHEIGHT, regs[MGA_TEXREG_HEIGHT] );   
+	PRIMOUTREG( 0x2d00 + 49*4, 0 );
 
-	PRIMOUTREG(0x2d00 + 57*4, 0);
-	PRIMOUTREG(0x2d00 + 53*4, 0);
-	PRIMOUTREG(0x2d00 + 61*4, 0);
+	PRIMOUTREG( 0x2d00 + 57*4, 0 );
+	PRIMOUTREG( 0x2d00 + 53*4, 0 );
+	PRIMOUTREG( 0x2d00 + 61*4, 0 );
 	PRIMOUTREG( MGAREG_DMAPAD, 0 );
 
 	if (!multitex) {
-		PRIMOUTREG(0x2d00 + 52*4, 0x40 ); 
-		PRIMOUTREG(0x2d00 + 60*4, 0x40 );
+		PRIMOUTREG( 0x2d00 + 52*4, 0x40 ); 
+		PRIMOUTREG( 0x2d00 + 60*4, 0x40 );
 		PRIMOUTREG( MGAREG_DMAPAD, 0 );
 		PRIMOUTREG( MGAREG_DMAPAD, 0 );
 	} 
@@ -178,85 +189,127 @@ static void mgaG400EmitTex1( drm_mga_private_t *dev_priv )
       	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
 	unsigned int *regs = sarea_priv->TexState[1];
 	PRIMLOCALS;
+	DRM_DEBUG("%s\n", __FUNCTION__);
    
-	PRIMGETPTR(dev_priv);
+	PRIMGETPTR( dev_priv );
 
    	/* This takes 25 dwords */
-	PRIMOUTREG(MGAREG_TEXCTL2, regs[MGA_TEXREG_CTL2] | TMC_map1_enable);
-	PRIMOUTREG(MGAREG_TEXCTL, regs[MGA_TEXREG_CTL] );
-	PRIMOUTREG(MGAREG_TEXFILTER, regs[MGA_TEXREG_FILTER] );
-	PRIMOUTREG(MGAREG_TEXBORDERCOL, regs[MGA_TEXREG_BORDERCOL] );
-
-	PRIMOUTREG(MGAREG_TEXORG, regs[MGA_TEXREG_ORG] );
-	PRIMOUTREG(MGAREG_TEXORG1, regs[MGA_TEXREG_ORG1] );
-	PRIMOUTREG(MGAREG_TEXORG2, regs[MGA_TEXREG_ORG2] );
-	PRIMOUTREG(MGAREG_TEXORG3, regs[MGA_TEXREG_ORG3] );
-
-	PRIMOUTREG(MGAREG_TEXORG4, regs[MGA_TEXREG_ORG4] );		
-	PRIMOUTREG(MGAREG_TEXWIDTH, regs[MGA_TEXREG_WIDTH] );
-	PRIMOUTREG(MGAREG_TEXHEIGHT, regs[MGA_TEXREG_HEIGHT] );   
-	PRIMOUTREG(0x2d00 + 49*4, 0);
-
-	PRIMOUTREG(0x2d00 + 57*4, 0);
-	PRIMOUTREG(0x2d00 + 53*4, 0);
-	PRIMOUTREG(0x2d00 + 61*4, 0);
-	PRIMOUTREG(0x2d00 + 52*4, regs[MGA_TEXREG_WIDTH] | 0x40 ); 
-
-	PRIMOUTREG(0x2d00 + 60*4, regs[MGA_TEXREG_HEIGHT] | 0x40 );
+
+	PRIMOUTREG( MGAREG_TEXCTL2, regs[MGA_TEXREG_CTL2] | TMC_map1_enable );
+	PRIMOUTREG( MGAREG_TEXCTL, regs[MGA_TEXREG_CTL] );
+	PRIMOUTREG( MGAREG_TEXFILTER, regs[MGA_TEXREG_FILTER] );
+	PRIMOUTREG( MGAREG_TEXBORDERCOL, regs[MGA_TEXREG_BORDERCOL] );
+
+	PRIMOUTREG( MGAREG_TEXORG, regs[MGA_TEXREG_ORG] );
+	PRIMOUTREG( MGAREG_TEXORG1, regs[MGA_TEXREG_ORG1] );
+	PRIMOUTREG( MGAREG_TEXORG2, regs[MGA_TEXREG_ORG2] );
+	PRIMOUTREG( MGAREG_TEXORG3, regs[MGA_TEXREG_ORG3] );
+
+	PRIMOUTREG( MGAREG_TEXORG4, regs[MGA_TEXREG_ORG4] );		
+	PRIMOUTREG( MGAREG_TEXWIDTH, regs[MGA_TEXREG_WIDTH] );
+	PRIMOUTREG( MGAREG_TEXHEIGHT, regs[MGA_TEXREG_HEIGHT] );   
+	PRIMOUTREG( 0x2d00 + 49*4, 0 );
+
+	PRIMOUTREG( 0x2d00 + 57*4, 0 );
+	PRIMOUTREG( 0x2d00 + 53*4, 0 );
+	PRIMOUTREG( 0x2d00 + 61*4, 0 );
+	PRIMOUTREG( 0x2d00 + 52*4, regs[MGA_TEXREG_WIDTH] | 0x40 ); 
+
+	PRIMOUTREG( 0x2d00 + 60*4, regs[MGA_TEXREG_HEIGHT] | 0x40 );
 	PRIMOUTREG( MGAREG_TEXTRANS, 0xffff );
 	PRIMOUTREG( MGAREG_TEXTRANSHIGH, 0xffff );
-	PRIMOUTREG(MGAREG_TEXCTL2, regs[MGA_TEXREG_CTL2] );
+	PRIMOUTREG( MGAREG_TEXCTL2, regs[MGA_TEXREG_CTL2] );
+
+	PRIMADVANCE( dev_priv );
+}
+
+/* Required when switching from multitexturing to single texturing.
+ */
+static void mgaG400EmitTexFlush( drm_mga_private_t *dev_priv )
+{
+	PRIMLOCALS;
+	DRM_DEBUG("%s\n", __FUNCTION__);
+   
+	PRIMGETPTR( dev_priv );
+
+   	/* This takes 15 dwords */
+
+	PRIMOUTREG( MGAREG_YDST, 0 );
+	PRIMOUTREG( MGAREG_FXLEFT, 0 );
+	PRIMOUTREG( MGAREG_FXRIGHT, 1 );
+	PRIMOUTREG( MGAREG_DWGCTL, MGA_FLUSH_CMD );
+
+	PRIMOUTREG( MGAREG_LEN + MGAREG_MGA_EXEC, 1 );
+	PRIMOUTREG( MGAREG_DMAPAD, 0 );
+	PRIMOUTREG( MGAREG_DWGSYNC, 0x7000 );
+	PRIMOUTREG( MGAREG_DMAPAD, 0 );
+
+	PRIMOUTREG( MGAREG_TEXCTL2, 0 );
+	PRIMOUTREG( MGAREG_LEN + MGAREG_MGA_EXEC, 0 );
+	PRIMOUTREG( MGAREG_TEXCTL2, 0x80 );
+	PRIMOUTREG( MGAREG_LEN + MGAREG_MGA_EXEC, 0 );
 
 	PRIMADVANCE( dev_priv );
 }
 
-static void mgaG400EmitPipe(drm_mga_private_t *dev_priv )
+static void mgaG400EmitPipe( drm_mga_private_t *dev_priv )
 {
       	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
 	unsigned int pipe = sarea_priv->WarpPipe;
 	float fParam = 12800.0f;
 	PRIMLOCALS;
+	DRM_DEBUG("%s\n", __FUNCTION__);
    
-	PRIMGETPTR(dev_priv);
+	PRIMGETPTR( dev_priv );
+
+   	/* This takes 30 dwords */
 
-   	/* This takes 25 dwords */
-   
 	/* Establish vertex size.  
 	 */
 	if (pipe & MGA_T2) {
-		PRIMOUTREG(MGAREG_WIADDR2, WIA_wmode_suspend);
-		PRIMOUTREG(MGAREG_WVRTXSZ, 0x00001e09);
-		PRIMOUTREG(MGAREG_WACCEPTSEQ, 0x1e000000);
-		PRIMOUTREG(MGAREG_WFLAG, 0);
+		PRIMOUTREG( MGAREG_WIADDR2, WIA_wmode_suspend );
+		PRIMOUTREG( MGAREG_WVRTXSZ, 0x00001e09 );
+		PRIMOUTREG( MGAREG_DMAPAD, 0 );
+		PRIMOUTREG( MGAREG_DMAPAD, 0 );
+
+		PRIMOUTREG( MGAREG_WACCEPTSEQ, 0 );
+		PRIMOUTREG( MGAREG_WACCEPTSEQ, 0 );
+		PRIMOUTREG( MGAREG_WACCEPTSEQ, 0 );
+		PRIMOUTREG( MGAREG_WACCEPTSEQ, 0x1e000000 );
 	} else {
-		PRIMOUTREG(MGAREG_WIADDR2, WIA_wmode_suspend);
-		PRIMOUTREG(MGAREG_WVRTXSZ, 0x00001807);
-		PRIMOUTREG(MGAREG_WACCEPTSEQ, 0x18000000);
-		PRIMOUTREG(MGAREG_WFLAG, 0);
+		PRIMOUTREG( MGAREG_WIADDR2, WIA_wmode_suspend );
+		PRIMOUTREG( MGAREG_WVRTXSZ, 0x00001807 );
+		PRIMOUTREG( MGAREG_DMAPAD, 0 );
+		PRIMOUTREG( MGAREG_DMAPAD, 0 );
+
+		PRIMOUTREG( MGAREG_WACCEPTSEQ, 0 );
+		PRIMOUTREG( MGAREG_WACCEPTSEQ, 0 );
+		PRIMOUTREG( MGAREG_WACCEPTSEQ, 0 );
+		PRIMOUTREG( MGAREG_WACCEPTSEQ, 0x18000000 );
 	}   
 
-	PRIMOUTREG(MGAREG_WFLAG1, 0);   
-	PRIMOUTREG(0x2d00 + 56*4, *((u32 *)(&fParam)));
-	PRIMOUTREG(MGAREG_DMAPAD, 0);
-	PRIMOUTREG(MGAREG_DMAPAD, 0);
+	PRIMOUTREG( MGAREG_WFLAG, 0 );
+	PRIMOUTREG( MGAREG_WFLAG1, 0 );
+	PRIMOUTREG( 0x2d00 + 56*4, *((u32 *)(&fParam)) );
+	PRIMOUTREG( MGAREG_DMAPAD, 0 );
    
-	PRIMOUTREG(0x2d00 + 49*4, 0);  /* Tex stage 0 */
-	PRIMOUTREG(0x2d00 + 57*4, 0);  /* Tex stage 0 */
-	PRIMOUTREG(0x2d00 + 53*4, 0);  /* Tex stage 1 */
-	PRIMOUTREG(0x2d00 + 61*4, 0);  /* Tex stage 1 */
+	PRIMOUTREG( 0x2d00 + 49*4, 0 );  /* Tex stage 0 */
+	PRIMOUTREG( 0x2d00 + 57*4, 0 );  /* Tex stage 0 */
+	PRIMOUTREG( 0x2d00 + 53*4, 0 );  /* Tex stage 1 */
+	PRIMOUTREG( 0x2d00 + 61*4, 0 );  /* Tex stage 1 */
    
-	PRIMOUTREG(0x2d00 + 54*4, 0x40); /* Tex stage 0 : w */
-	PRIMOUTREG(0x2d00 + 62*4, 0x40); /* Tex stage 0 : h */
-	PRIMOUTREG(0x2d00 + 52*4, 0x40); /* Tex stage 1 : w */
-	PRIMOUTREG(0x2d00 + 60*4, 0x40); /* Tex stage 1 : h */
+	PRIMOUTREG( 0x2d00 + 54*4, 0x40 ); /* Tex stage 0 : w */
+	PRIMOUTREG( 0x2d00 + 62*4, 0x40 ); /* Tex stage 0 : h */
+	PRIMOUTREG( 0x2d00 + 52*4, 0x40 ); /* Tex stage 1 : w */
+	PRIMOUTREG( 0x2d00 + 60*4, 0x40 ); /* Tex stage 1 : h */
    
 	/* Dma pading required due to hw bug */
-	PRIMOUTREG(MGAREG_DMAPAD, 0xffffffff);
-	PRIMOUTREG(MGAREG_DMAPAD, 0xffffffff);
-	PRIMOUTREG(MGAREG_DMAPAD, 0xffffffff);
-	PRIMOUTREG(MGAREG_WIADDR2, (__u32)(dev_priv->WarpIndex[pipe].phys_addr |
-					   WIA_wmode_start | WIA_wagp_agp));
-	PRIMADVANCE(dev_priv);
+	PRIMOUTREG( MGAREG_DMAPAD, 0xffffffff );
+	PRIMOUTREG( MGAREG_DMAPAD, 0xffffffff );
+	PRIMOUTREG( MGAREG_DMAPAD, 0xffffffff );
+	PRIMOUTREG( MGAREG_WIADDR2, (u32)(dev_priv->WarpIndex[pipe].phys_addr |
+					  WIA_wmode_start | WIA_wagp_agp) );
+	PRIMADVANCE( dev_priv );
 }
 
 static void mgaG200EmitPipe( drm_mga_private_t *dev_priv )
@@ -264,40 +317,45 @@ static void mgaG200EmitPipe( drm_mga_private_t *dev_priv )
       	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
 	unsigned int pipe = sarea_priv->WarpPipe;
 	PRIMLOCALS;
+	DRM_DEBUG("%s\n", __FUNCTION__);
 
-	PRIMGETPTR(dev_priv);
+	PRIMGETPTR( dev_priv );
    
    	/* This takes 15 dwords */
 
-	PRIMOUTREG(MGAREG_WIADDR, WIA_wmode_suspend);
-	PRIMOUTREG(MGAREG_WVRTXSZ, 7);
-	PRIMOUTREG(MGAREG_WFLAG, 0);
-	PRIMOUTREG(0x2d00 + 24*4, 0); /* tex w/h */
+	PRIMOUTREG( MGAREG_WIADDR, WIA_wmode_suspend );
+	PRIMOUTREG( MGAREG_WVRTXSZ, 7 );
+	PRIMOUTREG( MGAREG_WFLAG, 0 );
+	PRIMOUTREG( 0x2d00 + 24*4, 0 ); /* tex w/h */
    
-	PRIMOUTREG(0x2d00 + 25*4, 0x100);
-	PRIMOUTREG(0x2d00 + 34*4, 0); /* tex w/h */
-	PRIMOUTREG(0x2d00 + 42*4, 0xFFFF);
-	PRIMOUTREG(0x2d00 + 60*4, 0xFFFF);
+	PRIMOUTREG( 0x2d00 + 25*4, 0x100 );
+	PRIMOUTREG( 0x2d00 + 34*4, 0 ); /* tex w/h */
+	PRIMOUTREG( 0x2d00 + 42*4, 0xFFFF );
+	PRIMOUTREG( 0x2d00 + 60*4, 0xFFFF );
    
 	/* Dma pading required due to hw bug */
-	PRIMOUTREG(MGAREG_DMAPAD, 0xffffffff);
-	PRIMOUTREG(MGAREG_DMAPAD, 0xffffffff);
-	PRIMOUTREG(MGAREG_DMAPAD, 0xffffffff);
-	PRIMOUTREG(MGAREG_WIADDR, (__u32)(dev_priv->WarpIndex[pipe].phys_addr | 
-					  WIA_wmode_start | WIA_wagp_agp));
+	PRIMOUTREG( MGAREG_DMAPAD, 0xffffffff );
+	PRIMOUTREG( MGAREG_DMAPAD, 0xffffffff );
+	PRIMOUTREG( MGAREG_DMAPAD, 0xffffffff );
+	PRIMOUTREG( MGAREG_WIADDR, (u32)(dev_priv->WarpIndex[pipe].phys_addr | 
+					 WIA_wmode_start | WIA_wagp_agp) );
 
-	PRIMADVANCE(dev_priv);
+	PRIMADVANCE( dev_priv );
 }
 
 static void mgaEmitState( drm_mga_private_t *dev_priv )
 {
       	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
 	unsigned int dirty = sarea_priv->dirty;
+	DRM_DEBUG("%s\n", __FUNCTION__);
 
 	if (dev_priv->chipset == MGA_CARD_TYPE_G400) {	   
 		int multitex = sarea_priv->WarpPipe & MGA_T2;
 
-		if (sarea_priv->WarpPipe != dev_priv->WarpPipe) { 
+		if (sarea_priv->WarpPipe != dev_priv->WarpPipe) {
+			if ((dev_priv->WarpPipe & MGA_T2) && !multitex) {
+				mgaG400EmitTexFlush( dev_priv );
+			}
 			mgaG400EmitPipe( dev_priv );
 			dev_priv->WarpPipe = sarea_priv->WarpPipe;
 		}
@@ -342,6 +400,8 @@ static int mgaVerifyContext(drm_mga_private_t *dev_priv )
 	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
 	unsigned int *regs = sarea_priv->ContextState;
 
+   	DRM_DEBUG("%s\n", __FUNCTION__);
+
 	if (regs[MGA_CTXREG_DSTORG] != dev_priv->frontOffset &&
 	    regs[MGA_CTXREG_DSTORG] != dev_priv->backOffset) {
 		DRM_DEBUG("BAD DSTORG: %x (front %x, back %x)\n\n", 
@@ -361,6 +421,8 @@ static int mgaVerifyTex(drm_mga_private_t *dev_priv,
 {
 	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
 
+   	DRM_DEBUG("%s\n", __FUNCTION__);
+
 	if ((sarea_priv->TexState[unit][MGA_TEXREG_ORG] & 0x3) == 0x1) {
 		DRM_DEBUG("BAD TEXREG_ORG: %x, unit %d\n", 
 		       sarea_priv->TexState[unit][MGA_TEXREG_ORG],
@@ -378,6 +440,8 @@ static int mgaVerifyState( drm_mga_private_t *dev_priv )
 	unsigned int dirty = sarea_priv->dirty;
 	int rv = 0;
 
+   	DRM_DEBUG("%s\n", __FUNCTION__);
+
    	if (sarea_priv->nbox > MGA_NR_SAREA_CLIPRECTS)
      		sarea_priv->nbox = MGA_NR_SAREA_CLIPRECTS;
 
@@ -408,6 +472,8 @@ static int mgaVerifyIload( drm_mga_private_t *dev_priv,
 			   unsigned long bus_address, 
 			   unsigned int dstOrg, int length )
 {
+   	DRM_DEBUG("%s\n", __FUNCTION__);
+
    	if(dstOrg < dev_priv->textureOffset ||
 	   dstOrg + length > 
 	   (dev_priv->textureOffset + dev_priv->textureSize)) {
@@ -431,35 +497,34 @@ static void mga_dma_dispatch_tex_blit( drm_device_t *dev,
    	int use_agp = PDEA_pagpxfer_enable | 0x00000001;
    	u16 y2;
     	PRIMLOCALS;
+	DRM_DEBUG("%s\n", __FUNCTION__);
     
         y2 = length / 64;
  
-	PRIM_OVERFLOW(dev, dev_priv, 30);
+	PRIM_OVERFLOW( dev, dev_priv, 30 );
 	PRIMGETPTR( dev_priv );
 
-      	dev_priv->last_sync_tag = mga_create_sync_tag(dev);
-
-	PRIMOUTREG( MGAREG_DSTORG, destOrg);
-	PRIMOUTREG( MGAREG_MACCESS, 0x00000000);
+	PRIMOUTREG( MGAREG_DSTORG, destOrg );
+	PRIMOUTREG( MGAREG_MACCESS, 0x00000000 );
    	DRM_DEBUG("srcorg : %lx\n", bus_address | use_agp);
-	PRIMOUTREG( MGAREG_SRCORG, (u32) bus_address | use_agp);
-	PRIMOUTREG( MGAREG_AR5, 64);  
-
-	PRIMOUTREG( MGAREG_PITCH, 64);
-	PRIMOUTREG( MGAREG_DMAPAD, 0);
-	PRIMOUTREG( MGAREG_DMAPAD, 0);
-	PRIMOUTREG( MGAREG_DWGCTL, MGA_COPY_CMD); 
-
-   	PRIMOUTREG(MGAREG_AR0, 63);
-   	PRIMOUTREG(MGAREG_AR3, 0);		
-   	PRIMOUTREG(MGAREG_FXBNDRY, (63 << 16));
-   	PRIMOUTREG(MGAREG_YDSTLEN+MGAREG_MGA_EXEC, y2);
+	PRIMOUTREG( MGAREG_SRCORG, (u32) bus_address | use_agp );
+	PRIMOUTREG( MGAREG_AR5, 64 );
+
+	PRIMOUTREG( MGAREG_PITCH, 64 );
+	PRIMOUTREG( MGAREG_DMAPAD, 0 );
+	PRIMOUTREG( MGAREG_DMAPAD, 0 );
+	PRIMOUTREG( MGAREG_DWGCTL, MGA_COPY_CMD ); 
+
+   	PRIMOUTREG( MGAREG_AR0, 63 );
+   	PRIMOUTREG( MGAREG_AR3, 0 );		
+   	PRIMOUTREG( MGAREG_FXBNDRY, (63 << 16) );
+   	PRIMOUTREG( MGAREG_YDSTLEN+MGAREG_MGA_EXEC, y2 );
   
-	PRIMOUTREG( MGAREG_SRCORG, 0);
-	PRIMOUTREG( MGAREG_PITCH, dev_priv->stride / dev_priv->cpp);
-      	PRIMOUTREG( MGAREG_DMAPAD, 0);
-      	PRIMOUTREG( MGAREG_DWGSYNC, dev_priv->last_sync_tag);
-    	PRIMADVANCE(dev_priv);
+	PRIMOUTREG( MGAREG_SRCORG, 0 );
+	PRIMOUTREG( MGAREG_PITCH, dev_priv->stride / dev_priv->cpp );
+      	PRIMOUTREG( MGAREG_DMAPAD, 0 );
+      	PRIMOUTREG( MGAREG_DMAPAD, 0 );
+    	PRIMADVANCE( dev_priv );
 }
 
 static void mga_dma_dispatch_vertex(drm_device_t *dev, 
@@ -474,33 +539,100 @@ static void mga_dma_dispatch_vertex(drm_device_t *dev,
 	int i = 0;
    	int primary_needed;
       	PRIMLOCALS;
+	DRM_DEBUG("%s\n", __FUNCTION__);
 
 	DRM_DEBUG("dispatch vertex %d addr 0x%lx, "
 		  "length 0x%x nbox %d dirty %x\n", 
 		  buf->idx, address, length, 
 		  sarea_priv->nbox, sarea_priv->dirty);
 
+	DRM_DEBUG("used : %d, total : %d\n", buf->used, buf->total);
+	if(sarea_priv->WarpPipe & MGA_T2) { 
+	   if ((buf->used/4) % 10)
+	     DRM_DEBUG("Multitex Buf is not aligned properly!!!\n");
+	} else { 
+	   if ((buf->used/4) % 8)
+	     DRM_DEBUG("Buf is not aligned properly!!!\n");
+	}
 
-   	dev_priv->last_sync_tag = mga_create_sync_tag(dev);
+	if (buf->used) {
+		/* WARNING: if you change any of the state functions verify
+		 * these numbers (Overestimating this doesn't hurt).  
+		 */
+		buf_priv->dispatched = 1;
+		primary_needed = (30+15+15+30+25+ 
+				  10 + 
+				  15 * MGA_NR_SAREA_CLIPRECTS);
+		PRIM_OVERFLOW(dev, dev_priv, primary_needed);
+   		mgaEmitState( dev_priv );
+
+		do {
+			if (i < sarea_priv->nbox) {
+				DRM_DEBUG("idx %d Emit box %d/%d:"
+					  "%d,%d - %d,%d\n", 
+					  buf->idx,
+					  i, sarea_priv->nbox,
+					  sarea_priv->boxes[i].x1, 
+					  sarea_priv->boxes[i].y1,
+					  sarea_priv->boxes[i].x2, 
+					  sarea_priv->boxes[i].y2);
+				
+				mgaEmitClipRect( dev_priv, 
+						 &sarea_priv->boxes[i] );
+			}
+			
+			PRIMGETPTR( dev_priv );
+			PRIMOUTREG( MGAREG_DMAPAD, 0 );
+			PRIMOUTREG( MGAREG_DMAPAD, 0 );
+			PRIMOUTREG( MGAREG_SECADDRESS, 
+				    ((u32)address) | TT_VERTEX );
+			PRIMOUTREG( MGAREG_SECEND, 
+				    (((u32)(address + length)) | use_agp) );
+			PRIMADVANCE( dev_priv );
+		}  while (++i < sarea_priv->nbox);
+	}
 
-   	if (buf_priv->discard) { 
-		buf_priv->my_freelist->age = dev_priv->last_sync_tag;
+   	if (buf_priv->discard) {
+		if (buf_priv->dispatched == 1) AGEBUF(dev_priv, buf_priv);
+		buf_priv->dispatched = 0;
 	   	mga_freelist_put(dev, buf);
 	}
 
 
-	/* WARNING: if you change any of the state functions verify
-	 * these numbers (Overestimating this doesn't hurt).  
-	 */
-	primary_needed = (25+15+30+25+ 
-			  10 + 
-			  15 * MGA_NR_SAREA_CLIPRECTS);
+}
 
 
-   	PRIM_OVERFLOW(dev, dev_priv, primary_needed);
-   	mgaEmitState( dev_priv );
+static void mga_dma_dispatch_indices(drm_device_t *dev, 
+				     drm_buf_t *buf,
+				     unsigned int start,
+				     unsigned int end)
+{
+   	drm_mga_private_t *dev_priv = dev->dev_private;
+	drm_mga_buf_priv_t *buf_priv = buf->dev_private;
+   	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	unsigned int address = (unsigned int)buf->bus_address;
+	int use_agp = PDEA_pagpxfer_enable;
+	int i = 0;
+   	int primary_needed;
+      	PRIMLOCALS;
+	DRM_DEBUG("%s\n", __FUNCTION__);
+
+	DRM_DEBUG("dispatch indices %d addr 0x%x, "
+		  "start 0x%x end 0x%x nbox %d dirty %x\n", 
+		  buf->idx, address, start, end, 
+		  sarea_priv->nbox, sarea_priv->dirty);
+
+	if (start != end) {
+		/* WARNING: if you change any of the state functions verify
+		 * these numbers (Overestimating this doesn't hurt).  
+		 */
+		buf_priv->dispatched = 1;
+		primary_needed = (25+15+30+25+ 
+				  10 + 
+				  15 * MGA_NR_SAREA_CLIPRECTS);
+   		PRIM_OVERFLOW( dev, dev_priv, primary_needed );
+   		mgaEmitState( dev_priv );
 
-	if (buf->used) {
 		do {
 			if (i < sarea_priv->nbox) {
 				DRM_DEBUG("idx %d Emit box %d/%d:"
@@ -515,25 +647,23 @@ static void mga_dma_dispatch_vertex(drm_device_t *dev,
 				mgaEmitClipRect( dev_priv, 
 						 &sarea_priv->boxes[i] );
 			}
-			
-			PRIMGETPTR(dev_priv);
-			PRIMOUTREG( MGAREG_DMAPAD, 0);
-			PRIMOUTREG( MGAREG_DMAPAD, 0);
-			PRIMOUTREG( MGAREG_SECADDRESS, 
-				    ((__u32)address) | TT_VERTEX);
-			PRIMOUTREG( MGAREG_SECEND, 
-				    (((__u32)(address + length)) | 
-				     use_agp));
+
+			PRIMGETPTR( dev_priv );
+			PRIMOUTREG( MGAREG_DMAPAD, 0 );
+			PRIMOUTREG( MGAREG_DMAPAD, 0 );
+			PRIMOUTREG( MGAREG_SETUPADDRESS, 
+				    ((address + start) |
+				     SETADD_mode_vertlist) );
+			PRIMOUTREG( MGAREG_SETUPEND, 
+				    ((address + end) | use_agp) );
 			PRIMADVANCE( dev_priv );	       
 		}  while (++i < sarea_priv->nbox);
 	}
-
-	PRIMGETPTR( dev_priv );
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_DWGSYNC, dev_priv->last_sync_tag);   
-	PRIMADVANCE( dev_priv );
+   	if (buf_priv->discard) {
+		if (buf_priv->dispatched == 1) AGEBUF(dev_priv, buf_priv);
+		buf_priv->dispatched = 0;
+	   	mga_freelist_put(dev, buf);
+	}
 }
 
 
@@ -550,6 +680,7 @@ static void mga_dma_dispatch_clear( drm_device_t *dev, int flags,
 	int i;
    	int primary_needed;
    	PRIMLOCALS;
+	DRM_DEBUG("%s\n", __FUNCTION__);
 
 	if ( dev_priv->sgram ) 
 		cmd = MGA_CLEAR_CMD | DC_atype_blk;
@@ -557,10 +688,9 @@ static void mga_dma_dispatch_clear( drm_device_t *dev, int flags,
 		cmd = MGA_CLEAR_CMD | DC_atype_rstr;
    
    	primary_needed = nbox * 70;
-   	if(primary_needed == 0) primary_needed = 70;
-	PRIM_OVERFLOW(dev, dev_priv, primary_needed);
+   	if (primary_needed == 0) primary_needed = 70;
+	PRIM_OVERFLOW( dev, dev_priv, primary_needed );
 	PRIMGETPTR( dev_priv );
-      	dev_priv->last_sync_tag = mga_create_sync_tag(dev);
    
 	for (i = 0 ; i < nbox ; i++) {
 		unsigned int height = pbox[i].y2 - pbox[i].y1;
@@ -571,55 +701,50 @@ static void mga_dma_dispatch_clear( drm_device_t *dev, int flags,
 
 	   	if ( flags & MGA_FRONT ) {	    
 		   	DRM_DEBUG("clear front\n");
-			PRIMOUTREG( MGAREG_DMAPAD, 0);
-			PRIMOUTREG( MGAREG_DMAPAD, 0);
-			PRIMOUTREG(MGAREG_YDSTLEN, (pbox[i].y1<<16)|height);
-			PRIMOUTREG(MGAREG_FXBNDRY, (pbox[i].x2<<16)|pbox[i].x1);
-
-			PRIMOUTREG( MGAREG_DMAPAD, 0);
-			PRIMOUTREG(MGAREG_FCOL, clear_color);
-			PRIMOUTREG(MGAREG_DSTORG, dev_priv->frontOffset);
-			PRIMOUTREG(MGAREG_DWGCTL+MGAREG_MGA_EXEC, cmd );
+			PRIMOUTREG( MGAREG_DMAPAD, 0 );
+			PRIMOUTREG( MGAREG_DMAPAD, 0 );
+			PRIMOUTREG( MGAREG_YDSTLEN, (pbox[i].y1<<16)|height);
+			PRIMOUTREG( MGAREG_FXBNDRY, (pbox[i].x2<<16)|pbox[i].x1);
+
+			PRIMOUTREG( MGAREG_DMAPAD, 0 );
+			PRIMOUTREG( MGAREG_FCOL, clear_color );
+			PRIMOUTREG( MGAREG_DSTORG, dev_priv->frontOffset );
+			PRIMOUTREG( MGAREG_DWGCTL+MGAREG_MGA_EXEC, cmd );
 		}
 
 		if ( flags & MGA_BACK ) {
 			DRM_DEBUG("clear back\n");
 			PRIMOUTREG( MGAREG_DMAPAD, 0);
 			PRIMOUTREG( MGAREG_DMAPAD, 0);
-			PRIMOUTREG(MGAREG_YDSTLEN, (pbox[i].y1<<16)|height);
-			PRIMOUTREG(MGAREG_FXBNDRY, (pbox[i].x2<<16)|pbox[i].x1);
+			PRIMOUTREG( MGAREG_YDSTLEN, (pbox[i].y1<<16)|height );
+			PRIMOUTREG( MGAREG_FXBNDRY, (pbox[i].x2<<16)|pbox[i].x1 );
 
-			PRIMOUTREG( MGAREG_DMAPAD, 0);
-			PRIMOUTREG(MGAREG_FCOL, clear_color);
-			PRIMOUTREG(MGAREG_DSTORG, dev_priv->backOffset);
-			PRIMOUTREG(MGAREG_DWGCTL+MGAREG_MGA_EXEC, cmd );
+			PRIMOUTREG( MGAREG_DMAPAD, 0 );
+			PRIMOUTREG( MGAREG_FCOL, clear_color );
+			PRIMOUTREG( MGAREG_DSTORG, dev_priv->backOffset );
+			PRIMOUTREG( MGAREG_DWGCTL+MGAREG_MGA_EXEC, cmd );
 		}
 
 		if ( flags & MGA_DEPTH ) {
 			DRM_DEBUG("clear depth\n");
-			PRIMOUTREG( MGAREG_DMAPAD, 0);
-			PRIMOUTREG( MGAREG_DMAPAD, 0);
-			PRIMOUTREG(MGAREG_YDSTLEN, (pbox[i].y1<<16)|height);
-			PRIMOUTREG(MGAREG_FXBNDRY, (pbox[i].x2<<16)|pbox[i].x1);
-
-			PRIMOUTREG( MGAREG_DMAPAD, 0);
-			PRIMOUTREG(MGAREG_FCOL, clear_zval);
-			PRIMOUTREG(MGAREG_DSTORG, dev_priv->depthOffset);
-			PRIMOUTREG(MGAREG_DWGCTL+MGAREG_MGA_EXEC, cmd );
+			PRIMOUTREG( MGAREG_DMAPAD, 0 );
+			PRIMOUTREG( MGAREG_DMAPAD, 0 );
+			PRIMOUTREG( MGAREG_YDSTLEN, (pbox[i].y1<<16)|height );
+			PRIMOUTREG( MGAREG_FXBNDRY, (pbox[i].x2<<16)|pbox[i].x1 );
+
+			PRIMOUTREG( MGAREG_DMAPAD, 0 );
+			PRIMOUTREG( MGAREG_FCOL, clear_zval );
+			PRIMOUTREG( MGAREG_DSTORG, dev_priv->depthOffset );
+			PRIMOUTREG( MGAREG_DWGCTL+MGAREG_MGA_EXEC, cmd );
 		}
 	}
 
 	/* Force reset of DWGCTL */
-   	PRIMOUTREG( MGAREG_DMAPAD, 0);   
-	PRIMOUTREG( MGAREG_DMAPAD, 0);   
-      	PRIMOUTREG( MGAREG_DMAPAD, 0);
+   	PRIMOUTREG( MGAREG_DMAPAD, 0 );   
+	PRIMOUTREG( MGAREG_DMAPAD, 0 );   
+      	PRIMOUTREG( MGAREG_DMAPAD, 0 );
       	PRIMOUTREG( MGAREG_DWGCTL, regs[MGA_CTXREG_DWGCTL] );
-
-	PRIMOUTREG( MGAREG_DMAPAD, 0);
-	PRIMOUTREG( MGAREG_DMAPAD, 0);
-      	PRIMOUTREG( MGAREG_DMAPAD, 0);
-   	PRIMOUTREG( MGAREG_DWGSYNC, dev_priv->last_sync_tag);
-	PRIMADVANCE(dev_priv);
+	PRIMADVANCE( dev_priv );
 }
 
 static void mga_dma_dispatch_swap( drm_device_t *dev )
@@ -632,23 +757,22 @@ static void mga_dma_dispatch_swap( drm_device_t *dev )
 	int i;
    	int primary_needed;
    	PRIMLOCALS;
+	DRM_DEBUG("%s\n", __FUNCTION__);
 
    	primary_needed = nbox * 5;
    	primary_needed += 60;
 	PRIM_OVERFLOW(dev, dev_priv, primary_needed);
 	PRIMGETPTR( dev_priv );
    
-      	dev_priv->last_sync_tag = mga_create_sync_tag(dev);
+	PRIMOUTREG( MGAREG_DSTORG, dev_priv->frontOffset );
+	PRIMOUTREG( MGAREG_MACCESS, dev_priv->mAccess );
+	PRIMOUTREG( MGAREG_SRCORG, dev_priv->backOffset );
+	PRIMOUTREG( MGAREG_AR5, dev_priv->stride/2 );  
 
-	PRIMOUTREG(MGAREG_DSTORG, dev_priv->frontOffset);
-	PRIMOUTREG(MGAREG_MACCESS, dev_priv->mAccess);
-	PRIMOUTREG(MGAREG_SRCORG, dev_priv->backOffset);
-	PRIMOUTREG(MGAREG_AR5, dev_priv->stride/2);  
-
-	PRIMOUTREG( MGAREG_DMAPAD, 0);
-	PRIMOUTREG( MGAREG_DMAPAD, 0);
-	PRIMOUTREG( MGAREG_DMAPAD, 0);
-	PRIMOUTREG(MGAREG_DWGCTL, MGA_COPY_CMD); 
+	PRIMOUTREG( MGAREG_DMAPAD, 0 );
+	PRIMOUTREG( MGAREG_DMAPAD, 0 );
+	PRIMOUTREG( MGAREG_DMAPAD, 0 );
+	PRIMOUTREG( MGAREG_DWGCTL, MGA_COPY_CMD ); 
 	     
 	for (i = 0 ; i < nbox; i++) {
 		unsigned int h = pbox[i].y2 - pbox[i].y1;
@@ -658,23 +782,19 @@ static void mga_dma_dispatch_swap( drm_device_t *dev )
 			  pbox[i].x1, pbox[i].y1,
 			  pbox[i].x2, pbox[i].y2);
 
-		PRIMOUTREG(MGAREG_AR0, start + pbox[i].x2 - 1);
-		PRIMOUTREG(MGAREG_AR3, start + pbox[i].x1);		
-		PRIMOUTREG(MGAREG_FXBNDRY, pbox[i].x1|((pbox[i].x2 - 1)<<16));
-		PRIMOUTREG(MGAREG_YDSTLEN+MGAREG_MGA_EXEC, (pbox[i].y1<<16)|h);
+		PRIMOUTREG( MGAREG_AR0, start + pbox[i].x2 - 1 );
+		PRIMOUTREG( MGAREG_AR3, start + pbox[i].x1 );		
+		PRIMOUTREG( MGAREG_FXBNDRY, pbox[i].x1|((pbox[i].x2 - 1)<<16) );
+		PRIMOUTREG( MGAREG_YDSTLEN+MGAREG_MGA_EXEC, (pbox[i].y1<<16)|h );
 	}
 
    	/* Force reset of DWGCTL */
-   	PRIMOUTREG( MGAREG_DMAPAD, 0);
-      	PRIMOUTREG( MGAREG_DMAPAD, 0);
-   	PRIMOUTREG( MGAREG_DMAPAD, 0);
+   	PRIMOUTREG( MGAREG_DMAPAD, 0 );
+      	PRIMOUTREG( MGAREG_DMAPAD, 0 );
+   	PRIMOUTREG( MGAREG_SRCORG, 0 );
       	PRIMOUTREG( MGAREG_DWGCTL, regs[MGA_CTXREG_DWGCTL] );
 
-	PRIMOUTREG( MGAREG_SRCORG, 0);
-	PRIMOUTREG( MGAREG_DMAPAD, 0);
-      	PRIMOUTREG( MGAREG_DMAPAD, 0);
-      	PRIMOUTREG( MGAREG_DWGSYNC, dev_priv->last_sync_tag);
-   	PRIMADVANCE(dev_priv);
+   	PRIMADVANCE( dev_priv );
 }
 
 int mga_clear_bufs(struct inode *inode, struct file *filp,
@@ -684,12 +804,11 @@ int mga_clear_bufs(struct inode *inode, struct file *filp,
 	drm_device_t *dev = priv->dev;
    	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
    	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
-      	__volatile__ unsigned int *status = 
-     		(__volatile__ unsigned int *)dev_priv->status_page;
 	drm_mga_clear_t clear;
 
    	copy_from_user_ret(&clear, (drm_mga_clear_t *)arg, sizeof(clear), 
 			   -EFAULT);
+	DRM_DEBUG("%s\n", __FUNCTION__);
 
 	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
 		DRM_ERROR("mga_clear_bufs called without lock held\n");
@@ -706,8 +825,8 @@ int mga_clear_bufs(struct inode *inode, struct file *filp,
 			        clear.clear_color, 
 			        clear.clear_depth );
       	PRIMUPDATE(dev_priv);
+	mga_flush_write_combine();
 	mga_dma_schedule(dev, 1);
-   	sarea_priv->last_dispatch = status[1];
    	return 0;
 }
 
@@ -718,8 +837,7 @@ int mga_swap_bufs(struct inode *inode, struct file *filp,
 	drm_device_t *dev = priv->dev;
    	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
       	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
-      	__volatile__ unsigned int *status = 
-     		(__volatile__ unsigned int *)dev_priv->status_page;
+	DRM_DEBUG("%s\n", __FUNCTION__);
    
 	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
 		DRM_ERROR("mga_swap_bufs called without lock held\n");
@@ -734,10 +852,9 @@ int mga_swap_bufs(struct inode *inode, struct file *filp,
 	dev_priv->sarea_priv->dirty |= MGA_UPLOAD_CTX;
 	mga_dma_dispatch_swap( dev );
       	PRIMUPDATE(dev_priv);
-   	set_bit(0, &dev_priv->current_prim->swap_pending);
-   	dev_priv->current_prim->swap_pending = 1;
+   	set_bit(MGA_BUF_SWAP_PENDING, &dev_priv->current_prim->buffer_status);
+	mga_flush_write_combine();
 	mga_dma_schedule(dev, 1);
-      	sarea_priv->last_dispatch = status[1];
    	return 0;
 }
 
@@ -749,12 +866,11 @@ int mga_iload(struct inode *inode, struct file *filp,
    	drm_device_dma_t  *dma	    = dev->dma;
        	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
        	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
-       	__volatile__ unsigned int *status = 
-      		(__volatile__ unsigned int *)dev_priv->status_page;
    	drm_buf_t *buf;
    	drm_mga_buf_priv_t *buf_priv;
  	drm_mga_iload_t iload;
 	unsigned long bus_address;
+	DRM_DEBUG("%s\n", __FUNCTION__);
    
    	DRM_DEBUG("Starting Iload\n");
  	copy_from_user_ret(&iload, (drm_mga_iload_t *)arg, sizeof(iload),
@@ -783,11 +899,11 @@ int mga_iload(struct inode *inode, struct file *filp,
 
    	mga_dma_dispatch_tex_blit(dev, bus_address, iload.length, 
 				  iload.destOrg);
-   	buf_priv->my_freelist->age = dev_priv->last_sync_tag;
+	AGEBUF(dev_priv, buf_priv);
 	buf_priv->discard = 1;
     	mga_freelist_put(dev, buf);
+	mga_flush_write_combine();
 	mga_dma_schedule(dev, 1);
-    	sarea_priv->last_dispatch = status[1];
  	return 0; 
 }
 
@@ -797,13 +913,11 @@ int mga_vertex(struct inode *inode, struct file *filp,
 	drm_file_t *priv = filp->private_data;
 	drm_device_t *dev = priv->dev;
       	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
-   	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
-      	__volatile__ unsigned int *status = 
-     		(__volatile__ unsigned int *)dev_priv->status_page;
    	drm_device_dma_t *dma = dev->dma;
 	drm_buf_t *buf;
       	drm_mga_buf_priv_t *buf_priv;
 	drm_mga_vertex_t vertex;
+	DRM_DEBUG("%s\n", __FUNCTION__);
 
 	copy_from_user_ret(&vertex, (drm_mga_vertex_t *)arg, sizeof(vertex),
 			   -EFAULT);
@@ -822,25 +936,75 @@ int mga_vertex(struct inode *inode, struct file *filp,
 	buf_priv->discard = vertex.discard;
    
 	if (!mgaVerifyState(dev_priv)) {
-	   if (vertex.discard) { 
-	     	buf_priv->my_freelist->age = dev_priv->last_sync_tag;
-	      	mga_freelist_put(dev, buf);
+	   if (vertex.discard) {
+		if(buf_priv->dispatched == 1) AGEBUF(dev_priv, buf_priv);
+		buf_priv->dispatched = 0;
+	   	mga_freelist_put(dev, buf);
 	   }
+	   DRM_DEBUG("bad state\n");
 	   return -EINVAL;
 	}
 
 	mga_dma_dispatch_vertex(dev, buf);
 
    	PRIMUPDATE(dev_priv);
+	mga_flush_write_combine();
 	mga_dma_schedule(dev, 1);
-      	sarea_priv->last_dispatch = status[1];
    	return 0;
 }
 
+
+int mga_indices(struct inode *inode, struct file *filp,
+		unsigned int cmd, unsigned long arg)
+{
+	drm_file_t *priv = filp->private_data;
+	drm_device_t *dev = priv->dev;
+      	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
+   	drm_device_dma_t *dma = dev->dma;
+	drm_buf_t *buf;
+      	drm_mga_buf_priv_t *buf_priv;
+	drm_mga_indices_t indices;
+	DRM_DEBUG("%s\n", __FUNCTION__);
+
+	copy_from_user_ret(&indices, (drm_mga_indices_t *)arg, sizeof(indices),
+			   -EFAULT);
+   
+	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("mga_indices called without lock held\n");
+		return -EINVAL;
+	}
+
+	DRM_DEBUG("mga_indices\n");
+
+	buf = dma->buflist[ indices.idx ];
+   	buf_priv = buf->dev_private;
+
+	buf_priv->discard = indices.discard;
+   
+	if (!mgaVerifyState(dev_priv)) {
+	   if (indices.discard) {
+		if(buf_priv->dispatched == 1) AGEBUF(dev_priv, buf_priv);
+		buf_priv->dispatched = 0;
+	   	mga_freelist_put(dev, buf);
+	   }
+	   return -EINVAL;
+	}
+
+	mga_dma_dispatch_indices(dev, buf, indices.start, indices.end);
+
+   	PRIMUPDATE(dev_priv);
+	mga_flush_write_combine();
+	mga_dma_schedule(dev, 1);
+   	return 0;
+}
+
+
+
 static int mga_dma_get_buffers(drm_device_t *dev, drm_dma_t *d)
 {
 	int i;
 	drm_buf_t *buf;
+	DRM_DEBUG("%s\n", __FUNCTION__);
 
 	for (i = d->granted_count; i < d->request_count; i++) {
 		buf = mga_freelist_get(dev);
@@ -865,12 +1029,9 @@ int mga_dma(struct inode *inode, struct file *filp, unsigned int cmd,
 	drm_file_t	  *priv	    = filp->private_data;
 	drm_device_t	  *dev	    = priv->dev;
 	drm_device_dma_t  *dma	    = dev->dma;
-   	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
-   	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
-      	__volatile__ unsigned int *status = 
-     		(__volatile__ unsigned int *)dev_priv->status_page;
 	int		  retcode   = 0;
 	drm_dma_t	  d;
+	DRM_DEBUG("%s\n", __FUNCTION__);
 
    	copy_from_user_ret(&d, (drm_dma_t *)arg, sizeof(d), -EFAULT);
 	DRM_DEBUG("%d %d: %d send, %d req\n",
@@ -906,6 +1067,5 @@ int mga_dma(struct inode *inode, struct file *filp, unsigned int cmd,
 	DRM_DEBUG("%d returning, granted = %d\n",
 		  current->pid, d.granted_count);
 	copy_to_user_ret((drm_dma_t *)arg, &d, sizeof(d), -EFAULT);
-      	sarea_priv->last_dispatch = status[1];
 	return retcode;
 }
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/proc.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/proc.c
index bb8e18e0e..ba6dee00d 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/proc.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/proc.c
@@ -400,6 +400,7 @@ static int _drm_vma_info(char *buf, char **start, off_t offset, int len,
 			       pgprot & _PAGE_GLOBAL   ? 'g' : 'l' );
 #endif		
 		DRM_PROC_PRINT("\n");
+#if 0
 		for (i = vma->vm_start; i < vma->vm_end; i += PAGE_SIZE) {
 			pgd = pgd_offset(vma->vm_mm, i);
 			pmd = pmd_offset(pgd, i);
@@ -420,6 +421,7 @@ static int _drm_vma_info(char *buf, char **start, off_t offset, int len,
 				DRM_PROC_PRINT("      0x%08lx\n", i);
 			}
 		}
+#endif
 	}
 	
 	return len;
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_bufs.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_bufs.c
new file mode 100644
index 000000000..bad6c5714
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_bufs.c
@@ -0,0 +1,309 @@
+/* r128_bufs.c -- IOCTLs to manage buffers -*- linux-c -*-
+ * Created: Wed Apr 12 16:19:08 2000 by kevin@precisioninsight.com
+ *
+ * Copyright 2000 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * Authors: Kevin E. Martin <kevin@precisioninsight.com>
+ *          Rickard E. (Rik) Faith <faith@precisioninsight.com>
+ *	    Jeff Hartmann <jhartmann@precisioninsight.com>
+ * 
+ * $XFree86$
+ *
+ */
+
+#define __NO_VERSION__
+#include "drmP.h"
+#include "r128_drv.h"
+#include "linux/un.h"
+
+
+#ifdef DRM_AGP
+int r128_addbufs_agp(struct inode *inode, struct file *filp, unsigned int cmd,
+		     unsigned long arg)
+{
+	drm_file_t       *priv = filp->private_data;
+	drm_device_t     *dev  = priv->dev;
+	drm_device_dma_t *dma  = dev->dma;
+	drm_buf_desc_t    request;
+	drm_buf_entry_t  *entry;
+	drm_buf_t        *buf;
+	unsigned long     offset;
+	unsigned long     agp_offset;
+	int               count;
+	int               order;
+	int               size;
+	int               alignment;
+	int               page_order;
+	int               total;
+	int               byte_count;
+	int               i;
+
+	if (!dma) return -EINVAL;
+
+	copy_from_user_ret(&request,
+			   (drm_buf_desc_t *)arg,
+			   sizeof(request),
+			   -EFAULT);
+
+	count      = request.count;
+	order      = drm_order(request.size);
+	size       = 1 << order;
+
+	alignment  = (request.flags & _DRM_PAGE_ALIGN) ? PAGE_ALIGN(size):size;
+	page_order = order - PAGE_SHIFT > 0 ? order - PAGE_SHIFT : 0;
+	total      = PAGE_SIZE << page_order;
+
+	byte_count = 0;
+	agp_offset = dev->agp->base + request.agp_start;
+
+	DRM_DEBUG("count:      %d\n",  count);
+	DRM_DEBUG("order:      %d\n",  order);
+	DRM_DEBUG("size:       %d\n",  size);
+	DRM_DEBUG("agp_offset: %ld\n", agp_offset);
+	DRM_DEBUG("alignment:  %d\n",  alignment);
+	DRM_DEBUG("page_order: %d\n",  page_order);
+	DRM_DEBUG("total:      %d\n",  total);
+
+	if (order < DRM_MIN_ORDER || order > DRM_MAX_ORDER) return -EINVAL;
+	if (dev->queue_count) return -EBUSY; /* Not while in use */
+
+	spin_lock(&dev->count_lock);
+	if (dev->buf_use) {
+		spin_unlock(&dev->count_lock);
+		return -EBUSY;
+	}
+	atomic_inc(&dev->buf_alloc);
+	spin_unlock(&dev->count_lock);
+   
+	down(&dev->struct_sem);
+	entry = &dma->bufs[order];
+	if (entry->buf_count) {
+		up(&dev->struct_sem);
+		atomic_dec(&dev->buf_alloc);
+		return -ENOMEM; /* May only call once for each order */
+	}
+   
+	entry->buflist = drm_alloc(count * sizeof(*entry->buflist),
+				   DRM_MEM_BUFS);
+	if (!entry->buflist) {
+		up(&dev->struct_sem);
+		atomic_dec(&dev->buf_alloc);
+		return -ENOMEM;
+	}
+	memset(entry->buflist, 0, count * sizeof(*entry->buflist));
+   
+	entry->buf_size   = size;
+	entry->page_order = page_order;
+	offset            = 0;
+
+	for (offset = 0;
+	     entry->buf_count < count;
+	     offset += alignment, ++entry->buf_count) {
+		buf          = &entry->buflist[entry->buf_count];
+		buf->idx     = dma->buf_count + entry->buf_count;
+		buf->total   = alignment;
+		buf->order   = order;
+		buf->used    = 0;
+		buf->offset  = (dma->byte_count + offset);
+		buf->address = (void *)(agp_offset + offset);
+		buf->next    = NULL;
+		buf->waiting = 0;
+		buf->pending = 0;
+		init_waitqueue_head(&buf->dma_wait);
+		buf->pid     = 0;
+
+		buf->dev_priv_size = sizeof(drm_r128_buf_priv_t);
+		buf->dev_private   = drm_alloc(sizeof(drm_r128_buf_priv_t),
+					       DRM_MEM_BUFS);
+		memset(buf->dev_private, 0, buf->dev_priv_size);
+
+#if DRM_DMA_HISTOGRAM
+		buf->time_queued     = 0;
+		buf->time_dispatched = 0;
+		buf->time_completed  = 0;
+		buf->time_freed      = 0;
+#endif
+
+		byte_count += PAGE_SIZE << page_order;
+
+		DRM_DEBUG("buffer %d @ %p\n",
+			  entry->buf_count, buf->address);
+	}
+
+	DRM_DEBUG("byte_count: %d\n", byte_count);
+
+	dma->buflist = drm_realloc(dma->buflist,
+				   dma->buf_count * sizeof(*dma->buflist),
+				   (dma->buf_count + entry->buf_count)
+				   * sizeof(*dma->buflist),
+				   DRM_MEM_BUFS);
+	for (i = dma->buf_count; i < dma->buf_count + entry->buf_count; i++)
+		dma->buflist[i] = &entry->buflist[i - dma->buf_count];
+
+	dma->buf_count  += entry->buf_count;
+	dma->byte_count += byte_count;
+
+	drm_freelist_create(&entry->freelist, entry->buf_count);
+	for (i = 0; i < entry->buf_count; i++) {
+		drm_freelist_put(dev, &entry->freelist, &entry->buflist[i]);
+	}
+
+	up(&dev->struct_sem);
+
+	request.count = entry->buf_count;
+	request.size  = size;
+
+	copy_to_user_ret((drm_buf_desc_t *)arg,
+			 &request,
+			 sizeof(request),
+			 -EFAULT);
+
+	dma->flags = _DRM_DMA_USE_AGP;
+
+	atomic_dec(&dev->buf_alloc);
+	return 0;
+}
+#endif
+
+int r128_addbufs(struct inode *inode, struct file *filp, unsigned int cmd,
+		 unsigned long arg)
+{
+	drm_file_t		*priv		= filp->private_data;
+	drm_device_t		*dev		= priv->dev;
+	drm_r128_private_t	*dev_priv	= dev->dev_private;
+	drm_buf_desc_t		request;
+
+	if (!dev_priv || dev_priv->is_pci) return -EINVAL;
+
+	copy_from_user_ret(&request,
+			   (drm_buf_desc_t *)arg,
+			   sizeof(request),
+			   -EFAULT);
+
+#ifdef DRM_AGP
+	if (request.flags & _DRM_AGP_BUFFER)
+		return r128_addbufs_agp(inode, filp, cmd, arg);
+	else
+#endif
+		return -EINVAL;
+}
+
+int r128_mapbufs(struct inode *inode, struct file *filp, unsigned int cmd,
+		 unsigned long arg)
+{
+	drm_file_t		*priv		= filp->private_data;
+	drm_device_t		*dev		= priv->dev;
+	drm_r128_private_t	*dev_priv	= dev->dev_private;
+	drm_device_dma_t	*dma		= dev->dma;
+	int			 retcode	= 0;
+	const int		 zero		= 0;
+	unsigned long		 virtual;
+	unsigned long		 address;
+	drm_buf_map_t		 request;
+	int			 i;
+
+	if (!dma || !dev_priv || dev_priv->is_pci) return -EINVAL;
+
+	DRM_DEBUG("\n");
+
+	spin_lock(&dev->count_lock);
+	if (atomic_read(&dev->buf_alloc)) {
+		spin_unlock(&dev->count_lock);
+		return -EBUSY;
+	}
+	++dev->buf_use;		/* Can't allocate more after this call */
+	spin_unlock(&dev->count_lock);
+
+	copy_from_user_ret(&request,
+			   (drm_buf_map_t *)arg,
+			   sizeof(request),
+			   -EFAULT);
+
+	if (request.count >= dma->buf_count) {
+		if (dma->flags & _DRM_DMA_USE_AGP) {
+			drm_map_t *map;
+
+			map = dev_priv->agp_vertbufs;
+			if (!map) {
+				retcode = -EINVAL;
+				goto done;
+			}
+
+			down(&current->mm->mmap_sem);
+			virtual = do_mmap(filp, 0, map->size, 
+					  PROT_READ|PROT_WRITE,
+					  MAP_SHARED, 
+					  (unsigned long)map->offset);
+			up(&current->mm->mmap_sem);
+		} else {
+			down(&current->mm->mmap_sem);
+			virtual = do_mmap(filp, 0, dma->byte_count,
+					  PROT_READ|PROT_WRITE, MAP_SHARED, 0);
+			up(&current->mm->mmap_sem);
+		}
+		if (virtual > -1024UL) {
+			/* Real error */
+			retcode = (signed long)virtual;
+			goto done;
+		}
+		request.virtual = (void *)virtual;
+
+		for (i = 0; i < dma->buf_count; i++) {
+			if (copy_to_user(&request.list[i].idx,
+					 &dma->buflist[i]->idx,
+					 sizeof(request.list[0].idx))) {
+				retcode = -EFAULT;
+				goto done;
+			}
+			if (copy_to_user(&request.list[i].total,
+					 &dma->buflist[i]->total,
+					 sizeof(request.list[0].total))) {
+				retcode = -EFAULT;
+				goto done;
+			}
+			if (copy_to_user(&request.list[i].used,
+					 &zero,
+					 sizeof(zero))) {
+				retcode = -EFAULT;
+				goto done;
+			}
+			address = virtual + dma->buflist[i]->offset;
+			if (copy_to_user(&request.list[i].address,
+					 &address,
+					 sizeof(address))) {
+				retcode = -EFAULT;
+				goto done;
+			}
+		}
+	}
+ done:
+	request.count = dma->buf_count;
+	DRM_DEBUG("%d buffers, retcode = %d\n", request.count, retcode);
+
+	copy_to_user_ret((drm_buf_map_t *)arg,
+			 &request,
+			 sizeof(request),
+			 -EFAULT);
+
+	return retcode;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_context.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_context.c
new file mode 100644
index 000000000..d288fd284
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_context.c
@@ -0,0 +1,214 @@
+/* r128_context.c -- IOCTLs for r128 contexts -*- linux-c -*-
+ * Created: Mon Dec 13 09:51:35 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999, 2000 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * Author: Rickard E. (Rik) Faith <faith@precisioninsight.com>
+ *
+ * $XFree86$
+ *
+ */
+
+#include <linux/sched.h>
+
+#define __NO_VERSION__
+#include "drmP.h"
+#include "r128_drv.h"
+
+extern drm_ctx_t r128_res_ctx;
+
+static int r128_alloc_queue(drm_device_t *dev)
+{
+#if 0
+	static int context = 0;
+#endif
+
+	return drm_ctxbitmap_next(dev);
+}
+
+int r128_context_switch(drm_device_t *dev, int old, int new)
+{
+        char        buf[64];
+
+        atomic_inc(&dev->total_ctx);
+
+        if (test_and_set_bit(0, &dev->context_flag)) {
+                DRM_ERROR("Reentering -- FIXME\n");
+                return -EBUSY;
+        }
+
+#if DRM_DMA_HISTOGRAM
+        dev->ctx_start = get_cycles();
+#endif
+        
+        DRM_DEBUG("Context switch from %d to %d\n", old, new);
+
+        if (new == dev->last_context) {
+                clear_bit(0, &dev->context_flag);
+                return 0;
+        }
+        
+        if (drm_flags & DRM_FLAG_NOCTX) {
+                r128_context_switch_complete(dev, new);
+        } else {
+                sprintf(buf, "C %d %d\n", old, new);
+                drm_write_string(dev, buf);
+        }
+        
+        return 0;
+}
+
+int r128_context_switch_complete(drm_device_t *dev, int new)
+{
+        dev->last_context = new;  /* PRE/POST: This is the _only_ writer. */
+        dev->last_switch  = jiffies;
+        
+        if (!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+                DRM_ERROR("Lock isn't held after context switch\n");
+        }
+
+				/* If a context switch is ever initiated
+                                   when the kernel holds the lock, release
+                                   that lock here. */
+#if DRM_DMA_HISTOGRAM
+        atomic_inc(&dev->histo.ctx[drm_histogram_slot(get_cycles()
+                                                      - dev->ctx_start)]);
+                   
+#endif
+        clear_bit(0, &dev->context_flag);
+        wake_up(&dev->context_wait);
+        
+        return 0;
+}
+
+
+int r128_resctx(struct inode *inode, struct file *filp, unsigned int cmd,
+	       unsigned long arg)
+{
+	drm_ctx_res_t	res;
+	drm_ctx_t	ctx;
+	int		i;
+
+	DRM_DEBUG("%d\n", DRM_RESERVED_CONTEXTS);
+	copy_from_user_ret(&res, (drm_ctx_res_t *)arg, sizeof(res), -EFAULT);
+	if (res.count >= DRM_RESERVED_CONTEXTS) {
+		memset(&ctx, 0, sizeof(ctx));
+		for (i = 0; i < DRM_RESERVED_CONTEXTS; i++) {
+			ctx.handle = i;
+			copy_to_user_ret(&res.contexts[i],
+					 &i,
+					 sizeof(i),
+					 -EFAULT);
+		}
+	}
+	res.count = DRM_RESERVED_CONTEXTS;
+	copy_to_user_ret((drm_ctx_res_t *)arg, &res, sizeof(res), -EFAULT);
+	return 0;
+}
+
+
+int r128_addctx(struct inode *inode, struct file *filp, unsigned int cmd,
+	       unsigned long arg)
+{
+	drm_file_t	*priv	= filp->private_data;
+	drm_device_t	*dev	= priv->dev;
+	drm_ctx_t	ctx;
+
+	copy_from_user_ret(&ctx, (drm_ctx_t *)arg, sizeof(ctx), -EFAULT);
+	if ((ctx.handle = r128_alloc_queue(dev)) == DRM_KERNEL_CONTEXT) {
+				/* Skip kernel's context and get a new one. */
+		ctx.handle = r128_alloc_queue(dev);
+	}
+	DRM_DEBUG("%d\n", ctx.handle);
+	if (ctx.handle == -1) {
+		DRM_DEBUG("Not enough free contexts.\n");
+				/* Should this return -EBUSY instead? */
+		return -ENOMEM;
+	}
+
+	copy_to_user_ret((drm_ctx_t *)arg, &ctx, sizeof(ctx), -EFAULT);
+	return 0;
+}
+
+int r128_modctx(struct inode *inode, struct file *filp, unsigned int cmd,
+	unsigned long arg)
+{
+	drm_ctx_t ctx;
+
+	copy_from_user_ret(&ctx, (drm_ctx_t*)arg, sizeof(ctx), -EFAULT);
+	if (ctx.flags==_DRM_CONTEXT_PRESERVED)
+		r128_res_ctx.handle=ctx.handle;
+	return 0;
+}
+
+int r128_getctx(struct inode *inode, struct file *filp, unsigned int cmd,
+	unsigned long arg)
+{
+	drm_ctx_t ctx;
+
+	copy_from_user_ret(&ctx, (drm_ctx_t*)arg, sizeof(ctx), -EFAULT);
+	/* This is 0, because we don't hanlde any context flags */
+	ctx.flags = 0;
+	copy_to_user_ret((drm_ctx_t*)arg, &ctx, sizeof(ctx), -EFAULT);
+	return 0;
+}
+
+int r128_switchctx(struct inode *inode, struct file *filp, unsigned int cmd,
+		   unsigned long arg)
+{
+	drm_file_t	*priv	= filp->private_data;
+	drm_device_t	*dev	= priv->dev;
+	drm_ctx_t	ctx;
+
+	copy_from_user_ret(&ctx, (drm_ctx_t *)arg, sizeof(ctx), -EFAULT);
+	DRM_DEBUG("%d\n", ctx.handle);
+	return r128_context_switch(dev, dev->last_context, ctx.handle);
+}
+
+int r128_newctx(struct inode *inode, struct file *filp, unsigned int cmd,
+		unsigned long arg)
+{
+	drm_file_t	*priv	= filp->private_data;
+	drm_device_t	*dev	= priv->dev;
+	drm_ctx_t	ctx;
+
+	copy_from_user_ret(&ctx, (drm_ctx_t *)arg, sizeof(ctx), -EFAULT);
+	DRM_DEBUG("%d\n", ctx.handle);
+	r128_context_switch_complete(dev, ctx.handle);
+
+	return 0;
+}
+
+int r128_rmctx(struct inode *inode, struct file *filp, unsigned int cmd,
+	       unsigned long arg)
+{
+	drm_file_t	*priv	= filp->private_data;
+	drm_device_t	*dev	= priv->dev;
+	drm_ctx_t	ctx;
+
+	copy_from_user_ret(&ctx, (drm_ctx_t *)arg, sizeof(ctx), -EFAULT);
+	DRM_DEBUG("%d\n", ctx.handle);
+	drm_ctxbitmap_free(dev, ctx.handle);
+
+	return 0;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_dma.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_dma.c
new file mode 100644
index 000000000..860c41885
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_dma.c
@@ -0,0 +1,908 @@
+/* r128_drv.c -- ATI Rage 128 driver -*- linux-c -*-
+ * Created: Wed Apr  5 19:24:19 2000 by kevin@precisioninsight.com
+ *
+ * Copyright 2000 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * Authors: Kevin E. Martin <kevin@precisioninsight.com>
+ * 
+ * $XFree86$
+ *
+ */
+
+#define __NO_VERSION__
+#include "drmP.h"
+#include "r128_drv.h"
+
+#include <linux/interrupt.h>	/* For task queue support */
+#include <linux/delay.h>
+
+
+
+#define DO_REMAP(_m) (_m)->handle = drm_ioremap((_m)->offset, (_m)->size)
+
+#define DO_REMAPFREE(_m)                                                    \
+	do {                                                                \
+		if ((_m)->handle && (_m)->size)                             \
+			drm_ioremapfree((_m)->handle, (_m)->size);          \
+	} while (0)
+
+#define DO_FIND_MAP(_m, _o)                                                 \
+	do {                                                                \
+		int _i;                                                     \
+		for (_i = 0; _i < dev->map_count; _i++) {                   \
+			if (dev->maplist[_i]->offset == _o) {               \
+				_m = dev->maplist[_i];                      \
+				break;                                      \
+			}                                                   \
+		}                                                           \
+	} while (0)
+
+
+#define R128_MAX_VBUF_AGE	0x10000000
+#define R128_VB_AGE_REG		R128_GUI_SCRATCH_REG0
+
+int R128_READ_PLL(drm_device_t *dev, int addr)
+{
+	drm_r128_private_t *dev_priv = dev->dev_private;
+
+	R128_WRITE8(R128_CLOCK_CNTL_INDEX, addr & 0x1f);
+	return R128_READ(R128_CLOCK_CNTL_DATA);
+}
+
+static void r128_flush_write_combine(void)
+{
+	int xchangeDummy;
+
+	__asm__ volatile("push %%eax ;"
+			 "xchg %%eax, %0 ;"
+			 "pop %%eax" : : "m" (xchangeDummy));
+	__asm__ volatile("push %%eax ;"
+			 "push %%ebx ;"
+			 "push %%ecx ;"
+			 "push %%edx ;"
+			 "movl $0,%%eax ;"
+			 "cpuid ;"
+			 "pop %%edx ;"
+			 "pop %%ecx ;"
+			 "pop %%ebx ;"
+			 "pop %%eax" : /* no outputs */ :  /* no inputs */ );
+}
+
+static int r128_do_cleanup_cce(drm_device_t *dev)
+{
+	if (dev->dev_private) {
+		drm_r128_private_t *dev_priv = dev->dev_private;
+
+		if (!dev_priv->is_pci) {
+			DO_REMAPFREE(dev_priv->agp_ring);
+			DO_REMAPFREE(dev_priv->agp_read_ptr);
+			DO_REMAPFREE(dev_priv->agp_vertbufs);
+			DO_REMAPFREE(dev_priv->agp_indbufs);
+			DO_REMAPFREE(dev_priv->agp_textures);
+		}
+
+		drm_free(dev->dev_private, sizeof(drm_r128_private_t),
+			 DRM_MEM_DRIVER);
+		dev->dev_private = NULL;
+	}
+
+	return 0;
+}
+
+static int r128_do_init_cce(drm_device_t *dev, drm_r128_init_t *init)
+{
+	drm_r128_private_t *dev_priv;
+        int                 i;
+
+	dev_priv = drm_alloc(sizeof(drm_r128_private_t), DRM_MEM_DRIVER);
+	if (dev_priv == NULL) return -ENOMEM;
+	dev->dev_private = (void *)dev_priv;
+
+	memset(dev_priv, 0, sizeof(drm_r128_private_t));
+
+	dev_priv->is_pci         = init->is_pci;
+
+	dev_priv->usec_timeout   = init->usec_timeout;
+	if (dev_priv->usec_timeout < 1 ||
+	    dev_priv->usec_timeout > R128_MAX_USEC_TIMEOUT) {
+		drm_free(dev_priv, sizeof(*dev_priv), DRM_MEM_DRIVER);
+		dev->dev_private = NULL;
+		return -EINVAL;
+	}
+
+	dev_priv->cce_mode       = init->cce_mode;
+	dev_priv->cce_fifo_size  = init->cce_fifo_size;
+	dev_priv->cce_is_bm_mode =
+		((init->cce_mode == R128_PM4_192BM) ||
+		 (init->cce_mode == R128_PM4_128BM_64INDBM) ||
+		 (init->cce_mode == R128_PM4_64BM_128INDBM) ||
+		 (init->cce_mode == R128_PM4_64BM_64VCBM_64INDBM));
+	dev_priv->cce_secure     = init->cce_secure;
+
+	if (dev_priv->cce_is_bm_mode && dev_priv->is_pci) {
+		drm_free(dev_priv, sizeof(*dev_priv), DRM_MEM_DRIVER);
+		dev->dev_private = NULL;
+		return -EINVAL;
+	}
+
+	for (i = 0; i < dev->map_count; i++) {
+		if (dev->maplist[i]->type == _DRM_SHM) {
+			dev_priv->sarea = dev->maplist[i];
+			break;
+		}
+	}
+
+	DO_FIND_MAP(dev_priv->fb,           init->fb_offset);
+	if (!dev_priv->is_pci) {
+		DO_FIND_MAP(dev_priv->agp_ring,     init->agp_ring_offset);
+		DO_FIND_MAP(dev_priv->agp_read_ptr, init->agp_read_ptr_offset);
+		DO_FIND_MAP(dev_priv->agp_vertbufs, init->agp_vertbufs_offset);
+		DO_FIND_MAP(dev_priv->agp_indbufs,  init->agp_indbufs_offset);
+		DO_FIND_MAP(dev_priv->agp_textures, init->agp_textures_offset);
+	}
+	DO_FIND_MAP(dev_priv->mmio,         init->mmio_offset);
+
+	dev_priv->sarea_priv =
+		(drm_r128_sarea_t *)((u8 *)dev_priv->sarea->handle +
+				     init->sarea_priv_offset);
+
+	if (!dev_priv->is_pci) {
+		DO_REMAP(dev_priv->agp_ring);
+		DO_REMAP(dev_priv->agp_read_ptr);
+		DO_REMAP(dev_priv->agp_vertbufs);
+#if 0
+		DO_REMAP(dev_priv->agp_indirectbufs);
+		DO_REMAP(dev_priv->agp_textures);
+#endif
+
+		dev_priv->ring_size     = init->ring_size;
+		dev_priv->ring_sizel2qw = drm_order(init->ring_size/8);
+		dev_priv->ring_entries  = init->ring_size/sizeof(u32);
+		dev_priv->ring_read_ptr = ((__volatile__ u32 *)
+					   dev_priv->agp_read_ptr->handle);
+		dev_priv->ring_start    = (u32 *)dev_priv->agp_ring->handle;
+		dev_priv->ring_end      = ((u32 *)dev_priv->agp_ring->handle
+					   + dev_priv->ring_entries);
+	}
+
+	dev_priv->submit_age    = 0;
+	R128_WRITE(R128_VB_AGE_REG, dev_priv->submit_age);
+
+	return 0;
+}
+
+int r128_init_cce(struct inode *inode, struct file *filp,
+		  unsigned int cmd, unsigned long arg)
+{
+        drm_file_t        *priv   = filp->private_data;
+        drm_device_t      *dev    = priv->dev;
+	drm_r128_init_t    init;
+
+	copy_from_user_ret(&init, (drm_r128_init_t *)arg, sizeof(init),
+			   -EFAULT);
+
+	switch (init.func) {
+	case R128_INIT_CCE:
+		return r128_do_init_cce(dev, &init);
+	case R128_CLEANUP_CCE:
+		return r128_do_cleanup_cce(dev);
+	}
+
+	return -EINVAL;
+}
+
+static void r128_mark_vertbufs_done(drm_device_t *dev)
+{
+	drm_device_dma_t   *dma      = dev->dma;
+	int                 i;
+
+	for (i = 0; i < dma->buf_count; i++) {
+		drm_buf_t           *buf      = dma->buflist[i];
+		drm_r128_buf_priv_t *buf_priv = buf->dev_private;
+		buf_priv->age = 0;
+	}
+}
+
+static int r128_do_pixcache_flush(drm_device_t *dev)
+{
+	drm_r128_private_t *dev_priv = dev->dev_private;
+	u32                 tmp;
+	int                 i;
+
+	tmp = R128_READ(R128_PC_NGUI_CTLSTAT) | R128_PC_FLUSH_ALL;
+	R128_WRITE(R128_PC_NGUI_CTLSTAT, tmp);
+
+	for (i = 0; i < dev_priv->usec_timeout; i++) {
+		if (!(R128_READ(R128_PC_NGUI_CTLSTAT) & R128_PC_BUSY))
+			return 0;
+		udelay(1);
+	}
+
+	return -EBUSY;
+}
+
+static int r128_do_wait_for_fifo(drm_device_t *dev, int entries)
+{
+	drm_r128_private_t *dev_priv = dev->dev_private;
+	int                 i;
+
+	for (i = 0; i < dev_priv->usec_timeout; i++) {
+		int slots = R128_READ(R128_GUI_STAT) & R128_GUI_FIFOCNT_MASK;
+		if (slots >= entries) return 0;
+		udelay(1);
+	}
+	return -EBUSY;
+}
+
+static int r128_do_wait_for_idle(drm_device_t *dev)
+{
+	drm_r128_private_t *dev_priv = dev->dev_private;
+	int                 i, ret;
+
+	if (!(ret = r128_do_wait_for_fifo(dev, 64))) return ret;
+
+	for (i = 0; i < dev_priv->usec_timeout; i++) {
+		if (!(R128_READ(R128_GUI_STAT) & R128_GUI_ACTIVE)) {
+			(void)r128_do_pixcache_flush(dev);
+			return 0;
+		}
+		udelay(1);
+	}
+	return -EBUSY;
+}
+
+int r128_do_engine_reset(drm_device_t *dev)
+{
+	drm_r128_private_t *dev_priv = dev->dev_private;
+	u32                 clock_cntl_index, mclk_cntl, gen_reset_cntl;
+
+	(void)r128_do_pixcache_flush(dev);
+
+	clock_cntl_index = R128_READ(R128_CLOCK_CNTL_INDEX);
+	mclk_cntl        = R128_READ_PLL(dev, R128_MCLK_CNTL);
+
+	R128_WRITE_PLL(R128_MCLK_CNTL,
+		       mclk_cntl | R128_FORCE_GCP | R128_FORCE_PIPE3D_CPP);
+
+	gen_reset_cntl   = R128_READ(R128_GEN_RESET_CNTL);
+
+	R128_WRITE(R128_GEN_RESET_CNTL, gen_reset_cntl | R128_SOFT_RESET_GUI);
+	(void)R128_READ(R128_GEN_RESET_CNTL);
+	R128_WRITE(R128_GEN_RESET_CNTL, gen_reset_cntl & ~R128_SOFT_RESET_GUI);
+	(void)R128_READ(R128_GEN_RESET_CNTL);
+
+	R128_WRITE_PLL(R128_MCLK_CNTL,    mclk_cntl);
+	R128_WRITE(R128_CLOCK_CNTL_INDEX, clock_cntl_index);
+	R128_WRITE(R128_GEN_RESET_CNTL,   gen_reset_cntl);
+
+	/* For CCE ring buffer only */
+	if (dev_priv->cce_is_bm_mode) {
+		R128_WRITE(R128_PM4_BUFFER_DL_WPTR, 0);
+		R128_WRITE(R128_PM4_BUFFER_DL_RPTR, 0);
+		*dev_priv->ring_read_ptr = 0;
+		dev_priv->sarea_priv->ring_write = 0;
+	}
+
+	/* Reset the CCE mode */
+	(void)r128_do_wait_for_idle(dev);
+	R128_WRITE(R128_PM4_BUFFER_CNTL,
+		   dev_priv->cce_mode | dev_priv->ring_sizel2qw);
+	(void)R128_READ(R128_PM4_BUFFER_ADDR); /* as per the sample code */
+	R128_WRITE(R128_PM4_MICRO_CNTL, R128_PM4_MICRO_FREERUN);
+
+	r128_mark_vertbufs_done(dev);
+	return 0;
+}
+
+int r128_eng_reset(struct inode *inode, struct file *filp,
+		   unsigned int cmd, unsigned long arg)
+{
+        drm_file_t        *priv   = filp->private_data;
+        drm_device_t      *dev    = priv->dev;
+
+	if (!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock) ||
+	    dev->lock.pid != current->pid) {
+		DRM_ERROR("r128_eng_reset called without holding the lock\n");
+		return -EINVAL;
+	}
+
+	return r128_do_engine_reset(dev);
+}
+
+static int r128_do_engine_flush(drm_device_t *dev)
+{
+	drm_r128_private_t *dev_priv = dev->dev_private;
+	u32                 tmp;
+
+	tmp = R128_READ(R128_PM4_BUFFER_DL_WPTR);
+	R128_WRITE(R128_PM4_BUFFER_DL_WPTR, tmp | R128_PM4_BUFFER_DL_DONE);
+
+	return 0;
+}
+
+int r128_eng_flush(struct inode *inode, struct file *filp,
+		   unsigned int cmd, unsigned long arg)
+{
+        drm_file_t        *priv   = filp->private_data;
+        drm_device_t      *dev    = priv->dev;
+
+	if (!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock) ||
+	    dev->lock.pid != current->pid) {
+		DRM_ERROR("r128_eng_flush called without holding the lock\n");
+		return -EINVAL;
+	}
+
+	return r128_do_engine_flush(dev);
+}
+
+static int r128_do_cce_wait_for_fifo(drm_device_t *dev, int entries)
+{
+	drm_r128_private_t *dev_priv = dev->dev_private;
+	int                 i;
+
+	for (i = 0; i < dev_priv->usec_timeout; i++) {
+		int slots = R128_READ(R128_PM4_STAT) & R128_PM4_FIFOCNT_MASK;
+		if (slots >= entries) return 0;
+		udelay(1);
+	}
+	return -EBUSY;
+}
+
+int r128_do_cce_wait_for_idle(drm_device_t *dev)
+{
+	drm_r128_private_t *dev_priv = dev->dev_private;
+	int                 i;
+
+	if (dev_priv->cce_is_bm_mode) {
+		for (i = 0; i < dev_priv->usec_timeout; i++) {
+			if (*dev_priv->ring_read_ptr == dev_priv->sarea_priv->ring_write) {
+				int pm4stat = R128_READ(R128_PM4_STAT);
+				if ((pm4stat & R128_PM4_FIFOCNT_MASK) >= dev_priv->cce_fifo_size &&
+				    !(pm4stat & (R128_PM4_BUSY | R128_PM4_GUI_ACTIVE))) {
+					return r128_do_pixcache_flush(dev);
+				}
+			}
+			udelay(1);
+		}
+		return -EBUSY;
+	} else {
+		int ret = r128_do_cce_wait_for_fifo(dev, dev_priv->cce_fifo_size);
+		if (ret < 0) return ret;
+
+		for (i = 0; i < dev_priv->usec_timeout; i++) {
+			int pm4stat = R128_READ(R128_PM4_STAT);
+			if (!(pm4stat & (R128_PM4_BUSY | R128_PM4_GUI_ACTIVE))) {
+				return r128_do_pixcache_flush(dev);
+			}
+			udelay(1);
+		}
+		return -EBUSY;
+	}
+}
+
+int r128_cce_idle(struct inode *inode, struct file *filp,
+		  unsigned int cmd, unsigned long arg)
+{
+        drm_file_t         *priv     = filp->private_data;
+        drm_device_t       *dev      = priv->dev;
+
+	if (!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock) ||
+	    dev->lock.pid != current->pid) {
+		DRM_ERROR("r128_wait_idle called without holding the lock\n");
+		return -EINVAL;
+	}
+
+	return r128_do_cce_wait_for_idle(dev);
+}
+
+static int r128_submit_packets_ring_secure(drm_device_t *dev,
+					   u32 *commands, int *count)
+{
+	drm_r128_private_t *dev_priv  = dev->dev_private;
+	int                 write     = dev_priv->sarea_priv->ring_write;
+	int                *write_ptr = dev_priv->ring_start + write;
+	int                 c         = *count;
+	u32                 tmp       = 0;
+	int                 psize     = 0;
+	int                 writing   = 1;
+	int                 timeout;
+
+	while (c > 0) {
+		tmp = *commands++;
+		if (!psize) {
+			writing = 1;
+
+			if ((tmp & R128_CCE_PACKET_MASK) == R128_CCE_PACKET0) {
+				if ((tmp & R128_CCE_PACKET0_REG_MASK) <= (0x1004 >> 2)) {
+					if ((tmp & R128_CCE_PACKET0_REG_MASK) !=
+					    (R128_PM4_VC_FPU_SETUP >> 2)) {
+						writing = 0;
+					}
+				}
+				psize = ((tmp & R128_CCE_PACKET_COUNT_MASK) >> 16) + 2;
+			} else if ((tmp & R128_CCE_PACKET_MASK) == R128_CCE_PACKET1) {
+				if ((tmp & R128_CCE_PACKET1_REG0_MASK) <= (0x1004 >> 2)) {
+					if ((tmp & R128_CCE_PACKET1_REG0_MASK) !=
+					    (R128_PM4_VC_FPU_SETUP >> 2)) {
+						writing = 0;
+					}
+				} else if ((tmp & R128_CCE_PACKET1_REG1_MASK) <=
+					   (0x1004 << 9)) {
+					if ((tmp & R128_CCE_PACKET1_REG1_MASK) !=
+					    (R128_PM4_VC_FPU_SETUP << 9)) {
+						writing = 0;
+					}
+				}
+				psize = 3;
+			} else {
+				psize = ((tmp & R128_CCE_PACKET_COUNT_MASK) >> 16) + 2;
+			}
+		}
+		psize--;
+
+		if (writing) {
+			write++;
+			*write_ptr++ = tmp;
+		}
+		if (write >= dev_priv->ring_entries) {
+			write     = 0;
+			write_ptr = dev_priv->ring_start;
+		}
+		timeout = 0;
+		while (write == *dev_priv->ring_read_ptr) {
+			(void)R128_READ(R128_PM4_BUFFER_DL_RPTR);
+			if (timeout++ >= dev_priv->usec_timeout)
+				return -EBUSY;
+			udelay(1);
+		}
+		c--;
+	}
+
+	if (write < 32)
+	    memcpy(dev_priv->ring_end,
+		   dev_priv->ring_start,
+		   write * sizeof(u32));
+
+	/* Make sure WC cache has been flushed */
+	r128_flush_write_combine();
+
+	dev_priv->sarea_priv->ring_write = write;
+	R128_WRITE(R128_PM4_BUFFER_DL_WPTR, write);
+
+	*count = 0;
+
+	return 0;
+}
+
+static int r128_submit_packets_pio_secure(drm_device_t *dev,
+					  u32 *commands, int *count)
+{
+	drm_r128_private_t *dev_priv = dev->dev_private;
+	u32                 tmp      = 0;
+	int                 psize    = 0;
+	int                 writing  = 1;
+	int                 addr     = R128_PM4_FIFO_DATA_EVEN;
+	int                 ret;
+
+	while (*count > 0) {
+		tmp = *commands++;
+		if (!psize) {
+			writing = 1;
+
+			if ((tmp & R128_CCE_PACKET_MASK) == R128_CCE_PACKET0) {
+				if ((tmp & R128_CCE_PACKET0_REG_MASK) <= (0x1004 >> 2)) {
+					if ((tmp & R128_CCE_PACKET0_REG_MASK) !=
+					    (R128_PM4_VC_FPU_SETUP >> 2)) {
+						writing = 0;
+					}
+				}
+				psize = ((tmp & R128_CCE_PACKET_COUNT_MASK) >> 16) + 2;
+			} else if ((tmp & R128_CCE_PACKET_MASK) == R128_CCE_PACKET1) {
+				if ((tmp & R128_CCE_PACKET1_REG0_MASK) <= (0x1004 >> 2)) {
+					if ((tmp & R128_CCE_PACKET1_REG0_MASK) !=
+					    (R128_PM4_VC_FPU_SETUP >> 2)) {
+						writing = 0;
+					}
+				} else if ((tmp & R128_CCE_PACKET1_REG1_MASK) <=
+					   (0x1004 << 9)) {
+					if ((tmp & R128_CCE_PACKET1_REG1_MASK) !=
+					    (R128_PM4_VC_FPU_SETUP << 9)) {
+						writing = 0;
+					}
+				}
+				psize = 3;
+			} else {
+				psize = ((tmp & R128_CCE_PACKET_COUNT_MASK) >> 16) + 2;
+			}
+		}
+		psize--;
+
+		if (writing) {
+			if ((ret = r128_do_cce_wait_for_fifo(dev, 1)) < 0)
+				return ret;
+			R128_WRITE(addr, tmp);
+			addr ^= 0x0004;
+		}
+
+		*count -= 1;
+	}
+
+	if (addr == R128_PM4_FIFO_DATA_ODD) {
+		if ((ret = r128_do_cce_wait_for_fifo(dev, 1)) < 0) return ret;
+		R128_WRITE(addr, R128_CCE_PACKET2);
+	}
+
+	return 0;
+}
+
+static int r128_submit_packets_ring(drm_device_t *dev,
+				    u32 *commands, int *count)
+{
+	drm_r128_private_t *dev_priv  = dev->dev_private;
+	int                 write     = dev_priv->sarea_priv->ring_write;
+	int                *write_ptr = dev_priv->ring_start + write;
+	int                 c         = *count;
+	int                 timeout;
+
+	while (c > 0) {
+		write++;
+		*write_ptr++ = *commands++;
+		if (write >= dev_priv->ring_entries) {
+			write     = 0;
+			write_ptr = dev_priv->ring_start;
+		}
+		timeout = 0;
+		while (write == *dev_priv->ring_read_ptr) {
+			(void)R128_READ(R128_PM4_BUFFER_DL_RPTR);
+			if (timeout++ >= dev_priv->usec_timeout)
+				return -EBUSY;
+			udelay(1);
+		}
+		c--;
+	}
+
+	if (write < 32)
+	    memcpy(dev_priv->ring_end,
+		   dev_priv->ring_start,
+		   write * sizeof(u32));
+
+	/* Make sure WC cache has been flushed */
+	r128_flush_write_combine();
+
+	dev_priv->sarea_priv->ring_write = write;
+	R128_WRITE(R128_PM4_BUFFER_DL_WPTR, write);
+
+	*count = 0;
+
+	return 0;
+}
+
+static int r128_submit_packets_pio(drm_device_t *dev,
+				   u32 *commands, int *count)
+{
+	drm_r128_private_t *dev_priv = dev->dev_private;
+	int                 ret;
+
+	while (*count > 1) {
+		if ((ret = r128_do_cce_wait_for_fifo(dev, 2)) < 0) return ret;
+		R128_WRITE(R128_PM4_FIFO_DATA_EVEN, *commands++);
+		R128_WRITE(R128_PM4_FIFO_DATA_ODD,  *commands++);
+		*count -= 2;
+	}
+
+	if (*count) {
+		if ((ret = r128_do_cce_wait_for_fifo(dev, 2)) < 0) return ret;
+		R128_WRITE(R128_PM4_FIFO_DATA_EVEN, *commands++);
+		R128_WRITE(R128_PM4_FIFO_DATA_ODD,  R128_CCE_PACKET2);
+		*count = 0;
+	}
+
+	return 0;
+}
+
+static int r128_do_submit_packets(drm_device_t *dev, u32 *buffer, int count)
+{
+	drm_r128_private_t *dev_priv = dev->dev_private;
+	int                 c = count;
+	int                 ret;
+
+	if (dev_priv->cce_is_bm_mode) {
+		int left = 0;
+
+		if (c >= dev_priv->ring_entries) {
+			c    = dev_priv->ring_entries-1;
+			left = count - c;
+		}
+
+		/* Since this is only used by the kernel we can use the
+                   insecure ring buffer submit packet routine */
+		ret = r128_submit_packets_ring(dev, buffer, &c);
+
+		c += left;
+	} else {
+		/* Since this is only used by the kernel we can use the
+                   insecure PIO submit packet routine */
+		ret = r128_submit_packets_pio(dev, buffer, &c);
+	}
+
+	if (ret < 0) return ret;
+	else         return c;
+}
+
+int r128_submit_pkt(struct inode *inode, struct file *filp,
+		    unsigned int cmd, unsigned long arg)
+{
+        drm_file_t         *priv     = filp->private_data;
+        drm_device_t       *dev      = priv->dev;
+	drm_r128_private_t *dev_priv = dev->dev_private;
+	drm_r128_packet_t   packet;
+	u32                *buffer;
+	int                 c;
+	int                 size;
+	int                 ret = 0;
+
+	if (!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock) ||
+	    dev->lock.pid != current->pid) {
+		DRM_ERROR("r128_submit_pkt called without holding the lock\n");
+		return -EINVAL;
+	}
+
+	copy_from_user_ret(&packet, (drm_r128_packet_t *)arg, sizeof(packet),
+			   -EFAULT);
+
+	c    = packet.count;
+	size = c * sizeof(*buffer);
+
+	if (dev_priv->cce_is_bm_mode) {
+		int left = 0;
+
+		if (c >= dev_priv->ring_entries) {
+			c    = dev_priv->ring_entries-1;
+			size = c * sizeof(*buffer);
+			left = packet.count - c;
+		}
+
+		if ((buffer = kmalloc(size, 0)) == NULL) return -ENOMEM;
+		copy_from_user_ret(buffer, packet.buffer, size, -EFAULT);
+
+		if (dev_priv->cce_secure)
+			ret = r128_submit_packets_ring_secure(dev, buffer, &c);
+		else
+			ret = r128_submit_packets_ring(dev, buffer, &c);
+
+		c += left;
+	} else {
+		if ((buffer = kmalloc(size, 0)) == NULL) return -ENOMEM;
+		copy_from_user_ret(buffer, packet.buffer, size, -EFAULT);
+
+		if (dev_priv->cce_secure)
+			ret = r128_submit_packets_pio_secure(dev, buffer, &c);
+		else
+			ret = r128_submit_packets_pio(dev, buffer, &c);
+	}
+
+	kfree(buffer);
+
+	packet.count = c;
+	copy_to_user_ret((drm_r128_packet_t *)arg, &packet, sizeof(packet),
+			 -EFAULT);
+
+	if (ret)        return ret;
+	else if (c > 0) return -EAGAIN;
+
+	return 0;
+}
+
+static int r128_send_vertbufs(drm_device_t *dev, drm_r128_vertex_t *v)
+{
+	drm_device_dma_t    *dma      = dev->dma;
+	drm_r128_private_t  *dev_priv = dev->dev_private;
+	drm_r128_buf_priv_t *buf_priv;
+	drm_buf_t           *buf;
+	int                  i, ret;
+	u32                  cce[2];
+
+	/* Make sure we have valid data */
+	for (i = 0; i < v->send_count; i++) {
+		int idx = v->send_indices[i];
+
+		if (idx < 0 || idx >= dma->buf_count) {
+			DRM_ERROR("Index %d (of %d max)\n",
+				  idx, dma->buf_count - 1);
+			return -EINVAL;
+		}
+		buf = dma->buflist[idx];
+		if (buf->pid != current->pid) {
+			DRM_ERROR("Process %d using buffer owned by %d\n",
+				  current->pid, buf->pid);
+			return -EINVAL;
+		}
+		if (buf->pending) {
+			DRM_ERROR("Sending pending buffer:"
+				  " buffer %d, offset %d\n",
+				  v->send_indices[i], i);
+			return -EINVAL;
+		}
+	}
+
+	/* Wait for idle, if we've wrapped to make sure that all pending
+           buffers have been processed */
+	if (dev_priv->submit_age == R128_MAX_VBUF_AGE) {
+		if ((ret = r128_do_cce_wait_for_idle(dev)) < 0) return ret;
+		dev_priv->submit_age = 0;
+		r128_mark_vertbufs_done(dev);
+	}
+
+	/* Make sure WC cache has been flushed (if in PIO mode) */
+	if (!dev_priv->cce_is_bm_mode) r128_flush_write_combine();
+
+	/* FIXME: Add support for sending vertex buffer to the CCE here
+	   instead of in client code.  The v->prim holds the primitive
+	   type that should be drawn.  Loop over the list buffers in
+	   send_indices[] and submit a packet for each VB.
+
+	   This will require us to loop over the clip rects here as
+	   well, which implies that we extend the kernel driver to allow
+	   cliprects to be stored here.  Note that the cliprects could
+	   possibly come from the X server instead of the client, but
+	   this will require additional changes to the DRI to allow for
+	   this optimization. */
+
+	/* Submit a CCE packet that writes submit_age to R128_VB_AGE_REG */
+	cce[0] = R128CCE0(R128_CCE_PACKET0, R128_VB_AGE_REG, 0);
+	cce[1] = dev_priv->submit_age;
+	if ((ret = r128_do_submit_packets(dev, cce, 2)) < 0) {
+		/* Until we add support for sending VBs to the CCE in
+		   this routine, we can recover from this error.  After
+		   we add that support, we won't be able to easily
+		   recover, so we will probably have to implement
+		   another mechanism for handling timeouts from packets
+		   submitted directly by the kernel. */
+		return ret;
+	}
+
+	/* Now that the submit packet request has succeeded, we can mark
+           the buffers as pending */
+	for (i = 0; i < v->send_count; i++) {
+		buf = dma->buflist[v->send_indices[i]];
+		buf->pending = 1;
+
+		buf_priv      = buf->dev_private;
+		buf_priv->age = dev_priv->submit_age;
+	}
+
+	dev_priv->submit_age++;
+
+	return 0;
+}
+
+static drm_buf_t *r128_freelist_get(drm_device_t *dev)
+{
+	drm_device_dma_t    *dma      = dev->dma;
+	drm_r128_private_t  *dev_priv = dev->dev_private;
+	drm_r128_buf_priv_t *buf_priv;
+	drm_buf_t           *buf;
+	int                  i, t;
+
+	/* FIXME: Optimize -- use freelist code */
+
+	for (i = 0; i < dma->buf_count; i++) {
+		buf = dma->buflist[i];
+		buf_priv = buf->dev_private;
+		if (buf->pid == 0) return buf;
+	}
+
+	for (t = 0; t < dev_priv->usec_timeout; t++) {
+		u32 done_age = R128_READ(R128_VB_AGE_REG);
+
+		for (i = 0; i < dma->buf_count; i++) {
+			buf = dma->buflist[i];
+			buf_priv = buf->dev_private;
+			if (buf->pending && buf_priv->age <= done_age) {
+				/* The buffer has been processed, so it
+                                   can now be used */
+				buf->pending = 0;
+				return buf;
+			}
+		}
+		udelay(1);
+	}
+
+	return NULL;
+}
+
+
+static int r128_get_vertbufs(drm_device_t *dev, drm_r128_vertex_t *v)
+{
+	drm_buf_t *buf;
+	int        i;
+
+	for (i = v->granted_count; i < v->request_count; i++) {
+		buf = r128_freelist_get(dev);
+		if (!buf) break;
+		buf->pid = current->pid;
+		copy_to_user_ret(&v->request_indices[i],
+				 &buf->idx,
+				 sizeof(buf->idx),
+				 -EFAULT);
+		copy_to_user_ret(&v->request_sizes[i],
+				 &buf->total,
+				 sizeof(buf->total),
+				 -EFAULT);
+		++v->granted_count;
+	}
+	return 0;
+}
+
+int r128_vertex_buf(struct inode *inode, struct file *filp, unsigned int cmd,
+		    unsigned long arg)
+{
+	drm_file_t	   *priv     = filp->private_data;
+	drm_device_t	   *dev	     = priv->dev;
+	drm_r128_private_t *dev_priv = dev->dev_private;
+	drm_device_dma_t   *dma	     = dev->dma;
+	int		    retcode  = 0;
+	drm_r128_vertex_t   v;
+
+	if (!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock) ||
+	    dev->lock.pid != current->pid) {
+		DRM_ERROR("r128_vertex_buf called without holding the lock\n");
+		return -EINVAL;
+	}
+
+	if (!dev_priv || dev_priv->is_pci) {
+		DRM_ERROR("r128_vertex_buf called with a PCI card\n");
+		return -EINVAL;
+	}
+
+	copy_from_user_ret(&v, (drm_r128_vertex_t *)arg, sizeof(v), -EFAULT);
+	DRM_DEBUG("%d: %d send, %d req\n",
+		  current->pid, v.send_count, v.request_count);
+
+	if (v.send_count < 0 || v.send_count > dma->buf_count) {
+		DRM_ERROR("Process %d trying to send %d buffers (of %d max)\n",
+			  current->pid, v.send_count, dma->buf_count);
+		return -EINVAL;
+	}
+	if (v.request_count < 0 || v.request_count > dma->buf_count) {
+		DRM_ERROR("Process %d trying to get %d buffers (of %d max)\n",
+			  current->pid, v.request_count, dma->buf_count);
+		return -EINVAL;
+	}
+
+	if (v.send_count) {
+		retcode = r128_send_vertbufs(dev, &v);
+	}
+
+	v.granted_count = 0;
+
+	if (!retcode && v.request_count) {
+		retcode = r128_get_vertbufs(dev, &v);
+	}
+
+	DRM_DEBUG("%d returning, granted = %d\n",
+		  current->pid, v.granted_count);
+	copy_to_user_ret((drm_r128_vertex_t *)arg, &v, sizeof(v), -EFAULT);
+
+	return retcode;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_drm.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_drm.h
new file mode 100644
index 000000000..fa90d72db
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_drm.h
@@ -0,0 +1,111 @@
+/* r128_drm.h -- Public header for the r128 driver -*- linux-c -*-
+ * Created: Wed Apr  5 19:24:19 2000 by kevin@precisioninsight.com
+ *
+ * Copyright 2000 Precision Insight, Inc., Cedar Park, Texas.
+ * All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ * $XFree86$
+ */
+
+#ifndef _R128_DRM_H_
+#define _R128_DRM_H_
+
+/* WARNING: If you change any of these defines, make sure to change the
+ * defines in the Xserver file (xf86drmR128.h)
+ */
+typedef struct drm_r128_init {
+	enum { 
+		R128_INIT_CCE    = 0x01,
+		R128_CLEANUP_CCE = 0x02
+	} func;
+	int sarea_priv_offset;
+	int is_pci;
+	int cce_mode;
+	int cce_fifo_size;
+	int cce_secure;
+	int ring_size;
+	int usec_timeout;
+
+	int fb_offset;
+	int agp_ring_offset;
+	int agp_read_ptr_offset;
+	int agp_vertbufs_offset;
+	int agp_indbufs_offset;
+	int agp_textures_offset;
+	int mmio_offset;
+} drm_r128_init_t;
+
+typedef struct drm_r128_packet {
+	unsigned long *buffer;
+	int            count;
+	int            flags;
+} drm_r128_packet_t;
+
+typedef enum drm_r128_prim {
+	_DRM_R128_PRIM_NONE		= 0x0001,
+	_DRM_R128_PRIM_POINT		= 0x0002,
+	_DRM_R128_PRIM_LINE		= 0x0004,
+	_DRM_R128_PRIM_POLY_LINE	= 0x0008,
+	_DRM_R128_PRIM_TRI_LIST		= 0x0010,
+	_DRM_R128_PRIM_TRI_FAN		= 0x0020,
+	_DRM_R128_PRIM_TRI_STRIP	= 0x0040,
+	_DRM_R128_PRIM_TRI_TYPE2	= 0x0080
+} drm_r128_prim_t;
+
+typedef struct drm_r128_vertex {
+				/* Indices here refer to the offset into
+				   buflist in drm_buf_get_t.  */
+	int		send_count;	  /* Number of buffers to send	    */
+	int		*send_indices;	  /* List of handles to buffers	    */
+	int		*send_sizes;	  /* Lengths of data to send	    */
+	drm_r128_prim_t	prim;		  /* Primitive type		    */
+	int		request_count;	  /* Number of buffers requested    */
+	int		*request_indices; /* Buffer information		    */
+	int		*request_sizes;
+	int		granted_count;	  /* Number of buffers granted	    */
+} drm_r128_vertex_t;
+
+/* WARNING: If you change any of these defines, make sure to change the
+ * defines in the Xserver file (r128_sarea.h)
+ */
+#define R128_LOCAL_TEX_HEAP       0
+#define R128_AGP_TEX_HEAP         1
+#define R128_NR_TEX_HEAPS         2
+#define R128_NR_TEX_REGIONS      64
+#define R128_LOG_TEX_GRANULARITY 16
+
+typedef struct drm_tex_region {
+	unsigned char next, prev;       
+	unsigned char in_use;   
+	int age;                        
+} drm_tex_region_t;
+
+typedef struct drm_r128_sarea {
+	drm_tex_region_t tex_list[R128_NR_TEX_HEAPS][R128_NR_TEX_REGIONS+1];
+	int              tex_age[R128_NR_TEX_HEAPS];
+	int              ctx_owner;
+	int              ring_write;
+} drm_r128_sarea_t;
+
+#endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_drv.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_drv.c
new file mode 100644
index 000000000..45ade1def
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_drv.c
@@ -0,0 +1,737 @@
+/* r128_drv.c -- ATI Rage 128 driver -*- linux-c -*-
+ * Created: Mon Dec 13 09:47:27 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999, 2000 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * Authors: Rickard E. (Rik) Faith <faith@precisioninsight.com>
+ *          Kevin E. Martin <kevin@precisioninsight.com>
+ * 
+ * $XFree86$
+ *
+ */
+
+#define EXPORT_SYMTAB
+#include "drmP.h"
+#include "r128_drv.h"
+EXPORT_SYMBOL(r128_init);
+EXPORT_SYMBOL(r128_cleanup);
+
+#define R128_NAME	 "r128"
+#define R128_DESC	 "r128"
+#define R128_DATE	 "20000422"
+#define R128_MAJOR	 0
+#define R128_MINOR	 0
+#define R128_PATCHLEVEL  5
+
+static drm_device_t	      r128_device;
+drm_ctx_t	              r128_res_ctx;
+
+static struct file_operations r128_fops = {
+	open:	 r128_open,
+	flush:	 drm_flush,
+	release: r128_release,
+	ioctl:	 r128_ioctl,
+	mmap:	 drm_mmap,
+	read:	 drm_read,
+	fasync:	 drm_fasync,
+	poll:	 drm_poll,
+};
+
+static struct miscdevice      r128_misc = {
+	minor: MISC_DYNAMIC_MINOR,
+	name:  R128_NAME,
+	fops:  &r128_fops,
+};
+
+static drm_ioctl_desc_t	      r128_ioctls[] = {
+	[DRM_IOCTL_NR(DRM_IOCTL_VERSION)]     = { r128_version,	   0, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_GET_UNIQUE)]  = { drm_getunique,   0, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_GET_MAGIC)]   = { drm_getmagic,	   0, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_IRQ_BUSID)]   = { drm_irq_busid,   0, 1 },
+
+	[DRM_IOCTL_NR(DRM_IOCTL_SET_UNIQUE)]  = { drm_setunique,   1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_BLOCK)]	      = { drm_block,	   1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_UNBLOCK)]     = { drm_unblock,	   1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_AUTH_MAGIC)]  = { drm_authmagic,   1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_ADD_MAP)]     = { drm_addmap,	   1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_ADD_BUFS)]    = { r128_addbufs,	   1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_MARK_BUFS)]   = { drm_markbufs,    1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_INFO_BUFS)]   = { drm_infobufs,    1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_MAP_BUFS)]    = { r128_mapbufs,	   1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_FREE_BUFS)]   = { drm_freebufs,    1, 0 },
+
+	[DRM_IOCTL_NR(DRM_IOCTL_ADD_CTX)]     = { r128_addctx,	   1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_RM_CTX)]      = { r128_rmctx,	   1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_MOD_CTX)]     = { r128_modctx,	   1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_GET_CTX)]     = { r128_getctx,	   1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_SWITCH_CTX)]  = { r128_switchctx,  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_NEW_CTX)]     = { r128_newctx,	   1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_RES_CTX)]     = { r128_resctx,	   1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_ADD_DRAW)]    = { drm_adddraw,	   1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_RM_DRAW)]     = { drm_rmdraw,	   1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_LOCK)]	      = { r128_lock,	   1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_UNLOCK)]      = { r128_unlock,	   1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_FINISH)]      = { drm_finish,	   1, 0 },
+
+#ifdef DRM_AGP
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_ACQUIRE)] = { drm_agp_acquire, 1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_RELEASE)] = { drm_agp_release, 1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_ENABLE)]  = { drm_agp_enable,  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_INFO)]    = { drm_agp_info,    1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_ALLOC)]   = { drm_agp_alloc,   1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_FREE)]    = { drm_agp_free,    1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_BIND)]    = { drm_agp_bind,    1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_UNBIND)]  = { drm_agp_unbind,  1, 1 },
+#endif
+
+	[DRM_IOCTL_NR(DRM_IOCTL_R128_INIT)]   = { r128_init_cce,   1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_R128_RESET)]  = { r128_eng_reset,  1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_R128_FLUSH)]  = { r128_eng_flush,  1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_R128_PACKET)] = { r128_submit_pkt, 1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_R128_CCEIDL)] = { r128_cce_idle,   1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_R128_VERTEX)] = { r128_vertex_buf, 1, 0 },
+};
+#define R128_IOCTL_COUNT DRM_ARRAY_SIZE(r128_ioctls)
+
+#ifdef MODULE
+int			      init_module(void);
+void			      cleanup_module(void);
+static char		      *r128 = NULL;
+
+MODULE_AUTHOR("Precision Insight, Inc., Cedar Park, Texas.");
+MODULE_DESCRIPTION("r128");
+MODULE_PARM(r128, "s");
+
+/* init_module is called when insmod is used to load the module */
+
+int init_module(void)
+{
+	return r128_init();
+}
+
+/* cleanup_module is called when rmmod is used to unload the module */
+
+void cleanup_module(void)
+{
+	r128_cleanup();
+}
+#endif
+
+#ifndef MODULE
+/* r128_setup is called by the kernel to parse command-line options passed
+ * via the boot-loader (e.g., LILO).  It calls the insmod option routine,
+ * drm_parse_drm.
+ *
+ * This is not currently supported, since it requires changes to
+ * linux/init/main.c. */
+ 
+
+void __init r128_setup(char *str, int *ints)
+{
+	if (ints[0] != 0) {
+		DRM_ERROR("Illegal command line format, ignored\n");
+		return;
+	}
+	drm_parse_options(str);
+}
+#endif
+
+static int r128_setup(drm_device_t *dev)
+{
+	int i;
+	
+	atomic_set(&dev->ioctl_count, 0);
+	atomic_set(&dev->vma_count, 0);
+	dev->buf_use	  = 0;
+	atomic_set(&dev->buf_alloc, 0);
+
+	drm_dma_setup(dev);
+
+	atomic_set(&dev->total_open, 0);
+	atomic_set(&dev->total_close, 0);
+	atomic_set(&dev->total_ioctl, 0);
+	atomic_set(&dev->total_irq, 0);
+	atomic_set(&dev->total_ctx, 0);
+	atomic_set(&dev->total_locks, 0);
+	atomic_set(&dev->total_unlocks, 0);
+	atomic_set(&dev->total_contends, 0);
+	atomic_set(&dev->total_sleeps, 0);
+
+	for (i = 0; i < DRM_HASH_SIZE; i++) {
+		dev->magiclist[i].head = NULL;
+		dev->magiclist[i].tail = NULL;
+	}
+	dev->maplist	    = NULL;
+	dev->map_count	    = 0;
+	dev->vmalist	    = NULL;
+	dev->lock.hw_lock   = NULL;
+	init_waitqueue_head(&dev->lock.lock_queue);
+	dev->queue_count    = 0;
+	dev->queue_reserved = 0;
+	dev->queue_slots    = 0;
+	dev->queuelist	    = NULL;
+	dev->irq	    = 0;
+	dev->context_flag   = 0;
+	dev->interrupt_flag = 0;
+	dev->dma_flag	    = 0;
+	dev->last_context   = 0;
+	dev->last_switch    = 0;
+	dev->last_checked   = 0;
+	init_timer(&dev->timer);
+	init_waitqueue_head(&dev->context_wait);
+
+	dev->ctx_start	    = 0;
+	dev->lck_start	    = 0;
+	
+	dev->buf_rp	  = dev->buf;
+	dev->buf_wp	  = dev->buf;
+	dev->buf_end	  = dev->buf + DRM_BSZ;
+	dev->buf_async	  = NULL;
+	init_waitqueue_head(&dev->buf_readers);
+	init_waitqueue_head(&dev->buf_writers);
+
+	r128_res_ctx.handle=-1;
+			
+	DRM_DEBUG("\n");
+			
+	/* The kernel's context could be created here, but is now created
+	   in drm_dma_enqueue.	This is more resource-efficient for
+	   hardware that does not do DMA, but may mean that
+	   drm_select_queue fails between the time the interrupt is
+	   initialized and the time the queues are initialized. */
+			
+	return 0;
+}
+
+
+static int r128_takedown(drm_device_t *dev)
+{
+	int		  i;
+	drm_magic_entry_t *pt, *next;
+	drm_map_t	  *map;
+	drm_vma_entry_t	  *vma, *vma_next;
+
+	DRM_DEBUG("\n");
+
+	down(&dev->struct_sem);
+	del_timer(&dev->timer);
+	
+	if (dev->devname) {
+		drm_free(dev->devname, strlen(dev->devname)+1, DRM_MEM_DRIVER);
+		dev->devname = NULL;
+	}
+	
+	if (dev->unique) {
+		drm_free(dev->unique, strlen(dev->unique)+1, DRM_MEM_DRIVER);
+		dev->unique = NULL;
+		dev->unique_len = 0;
+	}
+				/* Clear pid list */
+	for (i = 0; i < DRM_HASH_SIZE; i++) {
+		for (pt = dev->magiclist[i].head; pt; pt = next) {
+			next = pt->next;
+			drm_free(pt, sizeof(*pt), DRM_MEM_MAGIC);
+		}
+		dev->magiclist[i].head = dev->magiclist[i].tail = NULL;
+	}
+
+#ifdef DRM_AGP
+				/* Clear AGP information */
+	if (dev->agp) {
+		drm_agp_mem_t *entry;
+		drm_agp_mem_t *nexte;
+		
+				/* Remove AGP resources, but leave dev->agp
+                                   intact until r128_cleanup is called. */
+		for (entry = dev->agp->memory; entry; entry = nexte) {
+			nexte = entry->next;
+			if (entry->bound) drm_unbind_agp(entry->memory);
+			drm_free_agp(entry->memory, entry->pages);
+			drm_free(entry, sizeof(*entry), DRM_MEM_AGPLISTS);
+		}
+		dev->agp->memory = NULL;
+		
+		if (dev->agp->acquired && drm_agp.release)
+			(*drm_agp.release)();
+		
+		dev->agp->acquired = 0;
+		dev->agp->enabled  = 0;
+	}
+#endif
+	
+				/* Clear vma list (only built for debugging) */
+	if (dev->vmalist) {
+		for (vma = dev->vmalist; vma; vma = vma_next) {
+			vma_next = vma->next;
+			drm_free(vma, sizeof(*vma), DRM_MEM_VMAS);
+		}
+		dev->vmalist = NULL;
+	}
+	
+				/* Clear map area and mtrr information */
+	if (dev->maplist) {
+		for (i = 0; i < dev->map_count; i++) {
+			map = dev->maplist[i];
+			switch (map->type) {
+			case _DRM_REGISTERS:
+			case _DRM_FRAME_BUFFER:
+#ifdef CONFIG_MTRR
+				if (map->mtrr >= 0) {
+					int retcode;
+					retcode = mtrr_del(map->mtrr,
+							   map->offset,
+							   map->size);
+					DRM_DEBUG("mtrr_del = %d\n", retcode);
+				}
+#endif
+				drm_ioremapfree(map->handle, map->size);
+				break;
+			case _DRM_SHM:
+				drm_free_pages((unsigned long)map->handle,
+					       drm_order(map->size)
+					       - PAGE_SHIFT,
+					       DRM_MEM_SAREA);
+				break;
+			case _DRM_AGP:
+				/* Do nothing here, because this is all
+                                   handled in the AGP/GART driver. */
+				break;
+			}
+			drm_free(map, sizeof(*map), DRM_MEM_MAPS);
+		}
+		drm_free(dev->maplist,
+			 dev->map_count * sizeof(*dev->maplist),
+			 DRM_MEM_MAPS);
+		dev->maplist   = NULL;
+		dev->map_count = 0;
+	}
+	
+	drm_dma_takedown(dev);
+
+	dev->queue_count     = 0;
+	if (dev->lock.hw_lock) {
+		dev->lock.hw_lock    = NULL; /* SHM removed */
+		dev->lock.pid	     = 0;
+		wake_up_interruptible(&dev->lock.lock_queue);
+	}
+	up(&dev->struct_sem);
+	
+	return 0;
+}
+
+/* r128_init is called via init_module at module load time, or via
+ * linux/init/main.c (this is not currently supported). */
+
+int r128_init(void)
+{
+	int		      retcode;
+	drm_device_t	      *dev = &r128_device;
+
+	DRM_DEBUG("\n");
+
+	memset((void *)dev, 0, sizeof(*dev));
+	dev->count_lock	  = SPIN_LOCK_UNLOCKED;
+	sema_init(&dev->struct_sem, 1);
+	
+#ifdef MODULE
+	drm_parse_options(r128);
+#endif
+
+	if ((retcode = misc_register(&r128_misc))) {
+		DRM_ERROR("Cannot register \"%s\"\n", R128_NAME);
+		return retcode;
+	}
+	dev->device = MKDEV(MISC_MAJOR, r128_misc.minor);
+	dev->name   = R128_NAME;
+
+	drm_mem_init();
+	drm_proc_init(dev);
+
+#ifdef DRM_AGP
+	dev->agp    = drm_agp_init();
+
+#ifdef CONFIG_MTRR
+	dev->agp->agp_mtrr = mtrr_add(dev->agp->agp_info.aper_base,
+				      dev->agp->agp_info.aper_size*1024*1024,
+				      MTRR_TYPE_WRCOMB,
+				      1);
+#endif
+#endif
+
+	if((retcode = drm_ctxbitmap_init(dev))) {
+		DRM_ERROR("Cannot allocate memory for context bitmap.\n");
+		drm_proc_cleanup();
+		misc_deregister(&r128_misc);
+		r128_takedown(dev);
+		return retcode;
+	}
+
+	DRM_INFO("Initialized %s %d.%d.%d %s on minor %d\n",
+		 R128_NAME,
+		 R128_MAJOR,
+		 R128_MINOR,
+		 R128_PATCHLEVEL,
+		 R128_DATE,
+		 r128_misc.minor);
+
+	return 0;
+}
+
+/* r128_cleanup is called via cleanup_module at module unload time. */
+
+void r128_cleanup(void)
+{
+	drm_device_t	      *dev = &r128_device;
+
+	DRM_DEBUG("\n");
+	
+	drm_proc_cleanup();
+	if (misc_deregister(&r128_misc)) {
+		DRM_ERROR("Cannot unload module\n");
+	} else {
+		DRM_INFO("Module unloaded\n");
+	}
+	drm_ctxbitmap_cleanup(dev);
+	r128_takedown(dev);
+#ifdef DRM_AGP
+	if (dev->agp) {
+				/* FIXME -- free other information, too */
+		drm_free(dev->agp, sizeof(*dev->agp), DRM_MEM_AGPLISTS);
+		dev->agp = NULL;
+	}
+#endif
+}
+
+int r128_version(struct inode *inode, struct file *filp, unsigned int cmd,
+		  unsigned long arg)
+{
+	drm_version_t version;
+	int	      len;
+
+	copy_from_user_ret(&version,
+			   (drm_version_t *)arg,
+			   sizeof(version),
+			   -EFAULT);
+
+#define DRM_COPY(name,value)				     \
+	len = strlen(value);				     \
+	if (len > name##_len) len = name##_len;		     \
+	name##_len = strlen(value);			     \
+	if (len && name) {				     \
+		copy_to_user_ret(name, value, len, -EFAULT); \
+	}
+
+	version.version_major	   = R128_MAJOR;
+	version.version_minor	   = R128_MINOR;
+	version.version_patchlevel = R128_PATCHLEVEL;
+
+	DRM_COPY(version.name, R128_NAME);
+	DRM_COPY(version.date, R128_DATE);
+	DRM_COPY(version.desc, R128_DESC);
+
+	copy_to_user_ret((drm_version_t *)arg,
+			 &version,
+			 sizeof(version),
+			 -EFAULT);
+	return 0;
+}
+
+int r128_open(struct inode *inode, struct file *filp)
+{
+	drm_device_t  *dev    = &r128_device;
+	int	      retcode = 0;
+	
+	DRM_DEBUG("open_count = %d\n", dev->open_count);
+	if (!(retcode = drm_open_helper(inode, filp, dev))) {
+		MOD_INC_USE_COUNT;
+		atomic_inc(&dev->total_open);
+		spin_lock(&dev->count_lock);
+		if (!dev->open_count++) {
+			spin_unlock(&dev->count_lock);
+			return r128_setup(dev);
+		}
+		spin_unlock(&dev->count_lock);
+	}
+	return retcode;
+}
+
+int r128_release(struct inode *inode, struct file *filp)
+{
+	drm_file_t    *priv   = filp->private_data;
+	drm_device_t  *dev    = priv->dev;
+	int	      retcode = 0;
+
+	DRM_DEBUG("open_count = %d\n", dev->open_count);
+	if (!(retcode = drm_release(inode, filp))) {
+		MOD_DEC_USE_COUNT;
+		atomic_inc(&dev->total_close);
+		spin_lock(&dev->count_lock);
+		if (!--dev->open_count) {
+			if (atomic_read(&dev->ioctl_count) || dev->blocked) {
+				DRM_ERROR("Device busy: %d %d\n",
+					  atomic_read(&dev->ioctl_count),
+					  dev->blocked);
+				spin_unlock(&dev->count_lock);
+				return -EBUSY;
+			}
+			spin_unlock(&dev->count_lock);
+			return r128_takedown(dev);
+		}
+		spin_unlock(&dev->count_lock);
+	}
+	return retcode;
+}
+
+/* r128_ioctl is called whenever a process performs an ioctl on /dev/drm. */
+
+int r128_ioctl(struct inode *inode, struct file *filp, unsigned int cmd,
+		unsigned long arg)
+{
+	int		 nr	 = DRM_IOCTL_NR(cmd);
+	drm_file_t	 *priv	 = filp->private_data;
+	drm_device_t	 *dev	 = priv->dev;
+	int		 retcode = 0;
+	drm_ioctl_desc_t *ioctl;
+	drm_ioctl_t	 *func;
+
+	atomic_inc(&dev->ioctl_count);
+	atomic_inc(&dev->total_ioctl);
+	++priv->ioctl_count;
+	
+	DRM_DEBUG("pid = %d, cmd = 0x%02x, nr = 0x%02x, dev 0x%x, auth = %d\n",
+		  current->pid, cmd, nr, dev->device, priv->authenticated);
+
+	if (nr >= R128_IOCTL_COUNT) {
+		retcode = -EINVAL;
+	} else {
+		ioctl	  = &r128_ioctls[nr];
+		func	  = ioctl->func;
+
+		if (!func) {
+			DRM_DEBUG("no function\n");
+			retcode = -EINVAL;
+		} else if ((ioctl->root_only && !capable(CAP_SYS_ADMIN))
+			    || (ioctl->auth_needed && !priv->authenticated)) {
+			retcode = -EACCES;
+		} else {
+			retcode = (func)(inode, filp, cmd, arg);
+		}
+	}
+	
+	atomic_dec(&dev->ioctl_count);
+	return retcode;
+}
+
+int r128_lock(struct inode *inode, struct file *filp, unsigned int cmd,
+	      unsigned long arg)
+{
+        drm_file_t        *priv   = filp->private_data;
+        drm_device_t      *dev    = priv->dev;
+        DECLARE_WAITQUEUE(entry, current);
+        int               ret   = 0;
+        drm_lock_t        lock;
+#if DRM_DMA_HISTOGRAM
+        cycles_t          start;
+
+        dev->lck_start = start = get_cycles();
+#endif
+
+        copy_from_user_ret(&lock, (drm_lock_t *)arg, sizeof(lock), -EFAULT);
+
+        if (lock.context == DRM_KERNEL_CONTEXT) {
+                DRM_ERROR("Process %d using kernel context %d\n",
+                          current->pid, lock.context);
+                return -EINVAL;
+        }
+
+        DRM_DEBUG("%d (pid %d) requests lock (0x%08x), flags = 0x%08x\n",
+                  lock.context, current->pid, dev->lock.hw_lock->lock,
+                  lock.flags);
+
+#if 0
+				/* dev->queue_count == 0 right now for
+                                   r128.  FIXME? */
+        if (lock.context < 0 || lock.context >= dev->queue_count)
+                return -EINVAL;
+#endif
+        
+        if (!ret) {
+#if 0
+                if (_DRM_LOCKING_CONTEXT(dev->lock.hw_lock->lock)
+                    != lock.context) {
+                        long j = jiffies - dev->lock.lock_time;
+
+                        if (lock.context == r128_res_ctx.handle &&
+				j >= 0 && j < DRM_LOCK_SLICE) {
+                                /* Can't take lock if we just had it and
+                                   there is contention. */
+                                DRM_DEBUG("%d (pid %d) delayed j=%d dev=%d jiffies=%d\n",
+					lock.context, current->pid, j, 
+					dev->lock.lock_time, jiffies);
+                                current->state = TASK_INTERRUPTIBLE;
+				current->policy |= SCHED_YIELD;
+                                schedule_timeout(DRM_LOCK_SLICE-j);
+				DRM_DEBUG("jiffies=%d\n", jiffies);
+                        }
+                }
+#endif
+                add_wait_queue(&dev->lock.lock_queue, &entry);
+                for (;;) {
+                        if (!dev->lock.hw_lock) {
+                                /* Device has been unregistered */
+                                ret = -EINTR;
+                                break;
+                        }
+                        if (drm_lock_take(&dev->lock.hw_lock->lock,
+                                          lock.context)) {
+                                dev->lock.pid       = current->pid;
+                                dev->lock.lock_time = jiffies;
+                                atomic_inc(&dev->total_locks);
+                                break;  /* Got lock */
+                        }
+                        
+                                /* Contention */
+                        atomic_inc(&dev->total_sleeps);
+                        current->state = TASK_INTERRUPTIBLE;
+#if 1
+			current->policy |= SCHED_YIELD;
+#endif
+                        schedule();
+                        if (signal_pending(current)) {
+                                ret = -ERESTARTSYS;
+                                break;
+                        }
+                }
+                current->state = TASK_RUNNING;
+                remove_wait_queue(&dev->lock.lock_queue, &entry);
+        }
+
+#if 0
+	if (!ret && dev->last_context != lock.context &&
+		lock.context != r128_res_ctx.handle &&
+		dev->last_context != r128_res_ctx.handle) {
+		add_wait_queue(&dev->context_wait, &entry);
+	        current->state = TASK_INTERRUPTIBLE;
+                /* PRE: dev->last_context != lock.context */
+	        r128_context_switch(dev, dev->last_context, lock.context);
+		/* POST: we will wait for the context
+                   switch and will dispatch on a later call
+                   when dev->last_context == lock.context
+                   NOTE WE HOLD THE LOCK THROUGHOUT THIS
+                   TIME! */
+		current->policy |= SCHED_YIELD;
+	        schedule();
+	        current->state = TASK_RUNNING;
+	        remove_wait_queue(&dev->context_wait, &entry);
+	        if (signal_pending(current)) {
+	                ret = -EINTR;
+	        } else if (dev->last_context != lock.context) {
+			DRM_ERROR("Context mismatch: %d %d\n",
+                        	dev->last_context, lock.context);
+	        }
+	}
+#endif
+
+        if (!ret) {
+                if (lock.flags & _DRM_LOCK_READY) {
+				/* Wait for space in DMA/FIFO */
+		}
+                if (lock.flags & _DRM_LOCK_QUIESCENT) {
+				/* Make hardware quiescent */
+#if 0
+                        r128_quiescent(dev);
+#endif
+		}
+        }
+
+#if 0
+	DRM_ERROR("pid = %5d, old counter = %5ld\n", 
+		current->pid, current->counter);
+#endif
+	if (lock.context != r128_res_ctx.handle) {
+		current->counter = 5;
+		current->priority = DEF_PRIORITY/4;
+	}
+#if 0
+	while (current->counter > 25)
+		current->counter >>= 1; /* decrease time slice */
+	DRM_ERROR("pid = %5d, new counter = %5ld\n",
+		 current->pid, current->counter);
+#endif
+        DRM_DEBUG("%d %s\n", lock.context, ret ? "interrupted" : "has lock");
+
+#if DRM_DMA_HISTOGRAM
+        atomic_inc(&dev->histo.lacq[drm_histogram_slot(get_cycles() - start)]);
+#endif
+        
+        return ret;
+}
+
+
+int r128_unlock(struct inode *inode, struct file *filp, unsigned int cmd,
+		 unsigned long arg)
+{
+	drm_file_t	  *priv	  = filp->private_data;
+	drm_device_t	  *dev	  = priv->dev;
+	drm_lock_t	  lock;
+
+	copy_from_user_ret(&lock, (drm_lock_t *)arg, sizeof(lock), -EFAULT);
+	
+	if (lock.context == DRM_KERNEL_CONTEXT) {
+		DRM_ERROR("Process %d using kernel context %d\n",
+			  current->pid, lock.context);
+		return -EINVAL;
+	}
+
+	DRM_DEBUG("%d frees lock (%d holds)\n",
+		  lock.context,
+		  _DRM_LOCKING_CONTEXT(dev->lock.hw_lock->lock));
+	atomic_inc(&dev->total_unlocks);
+	if (_DRM_LOCK_IS_CONT(dev->lock.hw_lock->lock))
+		atomic_inc(&dev->total_contends);
+	drm_lock_transfer(dev, &dev->lock.hw_lock->lock, DRM_KERNEL_CONTEXT);
+				/* FIXME: Try to send data to card here */
+	if (!dev->context_flag) {
+		if (drm_lock_free(dev, &dev->lock.hw_lock->lock,
+				  DRM_KERNEL_CONTEXT)) {
+			DRM_ERROR("\n");
+		}
+	}
+
+#if 0
+	current->policy |= SCHED_YIELD;
+	current->state = TASK_INTERRUPTIBLE;
+	schedule_timeout(1000);
+#endif
+
+	if (lock.context != r128_res_ctx.handle) {
+		current->counter = 5;
+		current->priority = DEF_PRIORITY;
+	}
+#if 0
+	current->state = TASK_INTERRUPTIBLE;
+	schedule_timeout(10);
+#endif
+	
+	return 0;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_drv.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_drv.h
new file mode 100644
index 000000000..3b888c493
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/r128_drv.h
@@ -0,0 +1,226 @@
+/* r128_drv.h -- Private header for r128 driver -*- linux-c -*-
+ * Created: Mon Dec 13 09:51:11 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * Authors: Rickard E. (Rik) Faith <faith@precisioninsight.com>
+ *          Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ * $XFree86$
+ * 
+ */
+
+#ifndef _R128_DRV_H_
+#define _R128_DRV_H_
+
+typedef struct drm_r128_private {
+	int               is_pci;
+
+	int               cce_mode;
+	int               cce_fifo_size;
+	int               cce_is_bm_mode;
+	int               cce_secure;
+
+	drm_r128_sarea_t *sarea_priv;
+
+	__volatile__ u32 *ring_read_ptr;
+
+	u32              *ring_start;
+	u32              *ring_end;
+	int               ring_size;
+	int               ring_sizel2qw;
+	int               ring_entries;
+
+	int               submit_age;
+
+	int               usec_timeout;
+
+	drm_map_t        *sarea;
+	drm_map_t        *fb;
+	drm_map_t        *agp_ring;
+	drm_map_t        *agp_read_ptr;
+	drm_map_t        *agp_vertbufs;
+	drm_map_t        *agp_indbufs;
+	drm_map_t        *agp_textures;
+	drm_map_t        *mmio;
+} drm_r128_private_t;
+
+typedef struct drm_r128_buf_priv {
+	u32               age;
+} drm_r128_buf_priv_t;
+
+				/* r128_drv.c */
+extern int  r128_init(void);
+extern void r128_cleanup(void);
+extern int  r128_version(struct inode *inode, struct file *filp,
+			  unsigned int cmd, unsigned long arg);
+extern int  r128_open(struct inode *inode, struct file *filp);
+extern int  r128_release(struct inode *inode, struct file *filp);
+extern int  r128_ioctl(struct inode *inode, struct file *filp,
+			unsigned int cmd, unsigned long arg);
+extern int  r128_lock(struct inode *inode, struct file *filp,
+		       unsigned int cmd, unsigned long arg);
+extern int  r128_unlock(struct inode *inode, struct file *filp,
+			 unsigned int cmd, unsigned long arg);
+
+				/* r128_dma.c */
+extern int r128_init_cce(struct inode *inode, struct file *filp,
+			 unsigned int cmd, unsigned long arg);
+extern int r128_eng_reset(struct inode *inode, struct file *filp,
+			  unsigned int cmd, unsigned long arg);
+extern int r128_eng_flush(struct inode *inode, struct file *filp,
+			  unsigned int cmd, unsigned long arg);
+extern int r128_submit_pkt(struct inode *inode, struct file *filp,
+			   unsigned int cmd, unsigned long arg);
+extern int r128_cce_idle(struct inode *inode, struct file *filp,
+			 unsigned int cmd, unsigned long arg);
+extern int r128_vertex_buf(struct inode *inode, struct file *filp,
+			   unsigned int cmd, unsigned long arg);
+
+				/* r128_bufs.c */
+extern int r128_addbufs(struct inode *inode, struct file *filp,
+			unsigned int cmd, unsigned long arg);
+extern int r128_mapbufs(struct inode *inode, struct file *filp,
+			unsigned int cmd, unsigned long arg);
+
+				/* r128_context.c */
+extern int  r128_resctx(struct inode *inode, struct file *filp,
+			unsigned int cmd, unsigned long arg);
+extern int  r128_addctx(struct inode *inode, struct file *filp,
+		        unsigned int cmd, unsigned long arg);
+extern int  r128_modctx(struct inode *inode, struct file *filp,
+		        unsigned int cmd, unsigned long arg);
+extern int  r128_getctx(struct inode *inode, struct file *filp,
+		        unsigned int cmd, unsigned long arg);
+extern int  r128_switchctx(struct inode *inode, struct file *filp,
+			   unsigned int cmd, unsigned long arg);
+extern int  r128_newctx(struct inode *inode, struct file *filp,
+			unsigned int cmd, unsigned long arg);
+extern int  r128_rmctx(struct inode *inode, struct file *filp,
+		       unsigned int cmd, unsigned long arg);
+
+extern int  r128_context_switch(drm_device_t *dev, int old, int new);
+extern int  r128_context_switch_complete(drm_device_t *dev, int new);
+
+
+/* Register definitions, register access macros and drmAddMap constants
+ * for Rage 128 kernel driver.
+ */
+
+#define R128_PC_NGUI_CTLSTAT	0x0184
+#       define R128_PC_FLUSH_ALL	0x00ff
+#       define R128_PC_BUSY		(1 << 31)
+
+#define R128_CLOCK_CNTL_INDEX	0x0008
+#define R128_CLOCK_CNTL_DATA	0x000c
+#       define R128_PLL_WR_EN		(1 << 7)
+
+#define R128_MCLK_CNTL		0x000f
+#       define R128_FORCE_GCP		(1 << 16)
+#       define R128_FORCE_PIPE3D_CPP	(1 << 17)
+
+#define R128_GEN_RESET_CNTL	0x00f0
+#       define R128_SOFT_RESET_GUI	(1 <<  0)
+
+#define R128_PM4_BUFFER_CNTL	0x0704
+#       define R128_PM4_NONPM4			(0  << 28)
+#       define R128_PM4_192PIO			(1  << 28)
+#       define R128_PM4_192BM			(2  << 28)
+#       define R128_PM4_128PIO_64INDBM		(3  << 28)
+#       define R128_PM4_128BM_64INDBM		(4  << 28)
+#       define R128_PM4_64PIO_128INDBM		(5  << 28)
+#       define R128_PM4_64BM_128INDBM		(6  << 28)
+#       define R128_PM4_64PIO_64VCBM_64INDBM	(7  << 28)
+#       define R128_PM4_64BM_64VCBM_64INDBM	(8  << 28)
+#       define R128_PM4_64PIO_64VCPIO_64INDPIO	(15 << 28)
+
+
+#define R128_PM4_BUFFER_DL_RPTR	0x0710
+#define R128_PM4_BUFFER_DL_WPTR	0x0714
+#       define R128_PM4_BUFFER_DL_DONE		(1 << 31)
+
+#define R128_PM4_VC_FPU_SETUP	0x071c
+
+#define R128_PM4_STAT		0x07b8
+#       define R128_PM4_FIFOCNT_MASK		0x0fff
+#       define R128_PM4_BUSY			(1 << 16)
+#       define R128_PM4_GUI_ACTIVE		(1 << 31)
+
+#define R128_PM4_BUFFER_ADDR	0x07f0
+#define R128_PM4_MICRO_CNTL	0x07fc
+#       define R128_PM4_MICRO_FREERUN		(1 << 30)
+
+#define R128_PM4_FIFO_DATA_EVEN	0x1000
+#define R128_PM4_FIFO_DATA_ODD	0x1004
+
+#define R128_GUI_SCRATCH_REG0	0x15e0
+#define R128_GUI_SCRATCH_REG1	0x15e4
+#define R128_GUI_SCRATCH_REG2	0x15e8
+#define R128_GUI_SCRATCH_REG3	0x15ec
+#define R128_GUI_SCRATCH_REG4	0x15f0
+#define R128_GUI_SCRATCH_REG5	0x15f4
+
+#define R128_GUI_STAT		0x1740
+#       define R128_GUI_FIFOCNT_MASK		0x0fff
+#       define R128_GUI_ACTIVE			(1 << 31)
+
+
+/* CCE command packets */
+#define R128_CCE_PACKET0	0x00000000
+#define R128_CCE_PACKET1	0x40000000
+#define R128_CCE_PACKET2	0x80000000
+#       define R128_CCE_PACKET_MASK		0xC0000000
+#       define R128_CCE_PACKET_COUNT_MASK	0x3fff0000
+#       define R128_CCE_PACKET0_REG_MASK	0x000007ff
+#       define R128_CCE_PACKET1_REG0_MASK	0x000007ff
+#       define R128_CCE_PACKET1_REG1_MASK	0x003ff800
+
+
+#define R128_MAX_USEC_TIMEOUT	100000	/* 100 ms */
+
+
+#define R128_BASE(reg)		((u32)(dev_priv->mmio->handle))
+#define R128_ADDR(reg)		(R128_BASE(reg) + reg)
+
+#define R128_DEREF(reg)		*(__volatile__ int *)R128_ADDR(reg)
+#define R128_READ(reg)		R128_DEREF(reg)
+#define R128_WRITE(reg,val)	do { R128_DEREF(reg) = val; } while (0)
+
+#define R128_DEREF8(reg)	*(__volatile__ char *)R128_ADDR(reg)
+#define R128_READ8(reg)		R128_DEREF8(reg)
+#define R128_WRITE8(reg,val)	do { R128_DEREF8(reg) = val; } while (0)
+
+#define R128_WRITE_PLL(addr,val)                                              \
+do {                                                                          \
+	R128_WRITE8(R128_CLOCK_CNTL_INDEX, ((addr) & 0x1f) | R128_PLL_WR_EN); \
+	R128_WRITE(R128_CLOCK_CNTL_DATA, (val));                              \
+} while (0)
+
+extern int R128_READ_PLL(drm_device_t *dev, int addr);
+
+#define R128CCE0(p,r,n)   ((p) | ((n) << 16) | ((r) >> 2))
+#define R128CCE1(p,r1,r2) ((p) | (((r2) >> 2) << 11) | ((r1) >> 2))
+#define R128CCE2(p)       ((p))
+#define R128CCE3(p,n)     ((p) | ((n) << 16))
+
+#endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/vm.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/vm.c
index 534723a50..0b2b00e0a 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/vm.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/vm.c
@@ -247,6 +247,18 @@ int drm_mmap(struct file *filp, struct vm_area_struct *vma)
 				/* Check for valid size. */
 	if (map->size != vma->vm_end - vma->vm_start) return -EINVAL;
 	
+	if (!capable(CAP_SYS_ADMIN) && (map->flags & _DRM_READ_ONLY)) {
+		vma->vm_flags &= VM_MAYWRITE;
+#if defined(__i386__)
+		pgprot_val(vma->vm_page_prot) &= ~_PAGE_RW;
+#else
+				/* Ye gads this is ugly.  With more thought
+                                   we could move this up higher and use
+                                   `protection_map' instead.  */
+		vma->vm_page_prot = __pgprot(pte_val(pte_wrprotect(
+			__pte(pgprot_val(vma->vm_page_prot)))));
+#endif
+	}
 
 	switch (map->type) {
 	case _DRM_FRAME_BUFFER:
@@ -266,6 +278,10 @@ int drm_mmap(struct file *filp, struct vm_area_struct *vma)
 				     vma->vm_end - vma->vm_start,
 				     vma->vm_page_prot))
 				return -EAGAIN;
+		DRM_DEBUG("   Type = %d; start = 0x%lx, end = 0x%lx,"
+			  " offset = 0x%lx\n",
+			  map->type,
+			  vma->vm_start, vma->vm_end, VM_OFFSET(vma));
 		vma->vm_ops = &drm_vm_ops;
 		break;
 	case _DRM_SHM:
@@ -278,19 +294,7 @@ int drm_mmap(struct file *filp, struct vm_area_struct *vma)
 		return -EINVAL;	/* This should never happen. */
 	}
 	vma->vm_flags |= VM_LOCKED | VM_SHM; /* Don't swap */
-	if (map->flags & _DRM_READ_ONLY) {
-#if defined(__i386__)
-		pgprot_val(vma->vm_page_prot) &= ~_PAGE_RW;
-#else
-				/* Ye gads this is ugly.  With more thought
-                                   we could move this up higher and use
-                                   `protection_map' instead.  */
-		vma->vm_page_prot = __pgprot(pte_val(pte_wrprotect(
-			__pte(pgprot_val(vma->vm_page_prot)))));
-#endif
-	}
 
-	
 #if LINUX_VERSION_CODE < 0x020203 /* KERNEL_VERSION(2,2,3) */
 				/* In Linux 2.2.3 and above, this is
 				   handled in do_mmap() in mm/mmap.c. */
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmI810.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmI810.c
new file mode 100644
index 000000000..b2489eeed
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmI810.c
@@ -0,0 +1,87 @@
+#ifdef XFree86Server
+# include "xf86.h"
+# include "xf86_OSproc.h"
+# include "xf86_ansic.h"
+# include "xf86Priv.h"
+# define _DRM_MALLOC xalloc
+# define _DRM_FREE   xfree
+# ifndef XFree86LOADER
+#  include <sys/stat.h>
+#  include <sys/mman.h>
+# endif
+#else
+# include <stdio.h>
+# include <stdlib.h>
+# include <unistd.h>
+# include <string.h>
+# include <ctype.h>
+# include <fcntl.h>
+# include <errno.h>
+# include <signal.h>
+# include <sys/types.h>
+# include <sys/stat.h>
+# include <sys/ioctl.h>
+# include <sys/mman.h>
+# include <sys/time.h>
+# ifdef DRM_USE_MALLOC
+#  define _DRM_MALLOC malloc
+#  define _DRM_FREE   free
+extern int xf86InstallSIGIOHandler(int fd, void (*f)(int, void *), void *);
+extern int xf86RemoveSIGIOHandler(int fd);
+# else
+#  include <Xlibint.h>
+#  define _DRM_MALLOC Xmalloc
+#  define _DRM_FREE   Xfree
+# endif
+#endif
+
+/* Not all systems have MAP_FAILED defined */
+#ifndef MAP_FAILED
+#define MAP_FAILED ((void *)-1)
+#endif
+
+#include <sys/sysmacros.h>	/* for makedev() */
+#include "xf86drm.h"
+#include "xf86drmI810.h"
+#include "drm.h"
+
+Bool drmI810CleanupDma(int driSubFD)
+{
+   drm_i810_init_t init;
+   
+   memset(&init, 0, sizeof(drm_i810_init_t));
+   init.func = I810_CLEANUP_DMA;
+   
+   if(ioctl(driSubFD, DRM_IOCTL_I810_INIT, &init)) {
+      return FALSE;
+   }
+   
+   return TRUE;
+}
+
+Bool drmI810InitDma(int driSubFD, drmI810Init *info)
+{
+   drm_i810_init_t init;
+   
+   memset(&init, 0, sizeof(drm_i810_init_t));
+
+   init.func = I810_INIT_DMA;
+   init.ring_map_idx = info->ring_map_idx;
+   init.buffer_map_idx = info->buffer_map_idx;
+   init.ring_start = info->start;
+   init.ring_end = info->end;
+   init.ring_size = info->size;
+   init.sarea_priv_offset = info->sarea_off;
+   init.front_offset = info->front_offset;
+   init.back_offset = info->back_offset;
+   init.depth_offset = info->depth_offset;
+   init.w = info->w;
+   init.h = info->h;
+   init.pitch = info->pitch;
+   init.pitch_bits = info->pitch_bits;
+
+   if(ioctl(driSubFD, DRM_IOCTL_I810_INIT, &init)) {
+      return FALSE;
+   }
+   return TRUE;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmR128.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmR128.c
new file mode 100644
index 000000000..99f267f3d
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmR128.c
@@ -0,0 +1,199 @@
+/* xf86drmR128.c -- User-level interface to Rage 128 DRM device
+ * Created: Sun Apr  9 18:13:54 2000 by kevin@precisioninsight.com
+ *
+ * Copyright 1999, 2000 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * Author: Kevin E. Martin <kevin@precisioninsight.com>
+ * 
+ * $XFree86$
+ * 
+ */
+
+#ifdef XFree86Server
+# include "xf86.h"
+# include "xf86_OSproc.h"
+# include "xf86_ansic.h"
+# include "xf86Priv.h"
+# define _DRM_MALLOC xalloc
+# define _DRM_FREE   xfree
+# ifndef XFree86LOADER
+#  include <sys/stat.h>
+#  include <sys/mman.h>
+# endif
+#else
+# include <stdio.h>
+# include <stdlib.h>
+# include <unistd.h>
+# include <string.h>
+# include <ctype.h>
+# include <fcntl.h>
+# include <errno.h>
+# include <signal.h>
+# include <sys/types.h>
+# include <sys/stat.h>
+# include <sys/ioctl.h>
+# include <sys/mman.h>
+# include <sys/time.h>
+# ifdef DRM_USE_MALLOC
+#  define _DRM_MALLOC malloc
+#  define _DRM_FREE   free
+extern int xf86InstallSIGIOHandler(int fd, void (*f)(int, void *), void *);
+extern int xf86RemoveSIGIOHandler(int fd);
+# else
+#  include <Xlibint.h>
+#  define _DRM_MALLOC Xmalloc
+#  define _DRM_FREE   Xfree
+# endif
+#endif
+
+/* Not all systems have MAP_FAILED defined */
+#ifndef MAP_FAILED
+#define MAP_FAILED ((void *)-1)
+#endif
+
+#include <sys/sysmacros.h>	/* for makedev() */
+#include "xf86drm.h"
+#include "xf86drmR128.h"
+#include "drm.h"
+
+int drmR128InitCCE(int fd, drmR128Init *info)
+{
+    drm_r128_init_t init;
+   
+    memset(&init, 0, sizeof(drm_r128_init_t));
+
+    init.func                = R128_INIT_CCE;
+    init.sarea_priv_offset   = info->sarea_priv_offset;
+    init.is_pci              = info->is_pci;
+    init.cce_mode            = info->cce_mode;
+    init.cce_fifo_size       = info->cce_fifo_size;
+    init.cce_secure          = info->cce_secure;
+    init.ring_size           = info->ring_size;
+    init.usec_timeout        = info->usec_timeout;
+
+    init.fb_offset           = info->fb_offset;
+    init.agp_ring_offset     = info->agp_ring_offset;
+    init.agp_read_ptr_offset = info->agp_read_ptr_offset;
+    init.agp_vertbufs_offset = info->agp_vertbufs_offset;
+    init.agp_indbufs_offset  = info->agp_indbufs_offset;
+    init.agp_textures_offset = info->agp_textures_offset;
+    init.mmio_offset         = info->mmio_offset;
+
+    if (ioctl(fd, DRM_IOCTL_R128_INIT, &init)) return -errno;
+
+    return 0;
+}
+
+int drmR128CleanupCCE(int fd)
+{
+    drm_r128_init_t init;
+
+    memset(&init, 0, sizeof(drm_r128_init_t));
+
+    init.func = R128_CLEANUP_CCE;
+
+    if (ioctl(fd, DRM_IOCTL_R128_INIT, &init)) return -errno;
+
+    return 0;
+}
+
+int drmR128EngineReset(int fd)
+{
+    if (ioctl(fd, DRM_IOCTL_R128_RESET, NULL)) return -errno;
+
+    return 0;
+}
+
+int drmR128EngineFlush(int fd)
+{
+    if (ioctl(fd, DRM_IOCTL_R128_FLUSH, NULL)) return -errno;
+
+    return 0;
+}
+
+int drmR128CCEWaitForIdle(int fd)
+{
+    if (ioctl(fd, DRM_IOCTL_R128_CCEIDL, NULL)) return -errno;
+
+    return 0;
+}
+
+int drmR128SubmitPackets(int fd, CARD32 *buffer, int *count, int flags)
+{
+    drm_r128_packet_t packet;
+    int               ret;
+
+    memset(&packet, 0, sizeof(drm_r128_packet_t));
+
+    packet.count = *count;
+    packet.flags = flags;
+
+    while (packet.count > 0) {
+	packet.buffer = buffer + (*count - packet.count);
+	ret = ioctl(fd, DRM_IOCTL_R128_PACKET, &packet);
+	if (ret < 0 && ret != -EAGAIN) {
+	    *count = packet.count;
+	    return -errno;
+	}
+    }
+
+    *count = 0;
+    return 0;
+}
+
+int drmR128GetVertexBuffers(int fd, int count, int *indices, int *sizes)
+{
+    drm_r128_vertex_t v;
+
+    v.send_count      = 0;
+    v.send_indices    = NULL;
+    v.send_sizes      = NULL;
+    v.prim            = DRM_R128_PRIM_NONE;
+    v.request_count   = count;
+    v.request_indices = indices;
+    v.request_sizes   = sizes;
+    v.granted_count   = 0;
+
+    if (ioctl(fd, DRM_IOCTL_R128_VERTEX, &v)) return -errno;
+
+    return v.granted_count;
+}
+
+int drmR128FlushVertexBuffers(int fd, int count, int *indices,
+			      int *sizes, drmR128PrimType prim)
+{
+    drm_r128_vertex_t v;
+
+    v.send_count      = count;
+    v.send_indices    = indices;
+    v.send_sizes      = sizes;
+    v.prim            = prim;
+    v.request_count   = 0;
+    v.request_indices = NULL;
+    v.request_sizes   = NULL;
+    v.granted_count   = 0;
+
+    if (ioctl(fd, DRM_IOCTL_R128_VERTEX, &v) < 0) return -errno;
+
+    return 0;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/shared/drm/kernel/drm.h b/xc/programs/Xserver/hw/xfree86/os-support/shared/drm/kernel/drm.h
index 15491aeeb..c8c5581da 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/shared/drm/kernel/drm.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/shared/drm/kernel/drm.h
@@ -72,9 +72,10 @@ typedef struct drm_clip_rect {
            unsigned short y2;
 } drm_clip_rect_t;
 
-/* Seperate include files for the i810/mga specific structures */
+/* Seperate include files for the i810/mga/r128 specific structures */
 #include "mga_drm.h"
 #include "i810_drm.h"
+#include "r128_drm.h"
 
 typedef struct drm_version {
 	int    version_major;	  /* Major version			    */
@@ -297,7 +298,7 @@ typedef struct drm_agp_info {
 
 #define DRM_IOCTL_VERSION    DRM_IOWR(0x00, drm_version_t)
 #define DRM_IOCTL_GET_UNIQUE DRM_IOWR(0x01, drm_unique_t)
-#define DRM_IOCTL_GET_MAGIC  DRM_IOW( 0x02, drm_auth_t)
+#define DRM_IOCTL_GET_MAGIC  DRM_IOR( 0x02, drm_auth_t)
 #define DRM_IOCTL_IRQ_BUSID  DRM_IOWR(0x03, drm_irq_busid_t)
 
 #define DRM_IOCTL_SET_UNIQUE DRM_IOW( 0x10, drm_unique_t)
@@ -328,11 +329,11 @@ typedef struct drm_agp_info {
 
 #define DRM_IOCTL_AGP_ACQUIRE DRM_IO(  0x30)
 #define DRM_IOCTL_AGP_RELEASE DRM_IO(  0x31)
-#define DRM_IOCTL_AGP_ENABLE  DRM_IOR( 0x32, drm_agp_mode_t)
-#define DRM_IOCTL_AGP_INFO    DRM_IOW( 0x33, drm_agp_info_t)
+#define DRM_IOCTL_AGP_ENABLE  DRM_IOW( 0x32, drm_agp_mode_t)
+#define DRM_IOCTL_AGP_INFO    DRM_IOR( 0x33, drm_agp_info_t)
 #define DRM_IOCTL_AGP_ALLOC   DRM_IOWR(0x34, drm_agp_buffer_t)
 #define DRM_IOCTL_AGP_FREE    DRM_IOW( 0x35, drm_agp_buffer_t)
-#define DRM_IOCTL_AGP_BIND    DRM_IOWR(0x36, drm_agp_binding_t)
+#define DRM_IOCTL_AGP_BIND    DRM_IOW( 0x36, drm_agp_binding_t)
 #define DRM_IOCTL_AGP_UNBIND  DRM_IOW( 0x37, drm_agp_binding_t)
 
 /* Mga specific ioctls */
@@ -342,12 +343,23 @@ typedef struct drm_agp_info {
 #define DRM_IOCTL_MGA_ILOAD   DRM_IOW( 0x43, drm_mga_iload_t)
 #define DRM_IOCTL_MGA_VERTEX  DRM_IOW( 0x44, drm_mga_vertex_t)
 #define DRM_IOCTL_MGA_FLUSH   DRM_IOW( 0x45, drm_lock_t )
+#define DRM_IOCTL_MGA_INDICES DRM_IOW( 0x46, drm_mga_indices_t)
 
 /* I810 specific ioctls */
 #define DRM_IOCTL_I810_INIT    DRM_IOW( 0x40, drm_i810_init_t)
 #define DRM_IOCTL_I810_VERTEX  DRM_IOW( 0x41, drm_i810_vertex_t)
-#define DRM_IOCTL_I810_DMA     DRM_IOW( 0x42, drm_i810_general_t)
+#define DRM_IOCTL_I810_CLEAR   DRM_IOW( 0x42, drm_i810_clear_t)
 #define DRM_IOCTL_I810_FLUSH   DRM_IO ( 0x43)
 #define DRM_IOCTL_I810_GETAGE  DRM_IO ( 0x44)
+#define DRM_IOCTL_I810_GETBUF  DRM_IOW( 0x45, drm_i810_dma_t)
+#define DRM_IOCTL_I810_SWAP    DRM_IO ( 0x46)
+
+/* Rage 128 specific ioctls */
+#define DRM_IOCTL_R128_INIT	DRM_IOW( 0x40, drm_r128_init_t)
+#define DRM_IOCTL_R128_RESET	DRM_IO(  0x41)
+#define DRM_IOCTL_R128_FLUSH	DRM_IO(  0x42)
+#define DRM_IOCTL_R128_CCEIDL	DRM_IO(  0x43)
+#define DRM_IOCTL_R128_PACKET	DRM_IOW( 0x44, drm_r128_packet_t)
+#define DRM_IOCTL_R128_VERTEX	DRM_IOW( 0x45, drm_r128_vertex_t)
 
 #endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/xf86drm.h b/xc/programs/Xserver/hw/xfree86/os-support/xf86drm.h
index 61287e3e2..372ba595e 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/xf86drm.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/xf86drm.h
@@ -1,8 +1,7 @@
 /* xf86drm.h -- OS-independent header for DRM user-level library interface
  * Created: Tue Jan  5 08:17:23 1999 by faith@precisioninsight.com
- * Revised: Sun Feb 13 23:46:21 2000 by kevin@precisioninsight.com
  *
- * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * Copyright 1999, 2000 Precision Insight, Inc., Cedar Park, Texas.
  * All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -24,6 +23,8 @@
  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  * DEALINGS IN THE SOFTWARE.
  * 
+ * Author: Rickard E. (Rik) Faith <faith@precisioninsight.com>
+ *
  * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/xf86drm.h,v 1.7 2000/02/23 04:47:21 martin Exp $
  * 
  */
@@ -65,7 +66,8 @@ typedef struct _drmVersion {
 typedef enum {
     DRM_FRAME_BUFFER    = 0,      /* WC, no caching, no core dump           */
     DRM_REGISTERS       = 1,      /* no caching, no core dump               */
-    DRM_SHM             = 2       /* shared, cached                         */
+    DRM_SHM             = 2,      /* shared, cached                         */
+    DRM_AGP             = 3	  /* AGP/GART                               */
 } drmMapType;
 
 typedef enum {
@@ -95,6 +97,11 @@ typedef enum {			 /* These values *MUST* match drm.h         */
 } drmDMAFlags;
 
 typedef enum {
+    DRM_PAGE_ALIGN       = 0x01,
+    DRM_AGP_BUFFER       = 0x02
+} drmBufDescFlags;
+
+typedef enum {
     DRM_LOCK_READY      = 0x01, /* Wait until hardware is ready for DMA */
     DRM_LOCK_QUIESCENT  = 0x02, /* Wait until hardware quiescent        */
     DRM_LOCK_FLUSH      = 0x04, /* Flush this context's DMA queue first */
@@ -196,7 +203,7 @@ typedef struct { unsigned int a[100]; } __drm_dummy_lock_t;
 #endif
 
 #ifndef DRM_CAS
-#define DRM_CAS(lock,old,new,ret) /* FAST LOCK FAILS */
+#define DRM_CAS(lock,old,new,ret) do { ret=1; } while (0) /* FAST LOCK FAILS */
 #endif
 
 #define DRM_LIGHT_LOCK(fd,lock,context)                                \
@@ -291,7 +298,9 @@ extern int           drmAddMap(int fd,
 			       drmMapType type,
 			       drmMapFlags flags,
 			       drmHandlePtr handle);
-extern int           drmAddBufs(int fd, int count, int size, int flags);
+extern int           drmAddBufs(int fd, int count, int size, 
+				drmBufDescFlags flags,
+				int agp_offset);
 extern int           drmMarkBufs(int fd, double low, double high);
 extern int           drmCreateContext(int fd, drmContextPtr handle);
 extern int           drmSetContextFlags(int fd, drmContext context,
@@ -332,6 +341,29 @@ extern int           drmGetLock(int fd,
 extern int           drmUnlock(int fd, drmContext context);
 extern int           drmFinish(int fd, int context, drmLockFlags flags);
 
+/* AGP/GART support: X server (root) only */
+extern int           drmAgpAcquire(int fd);
+extern int           drmAgpRelease(int fd);
+extern int           drmAgpEnable(int fd, unsigned long mode);
+extern int           drmAgpAlloc(int fd, unsigned long size, 
+				 unsigned long type, unsigned long *address,
+				 unsigned long *handle);
+extern int           drmAgpFree(int fd, unsigned long handle);
+extern int 	     drmAgpBind(int fd, unsigned long handle,
+				unsigned long offset);
+extern int           drmAgpUnbind(int fd, unsigned long handle);
+
+/* AGP/GART info: authenticated client and/or X */
+extern int           drmAgpVersionMajor(int fd);
+extern int           drmAgpVersionMinor(int fd);
+extern unsigned long drmAgpGetMode(int fd);
+extern unsigned long drmAgpBase(int fd); /* Physical location */
+extern unsigned long drmAgpSize(int fd); /* Bytes */
+extern unsigned long drmAgpMemoryUsed(int fd);
+extern unsigned long drmAgpMemoryAvail(int fd);
+extern unsigned int  drmAgpVendorId(int fd);
+extern unsigned int  drmAgpDeviceId(int fd);
+
 /* Support routines */
 extern int           drmError(int err, const char *label);
 extern void          *drmMalloc(int size);
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/xf86drmI810.h b/xc/programs/Xserver/hw/xfree86/os-support/xf86drmI810.h
new file mode 100644
index 000000000..6c11b7a10
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/xf86drmI810.h
@@ -0,0 +1,46 @@
+
+/* WARNING: If you change any of these defines, make sure to change
+ * the kernel include file as well (i810_drm.h)
+ */
+
+#ifndef _XF86DRI_I810_H_
+#define _XF86DRI_I810_H_
+
+#ifndef _I810_DEFINES_
+#define _I810_DEFINES_
+#define I810_USE_BATCH 1
+
+#define I810_DMA_BUF_ORDER     12
+#define I810_DMA_BUF_SZ        (1<<I810_DMA_BUF_ORDER)
+#define I810_DMA_BUF_NR        256
+
+#define I810_NR_SAREA_CLIPRECTS 8
+
+/* Each region is a minimum of 64k, and there are at most 64 of them.
+ */
+#define I810_NR_TEX_REGIONS 64
+#define I810_LOG_MIN_TEX_REGION_SIZE 16
+#endif
+
+typedef struct _drmI810Init {
+   unsigned int start; 
+   unsigned int end; 
+   unsigned int size; 
+   int ring_map_idx; 
+   int buffer_map_idx; 
+   int sarea_off;
+
+   unsigned int front_offset;
+   unsigned int back_offset;
+   unsigned int depth_offset;
+   unsigned int w;
+   unsigned int h;
+   unsigned int pitch;
+   unsigned int pitch_bits;
+} drmI810Init;
+
+
+Bool drmI810CleanupDma(int driSubFD);
+Bool drmI810InitDma(int driSubFD, drmI810Init *info );
+
+#endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/xf86drmMga.h b/xc/programs/Xserver/hw/xfree86/os-support/xf86drmMga.h
new file mode 100644
index 000000000..3b0e4a6fc
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/xf86drmMga.h
@@ -0,0 +1,147 @@
+
+/* WARNING: If you change any of these defines, make sure to change
+ * the kernel include file as well (mga_drm.h)
+ */
+
+#ifndef _XF86DRI_MGA_H_
+#define _XF86DRI_MGA_H_
+#ifndef _MGA_DEFINES_
+#define _MGA_DEFINES_
+#define MGA_F  0x1		/* fog */
+#define MGA_A  0x2		/* alpha */
+#define MGA_S  0x4		/* specular */
+#define MGA_T2 0x8		/* multitexture */
+
+#define MGA_WARP_TGZ            0
+#define MGA_WARP_TGZF           (MGA_F)
+#define MGA_WARP_TGZA           (MGA_A)
+#define MGA_WARP_TGZAF          (MGA_F|MGA_A)
+#define MGA_WARP_TGZS           (MGA_S)
+#define MGA_WARP_TGZSF          (MGA_S|MGA_F)
+#define MGA_WARP_TGZSA          (MGA_S|MGA_A)
+#define MGA_WARP_TGZSAF         (MGA_S|MGA_F|MGA_A)
+#define MGA_WARP_T2GZ           (MGA_T2)
+#define MGA_WARP_T2GZF          (MGA_T2|MGA_F)
+#define MGA_WARP_T2GZA          (MGA_T2|MGA_A)
+#define MGA_WARP_T2GZAF         (MGA_T2|MGA_A|MGA_F)
+#define MGA_WARP_T2GZS          (MGA_T2|MGA_S)
+#define MGA_WARP_T2GZSF         (MGA_T2|MGA_S|MGA_F)
+#define MGA_WARP_T2GZSA         (MGA_T2|MGA_S|MGA_A)
+#define MGA_WARP_T2GZSAF        (MGA_T2|MGA_S|MGA_F|MGA_A)
+
+#define MGA_MAX_G400_PIPES 16
+#define MGA_MAX_G200_PIPES  8	/* no multitex */
+
+#define MGA_MAX_WARP_PIPES MGA_MAX_G400_PIPES
+
+#define MGA_CARD_TYPE_G200 1
+#define MGA_CARD_TYPE_G400 2
+#define MGA_FRONT   0x1
+#define MGA_BACK    0x2
+#define MGA_DEPTH   0x4
+
+/* 3d state excluding texture units:
+ */
+#define MGA_CTXREG_DSTORG   0	/* validated */
+#define MGA_CTXREG_MACCESS  1	
+#define MGA_CTXREG_PLNWT    2	
+#define MGA_CTXREG_DWGCTL    3	
+#define MGA_CTXREG_ALPHACTRL 4
+#define MGA_CTXREG_FOGCOLOR  5
+#define MGA_CTXREG_WFLAG     6
+#define MGA_CTXREG_TDUAL0    7
+#define MGA_CTXREG_TDUAL1    8
+#define MGA_CTXREG_FCOL      9
+#define MGA_CTX_SETUP_SIZE   10
+
+/* 2d state
+ */
+#define MGA_2DREG_PITCH 	0
+#define MGA_2D_SETUP_SIZE 	1
+
+/* Each texture unit has a state:
+ */
+#define MGA_TEXREG_CTL        0
+#define MGA_TEXREG_CTL2       1
+#define MGA_TEXREG_FILTER     2
+#define MGA_TEXREG_BORDERCOL  3
+#define MGA_TEXREG_ORG        4 /* validated */
+#define MGA_TEXREG_ORG1       5
+#define MGA_TEXREG_ORG2       6
+#define MGA_TEXREG_ORG3       7
+#define MGA_TEXREG_ORG4       8
+#define MGA_TEXREG_WIDTH      9
+#define MGA_TEXREG_HEIGHT     10
+#define MGA_TEX_SETUP_SIZE    11
+
+/* What needs to be changed for the current vertex dma buffer?
+ */
+#define MGA_UPLOAD_CTX        0x1
+#define MGA_UPLOAD_TEX0       0x2
+#define MGA_UPLOAD_TEX1       0x4
+#define MGA_UPLOAD_PIPE       0x8
+#define MGA_UPLOAD_TEX0IMAGE  0x10
+#define MGA_UPLOAD_TEX1IMAGE  0x20
+#define MGA_UPLOAD_2D 	      0x40
+#define MGA_WAIT_AGE          0x80 /* handled client-side */
+#define MGA_UPLOAD_CLIPRECTS  0x100 /* handled client-side */
+#define MGA_DMA_FLUSH	      0x200 /* set when someone gets the lock
+                                       quiescent */
+
+/* 32 buffers of 64k each, total 1 meg.
+ */
+#define MGA_DMA_BUF_ORDER     16
+#define MGA_DMA_BUF_SZ        (1<<MGA_DMA_BUF_ORDER)
+#define MGA_DMA_BUF_NR        31
+
+/* Keep these small for testing.
+ */
+#define MGA_NR_SAREA_CLIPRECTS 8
+
+/* 2 heaps (1 for card, 1 for agp), each divided into upto 128
+ * regions, subject to a minimum region size of (1<<16) == 64k. 
+ *
+ * Clients may subdivide regions internally, but when sharing between
+ * clients, the region size is the minimum granularity. 
+ */
+
+#define MGA_CARD_HEAP 0
+#define MGA_AGP_HEAP  1
+#define MGA_NR_TEX_HEAPS 2
+#define MGA_NR_TEX_REGIONS 16
+#define MGA_LOG_MIN_TEX_REGION_SIZE 16
+#endif
+
+typedef struct _drmMgaWarpIndex {
+   int installed;
+   unsigned long phys_addr;
+   int size;
+} drmMgaWarpIndex;
+
+typedef struct _drmMgaInit {
+   int reserved_map_agpstart;
+   int reserved_map_idx;
+   int buffer_map_idx;
+   int sarea_priv_offset;
+   int primary_size;
+   int warp_ucode_size;
+   unsigned int frontOffset;
+   unsigned int backOffset;
+   unsigned int depthOffset;
+   unsigned int textureOffset;
+   unsigned int textureSize;
+   unsigned int agpTextureSize;
+   unsigned int agpTextureOffset;
+   unsigned int cpp;
+   unsigned int stride;
+   int sgram;
+   int chipset;
+   drmMgaWarpIndex WarpIndex[MGA_MAX_WARP_PIPES];
+   unsigned int mAccess;
+} drmMgaInit;
+
+
+Bool drmMgaCleanupDma(int driSubFD);
+Bool drmMgaLockUpdate(int driSubFD, drmLockFlags flags);
+Bool drmMgaInitDma(int driSubFD, drmMgaInit *info);
+#endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/xf86drmR128.h b/xc/programs/Xserver/hw/xfree86/os-support/xf86drmR128.h
new file mode 100644
index 000000000..46898a53c
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/xf86drmR128.h
@@ -0,0 +1,80 @@
+/* xf86drm.h -- OS-independent header for DRM user-level library interface
+ * Created: Sun Apr  9 18:16:28 2000 by kevin@precisioninsight.com
+ *
+ * Copyright 1999, 2000 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * Author: Kevin E. Martin <kevin@precisioninsight.com>
+ *
+ * $XFree86$
+ * 
+ */
+
+#ifndef _XF86DRI_R128_H_
+#define _XF86DRI_R128_H_
+
+/* WARNING: If you change any of these defines, make sure to change
+ * the kernel include file as well (r128_drm.h)
+ */
+
+typedef struct _drmR128Init {
+    int sarea_priv_offset;
+    int is_pci;
+    int cce_mode;
+    int cce_fifo_size;
+    int cce_secure;
+    int ring_size;
+    int usec_timeout;
+
+    int fb_offset;
+    int agp_ring_offset;
+    int agp_read_ptr_offset;
+    int agp_vertbufs_offset;
+    int agp_indbufs_offset;
+    int agp_textures_offset;
+    int mmio_offset;
+} drmR128Init;
+
+typedef enum {
+    DRM_R128_PRIM_NONE		= 0x0001,
+    DRM_R128_PRIM_POINT		= 0x0002,
+    DRM_R128_PRIM_LINE		= 0x0004,
+    DRM_R128_PRIM_POLY_LINE	= 0x0008,
+    DRM_R128_PRIM_TRI_LIST	= 0x0010,
+    DRM_R128_PRIM_TRI_FAN	= 0x0020,
+    DRM_R128_PRIM_TRI_STRIP	= 0x0040,
+    DRM_R128_PRIM_TRI_TYPE2	= 0x0080
+} drmR128PrimType;
+
+
+extern int drmR128InitCCE(int fd, drmR128Init *info);
+extern int drmR128CleanupCCE(int fd);
+extern int drmR128EngineReset(int fd);
+extern int drmR128EngineFlush(int fd);
+extern int drmR128CCEWaitForIdle(int fd);
+extern int drmR128SubmitPackets(int fd, CARD32 *buffer, int *count, int flags);
+extern int drmR128GetVertexBuffers(int fd, int count, int *indices,
+				   int *sizes);
+extern int drmR128FlushVertexBuffers(int fd, int count, int *indices,
+				     int *sizes, drmR128PrimType prim);
+
+#endif