Merged mga branch with trunkmga-0-0-2-20000404-merge

author: jhartmann <jhartmann> 2000-04-04 22:08:12 +0000
committer: jhartmann <jhartmann> 2000-04-04 22:08:12 +0000
commit: 4fe5b7fbada2f73c1533e30f8abc5cf71e7c61c4 (patch)
tree: d526515e629abcae4e03e1c5c02b60243dc35f72 /xc
parent: d13b9c4ecba68ca9c63d94dbf48a395e3d39b847 (diff)
155 files changed, 12965 insertions, 9747 deletions
diff --git a/xc/config/cf/host.def b/xc/config/cf/host.def
index ef72a80d4..67b924bbf 100644
--- a/xc/config/cf/host.def
+++ b/xc/config/cf/host.def
@@ -1,13 +1,26 @@
 
+#define ProjectRoot /mnt/src/XF40
 #define DefaultGcc2i386Opt -O2
 #define LibraryCDebugFlags -O2
 #define BuildServersOnly YES
-#define XF86CardDrivers vga i810 tdfx
+#define XF86CardDrivers vga tdfx i810 mga 
 #define LinuxDistribution LinuxRedHat
-#define DefaultCCOptions -ansi GccWarningOptions -pipe
+#define GccWarningOptions -Wall -Wpointer-arith -Wstrict-prototypes \
+                          -Wmissing-prototypes -Wmissing-declarations \
+                          -Wnested-externs
+#define DefaultCCOptions -ansi GccWarningOptions -pipe -g
+#define NormalLibGlx NO
+
 #define BuildXF86DRI YES
 #define HasGlide3 YES
+
 /* Optionally turn these on for debugging */
 /* #define GlxBuiltInTdfx YES */
+/* #define GlxBuiltInI810 YES */
+/* #define GlxBuiltInMga YES */
 /* #define DoLoadableServer NO */
 #define SharedLibFont NO
+
+#define XnestServer NO
+#define XVirtualFramebufferServer NO
+#define XprtServer NO
diff --git a/xc/extras/Mesa/src/dlist.c b/xc/extras/Mesa/src/dlist.c
index cdb1a5ac3..b2cc13787 100644
--- a/xc/extras/Mesa/src/dlist.c
+++ b/xc/extras/Mesa/src/dlist.c
@@ -4808,6 +4808,7 @@ static void print_list( GLcontext *ctx, FILE *f, GLuint list )
 	 case OPCODE_END_OF_LIST:
             fprintf(f,"END-LIST %u\n", list);
 	    done = GL_TRUE;
+	    gl_print_cassette( (struct immediate *) n[1].data );
 	    break;
          default:
             if (opcode < 0 || opcode > OPCODE_END_OF_LIST) {
diff --git a/xc/extras/Mesa/src/fog.c b/xc/extras/Mesa/src/fog.c
index bac41cb1d..e35906c8f 100644
--- a/xc/extras/Mesa/src/fog.c
+++ b/xc/extras/Mesa/src/fog.c
@@ -35,6 +35,7 @@
 #include "types.h"
 #include "xform.h"
 #endif
+#include "xform.h"
 
 
 
diff --git a/xc/extras/Mesa/src/fog.h b/xc/extras/Mesa/src/fog.h
index 9a58f54a1..59d84517b 100644
--- a/xc/extras/Mesa/src/fog.h
+++ b/xc/extras/Mesa/src/fog.h
@@ -65,4 +65,6 @@ extern void
 _mesa_init_fog( void );
 
 
+extern struct gl_pipeline_stage gl_fog_coord_stage;
+
 #endif
diff --git a/xc/extras/Mesa/src/mmath.h b/xc/extras/Mesa/src/mmath.h
index 6048eb186..30a300cdb 100644
--- a/xc/extras/Mesa/src/mmath.h
+++ b/xc/extras/Mesa/src/mmath.h
@@ -22,7 +22,7 @@
  * 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.
  */
-
+/* $XFree86: xc/extras/Mesa/src/mmath.h,v 1.4 2000/03/03 18:30:08 tsi Exp $ */
 
 /*
  * Faster arithmetic functions.  If the FAST_MATH preprocessor symbol is
@@ -47,7 +47,10 @@
  * In the worst case, we force the compiler to use a memory access to
  * truncate the float, by specifying the 'volatile' keyword.
  */
-#if defined(__linux__) && defined(__i386__) 
+#if defined(__linux__) && defined(__i386__) && !defined(IN_MODULE)
+/*
+ * A libc interface is needed for this...
+ */
 #include <fpu_control.h>
 
 #if !defined(_FPU_SETCW)
diff --git a/xc/extras/Mesa/src/types.h b/xc/extras/Mesa/src/types.h
index dc1b5e103..e96740e78 100644
--- a/xc/extras/Mesa/src/types.h
+++ b/xc/extras/Mesa/src/types.h
@@ -1345,8 +1345,6 @@ struct gl_extensions {
 #define DD_STENCIL                  0x1000000 
 #define DD_CLIP_FOG_COORD           0x2000000
 
-
-
 #define DD_SW_SETUP           (DD_TRI_CULL|		\
                                DD_TRI_CULL_FRONT_BACK|	\
                                DD_TRI_OFFSET|		\
diff --git a/xc/extras/Mesa/src/varray.c b/xc/extras/Mesa/src/varray.c
index dc758594a..786793da7 100644
--- a/xc/extras/Mesa/src/varray.c
+++ b/xc/extras/Mesa/src/varray.c
@@ -682,6 +682,7 @@ _mesa_DrawArrays(GLenum mode, GLint start, GLsizei count)
 	 VB->Primitive[VB->CopyStart] = mode;
 	 ctx->Array.Flag[count] |= VERT_END_VB;
 
+	 ctx->Array.Flag[count] |= VERT_END_VB;
          /* Transform and render.
 	  */
          gl_run_pipeline( VB );
diff --git a/xc/include/GL/glx.h b/xc/include/GL/glx.h
index 367ce0a51..0254b6064 100644
--- a/xc/include/GL/glx.h
+++ b/xc/include/GL/glx.h
@@ -101,7 +101,7 @@ 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 );
 
 
 
diff --git a/xc/lib/GL/Imakefile b/xc/lib/GL/Imakefile
index a185089e4..c3dfb015d 100644
--- a/xc/lib/GL/Imakefile
+++ b/xc/lib/GL/Imakefile
@@ -27,6 +27,14 @@ DependSubdirs($(SUBDIRS))
      LIBNAME = GL
      SOREV   = $(SOGLREV)
 
+#if 0
+        LOWSRC = lowpc.c
+	LOWOBJ = lowpc.o
+
+	HISRC = highpc.c
+	HIOBJ = highpc.o
+#endif
+
      GLXOBJS = glx/?*.o
     GLXUOBJS = glx/unshared/?*.o
     GLXDOBJS = glx/debugger/?*.o
@@ -98,6 +106,87 @@ DRIMESADONES = mesa/dri/DONE
     DRVDONES = $(TDFXDONES) $(MESADONES) $(DRIMESADONES) $(DRMDONES)
 
  REQUIREDLIBS += -lglide3x
+
+#elif GlxBuiltInI810
+
+     DRMOBJS = dri/drm/?*.o
+    DRMUOBJS = dri/drm/unshared/?*.o
+    DRMDOBJS = dri/drm/debugger/?*.o
+    DRMPOBJS = dri/drm/profiled/?*.o
+    DRMDONES = dri/drm/DONE
+
+    I810OBJS = mesa/src/drv/i810/?*.o
+   I810UOBJS = mesa/src/drv/i810/unshared/?*.o
+   I810DOBJS = mesa/src/drv/i810/debugger/?*.o
+   I810POBJS = mesa/src/drv/i810/profiled/?*.o
+   I810DONES = mesa/src/drv/i810/DONE
+
+    COMMONOBJS = mesa/src/drv/common/?*.o
+   COMMONUOBJS = mesa/src/drv/common/unshared/?*.o
+   COMMONDOBJS = mesa/src/drv/common/debugger/?*.o
+   COMMONPOBJS = mesa/src/drv/common/profiled/?*.o
+   COMMONDONES = mesa/src/drv/common/DONE
+
+    MESAOBJS = mesa/src/?*.o mesa/src/X86/?*.o
+   MESAUOBJS = mesa/src/unshared/?*.o mesa/src/X86/unshared/?*.o
+   MESADOBJS = mesa/src/debugger/?*.o mesa/src/X86/debugger/?*.o
+   MESAPOBJS = mesa/src/profiled/?*.o mesa/src/X86/profiled/?*.o
+   MESADONES = mesa/src/DONE mesa/src/X86/DONE
+
+ DRIMESAOBJS = mesa/dri/?*.o
+DRIMESAUOBJS = mesa/dri/unshared/?*.o
+DRIMESADOBJS = mesa/dri/debugger/?*.o
+DRIMESAPOBJS = mesa/dri/profiled/?*.o
+DRIMESADONES = mesa/dri/DONE
+
+     DRVOBJS = $(I810OBJS) $(COMMONOBJS) $(MESAOBJS) $(DRIMESAOBJS) $(DRMOBJS)
+    DRVUOBJS = $(I810UOBJS) $(COMMONUOBJS) $(MESAUOBJS) $(DRIMESAUOBJS) $(DRMUOBJS)
+    DRVDOBJS = $(I810DOBJS) $(COMMONDOBJS) $(MESADOBJS) $(DRIMESADOBJS) $(DRMDOBJS)
+    DRVPOBJS = $(I810POBJS) $(COMMONPOBJS) $(MESAPOBJS) $(DRIMESAPOBJS) $(DRMPOBJS)
+    DRVDONES = $(I810DONES) $(COMMONDONES) $(MESADONES) $(DRIMESADONES) $(DRMDONES)
+
+
+
+
+#elif GlxBuiltInMga
+
+     DRMOBJS = dri/drm/?*.o
+    DRMUOBJS = dri/drm/unshared/?*.o
+    DRMDOBJS = dri/drm/debugger/?*.o
+    DRMPOBJS = dri/drm/profiled/?*.o
+    DRMDONES = dri/drm/DONE
+
+    MGAOBJS = mesa/src/drv/mga/?*.o
+   MGAUOBJS = mesa/src/drv/mga/unshared/?*.o
+   MGADOBJS = mesa/src/drv/mga/debugger/?*.o
+   MGAPOBJS = mesa/src/drv/mga/profiled/?*.o
+   MGADONES = mesa/src/drv/mga/DONE
+
+    COMMONOBJS = mesa/src/drv/common/?*.o
+   COMMONUOBJS = mesa/src/drv/common/unshared/?*.o
+   COMMONDOBJS = mesa/src/drv/common/debugger/?*.o
+   COMMONPOBJS = mesa/src/drv/common/profiled/?*.o
+   COMMONDONES = mesa/src/drv/common/DONE
+
+    MESAOBJS = mesa/src/?*.o mesa/src/X86/?*.o
+   MESAUOBJS = mesa/src/unshared/?*.o mesa/src/X86/unshared/?*.o
+   MESADOBJS = mesa/src/debugger/?*.o mesa/src/X86/debugger/?*.o
+   MESAPOBJS = mesa/src/profiled/?*.o mesa/src/X86/profiled/?*.o
+   MESADONES = mesa/src/DONE mesa/src/X86/DONE
+
+ DRIMESAOBJS = mesa/dri/?*.o
+DRIMESAUOBJS = mesa/dri/unshared/?*.o
+DRIMESADOBJS = mesa/dri/debugger/?*.o
+DRIMESAPOBJS = mesa/dri/profiled/?*.o
+DRIMESADONES = mesa/dri/DONE
+
+     DRVOBJS = $(MGAOBJS) $(COMMONOBJS) $(MESAOBJS) $(DRIMESAOBJS) $(DRMOBJS)
+    DRVUOBJS = $(MGAUOBJS) $(COMMONUOBJS) $(MESAUOBJS) $(DRIMESAUOBJS) $(DRMUOBJS)
+    DRVDOBJS = $(MGADOBJS) $(COMMONDOBJS) $(MESADOBJS) $(DRIMESADOBJS) $(DRMDOBJS)
+    DRVPOBJS = $(MGAPOBJS) $(COMMONPOBJS) $(MESAPOBJS) $(DRIMESAPOBJS) $(DRMPOBJS)
+    DRVDONES = $(MGADONES) $(COMMONDONES) $(MESADONES) $(DRIMESADONES) $(DRMDONES)
+
+
 #elif GlxBuiltInMesa
 #ifndef GlxDriverUsesMesa
      DRVOBJS = mesa/src/?*.o mesa/src/X/?*.o
@@ -113,17 +202,26 @@ DRIMESADONES = mesa/dri/DONE
 #ifdef OS2Architecture
         OBJS = $(LIBNAME).a
 #else
-       OBJS = $(GLXOBJS)  $(DRIOBJS)   $(DRVOBJS)
+       OBJS = $(LOWOBJ) $(GLXOBJS) $(DRIOBJS) $(DRVOBJS) $(HIOBJ)
 #endif
 #if HasSharedLibraries
+#if 1
        UOBJS = $(GLXUOBJS) $(DRIUOBJS) $(DRVUOBJS)
 #else
        UOBJS = $(OBJS)
 #endif
+#else
+       UOBJS = $(OBJS)
+#endif
        DOBJS = $(GLXDOBJS) $(DRIDOBJS) $(DRVDOBJS)
        POBJS = $(GLXPOBJS) $(DRIPOBJS) $(DRVPOBJS)
        DONES = $(GLXDONES) $(DRIDONES) $(DRVDONES)
 
+#if 0
+SubdirLibraryRule(highpc.o lowpc.o)
+NormalLintTarget(highpc.c lowpc.c)
+#endif
+
 #if LocalThreads
   THREADOBJS = $(THREADS_LIBS)
 #endif
diff --git a/xc/lib/GL/dri/XF86dri.c b/xc/lib/GL/dri/XF86dri.c
index 77654ac76..cd485680b 100644
--- a/xc/lib/GL/dri/XF86dri.c
+++ b/xc/lib/GL/dri/XF86dri.c
@@ -1,4 +1,4 @@
-/* $XFree86: xc/lib/GL/dri/XF86dri.c,v 1.2 1999/06/27 14:07:22 dawes Exp $ */
+/* $XFree86: xc/lib/GL/dri/XF86dri.c,v 1.5 2000/02/23 04:46:33 martin Exp $ */
 /**************************************************************************
 
 Copyright 1998-1999 Precision Insight, Inc., Cedar Park, Texas.
@@ -31,7 +31,6 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  *   Kevin E. Martin <kevin@precisioninsight.com>
  *   Jens Owen <jens@precisioninsight.com>
  *
- * $PI: xc/lib/GL/dri/XF86dri.c,v 1.14 1999/06/16 20:08:33 faith Exp $
  */
 
 /* THIS IS NOT AN X CONSORTIUM STANDARD */
@@ -385,7 +384,11 @@ Bool XF86DRIDestroyDrawable(dpy, screen, drawable)
 }
 
 Bool XF86DRIGetDrawableInfo(dpy, screen, drawable, 
-	index, stamp, X, Y, W, H, numClipRects, pClipRects)
+			    index, stamp, X, Y, W, H, 
+			    numClipRects, pClipRects,
+			    backX, backY, 
+			    numBackClipRects, pBackClipRects
+			    )
     Display* dpy;
     int screen;
     Drawable drawable;
@@ -397,10 +400,15 @@ Bool XF86DRIGetDrawableInfo(dpy, screen, drawable,
     int* H;
     int* numClipRects;
     XF86DRIClipRectPtr* pClipRects;
+    int* backX;
+    int* backY;
+    int* numBackClipRects;
+    XF86DRIClipRectPtr*	pBackClipRects;    
 {
     XExtDisplayInfo *info = find_display (dpy);
     xXF86DRIGetDrawableInfoReply rep;
     xXF86DRIGetDrawableInfoReq *req;
+    int total_rects;
 
     XF86DRICheckExtension (dpy, info, False);
 
@@ -410,7 +418,9 @@ Bool XF86DRIGetDrawableInfo(dpy, screen, drawable,
     req->driReqType = X_XF86DRIGetDrawableInfo;
     req->screen = screen;
     req->drawable = drawable;
-    if (!_XReply(dpy, (xReply *)&rep, 0, xFalse)) {
+
+    if (!_XReply(dpy, (xReply *)&rep, 1, xFalse)) 
+    {
 	UnlockDisplay(dpy);
 	SyncHandle();
 	return False;
@@ -422,16 +432,41 @@ Bool XF86DRIGetDrawableInfo(dpy, screen, drawable,
     *W = (int)rep.drawableWidth;
     *H = (int)rep.drawableHeight;
     *numClipRects = rep.numClipRects;
+    total_rects = *numClipRects;
+
+    *backX = rep.backX;
+    *backY = rep.backY;
+    *numBackClipRects = rep.numBackClipRects;
+    total_rects += *numBackClipRects;
+
+    if (rep.length != (SIZEOF(xXF86DRIGetDrawableInfoReply) - 
+		       SIZEOF(xGenericReply) + 
+		       total_rects * sizeof(XF86DRIClipRectRec))) {
+       _XEatData(dpy, rep.length);
+       return False;
+    }
+       
 
-    if (rep.length) {
-        if (!(*pClipRects = (XF86DRIClipRectPtr)Xcalloc(rep.length, 1))) {
-            _XEatData(dpy, rep.length);
-            return False;
-        }
-	_XRead32(dpy, *pClipRects, rep.length);
+    if (*numClipRects) {
+       int len = sizeof(XF86DRIClipRectRec) * (*numClipRects);
+
+       *pClipRects = (XF86DRIClipRectPtr)Xcalloc(len, 1);
+       if (*pClipRects) 
+	  _XRead32(dpy, *pClipRects, len);
     } else {
         *pClipRects = NULL;
     }
+
+    if (*numBackClipRects) {
+       int len = sizeof(XF86DRIClipRectRec) * (*numBackClipRects);
+
+       *pBackClipRects = (XF86DRIClipRectPtr)Xcalloc(len, 1);
+       if (*pBackClipRects) 
+	  _XRead32(dpy, *pBackClipRects, len);
+    } else {
+        *pBackClipRects = NULL;
+    }
+
     UnlockDisplay(dpy);
     SyncHandle();
     return True;
diff --git a/xc/lib/GL/dri/drm/Imakefile b/xc/lib/GL/dri/drm/Imakefile
index 18619461b..4fd493e1d 100644
--- a/xc/lib/GL/dri/drm/Imakefile
+++ b/xc/lib/GL/dri/drm/Imakefile
@@ -25,6 +25,8 @@ 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)
 
 #include <Library.tmpl>
 
diff --git a/xc/lib/GL/dri/xf86dri.h b/xc/lib/GL/dri/xf86dri.h
index 8e234fcb2..8b7cd3aa2 100644
--- a/xc/lib/GL/dri/xf86dri.h
+++ b/xc/lib/GL/dri/xf86dri.h
@@ -1,4 +1,4 @@
-/* $XFree86: xc/lib/GL/dri/xf86dri.h,v 1.2 1999/06/27 14:07:24 dawes Exp $ */
+/* $XFree86: xc/lib/GL/dri/xf86dri.h,v 1.5 2000/02/23 04:46:34 martin Exp $ */
 /**************************************************************************
 
 Copyright 1998-1999 Precision Insight, Inc., Cedar Park, Texas.
@@ -32,7 +32,6 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  *   Jens Owen <jens@precisioninsight.com>
  *   Rickard E. Faith <faith@precisioninsight.com>
  *
- * $PI: xc/lib/GL/dri/xf86dri.h,v 1.12 1999/06/16 20:08:34 faith Exp $
  */
 
 #ifndef _XF86DRI_H_
@@ -60,6 +59,8 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define XF86DRIOperationNotSupported	1
 #define XF86DRINumberErrors		(XF86DRIOperationNotSupported + 1)
 
+/* Warning : Do not change XF86DRIClipRect without changing the kernel 
+ * structure! */
 typedef struct _XF86DRIClipRect {
     unsigned short	x1; /* Upper left: inclusive */
     unsigned short	y1;
@@ -178,7 +179,11 @@ Bool XF86DRIGetDrawableInfo(
     int*		/* W */,
     int*		/* H */,
     int*		/* numClipRects */,
-    XF86DRIClipRectPtr*	/* pClipRects */
+    XF86DRIClipRectPtr*,/* pClipRects */
+    int*		/* backX */,
+    int*		/* backY */,
+    int*		/* numBackClipRects */,
+    XF86DRIClipRectPtr*	/* pBackClipRects */    
 #endif
 );
 
diff --git a/xc/lib/GL/dri/xf86dristr.h b/xc/lib/GL/dri/xf86dristr.h
index 32969834c..dd246bac8 100644
--- a/xc/lib/GL/dri/xf86dristr.h
+++ b/xc/lib/GL/dri/xf86dristr.h
@@ -1,4 +1,4 @@
-/* $XFree86: xc/lib/GL/dri/xf86dristr.h,v 1.2 1999/06/27 14:07:24 dawes Exp $ */
+/* $XFree86: xc/lib/GL/dri/xf86dristr.h,v 1.5 2000/02/23 04:46:34 martin Exp $ */
 /**************************************************************************
 
 Copyright 1998-1999 Precision Insight, Inc., Cedar Park, Texas.
@@ -31,7 +31,6 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  *   Kevin E. Martin <kevin@precisioninsight.com>
  *   Jens Owen <jens@precisioninsight.com>
  *
- * $PI: xc/lib/GL/dri/xf86dristr.h,v 1.11 1999/06/22 03:09:44 jens Exp $
  */
 
 #ifndef _XF86DRISTR_H_
@@ -41,7 +40,7 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 #define XF86DRINAME "XFree86-DRI"
 
-#define XF86DRI_MAJOR_VERSION	1	/* current version numbers */
+#define XF86DRI_MAJOR_VERSION	2	/* current version numbers */
 #define XF86DRI_MINOR_VERSION	0
 #define XF86DRI_PATCH_VERSION	0
 
@@ -239,9 +238,13 @@ typedef struct {
     INT16	drawableWidth B16;
     INT16	drawableHeight B16;
     CARD32	numClipRects B32;
-    CARD32	pad6 B32;
+    INT16       backX B16;
+    INT16       backY B16;
+    CARD32      numBackClipRects B32;
 } xXF86DRIGetDrawableInfoReply;
-#define sz_xXF86DRIGetDrawableInfoReply	32
+
+#define sz_xXF86DRIGetDrawableInfoReply	36
+
 
 typedef struct _XF86DRIGetDeviceInfo {
     CARD8	reqType;		/* always DRIReqCode */
diff --git a/xc/lib/GL/highpc.c b/xc/lib/GL/highpc.c
new file mode 100644
index 000000000..c9218d02a
--- /dev/null
+++ b/xc/lib/GL/highpc.c
@@ -0,0 +1,75 @@
+#include <stdio.h>
+#include <stdlib.h>
+
+
+/* Mesa shared lib profiling code by Josh Vanderhoof.  Adapted
+ * (trivially) to the DRI by Keith Whitwell. 
+ */
+
+/*
+ * This is the highest address in libGL.so
+ */
+void glx_highpc(void) { }
+
+#if defined(__GNUC__) && defined(__linux__)
+
+void monstartup( char *lowpc, char *highpc );
+void _mcleanup( void );
+void glx_lowpc( void );
+
+static int profile = 0;
+
+
+/*
+ * Start profiling
+ */
+void __attribute__ ((constructor))
+glx_init_prof( void )
+{
+   FILE *fp;
+   char *s = getenv("GLX_SO_MON");
+
+   fprintf(stderr, "\n\n\nIn glx_init_prof\n\n\n");
+
+   if (!s) return;
+
+   profile = 1;
+   monstartup( (char *)glx_lowpc, (char *)glx_highpc );
+
+   fprintf(stderr, "Starting profiling, %x %x\n", 
+	   (unsigned int)glx_lowpc, 
+	   (unsigned int)glx_highpc);
+
+   if ((fp = fopen( "glx_lowpc", "w" )) != NULL) {
+      fprintf( fp, "0x%08x ", (unsigned int)glx_lowpc );
+      fclose( fp );
+   }
+}
+
+
+
+/*
+ * Finish profiling
+ */
+void __attribute__ ((destructor))
+glx_fini_prof( void )
+{
+   fprintf(stderr, "in glx_fini_prof\n");
+
+   if (profile) {
+      _mcleanup();
+      profile = 0;
+      fprintf(stderr, "Finished profiling\n");
+   }
+}
+
+
+
+
+#else
+
+void force_init_prof( void )
+{
+}
+
+#endif
diff --git a/xc/lib/GL/lowpc.c b/xc/lib/GL/lowpc.c
new file mode 100644
index 000000000..7c580a02c
--- /dev/null
+++ b/xc/lib/GL/lowpc.c
@@ -0,0 +1,4 @@
+/*
+ * This is the lowest address in Mesa
+ */
+void glx_lowpc(void) { }
diff --git a/xc/lib/GL/makeprofile.sh b/xc/lib/GL/makeprofile.sh
new file mode 100755
index 000000000..05d1b221a
--- /dev/null
+++ b/xc/lib/GL/makeprofile.sh
@@ -0,0 +1,13 @@
+#!/bin/sh
+
+RUNDIR=$1
+
+[ -f ${RUNDIR}/glx_lowpc ] || { echo "no file ${RUNDIR}/glx_lowpc" ; exit 1 } 
+[ -f ${RUNDIR}/gmon.out ] || { echo "no file ${RUNDIR}/gmon.out" ; exit 1 }
+
+rm -f glx_lowpc gmon.out glxsyms profile
+ln -s ${RUNDIR}/glx_lowpc . || exit 1
+ln -s ${RUNDIR}/gmon.out . || exit 1
+
+ld -o glxsyms -noinhibit-exec --whole-archive -Ttext=`cat glx_lowpc` libGL.a 2> /dev/null && \
+gprof glxsyms < gmon.out > profile
diff --git a/xc/lib/GL/mesa/dri/dri_mesa.c b/xc/lib/GL/mesa/dri/dri_mesa.c
index 08f3dfcfd..b2ee7e3e0 100644
--- a/xc/lib/GL/mesa/dri/dri_mesa.c
+++ b/xc/lib/GL/mesa/dri/dri_mesa.c
@@ -307,14 +307,26 @@ void driMesaUpdateDrawableInfo(Display *dpy, int scrn,
 	Xfree(pdp->pClipRects); 
     }
 
+    if (pdp->pBackClipRects) {
+	Xfree(pdp->pBackClipRects); 
+    }
+
+
     DRM_SPINUNLOCK(&psp->pSAREA->drawable_lock, psp->drawLockID);
 
     if (!XF86DRIGetDrawableInfo(dpy, scrn, pdp->draw,
 				&pdp->index, &pdp->lastStamp,
 				&pdp->x, &pdp->y, &pdp->w, &pdp->h,
-				&pdp->numClipRects, &pdp->pClipRects)) {
+				&pdp->numClipRects, &pdp->pClipRects,
+				&pdp->backX,
+				&pdp->backY,
+				&pdp->numBackClipRects,
+				&pdp->pBackClipRects
+                                )) {
 	pdp->numClipRects = 0;
 	pdp->pClipRects = NULL;
+	pdp->numBackClipRects = 0;
+	pdp->pBackClipRects = 0;
 	/* ERROR!!! */
     }
 
@@ -354,7 +366,9 @@ static void *driMesaCreateDrawable(Display *dpy, int scrn, GLXDrawable draw,
     pdp->w = 0;
     pdp->h = 0;
     pdp->numClipRects = 0;
+    pdp->numBackClipRects = 0;
     pdp->pClipRects = NULL;
+    pdp->pBackClipRects = NULL;
 
     pDRIScreen = __glXFindDRIScreen(dpy, scrn);
     pdp->driScreenPriv = psp = (__DRIscreenPrivate *)pDRIScreen->private;
diff --git a/xc/lib/GL/mesa/dri/dri_mesaint.h b/xc/lib/GL/mesa/dri/dri_mesaint.h
index 6b64114d0..16c483f01 100644
--- a/xc/lib/GL/mesa/dri/dri_mesaint.h
+++ b/xc/lib/GL/mesa/dri/dri_mesaint.h
@@ -104,6 +104,16 @@ struct __DRIdrawablePrivateRec {
     XF86DRIClipRectPtr pClipRects;
 
     /*
+    ** Information about the back and depthbuffer where different
+    ** from above.
+    */
+    int backX;
+    int backY;
+    int backClipRectType;
+    int numBackClipRects;
+    XF86DRIClipRectPtr pBackClipRects;
+
+    /*
     ** Pointer to context to which this drawable is currently bound.
     */
     __DRIcontextPrivate *driContextPriv;
diff --git a/xc/lib/GL/mesa/src/drv/Imakefile b/xc/lib/GL/mesa/src/drv/Imakefile
index fadb29bb2..306bf16dc 100644
--- a/xc/lib/GL/mesa/src/drv/Imakefile
+++ b/xc/lib/GL/mesa/src/drv/Imakefile
@@ -16,6 +16,12 @@ DRIVER += gamma
 #if GlxBuiltInTdfx
 DRIVER += tdfx
 #endif
+#if GlxBuiltInMga
+DRIVER += common mga
+#endif
+#if GlxBuiltInI810
+DRIVER += common i810
+#endif
 
 SUBDIRS = $(DRIVER)
 #else
@@ -23,6 +29,9 @@ SUBDIRS += gamma
 #if HasGlide3
 SUBDIRS += tdfx
 #endif
+SUBDIRS += common
+SUBDIRS += mga
+SUBDIRS += i810
 #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..e96291842 100644
--- a/xc/lib/GL/mesa/src/drv/common/Imakefile
+++ b/xc/lib/GL/mesa/src/drv/common/Imakefile
@@ -22,8 +22,7 @@ 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
 
@@ -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/common/depthtmp.h b/xc/lib/GL/mesa/src/drv/common/depthtmp.h
new file mode 100644
index 000000000..3aa307904
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/common/depthtmp.h
@@ -0,0 +1,133 @@
+#ifndef DBG
+#define DBG 0
+#endif
+
+
+
+static void TAG(WriteDepthSpan)( GLcontext *ctx,
+				 GLuint n, GLint x, GLint y,
+				 const GLdepth *depth, 
+				 const GLubyte mask[] )
+{
+   HW_LOCK()
+      {
+	 GLint x1;
+	 GLint n1;
+	 LOCAL_DEPTH_VARS;
+
+	 y = Y_FLIP(y);
+
+	 HW_CLIPLOOP() 
+	    {
+	       GLint i = 0;
+	       CLIPSPAN(x,y,n,x1,n1,i);
+
+	       if (DBG) fprintf(stderr, "WriteDepthSpan %d..%d (x1 %d)\n",
+				(int)i, (int)n1, (int)x1);
+
+	       if (mask)
+	       {
+		  for (;i<n1;i++,x1++)
+		     if (mask[i])
+			WRITE_DEPTH( x1, y, depth[i] );
+	       }
+	       else
+	       {
+		  for (;i<n1;i++,x1++)
+		     WRITE_DEPTH( x1, y, depth[i] );
+	       }
+	    }
+	 HW_ENDCLIPLOOP();
+      }
+   HW_UNLOCK();
+}
+
+
+static void TAG(WriteDepthPixels)( GLcontext *ctx,
+				   GLuint n, 
+				   const GLint x[], 
+				   const GLint y[],
+				   const GLdepth depth[], 
+				   const GLubyte mask[] )
+{
+   HW_LOCK()
+      {
+	 GLint i;
+	 LOCAL_DEPTH_VARS;
+
+	 if (DBG) fprintf(stderr, "WriteDepthPixels\n");
+
+	 HW_CLIPLOOP()
+	    {
+	       for (i=0;i<n;i++)
+	       {
+		  if (mask[i]) {
+		     const int fy = Y_FLIP(y[i]);
+		     if (CLIPPIXEL(x[i],fy))
+			WRITE_DEPTH( x[i], fy, depth[i] );
+		  }
+	       }
+	    }
+	 HW_ENDCLIPLOOP();
+      }
+   HW_UNLOCK();
+}
+
+
+
+
+/* Read depth spans and pixels
+ */
+static void TAG(ReadDepthSpan)( GLcontext *ctx,
+				GLuint n, GLint x, GLint y,
+				GLdepth depth[])
+{
+   HW_LOCK()
+      {
+	 GLint x1,n1;
+	 LOCAL_DEPTH_VARS;
+
+	 y = Y_FLIP(y);
+
+	 if (DBG) fprintf(stderr, "ReadDepthSpan\n");
+
+	 HW_CLIPLOOP() 
+	    {
+	       GLint i = 0;
+	       CLIPSPAN(x,y,n,x1,n1,i);
+	       for (;i<n1;i++)
+		  READ_DEPTH( depth[i], (x1+i), y );
+	    }
+	 HW_ENDCLIPLOOP();
+      }
+   HW_UNLOCK();
+}
+
+static void TAG(ReadDepthPixels)( GLcontext *ctx, GLuint n, 
+				  const GLint x[], const GLint y[],
+				  GLdepth depth[] )
+{
+   HW_LOCK()
+      {
+	 GLint i;
+	 LOCAL_DEPTH_VARS;
+
+	 if (DBG) fprintf(stderr, "ReadDepthPixels\n");
+ 
+	 HW_CLIPLOOP()
+	    {
+	       for (i=0;i<n;i++) {
+		  int fy = Y_FLIP( y[i] );
+		  if (CLIPPIXEL( x[i], fy ))
+		     READ_DEPTH( depth[i], x[i], fy );
+	       }
+	    }
+	 HW_ENDCLIPLOOP();
+      }
+   HW_UNLOCK();
+}
+
+
+#undef WRITE_DEPTH
+#undef READ_DEPTH
+#undef TAG
diff --git a/xc/lib/GL/mesa/src/drv/common/shared_texture_lru.c b/xc/lib/GL/mesa/src/drv/common/shared_texture_lru.c
new file mode 100644
index 000000000..2fd80eb3e
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/common/shared_texture_lru.c
@@ -0,0 +1,88 @@
+#include <stdio.h>
+#include "shared_texture_lru.h"
+
+/* (Re)initialize the global circular LRU list.  The last element
+ * in the array (heap->nrRegions) is the sentinal.  Keeping it
+ * at the end of the array allows the other elements of the array
+ * to be addressed rationally when looking up objects at a
+ * particular location in texture memory.  
+ */
+static void resetGlobalLRU( driHeapPtr heap )
+{
+   driTexRegion *list = heap->shared->list;
+   int sz = 1 << heap->logGranularity;
+   int i;
+
+   heap->localAge = ++heap->shared->texAge;
+
+   for (i = 0 ; (i+1) * sz <= heap->size ; i++) {
+      list[i].prev = i-1;
+      list[i].next = i+1;
+      list[i].age = heap->shared->texAge;   
+   }
+
+   i--;
+   list[0].prev = heap->nrRegions;
+   list[i].prev = i-1;
+   list[i].next = heap->nrRegions;
+   list[heap->nrRegions].prev = i;
+   list[heap->nrRegions].next = 0;
+}
+
+/* Called by the client whenever it touches a local texture.  
+ */
+void driUpdateHeap( driHeapPtr heap, int start, int end )
+{
+   driTexRegion *list = heap->shared->list;
+   int i;
+
+   heap->localAge = ++heap->shared->globalAge;
+
+   for (i = start ; i <= end ; i++) 
+   {
+      list[i].in_use = 1;
+      list[i].age = heap->localAge;
+
+      /* remove_from_list(i)
+       */
+      list[(unsigned)list[i].next].prev = list[i].prev;
+      list[(unsigned)list[i].prev].next = list[i].next;
+      
+      /* insert_at_head(list, i)
+       */
+      list[i].prev = heap->nrRegions;
+      list[i].next = list[heap->nrRegions].next;
+      list[(unsigned)list[heap->nrRegions].next].prev = i;
+      list[heap->nrRegions].next = i;
+   }
+}
+
+
+/* Called by the client on lock contention to determine whether
+ * textures have been stolen
+ */
+void driAgeTextures( driHeapPtr heap )
+{
+   driTexRegion *list = heap->shared->list;
+   int sz = 1 << (heap->logGranularity);
+   int i, nr = 0;
+
+   /* Have to go right round from the back to ensure stuff ends up
+    * LRU in the local list...  Fix with a cursor pointer.
+    */
+   for (i = list[heap->nrRegions].prev ; 
+	i != heap->nrRegions && nr < heap->nrRegions ; 
+	i = list[i].prev, nr++)
+   {
+      if (list[i].age > heap->localAge) 
+	 heap->texturesGone( heap->driverContext, heap->heapId, i * sz, sz, 1); 
+   }
+
+   /* Loop or uninitialized heap detected.  Reset.
+    */
+   if (nr == heap->nrRegions) {
+      heap->texturesGone( heap->driverContext, heap->heapId, 0, heap->size, 0);
+      resetGlobalLRU( heap );
+   }
+}
+
diff --git a/xc/lib/GL/mesa/src/drv/common/shared_texture_lru.h b/xc/lib/GL/mesa/src/drv/common/shared_texture_lru.h
new file mode 100644
index 000000000..e6cee567d
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/common/shared_texture_lru.h
@@ -0,0 +1,76 @@
+#ifndef DRI_TEX_HEAP_H
+#define DRI_TEX_HEAP_H
+
+
+/* Private struct.
+ */
+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 */
+} driTexRegion;
+
+
+/* This is the global part of the shared texture mechanism.
+ *
+ * Do not use this struct directly - declare an equivalent one with a 
+ * larger list[] array, tuned to suit your application.
+ *
+ * Your expanded struct should be placed in the driver-specific
+ * portion of the sarea.
+ */
+struct {
+	int globalAge;
+	driTexRegion list[1];	/* drivers will want to define a larger list */
+} driGlobalList; 
+
+
+/* This is the client-private part of the mechanism.
+ *
+ * Clients will place one or more of these structs in their driver
+ * context struct to manage one or more global texture heaps.  All
+ * fields except print_local_lru must be filled in.
+ */
+struct dri_tex_heap_t {
+
+	int heapId;		/* client-supplied identifier */
+	void *driverContext;	/* pointer to the client's context private */
+	int size;		/* heap size in bytes */
+	int logGranularity;	/* log base 2 of size of single heap region */
+	int nrRegions;	        /* number of elements in global list */
+	driGlobalList *shared;	/* pointer to sarea driGlobalList struct */
+	int localAge;	        /* initialize to zero */
+	
+	/* Callback to the client to let it know a region of texture
+	 * space has changed age.  The client must integrate this
+	 * information with its local texture knowledge, in particular
+	 * checking whether any of its own textures have been
+	 * invalidated.
+	 */
+	void (*textures_gone)( void *driverContext,
+			       int heapId,
+			       int offset, 
+			       int size,
+			       int inUse );
+
+	/* Optional hook for debugging.
+	 */
+	void (*print_local_lru)( void *driverContext,
+				 int heapId );
+} driTexHeap;
+
+
+#define DRI_AGE_TEXTURES( heap )			\
+   if ((heap)->localAge > (heap)->shared->globalAge)	\
+      driAgeTextures( heap );
+
+
+/* This should be called whenever there has been contention on the
+ * hardware lock.  Clients can shortcircuit this slightly by using
+ * DRI_AGE_TEXTURES, above.
+ */
+void driAgeTextures( driTexHeap *heap );
+
+#endif
+
+				
diff --git a/xc/lib/GL/mesa/src/drv/common/spantmp.h b/xc/lib/GL/mesa/src/drv/common/spantmp.h
index e10dca1a2..b475c6b18 100644
--- a/xc/lib/GL/mesa/src/drv/common/spantmp.h
+++ b/xc/lib/GL/mesa/src/drv/common/spantmp.h
@@ -1,37 +1,48 @@
+#ifndef DBG
+#define DBG 0
+#endif
 
 static void TAG(WriteRGBASpan)( const GLcontext *ctx, 
 				GLuint n, GLint x, GLint y, 
 				const GLubyte rgba[][4], 
 				const GLubyte mask[] )
 {
-   GLuint x1,n1;
-   LOCAL_VARS;
-
-   y = Y_FLIP(y);
-
-   if (DBG) fprintf(stderr, "WriteRGBASpan\n");
-
-   HW_CLIPLOOP() 
+   HW_LOCK()
       {
-	 GLuint i = 0;
-	 CLIPSPAN(x,y,n,x1,n1,i);
-	 if (mask)
-	 {
-	    for (i=0;i<n1;i++,x1++)
-	       if (mask[i])
-		  WRITE_RGBA( x1, y, 
-			      rgba[i][0], rgba[i][1], 
-			      rgba[i][2], rgba[i][3] );
-	 }
-	 else
-	 {
-	    for (i=0;i<n1;i++,x1++)
-	       WRITE_RGBA( x1, y, 
-			   rgba[i][0], rgba[i][1], 
-			   rgba[i][2], rgba[i][3] );
-	 }
+	 GLint x1;
+	 GLint n1;
+	 LOCAL_VARS;
+	 
+	 y = Y_FLIP(y);
+
+
+	 HW_CLIPLOOP() 
+	    {
+	       GLint i = 0;
+	       CLIPSPAN(x,y,n,x1,n1,i);
+
+	       if (DBG) fprintf(stderr, "WriteRGBASpan %d..%d (x1 %d)\n",
+				(int)i, (int)n1, (int)x1);
+
+	       if (mask)
+	       {
+		  for (;i<n1;i++,x1++)
+		     if (mask[i])
+			WRITE_RGBA( x1, y, 
+				    rgba[i][0], rgba[i][1], 
+				    rgba[i][2], rgba[i][3] );
+	       }
+	       else
+	       {
+		  for (;i<n1;i++,x1++)
+		     WRITE_RGBA( x1, y, 
+				 rgba[i][0], rgba[i][1], 
+				 rgba[i][2], rgba[i][3] );
+	       }
+	    }
+	 HW_ENDCLIPLOOP();
       }
-   HW_ENDCLIPLOOP();
+   HW_UNLOCK();
 }
 
 static void TAG(WriteRGBSpan)( const GLcontext *ctx,
@@ -39,32 +50,37 @@ static void TAG(WriteRGBSpan)( const GLcontext *ctx,
 			       const GLubyte rgb[][3], 
 			       const GLubyte mask[] )
 {
-   GLuint x1,n1;
-   LOCAL_VARS;
-
-   y = Y_FLIP(y);
-
-
-   if (DBG) fprintf(stderr, "WriteRGBSpan\n");
-
-   HW_CLIPLOOP() 
+   HW_LOCK()
       {
-	 GLuint i = 0;
-	 CLIPSPAN(x,y,n,x1,n1,i);
-
-	 if (mask)
-	 {
-	    for (;i<n1;i++,x1++)
-	       if (mask[i])
-		  WRITE_RGBA( x1, y, rgb[i][0], rgb[i][1], rgb[i][2], 0 );
-	 }
-	 else
-	 {
-	    for (;i<n1;i++,x1++)
-	       WRITE_RGBA( x1, y, rgb[i][0], rgb[i][1], rgb[i][2], 0 );
-	 }
+	 GLint x1;
+	 GLint n1;
+	 LOCAL_VARS;
+
+	 y = Y_FLIP(y);
+
+	 HW_CLIPLOOP() 
+	    {
+	       GLint i = 0;
+	       CLIPSPAN(x,y,n,x1,n1,i);
+
+	       if (DBG) fprintf(stderr, "WriteRGBSpan %d..%d (x1 %d)\n",
+				(int)i, (int)n1, (int)x1);
+
+	       if (mask)
+	       {
+		  for (;i<n1;i++,x1++)
+		     if (mask[i])
+			WRITE_RGBA( x1, y, rgb[i][0], rgb[i][1], rgb[i][2], 0 );
+	       }
+	       else
+	       {
+		  for (;i<n1;i++,x1++)
+		     WRITE_RGBA( x1, y, rgb[i][0], rgb[i][1], rgb[i][2], 0 );
+	       }
+	    }
+	 HW_ENDCLIPLOOP();
       }
-   HW_ENDCLIPLOOP();
+   HW_UNLOCK();
 }
 
 static void TAG(WriteRGBAPixels)( const GLcontext *ctx,
@@ -74,25 +90,29 @@ static void TAG(WriteRGBAPixels)( const GLcontext *ctx,
 			       const GLubyte rgba[][4], 
 			       const GLubyte mask[] )
 {
-   GLuint i;
-   LOCAL_VARS;
-
-   if (DBG) fprintf(stderr, "WriteRGBAPixels\n");
-
-   HW_CLIPLOOP()
+   HW_LOCK()
       {
-	 for (i=0;i<n;i++)
-	 {
-	    if (mask[i]) {
-	       const int fy = Y_FLIP(y[i]);
-	       if (CLIPPIXEL(x[i],fy))
-		  WRITE_RGBA( x[i], fy, 
-			      rgba[i][0], rgba[i][1], 
-			      rgba[i][2], rgba[i][3] );
+	 GLint i;
+	 LOCAL_VARS;
+
+	 if (DBG) fprintf(stderr, "WriteRGBAPixels\n");
+
+	 HW_CLIPLOOP()
+	    {
+	       for (i=0;i<n;i++)
+	       {
+		  if (mask[i]) {
+		     const int fy = Y_FLIP(y[i]);
+		     if (CLIPPIXEL(x[i],fy))
+			WRITE_RGBA( x[i], fy, 
+				    rgba[i][0], rgba[i][1], 
+				    rgba[i][2], rgba[i][3] );
+		  }
+	       }
 	    }
-	 }
+	 HW_ENDCLIPLOOP();
       }
-   HW_ENDCLIPLOOP();
+   HW_UNLOCK();
 }
 
 
@@ -100,23 +120,28 @@ static void TAG(WriteMonoRGBASpan)( const GLcontext *ctx,
 				    GLuint n, GLint x, GLint y, 
 				    const GLubyte mask[] )
 {
-   GLuint x1,n1;
-   LOCAL_VARS;
-   INIT_MONO_PIXEL(p);
-
-   y = Y_FLIP( y );
-
-   if (DBG) fprintf(stderr, "WriteMonoRGBASpan\n");
-
-   HW_CLIPLOOP() 
+   HW_LOCK()
       {
-	 GLuint i = 0;
-	 CLIPSPAN(x,y,n,x1,n1,i);
-	 for (;i<n1;i++,x1++)
-	    if (mask[i])
-	       WRITE_PIXEL( x1, y, p );
+	 GLint x1;
+	 GLint n1;
+	 LOCAL_VARS;
+	 INIT_MONO_PIXEL(p);
+
+	 y = Y_FLIP( y );
+
+	 if (DBG) fprintf(stderr, "WriteMonoRGBASpan\n");
+
+	 HW_CLIPLOOP() 
+	    {
+	       GLint i = 0;
+	       CLIPSPAN(x,y,n,x1,n1,i);
+	       for (;i<n1;i++,x1++)
+		  if (mask[i])
+		     WRITE_PIXEL( x1, y, p );
+	    }
+	 HW_ENDCLIPLOOP();
       }
-   HW_ENDCLIPLOOP();
+   HW_UNLOCK();
 }
 
 
@@ -125,22 +150,26 @@ static void TAG(WriteMonoRGBAPixels)( const GLcontext *ctx,
 				      const GLint x[], const GLint y[],
 				      const GLubyte mask[] ) 
 {
-   GLuint i;
-   LOCAL_VARS;
-   INIT_MONO_PIXEL(p);
-
-   if (DBG) fprintf(stderr, "WriteMonoRGBAPixels\n");
-
-   HW_CLIPLOOP()
+   HW_LOCK()
       {
-	 for (i=0;i<n;i++)
-	    if (mask[i]) {
-	       int fy = Y_FLIP(y[i]);
-	       if (CLIPPIXEL( x[i], fy ))
-		  WRITE_PIXEL( x[i], fy, p );
+	 GLint i;
+	 LOCAL_VARS;
+	 INIT_MONO_PIXEL(p);
+
+	 if (DBG) fprintf(stderr, "WriteMonoRGBAPixels\n");
+
+	 HW_CLIPLOOP()
+	    {
+	       for (i=0;i<n;i++)
+		  if (mask[i]) {
+		     int fy = Y_FLIP(y[i]);
+		     if (CLIPPIXEL( x[i], fy ))
+			WRITE_PIXEL( x[i], fy, p );
+		  }
 	    }
-      }
-   HW_ENDCLIPLOOP();
+	 HW_ENDCLIPLOOP();
+      } 
+   HW_UNLOCK();
 }
 
 
@@ -152,42 +181,50 @@ static void TAG(ReadRGBASpan)( const GLcontext *ctx,
 			       GLuint n, GLint x, GLint y,
 			       GLubyte rgba[][4])
 {
-   GLuint x1,n1;
-   LOCAL_VARS;
+   HW_LOCK()
+      {
+	 GLint x1,n1;
+	 LOCAL_VARS;
 
-   y = Y_FLIP(y);
+	 y = Y_FLIP(y);
 
-   if (DBG) fprintf(stderr, "ReadRGBASpan\n");
+	 if (DBG) fprintf(stderr, "ReadRGBASpan\n");
 
-   HW_CLIPLOOP() 
-      {
-	 GLuint i = 0;
-	 CLIPSPAN(x,y,n,x1,n1,i);
-	 for (;i<n1;i++)
-	    READ_RGBA( rgba[i], x1+i, y );
+	 HW_CLIPLOOP() 
+	    {
+	       GLint i = 0;
+	       CLIPSPAN(x,y,n,x1,n1,i);
+	       for (;i<n1;i++)
+		  READ_RGBA( rgba[i], (x1+i), y );
+	    }
+	 HW_ENDCLIPLOOP();
       }
-   HW_ENDCLIPLOOP();
+   HW_UNLOCK();
 }
 
 static void TAG(ReadRGBAPixels)( const GLcontext *ctx,
 				 GLuint n, const GLint x[], const GLint y[],
 				 GLubyte rgba[][4], const GLubyte mask[] )
 {
-   GLuint i;
-   LOCAL_VARS;
+   HW_LOCK()
+      {
+	 GLint i;
+	 LOCAL_VARS;
 
-   if (DBG) fprintf(stderr, "ReadRGBAPixels\n");
+	 if (DBG) fprintf(stderr, "ReadRGBAPixels\n");
  
-   HW_CLIPLOOP()
-      {
-	 for (i=0;i<n;i++)
-	    if (mask[i]) {
-	       int fy = Y_FLIP( y[i] );
-	       if (CLIPPIXEL( x[i], fy ))
-		  READ_RGBA( rgba[i], x[i], fy );
+	 HW_CLIPLOOP()
+	    {
+	       for (i=0;i<n;i++)
+		  if (mask[i]) {
+		     int fy = Y_FLIP( y[i] );
+		     if (CLIPPIXEL( x[i], fy ))
+			READ_RGBA( rgba[i], x[i], fy );
+		  }
 	    }
+	 HW_ENDCLIPLOOP();
       }
-   HW_ENDCLIPLOOP();
+   HW_UNLOCK();
 }
 
 
diff --git a/xc/lib/GL/mesa/src/drv/i810/Imakefile b/xc/lib/GL/mesa/src/drv/i810/Imakefile
index 5f1d93072..5c6a33796 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
@@ -10,12 +12,14 @@ ALLOC_DEFINES = -DMALLOC_0_RETURNS_NULL
 #endif
 
 #if BuildXF86DRI
-  DRI_DEFINES = GlxDefines -DFX -DFX_GLIDE3 -DGLIDE3 -DDRIVERTS
+  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$(XF86DRIVERSRC)/i810 \
-		-I../../../include -I../.. -I../../X
+		-I$(XF86DRIVERSRC)/i810  \
+		-I../../../include -I../.. -I../common -I../../X \
+	        -I$(XF86OSSRC)/linux/drm/kernel
 #endif
 
 MESA_INCLUDES = -I. -I.. -I../../include
@@ -23,20 +27,234 @@ MESA_INCLUDES = -I. -I.. -I../../include
 
 
       DEFINES = $(ALLOC_DEFINES) $(DRI_DEFINES)
-     INCLUDES = -I$(XLIBSRC) -I$(EXTINCSRC) $(MESA_INCLUDES) $(DRI_INCLUDES) \
-		-I/usr/include/glide
-      DRISRCS = i810_init.c i810_inithw.c i810_xmesa.c i810clear.c	\
-		i810context.c i810dd.c i810depth.c i810dma.c		\
+     INCLUDES = -I$(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 i810clear.c	\
+		i810dd.c  	\
 		i810pipeline.c i810span.c i810state.c i810swap.c	\
-		i810tex.c i810tris.c i810vb.c
+		i810tex.c i810tris.c i810vb.c i810fastpath.c i810ioctl.c
 
-      DRIOBJS = i810_init.o i810_inithw.o i810_xmesa.o i810clear.o	\
-		i810context.o i810dd.o i810depth.o i810dma.o		\
+     I810OBJS = i810_xmesa.o i810clear.o	\
+		i810dd.o  		\
 		i810pipeline.o i810span.o i810state.o i810swap.o	\
-		i810tex.o i810tris.o i810vb.o
+		i810tex.o i810tris.o i810vb.o i810fastpath.o i810ioctl.o
+
+     MESASRCS = ../../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.o \
+		../../light.c \
+		../../lines.c \
+		../../logic.c \
+		../../masking.c \
+		../../matrix.c \
+		../../mem.c \
+		../../mmath.c \
+		../../pb.c \
+		../../pipeline.c \
+		../../pixel.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 \
+		../../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 = ../../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 \
+		../../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 \
+		../../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
+
+      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)
 
-	 SRCS = $(DRISRCS) 
-	 OBJS = $(DRIOBJS) 
+REQUIREDLIBS += -lm
 
 #if !GlxUseBuiltInDRIDriver
 #undef DoNormalLib NormalLibGlx
@@ -57,7 +275,7 @@ 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 677a35222..e4e274e9a 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
@@ -75,7 +75,8 @@
  *     4: DR4_*
  */
 #define GFX_OP_DRAWRECT_INFO   ((0x3<<29)|(0x1d<<24)|(0x80<<16)|(0x3))
-#define DR1_RECT_CLIP_ENABLE   (0x1<<31)
+#define DR1_RECT_CLIP_ENABLE   (0x0<<31)
+#define DR1_RECT_CLIP_DISABLE  (0x1<<31)
 #define DR1_X_DITHER_BIAS_MASK (0x3<<26)
 #define DR1_X_DITHER_BIAS_SHIFT      26
 #define DR1_Y_DITHER_BIAS_MASK (0x3<<24)
@@ -107,6 +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)
@@ -238,18 +241,6 @@
 #define SDM_DST_BOTH_SRC_ALPHA     (0xc<<0)
 #define SDM_DST_BOTH_INV_SRC_ALPHA (0xd<<0)
 
-/* GFXRENDERSTATE_COLOR_CHROMA_KEY, p135
- */
-#define GFX_OP_COLOR_CHROMA_KEY  ((0x3<<29)|(0x1d<<24)|(0x2<<16)|0x1)
-#define CC1_UPDATE_CC_TEST_ENABLE  (1<<28)
-#define CC1_CC_TEST_ENABLE         (1<<27)
-#define CC1_UPDATE_CI_VALUE        (1<<26)
-#define CC1_UPDATE_CK_LOW_VALUE    (1<<25)
-#define CC1_UPDATE_CK_HI_VALUE     (1<<24)
-#define CC1_CK_LOW_VALUE_MASK      ((1<<24)-1)
-#define CC2_CI_VALUE_MASK          (0xff<<24)
-#define CC2_CI_VALUE_SHIFT         24
-#define CC2_CK_HIGH_VALUE_MASK     ((1<<24)-1)
 
 /* GFXRENDERSTATE_COLOR_FACTOR, p134
  *
@@ -628,13 +619,43 @@ typedef struct {
 #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)
+#define AA_UPDATE_POLYWIDTH      (1<<11)
+#define AA_POLYWIDTH_05          (1<<9)
+#define AA_POLYWIDTH_10          (2<<9)
+#define AA_POLYWIDTH_20          (3<<9)
+#define AA_POLYWIDTH_40          (4<<9)
+#define AA_UPDATE_LINEWIDTH      (1<<8)
+#define AA_LINEWIDTH_05          (1<<6)
+#define AA_LINEWIDTH_10          (2<<6)
+#define AA_LINEWIDTH_20          (3<<6)
+#define AA_LINEWIDTH_40          (4<<6)
+#define AA_UPDATE_BB_EXPANSION   (1<<5)
+#define AA_BB_EXPANSION_SHIFT    2
+#define AA_UPDATE_AA_ENABLE      (1<<1)
+#define AA_ENABLE                (1<<0)
+
+#define GFX_OP_STIPPLE           ((0x3<<29)|(0x1d<<24)|(0x83<<16))
+#define ST1_ENABLE               (1<<16)
+#define ST1_MASK                 (0xffff)
+
+
 /* 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 (includes zbuffer).  Drawrect is the only thing
- * that needs to be restored for 2d???
+/* 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
@@ -667,14 +688,24 @@ typedef struct {
 #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_SCI0  15 	/* GFX_OP_SCISSOR_INFO (3 dwords) */
-#define I810_CTXREG_SCI1  16	
-#define I810_CTXREG_SCI2  17	
-#define I810_CTXREG_SC    18	/* GFX_OP_SCISSOR */
-#define I810_CTXREG_PV    19	/* GFX_OP_PV_RULE -- Invarient! */
-#define I810_CTXREG_ZA    20	/* GFX_OP_ZBIAS_ALPHAFUNC */
-
-#define I810_CTX_SETUP_SIZE (21+1) /* pad to qword */
+#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)
  */
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810_init.c b/xc/lib/GL/mesa/src/drv/i810/i810_init.c
deleted file mode 100644
index 5f37cd769..000000000
--- a/xc/lib/GL/mesa/src/drv/i810/i810_init.c
+++ /dev/null
@@ -1,99 +0,0 @@
-/**************************************************************************
-
-Copyright 1998-1999 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, 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 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:
- *   Keith Whitwell <keithw@precisioninsight.com>
- *
- *   based on tdfx driver by
- *      Daryll Strauss <daryll@precisioninsight.com>
- *
- * $PI: #
- */
-
-#ifdef GLX_DIRECT_RENDERING
-
-#include <X11/Xlibint.h>
-#include "xf86dri.h"
-#include "i810_dri.h"
-#include "i810_init.h"
-
-#ifdef DEBUG_LOCKING
-char *prevLockFile=0;
-int prevLockLine=0;
-#endif
-
-static void performMagic(__DRIscreenPrivate *driScrnPriv)
-{
-    i810ScreenPrivate *gPriv = (i810ScreenPrivate *)driScrnPriv->private;
-    I810DRIPtr         gDRIPriv = (I810DRIPtr)driScrnPriv->pDevPriv;
-
-    gPriv->regs.handle=gDRIPriv->regs;
-    gPriv->regs.size=gDRIPriv->regsSize;
-    gPriv->deviceID=gDRIPriv->deviceID;
-    gPriv->width=gDRIPriv->width;
-    gPriv->height=gDRIPriv->height;
-    gPriv->mem=gDRIPriv->mem;
-    gPriv->cpp=gDRIPriv->cpp;
-    gPriv->stride=gDRIPriv->stride;
-    gPriv->fbOffset=gDRIPriv->fbOffset;
-    gPriv->backOffset=gDRIPriv->backOffset;
-    gPriv->depthOffset=gDRIPriv->depthOffset;
-    gPriv->textureOffset=gDRIPriv->textureOffset;
-    gPriv->textureSize=gDRIPriv->textureSize;
-}
-
-GLboolean i810MapAllRegions(__DRIscreenPrivate *driScrnPriv)
-{
-    i810ScreenPrivate *gPriv = (i810ScreenPrivate *)driScrnPriv->private;
-    
-    /* First, pick apart pDevPriv & friends */
-    performMagic(driScrnPriv);
-
-    if (drmMap(driScrnPriv->fd, gPriv->regs.handle, gPriv->regs.size, 
-	       &gPriv->regs.map)) {
-      return GL_FALSE;
-    }
-
-    return GL_TRUE;
-}
-
-void i810UnmapAllRegions(__DRIscreenPrivate *driScrnPriv)
-{
-    i810ScreenPrivate *gPriv = (i810ScreenPrivate *)driScrnPriv->private;
-
-    drmUnmap(gPriv->regs.map, gPriv->regs.size);
-}
-
-/*
- * Shutdown hardware - nothing to do...
- */
-void i810CloseHardware(void)
-{
-}
-
-#endif
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810_init.h b/xc/lib/GL/mesa/src/drv/i810/i810_init.h
index c7af9c443..804f66889 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810_init.h
+++ b/xc/lib/GL/mesa/src/drv/i810/i810_init.h
@@ -27,7 +27,7 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 /*
  * Authors:
- *   Daryll Strauss <daryll@precisioninsight.com>
+ *   Keith Whitwell <keithw@precisioninsight.com>
  *
  */
 
@@ -37,7 +37,6 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #ifdef GLX_DIRECT_RENDERING
 
 #include <sys/time.h>
-#include <glide.h>
 #include "dri_tmm.h"
 #include "dri_mesaint.h"
 #include "dri_mesa.h"
@@ -45,148 +44,88 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "xmesaP.h"
 
 typedef struct {
-  drmHandle handle;
-  drmSize size;
-  drmAddress map;
+   drmHandle handle;
+   drmSize size;
+   char *map;
 } i810Region, *i810RegionPtr;
 
 typedef struct {
-  i810Region regs;
-  int deviceID;
-  int width;
-  int height;
-  int mem;
-  int cpp;
-  int stride;
-  int fbOffset;
-  int backOffset;
-  int depthOffset;
-  int textureOffset;
-  int textureSize;
-  __DRIscreenPrivate *driScrnPriv;
-} i810ScreenPrivate;
+   i810Region front;
+   i810Region back;
+   i810Region depth;
+   i810Region tex;
 
-typedef struct {
-  volatile int *fifoPtr;
-  volatile int *fifoRead;
-  volatile int fifoOwner;
-  volatile int ctxOwner;
-  volatile int texOwner;
-} I810SAREAPriv;
+   int deviceID;
+   int width;
+   int height;
+   int mem;
+
+   int cpp;			/* for front and back buffers */
+   int bitsPerPixel;
+
+   int fbFormat;
+   int fbOffset;
+   int fbStride;
+
+   int backOffset;
+   int depthOffset;
+
+   int backPitch;
+   int backPitchBits;
 
+   int textureOffset;
+   int textureSize;
+   int logTextureGranularity;
 
+   __DRIscreenPrivate *driScrnPriv;
+   drmBufMapPtr  bufs;
+
+} i810ScreenPrivate;
 
 
 #include "i810context.h"
 
-extern GLboolean i810MapAllRegions(__DRIscreenPrivate *driScrnPriv);
-extern void i810UnmapAllRegions(__DRIscreenPrivate *driScrnPriv);
-extern GLboolean i810InitHW(XMesaContext c);
-
-extern void XMesaWindowMoved(XMesaContext *c);
-extern void XMesaUpdateState(XMesaContext *c, int windowMoved);
-extern void XMesaSetSAREA(XMesaContext *c);
-
-extern XMesaContext gCC;
-extern i810ContextPtr i810Ctx;
-
-/* 
-You can turn this on to find locking conflicts.
-#define DEBUG_LOCKING 
-*/
-
-#ifdef 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
+extern void i810GetLock( i810ContextPtr imesa, GLuint flags );
+extern void i810EmitHwStateLocked( i810ContextPtr imesa );
+extern void i810EmitScissorValues( i810ContextPtr imesa, int box_nr, int emit );
+extern void i810EmitDrawingRectangle( i810ContextPtr imesa );
+extern void i810XMesaSetBackClipRects( i810ContextPtr imesa );
+extern void i810XMesaSetFrontClipRects( i810ContextPtr imesa );
 
-/* !!! 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 !!! */
-
-#define DRM_LIGHT_LOCK_RETURN(fd,lock,context,__ret)                   \
-	do {                                                           \
-		DRM_CAS(lock,context,DRM_LOCK_HELD|context,__ret);     \
-                if (__ret) drmGetLock(fd,context,0);                   \
-        } while(0)
-
-/* Lock the hardware using the global current context */
-#define LOCK_HARDWARE() \
-  do { \
-    int stamp; \
-    char __ret=0; \
-    __DRIdrawablePrivate *dPriv = gCC->driContextPriv->driDrawablePriv; \
-    __DRIscreenPrivate *sPriv = dPriv->driScreenPriv; \
-    DEBUG_CHECK_LOCK(); \
-    DRM_LIGHT_LOCK_RETURN(sPriv->fd, &sPriv->pSAREA->lock, \
-		   dPriv->driContextPriv->hHWContext, __ret); \
-    if (__ret) __ret=1; \
-    stamp=dPriv->lastStamp; \
-    XMESA_VALIDATE_DRAWABLE_INFO(gCC->display, sPriv, dPriv); \
-    if (*(dPriv->pStamp)!=stamp) __ret=2; \
-    if (__ret) XMesaUpdateState(gCC, __ret==2); \
-    DEBUG_LOCK(); \
-  } while (0)
 
-/* Unlock the hardware using the global current context */
-#define UNLOCK_HARDWARE() \
-  do { \
-    __DRIdrawablePrivate *dPriv = gCC->driContextPriv->driDrawablePriv; \
-    __DRIscreenPrivate *sPriv = dPriv->driScreenPriv; \
-    XMesaSetSAREA(gCC); \
-    DRM_UNLOCK(sPriv->fd, &sPriv->pSAREA->lock,  \
-	       dPriv->driContextPriv->hHWContext); \
-    DEBUG_RESET(); \
-  } while (0)
+#define GET_DISPATCH_AGE( imesa ) imesa->sarea->last_dispatch
+#define GET_ENQUEUE_AGE( imesa ) imesa->sarea->last_enqueue
 
-#define BEGIN_BOARD_LOCK() LOCK_HARDWARE()
-#define END_BOARD_LOCK() UNLOCK_HARDWARE()
 
-/*
-  This pair of macros makes a loop over the drawing operations
-  so it is not self contained and doesn't have the nice single 
-  statement semantics of most macros
-*/
-#define BEGIN_CLIP_LOOP()	\
-  do {				\
-    __DRIdrawablePrivate *dPriv = gCC->driContextPriv->driDrawablePriv; \
-    int _nc = dPriv->numClipRects; \
-    LOCK_HARDWARE(); \
-    while (_nc--) { \
-      if (i810Ctx->needClip) { \
-        i810Ctx->clipMinX=dPriv->pClipRects[_nc].x1; \
-        i810Ctx->clipMaxX=dPriv->pClipRects[_nc].x2; \
-        i810Ctx->clipMinY=dPriv->pClipRects[_nc].y1; \
-        i810Ctx->clipMaxY=dPriv->pClipRects[_nc].y2; \
-        i810SetScissorValues(i810Ctx->glCtx); \
-      }
-
-#define END_CLIP_LOOP() \
-    } \
-    UNLOCK_HARDWARE(); \
+/* Lock the hardware and validate our state.  
+ */
+#define LOCK_HARDWARE( imesa )				\
+  do {							\
+    char __ret=0;					\
+    DRM_CAS(imesa->driHwLock, imesa->hHWContext,	\
+	    (DRM_LOCK_HELD|imesa->hHWContext), __ret);	\
+    if (__ret)						\
+        i810GetLock( imesa, 0 );			\
   } while (0)
 
+
+
+/* Unlock the hardware using the global current context 
+ */
+#define UNLOCK_HARDWARE(imesa)					\
+    DRM_UNLOCK(imesa->driFd, imesa->driHwLock, imesa->hHWContext);	
+
+
+/* This is the wrong way to do it, I'm sure.  Otherwise the drm
+ * bitches that I've already got the heavyweight lock.  At worst,
+ * this is 3 ioctls.  The best solution probably only gets me down 
+ * to 2 ioctls in the worst case.
+ */
+#define LOCK_HARDWARE_QUIESCENT( imesa ) do {	\
+   LOCK_HARDWARE( imesa );			\
+   i810RegetLockQuiescent( imesa );		\
+} while(0)
+     
+
 #endif
 #endif
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810_inithw.c b/xc/lib/GL/mesa/src/drv/i810/i810_inithw.c
deleted file mode 100644
index 6bcae44c7..000000000
--- a/xc/lib/GL/mesa/src/drv/i810/i810_inithw.c
+++ /dev/null
@@ -1,99 +0,0 @@
-/**************************************************************************
-
-Copyright 1998-1999 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, 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 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:
- *   Daryll Strauss <daryll@precisioninsight.com>
- *
- * $PI: $
- */
-
-#ifdef GLX_DIRECT_RENDERING
-
-#include "i810_init.h"
-#include <glide.h>
-
-GLboolean i810InitHW(XMesaContext c)
-{
-  /* KW: Would be nice to make one of these a member of the other.
-   */
-  i810ContextPtr cPriv = (i810ContextPtr)c->private;
-  __DRIdrawablePrivate *driDrawPriv = c->driContextPriv->driDrawablePriv;
-  __DRIscreenPrivate *driScrnPriv = driDrawPriv->driScreenPriv;
-  i810ScreenPrivate *sPriv = (i810ScreenPrivate*)driScrnPriv->private;
-
-#ifdef DEBUG_LOCKING
-  fprintf(stderr, "Debug locking enabled\n");
-#endif
-    
-  if (cPriv->initDone) return GL_TRUE;
-
-  cPriv->width=driDrawPriv->w;
-  cPriv->height=driDrawPriv->h;
-
-  DRM_LIGHT_LOCK(driScrnPriv->fd, &driScrnPriv->pSAREA->lock, 
-		 driDrawPriv->driContextPriv->hHWContext);
-  FX_grGlideInit_NoLock();
-
-
-  if (sPriv->deviceID==0x3)
-    cPriv->haveTwoTMUs=GL_FALSE;
-  else
-    cPriv->haveTwoTMUs=GL_TRUE;
-
-  /* !!! We are forcing these !!! */
-  cPriv->haveDoubleBuffer=GL_TRUE; 
-  cPriv->haveAlphaBuffer=GL_FALSE; 
-  cPriv->haveZBuffer=GL_TRUE;
-  cPriv->haveGlobalPaletteTexture=GL_FALSE;
-
-  cPriv->glideContext =  FX_grSstWinOpen_NoLock((FxU32)-1, GR_RESOLUTION_NONE,
-						 GR_REFRESH_NONE,
-						 GR_COLORFORMAT_ARGB,
-						 GR_ORIGIN_LOWER_LEFT, 2, 1);
-
-
-  grDRIResetSAREA();
-
-  DRM_UNLOCK(driScrnPriv->fd, &driScrnPriv->pSAREA->lock, 
-	     driDrawPriv->driContextPriv->hHWContext);
-
-  cPriv->needClip=1;
-
-  if (!fxDDInitFxMesaContext( fxMesa ))
-    return GL_FALSE;
-
-  cPriv->initDone=GL_TRUE;
-  return GL_TRUE;
-
-}
-
-#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 d4b540507..7716b4634 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810_xmesa.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810_xmesa.c
@@ -27,21 +27,54 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 /*
  * Authors:
- *   Daryll Strauss <daryll@precisioninsight.com>
+ *   Keith Whitwell <keithw@precisioninsight.com>
  *
  */
 
 #ifdef GLX_DIRECT_RENDERING
 
 #include <X11/Xlibint.h>
-#include <glide.h>
+#include <stdio.h>
+
 #include "i810_init.h"
 #include "context.h"
 #include "vbxform.h"
 #include "matrix.h"
+#include "simple_list.h"
+
+#include "i810dd.h"
+#include "i810state.h"
+#include "i810tex.h"
+#include "i810span.h"
+#include "i810tris.h"
+#include "i810swap.h"
+#include "i810pipeline.h"
+#include "i810ioctl.h"
+
+#include "i810_dri.h"
+
+
+
+#ifndef I810_DEBUG
+int I810_DEBUG = (0
+/*     		  | DEBUG_ALWAYS_SYNC  */
+/*  		  | DEBUG_VERBOSE_RING    */
+/*  		  | 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
+
+
+static i810ContextPtr      i810Ctx = 0;
 
-XMesaContext        gCC     = 0;
-i810ContextPtr     *i810Ctx = 0;
+i810Glx_t	i810glx;
 
 static int count_bits(unsigned int n)
 {
@@ -76,32 +109,106 @@ static int count_bits(unsigned int n)
  */
 
 
-GLboolean XMesaInitDriver(__DRIscreenPrivate *driScrnPriv)
+GLboolean XMesaInitDriver(__DRIscreenPrivate *sPriv)
 {
-   i810ScreenPrivate *gsp;
+   i810ScreenPrivate *i810Screen;
+   I810DRIPtr         gDRIPriv = (I810DRIPtr)sPriv->pDevPriv;
 
    /* Allocate the private area */
-   gsp = (i810ScreenPrivate *)Xmalloc(sizeof(i810ScreenPrivate));
-   if (!gsp) return GL_FALSE;
+   i810Screen = (i810ScreenPrivate *)Xmalloc(sizeof(i810ScreenPrivate));
+   if (!i810Screen) return GL_FALSE;
+
+   i810Screen->driScrnPriv = sPriv;
+   sPriv->private = (void *)i810Screen;
+
+   i810Screen->deviceID=gDRIPriv->deviceID;
+   i810Screen->width=gDRIPriv->width;
+   i810Screen->height=gDRIPriv->height;
+   i810Screen->mem=gDRIPriv->mem;
+   i810Screen->cpp=gDRIPriv->cpp;
+   i810Screen->fbStride=gDRIPriv->fbStride;
+   i810Screen->fbOffset=gDRIPriv->fbOffset;
+
+   if (gDRIPriv->bitsPerPixel == 15) 
+      i810Screen->fbFormat = DV_PF_555;
+   else
+      i810Screen->fbFormat = DV_PF_565;
+
+   i810Screen->backOffset=gDRIPriv->backOffset; 
+   i810Screen->depthOffset=gDRIPriv->depthOffset;
+   i810Screen->backPitch = gDRIPriv->auxPitch;
+   i810Screen->backPitchBits = gDRIPriv->auxPitchBits;
+   i810Screen->textureOffset=gDRIPriv->textureOffset;
+   i810Screen->textureSize=gDRIPriv->textureSize;
+   i810Screen->logTextureGranularity = gDRIPriv->logTextureGranularity;
+
+   if (1)
+      fprintf(stderr, "Tex heap size %x, granularity %x bytes\n",
+	      i810Screen->textureSize, 1<<(i810Screen->logTextureGranularity));
+
+   i810Screen->bufs = drmMapBufs(sPriv->fd);
+    
+
+   i810Screen->back.handle = gDRIPriv->backbuffer;
+   i810Screen->back.size = gDRIPriv->backbufferSize;
+
+   if (drmMap(sPriv->fd, 
+	      i810Screen->back.handle, 
+	      i810Screen->back.size, 
+	      (drmAddress *)&i810Screen->back.map) != 0) 
+   {
+      Xfree(i810Screen);
+      return GL_FALSE;
+   }
+
+   i810Screen->depth.handle = gDRIPriv->depthbuffer;
+   i810Screen->depth.size = gDRIPriv->depthbufferSize;
 
-   gsp->driScrnPriv = driScrnPriv;
+   if (drmMap(sPriv->fd, 
+	      i810Screen->depth.handle, 
+	      i810Screen->depth.size, 
+	      (drmAddress *)&i810Screen->depth.map) != 0) 
+   {
+      Xfree(i810Screen);
+      return GL_FALSE;
+   }
 
-   driScrnPriv->private = (void *)gsp;
+   i810Screen->tex.handle = gDRIPriv->textures;
+   i810Screen->tex.size = gDRIPriv->textureSize;
 
-   if (!i810MapAllRegions(driScrnPriv)) {
-      Xfree(driScrnPriv->private);
+   if (drmMap(sPriv->fd, 
+	      i810Screen->tex.handle, 
+	      i810Screen->tex.size, 
+	      (drmAddress *)&i810Screen->tex.map) != 0) 
+   {
+      Xfree(i810Screen);
       return GL_FALSE;
    }
 
+
+   /* Ditch i810glx in favor of i810Screen?
+    */
+   memset(&i810glx, 0, sizeof(i810glx));
+   i810glx.texVirtual = i810Screen->tex.map;
+   
+   i810DDFastPathInit();
+   i810DDTrifuncInit();
+   i810DDSetupInit();
+
    return GL_TRUE;
 }
 
 /* Accessed by dlsym from dri_mesa_init.c
  */
-void XMesaResetDriver(__DRIscreenPrivate *driScrnPriv)
+void XMesaResetDriver(__DRIscreenPrivate *sPriv)
 {
-   i810UnmapAllRegions(driScrnPriv); /* in i810_init.c */
-   Xfree(driScrnPriv->private);
+   i810ScreenPrivate *i810Screen = (i810ScreenPrivate *)sPriv->private;
+
+   /* Need to unmap all the bufs and maps here:
+    */
+
+
+   Xfree(i810Screen);
 }
 
 /* Accessed by dlsym from dri_mesa_init.c
@@ -135,27 +242,18 @@ XMesaVisual XMesaCreateVisual(XMesaDisplay *display,
 
    v->display = display;
    v->level = level;
-  
-   v->gl_visual = (GLvisual *)Xmalloc(sizeof(GLvisual));
+   v->gl_visual = gl_create_visual(rgb_flag, GL_FALSE, db_flag, stereo_flag,
+				   depth_size, stencil_size, accum_size, 0,
+				   count_bits(visinfo->red_mask),
+				   count_bits(visinfo->green_mask),
+				   count_bits(visinfo->blue_mask),
+				   GL_FALSE);
    if (!v->gl_visual) {
       Xfree(v->visinfo);
-      XFree(v);
-      return 0;
+      Xfree(v);
+      return NULL;
    }
-
-   v->gl_visual->RGBAflag   = rgb_flag;
-   v->gl_visual->DBflag     = db_flag;
-   v->gl_visual->StereoFlag = stereo_flag;
-
-   v->gl_visual->RedBits   = count_bits(visinfo->red_mask);
-   v->gl_visual->GreenBits = count_bits(visinfo->green_mask);
-   v->gl_visual->BlueBits  = count_bits(visinfo->blue_mask);
-   v->gl_visual->AlphaBits = 0; /* Not currently supported */
-
-   v->gl_visual->AccumBits   = accum_size;
-   v->gl_visual->DepthBits   = depth_size;
-   v->gl_visual->StencilBits = stencil_size;
-
+  
    return v;
 }
 
@@ -169,71 +267,143 @@ void XMesaDestroyVisual(XMesaVisual v)
 XMesaContext XMesaCreateContext(XMesaVisual v, XMesaContext share_list,
 				__DRIcontextPrivate *driContextPriv)
 {
+   GLcontext *ctx;
    XMesaContext c;
-   i810ContextPtr cPriv;
-   __DRIscreenPrivate *driScrnPriv = driContextPriv->driScreenPriv;
-   i810ScreenPrivate *sPriv = (i810ScreenPrivate *)driScrnPriv->private;
-   I810SAREAPriv *saPriv;
-   GLcontext *shareCtx;
-/*     int **fifoPtr; */
+   i810ContextPtr imesa;
+   __DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv;
+   i810ScreenPrivate *i810Screen = (i810ScreenPrivate *)sPriv->private;
+   drm_i810_sarea_t *saPriv=(drm_i810_sarea_t *)(((char*)sPriv->pSAREA)+
+						 sizeof(XF86DRISAREARec));
+
+   GLcontext *shareCtx = 0;
 
    c = (XMesaContext)Xmalloc(sizeof(struct xmesa_context));
-   if (!c) return 0;
+   if (!c) {
+      return 0;
+   }
+
+   imesa = (i810ContextPtr)Xcalloc(sizeof(i810Context), 1);
+   if (!imesa) {
+      Xfree(c);
+      return 0;
+   }
 
    c->driContextPriv = driContextPriv;
    c->xm_visual = v;
    c->xm_buffer = 0; /* Set by MakeCurrent */
    c->display = v->display;
+   c->private = (void *)imesa;
 
-   cPriv = (i810ContextPtr)Xmalloc(sizeof(i810Context));
-   if (!cPriv) {
-      Xfree(c);
-      return NULL;
-   }
+   if (share_list)
+      shareCtx=((i810ContextPtr)(share_list->private))->glCtx;
 
-   cPriv->hHWContext = driContextPriv->hHWContext;
-   cPriv->i810ScrnPriv = sPriv;
-   c->private = (void *)cPriv;
+   ctx = imesa->glCtx = gl_create_context(v->gl_visual, shareCtx, 
+					  (void*)imesa, GL_TRUE);
 
-   cPriv->glVis=v->gl_visual;
-   cPriv->glBuffer=gl_create_framebuffer(v->gl_visual);
-   cPriv->screen_width=sPriv->width;
-   cPriv->screen_height=sPriv->height;
 
-   cPriv->new_state = ~0;
+   /* Set the maximum texture size small enough that we can guarentee
+    * that both texture units can bind a maximal texture and have them
+    * in memory at once.
+    */
+   if (i810Screen->textureSize < 2*1024*1024) {
+      ctx->Const.MaxTextureLevels = 8;
+      ctx->Const.MaxTextureSize = 1<<8;
+   } else if (i810Screen->textureSize < 8*1024*1024) {
+      ctx->Const.MaxTextureLevels = 9;
+      ctx->Const.MaxTextureSize = 1<<9;     
+   } else {
+      ctx->Const.MaxTextureLevels = 10;
+      ctx->Const.MaxTextureSize = 1<<10;
+   }      
 
-   if (share_list)
-      shareCtx=((i810ContextPtr)(share_list->private))->glCtx;
-   else
-      shareCtx=0;
-   cPriv->glCtx=gl_create_context(v->gl_visual, shareCtx, (void*)cPriv, GL_TRUE);
-   cPriv->initDone=GL_FALSE;
 
-   saPriv=(I810SAREAPriv*)((void*)driScrnPriv->pSAREA+sizeof(XF86DRISAREARec));
+   /* Dri stuff
+    */
+   imesa->display = v->display;
+   imesa->hHWContext = driContextPriv->hHWContext;
+   imesa->driFd = sPriv->fd;
+   imesa->driHwLock = &sPriv->pSAREA->lock;
 
+   imesa->i810Screen = i810Screen;
+   imesa->driScreen = sPriv;
+   imesa->sarea = saPriv;
 
-   fprintf(stderr, "would call grDRIOpen here\n");
+   imesa->glBuffer = gl_create_framebuffer(v->gl_visual,
+					   GL_FALSE,
+					   v->gl_visual->StencilBits > 0,
+					   v->gl_visual->AccumBits > 0,
+					   v->gl_visual->AlphaBits > 0);
 
 
-/*
-  grDRIOpen(driScrnPriv->pFB, sPriv->regs.map, sPriv->deviceID, 
-  sPriv->width, sPriv->height, sPriv->mem, sPriv->cpp, sPriv->stride,
-  sPriv->fifoOffset, sPriv->fifoSize, sPriv->fbOffset,
-  sPriv->backOffset, sPriv->depthOffset, sPriv->textureOffset, 
-  sPriv->textureSize, &saPriv->fifoPtr, &saPriv->fifoRead);
-*/
+   imesa->texHeap = mmInit( 0, i810Screen->textureSize );
+
+
+   /* Utah stuff
+    */
+   imesa->renderindex = -1;		/* impossible value */
+   imesa->new_state = ~0;
+   imesa->dirty = ~0;
+
+   make_empty_list(&imesa->TexObjList);
+   make_empty_list(&imesa->SwappedOut);
+   
+   imesa->TextureMode = ctx->Texture.Unit[0].EnvMode;
+   imesa->CurrentTexObj[0] = 0;
+   imesa->CurrentTexObj[1] = 0;
+   
+   i810DDExtensionsInit( ctx );
+
+   i810DDInitStateFuncs( ctx );
+   i810DDInitTextureFuncs( ctx );
+   i810DDInitSpanFuncs( ctx );
+   i810DDInitDriverFuncs( ctx );
+   i810DDInitIoctlFuncs( ctx );
+
+   ctx->Driver.TriangleCaps = (DD_TRI_CULL|
+			       DD_TRI_LIGHT_TWOSIDE|
+			       DD_TRI_STIPPLE|
+			       DD_TRI_OFFSET);
+
+   /* Ask mesa to clip fog coordinates for us.
+    */
+   ctx->TriangleCaps |= DD_CLIP_FOG_COORD;
+
+   ctx->Shared->DefaultD[2][0].DriverData = 0;
+   ctx->Shared->DefaultD[2][1].DriverData = 0;
+
+   if (ctx->VB) 
+      i810DDRegisterVB( ctx->VB );
+
+   if (ctx->NrPipelineStages)
+      ctx->NrPipelineStages =
+	 i810DDRegisterPipelineStages(ctx->PipelineStage,
+				      ctx->PipelineStage,
+				      ctx->NrPipelineStages);
+
+   i810DDInitState( imesa );
 
    return c;
 }
 
 void XMesaDestroyContext(XMesaContext c)
 {
-   i810ContextPtr cPriv;
+   i810ContextPtr imesa = (i810ContextPtr) c->private;
+
+   if (imesa) {
+      i810TextureObjectPtr next_t, t;
 
-   cPriv=(i810ContextPtr)c->private;
-   if (cPriv) {
-      gl_destroy_context(cPriv->glCtx);
-      gl_destroy_framebuffer(cPriv->glBuffer);
+      gl_destroy_context(imesa->glCtx);
+      gl_destroy_framebuffer(imesa->glBuffer);
+
+      foreach_s (t, next_t, &(imesa->TexObjList))
+	 i810DestroyTexObj(imesa, t);
+
+      foreach_s (t, next_t, &(imesa->SwappedOut))
+	 i810DestroyTexObj(imesa, t);
+
+      Xfree(imesa);
+
+      c->private = 0;
    }
 }
 
@@ -254,137 +424,232 @@ void XMesaDestroyBuffer(XMesaBuffer b)
 {
 }
 
-void XMesaSwapBuffers(XMesaBuffer b)
+void XMesaSwapBuffers(XMesaBuffer bogus)
 {
-   FxI32 result;
-   XMesaContext *xmc = b->xm_context;
-   i810ContextPtr imesa;
-   
-   if (!xmc || !xmc->private) return;
+   i810ContextPtr imesa = i810Ctx; 
 
-   imesa = (i810ContextPtr) xmc->private;
+   FLUSH_VB( imesa->glCtx, "swap buffers" );
+   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;
+}
 
-   fprintf(stderr, "imesa %x i810Ctx %x\n", imesa, i810Ctx );
 
-   FLUSH_VB( imesa->glCtx, "swap buffers" );
 
-   if (imesa->haveDoubleBuffer) {
-      i810BackToFront(imesa);
-      i810Ctx->stats.swapBuffer++;
+void i810XMesaSetFrontClipRects( i810ContextPtr imesa )
+{
+   __DRIdrawablePrivate *dPriv = imesa->driDrawable;
+
+   imesa->numClipRects = dPriv->numClipRects;
+   imesa->pClipRects = dPriv->pClipRects;
+   imesa->drawX = dPriv->x;
+   imesa->drawY = dPriv->y;
+
+   imesa->drawMap = imesa->driScreen->pFB;
+   i810EmitDrawingRectangle( imesa );
+   i810InitClipRects( imesa );
+}
+
+
+void i810XMesaSetBackClipRects( i810ContextPtr imesa )
+{
+   __DRIdrawablePrivate *dPriv = imesa->driDrawable;
+   int i;
+
+   if (dPriv->numBackClipRects == 0) 
+   {
+      if (I810_DEBUG & DEBUG_VERBOSE_DRI)
+	 fprintf(stderr, "FRONT_CLIPRECTS, %d rects\n", 
+		 dPriv->numClipRects);
+
+      imesa->numClipRects = dPriv->numClipRects;
+      imesa->pClipRects = dPriv->pClipRects;
+      imesa->drawX = dPriv->x;
+      imesa->drawY = dPriv->y;
    } else {
-      fprintf(stderr, "No double buffer\n");
+      if (I810_DEBUG & DEBUG_VERBOSE_DRI)
+	 fprintf(stderr, "BACK_RECTS, %d rects\n", 
+		 dPriv->numBackClipRects);
+
+      imesa->numClipRects = dPriv->numBackClipRects;
+      imesa->pClipRects = dPriv->pBackClipRects;
+      imesa->drawX = dPriv->backX;
+      imesa->drawY = dPriv->backY;
    }
+
+   imesa->drawMap = imesa->i810Screen->back.map;
+   i810EmitDrawingRectangle( imesa );
+   i810InitClipRects( imesa );
+
+   if (I810_DEBUG & DEBUG_VERBOSE_DRI)
+      for (i = 0 ; i < imesa->numClipRects ; i++) 
+	 fprintf(stderr, "cliprect %d: %d,%d - %d,%d\n",
+		 i,
+		 imesa->pClipRects[i].x1,
+		 imesa->pClipRects[i].y1,
+		 imesa->pClipRects[i].x2,
+		 imesa->pClipRects[i].y2);
+}
+
+
+static void i810XMesaWindowMoved( i810ContextPtr imesa ) 
+{
+   if (0)
+   fprintf(stderr, "i810XMesaWindowMoved\n\n");
+
+   switch (imesa->glCtx->Color.DriverDrawBuffer) {
+   case GL_FRONT_LEFT:
+      i810XMesaSetFrontClipRects( imesa );
+      break;
+   case GL_BACK_LEFT:
+      i810XMesaSetBackClipRects( imesa );
+      break;
+   default:
+      fprintf(stderr, "fallback buffer\n");
+      break;
+   }
+}
+
+
+GLboolean XMesaUnbindContext(XMesaContext c)
+{
+   if (c->private)
+      ((i810ContextPtr)c->private)->dirty = ~0;
+
+   return GL_TRUE;
 }
 
-/* This looks buggy to me:
+
+/* 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(XMesaContext c, XMesaBuffer b)
 {
-   __DRIdrawablePrivate *driDrawPriv;
+
+   if (c->private==(void *)i810Ctx) return GL_TRUE;
 
    if (c) {
-      if (c==gCC) return GL_TRUE;
-      gCC     = c;
-      i810Ctx = (i810ContextPtr)c->private;
+      __DRIdrawablePrivate *dPriv = c->driContextPriv->driDrawablePriv;
 
-      fprintf(stderr, "Make current\n");
-      driDrawPriv = c->driContextPriv->driDrawablePriv;
-      if (!i810Ctx->initDone) {
-	 i810Ctx->width=driDrawPriv->w;
-	 i810Ctx->height=driDrawPriv->h;
-	 if (!i810InitHW(c)) return GL_FALSE;
-	 /* Zap the width to force XMesaWindowMoved to execute */
-	 i810Ctx->width=0;
-      }
-      FX_grSstSelect(i810Ctx->board);
-      XMesaWindowMoved();
+      i810Ctx = (i810ContextPtr)c->private;
 
-      FX_grGlideSetState((GrState*)i810Ctx->state);
       gl_make_current(i810Ctx->glCtx, i810Ctx->glBuffer);
-      fxSetupDDPointers(i810Ctx->glCtx);
+
+      i810Ctx->driDrawable = dPriv;
+      i810Ctx->dirty = ~0;
+   
+      i810XMesaWindowMoved( i810Ctx );
+
       if (!i810Ctx->glCtx->Viewport.Width)
-	 gl_Viewport(i810Ctx->glCtx, 0, 0, driDrawPriv->w, driDrawPriv->h);
+	 gl_Viewport(i810Ctx->glCtx, 0, 0, dPriv->w, dPriv->h);
+
    } else {
       gl_make_current(0,0);
-      gCC     = NULL;
       i810Ctx = NULL;
    }
    return GL_TRUE;
 }
 
-void i810XMesaWindowMoved( XMesaContext xmc ) 
+
+void i810GetLock( i810ContextPtr imesa, GLuint flags ) 
 {
-   i810ContextPtr imesa = (i810ContextPtr) xmc->private;
-   __DRIdrawablePrivate *dPriv = xmc->driContextPriv->driDrawablePriv;
-   GLcontext *ctx=imesa->glCtx;
-
-/*     grDRIPosition(dPriv->x, dPriv->y, dPriv->w, dPriv->h, */
-/*  		 dPriv->numClipRects, dPriv->pClipRects); */
-
-
-   imesa->numClipRects=dPriv->numClipRects;
-   imesa->pClipRects=dPriv->pClipRects;
-   if (dPriv->x!=imesa->x_offset || dPriv->y!=imesa->y_offset ||
-       dPriv->w!=imesa->width || dPriv->h!=imesa->height) {
-      imesa->x_offset=dPriv->x;
-      imesa->y_offset=dPriv->y;
-      imesa->width=dPriv->w;
-      imesa->height=dPriv->h;
-      imesa->y_delta=imesa->screen_height-imesa->y_offset-imesa->height;
+   __DRIdrawablePrivate *dPriv = imesa->driDrawable;
+   __DRIscreenPrivate *sPriv = imesa->driScreen;
+   drm_i810_sarea_t *sarea = imesa->sarea;
+   int me = imesa->hHWContext;
+   int stamp = dPriv->lastStamp; 
+
+
+   if (0) fprintf(stderr, ".\n");
+
+   /* We know there has been contention.
+    */
+   drmGetLock(imesa->driFd, imesa->hHWContext, flags);	
+
+
+   /* Note contention for throttling hint
+    */
+   imesa->any_contend = 1;
+
+   /* If the window moved, may need to set a new cliprect now.
+    *
+    * NOTE: This releases and regains the hw lock, so all state
+    * checking must be done *after* this call:
+    */
+   XMESA_VALIDATE_DRAWABLE_INFO(imesa->display, sPriv, dPriv);		
+
+
+   if (0)
+   fprintf(stderr, "i810GetLock, last enque: %d last dispatch: %d\n",
+	   sarea->last_enqueue, 
+	   sarea->last_dispatch);
+
+   /* If we lost context, need to dump all registers to hardware.
+    * Note that we don't care about 2d contexts, even if they perform
+    * accelerated commands, so the DRI locking in the X server is even
+    * 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;
+      sarea->ctxOwner = me;
    }
-   switch (dPriv->numClipRects) {
-   case 0:
-      imesa->clipMinX=dPriv->x;
-      imesa->clipMaxX=dPriv->x+dPriv->w;
-      imesa->clipMinY=dPriv->y;
-      imesa->clipMaxY=dPriv->y+dPriv->h;
-      fxSetScissorValues(ctx);
-      imesa->needClip=0;
-      break;
-   case 1:
-      imesa->clipMinX=dPriv->pClipRects[0].x1;
-      imesa->clipMaxX=dPriv->pClipRects[0].x2;
-      imesa->clipMinY=dPriv->pClipRects[0].y1;
-      imesa->clipMaxY=dPriv->pClipRects[0].y2;
-      fxSetScissorValues(ctx);
-      imesa->needClip=0;
-      break;
-   default:
-      imesa->needClip=1;
-   }
-}
 
-void i810XMesaUpdateState(XMesaContext *xmc, int windowMoved) {
-  __DRIdrawablePrivate *dPriv = xmc->driContextPriv->driDrawablePriv;
-  __DRIscreenPrivate *sPriv = dPriv->driScreenPriv;
-  I810SAREAPriv *saPriv=(I810SAREAPriv*)(((char*)sPriv->pSAREA)+sizeof(XF86DRISAREARec));
-
-  /* fprintf(stderr, "In FifoPtr=%d FifoRead=%d\n", saPriv->fifoPtr, saPriv->fifoRead); */
-  if (saPriv->fifoOwner!=dPriv->driContextPriv->hHWContext) {
-    grDRIImportFifo(saPriv->fifoPtr, saPriv->fifoRead);
-  }
-  if (saPriv->ctxOwner!=dPriv->driContextPriv->hHWContext) {
-    grDRIInvalidateAll();
-    /* This shouldn't be needed if the state is reset by Glide */
-    /* FX_grConstantColorValue_NoLock(FXCOLOR4(&i810Ctx->constColor)); */
-  }
-  if (saPriv->texOwner!=dPriv->driContextPriv->hHWContext) {
-    fxTMRestoreTextures(i810Ctx);
-  }
-  if (windowMoved)
-    XMesaWindowMoved( xmc );
-}
+   /* Shared texture managment - if another client has played with
+    * texture space, figure out which if any of our textures have been
+    * ejected, and update our global LRU.
+    */
+   if (sarea->texAge != imesa->texAge) {
+      int sz = 1 << (imesa->i810Screen->logTextureGranularity);
+      int idx, nr = 0;
+
+      /* Have to go right round from the back to ensure stuff ends up
+       * LRU in our local list...
+       */
+      for (idx = sarea->texList[I810_NR_TEX_REGIONS].prev ; 
+	   idx != I810_NR_TEX_REGIONS && nr < I810_NR_TEX_REGIONS ; 
+	   idx = sarea->texList[idx].prev, nr++)
+      {
+	 if (sarea->texList[idx].age > imesa->texAge)
+	    i810TexturesGone(imesa, idx * sz, sz, sarea->texList[idx].in_use);
+      }
 
-void i810XMesaSetSAREA( XMesaContext *xmc ) {
-  __DRIdrawablePrivate *dPriv = xmc->driContextPriv->driDrawablePriv;
-  __DRIscreenPrivate *sPriv = dPriv->driScreenPriv;
-  I810SAREAPriv *saPriv=(I810SAREAPriv*)(((char*)sPriv->pSAREA)+sizeof(XF86DRISAREARec));
+      if (nr == I810_NR_TEX_REGIONS) {
+	 i810TexturesGone(imesa, 0, imesa->i810Screen->textureSize, 0);
+	 i810ResetGlobalLRU( imesa );
+      }
+
+      if (0) fprintf(stderr, "imesa %d sarea %d\n", imesa->texAge, sarea->texAge);
+      imesa->dirty |= I810_UPLOAD_TEX0IMAGE;
+      imesa->dirty |= I810_UPLOAD_TEX1IMAGE;
+      imesa->texAge = sarea->texAge;
+   }
 
-  saPriv->fifoOwner=dPriv->driContextPriv->hHWContext;
-  saPriv->ctxOwner=dPriv->driContextPriv->hHWContext;
-  saPriv->texOwner=dPriv->driContextPriv->hHWContext;
-  grDRIResetSAREA();
-  /* fprintf(stderr, "Out FifoPtr=%d FifoRead=%d\n", saPriv->fifoPtr, saPriv->fifoRead); */
+   
+   if (dPriv->lastStamp != stamp)
+      i810XMesaWindowMoved( imesa );
+
+   
+   sarea->last_quiescent = -1;	/* just kill it for now */
 }
 
+
 #endif
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810buf.h b/xc/lib/GL/mesa/src/drv/i810/i810buf.h
deleted file mode 100644
index e72b0f225..000000000
--- a/xc/lib/GL/mesa/src/drv/i810/i810buf.h
+++ /dev/null
@@ -1,90 +0,0 @@
-/*
- * GLX Hardware Device Driver for Intel 810
- * 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.
- *
- *    Keith Whitwell <keithw@precisioninsight.com>
- *
- * based on matrox driver by
- *    Wittawat Yamwong <Wittawat.Yamwong@stud.uni-hannover.de>
- * and others.
- */
-
-#ifndef I810BUF_INC
-#define I810BUF_INC
-
-#include "i810common.h"
-#include "i810_3d_reg.h"
-#include "types.h"
-#include "mm.h"
-
-#define VALID_I810_BUFFER(b) (b)
-
-
-struct i810_dest_buffer 
-{
-   int refcount;
-   PMemBlock MemBlock;
-   GLuint Setup[I810_DEST_SETUP_SIZE];
-
-   int Format;
-   int Width;			/* in pixels */
-   int Height; 
-   int Pitch;  
-   int BytesPerPixel;
-   int Size;
-   int Drawable;
-   char *BufAddr;
-   struct i810_z_buffer *ZBuffer;
-};
-
-struct i810_z_buffer 
-{
-   int refcount;
-   PMemBlock MemBlock;
-
-   int Format;
-   char *BufAddr;
-
-   int Pitch;
-};
-
-
-
-/* Don't bother trying to hide the differences between these types of
- * buffer.
- */
-struct i810_dest_buffer *i810CreateDestBuffer(int Format, 
-					      int Width, 
-					      int Height);
-void i810DestroyDestBuffer(struct i810_dest_buffer *buf);
-
-
-
-/* Parent is a dest-buffer, supplies geometry info.
- */
-struct i810_z_buffer *i810CreateZBuffer( struct i810_dest_buffer *parent );
-void i810FreeZBuffer(struct i810_z_buffer *buf);
-
-
-
-
-
-#endif
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810clear.c b/xc/lib/GL/mesa/src/drv/i810/i810clear.c
index 3749b9659..f6e74beb8 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810clear.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810clear.c
@@ -1,144 +1,145 @@
 
 
-#include "xsmesaP.h"
-#include "mesaglx/types.h"
+#include "types.h"
 #include "vbrender.h"
-#include "glx_log.h"
-
-#include "vga.h"
-#include "vgaPCI.h"
+#include "i810log.h"
 
 #include <stdio.h>
 #include <stdlib.h>
+#include <unistd.h>
 
 #include "mm.h"
 #include "i810lib.h"
 #include "i810dd.h"
-#include "i810glx.h"
 #include "i810clear.h"
 #include "i810state.h"
 #include "i810tris.h"
+#include "i810ioctl.h"
+
+#define DEPTH_SCALE 65535.0F
+
+static void i810_clear_buffer( GLcontext *ctx, 
+			       GLboolean all,
+			       GLint cx, GLint cy, 
+			       GLint cwidth, 
+			       GLint cheight,
+			       GLushort value,
+			       GLuint offset ) 
+{
+   i810ContextPtr imesa = I810_CONTEXT( ctx );
+   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 = (offset +  
+		      y * i810Screen->backPitch + 
+		      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->backPitch );
+	 OUT_BATCH( (height << 16) | (width * i810Screen->cpp));
+	 OUT_BATCH( start );
+	 OUT_BATCH( value );
+	 OUT_BATCH( 0 );
+
+	 ADVANCE_BATCH();
+      }
+   }
+}
+
 
 
 
 /*
  * i810Clear
- * perform hardware accelerated clearing of the color and/or depth
- * buffer.  Software may still clear stencil buffers.
- * If all==GL_TRUE, clear whole buffer, else just clear region defined
- * by x,y,width,height
+ *
+ * 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 x, GLint y, GLint width, GLint height ) 
+		      GLint cx, GLint cy, GLint cwidth, GLint cheight ) 
 {
-   XSMesaContext xsmesa = (XSMesaContext) ctx->DriverCtx;
-   GLubyte	*c;
-   GLuint	zval;
+   i810ContextPtr imesa = I810_CONTEXT( ctx );
+   const GLuint colorMask = *((GLuint *) &ctx->Color.ColorMask);
 
-   i810Msg(10, "i810Clear( %i, %i, %i, %i, %i )\n",
-	   mask, x, y, width, height );
+   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 );
 	
-   CHECK_CONTEXT( return mask; );
-
-   if (all == GL_TRUE) {
-      x = 0;
-      y = 0;
-      width = i810DB->Width;
-      height = i810DB->Height;
-   }
-
-   if ( y + height > i810DB->Height ) {
-      height = i810DB->Height - y;
-   }
-   if ( x + width > i810DB->Width ) {
-      width = i810DB->Width - x;
-   }
-   if ( x < 0 ) {
-      width += x;
-      x = 0;
-   }
-   if ( y < 0 ) {
-      height += y;
-      y = 0;
-   }
-   if ( x >= i810DB->Width || y >= i810DB->Height || width < 1 || height < 1 ) {
-      return 0;
-   }
-
-   /* flip top to bottom */
-   y = i810DB->Height-y-height;
-   c = xsmesa->clearcolor;
-   zval = (GLuint) (ctx->Depth.Clear * DEPTH_SCALE);
-
-   i810FinishPrimitive();
+   FLUSH_BATCH(imesa);
+   
+   LOCK_HARDWARE(imesa);
+   i810GetGeneralDmaBufferLocked( imesa );
 
-   if (1)
    {
-      BEGIN_BATCH(2);   
+      BEGIN_BATCH( imesa, 2 );   
       OUT_BATCH( INST_PARSER_CLIENT | INST_OP_FLUSH );
-      OUT_BATCH( 0 );		/* pad to quadword */
+      OUT_BATCH( 0 );
       ADVANCE_BATCH();      
    }
 
-   /* color buffer */
-   if (mask & GL_COLOR_BUFFER_BIT) 
-   {
-      int start = (i810DB->MemBlock->ofs +  
-		   y * i810DB->Pitch + 
-		   x * vgaBytesPerPixel);
-
-      BEGIN_BATCH(6);
-   
-      OUT_BATCH( BR00_BITBLT_CLIENT | BR00_OP_COLOR_BLT | 0x3 );
-      OUT_BATCH( BR13_SOLID_PATTERN | 
-		 (0xF0 << 16) |
-		 i810DB->Pitch );
-      OUT_BATCH( (height << 16) | (width * vgaBytesPerPixel));
-      OUT_BATCH( start );
-      OUT_BATCH( i810PackColor( i810DB->Format, c[0], c[1], c[2], c[3] ) );
-      OUT_BATCH( 0 );
-
-      ADVANCE_BATCH();
-      
-      mask &= ~GL_COLOR_BUFFER_BIT;
+   if ( (mask & DD_FRONT_LEFT_BIT) && colorMask == ~0) {
+      i810_clear_buffer( ctx, all, cx, cy, cwidth, cheight,
+			 imesa->ClearColor,
+			 imesa->i810Screen->fbOffset);
+      mask &= ~DD_FRONT_LEFT_BIT;
    }
 
-   /* depth buffer */
-   if ( (mask & GL_DEPTH_BUFFER_BIT) && 
-	i810DB->ZBuffer && 
-	ctx->Depth.Mask ) 
-   {
-      int start = (i810DB->ZBuffer->MemBlock->ofs + 
-		   y * i810DB->ZBuffer->Pitch + 
-		   x * 2);
-
-      BEGIN_BATCH(6);
-   
-      OUT_BATCH( BR00_BITBLT_CLIENT | BR00_OP_COLOR_BLT | 0x3 );
-      OUT_BATCH( BR13_SOLID_PATTERN | 
-		 (0xF0 << 16) |
-		 i810DB->ZBuffer->Pitch );
-      OUT_BATCH( (height << 16) | (width * 2));
-      OUT_BATCH( start );
-      OUT_BATCH( zval );
-      OUT_BATCH( 0 );
-
-      ADVANCE_BATCH();   
-
-      mask &= ~GL_DEPTH_BUFFER_BIT;
+   if ( (mask & DD_BACK_LEFT_BIT) && colorMask == ~0) {
+      i810_clear_buffer( ctx, all, cx, cy, cwidth, cheight, 
+			 imesa->ClearColor,
+			 imesa->i810Screen->backOffset);
+      mask &= ~DD_BACK_LEFT_BIT;
    }
 
-   if (1)
+   if ( (mask & DD_DEPTH_BIT) && ctx->Depth.Mask ) {
+      i810_clear_buffer( ctx, all, cx, cy, cwidth, cheight, 
+			 ctx->Depth.Clear * DEPTH_SCALE,
+			 imesa->i810Screen->depthOffset);
+      mask &= ~DD_DEPTH_BIT;
+   }      
+
    {
-      BEGIN_BATCH(2);   
+      BEGIN_BATCH( imesa, 2 );   
       OUT_BATCH( INST_PARSER_CLIENT | INST_OP_FLUSH );
-      OUT_BATCH( 0 );		/* pad to quadword */
+      OUT_BATCH( 0 );
       ADVANCE_BATCH();      
    }
 
+   i810FlushGeneralLocked(imesa);
+   UNLOCK_HARDWARE( imesa );
    return mask;
 }
 
-
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810context.c b/xc/lib/GL/mesa/src/drv/i810/i810context.c
deleted file mode 100644
index a9bd855d5..000000000
--- a/xc/lib/GL/mesa/src/drv/i810/i810context.c
+++ /dev/null
@@ -1,126 +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.
- *
- */
-
-#include "xsmesaP.h"
-#include <stdlib.h>
-#include <stdio.h>
-#include <time.h>
-#include <GL/gl.h>
-
-#include "i810context.h"
-#include "i810lib.h"
-#include "i810log.h"
-#include "i810dd.h"
-#include "i810pipeline.h"
-/*  #include "i810render.h" */
-#include "i810state.h"
-#include "i810tris.h"
-#include "simple_list.h"
-
-
-
-i810ContextPtr i810CreateContext(GLcontext *ctx)
-{
-   i810ContextPtr c;
-
-   c = (i810ContextPtr) calloc(1,sizeof(i810Context));
-   if (!c)
-      return NULL;
-
-   c->gl_ctx = ctx;
-
-   c->renderindex = -1;		/* impossible value */
-   c->new_state = ~0;
-   c->reg_dirty = ~0;
-
-   make_empty_list(&c->SwappedOut);
-   make_empty_list(&c->TexObjList);
-   
-   c->TextureMode = ctx->Texture.Unit[0].EnvMode;
-
-   c->CurrentTex0Obj = 0;
-   c->CurrentTex1Obj = 0;
-   ctx->Shared->DefaultD[2][0].DriverData = 0;
-   ctx->Shared->DefaultD[2][1].DriverData = 0;
-
-   if (ctx->VB) 
-      i810DDRegisterVB( ctx->VB );
-
-   if (ctx->NrPipelineStages)
-      ctx->NrPipelineStages =
-	 i810DDRegisterPipelineStages(ctx->PipelineStage,
-				      ctx->PipelineStage,
-				      ctx->NrPipelineStages);
-
-   
-   /* Ask mesa to clip fog coordinates for us.
-    */
-   ctx->TriangleCaps |= DD_CLIP_FOG_COORD;
-
-
-#if 0
-   if (!getenv("I810_NO_FAST_PATH")) 
-      ctx->Driver.BuildPrecalcPipeline = i810DDBuildPrecalcPipeline;
-
-   i810DDRenderInit();
-   i810DDFastPathInit();
-#endif
-
-
-   i810DDTrifuncInit();
-   i810DDSetupInit();
-   i810DDExtensionsInit( ctx );
-
-   i810_setup_DD_pointers( ctx );
-   i810DDInitState( c );
-
-   if (MESA_VERBOSE&VERBOSE_DRIVER)
-      fprintf(stderr,"i810CreateContext(): successful.\n");
-
-   return c;
-}
-
-int i810DestroyContext(i810ContextPtr ctx)
-{
-   i810TextureObjectPtr next_t, t;
-
-   if (!ctx)
-      return 0;
-
-   if (--(ctx->refcount) > 0)
-      return 0;
-
-   foreach_s (t, next_t, &(ctx->TexObjList))
-      i810DestroyTexObj(ctx, t);
-
-   foreach_s (t, next_t, &(ctx->SwappedOut))
-      i810DestroyTexObj(ctx, t);
-
-   free(ctx);
-
-   if (MESA_VERBOSE&VERBOSE_DRIVER)
-      fprintf(stderr,"i810DestroyContext(): successfully destroyed.\n");
-
-   return 0;
-}
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810context.h b/xc/lib/GL/mesa/src/drv/i810/i810context.h
index 754a0e3f1..6a7ffa968 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810context.h
+++ b/xc/lib/GL/mesa/src/drv/i810/i810context.h
@@ -26,60 +26,82 @@
 #ifndef I810CONTEXT_INC
 #define I810CONTEXT_INC
 
+typedef struct i810_context_t i810Context;
+typedef struct i810_context_t *i810ContextPtr;
+
+#include <X11/Xlibint.h>
+#include "dri_tmm.h"
+#include "dri_mesaint.h"
+#include "dri_mesa.h"
+#include "xmesaP.h"
 
-#include "mesaglx/types.h"
+#include "types.h"
+#include "i810_init.h"
+#include "drm.h"
 
-#include "i810common.h"
-#include "i810buf.h"
 #include "i810tex.h"
 #include "i810vb.h"
 
-#include "xsmesaP.h"
 
-/* Hmmm */
-#define FALLBACK_TEXTURE   1
+
+#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 
+
 
 
+/* for i810ctx.new_state - manage GL->driver state changes
+ */
+#define I810_NEW_TEXTURE 0x1
+
+/* for i810ctx.dirty - manage driver->hw state changes, including
+ * lost contexts.
+ */
+#define I810_WAIT_AGE          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
 
-#define I810_NEW_TEXTURE 0x20
-#define I810_NEW_CONTEXT 0x100
 
+typedef void (*i810_interp_func)( GLfloat t, 
+				  GLfloat *result,
+				  const GLfloat *in,
+				  const GLfloat *out );
+				 
 
 
 struct i810_context_t {
    GLint refcount;
 
-   struct i810_dest_buffer *DB;
-   GLcontext *gl_ctx;
+   GLcontext *glCtx;
 
-   i810TextureObjectPtr CurrentTex0Obj;
-   i810TextureObjectPtr CurrentTex1Obj;
+   i810TextureObjectPtr CurrentTexObj[2];
 
-   struct i810_texture_object_t TexObjList, SwappedOut;
+   struct i810_texture_object_t TexObjList;
+   struct i810_texture_object_t SwappedOut; 
 
    int TextureMode;
 
-   /* flags 
-    */
-   int GlobalPaletteUpdated:1; 
-
-
    /* Hardware state
     */
    GLuint Setup[I810_CTX_SETUP_SIZE];
+   GLuint BufferSetup[I810_DEST_SETUP_SIZE];
+   GLuint ClipSetup[I810_CLIP_SETUP_SIZE];
    
 
-   /* Context hardware state mirrors.
-    */
-   GLushort GlobalPalette[256];
-
    /* Support for CVA and the fast paths.
     */
    GLuint setupdone;
    GLuint setupindex;
    GLuint renderindex;
    GLuint using_fast_path;
-/*     gl_vertex_interp_func interp; */
+   i810_interp_func interp;
 
    /* Shortcircuit some state changes.
     */
@@ -90,39 +112,98 @@ struct i810_context_t {
 
    /* Manage our own state */
    GLuint new_state; 
-   GLuint reg_dirty;
-   
+
+   /* Manage hardware state */
+   GLuint dirty;
+   memHeap_t *texHeap;
+
    /* One of the few bits of hardware state that can't be calculated
     * completely on the fly:
     */
    GLuint LcsCullMode;
 
-   /* For mono span and pixel funcs.
+   /* Funny mesa mirrors
     */
-   GLushort CurrentColor;
+   GLushort MonoColor;
+   GLushort ClearColor;
 
    /* DRI stuff
     */
-   GLuint needClip;
-   GLuint clipMinX;
-   GLuint clipMaxX;
-   GLuint clipMinY;
-   GLuint clipMaxY;
+   drmBufPtr  dma_buffer;
+   drmBufPtr  vertex_dma_buffer;
+
+   GLframebuffer *glBuffer;
    
+   /* Two flags to keep track of fallbacks.
+    */
+   GLuint IndirectTriangles;
+   GLuint Fallback;
 
-   drmContext hHWContext;
-   int numClipRects;
+
+   GLuint needClip;
+
+   /* These refer to the current draw (front vs. back) buffer:
+    */
+   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;
-   i810ScreenPrivate *i810ScrnPriv; 
+
+   int lastSwap;
+   int secondLastSwap;
+   int texAge;
+   int ctxAge;
+   int dirtyAge;
+   int any_contend;		/* throttle me harder */
+
+   int scissor;
+
+   int lines_are_tris;		/* not used */
+   int horiz_line_stipple;
+   int vert_line_stipple;
+
+   int poly_stipple;
+
+   drm_clip_rect_t draw_rect;
+   drm_clip_rect_t scissor_rect;
+
+   drmContext hHWContext;
+   drmLock *driHwLock;
+   int driFd;
+   Display *display;
+
+   __DRIdrawablePrivate *driDrawable;
+   __DRIscreenPrivate *driScreen;
+   i810ScreenPrivate *i810Screen; 
+   drm_i810_sarea_t *sarea;
 };
 
-typedef struct i810_context_t i810Context;
-typedef struct i810_context_t *i810ContextPtr;
+
+/* To remove all debugging, make sure I810_DEBUG is defined as a
+ * preprocessor symbol, and equal to zero.  
+ */
+#define I810_DEBUG 0   
+#ifndef I810_DEBUG
+#warning "Debugging enabled - expect reduced performance"
+extern int I810_DEBUG;
+#endif
+
+#define DEBUG_VERBOSE_2D     0x1
+#define DEBUG_VERBOSE_RING   0x8
+#define DEBUG_VERBOSE_OUTREG 0x10
+#define DEBUG_ALWAYS_SYNC    0x40
+#define DEBUG_VERBOSE_MSG    0x80
+#define DEBUG_NO_OUTRING     0x100
+#define DEBUG_NO_OUTREG      0x200
+#define DEBUG_VERBOSE_API    0x400
+#define DEBUG_VALIDATE_RING  0x800
+#define DEBUG_VERBOSE_LRU    0x1000
+#define DEBUG_VERBOSE_DRI    0x2000
+#define DEBUG_VERBOSE_IOCTL  0x4000
 
 
-i810ContextPtr i810CreateContext(GLcontext *ctx); /* Do I need visual info? */
-int i810DestroyContext(i810ContextPtr ctx);
-int i810BindBuffer(i810ContextPtr ctx, struct i810_dest_buffer *buf);
 
 extern GLuint i810DDRegisterPipelineStages( struct gl_pipeline_stage *out,
 					   const struct gl_pipeline_stage *in,
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810dd.c b/xc/lib/GL/mesa/src/drv/i810/i810dd.c
index 7a762ce9b..954f60a79 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810dd.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810dd.c
@@ -30,17 +30,15 @@
 
 #include "mm.h"
 #include "i810lib.h"
-#include "i810glx.h"
 #include "i810clear.h"
 #include "i810dd.h"
-#include "i810direct.h"
-#include "i810depth.h"
 #include "i810log.h"
 #include "i810state.h"
 #include "i810span.h"
 #include "i810tex.h"
 #include "i810tris.h"
 #include "i810vb.h"
+#include "i810pipeline.h"
 #include "extensions.h"
 #include "vb.h"
 #include "dd.h"
@@ -54,18 +52,39 @@ extern int xf86VTSema;
  ***************************************/
 
 
-const GLubyte *i810DDGetString( GLcontext *ctx, GLenum name )
+static const GLubyte *i810DDGetString( GLcontext *ctx, GLenum name )
 {
    switch (name) {
    case GL_VENDOR:
       return "Keith Whitwell, Precision Insight Inc.";
    case GL_RENDERER:
-      return "GLX-I810";
+      return "DRI-I810";
    default:
       return 0;
    }
 }
 
+static GLint i810GetParameteri(const GLcontext *ctx, GLint param)
+{
+   switch (param) {
+   case DD_HAVE_HARDWARE_FOG:  
+      return 1;
+   default:
+      return 0; 
+   }
+}
+
+
+
+static void i810BufferSize(GLcontext *ctx, GLuint *width, GLuint *height)
+{
+   i810ContextPtr imesa = I810_CONTEXT(ctx);  
+   *width = imesa->driDrawable->w;
+   *height = imesa->driDrawable->h;
+}
+
+
+
 
 void i810DDExtensionsInit( GLcontext *ctx )
 {
@@ -86,77 +105,26 @@ void i810DDExtensionsInit( GLcontext *ctx )
    gl_extensions_disable( ctx, "GL_INGR_blend_func_separate" );
 
 
+   if (0) gl_extensions_disable( ctx, "GL_ARB_multitexture" );
+
+
    /* We do support tex_env_add, however
     */
    gl_extensions_enable( ctx, "GL_EXT_texture_env_add" );
 }
 
 
-static GLint i810GetParameteri(const GLcontext *ctx, GLint param)
-{
-   switch (param) {
-   case DD_HAVE_HARDWARE_FOG:  
-      return 1;
-   default:
-      return 0; 
-   }
-}
 
 
-GLboolean i810IsTextureResident( GLcontext *ctx, struct gl_texture_object *t )
+void i810DDInitDriverFuncs( GLcontext *ctx )
 {
-   i810TextureObjectPtr mt = (i810TextureObjectPtr)t->DriverData;
-   return mt && mt->MemBlock;
-}
-
-
-
-
-void i810_setup_DD_pointers( GLcontext *ctx )
-{
-   i810DDInitStatePointers(ctx);
-   
+   ctx->Driver.GetBufferSize = i810BufferSize;
    ctx->Driver.GetString = i810DDGetString;
-   ctx->Driver.UpdateState = i810DDUpdateState;
+   ctx->Driver.GetParameteri = i810GetParameteri;
    ctx->Driver.RegisterVB = i810DDRegisterVB;
    ctx->Driver.UnregisterVB = i810DDUnregisterVB;
    ctx->Driver.Clear = i810Clear;
-   ctx->Driver.GetParameteri = i810GetParameteri;
-   ctx->Driver.TexEnv = i810TexEnv;
-   ctx->Driver.TexImage = i810TexImage;
-   ctx->Driver.TexSubImage = i810TexSubImage;
-   ctx->Driver.BindTexture = i810BindTexture;
-   ctx->Driver.DeleteTexture = i810DeleteTexture;
-   ctx->Driver.TexParameter = i810TexParameter;
-   ctx->Driver.UpdateTexturePalette = i810UpdateTexturePalette;
-   ctx->Driver.IsTextureResident = i810IsTextureResident;
-
-   ctx->Driver.TriangleCaps = (DD_TRI_CULL|
-			       DD_TRI_LIGHT_TWOSIDE|
-			       DD_TRI_OFFSET);
-
-   i810DDInitSpans( ctx );
-
-   ctx->Driver.ReadDepthSpanFloat = i810_read_depth_span_float;
-   ctx->Driver.ReadDepthSpanInt = i810_read_depth_span_int;
-   if (ctx->Depth.Test) {
-      switch (ctx->Depth.Func) {
-      case GL_LESS:
-	 ctx->Driver.DepthTestSpan = i810_depth_test_span_less;
-	 ctx->Driver.DepthTestPixels = i810_depth_test_pixels_less;
-	 break;
-      case GL_GREATER:
-	 ctx->Driver.DepthTestSpan = i810_depth_test_span_greater;
-	 ctx->Driver.DepthTestPixels = i810_depth_test_pixels_greater;
-	 break;
-      default:
-	 ctx->Driver.DepthTestSpan = i810_depth_test_span_generic;
-	 ctx->Driver.DepthTestPixels = i810_depth_test_pixels_generic;
-	 break;
-      }
-   }
 
-   /* Also do the normal GL state-change checks.
-    */
-   i810DDUpdateState( ctx );
+
+   ctx->Driver.BuildPrecalcPipeline = i810DDBuildPrecalcPipeline;
 }
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810dd.h b/xc/lib/GL/mesa/src/drv/i810/i810dd.h
index 1274e66e2..86a8618d0 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810dd.h
+++ b/xc/lib/GL/mesa/src/drv/i810/i810dd.h
@@ -25,23 +25,9 @@
 #ifndef I810DD_INC
 #define I810DD_INC
 
-#include "mesaglx/context.h"
+#include "context.h"
 
-#include "i810buf.h"
-
-void i810_setup_DD_pointers( GLcontext *ctx );
-void i810DDRegisterVB( struct vertex_buffer *VB );
-
-void i810DDFastPath( struct vertex_buffer *VB );
-void i810DDImmediateFastPath( struct vertex_buffer *VB );
-void i810DDFastPathInit();
 void i810DDExtensionsInit( GLcontext *ctx );
-
-void i810_setup_DD_pointers_no_accel( GLcontext *ctx );
-/*  void i810FinishPrimitive( void ); */
-
-#define i810FinishPrimitive()
-
-
+void i810DDInitDriverFuncs( GLcontext *ctx );
 
 #endif
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810depth.c b/xc/lib/GL/mesa/src/drv/i810/i810depth.c
deleted file mode 100644
index 9b089f7de..000000000
--- a/xc/lib/GL/mesa/src/drv/i810/i810depth.c
+++ /dev/null
@@ -1,735 +0,0 @@
-/*
- * Mesa 3-D graphics library
- * Version:  3.1
- * 
- * Copyright (C) 1999  Brian Paul   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 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
- * BRIAN PAUL 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.
- */
-
-/*
- * This file has been modified by Wittawat Yamwong for GLX module.
- */
-
-/*
- * Depth buffer functions
- */
-
-
-#ifdef PC_HEADER
-#include "all.h"
-#else
-#include <stdlib.h>
-#include <string.h>
-#include "mesaglx/context.h"
-#include "i810depth.h"
-#include "mesaglx/types.h"
-#include "mm.h"
-#include "i810lib.h"
-#endif
-
-
-
-/*
- * Return the address of the Z-buffer value for window coordinate (x,y):
- */
-#define Z_SETUP							\
-  GLdepth *zbstart = (GLdepth *)i810DB->ZBuffer->BufAddr;	\
-  GLint zbpitch = i810DB->ZBuffer->Pitch;			
-#define Z_ADDRESS( X, Y )  \
-  (zbstart + zbpitch * (Y) + (X))
-
-
-/**********************************************************************/
-/*****                   Depth Testing Functions                  *****/
-/**********************************************************************/
-
-
-/*
- * Depth test horizontal spans of fragments.  These functions are called
- * via ctx->Driver.depth_test_span only.
- *
- * Input:  n - number of pixels in the span
- *         x, y - location of leftmost pixel in span in window coords
- *         z - array [n] of integer depth values
- * In/Out:  mask - array [n] of flags (1=draw pixel, 0=don't draw) 
- * Return:  number of pixels which passed depth test
- */
-
-
-/*
- * glDepthFunc( any ) and glDepthMask( GL_TRUE or GL_FALSE ).
- */
-GLuint i810_depth_test_span_generic( GLcontext* ctx,
-                                   GLuint n, GLint x, GLint y,
-                                   const GLdepth z[],
-                                   GLubyte mask[] )
-{
-   Z_SETUP
-   GLdepth *zptr = Z_ADDRESS( x, y );
-   GLubyte *m = mask;
-   GLuint i;
-   GLuint passed = 0;
-
-  i810WaitDrawingEngine();
-
-   /* switch cases ordered from most frequent to less frequent */
-   switch (ctx->Depth.Func) {
-      case GL_LESS:
-         if (ctx->Depth.Mask) {
-	    /* Update Z buffer */
-	    for (i=0; i<n; i++,zptr++,m++) {
-	       if (*m) {
-		  if (z[i] < *zptr) {
-		     /* pass */
-		     *zptr = z[i];
-		     passed++;
-		  }
-		  else {
-		     /* fail */
-		     *m = 0;
-		  }
-	       }
-	    }
-	 }
-	 else {
-	    /* Don't update Z buffer */
-	    for (i=0; i<n; i++,zptr++,m++) {
-	       if (*m) {
-		  if (z[i] < *zptr) {
-		     /* pass */
-		     passed++;
-		  }
-		  else {
-		     *m = 0;
-		  }
-	       }
-	    }
-	 }
-	 break;
-      case GL_LEQUAL:
-	 if (ctx->Depth.Mask) {
-	    /* Update Z buffer */
-	    for (i=0;i<n;i++,zptr++,m++) {
-	       if (*m) {
-		  if (z[i] <= *zptr) {
-		     *zptr = z[i];
-		     passed++;
-		  }
-		  else {
-		     *m = 0;
-		  }
-	       }
-	    }
-	 }
-	 else {
-	    /* Don't update Z buffer */
-	    for (i=0;i<n;i++,zptr++,m++) {
-	       if (*m) {
-		  if (z[i] <= *zptr) {
-		     /* pass */
-		     passed++;
-		  }
-		  else {
-		     *m = 0;
-		  }
-	       }
-	    }
-	 }
-	 break;
-      case GL_GEQUAL:
-	 if (ctx->Depth.Mask) {
-	    /* Update Z buffer */
-	    for (i=0;i<n;i++,zptr++,m++) {
-	       if (*m) {
-		  if (z[i] >= *zptr) {
-		     *zptr = z[i];
-		     passed++;
-		  }
-		  else {
-		     *m = 0;
-		  }
-	       }
-	    }
-	 }
-	 else {
-	    /* Don't update Z buffer */
-	    for (i=0;i<n;i++,zptr++,m++) {
-	       if (*m) {
-		  if (z[i] >= *zptr) {
-		     /* pass */
-		     passed++;
-		  }
-		  else {
-		     *m = 0;
-		  }
-	       }
-	    }
-	 }
-	 break;
-      case GL_GREATER:
-	 if (ctx->Depth.Mask) {
-	    /* Update Z buffer */
-	    for (i=0;i<n;i++,zptr++,m++) {
-	       if (*m) {
-		  if (z[i] > *zptr) {
-		     *zptr = z[i];
-		     passed++;
-		  }
-		  else {
-		     *m = 0;
-		  }
-	       }
-	    }
-	 }
-	 else {
-	    /* Don't update Z buffer */
-	    for (i=0;i<n;i++,zptr++,m++) {
-	       if (*m) {
-		  if (z[i] > *zptr) {
-		     /* pass */
-		     passed++;
-		  }
-		  else {
-		     *m = 0;
-		  }
-	       }
-	    }
-	 }
-	 break;
-      case GL_NOTEQUAL:
-	 if (ctx->Depth.Mask) {
-	    /* Update Z buffer */
-	    for (i=0;i<n;i++,zptr++,m++) {
-	       if (*m) {
-		  if (z[i] != *zptr) {
-		     *zptr = z[i];
-		     passed++;
-		  }
-		  else {
-		     *m = 0;
-		  }
-	       }
-	    }
-	 }
-	 else {
-	    /* Don't update Z buffer */
-	    for (i=0;i<n;i++,zptr++,m++) {
-	       if (*m) {
-		  if (z[i] != *zptr) {
-		     /* pass */
-		     passed++;
-		  }
-		  else {
-		     *m = 0;
-		  }
-	       }
-	    }
-	 }
-	 break;
-      case GL_EQUAL:
-	 if (ctx->Depth.Mask) {
-	    /* Update Z buffer */
-	    for (i=0;i<n;i++,zptr++,m++) {
-	       if (*m) {
-		  if (z[i] == *zptr) {
-		     *zptr = z[i];
-		     passed++;
-		  }
-		  else {
-		     *m =0;
-		  }
-	       }
-	    }
-	 }
-	 else {
-	    /* Don't update Z buffer */
-	    for (i=0;i<n;i++,zptr++,m++) {
-	       if (*m) {
-		  if (z[i] == *zptr) {
-		     /* pass */
-		     passed++;
-		  }
-		  else {
-		     *m =0;
-		  }
-	       }
-	    }
-	 }
-	 break;
-      case GL_ALWAYS:
-	 if (ctx->Depth.Mask) {
-	    /* Update Z buffer */
-	    for (i=0;i<n;i++,zptr++,m++) {
-	       if (*m) {
-		  *zptr = z[i];
-		  passed++;
-	       }
-	    }
-	 }
-	 else {
-	    /* Don't update Z buffer or mask */
-	    passed = n;
-	 }
-	 break;
-      case GL_NEVER:
-	 for (i=0;i<n;i++) {
-	    mask[i] = 0;
-	 }
-	 break;
-      default:
-         gl_problem(ctx, "Bad depth func in gl_depth_test_span_generic");
-   } /*switch*/
-
-   return passed;
-}
-
-
-
-/*
- * glDepthFunc(GL_LESS) and glDepthMask(GL_TRUE).
- */
-GLuint i810_depth_test_span_less( GLcontext* ctx,
-                                GLuint n, GLint x, GLint y, const GLdepth z[],
-                                GLubyte mask[] )
-{
-   Z_SETUP
-   GLdepth *zptr = Z_ADDRESS( x, y );
-   GLuint i;
-   GLuint passed = 0;
-
-  i810WaitDrawingEngine();
-
-   for (i=0; i<n; i++) {
-      if (mask[i]) {
-         if (z[i] < zptr[i]) {
-            /* pass */
-            zptr[i] = z[i];
-            passed++;
-         }
-         else {
-            /* fail */
-            mask[i] = 0;
-         }
-      }
-   }
-   return passed;
-}
-
-
-/*
- * glDepthFunc(GL_GREATER) and glDepthMask(GL_TRUE).
- */
-GLuint i810_depth_test_span_greater( GLcontext* ctx,
-                                   GLuint n, GLint x, GLint y,
-                                   const GLdepth z[],
-                                   GLubyte mask[] )
-{
-   Z_SETUP
-   GLdepth *zptr = Z_ADDRESS( x, y );
-   GLuint i;
-   GLuint passed = 0;
-
-  i810WaitDrawingEngine();
-
-   for (i=0; i<n; i++) {
-      if (mask[i]) {
-         if (z[i] > zptr[i]) {
-            /* pass */
-            zptr[i] = z[i];
-            passed++;
-         }
-         else {
-            /* fail */
-            mask[i] = 0;
-         }
-      }
-   }
-   return passed;
-}
-
-
-
-/*
- * Depth test an array of randomly positioned fragments.
- */
-
-/*
- * glDepthFunc( any ) and glDepthMask( GL_TRUE or GL_FALSE ).
- */
-void i810_depth_test_pixels_generic( GLcontext* ctx,
-                                   GLuint n, const GLint x[], const GLint y[],
-                                   const GLdepth z[], GLubyte mask[] )
-{
-   Z_SETUP
-   register GLdepth *zptr;
-   register GLuint i;
-
-  i810WaitDrawingEngine();
-
-   /* switch cases ordered from most frequent to less frequent */
-   switch (ctx->Depth.Func) {
-      case GL_LESS:
-         if (ctx->Depth.Mask) {
-	    /* Update Z buffer */
-	    for (i=0; i<n; i++) {
-	       if (mask[i]) {
-		  zptr = Z_ADDRESS(x[i],y[i]);
-		  if (z[i] < *zptr) {
-		     /* pass */
-		     *zptr = z[i];
-		  }
-		  else {
-		     /* fail */
-		     mask[i] = 0;
-		  }
-	       }
-	    }
-	 }
-	 else {
-	    /* Don't update Z buffer */
-	    for (i=0; i<n; i++) {
-	       if (mask[i]) {
-		  zptr = Z_ADDRESS(x[i],y[i]);
-		  if (z[i] < *zptr) {
-		     /* pass */
-		  }
-		  else {
-		     /* fail */
-		     mask[i] = 0;
-		  }
-	       }
-	    }
-	 }
-	 break;
-      case GL_LEQUAL:
-         if (ctx->Depth.Mask) {
-	    /* Update Z buffer */
-	    for (i=0; i<n; i++) {
-	       if (mask[i]) {
-		  zptr = Z_ADDRESS(x[i],y[i]);
-		  if (z[i] <= *zptr) {
-		     /* pass */
-		     *zptr = z[i];
-		  }
-		  else {
-		     /* fail */
-		     mask[i] = 0;
-		  }
-	       }
-	    }
-	 }
-	 else {
-	    /* Don't update Z buffer */
-	    for (i=0; i<n; i++) {
-	       if (mask[i]) {
-		  zptr = Z_ADDRESS(x[i],y[i]);
-		  if (z[i] <= *zptr) {
-		     /* pass */
-		  }
-		  else {
-		     /* fail */
-		     mask[i] = 0;
-		  }
-	       }
-	    }
-	 }
-	 break;
-      case GL_GEQUAL:
-         if (ctx->Depth.Mask) {
-	    /* Update Z buffer */
-	    for (i=0; i<n; i++) {
-	       if (mask[i]) {
-		  zptr = Z_ADDRESS(x[i],y[i]);
-		  if (z[i] >= *zptr) {
-		     /* pass */
-		     *zptr = z[i];
-		  }
-		  else {
-		     /* fail */
-		     mask[i] = 0;
-		  }
-	       }
-	    }
-	 }
-	 else {
-	    /* Don't update Z buffer */
-	    for (i=0; i<n; i++) {
-	       if (mask[i]) {
-		  zptr = Z_ADDRESS(x[i],y[i]);
-		  if (z[i] >= *zptr) {
-		     /* pass */
-		  }
-		  else {
-		     /* fail */
-		     mask[i] = 0;
-		  }
-	       }
-	    }
-	 }
-	 break;
-      case GL_GREATER:
-         if (ctx->Depth.Mask) {
-	    /* Update Z buffer */
-	    for (i=0; i<n; i++) {
-	       if (mask[i]) {
-		  zptr = Z_ADDRESS(x[i],y[i]);
-		  if (z[i] > *zptr) {
-		     /* pass */
-		     *zptr = z[i];
-		  }
-		  else {
-		     /* fail */
-		     mask[i] = 0;
-		  }
-	       }
-	    }
-	 }
-	 else {
-	    /* Don't update Z buffer */
-	    for (i=0; i<n; i++) {
-	       if (mask[i]) {
-		  zptr = Z_ADDRESS(x[i],y[i]);
-		  if (z[i] > *zptr) {
-		     /* pass */
-		  }
-		  else {
-		     /* fail */
-		     mask[i] = 0;
-		  }
-	       }
-	    }
-	 }
-	 break;
-      case GL_NOTEQUAL:
-         if (ctx->Depth.Mask) {
-	    /* Update Z buffer */
-	    for (i=0; i<n; i++) {
-	       if (mask[i]) {
-		  zptr = Z_ADDRESS(x[i],y[i]);
-		  if (z[i] != *zptr) {
-		     /* pass */
-		     *zptr = z[i];
-		  }
-		  else {
-		     /* fail */
-		     mask[i] = 0;
-		  }
-	       }
-	    }
-	 }
-	 else {
-	    /* Don't update Z buffer */
-	    for (i=0; i<n; i++) {
-	       if (mask[i]) {
-		  zptr = Z_ADDRESS(x[i],y[i]);
-		  if (z[i] != *zptr) {
-		     /* pass */
-		  }
-		  else {
-		     /* fail */
-		     mask[i] = 0;
-		  }
-	       }
-	    }
-	 }
-	 break;
-      case GL_EQUAL:
-         if (ctx->Depth.Mask) {
-	    /* Update Z buffer */
-	    for (i=0; i<n; i++) {
-	       if (mask[i]) {
-		  zptr = Z_ADDRESS(x[i],y[i]);
-		  if (z[i] == *zptr) {
-		     /* pass */
-		     *zptr = z[i];
-		  }
-		  else {
-		     /* fail */
-		     mask[i] = 0;
-		  }
-	       }
-	    }
-	 }
-	 else {
-	    /* Don't update Z buffer */
-	    for (i=0; i<n; i++) {
-	       if (mask[i]) {
-		  zptr = Z_ADDRESS(x[i],y[i]);
-		  if (z[i] == *zptr) {
-		     /* pass */
-		  }
-		  else {
-		     /* fail */
-		     mask[i] = 0;
-		  }
-	       }
-	    }
-	 }
-	 break;
-      case GL_ALWAYS:
-	 if (ctx->Depth.Mask) {
-	    /* Update Z buffer */
-	    for (i=0; i<n; i++) {
-	       if (mask[i]) {
-		  zptr = Z_ADDRESS(x[i],y[i]);
-		  *zptr = z[i];
-	       }
-	    }
-	 }
-	 else {
-	    /* Don't update Z buffer or mask */
-	 }
-	 break;
-      case GL_NEVER:
-	 /* depth test never passes */
-	 for (i=0;i<n;i++) {
-	    mask[i] = 0;
-	 }
-	 break;
-      default:
-         gl_problem(ctx, "Bad depth func in gl_depth_test_pixels_generic");
-   } /*switch*/
-}
-
-
-
-/*
- * glDepthFunc( GL_LESS ) and glDepthMask( GL_TRUE ).
- */
-void i810_depth_test_pixels_less( GLcontext* ctx,
-                                GLuint n, const GLint x[], const GLint y[],
-                                const GLdepth z[], GLubyte mask[] )
-{
-   Z_SETUP
-   GLdepth *zptr;
-   GLuint i;
-
-  i810WaitDrawingEngine();
-
-   for (i=0; i<n; i++) {
-      if (mask[i]) {
-         zptr = Z_ADDRESS(x[i],y[i]);
-         if (z[i] < *zptr) {
-            /* pass */
-            *zptr = z[i];
-         }
-         else {
-            /* fail */
-            mask[i] = 0;
-         }
-      }
-   }
-}
-
-
-/*
- * glDepthFunc( GL_GREATER ) and glDepthMask( GL_TRUE ).
- */
-void i810_depth_test_pixels_greater( GLcontext* ctx,
-                                   GLuint n, const GLint x[], const GLint y[],
-                                   const GLdepth z[], GLubyte mask[] )
-{
-   Z_SETUP
-   GLdepth *zptr;
-   GLuint i;
-
-  i810WaitDrawingEngine();
-
-   for (i=0; i<n; i++) {
-      if (mask[i]) {
-         zptr = Z_ADDRESS(x[i],y[i]);
-         if (z[i] > *zptr) {
-            /* pass */
-            *zptr = z[i];
-         }
-         else {
-            /* fail */
-            mask[i] = 0;
-         }
-      }
-   }
-}
-
-
-
-
-/**********************************************************************/
-/*****                      Read Depth Buffer                     *****/
-/**********************************************************************/
-
-
-/*
- * Return a span of depth values from the depth buffer as floats in [0,1].
- * This function is only called through Driver.read_depth_span_float()
- * Input:  n - how many pixels
- *         x,y - location of first pixel
- * Output:  depth - the array of depth values
- */
-void i810_read_depth_span_float( GLcontext* ctx,
-                               GLuint n, GLint x, GLint y, GLfloat depth[] )
-{
-   Z_SETUP
-   GLdepth *zptr;
-   GLfloat scale;
-   GLuint i;
-
-  i810WaitDrawingEngine();
-
-   scale = 1.0F / DEPTH_SCALE;
-
-   if (ctx->Buffer->Depth) {
-      zptr = Z_ADDRESS( x, y );
-      for (i=0;i<n;i++) {
-	 depth[i] = (GLfloat) zptr[i] * scale;
-      }
-   }
-   else {
-      for (i=0;i<n;i++) {
-	 depth[i] = 0.0F;
-      }
-   }
-}
-
-
-/*
- * Return a span of depth values from the depth buffer as integers in
- * [0,MAX_DEPTH].
- * This function is only called through Driver.read_depth_span_int()
- * Input:  n - how many pixels
- *         x,y - location of first pixel
- * Output:  depth - the array of depth values
- */
-void i810_read_depth_span_int( GLcontext* ctx,
-                             GLuint n, GLint x, GLint y, GLdepth depth[] )
-{
-   Z_SETUP
-   
-   i810WaitDrawingEngine();
-
-   if (ctx->Buffer->Depth) {
-      GLdepth *zptr = Z_ADDRESS( x, y );
-      MEMCPY( depth, zptr, n * sizeof(GLdepth) );
-   }
-   else {
-      GLuint i;
-      for (i=0;i<n;i++) {
-	 depth[i] = 0;
-      }
-   }
-}
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810depth.h b/xc/lib/GL/mesa/src/drv/i810/i810depth.h
deleted file mode 100644
index 878fbefe8..000000000
--- a/xc/lib/GL/mesa/src/drv/i810/i810depth.h
+++ /dev/null
@@ -1,80 +0,0 @@
-/*
- * Mesa 3-D graphics library
- * Version:  3.1
- * 
- * Copyright (C) 1999  Brian Paul   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 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
- * BRIAN PAUL 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.
- */
-
-/*
- * This file has been modified by Wittawat Yamwong for GLX module.
- */
-
-
-#ifndef I810DEPTH_INC
-#define I810DEPTH_INC
-
-
-#include "mesaglx/types.h"
-
-
-
-
-
-extern GLuint
-i810_depth_test_span_generic( GLcontext* ctx, GLuint n, GLint x, GLint y,
-                            const GLdepth z[], GLubyte mask[] );
-
-extern GLuint
-i810_depth_test_span_less( GLcontext* ctx, GLuint n, GLint x, GLint y,
-                         const GLdepth z[], GLubyte mask[] );
-
-extern GLuint
-i810_depth_test_span_greater( GLcontext* ctx, GLuint n, GLint x, GLint y,
-                            const GLdepth z[], GLubyte mask[] );
-
-
-
-extern void
-i810_depth_test_pixels_generic( GLcontext* ctx,
-                              GLuint n, const GLint x[], const GLint y[],
-                              const GLdepth z[], GLubyte mask[] );
-
-extern void
-i810_depth_test_pixels_less( GLcontext* ctx,
-                           GLuint n, const GLint x[], const GLint y[],
-                           const GLdepth z[], GLubyte mask[] );
-
-extern void
-i810_depth_test_pixels_greater( GLcontext* ctx,
-                              GLuint n, const GLint x[], const GLint y[],
-                              const GLdepth z[], GLubyte mask[] );
-
-
-extern void i810_read_depth_span_float( GLcontext* ctx,
-                                      GLuint n, GLint x, GLint y,
-                                      GLfloat depth[] );
-
-
-extern void i810_read_depth_span_int( GLcontext* ctx, GLuint n, GLint x, GLint y,
-                                    GLdepth depth[] );
-
-
-
-#endif
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810dma.c b/xc/lib/GL/mesa/src/drv/i810/i810dma.c
deleted file mode 100644
index f745d272f..000000000
--- a/xc/lib/GL/mesa/src/drv/i810/i810dma.c
+++ /dev/null
@@ -1,363 +0,0 @@
-/* -*- mode: C; c-basic-offset:8 -*- */
-/*
- * 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.
- *
- * based on the original mgadma.c by Jeff Hartmann <slicer@ionet.net>
- * as rewritten by John Carmack <johnc@idsoftware.com>
- */
-
-#include <stdlib.h>
-#include <errno.h>
-#include <unistd.h>
-#include <sys/mman.h>
-#include <stdio.h>
-#include <sys/time.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <fcntl.h>
-#include <signal.h>
-
-#include "mm.h"
-#include "i810buf.h"
-#include "i810dd.h"
-#include "i810lib.h"
-#include "i810log.h"
-#include "i810tris.h"
-#include "i810direct.h"
-#include "i810state.h"
-
-#include "pb.h"
-
-/* X server include:
- */
-#include "i810.h"
-
-
-
-/* This will be overwritten from the default values when glx.so is
- * loaded on the client.
- */
-void i810ServerDmaFlush( int wait );
-GLuint i810ActiveDmaBuffer = 0;	
-
-
-
-#ifndef I810_DEBUG
-int I810_DEBUG = (0
-/*     		  | DEBUG_ALWAYS_SYNC  */
-/*    		  | DEBUG_VERBOSE_ACCEL  */
-/*  		  | DEBUG_VERBOSE_SYNC */
-/*  		  | DEBUG_VERBOSE_VGA */
-/*  		  | DEBUG_VERBOSE_RING    */
-/*  		  | DEBUG_VERBOSE_OUTREG  */
-/*  		  | DEBUG_VERBOSE_MEMORY */
-/*  		  | DEBUG_ALWAYS_SYNC */
-/*  		  | DEBUG_VERBOSE_MSG */
-		  );
-#endif
-
-
-
-static void delay( void ) {
-}
-
-
-
-int i810WaitForDmaCompletion( void ) 
-{
-	int start, end;
-
-	if (I810LpRing.head == I810LpRing.tail && I810LpRing.space) 
-		return 0;
-
-	start = i810_usec();
-	I810LpRing.head = INREG(I810LpRing.base_reg + RING_HEAD);
-	end = i810_usec();
-
-	if (0 && start + 100 < end)
-		fprintf(stderr, "INREG in WaitForDma: %d usec\n", end - start);
-
-	I810LpRing.head &= HEAD_ADDR;
-	I810LpRing.space = I810LpRing.head - (I810LpRing.tail + 8);
-	if (I810LpRing.space < 0) 
-		I810LpRing.space += I810LpRing.mem.Size;
-
-	if (I810LpRing.head == I810LpRing.tail) 
-		return 0;
-
-	start = i810_usec();
-  	I810Sync(); 
-	end = i810_usec();
-
-	return 1;
-}
-
-
-
-/*
- * i810DmaResetBuffer
- */
-void i810DmaResetBuffer( void ) {
-
-	i810glx.dma_buffer = dmaBuffers[ i810ActiveDmaBuffer ];
-	i810glx.dma_buffer->head = 0;
-	i810glx.dma_buffer->space = 0;
-	i810glx.dma_buffer->additional_space = 
-		i810glx.dma_buffer->mem.Size - 256;
-	
-	i810DmaOverflow( 0 );
-
-	if (i810glx.dma_buffer_age < i810glx.dma_buffer->texture_age) 
-		i810glx.dma_buffer_age = i810glx.dma_buffer->texture_age; 
-}
-
-
-
-
-/*
- * i810FlushRealDma
- */
-void i810FlushRealDma( void ) {
-
-	GLuint start;
-
-	if (MESA_VERBOSE&VERBOSE_DRIVER)
-		fprintf(stderr, "i810FlushRealDma()\n" );
-
-
-	if ( i810glx.skipDma || !I810_USE_BATCH )
-		return;
-	
-
-	if (i810glx.dma_buffer->head & 0x4) {
-		FatalError( "Misaligned batch buffer\n" );
-	}
-
-	if ( (I810_DEBUG&DEBUG_VERBOSE_OUTREG) && 0)
-	{
-		int i;
-		for (i = 0 ; i <= i810glx.dma_buffer->head ; i+=4) 
-			fprintf(stderr, 
-				" 0x%05x : 0x%08x\n", 
-				i/4, 
-				*(GLuint *)(i810glx.dma_buffer->virtual_start +
-					    i));
-	}
-
-
-	/* fire the batch buffer */       
-	for (start = 0 ; start < i810glx.dma_buffer->head ; start += MAX_BATCH)
-	{
-		GLuint ofs = i810glx.dma_buffer->mem.Start;
-		GLuint end = MIN2(start + MAX_BATCH, i810glx.dma_buffer->head);
-		
-		BEGIN_LP_RING(4);
-		OUT_RING( CMD_OP_BATCH_BUFFER );
-		OUT_RING( (ofs + start) | BB1_PROTECTED );
-		OUT_RING( (ofs + end) - 4 );
-		OUT_RING( 0 );
-		ADVANCE_LP_RING();
-	}
-
-	{
-		BEGIN_LP_RING(2);
-		OUT_RING( INST_PARSER_CLIENT | INST_OP_FLUSH | 
-			  INST_FLUSH_MAP_CACHE );
-		OUT_RING( 0 );
-	}
-}
-
-/*
- * i810DmaFlush
- * Send all pending commands off to the hardware.
- * If we are running async, the hardware will be drawing
- * while we return to do other things.
- */
-void i810ServerDmaFlush( int wait ) {
-	int		start, end;
-
-	if (I810_DEBUG)
-		fprintf(stderr, 
-			"i810ServerDmaFlush, buffer %d, head %x space %x\n",
-			i810ActiveDmaBuffer, 
-			i810glx.dma_buffer->head,
-			i810glx.dma_buffer->space);
-
-	if ( i810glx.dma_buffer->head == 0 ) 
-	{
-		if (wait && !i810WaitForDmaCompletion())
-			i810glx.hardwareWentIdle = 1;
-
-		return;
-	}
-
-	i810glx.c_dmaFlush++;
-	
-	/* wait for the last buffer to complete */
-	if ( !i810WaitForDmaCompletion() ) 
-		i810glx.hardwareWentIdle = 1;
-	
-
-	/* collect timing information if we are going syncronously */
-	if ( i810glx.dmaDriver != 3 ) {
- 		start = i810_usec();
- 	} else {
- 		start = end = 0;
- 	}
- 	
-	i810FlushRealDma();
- 
-	if ( i810glx.dmaDriver == 2 ) {
-		/* wait until the dma completes */
-		i810WaitForDmaCompletion();
-	}
-
- 	if ( i810glx.dmaDriver != 3 ) {
-		end = i810_usec();
-	}
-
-	i810Msg(10, "flushmode %i, buffer %i: prim dwords:%i usec:%i\n", 
-		i810glx.dmaDriver,  i810ActiveDmaBuffer,
-		i810glx.dma_buffer->head / 4,
-		end - start );	
-
-	/* swap to using the other buffer */
-	i810ActiveDmaBuffer ^= 1;
-	i810DmaResetBuffer();
-	
-	if ( wait ) 
-		i810WaitForDmaCompletion();
-}
-
-
-/*
- * i810DmaFlush
- */
-void i810DmaFlush( void ) {
-	i810FinishPrimitive();
-	i810glx.dma_buffer->texture_age = ++i810glx.current_texture_age;
-
-	if ( i810Ctx && i810Ctx->CurrentTex0Obj ) 
-		i810Ctx->CurrentTex0Obj->age = ++i810glx.current_texture_age; 
-
-	if ( i810Ctx && i810Ctx->CurrentTex1Obj ) 
-		i810Ctx->CurrentTex1Obj->age = ++i810glx.current_texture_age; 
-		
-	i810ServerDmaFlush( 0 );
-}
-
-
-/*
- * i810DmaFinish
- */
-void i810DmaFinish( void ) {
-	i810FinishPrimitive();
-	i810ServerDmaFlush( 1 );
-
-	i810glx.dma_buffer_age = ++i810glx.current_texture_age;
-  
-	if ( i810Ctx && i810Ctx->CurrentTex0Obj ) 
-		i810Ctx->CurrentTex0Obj->age = ++i810glx.current_texture_age; 
-
-	if ( i810Ctx && i810Ctx->CurrentTex1Obj ) 
-		i810Ctx->CurrentTex1Obj->age = ++i810glx.current_texture_age; 
-}
-
-
-/*
- * i810DmaOverflow
- * This is called when I810DMAGETPTR is at the end of the buffer
- */
-void i810DmaOverflow( int newDwords ) {
-
-	if (i810glx.dma_buffer->additional_space) {
-		GLuint incr = MAX_BATCH;
-
-		if (i810glx.dma_buffer->additional_space < incr) 
-			incr = i810glx.dma_buffer->additional_space;
-
-		while (i810glx.dma_buffer->head & (MAX_BATCH - 1)) {
-			GLuint outbatch = i810glx.dma_buffer->head; 
-			GLubyte *virt = i810glx.dma_buffer->virtual_start; 
-
-			*(volatile unsigned int *)(virt + outbatch) = 0; 
-			i810glx.dma_buffer->head += 4;
-		}
-
-		i810glx.dma_buffer->space += incr;
-		i810glx.dma_buffer->additional_space -= incr;
-
-		if (0)
-		fprintf(stderr, "overflow, head %x space %x\n",
-			i810glx.dma_buffer->head,
-			i810glx.dma_buffer->space);
-		return;
-	}
-	
-	i810Msg( 9, "i810DmaOverflow(%i)\n", newDwords );
-
-	/* flush all the current commands so we will have another
-           empty buffer */
-	i810DmaFlush();
-
-	i810glx.c_overflows++;
-
-	if ( newDwords > i810glx.dma_buffer->space ) 
- 		FatalError("i810DmaOverflow > maxPrimaryDwords");
-}
-
-
-/*
- * i810WaitDrawingEngine
- * This will not return until the drawing engine has completed
- * drawing pixels and it is safe to read or write the framebuffer
- * for software rendering.
- */
-int i810WaitDrawingEngine( void ) {
-	/* note this for the performance block display */
-	i810glx.c_drawWaits++;
-
-	/* make sure all pending dma has completed */
-	i810DmaFinish();
-	return 0;
-}
-
-
-/*
- * i810DmaExecute
- * Add a block of data to the dma buffer
- */
-void i810DmaExecute( GLuint *code, int dwords ) 
-{	
-	int i;
-	BEGIN_BATCH(dwords);
-
-	if (dwords & 1) 
-		FatalError( "Misaligned buffer in i810DmaExecute\n" );
-
-	for ( i = 0 ; i < dwords ; i++ )
-		OUT_BATCH( code[i] );
-	ADVANCE_BATCH();
-}
-
-
-
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810dma.h b/xc/lib/GL/mesa/src/drv/i810/i810dma.h
index 55adbcddd..3fd995cb3 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810dma.h
+++ b/xc/lib/GL/mesa/src/drv/i810/i810dma.h
@@ -1,138 +1,44 @@
-/*
-  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 "i810common.h"
 #include "i810lib.h"
-#include "i810.h"
+#include "i810_init.h"
 #include "mm.h"
 
-/* the main initialization of the entire i810 hardware driver */
-GLboolean i810InitGLX( void );
 
-/* 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(void);
-
-/* 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( void );
-
-/* a i810WaitDrawingEngine command will guarantee that the framebuffer
-is safe to read or write for software rendering */
-int i810WaitDrawingEngine( void );
-
-
-
-
-
-typedef struct {
-   I810MemRange mem;
-   char *virtual_start;
-   int head;
-   int space;
-   int additional_space;
-   int texture_age;
-} i810BatchBuffer;
-
-
-
-#define I810_USE_BATCH 1
-#if I810_USE_BATCH
-
-#define BEGIN_BATCH(n)							\
+/* Vertex data does not use these macros.
+ */
+#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/%d, head %x vstart %x start %x)\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", outbatch/4, val);	\
-   outbatch += 4;						\
-}
+      fprintf(stderr, "BEGIN_BATCH(%d) in %s\n", n, __FUNCTION__);	\
+   if (imesa->dma_buffer->total - imesa->dma_buffer->used < n*4) {	\
+      i810FlushGeneralLocked( imesa );					\
+      i810GetGeneralDmaBufferLocked( imesa );				\
+   }									\
+   outbatch = imesa->dma_buffer->used;					\
+   virt = imesa->dma_buffer->address;			
+
+
+#define OUT_BATCH(val) do {				\
+   *(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;					\
+} while (0)
    
 
-#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 ADVANCE_BATCH() do {					\
+   if (I810_DEBUG & DEBUG_VERBOSE_RING)				\
+      fprintf(stderr, "ADVANCE_BATCH(%.1f) in %s\n",		\
+                  (outbatch - imesa->dma_buffer->used) / 4.0,	\
+                  __FUNCTION__);				\
+   imesa->dma_buffer->used = outbatch;				\
+} while(0)
 
 #define FINISH_PRIM()
 
-#else
-
-/* Need some work to be able to do primitives this way.
- */
-#define BEGIN_BATCH(n) BEGIN_LP_RING(n)
-#define ADVANCE_BATCH() ADVANCE_LP_RING()
-#define OUT_BATCH(val) OUT_RING(val)
-#define FINISH_PRIM() OUTREG(I810LpRing.base_reg + RING_TAIL, I810LpRing.tail)
-
-#endif
-
-
-/* These are only required by i810direct.c:
- */
-extern GLuint i810ActiveDmaBuffer;
-extern void (*i810DoDmaFlush)( int );
-extern i810BatchBuffer *dmaBuffers[2];
-
-extern int i810WaitForDmaCompletion( void );
-extern void i810DmaResetBuffer( void );
-
-extern void i810DmaExecute( GLuint *code, int dwords );
-
-extern void I810Sync( void ) ;
-extern void I810WaitLpRing( int n );
-
-
 #endif
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810dmainit.c b/xc/lib/GL/mesa/src/drv/i810/i810dmainit.c
deleted file mode 100644
index 335c4a612..000000000
--- a/xc/lib/GL/mesa/src/drv/i810/i810dmainit.c
+++ /dev/null
@@ -1,389 +0,0 @@
-
-/* -*- mode: C; c-basic-offset:8 -*- */
-/*
- * 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.
- *
- *
- * based on the original by Jeff Hartmann <slicer@ionet.net>
- * as rewritten by John Carmack <johnc@idsoftware.com>
- */
-
-/*
-
-This file is only entered at startup.  After i810GlxInit completes,
-nothing here will be executed again.
-
-*/
-
-#include <stdlib.h>
-#include <errno.h>
-#include <unistd.h>
-#include <sys/mman.h>
-#include <stdio.h>
-#include <sys/stat.h>
-#include <sys/ioctl.h>
-#include <fcntl.h>
-#include <signal.h>
-
-#include "context.h"
-#include "depth.h"
-#include "macros.h"
-#include "texstate.h"
-#include "triangle.h"
-#include "vb.h"
-#include "types.h"
-
-#include "xsmesaP.h"
-#include "glx_log.h"
-#include "mesaglx/context.h"
-#include "mesaglx/matrix.h"
-#include "mesaglx/types.h"
-
-#define GC XXGC
-#include "gcstruct.h"
-#include "pixmapstr.h"
-#include "servermd.h" /* PixmapBytePad */
-#include "scrnintstr.h"
-#include "regionstr.h"
-#include "windowstr.h"
-#undef GC
-
-#include "vga.h"
-#include "vgaPCI.h"
-#include "i810.h"
-#include "xaa/xf86xaa.h"
-
-#include "mm.h"
-#include "i810buf.h"
-#include "i810dd.h"
-#include "i810lib.h"
-#include "i810log.h"
-#include "i810direct.h"
-#include "i810glx.h"
-
-#include <agpgart.h>
-
-/* private vars */
-i810BatchBuffer	*dmaBuffers[2];
-static I810MemRange I810SysMem2;
-
-
-static const char *getenvSafe(const char *name ) {
-   const char	*r;
-	
-   r = getenv( name );
-   if ( !r ) {
-      return "";
-   }
-   return r;
-}
-
-
-static void AllocateCommandBuffers( void ) 
-{
-   PMemBlock block;
-   GLuint bufferBytes = i810glx.cmdSize;
-   GLuint start, i;
-
-   if ( 1 || !bufferBytes ) {
-      fprintf(stderr,"temporarily hardwiring GLX_I810_CMDSIZE = 1\n" );
-      bufferBytes = 1;
-   }
-
-   bufferBytes *= 0x100000;
-	
-   block = mmAllocMem( i810glx.sysmemHeap, bufferBytes, 8, 0 );
-
-   if ( !block ) {
-      fprintf(stderr, 
-	      "failed to allocate 0x%x bytes from "
-	      "sysmemHeap for command buffers.\n", 
-	      bufferBytes );
-
-      FatalError("Couldn't get dma buffer\n");
-   }
-
-   start = block->ofs;
-		
-   /* setup the two buffers that will be ping-ponged 
-    */
-   for (i = 0 ; i < 2 ; i++) {
-      dmaBuffers[i] = calloc(1,sizeof(i810BatchBuffer));
-      dmaBuffers[i]->virtual_start = i810glx.sysmemVirtual + start;
-      dmaBuffers[i]->mem.Start = start;
-      dmaBuffers[i]->mem.Size = bufferBytes / 2;
-      dmaBuffers[i]->mem.End = (start += bufferBytes / 2);
-   }
-	
-   i810DmaResetBuffer();
-}
-
-
-/* Another bit of duplication caused by 'static' code in the X server.
- */
-static int    gartfd;
-static struct gart_info gartinf;
-static char *gart_buf;
-
-static int AllocateGARTMemory( int size ) 
-{
-   int i, pages = size / 4096; 
-   int start = vga256InfoRec.videoRam / 4; /* a crock! */
-   struct stat sb;
-
-   if (stat("/dev/agpgart", &sb) != 0) {
-      ErrorF("Stat failed on /dev/agpgart: %s\n", 
-	     sys_errlist[errno]);
-      return -1;
-   }
-
-   gartfd = open("/dev/agpgart", O_RDWR);
-   if (gartfd == -1) {	
-      ErrorF("unable to open /dev/agpgart: %s\n", 
-	     sys_errlist[errno]);
-      FatalError("Aborting");
-   }
-
-   if (ioctl(gartfd, GARTIOCINFO, &gartinf) != 0) {
-      ErrorF("error doing ioctl(GARTIOCINFO): %s\n", 
-	     sys_errlist[errno]);
-      FatalError("Aborting");
-   }
-
-   /* Treat the gart like video memory - we assume we own all that is
-    * there, so ignore EBUSY errors.  Don't try to remove it on
-    * failure, either.
-    */
-   for (i = start; i < start+pages; i++) 
-      if (ioctl(gartfd, GARTIOCINSERT, &i) != 0) {
-	 if (errno != EBUSY) 
-	 {	
-	    perror("gart insert");
-	    ErrorF("GART: allocation of %d pages failed at page %d\n", 
-		   pages, i);
-	    FatalError("Aborting");
-	 }	
-      } 
-
-   ErrorF("GART: allocated %dK system ram\n", pages * 4);
-   
-   I810SysMem2.Start = start * 4096;
-   I810SysMem2.End = (start + pages) * 4096;
-   I810SysMem2.Size = pages * 4096;
-
-   gart_buf = mmap( 0, 
-		    gartinf.size * 0x100000,
-		    PROT_READ | PROT_WRITE,
-		    MAP_SHARED,		    
-		    gartfd,
-		    0 );
-   
-   if ((unsigned int)gart_buf == ~0) {
-      perror("/dev/agpgart");
-      FatalError("Couldn't mmap /dev/agpgart - aborting");
-   }
-
-   return 0;
-}
-
-
-/*
- * AllocateSystemMemory
- * Looks at environment variables to determine if a block
- * of physical memory has been left for graphics after the
- * memory available to the kernel.
- * System memory can be used for dma command buffers or
- * textures.
- */
-static void AllocateSystemMemory( void ) 
-{
-   GLuint sysmemBytes = i810glx.dmaSize;
-
-   if ( !sysmemBytes ) 
-      sysmemBytes = 8;
-
-   sysmemBytes *= 0x100000;
-
-   if (AllocateGARTMemory(sysmemBytes) != 0) 
-      FatalError( "AllocateGARTMemory failed.\n" );
-
-   
-   i810glx.sysmemVirtual = (unsigned char *)gart_buf;
-
-   /* Suck in any leftover memory from the 2d server.
-
-   i810glx.sysmemHeap = mmInit( I810SysMem.Start,
-				I810SysMem.Size );
-    */
-
-   /* Add our new memory.
-    */
-   i810glx.sysmemHeap = 
-/*        mmAddRange( i810glx.sysmemHeap,  */
-      mmInit(
-	 I810SysMem2.Start,
-	 I810SysMem2.Size );
-
-      			
-   /* Manage dcache mem via. a seperate heap.
-    */
-   i810glx.cardHeap = mmInit( I810DcacheMem.Start,
-			      I810DcacheMem.Size );
-
-   i810Msg( 1, "sysmemSize: 0x%x\n", sysmemBytes );
-   i810Msg( 1, "cardSize: 0x%lx\n", I810DcacheMem.Size );
-}
-
-
-
-/*
- * i810DmaInit
- *
-*/
-void i810DmaInit(void) {
-
-   /* Server init - queries environment variables.  The client
-    * gets these values from the sever and initializes them in
-    * i810direct.c 
-    */
-   if (__glx_is_server) {
-      i810glx.dmaDriver = atoi( getenvSafe("GLX_I810_DMA") );
-      i810glx.dmaSize = atoi( getenvSafe("GLX_I810_DMASIZE") );
-      i810glx.cmdSize = atoi( getenvSafe("GLX_I810_CMDSIZE") );
-
-      if (i810glx.dmaDriver != 3)
-	 FatalError("GLX_I810_DMA not set\n");
-   }
-  
-   /* get some system memory and make it write combining if we can */
-   AllocateSystemMemory();
-
-   /* read the command environment variable */
-   i810Msg(1,"i810DmaInit: GLX_I810_DMA = %i\n", i810glx.dmaDriver );
-
-   /* setup the two command buffers in the apropriate memory space */
-   AllocateCommandBuffers();
-	
-   /* prepare the first buffer for use */
-   i810DmaResetBuffer();
-}
-
-
-/*
- * This function should only verify that the current hardware is supported.
- * It should do no setup.
- */
-static GLboolean i810_detect_hw( void ) 
-{  
-   if (I810Chipset == -1)
-      return GL_FALSE;
-
-   if (vga256InfoRec.depth != 15 &&
-       vga256InfoRec.depth != 16) 
-   {
-      i810Error("Unsupported depth: %d, only 15 and 16d bpp "
-		"are supported right now\n",
-		vga256InfoRec.depth);
-      return GL_FALSE;
-   }
-
-   return GL_TRUE;
-}
-
-/*
- * i810InitLogging
- *
- */
-void i810InitLogging( void )
-{		
-   i810glx.logLevel = 100;
-}
-
-
-
-/*
- * i810InitGLX
- * This is the initial entry point for the i810 hardware driver,
- * called at X server module load time, or libGL direct rendering
- * init time.
- */
-GLboolean i810InitGLX( void ) {
-
-   fprintf(stderr, "\n\n\n\ni810InitGLX\n");
-
-   i810InitLogging();
-
-   i810Msg(1,"virtual (x, y) (%d, %d)\n", vga256InfoRec.virtualX,
-	   vga256InfoRec.virtualY);
-   i810Msg(1,"width: %d\n", vga256InfoRec.displayWidth);
-   i810Msg(1,"depth: %d\n", vga256InfoRec.depth);
-   i810Msg(1,"memBase: %p\n", vgaLinearBase);
-   i810Msg(1,"videoRam: 0x%08x\n", vga256InfoRec.videoRam);
-
-   if (!i810_detect_hw()) {
-      ErrorF("Couldn't find i810 hardware\n\n\n");
-      return GL_FALSE;
-   }
-
-   /* init the dma system */  
-   i810DmaInit();
-
-   /* Register as a glx driver */
-   GLXProcs.CreateContext = i810GLXCreateContext;
-   GLXProcs.DestroyContext = i810GLXDestroyContext;
-   GLXProcs.SwapBuffers = i810GLXSwapBuffers;
-   GLXProcs.CreateImage = i810GLXCreateImage;
-   GLXProcs.DestroyImage = i810GLXDestroyImage;
-   GLXProcs.CreateDepthBuffer = i810GLXCreateDepthBuffer;
-   GLXProcs.MakeCurrent = i810GLXMakeCurrent;
-   GLXProcs.BindBuffer = i810GLXBindBuffer;
-   GLXProcs.SwapBuffers = i810GLXSwapBuffers;
-   GLXProcs.VendorPrivate = i810GLXVendorPrivate;
-   GLXProcs.AllowDirect = i810GLXAllowDirect;
-
-   if (!__glx_is_server) {
-      GLXProcs.ValidateFrontBuffer = i810ClientGetGeometry;
-   }
-
-   /* these vars can be changed between invocations of direct clients */
-   if (getenv("GLX_I810_NULLPRIMS") ) {
-      i810Msg( 1, "enabling GLX_I810_NULLPRIMS\n" );
-      i810glx.nullprims = 1;
-   }
-   if (getenv("GLX_I810_SKIPDMA") ) {
-      i810Msg( 1, "enabling GLX_I810_SKIPDMA\n" );
-      i810glx.skipDma = 1;
-   }
-   if (getenv("GLX_I810_BOXES") ) {
-      i810Msg( 1, "enabling GLX_I810_BOXES\n" );
-      i810glx.boxes = 1;
-   }
-   if (getenv("GLX_I810_NOFALLBACK") ) {
-      i810Msg( 1, "enabling GLX_I810_NOFALLBACK\n" );
-      i810glx.noFallback = 1;
-   }
-	
-   i810Error("i810InitGLX completed\n");
-   return GL_TRUE;
-}
-
-
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810fastpath.c b/xc/lib/GL/mesa/src/drv/i810/i810fastpath.c
index 9223060e7..3e4887619 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810fastpath.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810fastpath.c
@@ -116,14 +116,10 @@ static void i810_render_elements_direct( struct vertex_buffer *VB )
    if (imesa->new_state)
       i810DDUpdateHwState( ctx );
 
-   BEGIN_CLIP_LOOP(imesa) 
-      {
-	 do {
-	    func( VB, 0, nr, 0 );
-	 } while (ctx->Driver.MultipassFunc &&
-		  ctx->Driver.MultipassFunc( VB, ++p ));
-      }
-   END_CLIP_LOOP(imesa);
+   do {
+      func( VB, 0, nr, 0 );
+   } while (ctx->Driver.MultipassFunc &&
+	    ctx->Driver.MultipassFunc( VB, ++p ));
 }
 
 
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810glx.h b/xc/lib/GL/mesa/src/drv/i810/i810glx.h
deleted file mode 100644
index df5c08674..000000000
--- a/xc/lib/GL/mesa/src/drv/i810/i810glx.h
+++ /dev/null
@@ -1,44 +0,0 @@
-/*
- * GLX Hardware Device Driver for Intel i810
- * Copyright (C) 1999 Wittawat Yamwong
- *
- * 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
- * WITTAWAT YAMWONG, 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 I810GLX_INC
-#define I810GLX_INC
-
-void i810DumpDB(struct i810_dest_buffer *buf);
-void i810GLXCreateDepthBuffer(GLcontext* ctx);
-void i810GLXDestroyImage(GLXImage* image);
-GLXImage* i810GLXCreateImage(WindowPtr pwindow, int depth, int width, int height);
-GLboolean i810GLXMakeCurrent( XSMesaContext c );
-GLboolean i810GLXBindBuffer( XSMesaContext c, XSMesaBuffer b );
-XSMesaContext i810GLXCreateContext( XSMesaVisual v,  
-                                   XSMesaContext share_list );
-void i810GLXDestroyContext( XSMesaContext c );
-
-extern int i810BackToFront(DrawablePtr drawable, struct i810_dest_buffer *buf);
-extern void (*i810GLXSwapBuffers)(XSMesaBuffer b);
-extern void i810PerformanceBoxes( int is_direct );
-
-
-#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..93e2f4492
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/i810/i810ioctl.c
@@ -0,0 +1,436 @@
+#include <stdio.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 int _tot_used, _tot_size;
+
+void i810FlushGeneralLocked( i810ContextPtr imesa )
+{
+   int retcode;
+   drm_i810_general_t dma;
+   drmBufPtr buf = imesa->dma_buffer;
+
+   if (!buf) {
+      fprintf(stderr, "i810FlushGeneralLocked: no buffer\n");
+      return;
+   }
+   
+   _tot_used += buf->used;
+   _tot_size += buf->total;
+
+   dma.idx = buf->idx;
+   dma.used = buf->used;
+
+   if (I810_DEBUG&DEBUG_VERBOSE_IOCTL)
+      fprintf(stderr, "DRM_IOCTL_I810_DMA idx %d used %d\n",
+	   dma.idx, dma.used);
+	      
+   if ((retcode = ioctl(imesa->driFd, DRM_IOCTL_I810_DMA, &dma))) {
+      printf("send dma retcode = %d\n", retcode);
+      exit(1);
+   }
+
+   imesa->dma_buffer = 0;
+
+   if (I810_DEBUG&DEBUG_VERBOSE_IOCTL)
+      fprintf(stderr, "finished general dma put\n");
+}
+
+static drmBufPtr i810_get_buffer_ioctl( i810ContextPtr imesa )
+{
+   int idx = 0;
+   int size = 0;
+   drmDMAReq dma;
+   int retcode;
+   drmBufPtr buf;
+   int cnt = 0;
+   
+
+   if (I810_DEBUG&DEBUG_VERBOSE_IOCTL)
+      fprintf(stderr,  "Getting dma buffer\n");
+   
+   dma.context = imesa->hHWContext;
+   dma.send_count = 0;
+   dma.send_list = NULL;
+   dma.send_sizes = NULL;
+   dma.flags = 0;
+   dma.request_count = 1;
+   dma.request_size = I810_DMA_BUF_SZ;
+   dma.request_list = &idx;
+   dma.request_sizes = &size;
+   dma.granted_count = 0;
+
+   if (I810_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(imesa->driFd, &dma);
+
+      if (I810_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);
+
+      if (retcode == 0 && 
+	  dma.request_list[0] && 
+	  dma.request_sizes[0] &&
+	  dma.granted_count) 
+	 break;
+
+      if (++cnt > 1000) {
+
+	 if (I810_DEBUG&DEBUG_VERBOSE_IOCTL) {
+	    fprintf(stderr, "\n\nflush retcode %d idx %d sz %d count %d\n", 
+		    retcode, dma.request_list[0], 
+		    dma.request_sizes[0], dma.granted_count);
+	 
+	    fprintf(stderr, "used %d size %d (ratio %.3f)\n", 
+		    _tot_used,  _tot_size, (float)_tot_size / (float)_tot_used);
+	 }
+
+	 ioctl(imesa->driFd, DRM_IOCTL_I810_FLUSH);
+      }
+   }
+
+
+
+   buf = &(imesa->i810Screen->bufs->list[idx]);
+   buf->used = 0;
+   
+
+   if (I810_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);
+
+   return buf;
+}
+
+
+
+void i810GetGeneralDmaBufferLocked( i810ContextPtr imesa )
+{
+   if (imesa->dma_buffer) {
+      fprintf(stderr, "i810GetGeneralDmaBufferLocked - dma_buffer not zero\n");
+      exit(1);
+   }
+
+   imesa->dma_buffer = i810_get_buffer_ioctl( imesa );
+}
+
+
+
+/* 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;
+
+   if (0) fprintf(stderr, "waitagelocked\n");
+
+   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 %p\n", buffer);
+
+   if (!buffer)
+      return;
+
+   if (imesa->dirty & ~I810_EMIT_CLIPRECT)
+      i810EmitHwStateLocked( imesa );
+
+
+   if (I810_DEBUG&DEBUG_VERBOSE_IOCTL)
+      fprintf(stderr, "i810FlushVerticesLocked, used %d\n",
+	      buffer->used);
+   
+   _tot_used += buffer->used;
+   _tot_size += buffer->total;
+
+   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_EMIT_CLIPRECT;
+
+   if (!vertex.used || !(imesa->dirty & I810_EMIT_CLIPRECT)) 
+   {
+      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 orig_dwords = dwords;
+   GLuint *start;
+
+   dwords++;
+   dwords&=~1;
+
+   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 );
+      UNLOCK_HARDWARE(imesa);
+   } 
+   else if (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 );
+      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;
+
+   if (orig_dwords & 1) 
+      *start++ = 0;
+
+   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..ac213bf1e
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/i810/i810ioctl.h
@@ -0,0 +1,32 @@
+#ifndef MGA_IOCTL_H
+#define MGA_IOCTL_H
+
+#include "i810context.h"
+
+
+GLuint *i810AllocDwords( i810ContextPtr imesa, int dwords );
+
+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 );
+
+
+#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
index ec07995a0..ad80674f0 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810lib.h
+++ b/xc/lib/GL/mesa/src/drv/i810/i810lib.h
@@ -28,29 +28,34 @@
 
 #include <stdio.h>
 
-#include "i810common.h"
-#include "i810buf.h"
 #include "i810context.h"
 #include "mm.h"
 #include "i810log.h"
-#include "i810dma.h"
 
-#ifndef MESA31
-#error "The I810 driver now requires Mesa 3.1 or higher"
-#endif
 
+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;
 
-   /* dma stuff */
-   GLuint dmaDriver;	
-   GLuint dmaSize;
-   GLuint dmaAdr;
-   GLuint cmdSize;
-
    /* bookkeeping for texture swaps */
    GLuint dma_buffer_age;
    GLuint current_texture_age;
@@ -68,37 +73,28 @@ typedef struct {
    GLuint c_lines;
    GLuint c_drawWaits;
    GLuint c_textureSwaps;
-   GLuint c_signals;
    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 */
 
-   /* Primitive managment (like warpSerie) */
-   GLuint prim_start;
-   GLuint prim_dwords;
-
-   /* card == dcache
-    * sysmem == normal gart memory
-    */
-   memHeap_t *cardHeap;
-   memHeap_t *sysmemHeap;
-   unsigned char *sysmemVirtual;
+   unsigned char *texVirtual;
 
-   i810BatchBuffer	*dma_buffer;
+   
+   struct i810_batch_buffer *dma_buffer;
+   struct i810_ring_buffer LpRing;
+   unsigned char *MMIOBase;
 
 } i810Glx_t;
 
 extern i810Glx_t	i810glx;
 
-int	i810_usec( void );
-void i810LibInit();
-void i810SoftReset();
-
-int i810MakeCurrent(i810ContextPtr ctx, struct i810_dest_buffer *buf);
-void i810LoadTexturePalette(unsigned short *pal, int start, int len);
 
 #define I810PACKCOLOR1555(r,g,b,a) \
   ((((r) & 0xf8) << 7) | (((g) & 0xf8) << 2) | (((b) & 0xf8) >> 3) | \
@@ -123,6 +119,7 @@ static __inline__ GLuint i810PackColor(GLuint format,
   case DV_PF_565:
     return I810PACKCOLOR565(r,g,b);
   default:
+     fprintf(stderr, "unknown format %d\n", (int)format);
     return 0;
   }
 }
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810pipeline.c b/xc/lib/GL/mesa/src/drv/i810/i810pipeline.c
index 2cd53b42e..06269a818 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810pipeline.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810pipeline.c
@@ -1,15 +1,16 @@
 /*  #include "i810pipeline.h" */
 
 #include <stdio.h>
-#include "xsmesaP.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"
-
-#if 0
 
 static struct gl_pipeline_stage i810_fast_stage = {
    "I810 fast path",
@@ -68,11 +69,6 @@ GLboolean i810DDBuildPrecalcPipeline( GLcontext *ctx )
 }
 
 
-#endif
-
-
-
-
 
 GLuint i810DDRegisterPipelineStages( struct gl_pipeline_stage *out,
 				    const struct gl_pipeline_stage *in,
@@ -83,7 +79,6 @@ GLuint i810DDRegisterPipelineStages( struct gl_pipeline_stage *out,
    for (i = o = 0 ; i < nr ; i++) {
       switch (in[i].ops) {
 
-#if 0
       case PIPE_OP_RAST_SETUP_0:
 	 out[o] = in[i];
 	 out[o].cva_state_change = NEW_LIGHTING|NEW_TEXTURING|NEW_RASTER_OPS;
@@ -98,7 +93,6 @@ GLuint i810DDRegisterPipelineStages( struct gl_pipeline_stage *out,
 	 out[o].run = i810DDDoRasterSetup;
 	 o++;
 	 break;
-#endif
 
 	 /* Completely replace Mesa's fog processing to generate fog
 	  * coordinates instead of messing with colors.
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810pipeline.h b/xc/lib/GL/mesa/src/drv/i810/i810pipeline.h
index 350f23431..87b95455d 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810pipeline.h
+++ b/xc/lib/GL/mesa/src/drv/i810/i810pipeline.h
@@ -7,4 +7,8 @@ extern GLuint i810DDRegisterPipelineStages( struct gl_pipeline_stage *out,
 
 extern GLboolean i810DDBuildPrecalcPipeline( GLcontext *ctx ); 
 
+extern void i810DDFastPath( struct vertex_buffer *VB );
+extern void i810DDFastPathInit( void );
+
+
 #endif
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810ring.c b/xc/lib/GL/mesa/src/drv/i810/i810ring.c
deleted file mode 100644
index 2e5909e9f..000000000
--- a/xc/lib/GL/mesa/src/drv/i810/i810ring.c
+++ /dev/null
@@ -1,162 +0,0 @@
-#include <stdlib.h>
-#include <errno.h>
-#include <unistd.h>
-#include <sys/mman.h>
-#include <stdio.h>
-#include <sys/time.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <fcntl.h>
-#include <signal.h>
-
-#include "mm.h"
-#include "i810dd.h"
-#include "i810lib.h"
-#include "i810state.h"
-
-
-#include "i810dma.h"
-
-static void 
-I810PrintErrorState( void )
-{
-   fprintf(stderr,  "pgetbl_ctl: 0x%x pgetbl_err: 0x%x\n", 
-	   INREG(PGETBL_CTL),
-	   INREG(PGE_ERR));
-
-   fprintf(stderr,  "ipeir: %x iphdr: %x\n", 
-	   INREG(IPEIR),
-	   INREG(IPEHR));
-
-   fprintf(stderr,  "LP ring tail: %x head: %x len: %x start %x\n",
-	   INREG(LP_RING + RING_TAIL),
-	   INREG(LP_RING + RING_HEAD) & HEAD_ADDR,
-	   INREG(LP_RING + RING_LEN),
-	   INREG(LP_RING + RING_START));
-
-   fprintf(stderr,  "eir: %x esr: %x emr: %x\n",
-	   INREG16(EIR),
-	   INREG16(ESR),
-	   INREG16(EMR));
-
-   fprintf(stderr,  "instdone: %x instpm: %x\n",
-	   INREG16(INST_DONE),
-	   INREG8(INST_PM));
-
-   fprintf(stderr,  "memmode: %x instps: %x\n",
-	   INREG(MEMMODE),
-	   INREG(INST_PS));
-
-   fprintf(stderr,  "hwstam: %x ier: %x imr: %x iir: %x\n",
-	   INREG16(HWSTAM),
-	   INREG16(IER),
-	   INREG16(IMR),
-	   INREG16(IIR));
-}
-
-
-
-
-static int
-I810USec( void ) 
-{
-   struct timeval tv;
-   struct timezone tz;	
-   gettimeofday( &tv, &tz );
-   return (tv.tv_sec & 2047) * 1000000 + tv.tv_usec;
-}
-
-
-
-void
-_I810RefreshLpRing( i810ContextPtr imesa, int update )
-{
-	struct i810_ring_buffer *ring = &(i810glx.LpRing);
-	
-	ring->head = INREG(LP_RING + RING_HEAD) & HEAD_ADDR;
-	ring->tail = INREG(LP_RING + RING_TAIL);
-	ring->space = ring->head - (ring->tail+8);
-
-	if (ring->space < 0) {
-		ring->space += ring->mem.Size;
-		if (update)
-			imesa->sarea->lastWrap = ++imesa->sarea->ringAge;
-	}
-}	
-
-int
-_I810WaitLpRing( i810ContextPtr imesa, int n, int timeout_usec )
-{
-	struct i810_ring_buffer *ring = &(i810glx.LpRing);
-	int iters = 0;
-	int startTime = 0;
-	int curTime = 0;
-   
-	if (timeout_usec == 0)
-		timeout_usec = 15000000;
-
-	if (I810_DEBUG & DEBUG_VERBOSE_API)
-		fprintf(stderr, "I810WaitLpRing %d\n", n);
-
-	while (ring->space < n) 
-	{
-		int i;
-
-		ring->head = INREG(LP_RING + RING_HEAD) & HEAD_ADDR;
-		ring->space = ring->head - (ring->tail+8);
-
-		if (ring->space < 0) {
-			imesa->sarea->lastWrap = ++imesa->sarea->ringAge;
-			ring->space += ring->mem.Size;
-		}
-      
-		iters++;
-		curTime = I810USec();
-		if ( startTime == 0 || curTime < startTime /*wrap case*/) {
-			startTime = curTime;
-		} else if ( curTime - startTime > timeout_usec ) { 
-			I810PrintErrorState();
-			fprintf(stderr, "space: %d wanted %d\n", 
-				ring->space, n );
-			UNLOCK_HARDWARE(imesa);
-			exit(1);
-		}
-
-		for (i = 0 ; i < 2000 ; i++)
-			;
-	}
-
-	return iters;
-}
-
-int
-_I810Sync( i810ContextPtr imesa ) 
-{
-	int rv;
-
-	if (I810_DEBUG & DEBUG_VERBOSE_API)
-		fprintf(stderr, "I810Sync\n");
-   
-	/* Send a flush instruction and then wait till the ring is empty.
-	 * This is stronger than waiting for the blitter to finish as it also
-	 * flushes the internal graphics caches.
-	 */
-	{
-		BEGIN_LP_RING( imesa, 2 );   
-		OUT_RING( INST_PARSER_CLIENT | INST_OP_FLUSH );
-		OUT_RING( 0 );		/* pad to quadword */
-		ADVANCE_LP_RING();
-	}
-
-
-	i810glx.LpRing.synced = 1;	/* ?? */
-
-	rv =  _I810WaitLpRing( imesa, i810glx.LpRing.mem.Size - 8, 0 );	
-	imesa->sarea->lastSync = ++imesa->sarea->ringAge;
-
-	return rv;
-}
-
-
-
-
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810span.c b/xc/lib/GL/mesa/src/drv/i810/i810span.c
index ce55c6f70..c47bf2f40 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810span.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810span.c
@@ -1,149 +1,174 @@
 #include "types.h"
-#include "i810buf.h"
 #include "i810dd.h"
 #include "i810lib.h"
 #include "i810dma.h"
 #include "i810log.h"
+#include "i810span.h"
+#include "i810ioctl.h"
 
-static void (*xsmWriteRGBASpan)( const GLcontext *ctx,
-				 GLuint n, GLint x, GLint y,
-				 CONST GLubyte rgba[][4], 
-				 const GLubyte mask[] );
-static void (*xsmWriteRGBSpan)( const GLcontext *ctx,
-				GLuint n, GLint x, GLint y,
-				CONST GLubyte rgb[][3], const GLubyte mask[] );
-static void (*xsmWriteMonoRGBASpan)(const GLcontext *ctx, GLuint n, 
-				    GLint x, GLint y,const GLubyte mask[] );
-static void (*xsmWriteRGBAPixels)(const GLcontext *ctx,
-				  GLuint n, const GLint x[], const GLint y[],
-				  CONST GLubyte rgba[][4], 
-				  const GLubyte mask[] );
-static void (*xsmWriteMonoRGBAPixels)(const GLcontext *ctx,GLuint n, 
-				      const GLint x[], const GLint y[],
-				      const GLubyte mask[] );
-static void (*xsmReadRGBASpan)(const GLcontext *ctx, GLuint n, 
-			       GLint x, GLint y,GLubyte rgba[][4] );
-static void (*xsmReadRGBAPixels)(const GLcontext *ctx,GLuint n, 
-				 const GLint x[], const GLint y[],
-				 GLubyte rgba[][4], const GLubyte mask[] );
-
-/* Wrapper functions */
-
-static void WriteRGBASpan( const GLcontext *ctx,
-			   GLuint n, GLint x, GLint y,
-			   CONST GLubyte rgba[][4], 
-			   const GLubyte mask[] )
-{
-  i810Msg(12,__FUNCTION__ "\n"); 
-  i810WaitDrawingEngine();
-  (*xsmWriteRGBASpan)(ctx,n,x,y,rgba,mask);
-}
 
-static void WriteRGBSpan( const GLcontext *ctx,
-			  GLuint n, GLint x, GLint y,
-			  CONST GLubyte rgb[][3], const GLubyte mask[] )
-{
-  i810Msg(12,__FUNCTION__ "\n");
-  i810WaitDrawingEngine();
-  (*xsmWriteRGBSpan)(ctx,n,x,y,rgb,mask);
-}
+#define DBG 0
 
-static void WriteMonoRGBASpan(const GLcontext *ctx, GLuint n, 
-			      GLint x, GLint y,const GLubyte mask[] )
-{
-  i810Msg(12,__FUNCTION__ "\n");
-  i810WaitDrawingEngine();
-  (*xsmWriteMonoRGBASpan)(ctx,n,x,y,mask);
-}
+#define LOCAL_VARS					\
+   __DRIdrawablePrivate *dPriv = imesa->driDrawable;	\
+   i810ScreenPrivate *i810Screen = imesa->i810Screen;	\
+   GLuint pitch = i810Screen->backPitch;		\
+   GLuint height = dPriv->h;				\
+   char *buf = (char *)(imesa->drawMap +		\
+			dPriv->x * 2 +			\
+			dPriv->y * pitch);		\
+   char *read_buf = (char *)(imesa->readMap +		\
+			     dPriv->x * 2 +		\
+			     dPriv->y * pitch); 	\
+   GLushort p = I810_CONTEXT( ctx )->MonoColor;         \
+   (void) read_buf; (void) buf; (void) p
 
-static void WriteRGBAPixels(const GLcontext *ctx,
-			    GLuint n, const GLint x[], const GLint y[],
-			    CONST GLubyte rgba[][4], 
-			    const GLubyte mask[] )
-{
-  i810Msg(12,__FUNCTION__ "\n");
-  i810WaitDrawingEngine();
-  (*xsmWriteRGBAPixels)(ctx,n,x,y,rgba,mask);
-}
+#define LOCAL_DEPTH_VARS				\
+   __DRIdrawablePrivate *dPriv = imesa->driDrawable;	\
+   i810ScreenPrivate *i810Screen = imesa->i810Screen;	\
+   GLuint pitch = i810Screen->backPitch;		\
+   GLuint height = dPriv->h;				\
+   char *buf = (char *)(i810Screen->depth.map +	\
+			dPriv->x * 2 +			\
+			dPriv->y * pitch)
 
-static void WriteMonoRGBAPixels(const GLcontext *ctx,GLuint n, 
-				const GLint x[], const GLint y[],
-				const GLubyte mask[] )
-{
-  i810Msg(12,__FUNCTION__ "\n");
-  i810WaitDrawingEngine();
-  (*xsmWriteMonoRGBAPixels)(ctx,n,x,y,mask);
-}
+#define INIT_MONO_PIXEL(p) 
 
-static void ReadRGBASpan(const GLcontext *ctx, GLuint n, 
-			 GLint x, GLint y,GLubyte rgba[][4] )
-{
-  i810Msg(12,__FUNCTION__ "\n");
-  i810WaitDrawingEngine();
-  (*xsmReadRGBASpan)(ctx,n,x,y,rgba);
-}
+#define CLIPPIXEL(_x,_y) (_x >= minx && _x < maxx && \
+			  _y >= miny && _y < maxy)
 
-static void ReadRGBAPixels(const GLcontext *ctx,GLuint n, 
-			   const GLint x[], const GLint y[],
-			   GLubyte rgba[][4], const GLubyte mask[] )
-{
-  i810Msg(12,__FUNCTION__ "\n");
-  i810WaitDrawingEngine();
-  (*xsmReadRGBAPixels)(ctx,n,x,y,rgba,mask);
-}
 
-#define WRAP(x) 				\
-    xsm ## x = ctx->Driver.x;			\
-    ctx->Driver.x = x
+#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;	\
+         }
 
-void i810DDInitSpans( GLcontext *ctx )
-{
-   /* Need another hook from mesa to synchronize hardware before
-    * write-pixels and read-spans/pixels.
-    */ 
-   WRAP(WriteRGBASpan);
-   WRAP(WriteRGBSpan);
-   WRAP(WriteMonoRGBASpan);
-   WRAP(WriteRGBAPixels);
-   WRAP(WriteMonoRGBAPixels);
-   WRAP(ReadRGBASpan);
-   WRAP(ReadRGBAPixels);
-}
+#define Y_FLIP(_y) (height - _y - 1)
 
 
+#define HW_LOCK()				\
+   i810ContextPtr imesa = I810_CONTEXT(ctx);	\
+   LOCK_HARDWARE_QUIESCENT(imesa);
 
+#define HW_CLIPLOOP()						\
+  do {								\
+    __DRIdrawablePrivate *dPriv = imesa->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(imesa);
 
 
 
-void fxUpdateDDSpanPointers(GLcontext *ctx)
-{
-  ctx->Driver.DepthTestSpan=fxDDDepthTestSpanGeneric;
-  ctx->Driver.DepthTestPixels=fxDDDepthTestPixelsGeneric;
 
-  ctx->Driver.ReadDepthSpanFloat=fxDDReadDepthSpanFloat;
-  ctx->Driver.ReadDepthSpanInt=fxDDReadDepthSpanInt;
-}
+/* 16 bit, 565 rgb 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] = 0;			/* or 255? */			\
+} while(0)
+
+#define TAG(x) i810##x##_565
+#include "spantmp.h"
+
+
+
+
+/* 15 bit, 555 rgb 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) << 3) |	\
+                         		    ((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 >> 7) & 0xf8;					\
+   rgba[1] = (p >> 3) & 0xf8;					\
+   rgba[2] = (p << 3) & 0xf8;					\
+   rgba[3] = 0;			/* or 255? */			\
+} while(0)
+
+#define TAG(x) i810##x##_555
+#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);	 
+
+/*     d = 0xffff; */
+
+#define TAG(x) i810##x##_16
+#include "depthtmp.h"
+
+
 
-void fxSetupDDSpanPointers(GLcontext *ctx)
+void i810DDInitSpanFuncs( GLcontext *ctx )
 {
-  ctx->Driver.WriteRGBASpan       =fxDDWriteRGBASpan;
-  ctx->Driver.WriteRGBSpan        =fxDDWriteRGBSpan;
-  ctx->Driver.WriteMonoRGBASpan   =fxDDWriteMonoRGBASpan;
-  ctx->Driver.WriteRGBAPixels     =fxDDWriteRGBAPixels;
-  ctx->Driver.WriteMonoRGBAPixels =fxDDWriteMonoRGBAPixels;
-
-  ctx->Driver.WriteCI8Span        =NULL;
-  ctx->Driver.WriteCI32Span       =NULL;
-  ctx->Driver.WriteMonoCISpan     =NULL;
-  ctx->Driver.WriteCI32Pixels     =NULL;
-  ctx->Driver.WriteMonoCIPixels   =NULL;
-
-  ctx->Driver.ReadRGBASpan        =fxDDReadRGBASpan;
-  ctx->Driver.ReadRGBAPixels      =fxDDReadRGBAPixels;
-
-  ctx->Driver.ReadCI32Span        =NULL;
-  ctx->Driver.ReadCI32Pixels      =NULL;
+   if (1) {
+      ctx->Driver.WriteRGBASpan = i810WriteRGBASpan_565;
+      ctx->Driver.WriteRGBSpan = i810WriteRGBSpan_565;
+      ctx->Driver.WriteMonoRGBASpan = i810WriteMonoRGBASpan_565;
+      ctx->Driver.WriteRGBAPixels = i810WriteRGBAPixels_565;
+      ctx->Driver.WriteMonoRGBAPixels = i810WriteMonoRGBAPixels_565; 
+      ctx->Driver.ReadRGBASpan = i810ReadRGBASpan_565;
+      ctx->Driver.ReadRGBAPixels = i810ReadRGBAPixels_565;
+   } else {
+      ctx->Driver.WriteRGBASpan = i810WriteRGBASpan_555;
+      ctx->Driver.WriteRGBSpan = i810WriteRGBSpan_555;
+      ctx->Driver.WriteMonoRGBASpan = i810WriteMonoRGBASpan_555;
+      ctx->Driver.WriteRGBAPixels = i810WriteRGBAPixels_555;
+      ctx->Driver.WriteMonoRGBAPixels = i810WriteMonoRGBAPixels_555;
+      ctx->Driver.ReadRGBASpan = i810ReadRGBASpan_555;
+      ctx->Driver.ReadRGBAPixels = i810ReadRGBAPixels_555;
+   }
+
+   ctx->Driver.ReadDepthSpan = i810ReadDepthSpan_16;
+   ctx->Driver.WriteDepthSpan = i810WriteDepthSpan_16;
+   ctx->Driver.ReadDepthPixels = i810ReadDepthPixels_16;
+   ctx->Driver.WriteDepthPixels = i810WriteDepthPixels_16;
+
+   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/i810/i810span.h b/xc/lib/GL/mesa/src/drv/i810/i810span.h
index 07276cd86..17a52085b 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810span.h
+++ b/xc/lib/GL/mesa/src/drv/i810/i810span.h
@@ -1,6 +1,6 @@
 #ifndef _I810_SPAN_H
 #define _I810_SPAN_H
 
-extern void i810DDInitSpans( GLcontext *ctx );
+extern void i810DDInitSpanFuncs( GLcontext *ctx );
 
 #endif
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810state.c b/xc/lib/GL/mesa/src/drv/i810/i810state.c
index e4a77efc1..caf9fafd6 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810state.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810state.c
@@ -1,33 +1,33 @@
 
 #include <stdio.h>
 
-#include "xsmesaP.h"
-
 #include "types.h"
+#include "enums.h"
 #include "pb.h"
 #include "dd.h"
 
-#include "glx_log.h"
 #include "mm.h"
 #include "i810lib.h"
-#include "i810glx.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 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;
@@ -42,21 +42,17 @@ static void i810DDAlphaFunc(GLcontext *ctx, GLenum func, GLclampf ref)
 
    a |= ctx->Color.AlphaRef << ZA_ALPHAREF_SHIFT;
 
-   imesa->reg_dirty |= (1<<I810_CTXREG_ZA);
+   imesa->dirty |= I810_UPLOAD_CTX;
    imesa->Setup[I810_CTXREG_ZA] &= ~(ZA_ALPHA_MASK|ZA_ALPHAREF_MASK);
    imesa->Setup[I810_CTXREG_ZA] |= a;
 }
 
-/* This shouldn't get called, as the extension is disabled.  However,
- * there are internal Mesa calls, and rouge use of the api which must be
- * caught.
- */
 static void i810DDBlendEquation(GLcontext *ctx, GLenum mode) 
 {
    if (mode != GL_FUNC_ADD_EXT) {
       ctx->Color.BlendEquation = GL_FUNC_ADD_EXT;
-      FatalError("Unsupported blend equation");
-      gl_error( ctx, GL_INVALID_OPERATION, "glBlendEquation (disabled)");
+      i810Error("Unsupported blend equation");
+      exit(1);
    }
 }
 
@@ -65,6 +61,9 @@ static void i810DDBlendFunc(GLcontext *ctx, GLenum sfactor, GLenum dfactor)
    i810ContextPtr imesa = I810_CONTEXT(ctx);
    GLuint a;
 
+   FLUSH_BATCH(imesa);
+
+
    a = SDM_UPDATE_SRC_BLEND | SDM_UPDATE_DST_BLEND;
 
    switch (ctx->Color.BlendSrcRGB) {
@@ -80,7 +79,8 @@ static void i810DDBlendFunc(GLcontext *ctx, GLenum sfactor, GLenum dfactor)
       a |= SDM_SRC_SRC_ALPHA;	/* use GFXRENDERSTATE_COLOR_FACTOR ??? */
       break;
    default: 
-      FatalError("unknown blend source func");
+      i810Error("unknown blend source func");
+      exit(1);
       return;
    }
 
@@ -94,11 +94,12 @@ static void i810DDBlendFunc(GLcontext *ctx, GLenum sfactor, GLenum dfactor)
    case GL_DST_ALPHA:           a |= SDM_DST_ONE; break;
    case GL_ONE_MINUS_DST_ALPHA: a |= SDM_DST_ZERO; break;
    default: 
-      FatalError("unknown blend dst func");
+      i810Error( "unknown blend dst func");
+      exit(1);
       return;      
    }  
 
-   imesa->reg_dirty |= (1<<I810_CTXREG_SDM);
+   imesa->dirty |= I810_UPLOAD_CTX;
    imesa->Setup[I810_CTXREG_SDM] &= ~(SDM_SRC_MASK|SDM_DST_MASK);
    imesa->Setup[I810_CTXREG_SDM] |= a;
 }
@@ -124,6 +125,8 @@ 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;
@@ -135,43 +138,161 @@ static void i810DDDepthFunc(GLcontext *ctx, GLenum func)
    case GL_NOTEQUAL: zmode = LCS_Z_NOTEQUAL; break;
    default: return;
    }
-
-   /* Could just fire it off now...
-    */
+   
    imesa->Setup[I810_CTXREG_LCS] &= ~LCS_Z_MASK;
    imesa->Setup[I810_CTXREG_LCS] |= LCS_UPDATE_ZMODE | zmode;
-   imesa->reg_dirty |= (1<<I810_CTXREG_LCS);
+   
+   imesa->dirty |= I810_UPLOAD_CTX;
 }
 
 static void i810DDDepthMask(GLcontext *ctx, GLboolean flag)
 {
    i810ContextPtr imesa = I810_CONTEXT(ctx);
 
-   imesa->reg_dirty |= (1<<I810_CTXREG_B2);
+   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;
+}
 
-/*
-   Possible technique to reduce the amount of checking for small state
-   changes.  Large ones detect wrap on update_count & just send the
-   whole state...
 
-   {
-      imesa->Update[imesa->update_count & MAX_UPDATE] = 
-	 ( imesa->Setup[I810_CTXREG_B2] & 
-	   (B1_ZB_WRITE_ENABLE | B1_UPDATE_ZB_WRITE_ENABLE | GFX_OP_BOOL_1)
-	 ); 
-      imesa->update_count++;
-   } 
-*/
+/* =============================================================
+ * 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->poly_stipple = ( ((p[0] & 0xf) << 0) |
+			   ((p[1] & 0xf) << 4) |
+			   ((p[2] & 0xf) << 8) |
+			   ((p[3] & 0xf) << 12) );   
+   
+   if (1) {
+      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;
 }
 
 
+/* ============================================================
+ * Line stipple -- not completed 
+ *  - need to cope with stipplecounter
+ */
+#if 0
+static void i810DDLineStipple( GLcontext *ctx, GLint factor, GLushort pattern )
+{
+   i810ContextPtr imesa = I810_CONTEXT(ctx);
+   GLushort p, h, test;
+
+   FLUSH_BATCH(imesa);
+   ctx->Driver.TriangleCaps |= DD_LINE_STIPPLE;
+   
+   if (ctx->Line.StippleFlag) {
+      imesa->lines_are_tris |= I810_LINE_STIPPLE;
+
+      if (ctx->PB->primitive == GL_LINE) {
+	 imesa->dirty |= I810_UPLOAD_CTX;
+	 imesa->Setup[I810_CTXREG_ST1] &= ~ST1_ENABLE;
+      }
+   }
+ 
+   switch (factor) {
+   case 1:
+      p = pattern & 0xf;
+      p |= p << 4;
+      p |= p << 8;
+      test = p;
+      break;
+   case 2:
+      p = pattern & 1;
+      p |= (pattern & 2) << 2;
+      p |= p << 1;
+      p |= p << 4;
+      p |= p << 8;
+      test = pattern & 3;
+      test |= test << 2;
+      test |= test << 4;
+      test |= test << 8;
+      break;
+   default:
+      p = 0;
+      if (pattern)
+	 p = 0xffff;
+      test = p;
+      break;
+   }
+      
+   if (pattern != test) {
+      ctx->Driver.TriangleCaps &= ~DD_LINE_STIPPLE;
+      imesa->lines_are_tris &= ~I810_LINE_STIPPLE;
+      return;
+   }
+
+   h = (((p & 1) << 0) |
+	((p & 2) << 4) |
+	((p & 3) << 8) |
+	((p & 4) << 12));
+
+   h |= h << 1;
+   h |= h << 2;
+      
+   imesa->vert_line_stipple = p;
+   imesa->horiz_line_stipple = h;
+   
+   if (ctx->Line.Stipple && ctx->PB->primitive == GL_LINE)
+      imesa->Setup[I810_CTXREG_ST1] |= ST1_ENABLE;
+}
+#endif 
+
 
+/* ============================================================
+ * Antialiased lines and triangles
+ */
 
+/* Nothing to do...
+ */
 
 
 
@@ -184,36 +305,105 @@ 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);
+
+   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;
+
+
+   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);
+
+
+   imesa->dirty |= I810_EMIT_CLIPRECT;
+}
+
+
+static void i810DDDither(GLcontext *ctx, GLboolean enable)
+{
+}
+
+
+static GLboolean i810DDSetDrawBuffer(GLcontext *ctx, GLenum mode )
 {
    i810ContextPtr imesa = I810_CONTEXT(ctx);
-   int x1,x2,y1,y2;
 
-   x1 = ctx->Scissor.X;
-   x2 = ctx->Scissor.X + ctx->Scissor.Width - 1;
-   y1 = i810DB->Height - ctx->Scissor.Y - ctx->Scissor.Height;
-   y2 = i810DB->Height - ctx->Scissor.Y - 1;
-  
-   if (x1 < 0) x1 = 0;
-   if (y1 < 0) y1 = 0;
-   if (x2 >= i810DB->Width) x2 = i810DB->Width-1;
-   if (y2 >= i810DB->Height) y2 = i810DB->Height-1;
-
-   if (x1 > x2 || y1 > y2) {
-      x1 = 0; x2 = 0;
-      y2 = 0; y1 = 1;
+   FLUSH_BATCH(imesa);
+
+   imesa->Fallback &= ~I810_FALLBACK_DRAW_BUFFER;
+
+   if (mode == GL_FRONT_LEFT) 
+   {
+      imesa->drawMap = imesa->driScreen->pFB;
+      imesa->BufferSetup[I810_DESTREG_DI1] = (imesa->i810Screen->fbOffset | 
+					      imesa->i810Screen->backPitchBits);
+      imesa->dirty |= I810_UPLOAD_BUFFERS;
+      i810XMesaSetFrontClipRects( imesa );
+      return GL_TRUE;
+   } 
+   else if (mode == GL_BACK_LEFT) 
+   {
+      imesa->drawMap = imesa->i810Screen->back.map;
+      imesa->BufferSetup[I810_DESTREG_DI1] = (imesa->i810Screen->backOffset | 
+					      imesa->i810Screen->backPitchBits);
+      imesa->dirty |= I810_UPLOAD_BUFFERS;
+      i810XMesaSetBackClipRects( imesa );
+      return GL_TRUE;
    }
 
-   /* Turn it off if full screen?  
-    */
-   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;
+   imesa->Fallback |= I810_FALLBACK_DRAW_BUFFER;
+   return GL_FALSE;
+}
 
+static void i810DDSetReadBuffer(GLcontext *ctx, GLframebuffer *colorBuffer,
+				GLenum mode )
+{
+   i810ContextPtr imesa = I810_CONTEXT(ctx);
 
-   imesa->reg_dirty |= ((1<<I810_CTXREG_SCI0) |
-			  (1<<I810_CTXREG_SCI1) |
-			  (1<<I810_CTXREG_SCI2));
+   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, 
+			   GLubyte r, GLubyte g,
+			   GLubyte b, GLubyte a )
+{
+   i810ContextPtr imesa = I810_CONTEXT(ctx);
+
+   imesa->MonoColor = i810PackColor( imesa->i810Screen->fbFormat,
+				      r, g, b, a );
+}
+
+
+static void i810DDClearColor(GLcontext *ctx, 
+			     GLubyte r, GLubyte g,
+			     GLubyte b, GLubyte a )
+{
+   i810ContextPtr imesa = I810_CONTEXT(ctx);
+
+   imesa->ClearColor = i810PackColor( imesa->i810Screen->fbFormat,
+				      r, g, b, a );
 }
 
 
@@ -226,6 +416,8 @@ 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)
@@ -238,25 +430,55 @@ static void i810DDCullFaceFrontFace(GLcontext *ctx, GLenum unused)
 
    if (ctx->Polygon.CullFlag && ctx->PB->primitive == GL_POLYGON)
    {
-      imesa->reg_dirty |= (1<<I810_CTXREG_LCS);
+      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->reg_dirty |= (1<<I810_CTXREG_LCS);
-      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);
+
+   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:
+#if 0      
+      if (ctx->Line.StippleFlag && (ctx->Driver.TriangleCaps & DD_LINE_STIPPLE))
+	 imesa->Setup[I810_CTXREG_ST1] |= ST1_ENABLE;
+#endif
+      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;
    }
 }
 
@@ -266,15 +488,36 @@ 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
@@ -284,7 +527,16 @@ static GLboolean i810DDColorMask(GLcontext *ctx,
 static void i810DDLightModelfv(GLcontext *ctx, GLenum pname, 
 			      const GLfloat *param)
 {
-   if (pname == GL_LIGHT_MODEL_COLOR_CONTROL) {
+   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;
+
       I810_CONTEXT(ctx)->new_state |= I810_NEW_TEXTURE;
    }
 }
@@ -295,7 +547,7 @@ static void i810DDLightModelfv(GLcontext *ctx, GLenum pname,
  * Fog
  */
 
-void i810DDFogfv(GLcontext *ctx, GLenum pname, const GLfloat *param)
+static void i810DDFogfv(GLcontext *ctx, GLenum pname, const GLfloat *param)
 {
    i810ContextPtr imesa = I810_CONTEXT(ctx);
 
@@ -304,7 +556,7 @@ void i810DDFogfv(GLcontext *ctx, GLenum pname, const GLfloat *param)
 		      ((GLubyte)(ctx->Fog.Color[1]*255.0F) << 8) |
 		      ((GLubyte)(ctx->Fog.Color[2]*255.0F) << 0));
 
-      imesa->reg_dirty |= (1<<I810_CTXREG_FOG);
+      imesa->dirty |= I810_UPLOAD_CTX;
       imesa->Setup[I810_CTXREG_FOG] = ((GFX_OP_FOG_COLOR | color) &
 				      ~FOG_RESERVED_MASK);
    }
@@ -320,66 +572,126 @@ static void i810DDEnable(GLcontext *ctx, GLenum cap, GLboolean state)
 
    switch(cap) {
    case GL_ALPHA_TEST:
-      imesa->reg_dirty |= (1<<I810_CTXREG_B1);
+      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:
-      imesa->reg_dirty |= (1<<I810_CTXREG_B1);
+      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:
-      imesa->reg_dirty |= (1<<I810_CTXREG_B1);
+      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:
-      imesa->reg_dirty |= (1<<I810_CTXREG_SC);
-      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;
+      FLUSH_BATCH(imesa);
+      imesa->scissor = state;
+      imesa->dirty |= I810_EMIT_CLIPRECT;
+      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_STIPPLE:
+#if 0
+      if (ctx->Driver.TriangleCaps & DD_LINE_STIPPLE) 
+      {
+	 imesa->lines_are_tris &= ~I810_LINE_STIPPLE;
+	 if (state)
+	    imesa->lines_are_tris |= I810_LINE_STIPPLE;
+
+	 if (ctx->PB->primitive == GL_LINE) {
+	    FLUSH_BATCH(imesa);
+	    imesa->dirty |= I810_UPLOAD_CTX;
+	    imesa->Setup[I810_CTXREG_ST1] &= ~ST1_ENABLE;
+	    if (state) 
+	       imesa->Setup[I810_CTXREG_ST1] |= ST1_ENABLE;
+	 }
+      }
+#endif
+      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:
-      imesa->reg_dirty |= (1<<I810_CTXREG_B1);
+      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) {
-	 imesa->reg_dirty |= (1<<I810_CTXREG_LCS);
+	 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:      
-      imesa->reg_dirty |= (1<<I810_CTXREG_MT);
+      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;
       }
-
-      I810_CONTEXT(ctx)->new_state |= I810_NEW_TEXTURE;
       break;
    default:
       ; 
@@ -387,26 +699,6 @@ static void i810DDEnable(GLcontext *ctx, GLenum cap, GLboolean state)
 }
 
 
-/* Don't use the regs parm yet.
- */
-static void i810UpdateRegs( GLcontext *ctx, GLuint regs ) 
-{
-   i810ContextPtr imesa = I810_CONTEXT(ctx);
-   int i;
-
-   {
-      BEGIN_BATCH( I810_CTX_SETUP_SIZE ); 
-   
-      for (i = 0 ; i < I810_CTX_SETUP_SIZE ; i++) {
-	    OUT_BATCH( imesa->Setup[i] );
-      }
-
-      ADVANCE_BATCH();
-   }
-   
-   imesa->reg_dirty = 0;
-}
-
 
 /* =============================================================
  */
@@ -417,50 +709,158 @@ void i810DDPrintState( const char *msg, GLuint state )
 {
    fprintf(stderr, "%s (0x%x): %s\n",
 	   msg,
-	   state,
+	   (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->reg_dirty) {
-      i810FinishPrimitive();
-      i810UpdateRegs( ctx, imesa->reg_dirty );
+
+/*
+ * 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++ ) {
+	 OUT_BATCH( code[i] );
+      }
+
+      ADVANCE_BATCH();
    }
 }
 
 
 
-void i810DDInitStatePointers(GLcontext *ctx)
+
+void i810EmitDrawingRectangle( i810ContextPtr imesa )
 {
-   ctx->Driver.Enable = i810DDEnable;
-   ctx->Driver.LightModelfv = i810DDLightModelfv;
-   ctx->Driver.AlphaFunc = i810DDAlphaFunc;
-   ctx->Driver.BlendEquation = i810DDBlendEquation;
-   ctx->Driver.BlendFunc = i810DDBlendFunc;
-   ctx->Driver.BlendFuncSeparate = i810DDBlendFuncSeparate;
-   ctx->Driver.DepthFunc = i810DDDepthFunc;
-   ctx->Driver.DepthMask = i810DDDepthMask;
-   ctx->Driver.Fogfv = i810DDFogfv;
-   ctx->Driver.Scissor = i810DDScissor;
-   ctx->Driver.CullFace = i810DDCullFaceFrontFace;
-   ctx->Driver.FrontFace = i810DDCullFaceFrontFace;
-   ctx->Driver.ColorMask = i810DDColorMask;
-   ctx->Driver.ReducedPrimitiveChange = i810DDReducedPrimitiveChange;
-   ctx->Driver.RenderStart = i810DDUpdateHwState; 
-   ctx->Driver.RenderFinish = 0; 
+   __DRIdrawablePrivate *dPriv = imesa->driDrawable;
+   i810ScreenPrivate *i810Screen = imesa->i810Screen;
+
+   int x0 = imesa->drawX;
+   int y0 = imesa->drawY;
+   int x1 = x0 + dPriv->w;
+   int y1 = y0 + dPriv->h;
+
+
+   /* Coordinate origin of the window - may be offscreen.
+    */
+   imesa->BufferSetup[I810_DESTREG_DR4] = ((y0<<16) | 
+					   (((unsigned)x0)&0xFFFF));
+  
+   /* Clip to screen.
+    */
+   if (x0 < 0) x0 = 0;
+   if (y0 < 0) y0 = 0;
+   if (x1 > i810Screen->width-1) x1 = i810Screen->width-1;
+   if (y1 > i810Screen->height-1) y1 = i810Screen->height-1;
+
+
+   /* Onscreen drawing rectangle.
+    */
+   imesa->BufferSetup[I810_DESTREG_DR2] = ((y0<<16) | x0);
+   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",	   
+	   msg,
+	   (unsigned int) state,
+	   (state & I810_REFRESH_RING)      ? "read-lp-ring, " : "",
+	   (state & I810_WAIT_AGE)          ? "wait-age, " : "",
+	   (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, " : ""
+	   );
 }
 
 
+/* Spew the state onto the ring/batch buffer and/or texture memory.
+ */
+void i810EmitHwStateLocked( i810ContextPtr imesa )
+{
+   if (I810_DEBUG & DEBUG_VERBOSE_API)
+      i810DDPrintDirty( "\n\n\ni810EmitHwStateLocked", imesa->dirty );
+
+   if (imesa->dirty & ~(I810_WAIT_AGE|I810_EMIT_CLIPRECT))
+   {
+      i810GetGeneralDmaBufferLocked( imesa );
+
+      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_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->CurrentTexObj[1]) 
+	    i810DmaExecute( imesa, imesa->CurrentTexObj[1]->Setup,
+			    I810_TEX_SETUP_SIZE, "tex-1");
+      }
+      
+      if (imesa->dirty & I810_UPLOAD_BUFFERS) 
+	 i810DmaExecute( imesa, imesa->BufferSetup, 
+			 I810_DEST_SETUP_SIZE, "buffers" );
+      
+      i810FlushGeneralLocked( imesa );
+   }
+
+   /* Clear everything but the cliprect dirty flag:
+    */
+   imesa->dirty &= (I810_EMIT_CLIPRECT|I810_WAIT_AGE);
+}
+
+
+
 void i810DDInitState( i810ContextPtr imesa )
 {
+   i810ScreenPrivate *i810Screen = imesa->i810Screen;
+
+   memset(imesa->Setup, 0, sizeof(imesa->Setup));
 
    imesa->Setup[I810_CTXREG_VF] = (GFX_OP_VERTEX_FMT |
 				   VF_TEXCOORD_COUNT_2 |
@@ -562,6 +962,8 @@ void i810DDInitState( i810ContextPtr imesa )
 
 
    imesa->Setup[I810_CTXREG_SDM] = ( GFX_OP_SRC_DEST_MONO |
+				     SDM_UPDATE_MONO_ENABLE |
+				     0 |
 				     SDM_UPDATE_SRC_BLEND | 
 				     SDM_SRC_ONE |
 				     SDM_UPDATE_DST_BLEND |
@@ -627,7 +1029,7 @@ 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 |
@@ -641,14 +1043,6 @@ void i810DDInitState( i810ContextPtr imesa )
 
    imesa->LcsCullMode = LCS_CULL_CW;
    
-   imesa->Setup[I810_CTXREG_SCI0] = GFX_OP_SCISSOR_INFO;
-   imesa->Setup[I810_CTXREG_SCI1] = 0;
-   imesa->Setup[I810_CTXREG_SCI2] = 0;
-
-   imesa->Setup[I810_CTXREG_SC] = ( GFX_OP_SCISSOR | 
-				    SC_UPDATE_SCISSOR |
-				    0 );
-
    imesa->Setup[I810_CTXREG_PV] = ( GFX_OP_PV_RULE |
 				    PV_UPDATE_PIXRULE |
 				    PV_PIXRULE_ENABLE |
@@ -659,8 +1053,59 @@ void i810DDInitState( i810ContextPtr imesa )
 				    PV_UPDATE_TRISTRIP |
 				    PV_TRISTRIP_PV0 );
 
-}
 
+   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 |
+				    AA_ENABLE_EDGEFLAG | /* ? */
+				    AA_UPDATE_POLYWIDTH |
+				    AA_POLYWIDTH_05 |
+				    AA_UPDATE_LINEWIDTH |
+				    AA_LINEWIDTH_05 |
+				    AA_UPDATE_BB_EXPANSION |
+				    0 |
+				    AA_UPDATE_AA_ENABLE |
+				    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->BufferSetup[I810_DESTREG_DI0] = CMD_OP_DESTBUFFER_INFO;
+
+   if (imesa->glCtx->Color.DriverDrawBuffer == GL_BACK_LEFT) {
+      imesa->drawMap = i810Screen->back.map;
+      imesa->BufferSetup[I810_DESTREG_DI1] = (i810Screen->backOffset | 
+					      i810Screen->backPitchBits);
+   } else {
+      imesa->drawMap = imesa->driScreen->pFB;
+      imesa->BufferSetup[I810_DESTREG_DI1] = (i810Screen->fbOffset | 
+					      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->backPitchBits);
+
+   imesa->BufferSetup[I810_DESTREG_DR0] = GFX_OP_DRAWRECT_INFO;
+   imesa->BufferSetup[I810_DESTREG_DR1] = DR1_RECT_CLIP_ENABLE;
+}
 
 
 #define INTERESTED (~(NEW_MODELVIEW|NEW_PROJECTION|\
@@ -672,16 +1117,66 @@ 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) {
-      i810FinishPrimitive();
       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;
+   }
+}
+
+
+void i810DDInitStateFuncs(GLcontext *ctx)
+{
+   ctx->Driver.UpdateState = i810DDUpdateState;
+   ctx->Driver.Enable = i810DDEnable;
+   ctx->Driver.LightModelfv = i810DDLightModelfv;
+   ctx->Driver.AlphaFunc = i810DDAlphaFunc;
+   ctx->Driver.BlendEquation = i810DDBlendEquation;
+   ctx->Driver.BlendFunc = i810DDBlendFunc;
+   ctx->Driver.BlendFuncSeparate = i810DDBlendFuncSeparate;
+   ctx->Driver.DepthFunc = i810DDDepthFunc;
+   ctx->Driver.DepthMask = i810DDDepthMask;
+   ctx->Driver.Fogfv = i810DDFogfv;
+   ctx->Driver.Scissor = i810DDScissor;
+   ctx->Driver.CullFace = i810DDCullFaceFrontFace;
+   ctx->Driver.FrontFace = i810DDCullFaceFrontFace;
+   ctx->Driver.ColorMask = i810DDColorMask;
+   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.Color = i810DDSetColor;
+   ctx->Driver.ClearColor = i810DDClearColor;
+   ctx->Driver.Dither = i810DDDither;
+
+   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 1cccb44bb..d0551631e 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810state.h
+++ b/xc/lib/GL/mesa/src/drv/i810/i810state.h
@@ -2,10 +2,12 @@
 #define _I810_STATE_H
 
 
-extern void i810DDInitStatePointers(GLcontext *ctx);
 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
index 3c78921e3..a0f6d8938 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810swap.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810swap.c
@@ -1,66 +1,71 @@
 
+
+#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"
+#include "i810ioctl.h"
+
 /*
  * i810BackToFront
  *
- * Blit the visible rectangles from the back buffer to the screen
- * Can't really do this from a direct context - don't have the clip
- * info, and getting it would be a lot slower than just sending a
- * request to the server to do the blit there.
- *
- * drawable -- front buffer 
- * buf      -- back buffer
+ * 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.
  */
-int i810BackToFront(DrawablePtr drawable, 
-		    struct i810_dest_buffer *buf) 
+static int i810BackToFront( i810ContextPtr imesa ) 
 {
-   RegionPtr		prgnClip;
-   BoxPtr		pbox;
-   int 			i,nbox;
-   int 			ofs, xorg, yorg, pitch;
-
-   if ( !xf86VTSema ) {
-      i810Error("BackToFront(): !xf86VTSema\n");
-      return BadMatch;	/* Is this really an error? */
-   }
-
-   if ( drawable->width != buf->Width || 
-	drawable->height != buf->Height || 
-	drawable->type != DRAWABLE_WINDOW ) {
-      i810Error("BackToFront(): bad drawable\n");
-      return BadMatch;
-   }
+   __DRIdrawablePrivate *dPriv = imesa->driDrawable;
+   i810ScreenPrivate *i810Screen = imesa->i810Screen;
 
-   prgnClip = &((WindowPtr)drawable)->clipList;
-   pbox = REGION_RECTS(prgnClip);
-   nbox = REGION_NUM_RECTS(prgnClip);
-   if( !nbox ) return Success;
 
-   xorg = drawable->x;
-   yorg = drawable->y;
-   pitch = buf->Pitch;
-   ofs = buf->MemBlock->ofs; 
+   /* Use the frontbuffer cliprects:
+    */
+   XF86DRIClipRectPtr pbox = dPriv->pClipRects;
+   int nbox = dPriv->numClipRects;
 
+   if( nbox ) 
    {
-      unsigned int BR13 = ((vga256InfoRec.displayWidth * vgaBytesPerPixel) |
+      int i;
+      int pitch = i810Screen->backPitch;
+
+      int dx = dPriv->backX - dPriv->x;
+      int dy = dPriv->backY - dPriv->y;
+      int ofs = ( i810Screen->backOffset + 
+		  dy * i810Screen->backPitch +
+		  dx * i810Screen->cpp);
+
+      unsigned int BR13 = ((i810Screen->fbStride) |
 			   (0xCC << 16));
 
 
       for (i=0; i < nbox; i++, pbox++) {
-	 int x = pbox->x1 - xorg;
-	 int y = pbox->y1 - yorg;
 	 int w = pbox->x2 - pbox->x1;
 	 int h = pbox->y2 - pbox->y1;
 
-	 int start = ofs + x*vgaBytesPerPixel + y*pitch;
-	 int dst = ((pbox->x1 + pbox->y1 * vga256InfoRec.displayWidth) * 
-		    vgaBytesPerPixel);
+	 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( 6 );
+	    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 * vgaBytesPerPixel));
+	    OUT_BATCH( (h << 16) | (w * i810Screen->cpp));
 	    OUT_BATCH( dst );
 
 	    OUT_BATCH( pitch );	/* src pitch */
@@ -76,33 +81,70 @@ int i810BackToFront(DrawablePtr drawable,
 
 /*
  * ClearBox
- * Add hardware commands to draw a filled box for the
- * debugging display.
+ *
+ * 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( int x, int y, int w, int h, int r, int g, int b ) 
+static void ClearBox( i810ContextPtr imesa,
+		      int cx, int cy, int cw, int ch, 
+		      int r, int g, int b ) 
 {
-   int start = (i810DB->MemBlock->ofs + 
-		y * i810DB->Pitch + 
-		x * vgaBytesPerPixel);
-
-   BEGIN_BATCH(6);
-   
-   OUT_BATCH( BR00_BITBLT_CLIENT | BR00_OP_COLOR_BLT | 0x3 );
-   OUT_BATCH( BR13_SOLID_PATTERN | 
-	      (0xF0 << 16) |
-	      i810DB->Pitch );
-   OUT_BATCH( (h << 16) | (w * vgaBytesPerPixel));
-   OUT_BATCH( start );
-   OUT_BATCH( i810PackColor( i810DB->Format, r, g, b, 0 ) );
-   OUT_BATCH( 0 );
+   __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;
+
+      {
+	 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;
+      }
 
-   if (0) 
-      fprintf(stderr, "box %d,%x %dx%d col %x (%d,%d,%d)\n", x,y,w,h,
-	      i810PackColor( i810DB->Format, r, g, b, 0 ), r, g, b);
 
-   ADVANCE_BATCH();
+      {
+	 int start = (0 + 
+		      y1 * i810Screen->backPitch + 
+		      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->backPitch );
+	 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();
+      }
+   }
 }
 
 
@@ -111,143 +153,110 @@ static void ClearBox( int x, int y, int w, int h, int r, int g, int b )
  * Draw some small boxesin the corner of the buffer
  * based on some performance information
  */
-void i810PerformanceBoxes( int is_direct ) {
-   int		w, t;
+static void i810PerformanceBoxes( i810ContextPtr imesa ) {
 	
-   if ( !i810glx.boxes ) 
-      return;
-
-
-   /* draw a box to show we are active, so if it is't seen
-      it means that it is completely software based rendering  */
-   /* draw a purple box if we are direct rendering */
-   if ( is_direct ) {			/* purple = direct (client dma) rendering */
-      ClearBox( 4, 4, 8, 8, 255, 0, 255 );
-   } else if ( i810glx.dmaDriver ) {	/* white = server dma rendering */
-      ClearBox( 4, 4, 8, 8, 255, 255, 255 );
-   } else {				/* grey = servery PDMA */
-      ClearBox( 4, 4, 8, 8, 128, 128, 128 );
-   }
+
+   /* 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( 16, 4, 8, 8, 255, 0, 0 );
+      ClearBox( imesa, 16, 4, 8, 8, 255, 0, 0 );
       i810glx.c_drawWaits = 0;
    }
 	
-   /* draw a blue box if the register protection signal was hit */
-   if ( i810glx.c_signals ) {
-      ClearBox( 28, 4, 8, 8, 0, 0, 255 );
-      i810glx.c_signals = 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( 40, 4, 8, 8, 255, 255, 0 );
+      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( 64, 4, 8, 8, 0, 255, 0 );
+      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( 4, 16, 252, 4, 255, 32, 32 );
+      ClearBox( imesa, 4, 16, 252, 4, 255, 32, 32 );
    } else {
-      /* draw bars to represent the utilization of primary and secondary buffers */
-      ClearBox( 4, 16, 252, 4, 32, 32, 32 );
+      /* 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( 4, 16, w, 4, 196, 128, 128 );
+      ClearBox( imesa, 4, 16, w, 4, 196, 128, 128 );
    }
+#endif
+
    i810glx.c_dmaFlush = 0;
 }
 
-static void i810SendDmaFlush( void )
+
+static void i810EmitFlush( i810ContextPtr imesa )
 {
-   BEGIN_BATCH(2);
+   BEGIN_BATCH( imesa, 2 );
    OUT_BATCH( INST_PARSER_CLIENT | INST_OP_FLUSH );
    OUT_BATCH( 0 );
    ADVANCE_BATCH();
 }
 
-
-/*
- * Copy the back buffer to the front buffer.  If there's no back buffer
- * this is a no-op.  Only called in indirect contexts.
- */
-static void i810ServerSwapBuffers( XSMesaBuffer b )
+void i810SwapBuffers( i810ContextPtr imesa )
 {
-   /* make sure mesa gives us everything */
-   if (i810Ctx && i810Ctx->gl_ctx)
-      glFlush();
+   int tmp;
 
-   if ( !b->db_state ) {
-      /* not double buffered, so do nothing */
-   } else {
-      ValidateGC(b->frontbuffer, b->cleargc);
-      if ( b->backimage ) {
-
-	 struct i810_dest_buffer *buf = 
-	    (struct i810_dest_buffer *)b->backimage->devPriv;
+   if (I810_DEBUG & DEBUG_VERBOSE_API)
+      fprintf(stderr, "i810SwapBuffers()\n");
 
-	 if ( buf && buf->Drawable ) {
+   FLUSH_BATCH( imesa );
+   LOCK_HARDWARE( imesa );
+   
+   i810GetGeneralDmaBufferLocked( imesa );
+   i810EmitFlush( imesa );
 
-	    /* flush the last primitive */
-	    i810FinishPrimitive();
+   if ( i810glx.boxes )   
+      i810PerformanceBoxes( imesa );
 
-	    /* Drop a flush op into the dma stream */
-	    i810SendDmaFlush();
+   i810BackToFront( imesa );
+   i810EmitFlush( imesa );
+   i810FlushGeneralLocked( imesa );
 
-	    /* diagnostic drawing tools */
-	    i810PerformanceBoxes( 0 );
+   /* Throttle app if the last swap hasn't dispatched yet
+    */
+   if (0) 
+      fprintf(stderr, "dispatch age: %d last swap: %d\n", 
+	      GET_DISPATCH_AGE(imesa), imesa->lastSwap);
+   
+   tmp = GET_ENQUEUE_AGE(imesa);
 
-	    /* hardware accelerated back to front blit */	
-	    i810BackToFront( (DrawablePtr)b->frontbuffer,buf );
+   UNLOCK_HARDWARE( imesa );
 
-	    /* make sure all dma is going to get executed if
-	     * everything has gone well, this will be the only flush
-	     * each frame 
-	     */
-	    i810DmaFlush();
+   if (GET_DISPATCH_AGE(imesa) < imesa->lastSwap)
+      i810WaitAge(imesa, imesa->lastSwap);
 
-	 } else {
-	    /* Use backimage's dimension (not buffer's) */
-	    (*b->cleargc->ops->PutImage)((DrawablePtr)b->frontbuffer,
-					 b->cleargc,
-					 b->frontbuffer->depth,
-					 0, 0,
-					 b->backimage->width, 
-					 b->backimage->height,
-					 0, ZPixmap, b->backimage->data);
-	 }
-      } else {
-	 /* Copy pixmap to window on server */
-	 (*b->cleargc->ops->CopyArea)((DrawablePtr)b->backpixmap,
-				      b->frontbuffer,
-				      b->cleargc,
-				      0, 0, b->width, b->height,
-				      0, 0);
-      }
-   }
-  
-   /* report performance counters */
-   i810Msg(10, "swapBuffers: c_triangles:%i  c_lines:%i c_points:%i c_setup:%i c_textures:%i\n", 
-	   i810glx.c_triangles, i810glx.c_lines, i810glx.c_points, i810glx.c_setupPointers,
-	   i810glx.c_textureSwaps );
-   i810glx.c_triangles = 0;
-   i810glx.c_lines = 0;
-   i810glx.c_points = 0;
-   i810glx.c_setupPointers = 0;
-   i810Msg(10, "---------------------------------------------------------\n" );
+   imesa->secondLastSwap = imesa->lastSwap;
+   imesa->lastSwap = tmp;
 }
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810tex.c b/xc/lib/GL/mesa/src/drv/i810/i810tex.c
index 5b1d00398..35c395296 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810tex.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810tex.c
@@ -22,7 +22,6 @@
  *
  */
 
-#include "xsmesaP.h"
 #include <stdlib.h>
 #include <stdio.h>
 
@@ -32,6 +31,7 @@
 #include "i810lib.h"
 #include "i810tex.h"
 #include "i810log.h"
+#include "i810ioctl.h"
 #include "simple_list.h"
 #include "enums.h"
 
@@ -49,7 +49,6 @@ static void i810SetTexWrapping(i810TextureObjectPtr tex, GLenum s, GLenum t)
       val ^= (MCS_V_WRAP^MCS_V_CLAMP);
 
    tex->Setup[I810_TEXREG_MCS] = val;
-   tex->reg_dirty |= (1 << I810_TEXREG_MCS);
 }
 
 static void i810SetTexFilter(i810TextureObjectPtr t, GLenum minf, GLenum magf)
@@ -136,10 +135,6 @@ static void i810SetTexFilter(i810TextureObjectPtr t, GLenum minf, GLenum magf)
 		     MLC_LOD_BIAS_MASK,
 		     0x0);
    }
-
-   t->reg_dirty |= ((1<<I810_TEXREG_MLC) |
-		    (1<<I810_TEXREG_MLL) |
-		    (1<<I810_TEXREG_MF));
 }
 
 
@@ -161,8 +156,8 @@ static void ReplicateMesaTexState(i810TextureObjectPtr t,
    i810SetTexBorderColor(t,mesatex->BorderColor);
 }
 
-i810TextureObjectPtr i810CreateTexObj(i810ContextPtr ctx,
-				      struct gl_texture_object *tObj)
+static i810TextureObjectPtr i810CreateTexObj(i810ContextPtr imesa,
+					     struct gl_texture_object *tObj)
 {
    i810TextureObjectPtr t;
    GLuint height, width, pitch, i, textureFormat, log_pitch;
@@ -194,7 +189,6 @@ i810TextureObjectPtr i810CreateTexObj(i810ContextPtr ctx,
    case GL_COLOR_INDEX:
       textureFormat = MI1_FMT_8CI | MI1_PF_8CI_ARGB4444;
       t->texelBytes = 1;
-      t->UsePalette = 1;
       break;
    default:
       i810Error( "i810CreateTexObj: bad image->Format\n" );
@@ -211,6 +205,7 @@ i810TextureObjectPtr i810CreateTexObj(i810ContextPtr ctx,
       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];
@@ -226,7 +221,6 @@ i810TextureObjectPtr i810CreateTexObj(i810ContextPtr ctx,
    t->min_level = 0;
    t->globj = tObj;
    t->age = 0;
-   t->reg_dirty = ~0;
 
    t->Setup[I810_TEXREG_MI0] = GFX_OP_MAP_INFO;
 
@@ -279,89 +273,72 @@ i810TextureObjectPtr i810CreateTexObj(i810ContextPtr ctx,
 
    ReplicateMesaTexState(t,tObj);
    tObj->DriverData = t;
-   insert_at_head(&ctx->SwappedOut, t);
+   imesa->dirty |= I810_UPLOAD_CTX;
+   make_empty_list( t );
    return t;
 }
 
-int i810DestroyTexObj(i810ContextPtr ctx, i810TextureObjectPtr t)
+void i810DestroyTexObj(i810ContextPtr imesa, i810TextureObjectPtr t)
 {
-   if (t) {
-      if (t->age > i810glx.dma_buffer_age)
-	 i810WaitDrawingEngine();
+   if (!t) return;
 
+   /* This is sad - need to sync *in case* we upload a texture
+    * to this newly free memory...
+    */
+   if (t->MemBlock) {
       mmFreeMem(t->MemBlock);
-      t->MemBlock = 0;      
-      remove_from_list(t);
-      free(t);
+      t->MemBlock = 0;
+
+      if (t->age > imesa->dirtyAge)
+	 imesa->dirtyAge = t->age;
    }
-   return 0;
+
+   if (t->globj)
+      t->globj->DriverData = 0;
+
+   if (t->bound)
+      imesa->CurrentTexObj[t->bound - 1] = 0; 
+
+   remove_from_list(t);
+   free(t);
 }
 
 
-/* Card memory management
- */
-static void i810SwapOutOldestTexObj( i810ContextPtr ctx )
+static void i810SwapOutTexObj(i810ContextPtr imesa, i810TextureObjectPtr t)
 {
-   i810TextureObjectPtr t = ctx->TexObjList.prev;
-
-   fprintf(stderr, "swap out oldest, age %d, dma age %d\n",
-	   t->age, i810glx.dma_buffer_age);
+   if (t->MemBlock) {
+      mmFreeMem(t->MemBlock);
+      t->MemBlock = 0;      
 
-   if (t->age > i810glx.dma_buffer_age)
-      i810WaitDrawingEngine();	
+      if (t->age > imesa->dirtyAge)
+	 imesa->dirtyAge = t->age;
+   }
 
-   mmFreeMem( t->MemBlock );
-   t->MemBlock = 0;      
    t->dirty_images = ~0;
-   move_to_tail(&ctx->SwappedOut, t);
+   move_to_tail(&(imesa->SwappedOut), t);
 }
 
-/* Just allocates the memory - doesn't do the upload.
- */
-static int i810SwapInTexObj( i810ContextPtr ctx, i810TextureObjectPtr t)
-{
-   i810Msg(10,"  Swapping in texture.\n");
-   i810glx.c_textureSwaps++;
-
-   /* Try to allocate textureHeap memory, swapping out stuff until we
-    * can.
-    */
-   while (1)
-   {
-      t->MemBlock = mmAllocMem( i810glx.sysmemHeap, t->totalSize, 12, 0 );
-
-      if (t->MemBlock)
-	 break;
 
-      if (is_empty_list(&ctx->TexObjList))
-	 return -1;
-
-      i810SwapOutOldestTexObj( ctx );
-   }
- 
-   t->Setup[I810_TEXREG_MI3] = t->MemBlock->ofs;
-   t->BufAddr = i810glx.sysmemVirtual + t->MemBlock->ofs;
-   t->reg_dirty |= ((1<<I810_TEXREG_MI0) |
-		    (1<<I810_TEXREG_MI1) |
-		    (1<<I810_TEXREG_MI2) |
-		    (1<<I810_TEXREG_MI3));
-   return 0;
-}
 
 /* Upload an image from mesa's internal copy.
  */
-static int i810UploadTexLevel( i810TextureObjectPtr t, int level )
+static void i810UploadTexLevel( i810TextureObjectPtr t, int level )
 {
    const struct gl_texture_image *image = t->image[level].image;
    int i,j;
 
-   i810Msg(10,"CopyImage():\n");
+   if (I810_DEBUG & DEBUG_VERBOSE_LRU)
+      fprintf(stderr, "i810UploadTexLevel %d, BufAddr %p offset %x\n",
+	      level, t->BufAddr, t->image[level].offset);
 
    /* Need triangle (rather than pixel) fallbacks to simulate this using
     * normal textured triangles.
-    */
+    *
+    * DO THIS IN DRIVER STATE MANAGMENT, not hardware state.
+    *
    if (image->Border != 0) 
-      i810Error("Not supported texture border %d.\n",image->Border);
+      i810Error("Not supported texture border %d.\n", (int) image->Border);
+    */
 
    switch (t->image[level].internalFormat) {
    case GL_RGB:
@@ -440,6 +417,8 @@ static int i810UploadTexLevel( i810TextureObjectPtr t, int level )
    }
    break;
 
+   /* TODO: Translate color indices *now*:
+    */
    case GL_COLOR_INDEX:
       {
 	 GLubyte *dst = (GLubyte *)(t->BufAddr + t->image[level].offset);
@@ -452,38 +431,221 @@ static int i810UploadTexLevel( i810TextureObjectPtr t, int level )
 	       src += 1;
 	    }
 	 }
-
-	 t->UsePalette = I810_USE_PALETTE;
       }
    break;
       
    default:
-      i810Error("Not supported texture format %d\n",image->Format);
-      FatalError("bummer");
-      return -1;
+      i810Error("Not supported texture format %d\n",(int)image->Format);
+      exit(1);
    }
+}
 
-   return 0;
+
+
+void i810PrintLocalLRU( i810ContextPtr imesa ) 
+{
+   i810TextureObjectPtr t;
+   int sz = 1 << (imesa->i810Screen->logTextureGranularity);
+
+   foreach( t, &imesa->TexObjList ) {
+      if (!t->globj)
+	 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);      
+
+   }
+}
+
+void i810PrintGlobalLRU( i810ContextPtr imesa )
+{
+   int i, j;
+   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",
+	      j, list[j].age, list[j].next, list[j].prev);
+      j = list[j].next;
+      if (j == I810_NR_TEX_REGIONS) break;
+   }
+   
+   if (j != I810_NR_TEX_REGIONS)
+      fprintf(stderr, "Loop detected in global LRU\n");
+}
+
+
+void i810ResetGlobalLRU( i810ContextPtr imesa )
+{
+   drm_i810_tex_region_t *list = imesa->sarea->texList;
+   int sz = 1 << imesa->i810Screen->logTextureGranularity;
+   int i;
+
+   /* (Re)initialize the global circular LRU list.  The last element
+    * in the array (I810_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 <= imesa->i810Screen->textureSize ; i++) {
+      list[i].prev = i-1;
+      list[i].next = i+1;
+      list[i].age = 0;
+   }
+
+   i--;
+   list[0].prev = I810_NR_TEX_REGIONS;
+   list[i].prev = i-1;
+   list[i].next = I810_NR_TEX_REGIONS;
+   list[I810_NR_TEX_REGIONS].prev = i;
+   list[I810_NR_TEX_REGIONS].next = 0;
+   imesa->sarea->texAge = 0;
 }
 
 
+static void i810UpdateTexLRU( i810ContextPtr imesa, i810TextureObjectPtr t ) 
+{
+   int i;
+   int logsz = imesa->i810Screen->logTextureGranularity;
+   int start = t->MemBlock->ofs >> logsz;
+   int end = (t->MemBlock->ofs + t->MemBlock->size - 1) >> logsz;
+   drm_i810_tex_region_t *list = imesa->sarea->texList;
+   
+   imesa->texAge = ++imesa->sarea->texAge;
+
+   /* Update our local LRU
+    */
+   move_to_head( &(imesa->TexObjList), t );
+
+   /* Update the global LRU
+    */
+   for (i = start ; i <= end ; i++) {
+
+      list[i].in_use = 1;
+      list[i].age = imesa->texAge;
+
+      /* remove_from_list(i)
+       */
+      list[(unsigned)list[i].next].prev = list[i].prev;
+      list[(unsigned)list[i].prev].next = list[i].next;
+      
+      /* insert_at_head(list, i)
+       */
+      list[i].prev = I810_NR_TEX_REGIONS;
+      list[i].next = list[I810_NR_TEX_REGIONS].next;
+      list[(unsigned)list[I810_NR_TEX_REGIONS].next].prev = i;
+      list[I810_NR_TEX_REGIONS].next = i;
+   }
+}
+
 
-static int i810UploadTexImages( i810ContextPtr ctx, i810TextureObjectPtr t )
+/* Called for every shared texture region which has increased in age
+ * since we last held the lock.
+ *
+ * Figures out which of our textures have been ejected by other clients,
+ * and pushes a placeholder texture onto the LRU list to represent 
+ * the other client's textures.  
+ */
+void i810TexturesGone( i810ContextPtr imesa,
+		       GLuint offset, 
+		       GLuint size,
+		       GLuint in_use ) 
+{
+   i810TextureObjectPtr t, tmp;
+   
+   foreach_s ( t, tmp, &imesa->TexObjList ) {
+
+      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->bound)
+	 i810SwapOutTexObj( imesa, t );
+      else
+	 i810DestroyTexObj( imesa, t );
+   }
+
+   
+   if (in_use) {
+      t = (i810TextureObjectPtr) calloc(1,sizeof(*t));
+      if (!t) return;
+
+      t->MemBlock = mmAllocMem( imesa->texHeap, size, 0, offset);      
+      insert_at_head( &imesa->TexObjList, t );
+   }
+}
+
+
+
+
+
+/* This is called with the lock held.  May have to eject our own and/or
+ * other client's texture objects to make room for the upload.
+ */
+int i810UploadTexImages( i810ContextPtr imesa, i810TextureObjectPtr t )
 {
    int i;
+   int ofs;
    i810glx.c_textureSwaps++;
 
-   if (!t->MemBlock) 
-      i810SwapInTexObj( ctx, t ); 
+   /* Do we need to eject LRU texture objects?
+    */
+   if (!t->MemBlock) {
+      while (1)
+      {
+	 t->MemBlock = mmAllocMem( imesa->texHeap, t->totalSize, 12, 0 ); 
+	 if (t->MemBlock)
+	    break;
 
-   if (t->age > i810glx.dma_buffer_age)
-      i810WaitDrawingEngine();
+	 if (imesa->TexObjList.prev->bound) {
+  	    fprintf(stderr, "Hit bound texture in upload\n"); 
+	    i810PrintLocalLRU( imesa );
+	    return -1;
+	 }
 
-   for (i = t->min_level ; i <= t->max_level ; i++)
-      if (t->dirty_images & (1<<i)) {
-	 if (i810UploadTexLevel( t, i ) != 0)
+	 if (imesa->TexObjList.prev == &(imesa->TexObjList)) {
+ 	    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;
+      imesa->dirty |= I810_UPLOAD_CTX;
+   }
+
+   /* Let the world know we've used this memory recently.
+    */
+   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, "*");
+
+      for (i = t->min_level ; i <= t->max_level ; i++)
+	 if (t->dirty_images & (1<<i)) 
+	    i810UploadTexLevel( t, i );
+   }
+
 
    t->dirty_images = 0;
    return 0;
@@ -499,7 +661,6 @@ static void i810TexSetUnit( i810TextureObjectPtr t, GLuint unit )
    t->Setup[I810_TEXREG_MCS] ^= (MCS_COORD_0 ^ MCS_COORD_1);
    t->Setup[I810_TEXREG_MF]  ^= (MF_MAP_0 ^ MF_MAP_1);
 
-   t->reg_dirty = ~0;
    t->current_unit = unit;
 }
 
@@ -519,30 +680,29 @@ static void i810UpdateTex0State( GLcontext *ctx )
    if ( tObj != ctx->Texture.Unit[0].CurrentD[2] ) 
       tObj = 0;
 
-   if (!(ctx->Texture.Enabled & 0xf) || !tObj || !tObj->Complete) {
-
-      imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
-					MC_STAGE_0 |
-					MC_UPDATE_DEST |
-					MC_DEST_CURRENT |
-					MC_UPDATE_ARG1 |
-					MC_ARG1_ITERATED_COLOR | 
-					MC_UPDATE_ARG2 |
-					MC_ARG2_ONE |
-					MC_UPDATE_OP |
-					MC_OP_ARG1 );
 
-      imesa->Setup[I810_CTXREG_MA0] = ( GFX_OP_MAP_ALPHA_STAGES |
-					MA_STAGE_0 |
-					MA_UPDATE_ARG1 |
-					MA_ARG1_ITERATED_ALPHA |
-					MA_UPDATE_ARG2 |
-					MA_ARG2_TEX0_ALPHA |
-					MA_UPDATE_OP |
-					MA_OP_ARG1 );
+   imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
+				     MC_STAGE_0 |
+				     MC_UPDATE_DEST |
+				     MC_DEST_CURRENT |
+				     MC_UPDATE_ARG1 |
+				     MC_ARG1_ITERATED_COLOR | 
+				     MC_UPDATE_ARG2 |
+				     MC_ARG2_ONE |
+				     MC_UPDATE_OP |
+				     MC_OP_ARG1 );
+
+   imesa->Setup[I810_CTXREG_MA0] = ( GFX_OP_MAP_ALPHA_STAGES |
+				     MA_STAGE_0 |
+				     MA_UPDATE_ARG1 |
+				     MA_ARG1_ITERATED_ALPHA |
+				     MA_UPDATE_ARG2 |
+				     MA_ARG2_TEX0_ALPHA |
+				     MA_UPDATE_OP |
+				     MA_OP_ARG1 );
     
-      imesa->reg_dirty |= ((1<<I810_CTXREG_MA0) |
-			   (1<<I810_CTXREG_MC0));
+
+   if (!(ctx->Texture.Enabled & 0xf) || !tObj || !tObj->Complete) {
       return;
    }
 
@@ -557,30 +717,13 @@ static void i810UpdateTex0State( GLcontext *ctx )
       i810TexSetUnit( t, 0 );
     
    if (t->dirty_images) 
-      i810UploadTexImages(imesa, t);
-
-   t->age = ++i810glx.current_texture_age;
-   move_to_head(&imesa->TexObjList, t);
-
-   imesa->CurrentTex0Obj = t;
-
-   if (0)
-      if (t->UsePalette == I810_USE_PALETTE) {      
-	 if (ctx->Texture.SharedPalette) {
-	    if (imesa->GlobalPaletteUpdated)
-	       i810LoadTexturePalette(imesa->GlobalPalette,0,256);
-	    imesa->GlobalPaletteUpdated = 0;
-	 }
-	 else {
-	    i810LoadTexturePalette(t->Palette,0,256);
-	    imesa->GlobalPaletteUpdated = 1;
-	 }
-      }		      
-
-
-   if (t->reg_dirty) 
-      i810DmaExecute(t->Setup,I810_TEX_SETUP_SIZE);
+      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:
@@ -730,15 +873,10 @@ static void i810UpdateTex0State( GLcontext *ctx )
       break;
 
    default:
-      FatalError("unkown tex env mode");
+      fprintf(stderr, "unknown tex env mode");
+      exit(1);
       break;			
    }
-
-
-
-   imesa->reg_dirty |= ((1<<I810_CTXREG_MA0) |
-			(1<<I810_CTXREG_MC0));
-
 }
 
 
@@ -756,36 +894,34 @@ static void i810UpdateTex1State( GLcontext *ctx )
    if ( tObj != ctx->Texture.Unit[1].CurrentD[2] ) 
       tObj = 0;
 
-   if (!(ctx->Texture.Enabled & 0xf0) || !tObj || !tObj->Complete) {
 
-      imesa->Setup[I810_CTXREG_MC1] = ( GFX_OP_MAP_COLOR_STAGES |
-					MC_STAGE_1 |
-					MC_UPDATE_DEST |
-					MC_DEST_CURRENT |
-					MC_UPDATE_ARG1 |
-					MC_ARG1_ONE | 
-					MC_ARG1_DONT_REPLICATE_ALPHA |
-					MC_ARG1_DONT_INVERT |
-					MC_UPDATE_ARG2 |
-					MC_ARG2_ONE |
-					MC_ARG2_DONT_REPLICATE_ALPHA |
-					MC_ARG2_DONT_INVERT |
-					MC_UPDATE_OP |
-					MC_OP_DISABLE );
+   imesa->Setup[I810_CTXREG_MC1] = ( GFX_OP_MAP_COLOR_STAGES |
+				     MC_STAGE_1 |
+				     MC_UPDATE_DEST |
+				     MC_DEST_CURRENT |
+				     MC_UPDATE_ARG1 |
+				     MC_ARG1_ONE | 
+				     MC_ARG1_DONT_REPLICATE_ALPHA |
+				     MC_ARG1_DONT_INVERT |
+				     MC_UPDATE_ARG2 |
+				     MC_ARG2_ONE |
+				     MC_ARG2_DONT_REPLICATE_ALPHA |
+				     MC_ARG2_DONT_INVERT |
+				     MC_UPDATE_OP |
+				     MC_OP_DISABLE );
+
+   imesa->Setup[I810_CTXREG_MA1] = ( GFX_OP_MAP_ALPHA_STAGES |
+				     MA_STAGE_1 |
+				     MA_UPDATE_ARG1 |
+				     MA_ARG1_CURRENT_ALPHA |
+				     MA_ARG1_DONT_INVERT |
+				     MA_UPDATE_ARG2 |
+				     MA_ARG2_CURRENT_ALPHA |
+				     MA_ARG2_DONT_INVERT |
+				     MA_UPDATE_OP |
+				     MA_OP_ARG1 );
 
-      imesa->Setup[I810_CTXREG_MA1] = ( GFX_OP_MAP_ALPHA_STAGES |
-					MA_STAGE_1 |
-					MA_UPDATE_ARG1 |
-					MA_ARG1_CURRENT_ALPHA |
-					MA_ARG1_DONT_INVERT |
-					MA_UPDATE_ARG2 |
-					MA_ARG2_CURRENT_ALPHA |
-					MA_ARG2_DONT_INVERT |
-					MA_UPDATE_OP |
-					MA_OP_ARG1 );
-
-      imesa->reg_dirty |= ((1<<I810_CTXREG_MA1) |
-			   (1<<I810_CTXREG_MC1));
+   if (!(ctx->Texture.Enabled & 0xf0) || !tObj || !tObj->Complete) {
       return;
    }
 
@@ -796,35 +932,18 @@ static void i810UpdateTex1State( GLcontext *ctx )
       if (!t) return;
    }
     
-   if (t->dirty_images) 
-      i810UploadTexImages(imesa, t);
-
    if (t->current_unit != 1)
       i810TexSetUnit( t, 1 );
 
-   t->age = ++i810glx.current_texture_age;
-   move_to_head(&imesa->TexObjList, t);
-   imesa->CurrentTex1Obj = t;
-
-   if (0)
-      if (t->UsePalette == I810_USE_PALETTE) {      
-	 if (ctx->Texture.SharedPalette) {
-	    if (imesa->GlobalPaletteUpdated)
-	       i810LoadTexturePalette(imesa->GlobalPalette,0,256);
-	    imesa->GlobalPaletteUpdated = 0;
-	 }
-	 else {
-	    i810LoadTexturePalette(t->Palette,0,256);
-	    imesa->GlobalPaletteUpdated = 1;
-	 }
-      }		      
-
+   if (t->dirty_images) 
+      imesa->dirty |= I810_UPLOAD_TEX1IMAGE;
 
-   if (t->reg_dirty) 
-      i810DmaExecute(t->Setup,I810_TEX_SETUP_SIZE);
+   imesa->CurrentTexObj[1] = t;
+   t->bound = 2;
 
+   if (t->MemBlock)
+      i810UpdateTexLRU( imesa, t );
 
-   
    switch (ctx->Texture.Unit[1].EnvMode) {
    case GL_REPLACE:
       imesa->Setup[I810_CTXREG_MC1] = ( GFX_OP_MAP_COLOR_STAGES |
@@ -980,64 +1099,70 @@ static void i810UpdateTex1State( GLcontext *ctx )
       break;
 
    default:
-      FatalError("unkown tex 1 env mode");
+      fprintf(stderr, "unkown tex 1 env mode\n");
+      exit(1);
       break;			
    }
-
-   imesa->reg_dirty |= ((1<<I810_CTXREG_MA1) |
-			(1<<I810_CTXREG_MC1));
 }
 
 
 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;   
    i810UpdateTex0State( ctx );
    i810UpdateTex1State( ctx );
+   I810_CONTEXT( ctx )->dirty |= I810_UPLOAD_CTX;
 }
 
 
 
 /*****************************************
- * Driver functions
+ * DRIVER functions
  *****************************************/
 
-void i810TexEnv( GLcontext *ctx, GLenum pname, const GLfloat *param )
+static void i810TexEnv( GLcontext *ctx, 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->reg_dirty |= (1<<I810_CTXREG_CF1);
+	 imesa->dirty |= I810_UPLOAD_CTX;
       }
    } 
 }
 
-void i810TexImage( GLcontext *ctx, 
-		   GLenum target,
-		   struct gl_texture_object *tObj, 
-		   GLint level,
-		   GLint internalFormat,
-		   const struct gl_texture_image *image )
+static void i810TexImage( GLcontext *ctx, 
+			  GLenum target,
+			  struct gl_texture_object *tObj, 
+			  GLint level,
+			  GLint internalFormat,
+			  const struct gl_texture_image *image )
 {
    i810ContextPtr imesa = I810_CONTEXT( ctx );
    i810TextureObjectPtr t;
 
-   CHECK_CONTEXT( return; );
-  
    i810Msg(10,"i810TexImage(%d): level %d internalFormat %x\n", 
 	   tObj->Name, level, internalFormat);
 
@@ -1049,6 +1174,7 @@ 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;
@@ -1056,12 +1182,12 @@ void i810TexImage( GLcontext *ctx,
    }
 }
 
-void i810TexSubImage( 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 ) 
+static void i810TexSubImage( 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 ) 
 {
    i810ContextPtr imesa = I810_CONTEXT( ctx );
    i810TextureObjectPtr t;
@@ -1070,13 +1196,12 @@ void i810TexSubImage( GLcontext *ctx, GLenum target,
 	   width, height, image->Width,image->Height,
 	   level);
 
-   CHECK_CONTEXT( return; );
-
    if ( target != GL_TEXTURE_2D ) 
       return;
    
    t = (i810TextureObjectPtr) tObj->DriverData;
    if (t) {
+      if (t->bound) FLUSH_BATCH( imesa );
       i810DestroyTexObj( imesa, t );
       tObj->DriverData = 0;
       imesa->new_state |= I810_NEW_TEXTURE;
@@ -1084,9 +1209,9 @@ void i810TexSubImage( GLcontext *ctx, GLenum target,
    }
 }
 
-void i810TexParameter( GLcontext *ctx, GLenum target,
-		      struct gl_texture_object *tObj,
-		      GLenum pname, const GLfloat *params )
+static void i810TexParameter( GLcontext *ctx, GLenum target,
+			      struct gl_texture_object *tObj,
+			      GLenum pname, const GLfloat *params )
 {
    i810TextureObjectPtr t = (i810TextureObjectPtr) tObj->DriverData;
    i810ContextPtr imesa = I810_CONTEXT( ctx );
@@ -1097,97 +1222,82 @@ 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;
 
    default:
       return;
    }
+
    imesa->new_state |= I810_NEW_TEXTURE;
 }
 
-void i810BindTexture( GLcontext *ctx, GLenum target,
-		     struct gl_texture_object *tObj )
+static void i810BindTexture( GLcontext *ctx, GLenum target,
+			     struct gl_texture_object *tObj )
 {
    i810ContextPtr imesa = I810_CONTEXT( ctx );
+   
+   FLUSH_BATCH(imesa);
 
-   if (target != GL_TEXTURE_2D)
-      return;
+   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;
 }
 
-void i810DeleteTexture( GLcontext *ctx, struct gl_texture_object *tObj )
+static void i810DeleteTexture( GLcontext *ctx, struct gl_texture_object *tObj )
 {
    i810TextureObjectPtr t = (i810TextureObjectPtr)tObj->DriverData;
    i810ContextPtr imesa = I810_CONTEXT( ctx );
 
-   i810DestroyTexObj(imesa,t);
-   tObj->DriverData=0;
-}
-
-void i810UpdateTexturePalette( GLcontext *ctx, struct gl_texture_object *tObj )
-{
-   i810ContextPtr imesa = I810_CONTEXT( ctx );
-   i810TextureObjectPtr t;
-   GLushort *dst;
-   GLubyte *src;
-   int i, size, next;
-   GLenum format;
+   if (t) {
 
+      if (t->bound) {
+	 FLUSH_BATCH(imesa);
+	 imesa->CurrentTexObj[t->bound-1] = 0;
+	 imesa->new_state |= I810_NEW_TEXTURE;
+      }
 
-   return;
+      i810DestroyTexObj(imesa,t);
+      tObj->DriverData=0;
+   }
+}
 
 
-   if (tObj) 
-   {
-      i810Msg(8,"i810UpdateTexturePalette(): size %d\n", tObj->PaletteSize);
-      t = (i810TextureObjectPtr) tObj->DriverData;;
-      if (!t) return;
-      size = tObj->PaletteSize;
-      format = tObj->PaletteFormat;
-      dst = t->Palette;
-      src = (GLubyte *) tObj->Palette;
-   }
-   else
-   {
-      /* shared palette */
-      imesa->GlobalPaletteUpdated = 1;
-      size = ctx->Texture.PaletteSize;
-      format = ctx->Texture.PaletteFormat;
-      dst = imesa->GlobalPalette;
-      src = (GLubyte *) ctx->Texture.Palette;
-    
-      i810Msg(8, "i810UpdateTexturePalette(): size %d intFormat %x format %x\n",
-	      size, ctx->Texture.PaletteIntFormat, ctx->Texture.PaletteFormat);
-   }
+static GLboolean i810IsTextureResident( GLcontext *ctx, 
+					struct gl_texture_object *t )
+{
+   i810TextureObjectPtr mt;
 
-   if (size > 256) {
-      i810Error("i810UpdateTexturePalette(): palette > 256 entries!\n");
-      return;
-   }
+/*     LOCK_HARDWARE; */
+   mt = (i810TextureObjectPtr)t->DriverData;
+/*     UNLOCK_HARDWARE; */
 
-   switch (format)
-   { 
-   case GL_RGB: next = 3; break;
-   case GL_RGBA:next = 4; break;
-   default : 
-      i810Error("i810UpdateTexturePalette(): unsupported palette format %x\n");
-      next = 4;
-   }
+   return mt && mt->MemBlock;
+}
 
-   for (i=0; i < size; i++) {
-      *dst = I810PACKCOLOR565(src[RCOMP],src[GCOMP],src[BCOMP]);
-      dst++;
-      src += next;
-   }  
+void i810DDInitTextureFuncs( GLcontext *ctx )
+{
+   ctx->Driver.TexEnv = i810TexEnv;
+   ctx->Driver.TexImage = i810TexImage;
+   ctx->Driver.TexSubImage = i810TexSubImage;
+   ctx->Driver.BindTexture = i810BindTexture;
+   ctx->Driver.DeleteTexture = i810DeleteTexture;
+   ctx->Driver.TexParameter = i810TexParameter;
+   ctx->Driver.UpdateTexturePalette = 0;
+   ctx->Driver.IsTextureResident = i810IsTextureResident;
 }
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810tex.h b/xc/lib/GL/mesa/src/drv/i810/i810tex.h
index ed9875be8..5261d0b1b 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810tex.h
+++ b/xc/lib/GL/mesa/src/drv/i810/i810tex.h
@@ -26,16 +26,31 @@
 #ifndef I810TEX_INC
 #define I810TEX_INC
 
-#include "mesaglx/types.h"
+#include "types.h"
 #include "mmath.h"
+#include "mm.h"
 
-#include "i810common.h"
-#include "i810buf.h"
+#include "i810context.h"
+#include "i810_3d_reg.h"
 
 #define VALID_I810_TEXTURE_OBJECT(tobj)  (tobj) 
 
 #define I810_TEX_MAXLEVELS 10
 
+
+
+/* For shared texture space managment, these texture objects may also
+ * be used as proxies for regions of texture memory containing other
+ * client's textures.  Such proxy textures (not to be confused with GL
+ * proxy textures) are subject to the same LRU aging we use for our
+ * own private textures, and thus we have a mechanism where we can
+ * fairly decide between kicking out our own textures and those of
+ * other clients.
+ *
+ * Non-local texture objects have a valid MemBlock to describe the
+ * region managed by the other client, and can be identified by
+ * 't->globj == 0' 
+ */
 struct i810_texture_object_t {
    struct i810_texture_object_t *next, *prev;
 
@@ -46,11 +61,11 @@ struct i810_texture_object_t {
    int Height;
    int texelBytes;
    int totalSize;
+   int bound;
 
-   PMemBlock MemBlock;
-   
+   PMemBlock MemBlock;   
    char *BufAddr;
-
+   
    GLuint min_level;
    GLuint max_level;
    GLuint dirty_images;
@@ -64,13 +79,8 @@ struct i810_texture_object_t {
 
    /* Support for multitexture.
     */
-   GLuint current_unit;		
-   
+   GLuint current_unit;   
    GLuint Setup[I810_TEX_SETUP_SIZE];
-   GLuint reg_dirty;
-
-   GLushort Palette[256];
-   int UsePalette;
 };		
 
 #define I810_NO_PALETTE        0x0
@@ -80,38 +90,20 @@ struct i810_texture_object_t {
 
 typedef struct i810_texture_object_t *i810TextureObjectPtr;
 
-i810TextureObjectPtr i810CreateTexObj(i810ContextPtr ctx,
-				    struct gl_texture_object *tObj);
-int i810DestroyTexObj(i810ContextPtr ctx, i810TextureObjectPtr t);
-void i810UpdateTextureState(GLcontext *ctx);
-
-
-/* Driver functions */
-
-void i810TexEnv( GLcontext *ctx, GLenum pname, const GLfloat *param );
-
-void i810TexImage( GLcontext *ctx, GLenum target,
-		  struct gl_texture_object *tObj, GLint level,
-		  GLint internalFormat,
-		  const struct gl_texture_image *image );
+void i810UpdateTextureState( GLcontext *ctx );
+void i810DDInitTextureFuncs( GLcontext *ctx );
 
-void i810TexSubImage( 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 );
+void i810DestroyTexObj( i810ContextPtr imesa, i810TextureObjectPtr t);
+int i810UploadTexImages( i810ContextPtr imesa, i810TextureObjectPtr t );
 
-void i810TexParameter( GLcontext *ctx, GLenum target,
-		      struct gl_texture_object *tObj,
-		      GLenum pname, const GLfloat *params );
+void i810ResetGlobalLRU( i810ContextPtr imesa );
+void i810TexturesGone( i810ContextPtr imesa, 
+		       GLuint start, GLuint end, 
+		       GLuint in_use ); 
 
-void i810BindTexture( GLcontext *ctx, GLenum target,
-		     struct gl_texture_object *tObj );
+void i810PrintLocalLRU( i810ContextPtr imesa );
+void i810PrintGlobalLRU( i810ContextPtr imesa );
 
-void i810DeleteTexture( GLcontext *ctx, struct gl_texture_object *tObj );
 
-void i810UpdateTexturePalette( GLcontext *ctx, struct gl_texture_object *tObj );
-void i810UseGlobalTexturePalette( GLcontext *ctx, GLboolean state );
 
 #endif
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810tris.c b/xc/lib/GL/mesa/src/drv/i810/i810tris.c
index 1c81b9549..d970e3a36 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810tris.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810tris.c
@@ -26,6 +26,7 @@
 #include <stdio.h>
 #include <math.h>
 
+#include "types.h"
 #include "vb.h"
 #include "pipeline.h"
 
@@ -34,28 +35,9 @@
 #include "i810tris.h"
 #include "i810vb.h"
 #include "i810log.h"
-#include "xsmesaP.h"
-
-
-static void i810_null_quad( GLcontext *ctx, GLuint v0,
-			    GLuint v1, GLuint v2, GLuint v3, GLuint pv )
-{
-}     
-
-static void i810_null_triangle( GLcontext *ctx, GLuint v0,
-				GLuint v1, GLuint v2, GLuint pv )
-{
-}     
-
-static void i810_null_line( GLcontext *ctx, GLuint v1, GLuint v2, GLuint pv ) 
-{
-}
-
-static void i810_null_points( GLcontext *ctx, GLuint first, GLuint last ) 
-{
-}
-
 
+/* Used in i810tritmp.h
+ */
 #define I810_COLOR(to, from) {			\
   (to)[0] = (from)[2];				\
   (to)[1] = (from)[1];				\
@@ -70,17 +52,6 @@ static quad_func     quad_tab[0x20];
 static line_func     line_tab[0x20];  
 static points_func   points_tab[0x20];
 
-void i810PrintRenderState( const char *msg, GLuint state )
-{
-   fprintf(stderr, "%s: (%x) %s%s%s%s%s%s\n",
-	   msg, state,
-	   (state & I810_FLAT_BIT)       ? "flat, "       : "",
-	   (state & I810_OFFSET_BIT)     ? "offset, "     : "",
-	   (state & I810_TWOSIDE_BIT)    ? "twoside, "    : "",
-	   (state & I810_ANTIALIAS_BIT)  ? "antialias, "  : "",
-	   (state & I810_NODRAW_BIT)     ? "no-draw, "    : "",
-	   (state & I810_FALLBACK_BIT)   ? "fallback"     : "");
-}
 
 #define IND (0)
 #define TAG(x) x
@@ -90,33 +61,32 @@ void i810PrintRenderState( const char *msg, GLuint state )
 #define TAG(x) x##_flat
 #include "i810tritmp.h"
 
-#define IND (I810_OFFSET_BIT)
+#define IND (I810_OFFSET_BIT)	/* wide */
 #define TAG(x) x##_offset
 #include "i810tritmp.h"
 
-#define IND (I810_OFFSET_BIT|I810_FLAT_BIT)
+#define IND (I810_OFFSET_BIT|I810_FLAT_BIT) /* wide|flat */
 #define TAG(x) x##_offset_flat
 #include "i810tritmp.h"
 
-#define IND (I810_TWOSIDE_BIT)
+#define IND (I810_TWOSIDE_BIT)	/* stipple */
 #define TAG(x) x##_twoside
 #include "i810tritmp.h"
 
-#define IND (I810_TWOSIDE_BIT|I810_FLAT_BIT)
+#define IND (I810_TWOSIDE_BIT|I810_FLAT_BIT) /* stipple|flat */
 #define TAG(x) x##_twoside_flat
 #include "i810tritmp.h"
 
-#define IND (I810_TWOSIDE_BIT|I810_OFFSET_BIT)
+#define IND (I810_TWOSIDE_BIT|I810_OFFSET_BIT) /* stipple|wide */
 #define TAG(x) x##_twoside_offset
 #include "i810tritmp.h"
 
-#define IND (I810_TWOSIDE_BIT|I810_OFFSET_BIT|I810_FLAT_BIT)
+#define IND (I810_TWOSIDE_BIT|I810_OFFSET_BIT|I810_FLAT_BIT) /* stip|wide|flat*/
 #define TAG(x) x##_twoside_offset_flat
 #include "i810tritmp.h"
 
 void i810DDTrifuncInit()
 {
-   int i;
    init();
    init_flat();
    init_offset();
@@ -125,113 +95,9 @@ void i810DDTrifuncInit()
    init_twoside_flat();
    init_twoside_offset();
    init_twoside_offset_flat();
-
-   /* Hmmm...
-    */
-   for (i = 0 ; i < 0x20 ; i++) {
-      if (i & ~I810_FLAT_BIT) {
-	 points_tab[i] = points_tab[i&I810_FLAT_BIT];
-	 line_tab[i] = line_tab[i&I810_FLAT_BIT];
-      }
-   }
-
-   for (i = 0 ; i < 0x20 ; i++) 
-      if ((i & (I810_NODRAW_BIT|I810_FALLBACK_BIT)) == I810_NODRAW_BIT ||
-	  i810glx.nullprims) 
-      {
-	 quad_tab[i] = i810_null_quad; 
-	 tri_tab[i] = i810_null_triangle; 
-	 line_tab[i] = i810_null_line;
-	 points_tab[i] = i810_null_points;
-      }
-
-   if (i810glx.noFallback) {
-      for (i = 0 ; i < 0x10 ; i++) {
-	 points_tab[i|I810_FALLBACK_BIT] = points_tab[i];
-	 line_tab[i|I810_FALLBACK_BIT] = line_tab[i];
-	 tri_tab[i|I810_FALLBACK_BIT] = tri_tab[i];
-	 quad_tab[i|I810_FALLBACK_BIT] = quad_tab[i];
-      }
-   }
-
 }
 
 
-/* Everything is done via single triangle instructiopns at the moment;
- * this can change fairly easily.
- */
-#if I810_USE_BATCH
-
-GLuint *i810AllocPrimitiveVerts( int dwords ) 
-{
-   GLuint orig_dwords = dwords;
-
-   dwords+=2;
-   dwords&=~1;
-
-   while (1) {
-      if (i810glx.dma_buffer->space < dwords * 4) 
-      {
-	 if (I810_DEBUG & DEBUG_VERBOSE_RING)
-	    fprintf(stderr, "i810AllocPrimitiveVerts: dma buffer overflow\n");
-	 i810DmaOverflow( dwords );
-      }
-      else
-      {
-	 GLuint start = i810glx.dma_buffer->head;
-	 i810glx.dma_buffer->head += dwords * 4;
-	 i810glx.dma_buffer->space -= dwords * 4;
-
-	 if ((orig_dwords & 1) == 0) {
-	    *(GLuint *)(i810glx.dma_buffer->virtual_start + start ) = 0;
-	    start += 4;
-	 }
-
-	 *(GLuint *)(i810glx.dma_buffer->virtual_start + start ) = 
-	    GFX_OP_PRIMITIVE | PR_TRIANGLES | (orig_dwords-1);
-
-	 return (GLuint *)(i810glx.dma_buffer->virtual_start + start + 4);
-      }
-   }
-}
-
-#else
-GLuint *i810AllocPrimitiveVerts( int dwords ) 
-{
-   GLuint orig_dwords = dwords;
-   
-   dwords+=2;
-   dwords&=~1;
-
-   while (1)
-   {
-      BEGIN_LP_RING( dwords );
-
-      if (outring + dwords * 4 != ((outring + dwords * 4) & ringmask)) 
-      {
-	 int i;
-
-	 if (I810_DEBUG & DEBUG_VERBOSE_RING)
-	    fprintf(stderr, "\n\nwrap case in i810AllocPrimitiveVerts\n\n");
-
-	 for (i = 0 ; i < dwords ; i++)
-	    OUT_RING( 0 );
-	 ADVANCE_LP_RING();
-      }
-      else 
-      {
-	 I810LpRing.tail = outring + dwords * 4;
-
-	 if ((orig_dwords & 1) == 0)
-	    OUT_RING(0);
-
-	 OUT_RING( GFX_OP_PRIMITIVE | PR_TRIANGLES | (orig_dwords-1) );
-	 return (GLuint *)(virt + outring);
-      }
-   }
-}
-#endif
-
 
 
 void i810DDChooseRenderState( GLcontext *ctx )
@@ -239,50 +105,41 @@ void i810DDChooseRenderState( GLcontext *ctx )
    i810ContextPtr imesa = I810_CONTEXT( ctx );
    GLuint flags = ctx->TriangleCaps;
 
-   ctx->IndirectTriangles &= ~DD_SW_RASTERIZE;
+   imesa->IndirectTriangles = 0;
 
    if (flags) {
       GLuint ind = 0;
       GLuint shared = 0;
-      GLuint setup = imesa->setupindex;
-      GLuint fallback = I810_FALLBACK_BIT; 
-
-      if (i810glx.noFallback) fallback = 0; 
 
-      if ((flags & DD_FLATSHADE) && (setup & I810_RGBA_BIT)) shared |= I810_FLAT_BIT;
-      if (flags & DD_MULTIDRAW)                    shared |= fallback;
+      if (flags & DD_FLATSHADE)                    shared |= I810_FLAT_BIT;
+      if (flags & DD_MULTIDRAW)                    shared |= I810_FALLBACK_BIT;
       if (flags & DD_SELECT)                       shared |= I810_FALLBACK_BIT;
       if (flags & DD_FEEDBACK)                     shared |= I810_FALLBACK_BIT;
 
-      ind = shared;
-      if (flags & DD_POINT_SMOOTH)                 ind |= I810_ANTIALIAS_BIT;
-      if (flags & DD_POINT_ATTEN)                  ind |= fallback;
-
-      imesa->renderindex = ind;
-      imesa->PointsFunc = points_tab[ind]; 
-      if (ind & I810_FALLBACK_BIT)  
-	 ctx->IndirectTriangles |= DD_POINT_SW_RASTERIZE;
+      imesa->renderindex = shared;
+      imesa->PointsFunc = points_tab[shared]; 
 
       ind = shared;
-      if (flags & DD_LINE_SMOOTH)                   ind |= I810_ANTIALIAS_BIT;
-      if (flags & DD_LINE_STIPPLE)                  ind |= fallback;
+      if (flags & DD_LINE_WIDTH)                    ind |= I810_WIDE_LINE_BIT;
+      if (flags & DD_LINE_STIPPLE)                  ind |= I810_FALLBACK_BIT;
 
       imesa->renderindex |= ind;
       imesa->LineFunc = line_tab[ind];
       if (ind & I810_FALLBACK_BIT) 
-	 ctx->IndirectTriangles |= DD_LINE_SW_RASTERIZE;
+	 imesa->IndirectTriangles |= DD_LINE_SW_RASTERIZE;
 
       ind = shared;
-      if (flags & DD_TRI_SMOOTH)                    ind |= I810_ANTIALIAS_BIT;
       if (flags & DD_TRI_OFFSET)                    ind |= I810_OFFSET_BIT;
       if (flags & DD_TRI_LIGHT_TWOSIDE)             ind |= I810_TWOSIDE_BIT;
-      if (flags & (DD_TRI_UNFILLED|DD_TRI_STIPPLE)) ind |= fallback;	
+      if (flags & DD_TRI_UNFILLED)                  ind |= I810_FALLBACK_BIT;
+      if ((flags & DD_TRI_STIPPLE) &&
+	  (ctx->IndirectTriangles & DD_TRI_STIPPLE)) ind |= I810_FALLBACK_BIT;	
 
       imesa->renderindex |= ind;
       imesa->TriangleFunc = tri_tab[ind];
       imesa->QuadFunc = quad_tab[ind];
       if (ind & I810_FALLBACK_BIT)
-	 ctx->IndirectTriangles |= (DD_TRI_SW_RASTERIZE|DD_QUAD_SW_RASTERIZE);
+	 imesa->IndirectTriangles |= (DD_TRI_SW_RASTERIZE|DD_QUAD_SW_RASTERIZE);
    } 
    else if (imesa->renderindex)
    {
@@ -293,9 +150,9 @@ void i810DDChooseRenderState( GLcontext *ctx )
       imesa->QuadFunc     = quad_tab[0];
    }
 
-   if (MESA_VERBOSE&VERBOSE_DRIVER) {
+
+   if (I810_DEBUG&DEBUG_VERBOSE_API) {
       gl_print_tri_caps("tricaps", ctx->TriangleCaps);
-      i810PrintRenderState("i810: Render state", imesa->renderindex);
    }
 }
 
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810tris.h b/xc/lib/GL/mesa/src/drv/i810/i810tris.h
index 12e0ec7fd..5b73b7676 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810tris.h
+++ b/xc/lib/GL/mesa/src/drv/i810/i810tris.h
@@ -23,107 +23,76 @@
  *
  */
 
-#ifndef I810TIS_INC
-#define I810TIS_INC
+#ifndef I810TRIS_INC
+#define I810TRIS_INC
 
-#include "mesaglx/types.h"
+#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();
+extern void i810DDTrifuncInit( void );
 
-/*  extern void i810FinishPrimitive( void ); */
-extern void i810NaughtyFinishPrimitive( void );
-extern GLuint *i810AllocPrimitiveVerts( int dwords );
 
+/* shared */
+#define I810_FLAT_BIT 	     0x1
 
+/* triangle */
+#define I810_OFFSET_BIT	     0x2	
+#define I810_TWOSIDE_BIT     0x4
 
-/* Todo: 
- *    - multidraw, ...
- *    - Antialiasing (?)
- *    - line and polygon stipple
- *    - select and feedback 
- *    - stencil 
- *    - point parameters
- *    - 
- */
-#define I810_ANTIALIAS_BIT   0       /* ignored for now, no fallback */
-#define I810_FLAT_BIT	0x1
-#define I810_OFFSET_BIT	0x2	/* 3.1 only */
-#define I810_TWOSIDE_BIT	0x4	/* 3.1 only */
-#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 inline void i810_draw_triangle( i810_vertex *v0, 
-				       i810_vertex *v1, 
-				       i810_vertex *v2 )
-{
 
-#if 0
-   fprintf(stderr, "i810_draw_triangle( %p, %p, %p )\n", v0, v1, v2);
-#endif
+static i810_vertex __inline__ *i810AllocTriangles( i810ContextPtr imesa, int nr)
+{
+   GLuint *start = i810AllocDwords( imesa, 30*nr+1 );
+   *start++ = GFX_OP_PRIMITIVE | PR_TRIANGLES | (30*nr-1);
+   return (i810_vertex *)start;
+}
 
-#if 0
-#define WID 640
-#define HI 480
-   {
-      GLuint print = 1, draw = 1;
-      GLfloat area = ((v0->x - v2->x) * (v1->y - v2->y) -
-		      (v0->y - v2->y) * (v1->x - v2->x));
-
-      if (v0->x < -.501 || v0->x > WID || v0->y < -.501 || v0->y > HI ||
-	  v1->x < -.501 || v1->x > WID || v1->y < -.501 || v1->y > HI ||
-	  v2->x < -.501 || v2->x > WID || v2->y < -.501 || v2->y > HI) {
-	 fprintf(stderr, "not clipped\n");
-	 print = 1;
-      } 
-
-      if (area == 0 && ((v0 == v1) || (v1 == v2))) {
-	 fprintf(stderr, "zero area %p %p %p\n", v0, v1, v2);
-	 draw = 0;
-      }
-
-      if (print) {
-	 fprintf(stderr," v0: %f %f %f %f tex: %f %f\n",
-		 v0->x, v0->y, v0->z, v0->oow, v0->tu0, v0->tv0);
-	 fprintf(stderr," v1: %f %f %f %f tex: %f %f\n",
-		 v1->x, v1->y, v1->z, v1->oow, v1->tu0, v1->tv0);
-	 fprintf(stderr," v2: %f %f %f %f tex: %f %f\n",
-		 v2->x, v2->y, v2->z, v2->oow, v2->tu0, v2->tv0);
-	 return;
-      }
-      
-      if (!draw) return;
-   }
-#endif
+static i810_vertex __inline__ *i810AllocLine( i810ContextPtr imesa ) 
+{
+   GLuint *start = i810AllocDwords( imesa, 21 );
+   *start++ = GFX_OP_PRIMITIVE | PR_LINESTRIP | (20-1);
+   return (i810_vertex *)start;
+}
 
-   {
-      i810_vertex *wv = (i810_vertex *)i810AllocPrimitiveVerts( 3 * 10 );
-      wv[0] = *v0;
-      wv[1] = *v1;
-      wv[2] = *v2;
-      FINISH_PRIM();
-   }
+static i810_vertex __inline__ *i810AllocRect( i810ContextPtr imesa ) 
+{
+   GLuint *start = i810AllocDwords( imesa, 31 );
+   *start++ = GFX_OP_PRIMITIVE | PR_RECTS | (30-1);
+   return (i810_vertex *)start;
 }
 
 
 
+static void __inline__ i810_draw_triangle( i810ContextPtr imesa,
+					   i810_vertex *v0, 
+					   i810_vertex *v1, 
+					   i810_vertex *v2 )
+{
+   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( i810_vertex *tmp, float sz )
+
+static __inline__ void i810_draw_point( i810ContextPtr imesa,
+					i810_vertex *tmp, float sz )
 {
-   i810_vertex *wv = (i810_vertex *)i810AllocPrimitiveVerts(6*10);
+   i810_vertex *wv = i810AllocTriangles( imesa, 2 );
 
    wv[0] = *tmp;
    wv[0].x = tmp->x - sz;
@@ -153,11 +122,23 @@ static inline void i810_draw_point( i810_vertex *tmp, float sz )
 }
 
 
-static inline void i810_draw_line( 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 = i810AllocLine( imesa );
+   wv[0] = *tmp0;
+   wv[1] = *tmp1;
+   FINISH_PRIM();
+}
+
+static __inline__ void i810_draw_tri_line( i810ContextPtr imesa, 
+					   i810_vertex *tmp0, 
+					   i810_vertex *tmp1,
+					   float width )
 {
-   i810_vertex *wv = (i810_vertex *)i810AllocPrimitiveVerts( 6 * 10 );
+   i810_vertex *wv = i810AllocTriangles( imesa, 2 );
+      
    float dx, dy, ix, iy;
 
    dx = tmp0->x - tmp1->x;
@@ -168,15 +149,6 @@ static inline void i810_draw_line( i810_vertex *tmp0,
       iy = ix; ix = 0;
    }
 
-#if 0
-   fprintf(stderr,"tmp0: %f %f %f %f col: %x tex: %f %f\n",
-	   tmp0->x, tmp0->y, tmp0->z, tmp0->oow, *(GLuint*)&tmp0->color, tmp0->tu0, tmp0->tv0);
-   fprintf(stderr,"tmp1: %f %f %f %f col: %x tex: %f %f\n",
-	   tmp1->x, tmp1->y, tmp1->z, tmp1->oow, *(GLuint*)&tmp1->color, tmp1->tu0, tmp1->tv0);
-   fprintf(stderr, "ix: %f, iy: %f\n", ix, iy);
-#endif
-
-
    wv[0] = *tmp0;
    wv[0].x = tmp0->x - ix;
    wv[0].y = tmp0->y - iy;
@@ -200,8 +172,16 @@ static inline void i810_draw_line( i810_vertex *tmp0,
    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 c8ec01d49..ebe1ef5ea 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810tritmp.h
+++ b/xc/lib/GL/mesa/src/drv/i810/i810tritmp.h
@@ -1,19 +1,20 @@
 
-static inline void TAG(triangle)( GLcontext *ctx, GLuint e0,
-				  GLuint e1, GLuint e2, GLuint pv )
+static __inline__ void TAG(triangle)( GLcontext *ctx, GLuint e0,
+				      GLuint e1, GLuint e2, GLuint pv )
 {
+   i810ContextPtr imesa = I810_CONTEXT(ctx);
    struct vertex_buffer *VB = ctx->VB;
    i810VertexPtr i810VB = I810_DRIVER_DATA(VB)->verts;
-   const i810_vertex *v0 = &i810VB[e0].v.v;  
-   const i810_vertex *v1 = &i810VB[e1].v.v;  
-   const i810_vertex *v2 = &i810VB[e2].v.v;  
+   const i810_vertex *v0 = &i810VB[e0].v;  
+   const i810_vertex *v1 = &i810VB[e1].v;  
+   const i810_vertex *v2 = &i810VB[e2].v;  
 
 #if (IND & I810_OFFSET_BIT)
    GLfloat offset = ctx->Polygon.OffsetUnits * 1.0/0x10000;
 #endif
 
 #if (IND & (I810_FLAT_BIT|I810_TWOSIDE_BIT))
-   int c0 = *(int *)&i810VB[pv].v.v.color; 
+   int c0 = *(int *)&i810VB[pv].v.color; 
    int c1 = c0;
    int c2 = c0;
 #endif
@@ -64,78 +65,39 @@ static inline void TAG(triangle)( GLcontext *ctx, GLuint e0,
 
    i810glx.c_triangles++;
 
-#if 0
-#define WID 512
-#define HI 384
-   {
-      GLuint print = 0, draw = 1;
-      GLfloat area = ((v0->x - v2->x) * (v1->y - v2->y) -
-		      (v0->y - v2->y) * (v1->x - v2->x));
-
-      if (v0->x < -.501 || v0->x > WID || v0->y < -.501 || v0->y > HI ||
-	  v1->x < -.501 || v1->x > WID || v1->y < -.501 || v1->y > HI ||
-	  v2->x < -.501 || v2->x > WID || v2->y < -.501 || v2->y > HI) {
-	 fprintf(stderr, "not clipped\n");
-	 print = 1;
-      } 
-
-      if (area == 0 && ((v0 == v1) || (v1 == v2))) {
-	 fprintf(stderr, "zero area %p %p %p\n", v0, v1, v2);
-	 draw = 0;
-      }
-
-      if (print) {
-	 fprintf(stderr,"v0: %f %f %f %f col: %x tex: %f %f\n",
-		 v0->x, v0->y, v0->z, v0->oow, 
-		 *(GLuint*)&v0->color, v0->tu0, v0->tv0);
-	 fprintf(stderr,"v1: %f %f %f %f col: %x tex: %f %f\n",
-		 v1->x, v1->y, v1->z, v1->oow, 
-		 *(GLuint*)&v1->color, v1->tu0, v1->tv0);
-	 fprintf(stderr,"v2: %f %f %f %f col: %x tex: %f %f\n",
-		 v2->x, v2->y, v2->z, v2->oow, 
-		 *(GLuint*)&v2->color, v2->tu0, v2->tv0);
-/*  	 return; */
-      }
-      
-      if (!draw) return;
-   }
-#endif
    
-   BEGIN_CLIP_LOOP()
-      {
-	 i810_vertex *wv = (i810_vertex *)i810AllocPrimitiveVerts( 3 * 10 );
-	 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();
+      FINISH_PRIM();
+   }
 }
 
 
-
 static void TAG(quad)( GLcontext *ctx, GLuint v0,
 		       GLuint v1, GLuint v2, GLuint v3, 
 		       GLuint pv )
@@ -144,60 +106,66 @@ 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.v;
-   i810_vertex tmp1 = i810VB[v1].v.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.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()
-      i810_draw_line( &tmp0, &tmp1, width );
-   END_CLIP_LOOP();
+   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;
+   }      
 }
 
 
 static void TAG(points)( GLcontext *ctx, GLuint first, GLuint last )
 {
-   int i;
+   i810ContextPtr imesa = I810_CONTEXT( ctx );
    struct vertex_buffer *VB = ctx->VB;
    i810VertexPtr i810VB = I810_DRIVER_DATA(VB)->verts;
    GLfloat sz = ctx->Point.Size * .5;
+   int i;
 
    /* Culling is disabled automatically via. the
     * ctx->Driver.ReducedPrimitiveChange() callback.  
     */
-
+   
    for(i=first;i<=last;i++) {
       if(VB->ClipMask[i]==0) {
-	 i810_vertex *tmp = &i810VB[i].v.v;
-	 BEGIN_CLIP_LOOP()
-	    i810_draw_point( tmp, sz );
-	 END_CLIP_LOOP();
+	 i810_vertex *tmp = &i810VB[i].v;
+	 i810_draw_point( imesa, tmp, sz );
       }
    }
 }
 
-#endif
 
 
-static void TAG(init)()
+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 b41ad0960..6ba7eb9a5 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810vb.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810vb.c
@@ -28,46 +28,43 @@
 #include "i810vb.h"
 #include "i810log.h"
 #include "i810dma.h"
-#include "xsmesaP.h"
-#include "xsmesa.h"
 
 #include "stages.h"
 
-#include "os.h"
 #include <stdio.h>
 #include <stdlib.h>
 
 
 #define TEX0 {					\
-  v->v.v.tu0 = tc0[i][0];			\
-  v->v.v.tv0 = tc0[i][1];			\
+  v->v.tu0 = tc0[i][0];			\
+  v->v.tv0 = tc0[i][1];			\
 }
 
 #define TEX1 {					\
-  v->v.v.tu1 = tc1[i][0];			\
-  v->v.v.tv1 = tc1[i][1];			\
+  v->v.tu1 = tc1[i][0];			\
+  v->v.tv1 = tc1[i][1];			\
 }
 
 /* Doesn't seem to work very well (golly).
  */
 #define SPC {					\
   GLubyte *spec = &(VB->Spec[0][i][0]);		\
-  v->v.v.specular.red = spec[0];		\
-  v->v.v.specular.green = spec[1];		\
-  v->v.v.specular.blue = spec[2];		\
+  v->v.specular.red = spec[0];		\
+  v->v.specular.green = spec[1];		\
+  v->v.specular.blue = spec[2];		\
 }
 
 #define FOG {					\
   GLubyte *spec = &(VB->Spec[0][i][0]);		\
-  v->v.v.specular.alpha = spec[3];		\
+  v->v.specular.alpha = spec[3];		\
 }
 
 #define COL {					\
   GLubyte *col = &(VB->Color[0]->data[i][0]);	\
-  v->v.v.color.blue  = col[2];		\
-  v->v.v.color.green = col[1];		\
-  v->v.v.color.red   = col[0];		\
-  v->v.v.color.alpha = col[3];		\
+  v->v.color.blue  = col[2];		\
+  v->v.color.green = col[1];		\
+  v->v.color.red   = col[0];		\
+  v->v.color.alpha = col[3];		\
 }
 
 /* The vertex formats we have don't seem to support projective texturing
@@ -82,19 +79,19 @@
      imesa->setupdone &= ~I810_WIN_BIT;		\
      for (i=start; i < end; i++, v++)	{		\
         float oow = 1.0 / tc[i][3];			\
-	v->v.v.oow *= tc[i][3];				\
-	v->v.v.tu0 *= oow;				\
-	v->v.v.tv0 *= oow;				\
+	v->v.oow *= tc[i][3];				\
+	v->v.tu0 *= oow;				\
+	v->v.tv0 *= oow;				\
      }							\
   }
 
 
 #define COORD					\
       GLfloat *win = VB->Win.data[i];		\
-      v->v.v.x =                 win[0];	\
-      v->v.v.y =    i810height - win[1];	\
-      v->v.v.z = (1.0/0x10000) * win[2];	\
-      v->v.v.oow =               win[3]; 
+      v->v.x =                 win[0];	\
+      v->v.y =    i810height - win[1];	\
+      v->v.z = (1.0/0x10000) * win[2];	\
+      v->v.oow =               win[3]; 
 
 
 
@@ -104,15 +101,15 @@
 #define SETUPFUNC(name,win,col,tex0,tex1,tex0_4,spec,fog)		\
 static void name(struct vertex_buffer *VB, GLuint start, GLuint end)	\
 {									\
-   i810ContextPtr imesa = I810_CONTEXT( ctx );                          \
+   i810ContextPtr imesa = I810_CONTEXT( VB->ctx );			\
+   __DRIdrawablePrivate *dPriv = imesa->driDrawable;			\
    i810VertexPtr v;							\
    GLfloat (*tc0)[4];							\
    GLfloat (*tc1)[4];							\
-   GLfloat i810height = i810DB->Height;					\
+   GLfloat i810height = dPriv->h;					\
    int i;								\
-   (void) i810height;	                				\
+   (void) i810height; (void) imesa;					\
 									\
-   CHECK_CONTEXT( return; );						\
 									\
    gl_import_client_data( VB, VB->ctx->RenderFlags,			\
 			  (VB->ClipOrMask				\
@@ -143,7 +140,7 @@ static void name(struct vertex_buffer *VB, GLuint start, GLuint end)	\
 	    tex1;							\
 	 }								\
 	    col;							\
-      }						  		        \
+      }									\
    tex0_4;								\
 }
 
@@ -155,11 +152,13 @@ SETUPFUNC(rs_wt0t1,	COORD,NOP,TEX0,TEX1,TEX0_4,NOP,NOP)
 SETUPFUNC(rs_wft0,	COORD,NOP,TEX0,NOP,TEX0_4,NOP,FOG)
 SETUPFUNC(rs_wft0t1,	COORD,NOP,TEX0,TEX1,TEX0_4,NOP,FOG)
 SETUPFUNC(rs_wg,	COORD,COL,NOP,NOP,NOP,NOP,NOP)
+SETUPFUNC(rs_wgs,	COORD,COL,NOP,NOP,NOP,SPC,NOP)
 SETUPFUNC(rs_wgt0,	COORD,COL,TEX0,NOP,TEX0_4,NOP,NOP)
 SETUPFUNC(rs_wgt0t1,	COORD,COL,TEX0,TEX1,TEX0_4,NOP,NOP)
 SETUPFUNC(rs_wgst0,	COORD,COL,TEX0,NOP,TEX0_4,SPC,NOP)
 SETUPFUNC(rs_wgst0t1,	COORD,COL,TEX0,TEX1,TEX0_4,SPC,NOP)
 SETUPFUNC(rs_wgf,	COORD,COL,NOP,NOP,NOP,NOP,FOG)
+SETUPFUNC(rs_wgfs,	COORD,COL,NOP,NOP,NOP,SPC,FOG)
 SETUPFUNC(rs_wgft0,	COORD,COL,TEX0,NOP,TEX0_4,NOP,FOG)
 SETUPFUNC(rs_wgft0t1,	COORD,COL,TEX0,TEX1,TEX0_4,NOP,FOG)
 SETUPFUNC(rs_wgfst0,	COORD,COL,TEX0,NOP,TEX0_4,SPC,FOG)
@@ -171,11 +170,13 @@ SETUPFUNC(rs_f,		NOP,NOP,NOP,NOP,NOP,NOP,FOG)
 SETUPFUNC(rs_ft0,	NOP,NOP,TEX0,NOP,TEX0_4,NOP,FOG)
 SETUPFUNC(rs_ft0t1,	NOP,NOP,TEX0,TEX1,TEX0_4,NOP,FOG)
 SETUPFUNC(rs_g,		NOP,COL,NOP,NOP,NOP,NOP,NOP)
+SETUPFUNC(rs_gs,	NOP,COL,NOP,NOP,NOP,SPC,NOP)
 SETUPFUNC(rs_gt0,	NOP,COL,TEX0,NOP,TEX0_4,NOP,NOP)
 SETUPFUNC(rs_gt0t1,	NOP,COL,TEX0,TEX1,TEX0_4,NOP,NOP)
 SETUPFUNC(rs_gst0,	NOP,COL,TEX0,NOP,TEX0_4,SPC,NOP)
 SETUPFUNC(rs_gst0t1,	NOP,COL,TEX0,TEX1,TEX0_4,SPC,NOP)
 SETUPFUNC(rs_gf,	NOP,COL,NOP,NOP,NOP,NOP,FOG)
+SETUPFUNC(rs_gfs,	NOP,COL,NOP,NOP,NOP,SPC,FOG)
 SETUPFUNC(rs_gft0,	NOP,COL,TEX0,NOP,TEX0_4,NOP,FOG)
 SETUPFUNC(rs_gft0t1,	NOP,COL,TEX0,TEX1,TEX0_4,NOP,FOG)
 SETUPFUNC(rs_gfst0,	NOP,COL,TEX0,NOP,TEX0_4,SPC,FOG)
@@ -206,11 +207,13 @@ void i810DDSetupInit( void )
    setup_func[I810_WIN_BIT|I810_FOG_BIT|I810_TEX0_BIT] = rs_wft0;
    setup_func[I810_WIN_BIT|I810_FOG_BIT|I810_TEX0_BIT|I810_TEX1_BIT] = rs_wft0t1;
    setup_func[I810_WIN_BIT|I810_RGBA_BIT] = rs_wg;
+   setup_func[I810_WIN_BIT|I810_RGBA_BIT|I810_SPEC_BIT] = rs_wgs;
    setup_func[I810_WIN_BIT|I810_RGBA_BIT|I810_TEX0_BIT] = rs_wgt0;
    setup_func[I810_WIN_BIT|I810_RGBA_BIT|I810_TEX0_BIT|I810_TEX1_BIT] = rs_wgt0t1;
    setup_func[I810_WIN_BIT|I810_RGBA_BIT|I810_SPEC_BIT|I810_TEX0_BIT] = rs_wgst0;
    setup_func[I810_WIN_BIT|I810_RGBA_BIT|I810_SPEC_BIT|I810_TEX0_BIT|I810_TEX1_BIT] = rs_wgst0t1;
    setup_func[I810_WIN_BIT|I810_RGBA_BIT|I810_FOG_BIT] = rs_wgf;
+   setup_func[I810_WIN_BIT|I810_RGBA_BIT|I810_FOG_BIT|I810_SPEC_BIT] = rs_wgfs;
    setup_func[I810_WIN_BIT|I810_RGBA_BIT|I810_FOG_BIT|I810_TEX0_BIT] = rs_wgft0;
    setup_func[I810_WIN_BIT|I810_RGBA_BIT|I810_FOG_BIT|I810_TEX0_BIT|I810_TEX1_BIT] = rs_wgft0t1;
    setup_func[I810_WIN_BIT|I810_RGBA_BIT|I810_FOG_BIT|I810_SPEC_BIT|I810_TEX0_BIT] = rs_wgfst0;
@@ -223,11 +226,13 @@ void i810DDSetupInit( void )
    setup_func[I810_FOG_BIT|I810_TEX0_BIT] = rs_ft0;
    setup_func[I810_FOG_BIT|I810_TEX0_BIT|I810_TEX1_BIT] = rs_ft0t1;
    setup_func[I810_RGBA_BIT] = rs_g;
+   setup_func[I810_RGBA_BIT|I810_SPEC_BIT] = rs_gs;
    setup_func[I810_RGBA_BIT|I810_TEX0_BIT] = rs_gt0;
    setup_func[I810_RGBA_BIT|I810_TEX0_BIT|I810_TEX1_BIT] = rs_gt0t1;
    setup_func[I810_RGBA_BIT|I810_SPEC_BIT|I810_TEX0_BIT] = rs_gst0;
    setup_func[I810_RGBA_BIT|I810_SPEC_BIT|I810_TEX0_BIT|I810_TEX1_BIT] = rs_gst0t1;
    setup_func[I810_RGBA_BIT|I810_FOG_BIT] = rs_gf;
+   setup_func[I810_RGBA_BIT|I810_FOG_BIT|I810_SPEC_BIT] = rs_gfs;
    setup_func[I810_RGBA_BIT|I810_FOG_BIT|I810_TEX0_BIT] = rs_gft0;
    setup_func[I810_RGBA_BIT|I810_FOG_BIT|I810_TEX0_BIT|I810_TEX1_BIT] = rs_gft0t1;
    setup_func[I810_RGBA_BIT|I810_FOG_BIT|I810_SPEC_BIT|I810_TEX0_BIT] = rs_gfst0;
@@ -240,7 +245,7 @@ void i810PrintSetupFlags(char *msg, GLuint flags )
 {
    fprintf(stderr, "%s: %d %s%s%s%s%s%s%s\n",
 	   msg,
-	   flags,
+	   (int)flags,
 	   (flags & I810_WIN_BIT)      ? " xyzw," : "", 
 	   (flags & I810_RGBA_BIT)     ? " rgba," : "",
 	   (flags & I810_SPEC_BIT)     ? " spec," : "",
@@ -274,7 +279,7 @@ void i810ChooseRasterSetupFunc(GLcontext *ctx)
   if (ctx->FogMode == FOG_FRAGMENT)
      funcindex |= I810_FOG_BIT;
     
-  if (MESA_VERBOSE || I810_DEBUG)
+  if (MESA_VERBOSE)
      i810PrintSetupFlags("xsmesa: full setup function", funcindex); 
 
   imesa->setupindex = funcindex;
@@ -312,6 +317,7 @@ void i810DDCheckPartialRasterSetup( GLcontext *ctx,
  */
 void i810DDPartialRasterSetup( struct vertex_buffer *VB )
 {
+   i810ContextPtr imesa = I810_CONTEXT( VB->ctx );
    GLuint new = VB->pipeline->new_outputs;
    GLuint available = VB->pipeline->outputs;
    GLuint ind = 0;
@@ -337,7 +343,7 @@ void i810DDPartialRasterSetup( struct vertex_buffer *VB )
    ind &= imesa->setupindex;
    imesa->setupdone |= ind;
 
-   i810PrintSetupFlags("xsmesa: partial setup function", ind);
+   if (0) i810PrintSetupFlags("xsmesa: partial setup function", ind);
 
    if (ind) 
       setup_func[ind&~I810_ALPHA_BIT]( VB, VB->Start, VB->Count );   
@@ -366,28 +372,37 @@ void i810DDResizeVB( struct vertex_buffer *VB, GLuint size )
 
    free( mvb->vert_store );
    mvb->vert_store = malloc( sizeof(i810Vertex) * mvb->size + 31);
-   if (!mvb->vert_store) 
-      FatalError("i810-glx: out of memory !\n");
+   if (!mvb->vert_store) {
+      fprintf(stderr, "i810-glx: out of memory !\n");
+      exit(1);
+   }
 
    mvb->verts = (i810VertexPtr)(((unsigned long)mvb->vert_store + 31) & ~31);
 
    gl_vector1ui_free( &mvb->clipped_elements );
    gl_vector1ui_alloc( &mvb->clipped_elements, VEC_WRITABLE, mvb->size, 32 );   
-   if (!mvb->clipped_elements.start) 
-      FatalError("i810-glx: out of memory !\n");
+   if (!mvb->clipped_elements.start) {
+      fprintf(stderr, "i810-glx: out of memory !\n");
+      exit(1);
+   }
 
    free( VB->ClipMask );
    VB->ClipMask = (GLubyte *)malloc(sizeof(GLubyte) * mvb->size);
-   if (!VB->ClipMask) 
-      FatalError("i810-glx: out of memory !\n");
+   if (!VB->ClipMask) {
+      fprintf(stderr, "i810-glx: out of memory !\n");
+      exit(1);
+   }
+
 
    if (VB->Type == VB_IMMEDIATE) {
       free( mvb->primitive );
       free( mvb->next_primitive );
       mvb->primitive = (GLuint *)malloc( sizeof(GLuint) * mvb->size );
       mvb->next_primitive = (GLuint *)malloc( sizeof(GLuint) * mvb->size );
-      if (!mvb->primitive || !mvb->next_primitive)
-	 FatalError("i810-glx: out of memory!");
+      if (!mvb->primitive || !mvb->next_primitive) {
+	 fprintf(stderr, "i810-glx: out of memory !\n");
+	 exit(1);
+      }
    }
 }
 
@@ -398,31 +413,35 @@ void i810DDRegisterVB( struct vertex_buffer *VB )
 
    mvb = (i810VertexBufferPtr)calloc( 1, sizeof(*mvb) );
 
-   /* This looks like it allocates a lot of memory, but it basically
-    * just sets an upper limit on how much can be used - nothing like
-    * this amount will ever be turned into 'real' memory.
-    */
-   mvb->size = VB->Size * 5;
+   mvb->size = VB->Size * 2;
    mvb->vert_store = malloc( sizeof(i810Vertex) * mvb->size + 31);
-   if (!mvb->vert_store) 
-      FatalError("i810-glx: out of memory !\n");
+   if (!mvb->vert_store) {
+      fprintf(stderr, "i810-glx: out of memory !\n");
+      exit(1);
+   }
    
    mvb->verts = (i810VertexPtr)(((unsigned long)mvb->vert_store + 31) & ~31);
 
    gl_vector1ui_alloc( &mvb->clipped_elements, VEC_WRITABLE, mvb->size, 32 );
-   if (!mvb->clipped_elements.start) 
-      FatalError("i810-glx: out of memory !\n");
-
+   if (!mvb->clipped_elements.start) {
+      fprintf(stderr, "i810-glx: out of memory !\n");
+      exit(1);
+   }
+      
    free( VB->ClipMask );
    VB->ClipMask = (GLubyte *)malloc(sizeof(GLubyte) * mvb->size);
-   if (!VB->ClipMask) 
-      FatalError("i810-glx: out of memory !\n");
+   if (!VB->ClipMask) {
+      fprintf(stderr, "i810-glx: out of memory !\n");
+      exit(1);
+   }
 
    mvb->primitive = (GLuint *)malloc( sizeof(GLuint) * mvb->size );
    mvb->next_primitive = (GLuint *)malloc( sizeof(GLuint) * mvb->size );
-   if (!mvb->primitive || !mvb->next_primitive)
-      FatalError("i810-glx: out of memory!");
-   
+   if (!mvb->primitive || !mvb->next_primitive) {
+      fprintf(stderr, "i810-glx: out of memory !\n");
+      exit(1);
+   }
+
    VB->driver_data = mvb;
 }
 
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810vb.h b/xc/lib/GL/mesa/src/drv/i810/i810vb.h
index 640afa678..23685f106 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810vb.h
+++ b/xc/lib/GL/mesa/src/drv/i810/i810vb.h
@@ -26,7 +26,8 @@
 #ifndef I810VB_INC
 #define I810VB_INC
 
-#include "mesaglx/vb.h"
+#include "vb.h"
+#include "types.h"
 
 /* 
  *  color type for the vertex data
@@ -56,15 +57,13 @@ typedef struct {
 
 /* Unfortunately only have assembly for 16-stride vertices.
  */
-struct i810_vertex_t {
-   union {
-      i810_vertex v;
-      float f[16];
-   } v;
+union i810_vertex_t {
+   i810_vertex v;
+   float f[16];
 };
 
-typedef struct i810_vertex_t i810Vertex;
-typedef struct i810_vertex_t *i810VertexPtr;
+typedef union i810_vertex_t i810Vertex;
+typedef union i810_vertex_t *i810VertexPtr;
 
 struct i810_vertex_buffer_t {
    GLvector1ui clipped_elements;
@@ -79,7 +78,7 @@ struct i810_vertex_buffer_t {
 typedef struct i810_vertex_buffer_t *i810VertexBufferPtr;
 
 
-#define I810_CONTEXT(ctx)    ((i810ContextPtr)(((XSMesaContext)(ctx)->DriverCtx)->hw_ctx))
+#define I810_CONTEXT(ctx)    ((i810ContextPtr)(ctx->DriverCtx))
 #define I810_DRIVER_DATA(vb) ((i810VertexBufferPtr)((vb)->driver_data))
 
 
@@ -97,7 +96,7 @@ extern void i810PrintSetupFlags(char *msg, GLuint flags );
 extern void i810DDDoRasterSetup( struct vertex_buffer *VB );
 extern void i810DDPartialRasterSetup( struct vertex_buffer *VB );
 extern void i810DDCheckPartialRasterSetup( GLcontext *ctx, 
-					  struct gl_pipeline_stage *d );
+					   struct gl_pipeline_stage *d );
 
 extern void i810DDViewport( GLcontext *ctx, 
 			   GLint x, GLint y, 
diff --git a/xc/lib/GL/mesa/src/drv/mga/Imakefile b/xc/lib/GL/mesa/src/drv/mga/Imakefile
index e4e39b747..1ca3bde8a 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
@@ -12,6 +14,7 @@ ALLOC_DEFINES = -DMALLOC_0_RETURNS_NULL
 #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 \
@@ -24,23 +27,244 @@ MESA_INCLUDES = -I. -I.. -I../../include
 
 
       DEFINES = $(ALLOC_DEFINES) $(DRI_DEFINES)
-     INCLUDES = -I$(XLIBSRC) -I$(EXTINCSRC) $(MESA_INCLUDES) $(DRI_INCLUDES) \
-		-I/usr/include/glide
-      DRISRCS = mgaclear.c mgacnvtex.c mgadd.c mgadepth.c	\
+     INCLUDES = -I$(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					\
 		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	\
+      MGAOBJS = mgaclear.o mgacnvtex.o mgadd.o \
 		mgafastpath.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 = ../../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.o \
+		../../light.c \
+		../../lines.c \
+		../../logic.c \
+		../../masking.c \
+		../../matrix.c \
+		../../mem.c \
+		../../mmath.c \
+		../../pb.c \
+		../../pipeline.c \
+		../../pixel.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 \
+		../../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 = ../../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 \
+		../../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 \
+		../../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
+
+      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)
 
-	 SRCS = $(DRISRCS) 
-	 OBJS = $(DRIOBJS) 
+REQUIREDLIBS += -lm
 
 #if !GlxUseBuiltInDRIDriver
 #undef DoNormalLib NormalLibGlx
@@ -61,7 +285,7 @@ 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 496c29a05..996dc3819 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mga_xmesa.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mga_xmesa.c
@@ -27,7 +27,7 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 /*
  * Authors:
- *   Daryll Strauss <daryll@precisioninsight.com>
+ *   Keith Whitwell <keithw@precisioninsight.com>
  *
  */
 
@@ -36,6 +36,7 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include <X11/Xlibint.h>
 #include <stdio.h>
 
+#include "drm.h"
 #include "mga_xmesa.h"
 #include "context.h"
 #include "vbxform.h"
@@ -49,20 +50,25 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "mgadepth.h"
 #include "mgatris.h"
 #include "mgapipeline.h"
+#include "mgabuffers.h"
 
 #include "xf86dri.h"
-#include "mga_dri.h"
-#include "mga_drm_public.h"
 #include "mga_xmesa.h"
 
+#include "mga_dri.h"
+
 
 
 
 
 #ifndef MGA_DEBUG
 int MGA_DEBUG = (0
-/*  		  | MGA_DEBUG_ALWAYS_SYNC */
-/*  		  | MGA_DEBUG_VERBOSE_MSG */
+/*  		  | DEBUG_ALWAYS_SYNC */
+/*  		  | DEBUG_VERBOSE_MSG */
+/*   		  | DEBUG_VERBOSE_LRU */
+/*  		  | DEBUG_VERBOSE_DRI */
+/*  		  | DEBUG_VERBOSE_IOCTL */
+/*   		  | DEBUG_VERBOSE_2D */
 		  );
 #endif
 
@@ -107,7 +113,10 @@ static int count_bits(unsigned int n)
 GLboolean XMesaInitDriver(__DRIscreenPrivate *sPriv)
 {
    mgaScreenPrivate *mgaScreen;
-   MGADRIPtr         gDRIPriv = (MGADRIPtr)sPriv->pDevPriv;
+   MGADRIPtr         serverInfo = (MGADRIPtr)sPriv->pDevPriv;
+
+
+   fprintf(stderr, "XMesaInitDriver\n");
 
    /* Allocate the private area */
    mgaScreen = (mgaScreenPrivate *)Xmalloc(sizeof(mgaScreenPrivate));
@@ -116,20 +125,68 @@ GLboolean XMesaInitDriver(__DRIscreenPrivate *sPriv)
    mgaScreen->sPriv = sPriv;
    sPriv->private = (void *)mgaScreen;
 
-   mgaScreen->chipset=gDRIPriv->chipset;
-   mgaScreen->width=gDRIPriv->width;
-   mgaScreen->height=gDRIPriv->height;
-   mgaScreen->mem=gDRIPriv->mem;
-   mgaScreen->cpp=gDRIPriv->cpp;
-   mgaScreen->frontPitch=gDRIPriv->frontPitch;
-   mgaScreen->frontOffset=gDRIPriv->frontOffset;
-   mgaScreen->backOffset=gDRIPriv->backOffset; 
-   mgaScreen->backPitch = gDRIPriv->backPitch;
-   mgaScreen->depthOffset=gDRIPriv->depthOffset;
-   mgaScreen->depthPitch = gDRIPriv->depthPitch;
-   mgaScreen->textureOffset=gDRIPriv->textureOffset;
-   mgaScreen->textureSize=gDRIPriv->textureSize;
-   mgaScreen->logTextureGranularity = gDRIPriv->logTextureGranularity;
+   fprintf(stderr, "serverInfo->chipset: %d\n", serverInfo->chipset);
+
+   if (serverInfo->chipset != MGA_CARD_TYPE_G200 &&
+       serverInfo->chipset != MGA_CARD_TYPE_G400)
+	   return GL_FALSE;
+   
+   mgaScreen->chipset = serverInfo->chipset;
+   mgaScreen->width = serverInfo->width;
+   mgaScreen->height = serverInfo->height;
+   mgaScreen->mem = serverInfo->mem;
+   mgaScreen->cpp = serverInfo->cpp;
+   mgaScreen->frontPitch = serverInfo->frontPitch;
+   mgaScreen->frontOffset = serverInfo->frontOffset;
+   mgaScreen->backOffset = serverInfo->backOffset; 
+   mgaScreen->backPitch  =  serverInfo->backPitch;
+   mgaScreen->depthOffset = serverInfo->depthOffset;
+   mgaScreen->depthPitch  =  serverInfo->depthPitch;
+
+
+   mgaScreen->agp_tex.handle = serverInfo->agp;
+   mgaScreen->agp_tex.size = serverInfo->agpSize;
+
+   if (drmMap(sPriv->fd, 
+	      mgaScreen->agp_tex.handle, 
+	      mgaScreen->agp_tex.size, 
+	      (drmAddress *)&mgaScreen->agp_tex.map) != 0) 
+   {
+      Xfree(mgaScreen);
+      return GL_FALSE;
+   }
+
+
+
+   mgaScreen->textureOffset[MGA_CARD_HEAP] = serverInfo->textureOffset;
+   mgaScreen->textureOffset[MGA_AGP_HEAP] = (mgaScreen->agp_tex.handle |
+					     PDEA_pagpxfer_enable | 1);
+
+   fprintf(stderr, "CARD texture size %x, granul %d --> %x\n", 
+	   serverInfo->textureSize, serverInfo->logTextureGranularity,
+	   1<<serverInfo->logTextureGranularity);
+
+   mgaScreen->textureSize[MGA_CARD_HEAP] = serverInfo->textureSize;
+   mgaScreen->textureSize[MGA_AGP_HEAP] = serverInfo->agpTextureSize;
+
+   mgaScreen->logTextureGranularity[MGA_CARD_HEAP] = 
+      serverInfo->logTextureGranularity;
+   mgaScreen->logTextureGranularity[MGA_AGP_HEAP] = 
+      serverInfo->logAgpTextureGranularity;
+
+   mgaScreen->texVirtual[MGA_CARD_HEAP] = (mgaScreen->sPriv->pFB + 
+					   mgaScreen->textureOffset[0]);
+   mgaScreen->texVirtual[MGA_AGP_HEAP] = mgaScreen->agp_tex.map;
+
+   mgaScreen->mAccess = serverInfo->mAccess;
+
+
+   fprintf(stderr, "\n\n\nbackOffset: %x pitch %x\n", 
+	   mgaScreen->backOffset,
+	   mgaScreen->backPitch);
+
+   
+   mgaScreen->Attrib = MGA_PF_565;
 
 
    mgaScreen->bufs = drmMapBufs(sPriv->fd);
@@ -169,6 +226,8 @@ XMesaVisual XMesaCreateVisual(XMesaDisplay *display,
 {
    XMesaVisual v;
 
+   fprintf(stderr, "XMesaCreateVisual\n");
+
    /* Only RGB visuals are supported on the MGA boards */
    if (!rgb_flag) return 0;
 
@@ -183,28 +242,19 @@ XMesaVisual XMesaCreateVisual(XMesaDisplay *display,
    memcpy(v->visinfo, visinfo, sizeof(*visinfo));
 
    v->display = display;
-   v->level = level;
-  
-   v->gl_visual = (GLvisual *)Xmalloc(sizeof(GLvisual));
+   v->level = level;  
+   v->gl_visual = gl_create_visual(rgb_flag, GL_FALSE, db_flag, stereo_flag,
+				   depth_size, stencil_size, accum_size, 0,
+				   count_bits(visinfo->red_mask),
+				   count_bits(visinfo->green_mask),
+				   count_bits(visinfo->blue_mask),
+				   GL_FALSE);
    if (!v->gl_visual) {
       Xfree(v->visinfo);
-      XFree(v);
-      return 0;
+      Xfree(v);
+      return NULL;
    }
 
-   v->gl_visual->RGBAflag   = rgb_flag;
-   v->gl_visual->DBflag     = db_flag;
-   v->gl_visual->StereoFlag = stereo_flag;
-
-   v->gl_visual->RedBits   = count_bits(visinfo->red_mask);
-   v->gl_visual->GreenBits = count_bits(visinfo->green_mask);
-   v->gl_visual->BlueBits  = count_bits(visinfo->blue_mask);
-   v->gl_visual->AlphaBits = 0; /* Not currently supported */
-
-   v->gl_visual->AccumBits   = accum_size;
-   v->gl_visual->DepthBits   = depth_size;
-   v->gl_visual->StencilBits = stencil_size;
-
    return v;
 }
 
@@ -218,6 +268,7 @@ void XMesaDestroyVisual(XMesaVisual v)
 XMesaContext XMesaCreateContext(XMesaVisual v, XMesaContext share_list,
 				__DRIcontextPrivate *driContextPriv)
 {
+   int i;
    GLcontext *ctx;
    XMesaContext c;
    mgaContextPtr mmesa;
@@ -227,12 +278,15 @@ XMesaContext XMesaCreateContext(XMesaVisual v, XMesaContext share_list,
 					      sizeof(XF86DRISAREARec));
    GLcontext *shareCtx = 0;
 
+   fprintf(stderr, "XMesaCreateContext\n");
+
+
    c = (XMesaContext)Xmalloc(sizeof(struct xmesa_context));
    if (!c) {
       return 0;
    }
 
-   mmesa = (mgaContextPtr)Xmalloc(sizeof(mgaContext));
+   mmesa = (mgaContextPtr)Xcalloc(sizeof(mgaContext), 1);
    if (!mmesa) {
       Xfree(c);
       return 0;
@@ -250,8 +304,6 @@ XMesaContext XMesaCreateContext(XMesaVisual v, XMesaContext share_list,
    ctx = mmesa->glCtx = gl_create_context(v->gl_visual, shareCtx, 
 					  (void*)mmesa, GL_TRUE);
 
-   /* Dri stuff
-    */
    mmesa->display = v->display;
    mmesa->hHWContext = driContextPriv->hHWContext;
    mmesa->driFd = sPriv->fd;
@@ -261,40 +313,45 @@ XMesaContext XMesaCreateContext(XMesaVisual v, XMesaContext share_list,
    mmesa->driScreen = sPriv;
    mmesa->sarea = saPriv;
 
-   mmesa->glBuffer=gl_create_framebuffer(v->gl_visual);
+   mmesa->glBuffer=gl_create_framebuffer(v->gl_visual,
+					 GL_FALSE,  /* software depth buffer? */
+					 v->gl_visual->StencilBits > 0,
+					 v->gl_visual->AccumBits > 0,
+					 v->gl_visual->AlphaBits > 0);
+
 
+   make_empty_list(&mmesa->SwappedOut);
 
-   mmesa->needClip=1;
+   mmesa->lastTexHeap = mgaScreen->texVirtual[MGA_AGP_HEAP] ? 2 : 1;
 
-   mmesa->texHeap = mmInit( 0, mgaScreen->textureSize );
+   for (i = 0 ; i < mmesa->lastTexHeap ; i++) {
+      mmesa->texHeap[i] = mmInit( 0, mgaScreen->textureSize[i]);
+      make_empty_list(&mmesa->TexObjList[i]);
+   }
 
-   /* Utah stuff
-    */
    mmesa->renderindex = -1;		/* impossible value */
    mmesa->new_state = ~0;
    mmesa->dirty = ~0;
-
-   mmesa->warp_pipe = 0;
-
-
-   make_empty_list(&mmesa->SwappedOut);
-   make_empty_list(&mmesa->TexObjList);
-   
+   mmesa->warp_pipe = 0;   
    mmesa->CurrentTexObj[0] = 0;
    mmesa->CurrentTexObj[1] = 0;
    
+   mmesa->texAge[0] = 0;
+   mmesa->texAge[1] = 0;
+   
+
    mgaDDExtensionsInit( ctx );
 
    mgaDDInitStateFuncs( ctx );
    mgaDDInitTextureFuncs( ctx );
    mgaDDInitSpanFuncs( ctx );
-   mgaDDInitDepthFuncs( ctx );
    mgaDDInitDriverFuncs( ctx );
    mgaDDInitIoctlFuncs( ctx );
 
 
    ctx->Driver.TriangleCaps = (DD_TRI_CULL|
 			       DD_TRI_LIGHT_TWOSIDE|
+			       DD_TRI_STIPPLE|
 			       DD_TRI_OFFSET);
 
    /* Ask mesa to clip fog coordinates for us.
@@ -312,6 +369,9 @@ XMesaContext XMesaCreateContext(XMesaVisual v, XMesaContext share_list,
 	 mgaDDRegisterPipelineStages(ctx->PipelineStage,
 				      ctx->PipelineStage,
 				      ctx->NrPipelineStages);
+
+   mgaInitState( mmesa );
+
    return c;
 }
 
@@ -364,59 +424,12 @@ void XMesaSwapBuffers(XMesaBuffer bogus_value_do_not_use)
 
 
 
-
-void mgaXMesaSetFrontClipRects( mgaContextPtr mmesa )
+GLboolean XMesaUnbindContext(XMesaContext c)
 {
-   __DRIdrawablePrivate *dPriv = mmesa->driDrawable;
-
-   mmesa->numClipRects = dPriv->numClipRects;
-   mmesa->pClipRects = dPriv->pClipRects;
-   mmesa->drawX = dPriv->x;
-   mmesa->drawY = dPriv->y;
-
-   mmesa->drawOffset = mmesa->mgaScreen->frontOffset;
-}
-
-
-void mgaXMesaSetBackClipRects( mgaContextPtr mmesa )
-{
-   __DRIdrawablePrivate *dPriv = mmesa->driDrawable;
-
-   if (dPriv->numAuxClipRects == 0) 
-   {
-      mmesa->numClipRects = dPriv->numClipRects;
-      mmesa->pClipRects = dPriv->pClipRects;
-      mmesa->drawX = dPriv->x;
-      mmesa->drawY = dPriv->y;
-   } else {
-      mmesa->numClipRects = dPriv->numAuxClipRects;
-      mmesa->pClipRects = dPriv->pAuxClipRects;
-      mmesa->drawX = dPriv->auxX;
-      mmesa->drawY = dPriv->auxY;
-   }
-
-   mmesa->drawOffset = mmesa->mgaScreen->backOffset;
-}
-
-
-static void mgaXMesaWindowMoved( mgaContextPtr mmesa ) 
-{
-   /* Clear any contaminated CVA data.
-    */
-   mmesa->setupdone = 0;
-
-   switch (mmesa->glCtx->Color.DriverDrawBuffer) {
-   case GL_FRONT_LEFT:
-      mgaXMesaSetFrontClipRects( mmesa );
-      break;
-   case GL_BACK_LEFT:
-      mgaXMesaSetBackClipRects( mmesa );
-      break;
-   default:
-      fprintf(stderr, "fallback buffer\n");
-      break;
-   }
+   if (c->private)
+      ((mgaContextPtr)c->private)->dirty = ~0;
 
+   return GL_TRUE;
 }
 
 
@@ -439,9 +452,9 @@ GLboolean XMesaMakeCurrent(XMesaContext c, XMesaBuffer b)
 
       gl_make_current(mgaCtx->glCtx, mgaCtx->glBuffer);
 
-      mgaXMesaWindowMoved( mgaCtx );
       mgaCtx->driDrawable = dPriv;
       mgaCtx->dirty = ~0;
+      mgaCtx->dirty_cliprects = (MGA_FRONT|MGA_BACK);
 
       if (!mgaCtx->glCtx->Viewport.Width)
 	 gl_Viewport(mgaCtx->glCtx, 0, 0, dPriv->w, dPriv->h);
@@ -455,54 +468,36 @@ GLboolean XMesaMakeCurrent(XMesaContext c, XMesaBuffer b)
 }
 
 
-void mgaXMesaUpdateState( mgaContextPtr mmesa ) 
+void mgaGetLock( mgaContextPtr mmesa, GLuint flags ) 
 {
    __DRIdrawablePrivate *dPriv = mmesa->driDrawable;
-   __DRIscreenPrivate *sPriv = mmesa->driScreen;
    drm_mga_sarea_t *sarea = mmesa->sarea;
-
    int me = mmesa->hHWContext;
-   int stamp = dPriv->lastStamp; 
+   int i;
 
-   /* If the window moved, may need to set a new cliprect now.
-    *
-    * NOTE: This releases and regains the hw lock, so all state
-    * checking must be done *after* this call:
-    */
-   XMESA_VALIDATE_DRAWABLE_INFO(mmesa->display, sPriv, dPriv);		
+   drmGetLock(mmesa->driFd, mmesa->hHWContext, flags);	
 
-   if (sarea->ctxOwner != me) {
-      mmesa->dirty |= MGA_UPLOAD_CTX;
+   if (*(dPriv->pStamp) != dPriv->lastStamp) { 
+      mmesa->setupdone = 0;
+      mmesa->dirty_cliprects = (MGA_FRONT|MGA_BACK);
+      mgaUpdateRects( mmesa, (MGA_FRONT|MGA_BACK) );
    }
 
-   if (sarea->texAge != mmesa->texAge) {
-      int sz = 1 << (mmesa->mgaScreen->logTextureGranularity);
-      int idx, nr = 0;
-
-      /* Have to go right round from the back to ensure stuff ends up
-       * LRU in our local list...
-       */
-      for (idx = sarea->texList[MGA_NR_TEX_REGIONS].prev ; 
-	   idx != MGA_NR_TEX_REGIONS && nr < MGA_NR_TEX_REGIONS ; 
-	   idx = sarea->texList[idx].prev, nr++)
-      {
-	 if (sarea->texList[idx].age > mmesa->texAge)
-	    mgaTexturesGone(mmesa, idx * sz, sz, 1);
-      }
-
-      if (nr == MGA_NR_TEX_REGIONS) {
-	 mgaTexturesGone(mmesa, 0, mmesa->mgaScreen->textureSize, 0);
-	 mgaResetGlobalLRU( mmesa );
-      }
-
-      mmesa->texAge = sarea->texAge;
-      mmesa->dirty |= MGA_UPLOAD_TEX0IMAGE | MGA_UPLOAD_TEX1IMAGE;
+   mmesa->dirty |= MGA_UPLOAD_CTX | MGA_UPLOAD_CLIPRECTS;
+
+    mmesa->sarea->dirty |= MGA_UPLOAD_CTX; 
+
+   if (sarea->ctxOwner != me) {
+      mmesa->dirty |= (MGA_UPLOAD_CTX | MGA_UPLOAD_TEX0 | 
+		       MGA_UPLOAD_TEX1 | MGA_UPLOAD_PIPE);
+      sarea->ctxOwner=me;
    }
 
-   if (dPriv->lastStamp != stamp) 
-      mgaXMesaWindowMoved( mmesa );
+   for (i = 0 ; i < mmesa->lastTexHeap ; i++)
+      if (sarea->texAge[i] != mmesa->texAge[i])
+	 mgaAgeTextures( mmesa, i );
 
-   sarea->ctxOwner=me;
+   sarea->last_quiescent = -1;	/* just kill it for now */
 }
 
 
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..1379f48d9 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).
  *
  */
 
@@ -43,6 +42,13 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #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 {
 
@@ -54,6 +60,7 @@ typedef struct {
    int cpp;			/* for front and back buffers */
 
    int Attrib;
+   int mAccess;
 
    int frontOffset;
    int frontPitch;
@@ -64,43 +71,54 @@ typedef struct {
    int depthPitch;
    int depthCpp;
 
-   int textureOffset;
-   int textureSize;
-   int logTextureGranularity;
+   int textureOffset[MGA_NR_TEX_HEAPS];
+   int textureSize[MGA_NR_TEX_HEAPS];
+   int logTextureGranularity[MGA_NR_TEX_HEAPS];
+   char *texVirtual[MGA_NR_TEX_HEAPS];
+
 
    __DRIscreenPrivate *sPriv;
    drmBufMapPtr  bufs;
 
+   mgaRegion agp_tex;
+
 } 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 +127,8 @@ 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/mgabuffers.c b/xc/lib/GL/mesa/src/drv/mga/mgabuffers.c
new file mode 100644
index 000000000..276c5df85
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/mga/mgabuffers.c
@@ -0,0 +1,315 @@
+/**************************************************************************
+
+Copyright 1998-1999 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, 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 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:
+ *   Keith Whitwell <keithw@precisioninsight.com>
+ *
+ * $PI: $
+ */
+
+#include <stdio.h>
+#include "mgalib.h"
+#include "mgabuffers.h"
+#include "mgastate.h"
+
+static void mgaXMesaSetFrontClipRects( mgaContextPtr mmesa )
+{
+   __DRIdrawablePrivate *driDrawable = mmesa->driDrawable;
+
+   mmesa->numClipRects = driDrawable->numClipRects;
+   mmesa->pClipRects = driDrawable->pClipRects;
+   mmesa->drawX = driDrawable->x;
+   mmesa->drawY = driDrawable->y;
+
+   mmesa->drawOffset = mmesa->mgaScreen->frontOffset;
+   mmesa->Setup[MGA_CTXREG_DSTORG] = mmesa->drawOffset;
+   mmesa->dirty |= MGA_UPLOAD_CTX;
+   mmesa->dirty |= MGA_UPLOAD_CLIPRECTS;
+}
+
+
+static void mgaXMesaSetBackClipRects( mgaContextPtr mmesa )
+{
+   __DRIdrawablePrivate *driDrawable = mmesa->driDrawable;
+
+   if (driDrawable->numBackClipRects == 0) 
+   {
+      mmesa->numClipRects = driDrawable->numClipRects;
+      mmesa->pClipRects = driDrawable->pClipRects;
+      mmesa->drawX = driDrawable->x;
+      mmesa->drawY = driDrawable->y;
+   } else {
+      mmesa->numClipRects = driDrawable->numBackClipRects;
+      mmesa->pClipRects = driDrawable->pBackClipRects;
+      mmesa->drawX = driDrawable->backX;
+      mmesa->drawY = driDrawable->backY;
+   }
+
+   mmesa->drawOffset = mmesa->mgaScreen->backOffset;
+   mmesa->Setup[MGA_CTXREG_DSTORG] = mmesa->drawOffset;
+
+   mmesa->dirty |= MGA_UPLOAD_CTX;
+   mmesa->dirty |= MGA_UPLOAD_CLIPRECTS;
+}
+
+
+
+static void mgaUpdateRectsFromSarea( mgaContextPtr mmesa )
+{
+   __DRIdrawablePrivate *driDrawable = mmesa->driDrawable;
+   __DRIscreenPrivate *driScreen = mmesa->driScreen;
+   drm_mga_sarea_t *sarea = mmesa->sarea;
+   int i = 0, top = 0;
+
+
+   if (sarea->exported_buffers & MGA_BACK) {
+      XF86DRIClipRectPtr boxes = 
+	 (XF86DRIClipRectPtr)malloc( sarea->exported_nback * sizeof(*boxes) );
+
+      if (driDrawable->pBackClipRects)
+	 free(driDrawable->pBackClipRects);
+
+      driDrawable->numBackClipRects = sarea->exported_nback;
+      driDrawable->pBackClipRects = boxes;
+
+      top = sarea->exported_nback;
+      for (i = 0 ; i < top ; i++)
+	 boxes[i] = *(XF86DRIClipRectPtr)&(sarea->exported_boxes[i]);
+   }
+
+
+   if (sarea->exported_buffers & MGA_FRONT) 
+   {
+      int start = top;
+      XF86DRIClipRectPtr boxes = 
+	 (XF86DRIClipRectPtr)malloc( sarea->exported_nfront * sizeof(*boxes) );
+
+      if (driDrawable->pClipRects)
+	 free(driDrawable->pClipRects);
+
+      driDrawable->numClipRects = sarea->exported_nfront;
+      driDrawable->pClipRects = boxes;
+
+      top += sarea->exported_nfront;
+      for ( ; i < top ; i++)
+	 boxes[i-start] = *(XF86DRIClipRectPtr)&(sarea->exported_boxes[i]);
+
+   } 
+
+
+
+   driDrawable->index = sarea->exported_index;
+   driDrawable->lastStamp = sarea->exported_stamp;
+   driDrawable->x = sarea->exported_front_x;
+   driDrawable->y = sarea->exported_front_y;
+   driDrawable->backX = sarea->exported_back_x;
+   driDrawable->backY = sarea->exported_back_y;
+   driDrawable->w = sarea->exported_w;
+   driDrawable->h = sarea->exported_h;
+   driDrawable->pStamp = 
+      &(driScreen->pSAREA->drawableTable[driDrawable->index].stamp);
+
+   mmesa->dirty_cliprects = (MGA_FRONT|MGA_BACK) & ~(sarea->exported_buffers);
+}
+
+
+
+static void printSareaRects( mgaContextPtr mmesa )
+{
+   __DRIscreenPrivate *driScreen = mmesa->driScreen;
+   drm_mga_sarea_t *sarea = mmesa->sarea;
+   int i;
+
+   fprintf(stderr, "sarea->exported: %d\n", sarea->exported_drawable);
+   fprintf(stderr, "sarea->exported_index: %d\n", sarea->exported_index);
+   fprintf(stderr, "sarea->exported_stamp: %d\n", sarea->exported_stamp);
+   fprintf(stderr, "sarea->exported_front_x: %d\n", sarea->exported_front_x);
+   fprintf(stderr, "sarea->exported_front_y: %d\n", sarea->exported_front_y);
+   fprintf(stderr, "sarea->exported_back_x: %d\n", sarea->exported_back_x);
+   fprintf(stderr, "sarea->exported_back_y: %d\n", sarea->exported_back_y);
+   fprintf(stderr, "sarea->exported_w: %d\n", sarea->exported_w);
+   fprintf(stderr, "sarea->exported_h: %d\n", sarea->exported_h);
+   fprintf(stderr, "sarea->exported_buffers: %d\n", sarea->exported_buffers);
+   fprintf(stderr, "sarea->exported_nfront: %d\n", sarea->exported_nfront);
+   fprintf(stderr, "sarea->exported_nback: %d\n", sarea->exported_nback);
+
+   i = 0;
+   if (sarea->exported_buffers & MGA_BACK)
+      for ( ; i < sarea->exported_nback ; i++) 
+	 fprintf(stderr, "back %d: %d,%d-%d,%d\n", i,
+		 sarea->exported_boxes[i].x1, sarea->exported_boxes[i].y1,
+		 sarea->exported_boxes[i].x2, sarea->exported_boxes[i].y2);
+
+   if (sarea->exported_buffers & MGA_FRONT) {
+      int start = i;
+      int top = i + sarea->exported_nfront;
+      for ( ; i < top ; i++) 
+	 fprintf(stderr, "front %d: %d,%d-%d,%d\n", 
+		 i - start,
+		 sarea->exported_boxes[i].x1, sarea->exported_boxes[i].y1,
+		 sarea->exported_boxes[i].x2, sarea->exported_boxes[i].y2);
+   }
+
+   fprintf(stderr, "drawableTable[%d].stamp: %d\n", 
+	   sarea->exported_index,
+	   driScreen->pSAREA->drawableTable[sarea->exported_index].stamp);
+}
+
+static void printMmesaRects( mgaContextPtr mmesa )
+{
+   __DRIscreenPrivate *driScreen = mmesa->driScreen;
+   __DRIdrawablePrivate *driDrawable = mmesa->driDrawable;
+   int nr = mmesa->numClipRects;
+   int i;
+
+   fprintf(stderr, "driDrawable->draw: %ld\n", driDrawable->draw);
+   fprintf(stderr, "driDrawable->index: %d\n", driDrawable->index);
+   fprintf(stderr, "driDrawable->lastStamp: %d\n", driDrawable->lastStamp);
+   fprintf(stderr, "mmesa->drawX: %d\n", mmesa->drawX);
+   fprintf(stderr, "mmesa->drawY: %d\n", mmesa->drawY);
+   fprintf(stderr, "driDrawable->w: %d\n", driDrawable->w);
+   fprintf(stderr, "driDrawable->h: %d\n", driDrawable->h);
+
+   for (i = 0 ; i < nr ; i++) 
+      fprintf(stderr, "box %d: %d,%d-%d,%d\n", i,
+	      mmesa->pClipRects[i].x1, mmesa->pClipRects[i].y1,
+	      mmesa->pClipRects[i].x2, mmesa->pClipRects[i].y2);
+
+   fprintf(stderr, "mmesa->draw_buffer: %d\n", mmesa->draw_buffer);
+   fprintf(stderr, "drawableTable[%d].stamp: %d\n", 
+	   driDrawable->index,
+	   driScreen->pSAREA->drawableTable[driDrawable->index].stamp);
+}
+
+
+
+void mgaUpdateRects( mgaContextPtr mmesa, GLuint buffers )
+{
+   __DRIdrawablePrivate *driDrawable = mmesa->driDrawable;
+   drm_mga_sarea_t *sarea = mmesa->sarea;
+
+   /* The MGA X driver will try to push one set of cliprects (back or
+    * front, the last active) for one drawable (the last used) into
+    * the sarea.  See if that window was ours, else retrieve both back
+    * and front rects from the X server via a protocol request.  
+    *
+    * If there isn't room for the cliprects in the sarea, the X server
+    * clears the drawable value to indicate failure.  
+    */
+   if (0) printSareaRects(mmesa);
+
+   if (sarea->exported_drawable == driDrawable->draw && 
+       (sarea->exported_buffers & buffers) == buffers)
+   {
+      mgaUpdateRectsFromSarea( mmesa );
+   } 
+   else 
+   {
+      if(MGA_DEBUG & DEBUG_VERBOSE_MSG)
+		fprintf(stderr, "^");
+      driDrawable->lastStamp = 0;
+
+      XMESA_VALIDATE_DRAWABLE_INFO(mmesa->display, 
+				   mmesa->driScreen, 
+				   driDrawable); 
+      mmesa->dirty_cliprects = 0;	
+   }
+
+   if (mmesa->draw_buffer == MGA_FRONT)
+      mgaXMesaSetFrontClipRects( mmesa );
+   else
+      mgaXMesaSetBackClipRects( mmesa );
+
+
+   if (0) printMmesaRects(mmesa);
+
+   sarea->req_drawable = driDrawable->draw;
+   sarea->req_draw_buffer = mmesa->draw_buffer;
+
+   
+   mgaUpdateClipping( mmesa->glCtx );
+   
+   mmesa->dirty |= MGA_UPLOAD_CLIPRECTS;
+}
+
+
+
+GLboolean mgaDDSetDrawBuffer(GLcontext *ctx, GLenum mode )
+{
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
+
+   FLUSH_BATCH( MGA_CONTEXT(ctx) );
+
+   mmesa->Fallback &= ~MGA_FALLBACK_BUFFER;
+
+   if (mode == GL_FRONT_LEFT) 
+   {
+      mmesa->drawOffset = mmesa->mgaScreen->frontOffset;
+      mmesa->Setup[MGA_CTXREG_DSTORG] = mmesa->mgaScreen->frontOffset;
+      mmesa->dirty |= MGA_UPLOAD_CTX;
+      mmesa->draw_buffer = MGA_FRONT;
+      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->draw_buffer = MGA_BACK;
+      mmesa->dirty |= MGA_UPLOAD_CTX;
+      mgaXMesaSetBackClipRects( mmesa );
+      return GL_TRUE;
+   }
+   else 
+   {
+      mmesa->Fallback |= MGA_FALLBACK_BUFFER;
+      return GL_FALSE;
+   }
+}
+
+GLboolean mgaDDSetReadBuffer(GLcontext *ctx, GLenum mode )
+{
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
+
+   if (mode == GL_FRONT_LEFT) 
+   {
+      mmesa->readOffset = mmesa->mgaScreen->frontOffset;
+      mmesa->read_buffer = MGA_FRONT;
+      return GL_TRUE;
+   } 
+   else if (mode == GL_BACK_LEFT) 
+   {
+      mmesa->readOffset = mmesa->mgaScreen->backOffset;
+      mmesa->read_buffer = MGA_BACK;
+      return GL_TRUE;
+   }
+   else 
+   {
+      return GL_FALSE;
+   }
+}
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgabuffers.h b/xc/lib/GL/mesa/src/drv/mga/mgabuffers.h
new file mode 100644
index 000000000..9345fd9c4
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/mga/mgabuffers.h
@@ -0,0 +1,9 @@
+#ifndef MGA_BUFFERS_H
+#define MGA_BUFFERS_H
+
+GLboolean mgaDDSetReadBuffer(GLcontext *ctx, GLenum mode );
+GLboolean mgaDDSetDrawBuffer(GLcontext *ctx, GLenum mode );
+
+void mgaUpdateRects( mgaContextPtr mmesa, GLuint buffers );
+
+#endif
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgacnvtex.c b/xc/lib/GL/mesa/src/drv/mga/mgacnvtex.c
index ae8d95037..428b65d06 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgacnvtex.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgacnvtex.c
@@ -46,205 +46,206 @@
  */
 void mgaConvertTexture( mgaUI32 *destPtr, int texelBytes, 
 			struct gl_texture_image *image,
-			int x, int y, int width, int height ) {
+			int x, int y, int width, int height ) 
+{
 	register int		i, j;
 	mgaUI8		*src;
 	int stride;
 	
 	/* FIXME: g400 luminance_alpha internal format */
 	switch (texelBytes) {
-	  case 1:
-	    switch (image->Format) {
-		   case GL_COLOR_INDEX:
-		   case GL_INTENSITY:
-		   case GL_LUMINANCE:
-		   case GL_ALPHA:
- 	        src = (mgaUI8 *)image->Data + ( y * image->Width + x );
-			  stride = (image->Width - width);
-		     for ( i = height ; i ; i-- ) {
-			    for ( j = width >> 2  ; j ; j-- ) {
+	case 1:
+		switch (image->Format) {
+		case GL_COLOR_INDEX:
+		case GL_INTENSITY:
+		case GL_LUMINANCE:
+		case GL_ALPHA:
+			src = (mgaUI8 *)image->Data + ( y * image->Width + x );
+			stride = (image->Width - width);
+			for ( i = height ; i ; i-- ) {
+				for ( j = width >> 2  ; j ; j-- ) {
 			
-				   *destPtr++ = src[0] | ( src[1] << 8 ) | ( src[2] << 16 ) | ( src[3] << 24 );
-				   src += 4;
-			    }
-				 src += stride;
-		     }
-			  break;
-			default:
-			  goto format_error;
-		 }
-		 break;
-	  case 2:
-	    switch (image->Format) {
-			case GL_RGB:
- 	        src = (mgaUI8 *)image->Data + ( y * image->Width + x ) * 3;
-			  stride = (image->Width - width) * 3;
-		     for ( i = height ; i ; i-- ) {
-			    for ( j = width >> 1  ; j ; j-- ) {
+					*destPtr++ = src[0] | ( src[1] << 8 ) | ( src[2] << 16 ) | ( src[3] << 24 );
+					src += 4;
+				}
+				src += stride;
+			}
+			break;
+		default:
+			goto format_error;
+		}
+		break;
+	case 2:
+		switch (image->Format) {
+		case GL_RGB:
+			src = (mgaUI8 *)image->Data + ( y * image->Width + x ) * 3;
+			stride = (image->Width - width) * 3;
+			for ( i = height ; i ; i-- ) {
+				for ( j = width >> 1  ; j ; j-- ) {
 			
-				   *destPtr++ = MGAPACKCOLOR565(src[0],src[1],src[2]) |  
-					   ( MGAPACKCOLOR565(src[3],src[4],src[5]) << 16 );
-				   src += 6;
-			    }
-				 src += stride;
-		     }
-			  break;
-			case GL_RGBA:
- 	        src = (mgaUI8 *)image->Data + ( y * image->Width + x ) * 4;
-			  stride = (image->Width - width) * 4;
-		     for ( i = height ; i ; i-- ) {
-			    for ( j = width >> 1  ; j ; j-- ) {
+					*destPtr++ = MGAPACKCOLOR565(src[0],src[1],src[2]) |  
+						( MGAPACKCOLOR565(src[3],src[4],src[5]) << 16 );
+					src += 6;
+				}
+				src += stride;
+			}
+			break;
+		case GL_RGBA:
+			src = (mgaUI8 *)image->Data + ( y * image->Width + x ) * 4;
+			stride = (image->Width - width) * 4;
+			for ( i = height ; i ; i-- ) {
+				for ( j = width >> 1  ; j ; j-- ) {
 			
-				   *destPtr++ = MGAPACKCOLOR4444(src[0],src[1],src[2],src[3]) |  
-				   	( MGAPACKCOLOR4444(src[4],src[5],src[6],src[7]) << 16 );
-				   src += 8;
-			    }
-				 src += stride;
-		     }
-			  break;
-			case GL_LUMINANCE:
- 	        src = (mgaUI8 *)image->Data + ( y * image->Width + x );
-			  stride = (image->Width - width);
-		     for ( i = height ; i ; i-- ) {
-			    for ( j = width >> 1  ; j ; j-- ) {
-				   /* FIXME: should probably use 555 texture to get true grey */
-				   *destPtr++ = MGAPACKCOLOR565(src[0],src[0],src[0]) |  
-					   ( MGAPACKCOLOR565(src[1],src[1],src[1]) << 16 );
-				   src += 2;
-			    }
-				 src += stride;
-		     }
-			  break;
-			case GL_INTENSITY:
- 	        src = (mgaUI8 *)image->Data + ( y * image->Width + x );
-			  stride = (image->Width - width);
-		     for ( i = height ; i ; i-- ) {
-			    for ( j = width >> 1  ; j ; j-- ) {
+					*destPtr++ = MGAPACKCOLOR4444(src[0],src[1],src[2],src[3]) |  
+						( MGAPACKCOLOR4444(src[4],src[5],src[6],src[7]) << 16 );
+					src += 8;
+				}
+				src += stride;
+			}
+			break;
+		case GL_LUMINANCE:
+			src = (mgaUI8 *)image->Data + ( y * image->Width + x );
+			stride = (image->Width - width);
+			for ( i = height ; i ; i-- ) {
+				for ( j = width >> 1  ; j ; j-- ) {
+					/* FIXME: should probably use 555 texture to get true grey */
+					*destPtr++ = MGAPACKCOLOR565(src[0],src[0],src[0]) |  
+						( MGAPACKCOLOR565(src[1],src[1],src[1]) << 16 );
+					src += 2;
+				}
+				src += stride;
+			}
+			break;
+		case GL_INTENSITY:
+			src = (mgaUI8 *)image->Data + ( y * image->Width + x );
+			stride = (image->Width - width);
+			for ( i = height ; i ; i-- ) {
+				for ( j = width >> 1  ; j ; j-- ) {
 			
-				   *destPtr++ = MGAPACKCOLOR4444(src[0],src[0],src[0],src[0]) |  
-					   ( MGAPACKCOLOR4444(src[1],src[1],src[1],src[1]) << 16 );
-				   src += 2;
-			    }
-				 src += stride;
-		     }
-			  break;
-			case GL_ALPHA:
- 	        src = (mgaUI8 *)image->Data + ( y * image->Width + x );
-			  stride = (image->Width - width);
-		     for ( i = height ; i ; i-- ) {
-			    for ( j = width >> 1  ; j ; j-- ) {
+					*destPtr++ = MGAPACKCOLOR4444(src[0],src[0],src[0],src[0]) |  
+						( MGAPACKCOLOR4444(src[1],src[1],src[1],src[1]) << 16 );
+					src += 2;
+				}
+				src += stride;
+			}
+			break;
+		case GL_ALPHA:
+			src = (mgaUI8 *)image->Data + ( y * image->Width + x );
+			stride = (image->Width - width);
+			for ( i = height ; i ; i-- ) {
+				for ( j = width >> 1  ; j ; j-- ) {
 			
-				   *destPtr++ = MGAPACKCOLOR4444(255,255,255,src[0]) |  
-					   ( MGAPACKCOLOR4444(255,255,255,src[1]) << 16 );
-				   src += 2;
-			    }
-				 src += stride;
-		     }
-			  break;
-			case GL_LUMINANCE_ALPHA:
- 	        src = (mgaUI8 *)image->Data + ( y * image->Width + x ) * 2;
-			  stride = (image->Width - width) * 2;
-		     for ( i = height ; i ; i-- ) {
-			    for ( j = width >> 1  ; j ; j-- ) {
+					*destPtr++ = MGAPACKCOLOR4444(255,255,255,src[0]) |  
+						( MGAPACKCOLOR4444(255,255,255,src[1]) << 16 );
+					src += 2;
+				}
+				src += stride;
+			}
+			break;
+		case GL_LUMINANCE_ALPHA:
+			src = (mgaUI8 *)image->Data + ( y * image->Width + x ) * 2;
+			stride = (image->Width - width) * 2;
+			for ( i = height ; i ; i-- ) {
+				for ( j = width >> 1  ; j ; j-- ) {
 			
-				   *destPtr++ = MGAPACKCOLOR4444(src[0],src[0],src[0],src[1]) |  
-					   ( MGAPACKCOLOR4444(src[2],src[2],src[2],src[3]) << 16 );
-				   src += 4;
-			    }
-				 src += stride;
-		     }
-			  break;
-			default:
-			  goto format_error;
-		 }
-		 break;
-	  case 4:
-		 switch (image->Format) {
-			case GL_RGB:
- 	        src = (mgaUI8 *)image->Data + (  y * image->Width + x ) * 3;
-			  stride = (image->Width - width) * 3;
-		     for ( i = height ; i ; i-- ) {
-			    for ( j = width ; j ; j-- ) {
+					*destPtr++ = MGAPACKCOLOR4444(src[0],src[0],src[0],src[1]) |  
+						( MGAPACKCOLOR4444(src[2],src[2],src[2],src[3]) << 16 );
+					src += 4;
+				}
+				src += stride;
+			}
+			break;
+		default:
+			goto format_error;
+		}
+		break;
+	case 4:
+		switch (image->Format) {
+		case GL_RGB:
+			src = (mgaUI8 *)image->Data + (  y * image->Width + x ) * 3;
+			stride = (image->Width - width) * 3;
+			for ( i = height ; i ; i-- ) {
+				for ( j = width ; j ; j-- ) {
 			
-				   *destPtr++ = MGAPACKCOLOR8888(src[0],src[1],src[2], 255);
-				   src += 3;
-			    }
-				 src += stride;
-		     }
-			  break;
-			case GL_RGBA:
- 	        src = (mgaUI8 *)image->Data + (  y * image->Width + x ) * 4;
-			  stride = (image->Width - width) * 4;
-		     for ( i = height ; i ; i-- ) {
-			    for ( j = width  ; j ; j-- ) {
+					*destPtr++ = MGAPACKCOLOR8888(src[0],src[1],src[2], 255);
+					src += 3;
+				}
+				src += stride;
+			}
+			break;
+		case GL_RGBA:
+			src = (mgaUI8 *)image->Data + (  y * image->Width + x ) * 4;
+			stride = (image->Width - width) * 4;
+			for ( i = height ; i ; i-- ) {
+				for ( j = width  ; j ; j-- ) {
 			
-				   *destPtr++ = MGAPACKCOLOR8888(src[0],src[1],src[2],src[3]);
-				   src += 4;
-			    }
-				 src += stride;
-		     }
-			  break;
-			case GL_LUMINANCE:
- 	        src = (mgaUI8 *)image->Data + ( y * image->Width + x );
-			  stride = (image->Width - width);
-		     for ( i = height ; i ; i-- ) {
-			    for ( j = width ; j ; j-- ) {
+					*destPtr++ = MGAPACKCOLOR8888(src[0],src[1],src[2],src[3]);
+					src += 4;
+				}
+				src += stride;
+			}
+			break;
+		case GL_LUMINANCE:
+			src = (mgaUI8 *)image->Data + ( y * image->Width + x );
+			stride = (image->Width - width);
+			for ( i = height ; i ; i-- ) {
+				for ( j = width ; j ; j-- ) {
 			
-				   *destPtr++ = MGAPACKCOLOR8888(src[0],src[0],src[0], 255);
-				   src += 1;
-			    }
-				 src += stride;
-		     }
-			  break;
-			case GL_INTENSITY:
- 	        src = (mgaUI8 *)image->Data + (  y * image->Width + x );
-			  stride = (image->Width - width);
-		     for ( i = height ; i ; i-- ) {
-			    for ( j = width ; j ; j-- ) {
+					*destPtr++ = MGAPACKCOLOR8888(src[0],src[0],src[0], 255);
+					src += 1;
+				}
+				src += stride;
+			}
+			break;
+		case GL_INTENSITY:
+			src = (mgaUI8 *)image->Data + (  y * image->Width + x );
+			stride = (image->Width - width);
+			for ( i = height ; i ; i-- ) {
+				for ( j = width ; j ; j-- ) {
 			
-				   *destPtr++ = MGAPACKCOLOR8888(src[0],src[0],src[0],src[0]);
-				   src += 1;
-			    }
-				 src += stride;
-		     }
-			  break;
-			case GL_ALPHA:
- 	        src = (mgaUI8 *)image->Data + ( y * image->Width + x );
-			  stride = (image->Width - width);
-		     for ( i = height ; i ; i-- ) {
-			    for ( j = width ; j ; j-- ) {
+					*destPtr++ = MGAPACKCOLOR8888(src[0],src[0],src[0],src[0]);
+					src += 1;
+				}
+				src += stride;
+			}
+			break;
+		case GL_ALPHA:
+			src = (mgaUI8 *)image->Data + ( y * image->Width + x );
+			stride = (image->Width - width);
+			for ( i = height ; i ; i-- ) {
+				for ( j = width ; j ; j-- ) {
 			
-				   *destPtr++ = MGAPACKCOLOR8888(255,255,255,src[0]);
-				   src += 1;
-			    }
-				 src += stride;
-		     }
-			  break;
-			case GL_LUMINANCE_ALPHA:
- 	        src = (mgaUI8 *)image->Data + ( y * image->Width + x ) * 2;
-			  stride = (image->Width - width) * 2;
-		     for ( i = height ; i ; i-- ) {
-			    for ( j = width ; j ; j-- ) {
+					*destPtr++ = MGAPACKCOLOR8888(255,255,255,src[0]);
+					src += 1;
+				}
+				src += stride;
+			}
+			break;
+		case GL_LUMINANCE_ALPHA:
+			src = (mgaUI8 *)image->Data + ( y * image->Width + x ) * 2;
+			stride = (image->Width - width) * 2;
+			for ( i = height ; i ; i-- ) {
+				for ( j = width ; j ; j-- ) {
 			
-				   *destPtr++ = MGAPACKCOLOR8888(src[0],src[0],
-						       src[0],src[1]);
-				   src += 2;
-			    }
-				 src += stride;
-		     }
-			  break;
-			default:
-			  goto format_error;
-		 }
-		 break;
-	  default:
-		 goto format_error;
+					*destPtr++ = MGAPACKCOLOR8888(src[0],src[0],
+								      src[0],src[1]);
+					src += 2;
+				}
+				src += stride;
+			}
+			break;
+		default:
+			goto format_error;
+		}
+		break;
+	default:
+		goto format_error;
 	}
 
 	return;
 
-format_error:
+ format_error:
 
-		mgaError( "Unsupported texelBytes %i, image->Format %i\n", 
-			  texelBytes, image->Format );
+	mgaError( "Unsupported texelBytes %i, image->Format %i\n", 
+		  (int)texelBytes, (int)image->Format );
 }
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgadd.c b/xc/lib/GL/mesa/src/drv/mga/mgadd.c
index ad56cbbc3..bdae9a2bf 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgadd.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgadd.c
@@ -61,11 +61,11 @@ static const GLubyte *mgaDDGetString( GLcontext *ctx, GLenum name )
    mgaContextPtr mmesa = MGA_CONTEXT( ctx );
    switch (name) {
    case GL_VENDOR:
-      return "Utah GLX";
+      return "Precision Insight";
    case GL_RENDERER:
-      if (MGA_IS_G200(mmesa)) return "GLX-MGA-G200";
-      if (MGA_IS_G400(mmesa)) return "GLX-MGA-G400";
-      return "GLX-MGA";
+      if (MGA_IS_G200(mmesa)) return "DRI-MGA-G200";
+      if (MGA_IS_G400(mmesa)) return "DRI-MGA-G400";
+      return "DRI-MGA";
    default:
       return 0;
    }
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgadepth.c b/xc/lib/GL/mesa/src/drv/mga/mgadepth.c
index 0d55137d1..2dfc3e00c 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgadepth.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgadepth.c
@@ -43,20 +43,24 @@
 #include "mgalib.h"
 #endif
 
-
+#define DEPTH_SCALE 65535.0F
 
 /*
  * Return the address of the Z-buffer value for window coordinate (x,y):
  */
-#define Z_SETUP								\
-   mgaContextPtr mmesa = MGA_CONTEXT(ctx);				\
-   __DRIscreenPrivate *sPriv = mmesa->driScreen;			\
-   mgaScreenPrivate *mgaScreen = mmesa->mgaScreen;			\
-   GLdepth *zbstart = (GLdepth *)(sPriv->pFB + mgaScreen->depthOffset);\
-   GLint zbpitch = mgaScreen->depthPitch
+#define Z_SETUP						\
+   __DRIscreenPrivate *sPriv = mmesa->driScreen;	\
+   __DRIdrawablePrivate *dPriv = mmesa->driDrawable;	\
+   GLuint height = dPriv->h;				\
+   mgaScreenPrivate *mgaScreen = mmesa->mgaScreen;	\
+   GLint zbpitch = mgaScreen->depthPitch;		\
+   char *zbstart = (char *)(sPriv->pFB + 		\
+			       mgaScreen->depthOffset +	\
+			       dPriv->x * 2 +		\
+			       dPriv->y * zbpitch)
 
 #define Z_ADDRESS( X, Y )  \
-  (zbstart + zbpitch * (Y) + (X))
+  (GLdepth *)(zbstart + zbpitch * (height - (Y)) + (X) * 2)
 
 
 /**********************************************************************/
@@ -84,16 +88,18 @@ static GLuint mga_depth_test_span_generic( GLcontext* ctx,
 					   const GLdepth z[],
 					   GLubyte mask[] )
 {
-   Z_SETUP;
-   GLdepth *zptr = Z_ADDRESS( x, y );
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
    GLubyte *m = mask;
    GLuint i;
    GLuint passed = 0;
 
-   LOCK_HARDWARE(mmesa);
+   LOCK_HARDWARE_QUIESCENT(mmesa);
+   {
+      Z_SETUP;
+      GLdepth *zptr = Z_ADDRESS( x, y );
 
-   /* switch cases ordered from most frequent to less frequent */
-   switch (ctx->Depth.Func) {
+      /* switch cases ordered from most frequent to less frequent */
+      switch (ctx->Depth.Func) {
       case GL_LESS:
          if (ctx->Depth.Mask) {
 	    /* Update Z buffer */
@@ -297,8 +303,8 @@ static GLuint mga_depth_test_span_generic( GLcontext* ctx,
 	 }
 	 break;
       default:
+      }
    }
-
    UNLOCK_HARDWARE(mmesa);
 
    return passed;
@@ -320,14 +326,16 @@ static void mga_depth_test_pixels_generic( GLcontext* ctx,
 					   const GLint x[], const GLint y[],
 					   const GLdepth z[], GLubyte mask[] )
 {
-   Z_SETUP;
-   register GLdepth *zptr;
-   register GLuint i;
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
+   LOCK_HARDWARE_QUIESCENT(mmesa);
+   {
+      Z_SETUP;
+      register GLdepth *zptr;
+      register GLuint i;
 
-   LOCK_HARDWARE(mmesa);
 
-   /* switch cases ordered from most frequent to less frequent */
-   switch (ctx->Depth.Func) {
+      /* switch cases ordered from most frequent to less frequent */
+      switch (ctx->Depth.Func) {
       case GL_LESS:
          if (ctx->Depth.Mask) {
 	    /* Update Z buffer */
@@ -547,8 +555,8 @@ static void mga_depth_test_pixels_generic( GLcontext* ctx,
 	 }
 	 break;
       default:
-    } 
-
+      } 
+   }
    UNLOCK_HARDWARE(mmesa);
 }
 
@@ -570,27 +578,29 @@ static void mga_read_depth_span_float( GLcontext* ctx,
 				       GLuint n, GLint x, GLint y, 
 				       GLfloat depth[] )
 {
-   Z_SETUP;
-   GLdepth *zptr;
-   GLfloat scale;
-   GLuint i;
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
+   LOCK_HARDWARE_QUIESCENT(mmesa);
+   {
 
-   LOCK_HARDWARE(mmesa);
+      Z_SETUP;
+      GLdepth *zptr;
+      GLfloat scale;
+      GLuint i;
 
-   scale = 1.0F / DEPTH_SCALE;
+      scale = 1.0F / DEPTH_SCALE;
 
-   if (ctx->Buffer->Depth) {
-      zptr = Z_ADDRESS( x, y );
-      for (i=0;i<n;i++) {
-	 depth[i] = (GLfloat) zptr[i] * scale;
+      if (ctx->ReadBuffer->Depth) {
+	 zptr = Z_ADDRESS( x, y );
+	 for (i=0;i<n;i++) {
+	    depth[i] = (GLfloat) zptr[i] * scale;
+	 }
       }
-   }
-   else {
-      for (i=0;i<n;i++) {
-	 depth[i] = 0.0F;
+      else {
+	 for (i=0;i<n;i++) {
+	    depth[i] = 0.0F;
+	 }
       }
    }
-   
    UNLOCK_HARDWARE(mmesa);
 }
 
@@ -607,30 +617,32 @@ static void mga_read_depth_span_int( GLcontext* ctx,
 				     GLuint n, GLint x, GLint y, 
 				     GLdepth depth[] )
 {
-   Z_SETUP;
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
+   LOCK_HARDWARE_QUIESCENT(mmesa);
+   {
+      Z_SETUP;
    
-   LOCK_HARDWARE(mmesa);
       
-   if (ctx->Buffer->Depth) {
-      GLdepth *zptr = Z_ADDRESS( x, y );
-      MEMCPY( depth, zptr, n * sizeof(GLdepth) );
-   }
-   else {
-      GLuint i;
-      for (i=0;i<n;i++) {
-	 depth[i] = 0;
+      if (ctx->ReadBuffer->Depth) {
+	 GLdepth *zptr = Z_ADDRESS( x, y );
+	 MEMCPY( depth, zptr, n * sizeof(GLdepth) );
+      }
+      else {
+	 GLuint i;
+	 for (i=0;i<n;i++) {
+	    depth[i] = 0;
+	 }
       }
    }
-   
    UNLOCK_HARDWARE(mmesa);
 }
 
 
 void mgaDDInitDepthFuncs( GLcontext *ctx )
 {
-   ctx->Driver.ReadDepthSpanFloat = mga_read_depth_span_float;
-   ctx->Driver.ReadDepthSpanInt = mga_read_depth_span_int;
-   ctx->Driver.DepthTestSpan = mga_depth_test_span_generic;
-   ctx->Driver.DepthTestPixels = mga_depth_test_pixels_generic;
+   ctx->Driver.ReadDepthSpan = mga_read_depth_span_float;
+   ctx->Driver.WriteDepthSpan = mga_read_depth_span_float;
+   ctx->Driver.ReadDepthPixels mga_read_depth_span_float;
+   ctx->Driver.WriteDepthPixels = mga_read_depth_span_float;
 }
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgafastpath.c b/xc/lib/GL/mesa/src/drv/mga/mgafastpath.c
index 8017e7f02..5e237bb01 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgafastpath.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgafastpath.c
@@ -380,6 +380,8 @@ static void mga_project_vertices( struct vertex_buffer *VB )
    GLmatrix *mat = &ctx->Viewport.WindowMap;
    GLfloat m[16];
 
+   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]);
@@ -401,6 +403,8 @@ static void mga_project_clipped_vertices( struct vertex_buffer *VB )
    GLmatrix *mat = &ctx->Viewport.WindowMap;
    GLfloat m[16];
 
+   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]);
@@ -507,18 +511,12 @@ void mgaDDFastPath( struct vertex_buffer *VB )
 	 ctx->CVA.elt_mode = gl_reduce_prim[prim];
 	 VB->EltPtr = &(MGA_DRIVER_DATA(VB)->clipped_elements);
 
-	 LOCK_HARDWARE( mmesa );
 	 mga_project_clipped_vertices( VB );    /* clip->device space */
 	 mga_render_elements_direct( VB );        /* render using new list */
-	 mgaFlushVerticesLocked( mmesa );
-	 UNLOCK_HARDWARE( mmesa );
       }
    } else {
-      LOCK_HARDWARE( mmesa );
       mga_project_vertices( VB );               /* clip->device space  */
       mga_render_elements_direct( VB );           /* render using orig list */
-      mgaFlushVerticesLocked( mmesa );
-      UNLOCK_HARDWARE( mmesa );
    }
 
    /* This indicates that there is no cached data to reuse.  
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgafasttmp.h b/xc/lib/GL/mesa/src/drv/mga/mgafasttmp.h
index 43caddcac..1fb177694 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgafasttmp.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgafasttmp.h
@@ -87,6 +87,8 @@ static void TAG(mga_setup_full)( struct vertex_buffer *VB, GLuint do_cliptest )
 	    STRIDE_F(tex0_data, tex0_stride);
 	 }
 	 if (TYPE & MGA_TEX1_BIT) {
+	    /* Hits a second cache line.
+	     */
 	    f[CLIP_S1] = tex1_data[0];
 	    f[CLIP_T1] = tex1_data[1];
 	    STRIDE_F(tex1_data, tex1_stride);
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgaioctl.c b/xc/lib/GL/mesa/src/drv/mga/mgaioctl.c
index c1819924d..f9ef24310 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)) {
+      printf("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 +153,57 @@ 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 );
-
-
-   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 +224,37 @@ 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 +267,62 @@ 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 (MGA_DEBUG&DEBUG_VERBOSE_IOCTL)
+      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;
 }
 
 
@@ -269,53 +330,243 @@ 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);
+
+   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);
+   }
+}
 
-void mgaFlushVertices( mgaContextPtr mmesa ) 
+
+static int intersect_rect( drm_clip_rect_t *out,
+			   drm_clip_rect_t *a,
+			   drm_clip_rect_t *b )
 {
-   LOCK_HARDWARE( mmesa );
-   mgaFlushVerticesLocked( mmesa );
-   UNLOCK_HARDWARE( mmesa );
+   *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 )
 {
-   XF86DRIClipRectPtr pbox = mmesa->pClipRects;
+   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;
 
-   if (mmesa->dirty)
+   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; 
+      
+   /* 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);
+
+
+   mmesa->vertex_dma_buffer = 0;
+   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)) 
    {
-      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 (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];
+	 }
 
-      for ( ; i < nr ; i++) 
-	 *b++ = pbox[i];
-	 
-      mga_vertex_dma_ioctl( mmesa );
+	 /* 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 ) 
+{
+   LOCK_HARDWARE( mmesa );
+   mgaFlushVerticesLocked( mmesa );
+   UNLOCK_HARDWARE( mmesa );
+}
+
+
+mgaUI32 *mgaAllocVertexDwords( mgaContextPtr mmesa, int dwords )
+{
+   int bytes = dwords * 4;
+   mgaUI32 *head;
+
+   if (!mmesa->vertex_dma_buffer) {
+      LOCK_HARDWARE( 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;
+}
 
-      break;			/* fix dma multiple dispatch */
+
+void mgaFireILoadLocked( mgaContextPtr mmesa, 
+			 GLuint offset, GLuint length )
+{
+   if (!mmesa->iload_buffer) {
+      fprintf(stderr, "mgaFireILoad: no buffer\n");
+      return;
    }
 
-   mga_get_buffer_ioctl( mmesa );
+   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 );
+}
+
+
+static void mgaDDFlush( GLcontext *ctx )
+{
+   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\n");
+      mgaUpdateLock( mmesa, DRM_LOCK_FLUSH );
+      UNLOCK_HARDWARE( mmesa );
+   }
+}
 
 
 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..13cec59d7 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgaioctl.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgaioctl.h
@@ -9,6 +9,21 @@ 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 );
 
@@ -19,5 +34,11 @@ 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);		\
+} 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 4600a69a0..176c126bd 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgalib.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgalib.h
@@ -36,6 +36,7 @@
 
 #include "types.h"
 
+#include "drm.h"
 #include "mgacommon.h"
 #include "mm.h"
 #include "mgalog.h"
@@ -43,7 +44,6 @@
 #include "mgatex.h"
 #include "mgavb.h"
 
-#include "mga_drm_public.h"
 #include "mga_xmesa.h"
 
 
@@ -51,11 +51,8 @@
 #define MGA_FIELD(field,val) (((val) << (field ## _SHIFT)) & ~(field ## _MASK))
 #define MGA_GET_FIELD(field, val) ((val & ~(field ## _MASK)) >> (field ## _SHIFT))
 
-#define MGA_CHIP_MGAG200 0
-#define MGA_CHIP_MGAG400 1
-
-#define MGA_IS_G200(mmesa) (mmesa->mgaScreen->chipset == MGA_CHIP_MGAG200)
-#define MGA_IS_G400(mmesa) (mmesa->mgaScreen->chipset == MGA_CHIP_MGAG400)
+#define MGA_IS_G200(mmesa) (mmesa->mgaScreen->chipset == MGA_CARD_TYPE_G200)
+#define MGA_IS_G400(mmesa) (mmesa->mgaScreen->chipset == MGA_CARD_TYPE_G400)
 
 
 /* SoftwareFallback 
@@ -65,6 +62,7 @@
  */
 #define MGA_FALLBACK_TEXTURE   0x1
 #define MGA_FALLBACK_BUFFER    0x2
+#define MGA_FALLBACK_STIPPLE   0x3
 
 
 /* For mgaCtx->new_state.
@@ -97,17 +95,30 @@ typedef void (*mga_interp_func)( GLfloat t,
 #define MGA_PF_8888  (10 << 4)
 #define MGA_PF_HASALPHA (8 << 4)
 
+
+
+/* Reasons why the GL_BLEND fallback mightn't work:
+ */
+#define MGA_BLEND_ENV_COLOR 0x1
+#define MGA_BLEND_MULTITEX  0x2
+
+
 struct mga_context_t {
 
    GLcontext *glCtx;
 
-   /* Hardware state - moved from mgabuf.h
-    */
-   mgaUI32 Setup[MGA_CTX_SETUP_SIZE];
 
-   /* Variable sized vertices
+   /* Bookkeeping for texturing 
     */
-   mgaUI32 vertsize;
+   int lastTexHeap;
+   struct mga_texture_object_s TexObjList[MGA_NR_TEX_HEAPS];
+   struct mga_texture_object_s SwappedOut;
+   struct mga_texture_object_s *CurrentTexObj[2];
+   memHeap_t *texHeap[MGA_NR_TEX_HEAPS];
+   int c_texupload;
+   int c_texusage;
+   int tex_thrash;
+
 
    /* Map GL texture units onto hardware.
     */
@@ -116,68 +127,84 @@ struct mga_context_t {
    mgaUI32 tex_dest[2];
 
 
+   /* Manage fallbacks 
+    */
+   mgaUI32 IndirectTriangles;
+   int Fallback;  
 
-   /* bookkeeping for textureing */
-   struct mga_texture_object_s TexObjList;
-   struct mga_texture_object_s SwappedOut;
-   struct mga_texture_object_s *CurrentTexObj[2];
-
-
-   /* shared texture palette */
-   mgaUI16	GlobalPalette[256];
-  
-   int Fallback;  /* or'ed values of FALLBACK_* */
 
-   /* Support for CVA and the fast paths */
+   /* Support for CVA and the fastpath 
+    */
    unsigned int setupdone;
    unsigned int setupindex;
    unsigned int renderindex;
    unsigned int using_fast_path;
-   unsigned int using_immediate_fast_path;
    mga_interp_func interp;
 
-   /* Shortcircuit some state changes */
+
+   /* Support for limited GL_BLEND fallback
+    */
+   unsigned int blend_flags;
+   unsigned int envcolor;
+
+
+   /* Shortcircuit some state changes 
+    */
    points_func   PointsFunc;
    line_func     LineFunc;
    triangle_func TriangleFunc;
    quad_func     QuadFunc;
 
-   /* Manage our own state */
+
+   /* Manage driver and hardware state 
+    */
    GLuint        new_state; 
    GLuint        dirty;
-
-   GLubyte       clearcolor[4];
-   GLushort MonoColor;
-   GLushort ClearColor;
+   GLuint        Setup[MGA_CTX_SETUP_SIZE];
+   GLuint        warp_pipe;
+   GLuint        vertsize;
+   GLushort      MonoColor;
+   GLushort      ClearColor;
+   GLuint        poly_stipple;
 
    
-   /* DRI stuff
+   /* Dma buffers
     */
-   drmBufPtr  dma_buffer;
+   drmBufPtr  vertex_dma_buffer;
+   drmBufPtr  iload_buffer;
 
-   GLframebuffer *glBuffer;
-   memHeap_t *texHeap;
+   
+   /* Drawable, cliprect and scissor information
+    */
+   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 */
+   int read_buffer;	
+   int readOffset;	
+   int drawX, drawY;		/* origin of drawable in draw buffer */
+   int lastX, lastY;		/* detect DSTORG bug */
+   GLuint numClipRects;		/* cliprects for the draw buffer */
+   XF86DRIClipRectPtr pClipRects;
+   XF86DRIClipRectRec draw_rect;
+   drm_clip_rect_t scissor_rect;
+   int scissor;
 
-   GLuint needClip;
-   GLuint warp_pipe;
 
-   /* These refer to the current draw (front vs. back) buffer:
+   /* Texture aging and DMA based aging.
     */
-   int drawOffset;		/* draw buffer address in agp space */
-   int drawX;			/* origin of drawable in draw buffer */
-   int drawY;
-   GLuint numClipRects;		/* cliprects for that buffer */
-   XF86DRIClipRectPtr pClipRects;
+   unsigned int texAge[MGA_NR_TEX_HEAPS];/* texture LRU age  */
+   int dirtyAge;		/* buffer age for synchronization */
+   int lastSwap;		/* throttling runaway apps */
 
-   int texAge;
 
-   XF86DRIClipRectRec draw_rect;
 
+   /* Mirrors of some DRI state.
+    */
+   GLframebuffer *glBuffer;
    drmContext hHWContext;
    drmLock *driHwLock;
    int driFd;
    Display *display;
-
    __DRIdrawablePrivate *driDrawable;
    __DRIscreenPrivate *driScreen;
    mgaScreenPrivate *mgaScreen; 
@@ -191,16 +218,12 @@ typedef struct {
 
 	/* dma stuff */
         mgaUI32         systemTexture;
-        mgaUI32         noSetupDma;
 
         mgaUI32         default32BitTextures;
-        mgaUI32         swapBuffersCount;
         
 	/* options */
 	mgaUI32		nullprims;  /* skip all primitive generation */
-	mgaUI32		noFallback; /* don't fall back to software, do
-                                       best-effort rendering */
- 	mgaUI32		skipDma;    /* don't send anything to the hardware */
+        mgaUI32         noFallback;
  	
 	/* performance counters */	
 	mgaUI32		c_textureUtilization;
@@ -240,10 +263,12 @@ extern mgaGlx_t	mgaglx;
 extern int MGA_DEBUG;
 #endif
 
-#define MGA_DEBUG_ALWAYS_SYNC    0x1
-#define MGA_DEBUG_VERBOSE_MSG    0x2
-#define MGA_DEBUG_VERBOSE_LRU    0x4
-#define MGA_DEBUG_VERBOSE_DRI    0x8
+#define DEBUG_ALWAYS_SYNC    0x1
+#define DEBUG_VERBOSE_MSG    0x2
+#define DEBUG_VERBOSE_LRU    0x4
+#define DEBUG_VERBOSE_DRI    0x8
+#define DEBUG_VERBOSE_IOCTL  0x10
+#define DEBUG_VERBOSE_2D     0x20
 
 static __inline__ mgaUI32 mgaPackColor(mgaUI32 format,
 				       mgaUI8 r, mgaUI8 g, 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgaspan.c b/xc/lib/GL/mesa/src/drv/mga/mgaspan.c
index 38e3cdde8..9bf62391c 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgaspan.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgaspan.c
@@ -5,70 +5,99 @@
 #include "mgalog.h"
 #include "mgaspan.h"
 
+#define DBG 0
+
 
 #define LOCAL_VARS					\
-   mgaContextPtr mmesa = MGA_CONTEXT( ctx );		\
    __DRIdrawablePrivate *dPriv = mmesa->driDrawable;	\
-   __DRIscreenPrivate *sPriv = mmesa->driScreen;	\
    mgaScreenPrivate *mgaScreen = mmesa->mgaScreen;	\
-   GLuint pitch = mgaScreen->backPitch;			\
+   __DRIscreenPrivate *sPriv = mmesa->driScreen;	\
+   GLuint pitch = mgaScreen->frontPitch;		\
    GLuint height = dPriv->h;				\
+   char *read_buf = (char *)(sPriv->pFB +		\
+			mmesa->readOffset +		\
+			dPriv->x * 2 +			\
+			dPriv->y * pitch);              \
    char *buf = (char *)(sPriv->pFB +			\
 			mmesa->drawOffset +		\
 			dPriv->x * 2 +			\
+			dPriv->y * pitch);              \
+   GLushort p = MGA_CONTEXT( ctx )->MonoColor;          \
+   (void) read_buf; (void) buf; (void) p
+   
+
+
+#define LOCAL_DEPTH_VARS				\
+   __DRIdrawablePrivate *dPriv = mmesa->driDrawable;	\
+   mgaScreenPrivate *mgaScreen = mmesa->mgaScreen;	\
+   __DRIscreenPrivate *sPriv = mmesa->driScreen;	\
+   GLuint pitch = mgaScreen->frontPitch;		\
+   GLuint height = dPriv->h;				\
+   char *buf = (char *)(sPriv->pFB +			\
+			mgaScreen->depthOffset +	\
+			dPriv->x * 2 +			\
 			dPriv->y * pitch)
 
-#define INIT_MONO_PIXEL(p) \
-   GLushort p = MGA_CONTEXT( ctx )->MonoColor;
+#define INIT_MONO_PIXEL(p) 
 
-#define CLIPPIXEL(_x,_y) (_x >= minx && _x <= maxx && \
-			  _y >= miny && _y <= maxy)
+#define CLIPPIXEL(_x,_y) (_x >= minx && _x < maxx && \
+			  _y >= miny && _y < maxy)
 
 
-#define CLIPSPAN(_x,_y,_n,_x1,_n1)				\
+#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) _n1 -= (minx - _x1), _x1 = minx;	\
-	    if (_x1 + _n1 > maxx) n1 -= (_x1 + n1 - maxx);	\
+	    if (_x1 < minx) _i += (minx - _x1), _x1 = minx;	\
+	    if (_x1 + _n1 >= maxx) n1 -= (_x1 + n1 - maxx) + 1;	\
          }
 
 
 
-
+#define HW_LOCK()				\
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);	\
+   LOCK_HARDWARE_QUIESCENT(mmesa);
 
 #define HW_CLIPLOOP()						\
   do {								\
     int _nc = mmesa->numClipRects;				\
-    LOCK_HARDWARE_QUIESCENT(mmesa);				\
     while (_nc--) {						\
-       int minx = mmesa->pClipRects[_nc].x1 - mmesa->drawX;		\
-       int miny = mmesa->pClipRects[_nc].y1 - mmesa->drawY; 	\
-       int maxx = mmesa->pClipRects[_nc].x2 - mmesa->drawX;		\
+       int minx = mmesa->pClipRects[_nc].x1 - mmesa->drawX;	\
+       int miny = mmesa->pClipRects[_nc].y1 - mmesa->drawY;	\
+       int maxx = mmesa->pClipRects[_nc].x2 - mmesa->drawX;	\
        int maxy = mmesa->pClipRects[_nc].y2 - mmesa->drawY;
 
-
 #define HW_ENDCLIPLOOP()			\
     }						\
-    UNLOCK_HARDWARE(mmesa);			\
   } while (0)
 
+#define HW_UNLOCK()				\
+    UNLOCK_HARDWARE(mmesa);
+
+
+
 
-#define Y_FLIP(_y) (height - _y)
+
+
+
+/* 16 bit, 565 rgb color spanline and pixel functions
+ */
+#define Y_FLIP(_y) (height - _y - 1)
 #define WRITE_RGBA( _x, _y, r, g, b, a )				\
-   *(GLushort *)(buf + _x*2 + _y*pitch)  = ( ((r & 0x1f) << 11) |	\
-		                             ((g & 0x3f) << 5) |	\
-		                             ((b & 0x1f)))
+   *(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 *)(buf + _x*2 + _y*pitch);	\
-   rgba[0] = (p >> 11) & 0x1f;				\
-   rgba[1] = (p >> 5) & 0x3f;				\
-   rgba[2] = (p >> 0) & 0x1f;				\
+   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] = 0;			/* or 255? */		\
 } while(0)
 
@@ -78,21 +107,21 @@ do {							\
 
 
 
-
-#define WRITE_RGBA( _x, _y, r, g, b, a )				\
-   *(GLushort *)(buf + _x*2 + _y*pitch)  = ( ((r & 0x1f) << 10) |	\
-		                            ((g & 0x1f) << 5) |		\
-                         		    ((b & 0x1f)))
-
+/* 15 bit, 555 rgb 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) << 3) |	\
+                         		    ((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 *)(buf + _x*2 + _y*pitch);	\
-   rgba[0] = (p >> 10) & 0x1f;				\
-   rgba[1] = (p >> 5) & 0x1f;				\
-   rgba[2] = (p >> 0) & 0x1f;				\
+   GLushort p = *(GLushort *)(read_buf + _x*2 + _y*pitch);	\
+   rgba[0] = (p >> 7) & 0xf8;				\
+   rgba[1] = (p >> 3) & 0xf8;				\
+   rgba[2] = (p << 3) & 0xf8;				\
    rgba[3] = 0;			/* or 255? */		\
 } while(0)
 
@@ -101,27 +130,19 @@ do {							\
 
 
 
-#if 0
-#define WRITE_RGBA( _x, _y, r, g, b, a )				\
-   *(GLuint *)(buf + _x*4 + _y*pitch)  = ( ((r) << 16) |	\
-		                           ((g) << 8) |		\
-                         		   ((b)))
+/* 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) mga##x##_16
+#include "depthtmp.h"
 
-#define WRITE_PIXEL( _x, _y, p )  \
-   *(GLuint *)(buf + _x*4 + _y*pitch)  = p
 
-#define READ_RGBA( rgba, _x, _y )			\
-do {							\
-   GLuint p = *(GLuint *)(buf + _x*4 + _y*pitch);	\
-   rgba[0] = (p >> 16) & 0xff;				\
-   rgba[1] = (p >> 8) & 0xff;				\
-   rgba[2] = (p >> 0) & 0xff;				\
-   rgba[3] = 0;			/* or 255? */		\
-} while(0)
 
-#define TAG(x) mga##x##_888
-#include "spantmp.h"
-#endif
 
 void mgaDDInitSpanFuncs( GLcontext *ctx )
 {
@@ -143,6 +164,11 @@ void mgaDDInitSpanFuncs( GLcontext *ctx )
       ctx->Driver.ReadRGBAPixels = mgaReadRGBAPixels_555;
    }
 
+   ctx->Driver.ReadDepthSpan = mgaReadDepthSpan_16;
+   ctx->Driver.WriteDepthSpan = mgaWriteDepthSpan_16;
+   ctx->Driver.ReadDepthPixels = mgaReadDepthPixels_16;
+   ctx->Driver.WriteDepthPixels = mgaWriteDepthPixels_16;
+
    ctx->Driver.WriteCI8Span        =NULL;
    ctx->Driver.WriteCI32Span       =NULL;
    ctx->Driver.WriteMonoCISpan     =NULL;
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgastate.c b/xc/lib/GL/mesa/src/drv/mga/mgastate.c
index ff9063e0b..0dacd7646 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 +447,85 @@ 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.
+ * Could also consider using a multitexturing mechanism for this, but
+ * that has real issues, too.  
+ */
+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 +533,16 @@ 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 +551,7 @@ 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 +560,28 @@ 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;      
    mmesa->sarea->dirty |= mmesa->dirty;
-   mmesa->dirty = 0;
+#if 0
+   fprintf(stderr, "in mgaEmitHwStateLocked: dirty now %x\n",
+	   mmesa->sarea->dirty);
+#endif
+   mmesa->dirty &= (MGA_UPLOAD_CLIPRECTS|MGA_WAIT_AGE);
 }
 
 
@@ -533,29 +591,50 @@ 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 +660,7 @@ static void mgaWarpUpdateState( GLcontext *ctx )
    {
       mmesa->warp_pipe = index;
       mmesa->new_state |= MGA_NEW_WARP;
+      mmesa->dirty |= MGA_UPLOAD_PIPE;
    }
 }
 
@@ -611,21 +691,15 @@ 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 +715,6 @@ 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 +726,17 @@ 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 +758,16 @@ 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 +776,60 @@ 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 +851,14 @@ 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/mgastate.h b/xc/lib/GL/mesa/src/drv/mga/mgastate.h
index 5ada1b23c..ac92cb0c2 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgastate.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgastate.h
@@ -7,9 +7,10 @@ extern void mgaDDUpdateHwState( GLcontext *ctx );
 extern void mgaDDUpdateState( GLcontext *ctx );
 extern void mgaDDReducedPrimitiveChange( GLcontext *ctx, GLenum prim );
 
-/* reprograms the current registers without updating them, used to
-reset state after a dma buffer overflow */
-void mgaUpdateRegs( GLuint regs );
+extern void mgaInitState( mgaContextPtr mmesa );
+
+extern void mgaUpdateClipping(const GLcontext *ctx);
+
 
 
 #endif
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgatex.c b/xc/lib/GL/mesa/src/drv/mga/mgatex.c
index be7ad6018..3afb70bad 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) {
@@ -804,7 +1004,6 @@ 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 +1014,8 @@ 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 +1023,11 @@ 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 +1051,31 @@ 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 +1091,57 @@ 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 +1161,7 @@ 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 +1191,7 @@ 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 +1212,9 @@ 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 +1231,68 @@ 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/mgatex.h b/xc/lib/GL/mesa/src/drv/mga/mgatex.h
index 1e8e7e86f..7873146c0 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgatex.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgatex.h
@@ -28,7 +28,6 @@
 #ifndef MGATEX_INC
 #define MGATEX_INC
 
-#include "mga_drm_public.h"
 #include "types.h"
 #include "mgacommon.h"
 #include "mm.h"
@@ -48,7 +47,9 @@ typedef struct mga_texture_object_s {
         mgaUI32         dirty_images;
 	mgaUI32		totalSize;		
 	int		texelBytes;
-	mgaUI32 	age;   			
+	mgaUI32 	age;
+        int             bound;
+        int             heap;	/* agp or card */
         mgaUI32         Setup[MGA_TEX_SETUP_SIZE];
 } mgaTextureObject_t;
 
@@ -96,7 +97,7 @@ void mgaUpdateTexturePalette( GLcontext *ctx, struct gl_texture_object *tObj );
 
 GLboolean mgaIsTextureResident( GLcontext *ctx, struct gl_texture_object *t );
 
-void mgaConvertTexture( mgaUI32 *destPtr, int texelBytes,
+void mgaConvertTexture( mgaUI32 *dest, int texelBytes,
 			struct gl_texture_image *image,
 			int x, int y, int width, int height );
 
@@ -107,9 +108,7 @@ int mgaUploadTexImages( mgaContextPtr mmesa, mgaTextureObjectPtr t );
 
 
 
-void mgaResetGlobalLRU( mgaContextPtr mmesa );
-void mgaTexturesGone( mgaContextPtr mmesa, GLuint offset, 
-		      GLuint size, GLuint in_use );
+void mgaAgeTextures( mgaContextPtr mmesa, int heap );
 
 void mgaDDInitTextureFuncs( GLcontext *ctx );
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgatris.c b/xc/lib/GL/mesa/src/drv/mga/mgatris.c
index bd8ef083f..b83aaf2ea 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 +143,53 @@ void mgaDDChooseRenderState( GLcontext *ctx )
    mgaContextPtr mmesa = MGA_CONTEXT( ctx );
    GLuint flags = ctx->TriangleCaps;
 
-   ctx->IndirectTriangles &= ~DD_SW_RASTERIZE;
+   if (mmesa->Fallback)
+      return;
+
+   mmesa->IndirectTriangles &= ~DD_SW_RASTERIZE;
 
    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 (flags & DD_POINT_SMOOTH)                 ind |= MGA_ANTIALIAS_BIT;
-      if (flags & DD_POINT_ATTEN)                  ind |= fallback;
 
       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 (flags & DD_LINE_SMOOTH)                   ind |= MGA_ANTIALIAS_BIT;
-      if (flags & DD_LINE_STIPPLE)                  ind |= fallback;
+      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/mgatris.h b/xc/lib/GL/mesa/src/drv/mga/mgatris.h
index 3b41900f1..dcf697409 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgatris.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgatris.h
@@ -34,26 +34,16 @@ extern void mgaDDTrifuncInit( void );
 
 
 /* Todo: 
- *    - multidraw, ...
  *    - Antialiasing (?)
  *    - line and polygon stipple
- *    - select and feedback 
  *    - stencil 
- *    - point parameters
- *    - 
  */
 #define MGA_ANTIALIAS_BIT   0       /* ignored for now, no fallback */
-#define MGA_FLAT_BIT	0x1
-#define MGA_OFFSET_BIT	0x2	/* 3.1 only */
-#define MGA_TWOSIDE_BIT	0x4	/* 3.1 only */
-#define MGA_NODRAW_BIT	0x8
-#define MGA_FALLBACK_BIT	0x10
-
-/* Not in use:
- */
-#define MGA_FEEDBACK_BIT    0x20
-#define MGA_SELECT_BIT      0x40
-#define MGA_POINT_PARAM_BIT 0x80	/* not needed? */
+#define MGA_FLAT_BIT	    0x1
+#define MGA_OFFSET_BIT	    0x2	
+#define MGA_TWOSIDE_BIT	    0x4	
+#define MGA_NODRAW_BIT	    0x8
+#define MGA_FALLBACK_BIT    0x10
 
 
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgatritmp.h b/xc/lib/GL/mesa/src/drv/mga/mgatritmp.h
index ac5a545d2..ba2410914 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgatritmp.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgatritmp.h
@@ -98,8 +98,6 @@ 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 );
@@ -131,7 +129,6 @@ static void TAG(points)( GLcontext *ctx, GLuint first, GLuint last )
 	 mga_draw_point( mmesa, &mgaVB[i], sz );
 }
 
-#endif
 
 
 static void TAG(init)( void )
@@ -139,10 +136,9 @@ 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 f78c542c8..0b703a184 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgavb.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgavb.c
@@ -260,11 +260,21 @@ void mgaChooseRasterSetupFunc(GLcontext *ctx)
    mmesa->tex_dest[0] = MGA_TEX0_BIT;
    mmesa->tex_dest[1] = MGA_TEX1_BIT;
    mmesa->multitex = 0;
+   mmesa->blend_flags &= ~MGA_BLEND_MULTITEX;
 
    if (ctx->Texture.Enabled & 0xf) {
       if (ctx->Texture.Unit[0].EnvMode == GL_REPLACE)
 	 funcindex &= ~MGA_RGBA_BIT;
 
+      if (ctx->Texture.Unit[0].EnvMode == GL_BLEND &&
+	  mmesa->envcolor) 
+      {
+	 mmesa->multitex = 1;
+	 mmesa->vertsize = 10;
+	 mmesa->tmu_source[1] = 0;
+	 funcindex |= MGA_TEX1_BIT;
+      }	 
+
       funcindex |= MGA_TEX0_BIT;
    }
 
@@ -272,7 +282,8 @@ void mgaChooseRasterSetupFunc(GLcontext *ctx)
       if (ctx->Texture.Enabled & 0xf) {
 	 mmesa->multitex = 1;
 	 mmesa->vertsize = 10;
-	 funcindex |= MGA_TEX1_BIT;
+	 mmesa->blend_flags |= MGA_BLEND_MULTITEX;
+	 funcindex |= MGA_TEX1_BIT;	 
       } else {
 	 /* Just a funny way of doing single texturing 
 	  */
@@ -281,11 +292,25 @@ void mgaChooseRasterSetupFunc(GLcontext *ctx)
 
 	 if (ctx->Texture.Unit[1].EnvMode == GL_REPLACE)
 	    funcindex &= ~MGA_RGBA_BIT;
+	 
+	 if (ctx->Texture.Unit[0].EnvMode == GL_BLEND &&
+	     mmesa->envcolor) 
+	 {
+	    mmesa->multitex = 1;
+	    mmesa->vertsize = 10;
+	    mmesa->tmu_source[1] = 1;
+	    funcindex |= MGA_TEX1_BIT;
+	 }	 
 
 	 funcindex |= MGA_TEX0_BIT;
       }
    }
 
+   /* Not really a good place to do this - need to make the mga state
+    * management code more event-driven so this can be calculated for
+    * free.
+    */
+
    if (ctx->Color.BlendEnabled)
       funcindex |= MGA_ALPHA_BIT;
 
@@ -298,6 +323,7 @@ void mgaChooseRasterSetupFunc(GLcontext *ctx)
    if (0)
       mgaPrintSetupFlags("xsmesa: full setup function", funcindex); 
 
+   mmesa->dirty |= MGA_UPLOAD_PIPE;
    mmesa->setupindex = funcindex;
 
    /* Called by mesa's clip functons:
@@ -477,17 +503,3 @@ void mgaDDUnregisterVB( struct vertex_buffer *VB )
 }
 
 
-mgaUI32 *mgaAllocVertexDwords( mgaContextPtr mmesa, int dwords )
-{
-   int bytes = dwords * 4;
-   mgaUI32 *head;
-
-   if (mmesa->dma_buffer->used + bytes > mmesa->dma_buffer->total)
-      mgaFlushVertices( mmesa );
-
-   head = (mgaUI32 *)((char *)mmesa->dma_buffer->address + 
-		      mmesa->dma_buffer->used);
-
-   mmesa->dma_buffer->used += bytes;
-   return head;
-}
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgavb.h b/xc/lib/GL/mesa/src/drv/mga/mgavb.h
index 8d7c65207..675c1779c 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgavb.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgavb.h
@@ -27,6 +27,7 @@
 #ifndef MGAVB_INC
 #define MGAVB_INC
 
+#include "types.h"
 #include "vb.h"
 #include "mgacommon.h"
 
@@ -126,7 +127,4 @@ extern void mgaDDResizeVB( struct vertex_buffer *VB, GLuint size );
 extern void mgaDDSetupInit( void );
 
 
-extern mgaUI32 *mgaAllocVertexDwords( mgaContextPtr mmesa, int dwords );
-
-
 #endif
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/Imakefile b/xc/lib/GL/mesa/src/drv/tdfx/Imakefile
index 982ca93e8..a8fe76fe7 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/Imakefile
+++ b/xc/lib/GL/mesa/src/drv/tdfx/Imakefile
@@ -278,7 +278,7 @@ XCOMM Disabling 3Dnow code for the time being.
 	 SRCS = $(DRISRCS) $(DRMSRCS) $(TDFXSRCS) $(MESASRCS) $(ASMSRCS)
 	 OBJS = $(DRIOBJS) $(DRMOBJS) $(TDFXOBJS) $(MESAOBJS) $(ASMOBJS)
 
-REQUIREDLIBS += -lglide3x
+REQUIREDLIBS += -lglide3x -lm
 
 #if !GlxUseBuiltInDRIDriver
 #undef DoNormalLib NormalLibGlx
diff --git a/xc/programs/Xserver/GL/dri/dri.c b/xc/programs/Xserver/GL/dri/dri.c
index 68e638f27..22780d753 100644
--- a/xc/programs/Xserver/GL/dri/dri.c
+++ b/xc/programs/Xserver/GL/dri/dri.c
@@ -120,6 +120,7 @@ DRIScreenInit(
 
     pDRIPriv->directRenderingSupport = TRUE;
     pDRIPriv->pDriverInfo = pDRIInfo;
+    pDRIPriv->nrWindows = 0;
 
     /* setup device independent direct rendering memory maps */
 
@@ -334,8 +335,9 @@ DRIFinishScreenInit(ScreenPtr pScreen)
 	pDRIPriv->wrap.CopyWindow = pScreen->CopyWindow;
 	pScreen->CopyWindow = pDRIInfo->wrap.CopyWindow;
     }
-    miClipNotify(DRIClipNotify);
-
+    if (pDRIInfo->wrap.ClipNotify)
+       miClipNotify(pDRIInfo->wrap.ClipNotify);
+   
     DRIDrvMsg(pScreen->myNum, X_INFO, "[DRI] installation complete\n");
 
     return TRUE;
@@ -492,8 +494,6 @@ DRICloseConnection(
     ScreenPtr pScreen
 )
 {
-    DRIScreenPrivPtr pDRIPriv = DRI_SCREEN_PRIV(pScreen);
-
     return TRUE;
 }
 
@@ -652,7 +652,7 @@ DRICreateContext(
 				                      visual, 
 				                      *pHWContext, 
 				                      *pVisualConfigPriv,
-						      contextStore))) {
+						      (int)contextStore))) {
 	    DRIDestroyContextPriv(pDRIContextPriv);
 	    return FALSE;
 	}
@@ -689,12 +689,72 @@ DRIContextPrivDelete(
     if (pDRIPriv->pDriverInfo->DestroyContext) {
       contextStore=DRIGetContextStore(pDRIContextPriv);
       (pDRIPriv->pDriverInfo->DestroyContext)(pDRIContextPriv->pScreen,
-					     pDRIContextPriv->hwContext,
-					     contextStore);
+					      pDRIContextPriv->hwContext,
+					      (int)contextStore);
     }
     return DRIDestroyContextPriv(pDRIContextPriv);
 }
 
+
+/* This walks the drawable timestamp array and invalidates all of them
+ * in the case of transition from private to shared backbuffers.  It's
+ * not necessary for correctness, because DRIClipNotify gets called in
+ * time to prevent any conflict, but the transition from
+ * shared->private is sometimes missed if we don't do this.  
+ */
+static void 
+DRIClipNotifyAllDrawables( ScreenPtr pScreen )
+{
+   int i;
+   DRIScreenPrivPtr pDRIPriv = DRI_SCREEN_PRIV(pScreen);
+
+   for( i=0; i < pDRIPriv->pDriverInfo->maxDrawableTableEntry; i++) {
+      pDRIPriv->pSAREA->drawableTable[i].stamp = DRIDrawableValidationStamp++;
+   }   
+}
+
+
+static void 
+DRITransitionToSharedBuffers( ScreenPtr pScreen )
+{
+/*     ErrorF("DRITransitionToSharedBuffers\n"); */
+   DRIClipNotifyAllDrawables( pScreen );
+}
+
+
+static void 
+DRITransitionToPrivateBuffers( ScreenPtr pScreen )
+{
+/*     ErrorF("DRITransitionToPrivateBuffers\n"); */
+   DRIClipNotifyAllDrawables( pScreen );
+}
+
+
+static void
+DRITransitionTo3d( ScreenPtr pScreen )
+{
+   DRIScreenPrivPtr pDRIPriv = DRI_SCREEN_PRIV(pScreen);
+   DRIInfoPtr pDRIInfo = pDRIPriv->pDriverInfo;
+
+/*     ErrorF("DRITransitionTo3d\n"); */
+
+   if (pDRIInfo->TransitionTo3d)
+      pDRIInfo->TransitionTo3d( pScreen );
+}
+
+static void
+DRITransitionTo2d( ScreenPtr pScreen )
+{
+   DRIScreenPrivPtr pDRIPriv = DRI_SCREEN_PRIV(pScreen);
+   DRIInfoPtr pDRIInfo = pDRIPriv->pDriverInfo;
+
+/*     ErrorF("DRITransitionTo2d\n"); */
+
+   if (pDRIInfo->TransitionTo2d)
+      pDRIInfo->TransitionTo2d( pScreen );
+}
+
+
 Bool
 DRICreateDrawable(
     ScreenPtr pScreen,
@@ -734,6 +794,17 @@ DRICreateDrawable(
 	    pWin->devPrivates[DRIWindowPrivIndex].ptr = 
 						(pointer)pDRIDrawablePriv;
 
+	    switch (++pDRIPriv->nrWindows) {
+	    case 1:
+	       DRITransitionTo3d( pScreen );
+	       break;
+	    case 2:
+	       DRITransitionToSharedBuffers( pScreen );
+	       break;
+	    default:
+	       break;
+	    }
+
 	    /* track this in case this window is destroyed */
 	    AddResource(id, DRIDrawablePrivResType, (pointer)pWin);
 	}
@@ -756,13 +827,14 @@ DRIDestroyDrawable(
     DRIDrawablePrivPtr	pDRIDrawablePriv;
     WindowPtr		pWin;
 
+
     if (pDrawable->type == DRAWABLE_WINDOW) {
 	pWin = (WindowPtr)pDrawable;
 	pDRIDrawablePriv = DRI_DRAWABLE_PRIV_FROM_WINDOW(pWin);
 	pDRIDrawablePriv->refCount--;
 	if (pDRIDrawablePriv->refCount <= 0) {
 	    /* This calls back DRIDrawablePrivDelete which frees private area */
-	    FreeResourceByType(id, DRIDrawablePrivResType, FALSE);
+	    FreeResourceByType(id, DRIDrawablePrivResType, FALSE);	    
 	}
     }
     else { /* pixmap (or for GLX 1.3, a PBuffer) */
@@ -798,6 +870,17 @@ DRIDrawablePrivDelete(
 	}
 	xfree(pDRIDrawablePriv);
 	pWin->devPrivates[DRIWindowPrivIndex].ptr = NULL;
+
+	switch (--pDRIPriv->nrWindows) {
+	case 0:
+	   DRITransitionTo2d( pDrawable->pScreen );
+	   break;
+	case 1:
+	   DRITransitionToPrivateBuffers( pDrawable->pScreen );
+	   break;
+	default:
+	   break;
+	}
     }
     else { /* pixmap (or for GLX 1.3, a PBuffer) */
 	/* NOT_DONE */
@@ -818,7 +901,11 @@ DRIGetDrawableInfo(
     int* W,
     int* H,
     int* numClipRects,
-    XF86DRIClipRectPtr* pClipRects
+    XF86DRIClipRectPtr* pClipRects,
+    int* backX,
+    int* backY,
+    int* numBackClipRects,
+    XF86DRIClipRectPtr* pBackClipRects
 )
 {
     DRIScreenPrivPtr pDRIPriv = DRI_SCREEN_PRIV(pScreen);
@@ -907,6 +994,35 @@ DRIGetDrawableInfo(
 	    *H = (int)(pWin->drawable.height);
 	    *numClipRects = REGION_NUM_RECTS(&pWin->clipList);
 	    *pClipRects = (XF86DRIClipRectPtr)REGION_RECTS(&pWin->clipList);
+	    
+	    *backX = *X;
+	    *backY = *Y;
+
+	    if (pDRIPriv->nrWindows == 1 && *numClipRects) {
+	       /* Use a single cliprect. */
+
+	       int x0 = *X;
+	       int y0 = *Y;
+	       int x1 = x0 + *W;
+	       int y1 = y0 + *H;
+
+	       if (x0 < 0) x0 = 0;
+	       if (y0 < 0) y0 = 0;
+	       if (x1 > pScreen->width-1) x1 = pScreen->width-1;
+	       if (y1 > pScreen->height-1) y1 = pScreen->height-1;
+
+	       pDRIPriv->private_buffer_rect.x1 = x0;
+	       pDRIPriv->private_buffer_rect.y1 = y0;
+	       pDRIPriv->private_buffer_rect.x2 = x1;
+	       pDRIPriv->private_buffer_rect.y2 = y1;
+
+	       *numBackClipRects = 1;
+	       *pBackClipRects = &(pDRIPriv->private_buffer_rect);
+	    } else {
+	       /* Use the frontbuffer cliprects for back buffers.  */
+	       *numBackClipRects = 0;
+	       *pBackClipRects = 0;
+	    }
 	}
 	else {
 	    /* Not a DRIDrawable */
@@ -961,6 +1077,10 @@ DRICreateInfoRec(void)
     inforec->wrap.CopyWindow            = DRICopyWindow;
     inforec->wrap.ValidateTree          = DRIValidateTree;
     inforec->wrap.PostValidateTree      = DRIPostValidateTree;
+    inforec->wrap.ClipNotify            = DRIClipNotify;
+
+    inforec->TransitionTo2d = 0;
+    inforec->TransitionTo3d = 0;
 
     return inforec;
 }
@@ -972,6 +1092,7 @@ DRIDestroyInfoRec(DRIInfoPtr DRIInfo)
     xfree((char*)DRIInfo);
 }
 
+
 void
 DRIWakeupHandler(
     pointer wakeupData,
@@ -1502,10 +1623,27 @@ DRIGetDrawableIndex(
     return index;
 }
 
+unsigned int 
+DRIGetDrawableStamp(
+   ScreenPtr pScreen,
+   CARD32 drawable_index)
+{
+    DRIScreenPrivPtr pDRIPriv = DRI_SCREEN_PRIV(pScreen);
+    return pDRIPriv->pSAREA->drawableTable[drawable_index].stamp;
+}
+
+
+void DRIPrintDrawableLock(ScreenPtr pScreen, char *msg)
+{
+    DRIScreenPrivPtr pDRIPriv = DRI_SCREEN_PRIV(pScreen);
+    
+    ErrorF("%s: %d\n", msg,  pDRIPriv->pSAREA->drawable_lock.lock);
+}
+
 void
 DRILock(ScreenPtr pScreen, int flags) {
     DRIScreenPrivPtr pDRIPriv = DRI_SCREEN_PRIV(pScreen);
-
+   
     if (!lockRefCount)
       DRM_LOCK(pDRIPriv->drmFD, pDRIPriv->pSAREA, pDRIPriv->myContext, flags);
     lockRefCount++;
@@ -1532,7 +1670,7 @@ void *DRIGetSAREAPrivate(ScreenPtr pScreen)
 {
   DRIScreenPrivPtr pDRIPriv = DRI_SCREEN_PRIV(pScreen);
   if (!pDRIPriv) return 0;
-  return ((void*)pDRIPriv->pSAREA)+sizeof(XF86DRISAREARec);
+  return (void *)(((char*)pDRIPriv->pSAREA)+sizeof(XF86DRISAREARec));
 }
 
 drmContext DRIGetContext(ScreenPtr pScreen)
@@ -1542,3 +1680,11 @@ drmContext DRIGetContext(ScreenPtr pScreen)
   return pDRIPriv->myContext;
 }
   
+/* This lets get at the unwrapped functions so that they can correctly
+ * call the lowerlevel functions, and choose whether they will be
+ * called at every level of recursion (eg in validatetree).  
+ */
+DRIWrappedFuncsRec *DRIGetWrappedFuncs( ScreenPtr pScreen )
+{
+   return &(DRI_SCREEN_PRIV(pScreen)->wrap);
+}
diff --git a/xc/programs/Xserver/GL/dri/dri.h b/xc/programs/Xserver/GL/dri/dri.h
index d82ec5359..367351836 100644
--- a/xc/programs/Xserver/GL/dri/dri.h
+++ b/xc/programs/Xserver/GL/dri/dri.h
@@ -69,6 +69,10 @@ typedef int DRIWindowRequests;
 #define DRI_3D_WINDOWS_ONLY  1
 #define DRI_ALL_WINDOWS      2
 
+
+typedef void (*ClipNotifyPtr)( WindowPtr, int, int );
+
+
 /*
  * These functions can be wrapped by the DRI.  Each of these have
  * generic default funcs (initialized in DRICreateInfoRec) and can be
@@ -82,8 +86,12 @@ typedef struct {
     CopyWindowProcPtr            CopyWindow;
     ValidateTreeProcPtr          ValidateTree;
     PostValidateTreeProcPtr      PostValidateTree;
+    ClipNotifyPtr                ClipNotify;
 } DRIWrappedFuncsRec, *DRIWrappedFuncsPtr;
 
+
+
+
 typedef struct {
     /* driver call back functions */
     Bool	(*CreateContext)(ScreenPtr pScreen,
@@ -107,6 +115,8 @@ typedef struct {
 			       DDXPointRec ptOldOrg,
 			       RegionPtr prgnSrc,
 			       CARD32 index);
+    void        (*TransitionTo3d)(ScreenPtr pScreen);
+    void        (*TransitionTo2d)(ScreenPtr pScreen);
 
     /* wrapped functions */
     DRIWrappedFuncsRec  wrap;
@@ -185,7 +195,11 @@ Bool DRIGetDrawableInfo(
     int* W,
     int* H,
     int* numClipRects,
-    XF86DRIClipRectPtr* pClipRects
+    XF86DRIClipRectPtr* pClipRects,
+    int* backX,
+    int* backY,
+    int* numBackClipRects,
+    XF86DRIClipRectPtr* pBackClipRects
 );
 Bool DRIGetDeviceInfo(
     ScreenPtr pScreen,
@@ -199,6 +213,7 @@ Bool DRIGetDeviceInfo(
 DRIInfoPtr DRICreateInfoRec(void);
 void DRIDestroyInfoRec(DRIInfoPtr DRIInfo);
 Bool DRIFinishScreenInit(ScreenPtr pScreen);
+
 void DRIWakeupHandler(
     pointer wakeupData,
     int result,
@@ -246,7 +261,13 @@ CARD32 DRIGetDrawableIndex(
     WindowPtr pWin);
 void DRILock(ScreenPtr pScreen, int flags);
 void DRIUnlock(ScreenPtr pScreen);
+
+DRIWrappedFuncsRec *DRIGetWrappedFuncs( ScreenPtr pScreen );
+
 void *DRIGetSAREAPrivate(ScreenPtr pScreen);
+
+unsigned int DRIGetDrawableStamp(ScreenPtr pScreen, CARD32 drawable_index);
+
 DRIContextPrivPtr
 DRICreateContextPriv(ScreenPtr pScreen,
 		     drmContextPtr pHWContext,
@@ -256,6 +277,7 @@ DRICreateContextPrivFromHandle(ScreenPtr pScreen,
 			       drmContext hHWContext,
 			       DRIContextFlags flags);
 Bool DRIDestroyContextPriv(DRIContextPrivPtr pDRIContextPriv);
+drmContext DRIGetContext(ScreenPtr pScreen);
 
 #define _DRI_H_
 
diff --git a/xc/programs/Xserver/GL/dri/dristruct.h b/xc/programs/Xserver/GL/dri/dristruct.h
index 392be3d97..d576f40af 100644
--- a/xc/programs/Xserver/GL/dri/dristruct.h
+++ b/xc/programs/Xserver/GL/dri/dristruct.h
@@ -87,6 +87,8 @@ typedef struct _DRIScreenPrivRec
     void**		hiddenContextStore;    /* hidden X context          */
     void**		partial3DContextStore; /* parital 3D context        */
     DRIInfoPtr		pDriverInfo;
+    int                 nrWindows;
+    XF86DRIClipRectRec  private_buffer_rect; /* management of private buffers */
     DRIWrappedFuncsRec	wrap;
     DrawablePtr		DRIDrawables[SAREA_MAX_DRAWABLES];
 } DRIScreenPrivRec, *DRIScreenPrivPtr;
diff --git a/xc/programs/Xserver/GL/dri/xf86dri.c b/xc/programs/Xserver/GL/dri/xf86dri.c
index 77f74e0a8..2619fba87 100644
--- a/xc/programs/Xserver/GL/dri/xf86dri.c
+++ b/xc/programs/Xserver/GL/dri/xf86dri.c
@@ -1,4 +1,4 @@
-/* $XFree86: xc/programs/Xserver/GL/dri/xf86dri.c,v 1.2 1999/06/27 14:07:39 dawes Exp $ */
+/* $XFree86: xc/programs/Xserver/GL/dri/xf86dri.c,v 1.6 2000/02/23 04:46:52 martin Exp $ */
 /**************************************************************************
 
 Copyright 1998-1999 Precision Insight, Inc., Cedar Park, Texas.
@@ -31,7 +31,6 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  *   Kevin E. Martin <kevin@precisioninsight.com>
  *   Jens Owen <jens@precisioninsight.com>
  *
- * $PI: xc/programs/Xserver/GL/dri/xf86dri.c,v 1.20 1999/06/24 19:10:40 faith Exp $
  */
 
 #if XFree86LOADER
@@ -389,6 +388,8 @@ ProcXF86DRIGetDrawableInfo(
     DrawablePtr pDrawable;
     int X, Y, W, H;
     XF86DRIClipRectPtr pClipRects;
+    XF86DRIClipRectPtr pBackClipRects;
+    int backX, backY;
 
     REQUEST(xXF86DRIGetDrawableInfoReq);
     REQUEST_SIZE_MATCH(xXF86DRIGetDrawableInfoReq);
@@ -412,7 +413,11 @@ ProcXF86DRIGetDrawableInfo(
 			     (int*)&W,
 			     (int*)&H,
 			     (int*)&rep.numClipRects,
-			     &pClipRects)) {
+			     &pClipRects,
+			     &backX, 
+			     &backY,
+			     (int*)&rep.numBackClipRects,
+			     &pBackClipRects)) {
 	return BadValue;
     }
 
@@ -420,18 +425,32 @@ ProcXF86DRIGetDrawableInfo(
     rep.drawableY = Y;
     rep.drawableWidth = W;
     rep.drawableHeight = H;
+    rep.length = (SIZEOF(xXF86DRIGetDrawableInfoReply) - 
+		  SIZEOF(xGenericReply));
+
+    rep.backX = backX;
+    rep.backY = backY;
+        
+    if (rep.numBackClipRects) 
+       rep.length += sizeof(XF86DRIClipRectRec) * rep.numBackClipRects;    
+
+    if (rep.numClipRects) 
+       rep.length += sizeof(XF86DRIClipRectRec) * rep.numClipRects;
+    
+    WriteToClient(client, sizeof(xXF86DRIGetDrawableInfoReply), (char *)&rep);
 
-    rep.length = 0;
     if (rep.numClipRects) {
-	rep.length = (SIZEOF(xXF86DRIGetDrawableInfoReply) - 
-		      SIZEOF(xGenericReply) +
-		      sizeof(XF86DRIClipRectRec) * rep.numClipRects);
+	WriteToClient(client,  
+		      sizeof(XF86DRIClipRectRec) * rep.numClipRects,
+		      (char *)pClipRects);
     }
 
-    WriteToClient(client, sizeof(xXF86DRIGetDrawableInfoReply), (char *)&rep);
-    if (rep.length) {
-	WriteToClient(client, rep.length, (char *)pClipRects);
+    if (rep.numBackClipRects) {
+       WriteToClient(client, 
+		     sizeof(XF86DRIClipRectRec) * rep.numBackClipRects,
+		     (char *)pBackClipRects);
     }
+
     return (client->noClientException);
 }
 
diff --git a/xc/programs/Xserver/hw/xfree86/doc/sgml/DRI.sgml b/xc/programs/Xserver/hw/xfree86/doc/sgml/DRI.sgml
index 5306e50a9..895ce09a7 100644
--- a/xc/programs/Xserver/hw/xfree86/doc/sgml/DRI.sgml
+++ b/xc/programs/Xserver/hw/xfree86/doc/sgml/DRI.sgml
@@ -7,11 +7,11 @@
 
   <article>
 
-      <title>DRI User Guide
+      <title>DRI Users Guide
       <author><htmlurl url="http://www.precisioninsight.com/"
                        name="Precision Insight, Inc.">
-      <date>20 March 2000
-        
+      <date>6 March 2000
+
     <ident>
     $XFree86: xc/programs/Xserver/hw/xfree86/doc/sgml/DRI.sgml,v 1.3 2000/03/08 05:38:41 dawes Exp $
     </ident>
@@ -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>
@@ -60,13 +60,9 @@
 	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 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">
+        XFree86 4.0;
+        it is assumed that you've already installed a Linux distribution which
+        includes XFree86 4.0.  
 
     <sect>Supported Hardware
 <p>
@@ -108,13 +104,8 @@
 <p>
         <itemize>
           <item>XFree86 4.0
-          <item>Linux kernel 2.2.x.
-            Later kernels will be supported in the near future, and may be
-            required for some chipsets.
-            The DRI modules currently in the 2.3.x kernel tree will not
-            currently work.
-            DRI updates for the 2.3.x kernel series are under development
-            at the time of writing.
+          <item>Linux kernel 2.2.x (later kernels will be supported in
+                the near future, and may be required for some chipsets)
         </itemize>
       <p>
         Mesa 3.3 (beta) is included with XFree86 4.0; there is no need to
@@ -130,8 +121,6 @@
 <p>
           Before starting the X server you must install the correct kernel
           module for your hardware.
-          In the future, this should be automatically done by the system.
-          But until then, it must be installed manually.
         <p>
           This can be done by executing the following as root:
         <verb>
@@ -140,17 +129,7 @@
         <p>
           For example, on 3dfx hardware, the kernel module is called tdfx.o
           so you you would type insmod XXX/tdfx.o
-        <p>
-          Replace XXX with the location of DRI modules on your system.
-          This may vary between distributions or depend on where you
-          have your DRI build tree if you compiled it yourself.
-        <p>
-          If you compiled the XFree86 source tree yourself, the kernel
-          modules should be in
-          <tt>$BUILDROOT/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/</tt>.
-          Another place to look is
-          <tt>/lib/modules/</tt><em>kernel-version</em><tt>/misc/</tt> or
-          <tt>/lib/modules/</tt><em>kernel-version</em><tt>/dri/</tt>.
+
         <p>
           Verify that the kernel module was installed by checking that
           /proc/dri/0 exists. 
@@ -269,19 +248,6 @@
           The documentation included with your card should have information
           about maximum screen size when using 3D.
 
-      <sect1>Start-up
-<p>
-          Start your X server in the normal way (typically startx).
-          If the DRM kernel module loaded correctly there will be a
-          message to indicate so:
-        <verb>
-        (0): [DRI] installation complete
-        </verb>
-          When the X server is running, you can use <tt>xdpyinfo</tt> to
-          check that the GLX, SGI-GLX and XFree86-DRI extensions have loaded.
-        <p>
-          Go to the troubleshooting section if this fails.
-
 
     <sect>Using 3D Acceleration
 <p>
@@ -538,13 +504,14 @@
 
           <item>
             Older Linux OpenGL applications may have been linked against
-            Mesa's GL library and will not automatically use the new libGL.so.
-            In some cases, making symbolic links will solve the problem:
+            Mesa's GL library and will not automatically use libGL.so.
+            In some cases, making symbolic links from the Mesa GL library
+            to libGL.so.1 will solve the problem:
         <verb>
         ln -s libGL.so.1 libMesaGL.so.3
         </verb>
             In other cases, the application will have to be relinked
-            against the new libGL.so from XFree86.
+            against the new XFree86 libGL.so.
             <P>
             It is reported that part of the problem is that running
             <tt/ldconfig/ will silently rewrite symbolic links based
@@ -564,42 +531,6 @@
 
       <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
-            must be installed with:
-            <verb>
-        insmod tdfx.o
-            </verb>
-          <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>
@@ -608,9 +539,6 @@
               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
@@ -633,14 +561,10 @@
               demo, may become sluggish when falling back to software
               rendering to render in that mode.
           </itemize>
-
         <sect2>Known Problems
 <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.
@@ -649,16 +573,10 @@
 
       <sect1>Intel i810
 <p>
-          Information to be added in the future.
-
       <sect1>Matrox G400
 <p>
-          Information to be added in the future.
-
       <sect1>ATI Rage 128
 <p>
-          Information to be added in the future.
-
       <sect1>3DLabs Oxygen GMX 2000
 <p>
           The driver for this hardware was experimental and is no longer being
@@ -736,9 +654,6 @@
           <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
@@ -749,8 +664,9 @@
             <htmlurl url="http://sourceforge.net/mail/?group_id=387"
             name="SourceForge"> is a forum for people to discuss DRI problems.
           <item>
-            In the future there may be IHV and Linux vendor support resources
-            for the DRI.
+            XXX IHV support?
+          <item>
+            XXX Linux distro support?
         </itemize>
 
   </article>
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 9acce4b54..b90ed5511 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_dri.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/glint/glint_dri.c
@@ -510,7 +510,8 @@ GLINTDRIScreenInit(ScreenPtr pScreen)
 	    if ((bufs = drmAddBufs(pGlint->drmSubFD,
 				   xf86ConfigDRI.bufs[i].count,
 				   xf86ConfigDRI.bufs[i].size,
-				   0 /* flags */)) <= 0) {
+				   0, /* flags */
+				   0  /* offset */)) <= 0) {
 		xf86DrvMsg(pScrn->scrnIndex, X_ERROR, 
 			   "[drm] failure adding %d %d byte DMA buffers\n",
 			   xf86ConfigDRI.bufs[i].count,
@@ -530,7 +531,8 @@ GLINTDRIScreenInit(ScreenPtr pScreen)
 	if ((bufs = drmAddBufs(pGlint->drmSubFD,
 			       GLINT_DRI_BUF_COUNT,
 			       GLINT_DRI_BUF_SIZE,
-			       0 /* flags */)) <= 0) {
+			       0, /* flags */
+			       0  /* offset */)) <= 0) {
 	    xf86DrvMsg(pScrn->scrnIndex, X_INFO,
 		       "[drm] failure adding %d %d byte DMA buffers\n",
 		       GLINT_DRI_BUF_COUNT,
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/i810/Imakefile b/xc/programs/Xserver/hw/xfree86/drivers/i810/Imakefile
index 331396f38..1a960e964 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/i810/Imakefile
+++ b/xc/programs/Xserver/hw/xfree86/drivers/i810/Imakefile
@@ -11,8 +11,8 @@ XCOMM
 #
 #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 = $(GLX_DEFINES)
@@ -63,7 +63,6 @@ InstallDriverSDKNonExecFile(i810_cursor.c,$(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 6aceea540..6bb176f9f 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/i810/i810.h
+++ b/xc/programs/Xserver/hw/xfree86/drivers/i810/i810.h
@@ -36,7 +36,7 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #ifndef _I810_H_
 #define _I810_H_
 
-
+#include "compiler.h"
 #include "xf86PciInfo.h"
 #include "xf86Pci.h"
 #include "i810_reg.h"
@@ -211,8 +211,8 @@ typedef struct _I810Rec {
 extern Bool I810DRIScreenInit(ScreenPtr pScreen);
 extern void I810DRICloseScreen(ScreenPtr pScreen);
 extern Bool I810DRIFinishScreenInit(ScreenPtr pScreen);
-extern Bool I810drmInitDma(ScrnInfoPtr pScrn);
-extern Bool I810drmCleanupDma(ScrnInfoPtr pScrn);
+extern Bool I810InitDma(ScrnInfoPtr pScrn);
+extern Bool I810CleanupDma(ScrnInfoPtr pScrn);
 
 #define I810PTR(p) ((I810Ptr)((p)->driverPrivate))
 #define I810REGPTR(p) (&(I810PTR(p)->ModeReg))
@@ -312,7 +312,7 @@ extern void I810EmitInvarientState(ScrnInfoPtr pScrn);
 /* To remove all debugging, make sure I810_DEBUG is defined as a
  * preprocessor symbol, and equal to zero.  
  */
-#define I810_DEBUG 0
+#define I810_DEBUG 0 
 #ifndef I810_DEBUG
 #warning "Debugging enabled - expect reduced performance"
 extern int I810_DEBUG;
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..5c90fc88a 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_accel.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_accel.c
@@ -301,13 +301,9 @@ 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.
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 5bb1b3191..48adc170a 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_dri.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_dri.c
@@ -20,10 +20,6 @@
 #include "i810.h"
 #include "i810_dri.h"
 
-#include "xf86drm.h"
-#include "dristruct.h"
-
-
 static char I810KernelDriverName[] = "i810";
 static char I810ClientDriverName[] = "i810";
 
@@ -62,6 +58,29 @@ static int i810_pitch_flags[] = {
     0
 };
 
+Bool I810CleanupDma(ScrnInfoPtr pScrn)
+{
+   I810Ptr pI810 = I810PTR(pScrn);
+   Bool ret_val;
+   
+   ret_val = drmI810CleanupDma(pI810->drmSubFD);
+   if(ret_val == FALSE) ErrorF("I810 Dma Cleanup Failed\n");
+   return ret_val;
+}
+
+Bool I810InitDma(ScrnInfoPtr pScrn)
+{
+   I810Ptr pI810 = I810PTR(pScrn);
+   I810RingBuffer *ring = &(pI810->LpRing);
+   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) ErrorF("I810 Dma Initialization Failed\n");
+   return ret_val;
+}
+
 static Bool
 I810InitVisualConfigs(ScreenPtr pScreen)
 {
@@ -260,8 +279,13 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
    int width = pScrn->displayWidth * pI810->cpp;
    int i;
    
-   /* ToDo : save agpHandles? */
-
+#if XFree86LOADER
+   /* Check that the GLX, DRI, and DRM modules have been loaded by testing
+    * for canonical symbols in each module. */
+   if (!LoaderSymbol("GlxSetVisualConfigs")) return FALSE;
+   if (!LoaderSymbol("DRIScreenInit"))       return FALSE;
+   if (!LoaderSymbol("drmAvailable"))        return FALSE;
+#endif     
    
    pDRIInfo = DRICreateInfoRec();
    if (!pDRIInfo) {
@@ -302,7 +326,7 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
    /* For now the mapping works by using a fixed size defined
     * in the SAREA header
     */
-   if (sizeof(XF86DRISAREARec)+sizeof(drm_i810_sarea_t)>SAREA_MAX) {
+   if (sizeof(XF86DRISAREARec)+sizeof(I810SAREARec)>SAREA_MAX) {
       ErrorF("Data does not fit in SAREA\n");
       return FALSE;
    }
@@ -378,9 +402,10 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
     * under the DRI.  
     */
 
-   dcacheHandle = drmAgpAlloc(pI810->drmSubFD, 4096 * 1024, 1, NULL);
+   drmAgpAlloc(pI810->drmSubFD, 4096 * 1024, 1, NULL, &dcacheHandle);
    pI810->dcacheHandle = dcacheHandle;
 
+   ErrorF("dcacheHandle : %p\n", dcacheHandle);
    
 #define Elements(x) sizeof(x)/sizeof(*x)
    for (pitch_idx = 0 ; pitch_idx < Elements(i810_pitches) ; pitch_idx++) 
@@ -392,10 +417,12 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
 		 "Couldn't find depth/back buffer pitch");
       DRICloseScreen(pScreen);
       return FALSE;
-   }
-   else {
-      back_size = i810_pitches[pitch_idx] * pScrn->virtualY;
+   } else {
+      back_size = i810_pitches[pitch_idx] * (pScrn->virtualY + 4);
+      ErrorF("back_size: %d\n", back_size);
+      /* Round to 4k */
       back_size = ((back_size + 4096 - 1) / 4096) * 4096;
+      ErrorF("back_size 2 : %d\n", back_size);
    }
    
    sysmem_size = pScrn->videoRam * 1024;
@@ -411,6 +438,7 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
    } else {
       sysmem_size = sysmem_size - 2*back_size;
    }
+   ErrorF("sysmem_size : %d\n", sysmem_size);
    
    sysmem_size -= 4096;
    if(sysmem_size > ((48*1024*1024) - 1) ) {
@@ -428,6 +456,8 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
 
    if (dcacheHandle != 0) {
       /* The Z buffer is always aligned to the 48 mb mark in the aperture */
+      ErrorF("dcacheHandle : %p, aligned to : %lx\n",
+	     dcacheHandle, 48*1024*1024);
 
       if(drmAgpBind(pI810->drmSubFD, dcacheHandle, 48*1024*1024) == 0) {
 	 xf86memset (&pI810->DcacheMem, 0, sizeof(I810MemRange));
@@ -454,11 +484,12 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
       xf86DrvMsg(pScrn->scrnIndex, X_INFO, "GART: no dcache memory found\n");
    }
 
-   agpHandle = drmAgpAlloc(pI810->drmSubFD, back_size, 0, NULL);
+   drmAgpAlloc(pI810->drmSubFD, back_size, 0, NULL, &agpHandle);
    pI810->backHandle = agpHandle;
    
    if(agpHandle != 0) {
       /* The backbuffer is always aligned to the 56 mb mark in the aperture */
+      ErrorF("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");
@@ -481,11 +512,14 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
       
    if(dcacheHandle == 0) {
       /* The Z buffer is always aligned to the 48 mb mark in the aperture */
-      agpHandle = drmAgpAlloc(pI810->drmSubFD, back_size, 0,
-			      NULL);
+      drmAgpAlloc(pI810->drmSubFD, back_size, 0,
+		  NULL, &agpHandle);
       pI810->zHandle = agpHandle;
 
       if(agpHandle != 0) {
+	 ErrorF("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;
@@ -510,7 +544,7 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
     * regular framebuffer as well as texture memory.
     */
    
-   agpHandle = drmAgpAlloc(pI810->drmSubFD, sysmem_size, 0, NULL);
+   drmAgpAlloc(pI810->drmSubFD, sysmem_size, 0, NULL, &agpHandle);
    if(agpHandle == 0) {
       ErrorF("drmAgpAlloc failed\n");
       DRICloseScreen(pScreen);
@@ -524,8 +558,8 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
       return FALSE;
    }
    
-   agpHandle = drmAgpAlloc(pI810->drmSubFD, 4096, 2, 
-			   (unsigned long *)&pI810->CursorPhysical); 
+   drmAgpAlloc(pI810->drmSubFD, 4096, 2, 
+	       (unsigned long *)&pI810->CursorPhysical, &agpHandle); 
    pI810->cursorHandle = agpHandle;
 
    if (agpHandle != 0) {
@@ -536,6 +570,7 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
 		    "GART: Allocated 4K for mouse cursor image\n");
 	 pI810->CursorStart = tom;	 
 	 tom += 4096;
+	 ErrorF("tom : %lx\n", tom);
       } else {
 	 xf86DrvMsg(pScrn->scrnIndex, X_INFO, "GART: cursor bind failed\n");
 	 pI810->CursorPhysical = 0;    
@@ -545,16 +580,19 @@ 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];
@@ -562,6 +600,7 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
    pI810->SavedDcacheMem = pI810->DcacheMem;
    
    pI810DRI->backbufferSize = pI810->BackBuffer.Size;
+   ErrorF("Backbuffer map : %lx\n", pI810->BackBuffer.Start);
    if (drmAddMap(pI810->drmSubFD, (drmHandle)pI810->BackBuffer.Start,
 		 pI810->BackBuffer.Size, DRM_AGP, 0, 
 		 &pI810DRI->backbuffer) < 0) {
@@ -570,6 +609,7 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
       return FALSE;
    }
    
+   ErrorF("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, 
@@ -586,6 +626,15 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
    I810AllocHigh( &(pI810->BufferMem), &(pI810->SysMem), 
 		  I810_DMA_BUF_NR * I810_DMA_BUF_SZ);
    
+   ErrorF("Buffer map : %lx\n", pI810->BufferMem.Start);
+   
+   if(pI810->BufferMem.Start == 0 || 
+      pI810->BufferMem.End - pI810->BufferMem.Start > 
+      I810_DMA_BUF_NR * I810_DMA_BUF_SZ) {
+      ErrorF("Not enough memory for dma buffers\n");
+      DRICloseScreen(pScreen);
+      return FALSE;
+   }
    if(drmAddMap(pI810->drmSubFD, (drmHandle)pI810->BufferMem.Start,
 		pI810->BufferMem.Size, DRM_AGP, 0,
 		&pI810->buffer_map) < 0) {
@@ -597,6 +646,7 @@ 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) {
@@ -608,6 +658,7 @@ 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);
 
@@ -617,6 +668,7 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
    pI810DRI->logTextureGranularity = i;
    pI810DRI->textureSize = (pI810DRI->textureSize >> i) << i; /* truncate */
 
+   ErrorF("REDUCED texsize: %x (i: %d)\n", pI810DRI->textureSize, i);
 
    if(pI810DRI->textureSize < 512*1024) {
       ErrorF("Less then 512k for textures\n");
@@ -625,8 +677,8 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
    
    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) {
@@ -635,6 +687,7 @@ 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,
@@ -650,7 +703,7 @@ Bool I810DRIScreenInit(ScreenPtr pScreen)
    xf86DrvMsg(pScrn->scrnIndex, X_INFO, "[drm] added %d %d byte DMA buffers\n",
 	      bufs, I810_DMA_BUF_SZ);
 
-   I810drmInitDma(pScrn);
+   I810InitDma(pScrn);
    
    /* Okay now initialize the dma engine */
 
@@ -718,7 +771,7 @@ I810DRICloseScreen(ScreenPtr pScreen)
    ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
    I810Ptr pI810 = I810PTR(pScrn);
 
-   I810drmCleanupDma(pScrn);
+   I810CleanupDma(pScrn);
 
    if(pI810->dcacheHandle) drmAgpFree(pI810->drmSubFD, pI810->dcacheHandle);
    if(pI810->backHandle) drmAgpFree(pI810->drmSubFD, pI810->backHandle);
@@ -768,7 +821,7 @@ I810DestroyContext(ScreenPtr pScreen, drmContext hwContext,
 Bool
 I810DRIFinishScreenInit(ScreenPtr pScreen)
 {
-   drm_i810_sarea_t *sPriv = (drm_i810_sarea_t *)DRIGetSAREAPrivate(pScreen);
+   I810SAREARec *sPriv = (I810SAREARec *)DRIGetSAREAPrivate(pScreen);
    xf86memset( sPriv, 0, sizeof(sPriv) );
    return DRIFinishScreenInit(pScreen);
 }
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_dri.h b/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_dri.h
index e593f36b2..506227222 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_dri.h
+++ b/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_dri.h
@@ -4,7 +4,7 @@
 #define _I810_DRI_
 
 #include <xf86drm.h>
-#include "i810_drm_public.h"
+#include <xf86drmI810.h>
 
 #define I810_MAX_DRAWABLES 256
 
@@ -53,6 +53,45 @@ typedef struct {
 
 } I810DRIRec, *I810DRIPtr;
 
+/* WARNING: Do not change the SAREA structure without changing the kernel
+ * as well */
+
+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 */
+} I810TexRegionRec, *I810TexRegionPtr;
+
+typedef struct {
+   unsigned int nbox;
+   XF86DRIClipRectRec 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.  
+    */
+   I810TexRegionRec 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 */
+} I810SAREARec, *I810SAREAPtr;
+
 typedef struct {
   /* Nothing here yet */
   int dummy;
@@ -63,5 +102,4 @@ typedef struct {
   int dummy;
 } I810DRIContextRec, *I810DRIContextPtr;
 
-
 #endif
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_dripriv.h b/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_dripriv.h
deleted file mode 100644
index b4bbab133..000000000
--- a/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_dripriv.h
+++ /dev/null
@@ -1,24 +0,0 @@
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/i810/i810_dripriv.h,v 1.1 2000/03/02 16:07:50 martin Exp $ */
-
-#ifndef _I810_DRIPRIV_H_
-#define _I810_DRIPRIV_H_
-
-#define I810_MAX_DRAWABLES 256
-
-extern void GlxSetVisualConfigs(
-    int nconfigs,
-    __GLXvisualConfig *configs,
-    void **configprivs
-);
-
-typedef struct {
-  /* Nothing here yet */
-  int dummy;
-} I810ConfigPrivRec, *I810ConfigPrivPtr;
-
-typedef struct {
-  /* Nothing here yet */
-  int dummy;
-} I810DRIContextRec, *I810DRIContextPtr;
-
-#endif
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 e489d17c9..e43685559 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_driver.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_driver.c
@@ -207,31 +207,19 @@ static const char *ramdacSymbols[] = {
 
 #ifdef XF86DRI
 static const char *drmSymbols[] = {
+   "drmAvailable",
    "drmAddBufs",
    "drmAddMap",
-   "drmAvailable",
-   "drmCtlAddCommand",
    "drmCtlInstHandler",
    "drmGetInterruptFromBusID",
-   "drmMapBufs",
-   "drmMarkBufs",
-   "drmUnmapBufs",
    "drmAgpAcquire",
    "drmAgpRelease",
    "drmAgpEnable",
    "drmAgpAlloc",
    "drmAgpFree",
    "drmAgpBind",
-   "drmAgpUnbind",
-   "drmAgpVersionMajor",
-   "drmAgpVersionMinor",
-   "drmAgpGetMode",
-   "drmAgpBase",
-   "drmAgpSize",
-   "drmAgpMemoryUsed",
-   "drmAgpMemoryAvail",
-   "drmAgpVendorId",
-   "drmAgpDeviceId",
+   "drmI810CleanupDma",
+   "drmI810InitDma",
    NULL
 };
 
@@ -1551,9 +1539,16 @@ 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)
 	    ErrorF( "ring buffer at local %lx\n", 
 		    pI810->LpRing.mem.Start);
@@ -1607,6 +1602,7 @@ I810ScreenInit(int scrnIndex, ScreenPtr pScreen, int argc, char **argv) {
    /* Have to init the DRM earlier than in other drivers to get agp
     * memory.  Wonder if this is going to be a problem...
     */
+
 #ifdef XF86DRI
    /*
     * Setup DRI after visuals have been established, but before cfbScreenInit
@@ -1625,7 +1621,7 @@ I810ScreenInit(int scrnIndex, ScreenPtr pScreen, int argc, char **argv) {
      return FALSE;
    I810AllocateFront(pScrn);
 #endif
-   
+
    if (!I810MapMem(pScrn)) return FALSE;
 
    pScrn->memPhysBase = (int)pI810->FbBase;
@@ -1729,6 +1725,7 @@ I810ScreenInit(int scrnIndex, ScreenPtr pScreen, int argc, char **argv) {
       }
    }
 #endif
+   
 #ifdef XF86DRI
    if (!pI810->directRenderingEnabled) {
       pI810->DoneFrontAlloc = FALSE;
@@ -1738,6 +1735,14 @@ 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, 
 		 "Failed to init memory manager\n");
@@ -1762,7 +1767,7 @@ I810ScreenInit(int scrnIndex, ScreenPtr pScreen, int argc, char **argv) {
       }
    }
 
-#if XF86DRI
+#ifdef XF86DRI
    if (pI810->LpRing.mem.Start == 0) {
       pI810->directRenderingEnabled = 0;
       I810DRICloseScreen(pScreen);
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_sarea.h b/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_sarea.h
deleted file mode 100644
index 0f70d3535..000000000
--- a/xc/programs/Xserver/hw/xfree86/drivers/i810/i810_sarea.h
+++ /dev/null
@@ -1,6 +0,0 @@
-#ifndef I810_SAPRIV_H
-#define I810_SAPRIV_H
-
-#include "i810_drm_public.h"
-
-#endif
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/mga/Imakefile b/xc/programs/Xserver/hw/xfree86/drivers/mga/Imakefile
index e9a42442d..c11f6a457 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/mga/Imakefile
+++ b/xc/programs/Xserver/hw/xfree86/drivers/mga/Imakefile
@@ -1,4 +1,4 @@
-XCOMM $XFree86: xc/programs/Xserver/hw/xfree86/drivers/mga/Imakefile,v 1.26 1999/08/21 13:48:34 dawes Exp $
+XCOMM $XFree86: xc/programs/Xserver/hw/xfree86/drivers/mga/Imakefile,v 1.32 2000/02/28 19:53:12 alanh Exp $
 XCOMM
 XCOMM This is an Imakefile for the MGA driver.  
 XCOMM
@@ -6,26 +6,37 @@ XCOMM
 #define IHaveModules
 #include <Server.tmpl>
 
+#if BuildXF86DRI
+DRISRCS = mga_dri.c mga_warp.c mga_wrap.c
+DRIOBJS = mga_dri.o mga_warp.o mga_wrap.o
+DRIINCLUDES = -I$(SERVERSRC)/GL/dri -I$(LIBSRC)/GL/dri \
+	-I$(XF86OSSRC)/linux/drm/kernel
+DRIDEFINES = $(GLX_DEFINES)
+#endif
+
 SRCS = mga_driver.c mga_hwcurs.c /* mga_cmap.c */ mga_dac3026.c mga_dacG.c \
        mga_storm8.c mga_storm16.c mga_storm24.c mga_storm32.c mga_arc.c \
-       mga_dga.c mga_shadow.c
+       mga_dga.c mga_shadow.c mga_video.c $(DRISRCS)
 OBJS = mga_driver.o mga_hwcurs.o /* mga_cmap.o */ mga_dac3026.o mga_dacG.o \
        mga_storm8.o mga_storm16.o mga_storm24.o mga_storm32.o mga_arc.o \
-       mga_dga.o mga_shadow.o
+       mga_dga.o mga_shadow.o mga_video.o $(DRIOBJS)
 
 #if defined(XF86DriverSDK)
 INCLUDES = -I. -I../../include
 #else
 INCLUDES = -I. -I$(XF86COMSRC) -I$(XF86OSSRC) \
-           -I$(SERVERSRC)/mfb -I$(SERVERSRC)/mi \
+           -I$(SERVERSRC)/mfb -I$(SERVERSRC)/mi -I$(XF86SRC)/int10 \
            -I$(SERVERSRC)/cfb -I$(XF86SRC)/xaa -I$(XF86SRC)/rac \
            -I$(XF86SRC)/vgahw -I$(XF86SRC)/fbdevhw \
            -I$(XF86SRC)/ramdac -I$(XF86SRC)/ddc -I$(XF86SRC)/i2c \
            -I$(SERVERSRC)/Xext -I$(XF86SRC)/xf8_32bpp\
            -I$(FONTINCSRC) -I$(SERVERSRC)/include -I$(XINCLUDESRC) \
-           -I$(XF86SRC)/xf24_32bpp -I$(XF86SRC)/shadowfb -I$(EXTINCSRC)
+           -I$(XF86SRC)/xf24_32bpp -I$(XF86SRC)/shadowfb -I$(EXTINCSRC) \
+	   -I$(XF86OSSRC)/vbe $(DRIINCLUDES)
 #endif
 
+DEFINES = $(DRIDEFINES)
+
 #if MakeHasPosixVariableSubstitutions
 SubdirLibraryRule($(OBJS))
 #endif
@@ -62,6 +73,7 @@ InstallDriverSDKNonExecFile(mga_map.h,$(DRIVERSDKDIR)/drivers/mga)
 InstallDriverSDKNonExecFile(mga_reg.h,$(DRIVERSDKDIR)/drivers/mga)
 InstallDriverSDKNonExecFile(mga_shadow.c,$(DRIVERSDKDIR)/drivers/mga)
 InstallDriverSDKNonExecFile(mga_storm.c,$(DRIVERSDKDIR)/drivers/mga)
+InstallDriverSDKNonExecFile(mga_video.c,$(DRIVERSDKDIR)/drivers/mga)
 
 InstallDriverSDKObjectModule(mga,$(DRIVERSDKMODULEDIR),drivers)
 
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga.h b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga.h
index bc927c8a4..5e0430616 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga.h
+++ b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga.h
@@ -1,4 +1,4 @@
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/mga/mga.h,v 1.47 1999/08/22 05:57:33 dawes Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/mga/mga.h,v 1.57 2000/02/27 02:50:47 mvojkovi Exp $ */
 /*
  * MGA Millennium (MGA2064W) functions
  *
@@ -14,21 +14,32 @@
 #ifndef MGA_H
 #define MGA_H
 
+#include "compiler.h"
 #include "xaa.h"
 #include "xf86Cursor.h"
 #include "vgaHW.h"
 #include "colormapst.h"
 #include "xf86DDC.h"
+#include "xf86xv.h"
 
-#if defined(__alpha__)
-#define INREG8(addr) xf86ReadSparse8(pMga->IOBase, (addr))
-#define INREG16(addr) xf86ReadSparse16(pMga->IOBase, (addr))
-#define INREG(addr) xf86ReadSparse32(pMga->IOBase, (addr))
-#define OUTREG8(addr,val) xf86WriteSparse8((val),pMga->IOBase,(addr))
-#define OUTREG16(addr,val) xf86WriteSparse16((val),pMga->IOBase,(addr))
-#define OUTREG(addr, val) xf86WriteSparse32((val),pMga->IOBase,(addr))
-#else /* __alpha__ */
-#if defined(EXTRADEBUG)
+#ifdef XF86DRI
+#include "xf86drm.h"
+#include "sarea.h"
+#define _XF86DRI_SERVER_
+#include "xf86dri.h"
+#include "dri.h"
+#include "GL/glxint.h"
+#include "mga_dri.h"
+#endif
+
+#if !defined(EXTRADEBUG)
+#define INREG8(addr) MMIO_IN8(pMga->IOBase, addr)
+#define INREG16(addr) MMIO_IN16(pMga->IOBase, addr)
+#define INREG(addr) MMIO_IN32(pMga->IOBase, addr)
+#define OUTREG8(addr, val) MMIO_OUT8(pMga->IOBase, addr, val)
+#define OUTREG16(addr, val) MMIO_OUT16(pMga->IOBase, addr, val)
+#define OUTREG(addr, val) MMIO_OUT32(pMga->IOBase, addr, val)
+#else /* !EXTRADEBUG */
 CARD8 dbg_inreg8(ScrnInfoPtr,int,int);
 CARD16 dbg_inreg16(ScrnInfoPtr,int,int);
 CARD32 dbg_inreg32(ScrnInfoPtr,int,int);
@@ -41,15 +52,7 @@ void dbg_outreg32(ScrnInfoPtr,int,int);
 #define OUTREG8(addr,val) dbg_outreg8(pScrn,addr,val)
 #define OUTREG16(addr,val) dbg_outreg16(pScrn,addr,val)
 #define OUTREG(addr,val) dbg_outreg32(pScrn,addr,val)
-#else /* EXTRADEBUG */
-#define INREG8(addr) *(volatile CARD8 *)(pMga->IOBase + (addr))
-#define INREG16(addr) *(volatile CARD16 *)(pMga->IOBase + (addr))
-#define INREG(addr) *(volatile CARD32 *)(pMga->IOBase + (addr))
-#define OUTREG8(addr, val) *(volatile CARD8 *)(pMga->IOBase + (addr)) = (val)
-#define OUTREG16(addr, val) *(volatile CARD16 *)(pMga->IOBase + (addr)) = (val)
-#define OUTREG(addr, val) *(volatile CARD32 *)(pMga->IOBase + (addr)) = (val)
 #endif /* EXTRADEBUG */
-#endif /* __alpha__ */
 
 #define PORT_OFFSET 	(0x1F00 - 0x300)
 
@@ -62,6 +65,7 @@ typedef struct {
     CARD32		Option3;
 } MGARegRec, *MGARegPtr;
 
+
 typedef struct {
     Bool	isHwCursor;
     int		CursorMaxWidth;
@@ -93,6 +97,7 @@ typedef struct {
     int displayWidth;
     rgb weight;
     Bool Overlay8Plus24;
+    DisplayModePtr mode;
 } MGAFBLayout;
 
 
@@ -133,9 +138,7 @@ typedef struct {
     CARD32		BiosAddress;
     MessageType		BiosFrom;
     unsigned char *     IOBase;
-#ifdef __alpha__
     unsigned char *     IOBaseDense;
-#endif
     unsigned char *	FbBase;
     unsigned char *	ILOADBase;
     unsigned char *	FbStart;
@@ -170,6 +173,8 @@ typedef struct {
     CARD32		PlaneMask;
     CARD32		FgColor;
     CARD32		BgColor;
+    CARD32		MAccess;
+    int			FifoSize;
     int			StyleLen;
     XAAInfoRecPtr	AccelInfoRec;
     xf86CursorInfoPtr	CursorInfoRec;
@@ -185,19 +190,43 @@ typedef struct {
     Bool		(*ModeInit)(ScrnInfoPtr, DisplayModePtr);
     void		(*PointerMoved)(int index, int x, int y);
     CloseScreenProcPtr	CloseScreen;
+    ScreenBlockHandlerProcPtr BlockHandler;
     unsigned int	(*ddc1Read)(ScrnInfoPtr);
     void (*DDC1SetSpeed)(ScrnInfoPtr, xf86ddcSpeed);
     Bool		(*i2cInit)(ScrnInfoPtr);
     I2CBusPtr		I2C;
     Bool		FBDev;
     int			colorKey;
+    int			videoKey;
     int			fifoCount;
     int			Rotate;
     MGAFBLayout		CurrentLayout;
     Bool		DrawTransparent;
     int			MaxBlitDWORDS;
+
+#ifdef XF86DRI
+   Bool directRenderingEnabled;
+   DRIInfoPtr pDRIInfo;
+   int drmSubFD;
+   int numVisualConfigs;
+   __GLXvisualConfig* pVisualConfigs;
+   MGAConfigPrivPtr pVisualConfigsPriv;
+   MGARegRec DRContextRegs;
+   MGADRIServerPrivatePtr  DRIServerInfo;
+   Bool have_quiescense;
+#endif
+
+    XF86VideoAdaptorPtr adaptor;
 } MGARec, *MGAPtr;
 
+#ifdef XF86DRI
+extern void GlxSetVisualConfigs(
+    int nconfigs,
+    __GLXvisualConfig *configs,
+    void **configprivs
+);
+#endif
+
 extern CARD32 MGAAtype[16];
 extern CARD32 MGAAtypeNoBLK[16];
 
@@ -211,11 +240,17 @@ extern CARD32 MGAAtypeNoBLK[16];
 #define	MGA_NO_PLANEMASK	0x00000080
 #define USE_LINEAR_EXPANSION	0x00000100
 #define LARGE_ADDRESSES		0x00000200
-
+#define MGAIOMAPSIZE		0x00004000
+#define MGAILOADMAPSIZE		0x00400000
 
 #define TRANSPARENCY_KEY	255
 #define KEY_COLOR		0
 
+#define MGA_FRONT	0x1
+#define MGA_BACK	0x2
+#define MGA_DEPTH	0x4
+
+
 /* Prototypes */
 
 void MGAAdjustFrame(int scrnIndex, int x, int y, int flags);
@@ -238,6 +273,18 @@ void MGAPolyArcThinSolid(DrawablePtr, GCPtr, int, xArc*);
 
 Bool MGADGAInit(ScreenPtr pScreen);
 
+Bool MGADRIScreenInit(ScreenPtr pScreen);
+void MGADRICloseScreen(ScreenPtr pScreen);
+Bool MGADRIFinishScreenInit(ScreenPtr pScreen);
+void MGASwapContext(ScreenPtr pScreen);
+void MGASelectBuffer(ScrnInfoPtr pScrn, int which);
+Bool mgaConfigureWarp(ScrnInfoPtr pScrn);
+unsigned int mgaInstallMicrocode(ScreenPtr pScreen, int agp_offset);
+unsigned int mgaGetMicrocodeSize(ScreenPtr pScreen);
+Bool MgaCleanupDma(ScrnInfoPtr pScrn);
+Bool MgaInitDma(ScrnInfoPtr pScrn, int prim_size);
+Bool MgaLockUpdate(ScrnInfoPtr pScrn, drmLockFlags flags);
+
 void MGARefreshArea(ScrnInfoPtr pScrn, int num, BoxPtr pbox);
 void MGARefreshArea8(ScrnInfoPtr pScrn, int num, BoxPtr pbox);
 void MGARefreshArea16(ScrnInfoPtr pScrn, int num, BoxPtr pbox);
@@ -268,4 +315,7 @@ void Mga32SetupForSolidFill(ScrnInfoPtr pScrn, int color, int rop,
 
 void MGAPointerMoved(int index, int x, int y);
 
+void MGAInitVideo(ScreenPtr pScreen);
+void MGAResetVideo(ScrnInfoPtr pScrn); 
+
 #endif
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_arc.c b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_arc.c
index 2844302ff..993d763a8 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_arc.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_arc.c
@@ -69,6 +69,8 @@ MGAZeroArc(
     Bool do360;
     DDXPointRec org, orgo;
 
+    CHECK_DMA_QUIESCENT( pMga, infoRec->pScrn );
+
     (*infoRec->SetupForSolidFill)(infoRec->pScrn, pGC->fgPixel, 
 		pGC->alu, pGC->planemask);
 
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dac3026.c b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dac3026.c
index f9a07ade3..fb7952211 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dac3026.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dac3026.c
@@ -1,4 +1,4 @@
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dac3026.c,v 1.48 1999/08/21 13:48:35 dawes Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dac3026.c,v 1.52 2000/01/21 01:12:20 dawes Exp $ */
 /*
  * Copyright 1994 by Robin Cutshaw <robin@XFree86.org>
  *
@@ -49,6 +49,7 @@
 
 #include "mga_bios.h"
 #include "mga_reg.h"
+#include "mga_macros.h"
 #include "mga.h"
 
 #include "xf86DDC.h"
@@ -85,17 +86,33 @@ const static unsigned char MGADACregs[] = {
 	0x06
 };
 
+/* note: to fix a cursor hw glitch, register 0x37 (blue color key) needs
+   to be set to magic numbers, even though they are "never" used because 
+   blue keying disabled in 0x38.
+
+   Matrox sez:
+
+   ...The more precise statement of the software workaround is to insure
+   that bits 7-5 of register 0x37 (Blue Color Key High) and bits 7-5 of
+   register 0x38 (HZOOM)are the same...
+*/
+
+/* also note: the values of the MUX control register 0x19 (index [2]) can be
+   found in table 2-17 of the 3026 manual. If interlace is set, the values
+   listed here are incremented by one. 
+*/
+   
 #define DACREGSIZE sizeof(MGADACregs)
 /*
  * initial values of ti3026 registers
  */
 const static unsigned char MGADACbpp8[DACREGSIZE] = {
-	0x06, 0x80, 0x48, 0x25, 0x00,   0x00, 0x0C, 0x00, 0x1E, 0xFF,
+	0x06, 0x80, 0x4b, 0x25, 0x00,   0x00, 0x0C, 0x00, 0x1E, 0xFF,
 	0xFF, 0xFF, 0xFF, 0xFF, 0xFF,   0xFF, 0x00, 0x00,    0, 0x00,
 	0x00
 };
 const static unsigned char MGADACbpp16[DACREGSIZE] = {
-	0x07, 0x45, 0x50, 0x15, 0x00,   0x00, 0x2C, 0x00, 0x1E, 0xFF,
+	0x07, 0x45, 0x53, 0x15, 0x00,   0x00, 0x2C, 0x00, 0x1E, 0xFF,
 	0xFF, 0xFF, 0xFF, 0xFF, 0xFF,   0xFF, 0x00, 0x10,    0, 0x00,
 	0x00
 };
@@ -103,18 +120,29 @@ const static unsigned char MGADACbpp16[DACREGSIZE] = {
  * [0] value was 0x07, but changed to 0x06 by Doug Merrit to fix high res
  * stripe glitches and clock glitches at 24bpp.
  */
+/* [0] value is now set inside of MGA3026Init, based on the silicon revision
+   It is still set to 7 or 6 based on the revision, though, since setting to
+   8 as in the documentation makes (some?) revB chips get the colors wrong...
+   maybe BGR instead of RGB? This only applies to 24bpp, since it is the only
+   one documented as depending on revision.
+ */
+
 const static unsigned char MGADACbpp24[DACREGSIZE] = {
-	0x06, 0x56, 0x58, 0x25, 0x00,   0x00, 0x2C, 0x00, 0x1E, 0xFF,
+	0x06, 0x56, 0x5b, 0x25, 0x00,   0x00, 0x2C, 0x00, 0x1E, 0xFF,
 	0xFF, 0xFF, 0xFF, 0xFF, 0xFF,   0xFF, 0x00, 0x10,    0, 0x00,
 	0x00
 };
 const static unsigned char MGADACbpp32[DACREGSIZE] = {
-	0x07, 0x46, 0x58, 0x05, 0x00,   0x00, 0x2C, 0x00, 0x1E, 0xFF,
+	0x07, 0x46, 0x5b, 0x05, 0x00,   0x00, 0x2C, 0x00, 0x1E, 0xFF,
 	0xFF, 0xFF, 0xFF, 0xFF, 0xFF,   0xFF, 0x00, 0x10,    0, 0x00,
 	0x00
 };
+
+/* on at least some 2064Ws, the PSEL line flips at 4MB or so, so PSEL keying
+   has to be off in register 0x1e -> bit4 clear */
+   
 const static unsigned char MGADACbpp8plus24[DACREGSIZE] = {
-	0x07, 0x06, 0x58, 0x05, 0x00,   0x00, 0x3C, 0x00, 0x1E, 0xFF, 
+	0x07, 0x06, 0x5b, 0x05, 0x00,   0x00, 0x2C, 0x00, 0x1E, 0xFF, 
 	0xFF, 0xFF, 0xFF, 0xFF, 0xFF,   0xFF, 0x00, 0x01, 0x00, 0x00, 
 	0x00
 };
@@ -523,7 +551,31 @@ MGA3026Init(ScrnInfoPtr pScrn, DisplayModePtr mode)
 	    && (pLayout->weight.green == 5) && (pLayout->weight.blue == 5) ) {
 	    pReg->DacRegs[1] &= ~0x01;
 	}
-	pReg->DacRegs[2] |= pMga->Interleave? 0x04 : 0x03;	
+	if (pMga->Interleave ) 	pReg->DacRegs[2] += 1;
+
+
+	if ( pLayout->bitsPerPixel == 24 ) {
+	  int silicon_rev;
+	  /* we need to set DacRegs[0] differently based on the silicon
+	   * revision of the 3026 RAMDAC, as per page 2-14 of tvp3026.pdf.
+	   * If we have rev A silicon, we want 0x07; rev B silicon wants
+	   * 0x06.
+	   */
+	  silicon_rev = inTi3026(TVP3026_SILICON_REV);
+	  
+#ifdef DEBUG
+	    ErrorF("TVP3026 revision 0x%x (rev %s)\n",
+		   silicon_rev, (silicon_rev <= 0x20) ? "A" : "B");
+#endif
+	    
+	    if(silicon_rev <= 0x20) {
+	      /* rev A */
+	      pReg->DacRegs[0] = 0x07;
+	    } else {
+	      /* rev B */
+	      pReg->DacRegs[0] = 0x06;
+	    }
+	}
 
 	/*
 	 * This will initialize all of the generic VGA registers.
@@ -652,10 +704,9 @@ MGA3026Init(ScrnInfoPtr pScrn, DisplayModePtr mode)
 	else
 	  pReg->Option &= ~0x1000;
 
-        if(pMga->UsePCIRetry)
-	    pReg->Option &= ~0x20000000;
-	else
-	    pReg->Option |= 0x20000000;
+	/* must always have the pci retries on but rely on 
+	   polling to keep them from occuring */
+	pReg->Option &= ~0x20000000;
 
 	pVga->MiscOutReg |= 0x0C; 
 	/* XXX Need to check the first argument */
@@ -694,6 +745,8 @@ MGA3026Restore(ScrnInfoPtr pScrn, vgaRegPtr vgaReg, MGARegPtr mgaReg,
 	int i;
 	MGAPtr pMga = MGAPTR(pScrn);
 
+	CHECK_DMA_QUIESCENT( pMga, pScrn );
+
 	/*
 	 * Code is needed to get things back to bank zero.
 	 */
@@ -771,6 +824,8 @@ MGA3026Save(ScrnInfoPtr pScrn, vgaRegPtr vgaReg, MGARegPtr mgaReg,
 	int i;
 	MGAPtr pMga = MGAPTR(pScrn);
 	
+	CHECK_DMA_QUIESCENT( pMga, pScrn );
+
 	/* Allocate the DacRegs space if not done already */
 	if (mgaReg->DacRegs == NULL) {
 		mgaReg->DacRegs = xnfcalloc(DACREGSIZE, 1);
@@ -1034,14 +1089,29 @@ MGA3026RamdacInit(ScrnInfoPtr pScrn)
 	    break;
 	}
 	/* Set MCLK based on amount of memory */
-    	if ( pScrn->videoRam < 4096 )
+	if(pMga->OverclockMem) {
+	  if ( pScrn->videoRam < 4096 )
+            MGAdac->MemoryClock = pMga->Bios.ClkBase * 12;
+	  else if ( pScrn->videoRam < 8192 )
+            MGAdac->MemoryClock = pMga->Bios.Clk4MB * 12;
+	  else
+            MGAdac->MemoryClock = pMga->Bios.Clk8MB * 12;
+	  MGAdac->MemClkFrom = X_CONFIG;
+	  MGAdac->SetMemClk = TRUE;
+#if 0
+	  ErrorF("BIOS Memory clock settings: 2Mb %d, 4Mb %d, 8MB %d\n",
+		 pMga->Bios.ClkBase, pMga->Bios.Clk4MB, pMga->Bios.Clk8MB);
+#endif
+	} else {
+	  if ( pScrn->videoRam < 4096 )
             MGAdac->MemoryClock = pMga->Bios.ClkBase * 10;
-   	else if ( pScrn->videoRam < 8192 )
+	  else if ( pScrn->videoRam < 8192 )
             MGAdac->MemoryClock = pMga->Bios.Clk4MB * 10;
-    	else
+	  else
             MGAdac->MemoryClock = pMga->Bios.Clk8MB * 10;
-    	MGAdac->MemClkFrom = X_PROBED;
-    	MGAdac->SetMemClk = TRUE;
+	  MGAdac->MemClkFrom = X_PROBED;
+	  MGAdac->SetMemClk = TRUE;
+	}
     }
     else
     {
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dacG.c b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dacG.c
index 1cfa03f31..d81d12b3a 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dacG.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dacG.c
@@ -2,7 +2,7 @@
  * MGA-1064, MGA-G100, MGA-G200, MGA-G400 RAMDAC driver
  */
 
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dacG.c,v 1.30 1999/08/22 05:57:33 dawes Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dacG.c,v 1.33 1999/12/03 04:03:52 mvojkovi Exp $ */
 
 /*
  * This is a first cut at a non-accelerated version to work with the
@@ -24,6 +24,7 @@
 
 #include "mga_bios.h"
 #include "mga_reg.h"
+#include "mga_macros.h"
 #include "mga.h"
 
 #include "xf86DDC.h"
@@ -330,24 +331,40 @@ MGAGInit(ScrnInfoPtr pScrn, DisplayModePtr mode)
 		pReg->Option2 = 0x0000007;
 		break;
 	case PCI_CHIP_MGAG400:
-	        if(pMga->OverclockMem) {
-		    /* 166 Mhz but faster graphics engine clock */
-		    pReg->DacRegs[ MGA1064_SYS_PLL_M ] = 0x13;
-		    pReg->DacRegs[ MGA1064_SYS_PLL_N ] = 0x7A;
-		    pReg->DacRegs[ MGA1064_SYS_PLL_P ] = 0x08;
-		    pReg->Option3 = 0x0190a421;
+	       if(pMga->Dac.maxPixelClock == 360000) {  /* G400 MAX */
+	           if(pMga->OverclockMem) {
+			/* 150/200  */
+			pReg->DacRegs[ MGA1064_SYS_PLL_M ] = 0x05;
+			pReg->DacRegs[ MGA1064_SYS_PLL_N ] = 0x42;
+			pReg->DacRegs[ MGA1064_SYS_PLL_P ] = 0x18;
+			pReg->Option3 = 0x019B8419;
+			pReg->Option = 0x50574120;
+		   } else {
+			/* 125/166  */
+			pReg->DacRegs[ MGA1064_SYS_PLL_M ] = 0x02;
+			pReg->DacRegs[ MGA1064_SYS_PLL_N ] = 0x1B;
+			pReg->DacRegs[ MGA1064_SYS_PLL_P ] = 0x18;
+			pReg->Option3 = 0x019B8419;
+			pReg->Option = 0x5053C120;
+		   } 
 		} else {
-		    /* 165 Mhz */
-		    pReg->DacRegs[ MGA1064_SYS_PLL_M ] = 0x09;
-		    pReg->DacRegs[ MGA1064_SYS_PLL_N ] = 0x3C;
-		    pReg->DacRegs[ MGA1064_SYS_PLL_P ] = 0x10;
-		    pReg->Option3 = 0x0190a419;
-		} 
+	           if(pMga->OverclockMem) {
+			/* 125/166  */
+			pReg->DacRegs[ MGA1064_SYS_PLL_M ] = 0x02;
+			pReg->DacRegs[ MGA1064_SYS_PLL_N ] = 0x1B;
+			pReg->DacRegs[ MGA1064_SYS_PLL_P ] = 0x18;
+			pReg->Option3 = 0x019B8419;
+			pReg->Option = 0x5053C120;
+		   } else {
+			/* 110/166  */
+			pReg->DacRegs[ MGA1064_SYS_PLL_M ] = 0x13;
+			pReg->DacRegs[ MGA1064_SYS_PLL_N ] = 0x7A;
+			pReg->DacRegs[ MGA1064_SYS_PLL_P ] = 0x08;
+			pReg->Option3 = 0x0190a421;
+			pReg->Option = 0x50044120;
+		   } 
+		}
 		pReg->Option2 = 0x01003000;
-		if(pMga->HasSDRAM)
-		    pReg->Option = 0x50040120;
-		else
-		    pReg->Option = 0x50044120;
 		break;
 	case PCI_CHIP_MGAG200:
 	case PCI_CHIP_MGAG200_PCI:
@@ -371,10 +388,9 @@ MGAGInit(ScrnInfoPtr pScrn, DisplayModePtr mode)
 		break;
 	}
 
-	if(pMga->UsePCIRetry)
-		pReg->Option &= ~0x20000000;
-	else
-		pReg->Option |= 0x20000000;
+	/* must always have the pci retries on but rely on 
+	   polling to keep them from occuring */
+	pReg->Option &= ~0x20000000;
 
 	switch(pLayout->bitsPerPixel)
 	{
@@ -579,6 +595,8 @@ MGAGRestore(ScrnInfoPtr pScrn, vgaRegPtr vgaReg, MGARegPtr mgaReg,
 	int i;
 	MGAPtr pMga = MGAPTR(pScrn);
 
+	CHECK_DMA_QUIESCENT( pMga, pScrn );
+
 	/*
 	 * Code is needed to get things back to bank zero.
 	 */
@@ -656,6 +674,8 @@ MGAGSave(ScrnInfoPtr pScrn, vgaRegPtr vgaReg, MGARegPtr mgaReg,
 	int i;
 	MGAPtr pMga = MGAPTR(pScrn);
 	
+	CHECK_DMA_QUIESCENT( pMga, pScrn );
+
 	/* Allocate the DacRegs space if not done already */
 	if (mgaReg->DacRegs == NULL) {
 		mgaReg->DacRegs = xnfcalloc(DACREGSIZE, 1);
@@ -749,6 +769,9 @@ MGAGSetCursorPosition(ScrnInfoPtr pScrn, int x, int y)
     x += 64;
     y += 64;
 
+    /* This doesn't require DMA quiescence (what about flush?)
+     */
+
     /* cursor update must never occurs during a retrace period (pp 4-160) */
     while( INREG( MGAREG_Status ) & 0x08 );
     
@@ -915,7 +938,12 @@ MGAGRamdacInit(ScrnInfoPtr pScrn)
 	        MGAdac->maxPixelClock = 220000;
 	    break;
     	case PCI_CHIP_MGAG400:
-	    MGAdac->maxPixelClock = 300000;
+	    /* We don't know the new pins format but we know that
+	       the maxclock / 4 is where the RamdacType was in the
+	       old pins format */
+	    MGAdac->maxPixelClock = pMga->Bios2.RamdacType * 4000;
+	    if(MGAdac->maxPixelClock < 300000)
+		MGAdac->maxPixelClock = 300000;
 	    break;
 	default:
 	    MGAdac->maxPixelClock = 250000;
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 df23db4f6..ca69d37c1 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dri.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dri.c
@@ -22,7 +22,7 @@
 #include "mga.h"
 #include "mga_macros.h"
 #include "mga_dri.h"
-#include "mga_dripriv.h"
+#include "mga_wrap.h"
 
 static char MGAKernelDriverName[] = "mga";
 static char MGAClientDriverName[] = "mga";
@@ -51,172 +51,252 @@ extern void Mga32DRIInitBuffers(WindowPtr pWin, RegionPtr prgn, CARD32 index);
 extern void Mga32DRIMoveBuffers(WindowPtr pParent, DDXPointRec ptOldOrg, 
 			       RegionPtr prgnSrc, CARD32 index);
 
-static Bool
-MGAInitVisualConfigs(ScreenPtr pScreen)
+Bool MgaCleanupDma(ScrnInfoPtr pScrn)
 {
-  ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
-  MGAPtr pMGA = MGAPTR(pScrn);
-  int numConfigs = 0;
-  __GLXvisualConfig *pConfigs = 0;
-  MGAConfigPrivPtr pMGAConfigs = 0;
-  MGAConfigPrivPtr *pMGAConfigPtrs = 0;
-  int i;
+   MGAPtr pMGA = MGAPTR(pScrn);
+   Bool ret_val;
+   
+   ret_val = drmMgaCleanupDma(pMGA->drmSubFD);
+   if(ret_val == FALSE) ErrorF("Mga Dma Cleanup Failed\n");
+   
+   return ret_val;
+}
 
-  switch (pScrn->bitsPerPixel) {
-  case 8:
-  case 24:
-  case 32:
-    break;
-  case 16:
-    numConfigs = 4;
+Bool MgaLockUpdate(ScrnInfoPtr pScrn, drmLockFlags flags)
+{
+   MGAPtr pMGA = MGAPTR(pScrn);
+   Bool ret_val;
+   
+   ret_val = drmMgaLockUpdate(pMGA->drmSubFD, flags);
+   if(ret_val == FALSE) ErrorF("LockUpdate failed\n");
+   
+   return ret_val;
+}
 
-    if (!(pConfigs = (__GLXvisualConfig*)xnfcalloc(sizeof(__GLXvisualConfig),
-						   numConfigs))) {
+Bool MgaInitDma(ScrnInfoPtr pScrn, int prim_size)
+{
+   MGAPtr pMGA = MGAPTR(pScrn);
+   MGADRIPtr pMGADRI = (MGADRIPtr)pMGA->pDRIInfo->devPrivate;
+   MGADRIServerPrivatePtr pMGADRIServer = pMGA->DRIServerInfo;
+   drmMgaInit init;
+   Bool ret_val;
+   
+   memset(&init, 0, sizeof(drmMgaInit));
+   init.reserved_map_agpstart = 0;
+   init.reserved_map_idx = 3;
+   init.buffer_map_idx = 4;
+   init.sarea_priv_offset = sizeof(XF86DRISAREARec);
+   init.primary_size = prim_size;
+   init.warp_ucode_size = pMGADRIServer->warp_ucode_size;
+
+   switch(pMGA->Chipset) {
+   case PCI_CHIP_MGAG400:
+      init.chipset = MGA_CARD_TYPE_G400;
+      break;
+   case PCI_CHIP_MGAG200:
+      init.chipset = MGA_CARD_TYPE_G200;
+      break;
+   case PCI_CHIP_MGAG200_PCI:
+   default:
+      ErrorF("Direct rendering not supported on this card/chipset\n");
       return FALSE;
-    }
-    if (!(pMGAConfigs = (MGAConfigPrivPtr)xnfcalloc(sizeof(MGAConfigPrivRec),
+   }
+
+   init.frontOffset = pMGADRI->frontOffset;
+   init.backOffset = pMGADRI->backOffset;
+   init.depthOffset = pMGADRI->depthOffset;
+   init.textureOffset = pMGADRI->textureOffset;
+   init.textureSize = pMGADRI->textureSize;
+   init.agpTextureSize = pMGADRI->agpTextureSize;
+   init.cpp = pMGADRI->cpp;
+   init.stride = pMGADRI->frontPitch;
+   init.mAccess = pMGA->MAccess;
+   init.sgram = !pMGA->HasSDRAM;
+   
+   memcpy(&init.WarpIndex, &pMGADRIServer->WarpIndex, 
+	  sizeof(drmMgaWarpIndex) * MGA_MAX_WARP_PIPES);
+
+   ErrorF("Mga Dma Initialization start\n");
+
+   ret_val = drmMgaInitDma(pMGA->drmSubFD, &init);
+   if(ret_val == FALSE) ErrorF("Mga Dma Initialization Failed\n");
+   else ErrorF("Mga Dma Initialization done\n");
+   return ret_val;
+}
+
+static Bool
+MGAInitVisualConfigs(ScreenPtr pScreen)
+{
+   ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
+   MGAPtr pMGA = MGAPTR(pScrn);
+   int numConfigs = 0;
+   __GLXvisualConfig *pConfigs = 0;
+   MGAConfigPrivPtr pMGAConfigs = 0;
+   MGAConfigPrivPtr *pMGAConfigPtrs = 0;
+   int i;
+
+   switch (pScrn->bitsPerPixel) {
+   case 8:
+   case 24:
+   case 32:
+      break;
+   case 16:
+      numConfigs = 4;
+
+      if (!(pConfigs = (__GLXvisualConfig*)xnfcalloc(sizeof(__GLXvisualConfig),
 						     numConfigs))) {
-      xfree(pConfigs);
-      return FALSE;
-    }
-    if (!(pMGAConfigPtrs = (MGAConfigPrivPtr*)xnfcalloc(sizeof(MGAConfigPrivPtr),
-							 numConfigs))) {
-      xfree(pConfigs);
-      xfree(pMGAConfigs);
-      return FALSE;
-    }
-    for (i=0; i<numConfigs; i++) 
-      pMGAConfigPtrs[i] = &pMGAConfigs[i];
+	 return FALSE;
+      }
+      if (!(pMGAConfigs = (MGAConfigPrivPtr)xnfcalloc(sizeof(MGAConfigPrivRec),
+						      numConfigs))) {
+	 xfree(pConfigs);
+	 return FALSE;
+      }
+      if (!(pMGAConfigPtrs = (MGAConfigPrivPtr*)xnfcalloc(sizeof(MGAConfigPrivPtr),
+							  numConfigs))) {
+	 xfree(pConfigs);
+	 xfree(pMGAConfigs);
+	 return FALSE;
+      }
+      for (i=0; i<numConfigs; i++) 
+	 pMGAConfigPtrs[i] = &pMGAConfigs[i];
+
+      /* config 0: db=FALSE, depth=0
+	 config 1: db=FALSE, depth=16
+	 config 2: db=TRUE, depth=0;
+	 config 3: db=TRUE, depth=16
+      */
+      pConfigs[0].vid = -1;
+      pConfigs[0].class = -1;
+      pConfigs[0].rgba = TRUE;
+      pConfigs[0].redSize = 5;
+      pConfigs[0].greenSize = 6;
+      pConfigs[0].blueSize = 5;
+      pConfigs[0].redMask = 0x0000F800;
+      pConfigs[0].greenMask = 0x000007E0;
+      pConfigs[0].blueMask = 0x0000001F;
+      pConfigs[0].alphaMask = 0;
+      pConfigs[0].accumRedSize = 0;
+      pConfigs[0].accumGreenSize = 0;
+      pConfigs[0].accumBlueSize = 0;
+      pConfigs[0].accumAlphaSize = 0;
+      pConfigs[0].doubleBuffer = FALSE;
+      pConfigs[0].stereo = FALSE;
+      pConfigs[0].bufferSize = 16;
+      pConfigs[0].depthSize = 16;
+      pConfigs[0].stencilSize = 0;
+      pConfigs[0].auxBuffers = 0;
+      pConfigs[0].level = 0;
+      pConfigs[0].visualRating = 0;
+      pConfigs[0].transparentPixel = 0;
+      pConfigs[0].transparentRed = 0;
+      pConfigs[0].transparentGreen = 0;
+      pConfigs[0].transparentBlue = 0;
+      pConfigs[0].transparentAlpha = 0;
+      pConfigs[0].transparentIndex = 0;
+
+      pConfigs[1].vid = -1;
+      pConfigs[1].class = -1;
+      pConfigs[1].rgba = TRUE;
+      pConfigs[1].redSize = 5;
+      pConfigs[1].greenSize = 6;
+      pConfigs[1].blueSize = 5;
+      pConfigs[1].redMask = 0x0000F800;
+      pConfigs[1].greenMask = 0x000007E0;
+      pConfigs[1].blueMask = 0x0000001F;
+      pConfigs[1].alphaMask = 0;
+      pConfigs[1].accumRedSize = 0;
+      pConfigs[1].accumGreenSize = 0;
+      pConfigs[1].accumBlueSize = 0;
+      pConfigs[1].accumAlphaSize = 0;
+      pConfigs[1].doubleBuffer = FALSE;
+      pConfigs[1].stereo = FALSE;
+      pConfigs[1].bufferSize = 16;
+      pConfigs[1].depthSize = 16;
+      pConfigs[1].stencilSize = 0;
+      pConfigs[1].auxBuffers = 0;
+      pConfigs[1].level = 0;
+      pConfigs[1].visualRating = 0;
+      pConfigs[1].transparentPixel = 0;
+      pConfigs[1].transparentRed = 0;
+      pConfigs[1].transparentGreen = 0;
+      pConfigs[1].transparentBlue = 0;
+      pConfigs[1].transparentAlpha = 0;
+      pConfigs[1].transparentIndex = 0;
+
+      pConfigs[2].vid = -1;
+      pConfigs[2].class = -1;
+      pConfigs[2].rgba = TRUE;
+      pConfigs[2].redSize = 5;
+      pConfigs[2].greenSize = 6;
+      pConfigs[2].blueSize = 5;
+      pConfigs[2].redMask = 0x0000F800;
+      pConfigs[2].greenMask = 0x000007E0;
+      pConfigs[2].blueMask = 0x0000001F;
+      pConfigs[2].alphaMask = 0;
+      pConfigs[2].accumRedSize = 0;
+      pConfigs[2].accumGreenSize = 0;
+      pConfigs[2].accumBlueSize = 0;
+      pConfigs[2].accumAlphaSize = 0;
+      pConfigs[2].doubleBuffer = TRUE;
+      pConfigs[2].stereo = FALSE;
+      pConfigs[2].bufferSize = 16;
+      pConfigs[2].depthSize = 16;
+      pConfigs[2].stencilSize = 0;
+      pConfigs[2].auxBuffers = 0;
+      pConfigs[2].level = 0;
+      pConfigs[2].visualRating = 0;
+      pConfigs[2].transparentPixel = 0;
+      pConfigs[2].transparentRed = 0;
+      pConfigs[2].transparentGreen = 0;
+      pConfigs[2].transparentBlue = 0;
+      pConfigs[2].transparentAlpha = 0;
+      pConfigs[2].transparentIndex = 0;
+
+      pConfigs[3].vid = -1;
+      pConfigs[3].class = -1;
+      pConfigs[3].rgba = TRUE;
+      pConfigs[3].redSize = 5;
+      pConfigs[3].greenSize = 6;
+      pConfigs[3].blueSize = 5;
+      pConfigs[3].redMask = 0x0000F800;
+      pConfigs[3].greenMask = 0x000007E0;
+      pConfigs[3].blueMask = 0x0000001F;
+      pConfigs[3].alphaMask = 0;
+      pConfigs[3].accumRedSize = 0;
+      pConfigs[3].accumGreenSize = 0;
+      pConfigs[3].accumBlueSize = 0;
+      pConfigs[3].accumAlphaSize = 0;
+      pConfigs[3].doubleBuffer = TRUE;
+      pConfigs[3].stereo = FALSE;
+      pConfigs[3].bufferSize = 16;
+      pConfigs[3].depthSize = 16;
+      pConfigs[3].stencilSize = 0;
+      pConfigs[3].auxBuffers = 0;
+      pConfigs[3].level = 0;
+      pConfigs[3].visualRating = 0;
+      pConfigs[3].transparentPixel = 0;
+      pConfigs[3].transparentRed = 0;
+      pConfigs[3].transparentGreen = 0;
+      pConfigs[3].transparentBlue = 0;
+      pConfigs[3].transparentAlpha = 0;
+      pConfigs[3].transparentIndex = 0;
+   }
+   pMGA->numVisualConfigs = numConfigs;
+   pMGA->pVisualConfigs = pConfigs;
+   pMGA->pVisualConfigsPriv = pMGAConfigs;
+   GlxSetVisualConfigs(numConfigs, pConfigs, (void**)pMGAConfigPtrs);
+   return TRUE;
+}
 
-    /* config 0: db=FALSE, depth=0
-       config 1: db=FALSE, depth=16
-       config 2: db=TRUE, depth=0;
-       config 3: db=TRUE, depth=16
-    */
-    pConfigs[0].vid = -1;
-    pConfigs[0].class = -1;
-    pConfigs[0].rgba = TRUE;
-    pConfigs[0].redSize = 8;
-    pConfigs[0].greenSize = 8;
-    pConfigs[0].blueSize = 8;
-    pConfigs[0].redMask =   0x00FF0000;
-    pConfigs[0].greenMask = 0x0000FF00;
-    pConfigs[0].blueMask =  0x000000FF;
-    pConfigs[0].alphaMask = 0;
-    pConfigs[0].accumRedSize = 0;
-    pConfigs[0].accumGreenSize = 0;
-    pConfigs[0].accumBlueSize = 0;
-    pConfigs[0].accumAlphaSize = 0;
-    pConfigs[0].doubleBuffer = FALSE;
-    pConfigs[0].stereo = FALSE;
-    pConfigs[0].bufferSize = 16;
-    pConfigs[0].depthSize = 0;
-    pConfigs[0].stencilSize = 0;
-    pConfigs[0].auxBuffers = 0;
-    pConfigs[0].level = 0;
-    pConfigs[0].visualRating = 0;
-    pConfigs[0].transparentPixel = 0;
-    pConfigs[0].transparentRed = 0;
-    pConfigs[0].transparentGreen = 0;
-    pConfigs[0].transparentBlue = 0;
-    pConfigs[0].transparentAlpha = 0;
-    pConfigs[0].transparentIndex = 0;
-
-    pConfigs[1].vid = -1;
-    pConfigs[1].class = -1;
-    pConfigs[1].rgba = TRUE;
-    pConfigs[1].redSize = 8;
-    pConfigs[1].greenSize = 8;
-    pConfigs[1].blueSize = 8;
-    pConfigs[1].redMask = 0x00FF0000;
-    pConfigs[1].greenMask = 0x0000FF00;
-    pConfigs[1].blueMask = 0x000000FF;
-    pConfigs[1].alphaMask = 0;
-    pConfigs[1].accumRedSize = 0;
-    pConfigs[1].accumGreenSize = 0;
-    pConfigs[1].accumBlueSize = 0;
-    pConfigs[1].accumAlphaSize = 0;
-    pConfigs[1].doubleBuffer = FALSE;
-    pConfigs[1].stereo = FALSE;
-    pConfigs[1].bufferSize = 16;
-    pConfigs[1].depthSize = 16;
-    pConfigs[1].stencilSize = 0;
-    pConfigs[1].auxBuffers = 0;
-    pConfigs[1].level = 0;
-    pConfigs[1].visualRating = 0;
-    pConfigs[1].transparentPixel = 0;
-    pConfigs[1].transparentRed = 0;
-    pConfigs[1].transparentGreen = 0;
-    pConfigs[1].transparentBlue = 0;
-    pConfigs[1].transparentAlpha = 0;
-    pConfigs[1].transparentIndex = 0;
-
-    pConfigs[2].vid = -1;
-    pConfigs[2].class = -1;
-    pConfigs[2].rgba = TRUE;
-    pConfigs[2].redSize = 8;
-    pConfigs[2].greenSize = 8;
-    pConfigs[2].blueSize = 8;
-    pConfigs[2].redMask = 0x00FF0000;
-    pConfigs[2].greenMask = 0x0000FF00;
-    pConfigs[2].blueMask = 0x000000FF;
-    pConfigs[2].alphaMask = 0;
-    pConfigs[2].accumRedSize = 0;
-    pConfigs[2].accumGreenSize = 0;
-    pConfigs[2].accumBlueSize = 0;
-    pConfigs[2].accumAlphaSize = 0;
-    pConfigs[2].doubleBuffer = TRUE;
-    pConfigs[2].stereo = FALSE;
-    pConfigs[2].bufferSize = 16;
-    pConfigs[2].depthSize = 0;
-    pConfigs[2].stencilSize = 0;
-    pConfigs[2].auxBuffers = 0;
-    pConfigs[2].level = 0;
-    pConfigs[2].visualRating = 0;
-    pConfigs[2].transparentPixel = 0;
-    pConfigs[2].transparentRed = 0;
-    pConfigs[2].transparentGreen = 0;
-    pConfigs[2].transparentBlue = 0;
-    pConfigs[2].transparentAlpha = 0;
-    pConfigs[2].transparentIndex = 0;
-
-    pConfigs[3].vid = -1;
-    pConfigs[3].class = -1;
-    pConfigs[3].rgba = TRUE;
-    pConfigs[3].redSize = 8;
-    pConfigs[3].greenSize = 8;
-    pConfigs[3].blueSize = 8;
-    pConfigs[3].redMask = 0x00FF0000;
-    pConfigs[3].greenMask = 0x0000FF00;
-    pConfigs[3].blueMask = 0x000000FF;
-    pConfigs[3].alphaMask = 0;
-    pConfigs[3].accumRedSize = 0;
-    pConfigs[3].accumGreenSize = 0;
-    pConfigs[3].accumBlueSize = 0;
-    pConfigs[3].accumAlphaSize = 0;
-    pConfigs[3].doubleBuffer = TRUE;
-    pConfigs[3].stereo = FALSE;
-    pConfigs[3].bufferSize = 16;
-    pConfigs[3].depthSize = 16;
-    pConfigs[3].stencilSize = 0;
-    pConfigs[3].auxBuffers = 0;
-    pConfigs[3].level = 0;
-    pConfigs[3].visualRating = 0;
-    pConfigs[3].transparentPixel = 0;
-    pConfigs[3].transparentRed = 0;
-    pConfigs[3].transparentGreen = 0;
-    pConfigs[3].transparentBlue = 0;
-    pConfigs[3].transparentAlpha = 0;
-    pConfigs[3].transparentIndex = 0;
-    break;
-  }
-  pMGA->numVisualConfigs = numConfigs;
-  pMGA->pVisualConfigs = pConfigs;
-  pMGA->pVisualConfigsPriv = pMGAConfigs;
-  GlxSetVisualConfigs(numConfigs, pConfigs, (void**)pMGAConfigPtrs);
-  return TRUE;
+static unsigned int mylog2(unsigned int n)
+{
+   unsigned int log2 = 1;
+   while (n>1) n >>= 1, log2++;
+   return log2;
 }
 
+
 Bool MGADRIScreenInit(ScreenPtr pScreen)
 {
   ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
@@ -224,10 +304,24 @@ Bool MGADRIScreenInit(ScreenPtr pScreen)
   DRIInfoPtr pDRIInfo;
   MGADRIPtr pMGADRI;
   MGADRIServerPrivatePtr pMGADRIServer;
-  int bufs;
-   int prim_size;
-   int init_offset;
+  int bufs, size;
+  int prim_size;
+  int init_offset;
+  int i;
 
+#if XFree86LOADER
+  /* Check that the GLX, DRI, and DRM modules have been loaded by testing
+   * for canonical symbols in each module. */
+  if (!LoaderSymbol("GlxSetVisualConfigs")) return FALSE;
+  if (!LoaderSymbol("DRIScreenInit"))       return FALSE;
+  if (!LoaderSymbol("drmAvailable"))        return FALSE;
+#endif   
+     
+  if((pScrn->bitsPerPixel / 8) != 2) {
+     ErrorF("Direct Rendering only supported in 16 bpp mode\n");
+     return FALSE;
+  }
+   
   pDRIInfo = DRICreateInfoRec();
   if (!pDRIInfo) return FALSE;
   pMGA->pDRIInfo = pDRIInfo;
@@ -247,28 +341,25 @@ Bool MGADRIScreenInit(ScreenPtr pScreen)
   pDRIInfo->frameBufferStride = pScrn->displayWidth*(pScrn->bitsPerPixel/8);
   pDRIInfo->ddxDrawableTableEntry = MGA_MAX_DRAWABLES;
 
+  MGADRIWrapFunctions( pScreen, pDRIInfo );
+
   if (SAREA_MAX_DRAWABLES < MGA_MAX_DRAWABLES)
     pDRIInfo->maxDrawableTableEntry = SAREA_MAX_DRAWABLES;
   else
     pDRIInfo->maxDrawableTableEntry = MGA_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(drm_mga_sarea_t)>SAREA_MAX) {
+  if (sizeof(XF86DRISAREARec)+sizeof(MGASAREARec)>SAREA_MAX) {
     ErrorF("Data does not fit in SAREA\n");
     return FALSE;
   }
+  
+  ErrorF("\n\n\nSarea %d+%d: %d\n", sizeof(XF86DRISAREARec), 
+	 sizeof(MGASAREARec), sizeof(XF86DRISAREARec) + sizeof(MGASAREARec));
+
   pDRIInfo->SAREASize = SAREA_MAX;
-#endif
 
   if (!(pMGADRI = (MGADRIPtr)xnfcalloc(sizeof(MGADRIRec),1))) {
     DRIDestroyInfoRec(pMGA->pDRIInfo);
@@ -277,7 +368,7 @@ Bool MGADRIScreenInit(ScreenPtr pScreen)
     return FALSE;
   }
    if (!(pMGADRIServer = (MGADRIServerPrivatePtr)
-	 xnfcalloc(sizeof(MGADRIServerPrivate),1))) {
+	 xnfcalloc(sizeof(MGADRIServerPrivateRec),1))) {
     xfree(pMGADRI);
     DRIDestroyInfoRec(pMGA->pDRIInfo);
     pMGA->pDRIInfo=0;
@@ -334,25 +425,44 @@ Bool MGADRIScreenInit(ScreenPtr pScreen)
 	     pMGADRIServer->regs);
    
   /* Agp Support */
+  pMGADRIServer->agpAcquired = FALSE;
+  pMGADRIServer->agpHandle = 0;
+  pMGADRIServer->agpSizep = 0;
+  pMGADRIServer->agp_map = 0;
+   
   if(drmAgpAcquire(pMGA->drmSubFD) < 0) {
      DRICloseScreen(pScreen);
      ErrorF("drmAgpAcquire failed\n");
      return FALSE;
   }
+  pMGADRIServer->agpAcquired = TRUE;
+
   pMGADRIServer->warp_ucode_size = mgaGetMicrocodeSize(pScreen);
    if(pMGADRIServer->warp_ucode_size == 0) {
       ErrorF("microcodeSize is zero\n");
       DRICloseScreen(pScreen);
       return FALSE;
    }
+   
+  pMGADRIServer->agpMode = drmAgpGetMode(pMGA->drmSubFD);
+  /* Default to 1X agp mode */
+  pMGADRIServer->agpMode &= ~0x00000002;
+  if (drmAgpEnable(pMGA->drmSubFD, pMGADRIServer->agpMode) < 0) {
+    ErrorF("drmAgpEnable failed\n");
+    DRICloseScreen(pScreen);
+    return FALSE;
+  }
+  ErrorF("drmAgpEnabled succeeded\n");
 
   prim_size = 65536;
   init_offset = ((prim_size + pMGADRIServer->warp_ucode_size + 
 		  4096 - 1) / 4096) * 4096;
    
-  pMGADRIServer->agpSizep = drmAgpSize(pMGA->drmSubFD);
-  pMGADRIServer->agpBase = drmAgpBase(pMGA->drmSubFD);
-  if (drmAddMap(pMGA->drmSubFD, (drmHandle)pMGADRIServer->agpBase,
+  pMGADRIServer->agpSizep = init_offset;
+  pMGADRI->agpSize = (drmAgpSize(pMGA->drmSubFD)) - init_offset;
+
+  pMGADRIServer->agpBase = (drmAddress) drmAgpBase(pMGA->drmSubFD);
+  if (drmAddMap(pMGA->drmSubFD, 0,
 		init_offset, DRM_AGP, 0, 
 		&pMGADRIServer->agp_private) < 0) {
     DRICloseScreen(pScreen);
@@ -370,12 +480,15 @@ Bool MGADRIScreenInit(ScreenPtr pScreen)
    
   /* Now allocate and bind a default of 8 megs */
 
-  pMGADRIServer->agpHandle = drmAgpAlloc(pMGA->drmSubFD, 0x00800000, 0, 0);
+  drmAgpAlloc(pMGA->drmSubFD, 0x00800000, 0, 0,
+	      &pMGADRIServer->agpHandle);
+   
   if(pMGADRIServer->agpHandle == 0) {
     ErrorF("drmAgpAlloc failed\n");
     DRICloseScreen(pScreen);
     return FALSE;
   }
+   
   if(drmAgpBind(pMGA->drmSubFD, pMGADRIServer->agpHandle, 0) < 0) {
     DRICloseScreen(pScreen);
     ErrorF("drmAgpBind failed\n");
@@ -384,72 +497,115 @@ Bool MGADRIScreenInit(ScreenPtr pScreen)
 
   mgaInstallMicrocode(pScreen, prim_size);
 
-   ErrorF("init_offset: %x\n", init_offset);
-  pMGADRI->agpSize = pMGADRIServer->agpSizep - init_offset;
-   ErrorF("pMGADRI->agpSize: %x\n", pMGADRI->agpSize);
-	  
-  if(drmAddMap(pMGA->drmSubFD, (drmHandle)pMGADRIServer->agpBase + init_offset,
+  if(drmAddMap(pMGA->drmSubFD, (drmHandle)init_offset,
 	       pMGADRI->agpSize, DRM_AGP, 0, 
 	       &pMGADRI->agp) < 0) {
      ErrorF("Failed to map public agp area\n");
      DRICloseScreen(pScreen);
      return FALSE;
   }
-   ErrorF("Mapped public agp area\n");
+
+
+   switch(pMGA->Chipset) {
+   case PCI_CHIP_MGAG400:
+      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:
+      ErrorF("Direct rendering not supported on this card/chipset\n");
+      return FALSE;
+   }
+
+
+  pMGADRI->width = pScrn->virtualX;
+  pMGADRI->height = pScrn->virtualY;
+  pMGADRI->mem = pScrn->videoRam * 1024;
+  pMGADRI->cpp = pScrn->bitsPerPixel / 8;
+  pMGADRI->frontPitch = pScrn->displayWidth * (pScrn->bitsPerPixel / 8);
+
+
+  pMGADRI->frontOffset = 0; /* pMGA->YDstOrg * (pScrn->bitsPerPixel / 8) */
+  pMGADRI->backOffset = ((pScrn->virtualY+129) * pScrn->displayWidth *
+			 (pScrn->bitsPerPixel / 8) + 4095) & ~0xFFF;
+
+
+  fprintf(stderr, "calced backoffset: 0x%x\n",
+	  pMGADRI->backOffset);
+
+
+#if 0
+  size = 2 * pScrn->virtualX * pScrn->virtualY;
+  pMGADRI->depthOffset = (pMGADRI->backOffset + size + 4095) & ~0xFFF;
+  pMGADRI->textureOffset = pMGADRI->depthOffset + size;
+  pMGADRI->textureSize = pMGA->FbUsableSize - pMGADRI->textureOffset;
+#else
+  size = 2 * pScrn->virtualX * pScrn->virtualY;
+  size += 4095;
+  size &= ~4095;
+  pMGADRI->depthOffset = pMGA->FbUsableSize - size;
+  pMGADRI->depthOffset &= ~4095;  
+  pMGADRI->textureOffset = pMGADRI->backOffset + size;
+  pMGADRI->textureSize = pMGADRI->depthOffset - pMGADRI->textureOffset;
+
+  if (pMGADRI->depthOffset < pMGADRI->textureOffset + 512*1024) {
+     ErrorF("Insufficient memory for direct rendering\n");
+     DRICloseScreen(pScreen);
+     return FALSE;
+  }     
+#endif
+
+  pMGADRI->mAccess = pMGA->MAccess;
+
+  i = mylog2(pMGADRI->textureSize / MGA_NR_TEX_REGIONS);
+  if (i < MGA_LOG_MIN_TEX_REGION_SIZE)
+      i = MGA_LOG_MIN_TEX_REGION_SIZE;
+  
+  pMGADRI->logTextureGranularity = i;
+  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 */
-   
-  pMGADRIServer->agpMode = drmAgpGetMode(pMGA->drmSubFD);
-  /* Default to 1X agp mode */
-  pMGADRIServer->agpMode &= ~0x00000002;
-  if (drmAgpEnable(pMGA->drmSubFD, pMGADRIServer->agpMode) < 0) {
-    ErrorF("drmAgpEnable failed\n");
-    DRICloseScreen(pScreen);
-    return FALSE;
-  }
-   ErrorF("drmAgpEnabled succeeded\n");
-     if((bufs = drmAddBufs(pMGA->drmSubFD,
-			/*63*/ 15,
-			/* 65536 */ 524288,
+  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",
-	       /* 63 */ 15,
-	       /* 65536 */ 524288);
+	       MGA_DMA_BUF_NR,
+	       MGA_DMA_BUF_SZ);
     DRICloseScreen(pScreen);
     return FALSE;
   }
    
   xf86DrvMsg(pScrn->scrnIndex, X_INFO,
 	     "[drm] added %d %d byte DMA buffers\n",
-	     bufs, /* 65536 */ 524288);
+	     bufs, MGA_DMA_BUF_SZ);
 
 
-  if((mgadrmInitDma(pScrn, prim_size)) != TRUE) {
+  if((MgaInitDma(pScrn, prim_size)) != TRUE) {
     ErrorF("Failed to initialize dma engine\n");
     DRICloseScreen(pScreen);
     return FALSE;
   }
-   
-  ErrorF("Initialized Dma Engine\n");
 
+  ErrorF("Initialized Dma Engine\n");
    
-  if(drmMarkBufs(pMGA->drmSubFD, 0.133333, 0.266666) != 0) {
-     xf86DrvMsg(pScrn->scrnIndex, X_INFO,
-		"[drm] failure marking DMA buffers\n");
-     DRICloseScreen(pScreen);
-     return FALSE;
-  }
-  if (!(pMGADRIServer->drmBufs = drmMapBufs(pMGA->drmSubFD))) { 
-    xf86DrvMsg(pScrn->scrnIndex, X_INFO,
-	       "[drm] failure mapping DMA buffers\n");
-    DRICloseScreen(pScreen);
-    return FALSE;
-  }
-  xf86DrvMsg(pScrn->scrnIndex, X_INFO, "[drm] buffers mapped with %p\n",
-	     pMGADRIServer->drmBufs);
-  xf86DrvMsg(pScrn->scrnIndex, X_INFO, "[drm] %d DMA buffers mapped\n",
-	     pMGADRIServer->drmBufs->count);
   if (!pMGADRIServer->irq) {
     pMGADRIServer->irq = drmGetInterruptFromBusID(pMGA->drmSubFD,
 					    ((pciConfigPtr)pMGA->PciInfo
@@ -481,16 +637,26 @@ MGADRICloseScreen(ScreenPtr pScreen)
   ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
   MGAPtr pMGA = MGAPTR(pScrn);
   MGADRIServerPrivatePtr pMGADRIServer = pMGA->DRIServerInfo;
-   
-  xf86DrvMsg(pScrn->scrnIndex, X_INFO,
-	     "[drm] unmapping %d buffers\n",
-	     pMGADRIServer->drmBufs->count);
-  if (drmUnmapBufs(pMGADRIServer->drmBufs)) {
-     xf86DrvMsg(pScreen->myNum, X_INFO,
-		"[drm] unable to unmap DMA buffers\n");
+
+  MgaCleanupDma(pScrn);
+
+  if(pMGADRIServer->agp_map) {  
+     ErrorF("Unmapped agp region\n");
+     drmUnmap(pMGADRIServer->agp_map, pMGADRIServer->agpSizep);
+     pMGADRIServer->agp_map = 0;
   }
-  mgadrmCleanupDma(pScrn);
-   
+  if(pMGADRIServer->agpHandle) {
+     ErrorF("Freeing agp memory\n");
+     drmAgpFree(pMGA->drmSubFD, pMGADRIServer->agpHandle);
+     pMGADRIServer->agpHandle = 0;
+     pMGADRIServer->agpSizep = 0;
+  }
+  if(pMGADRIServer->agpAcquired == TRUE) {
+     ErrorF("releasing agp module\n");
+     drmAgpRelease(pMGA->drmSubFD);
+     pMGADRIServer->agpAcquired = FALSE;
+  }
+
   DRICloseScreen(pScreen);
 
   if (pMGA->pDRIInfo) {
@@ -518,9 +684,6 @@ MGACreateContext(ScreenPtr pScreen, VisualPtr visual,
 		  drmContext hwContext, void *pVisualConfigPriv,
 		  DRIContextType contextStore)
 {
-  MGADRIContextPtr ctx;
-
-  ctx = (MGADRIContextPtr)contextStore;
   return TRUE;
 }
 
@@ -528,106 +691,120 @@ static void
 MGADestroyContext(ScreenPtr pScreen, drmContext hwContext, 
 		   DRIContextType contextStore)
 {
-  MGADRIContextPtr ctx;
-  ctx = (MGADRIContextPtr)contextStore;
 }
 
 Bool
 MGADRIFinishScreenInit(ScreenPtr pScreen)
 {
   ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
+  MGASAREAPtr sPriv;
   MGAPtr pMGA = MGAPTR(pScrn);
-  MGADRIPtr pMGADRI;
-  int size;
 
+  if (!pMGA->pDRIInfo) return FALSE;
+
+  sPriv = (MGASAREAPtr)DRIGetSAREAPrivate(pScreen);
   pMGA->pDRIInfo->driverSwapMethod = DRI_HIDE_X_CONTEXT;
-  /* pMGA->pDRIInfo->driverSwapMethod = DRI_SERVER_SWAP; */
 
-  pMGADRI = (MGADRIPtr)pMGA->pDRIInfo->devPrivate;
-  pMGADRI->deviceID = pMGA->PciInfo->chipType;
-  pMGADRI->width = pScrn->virtualX;
-  pMGADRI->height = pScrn->virtualY;
-  pMGADRI->mem = pScrn->videoRam * 1024;
-  pMGADRI->cpp = pScrn->bitsPerPixel / 8;
-  pMGADRI->stride = pScrn->displayWidth * (pScrn->bitsPerPixel / 8);
-  pMGADRI->backOffset = ((pScrn->virtualY+129) * pScrn->displayWidth *
-			 (pScrn->bitsPerPixel / 8) + 4095) & ~0xFFF;
-  size = 2 * pScrn->virtualX * pScrn->virtualY;
-  pMGADRI->depthOffset = (pMGADRI->backOffset + size + 4095) & ~0xFFF;
-  pMGADRI->textureOffset = pMGADRI->depthOffset + size;
-  /*
-   * The rest of the framebuffer is for textures except for the
-   * memory for the hardware cursor. 
-   */
-  pMGADRI->textureSize = pMGA->FbUsableSize - pMGADRI->textureOffset;
-  pMGADRI->fbOffset = pMGA->YDstOrg * (pScrn->bitsPerPixel / 8);
+  xf86memset( sPriv, 0, sizeof(MGASAREARec) );
 
   return DRIFinishScreenInit(pScreen);
 }
 
-void MGASwapContext(ScreenPtr pScreen)
+
+void mgaGetQuiescence( ScrnInfoPtr pScrn )
 {
-#if 0
-  ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
-  MGAPtr pMga = MGAPTR(pScrn);
-  MGAFBLayout *pLayout = &pMga->CurrentLayout;
-
-  usleep(500);
-  ErrorF("Syncing : swap\n");
-  MGABUSYWAIT();
-  ErrorF("Sync : swap 1\n");
-  MGAStormSync(pScrn);
-  ErrorF("Syncing done\n");
-  pMga->AccelInfoRec->NeedToSync = TRUE;
-   
-  WAITFIFO(12);
-  OUTREG(MGAREG_MACCESS, pMga->MAccess);
-  OUTREG(MGAREG_PITCH, pLayout->displayWidth);
-  OUTREG(MGAREG_YDSTORG, pMga->YDstOrg);
-  OUTREG(MGAREG_PLNWT, pMga->PlaneMask);
-  OUTREG(MGAREG_BCOL, pMga->BgColor);
-  OUTREG(MGAREG_FCOL, pMga->FgColor);
-  OUTREG(MGAREG_SRCORG, pMga->SrcOrg);
-  OUTREG(MGAREG_DSTORG, pMga->DstOrg);
-  OUTREG(MGAREG_OPMODE, MGAOPM_DMA_BLIT);
-  OUTREG(MGAREG_CXBNDRY, 0xFFFF0000); /* (maxX << 16) | minX */
-  OUTREG(MGAREG_YTOP, 0x00000000);    /* minPixelPointer */
-  OUTREG(MGAREG_YBOT, 0x007FFFFF);    /* maxPixelPointer */ 
-  pMga->AccelFlags &= ~CLIPPER_ON;
-#endif
+   MGAPtr pMga = MGAPTR(pScrn);
+
+   pMga->have_quiescense = 1;					
+
+   if (pMga->directRenderingEnabled) {
+      MGAFBLayout *pLayout = &pMga->CurrentLayout;
+
+      MgaLockUpdate(pScrn, (DRM_LOCK_QUIESCENT | DRM_LOCK_FLUSH));	
+
+      WAITFIFO(12);
+      OUTREG(MGAREG_MACCESS, pMga->MAccess);
+      OUTREG(MGAREG_PITCH, pLayout->displayWidth);
+      OUTREG(MGAREG_YDSTORG, pMga->YDstOrg);
+      OUTREG(MGAREG_PLNWT, pMga->PlaneMask);
+      OUTREG(MGAREG_BCOL, pMga->BgColor);
+      OUTREG(MGAREG_FCOL, pMga->FgColor);
+      OUTREG(MGAREG_SRCORG, pMga->SrcOrg);
+      OUTREG(MGAREG_DSTORG, pMga->DstOrg);
+      OUTREG(MGAREG_OPMODE, MGAOPM_DMA_BLIT);
+      OUTREG(MGAREG_CXBNDRY, 0xFFFF0000); /* (maxX << 16) | minX */
+      OUTREG(MGAREG_YTOP, 0x00000000);    /* minPixelPointer */
+      OUTREG(MGAREG_YBOT, 0x007FFFFF);    /* maxPixelPointer */ 
+      pMga->AccelFlags &= ~CLIPPER_ON;
+   }
 }
 
-void MGALostContext(ScreenPtr pScreen)
+
+
+void MGASwapContext(ScreenPtr pScreen)
 {
-#if 0
-  ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
-  MGAPtr pMga = MGAPTR(pScrn);
-  MGAFBLayout *pLayout = &pMga->CurrentLayout;
-
-  ErrorF("Syncing : lost\n");
-  MGAStormSync(pScrn);
-  ErrorF("Sync : lost 1\n");
-  MGABUSYWAIT();
-  ErrorF("Syncing done\n");
-#endif
+   ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
+   MGAPtr pMga = MGAPTR(pScrn);
+
+   /* Arrange for dma_quiescence and xaa sync to be called as
+    * appropriate.
+    */
+   pMga->have_quiescense = 0;
+   pMga->AccelInfoRec->NeedToSync = TRUE;
 }
 
+
+
+/* This is really only called from validate/postvalidate as we
+ * override the dri lock/unlock.  Want to remove validate/postvalidate
+ * processing, but need to remove all client-side use of drawable lock
+ * first (otherwise there is noone recover when a client dies holding
+ * the drawable lock).
+ *
+ * What does this mean? 
+ *
+ *   - The above code gets executed every time a
+ *     window changes shape or the focus changes, which isn't really
+ *     optimal.  
+ *   - The X server therefore believes it needs to do an XAA sync
+ *     *and* a dma quiescense ioctl each time that happens.
+ *
+ * We don't wrap wakeuphandler any longer, so at least we can say that
+ * this doesn't happen *every time the mouse moves*...
+ */
 static void
 MGADRISwapContext(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 */
-    MGASwapContext(pScreen);
-  }
-  if ((syncType == DRI_2D_SYNC) && (oldContextType == DRI_NO_CONTEXT) &&
-      (newContextType == DRI_2D_CONTEXT)) { /* Exiting from Block Handler */
-    MGALostContext(pScreen);
-  }
+   if (syncType == DRI_3D_SYNC && 
+       oldContextType == DRI_2D_CONTEXT &&
+       newContextType == DRI_2D_CONTEXT)
+   {
+      MGASwapContext(pScreen);
+   }
 }
 
-/* Needs to be written */
-void MGASelectBuffer(MGAPtr pMGA, int which)
+
+void 
+MGASelectBuffer(ScrnInfoPtr pScrn, int which)
 {
+   MGAPtr pMga = MGAPTR(pScrn);
+   MGADRIPtr pMGADRI = (MGADRIPtr)pMga->pDRIInfo->devPrivate;
+
+   switch (which) {
+   case MGA_BACK:
+      OUTREG(MGAREG_DSTORG, pMGADRI->backOffset);
+      OUTREG(MGAREG_SRCORG, pMGADRI->backOffset);
+      break;
+   case MGA_DEPTH:
+      OUTREG(MGAREG_DSTORG, pMGADRI->depthOffset);
+      OUTREG(MGAREG_SRCORG, pMGADRI->depthOffset);
+      break;
+   default:
+   case MGA_FRONT:
+      OUTREG(MGAREG_DSTORG, pMGADRI->frontOffset);
+      OUTREG(MGAREG_SRCORG, pMGADRI->frontOffset);
+      break;
+   }
 }
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..2d35a6a89 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 +27,110 @@ 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;
+   int frontOffset;
+   int frontPitch;
+
+   int backOffset;
+   int backPitch;
+
+   int depthOffset;
+   int depthPitch;
+
+   int textureOffset;
+   int textureSize;
+   int logTextureGranularity;
+
+   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;
+   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.
+    */
+   int last_enqueue;       /* last time a buffer was enqueued */
+   int last_dispatch;      /* age of the most recently dispatched buffer */
+   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/mga/mga_dripriv.h b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dripriv.h
deleted file mode 100644
index 841c17cb9..000000000
--- a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dripriv.h
+++ /dev/null
@@ -1,24 +0,0 @@
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/mga/mga_dripriv.h,v 1.1 2000/02/11 17:25:55 dawes Exp $ */
-
-#ifndef _MGA_DRIPRIV_H_
-#define _MGA_DRIPRIV_H_
-
-#define MGA_MAX_DRAWABLES 256
-
-extern void GlxSetVisualConfigs(
-    int nconfigs,
-    __GLXvisualConfig *configs,
-    void **configprivs
-);
-
-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/mga/mga_driver.c b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_driver.c
index 85a529068..6f95b8d87 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_driver.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_driver.c
@@ -15,8 +15,8 @@
  * to be based on Radoslaw's original source
  *
  * Contributors:
- *		Andrew Vanderstock, Melbourne, Australia
- *			vanderaj@mail2.svhm.org.au
+ *		Andrew van der Stock
+ *			ajv@greebo.net
  *		additions, corrections, cleanups
  *
  *		Dirk Hohndel
@@ -43,7 +43,7 @@
  *		Fixed 32bpp hires 8MB horizontal line glitch at middle right
  */
  
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/mga/mga_driver.c,v 1.114 1999/08/28 14:32:47 dawes Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/mga/mga_driver.c,v 1.149 2000/03/07 01:37:49 dawes Exp $ */
 
 /*
  * This is a first cut at a non-accelerated version to work with the
@@ -76,8 +76,9 @@
 #include "micmap.h"
 
 #include "xf86DDC.h"
-
 #include "xf86RAC.h"
+#include "vbe.h"
+
 
 /*
  * If using cfb, cfb.h is required.  Select the others for the bpp values
@@ -102,8 +103,9 @@
 #include "shadowfb.h"
 #include "fbdevhw.h"
 
-#include "xf86xv.h"
-#include "Xv.h"
+#ifdef XF86DRI 
+#include "dri.h"
+#endif
 
 
 /*
@@ -111,6 +113,7 @@
  */
 
 /* Mandatory functions */
+static OptionInfoPtr	MGAAvailableOptions(int chipid, int busid);
 static void	MGAIdentify(int flags);
 static Bool	MGAProbe(DriverPtr drv, int flags);
 static Bool	MGAPreInit(ScrnInfoPtr pScrn, int flags);
@@ -120,7 +123,7 @@ static Bool	MGAEnterVT(int scrnIndex, int flags);
 static Bool	MGAEnterVTFBDev(int scrnIndex, int flags);
 static void	MGALeaveVT(int scrnIndex, int flags);
 static Bool	MGACloseScreen(int scrnIndex, ScreenPtr pScreen);
-static Bool	MGASaveScreen(ScreenPtr pScreen, Bool unblank);
+static Bool	MGASaveScreen(ScreenPtr pScreen, int mode);
 
 /* This shouldn't be needed since RAC will disable all I/O for MGA cards. */
 #ifdef DISABLE_VGA_IO
@@ -162,9 +165,13 @@ static Bool	MGAModeInit(ScrnInfoPtr pScrn, DisplayModePtr mode);
 
 DriverRec MGA = {
     VERSION,
+    MGA_DRIVER_NAME,
+#if 0
     "accelerated driver for Matrox Millennium and Mystique cards",
+#endif
     MGAIdentify,
     MGAProbe,
+    MGAAvailableOptions,
     NULL,
     0
 };
@@ -208,6 +215,7 @@ typedef enum {
     OPTION_COLOR_KEY,
     OPTION_SET_MCLK,
     OPTION_OVERCLOCK_MEM,
+    OPTION_VIDEO_KEY,
     OPTION_ROTATE
 } MGAOpts;
 
@@ -225,6 +233,7 @@ static OptionInfoRec MGAOptions[] = {
     { OPTION_COLOR_KEY,		"ColorKey",	OPTV_INTEGER,	{0}, FALSE },
     { OPTION_SET_MCLK,		"SetMclk",	OPTV_FREQ,	{0}, FALSE },
     { OPTION_OVERCLOCK_MEM,	"OverclockMem",	OPTV_BOOLEAN,	{0}, FALSE },
+    { OPTION_VIDEO_KEY,		"VideoKey",	OPTV_INTEGER,	{0}, FALSE },
     { OPTION_ROTATE,		"Rotate",	OPTV_ANYSTR,	{0}, FALSE },
     { -1,			NULL,		OPTV_NONE,	{0}, FALSE }
 };
@@ -287,6 +296,46 @@ static const char *ramdacSymbols[] = {
     NULL
 };
 
+#ifdef XF86DRI 
+static const char *drmSymbols[] = {
+    "drmAvailable",
+    "drmAddBufs",
+    "drmAddMap",
+    "drmCtlInstHandler",
+    "drmGetInterruptFromBusID",
+    "drmAgpAcquire",
+    "drmAgpRelease",
+    "drmAgpEnable",
+    "drmAgpAlloc",
+    "drmAgpFree",
+    "drmAgpBind",
+    "drmAgpGetMode",
+    "drmAgpBase",
+    "drmAgpSize",
+    "drmMgaCleanupDma",
+    "drmMgaLockUpdate",
+    "drmMgaInitDma",
+    NULL
+};
+
+static const char *driSymbols[] = {
+    "DRIGetDrawableIndex",
+    "DRIFinishScreenInit",
+    "DRIDestroyInfoRec",
+    "DRICloseScreen",
+    "DRIDestroyInfoRec",
+    "DRIScreenInit",
+    "DRIDestroyInfoRec",
+    "DRICreateInfoRec",
+    "DRILock",
+    "DRIUnlock",
+    "DRIGetSAREAPrivate",
+    "DRIGetContext",
+    "GlxSetVisualConfigs",
+    NULL
+};
+#endif
+
 #define MGAuseI2C 1
 
 static const char *ddcSymbols[] = {
@@ -309,6 +358,12 @@ static const char *shadowSymbols[] = {
     NULL
 };
 
+static const char *vbeSymbols[] = {
+    "VBEInit",
+    "vbeDoEDID",
+    NULL
+};
+
 static const char *fbdevHWSymbols[] = {
 	"fbdevHWInit",
 	"fbdevHWUseBuildinMode",
@@ -381,7 +436,12 @@ mgaSetup(pointer module, pointer opts, int *errmaj, int *errmin)
 	LoaderRefSymLists(vgahwSymbols, cfbSymbols, xaaSymbols, 
 			  xf8_32bppSymbols, ramdacSymbols,
 			  ddcSymbols, i2cSymbols, shadowSymbols,
-			  fbdevHWSymbols, NULL);
+			  fbdevHWSymbols, vbeSymbols,
+#ifdef XF86DRI 
+			  drmSymbols, driSymbols,
+#endif
+			  
+			  NULL);
 
 	/*
 	 * The return value must be non-NULL on success even though there
@@ -439,6 +499,12 @@ MGAFreeRec(ScrnInfoPtr pScrn)
     pScrn->driverPrivate = NULL;
 }
 
+static
+OptionInfoPtr
+MGAAvailableOptions(int chipid, int busid)
+{
+    return MGAOptions;
+}
 
 /* Mandatory */
 static void
@@ -453,7 +519,7 @@ static Bool
 MGAProbe(DriverPtr drv, int flags)
 {
     int i;
-    GDevPtr *devSections;
+    GDevPtr *devSections = NULL;
     int *usedChips;
     int numDevSections;
     int numUsed;
@@ -509,12 +575,15 @@ MGAProbe(DriverPtr drv, int flags)
 			MGAChipsets, MGAPciChipsets, devSections,
 			numDevSections, drv, &usedChips);
     /* Free it since we don't need that list after this */
-    xfree(devSections);
+    if (devSections)
+	xfree(devSections);
     devSections = NULL;
     if (numUsed <= 0)
 	return FALSE;
 
-    for (i = 0; i < numUsed; i++) {
+    if (flags & PROBE_DETECT)
+	foundScreen = TRUE;
+    else for (i = 0; i < numUsed; i++) {
 	ScrnInfoPtr pScrn;
 #ifdef DISABLE_VGA_IO
 	MgaSavePtr smga;
@@ -694,8 +763,6 @@ MGAReadBios(ScrnInfoPtr pScrn)
 	      ErrorF("Pins[0x%02x] is 0x%02x\n", i,
 			((unsigned char *)pBios2)[i]);
 #endif
-	   ErrorF("0x20 of Pins[0x34] is %s\n",
-		  (pBios2->VidCtrl & 0x20) ?  "set" : "cleared");
 	   return;
 	} else {
 	  /* Set default MCLK values (scaled by 10 kHz) */
@@ -719,7 +786,7 @@ static void
 MGASoftReset(ScrnInfoPtr pScrn)
 {
 	MGAPtr pMga = MGAPTR(pScrn);
-	int i;
+/*  	int i; */
 
 	pMga->FbMapSize = 8192 * 1024;
 	MGAMapMem(pScrn);
@@ -733,6 +800,9 @@ MGASoftReset(ScrnInfoPtr pScrn)
 	OUTREG(MGAREG_MACCESS, 1<<15);
 	usleep(10);
 
+#if 0
+	/* This will hang if the PLLs aren't on */
+
 	/* wait until drawing engine is ready */
 	while ( MGAISBUSY() )
 	    usleep(1000);
@@ -742,6 +812,7 @@ MGASoftReset(ScrnInfoPtr pScrn)
 	WAITFIFO(i);
 	while ( i-- )
 	    OUTREG(MGAREG_SHIFT, 0);
+#endif
 
 	MGAUnmapMem(pScrn);
 }
@@ -754,86 +825,99 @@ MGASoftReset(ScrnInfoPtr pScrn)
 static int
 MGACountRam(ScrnInfoPtr pScrn)
 {
-	MGAPtr pMga = MGAPTR(pScrn);
-
-	if (pMga->FbAddress)
-	{
-		volatile unsigned char* base;
-		unsigned char tmp, tmp3, tmp5;
-	
-		pMga->FbMapSize = 8192 * 1024;
-		MGAMapMem(pScrn);
-		base = pMga->FbBase;
-	
-		/* turn MGA mode on - enable linear frame buffer (CRTCEXT3) */
-		OUTREG8(0x1FDE, 3);
-		tmp = INREG8(0x1FDF);
-		OUTREG8(0x1FDF, tmp | 0x80);
-	
-		/* write, read and compare method */
-		base[0x500000] = 0x55;
-		base[0x300000] = 0x33;
-		base[0x100000] = 0x11;
-		tmp5 = base[0x500000];
-		tmp3 = base[0x300000];
-
-		/* restore CRTCEXT3 state */
-		OUTREG8(0x1FDE, 3);
-		OUTREG8(0x1FDF, tmp);
-	
-		MGAUnmapMem(pScrn);
-	
-		if(tmp5 == 0x55)
-			return 8192;
-		if(tmp3 == 0x33)
-			return 4096;
-	}
-	return 2048;
-}
+    MGAPtr pMga = MGAPTR(pScrn);
+    int ProbeSize = 8192;
+    int SizeFound = 2048;
+    CARD32 biosInfo = 0;
 
-/*
- * GetAccelPitchValues -
- *
- * This function returns a list of display width (pitch) values that can
- * be used in accelerated mode.
- */
 #if 0
-static int *
-GetAccelPitchValues(ScrnInfoPtr pScrn)
-{
-    int *linePitches = NULL;
-    int i, n = 0;
-    MGAPtr pMga = MGAPTR(pScrn);
-	
-    /* XXX ajv - 512, 576, and 1536 may not be supported
-       line pitches. see sdk pp 4-59 for more
-       details. Why anyone would want less than 640 is 
-       bizarre. (maybe lots of pixels tall?) */
-
-    /* The only line pitches the accelerator supports */
-#if 0		
-    int accelWidths[] = { 512, 576, 640, 768, 800, 960, 
-		    1024, 1152, 1280, 1536, 1600, 1920, 2048, 0 };
-#else
-    int accelWidths[] = { 640, 768, 800, 960, 1024, 1152, 1280,
-		    1600, 1920, 2048, 0 };
+    /* This isn't correct. It looks like this can have arbitrary
+	data for the memconfig even when the bios has initialized
+	it.  At least, my cards don't advertise the documented 
+	values (my 8 and 16 Meg G200s have the same values) */
+    if(pMga->Primary) /* can only trust this for primary cards */
+	biosInfo = pciReadLong(pMga->PciTag, PCI_OPTION_REG);
 #endif
 
-    for (i = 0; accelWidths[i] != 0; i++) {
-	if (accelWidths[i] % pMga->Rounding == 0) {
-	    n++;
-	    linePitches = xnfrealloc(linePitches, n * sizeof(int));
-	    linePitches[n - 1] = accelWidths[i];
+    switch(pMga->Chipset) {
+    case PCI_CHIP_MGA2164:
+    case PCI_CHIP_MGA2164_AGP:
+	xf86DrvMsg(pScrn->scrnIndex, X_WARNING, 
+		"Unable to probe memory amount due to hardware bug.  "
+		"Assuming 4096 KB\n");
+	return 4096;
+    case PCI_CHIP_MGAG400:
+	if(biosInfo) {
+	    switch((biosInfo >> 10) & 0x07) {
+	    case 0:
+		return (biosInfo & (1 << 14)) ? 32768 : 16384;
+	    case 1:
+	    case 2:	    
+		return 16384;
+	    case 3:	    
+	    case 5:	    
+		return 65536;
+	    case 4:	   
+		return 32768;
+	    }
 	}
+	ProbeSize = 32768;
+	break;
+    case PCI_CHIP_MGAG200:
+    case PCI_CHIP_MGAG200_PCI:
+	if(biosInfo) {
+	    switch((biosInfo >> 11) & 0x03) {
+	    case 0:
+		return 8192;
+	    default:
+		return 16384;
+	    }
+	}
+	ProbeSize = 16384;
+	break;
+    case PCI_CHIP_MGAG100:
+	if(biosInfo) /* I'm not sure if the docs are correct */
+	    return (biosInfo & (1 << 12)) ? 16384 : 8192;
+    case PCI_CHIP_MGA1064:
+    case PCI_CHIP_MGA2064:
+	ProbeSize = 8192;
+        break;
+    default:
+        break;
     }
-    /* Mark the end of the list */
-    if (n > 0) {
-	linePitches = xnfrealloc(linePitches, (n + 1) * sizeof(int));
-	linePitches[n] = 0;
-    }
-    return linePitches;
+
+    if (pMga->FbAddress) {
+	volatile unsigned char* base;
+	unsigned char tmp;
+	int i;
+	
+	pMga->FbMapSize = ProbeSize * 1024;
+	MGAMapMem(pScrn);
+	base = pMga->FbBase;
+
+	/* turn MGA mode on - enable linear frame buffer (CRTCEXT3) */
+	OUTREG8(0x1FDE, 3);
+	tmp = INREG8(0x1FDF);
+	OUTREG8(0x1FDF, tmp | 0x80);
+	
+	/* write, read and compare method */
+	for(i = ProbeSize; i > 2048; i -= 2048) {
+	    base[(i * 1024) - 1] = 0xAA;
+	    OUTREG8(MGAREG_CRTC_INDEX, 0);  /* flush the cache */
+	    if(base[(i * 1024) - 1] == 0xAA) {
+		SizeFound = i;
+		break;
+	    }
+	}
+	
+	/* restore CRTCEXT3 state */
+	OUTREG8(0x1FDE, 3);
+	OUTREG8(0x1FDF, tmp);
+	
+	MGAUnmapMem(pScrn);
+   }
+   return SizeFound;
 }
-#endif
 
 static xf86MonPtr
 MGAdoDDC(ScrnInfoPtr pScrn)
@@ -878,7 +962,6 @@ MGAdoDDC(ScrnInfoPtr pScrn)
   /* Initialize I2C bus - used by DDC if available */
   if (pMga->i2cInit) {
     pMga->i2cInit(pScrn);
-ErrorF("I2C initialized on %p\n",pMga->I2C);
   }
   /* Read and output monitor info using DDC2 over I2C bus */
   if (pMga->I2C) {
@@ -908,6 +991,8 @@ ErrorF("I2C initialized on %p\n",pMga->I2C);
     vgaHWUnmapMem(pScrn);
   }
 
+  xf86SetDDCproperties(pScrn, MonInfo);
+
   return MonInfo;
 }
 
@@ -970,6 +1055,18 @@ VgaIOEnable(void *arg)
 }
 #endif /* DISABLE_VGA_IO */
 
+extern xf86MonPtr ConfiguredMonitor;
+
+void
+MGAProbeDDC(ScrnInfoPtr pScrn, int index)
+{
+    vbeInfoPtr pVbe;
+    if (xf86LoadSubModule(pScrn, "vbe")) {
+	pVbe = VBEInit(NULL,index);
+	ConfiguredMonitor = vbeDoEDID(pVbe);
+    }
+}
+
 /* Mandatory */
 static Bool
 MGAPreInit(ScrnInfoPtr pScrn, int flags)
@@ -1002,6 +1099,23 @@ MGAPreInit(ScrnInfoPtr pScrn, int flags)
     if (pScrn->numEntities != 1)
 	return FALSE;
 
+    /* Allocate the MGARec driverPrivate */
+    if (!MGAGetRec(pScrn)) {
+	return FALSE;
+    }
+
+    pMga = MGAPTR(pScrn);
+
+    /* Get the entity, and make sure it is PCI. */
+    pMga->pEnt = xf86GetEntityInfo(pScrn->entityList[0]);
+    if (pMga->pEnt->location.type != BUS_PCI)
+	return FALSE;
+
+    if (flags & PROBE_DETECT) {
+	MGAProbeDDC(pScrn, pMga->pEnt->index);
+	return TRUE;
+    }
+
     /* The vgahw module should be loaded here when needed */
     if (!xf86LoadSubModule(pScrn, "vgahw"))
 	return FALSE;
@@ -1014,16 +1128,19 @@ MGAPreInit(ScrnInfoPtr pScrn, int flags)
     if (!vgaHWGetHWRec(pScrn))
 	return FALSE;
 
-    /* Allocate the MGARec driverPrivate */
-    if (!MGAGetRec(pScrn)) {
-	return FALSE;
-    }
-    pMga = MGAPTR(pScrn);
+#if 0
+    /* This is causing problems with restoring the card to it's
+       original state.  If this is to be done, it needs to happen
+       after we've saved the original state */
+    /* Initialize the card through int10 interface if needed */
+    if ( xf86LoadSubModule(pScrn, "int10")){
+        xf86Int10InfoPtr pInt;
 
-    /* Get the entity, and make sure it is PCI. */
-    pMga->pEnt = xf86GetEntityInfo(pScrn->entityList[0]);
-    if (pMga->pEnt->location.type != BUS_PCI)
-	return FALSE;
+        xf86DrvMsg(pScrn->scrnIndex, X_INFO, "Initializing int10\n");
+        pInt = xf86InitInt10(pMga->pEnt->index);
+        xf86FreeInt10(pInt);
+    }
+#endif
 
     /* Find the PCI info for this screen */
     pMga->PciInfo = xf86GetPciInfoForEntity(pMga->pEnt->index);
@@ -1060,8 +1177,6 @@ MGAPreInit(ScrnInfoPtr pScrn, int flags)
     xf86SetAccessFuncs(pMga->pEnt, &pMga->Access, &pMga->Access,
 			&pMga->Access, NULL);
 #endif
-
-    pScrn->racMemFlags = RAC_FB | RAC_COLORMAP | RAC_CURSOR | RAC_VIEWPORT;
   
     /* Set pScrn->monitor */
     pScrn->monitor = pScrn->confScreen->monitor;
@@ -1120,17 +1235,8 @@ MGAPreInit(ScrnInfoPtr pScrn, int flags)
         }
     }
 
-    if (!xf86SetDefaultVisual(pScrn, -1)) {
+    if (!xf86SetDefaultVisual(pScrn, -1))
 	return FALSE;
-    } else {
-	/* We don't currently support DirectColor at > 8bpp */
-	if (pScrn->depth > 8 && pScrn->defaultVisual != TrueColor) {
-	    xf86DrvMsg(pScrn->scrnIndex, X_ERROR, "Given default visual"
-		       " (%s) is not supported at depth %d\n",
-		       xf86GetVisualName(pScrn->defaultVisual), pScrn->depth);
-	    return FALSE;
-	}
-    }
 
     bytesPerPixel = pScrn->bitsPerPixel / 8;
 
@@ -1144,20 +1250,52 @@ MGAPreInit(ScrnInfoPtr pScrn, int flags)
     xf86ProcessOptions(pScrn->scrnIndex, pScrn->options, MGAOptions);
 
     /* Set the bits per RGB for 8bpp mode */
-    if (pScrn->depth == 8) {
-	/* XXX This is here just to test options. */
-	/* Default to 8 */
+    if (pScrn->depth == 8) 
 	pScrn->rgbBits = 8;
-#if 0
-	if (xf86GetOptValInteger(MGAOptions, OPTION_RGB_BITS,
-				 &pScrn->rgbBits)) {
-	    xf86DrvMsg(pScrn->scrnIndex, X_CONFIG, "Bits per RGB set to %d\n",
-		       pScrn->rgbBits);
-	}
-#endif
+
+    /*
+     * Set the Chipset and ChipRev, allowing config file entries to
+     * override.
+     */
+    if (pMga->pEnt->device->chipset && *pMga->pEnt->device->chipset) {
+	pScrn->chipset = pMga->pEnt->device->chipset;
+        pMga->Chipset = xf86StringToToken(MGAChipsets, pScrn->chipset);
+        from = X_CONFIG;
+    } else if (pMga->pEnt->device->chipID >= 0) {
+	pMga->Chipset = pMga->pEnt->device->chipID;
+	pScrn->chipset = (char *)xf86TokenToString(MGAChipsets, pMga->Chipset);
+	from = X_CONFIG;
+	xf86DrvMsg(pScrn->scrnIndex, X_CONFIG, "ChipID override: 0x%04X\n",
+		   pMga->Chipset);
+    } else {
+	from = X_PROBED;
+	pMga->Chipset = pMga->PciInfo->chipType;
+	pScrn->chipset = (char *)xf86TokenToString(MGAChipsets, pMga->Chipset);
+    }
+    if (pMga->pEnt->device->chipRev >= 0) {
+	pMga->ChipRev = pMga->pEnt->device->chipRev;
+	xf86DrvMsg(pScrn->scrnIndex, X_CONFIG, "ChipRev override: %d\n",
+		   pMga->ChipRev);
+    } else {
+	pMga->ChipRev = pMga->PciInfo->chipRev;
     }
 
+    /*
+     * This shouldn't happen because such problems should be caught in
+     * MGAProbe(), but check it just in case.
+     */
+    if (pScrn->chipset == NULL) {
+	xf86DrvMsg(pScrn->scrnIndex, X_ERROR,
+		   "ChipID 0x%04X is not recognised\n", pMga->Chipset);
+	return FALSE;
+    }
+    if (pMga->Chipset < 0) {
+	xf86DrvMsg(pScrn->scrnIndex, X_ERROR,
+		   "Chipset \"%s\" is not recognised\n", pScrn->chipset);
+	return FALSE;
+    }
 
+    xf86DrvMsg(pScrn->scrnIndex, from, "Chipset: \"%s\"\n", pScrn->chipset);
     from = X_DEFAULT;
     pMga->HWCursor = TRUE;
     /*
@@ -1199,10 +1337,7 @@ MGAPreInit(ScrnInfoPtr pScrn, int flags)
     }
     if ((s = xf86GetOptValString(MGAOptions, OPTION_OVERLAY))) {
       if (!*s || !xf86NameCmp(s, "8,24") || !xf86NameCmp(s, "24,8")) {
-	if(pMga->Chipset == PCI_CHIP_MGAG100) {
-	    xf86DrvMsg(pScrn->scrnIndex, X_CONFIG, 
-		"Option \"Overlay\" is not supported by the G100\n");
-	} else if(pScrn->bitsPerPixel == 32) {
+	if(pScrn->bitsPerPixel == 32) {
 	    pMga->Overlay8Plus24 = TRUE;
 	    if(!xf86GetOptValInteger(
 			MGAOptions, OPTION_COLOR_KEY,&(pMga->colorKey)))
@@ -1213,13 +1348,21 @@ MGAPreInit(ScrnInfoPtr pScrn, int flags)
 				"PseudoColor overlay enabled\n");
 	} else {
 	    xf86DrvMsg(pScrn->scrnIndex, X_CONFIG, 
-		"Option \"Overlay\" is not supported in this configuration\n");
+		"Option \"Overlay\" is only supported in 32 bits per pixel\n");
 	}
       } else {
 	    xf86DrvMsg(pScrn->scrnIndex, X_CONFIG, 
 		"\"%s\" is not a valid value for Option \"Overlay\"\n", s);
       }
-    } 
+    }
+    if(xf86GetOptValInteger(MGAOptions, OPTION_VIDEO_KEY, &(pMga->videoKey))) {
+	xf86DrvMsg(pScrn->scrnIndex, X_CONFIG, "video key set to 0x%x\n",
+				pMga->videoKey);
+    } else {
+	pMga->videoKey =  (1 << pScrn->offset.red) | 
+			  (1 << pScrn->offset.green) |
+        (((pScrn->mask.blue >> pScrn->offset.blue) - 1) << pScrn->offset.blue); 
+    }
     if (xf86ReturnOptValBool(MGAOptions, OPTION_SHADOW_FB, FALSE)) {
 	pMga->ShadowFB = TRUE;
 	pMga->NoAccel = TRUE;
@@ -1273,57 +1416,21 @@ MGAPreInit(ScrnInfoPtr pScrn, int flags)
       }
     }
 
-
-    /*
-     * Set the Chipset and ChipRev, allowing config file entries to
-     * override.
-     */
-    if (pMga->pEnt->device->chipset && *pMga->pEnt->device->chipset) {
-	pScrn->chipset = pMga->pEnt->device->chipset;
-        pMga->Chipset = xf86StringToToken(MGAChipsets, pScrn->chipset);
-        from = X_CONFIG;
-    } else if (pMga->pEnt->device->chipID >= 0) {
-	pMga->Chipset = pMga->pEnt->device->chipID;
-	pScrn->chipset = (char *)xf86TokenToString(MGAChipsets, pMga->Chipset);
-	from = X_CONFIG;
-	xf86DrvMsg(pScrn->scrnIndex, X_CONFIG, "ChipID override: 0x%04X\n",
-		   pMga->Chipset);
-    } else {
-	from = X_PROBED;
-	pMga->Chipset = pMga->PciInfo->chipType;
-	pScrn->chipset = (char *)xf86TokenToString(MGAChipsets, pMga->Chipset);
-    }
-    if (pMga->pEnt->device->chipRev >= 0) {
-	pMga->ChipRev = pMga->pEnt->device->chipRev;
-	xf86DrvMsg(pScrn->scrnIndex, X_CONFIG, "ChipRev override: %d\n",
-		   pMga->ChipRev);
-    } else {
-	pMga->ChipRev = pMga->PciInfo->chipRev;
-    }
-
-    /*
-     * This shouldn't happen because such problems should be caught in
-     * MGAProbe(), but check it just in case.
-     */
-    if (pScrn->chipset == NULL) {
-	xf86DrvMsg(pScrn->scrnIndex, X_ERROR,
-		   "ChipID 0x%04X is not recognised\n", pMga->Chipset);
-	return FALSE;
-    }
-    if (pMga->Chipset < 0) {
-	xf86DrvMsg(pScrn->scrnIndex, X_ERROR,
-		   "Chipset \"%s\" is not recognised\n", pScrn->chipset);
-	return FALSE;
-    }
-
-    xf86DrvMsg(pScrn->scrnIndex, from, "Chipset: \"%s\"\n", pScrn->chipset);
-
-    if ((pMga->PciInfo->subsysCard == PCI_CARD_MILL_G200_SD) ||
+    if(pMga->HasSDRAM) { /* don't bother checking */ }
+    else if ((pMga->PciInfo->subsysCard == PCI_CARD_MILL_G200_SD) ||
 	(pMga->PciInfo->subsysCard == PCI_CARD_MARV_G200_SD) ||
 	(pMga->PciInfo->subsysCard == PCI_CARD_MYST_G200_SD) ||
 	(pMga->PciInfo->subsysCard == PCI_CARD_PROD_G100_SD)) {
         pMga->HasSDRAM = TRUE;
 	xf86DrvMsg(pScrn->scrnIndex, X_PROBED, "Has SDRAM\n");
+    } 
+    else if (pMga->Primary && (pMga->Chipset != PCI_CHIP_MGA2064) && 
+		(pMga->Chipset != PCI_CHIP_MGA2164) &&
+		(pMga->Chipset != PCI_CHIP_MGA2164_AGP)) {	
+	if(!(pciReadLong(pMga->PciTag, PCI_OPTION_REG) & (1 << 14))) {
+	    pMga->HasSDRAM = TRUE;
+	    xf86DrvMsg(pScrn->scrnIndex, X_PROBED, "Has SDRAM\n");
+	}
     }
 
     switch (pMga->Chipset) {
@@ -1464,28 +1571,21 @@ MGAPreInit(ScrnInfoPtr pScrn, int flags)
     /*
      * Reset card if it isn't primary one
      */
-    if (!pMga->Primary && !pMga->FBDev)
+    if ( (!pMga->Primary && !pMga->FBDev) || xf86IsPc98() )
         MGASoftReset(pScrn);
+
     /*
      * If the user has specified the amount of memory in the XF86Config
      * file, we respect that setting.
      */
+    from = X_PROBED;
     if (pMga->pEnt->device->videoRam != 0) {
 	pScrn->videoRam = pMga->pEnt->device->videoRam;
 	from = X_CONFIG;
-    } else if((pMga->Chipset == PCI_CHIP_MGA2164) ||
-	(pMga->Chipset == PCI_CHIP_MGA2164_AGP)){
-	pScrn->videoRam = 4096;
-	from = X_DEFAULT;
-	xf86DrvMsg(pScrn->scrnIndex, X_WARNING, 
-	 "Skipping memory probe due to hardware bug.  Assuming 4096 kBytes\n");
+    } else if (pMga->FBDev) {
+	pScrn->videoRam = fbdevHWGetVidmem(pScrn)/1024;
     } else {
-	if (pMga->FBDev) {
-	    pScrn->videoRam = fbdevHWGetVidmem(pScrn)/1024;
-	} else {
-	    pScrn->videoRam = MGACountRam(pScrn);
-	}
-	from = X_PROBED;
+	pScrn->videoRam = MGACountRam(pScrn);
     }
     xf86DrvMsg(pScrn->scrnIndex, from, "VideoRAM: %d kByte\n",
                pScrn->videoRam);
@@ -1847,6 +1947,7 @@ MGAPreInit(ScrnInfoPtr pScrn, int flags)
     pMga->CurrentLayout.weight.green = pScrn->weight.green;
     pMga->CurrentLayout.weight.blue = pScrn->weight.blue;
     pMga->CurrentLayout.Overlay8Plus24 = pMga->Overlay8Plus24;
+    pMga->CurrentLayout.mode = pScrn->currentMode;
 
     return TRUE;
 }
@@ -1860,37 +1961,30 @@ static Bool
 MGAMapMem(ScrnInfoPtr pScrn)
 {
     MGAPtr pMga;
-    int mmioFlags;
 
     pMga = MGAPTR(pScrn);
 
     /*
      * Map IO registers to virtual address space
      */ 
-#if !defined(__alpha__)
-    mmioFlags = VIDMEM_MMIO | VIDMEM_READSIDEEFFECT;
-#else
     /*
      * For Alpha, we need to map SPARSE memory, since we need
-     * byte/short access.
+     * byte/short access.  This is taken care of automatically by the
+     * os-support layer.
      */
-    mmioFlags = VIDMEM_MMIO | VIDMEM_READSIDEEFFECT | VIDMEM_SPARSE;
-#endif
-    pMga->IOBase = xf86MapPciMem(pScrn->scrnIndex, mmioFlags, pMga->PciTag,
-				 pMga->IOAddress, 0x4000);
+    pMga->IOBase = xf86MapPciMem(pScrn->scrnIndex,
+				 VIDMEM_MMIO | VIDMEM_READSIDEEFFECT,
+				 pMga->PciTag, pMga->IOAddress, 0x4000);
     if (pMga->IOBase == NULL)
 	return FALSE;
 
 #ifdef __alpha__
-    /*
-     * for Alpha, we need to map DENSE memory as well, for
-     * setting CPUToScreenColorExpandBase.
-     */
-    pMga->IOBaseDense = xf86MapPciMem(pScrn->scrnIndex, VIDMEM_MMIO,
+    pMga->IOBaseDense = xf86MapPciMem(pScrn->scrnIndex,
+				      VIDMEM_MMIO | VIDMEM_MMIO_32BIT,
 				      pMga->PciTag, pMga->IOAddress, 0x4000);
     if (pMga->IOBaseDense == NULL)
 	return FALSE;
-#endif /* __alpha__ */
+#endif
 
     pMga->FbBase = xf86MapPciMem(pScrn->scrnIndex, VIDMEM_FRAMEBUFFER,
 				 pMga->PciTag, pMga->FbAddress,
@@ -1905,7 +1999,8 @@ MGAMapMem(ScrnInfoPtr pScrn)
 	DWORD access on DWORD boundaries to this window */
     if (pMga->ILOADAddress) {
 	pMga->ILOADBase = xf86MapPciMem(pScrn->scrnIndex,
-				VIDMEM_MMIO | VIDMEM_READSIDEEFFECT, 
+				VIDMEM_MMIO | VIDMEM_MMIO_32BIT |
+				    VIDMEM_READSIDEEFFECT, 
 				pMga->PciTag, pMga->ILOADAddress, 0x800000);
     } else
 	pMga->ILOADBase = NULL;
@@ -1961,11 +2056,6 @@ MGAUnmapMem(ScrnInfoPtr pScrn)
     xf86UnMapVidMem(pScrn->scrnIndex, (pointer)pMga->IOBase, 0x4000);
     pMga->IOBase = NULL;
 
-#ifdef __alpha__
-    xf86UnMapVidMem(pScrn->scrnIndex, (pointer)pMga->IOBaseDense, 0x4000);
-    pMga->IOBaseDense = NULL;
-#endif /* __alpha__ */
-
     xf86UnMapVidMem(pScrn->scrnIndex, (pointer)pMga->FbBase, pMga->FbMapSize);
     pMga->FbBase = NULL;
     pMga->FbStart = NULL;
@@ -2039,9 +2129,16 @@ MGAModeInit(ScrnInfoPtr pScrn, DisplayModePtr mode)
     vgaReg = &hwp->ModeReg;
     mgaReg = &pMga->ModeReg;
 
+#ifdef XF86DRI
+   if (pMga->directRenderingEnabled) {
+       DRILock(screenInfo.screens[pScrn->scrnIndex], 0);
+/*         MGADRISwapContext(screenInfo.screens[pScrn->scrnIndex]); */
+   }
+#endif
+     
     (*pMga->Restore)(pScrn, vgaReg, mgaReg, FALSE);
-
-    MGAStormSync(pScrn);
+     
+    MGAStormSync(pScrn);	/* JEFF - Check */
     MGAStormEngineInit(pScrn);
     vgaHWProtect(pScrn, FALSE);
 
@@ -2052,6 +2149,13 @@ MGAModeInit(ScrnInfoPtr pScrn, DisplayModePtr mode)
 	    outb(0xfac, 0x02);
     }
 
+    pMga->CurrentLayout.mode = mode;
+   
+#ifdef XF86DRI
+   if (pMga->directRenderingEnabled)
+     DRIUnlock(screenInfo.screens[pScrn->scrnIndex]);
+#endif
+
     return TRUE;
 }
 
@@ -2129,7 +2233,6 @@ MGAScreenInit(int scrnIndex, ScreenPtr pScreen, int argc, char **argv)
 	pciSetBitsLong(pMga->PciTag, PCI_OPTION_REG, 0x100, 0x000);
 	if (!fbdevHWModeInit(pScrn, pScrn->currentMode))
 	    return FALSE;
-	MGAStormSync(pScrn);
 	MGAStormEngineInit(pScrn);
     } else {
 	/* Save the current state */
@@ -2141,7 +2244,7 @@ MGAScreenInit(int scrnIndex, ScreenPtr pScreen, int argc, char **argv)
 
 
     /* Darken the screen for aesthetic reasons and set the viewport */
-    MGASaveScreen(pScreen, FALSE);
+    MGASaveScreen(pScreen, SCREEN_SAVER_ON);
     pScrn->AdjustFrame(scrnIndex, pScrn->frameX0, pScrn->frameY0, 0);
 
     /*
@@ -2195,7 +2298,7 @@ MGAScreenInit(int scrnIndex, ScreenPtr pScreen, int argc, char **argv)
 
     if(pMga->ShadowFB) {
  	pMga->ShadowPitch = BitmapBytePad(pScrn->bitsPerPixel * width);
-        pMga->ShadowPtr = xalloc(pMga->ShadowPitch * height);
+	pMga->ShadowPtr = xalloc(pMga->ShadowPitch * height);
 	displayWidth = pMga->ShadowPitch / (pScrn->bitsPerPixel >> 3);
         FBStart = pMga->ShadowPtr;
     } else {
@@ -2203,6 +2306,19 @@ MGAScreenInit(int scrnIndex, ScreenPtr pScreen, int argc, char **argv)
 	FBStart = pMga->FbStart;
     }
 
+#ifdef XF86DRI
+     /*
+      * Setup DRI after visuals have been established, but before cfbScreenInit
+      * is called.   cfbScreenInit will eventually call into the drivers
+      * InitGLXVisuals call back.
+      */
+    if (!pMga->NoAccel)
+     pMga->directRenderingEnabled = MGADRIScreenInit(pScreen);
+    else
+     pMga->directRenderingEnabled = FALSE;
+#endif
+     
+   
     switch (pScrn->bitsPerPixel) {
     case 8:
 	ret = cfbScreenInit(pScreen, FBStart,
@@ -2276,6 +2392,7 @@ MGAScreenInit(int scrnIndex, ScreenPtr pScreen, int argc, char **argv)
 
     miInitializeBackingStore(pScreen);
     xf86SetBackingStore(pScreen);
+    xf86SetSilkenMouse(pScreen);
 
     /* Initialize software cursor.  
 	Must precede creation of the default colormap */
@@ -2309,9 +2426,11 @@ MGAScreenInit(int scrnIndex, ScreenPtr pScreen, int argc, char **argv)
 	RefreshAreaFuncPtr refreshArea = MGARefreshArea;
 
 	if(pMga->Rotate) {
-   	   pMga->PointerMoved = pScrn->PointerMoved;
-	   pScrn->PointerMoved = MGAPointerMoved;
-
+	    if (!pMga->PointerMoved) {
+	    pMga->PointerMoved = pScrn->PointerMoved;
+	    pScrn->PointerMoved = MGAPointerMoved;
+	    }
+	    
 	   switch(pScrn->bitsPerPixel) {
 	   case 8:	refreshArea = MGARefreshArea8;	break;
 	   case 16:	refreshArea = MGARefreshArea16;	break;
@@ -2326,21 +2445,29 @@ MGAScreenInit(int scrnIndex, ScreenPtr pScreen, int argc, char **argv)
 #ifdef DPMSExtension
     xf86DPMSInit(pScreen, MGADisplayPowerManagementSet, 0);
 #endif
-
+   
+#ifdef XF86DRI
+    /* Initialize the Warp engine */
+   if (pMga->directRenderingEnabled) {
+    pMga->directRenderingEnabled = mgaConfigureWarp(pScrn);
+   }
+   if (pMga->directRenderingEnabled) {
+      /* Now that mi, cfb, drm and others have done their thing, 
+       * complete the DRI setup.
+       */
+       pMga->directRenderingEnabled = MGADRIFinishScreenInit(pScreen);
+   }
+   if (pMga->directRenderingEnabled) {
+       xf86DrvMsg(pScrn->scrnIndex, X_INFO, "direct rendering enabled\n");
+   } else {
+       xf86DrvMsg(pScrn->scrnIndex, X_INFO, "direct rendering disabled\n");
+   }
+#endif
+     
     pScrn->memPhysBase = pMga->FbAddress;
     pScrn->fbOffset = pMga->YDstOrg * (pScrn->bitsPerPixel / 8);
 
-#ifdef XvExtension
-    {
-	XF86VideoAdaptorPtr *ptr;
-	int n;
-	
-	n = xf86XVListGenericAdaptors(&ptr);
-	if (n) { 
-	    xf86XVScreenInit(pScreen, ptr, n);
-	}
-    }
-#endif
+    MGAInitVideo(pScreen);
 
     pScreen->SaveScreen = MGASaveScreen;
 
@@ -2384,14 +2511,8 @@ MGAAdjustFrame(int scrnIndex, int x, int y, int flags)
     pLayout = &pMga->CurrentLayout;
 
 
-    if(pMga->ShowCache && y && pScrn->vtSema) {
-	int lastline = pMga->FbUsableSize / 
-		(pScrn->displayWidth * pScrn->bitsPerPixel/8);
-
-	lastline -= pScrn->currentMode->VDisplay;
+    if(pMga->ShowCache && y && pScrn->vtSema)
 	y += pScrn->virtualY - 1;
-        if(y > lastline) y = lastline;
-    }
 
     Base = (y * pLayout->displayWidth + x + pMga->YDstOrg) >>
 		(3 - pMga->BppShifts[(pLayout->bitsPerPixel >> 3) - 1]);
@@ -2430,7 +2551,17 @@ static Bool
 MGAEnterVT(int scrnIndex, int flags)
 {
     ScrnInfoPtr pScrn = xf86Screens[scrnIndex];
+#ifdef XF86DRI 
+    ScreenPtr pScreen;
+    MGAPtr pMGA;
 
+    pMGA = MGAPTR(pScrn);
+    if (pMGA->directRenderingEnabled) {
+        pScreen = screenInfo.screens[scrnIndex];
+        DRIUnlock(pScreen);
+    }
+#endif
+     
     if (!MGAModeInit(pScrn, pScrn->currentMode))
 	return FALSE;
     MGAAdjustFrame(scrnIndex, pScrn->frameX0, pScrn->frameY0, 0);
@@ -2441,9 +2572,18 @@ static Bool
 MGAEnterVTFBDev(int scrnIndex, int flags)
 {
     ScrnInfoPtr pScrn = xf86Screens[scrnIndex];
+#ifdef XF86DRI 
+    ScreenPtr pScreen;
+    MGAPtr pMGA;
+
+    pMGA = MGAPTR(pScrn);
+    if (pMGA->directRenderingEnabled) {
+        pScreen = screenInfo.screens[scrnIndex];
+        DRIUnlock(pScreen);
+    }
+#endif
 
     fbdevHWEnterVT(scrnIndex,flags);
-    MGAStormSync(pScrn);
     MGAStormEngineInit(pScrn);
     return TRUE;
 }
@@ -2461,12 +2601,24 @@ MGALeaveVT(int scrnIndex, int flags)
 {
     ScrnInfoPtr pScrn = xf86Screens[scrnIndex];
     vgaHWPtr hwp = VGAHWPTR(pScrn);
+#ifdef XF86DRI
+    ScreenPtr pScreen;
+    MGAPtr pMGA;
+#endif
 
     MGARestore(pScrn);
     vgaHWLock(hwp);
 
     if (xf86IsPc98())
 	outb(0xfac, 0x00);
+#ifdef XF86DRI
+    pMGA = MGAPTR(pScrn);
+    if (pMGA->directRenderingEnabled) {
+        pScreen = screenInfo.screens[scrnIndex];
+        DRILock(pScreen, 0);
+    }
+#endif
+     
 }
 
 
@@ -2496,6 +2648,13 @@ MGACloseScreen(int scrnIndex, ScreenPtr pScreen)
 	    vgaHWUnmapMem(pScrn);
 	}
     }
+#ifdef XF86DRI 
+   if (pMga->directRenderingEnabled) {
+       MGADRICloseScreen(pScreen);
+       pMga->directRenderingEnabled=FALSE;
+   }
+#endif
+     
     if (pMga->AccelInfoRec)
 	XAADestroyInfoRec(pMga->AccelInfoRec);
     if (pMga->CursorInfoRec)
@@ -2504,11 +2663,17 @@ MGACloseScreen(int scrnIndex, ScreenPtr pScreen)
 	xfree(pMga->ShadowPtr);
     if (pMga->DGAModes)
 	xfree(pMga->DGAModes);
+    if (pMga->adaptor)
+	xfree(pMga->adaptor);
+
     pScrn->vtSema = FALSE;
 
     if (xf86IsPc98())
 	outb(0xfac, 0x00);
 
+    if(pMga->BlockHandler)
+	pScreen->BlockHandler = pMga->BlockHandler;
+
     pScreen->CloseScreen = pMga->CloseScreen;
     return (*pScreen->CloseScreen)(scrnIndex, pScreen);
 }
@@ -2524,7 +2689,8 @@ MGAFreeScreen(int scrnIndex, int flags)
      * This only gets called when a screen is being deleted.  It does not
      * get called routinely at the end of a server generation.
      */
-    vgaHWFreeHWRec(xf86Screens[scrnIndex]);
+    if (xf86LoaderCheckSymbol("vgaHWFreeHWRec"))
+	vgaHWFreeHWRec(xf86Screens[scrnIndex]);
     MGAFreeRec(xf86Screens[scrnIndex]);
 }
 
@@ -2558,9 +2724,9 @@ MGAValidMode(int scrnIndex, DisplayModePtr mode, Bool verbose, int flags)
 
 /* Mandatory */
 static Bool
-MGASaveScreen(ScreenPtr pScreen, Bool unblank)
+MGASaveScreen(ScreenPtr pScreen, int mode)
 {
-    return vgaHWSaveScreen(pScreen, unblank);
+    return vgaHWSaveScreen(pScreen, mode);
 }
 
 
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_drm.c b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_drm.c
deleted file mode 100644
index 6dbebb162..000000000
--- a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_drm.c
+++ /dev/null
@@ -1,100 +0,0 @@
-#include "xf86.h"
-#include "xf86_OSproc.h"
-#include "xf86Resources.h"
-#include "xf86_ansic.h"
-#include "compiler.h"
-#include "xf86PciInfo.h"
-#include "xf86Pci.h"
-#include "xf86Priv.h"
-#include "mga_bios.h"
-#include "mga_reg.h"
-#include "mga.h"
-#include "mga_macros.h"
-#include "mga_dri.h"
-
-#ifndef XFree86LOADER
-#include <sys/stat.h>
-#include <sys/mman.h>
-#endif
-
-#include "xf86drm.h"
-#include "drm.h"
-#include "mga_drm_public.h"
-
-Bool mgadrmCleanupDma(ScrnInfoPtr pScrn)
-{
-   MGAPtr pMGA = MGAPTR(pScrn);
-   drm_mga_init_t init;
-   
-   memset(&init, 0, sizeof(drm_mga_init_t));
-   init.func = MGA_CLEANUP_DMA;
-   
-   if(ioctl(pMGA->drmSubFD, DRM_IOCTL_MGA_INIT, &init)) {
-      ErrorF("Mga Dma Cleanup Failed\n");
-      return FALSE;
-   }
-   
-   return TRUE;
-}
-
-int mgadrmInitDma(ScrnInfoPtr pScrn, int prim_size)
-{
-   MGAPtr pMGA = MGAPTR(pScrn);
-   MGADRIPtr pMGADRI = (MGADRIPtr)pMGA->pDRIInfo->devPrivate;
-   MGADRIServerPrivatePtr pMGADRIServer = pMGA->DRIServerInfo;
-   drm_mga_init_t init;
-   
-   memset(&init, 0, sizeof(drm_mga_init_t));
-   init.func = MGA_INIT_DMA;
-   init.reserved_map_agpstart = 0;
-   init.reserved_map_idx = 3;
-   init.buffer_map_idx = 4;
-   init.sarea_priv_offset = sizeof(XF86DRISAREARec);
-   init.primary_size = prim_size;
-   init.warp_ucode_size = pMGADRIServer->warp_ucode_size;
-
-   switch(pMGA->Chipset) {
-   case PCI_CHIP_MGAG400:
-      init.chipset = MGA_CARD_TYPE_G400;
-      break;
-   case PCI_CHIP_MGAG200:
-      init.chipset = MGA_CARD_TYPE_G200;
-      break;
-   case PCI_CHIP_MGAG200_PCI:
-   default:
-      ErrorF("Direct rendering not supported on this card/chipset\n");
-      return FALSE;
-   }
-
-   init.fbOffset = pMGADRI->fbOffset;
-   init.backOffset = pMGADRI->backOffset;
-   init.depthOffset = pMGADRI->depthOffset;
-   init.textureOffset = pMGADRI->textureOffset;
-   init.textureSize = pMGADRI->textureSize;
-   init.cpp = pMGADRI->cpp;
-   init.stride = pMGADRI->stride;
-
-   init.frontOrg = pMGA->DstOrg;
-   init.backOrg = pMGADRI->backOffset; /*  */
-   init.depthOrg = pMGADRI->textureOffset;
-   init.mAccess = pMGA->MAccess;
-
-   if (pMGA->HasSDRAM) 
-      init.sgram = 0;
-   else 
-      init.sgram = 1;
-
-   memcpy(&init.WarpIndex, &pMGADRIServer->WarpIndex, 
-	  sizeof(mgaWarpIndex) * MGA_MAX_WARP_PIPES);
-
-
-
-   ErrorF("Mga Dma Initialization start\n");
-
-   if(ioctl(pMGA->drmSubFD, DRM_IOCTL_MGA_INIT, &init)) {
-      ErrorF("Mga Dma Initialization Failed\n");
-      return FALSE;
-   }
-   ErrorF("Mga Dma Initialization done\n");
-   return TRUE;
-}
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_macros.h b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_macros.h
index 15d57b287..19b137740 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_macros.h
+++ b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_macros.h
@@ -1,4 +1,4 @@
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/mga/mga_macros.h,v 1.10 1999/08/14 10:49:48 dawes Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/mga/mga_macros.h,v 1.12 2000/02/11 17:25:57 dawes Exp $ */
 
 #ifndef _MGA_MACROS_H_
 #define _MGA_MACROS_H_
@@ -19,10 +19,23 @@
 
 #define RGBEQUAL(c) (!((((c) >> 8) ^ (c)) & 0xffff)) 
 
+#ifdef XF86DRI
+#define MGAREG_DWGSYNC		      0x2c4c
+#define MGA_SYNC_XTAG                 0x275f4200
+
+#define MGABUSYWAIT() do { \
+OUTREG(MGAREG_DWGSYNC, MGA_SYNC_XTAG); \
+while(INREG(MGAREG_DWGSYNC) != MGA_SYNC_XTAG) ; \
+}while(0);
+
+#endif
+
 #define MGAISBUSY() (INREG8(MGAREG_Status + 2) & 0x01)
 
-#define WAITFIFO(n) \
+#define WAITFIFO(cnt) \
    if(!pMga->UsePCIRetry) {\
+	register int n = cnt; \
+	if(n > pMga->FifoSize) n = pMga->FifoSize; \
 	while(pMga->fifoCount < (n))\
 	    pMga->fifoCount = INREG8(MGAREG_FIFOSTATUS);\
 	pMga->fifoCount -= n;\
@@ -71,4 +84,18 @@
 	WAITFIFO(1); \
 	OUTREG(MGAREG_CXBNDRY, 0xFFFF0000); }
 
+
+#ifdef XF86DRI
+extern void mgaGetQuiescence( ScrnInfoPtr pScrn );
+
+#define CHECK_DMA_QUIESCENT(pMGA, pScrn) {	\
+   if (!pMGA->have_quiescense) {		\
+      mgaGetQuiescence( pScrn );		\
+   }						\
+}
+#else
+#define CHECK_DMA_QUIESCENT(pMGA, pScrn)
+#endif
+
+
 #endif /* _MGA_MACROS_H_ */
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_storm.c b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_storm.c
index e97030796..903f6d279 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_storm.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_storm.c
@@ -1,4 +1,4 @@
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/mga/mga_storm.c,v 1.59 1999/08/29 12:21:01 dawes Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/mga/mga_storm.c,v 1.64 2000/02/12 20:45:24 dawes Exp $ */
 
 
 /* All drivers should typically include these */
@@ -22,12 +22,21 @@
 #include "miline.h"
 #include "servermd.h"
 
+#ifdef XF86DRI
+#include "cfb.h"
+#include "GL/glxtokens.h"
+#endif
+
 #include "mga_bios.h"
 #include "mga.h"
 #include "mga_reg.h"
 #include "mga_map.h"
 #include "mga_macros.h"
 
+#ifdef XF86DRI
+#include "mga_dri.h"
+#endif
+
 static void MGANAME(SubsequentScreenToScreenCopy)(ScrnInfoPtr pScrn,
 				int srcX, int srcY, int dstX, int dstY,
 				int w, int h);
@@ -87,6 +96,14 @@ static void MGANAME(SubsequentDashedTwoPointLine)(ScrnInfoPtr pScrn,
 				int x1, int y1, int x2, int y2, 
 				int flags, int phase);
 
+#ifdef XF86DRI
+void MGANAME(DRIInitBuffers)(WindowPtr pWin, 
+			     RegionPtr prgn, CARD32 index);
+void MGANAME(DRIMoveBuffers)(WindowPtr pParent, DDXPointRec ptOldOrg,
+			     RegionPtr prgnSrc, CARD32 index);
+
+#endif
+
 extern void MGAWriteBitmapColorExpand(ScrnInfoPtr pScrn, int x, int y,
 				int w, int h, unsigned char *src, int srcwidth,
 				int skipleft, int fg, int bg, int rop,
@@ -113,6 +130,9 @@ extern void MGAValidatePolyArc(GCPtr, unsigned long, DrawablePtr);
 extern void MGAValidatePolyPoint(GCPtr, unsigned long, DrawablePtr);
 extern void MGAFillCacheBltRects(ScrnInfoPtr, int, unsigned int, int, BoxPtr,
 				int, int, XAACacheInfoPtr);
+extern void MGATEGlyphRenderer(ScrnInfoPtr, int, int, int, int, int, int, 
+				unsigned int **glyphs, int, int, int, int, 
+				unsigned planemask);
 
 Bool
 MGANAME(AccelInit)(ScreenPtr pScreen) 
@@ -122,6 +142,9 @@ MGANAME(AccelInit)(ScreenPtr pScreen)
     MGAPtr pMga = MGAPTR(pScrn);
     int maxFastBlitMem;
     BoxRec AvailFBArea;
+#ifdef XF86DRI
+    RegionRec MemRegion;
+#endif
 
     pMga->AccelInfoRec = infoPtr = XAACreateInfoRec();
     if(!infoPtr) return FALSE;
@@ -209,12 +232,10 @@ MGANAME(AccelInit)(ScreenPtr pScreen)
     /* clipping */
     infoPtr->SetClippingRectangle = MGASetClippingRectangle;
     infoPtr->DisableClipping = MGADisableClipping;
-    infoPtr->ClippingFlags = 	HARDWARE_CLIP_SOLID_LINE |
+    infoPtr->ClippingFlags = 	HARDWARE_CLIP_SOLID_LINE  |
 				HARDWARE_CLIP_DASHED_LINE |
-				HARDWARE_CLIP_SOLID_FILL |
-				HARDWARE_CLIP_MONO_8x8_FILL |
-				HARDWARE_CLIP_SCREEN_TO_SCREEN_COPY |
-				HARDWARE_CLIP_SCREEN_TO_SCREEN_COLOR_EXPAND;
+				HARDWARE_CLIP_SOLID_FILL  |
+				HARDWARE_CLIP_MONO_8x8_FILL; 
 
     /* dashed lines */
     infoPtr->DashedLineFlags = LINE_PATTERN_MSBFIRST_LSBJUSTIFIED;
@@ -247,7 +268,11 @@ MGANAME(AccelInit)(ScreenPtr pScreen)
 	infoPtr->ColorExpandBase = pMga->ILOADBase;
     } else {
 	infoPtr->ColorExpandRange = 0x1C00;
+#ifdef __alpha__
+	infoPtr->ColorExpandBase = pMga->IOBaseDense;
+#else
 	infoPtr->ColorExpandBase = pMga->IOBase;
+#endif
     }
     infoPtr->SetupForCPUToScreenColorExpandFill =
 		MGANAME(SetupForCPUToScreenColorExpandFill);
@@ -305,6 +330,7 @@ MGANAME(AccelInit)(ScreenPtr pScreen)
 			infoPtr->SubsequentCPUToScreenColorExpandFill) {
 	infoPtr->FillColorExpandRects = MGAFillColorExpandRects; 
 	infoPtr->WriteBitmap = MGAWriteBitmapColorExpand;
+	infoPtr->TEGlyphRenderer = MGATEGlyphRenderer;  
         if(BITMAP_SCANLINE_PAD == 32)
 	    infoPtr->NonTEGlyphRenderer = MGANonTEGlyphRenderer;  
     }
@@ -343,6 +369,7 @@ MGANAME(AccelInit)(ScreenPtr pScreen)
 	infoPtr->FillSolidSpansFlags |= NO_PLANEMASK;
 	infoPtr->FillMono8x8PatternRectsFlags |= NO_PLANEMASK;
 	infoPtr->NonTEGlyphRendererFlags |= NO_PLANEMASK;
+	infoPtr->TEGlyphRendererFlags |= NO_PLANEMASK;
 	infoPtr->FillCacheBltRectsFlags |= NO_PLANEMASK;
     }
 
@@ -355,10 +382,23 @@ MGANAME(AccelInit)(ScreenPtr pScreen)
 
 
     AvailFBArea.x1 = 0;
-    AvailFBArea.y1 = 0;
     AvailFBArea.x2 = pScrn->displayWidth;
-    AvailFBArea.y2 = pMga->FbUsableSize / (pScrn->displayWidth * PSZ / 8);
+
+#ifndef XF86DRI
+    AvailFBArea.y1 = 0;
+    AvailFBArea.y2 = (min(pMga->FbUsableSize, 16*1024*1024)) / 
+			(pScrn->displayWidth * PSZ / 8);
+   
     xf86InitFBManager(pScreen, &AvailFBArea); 
+#else
+    AvailFBArea.y1 = pScrn->virtualY;
+    AvailFBArea.y2 = pScrn->virtualY+128;
+   
+    REGION_INIT(pScreen, &MemRegion, &AvailFBArea, 1);
+    xf86InitFBManagerRegion(pScreen, &MemRegion);
+    REGION_UNINIT(pScreen, &MemRegion);
+#endif
+   
 
     return(XAAInit(pScreen, infoPtr));
 }
@@ -414,6 +454,8 @@ void
 MGAStormSync(ScrnInfoPtr pScrn) 
 {
     MGAPtr pMga = MGAPTR(pScrn);
+
+    CHECK_DMA_QUIESCENT(pMga, pScrn);
     
     while(MGAISBUSY());
     /* flush cache before a read (mga-1064g 5.1.6) */
@@ -430,6 +472,8 @@ void MGAStormEngineInit(ScrnInfoPtr pScrn)
     MGAPtr pMga = MGAPTR(pScrn);
     MGAFBLayout *pLayout = &pMga->CurrentLayout;
 
+    CHECK_DMA_QUIESCENT(pMga, pScrn);
+
     if (pMga->Chipset == PCI_CHIP_MGAG100)
     	maccess = 1 << 14;
     
@@ -451,10 +495,18 @@ void MGAStormEngineInit(ScrnInfoPtr pScrn)
     
     pMga->fifoCount = 0;
 
-    WAITFIFO(12);
+    while(MGAISBUSY());
+
+    if(!pMga->FifoSize) {
+	pMga->FifoSize = INREG8(MGAREG_FIFOSTATUS);
+	xf86DrvMsg(pScrn->scrnIndex, X_PROBED, "%i DWORD fifo\n",
+						pMga->FifoSize);
+    }
+
     OUTREG(MGAREG_PITCH, pLayout->displayWidth);
     OUTREG(MGAREG_YDSTORG, pMga->YDstOrg);
     OUTREG(MGAREG_MACCESS, maccess);
+    pMga->MAccess = maccess;
     pMga->PlaneMask = ~0;
     if(pMga->Chipset != PCI_CHIP_MGAG100)
 	OUTREG(MGAREG_PLNWT, pMga->PlaneMask);
@@ -490,6 +542,8 @@ void MGASetClippingRectangle(
 ){
     MGAPtr pMga = MGAPTR(pScrn);
 
+    CHECK_DMA_QUIESCENT(pMga, pScrn);
+
     WAITFIFO(3);
     OUTREG(MGAREG_CXBNDRY,(x2 << 16) | x1);      
     OUTREG(MGAREG_YTOP, (y1 * pScrn->displayWidth) + pMga->YDstOrg); 
@@ -501,6 +555,8 @@ void MGADisableClipping(ScrnInfoPtr pScrn)
 {
     MGAPtr pMga = MGAPTR(pScrn);
 
+    CHECK_DMA_QUIESCENT(pMga, pScrn);
+
     WAITFIFO(3);
     OUTREG(MGAREG_CXBNDRY, 0xFFFF0000);     /* (maxX << 16) | minX */ 
     OUTREG(MGAREG_YTOP, 0x00000000);        /* minPixelPointer */ 
@@ -531,6 +587,9 @@ MGANAME(SetupForScreenToScreenCopy)(
     CARD32 dwgctl = pMga->AtypeNoBLK[rop] | MGADWG_SHIFTZERO | 
 			MGADWG_BITBLT | MGADWG_BFCOL;
 
+    CHECK_DMA_QUIESCENT(pMga, pScrn);
+
+
     pMga->AccelInfoRec->SubsequentScreenToScreenCopy = 
 		MGANAME(SubsequentScreenToScreenCopy);
 
@@ -718,6 +777,8 @@ MGANAME(SetupForSolidFill)(
 {
     MGAPtr pMga = MGAPTR(pScrn);
 
+    CHECK_DMA_QUIESCENT(pMga, pScrn);
+
 #if PSZ == 24
     if(!RGBEQUAL(color))
     pMga->FilledRectCMD = MGADWG_TRAP | MGADWG_SOLID | MGADWG_ARZERO | 
@@ -843,6 +904,8 @@ MGANAME(SetupForMono8x8PatternFill)(
     MGAPtr pMga = MGAPTR(pScrn);
     XAAInfoRecPtr infoRec = pMga->AccelInfoRec;
 
+    CHECK_DMA_QUIESCENT(pMga, pScrn);
+
     pMga->PatternRectCMD = MGADWG_TRAP | MGADWG_ARZERO | MGADWG_SGNZERO | 
 						MGADWG_BMONOLEF;
 
@@ -957,6 +1020,8 @@ MGANAME(SetupForCPUToScreenColorExpandFill)(
     CARD32 mgaCMD = MGADWG_ILOAD | MGADWG_LINEAR | MGADWG_SGNZERO | 
 			MGADWG_SHIFTZERO | MGADWG_BMONOLEF;
         
+    CHECK_DMA_QUIESCENT(pMga, pScrn);
+
     if(bg == -1) {
 #if PSZ == 24
     	if(!RGBEQUAL(fg))
@@ -1017,6 +1082,8 @@ static void MGANAME(SetupForImageWrite)(
 ){
     MGAPtr pMga = MGAPTR(pScrn);
 
+    CHECK_DMA_QUIESCENT(pMga, pScrn);
+
     WAITFIFO(3);
     OUTREG(MGAREG_AR5, 0);
     SET_PLANEMASK(planemask);
@@ -1061,6 +1128,8 @@ MGANAME(SetupForDashedLine)(
     CARD32 NiceDashPattern = DashPattern[0];
     int dwords = (length + 31) >> 5;
 
+    CHECK_DMA_QUIESCENT(pMga, pScrn);
+
     pMga->DashCMD = MGADWG_BFCOL | pMga->AtypeNoBLK[rop];
     pMga->StyleLen = length - 1;
 
@@ -1168,6 +1237,8 @@ MGANAME(SetupForPlanarScreenToScreenColorExpandFill)(
     CARD32 mgaCMD = pMga->AtypeNoBLK[rop] | MGADWG_BITBLT | 
 				MGADWG_SGNZERO | MGADWG_BPLAN;
         
+    CHECK_DMA_QUIESCENT(pMga, pScrn);
+
     if(bg == -1) {
 	mgaCMD |= MGADWG_TRANSC;
 	WAITFIFO(4);
@@ -1218,6 +1289,8 @@ MGANAME(SetupForScreenToScreenColorExpandFill)(
 ){
     MGAPtr pMga = MGAPTR(pScrn);
     CARD32 mgaCMD = MGADWG_BITBLT | MGADWG_SGNZERO | MGADWG_SHIFTZERO;
+
+    CHECK_DMA_QUIESCENT(pMga, pScrn);
         
     if(bg == -1) {
 #if PSZ == 24
@@ -1259,32 +1332,35 @@ MGANAME(SubsequentScreenToScreenColorExpandFill)(
     MGAPtr pMga = MGAPTR(pScrn);
     int pitch = pScrn->displayWidth * PSZ;
     int start, end, next, num;
-    int SrcOrg = 0, DstOrg = 0;
+    Bool resetDstOrg = FALSE;
 
     if (pMga->AccelFlags & LARGE_ADDRESSES) {
-        DstOrg = ((y & ~1023) * pScrn->displayWidth * PSZ) >> 9;
-        SrcOrg = ((srcy & ~1023) * pScrn->displayWidth * PSZ) >> 9;
+        int DstOrg = ((y & ~1023) * pScrn->displayWidth * PSZ) >> 9;
+        int SrcOrg = ((srcy & ~1023) * pScrn->displayWidth * PSZ) >> 9;
+
 	y &= 1023;
+	srcy &= 1023;
 
 	WAITFIFO(2);
-	if(DstOrg)
+	if(DstOrg) {
             OUTREG(MGAREG_DSTORG, DstOrg << 6);
+	    resetDstOrg = TRUE;
+	}
         if(SrcOrg != pMga->SrcOrg) {
             pMga->SrcOrg = SrcOrg;
             OUTREG(MGAREG_SRCORG, SrcOrg << 6);
         }
-	SrcOrg <<= 9;
     }
 
     w--;
     start = (XYADDRESS(srcx, srcy) * PSZ) + skipleft;
     end = start + w + (pitch * (h - 1));
 
-    /* src cannot split a 2 Meg boundary */
+    /* src cannot split a 2 Meg boundary from SrcOrg */
     if(!((start ^ end) & 0xff000000)) {
 	WAITFIFO(4);
-	OUTREG(MGAREG_AR3, start - SrcOrg);
-	OUTREG(MGAREG_AR0, start + w - SrcOrg);
+	OUTREG(MGAREG_AR3, start);
+	OUTREG(MGAREG_AR0, start + w);
 	OUTREG(MGAREG_FXBNDRY, ((x + w) << 16) | (x & 0xffff));
 	OUTREG(MGAREG_YDSTLEN + MGAREG_EXEC, (y << 16) | h);
     } else {
@@ -1294,13 +1370,13 @@ MGANAME(SubsequentScreenToScreenColorExpandFill)(
 		num = next - start - 1;
 
 		WAITFIFO(7);
-		OUTREG(MGAREG_AR3, start - SrcOrg);
-		OUTREG(MGAREG_AR0, start + num - SrcOrg);
+		OUTREG(MGAREG_AR3, start);
+		OUTREG(MGAREG_AR0, start + num);
 		OUTREG(MGAREG_FXBNDRY, ((x + num) << 16) | (x & 0xffff));
-		OUTREG(MGAREG_YDSTLEN + MGAREG_EXEC, (y << 16) | h);
+		OUTREG(MGAREG_YDSTLEN + MGAREG_EXEC, (y << 16) | 1);
 		
-		OUTREG(MGAREG_AR3, next - SrcOrg);
-		OUTREG(MGAREG_AR0, start + w - SrcOrg);
+		OUTREG(MGAREG_AR3, next);
+		OUTREG(MGAREG_AR0, start + w );
 		OUTREG(MGAREG_FXBNDRY + MGAREG_EXEC, ((x + w) << 16) | 
                                                      ((x + num + 1) & 0xffff));
 		start += pitch;
@@ -1310,8 +1386,8 @@ MGANAME(SubsequentScreenToScreenColorExpandFill)(
 		if(num > h) num = h;
 
 		WAITFIFO(4);
-		OUTREG(MGAREG_AR3, start - SrcOrg);
-		OUTREG(MGAREG_AR0, start + w - SrcOrg);
+		OUTREG(MGAREG_AR3, start);
+		OUTREG(MGAREG_AR0, start + w);
 		OUTREG(MGAREG_FXBNDRY, ((x + w) << 16) | (x & 0xffff));
 		OUTREG(MGAREG_YDSTLEN + MGAREG_EXEC, (y << 16) | num);
 
@@ -1321,7 +1397,7 @@ MGANAME(SubsequentScreenToScreenColorExpandFill)(
 	}
     }
 
-    if(DstOrg) {
+    if(resetDstOrg) {
 	WAITFIFO(1);
 	OUTREG(MGAREG_DSTORG, 0);
     }
@@ -1375,6 +1451,8 @@ MGAWriteBitmapColorExpand(
     CARD32* maxptr;
     int dwords, maxlines, count;
 
+    CHECK_DMA_QUIESCENT(pMga, pScrn);
+
     (*infoRec->SetupForCPUToScreenColorExpandFill)(
 				pScrn, fg, bg, rop, planemask);
 
@@ -1389,6 +1467,7 @@ MGAWriteBitmapColorExpand(
 			(pScrn, x, y, w, maxlines, skipleft);
 	count = maxlines;
 	while(count--) {
+	    WAITFIFO(dwords);
 	    destptr = MoveDWORDS(destptr, (CARD32*)src, dwords);
 	    src += srcwidth;
 	    if(destptr > maxptr) 
@@ -1402,13 +1481,14 @@ MGAWriteBitmapColorExpand(
 				pScrn, x, y, w, h, skipleft);
 
     while(h--) {
+	WAITFIFO(dwords);
 	destptr = MoveDWORDS(destptr, (CARD32*)src, dwords);
 	src += srcwidth;
 	if(destptr > maxptr) 
 		destptr = (CARD32*)infoRec->ColorExpandBase;
     }
-    MGAStormSync(infoRec->pScrn);
-    infoRec->NeedToSync = FALSE;
+    DISABLE_CLIP();
+    SET_SYNC_FLAG(infoRec);
 }
 
 
@@ -1441,6 +1521,8 @@ MGAFillColorExpandRects(
     } else
     	StippleFunc = XAAStippleScanlineFuncLSBFirst[2];
 
+    CHECK_DMA_QUIESCENT(pMga, pScrn);
+
     (*infoRec->SetupForCPUToScreenColorExpandFill)(
 				pScrn, fg, bg, rop, planemask);
 
@@ -1464,6 +1546,7 @@ MGAFillColorExpandRects(
 			pBox->x1, y, pBox->x2 - pBox->x1, maxlines, 0);
 	   count = maxlines;
 	   while(count--) {
+		WAITFIFO(dwords);
 		destptr = (*StippleFunc)(
 			destptr, (CARD32*)srcp, srcx, stipplewidth, dwords);
 		if(destptr > maxptr) 
@@ -1483,6 +1566,7 @@ MGAFillColorExpandRects(
 			pBox->x1, y , pBox->x2 - pBox->x1, h, 0);
 
 	while(h--) {
+	    WAITFIFO(dwords);
 	    destptr = (*StippleFunc)(
 		destptr, (CARD32*)srcp, srcx, stipplewidth, dwords);
 	    if(destptr > maxptr) 
@@ -1497,8 +1581,8 @@ MGAFillColorExpandRects(
 
 	pBox++;
     }
-    MGAStormSync(infoRec->pScrn);
-    infoRec->NeedToSync = FALSE;
+    DISABLE_CLIP();
+    SET_SYNC_FLAG(infoRec);
 }
 
 
@@ -1514,6 +1598,8 @@ MGAFillSolidRectsDMA(
     XAAInfoRecPtr infoRec = pMga->AccelInfoRec;
     CARD32 *base = (CARD32*)pMga->ILOADBase;
 
+    CHECK_DMA_QUIESCENT(pMga, pScrn);
+
     SET_SYNC_FLAG(infoRec);
     (*infoRec->SetupForSolidFill)(pScrn, fg, rop, planemask);
 
@@ -1554,6 +1640,8 @@ MGAFillSolidSpansDMA(
     XAAInfoRecPtr infoRec = pMga->AccelInfoRec;
     CARD32 *base = (CARD32*)pMga->ILOADBase;
 
+    CHECK_DMA_QUIESCENT(pMga, pScrn);
+
     SET_SYNC_FLAG(infoRec);
     
     if(infoRec->ClipBox) {
@@ -1614,6 +1702,8 @@ MGAFillMono8x8PatternRectsTwoPass(
     int	nBox, SecondPassColor;
     BoxPtr pBox;
 
+    CHECK_DMA_QUIESCENT(pMga, pScrn);
+
     if((rop == GXcopy) && (bg != -1)) {
 	SecondPassColor = bg;
 	bg = -1;
@@ -1662,6 +1752,8 @@ void MGANonTEGlyphRenderer(
     int x1, x2, y1, y2, i, h, skiptop, dwords, maxlines;
     unsigned char *src;
 
+    CHECK_DMA_QUIESCENT(pMga, pScrn);
+
     (*infoRec->SetupForCPUToScreenColorExpandFill)(
 					pScrn, fg, -1, rop, planemask);
     WAITFIFO(1);
@@ -1697,6 +1789,7 @@ void MGANonTEGlyphRenderer(
 	    OUTREG(MGAREG_AR3, 0);
 	    OUTREG(MGAREG_FXBNDRY, ((x2 - 1) << 16) | (x1 & 0xFFFF));
 	    OUTREG(MGAREG_YDSTLEN + MGAREG_EXEC, (y1 << 16) | h);
+	    WAITFIFO(dwords * maxlines);
 	    MoveDWORDS((CARD32*)infoRec->ColorExpandBase, 
 				(CARD32*)src, dwords * maxlines);
 	    src += dwords * maxlines << 2;
@@ -1710,12 +1803,12 @@ void MGANonTEGlyphRenderer(
 	OUTREG(MGAREG_AR3, 0);
 	OUTREG(MGAREG_FXBNDRY, ((x2 - 1) << 16) | (x1 & 0xFFFF));
 	OUTREG(MGAREG_YDSTLEN + MGAREG_EXEC, (y1 << 16) | h);
-
+	WAITFIFO(dwords);
 	MoveDWORDS((CARD32*)infoRec->ColorExpandBase, (CARD32*)src, dwords);
     }  
 
-    MGAStormSync(infoRec->pScrn);
-    infoRec->NeedToSync = FALSE;
+    DISABLE_CLIP();
+    SET_SYNC_FLAG(infoRec);
 }
 
 void
@@ -1767,6 +1860,8 @@ MGAPolyPoint (
 
     pbox = REGION_RECTS(pGC->pCompositeClip);
 
+    CHECK_DMA_QUIESCENT(pMga, infoRec->pScrn);
+
     (*infoRec->SetClippingRectangle)(infoRec->pScrn,
                 pbox->x1, pbox->y1, pbox->x2 - 1, pbox->y2 - 1);
     (*infoRec->SetupForSolidFill)(infoRec->pScrn, pGC->fgPixel, pGC->alu,
@@ -1831,6 +1926,8 @@ MGAFillCacheBltRects(
     XAAInfoRecPtr infoRec = GET_XAAINFORECPTR_FROM_SCRNINFOPTR(pScrn);
     int x, y, phaseY, phaseX, skipleft, height, width, w, blit_w, blit_h, start;
 
+    CHECK_DMA_QUIESCENT(MGAPTR(pScrn), pScrn);
+
     (*infoRec->SetupForScreenToScreenCopy)(pScrn, 1, 1, rop, planemask,
 		pCache->trans_color);
 
@@ -1909,14 +2006,223 @@ MGAFillCacheBltRects(
     SET_SYNC_FLAG(infoRec);
 }
 
+void 
+MGATEGlyphRenderer(
+    ScrnInfoPtr pScrn,
+    int x, int y, int w, int h, int skipleft, int startline, 
+    unsigned int **glyphs, int glyphWidth,
+    int fg, int bg, int rop, unsigned planemask
+){
+    XAAInfoRecPtr infoRec = GET_XAAINFORECPTR_FROM_SCRNINFOPTR(pScrn);
+    GlyphScanlineFuncPtr GlyphFunc = 
+			XAAGlyphScanlineFuncLSBFirst[glyphWidth - 1];
+    MGAPtr pMga = MGAPTR(pScrn);
+    CARD32* base;
+    int dwords;
 
-#endif
+    CHECK_DMA_QUIESCENT(pMga, pScrn);
 
+    (*infoRec->SetupForCPUToScreenColorExpandFill)(
+				pScrn, fg, bg, rop, planemask);
 
+    w += skipleft;
+    x -= skipleft;
+    dwords = (w + 31) >> 5;
 
+    (*infoRec->SubsequentCPUToScreenColorExpandFill)(
+				pScrn, x, y, w, h, skipleft);
 
+    base = (CARD32*)infoRec->ColorExpandBase;
 
+    if((dwords * h) <= infoRec->ColorExpandRange)
+	while(h--) {
+	    WAITFIFO(dwords);
+	    base = (*GlyphFunc)(base, glyphs, startline++, w, glyphWidth);
+	}
+    else
+	while(h--) {
+	    WAITFIFO(dwords);
+	    (*GlyphFunc)(base, glyphs, startline++, w, glyphWidth);
+	}
 
+    DISABLE_CLIP();
+    SET_SYNC_FLAG(infoRec);
+}
 
+#endif
 
+#ifdef XF86DRI
+void
+MGANAME(DRIInitBuffers)(WindowPtr pWin, RegionPtr prgn, CARD32 index)
+{
+  ScreenPtr pScreen = pWin->drawable.pScreen;
+  ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
+  MGAPtr pMGA = MGAPTR(pScrn);
+  BoxPtr pbox = REGION_RECTS(prgn);
+  int nbox  = REGION_NUM_RECTS(prgn);
+
+  CHECK_DMA_QUIESCENT(MGAPTR(pScrn), pScrn);
+
+  MGANAME(SetupForSolidFill)(pScrn, 0, GXcopy, -1);
+  while (nbox--) {
+    MGASelectBuffer(pScrn, MGA_BACK);
+    MGANAME(SubsequentSolidFillRect)(pScrn, pbox->x1, pbox->y1,
+				     pbox->x2-pbox->x1, pbox->y2-pbox->y1);
+    MGASelectBuffer(pScrn, MGA_DEPTH);
+    MGANAME(SubsequentSolidFillRect)(pScrn, pbox->x1, pbox->y1,
+				     pbox->x2-pbox->x1, pbox->y2-pbox->y1);
+    pbox++;
+  }
+  MGASelectBuffer(pScrn, MGA_FRONT);
+
+  pMGA->AccelInfoRec->NeedToSync = TRUE;
+}
 
+/*
+  This routine is a modified form of XAADoBitBlt with the calls to
+  ScreenToScreenBitBlt built in. My routine has the prgnSrc as source
+  instead of destination. My origin is upside down so the ydir cases
+  are reversed.
+*/
+	   
+void
+MGANAME(DRIMoveBuffers)(WindowPtr pParent, DDXPointRec ptOldOrg, 
+		   RegionPtr prgnSrc, CARD32 index)
+{
+  ScreenPtr pScreen = pParent->drawable.pScreen;
+  ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
+  MGAPtr pMGA = MGAPTR(pScrn);
+  int nbox;
+  BoxPtr pbox, pboxTmp, pboxNext, pboxBase, pboxNew1, pboxNew2;
+  DDXPointPtr pptTmp, pptNew1, pptNew2;
+  int xdir, ydir;
+  int dx, dy;
+  DDXPointPtr pptSrc;
+
+  int screenwidth = pScrn->virtualX;
+  int screenheight = pScrn->virtualY;
+
+  CHECK_DMA_QUIESCENT(MGAPTR(pScrn), pScrn);
+
+  pbox = REGION_RECTS(prgnSrc);
+  nbox = REGION_NUM_RECTS(prgnSrc);
+  pboxNew1 = 0;
+  pptNew1 = 0;
+  pboxNew2 = 0;
+  pboxNew2 = 0;
+  pptSrc = &ptOldOrg;
+
+  dx = pParent->drawable.x - ptOldOrg.x;
+  dy = pParent->drawable.y - ptOldOrg.y;
+
+  /* If the copy will overlap in Y, reverse the order */
+  if (dy>0) {
+    ydir = -1;
+
+    if (nbox>1) {
+      /* Keep ordering in each band, reverse order of bands */
+      pboxNew1 = (BoxPtr)ALLOCATE_LOCAL(sizeof(BoxRec)*nbox);
+      if (!pboxNew1) return;
+      pptNew1 = (DDXPointPtr)ALLOCATE_LOCAL(sizeof(DDXPointRec)*nbox);
+      if (!pptNew1) {
+	DEALLOCATE_LOCAL(pboxNew1);
+	return;
+      }
+      pboxBase = pboxNext = pbox+nbox-1;
+      while (pboxBase >= pbox) {
+	while ((pboxNext >= pbox) && (pboxBase->y1 == pboxNext->y1))
+	  pboxNext--;
+	pboxTmp = pboxNext+1;
+	pptTmp = pptSrc + (pboxTmp - pbox);
+	while (pboxTmp <= pboxBase) {
+	  *pboxNew1++ = *pboxTmp++;
+	  *pptNew1++ = *pptTmp++;
+	}
+	pboxBase = pboxNext;
+      }
+      pboxNew1 -= nbox;
+      pbox = pboxNew1;
+      pptNew1 -= nbox;
+      pptSrc = pptNew1;
+    }
+  } else {
+    /* No changes required */
+    ydir = 1;
+  }
+
+  /* If the regions will overlap in X, reverse the order */
+  if (dx>0) {
+    xdir = -1;
+
+    if (nbox > 1) {
+      /*reverse orderof rects in each band */
+      pboxNew2 = (BoxPtr)ALLOCATE_LOCAL(sizeof(BoxRec)*nbox);
+      pptNew2 = (DDXPointPtr)ALLOCATE_LOCAL(sizeof(DDXPointRec)*nbox);
+      if (!pboxNew2 || !pptNew2) {
+	if (pptNew2) DEALLOCATE_LOCAL(pptNew2);
+	if (pboxNew2) DEALLOCATE_LOCAL(pboxNew2);
+	if (pboxNew1) {
+	  DEALLOCATE_LOCAL(pptNew1);
+	  DEALLOCATE_LOCAL(pboxNew1);
+	}
+	return;
+      }
+      pboxBase = pboxNext = pbox;
+      while (pboxBase < pbox+nbox) {
+	while ((pboxNext < pbox+nbox) && (pboxNext->y1 == pboxBase->y1))
+	  pboxNext++;
+	pboxTmp = pboxNext;
+	pptTmp = pptSrc + (pboxTmp - pbox);
+	while (pboxTmp != pboxBase) {
+	  *pboxNew2++ = *--pboxTmp;
+	  *pptNew2++ = *--pptTmp;
+	}
+	pboxBase = pboxNext;
+      }
+      pboxNew2 -= nbox;
+      pbox = pboxNew2;
+      pptNew2 -= nbox;
+      pptSrc = pptNew2;
+    }
+  } else {
+    /* No changes are needed */
+    xdir = 1;
+  }
+  
+  MGANAME(SetupForScreenToScreenCopy)(pScrn, xdir, ydir, GXcopy, -1, -1);
+  for ( ; nbox-- ; pbox++) 
+  {
+      int x1 = pbox->x1;
+      int y1 = pbox->y1;
+      int destx = x1 + dx;
+      int desty = y1 + dy;
+      int w = pbox->x2 - x1 + 1;
+      int h = pbox->y2 - y1 + 1;
+      
+      if ( destx < 0 ) x1 -= destx, w += destx, destx = 0; 
+      if ( desty < 0 ) y1 -= desty, h += desty, desty = 0;
+      if ( destx + w > screenwidth ) w = screenwidth - destx;
+      if ( desty + h > screenheight ) h = screenheight - desty;
+      if ( w <= 0 ) continue;
+      if ( h <= 0 ) continue;
+      
+    MGASelectBuffer(pScrn, MGA_BACK);
+    MGANAME(SubsequentScreenToScreenCopy)(pScrn, x1,y1, destx,desty, w, h);
+    MGASelectBuffer(pScrn, MGA_DEPTH);
+    MGANAME(SubsequentScreenToScreenCopy)(pScrn, x1,y1, destx,desty, w, h);
+  }
+  MGASelectBuffer(pScrn, MGA_FRONT);
+
+  if (pboxNew2) {
+    DEALLOCATE_LOCAL(pptNew2);
+    DEALLOCATE_LOCAL(pboxNew2);
+  }
+  if (pboxNew1) {
+    DEALLOCATE_LOCAL(pptNew1);
+    DEALLOCATE_LOCAL(pboxNew1);
+  }
+
+  pMGA->AccelInfoRec->NeedToSync = TRUE;
+}
+	   
+#endif /* XF86DRI */
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_warp.c b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_warp.c
index 9bd43318a..f7c81968c 100644
--- a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_warp.c
+++ b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_warp.c
@@ -18,6 +18,8 @@ Bool mgaConfigureWarp(ScrnInfoPtr pScrn)
    MGAPtr pMga = MGAPTR(pScrn);
    int wmisc;
    
+   CHECK_DMA_QUIESCENT( pMga, pScrn );
+
    WAITFIFO(3);
    
    switch(pMga->Chipset) {
@@ -108,7 +110,7 @@ static unsigned int mgaG400InstallMicrocode(MGAPtr pMGA, int agp_offset)
    unsigned int microcode_size = 0;
 
    memset(pMGADRIServer->WarpIndex, 0, 
-	  sizeof(mgaWarpIndex) * MGA_MAX_WARP_PIPES);
+	  sizeof(drmMgaWarpIndex) * MGA_MAX_WARP_PIPES);
    microcode_size = mgaG400GetMicrocodeSize(pMGA);
 
    mgaWarpInstallCode(tgz,     MGA_WARP_TGZ);
@@ -150,7 +152,7 @@ static unsigned int mgaG200InstallMicrocode(MGAPtr pMGA, int agp_offset)
    unsigned int microcode_size = 0;
 
    memset(pMGADRIServer->WarpIndex, 0, 
-	  sizeof(mgaWarpIndex) * MGA_MAX_WARP_PIPES);
+	  sizeof(drmMgaWarpIndex) * MGA_MAX_WARP_PIPES);
 
    microcode_size = mgaG400GetMicrocodeSize(pMGA);   
    mgaWarpInstallCode(tgz,    MGA_WARP_TGZ);
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_wrap.c b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_wrap.c
new file mode 100644
index 000000000..d812149e1
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_wrap.c
@@ -0,0 +1,198 @@
+/* $XFree86: xc/programs/Xserver/GL/dri/dri.c,v 1.4 1999/09/25 14:36:41 dawes Exp $ */
+/**************************************************************************
+
+Copyright 1998-1999 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, 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 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:
+ *   Keith Whitwell <keithw@precisioninsight.com>
+ *
+ */
+
+
+#include "xf86.h"
+#include "xf86_OSproc.h"
+#include "xf86_ansic.h"
+#include "xf86Priv.h"
+
+#include "xf86PciInfo.h"
+#include "xf86Pci.h"
+#define PSZ 8
+#include "cfb.h"
+#undef PSZ
+#include "cfb16.h"
+#include "cfb32.h"
+
+#include "miline.h"
+
+#include "GL/glxtokens.h"
+
+#include "mga_bios.h"
+#include "mga_reg.h"
+#include "mga.h"
+#include "mga_macros.h"
+#include "mga_dri.h"
+#include "mga_wrap.h"
+
+
+
+static void MGAWakeupHandler(int screenNum,
+			     pointer wakeupData,
+			     unsigned long result,
+			     pointer pReadmask)
+{
+    ScreenPtr pScreen = screenInfo.screens[screenNum];
+    DRIWrappedFuncsRec *pDRIWrap = DRIGetWrappedFuncs(pScreen);
+
+    if (0) ErrorF("MGAWakeupHandler (in)\n");
+
+    /* Disabled: Check contention like the 3d clients do before trying
+     *   to restore state.  
+     */
+    DRILock(pScreen, 0);
+    MGASwapContext( pScreen );
+
+}
+
+static void MGABlockHandler(int screenNum,
+			    pointer blockData,
+			    pointer pTimeout,
+			    pointer pReadmask)
+
+{
+   ScreenPtr pScreen = screenInfo.screens[screenNum];
+   DRIWrappedFuncsRec *pDRIWrap = DRIGetWrappedFuncs(pScreen);
+   MGASAREAPtr sa = (MGASAREAPtr)DRIGetSAREAPrivate( pScreen );
+
+   if (0) ErrorF("MGABlockHandler (out)\n");
+
+
+   /* Examine the cliprects for the most recently used 3d drawable.
+    * If they've changed, attempt to push the updated values into the
+    * sarea.
+    *
+    * This avoids the protocol round trip in all single-client
+    * frontbuffer rendering cases providing the cliprects fit into the
+    * sarea, and in all single-client backbuffer rendering with arbitary
+    * numbers of cliprects, for all operations except swapbuffers.
+    *
+    * Thus the number of round trips in the cases where comparison is
+    * possible is reduced to no more (and usually fewer) than the number
+    * Utah requires.  
+    */
+
+   if (sa->req_drawable != sa->exported_drawable ||
+       sa->exported_stamp != DRIGetDrawableStamp( pScreen, sa->exported_index ))
+   {
+      int i;
+      XF86DRIClipRectPtr frontboxes, backboxes;
+      XF86DRIClipRectPtr boxes = (XF86DRIClipRectPtr)sa->exported_boxes;
+      WindowPtr window = LookupIDByType( sa->req_drawable, RT_WINDOW );
+
+      if (0)
+	 ErrorF("Trying to update req_drawable: %d (exp %d), stamp %d/%d\n", 
+		sa->req_drawable, sa->exported_drawable);
+
+      sa->exported_drawable = 0;
+
+      if (!window) {
+	 if (0)
+	    ErrorF("Couldn't retreive window\n");
+	 sa->req_drawable = 0;
+	 goto finished;
+      }
+
+      if (!DRIGetDrawableInfo( pScreen, &(window->drawable), 
+			       &sa->exported_index, 
+			       &sa->exported_stamp, 
+			       &sa->exported_front_x, 
+			       &sa->exported_front_y, 
+			       &sa->exported_w, 
+			       &sa->exported_h,
+			       &sa->exported_nfront, 
+			       &frontboxes,
+			       &sa->exported_back_x, 
+			       &sa->exported_back_y,
+			       &sa->exported_nback, 
+			       &backboxes ))
+      {
+	 if (0)
+	    ErrorF("Couldn't get DRI info\n");
+	 sa->req_drawable = 0;
+	 goto finished;
+      }
+
+      /* If we can't fit both sets of cliprects into the sarea, try to
+       * fit the current draw buffer.  Otherwise return.
+       */
+      
+      sa->exported_buffers = MGA_FRONT | MGA_BACK;
+
+      if (sa->exported_nback + sa->exported_nfront >= MGA_NR_SAREA_CLIPRECTS) {
+	 if (sa->req_draw_buffer == MGA_FRONT) {
+	    sa->exported_buffers = MGA_FRONT;
+	    if (sa->exported_nfront >= MGA_NR_SAREA_CLIPRECTS)
+	       goto finished;
+	 } else {
+	    sa->exported_buffers = MGA_BACK;
+	    if (sa->exported_nback >= MGA_NR_SAREA_CLIPRECTS)
+	       goto finished;
+	 }
+      }
+
+      if (sa->exported_buffers & MGA_BACK) {
+	 for (i = 0 ; i < sa->exported_nback ; i++) 
+	    *boxes++ = backboxes[i];	 
+      }
+
+      if (sa->exported_buffers & MGA_FRONT) {
+	 for (i = 0 ; i < sa->exported_nfront ; i++) 
+	    *boxes++ = frontboxes[i];	 
+      }
+
+      sa->exported_drawable = sa->req_drawable;
+   } 
+
+ finished:
+   DRIUnlock(pScreen);
+}
+
+
+
+
+/* Just wrap validate tree for now to remove the quiescense hack.
+ * This is more of a win for the i810 than the mga, but should be useful
+ * here too.
+ */
+void MGADRIWrapFunctions(ScreenPtr pScreen, DRIInfoPtr pDRIInfo)
+{
+   pDRIInfo->wrap.BlockHandler = MGABlockHandler;   
+   pDRIInfo->wrap.WakeupHandler = MGAWakeupHandler;    
+/*     pDRIInfo->wrap.ValidateTree = NULL; */
+/*     pDRIInfo->wrap.PostValidateTree = NULL; */
+}
+
+
diff --git a/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_wrap.h b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_wrap.h
new file mode 100644
index 000000000..df65ad731
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/drivers/mga/mga_wrap.h
@@ -0,0 +1,6 @@
+#ifndef MGA_WRAP_H
+#define MGA_WRAP_H
+
+void MGADRIWrapFunctions(ScreenPtr pScreen, DRIInfoPtr pDRIInfo);
+
+#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..2c0ce97db 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/Imakefile
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/Imakefile
@@ -13,9 +13,8 @@ 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 $(MSRC)
+OBJS = xf86drm.o xf86drmHash.o xf86drmRandom.o xf86drmSL.o xf86drmI810.o xf86drmMga.o $(MOBJ)
 
 INCLUDES = -I$(XF86COMSRC) -I$(XF86OSSRC) -I. -I$(SERVERSRC)/include \
            -I$(XINCLUDESRC) -I$(EXTINCSRC) -I../.. -Ikernel
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..2ea6c7215 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 +42,10 @@ 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
+
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 8a41f880b..1db7a81fa 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
@@ -31,12 +31,14 @@
 
 # *** Setup
 
+# **** End of SMP/MODVERSIONS detection
+
 MODS=           gamma.o tdfx.o
 LIBS=           libdrm.a
 PROGS=		drmstat
 
 DRMOBJS=	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
 DRMHEADERS=	drm.h drmP.h
 
 GAMMAOBJS=      gamma_drv.o gamma_dma.o
@@ -48,6 +50,8 @@ TDFXHEADERS=    tdfx_drv.h $(DRMHEADERS)
 PROGOBJS=       drmstat.po xf86drm.po xf86drmHash.po xf86drmRandom.po sigio.po
 PROGHEADERS=    xf86drm.h $(DRMHEADERS)
 
+INC=		/usr/include
+
 CFLAGS=		-O2 $(WARNINGS)
 WARNINGS=	-Wall -Wwrite-strings -Wpointer-arith -Wcast-align \
 		-Wstrict-prototypes -Wshadow -Wnested-externs \
@@ -102,7 +106,27 @@ SMP := $(shell gcc -E -nostdinc -I$(TREE) picker.c 2>/dev/null \
 	| grep -s 'SMP = ' | cut -d' ' -f3)
 MODVERSIONS := $(shell gcc -E -I $(TREE) picker.c 2>/dev/null \
 	| grep -s 'MODVERSIONS = ' | cut -d' ' -f3)
-all::;@echo KERNEL HEADERS IN $(TREE): SMP=${SMP} MODVERSIONS=${MODVERSIONS}
+AGP := $(shell gcc -E -nostdinc -I$(TREE) picker.c 2>/dev/null \
+	| grep -s 'AGP = ' | cut -d' ' -f3)
+ifeq ($(AGP),0)
+AGP := $(shell gcc -E -nostdinc -I$(TREE) picker.c 2>/dev/null \
+	| grep -s 'AGP_MODULE = ' | cut -d' ' -f3)
+endif
+
+ifeq ($(AGP),1)
+MODCFLAGS += -DDRM_AGP
+DRMOBJS += agpsupport.o
+MODS += mga.o i810.o
+
+MGAOBJS=	mga_drv.o mga_dma.o mga_bufs.o mga_state.o mga_context.o
+MGAHEADERS=	mga_drv.h $(DRMHEADERS)
+
+I810OBJS=	i810_drv.o i810_dma.o i810_bufs.o i810_context.o
+I810HEADERS=	i810_drv.h $(DRMHEADERS)
+endif
+
+all::;@echo KERNEL HEADERS IN $(TREE): SMP=${SMP} MODVERSIONS=${MODVERSIONS} \
+	AGP=${AGP}
 all:: $(LIBS) $(MODS) $(PROGS)
 endif
 
@@ -116,6 +140,7 @@ ifeq ($(MODVERSIONS),1)
 MODCFLAGS += -DMODVERSIONS -include $(TREE)/linux/modversions.h
 endif
 
+
 # **** End of configuration
 
 libdrm.a: $(DRMOBJS)
@@ -128,6 +153,14 @@ gamma.o: $(GAMMAOBJS) $(LIBS)
 tdfx.o: $(TDFXOBJS) $(LIBS)
 	$(LD) -r $^ -o $@
 
+ifeq ($(AGP),1)
+mga.o: $(MGAOBJS) $(LIBS)
+	$(LD) -r $^ -o $@
+
+i810.o: $(I810OBJS) $(LIBS)
+	$(LD) -r $^ -o $@
+endif
+
 drmstat: $(PROGOBJS)
 	$(CC) $(PRGCFLAGS) $^ $(PRGLIBS) -o $@
 
@@ -149,8 +182,11 @@ ChangeLog:
 $(DRMOBJS): $(DRMHEADERS)
 $(GAMMAOBJS): $(GAMMAHEADERS)
 $(TDFXOBJS): $(TDFXHEADERS)
+ifeq ($(AGP),1)
+$(MGAOBJS): $(MGAHEADERS)
+$(I810OBJS): $(I810HEADERS)
+endif
 $(PROGOBJS): $(PROGHEADERS)
 
 clean:
-	rm -f *.o *.po *~ core $(PROGS)
-
+	rm -f *.o *.a *.po *~ core $(PROGS)
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 fb58154d6..c2da9ec7e 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
@@ -159,6 +159,8 @@ int drm_agp_alloc(struct inode *inode, struct file *filp, unsigned int cmd,
 			   -EFAULT);
 	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 = (u32) request.type;
@@ -254,8 +256,10 @@ int drm_agp_free(struct inode *inode, struct file *filp, unsigned int cmd,
 	if (!(entry = drm_agp_lookup_entry(dev, request.handle)))
 		return -EINVAL;
 	if (entry->bound) drm_unbind_agp(entry->memory);
-	entry->prev->next = entry->next;
-	entry->next->prev = entry->prev;
+   
+	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->memory, entry->pages);
 	drm_free(entry, sizeof(*entry), DRM_MEM_AGPLISTS);
 	return 0;
@@ -269,15 +273,12 @@ drm_agp_head_t *drm_agp_init(void)
 
 	for (fill = &drm_agp_fill[0]; fill->name; fill++) {
 		char *n  = (char *)fill->name;
-#if 0
-		*fill->f = (drm_agp_func_u)get_module_symbol(NULL, n);
-#endif
 		*fill->f = (drm_agp_func_u)get_module_symbol(NULL, n);
-		printk("%s resolves to 0x%08lx\n", n, (*fill->f).address);
+		DRM_DEBUG("%s resolves to 0x%08lx\n", n, (*fill->f).address);
 		if (!(*fill->f).address) agp_available = 0;
 	}
    
-	printk("agp_available = %d\n", agp_available);
+	DRM_DEBUG("agp_available = %d\n", agp_available);
 
 	if (agp_available) {
 		if (!(head = drm_alloc(sizeof(*head), DRM_MEM_AGPLISTS)))
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/bufs.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/bufs.c
index 85244c7dc..7902f0ccf 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/bufs.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/bufs.c
@@ -102,6 +102,11 @@ int drm_addmap(struct inode *inode, struct file *filp, unsigned int cmd,
 			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);
 		return -EINVAL;
@@ -172,7 +177,7 @@ int drm_addbufs(struct inode *inode, struct file *filp, unsigned int cmd,
 	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) ? PAGE_ALIGN(size) :size;
+	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;
 
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/ctxbitmap.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/ctxbitmap.c
index f67209d44..781984f11 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/ctxbitmap.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/ctxbitmap.c
@@ -53,7 +53,7 @@ int drm_ctxbitmap_next(drm_device_t *dev)
 	bit = find_first_zero_bit(dev->ctx_bitmap, DRM_MAX_CTXBITMAP);
 	if (bit < DRM_MAX_CTXBITMAP) {
 		set_bit(bit, dev->ctx_bitmap);
-	   	printk("drm_ctxbitmap_next bit : %d\n", bit);
+	   	DRM_DEBUG("drm_ctxbitmap_next bit : %d\n", bit);
 		return bit;
 	}
 	return -1;
@@ -64,15 +64,15 @@ int drm_ctxbitmap_init(drm_device_t *dev)
 	int i;
    	int temp;
 
-	dev->ctx_bitmap = (unsigned long *) drm_alloc(PAGE_SIZE * 4, 
+	dev->ctx_bitmap = (unsigned long *) drm_alloc(PAGE_SIZE, 
 						      DRM_MEM_CTXBITMAP);
 	if(dev->ctx_bitmap == NULL) {
 		return -ENOMEM;
 	}
-	memset((void *) dev->ctx_bitmap, 0, PAGE_SIZE * 4);
+	memset((void *) dev->ctx_bitmap, 0, PAGE_SIZE);
 	for(i = 0; i < DRM_RESERVED_CONTEXTS; i++) {
 		temp = drm_ctxbitmap_next(dev);
-	   	printk("drm_ctxbitmap_init : %d\n", temp);
+	   	DRM_DEBUG("drm_ctxbitmap_init : %d\n", temp);
 	}
 
 	return 0;
@@ -80,7 +80,7 @@ int drm_ctxbitmap_init(drm_device_t *dev)
 
 void drm_ctxbitmap_cleanup(drm_device_t *dev)
 {
-	drm_free((void *)dev->ctx_bitmap, PAGE_SIZE * 4,
+	drm_free((void *)dev->ctx_bitmap, PAGE_SIZE,
 		 DRM_MEM_CTXBITMAP);
 }
 
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/dma.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/dma.c
index 8291e52e4..37ab674f8 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/dma.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/dma.c
@@ -63,15 +63,24 @@ void drm_dma_takedown(drm_device_t *dev)
 					       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
+				 dma->bufs[i].seg_count
 				 * sizeof(*dma->bufs[0].seglist),
 				 DRM_MEM_SEGS);
-			drm_freelist_destroy(&dma->bufs[i].freelist);
+		}
+	   	if(dma->bufs[i].buf_count) {
+		   	for(j = 0; j < dma->bufs[i].buf_count; j++) {
+			   if(dma->bufs[i].buflist[j].dev_private) {
+			      drm_free(dma->bufs[i].buflist[j].dev_private,
+				       dma->bufs[i].buflist[j].dev_priv_size,
+				       DRM_MEM_BUFS);
+			   }
+			}
+		   	drm_free(dma->bufs[i].buflist,
+				 dma->bufs[i].buf_count *
+				 sizeof(*dma->bufs[0].buflist),
+				 DRM_MEM_BUFS);
+		   	drm_freelist_destroy(&dma->bufs[i].freelist);
 		}
 	}
 	
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 7b8e88265..ae4c65ca3 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
@@ -61,6 +61,19 @@ 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"
 
 typedef struct drm_version {
 	int    version_major;	  /* Major version			    */
@@ -101,7 +114,8 @@ typedef struct drm_control {
 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_SHM	  = 2,	  /* shared, cached			    */
+	_DRM_AGP          = 3	  /* AGP/GART                               */
 } drm_map_type_t;
 
 typedef enum drm_map_flags {
@@ -165,8 +179,11 @@ typedef struct drm_buf_desc {
 	int	      low_mark;	 /* Low water mark			     */
 	int	      high_mark; /* High water mark			     */
 	enum {
-		DRM_PAGE_ALIGN = 0x01  /* Align on page boundaries for DMA   */
+		_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 {
@@ -237,6 +254,38 @@ typedef struct drm_irq_busid {
 	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)	     _IOC_NR(n)
 #define DRM_IO(nr)	     _IO(DRM_IOCTL_BASE,nr)
@@ -276,4 +325,28 @@ typedef struct drm_irq_busid {
 #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_IOR( 0x32, drm_agp_mode_t)
+#define DRM_IOCTL_AGP_INFO    DRM_IOW( 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_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 )
+
+/* 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_FLUSH   DRM_IO ( 0x43)
+#define DRM_IOCTL_I810_GETAGE  DRM_IO ( 0x44)
+
 #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 312fba36f..f8e78eabd 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
@@ -48,8 +48,12 @@
 #ifdef CONFIG_MTRR
 #include <asm/mtrr.h>
 #endif
+#ifdef DRM_AGP
+#include <linux/types.h>
+#include <linux/agp_backend.h>
+#endif
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,1,0)
-#include <asm/spinlock.h>
+#include <linux/tqueue.h>
 #include <linux/poll.h>
 #endif
 #include "drm.h"
@@ -69,21 +73,27 @@
 #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_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 */
 				/* _PAGE_WT changed to _PAGE_PWT in 2.2.6 */
@@ -235,6 +245,7 @@ typedef struct drm_buf {
 	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		     */
@@ -250,6 +261,11 @@ typedef struct drm_buf {
 		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
 	cycles_t	  time_queued;	   /* Queued to kernel DMA queue     */
 	cycles_t	  time_dispatched; /* Dispatched to hardware	     */
@@ -376,6 +392,9 @@ typedef struct drm_device_dma {
 	int		  page_count;
 	unsigned long	  *pagelist;
 	unsigned long	  byte_count;
+	enum {
+	   _DRM_DMA_USE_AGP = 0x01
+	} flags;
 
 				/* DMA support */
 	drm_buf_t	  *this_buffer;	/* Buffer being sent		   */
@@ -384,6 +403,41 @@ typedef struct drm_device_dma {
 	wait_queue_head_t waiting;	/* Processes waiting on free bufs  */
 } drm_device_dma_t;
 
+#ifdef DRM_AGP
+typedef struct drm_agp_mem {
+	unsigned long      handle;
+	agp_memory         *memory;
+	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 {
+	agp_kern_info      agp_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;
+
+typedef struct {
+	void       (*free_memory)(agp_memory *);
+	agp_memory *(*allocate_memory)(size_t, u32);
+	int        (*bind_memory)(agp_memory *, off_t);
+	int        (*unbind_memory)(agp_memory *);
+	void       (*enable)(u32);
+	int        (*acquire)(void);
+	void       (*release)(void);
+	void       (*copy_info)(agp_kern_info *);
+} drm_agp_func_t;
+
+extern drm_agp_func_t drm_agp;
+#endif
+
 typedef struct drm_device {
 	const char	  *name;	/* Simple driver name		   */
 	char		  *unique;	/* Unique identifier: e.g., busid  */
@@ -462,6 +516,12 @@ typedef struct drm_device {
 	struct fasync_struct *buf_async;/* Processes waiting for SIGIO	   */
 	wait_queue_head_t buf_readers;	/* Processes waiting to read	   */
 	wait_queue_head_t buf_writers;	/* Processes waiting to ctx switch */
+	
+#ifdef DRM_AGP
+	drm_agp_head_t    *agp;
+#endif
+	unsigned long     *ctx_bitmap;
+	void		  *dev_private;
 } drm_device_t;
 
 
@@ -533,6 +593,14 @@ extern void	     drm_free_pages(unsigned long address, int order,
 extern void	     *drm_ioremap(unsigned long offset, unsigned long size);
 extern void	     drm_ioremapfree(void *pt, unsigned long size);
 
+#ifdef DRM_AGP
+extern agp_memory    *drm_alloc_agp(int pages, u32 type);
+extern int           drm_free_agp(agp_memory *handle, int pages);
+extern int           drm_bind_agp(agp_memory *handle, unsigned int start);
+extern int           drm_unbind_agp(agp_memory *handle);
+#endif
+
+
 				/* Buffer management support (bufs.c) */
 extern int	     drm_order(unsigned long size);
 extern int	     drm_addmap(struct inode *inode, struct file *filp,
@@ -642,5 +710,32 @@ 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 int            drm_agp_acquire(struct inode *inode, struct file *filp,
+				      unsigned int cmd, unsigned long arg);
+extern int            drm_agp_release(struct inode *inode, struct file *filp,
+				      unsigned int cmd, unsigned long arg);
+extern int            drm_agp_enable(struct inode *inode, struct file *filp,
+				     unsigned int cmd, unsigned long arg);
+extern int            drm_agp_info(struct inode *inode, struct file *filp,
+				   unsigned int cmd, unsigned long arg);
+extern int            drm_agp_alloc(struct inode *inode, struct file *filp,
+				    unsigned int cmd, unsigned long arg);
+extern int            drm_agp_free(struct inode *inode, struct file *filp,
+				   unsigned int cmd, unsigned long arg);
+extern int            drm_agp_unbind(struct inode *inode, struct file *filp,
+				     unsigned int cmd, unsigned long arg);
+extern int            drm_agp_bind(struct inode *inode, struct file *filp,
+				   unsigned int cmd, unsigned long arg);
+#endif
 #endif
 #endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drmstat.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drmstat.c
index 8fe6f7072..a96a38b26 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drmstat.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/drmstat.c
@@ -187,7 +187,7 @@ int main(int argc, char **argv)
 	case 'b':
 	    count   = strtoul(optarg, &pt, 0);
 	    size    = strtoul(pt+1, NULL, 0);
-	    if ((r = drmAddBufs(fd, count, size, 0)) < 0) {
+	    if ((r = drmAddBufs(fd, count, size, 0, 65536)) < 0) {
 		drmError(r, argv[0]);
 		return 1;
 	    }
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 ce970835a..a3c2bd4a0 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
@@ -271,6 +271,10 @@ static int gamma_takedown(drm_device_t *dev)
 					       - 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);
 		}
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 49455b434..c025e400b 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
@@ -32,152 +32,28 @@
 
 #define __NO_VERSION__
 #include "drmP.h"
+#include "i810_drv.h"
 #include "linux/un.h"
 
 int i810_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;
-   agp_offset = request.agp_start;
-   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;
-   
-   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;
-   
-   while(entry->buf_count < count) {
-      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 = agp_offset - dev->agp->base + offset;/* ?? */
-	 buf->bus_address = agp_offset + offset;
-	 buf->address = agp_offset + offset + dev->agp->base;
-	 buf->next = NULL;
-	 buf->waiting = 0;
-	 buf->pending = 0;
-	 init_waitqueue_head(&buf->dma_wait);
-	 buf->pid = 0;
-#if DRM_DMA_HISTOGRAM
-	 buf->time_queued = 0;
-	 buf->time_dispatched = 0;
-	 buf->time_completed = 0;
-	 buf->time_freed = 0;
-#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->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]);
-   }
-   
-   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);
-   
-   atomic_dec(&dev->buf_alloc);
-   return 0;
-}
-
-int i810_addbufs_pci(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;
-	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;
+	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;
 
@@ -186,20 +62,17 @@ int i810_addbufs_pci(struct inode *inode, struct file *filp, unsigned int cmd,
 			   sizeof(request),
 			   -EFAULT);
 
-	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 */
-
+	count = request.count;
+	order = drm_order(request.size);
+	size	= 1 << order;
+	agp_offset = request.agp_start;
 	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;
-
+	total = PAGE_SIZE << page_order;
+	byte_count = 0;
+   
+	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);
@@ -207,15 +80,15 @@ int i810_addbufs_pci(struct inode *inode, struct file *filp, unsigned int cmd,
 	}
 	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 */
+		return -ENOMEM; /* May only call once for each order */
 	}
-	
+   
 	entry->buflist = drm_alloc(count * sizeof(*entry->buflist),
 				   DRM_MEM_BUFS);
 	if (!entry->buflist) {
@@ -224,69 +97,44 @@ int i810_addbufs_pci(struct inode *inode, struct file *filp, unsigned int cmd,
 		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);
-		up(&dev->struct_sem);
-		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->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;
-			init_waitqueue_head(&buf->dma_wait);
-			buf->pid     = 0;
+	offset = 0;
+   
+	while(entry->buf_count < 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 = offset;
+		buf->bus_address = dev->agp->base + agp_offset + offset;
+		buf->address = (void *)(agp_offset + offset + dev->agp->base);
+		buf->next = NULL;
+		buf->waiting = 0;
+		buf->pending = 0;
+		init_waitqueue_head(&buf->dma_wait);
+		buf->pid = 0;
+
+		buf->dev_private = drm_alloc(sizeof(drm_i810_buf_priv_t), 
+					     DRM_MEM_BUFS);
+		buf->dev_priv_size = sizeof(drm_i810_buf_priv_t);
+
 #if DRM_DMA_HISTOGRAM
-			buf->time_queued     = 0;
-			buf->time_dispatched = 0;
-			buf->time_completed  = 0;
-			buf->time_freed	     = 0;
+		buf->time_queued = 0;
+		buf->time_dispatched = 0;
+		buf->time_completed = 0;
+		buf->time_freed = 0;
 #endif
-			DRM_DEBUG("buffer %d @ %p\n",
-				  entry->buf_count, buf->address);
-		}
+		offset = offset + alignment;
+		entry->buf_count++;
 		byte_count += PAGE_SIZE << page_order;
+      
+		DRM_DEBUG("buffer %d @ %p\n",
+			  entry->buf_count, buf->address);
 	}
-
+   
 	dma->buflist = drm_realloc(dma->buflist,
 				   dma->buf_count * sizeof(*dma->buflist),
 				   (dma->buf_count + entry->buf_count)
@@ -294,45 +142,43 @@ int i810_addbufs_pci(struct inode *inode, struct file *filp, unsigned int cmd,
 				   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);
-	
+   
+	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);
-	
+   
 	atomic_dec(&dev->buf_alloc);
+	dma->flags = _DRM_DMA_USE_AGP;
 	return 0;
 }
 
 int i810_addbufs(struct inode *inode, struct file *filp, unsigned int cmd,
 		unsigned long arg)
 {
-   drm_buf_desc_t	 request;
+	drm_buf_desc_t	 request;
 
-   copy_from_user_ret(&request,
-		      (drm_buf_desc_t *)arg,
-		      sizeof(request),
-		      -EFAULT);
+	copy_from_user_ret(&request,
+			   (drm_buf_desc_t *)arg,
+			   sizeof(request),
+			   -EFAULT);
 
-   if(request.flags & _DRM_AGP_BUFFER)
-     return i810_addbufs_agp(inode, filp, cmd, arg);
-   else
-     return i810_addbufs_pci(inode, filp, cmd, arg);
+	if(request.flags & _DRM_AGP_BUFFER)
+		return i810_addbufs_agp(inode, filp, cmd, arg);
+	else
+		return -EINVAL;
 }
 
 int i810_infobufs(struct inode *inode, struct file *filp, unsigned int cmd,
@@ -487,7 +333,7 @@ int i810_freebufs(struct inode *inode, struct file *filp, unsigned int cmd,
 }
 
 int i810_mapbufs(struct inode *inode, struct file *filp, unsigned int cmd,
-		unsigned long arg)
+		 unsigned long arg)
 {
 	drm_file_t	 *priv	 = filp->private_data;
 	drm_device_t	 *dev	 = priv->dev;
@@ -506,6 +352,7 @@ int i810_mapbufs(struct inode *inode, struct file *filp, unsigned int cmd,
 	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 */
@@ -515,34 +362,46 @@ int i810_mapbufs(struct inode *inode, struct file *filp, unsigned int cmd,
 			   (drm_buf_map_t *)arg,
 			   sizeof(request),
 			   -EFAULT);
-
-	if (request.count >= dma->buf_count) {
-	   if(dma->flags & _DRM_DMA_USE_AGP) {
-	      /* This is an ugly vicious hack */
-	      drm_map_t *map = NULL;
-	      for(i = 0; i < dev->map_count; i++) {
-		 map = dev->maplist[i];
-		 if(map->type == _DRM_AGP) break;
-	      }
-	      if (i >= dev->map_count || !map) {
-		 retcode = -EINVAL;
-		 goto done;
-	      }
+   	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;
 	      
-	      virtual = do_mmap(filp, 0, map->size, PROT_READ|PROT_WRITE,
-				MAP_SHARED, (unsigned long)map->handle);
-	   }
-	   else {
-	      virtual = do_mmap(filp, 0, dma->byte_count,
-				PROT_READ|PROT_WRITE, MAP_SHARED, 0);
-	   }
+			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 */
+			/* 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,
@@ -571,14 +430,15 @@ int i810_mapbufs(struct inode *inode, struct file *filp, unsigned int cmd,
 			}
 		}
 	}
-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_context.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_context.c
new file mode 100644
index 000000000..5503edf24
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_context.c
@@ -0,0 +1,205 @@
+/* i810_context.c -- IOCTLs for i810 contexts -*- linux-c -*-
+ * Created: Mon Dec 13 09:51:35 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.
+ * 
+ * Author: Rickard E. (Rik) Faith <faith@precisioninsight.com>
+ *
+ * $XFree86$
+ *
+ */
+
+#include <linux/sched.h>
+
+#define __NO_VERSION__
+#include "drmP.h"
+#include "i810_drv.h"
+
+static int i810_alloc_queue(drm_device_t *dev)
+{
+   	int temp = drm_ctxbitmap_next(dev);
+   	DRM_DEBUG("i810_alloc_queue: %d\n", temp);
+	return temp;
+}
+
+int i810_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) {
+                i810_context_switch_complete(dev, new);
+        } else {
+                sprintf(buf, "C %d %d\n", old, new);
+                drm_write_string(dev, buf);
+        }
+        
+        return 0;
+}
+
+int i810_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 i810_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 i810_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 = i810_alloc_queue(dev)) == DRM_KERNEL_CONTEXT) {
+				/* Skip kernel's context and get a new one. */
+		ctx.handle = i810_alloc_queue(dev);
+	}
+        if (ctx.handle == -1) {
+		DRM_DEBUG("Not enough free contexts.\n");
+				/* Should this return -EBUSY instead? */
+		return -ENOMEM;
+	}
+	DRM_DEBUG("%d\n", ctx.handle);
+	copy_to_user_ret((drm_ctx_t *)arg, &ctx, sizeof(ctx), -EFAULT);
+	return 0;
+}
+
+int i810_modctx(struct inode *inode, struct file *filp, unsigned int cmd,
+	unsigned long arg)
+{
+   	/* This does nothing for the i810 */
+	return 0;
+}
+
+int i810_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 i810_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 i810_context_switch(dev, dev->last_context, ctx.handle);
+}
+
+int i810_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);
+	i810_context_switch_complete(dev, ctx.handle);
+
+	return 0;
+}
+
+int i810_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);
+   	if(ctx.handle != DRM_KERNEL_CONTEXT) {
+	   	drm_ctxbitmap_free(dev, ctx.handle);
+	}
+	
+	return 0;
+}
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 09959b657..30fda5b8b 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
@@ -25,8 +25,9 @@
  *
  * Authors: Rickard E. (Rik) Faith <faith@precisioninsight.com>
  *	    Jeff Hartmann <jhartmann@precisioninsight.com>
+ *          Keith Whitwell <keithw@precisioninsight.com>
  *
- * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_dma.c,v 1.1 2000/02/11 17:26:04 dawes Exp $
+ * $XFree86$
  *
  */
 
@@ -36,6 +37,9 @@
 
 #include <linux/interrupt.h>	/* For task queue support */
 
+#define I810_BUF_FREE		1
+#define I810_BUF_USED		0
+
 #define I810_REG(reg)		2
 #define I810_BASE(reg)		((unsigned long) \
 				dev->maplist[I810_REG(reg)]->handle)
@@ -43,544 +47,457 @@
 #define I810_DEREF(reg)		*(__volatile__ int *)I810_ADDR(reg)
 #define I810_READ(reg)		I810_DEREF(reg)
 #define I810_WRITE(reg,val) 	do { I810_DEREF(reg) = val; } while (0)
-
-void i810_dma_init(drm_device_t *dev)
+#define I810_DEREF16(reg)	*(__volatile__ u16 *)I810_ADDR(reg)
+#define I810_READ16(reg)	I810_DEREF16(reg)
+#define I810_WRITE16(reg,val)	do { I810_DEREF16(reg) = val; } while (0)
+
+#define RING_LOCALS	unsigned int outring, ringmask; volatile char *virt;
+
+#define BEGIN_LP_RING(n) do {				\
+	if (I810_VERBOSE)				\
+		DRM_DEBUG("BEGIN_LP_RING(%d) in %s\n",	\
+			  n, __FUNCTION__);		\
+	if (dev_priv->ring.space < n*4) 		\
+		i810_wait_ring(dev, n*4);		\
+	dev_priv->ring.space -= n*4;			\
+	outring = dev_priv->ring.tail;			\
+	ringmask = dev_priv->ring.tail_mask;		\
+	virt = dev_priv->ring.virtual_start;		\
+} while (0)
+
+#define ADVANCE_LP_RING() do {					\
+	if (I810_VERBOSE) DRM_DEBUG("ADVANCE_LP_RING\n");	\
+	dev_priv->ring.tail = outring;				\
+	I810_WRITE(LP_RING + RING_TAIL, outring);		\
+} while(0)
+
+#define OUT_RING(n) do {						\
+	if (I810_VERBOSE) DRM_DEBUG("   OUT_RING %x\n", (int)(n));	\
+	*(volatile unsigned int *)(virt + outring) = n;			\
+	outring += 4;							\
+	outring &= ringmask;						\
+} while (0);
+
+static inline void i810_print_status_page(drm_device_t *dev)
 {
-        printk(KERN_INFO "i810_dma_init\n");
+   	drm_device_dma_t *dma = dev->dma;
+      	drm_i810_private_t *dev_priv = dev->dev_private;
+	u32 *temp = (u32 *)dev_priv->hw_status_page;
+   	int i;
+
+   	DRM_DEBUG(  "hw_status: Interrupt Status : %x\n", temp[0]);
+   	DRM_DEBUG(  "hw_status: LpRing Head ptr : %x\n", temp[1]);
+   	DRM_DEBUG(  "hw_status: IRing Head ptr : %x\n", temp[2]);
+      	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]);
+	}
 }
 
-void i810_dma_cleanup(drm_device_t *dev)
+static drm_buf_t *i810_freelist_get(drm_device_t *dev)
 {
-   printk(KERN_INFO "i810_dma_cleanup\n");
+   	drm_device_dma_t *dma = dev->dma;
+	int		 i;
+   	int 		 used;
+   
+	/* Linear search might not be the best solution */
+
+   	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;
+		/* In use is already a pointer */
+	   	used = cmpxchg(buf_priv->in_use, I810_BUF_FREE, 
+			       I810_BUF_USED);
+	   	if(used == I810_BUF_FREE) {
+			return buf;
+		}
+	}
+   	return NULL;
 }
 
-static inline void i810_dma_dispatch(drm_device_t *dev, unsigned long address,
-				    unsigned long length)
+/* This should only be called if the buffer is not sent to the hardware
+ * yet, the hardware updates in use for us once its on the ring buffer.
+ */
+
+static int i810_freelist_put(drm_device_t *dev, drm_buf_t *buf)
 {
-   printk(KERN_INFO "i810_dma_dispatch\n");
+   	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+   	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) {
+	   	DRM_ERROR("Freeing buffer thats not in use : %d\n", buf->idx);
+	   	return -EINVAL;
+	}
+   
+   	return 0;
 }
 
-static inline void i810_dma_quiescent(drm_device_t *dev)
+static int i810_dma_get_buffers(drm_device_t *dev, drm_dma_t *d)
 {
+	int		  i;
+	drm_buf_t	  *buf;
+
+	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;
+	}
+	return 0;
 }
 
-static inline void i810_dma_ready(drm_device_t *dev)
+static unsigned long i810_alloc_page(drm_device_t *dev)
 {
-   i810_dma_quiescent(dev);
-   printk(KERN_INFO "i810_dma_ready\n");
+	unsigned long address;
+   
+	address = __get_free_page(GFP_KERNEL);
+	if(address == 0UL) 
+		return 0;
+	
+	atomic_inc(&mem_map[MAP_NR((void *) address)].count);
+	set_bit(PG_locked, &mem_map[MAP_NR((void *) address)].flags);
+   
+	return address;
 }
 
-static inline int i810_dma_is_ready(drm_device_t *dev)
+static void i810_free_page(drm_device_t *dev, unsigned long page)
 {
-
-   i810_dma_quiescent(dev);
-
-   printk(KERN_INFO "i810_dma_is_ready\n");
-   return 1;
+	if(page == 0UL) 
+		return;
+	
+	atomic_dec(&mem_map[MAP_NR((void *) page)].count);
+	clear_bit(PG_locked, &mem_map[MAP_NR((void *) page)].flags);
+	wake_up(&mem_map[MAP_NR((void *) page)].wait);
+	free_page(page);
+	return;
 }
 
-
-static void i810_dma_service(int irq, void *device, struct pt_regs *regs)
+static int i810_dma_cleanup(drm_device_t *dev)
 {
-	drm_device_t	 *dev = (drm_device_t *)device;
-	drm_device_dma_t *dma = dev->dma;
-	
-	atomic_inc(&dev->total_irq);
-	if (i810_dma_is_ready(dev)) {
-				/* Free previous buffer */
-		if (test_and_set_bit(0, &dev->dma_flag)) {
-			atomic_inc(&dma->total_missed_free);
-			return;
+	if(dev->dev_private) {
+	   	drm_i810_private_t *dev_priv = 
+	     		(drm_i810_private_t *) dev->dev_private;
+	   
+	   	if(dev_priv->ring.virtual_start) {
+		   	drm_ioremapfree((void *) dev_priv->ring.virtual_start,
+					dev_priv->ring.Size);
 		}
-		if (dma->this_buffer) {
-			drm_free_buffer(dev, dma->this_buffer);
-			dma->this_buffer = NULL;
+	   	if(dev_priv->hw_status_page != 0UL) {
+		   	i810_free_page(dev, dev_priv->hw_status_page);
+		   	/* Need to rewrite hardware status page */
+		   	I810_WRITE(0x02080, 0x1ffff000);
 		}
-		clear_bit(0, &dev->dma_flag);
-
-				/* Dispatch new buffer */
-		queue_task(&dev->tq, &tq_immediate);
-		mark_bh(IMMEDIATE_BH);
+	   	drm_free(dev->dev_private, sizeof(drm_i810_private_t), 
+			 DRM_MEM_DRIVER);
+	   	dev->dev_private = NULL;
 	}
+   	return 0;
 }
 
-/* Only called by i810_dma_schedule. */
-static int i810_do_dma(drm_device_t *dev, int locked)
+static int i810_wait_ring(drm_device_t *dev, int n)
 {
-	unsigned long	 address;
-	unsigned long	 length;
-	drm_buf_t	 *buf;
-	int		 retcode = 0;
-	drm_device_dma_t *dma = dev->dma;
-#if DRM_DMA_HISTOGRAM
-	cycles_t	 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
-	dma_start = get_cycles();
-#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->bus_address;
-	length	= buf->used;
+   	drm_i810_private_t *dev_priv = dev->dev_private;
+   	drm_i810_ring_buffer_t *ring = &(dev_priv->ring);
+   	int iters = 0;
+   	unsigned long end;
+
+	end = jiffies + (HZ*3);
+   	while (ring->space < n) {
+	   	int i;
 	
-
-	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 (!i810_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;
+	   	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;
+	   
+	   	iters++;
+		if((signed)(end - jiffies) <= 0) {
+		   	DRM_ERROR("space: %d wanted %d\n", ring->space, n);
+		   	DRM_ERROR("lockup\n");
+		   	goto out_wait_ring;
 		}
-	}
-
-	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
-	buf->time_dispatched = get_cycles();
-#endif
 
-	i810_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");
-		}
+	   	for (i = 0 ; i < 2000 ; i++) ;
 	}
-cleanup:
-
-	clear_bit(0, &dev->dma_flag);
 
-#if DRM_DMA_HISTOGRAM
-	dma_stop = get_cycles();
-	atomic_inc(&dev->histo.dma[drm_histogram_slot(dma_stop - dma_start)]);
-#endif
-
-	return retcode;
-}
-
-static void i810_dma_schedule_timer_wrapper(unsigned long dev)
-{
-	i810_dma_schedule((drm_device_t *)dev, 0);
+out_wait_ring:   
+   	return iters;
 }
 
-static void i810_dma_schedule_tq_wrapper(void *dev)
+static void i810_kernel_lost_context(drm_device_t *dev)
 {
-	i810_dma_schedule(dev, 0);
+      	drm_i810_private_t *dev_priv = dev->dev_private;
+   	drm_i810_ring_buffer_t *ring = &(dev_priv->ring);
+      
+   	ring->head = I810_READ(LP_RING + RING_HEAD) & HEAD_ADDR;
+     	ring->tail = I810_READ(LP_RING + RING_TAIL);
+     	ring->space = ring->head - (ring->tail+8);
+     	if (ring->space < 0) ring->space += ring->Size;
 }
 
-int i810_dma_schedule(drm_device_t *dev, int locked)
+static int i810_freelist_init(drm_device_t *dev)
 {
-	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
-	cycles_t	 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
-	schedule_start = get_cycles();
-#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 = i810_do_dma(dev, locked))) {
-			atomic_inc(&dma->total_hit);
-			++processed;
-		}
-	} else {
-		do {
-			next = drm_select_queue(dev,
-					     i810_dma_schedule_timer_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 = i810_do_dma(dev, locked))) {
-				++processed;
-			}
-		}
+      	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;
+   
+   	if(dma->buf_count > 1019) {
+	   	/* Not enough space in the status page for the freelist */
+	   	return -EINVAL;
 	}
 
-	if (--expire) {
-		if (missed != atomic_read(&dma->total_missed_sched)) {
-			atomic_inc(&dma->total_lost);
-			if (i810_dma_is_ready(dev)) goto again;
-		}
-		if (processed && i810_dma_is_ready(dev)) {
-			atomic_inc(&dma->total_lost);
-			processed = 0;
-			goto again;
-		}
+   	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;
+	   
+	   	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->my_use_idx = my_idx;
+	   	my_idx += 4;
 	}
-	
-	clear_bit(0, &dev->interrupt_flag);
-	
-#if DRM_DMA_HISTOGRAM
-	atomic_inc(&dev->histo.schedule[drm_histogram_slot(get_cycles()
-							   - schedule_start)]);
-#endif
-	return retcode;
+	return 0;
 }
 
-static int i810_dma_priority(drm_device_t *dev, drm_dma_t *d)
+static int i810_dma_initialize(drm_device_t *dev, 
+			       drm_i810_private_t *dev_priv,
+			       drm_i810_init_t *init)
 {
-	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;
-	DECLARE_WAITQUEUE(entry, current);
-
-				/* Turn off interrupt handling */
-	while (test_and_set_bit(0, &dev->interrupt_flag)) {
-		schedule();
-		if (signal_pending(current)) return -EINTR;
-	}
-	if (!(d->flags & _DRM_DMA_WHILE_LOCKED)) {
-		while (!drm_lock_take(&dev->lock.hw_lock->lock,
-				      DRM_KERNEL_CONTEXT)) {
-			schedule();
-			if (signal_pending(current)) {
-				clear_bit(0, &dev->interrupt_flag);
-				return -EINTR;
-			}
-		}
-		++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 != current->pid) {
-			DRM_ERROR("Process %d using buffer owned by %d\n",
-				  current->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",
-				  current->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)) {
-			add_wait_queue(&dev->context_wait, &entry);
-			current->state = TASK_INTERRUPTIBLE;
-				/* 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! */
-			schedule();
-			current->state = TASK_RUNNING;
-			remove_wait_queue(&dev->context_wait, &entry);
-			if (signal_pending(current)) {
-				retcode = -EINTR;
-				goto cleanup;
-			}
-			if (dev->last_context != buf->context) {
-				DRM_ERROR("Context mismatch: %d %d\n",
-					  dev->last_context,
-					  buf->context);
-			}
-		}
-
-#if DRM_DMA_HISTOGRAM
-		buf->time_queued     = get_cycles();
-		buf->time_dispatched = buf->time_queued;
-#endif
-		i810_dma_dispatch(dev, address, length);
-	        if (drm_lock_free(dev, &dev->lock.hw_lock->lock,
-				  DRM_KERNEL_CONTEXT)) {
-		   DRM_ERROR("\n");
-		}
-
-		atomic_add(length, &dma->total_bytes);
-		atomic_inc(&dma->total_dmas);
-		
-		if (last_buf) {
-			drm_free_buffer(dev, last_buf);
-		}
-		last_buf = buf;
-	}
+	drm_map_t *sarea_map;
 
+   	dev->dev_private = (void *) dev_priv;
+   	memset(dev_priv, 0, sizeof(drm_i810_private_t));
 
-cleanup:
-	if (last_buf) {
-		i810_dma_ready(dev);
-		drm_free_buffer(dev, last_buf);
+   	if (init->ring_map_idx >= dev->map_count ||
+	    init->buffer_map_idx >= dev->map_count) {
+	   	i810_dma_cleanup(dev);
+	   	DRM_ERROR("ring_map or buffer_map are invalid\n");
+	   	return -EINVAL;
 	}
-	
-	if (must_free && !dev->context_flag) {
-		if (drm_lock_free(dev, &dev->lock.hw_lock->lock,
-				  DRM_KERNEL_CONTEXT)) {
-			DRM_ERROR("\n");
-		}
+   
+   	dev_priv->ring_map_idx = init->ring_map_idx;
+   	dev_priv->buffer_map_idx = init->buffer_map_idx;
+	sarea_map = dev->maplist[0];
+	dev_priv->sarea_priv = (drm_i810_sarea_t *) 
+		((u8 *)sarea_map->handle + 
+		 init->sarea_priv_offset);
+
+   	atomic_set(&dev_priv->flush_done, 0);
+	init_waitqueue_head(&dev_priv->flush_queue);
+   	
+   	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) {
+	   	i810_dma_cleanup(dev);
+	   	DRM_ERROR("can not ioremap virtual address for"
+			  " ring buffer\n");
+	   	return -ENOMEM;
 	}
-	clear_bit(0, &dev->interrupt_flag);
-	return retcode;
+   
+   	/* Program Hardware Status Page */
+   	dev_priv->hw_status_page = i810_alloc_page(dev);
+   	memset((void *) dev_priv->hw_status_page, 0, PAGE_SIZE);
+   	if(dev_priv->hw_status_page == 0UL) {
+		i810_dma_cleanup(dev);
+		DRM_ERROR("Can not allocate hardware status page\n");
+		return -ENOMEM;
+	}
+   	DRM_DEBUG("hw status page @ %lx\n", dev_priv->hw_status_page);
+   
+   	I810_WRITE(0x02080, virt_to_bus((void *)dev_priv->hw_status_page));
+   	DRM_DEBUG("Enabled hardware status page\n");
+   
+   	/* Now we need to init our freelist */
+   	if(i810_freelist_init(dev) != 0) {
+	   	i810_dma_cleanup(dev);
+	   	DRM_ERROR("Not enough space in the status page for"
+			  " the freelist\n");
+	   	return -ENOMEM;
+	}
+   	return 0;
 }
 
-static int i810_dma_send_buffers(drm_device_t *dev, drm_dma_t *d)
+int i810_dma_init(struct inode *inode, struct file *filp,
+		  unsigned int cmd, unsigned long arg)
 {
-	DECLARE_WAITQUEUE(entry, current);
-	drm_buf_t	  *last_buf = NULL;
-	int		  retcode   = 0;
-	drm_device_dma_t  *dma	    = dev->dma;
-
-	if (d->flags & _DRM_DMA_BLOCK) {
-		last_buf = dma->buflist[d->send_indices[d->send_count-1]];
-		add_wait_queue(&last_buf->dma_wait, &entry);
-	}
+   	drm_file_t *priv = filp->private_data;
+   	drm_device_t *dev = priv->dev;
+   	drm_i810_private_t *dev_priv;
+   	drm_i810_init_t init;
+   	int retcode = 0;
 	
-	if ((retcode = drm_dma_enqueue(dev, d))) {
-		if (d->flags & _DRM_DMA_BLOCK)
-			remove_wait_queue(&last_buf->dma_wait, &entry);
-		return retcode;
-	}
-	
-	i810_dma_schedule(dev, 0);
+   	copy_from_user_ret(&init, (drm_i810_init_t *)arg, 
+			   sizeof(init), -EFAULT);
 	
-	if (d->flags & _DRM_DMA_BLOCK) {
-		DRM_DEBUG("%d waiting\n", current->pid);
-		current->state = TASK_INTERRUPTIBLE;
-		for (;;) {
-			if (!last_buf->waiting
-			    && !last_buf->pending)
-				break; /* finished */
-			schedule();
-			if (signal_pending(current)) {
-				retcode = -EINTR; /* Can't restart */
-				break;
-			}
-		}
-		current->state = TASK_RUNNING;
-		DRM_DEBUG("%d running\n", current->pid);
-		remove_wait_queue(&last_buf->dma_wait, &entry);
-		if (!retcode
-		    || (last_buf->list==DRM_LIST_PEND && !last_buf->pending)) {
-			if (!waitqueue_active(&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,
-				  current->pid);
-		}
+   	switch(init.func) {
+	 	case I810_INIT_DMA:
+	   		dev_priv = drm_alloc(sizeof(drm_i810_private_t), 
+					     DRM_MEM_DRIVER);
+	   		if(dev_priv == NULL) return -ENOMEM;
+	   		retcode = i810_dma_initialize(dev, dev_priv, &init);
+	   	break;
+	 	case I810_CLEANUP_DMA:
+	   		retcode = i810_dma_cleanup(dev);
+	   	break;
+	 	default:
+	   		retcode = -EINVAL;
+	   	break;
 	}
-	return retcode;
+   
+   	return retcode;
 }
 
-int i810_dma(struct inode *inode, struct file *filp, unsigned int cmd,
-	      unsigned long arg)
+static void i810_dma_dispatch_general(drm_device_t *dev, drm_buf_t *buf,
+				      int used )
 {
-	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;
-
-        printk("i810_dma start\n");
-   	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_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;
+
+   	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);
+
+   	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 (d.context == DRM_KERNEL_CONTEXT || d.context >= dev->queue_slots) {
-		DRM_ERROR("Process %d using context %d\n",
-			  current->pid, d.context);
-		return -EINVAL;
-	}
+static void i810_dma_dispatch_vertex(drm_device_t *dev, 
+				     drm_buf_t *buf,
+				     int discard,
+				     int used)
+{
+   	drm_i810_private_t *dev_priv = dev->dev_private;
+	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+   	drm_i810_sarea_t *sarea_priv = dev_priv->sarea_priv;
+   	drm_clip_rect_t *box = sarea_priv->boxes;
+   	int nbox = sarea_priv->nbox;
+	unsigned long address = (unsigned long)buf->bus_address;
+	unsigned long start = address - dev->agp->base;     
+	int i = 0;
+   	RING_LOCALS;
 
-	if (d.send_count < 0 || d.send_count > dma->buf_count) {
-		DRM_ERROR("Process %d trying to send %d buffers (of %d max)\n",
-			  current->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",
-			  current->pid, d.request_count, dma->buf_count);
-		return -EINVAL;
+   
+   	if (nbox > I810_NR_SAREA_CLIPRECTS) 
+		nbox = I810_NR_SAREA_CLIPRECTS;
+   
+   	DRM_DEBUG("dispatch vertex addr 0x%lx, used 0x%x nbox %d\n", 
+		  address, used, nbox);
+
+   	dev_priv->counter++;
+   	DRM_DEBUG(  "dispatch counter : %ld\n", dev_priv->counter);
+   	DRM_DEBUG(  "i810_dma_dispatch\n");
+   	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 (used) {
+		do {
+			if (i < nbox) {
+				BEGIN_LP_RING(4);
+				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].x2-1) | ((box[i].y2-1)<<16) );
+				ADVANCE_LP_RING();
+			}
+			
+			BEGIN_LP_RING(4);
+			OUT_RING( CMD_OP_BATCH_BUFFER );
+			OUT_RING( start | BB1_PROTECTED );
+			OUT_RING( start + used - 4 );
+			OUT_RING( 0 );
+			ADVANCE_LP_RING();
+			
+		} while (++i < nbox);
 	}
 
-	if (d.send_count) {
-#if 0
-		if (d.flags & _DRM_DMA_PRIORITY)
-			retcode = i810_dma_priority(dev, &d);
-		else 
-			retcode = i810_dma_send_buffers(dev, &d);
-#endif
-	   printk("i810_dma priority\n");
+	BEGIN_LP_RING(10);
+	OUT_RING( CMD_STORE_DWORD_IDX );
+	OUT_RING( 20 );
+	OUT_RING( dev_priv->counter );
+	OUT_RING( 0 );
 
-	   retcode = i810_dma_priority(dev, &d);
+	if (discard) {
+		OUT_RING( CMD_STORE_DWORD_IDX );
+		OUT_RING( buf_priv->my_use_idx );
+		OUT_RING( I810_BUF_FREE );
+		OUT_RING( 0 );
 	}
 
-	d.granted_count = 0;
+      	OUT_RING( CMD_REPORT_HEAD );
+	OUT_RING( 0 );
+   	ADVANCE_LP_RING();
+}
 
-	if (!retcode && d.request_count) {
-		retcode = drm_dma_get_buffers(dev, &d);
-	}
 
-	DRM_DEBUG("%d returning, granted = %d\n",
-		  current->pid, d.granted_count);
-	copy_to_user_ret((drm_dma_t *)arg, &d, sizeof(d), -EFAULT);
+/* Interrupts are only for flushing */
+static void i810_dma_service(int irq, void *device, struct pt_regs *regs)
+{
+	drm_device_t	 *dev = (drm_device_t *)device;
+   	u16 temp;
+   
+	atomic_inc(&dev->total_irq);
+      	temp = I810_READ16(I810REG_INT_IDENTITY_R);
+   	temp = temp & ~(0x6000);
+   	if(temp != 0) I810_WRITE16(I810REG_INT_IDENTITY_R, 
+				   temp); /* Clear all interrupts */
+   
+   	queue_task(&dev->tq, &tq_immediate);
+   	mark_bh(IMMEDIATE_BH);
+}
 
-   	printk("i810_dma end (granted)\n");
-	return retcode;
+static void i810_dma_task_queue(void *device)
+{
+	drm_device_t *dev = (drm_device_t *) device;
+      	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
+
+   	atomic_set(&dev_priv->flush_done, 1);
+   	wake_up_interruptible(&dev_priv->flush_queue);
 }
 
 int i810_irq_install(drm_device_t *dev, int irq)
 {
 	int retcode;
-
+	u16 temp;
+   
 	if (!irq)     return -EINVAL;
 	
 	down(&dev->struct_sem);
@@ -591,6 +508,7 @@ int i810_irq_install(drm_device_t *dev, int irq)
 	dev->irq = irq;
 	up(&dev->struct_sem);
 	
+   	DRM_DEBUG(  "Interrupt Install : %d\n", irq);
 	DRM_DEBUG("%d\n", irq);
 
 	dev->context_flag     = 0;
@@ -603,12 +521,21 @@ int i810_irq_install(drm_device_t *dev, int irq)
 
 	dev->tq.next	      = NULL;
 	dev->tq.sync	      = 0;
-	dev->tq.routine	      = i810_dma_schedule_tq_wrapper;
+	dev->tq.routine	      = i810_dma_task_queue;
 	dev->tq.data	      = dev;
 
-
 				/* Before installing handler */
-	/* TODO */	
+   	temp = I810_READ16(I810REG_HWSTAM);
+   	temp = temp & 0x6000;
+   	I810_WRITE16(I810REG_HWSTAM, temp);
+   	
+      	temp = I810_READ16(I810REG_INT_MASK_R);
+   	temp = temp & 0x6000;
+   	I810_WRITE16(I810REG_INT_MASK_R, temp); /* Unmask interrupts */
+   	temp = I810_READ16(I810REG_INT_ENABLE_R);
+   	temp = temp & 0x6000;
+      	I810_WRITE16(I810REG_INT_ENABLE_R, temp); /* Disable all interrupts */
+
 				/* Install handler */
 	if ((retcode = request_irq(dev->irq,
 				   i810_dma_service,
@@ -620,15 +547,18 @@ int i810_irq_install(drm_device_t *dev, int irq)
 		up(&dev->struct_sem);
 		return retcode;
 	}
-
-				/* After installing handler */
-	/* TODO */
+   	temp = I810_READ16(I810REG_INT_ENABLE_R);
+   	temp = temp & 0x6000;
+   	temp = temp | 0x0003;
+   	I810_WRITE16(I810REG_INT_ENABLE_R, 
+		     temp); /* Enable bp & user interrupts */
 	return 0;
 }
 
 int i810_irq_uninstall(drm_device_t *dev)
 {
 	int irq;
+   	u16 temp;
 
 	down(&dev->struct_sem);
 	irq	 = dev->irq;
@@ -636,16 +566,25 @@ int i810_irq_uninstall(drm_device_t *dev)
 	up(&dev->struct_sem);
 	
 	if (!irq) return -EINVAL;
-	
+
+   	DRM_DEBUG(  "Interrupt UnInstall: %d\n", irq);	
 	DRM_DEBUG("%d\n", irq);
-	
-	/* TODO : Disable interrupts */
+   
+   	temp = I810_READ16(I810REG_INT_IDENTITY_R);
+   	temp = temp & ~(0x6000);
+   	if(temp != 0) I810_WRITE16(I810REG_INT_IDENTITY_R, 
+				   temp); /* Clear all interrupts */
+   
+   	temp = I810_READ16(I810REG_INT_ENABLE_R);
+   	temp = temp & 0x6000;
+   	I810_WRITE16(I810REG_INT_ENABLE_R, 
+		     temp);                     /* Disable all interrupts */
+
    	free_irq(irq, dev);
 
 	return 0;
 }
 
-
 int i810_control(struct inode *inode, struct file *filp, unsigned int cmd,
 		  unsigned long arg)
 {
@@ -654,8 +593,7 @@ int i810_control(struct inode *inode, struct file *filp, unsigned int cmd,
 	drm_control_t	ctl;
 	int		retcode;
    
-   	printk(KERN_INFO "i810_control\n");
-   	i810_dma_init(dev);
+   	DRM_DEBUG(  "i810_control\n");
 
 	copy_from_user_ret(&ctl, (drm_control_t *)arg, sizeof(ctl), -EFAULT);
 	
@@ -674,20 +612,134 @@ int i810_control(struct inode *inode, struct file *filp, unsigned int cmd,
 	return 0;
 }
 
+static inline void i810_dma_emit_flush(drm_device_t *dev)
+{
+   	drm_i810_private_t *dev_priv = dev->dev_private;
+   	RING_LOCALS;
+
+   	i810_kernel_lost_context(dev);
+   	BEGIN_LP_RING(2);
+      	OUT_RING( CMD_REPORT_HEAD );
+   	OUT_RING( GFX_OP_USER_INTERRUPT );
+      	ADVANCE_LP_RING();
+}
+
+static inline void i810_dma_quiescent_emit(drm_device_t *dev)
+{
+      	drm_i810_private_t *dev_priv = dev->dev_private;
+   	RING_LOCALS;
+
+  	i810_kernel_lost_context(dev);
+   	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 );
+   	ADVANCE_LP_RING();
+}
+
+static void i810_dma_quiescent(drm_device_t *dev)
+{
+      	DECLARE_WAITQUEUE(entry, current);
+  	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
+	unsigned long end;      
+
+   	if(dev_priv == NULL) {
+	   	return;
+	}
+      	atomic_set(&dev_priv->flush_done, 0);
+   	current->state = TASK_INTERRUPTIBLE;
+   	add_wait_queue(&dev_priv->flush_queue, &entry);
+   	end = jiffies + (HZ*3);
+   
+   	for (;;) {
+	      	i810_dma_quiescent_emit(dev);
+	   	if (atomic_read(&dev_priv->flush_done) == 1) break;
+		if((signed)(end - jiffies) <= 0) {
+		   	DRM_ERROR("lockup\n");
+		   	break;
+		}	   
+	      	schedule_timeout(HZ*3);
+	      	if (signal_pending(current)) {
+		   	break;
+		}
+	}
+   
+   	current->state = TASK_RUNNING;
+   	remove_wait_queue(&dev_priv->flush_queue, &entry);
+   
+   	return;
+}
+
+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;
+	unsigned long end;
+   	int ret = 0;      
+
+   	if(dev_priv == NULL) {
+	   	return 0;
+	}
+      	atomic_set(&dev_priv->flush_done, 0);
+   	current->state = TASK_INTERRUPTIBLE;
+   	add_wait_queue(&dev_priv->flush_queue, &entry);
+   	end = jiffies + (HZ*3);
+   	for (;;) {
+	      	i810_dma_emit_flush(dev);
+	   	if (atomic_read(&dev_priv->flush_done) == 1) break;
+		if((signed)(end - jiffies) <= 0) {
+		   	DRM_ERROR("lockup\n");
+		   	break;
+		}	   
+	      	schedule_timeout(HZ*3);
+	      	if (signal_pending(current)) {
+		   	ret = -EINTR; /* Can't restart */
+		   	break;
+		}
+	}
+   
+   	current->state = TASK_RUNNING;
+   	remove_wait_queue(&dev_priv->flush_queue, &entry);
+   
+   	return ret;
+}
+
+/* Must be called with the lock held */
+void i810_reclaim_buffers(drm_device_t *dev, pid_t pid)
+{
+	drm_device_dma_t *dma = dev->dma;
+	int		 i;
+
+	if (!dma) return;
+      	if(dev->dev_private == NULL) 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;
+
+		if (buf->pid == pid) {
+		   	/* Only buffers that need to get reclaimed ever 
+			 * get set to free */
+		   	if(buf_priv == NULL) return;
+		   	cmpxchg(buf_priv->in_use, 
+				I810_BUF_USED, I810_BUF_FREE);
+		}
+	}
+}
+
 int i810_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;
-	drm_queue_t	  *q;
-#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);
 
@@ -696,16 +748,20 @@ int i810_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 			  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 (lock.context < 0 || lock.context >= dev->queue_count) {
+	if (lock.context < 0) {
 		return -EINVAL;
 	}
-	q = dev->queuelist[lock.context];
-	
-	ret = drm_flush_block_and_flush(dev, lock.context, lock.flags);
+	/* Only one queue:
+	 */
 
 	if (!ret) {
-		if (_DRM_LOCKING_CONTEXT(dev->lock.hw_lock->lock)
+#if 0
+	   	if (_DRM_LOCKING_CONTEXT(dev->lock.hw_lock->lock)
 		    != lock.context) {
 			long j = jiffies - dev->lock.lock_time;
 
@@ -716,6 +772,7 @@ int i810_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 				schedule_timeout(j);
 			}
 		}
+#endif
 		add_wait_queue(&dev->lock.lock_queue, &entry);
 		for (;;) {
 			if (!dev->lock.hw_lock) {
@@ -728,13 +785,13 @@ int i810_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 				dev->lock.pid	    = current->pid;
 				dev->lock.lock_time = jiffies;
 				atomic_inc(&dev->total_locks);
-				atomic_inc(&q->total_locks);
 				break;	/* Got lock */
 			}
 			
 				/* Contention */
 			atomic_inc(&dev->total_sleeps);
 			current->state = TASK_INTERRUPTIBLE;
+		   	DRM_DEBUG("Calling lock schedule\n");
 			schedule();
 			if (signal_pending(current)) {
 				ret = -ERESTARTSYS;
@@ -744,19 +801,184 @@ int i810_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 		current->state = TASK_RUNNING;
 		remove_wait_queue(&dev->lock.lock_queue, &entry);
 	}
-
-	drm_flush_unblock(dev, lock.context, lock.flags); /* cleanup phase */
 	
 	if (!ret) {
-		if (lock.flags & _DRM_LOCK_READY)
-			i810_dma_ready(dev);
-		if (lock.flags & _DRM_LOCK_QUIESCENT)
-			i810_dma_quiescent(dev);
+		if (lock.flags & _DRM_LOCK_QUIESCENT) {
+		   DRM_DEBUG("_DRM_LOCK_QUIESCENT\n");
+		   DRM_DEBUG("fred\n");
+		   i810_dma_quiescent(dev);
+		}
 	}
+	DRM_DEBUG("%d %s\n", lock.context, ret ? "interrupted" : "has lock");
+	return ret;
+}
 
-#if DRM_DMA_HISTOGRAM
-	atomic_inc(&dev->histo.lacq[drm_histogram_slot(get_cycles() - start)]);
-#endif
+int i810_flush_ioctl(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_DEBUG("i810_flush_ioctl\n");
+   	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("i810_flush_ioctl called without lock held\n");
+		return -EINVAL;
+	}
+
+   	i810_flush_queue(dev);
+   	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)
+{
+	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;
+      	u32 *hw_status = (u32 *)dev_priv->hw_status_page;
+   	drm_i810_sarea_t *sarea_priv = (drm_i810_sarea_t *) 
+     					dev_priv->sarea_priv; 
+
+	int retcode = 0;
 	
-	return ret;
+	copy_from_user_ret(&general, (drm_i810_general_t *)arg, sizeof(general),
+			   -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");
+		return -EINVAL;
+	}
+
+	retcode = i810DmaGeneral(dev, &general);
+	sarea_priv->last_enqueue = dev_priv->counter-1;
+   	sarea_priv->last_dispatch = (int) hw_status[5];
+   
+	return retcode;
+}
+
+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_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(&vertex, (drm_i810_vertex_t *)arg, sizeof(vertex),
+			   -EFAULT);
+   	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("i810_dma_vertex 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);
+
+	retcode = i810DmaVertex(dev, &vertex);
+	sarea_priv->last_enqueue = dev_priv->counter-1;
+   	sarea_priv->last_dispatch = (int) hw_status[5];
+   
+	return retcode;
+
+}
+
+int i810_getage(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; 
+
+      	sarea_priv->last_dispatch = (int) hw_status[5];
+	return 0;
+}
+
+int i810_dma(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_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);
+   
+	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;
+
+	if (!retcode && d.request_count) {
+		retcode = i810_dma_get_buffers(dev, &d);
+	}
+
+	DRM_DEBUG("i810_dma: %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 = (int) hw_status[5];
+
+	return retcode;
 }
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..0754874c6
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_drm.h
@@ -0,0 +1,93 @@
+#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.
+ */
+
+/* Might one day want to support the client-side ringbuffer code again.
+ */
+#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 	2
+
+/* 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 _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 long ring_start;
+   	unsigned long ring_end;
+   	unsigned long ring_size;
+} 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 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 */
+} drm_i810_sarea_t;
+
+typedef struct _drm_i810_general {
+   	int idx;
+	int used;
+} drm_i810_general_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;
+
+#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 f33153a36..d3b35c493 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
@@ -33,11 +33,13 @@
 #define EXPORT_SYMTAB
 #include "drmP.h"
 #include "i810_drv.h"
+
+
 EXPORT_SYMBOL(i810_init);
 EXPORT_SYMBOL(i810_cleanup);
 
 #define I810_NAME	 "i810"
-#define I810_DESC	 "Matrox g200/g400"
+#define I810_DESC	 "Intel I810"
 #define I810_DATE	 "19991213"
 #define I810_MAJOR	 0
 #define I810_MINOR	 0
@@ -54,6 +56,7 @@ static struct file_operations i810_fops = {
 	mmap:	 drm_mmap,
 	read:	 drm_read,
 	fasync:	 drm_fasync,
+      	poll:	 drm_poll,
 };
 
 static struct miscdevice      i810_misc = {
@@ -80,13 +83,13 @@ static drm_ioctl_desc_t	      i810_ioctls[] = {
 	[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)]     = { 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_CTX)]     = { i810_addctx,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_RM_CTX)]      = { i810_rmctx,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_MOD_CTX)]     = { i810_modctx,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_GET_CTX)]     = { i810_getctx,	  1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_SWITCH_CTX)]  = { i810_switchctx,  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_NEW_CTX)]     = { i810_newctx,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_RES_CTX)]     = { i810_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 },
 
@@ -104,6 +107,11 @@ 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_FLUSH)]  = { i810_flush_ioctl,1, 0 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_I810_GETAGE)] = { i810_getage,     1, 0 },
 };
 
 #define I810_IOCTL_COUNT DRM_ARRAY_SIZE(i810_ioctls)
@@ -121,7 +129,7 @@ MODULE_PARM(i810, "s");
 
 int init_module(void)
 {
-	printk("doing i810_init()\n");
+	DRM_DEBUG("doing i810_init()\n");
 	return i810_init();
 }
 
@@ -364,7 +372,7 @@ int i810_init(void)
 #ifdef MODULE
 	drm_parse_options(i810);
 #endif
-	printk("doing misc_register\n");
+	DRM_DEBUG("doing misc_register\n");
 	if ((retcode = misc_register(&i810_misc))) {
 		DRM_ERROR("Cannot register \"%s\"\n", I810_NAME);
 		return retcode;
@@ -372,13 +380,22 @@ int i810_init(void)
 	dev->device = MKDEV(MISC_MAJOR, i810_misc.minor);
 	dev->name   = I810_NAME;
 
-   	printk("doing mem init\n");
+   	DRM_DEBUG("doing mem init\n");
 	drm_mem_init();
-	printk("doing proc init\n");
+	DRM_DEBUG("doing proc init\n");
 	drm_proc_init(dev);
-	printk("doing agp init\n");
+	DRM_DEBUG("doing agp init\n");
 	dev->agp    = drm_agp_init();
-	printk("doing ctxbitmap init\n");
+   	if(dev->agp == NULL) {
+	   	DRM_INFO("The i810 drm module requires the agpgart module"
+			 " to function correctly\nPlease load the agpgart"
+			 " module before you load the i810 module\n");
+	   	drm_proc_cleanup();
+	   	misc_deregister(&i810_misc);
+	   	i810_takedown(dev);
+	   	return -ENOMEM;
+	}
+	DRM_DEBUG("doing ctxbitmap init\n");
 	if((retcode = drm_ctxbitmap_init(dev))) {
 		DRM_ERROR("Cannot allocate memory for context bitmap.\n");
 		drm_proc_cleanup();
@@ -386,10 +403,6 @@ int i810_init(void)
 		i810_takedown(dev);
 		return retcode;
 	}
-#if 0
-	printk("doing i810_dma_init\n");
-	i810_dma_init(dev);
-#endif
 
 	DRM_INFO("Initialized %s %d.%d.%d %s on minor %d\n",
 		 I810_NAME,
@@ -417,7 +430,6 @@ void i810_cleanup(void)
 		DRM_INFO("Module unloaded\n");
 	}
 	drm_ctxbitmap_cleanup(dev);
-	i810_dma_cleanup(dev);
 	i810_takedown(dev);
 	if (dev->agp) {
 		drm_free(dev->agp, sizeof(*dev->agp), DRM_MEM_AGPLISTS);
@@ -484,24 +496,82 @@ int i810_release(struct inode *inode, struct file *filp)
 	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;
+	DRM_DEBUG("pid = %d, device = 0x%x, open_count = %d\n",
+		  current->pid, dev->device, dev->open_count);
+
+	if (_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)
+	    && dev->lock.pid == current->pid) {
+	      	i810_reclaim_buffers(dev, priv->pid);
+		DRM_ERROR("Process %d dead, freeing lock for context %d\n",
+			  current->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. */
+	} else {
+	   	/* The lock is required to reclaim buffers */
+	   	DECLARE_WAITQUEUE(entry, current);
+	   	add_wait_queue(&dev->lock.lock_queue, &entry);
+		for (;;) {
+			if (!dev->lock.hw_lock) {
+				/* Device has been unregistered */
+				retcode = -EINTR;
+				break;
+			}
+			if (drm_lock_take(&dev->lock.hw_lock->lock,
+					  DRM_KERNEL_CONTEXT)) {
+				dev->lock.pid	    = priv->pid;
+				dev->lock.lock_time = jiffies;
+				atomic_inc(&dev->total_locks);
+				break;	/* Got lock */
+			}			
+				/* Contention */
+			atomic_inc(&dev->total_sleeps);
+			current->state = TASK_INTERRUPTIBLE;
+			schedule();
+			if (signal_pending(current)) {
+				retcode = -ERESTARTSYS;
+				break;
 			}
-			spin_unlock(&dev->count_lock);
-			return i810_takedown(dev);
 		}
-		spin_unlock(&dev->count_lock);
+		current->state = TASK_RUNNING;
+		remove_wait_queue(&dev->lock.lock_queue, &entry);
+	   	if(!retcode) {
+		   	i810_reclaim_buffers(dev, priv->pid);
+		   	drm_lock_free(dev, &dev->lock.hw_lock->lock,
+				      DRM_KERNEL_CONTEXT);
+		}
+	}
+	drm_fasync(-1, filp, 0);
+
+	down(&dev->struct_sem);
+	if (priv->prev) priv->prev->next = priv->next;
+	else		dev->file_first	 = priv->next;
+	if (priv->next) priv->next->prev = priv->prev;
+	else		dev->file_last	 = priv->prev;
+	up(&dev->struct_sem);
+	
+	drm_free(priv, sizeof(*priv), DRM_MEM_FILES);
+   	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 i810_takedown(dev);
 	}
+   	spin_unlock(&dev->count_lock);
 	return retcode;
 }
 
@@ -566,8 +636,7 @@ int i810_unlock(struct inode *inode, struct file *filp, unsigned int cmd,
 	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);
-	i810_dma_schedule(dev, 1);
+   	drm_lock_transfer(dev, &dev->lock.hw_lock->lock, DRM_KERNEL_CONTEXT);
 	if (!dev->context_flag) {
 		if (drm_lock_free(dev, &dev->lock.hw_lock->lock,
 				  DRM_KERNEL_CONTEXT)) {
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 0f5f42bb6..1badd36b8 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 +32,31 @@
 #ifndef _I810_DRV_H_
 #define _I810_DRV_H_
 
+typedef struct _drm_i810_ring_buffer{
+	int tail_mask;
+	unsigned long Start;
+	unsigned long End;
+	unsigned long Size;
+	u8 *virtual_start;
+	int head;
+	int tail;
+	int space;
+} drm_i810_ring_buffer_t;
+
+typedef struct drm_i810_private {
+   	int ring_map_idx;
+   	int buffer_map_idx;
+
+   	drm_i810_ring_buffer_t ring;
+	drm_i810_sarea_t *sarea_priv;
+
+      	unsigned long hw_status_page;
+   	unsigned long counter;
+
+   	atomic_t flush_done;
+   	wait_queue_head_t flush_queue;	/* Processes waiting until flush    */
+} drm_i810_private_t;
+
 				/* i810_drv.c */
 extern int  i810_init(void);
 extern void i810_cleanup(void);
@@ -54,8 +79,13 @@ extern int  i810_control(struct inode *inode, struct file *filp,
 			  unsigned int cmd, unsigned long arg);
 extern int  i810_lock(struct inode *inode, struct file *filp,
 		       unsigned int cmd, unsigned long arg);
-extern void i810_dma_init(drm_device_t *dev);
-extern void i810_dma_cleanup(drm_device_t *dev);
+extern int  i810_dma_init(struct inode *inode, struct file *filp,
+			  unsigned int cmd, unsigned long arg);
+extern int  i810_flush_ioctl(struct inode *inode, struct file *filp,
+			     unsigned int cmd, unsigned long arg);
+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);
 
 
 				/* i810_bufs.c */
@@ -72,5 +102,103 @@ extern int  i810_mapbufs(struct inode *inode, struct file *filp,
 extern int  i810_addmap(struct inode *inode, struct file *filp,
 		       unsigned int cmd, unsigned long arg);
 
+				/* i810_context.c */
+extern int  i810_resctx(struct inode *inode, struct file *filp,
+		       unsigned int cmd, unsigned long arg);
+extern int  i810_addctx(struct inode *inode, struct file *filp,
+		       unsigned int cmd, unsigned long arg);
+extern int  i810_modctx(struct inode *inode, struct file *filp,
+		       unsigned int cmd, unsigned long arg);
+extern int  i810_getctx(struct inode *inode, struct file *filp,
+		       unsigned int cmd, unsigned long arg);
+extern int  i810_switchctx(struct inode *inode, struct file *filp,
+			  unsigned int cmd, unsigned long arg);
+extern int  i810_newctx(struct inode *inode, struct file *filp,
+		       unsigned int cmd, unsigned long arg);
+extern int  i810_rmctx(struct inode *inode, struct file *filp,
+		      unsigned int cmd, unsigned long arg);
+
+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);
+
+
+#define GFX_OP_USER_INTERRUPT 		((0<<29)|(2<<23))
+#define GFX_OP_BREAKPOINT_INTERRUPT	((0<<29)|(1<<23))
+#define CMD_REPORT_HEAD			(7<<23)
+#define CMD_STORE_DWORD_IDX		((0x21<<23) | 0x1)
+#define CMD_OP_BATCH_BUFFER  ((0x0<<29)|(0x30<<23)|0x1)
+
+#define INST_PARSER_CLIENT   0x00000000
+#define INST_OP_FLUSH        0x02000000
+#define INST_FLUSH_MAP_CACHE 0x00000001
+
+
+#define BB1_START_ADDR_MASK   (~0x7)
+#define BB1_PROTECTED         (1<<0)
+#define BB1_UNPROTECTED       (0<<0)
+#define BB2_END_ADDR_MASK     (~0x7)
+
+#define I810REG_HWSTAM		0x02098
+#define I810REG_INT_IDENTITY_R	0x020a4
+#define I810REG_INT_MASK_R 	0x020a8
+#define I810REG_INT_ENABLE_R	0x020a0
+
+#define LP_RING     		0x2030
+#define HP_RING     		0x2040
+#define RING_TAIL      		0x00
+#define TAIL_ADDR		0x000FFFF8
+#define RING_HEAD      		0x04
+#define HEAD_WRAP_COUNT     	0xFFE00000
+#define HEAD_WRAP_ONE       	0x00200000
+#define HEAD_ADDR           	0x001FFFFC
+#define RING_START     		0x08
+#define START_ADDR          	0x00FFFFF8
+#define RING_LEN       		0x0C
+#define RING_NR_PAGES       	0x000FF000 
+#define RING_REPORT_MASK    	0x00000006
+#define RING_REPORT_64K     	0x00000002
+#define RING_REPORT_128K    	0x00000004
+#define RING_NO_REPORT      	0x00000000
+#define RING_VALID_MASK     	0x00000001
+#define RING_VALID          	0x00000001
+#define RING_INVALID        	0x00000000
+
+#define GFX_OP_SCISSOR         ((0x3<<29)|(0x1c<<24)|(0x10<<19))
+#define SC_UPDATE_SCISSOR       (0x1<<1)
+#define SC_ENABLE_MASK          (0x1<<0)
+#define SC_ENABLE               (0x1<<0)
+
+#define GFX_OP_SCISSOR_INFO    ((0x3<<29)|(0x1d<<24)|(0x81<<16)|(0x1))
+#define SCI_YMIN_MASK      (0xffff<<16)
+#define SCI_XMIN_MASK      (0xffff<<0)
+#define SCI_YMAX_MASK      (0xffff<<16)
+#define SCI_XMAX_MASK      (0xffff<<0)
 
 #endif
+
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/memory.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/memory.c
index a8a81abd8..559ac7391 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/memory.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/memory.c
@@ -44,21 +44,25 @@ static spinlock_t	  drm_mem_lock	    = SPIN_LOCK_UNLOCKED;
 static unsigned long	  drm_ram_available = 0; /* In pages */
 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_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 */
 };
 
@@ -324,3 +328,120 @@ void drm_ioremapfree(void *pt, unsigned long size)
 			      free_count, alloc_count);
 	}
 }
+
+#ifdef DRM_AGP
+agp_memory *drm_alloc_agp(int pages, u32 type)
+{
+	agp_memory *handle;
+
+	if (!pages) {
+		DRM_MEM_ERROR(DRM_MEM_TOTALAGP, "Allocating 0 pages\n");
+		return NULL;
+	}
+	
+	if (drm_agp.allocate_memory) {
+		if ((handle = (*drm_agp.allocate_memory)(pages,
+							 type))) {
+			spin_lock(&drm_mem_lock);
+			++drm_mem_stats[DRM_MEM_TOTALAGP].succeed_count;
+			drm_mem_stats[DRM_MEM_TOTALAGP].bytes_allocated
+				+= pages << PAGE_SHIFT;
+			spin_unlock(&drm_mem_lock);
+			return handle;
+		}
+	}
+	spin_lock(&drm_mem_lock);
+	++drm_mem_stats[DRM_MEM_TOTALAGP].fail_count;
+	spin_unlock(&drm_mem_lock);
+	return NULL;
+}
+
+int drm_free_agp(agp_memory *handle, int pages)
+{
+	int           alloc_count;
+	int           free_count;
+	int           retval = -EINVAL;
+
+	if (!handle) {
+		DRM_MEM_ERROR(DRM_MEM_TOTALAGP,
+			      "Attempt to free NULL AGP handle\n");
+		return retval;;
+	}
+	
+	if (drm_agp.free_memory) {
+		(*drm_agp.free_memory)(handle);
+		spin_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;
+		spin_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;
+	}
+	return retval;
+}
+
+int drm_bind_agp(agp_memory *handle, unsigned int start)
+{
+	int retcode = -EINVAL;
+
+   DRM_DEBUG("drm_bind_agp called\n");
+	if (!handle) {
+		DRM_MEM_ERROR(DRM_MEM_BOUNDAGP,
+			      "Attempt to bind NULL AGP handle\n");
+		return retcode;
+	}
+
+   DRM_DEBUG("drm_agp.bind_memory : %p\n", drm_agp.bind_memory);
+	if (drm_agp.bind_memory) {
+		if (!(retcode = (*drm_agp.bind_memory)(handle, start))) {
+			spin_lock(&drm_mem_lock);
+			++drm_mem_stats[DRM_MEM_BOUNDAGP].succeed_count;
+			drm_mem_stats[DRM_MEM_BOUNDAGP].bytes_allocated
+				+= handle->page_count << PAGE_SHIFT;
+			spin_unlock(&drm_mem_lock);
+		   DRM_DEBUG("drm_agp.bind_memory: retcode %d\n", retcode);
+			return retcode;
+		}
+	}
+	spin_lock(&drm_mem_lock);
+	++drm_mem_stats[DRM_MEM_BOUNDAGP].fail_count;
+	spin_unlock(&drm_mem_lock);
+	return retcode;
+}
+
+int drm_unbind_agp(agp_memory *handle)
+{
+	int alloc_count;
+	int free_count;
+	int retcode = -EINVAL;
+	
+	if (!handle) {
+		DRM_MEM_ERROR(DRM_MEM_BOUNDAGP,
+			      "Attempt to unbind NULL AGP handle\n");
+		return retcode;
+	}
+
+	if (drm_agp.unbind_memory) {
+		int c = handle->page_count;
+		if ((retcode = (*drm_agp.unbind_memory)(handle)))
+			return retcode;
+		spin_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 += c << PAGE_SHIFT;
+		spin_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/linux/drm/kernel/mga_bufs.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_bufs.c
index 34f1112a6..3ce428b27 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_bufs.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_bufs.c
@@ -33,169 +33,168 @@
 #define __NO_VERSION__
 #include "drmP.h"
 #include "mga_drv.h"
-#include "mga_dma.h"
 #include "linux/un.h"
 
 
 int mga_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;
-   agp_offset = request.agp_start;
-   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;
-
-   DRM_DEBUG("count: %d\n", count);
-   DRM_DEBUG("order: %d\n", order);
-   DRM_DEBUG("size: %d\n", size);
-   DRM_DEBUG("agp_offset: %d\n", agp_offset);
-   DRM_DEBUG("alignment: %d\n", alignment);
-   DRM_DEBUG("page_order: %d\n", page_order);
-   DRM_DEBUG("total: %d\n", total);
-   DRM_DEBUG("byte_count: %d\n", byte_count);
-
-   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);
+	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;
+	agp_offset = request.agp_start;
+	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;
+
+	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);
+	DRM_DEBUG("byte_count: %d\n", byte_count);
+
+	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 */
-   }
+	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->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;
+	entry->buf_size   = size;
+	entry->page_order = page_order;
+	offset = 0;
 
    
-   while(entry->buf_count < count) {
-      buf = &entry->buflist[entry->buf_count];
-      buf->idx = dma->buf_count + entry->buf_count;
-      buf->total = alignment;
-      buf->order = order;
-      buf->used = 0;
-
-      DRM_DEBUG("offset : %d\n", offset);
-
-      buf->offset = offset; /* Hrm */
-      buf->bus_address = dev->agp->base + agp_offset + offset;
-      buf->address = (void *)(agp_offset + offset + dev->agp->base);
-      buf->next = NULL;
-      buf->waiting = 0;
-      buf->pending = 0;
-      init_waitqueue_head(&buf->dma_wait);
-      buf->pid = 0;
-
-      buf->dev_private = drm_alloc(sizeof(drm_mga_buf_priv_t), DRM_MEM_BUFS);
-      buf->dev_priv_size = sizeof(drm_mga_buf_priv_t);
+	while(entry->buf_count < count) {
+		buf = &entry->buflist[entry->buf_count];
+		buf->idx = dma->buf_count + entry->buf_count;
+		buf->total = alignment;
+		buf->order = order;
+		buf->used = 0;
+
+		DRM_DEBUG("offset : %ld\n", offset);
+
+		buf->offset = offset; /* Hrm */
+		buf->bus_address = dev->agp->base + agp_offset + offset;
+		buf->address = (void *)(agp_offset + offset + dev->agp->base);
+		buf->next = NULL;
+		buf->waiting = 0;
+		buf->pending = 0;
+		init_waitqueue_head(&buf->dma_wait);
+		buf->pid = 0;
+
+		buf->dev_private = drm_alloc(sizeof(drm_mga_buf_priv_t), DRM_MEM_BUFS);
+		buf->dev_priv_size = sizeof(drm_mga_buf_priv_t);
 
 #if DRM_DMA_HISTOGRAM
-      buf->time_queued = 0;
-      buf->time_dispatched = 0;
-      buf->time_completed = 0;
-      buf->time_freed = 0;
+		buf->time_queued = 0;
+		buf->time_dispatched = 0;
+		buf->time_completed = 0;
+		buf->time_freed = 0;
 #endif
-      offset = offset + alignment;
-      entry->buf_count++;
-      byte_count += PAGE_SIZE << page_order;
+		offset = offset + alignment;
+		entry->buf_count++;
+		byte_count += PAGE_SIZE << page_order;
       
-      DRM_DEBUG("buffer %d @ %p\n",
-	     entry->buf_count, buf->address);
-   }
+		DRM_DEBUG("buffer %d @ %p\n",
+			  entry->buf_count, buf->address);
+	}
    
-   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->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->buf_count  += entry->buf_count;
 
-   DRM_DEBUG("dma->buf_count : %d\n", dma->buf_count);
+	DRM_DEBUG("dma->buf_count : %d\n", dma->buf_count);
 
-   dma->byte_count += byte_count;
+	dma->byte_count += byte_count;
 
-   DRM_DEBUG("entry->buf_count : %d\n", entry->buf_count);
+	DRM_DEBUG("entry->buf_count : %d\n", entry->buf_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]);
-   }
+	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);
+	up(&dev->struct_sem);
    
-   request.count = entry->buf_count;
-   request.size  = size;
+	request.count = entry->buf_count;
+	request.size  = size;
    
-   copy_to_user_ret((drm_buf_desc_t *)arg,
-		    &request,
-		    sizeof(request),
-		    -EFAULT);
+	copy_to_user_ret((drm_buf_desc_t *)arg,
+			 &request,
+			 sizeof(request),
+			 -EFAULT);
    
-   atomic_dec(&dev->buf_alloc);
+	atomic_dec(&dev->buf_alloc);
 
-   DRM_DEBUG("count: %d\n", count);
-   DRM_DEBUG("order: %d\n", order);
-   DRM_DEBUG("size: %d\n", size);
-   DRM_DEBUG("agp_offset: %d\n", agp_offset);
-   DRM_DEBUG("alignment: %d\n", alignment);
-   DRM_DEBUG("page_order: %d\n", page_order);
-   DRM_DEBUG("total: %d\n", total);
-   DRM_DEBUG("byte_count: %d\n", byte_count);
+	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);
+	DRM_DEBUG("byte_count: %d\n", byte_count);
 
-   dma->flags = _DRM_DMA_USE_AGP;
+	dma->flags = _DRM_DMA_USE_AGP;
 
-   DRM_DEBUG("dma->flags : %lx\n", dma->flags);
+	DRM_DEBUG("dma->flags : %x\n", dma->flags);
 
-   return 0;
+	return 0;
 }
 
 int mga_addbufs_pci(struct inode *inode, struct file *filp, unsigned int cmd,
@@ -362,17 +361,17 @@ int mga_addbufs_pci(struct inode *inode, struct file *filp, unsigned int cmd,
 int mga_addbufs(struct inode *inode, struct file *filp, unsigned int cmd,
 		unsigned long arg)
 {
-   drm_buf_desc_t	 request;
+	drm_buf_desc_t	 request;
 
-   copy_from_user_ret(&request,
-		      (drm_buf_desc_t *)arg,
-		      sizeof(request),
-		      -EFAULT);
+	copy_from_user_ret(&request,
+			   (drm_buf_desc_t *)arg,
+			   sizeof(request),
+			   -EFAULT);
 
-   if(request.flags & _DRM_AGP_BUFFER)
-     return mga_addbufs_agp(inode, filp, cmd, arg);
-   else
-     return mga_addbufs_pci(inode, filp, cmd, arg);
+	if(request.flags & _DRM_AGP_BUFFER)
+		return mga_addbufs_agp(inode, filp, cmd, arg);
+	else
+		return mga_addbufs_pci(inode, filp, cmd, arg);
 }
 
 int mga_infobufs(struct inode *inode, struct file *filp, unsigned int cmd,
@@ -546,7 +545,7 @@ int mga_mapbufs(struct inode *inode, struct file *filp, unsigned int cmd,
 	spin_lock(&dev->count_lock);
 	if (atomic_read(&dev->buf_alloc)) {
 		spin_unlock(&dev->count_lock);
-	   DRM_DEBUG("Buzy\n");
+		DRM_DEBUG("Busy\n");
 		return -EBUSY;
 	}
 	++dev->buf_use;		/* Can't allocate more after this call */
@@ -558,79 +557,84 @@ int mga_mapbufs(struct inode *inode, struct file *filp, unsigned int cmd,
 			   -EFAULT);
 
 	DRM_DEBUG("mga_mapbufs\n");
-   	DRM_DEBUG("dma->flags : %lx\n", dma->flags);
+   	DRM_DEBUG("dma->flags : %x\n", dma->flags);
    
-   if (request.count >= dma->buf_count) {
-      if(dma->flags & _DRM_DMA_USE_AGP) {
-	 drm_mga_private_t *dev_priv = dev->dev_private;
-	 drm_map_t *map = NULL;
+	if (request.count >= dma->buf_count) {
+		if(dma->flags & _DRM_DMA_USE_AGP) {
+			drm_mga_private_t *dev_priv = 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);
-
-	 virtual = do_mmap(filp, 0, map->size, PROT_READ|PROT_WRITE,
-			   MAP_SHARED, (unsigned long)map->offset);
-      } else {
-	      virtual = do_mmap(filp, 0, dma->byte_count,
-				PROT_READ|PROT_WRITE, MAP_SHARED, 0);
-      }
-      if (virtual > -1024UL) {
-	 /* Real error */
-	 DRM_DEBUG("mmap error\n");
-	 retcode = (signed long)virtual;
-	 goto done;
-      }
-      request.virtual = (void *)virtual;
+			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 : %p\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);
+		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);
+	copy_to_user_ret((drm_buf_map_t *)arg,
+			 &request,
+			 sizeof(request),
+			 -EFAULT);
 
-   DRM_DEBUG("retcode : %d\n", retcode);
+	DRM_DEBUG("retcode : %d\n", retcode);
 
-   return retcode;
+	return retcode;
 }
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_clear.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_clear.c
deleted file mode 100644
index 05895415e..000000000
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_clear.c
+++ /dev/null
@@ -1,417 +0,0 @@
-/* mga_state.c -- State support for mga g200/g400 -*- linux-c -*-
- *
- * Created: February 2000 by keithw@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: 
- *           Keith Whitwell <keithw@precisioninsight.com>
- *
- */
- 
-#define __NO_VERSION__
-#include "drmP.h"
-#include "mga_drv.h"
-#include "mgareg_flags.h"
-#include "mga_dma.h"
-#include "mga_state.h"
-
-#define MGA_CLEAR_CMD (DC_opcod_trap | DC_arzero_enable | 		\
-		       DC_sgnzero_enable | DC_shftzero_enable | 	\
-		       (0xC << DC_bop_SHIFT) | DC_clipdis_enable | 	\
-		       DC_solid_enable | DC_transc_enable)
-	  
-
-#define MGA_COPY_CMD (DC_opcod_bitblt | DC_atype_rpl | DC_linear_xy |	\
-		      DC_solid_disable | DC_arzero_disable | 		\
-		      DC_sgnzero_enable | DC_shftzero_enable | 		\
-		      (0xC << DC_bop_SHIFT) | DC_bltmod_bfcol | 	\
-		      DC_pattern_disable | DC_transc_disable | 		\
-		      DC_clipdis_enable)				\
-
-
-
-/* Build and queue a TT_GENERAL secondary buffer to do the clears.
- * With Jeff's ringbuffer idea, it might make sense if there are only
- * one or two cliprects to emit straight to the primary buffer.
- */
-static int mgaClearBuffers(drm_device_t *dev,
-			   int clear_color,
-			   int clear_depth,
-			   int flags)
-{
-	int cmd, i;	
-	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;
-	xf86drmClipRectRec *pbox = sarea_priv->boxes;
-	int nbox = sarea_priv->nbox;
-	drm_buf_t *buf;
-	drm_dma_t d;
-	int order = 10;		/* ??? what orders do we have ???*/
-	DMALOCALS;
-
-
-	if (!nbox) 
-		return -EINVAL;
-
-	if ( dev_priv->sgram ) 
-		cmd = MGA_CLEAR_CMD | DC_atype_blk;
-	else
-		cmd = MGA_CLEAR_CMD | DC_atype_rstr;
-	    
-	buf = drm_freelist_get(&dma->bufs[order].freelist, _DRM_DMA_WAIT);
-
-
-	DMAGETPTR( buf );
-
-	for (i = 0 ; i < nbox ; i++) {
-		unsigned int height = pbox[i].y2 - pbox[i].y1;
-		
-		/* Is it necessary to be this paranoid?  I don't think so.
-		if (pbox[i].x1 > dev_priv->width) continue;
-		if (pbox[i].y1 > dev_priv->height) continue;
-		if (pbox[i].x2 > dev_priv->width) continue;
-		if (pbox[i].y2 > dev_priv->height) continue;
-		if (pbox[i].x2 <= pbox[i].x1) continue;
-		if (pbox[i].y2 <= pbox[i].x1) continue;
-		 */
-
-		DMAOUTREG(MGAREG_YDSTLEN, (pbox[i].y1<<16)|height);
-		DMAOUTREG(MGAREG_FXBNDRY, (pbox[i].x2<<16)|pbox[i].x1);
-
-		if ( flags & MGA_CLEAR_FRONT ) {	    
-			DMAOUTREG(MGAREG_FCOL, clear_color);
-			DMAOUTREG(MGAREG_DSTORG, dev_priv->frontOrg);
-			DMAOUTREG(MGAREG_DWGCTL+MGAREG_MGA_EXEC, cmd );
-		}
-
-		if ( flags & MGA_CLEAR_BACK ) {
-			DMAOUTREG(MGAREG_FCOL, clear_color);
-			DMAOUTREG(MGAREG_DSTORG, dev_priv->backOrg);
-			DMAOUTREG(MGAREG_DWGCTL+MGAREG_MGA_EXEC, cmd );
-		}
-
-		if ( flags & MGA_CLEAR_DEPTH ) 
-		{
-			DMAOUTREG(MGAREG_FCOL, clear_depth);
-			DMAOUTREG(MGAREG_DSTORG, dev_priv->depthOrg);
-			DMAOUTREG(MGAREG_DWGCTL+MGAREG_MGA_EXEC, cmd );
-		}
-	}
-
-	DMAADVANCE( buf );
-
-	/* Make sure we restore the 3D state next time.
-	 */
-	sarea_priv->dirty |= MGASAREA_NEW_CONTEXT;
-
-	((drm_mga_buf_priv_t *)buf->dev_private)->dma_type = MGA_DMA_GENERAL;
-
-	d.context = DRM_KERNEL_CONTEXT;
-	d.send_count = 1;
-	d.send_indices = &buf->idx;
-	d.send_sizes = &buf->used;
-   	d.flags = 0;
-	d.request_count = 0;
-	d.request_size = 0;
-	d.request_indices = NULL;
-	d.request_sizes = NULL;
-	d.granted_count = 0;	   
-
-      	atomic_inc(&dev_priv->pending_bufs);
-   	if((drm_dma_enqueue(dev, &d)) != 0) 
-     		atomic_dec(&dev_priv->pending_bufs);
-	mga_dma_schedule(dev, 1);
-   	return 0;
-}
-
-int mgaSwapBuffers(drm_device_t *dev, int flags) 
-{
-	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;
-	xf86drmClipRectRec *pbox = sarea_priv->boxes;
-	int nbox = sarea_priv->nbox;
-	drm_buf_t *buf;
-	drm_dma_t d;
-	int order = 10;		/* ??? */
-	int i;
-	DMALOCALS;	
-
-	if (!nbox) 
-		return -EINVAL;
-
-	buf = drm_freelist_get(&dma->bufs[order].freelist, _DRM_DMA_WAIT);
-
-	DMAGETPTR(buf);
-
-	DMAOUTREG(MGAREG_DSTORG, dev_priv->frontOrg);
-	DMAOUTREG(MGAREG_MACCESS, dev_priv->mAccess);
-	DMAOUTREG(MGAREG_SRCORG, dev_priv->backOrg);
-	DMAOUTREG(MGAREG_AR5, dev_priv->stride); /* unnecessary? */
-	DMAOUTREG(MGAREG_DWGCTL, MGA_COPY_CMD); 
-	     
-	for (i = 0 ; i < nbox; i++) {
-		unsigned int h = pbox[i].y2 - pbox[i].y1;
-		unsigned int start = pbox[i].y1 * dev_priv->stride;
-
-		/*
-		if (pbox[i].x1 > dev_priv->width) continue;
-		if (pbox[i].y1 > dev_priv->height) continue;
-		if (pbox[i].x2 > dev_priv->width) continue;
-		if (pbox[i].y2 > dev_priv->height) continue;
-		if (pbox[i].x2 <= pbox[i].x1) continue;
-		if (pbox[i].y2 <= pbox[i].x1) continue;		
-		*/
-
-		DMAOUTREG(MGAREG_AR0, start + pbox[i].x2 - 1);
-		DMAOUTREG(MGAREG_AR3, start + pbox[i].x1);		
-		DMAOUTREG(MGAREG_FXBNDRY, pbox[i].x1|((pbox[i].x2 - 1)<<16));
-		DMAOUTREG(MGAREG_YDSTLEN+MGAREG_MGA_EXEC, (pbox[i].y1<<16)|h);
-	}
-  
-	DMAOUTREG(MGAREG_SRCORG, 0);
-	DMAADVANCE( buf );
-
-	/* Make sure we restore the 3D state next time.
-	 */
-	sarea_priv->dirty |= MGASAREA_NEW_CONTEXT;
-
-	((drm_mga_buf_priv_t *)buf->dev_private)->dma_type = MGA_DMA_GENERAL;
-
-	d.context = DRM_KERNEL_CONTEXT;
-	d.send_count = 1;
-	d.send_indices = &buf->idx;
-	d.send_sizes = &buf->used;
-   	d.flags = 0;
-	d.request_count = 0;
-	d.request_size = 0;
-	d.request_indices = NULL;
-	d.request_sizes = NULL;
-	d.granted_count = 0;	 
-
-   	atomic_inc(&dev_priv->pending_bufs);
-      	if((drm_dma_enqueue(dev, &d)) != 0) 
-     		atomic_dec(&dev_priv->pending_bufs);
-	mga_dma_schedule(dev, 1);
-   	return 0;
-}
-
-
-static int mgaIload(drm_device_t *dev, drm_mga_iload_t *args)
-{
-	drm_device_dma_t *dma = dev->dma;
-   	drm_buf_t *buf = dma->buflist[ args->idx ];
-	drm_mga_buf_priv_t *buf_priv = (drm_mga_buf_priv_t *)buf->dev_private;
-   	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
-	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
-	drm_dma_t d;
-   	int pixperdword;
-   	
-   	buf_priv->dma_type = MGA_DMA_ILOAD;
-   	buf_priv->boxes[0].y1 = args->texture.y1;
-   	buf_priv->boxes[0].y2 = args->texture.y2;
-   	buf_priv->boxes[0].x1 = args->texture.x1;
-   	buf_priv->boxes[0].x2 = args->texture.x2;
-   	buf_priv->ContextState[MGA_CTXREG_DSTORG] = args->destOrg;
-   	buf_priv->ContextState[MGA_CTXREG_MACCESS] = args->mAccess;
-   	buf_priv->ServerState[MGA_2DREG_PITCH] = args->pitch;
-	buf_priv->nbox = 1;   
-   	sarea_priv->dirty |= (MGASAREA_NEW_CONTEXT | MGASAREA_NEW_2D);
-   	switch((args->mAccess & 0x00000003)) {
-	 	case 0:
-	   		pixperdword = 4;
-	   	break;
-	 	case 1:
-	   		pixperdword = 2;
-	   	break;
-	 	case 2:
-	   		pixperdword = 1;
-	  	break;
-		default:
-	   		DRM_ERROR("Invalid maccess value passed"
-				  " to mgaIload\n");
-	   	return -EINVAL;
-	}
-   	buf->used = ((args->texture.y2 - args->texture.y1) *
-		     (args->texture.x2 - args->texture.x1) /
-		     pixperdword);
-   	DRM_DEBUG("buf->used : %d\n", buf->used);
-	d.context = DRM_KERNEL_CONTEXT;
-	d.send_count = 1;
-	d.send_indices = &buf->idx;
-	d.send_sizes = &buf->used;
-	d.flags = 0;
-	d.request_count = 0;
-	d.request_size = 0;
-	d.request_indices = NULL;
-	d.request_sizes = NULL;
-	d.granted_count = 0;	 
-   
-      	atomic_inc(&dev_priv->pending_bufs);
-      	if((drm_dma_enqueue(dev, &d)) != 0) 
-     		atomic_dec(&dev_priv->pending_bufs);
-	mga_dma_schedule(dev, 1);
-
-	return 0; 
-}
-
-
-/* Necessary?  Not necessary??
- */
-static int check_lock(void)
-{
-	return 1;
-}
-
-int mga_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_mga_clear_t clear;
-	int retcode;
-   
-	copy_from_user_ret(&clear, (drm_mga_clear_t *)arg,
-			   sizeof(clear), -EFAULT);
-   
-/*  	if (!check_lock( dev )) */
-/*  		return -EIEIO; */
-		
-	retcode = mgaClearBuffers(dev, clear.clear_color,
-				  clear.clear_depth,
-				  clear.flags);
-   
-	return retcode;
-}
-
-int mga_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;
-	drm_mga_swap_t swap;
-	int retcode = 0;
-
-/*  	if (!check_lock( dev )) */
-/*  		return -EIEIO; */
-   
-	copy_from_user_ret(&swap, (drm_mga_swap_t *)arg,
-			   sizeof(swap), -EFAULT);
-   
-	retcode = mgaSwapBuffers(dev, swap.flags);
-   
-	return retcode;
-}
-
-int mga_iload(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_iload_t iload;
-	int retcode = 0;
-
-/*  	if (!check_lock( dev )) */
-/*  		return -EIEIO; */
-
-	copy_from_user_ret(&iload, (drm_mga_iload_t *)arg, 
-			   sizeof(iload), -EFAULT);
-   
-	retcode = mgaIload(dev, &iload);
-   
-	return retcode;
-}
-
-
-int mga_dma(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;
-
-   	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);
-
-	/* Per-context queues are unworkable if you are trying to do
-	 * state management from the client.
-	 */
-	d.context = DRM_KERNEL_CONTEXT;
-	d.flags &= ~_DRM_DMA_WHILE_LOCKED;
-
-	/* Maybe multiple buffers is useful for iload...
-	 * But this ioctl is only for *despatching* vertex data...
-	 */
-	if (d.send_count < 0 || d.send_count > 1) {
-		DRM_ERROR("Process %d trying to send %d buffers (max 1)\n",
-			  current->pid, d.send_count);
-		return -EINVAL;
-	}
-
-	
-	/* But it *is* used to request buffers for all types of dma:
-	 */
-	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;
-	}
-
-	if (d.send_count) {
-		int idx = d.send_indices[0];
-		drm_mga_buf_priv_t *buf_priv = dma->buflist[ idx ]->dev_private;
-		drm_mga_private_t *dev_priv = dev->dev_private;
-
-		buf_priv->dma_type = MGA_DMA_VERTEX;
-
-/*         	if (!check_lock( dev )) */
-/*  		        return -EIEIO; */
-
-		/* Snapshot the relevent bits of the sarea... 
-		 */
-		mgaCopyAndVerifyState( dev_priv, buf_priv );
-
-	      	atomic_inc(&dev_priv->pending_bufs);
-		retcode = drm_dma_enqueue(dev, &d);
-	      	if(retcode != 0) 
-     			atomic_dec(&dev_priv->pending_bufs);
-		mga_dma_schedule(dev, 1);
-	}
-	
-	d.granted_count = 0;
-
-	if (!retcode && d.request_count) {
-		retcode = drm_dma_get_buffers(dev, &d);
-	}
-
-	DRM_DEBUG("%d returning, granted = %d\n",
-		  current->pid, d.granted_count);
-	copy_to_user_ret((drm_dma_t *)arg, &d, sizeof(d), -EFAULT);
-
-	return retcode;
-}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_context.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_context.c
index fb0d336fa..2459b35bb 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_context.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_context.c
@@ -38,7 +38,7 @@
 static int mga_alloc_queue(drm_device_t *dev)
 {
    	int temp = drm_ctxbitmap_next(dev);
-   	printk("mga_alloc_queue: %d\n", temp);
+   	DRM_DEBUG("mga_alloc_queue: %d\n", temp);
 	return temp;
 }
 
@@ -57,7 +57,7 @@ int mga_context_switch(drm_device_t *dev, int old, int new)
         dev->ctx_start = get_cycles();
 #endif
         
-        printk("Context switch from %d to %d\n", old, new);
+        DRM_DEBUG("Context switch from %d to %d\n", old, new);
 
         if (new == dev->last_context) {
                 clear_bit(0, &dev->context_flag);
@@ -104,7 +104,7 @@ int mga_resctx(struct inode *inode, struct file *filp, unsigned int cmd,
 	drm_ctx_t	ctx;
 	int		i;
 
-	printk("%d\n", DRM_RESERVED_CONTEXTS);
+	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));
@@ -134,11 +134,11 @@ int mga_addctx(struct inode *inode, struct file *filp, unsigned int cmd,
 		ctx.handle = mga_alloc_queue(dev);
 	}
         if (ctx.handle == -1) {
-		printk("Not enough free contexts.\n");
+		DRM_DEBUG("Not enough free contexts.\n");
 				/* Should this return -EBUSY instead? */
 		return -ENOMEM;
 	}
-	printk("%d\n", ctx.handle);
+	DRM_DEBUG("%d\n", ctx.handle);
 	copy_to_user_ret((drm_ctx_t *)arg, &ctx, sizeof(ctx), -EFAULT);
 	return 0;
 }
@@ -170,7 +170,7 @@ int mga_switchctx(struct inode *inode, struct file *filp, unsigned int cmd,
 	drm_ctx_t	ctx;
 
 	copy_from_user_ret(&ctx, (drm_ctx_t *)arg, sizeof(ctx), -EFAULT);
-	printk("%d\n", ctx.handle);
+	DRM_DEBUG("%d\n", ctx.handle);
 	return mga_context_switch(dev, dev->last_context, ctx.handle);
 }
 
@@ -182,7 +182,7 @@ int mga_newctx(struct inode *inode, struct file *filp, unsigned int cmd,
 	drm_ctx_t	ctx;
 
 	copy_from_user_ret(&ctx, (drm_ctx_t *)arg, sizeof(ctx), -EFAULT);
-	printk("%d\n", ctx.handle);
+	DRM_DEBUG("%d\n", ctx.handle);
 	mga_context_switch_complete(dev, ctx.handle);
 
 	return 0;
@@ -194,51 +194,11 @@ int mga_rmctx(struct inode *inode, struct file *filp, unsigned int cmd,
 	drm_file_t	*priv	= filp->private_data;
 	drm_device_t	*dev	= priv->dev;
 	drm_ctx_t	ctx;
-	drm_queue_t	*q;
-	drm_buf_t	*buf;
 
 	copy_from_user_ret(&ctx, (drm_ctx_t *)arg, sizeof(ctx), -EFAULT);
-	printk("%d\n", ctx.handle);
-	if(ctx.handle == DRM_KERNEL_CONTEXT) {
-	   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) */
-	   
-	   while (test_and_set_bit(0, &dev->interrupt_flag)) {
-	      printk("Calling schedule from rmctx\n");
-	      schedule();
-	      if (signal_pending(current)) {
-		 clear_bit(0, &dev->interrupt_flag);
-		 return -EINTR;
-	      }
-	   }
-	   
-	   /* Remove queued buffers */
-	   while ((buf = drm_waitlist_get(&q->waitlist))) {
-	      drm_free_buffer(dev, buf);
-	   }
-	   clear_bit(0, &dev->interrupt_flag);
-	   
-	   /* Wakeup blocked processes */
-	   wake_up_interruptible(&q->read_queue);
-	   wake_up_interruptible(&q->write_queue);
-	   wake_up_interruptible(&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);
-	} else {
-	   drm_ctxbitmap_free(dev, ctx.handle);
+	DRM_DEBUG("%d\n", ctx.handle);
+      	if(ctx.handle != DRM_KERNEL_CONTEXT) {
+	   	drm_ctxbitmap_free(dev, ctx.handle);
 	}
 	
 	return 0;
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 2e24e5b48..59cbad6b4 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
@@ -25,17 +25,15 @@
  *
  * Authors: Rickard E. (Rik) Faith <faith@precisioninsight.com>
  *	    Jeff Hartmann <jhartmann@precisioninsight.com>
+ *	    Keith Whitwell <keithw@precisioninsight.com>
  *
- * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_dma.c,v 1.1 2000/02/11 17:26:07 dawes Exp $
+ * $XFree86$
  *
  */
 
 #define __NO_VERSION__
 #include "drmP.h"
 #include "mga_drv.h"
-#include "mgareg_flags.h"
-#include "mga_dma.h"
-#include "mga_state.h"
 
 #include <linux/interrupt.h>	/* For task queue support */
 
@@ -48,643 +46,599 @@
 #define MGA_WRITE(reg,val) 	do { MGA_DEREF(reg) = val; } while (0)
 
 #define PDEA_pagpxfer_enable 	     0x2
-#define MGA_SYNC_TAG         	     0x423f4200
 
-typedef enum {
-	TT_GENERAL,
-	TT_BLIT,
-	TT_VECTOR,
-	TT_VERTEX
-} transferType_t;
+static int mga_flush_queue(drm_device_t *dev);
 
+static unsigned long mga_alloc_page(drm_device_t *dev)
+{
+	unsigned long address;
+   
+	address = __get_free_page(GFP_KERNEL);
+	if(address == 0UL) {
+		return 0;
+	}
+	atomic_inc(&mem_map[MAP_NR((void *) address)].count);
+	set_bit(PG_locked, &mem_map[MAP_NR((void *) address)].flags);
+   
+	return address;
+}
+
+static void mga_free_page(drm_device_t *dev, unsigned long page)
+{
+	if(page == 0UL) {
+		return;
+	}
+	atomic_dec(&mem_map[MAP_NR((void *) page)].count);
+	clear_bit(PG_locked, &mem_map[MAP_NR((void *) page)].flags);
+	wake_up(&mem_map[MAP_NR((void *) page)].wait);
+	free_page(page);
+	return;
+}
 
 static void mga_delay(void)
 {
    	return;
 }
 
-int mga_dma_cleanup(drm_device_t *dev)
+static void mga_flush_write_combine(void)
 {
-	if(dev->dev_private) {
-		drm_mga_private_t *dev_priv = 
-			(drm_mga_private_t *) dev->dev_private;
-      
-		if(dev_priv->ioremap) {
-			int temp = (dev_priv->warp_ucode_size + 
-				    dev_priv->primary_size + 
-				    PAGE_SIZE - 1) / PAGE_SIZE * PAGE_SIZE;
-
-			drm_ioremapfree((void *) dev_priv->ioremap, temp);
-		}
-
-		drm_free(dev->dev_private, sizeof(drm_mga_private_t), 
-			 DRM_MEM_DRIVER);
-		dev->dev_private = NULL;
-	}
-
-	return 0;
+   	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 mga_alloc_kernel_queue(drm_device_t *dev)
+/* These are two age tags that will never be sent to
+ * the hardware */
+#define MGA_BUF_USED 	0xffffffff
+#define MGA_BUF_FREE	0
+
+static void mga_freelist_debug(drm_mga_freelist_t *item)
 {
-	drm_queue_t *queue = NULL;
-				/* Allocate a new queue */
-	down(&dev->struct_sem);
-	
-   	if(dev->queue_count != 0) {
-	   /* Reseting the kernel context here is not
-	    * a race, since it can only happen when that
-	    * queue is empty.
-	    */
-	   queue = dev->queuelist[DRM_KERNEL_CONTEXT];
-	   printk("Kernel queue already allocated\n");
+   	if(item->buf != NULL) {
+     		DRM_DEBUG("buf index : %d\n", item->buf->idx);
 	} else {
-	   queue = drm_alloc(sizeof(*queue), DRM_MEM_QUEUES);
-	   if(!queue) {
-	      up(&dev->struct_sem);
-	      printk("out of memory\n");
-	      return -ENOMEM;
-	   }
-	   ++dev->queue_count;
-	   dev->queuelist = drm_alloc(sizeof(*dev->queuelist), 
-				      DRM_MEM_QUEUES);
-	   if(!dev->queuelist) {
-	      up(&dev->struct_sem);
-	      drm_free(queue, sizeof(*queue), DRM_MEM_QUEUES);
-	      printk("out of memory\n");
-	      return -ENOMEM;
-	   }  
+	   	DRM_DEBUG("Freelist head\n");
 	}
-	   
-	memset(queue, 0, sizeof(*queue));
-	atomic_set(&queue->use_count, 1);
-   	atomic_set(&queue->finalization,  0);
-	atomic_set(&queue->block_count,   0);
-	atomic_set(&queue->block_read,    0);
-	atomic_set(&queue->block_write,   0);
-	atomic_set(&queue->total_queued,  0);
-	atomic_set(&queue->total_flushed, 0);
-	atomic_set(&queue->total_locks,   0);
-
-	init_waitqueue_head(&queue->write_queue);
-	init_waitqueue_head(&queue->read_queue);
-	init_waitqueue_head(&queue->flush_queue);
-
-	queue->flags = 0;
-
-	drm_waitlist_create(&queue->waitlist, dev->dma->buf_count);
-   
-   	dev->queue_slots = 1;
-   	dev->queuelist[DRM_KERNEL_CONTEXT] = queue;
-   	dev->queue_count--;
-	
-	up(&dev->struct_sem);
-	printk("%d (new)\n", dev->queue_count - 1);
-	return DRM_KERNEL_CONTEXT;
+   	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_dma_initialize(drm_device_t *dev, drm_mga_init_t *init) {
-	drm_mga_private_t *dev_priv;
-	drm_map_t *prim_map = NULL;
-	drm_map_t *sarea_map = NULL;
-	int temp;
-
-
-	dev_priv = drm_alloc(sizeof(drm_mga_private_t), DRM_MEM_DRIVER);
-	if(dev_priv == NULL) return -ENOMEM;
-	dev->dev_private = (void *) dev_priv;
-
-	printk("dev_private\n");
-
-	memset(dev_priv, 0, sizeof(drm_mga_private_t));
-      	atomic_set(&dev_priv->pending_bufs, 0);
-
-	if((init->reserved_map_idx >= dev->map_count) ||
-	   (init->buffer_map_idx >= dev->map_count)) {
-		mga_dma_cleanup(dev);
-		printk("reserved_map or buffer_map are invalid\n");
-		return -EINVAL;
-	}
+static int mga_freelist_init(drm_device_t *dev)
+{
+      	drm_device_dma_t *dma = dev->dma;
+   	drm_buf_t *buf;
+   	drm_mga_buf_priv_t *buf_priv;
+      	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
+   	drm_mga_freelist_t *item;
+   	int i;
+   
+   	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));
+   	dev_priv->head->age = MGA_BUF_USED;
    
-	if(mga_alloc_kernel_queue(dev) != DRM_KERNEL_CONTEXT) {
-	   mga_dma_cleanup(dev);
-	   DRM_ERROR("Kernel context queue not present\n");
+   	for (i = 0; i < dma->buf_count; i++) {
+	   	buf = dma->buflist[ i ];
+	        buf_priv = buf->dev_private;
+		item = drm_alloc(sizeof(drm_mga_freelist_t),
+				 DRM_MEM_DRIVER);
+	   	if(item == NULL) return -ENOMEM;
+	   	memset(item, 0, sizeof(drm_mga_freelist_t));
+	  	item->age = MGA_BUF_FREE;
+	   	item->prev = dev_priv->head;
+	   	item->next = dev_priv->head->next;
+	   	if(dev_priv->head->next != NULL)
+			dev_priv->head->next->prev = item;
+	   	if(item->next == NULL) dev_priv->tail = item;
+	   	item->buf = buf;
+	   	buf_priv->my_freelist = item;
+		buf_priv->discard = 0;
+	   	dev_priv->head->next = item;
 	}
-
-	dev_priv->reserved_map_idx = init->reserved_map_idx;
-	dev_priv->buffer_map_idx = init->buffer_map_idx;
-	sarea_map = dev->maplist[0];
-	dev_priv->sarea_priv = (drm_mga_sarea_t *) 
-		((u8 *)sarea_map->handle + 
-		 init->sarea_priv_offset);
-	printk("sarea_priv\n");
-
-	/* Scale primary size to the next page */
-	dev_priv->primary_size = ((init->primary_size + PAGE_SIZE - 1) / 
-				  PAGE_SIZE) * PAGE_SIZE;
-	dev_priv->warp_ucode_size = init->warp_ucode_size;
-	dev_priv->chipset = init->chipset;
-	dev_priv->fbOffset = init->fbOffset;
-	dev_priv->backOffset = init->backOffset;
-	dev_priv->depthOffset = init->depthOffset;
-	dev_priv->textureOffset = init->textureOffset;
-	dev_priv->textureSize = init->textureSize;
-	dev_priv->cpp = init->cpp;
-	dev_priv->sgram = init->sgram;
-	dev_priv->stride = init->stride;
-
-	dev_priv->frontOrg = init->frontOrg;
-	dev_priv->backOrg = init->backOrg;
-	dev_priv->depthOrg = init->depthOrg;
-	dev_priv->mAccess = init->mAccess;
-	
    
-	printk("memcpy\n");
-	memcpy(&dev_priv->WarpIndex, &init->WarpIndex, 
-	       sizeof(mgaWarpIndex) * MGA_MAX_WARP_PIPES);
-	printk("memcpy done\n");
-	prim_map = dev->maplist[init->reserved_map_idx];
-	dev_priv->prim_phys_head = dev->agp->base + init->reserved_map_agpstart;
-	temp = init->warp_ucode_size + dev_priv->primary_size;
-	temp = ((temp + PAGE_SIZE - 1) / 
-		PAGE_SIZE) * PAGE_SIZE;
-	printk("temp : %x\n", temp);
-	printk("dev->agp->base: %lx\n", dev->agp->base);
-	printk("init->reserved_map_agpstart: %x\n", init->reserved_map_agpstart);
-
-
-	dev_priv->ioremap = drm_ioremap(dev->agp->base + init->reserved_map_agpstart, 
-					temp);
-	if(dev_priv->ioremap == NULL) {
-		printk("Ioremap failed\n");
-		mga_dma_cleanup(dev);
-		return -ENOMEM;
+   	item = dev_priv->head;
+   	while(item) {
+		mga_freelist_debug(item);
+		item = item->next;
 	}
-
-
-
-	dev_priv->prim_head = (u32 *)dev_priv->ioremap;
-	printk("dev_priv->prim_head : %p\n", dev_priv->prim_head);
-	dev_priv->current_dma_ptr = dev_priv->prim_head;
-	dev_priv->prim_num_dwords = 0;
-	dev_priv->prim_max_dwords = dev_priv->primary_size / 4;
+   	DRM_DEBUG("Head\n");
+   	mga_freelist_debug(dev_priv->head);
+   	DRM_DEBUG("Tail\n");
+   	mga_freelist_debug(dev_priv->tail);
    
-	printk("dma initialization\n");
-
-	/* Private is now filled in, initialize the hardware */
-	{
-		PRIMLOCALS;
-		PRIMRESET( dev_priv );
-		PRIMGETPTR( dev_priv );
-		PRIMOUTREG(MGAREG_DMAPAD, 0);
-		PRIMOUTREG(MGAREG_DMAPAD, 0);
-		PRIMOUTREG(MGAREG_DWGSYNC, 0);
-		PRIMOUTREG(MGAREG_SOFTRAP, 0);
-		PRIMADVANCE( dev_priv );
+   	return 0;
+}
 
-		/* Poll for the first buffer to insure that
-		 * the status register will be correct
-		 */
-	   	printk("phys_head : %lx\n", phys_head);
+static void mga_freelist_cleanup(drm_device_t *dev)
+{
+      	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
+   	drm_mga_freelist_t *item;
+   	drm_mga_freelist_t *prev;
+
+   	item = dev_priv->head;
+   	while(item) {
+	   	prev = item;
+	   	item = item->next;
+	   	drm_free(prev, sizeof(drm_mga_freelist_t), DRM_MEM_DRIVER);
+	}
    
-		MGA_WRITE(MGAREG_DWGSYNC, MGA_SYNC_TAG);
+   	dev_priv->head = dev_priv->tail = NULL;
+}
 
-		while(MGA_READ(MGAREG_DWGSYNC) != MGA_SYNC_TAG) {
-			int i;
-			for(i = 0 ; i < 4096; i++) mga_delay();
+/* Frees dispatch lock */
+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;
+
+	end = jiffies + (HZ*3);
+    	while(1) {
+		if(!test_and_set_bit(0, &dev_priv->dispatch_lock)) {
+			break;
 		}
-
-	   	MGA_WRITE(MGAREG_PRIMADDRESS, phys_head | TT_GENERAL);
-
-		MGA_WRITE(MGAREG_PRIMEND, ((phys_head + num_dwords * 4) | 
-					   PDEA_pagpxfer_enable));
-
-		while(MGA_READ(MGAREG_DWGSYNC) == MGA_SYNC_TAG) {
-			int i;
-			for(i = 0; i < 4096; i++) mga_delay();
+	   	if((signed)(end - jiffies) <= 0) {
+			DRM_ERROR("irqs: %d wanted %d\n", 
+				  atomic_read(&dev->total_irq), 
+				  atomic_read(&dma->total_lost));
+			DRM_ERROR("lockup\n"); 
+			goto out_nolock;
 		}
-
+		for (i = 0 ; i < 2000 ; i++) mga_delay();
 	}
-   
-	printk("dma init was successful\n");
-	return 0;
+	end = jiffies + (HZ*3);
+    	DRM_DEBUG("quiescent status : %x\n", MGA_READ(MGAREG_STATUS));
+    	while((MGA_READ(MGAREG_STATUS) & 0x00030001) != 0x00020000) {
+		if((signed)(end - jiffies) <= 0) {
+			DRM_ERROR("irqs: %d wanted %d\n", 
+				  atomic_read(&dev->total_irq), 
+				  atomic_read(&dma->total_lost));
+			DRM_ERROR("lockup\n"); 
+			goto out_status;
+		}
+		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);
+out_nolock:
 }
 
-int mga_dma_init(struct inode *inode, struct file *filp,
-		 unsigned int cmd, unsigned long arg)
+#define FREELIST_INITIAL (MGA_DMA_BUF_NR * 2)
+#define FREELIST_COMPARE(age) ((age >> 2))
+
+unsigned int mga_create_sync_tag(drm_device_t *dev)
 {
-	drm_file_t *priv = filp->private_data;
-	drm_device_t *dev = priv->dev;
-	drm_mga_init_t init;
+      	drm_mga_private_t *dev_priv = 
+     		(drm_mga_private_t *) dev->dev_private;
+   	unsigned int temp;
+      	drm_buf_t *buf;
+   	drm_mga_buf_priv_t *buf_priv;
+      	drm_device_dma_t *dma = dev->dma;
+   	int i;
    
-	copy_from_user_ret(&init, (drm_mga_init_t *)arg, sizeof(init), -EFAULT);
+   	dev_priv->sync_tag++;
    
-	switch(init.func) {
-	case MGA_INIT_DMA:
-		return mga_dma_initialize(dev, &init);
-	case MGA_CLEANUP_DMA:
-		return mga_dma_cleanup(dev);
+   	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;
 
-	return -EINVAL;
-}
-
-#define MGA_ILOAD_CMD (DC_opcod_iload | DC_atype_rpl |          	\
-		       DC_linear_linear | DC_bltmod_bfcol |       	\
-		       (0xC << DC_bop_SHIFT) | DC_sgnzero_enable |	\
-		       DC_shftzero_enable | DC_clipdis_enable)
+	dev_priv->sarea_priv->last_enqueue = temp;
 
-static void __mga_iload_small(drm_device_t *dev,
-			      drm_buf_t *buf,
-			      int use_agp) 
-{
-   	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 long address = (unsigned long)buf->bus_address;
-	int length = buf->used;
-	int y1 = buf_priv->boxes[0].y1;
-	int x1 = buf_priv->boxes[0].x1;
-	int y2 = buf_priv->boxes[0].y2;
-	int x2 = buf_priv->boxes[0].x2;
-   	int dstorg = buf_priv->ContextState[MGA_CTXREG_DSTORG];
-   	int maccess = buf_priv->ContextState[MGA_CTXREG_MACCESS];
-   	PRIMLOCALS;
-
-	PRIMRESET(dev_priv);		
-   	PRIMGETPTR(dev_priv);
-   
-   	PRIMOUTREG(MGAREG_DSTORG, dstorg | use_agp);
-   	PRIMOUTREG(MGAREG_MACCESS, maccess);
-   	PRIMOUTREG(MGAREG_PITCH, (1 << 15));
-   	PRIMOUTREG(MGAREG_YDST, y1 * (x2 - x1));   
-   	PRIMOUTREG(MGAREG_LEN, 1);
-   	PRIMOUTREG(MGAREG_FXBNDRY, ((x2 - x1) * (y2 - y1) - 1) << 16);
-   	PRIMOUTREG(MGAREG_AR0, (x2 - x1) * (y2 - y1) - 1);
-   	PRIMOUTREG(MGAREG_AR3, 0);
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_DWGCTL+MGAREG_MGA_EXEC, MGA_ILOAD_CMD);
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_SECADDRESS, address | TT_BLIT);
-   	PRIMOUTREG(MGAREG_SECEND, (address + length) | use_agp);
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_DWGSYNC, 0);
-   	PRIMOUTREG(MGAREG_SOFTRAP, 0);
-   	PRIMADVANCE(dev_priv);
-#if 0
-   	/* For now we need to set this in the ioctl */
-	sarea_priv->dirty |= MGASAREA_NEW_CONTEXT;
-#endif
-   	MGA_WRITE(MGAREG_DWGSYNC, MGA_SYNC_TAG);
-   	while(MGA_READ(MGAREG_DWGSYNC) != MGA_SYNC_TAG) ;
-
-      	MGA_WRITE(MGAREG_PRIMADDRESS, dev_priv->prim_phys_head | TT_GENERAL);
-	MGA_WRITE(MGAREG_PRIMEND, (phys_head + num_dwords * 4) | use_agp);   
+   	DRM_DEBUG("sync_tag : %x\n", temp);
+   	return temp;
 }
 
-static void __mga_iload_xy(drm_device_t *dev,
-			   drm_buf_t *buf,
-			   int use_agp) 
+/* Least recently used :
+ * These operations are not atomic b/c they are protected by the 
+ * hardware lock */
+
+drm_buf_t *mga_freelist_get(drm_device_t *dev)
 {
-      	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 long address = (unsigned long)buf->bus_address;
-   	int length = buf->used;
-	int y1 = buf_priv->boxes[0].y1;
-	int x1 = buf_priv->boxes[0].x1;
-	int y2 = buf_priv->boxes[0].y2;
-	int x2 = buf_priv->boxes[0].x2;
-   	int dstorg = buf_priv->ContextState[MGA_CTXREG_DSTORG];
-   	int maccess = buf_priv->ContextState[MGA_CTXREG_MACCESS];
-   	int pitch = buf_priv->ServerState[MGA_2DREG_PITCH];
-   	int width, height;
-   	int texperdword = 0;
-   	PRIMLOCALS;
-   
-   	width = (x2 - x1);
-   	height = (y2 - y1);
-   	switch((maccess & 0x00000003)) {
-	 	case 0:
-	   	texperdword = 4;
-      		break;
-	 	case 1:
-      		texperdword = 2;
-	   	break;
-    		case 2:
-      		texperdword = 1;
-      		break;
-    		default:
-      		DRM_ERROR("Invalid maccess value passed to __mga_iload_xy\n");
-	   	return;  
-	}
-   
-   	x2 = x1 + width;
-   	x2 = (x2 + (texperdword - 1)) & ~(texperdword - 1);
-   	x1 = (x1 + (texperdword - 1)) & ~(texperdword - 1);
-   	width = x2 - x1;
-   
-	PRIMRESET(dev_priv);		
-   	PRIMGETPTR(dev_priv);
-   	PRIMOUTREG(MGAREG_DSTORG, dstorg | use_agp);
-   	PRIMOUTREG(MGAREG_MACCESS, maccess);
-   	PRIMOUTREG(MGAREG_PITCH, pitch);
-   	PRIMOUTREG(MGAREG_YDSTLEN, (y1 << 16) | height);
-   
-   	PRIMOUTREG(MGAREG_FXBNDRY, ((x1+width-1) << 16) | x1);
-   	PRIMOUTREG(MGAREG_AR0, width * height - 1);
-      	PRIMOUTREG(MGAREG_AR3, 0 );
-   	PRIMOUTREG(MGAREG_DWGCTL+MGAREG_MGA_EXEC, MGA_ILOAD_CMD);
-	   	   
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_SECADDRESS, address | TT_BLIT);
-   	PRIMOUTREG(MGAREG_SECEND, (address + length) | use_agp);
-	   
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_DWGSYNC, 0);
-   	PRIMOUTREG(MGAREG_SOFTRAP, 0);
-   	PRIMADVANCE(dev_priv);
-#if 0
-   	/* For now we need to set this in the ioctl */
-	sarea_priv->dirty |= MGASAREA_NEW_CONTEXT;
-#endif
-   	MGA_WRITE(MGAREG_DWGSYNC, MGA_SYNC_TAG);
-   	while(MGA_READ(MGAREG_DWGSYNC) != MGA_SYNC_TAG) ;
-
-      	MGA_WRITE(MGAREG_PRIMADDRESS, dev_priv->prim_phys_head | TT_GENERAL);
-	MGA_WRITE(MGAREG_PRIMEND, (phys_head + num_dwords * 4) | use_agp);
+   	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;
+   
+   	if((dev_priv->tail->age >> 2) <= FREELIST_COMPARE(status[1])) {
+		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;
+	   	return next->buf;
+	} 
+
+   	return NULL;
 }
 
-static void mga_dma_dispatch_iload(drm_device_t *dev, drm_buf_t *buf)
+int mga_freelist_put(drm_device_t *dev, drm_buf_t *buf)
 {
+      	drm_mga_private_t *dev_priv = 
+     		(drm_mga_private_t *) dev->dev_private;
    	drm_mga_buf_priv_t *buf_priv = buf->dev_private;
-
-	int use_agp = PDEA_pagpxfer_enable;
-	int x1 = buf_priv->boxes[0].x1;
-	int x2 = buf_priv->boxes[0].x2;
-   
-   	if((x2 - x1) < 32) {
-	   	printk("using iload small\n");
-	   	__mga_iload_small(dev, buf, use_agp);
+	drm_mga_freelist_t *prev;
+   	drm_mga_freelist_t *head;
+   	drm_mga_freelist_t *next;
+
+   	if(buf_priv->my_freelist->age == MGA_BUF_USED) {
+		/* Discarded buffer, put it on the tail */
+		next = buf_priv->my_freelist;
+		next->age = MGA_BUF_FREE;
+		prev = dev_priv->tail;
+		prev->next = next;
+		next->prev = prev;
+		next->next = NULL;
+		dev_priv->tail = next;
+		DRM_DEBUG("Discarded\n");
 	} else {
-	   	printk("using iload xy\n");
-	   	__mga_iload_xy(dev, buf, use_agp); 
-	}   
+		/* Normally aged buffer, put it on the head + 1,
+		 * as the real head is a sentinal element
+		 */
+		next = buf_priv->my_freelist;
+		head = dev_priv->head;
+		prev = head->next;
+		head->next = next;
+		prev->prev = next;
+		next->prev = head;
+		next->next = prev;
+	}
+   
+   	return 0;
 }
 
-static void mga_dma_dispatch_vertex(drm_device_t *dev, drm_buf_t *buf)
+static void mga_print_all_primary(drm_device_t *dev)
 {
-   	drm_mga_private_t *dev_priv = dev->dev_private;
-	drm_mga_buf_priv_t *buf_priv = buf->dev_private;
-	unsigned long address = (unsigned long)buf->bus_address;
-	int length = buf->used;
-	int use_agp = PDEA_pagpxfer_enable;
-	int i, count;
-   	PRIMLOCALS;
-
-	PRIMRESET(dev_priv);
-	
-	count = buf_priv->nbox;
-	if (count == 0) 
-		count = 1;
-
-	mgaEmitState( dev_priv, buf_priv );
-
-	for (i = 0 ; i < count ; i++) {		
-		if (i < buf_priv->nbox)
-			mgaEmitClipRect( dev_priv, &buf_priv->boxes[i] );
-
-		PRIMGETPTR(dev_priv);
-		PRIMOUTREG( MGAREG_DMAPAD, 0);
-		PRIMOUTREG( MGAREG_DMAPAD, 0);
-		PRIMOUTREG( MGAREG_SECADDRESS, address | TT_VERTEX);
-		PRIMOUTREG( MGAREG_SECEND, (address + length) | use_agp);
-
-		PRIMOUTREG( MGAREG_DMAPAD, 0);
-		PRIMOUTREG( MGAREG_DMAPAD, 0);
-		PRIMOUTREG( MGAREG_DWGSYNC, 0);
-		PRIMOUTREG( MGAREG_SOFTRAP, 0);
-		PRIMADVANCE(dev_priv);
+      	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));
 	}
-
-	PRIMGETPTR( dev_priv );
-
-   	MGA_WRITE(MGAREG_DWGSYNC, MGA_SYNC_TAG);
-   	while(MGA_READ(MGAREG_DWGSYNC) != MGA_SYNC_TAG) ;
-
-      	MGA_WRITE(MGAREG_PRIMADDRESS, dev_priv->prim_phys_head | TT_GENERAL);
-	MGA_WRITE(MGAREG_PRIMEND, (phys_head + num_dwords * 4) | use_agp);
+   
+   	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);
 }
 
-
-/* Used internally for the small buffers generated from client state
- * information. 
- */
-static void mga_dma_dispatch_general(drm_device_t *dev, drm_buf_t *buf)
+static int mga_init_primary_bufs(drm_device_t *dev, drm_mga_init_t *init)
 {
    	drm_mga_private_t *dev_priv = dev->dev_private;
-	unsigned long address = (unsigned long)buf->bus_address;
-	int length = buf->used;
-	int use_agp = PDEA_pagpxfer_enable;
-   	PRIMLOCALS;
-
-	PRIMRESET(dev_priv);
-   	PRIMGETPTR(dev_priv);
-
-	PRIMOUTREG( MGAREG_DMAPAD, 0);
-	PRIMOUTREG( MGAREG_DMAPAD, 0);
-	PRIMOUTREG( MGAREG_SECADDRESS, address | TT_GENERAL);
-	PRIMOUTREG( MGAREG_SECEND, (address + length) | use_agp);
-
-	PRIMOUTREG( MGAREG_DMAPAD, 0);
-	PRIMOUTREG( MGAREG_DMAPAD, 0);
-      	PRIMOUTREG( MGAREG_DWGSYNC, 0);
-   	PRIMOUTREG( MGAREG_SOFTRAP, 0);
-   	PRIMADVANCE(dev_priv);
-
-   	MGA_WRITE(MGAREG_DWGSYNC, MGA_SYNC_TAG);
-   	while(MGA_READ(MGAREG_DWGSYNC) != MGA_SYNC_TAG) ;
-
-      	MGA_WRITE(MGAREG_PRIMADDRESS, dev_priv->prim_phys_head | TT_GENERAL);
-	MGA_WRITE(MGAREG_PRIMEND, (phys_head + num_dwords * 4) | use_agp);
-}
+	drm_mga_prim_buf_t *prim_buffer;
+   	int i, temp, size_of_buf;
+   	int offset = init->reserved_map_agpstart;
 
-/* Frees dispatch lock */
-static inline void mga_dma_quiescent(drm_device_t *dev)
-{
-   	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
-
-   	while(1) {
-	   atomic_inc(&dev_priv->dispatch_lock);
-	   if(atomic_read(&dev_priv->dispatch_lock) == 1) {
-	      break;
-	   } else {
-	      atomic_dec(&dev_priv->dispatch_lock);
-	   }
-	}
-	while((MGA_READ(MGAREG_STATUS) & 0x00020001) != 0x00020000) ;
-#if 0
-   MGA_WRITE(MGAREG_DWGSYNC, MGA_SYNC_TAG);
-#endif
-   	while(MGA_READ(MGAREG_DWGSYNC) == MGA_SYNC_TAG) ;
-   	MGA_WRITE(MGAREG_DWGSYNC, MGA_SYNC_TAG);
-	while(MGA_READ(MGAREG_DWGSYNC) != MGA_SYNC_TAG) ;
-   	atomic_dec(&dev_priv->dispatch_lock);
+   	DRM_DEBUG("mga_init_primary_bufs\n");
+   	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 *) * 
+					(MGA_NUM_PRIM_BUFS + 1), 
+					DRM_MEM_DRIVER);
+   	if(dev_priv->prim_bufs == NULL) {
+		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) {
+		DRM_DEBUG("Ioremap failed\n");
+		return -ENOMEM;
+	}
+   	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 = 
+			prim_buffer->head = 
+			(u32 *) (dev_priv->ioremap + 
+				 offset - 
+				 init->reserved_map_agpstart);
+	   	prim_buffer->num_dwords = 0;
+	   	prim_buffer->max_dwords = size_of_buf / sizeof(u32);
+	   	prim_buffer->max_dwords -= 5; /* Leave room for the softrap */
+	   	prim_buffer->sec_used = 0;
+	   	prim_buffer->idx = i;
+	   	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");
+   	return 0;
 }
 
-/* Keeps dispatch lock held */
-
-static inline int mga_dma_is_ready(drm_device_t *dev)
+void mga_fire_primary(drm_device_t *dev, drm_mga_prim_buf_t *prim)
 {
-	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
+       	drm_mga_private_t *dev_priv = dev->dev_private;
+      	drm_device_dma_t  *dma	    = dev->dma;
+       	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+ 	int use_agp = PDEA_pagpxfer_enable;
+	unsigned long end;
+   	int i;
+   	int next_idx;
+       	PRIMLOCALS;
+   
+    	DRM_DEBUG("mga_fire_primary\n");
+    	dev_priv->last_sync_tag = mga_create_sync_tag(dev); 
+   	dev_priv->last_prim = prim;
    
-	atomic_inc(&dev_priv->dispatch_lock);
-	if(atomic_read(&dev_priv->dispatch_lock) == 1) {
-		/* We got the lock */
-		return 1;
+ 	/* We never check for overflow, b/c there is always room */
+    	PRIMPTR(prim);
+   	if(num_dwords <= 0) {
+		DRM_DEBUG("num_dwords == 0 when dispatched\n");
+		goto out_prim_wait;
+	}
+ 	PRIMOUTREG( MGAREG_DMAPAD, 0);
+ 	PRIMOUTREG( MGAREG_DMAPAD, 0);
+       	PRIMOUTREG( MGAREG_DWGSYNC, dev_priv->last_sync_tag);
+   	PRIMOUTREG( MGAREG_SOFTRAP, 0);
+    	PRIMFINISH(prim);
+
+	end = jiffies + (HZ*3);
+    	if(sarea_priv->dirty & MGA_DMA_FLUSH) {
+		DRM_DEBUG("Dma top flush\n");	   
+		while((MGA_READ(MGAREG_STATUS) & 0x00030001) != 0x00020000) {
+			if((signed)(end - jiffies) <= 0) {
+				DRM_ERROR("irqs: %d wanted %d\n", 
+					  atomic_read(&dev->total_irq), 
+					  atomic_read(&dma->total_lost));
+				DRM_ERROR("lockup in fire primary "
+					  "(Dma Top Flush)\n");
+				goto out_prim_wait;
+			}
+	      
+			for (i = 0 ; i < 4096 ; i++) mga_delay();
+		}
+		sarea_priv->dirty &= ~(MGA_DMA_FLUSH);
 	} else {
-		atomic_dec(&dev_priv->dispatch_lock);
-		return 0;
+		DRM_DEBUG("Status wait\n");
+		while((MGA_READ(MGAREG_STATUS) & 0x00020001) != 0x00020000) {
+			if((signed)(end - jiffies) <= 0) {
+				DRM_ERROR("irqs: %d wanted %d\n", 
+					  atomic_read(&dev->total_irq), 
+					  atomic_read(&dma->total_lost));
+				DRM_ERROR("lockup in fire primary "
+					  "(Status Wait)\n");
+				goto out_prim_wait;
+			}
+	   
+			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;
+    
+   	next_idx = prim->idx + 1;
+    	if(next_idx >= MGA_NUM_PRIM_BUFS) 
+		next_idx = 0;
+
+    	dev_priv->next_prim = dev_priv->prim_bufs[next_idx];
+	return;
+
+ out_prim_wait:
+	prim->num_dwords = 0;
+	prim->sec_used = 0;
+	clear_bit(0, &prim->in_use);
+   	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);
 }
 
-static inline int mga_dma_is_ready_no_hold(drm_device_t *dev)
+int mga_advance_primary(drm_device_t *dev)
 {
-	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
+   	DECLARE_WAITQUEUE(entry, current);
+   	drm_mga_private_t *dev_priv = dev->dev_private;
+   	drm_mga_prim_buf_t *prim_buffer;
+   	drm_device_dma_t  *dma      = dev->dma;
+   	int next_prim_idx;
+   	int ret = 0;
    
-	atomic_inc(&dev_priv->dispatch_lock);
-	if(atomic_read(&dev_priv->dispatch_lock) == 1) {
-		/* We got the lock, but free it */
-		atomic_dec(&dev_priv->dispatch_lock);
-		return 1;
-	} else {
-		atomic_dec(&dev_priv->dispatch_lock);
-		return 0;
+   	/* This needs to reset the primary buffer if available,
+	 * we should collect stats on how many times it bites
+	 * it's tail */
+   
+   	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);
+   
+      	/* In use is cleared in interrupt handler */
+   
+   	if(test_and_set_bit(0, &prim_buffer->in_use)) {
+	   	add_wait_queue(&dev_priv->wait_queue, &entry);
+	   	for (;;) {
+		   	current->state = TASK_INTERRUPTIBLE;
+		   	mga_dma_schedule(dev, 0);
+		   	if(!test_and_set_bit(0, &prim_buffer->in_use)) 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);
+		   	if (signal_pending(current)) {
+			   	ret = -ERESTARTSYS;
+			   	break;
+			}
+		}
+	   	current->state = TASK_RUNNING;
+	   	remove_wait_queue(&dev_priv->wait_queue, &entry);
+	   	if(ret) return ret;
 	}
+   	atomic_set(&dev_priv->in_wait, 0);
+   	/* 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);
+   	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;
 }
 
-static void mga_dma_service(int irq, void *device, struct pt_regs *regs)
+/* More dynamic performance decisions */
+static inline int mga_decide_to_fire(drm_device_t *dev)
 {
-	drm_device_t	 *dev = (drm_device_t *)device;
-	drm_device_dma_t *dma = dev->dma;
    	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))
+	{
+	   	atomic_inc(&dma->total_prio);
+	   	return 1;
+	}
 
-	atomic_dec(&dev_priv->dispatch_lock);   	
-	atomic_inc(&dev->total_irq);
-      	MGA_WRITE(MGAREG_ICLEAR, 0xfa7);
-
-				/* Free previous buffer */
-   	if (test_and_set_bit(0, &dev->dma_flag)) {
-	   	atomic_inc(&dma->total_missed_free);
-	   	return;
+	if (atomic_read(&dev_priv->in_flush) && dev_priv->next_prim->num_dwords)
+	{
+	   	atomic_inc(&dma->total_prio);
+	   	return 1;
 	}
-   	if (dma->this_buffer) {
-	   	drm_free_buffer(dev, dma->this_buffer);
-	   	dma->this_buffer = NULL;
+   
+   	if(atomic_read(&dev_priv->pending_bufs) <= MGA_NUM_PRIM_BUFS - 1) {
+		if(test_bit(0, &dev_priv->next_prim->swap_pending)) {
+			atomic_inc(&dma->total_dmas);
+			return 1;
+		}
 	}
-   	clear_bit(0, &dev->dma_flag);
 
-				/* Dispatch new buffer */
-   	queue_task(&dev->tq, &tq_immediate);
-   	mark_bh(IMMEDIATE_BH);
+   	if(atomic_read(&dev_priv->pending_bufs) <= MGA_NUM_PRIM_BUFS / 2) {
+		if(dev_priv->next_prim->sec_used >= MGA_DMA_BUF_NR / 8) {
+			atomic_inc(&dma->total_hit);
+			return 1;
+		}
+	}
 
+   	if(atomic_read(&dev_priv->pending_bufs) >= MGA_NUM_PRIM_BUFS / 2) {
+		if(dev_priv->next_prim->sec_used >= MGA_DMA_BUF_NR / 4) {
+			atomic_inc(&dma->total_missed_free);
+			return 1;
+		}
+	}
+
+   	atomic_inc(&dma->total_tried);
+   	return 0;
 }
 
-/* Only called by mga_dma_schedule. */
-static int mga_do_dma(drm_device_t *dev, int locked)
+int mga_dma_schedule(drm_device_t *dev, int locked)
 {
-	drm_buf_t	 *buf;
-	int		 retcode = 0;
-	drm_device_dma_t *dma = dev->dma;
       	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
-	drm_mga_buf_priv_t *buf_priv;
+      	drm_device_dma_t  *dma	    = dev->dma;
 
-   	printk("mga_do_dma\n");
-	if (test_and_set_bit(0, &dev->dma_flag)) {
+   	if (test_and_set_bit(0, &dev->dma_flag)) {
 		atomic_inc(&dma->total_missed_dma);
 		return -EBUSY;
 	}
-	
-	if (!dma->next_buffer) {
-		DRM_ERROR("No next_buffer\n");
-		clear_bit(0, &dev->dma_flag);
-		return -EINVAL;
-	}
-
-	buf	= dma->next_buffer;
-
-	printk("context %d, buffer %d\n", buf->context, buf->idx);
-
-	if (buf->list == DRM_LIST_RECLAIM) {
-		drm_clear_next_buffer(dev);
-		drm_free_buffer(dev, buf);
-		clear_bit(0, &dev->dma_flag);
-		return -EINVAL;
-	}
+   
+       	DRM_DEBUG("mga_dma_schedule\n");
 
-	if (!buf->used) {
-		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 (mga_dma_is_ready(dev) == 0) {
-		clear_bit(0, &dev->dma_flag);
-		return -EBUSY;
+   	if(atomic_read(&dev_priv->in_flush) || 
+	   atomic_read(&dev_priv->in_wait)) {
+		locked = 1;
 	}
    
-	/* Always hold the hardware lock while dispatching.
-	 */
-	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);
-		atomic_dec(&dev_priv->dispatch_lock);
-		return -EBUSY;
+   	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;
 	}
+   	DRM_DEBUG("I'm locked\n");
 
-   	dma->next_queue	 = dev->queuelist[DRM_KERNEL_CONTEXT];
-	drm_clear_next_buffer(dev);
-	buf->pending	 = 1;
-	buf->waiting	 = 0;
-	buf->list	 = DRM_LIST_PEND;
-
-	buf_priv = buf->dev_private;
-
-   	printk("dispatch!\n");
-	switch (buf_priv->dma_type) {
-	case MGA_DMA_GENERAL:
-		mga_dma_dispatch_general(dev, buf);
-		break;
-	case MGA_DMA_VERTEX:
-		mga_dma_dispatch_vertex(dev, buf);
-		break;
-/*  	case MGA_DMA_SETUP: */
-/*  		mga_dma_dispatch_setup(dev, address, length); */
-/*  		break; */
-	case MGA_DMA_ILOAD:
-		mga_dma_dispatch_iload(dev, buf);
-		break;
-	default:
-		printk("bad buffer type %x in dispatch\n", buf_priv->dma_type);
-		break;
+   
+   	if(!test_and_set_bit(0, &dev_priv->dispatch_lock)) {
+	   	/* 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) {
+				/* Schedule overflow for a later time */
+				atomic_set(
+					&dev_priv->current_prim->needs_overflow,
+					1);
+			}
+		   	mga_fire_primary(dev, dev_priv->next_prim);
+		} else {
+			clear_bit(0, &dev_priv->dispatch_lock);
+		}
+	} else {
+		DRM_DEBUG("I can't get the dispatch lock\n");
 	}
-   	atomic_dec(&dev_priv->pending_bufs);
-
-	drm_free_buffer(dev, dma->this_buffer);
-	dma->this_buffer = buf;
-
-	atomic_add(buf->used, &dma->total_bytes);
-	atomic_inc(&dma->total_dmas);
-
+   	
 	if (!locked) {
 		if (drm_lock_free(dev, &dev->lock.hw_lock->lock,
 				  DRM_KERNEL_CONTEXT)) {
@@ -692,90 +646,251 @@ static int mga_do_dma(drm_device_t *dev, int locked)
 		}
 	}
 
-	clear_bit(0, &dev->dma_flag);
-   
-   if(!atomic_read(&dev_priv->pending_bufs)) {
-      wake_up_interruptible(&dev->queuelist[DRM_KERNEL_CONTEXT]->flush_queue);
-   }
-   
-#if 0
-   wake_up_interruptible(&dev->lock.lock_queue);
-#endif
+   	clear_bit(0, &dev->dma_flag);
    
-	/* We hold the dispatch lock until the interrupt handler
-	 * frees it
-	 */
-	return retcode;
+      	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);
+	   	wake_up_interruptible(&dev_priv->flush_queue);
+	}
+
+	return 0;
 }
 
-static void mga_dma_schedule_timer_wrapper(unsigned long dev)
+static void mga_dma_service(int irq, void *device, struct pt_regs *regs)
 {
-	mga_dma_schedule((drm_device_t *)dev, 0);
+    	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;
+    
+    	atomic_inc(&dev->total_irq);
+      	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);
+   	wake_up_interruptible(&dev_priv->wait_queue);
+      	clear_bit(0, &last_prim_buffer->swap_pending);
+      	clear_bit(0, &dev_priv->dispatch_lock);
+      	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_schedule_tq_wrapper(void *dev)
+static void mga_dma_task_queue(void *device)
 {
+   	drm_device_t *dev = (drm_device_t *) device;
+
 	mga_dma_schedule(dev, 0);
 }
 
-int mga_dma_schedule(drm_device_t *dev, int locked)
+int mga_dma_cleanup(drm_device_t *dev)
 {
-	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;
-
-      	printk("mga_dma_schedule\n");
-
-	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(dev->dev_private) {
+		drm_mga_private_t *dev_priv = 
+			(drm_mga_private_t *) dev->dev_private;
+      
+		if(dev_priv->ioremap) {
+			int temp = (dev_priv->warp_ucode_size + 
+				    dev_priv->primary_size + 
+				    PAGE_SIZE - 1) / PAGE_SIZE * PAGE_SIZE;
 
-again:
-	/* There is only one queue:
-	 */
-	if (!dma->next_buffer && DRM_WAITCOUNT(dev, DRM_KERNEL_CONTEXT)) {
-		q   = dev->queuelist[DRM_KERNEL_CONTEXT];
-		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);
+			drm_ioremapfree((void *) dev_priv->ioremap, temp);
+		}
+	   	if(dev_priv->status_page != NULL) {
+		   	iounmap(dev_priv->status_page);
+		}
+	   	if(dev_priv->real_status_page != 0UL) {
+		   	mga_free_page(dev, dev_priv->real_status_page);
+		}
+	   	if(dev_priv->prim_bufs != NULL) {
+		   	int i;
+		   	for(i = 0; i < MGA_NUM_PRIM_BUFS; i++) {
+			   	if(dev_priv->prim_bufs[i] != NULL) {
+			     		drm_free(dev_priv->prim_bufs[i],
+						 sizeof(drm_mga_prim_buf_t),
+						 DRM_MEM_DRIVER);
+				}
+			}
+		   	drm_free(dev_priv->prim_bufs, sizeof(void *) *
+				 (MGA_NUM_PRIM_BUFS + 1), 
+				 DRM_MEM_DRIVER);
+		}
+		if(dev_priv->head != NULL) {
+		   	mga_freelist_cleanup(dev);
 		}
+
+
+		drm_free(dev->dev_private, sizeof(drm_mga_private_t), 
+			 DRM_MEM_DRIVER);
+		dev->dev_private = NULL;
 	}
 
-	if (dma->next_buffer) {
-		if (!(retcode = mga_do_dma(dev, locked))) 
-			++processed;
+	return 0;
+}
+
+static int mga_dma_initialize(drm_device_t *dev, drm_mga_init_t *init) {
+	drm_mga_private_t *dev_priv;
+	drm_map_t *sarea_map = NULL;
+	int i;
+
+	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)) {
+		mga_dma_cleanup(dev);
+		DRM_DEBUG("reserved_map or buffer_map are invalid\n");
+		return -EINVAL;
 	}
+   
+	dev_priv->reserved_map_idx = init->reserved_map_idx;
+	dev_priv->buffer_map_idx = init->buffer_map_idx;
+	sarea_map = dev->maplist[0];
+	dev_priv->sarea_priv = (drm_mga_sarea_t *) 
+		((u8 *)sarea_map->handle + 
+		 init->sarea_priv_offset);
+	DRM_DEBUG("sarea_priv\n");
 
-	/* Try again if we succesfully dispatched a buffer, or if someone 
-	 * tried to schedule while we were working.
-	 */
-	if (--expire) {
-		if (missed != atomic_read(&dma->total_missed_sched)) {
-			atomic_inc(&dma->total_lost);
-		   	if (mga_dma_is_ready_no_hold(dev)) 
-				goto again;
-		}
+	/* Scale primary size to the next page */
+	dev_priv->chipset = init->chipset;
+	dev_priv->frontOffset = init->frontOffset;
+	dev_priv->backOffset = init->backOffset;
+	dev_priv->depthOffset = init->depthOffset;
+	dev_priv->textureOffset = init->textureOffset;
+	dev_priv->textureSize = init->textureSize;
+	dev_priv->cpp = init->cpp;
+	dev_priv->sgram = init->sgram;
+	dev_priv->stride = init->stride;
 
-		if (processed && mga_dma_is_ready_no_hold(dev)) {
-			atomic_inc(&dma->total_lost);
-			processed = 0;
-			goto again;
-		}
+	dev_priv->mAccess = init->mAccess;
+   	init_waitqueue_head(&dev_priv->flush_queue);
+	dev_priv->WarpPipe = -1;
+
+   	DRM_DEBUG("chipset: %d ucode_size: %d backOffset: %x depthOffset: %x\n",
+		  dev_priv->chipset, dev_priv->warp_ucode_size, 
+		  dev_priv->backOffset, dev_priv->depthOffset);
+   	DRM_DEBUG("cpp: %d sgram: %d stride: %d maccess: %x\n",
+		  dev_priv->cpp, dev_priv->sgram, dev_priv->stride, 
+		  dev_priv->mAccess);
+   
+	memcpy(&dev_priv->WarpIndex, &init->WarpIndex, 
+	       sizeof(drm_mga_warp_index_t) * MGA_MAX_WARP_PIPES);
+
+   	for (i = 0 ; i < MGA_MAX_WARP_PIPES ; i++) 
+		DRM_DEBUG("warp pipe %d: installed: %d phys: %lx size: %x\n",
+			  i, 
+			  dev_priv->WarpIndex[i].installed,
+			  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;
 	}
-	
-	clear_bit(0, &dev->interrupt_flag);
-	
-	return retcode;
+   	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),
+				PAGE_SIZE);
+
+   	if(dev_priv->status_page == NULL) {
+		mga_dma_cleanup(dev);
+		DRM_ERROR("Can not remap status page\n");
+		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_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);
+
+		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]);
+	}
+
+	if(mga_freelist_init(dev) != 0) {
+	   	DRM_ERROR("Could not initialize freelist\n");
+	   	mga_dma_cleanup(dev);
+	   	return -ENOMEM;
+	}
+   	DRM_DEBUG("dma init was successful\n");
+	return 0;
+}
+
+int mga_dma_init(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_init_t init;
+   
+	copy_from_user_ret(&init, (drm_mga_init_t *)arg, sizeof(init), -EFAULT);
+   
+	switch(init.func) {
+	case MGA_INIT_DMA:
+		return mga_dma_initialize(dev, &init);
+	case MGA_CLEANUP_DMA:
+		return mga_dma_cleanup(dev);
+	}
+
+	return -EINVAL;
 }
 
 int mga_irq_install(drm_device_t *dev, int irq)
@@ -792,7 +907,7 @@ int mga_irq_install(drm_device_t *dev, int irq)
 	dev->irq = irq;
 	up(&dev->struct_sem);
 	
-	printk("install irq handler %d\n", irq);
+	DRM_DEBUG("install irq handler %d\n", irq);
 
 	dev->context_flag     = 0;
 	dev->interrupt_flag   = 0;
@@ -802,13 +917,11 @@ int mga_irq_install(drm_device_t *dev, int irq)
 	dev->dma->this_buffer = NULL;
 	dev->tq.next	      = NULL;
 	dev->tq.sync	      = 0;
-	dev->tq.routine	      = mga_dma_schedule_tq_wrapper;
+	dev->tq.routine	      = mga_dma_task_queue;
 	dev->tq.data	      = dev;
 
 				/* Before installing handler */
-	MGA_WRITE(MGAREG_ICLEAR, 0xfa7);
 	MGA_WRITE(MGAREG_IEN, 0);
-
    				/* Install handler */
 	if ((retcode = request_irq(dev->irq,
 				   mga_dma_service,
@@ -820,11 +933,9 @@ int mga_irq_install(drm_device_t *dev, int irq)
 		up(&dev->struct_sem);
 		return retcode;
 	}
-
 				/* After installing handler */
-   	MGA_WRITE(MGAREG_ICLEAR, 0xfa7);
+   	MGA_WRITE(MGAREG_ICLEAR, 0x00000001);
 	MGA_WRITE(MGAREG_IEN, 0x00000001);
-
 	return 0;
 }
 
@@ -838,19 +949,13 @@ int mga_irq_uninstall(drm_device_t *dev)
 	up(&dev->struct_sem);
 	
 	if (!irq) return -EINVAL;
-	
-	printk("remove irq handler %d\n", irq);
-
-      	MGA_WRITE(MGAREG_ICLEAR, 0xfa7);
+   	DRM_DEBUG("remove irq handler %d\n", irq);
+      	MGA_WRITE(MGAREG_ICLEAR, 0x00000001);
 	MGA_WRITE(MGAREG_IEN, 0);
-   	MGA_WRITE(MGAREG_ICLEAR, 0xfa7);
-
 	free_irq(irq, dev);
-
 	return 0;
 }
 
-
 int mga_control(struct inode *inode, struct file *filp, unsigned int cmd,
 		  unsigned long arg)
 {
@@ -870,36 +975,64 @@ int mga_control(struct inode *inode, struct file *filp, unsigned int cmd,
 	}
 }
 
-int mga_flush_queue(drm_device_t *dev)
+static int mga_flush_queue(drm_device_t *dev)
 {
    	DECLARE_WAITQUEUE(entry, current);
-	drm_queue_t	  *q = dev->queuelist[DRM_KERNEL_CONTEXT];
-   	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
+  	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
    	int ret = 0;
-   	
-   	printk("mga_flush_queue\n");
-   	if(atomic_read(&dev_priv->pending_bufs) != 0) {
-	   current->state = TASK_INTERRUPTIBLE;
-	   add_wait_queue(&q->flush_queue, &entry);
-	   for (;;) {
-	      	if (!atomic_read(&dev_priv->pending_bufs)) break;
-	      	printk("Calling schedule from flush_queue : %d\n",
-		       atomic_read(&dev_priv->pending_bufs));
-	      	mga_dma_schedule(dev, 1);
-	      	schedule();
-	      	if (signal_pending(current)) {
-		   	ret = -EINTR; /* Can't restart */
-		   	break;
-		}
-	   }
-	   printk("Exited out of schedule from flush_queue\n");
-	   current->state = TASK_RUNNING;
-	   remove_wait_queue(&q->flush_queue, &entry);
+
+   	if(dev_priv == NULL) {
+	   	return 0;
 	}
    
+   	if(dev_priv->next_prim->num_dwords != 0) {
+	   	atomic_set(&dev_priv->in_flush, 1);
+   		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) 
+				break;
+		   	atomic_inc(&dev->total_sleeps);
+		   	DRM_DEBUG("Schedule in flush_queue\n");
+	      		schedule_timeout(HZ*3);
+	      		if (signal_pending(current)) {
+		   		ret = -EINTR; /* Can't restart */
+		   		break;
+			}
+		}
+   		current->state = TASK_RUNNING;
+   		remove_wait_queue(&dev_priv->flush_queue, &entry);
+	}
+   	atomic_set(&dev_priv->in_flush, 0);
    	return ret;
 }
 
+/* Must be called with the lock held */
+void mga_reclaim_buffers(drm_device_t *dev, pid_t pid)
+{
+	drm_device_dma_t *dma = dev->dma;
+	int		 i;
+
+	if (!dma) return;
+      	if(dev->dev_private == NULL) return;
+
+        mga_flush_queue(dev);
+
+	for (i = 0; i < dma->buf_count; i++) {
+	   	drm_buf_t *buf = dma->buflist[ i ];
+	   	drm_mga_buf_priv_t *buf_priv = buf->dev_private;
+
+		if (buf->pid == pid) {
+		   	if(buf_priv == NULL) return;
+		   	/* Only buffers that need to get reclaimed ever 
+			 * get set to free */
+			if(buf_priv->my_freelist->age == MGA_BUF_USED) 
+		     		buf_priv->my_freelist->age = MGA_BUF_FREE;
+		}
+	}
+}
+
 int mga_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 	       unsigned long arg)
 {
@@ -916,16 +1049,15 @@ int mga_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 			  current->pid, lock.context);
 		return -EINVAL;
 	}
-
-   	printk("%d (pid %d) requests lock (0x%08x), flags = 0x%08x\n",
-		  lock.context, current->pid, dev->lock.hw_lock->lock,
-		  lock.flags);
-
+   
+   	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 (lock.context < 0) {
 		return -EINVAL;
 	}
-
+   
 	/* Only one queue:
 	 */
 
@@ -948,8 +1080,7 @@ int mga_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 				/* Contention */
 			atomic_inc(&dev->total_sleeps);
 			current->state = TASK_INTERRUPTIBLE;
-		   	current->policy |= SCHED_YIELD;
-		   	printk("Calling lock schedule\n");
+		   	DRM_DEBUG("Calling lock schedule\n");
 			schedule();
 			if (signal_pending(current)) {
 				ret = -ERESTARTSYS;
@@ -962,17 +1093,60 @@ int mga_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 	
 	if (!ret) {
 		if (lock.flags & _DRM_LOCK_QUIESCENT) {
-		   	printk("_DRM_LOCK_QUIESCENT\n");
-		   	ret = mga_flush_queue(dev);
-		   	if(ret != 0) {
-			   drm_lock_free(dev, &dev->lock.hw_lock->lock,
-					 lock.context);
-			} else {
-			   mga_dma_quiescent(dev);
-			}
+		   DRM_DEBUG("_DRM_LOCK_QUIESCENT\n");
+		   mga_flush_queue(dev);
+		   mga_dma_quiescent(dev);
 		}
 	}
-	printk("%d %s\n", lock.context, ret ? "interrupted" : "has lock");
+   
+	DRM_DEBUG("%d %s\n", lock.context, ret ? "interrupted" : "has lock");
 	return ret;
 }
 		
+int mga_flush_ioctl(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;
+      	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;
+ 
+	copy_from_user_ret(&lock, (drm_lock_t *)arg, sizeof(lock), -EFAULT);
+
+	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("mga_flush_ioctl called without lock held\n");
+		return -EINVAL;
+	}
+
+   	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];
+		}
+	}
+   	if(lock.flags & _DRM_LOCK_QUIESCENT) {
+		mga_dma_quiescent(dev);
+	}
+
+    	return 0;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_dma.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_dma.h
deleted file mode 100644
index 7f1c57957..000000000
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_dma.h
+++ /dev/null
@@ -1,112 +0,0 @@
-#ifndef MGA_DMA_H
-#define MGA_DMA_H
-
-#include "mga_drm_public.h"
-
-
-/* Isn't this fun.  This has to be fixed asap by emitting primary
- * dma commands in the 'do_dma' ioctl.
- */
-typedef struct {
-	int dma_type;
-
-   	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;
-   	xf86drmClipRectRec boxes[MGA_NR_SAREA_CLIPRECTS];
-} drm_mga_buf_priv_t;
-
-
-#define MGA_DMA_GENERAL 0
-#define MGA_DMA_VERTEX  1
-#define MGA_DMA_SETUP   2
-#define MGA_DMA_ILOAD   3
-
-
-#define DWGREG0 	0x1c00
-#define DWGREG0_END 	0x1dff
-#define DWGREG1		0x2c00
-#define DWGREG1_END	0x2dff
-
-#define ISREG0(r)	(r >= DWGREG0 && r <= DWGREG0_END)
-#define ADRINDEX0(r)	(u8)((r - DWGREG0) >> 2)
-#define ADRINDEX1(r)	(u8)(((r - DWGREG1) >> 2) | 0x80)
-#define ADRINDEX(r)	(ISREG0(r) ? ADRINDEX0(r) : ADRINDEX1(r)) 
-
-
-/* Macros for inserting commands into a secondary dma buffer.
- */
-
-#define DMALOCALS	u8 tempIndex[4]; u32 *dma_ptr; \
-			int outcount, num_dwords;
-
-#define	DMAGETPTR(buf) do {			\
-  dma_ptr = (u32 *)((u8 *)buf->address + buf->used);	\
-  outcount = 0;					\
-  num_dwords = buf->used / 4;			\
-} while(0)
-
-#define DMAADVANCE(buf)	do {			\
-  buf->used = num_dwords * 4;			\
-} while(0)
-
-#define DMAOUTREG(reg, val) do {		\
-  tempIndex[outcount]=ADRINDEX(reg);		\
-  dma_ptr[++outcount] = val;			\
-  if (outcount == 4) {				\
-     outcount = 0;				\
-     dma_ptr[0] = *(u32 *)tempIndex;		\
-     dma_ptr+=5;				\
-     num_dwords += 5;				\
-  }						\
-}while (0)
-
-
-
-#define VERBO 0
-
-
-/* Primary buffer versions of above -- pretty similar really.
- */
-#define PRIMLOCALS	u8 tempIndex[4]; u32 *dma_ptr; u32 phys_head; \
-			int outcount, num_dwords
-
-#define PRIMRESET(dev_priv) do {				\
-	dev_priv->prim_num_dwords = 0;				\
-   	dev_priv->current_dma_ptr = dev_priv->prim_head;	\
-} while (0)
-	
-#define	PRIMGETPTR(dev_priv) do {		\
-	dma_ptr = dev_priv->current_dma_ptr;	\
-	phys_head = dev_priv->prim_phys_head;	\
-	num_dwords = dev_priv->prim_num_dwords;	\
-	outcount = 0;				\
-} while (0)
-
-#define PRIMADVANCE(dev_priv)	do {		\
-	dev_priv->prim_num_dwords = num_dwords;	\
-	dev_priv->current_dma_ptr = dma_ptr;	\
-} while (0)
-
-#define PRIMOUTREG(reg, val) do {					\
-	tempIndex[outcount]=ADRINDEX(reg);				\
-	dma_ptr[1+outcount] = val;					\
-	if( ++outcount == 4) {						\
-		outcount = 0;						\
-		dma_ptr[0] = *(u32 *)tempIndex;				\
-		dma_ptr+=5;						\
-		num_dwords += 5;					\
-	}								\
-	if (VERBO) 							\
-		printk(KERN_INFO 					\
-		       "OUT %x val %x dma_ptr %p nr_dwords %d\n",	\
-		       outcount, ADRINDEX(reg), dma_ptr, 		\
-		       num_dwords);   					\
-}while (0)
-
-
-#endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drm_public.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drm.h
index 8b21df11f..12a858e7a 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drm_public.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drm.h
@@ -1,4 +1,4 @@
-/* mga_drm_public.h -- Public header for the Matrox g200/g400 driver -*- linux-c -*-
+/* mga_drm.h -- Public header for the Matrox g200/g400 driver -*- linux-c -*-
  * Created: Tue Jan 25 01:50:01 1999 by jhartmann@precisioninsight.com
  *
  * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
@@ -26,12 +26,17 @@
  * Authors: Jeff Hartmann <jhartmann@precisioninsight.com>
  *          Keith Whitwell <keithw@precisioninsight.com>
  *
- * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drm_public.h,v 1.1 2000/02/11 17:26:07 dawes Exp $
+ * $XFree86$
  */
 
-#ifndef _MGA_DRM_PUBLIC_H_
-#define _MGA_DRM_PUBLIC_H_
+#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 */
@@ -54,66 +59,18 @@
 #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
 
-
-typedef struct _drm_mga_warp_index {
-   	int installed;
-   	unsigned long phys_addr;
-   	int size;
-} mgaWarpIndex;
-
-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;
-   	int fbOffset;
-   	int backOffset;
-   	int depthOffset;
-   	int textureOffset;
-   	int textureSize;
-   	int cpp;
-   	int stride;
-   	int sgram;
-	int chipset;
-   	mgaWarpIndex WarpIndex[MGA_MAX_WARP_PIPES];
-
-	/* Redundant?
-	 */
-	int frontOrg;
-	int backOrg;
-	int depthOrg;
-	int mAccess;
-} drm_mga_init_t;
-
-typedef struct _xf86drmClipRectRec {
-   	unsigned short x1;
-   	unsigned short y1;
-   	unsigned short x2;
-   	unsigned short y2;
-} xf86drmClipRectRec;
-
-#define MGA_CLEAR_FRONT 0x1
-#define MGA_CLEAR_BACK  0x2
-#define MGA_CLEAR_DEPTH 0x4
-
-
-/* Each context has a state:
+/* 3d state excluding texture units:
  */
 #define MGA_CTXREG_DSTORG   0	/* validated */
 #define MGA_CTXREG_MACCESS  1	
@@ -124,7 +81,8 @@ typedef struct _xf86drmClipRectRec {
 #define MGA_CTXREG_WFLAG     6
 #define MGA_CTXREG_TDUAL0    7
 #define MGA_CTXREG_TDUAL1    8
-#define MGA_CTX_SETUP_SIZE   9
+#define MGA_CTXREG_FCOL      9
+#define MGA_CTX_SETUP_SIZE   10
 
 /* 2d state
  */
@@ -146,33 +104,89 @@ typedef struct _xf86drmClipRectRec {
 #define MGA_TEXREG_HEIGHT     10
 #define MGA_TEX_SETUP_SIZE    11
 
-
 /* What needs to be changed for the current vertex dma buffer?
  */
-#define MGASAREA_NEW_CONTEXT    0x1
-#define MGASAREA_NEW_TEX0       0x2
-#define MGASAREA_NEW_TEX1       0x4
-#define MGASAREA_NEW_PIPE       0x8
-#define MGASAREA_NEW_2D 	0x10
-
+#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 */
+
+/* 64 buffers of 16k each, total 1 meg.
+ */
+#define MGA_DMA_BUF_ORDER     14
+#define MGA_DMA_BUF_SZ        (1<<MGA_DMA_BUF_ORDER)
+#define MGA_DMA_BUF_NR        63
 
-/* Keep this small for testing
+/* Keep these small for testing.
  */
-#define MGA_NR_SAREA_CLIPRECTS 2
+#define MGA_NR_SAREA_CLIPRECTS 8
 
-/* Upto 128 regions.  Minimum region size of 256k.
+/* 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_NR_TEX_REGIONS 128
-#define MGA_MIN_LOG_TEX_GRANULARITY 18
 
-typedef struct {
-   unsigned char next, prev;	
-   unsigned char in_use;	
-   int age;			
-} mgaTexRegion;
+#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;
+   	int frontOffset;
+   	int backOffset;
+   	int depthOffset;
+   	int textureOffset;
+   	int textureSize;
+        int agpTextureOffset;
+        int agpTextureSize;
+   	int cpp;
+   	int stride;
+   	int sgram;
+	int chipset;
+   	drm_mga_warp_index_t WarpIndex[MGA_MAX_WARP_PIPES];
+	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;	
+	int age;			
+} drm_mga_tex_region_t;
 
-typedef struct 
-{
+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];
@@ -180,39 +194,68 @@ typedef struct
    	unsigned int dirty;
 
    	unsigned int nbox;
-   	xf86drmClipRectRec boxes[MGA_NR_SAREA_CLIPRECTS];
+   	drm_clip_rect_t boxes[MGA_NR_SAREA_CLIPRECTS];
 
-	/* kernel doesn't touch from here down */
-   	int ctxOwner;		
-	mgaTexRegion texList[MGA_NR_TEX_REGIONS+1];
-	int texAge;	                            
-} drm_mga_sarea_t;	
+
+	/* 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.
+	 */
+        int last_enqueue;	/* last time a buffer was enqueued */
+	int last_dispatch;	/* age of the most recently dispatched buffer */
+	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 {
+typedef struct _drm_mga_clear {
 	int clear_color;
 	int clear_depth;
 	int flags;
 } drm_mga_clear_t;
 
-
-typedef struct {
-	int flags;		/* not actually used? */
+typedef struct _drm_mga_swap {
+   	int dummy;
 } drm_mga_swap_t;
 
-typedef struct {
+typedef struct _drm_mga_iload {
+	int idx;
+	int length;
 	unsigned int destOrg;
-	unsigned int mAccess;
-   	unsigned int pitch;
-	xf86drmClipRectRec texture;
-   	int idx;
 } 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;
 
-#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)
 #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 8bc25617e..480413549 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
@@ -54,6 +54,7 @@ static struct file_operations mga_fops = {
 	mmap:	 drm_mmap,
 	read:	 drm_read,
 	fasync:	 drm_fasync,
+   	poll:	 drm_poll,
 };
 
 static struct miscdevice      mga_misc = {
@@ -105,9 +106,11 @@ static drm_ioctl_desc_t	      mga_ioctls[] = {
 	[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_MGA_INIT)]    = { mga_dma_init,    1, 1 },
-   	[DRM_IOCTL_NR(DRM_IOCTL_MGA_SWAP)]    = { mga_clear_bufs,  1, 1 },
-   	[DRM_IOCTL_NR(DRM_IOCTL_MGA_CLEAR)]   = { mga_swap_bufs,   1, 1 },
-   	[DRM_IOCTL_NR(DRM_IOCTL_MGA_ILOAD)]   = { mga_iload,       1, 1 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_MGA_SWAP)]    = { mga_swap_bufs,   1, 0 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_MGA_CLEAR)]   = { mga_clear_bufs,  1, 0 },
+   	[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 },
 };
 
 #define MGA_IOCTL_COUNT DRM_ARRAY_SIZE(mga_ioctls)
@@ -380,6 +383,21 @@ int mga_init(void)
 	drm_proc_init(dev);
 	DRM_DEBUG("doing agp init\n");
 	dev->agp    = drm_agp_init();
+      	if(dev->agp == NULL) {
+	   	DRM_DEBUG("The mga drm module requires the agpgart module"
+		          " to function correctly\nPlease load the agpgart"
+		          " module before you load the mga module\n");
+	   	drm_proc_cleanup();
+	   	misc_deregister(&mga_misc);
+	   	mga_takedown(dev);
+	   	return -ENOMEM;
+	}
+#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
 	DRM_DEBUG("doing ctxbitmap init\n");
 	if((retcode = drm_ctxbitmap_init(dev))) {
 		DRM_ERROR("Cannot allocate memory for context bitmap.\n");
@@ -416,6 +434,16 @@ void mga_cleanup(void)
 	}
 	drm_ctxbitmap_cleanup(dev);
 	mga_dma_cleanup(dev);
+#ifdef CONFIG_MTRR
+   	if(dev->agp && dev->agp->agp_mtrr) {
+	   	int retval;
+	   	retval = mtrr_del(dev->agp->agp_mtrr, 
+				  dev->agp->agp_info.aper_base,
+				  dev->agp->agp_info.aper_size * 1024*1024);
+	   	DRM_DEBUG("mtrr_del = %d\n", retval);
+	}
+#endif
+
 	mga_takedown(dev);
 	if (dev->agp) {
 		drm_free(dev->agp, sizeof(*dev->agp), DRM_MEM_AGPLISTS);
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 bc7808b0a..43deb613e 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
@@ -26,12 +26,32 @@
  * Authors: Rickard E. (Rik) Faith <faith@precisioninsight.com>
  * 	    Jeff Hartmann <jhartmann@precisioninsight.com>
  *
- * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drv.h,v 1.1 2000/02/11 17:26:08 dawes Exp $
+ * $XFree86$
  */
 
 #ifndef _MGA_DRV_H_
 #define _MGA_DRV_H_
-#include "mga_drm_public.h"
+
+typedef struct {
+   	unsigned int num_dwords;
+   	unsigned int max_dwords;
+   	u32 *current_dma_ptr;
+   	u32 *head;
+   	u32 phys_head;
+   	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 {
+   	unsigned int age;
+   	drm_buf_t *buf;
+   	struct _drm_mga_freelist *next;
+   	struct _drm_mga_freelist *prev;
+} drm_mga_freelist_t;
 
 typedef struct _drm_mga_private {
    	int reserved_map_idx;
@@ -40,7 +60,7 @@ typedef struct _drm_mga_private {
    	int primary_size;
    	int warp_ucode_size;
    	int chipset;
-   	int fbOffset;
+   	int frontOffset;
    	int backOffset;
    	int depthOffset;
    	int textureOffset;
@@ -49,25 +69,32 @@ typedef struct _drm_mga_private {
    	int stride;
    	int sgram;
 	int use_agp;
-   	mgaWarpIndex WarpIndex[MGA_MAX_G400_PIPES];
+   	drm_mga_warp_index_t WarpIndex[MGA_MAX_G400_PIPES];
+	unsigned int WarpPipe;
    	__volatile__ unsigned long softrap_age;
-   	atomic_t dispatch_lock;
+   	u32 dispatch_lock;
+   	atomic_t in_flush;
+   	atomic_t in_wait;
    	atomic_t pending_bufs;
-   	void *ioremap;
-   	u32 *prim_head;
-   	u32 *current_dma_ptr;
-   	u32 prim_phys_head;
-   	int prim_num_dwords;
-   	int prim_max_dwords;
-
+   	unsigned int last_sync_tag;
+   	unsigned int sync_tag;
+   	void *status_page;
+   	unsigned long real_status_page;
+   	u8 *ioremap;
+   	drm_mga_prim_buf_t **prim_bufs;
+   	drm_mga_prim_buf_t *next_prim;
+   	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 */
 
 	/* Some validated register values:
 	 */	
-	u32 frontOrg;
-	u32 backOrg;
-	u32 depthOrg;
 	u32 mAccess;
-
 } drm_mga_private_t;
 
 				/* mga_drv.c */
@@ -92,16 +119,18 @@ extern int  mga_control(struct inode *inode, struct file *filp,
 			  unsigned int cmd, unsigned long arg);
 extern int  mga_lock(struct inode *inode, struct file *filp,
 		       unsigned int cmd, unsigned long arg);
-#if 0
-extern void mga_dma_init(drm_device_t *dev);
-extern void mga_dma_cleanup(drm_device_t *dev);
 
-#endif
+/* mga_dma_init does init and release */
 extern int mga_dma_init(struct inode *inode, struct file *filp,
 			unsigned int cmd, unsigned long arg);
 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);
 
-/* mga_dma_init does init and release */
+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);
+extern int mga_advance_primary(drm_device_t *dev);
 
 
 				/* mga_bufs.c */
@@ -124,6 +153,8 @@ extern int  mga_swap_bufs(struct inode *inode, struct file *filp,
 			  unsigned int cmd, unsigned long arg);
 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);
 				/* mga_context.c */
 extern int  mga_resctx(struct inode *inode, struct file *filp,
 		       unsigned int cmd, unsigned long arg);
@@ -144,13 +175,115 @@ extern int  mga_context_switch(drm_device_t *dev, int old, int new);
 extern int  mga_context_switch_complete(drm_device_t *dev, int new);
 
 
+typedef enum {
+	TT_GENERAL,
+	TT_BLIT,
+	TT_VECTOR,
+	TT_VERTEX
+} transferType_t;
+
+typedef struct {
+   	drm_mga_freelist_t *my_freelist;
+	int discard;
+} drm_mga_buf_priv_t;
+
+#define DWGREG0 	0x1c00
+#define DWGREG0_END 	0x1dff
+#define DWGREG1		0x2c00
+#define DWGREG1_END	0x2dff
+
+#define ISREG0(r)	(r >= DWGREG0 && r <= DWGREG0_END)
+#define ADRINDEX0(r)	(u8)((r - DWGREG0) >> 2)
+#define ADRINDEX1(r)	(u8)(((r - DWGREG1) >> 2) | 0x80)
+#define ADRINDEX(r)	(ISREG0(r) ? ADRINDEX0(r) : ADRINDEX1(r)) 
+
+#define MGA_VERBOSE 0
+#define MGA_NUM_PRIM_BUFS 	8
+
+#define PRIMLOCALS	u8 tempIndex[4]; u32 *dma_ptr; u32 phys_head; \
+			int outcount, num_dwords
+
+#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);				\
+	}								\
+} while(0)
+
+#define PRIMGETPTR(dev_priv) do {					\
+	drm_mga_prim_buf_t *tmp_buf =					\
+		dev_priv->prim_bufs[dev_priv->current_prim_idx];	\
+	if(MGA_VERBOSE)							\
+		DRM_DEBUG("PRIMGETPTR in %s\n", __FUNCTION__);		\
+	dma_ptr = tmp_buf->current_dma_ptr;				\
+	num_dwords = tmp_buf->num_dwords;				\
+	phys_head = tmp_buf->phys_head;					\
+	outcount = 0;							\
+} while(0)
+
+#define PRIMPTR(prim_buf) do {					\
+	if(MGA_VERBOSE)						\
+		DRM_DEBUG("PRIMPTR in %s\n", __FUNCTION__);	\
+	dma_ptr = prim_buf->current_dma_ptr;			\
+	num_dwords = prim_buf->num_dwords;			\
+	phys_head = prim_buf->phys_head;			\
+	outcount = 0;						\
+} while(0)
+
+#define PRIMFINISH(prim_buf) do {				\
+	if (MGA_VERBOSE) {					\
+		DRM_DEBUG( "PRIMFINISH in %s\n", __FUNCTION__);	\
+                if (outcount & 3)				\
+                      DRM_DEBUG(" --- truncation\n");	        \
+        }							\
+	prim_buf->num_dwords = num_dwords;			\
+	prim_buf->current_dma_ptr = dma_ptr;			\
+} while(0)
+
+#define PRIMADVANCE(dev_priv)	do {				\
+drm_mga_prim_buf_t *tmp_buf = 					\
+	dev_priv->prim_bufs[dev_priv->current_prim_idx];	\
+	if (MGA_VERBOSE) {					\
+		DRM_DEBUG("PRIMADVANCE in %s\n", __FUNCTION__);	\
+                if (outcount & 3)				\
+                      DRM_DEBUG(" --- truncation\n");	\
+        }							\
+	tmp_buf->num_dwords = num_dwords;      			\
+	tmp_buf->current_dma_ptr = dma_ptr;    			\
+} while (0)
+
+#define PRIMUPDATE(dev_priv)	do {					\
+	drm_mga_prim_buf_t *tmp_buf =					\
+		dev_priv->prim_bufs[dev_priv->current_prim_idx];	\
+	tmp_buf->sec_used++;						\
+} while (0)
+
+#define PRIMOUTREG(reg, val) do {					\
+	tempIndex[outcount]=ADRINDEX(reg);				\
+	dma_ptr[1+outcount] = val;					\
+	if (MGA_VERBOSE)						\
+		DRM_DEBUG("   PRIMOUT %d: 0x%x -- 0x%x\n",		\
+		       num_dwords + 1 + outcount, ADRINDEX(reg), val);	\
+	if( ++outcount == 4) {						\
+		outcount = 0;						\
+		dma_ptr[0] = *(u32 *)tempIndex;				\
+		dma_ptr+=5;						\
+		num_dwords += 5;					\
+	}								\
+}while (0)
+
+/* A reduced set of the mga registers.
+ */
 
 #define MGAREG_MGA_EXEC 			0x0100
-#define MGAREG_AGP_PLL				0x1e4c
 #define MGAREG_ALPHACTRL 			0x2c7c
-#define MGAREG_ALPHASTART 			0x2c70
-#define MGAREG_ALPHAXINC 			0x2c74
-#define MGAREG_ALPHAYINC 			0x2c78
 #define MGAREG_AR0 				0x1c60
 #define MGAREG_AR1 				0x1c64
 #define MGAREG_AR2 				0x1c68
@@ -158,39 +291,16 @@ extern int  mga_context_switch_complete(drm_device_t *dev, int new);
 #define MGAREG_AR4 				0x1c70
 #define MGAREG_AR5 				0x1c74
 #define MGAREG_AR6 				0x1c78
-#define MGAREG_BCOL 				0x1c20
 #define MGAREG_CXBNDRY				0x1c80
 #define MGAREG_CXLEFT 				0x1ca0
 #define MGAREG_CXRIGHT				0x1ca4
 #define MGAREG_DMAPAD 				0x1c54
-#define MGAREG_DR0_Z32LSB 			0x2c50
-#define MGAREG_DR0_Z32MSB 			0x2c54
-#define MGAREG_DR2_Z32LSB 			0x2c60
-#define MGAREG_DR2_Z32MSB 			0x2c64
-#define MGAREG_DR3_Z32LSB 			0x2c68
-#define MGAREG_DR3_Z32MSB 			0x2c6c
-#define MGAREG_DR0 				0x1cc0
-#define MGAREG_DR2 				0x1cc8
-#define MGAREG_DR3 				0x1ccc
-#define MGAREG_DR4 				0x1cd0
-#define MGAREG_DR6 				0x1cd8
-#define MGAREG_DR7 				0x1cdc
-#define MGAREG_DR8 				0x1ce0
-#define MGAREG_DR10 				0x1ce8
-#define MGAREG_DR11 				0x1cec
-#define MGAREG_DR12 				0x1cf0
-#define MGAREG_DR14 				0x1cf8
-#define MGAREG_DR15 				0x1cfc
 #define MGAREG_DSTORG 				0x2cb8
-#define MGAREG_DWG_INDIR_WT 			0x1e80
 #define MGAREG_DWGCTL 				0x1c00
 #define MGAREG_DWGSYNC				0x2c4c
 #define MGAREG_FCOL 				0x1c24
 #define MGAREG_FIFOSTATUS 			0x1e10
 #define MGAREG_FOGCOL 				0x1cf4
-#define MGAREG_FOGSTART 			0x1cc4
-#define MGAREG_FOGXINC				0x1cd4
-#define MGAREG_FOGYINC				0x1ce4
 #define MGAREG_FXBNDRY				0x1c84
 #define MGAREG_FXLEFT 				0x1ca8
 #define MGAREG_FXRIGHT				0x1cac
@@ -198,44 +308,22 @@ extern int  mga_context_switch_complete(drm_device_t *dev, int new);
 #define MGAREG_IEN 				0x1e1c
 #define MGAREG_LEN 				0x1c5c
 #define MGAREG_MACCESS				0x1c04
-#define MGAREG_MCTLWTST 			0x1c08
-#define MGAREG_MEMRDBK				0x1e44
-#define MGAREG_OPMODE 				0x1e54
-#define MGAREG_PAT0 				0x1c10
-#define MGAREG_PAT1 				0x1c14
 #define MGAREG_PITCH 				0x1c8c
 #define MGAREG_PLNWT 				0x1c1c
 #define MGAREG_PRIMADDRESS 			0x1e58
 #define MGAREG_PRIMEND				0x1e5c
 #define MGAREG_PRIMPTR				0x1e50
-#define MGAREG_RST 				0x1e40
 #define MGAREG_SECADDRESS 			0x2c40
 #define MGAREG_SECEND 				0x2c44
 #define MGAREG_SETUPADDRESS 			0x2cd0
 #define MGAREG_SETUPEND 			0x2cd4
-#define MGAREG_SGN 				0x1c58
-#define MGAREG_SHIFT 				0x1c50
 #define MGAREG_SOFTRAP				0x2c48
-#define MGAREG_SPECBSTART 			0x2c98
-#define MGAREG_SPECBXINC 			0x2c9c
-#define MGAREG_SPECBYINC 			0x2ca0
-#define MGAREG_SPECGSTART 			0x2c8c
-#define MGAREG_SPECGXINC 			0x2c90
-#define MGAREG_SPECGYINC 			0x2c94
-#define MGAREG_SPECRSTART 			0x2c80
-#define MGAREG_SPECRXINC 			0x2c84
-#define MGAREG_SPECRYINC 			0x2c88
-#define MGAREG_SRC0 				0x1c30
-#define MGAREG_SRC1 				0x1c34
-#define MGAREG_SRC2 				0x1c38
-#define MGAREG_SRC3 				0x1c3c
 #define MGAREG_SRCORG 				0x2cb4
 #define MGAREG_STATUS 				0x1e14
 #define MGAREG_STENCIL				0x2cc8
 #define MGAREG_STENCILCTL 			0x2ccc
 #define MGAREG_TDUALSTAGE0 			0x2cf8
 #define MGAREG_TDUALSTAGE1 			0x2cfc
-#define MGAREG_TEST0 				0x1e48
 #define MGAREG_TEXBORDERCOL 			0x2c5c
 #define MGAREG_TEXCTL 				0x2c30
 #define MGAREG_TEXCTL2				0x2c3c
@@ -249,18 +337,6 @@ extern int  mga_context_switch_complete(drm_device_t *dev, int new);
 #define MGAREG_TEXTRANS 			0x2c34
 #define MGAREG_TEXTRANSHIGH 			0x2c38
 #define MGAREG_TEXWIDTH 			0x2c28
-#define MGAREG_TMR0 				0x2c00
-#define MGAREG_TMR1 				0x2c04
-#define MGAREG_TMR2 				0x2c08
-#define MGAREG_TMR3 				0x2c0c
-#define MGAREG_TMR4 				0x2c10
-#define MGAREG_TMR5 				0x2c14
-#define MGAREG_TMR6 				0x2c18
-#define MGAREG_TMR7 				0x2c1c
-#define MGAREG_TMR8 				0x2c20
-#define MGAREG_VBIADDR0 			0x3e08
-#define MGAREG_VBIADDR1 			0x3e0c
-#define MGAREG_VCOUNT 				0x1e20
 #define MGAREG_WACCEPTSEQ 			0x1dd4
 #define MGAREG_WCODEADDR 			0x1e6c
 #define MGAREG_WFLAG 				0x1dc4
@@ -270,18 +346,8 @@ extern int  mga_context_switch_complete(drm_device_t *dev, int new);
 #define MGAREG_WGETMSB				0x1dc8
 #define MGAREG_WIADDR 				0x1dc0
 #define MGAREG_WIADDR2				0x1dd8
-#define MGAREG_WIADDRNB 			0x1e60
-#define MGAREG_WIADDRNB1 			0x1e04
-#define MGAREG_WIADDRNB2 			0x1e00
-#define MGAREG_WIMEMADDR 			0x1e68
-#define MGAREG_WIMEMDATA 			0x2000
-#define MGAREG_WIMEMDATA1 			0x2100
 #define MGAREG_WMISC 				0x1e70
-#define MGAREG_WR 				0x2d00
 #define MGAREG_WVRTXSZ				0x1dcc
-#define MGAREG_XDST 				0x1cb0
-#define MGAREG_XYEND 				0x1c44
-#define MGAREG_XYSTRT 				0x1c40
 #define MGAREG_YBOT 				0x1c9c
 #define MGAREG_YDST 				0x1c90
 #define MGAREG_YDSTLEN				0x1c88
@@ -289,4 +355,40 @@ extern int  mga_context_switch_complete(drm_device_t *dev, int new);
 #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_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 MGA_CLEAR_CMD (DC_opcod_trap | DC_arzero_enable | 		\
+		       DC_sgnzero_enable | DC_shftzero_enable | 	\
+		       (0xC << DC_bop_SHIFT) | DC_clipdis_enable | 	\
+		       DC_solid_enable | DC_transc_enable)
+	  
+
+#define MGA_COPY_CMD (DC_opcod_bitblt | DC_atype_rpl | DC_linear_xy |	\
+		      DC_solid_disable | DC_arzero_disable | 		\
+		      DC_sgnzero_enable | DC_shftzero_enable | 		\
+		      (0xC << DC_bop_SHIFT) | DC_bltmod_bfcol | 	\
+		      DC_pattern_disable | DC_transc_disable | 		\
+		      DC_clipdis_enable)				\
+
 #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 d09881bad..a70f86d22 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
@@ -26,189 +26,217 @@
  * Authors: Jeff Hartmann <jhartmann@precisioninsight.com>
  * 	    Keith Whitwell <keithw@precisioninsight.com>
  *
- * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_state.c,v 1.1 2000/02/11 17:26:08 dawes Exp $
+ * $XFree86$
  *
  */
  
 #define __NO_VERSION__
 #include "drmP.h"
 #include "mga_drv.h"
-#include "mgareg_flags.h"
-#include "mga_dma.h"
-#include "mga_state.h"
 #include "drm.h"
 
-void mgaEmitClipRect( drm_mga_private_t *dev_priv, xf86drmClipRectRec *box )
+static void mgaEmitClipRect( drm_mga_private_t *dev_priv, 
+			     drm_clip_rect_t *box )
 {
+   	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	unsigned int *regs = sarea_priv->ContextState;
 	PRIMLOCALS;
 
+   	/* This takes 10 dwords */
 	PRIMGETPTR( dev_priv );
+   
+   	/* Force reset of dwgctl (eliminates clip disable) */
+   	PRIMOUTREG( MGAREG_DMAPAD, 0 );
+	PRIMOUTREG( MGAREG_DWGSYNC, dev_priv->last_sync_tag - 1 );
+	PRIMOUTREG( MGAREG_DWGSYNC, dev_priv->last_sync_tag - 1 );
+   	PRIMOUTREG( MGAREG_DWGCTL, regs[MGA_CTXREG_DWGCTL] );
 
-	/* The G400 seems to have an issue with the second WARP not
-	 * stalling clipper register writes.  This bothers me, but the only
-	 * way I could get it to never clip the last triangle under any
-	 * circumstances is by inserting TWO dwgsync commands.
-	 */
- 	if (dev_priv->chipset == MGA_CARD_TYPE_G400) { 
-		PRIMOUTREG( MGAREG_DWGSYNC, 0 );
-		PRIMOUTREG( MGAREG_DWGSYNC, 0 );
-	}
-
+   	PRIMOUTREG( MGAREG_DMAPAD, 0 );
 	PRIMOUTREG( MGAREG_CXBNDRY, ((box->x2)<<16)|(box->x1) );
-	PRIMOUTREG( MGAREG_YTOP, box->y1 * dev_priv->stride );
-	PRIMOUTREG( MGAREG_YBOT, box->y2 * dev_priv->stride );
+	PRIMOUTREG( MGAREG_YTOP, box->y1 * dev_priv->stride/2 );
+	PRIMOUTREG( MGAREG_YBOT, box->y2 * dev_priv->stride/2 );
+
 	PRIMADVANCE( dev_priv );
 }
 
-
-static void mgaEmitContext(drm_mga_private_t *dev_priv, 
-			   drm_mga_buf_priv_t *buf_priv)
+static void mgaEmitContext(drm_mga_private_t *dev_priv )
 {
-	unsigned int *regs = buf_priv->ContextState;
+      	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	unsigned int *regs = sarea_priv->ContextState;
 	PRIMLOCALS;
 	
+   	/* This takes a max of 15 dwords */
 	PRIMGETPTR( dev_priv );
+
 	PRIMOUTREG( MGAREG_DSTORG, regs[MGA_CTXREG_DSTORG] );
 	PRIMOUTREG( MGAREG_MACCESS, regs[MGA_CTXREG_MACCESS] );
 	PRIMOUTREG( MGAREG_PLNWT, regs[MGA_CTXREG_PLNWT] );
 	PRIMOUTREG( MGAREG_DWGCTL, regs[MGA_CTXREG_DWGCTL] );
+
 	PRIMOUTREG( MGAREG_ALPHACTRL, regs[MGA_CTXREG_ALPHACTRL] );
 	PRIMOUTREG( MGAREG_FOGCOL, regs[MGA_CTXREG_FOGCOLOR] );
 	PRIMOUTREG( MGAREG_WFLAG, regs[MGA_CTXREG_WFLAG] );
+	PRIMOUTREG( MGAREG_ZORG, dev_priv->depthOffset ); /* invarient */
 
  	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_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);
+	}
    
 	PRIMADVANCE( dev_priv );
 }
 
-static void mgaG200EmitTex( drm_mga_private_t *dev_priv, 
-			    drm_mga_buf_priv_t *buf_priv )
+static void mgaG200EmitTex( drm_mga_private_t *dev_priv )
 {
-	unsigned int *regs = buf_priv->TexState[0];
+      	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	unsigned int *regs = sarea_priv->TexState[0];
 	PRIMLOCALS;
 
 	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_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(MGAREG_TEXHEIGHT, regs[MGA_TEXREG_HEIGHT] );   
 	PRIMOUTREG(0x2d00 + 24*4, regs[MGA_TEXREG_WIDTH] );
+
 	PRIMOUTREG(0x2d00 + 34*4, regs[MGA_TEXREG_HEIGHT] );
+	PRIMOUTREG( MGAREG_TEXTRANS, 0xffff );
+	PRIMOUTREG( MGAREG_TEXTRANSHIGH, 0xffff );
+	PRIMOUTREG( MGAREG_DMAPAD, 0 );
 
 	PRIMADVANCE( dev_priv );  
 }
 
-static void mgaG400EmitTex0( drm_mga_private_t *dev_priv, 
-			     drm_mga_buf_priv_t *buf_priv )
+static void mgaG400EmitTex0( drm_mga_private_t *dev_priv )
 {
-	unsigned int *regs = buf_priv->TexState[0];
-	int multitex = buf_priv->WarpPipe & MGA_T2;
-
+      	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	unsigned int *regs = sarea_priv->TexState[0];
+	int multitex = sarea_priv->WarpPipe & MGA_T2;
 	PRIMLOCALS;
+   
 	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_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(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( MGAREG_DMAPAD, 0 );
 
 	if (!multitex) {
 		PRIMOUTREG(0x2d00 + 52*4, 0x40 ); 
 		PRIMOUTREG(0x2d00 + 60*4, 0x40 );
+		PRIMOUTREG( MGAREG_DMAPAD, 0 );
+		PRIMOUTREG( MGAREG_DMAPAD, 0 );
 	} 
 
-	PRIMOUTREG(0x2d00 + 54*4, regs[MGA_TEXREG_WIDTH] | 0x40 );
-	PRIMOUTREG(0x2d00 + 62*4, regs[MGA_TEXREG_HEIGHT] | 0x40 );
+	PRIMOUTREG( 0x2d00 + 54*4, regs[MGA_TEXREG_WIDTH] | 0x40 );
+	PRIMOUTREG( 0x2d00 + 62*4, regs[MGA_TEXREG_HEIGHT] | 0x40 );
+	PRIMOUTREG( MGAREG_TEXTRANS, 0xffff );
+	PRIMOUTREG( MGAREG_TEXTRANSHIGH, 0xffff );
 
 	PRIMADVANCE( dev_priv );  
 }
 
 #define TMC_map1_enable 		0x80000000 	
 
-
-static void mgaG400EmitTex1( drm_mga_private_t *dev_priv, 
-			     drm_mga_buf_priv_t *buf_priv )
+static void mgaG400EmitTex1( drm_mga_private_t *dev_priv )
 {
-	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+      	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
 	unsigned int *regs = sarea_priv->TexState[1];
-
 	PRIMLOCALS;
+   
 	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(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(0x2d00 + 60*4, regs[MGA_TEXREG_HEIGHT] | 0x40 );
+	PRIMOUTREG( MGAREG_TEXTRANS, 0xffff );
+	PRIMOUTREG( MGAREG_TEXTRANSHIGH, 0xffff );
 	PRIMOUTREG(MGAREG_TEXCTL2, regs[MGA_TEXREG_CTL2] );
 
-	PRIMADVANCE( dev_priv );     
+	PRIMADVANCE( dev_priv );
 }
 
-
-static void mgaG400EmitPipe(drm_mga_private_t *dev_priv, 
-			    drm_mga_buf_priv_t *buf_priv)
+static void mgaG400EmitPipe(drm_mga_private_t *dev_priv )
 {
-	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+      	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
 	unsigned int pipe = sarea_priv->WarpPipe;
 	float fParam = 12800.0f;
 	PRIMLOCALS;
    
 	PRIMGETPTR(dev_priv);
-	PRIMOUTREG(MGAREG_WIADDR2, WIA_wmode_suspend);
+
+   	/* 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);
 	} else {
+		PRIMOUTREG(MGAREG_WIADDR2, WIA_wmode_suspend);
 		PRIMOUTREG(MGAREG_WVRTXSZ, 0x00001807);
 		PRIMOUTREG(MGAREG_WACCEPTSEQ, 0x18000000);
-	}
-   
-	PRIMOUTREG(MGAREG_WFLAG, 0);
-	PRIMOUTREG(MGAREG_WFLAG1, 0);
-   
+		PRIMOUTREG(MGAREG_WFLAG, 0);
+	}   
+
+	PRIMOUTREG(MGAREG_WFLAG1, 0);   
 	PRIMOUTREG(0x2d00 + 56*4, *((u32 *)(&fParam)));
 	PRIMOUTREG(MGAREG_DMAPAD, 0);
 	PRIMOUTREG(MGAREG_DMAPAD, 0);
@@ -227,19 +255,21 @@ static void mgaG400EmitPipe(drm_mga_private_t *dev_priv,
 	PRIMOUTREG(MGAREG_DMAPAD, 0xffffffff);
 	PRIMOUTREG(MGAREG_DMAPAD, 0xffffffff);
 	PRIMOUTREG(MGAREG_DMAPAD, 0xffffffff);
-	PRIMOUTREG(MGAREG_WIADDR2, (dev_priv->WarpIndex[pipe].phys_addr | 
-				   WIA_wmode_start | WIA_wagp_agp));
+	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,
-			     drm_mga_buf_priv_t *buf_priv )
+static void mgaG200EmitPipe( drm_mga_private_t *dev_priv )
 {
-	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+      	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
 	unsigned int pipe = sarea_priv->WarpPipe;
 	PRIMLOCALS;
 
 	PRIMGETPTR(dev_priv);
+   
+   	/* This takes 15 dwords */
+
 	PRIMOUTREG(MGAREG_WIADDR, WIA_wmode_suspend);
 	PRIMOUTREG(MGAREG_WVRTXSZ, 7);
 	PRIMOUTREG(MGAREG_WFLAG, 0);
@@ -254,109 +284,629 @@ static void mgaG200EmitPipe( drm_mga_private_t *dev_priv,
 	PRIMOUTREG(MGAREG_DMAPAD, 0xffffffff);
 	PRIMOUTREG(MGAREG_DMAPAD, 0xffffffff);
 	PRIMOUTREG(MGAREG_DMAPAD, 0xffffffff);
-	PRIMOUTREG(MGAREG_WIADDR, (dev_priv->WarpIndex[pipe].phys_addr | 
-				  WIA_wmode_start | WIA_wagp_agp));
+	PRIMOUTREG(MGAREG_WIADDR, (__u32)(dev_priv->WarpIndex[pipe].phys_addr | 
+					  WIA_wmode_start | WIA_wagp_agp));
 
 	PRIMADVANCE(dev_priv);
 }
 
-void mgaEmitState( drm_mga_private_t *dev_priv, drm_mga_buf_priv_t *buf_priv )
+static void mgaEmitState( drm_mga_private_t *dev_priv )
 {
-	unsigned int dirty = buf_priv->dirty;
+      	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	unsigned int dirty = sarea_priv->dirty;
 
 	if (dev_priv->chipset == MGA_CARD_TYPE_G400) {	   
-		if (dirty & MGASAREA_NEW_CONTEXT)
-			mgaEmitContext( dev_priv, buf_priv );
-
-		if (dirty & MGASAREA_NEW_TEX1)
-			mgaG400EmitTex1( dev_priv, buf_priv );
-	   
-		if (dirty & MGASAREA_NEW_TEX0)
-			mgaG400EmitTex0( dev_priv, buf_priv );
-
-		if (dirty & MGASAREA_NEW_PIPE) 
-			mgaG400EmitPipe( dev_priv, buf_priv );
+		int multitex = sarea_priv->WarpPipe & MGA_T2;
+
+		if (sarea_priv->WarpPipe != dev_priv->WarpPipe) { 
+			mgaG400EmitPipe( dev_priv );
+			dev_priv->WarpPipe = sarea_priv->WarpPipe;
+		}
+
+		if (dirty & MGA_UPLOAD_CTX) {
+			mgaEmitContext( dev_priv );
+		   	sarea_priv->dirty &= ~MGA_UPLOAD_CTX;
+		}
+
+		if (dirty & MGA_UPLOAD_TEX0) {
+			mgaG400EmitTex0( dev_priv );
+		   	sarea_priv->dirty &= ~MGA_UPLOAD_TEX0;
+		}
+
+		if ((dirty & MGA_UPLOAD_TEX1) && multitex) {
+			mgaG400EmitTex1( dev_priv );
+		   	sarea_priv->dirty &= ~MGA_UPLOAD_TEX1;
+		}
 	} else {
-		if (dirty & MGASAREA_NEW_CONTEXT)
-			mgaEmitContext( dev_priv, buf_priv );
-
-		if (dirty & MGASAREA_NEW_TEX0)
-			mgaG200EmitTex( dev_priv, buf_priv );
-
-		if (dirty & MGASAREA_NEW_PIPE) 
-			mgaG200EmitPipe( dev_priv, buf_priv );
-	}	  
+		if (sarea_priv->WarpPipe != dev_priv->WarpPipe) { 
+			mgaG200EmitPipe( dev_priv );
+			dev_priv->WarpPipe = sarea_priv->WarpPipe;
+		}
+
+		if (dirty & MGA_UPLOAD_CTX) {
+			mgaEmitContext( dev_priv );
+		   	sarea_priv->dirty &= ~MGA_UPLOAD_CTX;
+		}
+
+		if (dirty & MGA_UPLOAD_TEX0) {
+			mgaG200EmitTex( dev_priv );
+		   	sarea_priv->dirty &= ~MGA_UPLOAD_TEX0;
+		}
+	}
 }
 
 
-
 /* Disallow all write destinations except the front and backbuffer.
  */
-static int mgaCopyContext(drm_mga_private_t *dev_priv, 
-			  drm_mga_buf_priv_t *buf_priv)
+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;
-	
-	if (regs[MGA_CTXREG_DSTORG] != dev_priv->frontOrg &&
-	    regs[MGA_CTXREG_DSTORG] != dev_priv->backOrg)
+
+	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", 
+		       regs[MGA_CTXREG_DSTORG], dev_priv->frontOffset,
+		       dev_priv->backOffset);
+  	        regs[MGA_CTXREG_DSTORG] = 0;
 		return -1;
+	}
 
-	memcpy(buf_priv->ContextState, sarea_priv->ContextState,
-	       sizeof(buf_priv->ContextState));
 	return 0;
 }
 
-
 /* Disallow texture reads from PCI space.
  */
-static int mgaCopyTex(drm_mga_private_t *dev_priv, 
-		      drm_mga_buf_priv_t *buf_priv,
+static int mgaVerifyTex(drm_mga_private_t *dev_priv, 
 		      int unit)
 {
 	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
 
-	if ((sarea_priv->TexState[unit][MGA_TEXREG_ORG] & 0x3) == 0x1)
+	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],
+		       unit);
+		sarea_priv->TexState[unit][MGA_TEXREG_ORG] = 0;
 		return -1;
-
-	memcpy(buf_priv->TexState[unit], sarea_priv->TexState[unit],
-	       sizeof(buf_priv->TexState[0]));
+	} 
 
 	return 0;
 }
 
-
-int mgaCopyAndVerifyState( drm_mga_private_t *dev_priv, 
-			   drm_mga_buf_priv_t *buf_priv )
+static int mgaVerifyState( drm_mga_private_t *dev_priv )
 {
 	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
-	unsigned int dirty = sarea_priv->dirty ;
+	unsigned int dirty = sarea_priv->dirty;
 	int rv = 0;
 
-	buf_priv->dirty = sarea_priv->dirty;
-	buf_priv->WarpPipe = sarea_priv->WarpPipe;
+   	if (sarea_priv->nbox > MGA_NR_SAREA_CLIPRECTS)
+     		sarea_priv->nbox = MGA_NR_SAREA_CLIPRECTS;
 
-	if (dirty & MGASAREA_NEW_CONTEXT)
-		rv |= mgaCopyContext( dev_priv, buf_priv );
+	if (dirty & MGA_UPLOAD_CTX)
+		rv |= mgaVerifyContext( dev_priv );
 
-	if (dirty & MGASAREA_NEW_TEX0)
-		rv |= mgaCopyTex( dev_priv, buf_priv, 0 );
+	if (dirty & MGA_UPLOAD_TEX0)
+		rv |= mgaVerifyTex( dev_priv, 0 );
 
 	if (dev_priv->chipset == MGA_CARD_TYPE_G400) 
 	{	   
-		if (dirty & MGASAREA_NEW_TEX1)
-			rv |= mgaCopyTex( dev_priv, buf_priv, 1 );
+		if (dirty & MGA_UPLOAD_TEX1)
+			rv |= mgaVerifyTex( dev_priv, 1 );
 	   
-		if (dirty & MGASAREA_NEW_PIPE) 
-			rv |= (buf_priv->WarpPipe > MGA_MAX_G400_PIPES); 
+		if (dirty & MGA_UPLOAD_PIPE) 
+			rv |= (sarea_priv->WarpPipe > MGA_MAX_G400_PIPES); 
 	} 
 	else 
 	{
-		if (dirty & MGASAREA_NEW_PIPE) 
-			rv |= (buf_priv->WarpPipe > MGA_MAX_G200_PIPES); 
+		if (dirty & MGA_UPLOAD_PIPE) 
+			rv |= (sarea_priv->WarpPipe > MGA_MAX_G200_PIPES); 
 	}	  
 
 	return rv == 0;
 }
 
+static int mgaVerifyIload( drm_mga_private_t *dev_priv, 
+			   unsigned long bus_address, 
+			   unsigned int dstOrg, int length )
+{
+   	if(dstOrg < dev_priv->textureOffset ||
+	   dstOrg + length > 
+	   (dev_priv->textureOffset + dev_priv->textureSize)) {
+	   	return -EINVAL;
+	}
+   	if(length % 64) {
+	   	return -EINVAL;
+	}
+   	return 0;
+}
+
+/* This copies a 64 byte aligned agp region to the frambuffer
+ * with a standard blit, the ioctl needs to do checking */
+
+static void mga_dma_dispatch_tex_blit( drm_device_t *dev, 
+				       unsigned long bus_address, 
+				       int length, 
+				       unsigned int destOrg )
+{
+   	drm_mga_private_t *dev_priv = dev->dev_private;
+   	int use_agp = PDEA_pagpxfer_enable | 0x00000001;
+   	u16 y2;
+    	PRIMLOCALS;
+    
+        y2 = length / 64;
+ 
+	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);
+   	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, 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);
+}
+
+static void mga_dma_dispatch_vertex(drm_device_t *dev, 
+				    drm_buf_t *buf)
+{
+   	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 long address = (unsigned long)buf->bus_address;
+	int length = buf->used;	
+	int use_agp = PDEA_pagpxfer_enable;
+	int i = 0;
+   	int primary_needed;
+      	PRIMLOCALS;
+
+	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);
+
+
+   	dev_priv->last_sync_tag = mga_create_sync_tag(dev);
+
+   	if (buf_priv->discard) { 
+		buf_priv->my_freelist->age = dev_priv->last_sync_tag;
+	   	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 );
+
+	if (buf->used) {
+		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);
+	}
+
+	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 );
+}
+
 
+static void mga_dma_dispatch_clear( drm_device_t *dev, int flags, 
+				    unsigned int clear_color,
+				    unsigned int clear_zval )
+{
+   	drm_mga_private_t *dev_priv = dev->dev_private;
+      	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	unsigned int *regs = sarea_priv->ContextState;
+	int nbox = sarea_priv->nbox;
+	drm_clip_rect_t *pbox = sarea_priv->boxes;
+	unsigned int cmd;
+	int i;
+   	int primary_needed;
+   	PRIMLOCALS;
+
+	if ( dev_priv->sgram ) 
+		cmd = MGA_CLEAR_CMD | DC_atype_blk;
+	else
+		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);
+	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;
+
+	   	DRM_DEBUG("dispatch clear %d,%d-%d,%d flags %x!\n",
+			  pbox[i].x1, pbox[i].y1, pbox[i].x2, 
+			  pbox[i].y2, 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 );
+		}
+
+		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_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 );
+		}
+	}
+
+	/* Force reset of DWGCTL */
+   	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);
+}
+
+static void mga_dma_dispatch_swap( drm_device_t *dev )
+{
+   	drm_mga_private_t *dev_priv = dev->dev_private;
+      	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	unsigned int *regs = sarea_priv->ContextState;
+	int nbox = sarea_priv->nbox;
+	drm_clip_rect_t *pbox = sarea_priv->boxes;
+	int i;
+   	int primary_needed;
+   	PRIMLOCALS;
+
+   	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_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;
+		unsigned int start = pbox[i].y1 * dev_priv->stride/2;
+		
+	   	DRM_DEBUG("dispatch swap %d,%d-%d,%d!\n",
+			  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);
+	}
+
+   	/* Force reset of DWGCTL */
+   	PRIMOUTREG( MGAREG_DMAPAD, 0);
+      	PRIMOUTREG( MGAREG_DMAPAD, 0);
+   	PRIMOUTREG( MGAREG_DMAPAD, 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);
+}
+
+int mga_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_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);
+
+	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("mga_clear_bufs called without lock held\n");
+		return -EINVAL;
+	}
+   
+   	if (sarea_priv->nbox > MGA_NR_SAREA_CLIPRECTS)
+     		sarea_priv->nbox = MGA_NR_SAREA_CLIPRECTS;
+
+	/* Make sure we restore the 3D state next time.
+	 */
+	dev_priv->sarea_priv->dirty |= MGA_UPLOAD_CTX;
+	mga_dma_dispatch_clear( dev, clear.flags, 
+			        clear.clear_color, 
+			        clear.clear_depth );
+      	PRIMUPDATE(dev_priv);
+	mga_dma_schedule(dev, 1);
+   	sarea_priv->last_dispatch = status[1];
+   	return 0;
+}
+
+int mga_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;
+   	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;
+   
+	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("mga_swap_bufs called without lock held\n");
+		return -EINVAL;
+	}
+
+      	if (sarea_priv->nbox > MGA_NR_SAREA_CLIPRECTS)
+     		sarea_priv->nbox = MGA_NR_SAREA_CLIPRECTS;
+
+	/* Make sure we restore the 3D state next time.
+	 */
+	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;
+	mga_dma_schedule(dev, 1);
+      	sarea_priv->last_dispatch = status[1];
+   	return 0;
+}
+
+int mga_iload(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_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("Starting Iload\n");
+ 	copy_from_user_ret(&iload, (drm_mga_iload_t *)arg, sizeof(iload),
+ 			   -EFAULT);
+
+	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("mga_iload called without lock held\n");
+		return -EINVAL;
+	}
+
+   	buf = dma->buflist[ iload.idx ];
+	buf_priv = buf->dev_private;
+   	bus_address = buf->bus_address;
+   	DRM_DEBUG("bus_address %lx, length %d, destorg : %x\n",
+	       bus_address, iload.length, iload.destOrg);
+   
+	if(mgaVerifyIload(dev_priv, 
+			  bus_address,
+			  iload.destOrg,  
+			  iload.length)) {
+	      	mga_freelist_put(dev, buf);
+	   	return -EINVAL;
+	}
+    
+   	sarea_priv->dirty |= MGA_UPLOAD_CTX;
+
+   	mga_dma_dispatch_tex_blit(dev, bus_address, iload.length, 
+				  iload.destOrg);
+   	buf_priv->my_freelist->age = dev_priv->last_sync_tag;
+	buf_priv->discard = 1;
+    	mga_freelist_put(dev, buf);
+	mga_dma_schedule(dev, 1);
+    	sarea_priv->last_dispatch = status[1];
+ 	return 0; 
+}
+
+int mga_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_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;
+
+	copy_from_user_ret(&vertex, (drm_mga_vertex_t *)arg, sizeof(vertex),
+			   -EFAULT);
+   
+	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("mga_vertex called without lock held\n");
+		return -EINVAL;
+	}
+
+	DRM_DEBUG("mga_vertex\n");
+
+	buf = dma->buflist[ vertex.idx ];
+   	buf_priv = buf->dev_private;
+
+	buf->used = vertex.used;
+	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);
+	   }
+	   return -EINVAL;
+	}
+
+	mga_dma_dispatch_vertex(dev, buf);
+
+   	PRIMUPDATE(dev_priv);
+	mga_dma_schedule(dev, 1);
+      	sarea_priv->last_dispatch = status[1];
+   	return 0;
+}
+
+static int mga_dma_get_buffers(drm_device_t *dev, drm_dma_t *d)
+{
+	int i;
+	drm_buf_t *buf;
+
+	for (i = d->granted_count; i < d->request_count; i++) {
+		buf = mga_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;
+	}
+	return 0;
+}
+
+int mga_dma(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_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;
+
+   	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);
+
+	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("mga_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;
+
+	if (d.request_count) {
+		retcode = mga_dma_get_buffers(dev, &d);
+	}
+
+	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/mga_state.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_state.h
deleted file mode 100644
index e7b952e0c..000000000
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_state.h
+++ /dev/null
@@ -1,13 +0,0 @@
-#ifndef MGA_STATE_H
-#define MGA_STATE_H
-
-#include "mga_drv.h"
-
-int mgaCopyAndVerifyState( drm_mga_private_t *dev_priv, 
-			   drm_mga_buf_priv_t *buf_priv );
-
-void mgaEmitClipRect( drm_mga_private_t *dev_priv, xf86drmClipRectRec *box );
-
-void mgaEmitState( drm_mga_private_t *dev_priv, drm_mga_buf_priv_t *buf_priv );
-
-#endif
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mgareg_flags.h b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mgareg_flags.h
deleted file mode 100644
index 901f18310..000000000
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mgareg_flags.h
+++ /dev/null
@@ -1,930 +0,0 @@
-/* author: stephen crowley, crow@debian.org */
-
-/*
- * 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
- * STEPHEN CROWLEY, 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 _MGAREGS_H_
-#define _MGAREGS_H_
-
-/*************** (START) AUTOMATICLY GENERATED REGISTER FILE *****************/
-/*
- * Generated on Sat Nov 20 21:25:36 CST 1999
- */
-
-
-
-/*
- * Power Graphic Mode Memory Space Registers
- */
-
-    #define AGP_PLL_agp2xpllen_MASK 	0xfffffffe 	/* bit 0 */
-    #define AGP_PLL_agp2xpllen_disable 	0x0 		
-    #define AGP_PLL_agp2xpllen_enable 	0x1 		
-
-    #define AC_src_MASK 		0xfffffff0 	/* bits 0-3 */
-    #define AC_src_zero 		0x0 		/* val 0, shift 0 */
-    #define AC_src_one 			0x1 		/* val 1, shift 0 */
-    #define AC_src_dst_color 		0x2 		/* val 2, shift 0 */
-    #define AC_src_om_dst_color 	0x3 		/* val 3, shift 0 */
-    #define AC_src_src_alpha 		0x4 		/* val 4, shift 0 */
-    #define AC_src_om_src_alpha 	0x5 		/* val 5, shift 0 */
-    #define AC_src_dst_alpha 		0x6 		/* val 6, shift 0 */
-    #define AC_src_om_dst_alpha 	0x7 		/* val 7, shift 0 */
-    #define AC_src_src_alpha_sat 	0x8 		/* val 8, shift 0 */
-    #define AC_dst_MASK 		0xffffff0f 	/* bits 4-7 */
-    #define AC_dst_zero 		0x0 		/* val 0, shift 4 */
-    #define AC_dst_one 			0x10 		/* val 1, shift 4 */
-    #define AC_dst_src_color 		0x20 		/* val 2, shift 4 */
-    #define AC_dst_om_src_color 	0x30 		/* val 3, shift 4 */
-    #define AC_dst_src_alpha 		0x40 		/* val 4, shift 4 */
-    #define AC_dst_om_src_alpha 	0x50 		/* val 5, shift 4 */
-    #define AC_dst_dst_alpha 		0x60 		/* val 6, shift 4 */
-    #define AC_dst_om_dst_alpha 	0x70 		/* val 7, shift 4 */
-    #define AC_amode_MASK 		0xfffffcff 	/* bits 8-9 */
-    #define AC_amode_FCOL 		0x0 		/* val 0, shift 8 */
-    #define AC_amode_alpha_channel 	0x100 		/* val 1, shift 8 */
-    #define AC_amode_video_alpha 	0x200 		/* val 2, shift 8 */
-    #define AC_amode_RSVD 		0x300 		/* val 3, shift 8 */
-    #define AC_astipple_MASK 		0xfffff7ff 	/* bit 11 */
-    #define AC_astipple_disable 	0x0 		
-    #define AC_astipple_enable 		0x800 		
-    #define AC_aten_MASK 		0xffffefff 	/* bit 12 */
-    #define AC_aten_disable 		0x0 		
-    #define AC_aten_enable 		0x1000 		
-    #define AC_atmode_MASK 		0xffff1fff 	/* bits 13-15 */
-    #define AC_atmode_noacmp 		0x0 		/* val 0, shift 13 */
-    #define AC_atmode_ae 		0x4000 		/* val 2, shift 13 */
-    #define AC_atmode_ane 		0x6000 		/* val 3, shift 13 */
-    #define AC_atmode_alt 		0x8000 		/* val 4, shift 13 */
-    #define AC_atmode_alte 		0xa000 		/* val 5, shift 13 */
-    #define AC_atmode_agt 		0xc000 		/* val 6, shift 13 */
-    #define AC_atmode_agte 		0xe000 		/* val 7, shift 13 */
-    #define AC_atref_MASK 		0xff00ffff 	/* bits 16-23 */
-    #define AC_atref_SHIFT 		16 		
-    #define AC_alphasel_MASK 		0xfcffffff 	/* bits 24-25 */
-    #define AC_alphasel_fromtex 	0x0 		/* val 0, shift 24 */
-    #define AC_alphasel_diffused 	0x1000000 	/* val 1, shift 24 */
-    #define AC_alphasel_modulated 	0x2000000 	/* val 2, shift 24 */
-    #define AC_alphasel_trans 		0x3000000 	/* val 3, shift 24 */
-
-    #define AR0_ar0_MASK 		0xfffc0000 	/* bits 0-17 */
-    #define AR0_ar0_SHIFT 		0 		
-
-    #define AR1_ar1_MASK 		0xff000000 	/* bits 0-23 */
-    #define AR1_ar1_SHIFT 		0 		
-
-    #define AR2_ar2_MASK 		0xfffc0000 	/* bits 0-17 */
-    #define AR2_ar2_SHIFT 		0 		
-
-    #define AR3_ar3_MASK 		0xff000000 	/* bits 0-23 */
-    #define AR3_ar3_SHIFT 		0 		
-    #define AR3_spage_MASK 		0xf8ffffff 	/* bits 24-26 */
-    #define AR3_spage_SHIFT 		24 		
-
-    #define AR4_ar4_MASK 		0xfffc0000 	/* bits 0-17 */
-    #define AR4_ar4_SHIFT 		0 		
-
-    #define AR5_ar5_MASK 		0xfffc0000 	/* bits 0-17 */
-    #define AR5_ar5_SHIFT 		0 		
-
-    #define AR6_ar6_MASK 		0xfffc0000 	/* bits 0-17 */
-    #define AR6_ar6_SHIFT 		0 		
-
-    #define BC_besen_MASK 		0xfffffffe 	/* bit 0 */
-    #define BC_besen_disable 		0x0 		
-    #define BC_besen_enable 		0x1 		
-    #define BC_besv1srcstp_MASK 	0xffffffbf 	/* bit 6 */
-    #define BC_besv1srcstp_even 	0x0 		
-    #define BC_besv1srcstp_odd 		0x40 		
-    #define BC_besv2srcstp_MASK 	0xfffffeff 	/* bit 8 */
-    #define BC_besv2srcstp_disable 	0x0 		
-    #define BC_besv2srcstp_enable 	0x100 		
-    #define BC_beshfen_MASK 		0xfffffbff 	/* bit 10 */
-    #define BC_beshfen_disable 		0x0 		
-    #define BC_beshfen_enable 		0x400 		
-    #define BC_besvfen_MASK 		0xfffff7ff 	/* bit 11 */
-    #define BC_besvfen_disable 		0x0 		
-    #define BC_besvfen_enable 		0x800 		
-    #define BC_beshfixc_MASK 		0xffffefff 	/* bit 12 */
-    #define BC_beshfixc_weight 		0x0 		
-    #define BC_beshfixc_coeff 		0x1000 		
-    #define BC_bescups_MASK 		0xfffeffff 	/* bit 16 */
-    #define BC_bescups_disable 		0x0 		
-    #define BC_bescups_enable 		0x10000 	
-    #define BC_bes420pl_MASK 		0xfffdffff 	/* bit 17 */
-    #define BC_bes420pl_422 		0x0 		
-    #define BC_bes420pl_420 		0x20000 	
-    #define BC_besdith_MASK 		0xfffbffff 	/* bit 18 */
-    #define BC_besdith_disable 		0x0 		
-    #define BC_besdith_enable 		0x40000 	
-    #define BC_beshmir_MASK 		0xfff7ffff 	/* bit 19 */
-    #define BC_beshmir_disable 		0x0 		
-    #define BC_beshmir_enable 		0x80000 	
-    #define BC_besbwen_MASK 		0xffefffff 	/* bit 20 */
-    #define BC_besbwen_color 		0x0 		
-    #define BC_besbwen_bw 		0x100000 	
-    #define BC_besblank_MASK 		0xffdfffff 	/* bit 21 */
-    #define BC_besblank_disable 	0x0 		
-    #define BC_besblank_enable 		0x200000 	
-    #define BC_besfselm_MASK 		0xfeffffff 	/* bit 24 */
-    #define BC_besfselm_soft 		0x0 		
-    #define BC_besfselm_hard 		0x1000000 	
-    #define BC_besfsel_MASK 		0xf9ffffff 	/* bits 25-26 */
-    #define BC_besfsel_a1 		0x0 		/* val 0, shift 25 */
-    #define BC_besfsel_a2 		0x2000000 	/* val 1, shift 25 */
-    #define BC_besfsel_b1 		0x4000000 	/* val 2, shift 25 */
-    #define BC_besfsel_b2 		0x6000000 	/* val 3, shift 25 */
-
-    #define BGC_beshzoom_MASK 		0xfffffffe 	/* bit 0 */
-    #define BGC_beshzoom_disable 	0x0 		
-    #define BGC_beshzoom_enable 	0x1 		
-    #define BGC_beshzoomf_MASK 		0xfffffffd 	/* bit 1 */
-    #define BGC_beshzoomf_disable 	0x0 		
-    #define BGC_beshzoomf_enable 	0x2 		
-    #define BGC_bescorder_MASK 		0xfffffff7 	/* bit 3 */
-    #define BGC_bescorder_even 		0x0 		
-    #define BGC_bescorder_odd 		0x8 		
-    #define BGC_besreghup_MASK 		0xffffffef 	/* bit 4 */
-    #define BGC_besreghup_disable 	0x0 		
-    #define BGC_besreghup_enable 	0x10 		
-    #define BGC_besvcnt_MASK 		0xf000ffff 	/* bits 16-27 */
-    #define BGC_besvcnt_SHIFT 		16 		
-
-    #define BHC_besright_MASK 		0xfffff800 	/* bits 0-10 */
-    #define BHC_besright_SHIFT 		0 		
-    #define BHC_besleft_MASK 		0xf800ffff 	/* bits 16-26 */
-    #define BHC_besleft_SHIFT 		16 		
-
-    #define BHISF_beshiscal_MASK 	0xffe00003 	/* bits 2-20 */
-    #define BHISF_beshiscal_SHIFT 	2 		
-
-    #define BHSE_beshsrcend_MASK 	0xfc000003 	/* bits 2-25 */
-    #define BHSE_beshsrcend_SHIFT 	2 		
-
-    #define BHSL_beshsrclst_MASK 	0xfc00ffff 	/* bits 16-25 */
-    #define BHSL_beshsrclst_SHIFT 	16 		
-
-    #define BHSS_beshsrcst_MASK 	0xfc000003 	/* bits 2-25 */
-    #define BHSS_beshsrcst_SHIFT 	2 		
-
-    #define BP_bespitch_MASK 		0xfffff000 	/* bits 0-11 */
-    #define BP_bespitch_SHIFT 		0 		
-
-    #define BS_besstat_MASK 		0xfffffffc 	/* bits 0-1 */
-    #define BS_besstat_a1 		0x0 		/* val 0, shift 0 */
-    #define BS_besstat_a2 		0x1 		/* val 1, shift 0 */
-    #define BS_besstat_b1 		0x2 		/* val 2, shift 0 */
-    #define BS_besstat_b2 		0x3 		/* val 3, shift 0 */
-
-    #define BSF_besv1srclast_MASK 	0xfffffc00 	/* bits 0-9 */
-    #define BSF_besv1srclast_SHIFT 	0 		
-
-    #define BSF_besv2srclst_MASK 	0xfffffc00 	/* bits 0-9 */
-    #define BSF_besv2srclst_SHIFT 	0 		
-
-    #define BSF_besv1wght_MASK 		0xffff0003 	/* bits 2-15 */
-    #define BSF_besv1wght_SHIFT 	2 		
-    #define BSF_besv1wghts_MASK 	0xfffeffff 	/* bit 16 */
-    #define BSF_besv1wghts_disable 	0x0 		
-    #define BSF_besv1wghts_enable 	0x10000 	
-
-    #define BSF_besv2wght_MASK 		0xffff0003 	/* bits 2-15 */
-    #define BSF_besv2wght_SHIFT 	2 		
-    #define BSF_besv2wghts_MASK 	0xfffeffff 	/* bit 16 */
-    #define BSF_besv2wghts_disable 	0x0 		
-    #define BSF_besv2wghts_enable 	0x10000 	
-
-    #define BVC_besbot_MASK 		0xfffff800 	/* bits 0-10 */
-    #define BVC_besbot_SHIFT 		0 		
-    #define BVC_bestop_MASK 		0xf800ffff 	/* bits 16-26 */
-    #define BVC_bestop_SHIFT 		16 		
-
-    #define BVISF_besviscal_MASK 	0xffe00003 	/* bits 2-20 */
-    #define BVISF_besviscal_SHIFT 	2 		
-
-    #define CXB_cxleft_MASK 		0xfffff000 	/* bits 0-11 */
-    #define CXB_cxleft_SHIFT 		0 		
-    #define CXB_cxright_MASK 		0xf000ffff 	/* bits 16-27 */
-    #define CXB_cxright_SHIFT 		16 		
-
-    #define DO_dstmap_MASK 		0xfffffffe 	/* bit 0 */
-    #define DO_dstmap_fb 		0x0 		
-    #define DO_dstmap_sys 		0x1 		
-    #define DO_dstacc_MASK 		0xfffffffd 	/* bit 1 */
-    #define DO_dstacc_pci 		0x0 		
-    #define DO_dstacc_agp 		0x2 		
-    #define DO_dstorg_MASK 		0x7 		/* bits 3-31 */
-    #define DO_dstorg_SHIFT 		3 		
-
-    #define DC_opcod_MASK 		0xfffffff0 	/* bits 0-3 */
-    #define DC_opcod_line_open 		0x0 		/* val 0, shift 0 */
-    #define DC_opcod_autoline_open 	0x1 		/* val 1, shift 0 */
-    #define DC_opcod_line_close 	0x2 		/* val 2, shift 0 */
-    #define DC_opcod_autoline_close 	0x3 		/* val 3, shift 0 */
-    #define DC_opcod_trap 		0x4 		/* val 4, shift 0 */
-    #define DC_opcod_texture_trap 	0x6 		/* val 6, shift 0 */
-    #define DC_opcod_bitblt 		0x8 		/* val 8, shift 0 */
-    #define DC_opcod_iload 		0x9 		/* val 9, shift 0 */
-    #define DC_atype_MASK 		0xffffff8f 	/* bits 4-6 */
-    #define DC_atype_rpl 		0x0 		/* val 0, shift 4 */
-    #define DC_atype_rstr 		0x10 		/* val 1, shift 4 */
-    #define DC_atype_zi 		0x30 		/* val 3, shift 4 */
-    #define DC_atype_blk 		0x40 		/* val 4, shift 4 */
-    #define DC_atype_i 			0x70 		/* val 7, shift 4 */
-    #define DC_linear_MASK 		0xffffff7f 	/* bit 7 */
-    #define DC_linear_xy 		0x0 		
-    #define DC_linear_linear 		0x80 		
-    #define DC_zmode_MASK 		0xfffff8ff 	/* bits 8-10 */
-    #define DC_zmode_nozcmp 		0x0 		/* val 0, shift 8 */
-    #define DC_zmode_ze 		0x200 		/* val 2, shift 8 */
-    #define DC_zmode_zne 		0x300 		/* val 3, shift 8 */
-    #define DC_zmode_zlt 		0x400 		/* val 4, shift 8 */
-    #define DC_zmode_zlte 		0x500 		/* val 5, shift 8 */
-    #define DC_zmode_zgt 		0x600 		/* val 6, shift 8 */
-    #define DC_zmode_zgte 		0x700 		/* val 7, shift 8 */
-    #define DC_solid_MASK 		0xfffff7ff 	/* bit 11 */
-    #define DC_solid_disable 		0x0 		
-    #define DC_solid_enable 		0x800 		
-    #define DC_arzero_MASK 		0xffffefff 	/* bit 12 */
-    #define DC_arzero_disable 		0x0 		
-    #define DC_arzero_enable 		0x1000 		
-    #define DC_sgnzero_MASK 		0xffffdfff 	/* bit 13 */
-    #define DC_sgnzero_disable 		0x0 		
-    #define DC_sgnzero_enable 		0x2000 		
-    #define DC_shftzero_MASK 		0xffffbfff 	/* bit 14 */
-    #define DC_shftzero_disable 	0x0 		
-    #define DC_shftzero_enable 		0x4000 		
-    #define DC_bop_MASK 		0xfff0ffff 	/* bits 16-19 */
-    #define DC_bop_SHIFT 		16 		
-    #define DC_trans_MASK 		0xff0fffff 	/* bits 20-23 */
-    #define DC_trans_SHIFT 		20 		
-    #define DC_bltmod_MASK 		0xe1ffffff 	/* bits 25-28 */
-    #define DC_bltmod_bmonolef 		0x0 		/* val 0, shift 25 */
-    #define DC_bltmod_bmonowf 		0x8000000 	/* val 4, shift 25 */
-    #define DC_bltmod_bplan 		0x2000000 	/* val 1, shift 25 */
-    #define DC_bltmod_bfcol 		0x4000000 	/* val 2, shift 25 */
-    #define DC_bltmod_bu32bgr 		0x6000000 	/* val 3, shift 25 */
-    #define DC_bltmod_bu32rgb 		0xe000000 	/* val 7, shift 25 */
-    #define DC_bltmod_bu24bgr 		0x16000000 	/* val 11, shift 25 */
-    #define DC_bltmod_bu24rgb 		0x1e000000 	/* val 15, shift 25 */
-    #define DC_pattern_MASK 		0xdfffffff 	/* bit 29 */
-    #define DC_pattern_disable 		0x0 		
-    #define DC_pattern_enable 		0x20000000 	
-    #define DC_transc_MASK 		0xbfffffff 	/* bit 30 */
-    #define DC_transc_disable 		0x0 		
-    #define DC_transc_enable 		0x40000000 	
-    #define DC_clipdis_MASK 		0x7fffffff 	/* bit 31 */
-    #define DC_clipdis_disable 		0x0 		
-    #define DC_clipdis_enable 		0x80000000 	
-
-    #define DS_dwgsyncaddr_MASK 	0x3 		/* bits 2-31 */
-    #define DS_dwgsyncaddr_SHIFT 	2 		
-
-    #define FS_fifocount_MASK 		0xffffff80 	/* bits 0-6 */
-    #define FS_fifocount_SHIFT 		0 		
-    #define FS_bfull_MASK 		0xfffffeff 	/* bit 8 */
-    #define FS_bfull_disable 		0x0 		
-    #define FS_bfull_enable 		0x100 		
-    #define FS_bempty_MASK 		0xfffffdff 	/* bit 9 */
-    #define FS_bempty_disable 		0x0 		
-    #define FS_bempty_enable 		0x200 		
-
-    #define XA_fxleft_MASK 		0xffff0000 	/* bits 0-15 */
-    #define XA_fxleft_SHIFT 		0 		
-    #define XA_fxright_MASK 		0xffff 		/* bits 16-31 */
-    #define XA_fxright_SHIFT 		16 		
-
-    #define IC_softrapiclr_MASK 	0xfffffffe 	/* bit 0 */
-    #define IC_softrapiclr_disable 	0x0 		
-    #define IC_softrapiclr_enable 	0x1 		
-    #define IC_pickiclr_MASK 		0xfffffffb 	/* bit 2 */
-    #define IC_pickiclr_disable 	0x0 		
-    #define IC_pickiclr_enable 		0x4 		
-    #define IC_vlineiclr_MASK 		0xffffffdf 	/* bit 5 */
-    #define IC_vlineiclr_disable 	0x0 		
-    #define IC_vlineiclr_enable 	0x20 		
-    #define IC_wiclr_MASK 		0xffffff7f 	/* bit 7 */
-    #define IC_wiclr_disable 		0x0 		
-    #define IC_wiclr_enable 		0x80 		
-    #define IC_wciclr_MASK 		0xfffffeff 	/* bit 8 */
-    #define IC_wciclr_disable 		0x0 		
-    #define IC_wciclr_enable 		0x100 		
-
-    #define IE_softrapien_MASK 		0xfffffffe 	/* bit 0 */
-    #define IE_softrapien_disable 	0x0 		
-    #define IE_softrapien_enable 	0x1 		
-    #define IE_pickien_MASK 		0xfffffffb 	/* bit 2 */
-    #define IE_pickien_disable 		0x0 		
-    #define IE_pickien_enable 		0x4 		
-    #define IE_vlineien_MASK 		0xffffffdf 	/* bit 5 */
-    #define IE_vlineien_disable 	0x0 		
-    #define IE_vlineien_enable 		0x20 		
-    #define IE_extien_MASK 		0xffffffbf 	/* bit 6 */
-    #define IE_extien_disable 		0x0 		
-    #define IE_extien_enable 		0x40 		
-    #define IE_wien_MASK 		0xffffff7f 	/* bit 7 */
-    #define IE_wien_disable 		0x0 		
-    #define IE_wien_enable 		0x80 		
-    #define IE_wcien_MASK 		0xfffffeff 	/* bit 8 */
-    #define IE_wcien_disable 		0x0 		
-    #define IE_wcien_enable 		0x100 		
-
-    #define MA_pwidth_MASK 		0xfffffffc 	/* bits 0-1 */
-    #define MA_pwidth_8 		0x0 		/* val 0, shift 0 */
-    #define MA_pwidth_16 		0x1 		/* val 1, shift 0 */
-    #define MA_pwidth_32 		0x2 		/* val 2, shift 0 */
-    #define MA_pwidth_24 		0x3 		/* val 3, shift 0 */
-    #define MA_zwidth_MASK 		0xffffffe7 	/* bits 3-4 */
-    #define MA_zwidth_16 		0x0 		/* val 0, shift 3 */
-    #define MA_zwidth_32 		0x8 		/* val 1, shift 3 */
-    #define MA_zwidth_15 		0x10 		/* val 2, shift 3 */
-    #define MA_zwidth_24 		0x18 		/* val 3, shift 3 */
-    #define MA_memreset_MASK 		0xffff7fff 	/* bit 15 */
-    #define MA_memreset_disable 	0x0 		
-    #define MA_memreset_enable 		0x8000 		
-    #define MA_fogen_MASK 		0xfbffffff 	/* bit 26 */
-    #define MA_fogen_disable 		0x0 		
-    #define MA_fogen_enable 		0x4000000 	
-    #define MA_tlutload_MASK 		0xdfffffff 	/* bit 29 */
-    #define MA_tlutload_disable 	0x0 		
-    #define MA_tlutload_enable 		0x20000000 	
-    #define MA_nodither_MASK 		0xbfffffff 	/* bit 30 */
-    #define MA_nodither_disable 	0x0 		
-    #define MA_nodither_enable 		0x40000000 	
-    #define MA_dit555_MASK 		0x7fffffff 	/* bit 31 */
-    #define MA_dit555_disable 		0x0 		
-    #define MA_dit555_enable 		0x80000000 	
-
-    #define MCWS_casltncy_MASK 		0xfffffff8 	/* bits 0-2 */
-    #define MCWS_casltncy_SHIFT 	0 		
-    #define MCWS_rrddelay_MASK 		0xffffffcf 	/* bits 4-5 */
-    #define MCWS_rcddelay_MASK 		0xfffffe7f 	/* bits 7-8 */
-    #define MCWS_rasmin_MASK 		0xffffe3ff 	/* bits 10-12 */
-    #define MCWS_rasmin_SHIFT 		10 		
-    #define MCWS_rpdelay_MASK 		0xffff3fff 	/* bits 14-15 */
-    #define MCWS_wrdelay_MASK 		0xfff3ffff 	/* bits 18-19 */
-    #define MCWS_rddelay_MASK 		0xffdfffff 	/* bit 21 */
-    #define MCWS_rddelay_disable 	0x0 		
-    #define MCWS_rddelay_enable 	0x200000 	
-    #define MCWS_smrdelay_MASK 		0xfe7fffff 	/* bits 23-24 */
-    #define MCWS_bwcdelay_MASK 		0xf3ffffff 	/* bits 26-27 */
-    #define MCWS_bpldelay_MASK 		0x1fffffff 	/* bits 29-31 */
-    #define MCWS_bpldelay_SHIFT 	29 		
-
-    #define MRB_mclkbrd0_MASK 		0xfffffff0 	/* bits 0-3 */
-    #define MRB_mclkbrd0_SHIFT 		0 		
-    #define MRB_mclkbrd1_MASK 		0xfffffe1f 	/* bits 5-8 */
-    #define MRB_mclkbrd1_SHIFT 		5 		
-    #define MRB_strmfctl_MASK 		0xff3fffff 	/* bits 22-23 */
-    #define MRB_mrsopcod_MASK 		0xe1ffffff 	/* bits 25-28 */
-    #define MRB_mrsopcod_SHIFT 		25 		
-
-    #define OM_dmamod_MASK 		0xfffffff3 	/* bits 2-3 */
-    #define OM_dmamod_general 		0x0 		/* val 0, shift 2 */
-    #define OM_dmamod_blit 		0x4 		/* val 1, shift 2 */
-    #define OM_dmamod_vector 		0x8 		/* val 2, shift 2 */
-    #define OM_dmamod_vertex 		0xc 		/* val 3, shift 2 */
-    #define OM_dmadatasiz_MASK 		0xfffffcff 	/* bits 8-9 */
-    #define OM_dmadatasiz_8 		0x0 		/* val 0, shift 8 */
-    #define OM_dmadatasiz_16 		0x100 		/* val 1, shift 8 */
-    #define OM_dmadatasiz_32 		0x200 		/* val 2, shift 8 */
-    #define OM_dirdatasiz_MASK 		0xfffcffff 	/* bits 16-17 */
-    #define OM_dirdatasiz_8 		0x0 		/* val 0, shift 16 */
-    #define OM_dirdatasiz_16 		0x10000 	/* val 1, shift 16 */
-    #define OM_dirdatasiz_32 		0x20000 	/* val 2, shift 16 */
-
-    #define P_iy_MASK 			0xffffe000 	/* bits 0-12 */
-    #define P_iy_SHIFT 			0 		
-    #define P_ylin_MASK 		0xffff7fff 	/* bit 15 */
-    #define P_ylin_disable 		0x0 		
-    #define P_ylin_enable 		0x8000 		
-
-    #define PDCA_primod_MASK 		0xfffffffc 	/* bits 0-1 */
-    #define PDCA_primod_general 	0x0 		/* val 0, shift 0 */
-    #define PDCA_primod_blit 		0x1 		/* val 1, shift 0 */
-    #define PDCA_primod_vector 		0x2 		/* val 2, shift 0 */
-    #define PDCA_primod_vertex 		0x3 		/* val 3, shift 0 */
-    #define PDCA_primaddress_MASK 	0x3 		/* bits 2-31 */
-    #define PDCA_primaddress_SHIFT 	2 		
-
-    #define PDEA_primnostart_MASK 	0xfffffffe 	/* bit 0 */
-    #define PDEA_primnostart_disable 	0x0 		
-    #define PDEA_primnostart_enable 	0x1 		
-    #define PDEA_pagpxfer_MASK 		0xfffffffd 	/* bit 1 */
-    #define PDEA_pagpxfer_disable 	0x0 		
-    #define PDEA_pagpxfer_enable 	0x2 		
-    #define PDEA_primend_MASK 		0x3 		/* bits 2-31 */
-    #define PDEA_primend_SHIFT 		2 		
-
-    #define PLS_primptren0_MASK 	0xfffffffe 	/* bit 0 */
-    #define PLS_primptren0_disable 	0x0 		
-    #define PLS_primptren0_enable 	0x1 		
-    #define PLS_primptren1_MASK 	0xfffffffd 	/* bit 1 */
-    #define PLS_primptren1_disable 	0x0 		
-    #define PLS_primptren1_enable 	0x2 		
-    #define PLS_primptr_MASK 		0x7 		/* bits 3-31 */
-    #define PLS_primptr_SHIFT 		3 		
-
-    #define R_softreset_MASK 		0xfffffffe 	/* bit 0 */
-    #define R_softreset_disable 	0x0 		
-    #define R_softreset_enable 		0x1 		
-    #define R_softextrst_MASK 		0xfffffffd 	/* bit 1 */
-    #define R_softextrst_disable 	0x0 		
-    #define R_softextrst_enable 	0x2 		
-
-    #define SDCA_secmod_MASK 		0xfffffffc 	/* bits 0-1 */
-    #define SDCA_secmod_general 	0x0 		/* val 0, shift 0 */
-    #define SDCA_secmod_blit 		0x1 		/* val 1, shift 0 */
-    #define SDCA_secmod_vector 		0x2 		/* val 2, shift 0 */
-    #define SDCA_secmod_vertex 		0x3 		/* val 3, shift 0 */
-    #define SDCA_secaddress_MASK 	0x3 		/* bits 2-31 */
-    #define SDCA_secaddress_SHIFT 	2 		
-
-    #define SDEA_sagpxfer_MASK 		0xfffffffd 	/* bit 1 */
-    #define SDEA_sagpxfer_disable 	0x0 		
-    #define SDEA_sagpxfer_enable 	0x2 		
-    #define SDEA_secend_MASK 		0x3 		/* bits 2-31 */
-    #define SDEA_secend_SHIFT 		2 		
-
-    #define SETDCA_setupmod_MASK 	0xfffffffc 	/* bits 0-1 */
-    #define SETDCA_setupmod_vertlist 	0x0 		/* val 0, shift 0 */
-    #define SETDCA_setupaddress_MASK 	0x3 		/* bits 2-31 */
-    #define SETDCA_setupaddress_SHIFT 	2 		
-
-    #define SETDEA_setupagpxfer_MASK 	0xfffffffd 	/* bit 1 */
-    #define SETDEA_setupagpxfer_disable 0x0 		
-    #define SETDEA_setupagpxfer_enable 	0x2 		
-    #define SETDEA_setupend_MASK 	0x3 		/* bits 2-31 */
-    #define SETDEA_setupend_SHIFT 	2 		
-
-    #define S_sdydxl_MASK 		0xfffffffe 	/* bit 0 */
-    #define S_sdydxl_y 			0x0 		
-    #define S_sdydxl_x 			0x1 		
-    #define S_scanleft_MASK 		0xfffffffe 	/* bit 0 */
-    #define S_scanleft_disable 		0x0 		
-    #define S_scanleft_enable 		0x1 		
-    #define S_sdxl_MASK 		0xfffffffd 	/* bit 1 */
-    #define S_sdxl_pos 			0x0 		
-    #define S_sdxl_neg 			0x2 		
-    #define S_sdy_MASK 			0xfffffffb 	/* bit 2 */
-    #define S_sdy_pos 			0x0 		
-    #define S_sdy_neg 			0x4 		
-    #define S_sdxr_MASK 		0xffffffdf 	/* bit 5 */
-    #define S_sdxr_pos 			0x0 		
-    #define S_sdxr_neg 			0x20 		
-    #define S_brkleft_MASK 		0xfffffeff 	/* bit 8 */
-    #define S_brkleft_disable 		0x0 		
-    #define S_brkleft_enable 		0x100 		
-    #define S_errorinit_MASK 		0x7fffffff 	/* bit 31 */
-    #define S_errorinit_disable 	0x0 		
-    #define S_errorinit_enable 		0x80000000 	
-
-    #define FSC_x_off_MASK 		0xfffffff0 	/* bits 0-3 */
-    #define FSC_x_off_SHIFT 		0 		
-    #define FSC_funcnt_MASK 		0xffffff80 	/* bits 0-6 */
-    #define FSC_funcnt_SHIFT 		0 		
-    #define FSC_y_off_MASK 		0xffffff8f 	/* bits 4-6 */
-    #define FSC_y_off_SHIFT 		4 		
-    #define FSC_funoff_MASK 		0xffc0ffff 	/* bits 16-21 */
-    #define FSC_funoff_SHIFT 		16 		
-    #define FSC_stylelen_MASK 		0xffc0ffff 	/* bits 16-21 */
-    #define FSC_stylelen_SHIFT 		16 		
-
-
-    #define STH_softraphand_MASK 	0x3 		/* bits 2-31 */
-    #define STH_softraphand_SHIFT 	2 		
-
-    #define SO_srcmap_MASK 		0xfffffffe 	/* bit 0 */
-    #define SO_srcmap_fb 		0x0 		
-    #define SO_srcmap_sys 		0x1 		
-    #define SO_srcacc_MASK 		0xfffffffd 	/* bit 1 */
-    #define SO_srcacc_pci 		0x0 		
-    #define SO_srcacc_agp 		0x2 		
-    #define SO_srcorg_MASK 		0x7 		/* bits 3-31 */
-    #define SO_srcorg_SHIFT 		3 		
-
-    #define STAT_softrapen_MASK 	0xfffffffe 	/* bit 0 */
-    #define STAT_softrapen_disable 	0x0 		
-    #define STAT_softrapen_enable 	0x1 		
-    #define STAT_pickpen_MASK 		0xfffffffb 	/* bit 2 */
-    #define STAT_pickpen_disable 	0x0 		
-    #define STAT_pickpen_enable 	0x4 		
-    #define STAT_vsyncsts_MASK 		0xfffffff7 	/* bit 3 */
-    #define STAT_vsyncsts_disable 	0x0 		
-    #define STAT_vsyncsts_enable 	0x8 		
-    #define STAT_vsyncpen_MASK 		0xffffffef 	/* bit 4 */
-    #define STAT_vsyncpen_disable 	0x0 		
-    #define STAT_vsyncpen_enable 	0x10 		
-    #define STAT_vlinepen_MASK 		0xffffffdf 	/* bit 5 */
-    #define STAT_vlinepen_disable 	0x0 		
-    #define STAT_vlinepen_enable 	0x20 		
-    #define STAT_extpen_MASK 		0xffffffbf 	/* bit 6 */
-    #define STAT_extpen_disable 	0x0 		
-    #define STAT_extpen_enable 		0x40 		
-    #define STAT_wpen_MASK 		0xffffff7f 	/* bit 7 */
-    #define STAT_wpen_disable 		0x0 		
-    #define STAT_wpen_enable 		0x80 		
-    #define STAT_wcpen_MASK 		0xfffffeff 	/* bit 8 */
-    #define STAT_wcpen_disable 		0x0 		
-    #define STAT_wcpen_enable 		0x100 		
-    #define STAT_dwgengsts_MASK 	0xfffeffff 	/* bit 16 */
-    #define STAT_dwgengsts_disable 	0x0 		
-    #define STAT_dwgengsts_enable 	0x10000 	
-    #define STAT_endprdmasts_MASK 	0xfffdffff 	/* bit 17 */
-    #define STAT_endprdmasts_disable 	0x0 		
-    #define STAT_endprdmasts_enable 	0x20000 	
-    #define STAT_wbusy_MASK 		0xfffbffff 	/* bit 18 */
-    #define STAT_wbusy_disable 		0x0 		
-    #define STAT_wbusy_enable 		0x40000 	
-    #define STAT_swflag_MASK 		0xfffffff 	/* bits 28-31 */
-    #define STAT_swflag_SHIFT 		28 		
-
-    #define S_sref_MASK 		0xffffff00 	/* bits 0-7 */
-    #define S_sref_SHIFT 		0 		
-    #define S_smsk_MASK 		0xffff00ff 	/* bits 8-15 */
-    #define S_smsk_SHIFT 		8 		
-    #define S_swtmsk_MASK 		0xff00ffff 	/* bits 16-23 */
-    #define S_swtmsk_SHIFT 		16 		
-
-    #define SC_smode_MASK 		0xfffffff8 	/* bits 0-2 */
-    #define SC_smode_salways 		0x0 		/* val 0, shift 0 */
-    #define SC_smode_snever 		0x1 		/* val 1, shift 0 */
-    #define SC_smode_se 		0x2 		/* val 2, shift 0 */
-    #define SC_smode_sne 		0x3 		/* val 3, shift 0 */
-    #define SC_smode_slt 		0x4 		/* val 4, shift 0 */
-    #define SC_smode_slte 		0x5 		/* val 5, shift 0 */
-    #define SC_smode_sgt 		0x6 		/* val 6, shift 0 */
-    #define SC_smode_sgte 		0x7 		/* val 7, shift 0 */
-    #define SC_sfailop_MASK 		0xffffffc7 	/* bits 3-5 */
-    #define SC_sfailop_keep 		0x0 		/* val 0, shift 3 */
-    #define SC_sfailop_zero 		0x8 		/* val 1, shift 3 */
-    #define SC_sfailop_replace 		0x10 		/* val 2, shift 3 */
-    #define SC_sfailop_incrsat 		0x18 		/* val 3, shift 3 */
-    #define SC_sfailop_decrsat 		0x20 		/* val 4, shift 3 */
-    #define SC_sfailop_invert 		0x28 		/* val 5, shift 3 */
-    #define SC_sfailop_incr 		0x30 		/* val 6, shift 3 */
-    #define SC_sfailop_decr 		0x38 		/* val 7, shift 3 */
-    #define SC_szfailop_MASK 		0xfffffe3f 	/* bits 6-8 */
-    #define SC_szfailop_keep 		0x0 		/* val 0, shift 6 */
-    #define SC_szfailop_zero 		0x40 		/* val 1, shift 6 */
-    #define SC_szfailop_replace 	0x80 		/* val 2, shift 6 */
-    #define SC_szfailop_incrsat 	0xc0 		/* val 3, shift 6 */
-    #define SC_szfailop_decrsat 	0x100 		/* val 4, shift 6 */
-    #define SC_szfailop_invert 		0x140 		/* val 5, shift 6 */
-    #define SC_szfailop_incr 		0x180 		/* val 6, shift 6 */
-    #define SC_szfailop_decr 		0x1c0 		/* val 7, shift 6 */
-    #define SC_szpassop_MASK 		0xfffff1ff 	/* bits 9-11 */
-    #define SC_szpassop_keep 		0x0 		/* val 0, shift 9 */
-    #define SC_szpassop_zero 		0x200 		/* val 1, shift 9 */
-    #define SC_szpassop_replace 	0x400 		/* val 2, shift 9 */
-    #define SC_szpassop_incrsat 	0x600 		/* val 3, shift 9 */
-    #define SC_szpassop_decrsat 	0x800 		/* val 4, shift 9 */
-    #define SC_szpassop_invert 		0xa00 		/* val 5, shift 9 */
-    #define SC_szpassop_incr 		0xc00 		/* val 6, shift 9 */
-    #define SC_szpassop_decr 		0xe00 		/* val 7, shift 9 */
-
-    #define TD1_color1arg2selMASK 	0xfffffffc 	/* bits 0-1 */
-    #define TD1_color1alphaselMASK 	0xffffffe3 	/* bits 2-4 */
-    #define TD1_color1alphaselSHIFT 	2 		
-    #define TD1_color1arg1alphaMASK 	0xffffffdf 	/* bit 5 */
-    #define TD1_color1arg1alphadisable 	0x0 		
-    #define TD1_color1arg1alphaenable 	0x20 		
-    #define TD1_color1arg1invMASK 	0xffffffbf 	/* bit 6 */
-    #define TD1_color1arg1invdisable 	0x0 		
-    #define TD1_color1arg1invenable 	0x40 		
-    #define TD1_color1arg2alphaMASK 	0xffffff7f 	/* bit 7 */
-    #define TD1_color1arg2alphadisable 	0x0 		
-    #define TD1_color1arg2alphaenable 	0x80 		
-    #define TD1_color1arg2invMASK 	0xfffffeff 	/* bit 8 */
-    #define TD1_color1arg2invdisable 	0x0 		
-    #define TD1_color1arg2invenable 	0x100 		
-    #define TD1_color1alpha1invMASK 	0xfffffdff 	/* bit 9 */
-    #define TD1_color1alpha1invdisable 	0x0 		
-    #define TD1_color1alpha1invenable 	0x200 		
-    #define TD1_color1alpha2invMASK 	0xfffffbff 	/* bit 10 */
-    #define TD1_color1alpha2invdisable 	0x0 		
-    #define TD1_color1alpha2invenable 	0x400 		
-    #define TD1_color1selMASK 		0xff9fffff 	/* bits 21-22 */
-    #define TD1_color1selarg1 		0x0 		/* val 0, shift 21 */
-    #define TD1_color1selarg2 		0x200000 	/* val 1, shift 21 */
-    #define TD1_color1seladd 		0x400000 	/* val 2, shift 21 */
-    #define TD1_color1selmul 		0x600000 	/* val 3, shift 21 */
-    #define TD1_alpha1selMASK 		0x3fffffff 	/* bits 30-31 */
-    #define TD1_alpha1selarg1 		0x0 		/* val 0, shift 30 */
-    #define TD1_alpha1selarg2 		0x40000000 	/* val 1, shift 30 */
-    #define TD1_alpha1seladd 		0x80000000 	/* val 2, shift 30 */
-    #define TD1_alpha1selmul 		0xc0000000 	/* val 3, shift 30 */
-
-    #define TST_ramtsten_MASK 		0xfffffffe 	/* bit 0 */
-    #define TST_ramtsten_disable 	0x0 		
-    #define TST_ramtsten_enable 	0x1 		
-    #define TST_ramtstdone_MASK 	0xfffffffd 	/* bit 1 */
-    #define TST_ramtstdone_disable 	0x0 		
-    #define TST_ramtstdone_enable 	0x2 		
-    #define TST_wramtstpass_MASK 	0xfffffffb 	/* bit 2 */
-    #define TST_wramtstpass_disable 	0x0 		
-    #define TST_wramtstpass_enable 	0x4 		
-    #define TST_tcachetstpass_MASK 	0xfffffff7 	/* bit 3 */
-    #define TST_tcachetstpass_disable 	0x0 		
-    #define TST_tcachetstpass_enable 	0x8 		
-    #define TST_tluttstpass_MASK 	0xffffffef 	/* bit 4 */
-    #define TST_tluttstpass_disable 	0x0 		
-    #define TST_tluttstpass_enable 	0x10 		
-    #define TST_luttstpass_MASK 	0xffffffdf 	/* bit 5 */
-    #define TST_luttstpass_disable 	0x0 		
-    #define TST_luttstpass_enable 	0x20 		
-    #define TST_besramtstpass_MASK 	0xffffffbf 	/* bit 6 */
-    #define TST_besramtstpass_disable 	0x0 		
-    #define TST_besramtstpass_enable 	0x40 		
-    #define TST_ringen_MASK 		0xfffffeff 	/* bit 8 */
-    #define TST_ringen_disable 		0x0 		
-    #define TST_ringen_enable 		0x100 		
-    #define TST_apllbyp_MASK 		0xfffffdff 	/* bit 9 */
-    #define TST_apllbyp_disable 	0x0 		
-    #define TST_apllbyp_enable 		0x200 		
-    #define TST_hiten_MASK 		0xfffffbff 	/* bit 10 */
-    #define TST_hiten_disable 		0x0 		
-    #define TST_hiten_enable 		0x400 		
-    #define TST_tmode_MASK 		0xffffc7ff 	/* bits 11-13 */
-    #define TST_tmode_SHIFT 		11 		
-    #define TST_tclksel_MASK 		0xfffe3fff 	/* bits 14-16 */
-    #define TST_tclksel_SHIFT 		14 		
-    #define TST_ringcnten_MASK 		0xfffdffff 	/* bit 17 */
-    #define TST_ringcnten_disable 	0x0 		
-    #define TST_ringcnten_enable 	0x20000 	
-    #define TST_ringcnt_MASK 		0xc003ffff 	/* bits 18-29 */
-    #define TST_ringcnt_SHIFT 		18 		
-    #define TST_ringcntclksl_MASK 	0xbfffffff 	/* bit 30 */
-    #define TST_ringcntclksl_disable 	0x0 		
-    #define TST_ringcntclksl_enable 	0x40000000 	
-    #define TST_biosboot_MASK 		0x7fffffff 	/* bit 31 */
-    #define TST_biosboot_disable 	0x0 		
-    #define TST_biosboot_enable 	0x80000000 	
-
-    #define TMC_tformat_MASK 		0xfffffff0 	/* bits 0-3 */
-    #define TMC_tformat_tw4 		0x0 		/* val 0, shift 0 */
-    #define TMC_tformat_tw8 		0x1 		/* val 1, shift 0 */
-    #define TMC_tformat_tw15 		0x2 		/* val 2, shift 0 */
-    #define TMC_tformat_tw16 		0x3 		/* val 3, shift 0 */
-    #define TMC_tformat_tw12 		0x4 		/* val 4, shift 0 */
-    #define TMC_tformat_tw32 		0x6 		/* val 6, shift 0 */
-    #define TMC_tformat_tw422 		0xa 		/* val 10, shift 0 */
-    #define TMC_tpitchlin_MASK 		0xfffffeff 	/* bit 8 */
-    #define TMC_tpitchlin_disable 	0x0 		
-    #define TMC_tpitchlin_enable 	0x100 		
-    #define TMC_tpitchext_MASK 		0xfff001ff 	/* bits 9-19 */
-    #define TMC_tpitchext_SHIFT 	9 		
-    #define TMC_tpitch_MASK 		0xfff8ffff 	/* bits 16-18 */
-    #define TMC_tpitch_SHIFT 		16 		
-    #define TMC_owalpha_MASK 		0xffbfffff 	/* bit 22 */
-    #define TMC_owalpha_disable 	0x0 		
-    #define TMC_owalpha_enable 		0x400000 	
-    #define TMC_azeroextend_MASK 	0xff7fffff 	/* bit 23 */
-    #define TMC_azeroextend_disable 	0x0 		
-    #define TMC_azeroextend_enable 	0x800000 	
-    #define TMC_decalckey_MASK 		0xfeffffff 	/* bit 24 */
-    #define TMC_decalckey_disable 	0x0 		
-    #define TMC_decalckey_enable 	0x1000000 	
-    #define TMC_takey_MASK 		0xfdffffff 	/* bit 25 */
-    #define TMC_takey_0 		0x0 		
-    #define TMC_takey_1 		0x2000000 	
-    #define TMC_tamask_MASK 		0xfbffffff 	/* bit 26 */
-    #define TMC_tamask_0 		0x0 		
-    #define TMC_tamask_1 		0x4000000 	
-    #define TMC_clampv_MASK 		0xf7ffffff 	/* bit 27 */
-    #define TMC_clampv_disable 		0x0 		
-    #define TMC_clampv_enable 		0x8000000 	
-    #define TMC_clampu_MASK 		0xefffffff 	/* bit 28 */
-    #define TMC_clampu_disable 		0x0 		
-    #define TMC_clampu_enable 		0x10000000 	
-    #define TMC_tmodulate_MASK 		0xdfffffff 	/* bit 29 */
-    #define TMC_tmodulate_disable 	0x0 		
-    #define TMC_tmodulate_enable 	0x20000000 	
-    #define TMC_strans_MASK 		0xbfffffff 	/* bit 30 */
-    #define TMC_strans_disable 		0x0 		
-    #define TMC_strans_enable 		0x40000000 	
-    #define TMC_itrans_MASK 		0x7fffffff 	/* bit 31 */
-    #define TMC_itrans_disable 		0x0 		
-    #define TMC_itrans_enable 		0x80000000 	
-
-    #define TMC_decalblend_MASK 	0xfffffffe 	/* bit 0 */
-    #define TMC_decalblend_disable 	0x0 		
-    #define TMC_decalblend_enable 	0x1 		
-    #define TMC_idecal_MASK 		0xfffffffd 	/* bit 1 */
-    #define TMC_idecal_disable 		0x0 		
-    #define TMC_idecal_enable 		0x2 		
-    #define TMC_decaldis_MASK 		0xfffffffb 	/* bit 2 */
-    #define TMC_decaldis_disable 	0x0 		
-    #define TMC_decaldis_enable 	0x4 		
-    #define TMC_ckstransdis_MASK 	0xffffffef 	/* bit 4 */
-    #define TMC_ckstransdis_disable 	0x0 		
-    #define TMC_ckstransdis_enable 	0x10 		
-    #define TMC_borderen_MASK 		0xffffffdf 	/* bit 5 */
-    #define TMC_borderen_disable 	0x0 		
-    #define TMC_borderen_enable 	0x20 		
-    #define TMC_specen_MASK 		0xffffffbf 	/* bit 6 */
-    #define TMC_specen_disable 		0x0 		
-    #define TMC_specen_enable 		0x40 		
-
-    #define TF_minfilter_MASK 		0xfffffff0 	/* bits 0-3 */
-    #define TF_minfilter_nrst 		0x0 		/* val 0, shift 0 */
-    #define TF_minfilter_bilin 		0x2 		/* val 2, shift 0 */
-    #define TF_minfilter_cnst 		0x3 		/* val 3, shift 0 */
-    #define TF_minfilter_mm1s 		0x8 		/* val 8, shift 0 */
-    #define TF_minfilter_mm2s 		0x9 		/* val 9, shift 0 */
-    #define TF_minfilter_mm4s 		0xa 		/* val 10, shift 0 */
-    #define TF_minfilter_mm8s 		0xc 		/* val 12, shift 0 */
-    #define TF_magfilter_MASK 		0xffffff0f 	/* bits 4-7 */
-    #define TF_magfilter_nrst 		0x0 		/* val 0, shift 4 */
-    #define TF_magfilter_bilin 		0x20 		/* val 2, shift 4 */
-    #define TF_magfilter_cnst 		0x30 		/* val 3, shift 4 */
-    #define TF_avgstride_MASK 		0xfff7ffff 	/* bit 19 */
-    #define TF_avgstride_disable 	0x0 		
-    #define TF_avgstride_enable 	0x80000 	
-    #define TF_filteralpha_MASK 	0xffefffff 	/* bit 20 */
-    #define TF_filteralpha_disable 	0x0 		
-    #define TF_filteralpha_enable 	0x100000 	
-    #define TF_fthres_MASK 		0xe01fffff 	/* bits 21-28 */
-    #define TF_fthres_SHIFT 		21 		
-    #define TF_mapnb_MASK 		0x1fffffff 	/* bits 29-31 */
-    #define TF_mapnb_SHIFT 		29 		
-
-    #define TH_th_MASK 			0xffffffc0 	/* bits 0-5 */
-    #define TH_th_SHIFT 		0 		
-    #define TH_rfh_MASK 		0xffff81ff 	/* bits 9-14 */
-    #define TH_rfh_SHIFT 		9 		
-    #define TH_thmask_MASK 		0xe003ffff 	/* bits 18-28 */
-    #define TH_thmask_SHIFT 		18 		
-
-    #define TO_texorgmap_MASK 		0xfffffffe 	/* bit 0 */
-    #define TO_texorgmap_fb 		0x0 		
-    #define TO_texorgmap_sys 		0x1 		
-    #define TO_texorgacc_MASK 		0xfffffffd 	/* bit 1 */
-    #define TO_texorgacc_pci 		0x0 		
-    #define TO_texorgacc_agp 		0x2 		
-    #define TO_texorg_MASK 		0x1f 		/* bits 5-31 */
-    #define TO_texorg_SHIFT 		5 		
-
-    #define TT_tckey_MASK 		0xffff0000 	/* bits 0-15 */
-    #define TT_tckey_SHIFT 		0 		
-    #define TT_tkmask_MASK 		0xffff 		/* bits 16-31 */
-    #define TT_tkmask_SHIFT 		16 		
-
-    #define TT_tckeyh_MASK 		0xffff0000 	/* bits 0-15 */
-    #define TT_tckeyh_SHIFT 		0 		
-    #define TT_tkmaskh_MASK 		0xffff 		/* bits 16-31 */
-    #define TT_tkmaskh_SHIFT 		16 		
-
-    #define TW_tw_MASK 			0xffffffc0 	/* bits 0-5 */
-    #define TW_tw_SHIFT 		0 		
-    #define TW_rfw_MASK 		0xffff81ff 	/* bits 9-14 */
-    #define TW_rfw_SHIFT 		9 		
-    #define TW_twmask_MASK 		0xe003ffff 	/* bits 18-28 */
-    #define TW_twmask_SHIFT 		18 		
-
-    #define WAS_seqdst0_MASK 		0xffffffc0 	/* bits 0-5 */
-    #define WAS_seqdst0_SHIFT 		0 		
-    #define WAS_seqdst1_MASK 		0xfffff03f 	/* bits 6-11 */
-    #define WAS_seqdst1_SHIFT 		6 		
-    #define WAS_seqdst2_MASK 		0xfffc0fff 	/* bits 12-17 */
-    #define WAS_seqdst2_SHIFT 		12 		
-    #define WAS_seqdst3_MASK 		0xff03ffff 	/* bits 18-23 */
-    #define WAS_seqdst3_SHIFT 		18 		
-    #define WAS_seqlen_MASK 		0xfcffffff 	/* bits 24-25 */
-    #define WAS_wfirsttag_MASK 		0xfbffffff 	/* bit 26 */
-    #define WAS_wfirsttag_disable 	0x0 		
-    #define WAS_wfirsttag_enable 	0x4000000 	
-    #define WAS_wsametag_MASK 		0xf7ffffff 	/* bit 27 */
-    #define WAS_wsametag_disable 	0x0 		
-    #define WAS_wsametag_enable 	0x8000000 	
-    #define WAS_seqoff_MASK 		0xefffffff 	/* bit 28 */
-    #define WAS_seqoff_disable 		0x0 		
-    #define WAS_seqoff_enable 		0x10000000 	
-
-    #define WMA_wcodeaddr_MASK 		0xff 		/* bits 8-31 */
-    #define WMA_wcodeaddr_SHIFT 	8 		
-
-    #define WF_walustsflag_MASK 	0xffffff00 	/* bits 0-7 */
-    #define WF_walustsflag_SHIFT 	0 		
-    #define WF_walucfgflag_MASK 	0xffff00ff 	/* bits 8-15 */
-    #define WF_walucfgflag_SHIFT 	8 		
-    #define WF_wprgflag_MASK 		0xffff 		/* bits 16-31 */
-    #define WF_wprgflag_SHIFT 		16 		
-
-    #define WF1_walustsflag1_MASK 	0xffffff00 	/* bits 0-7 */
-    #define WF1_walustsflag1_SHIFT 	0 		
-    #define WF1_walucfgflag1_MASK 	0xffff00ff 	/* bits 8-15 */
-    #define WF1_walucfgflag1_SHIFT 	8 		
-    #define WF1_wprgflag1_MASK 		0xffff 		/* bits 16-31 */
-    #define WF1_wprgflag1_SHIFT 	16 		
-
-    #define WGV_wgetmsbmin_MASK 	0xffffffe0 	/* bits 0-4 */
-    #define WGV_wgetmsbmin_SHIFT 	0 		
-    #define WGV_wgetmsbmax_MASK 	0xffffe0ff 	/* bits 8-12 */
-    #define WGV_wgetmsbmax_SHIFT 	8 		
-    #define WGV_wbrklefttop_MASK 	0xfffeffff 	/* bit 16 */
-    #define WGV_wbrklefttop_disable 	0x0 		
-    #define WGV_wbrklefttop_enable 	0x10000 	
-    #define WGV_wfastcrop_MASK 		0xfffdffff 	/* bit 17 */
-    #define WGV_wfastcrop_disable 	0x0 		
-    #define WGV_wfastcrop_enable 	0x20000 	
-    #define WGV_wcentersnap_MASK 	0xfffbffff 	/* bit 18 */
-    #define WGV_wcentersnap_disable 	0x0 		
-    #define WGV_wcentersnap_enable 	0x40000 	
-    #define WGV_wbrkrighttop_MASK 	0xfff7ffff 	/* bit 19 */
-    #define WGV_wbrkrighttop_disable 	0x0 		
-    #define WGV_wbrkrighttop_enable 	0x80000 	
-
-    #define WIA_wmode_MASK 		0xfffffffc 	/* bits 0-1 */
-    #define WIA_wmode_suspend 		0x0 		/* val 0, shift 0 */
-    #define WIA_wmode_resume 		0x1 		/* val 1, shift 0 */
-    #define WIA_wmode_jump 		0x2 		/* val 2, shift 0 */
-    #define WIA_wmode_start 		0x3 		/* val 3, shift 0 */
-    #define WIA_wagp_MASK 		0xfffffffb 	/* bit 2 */
-    #define WIA_wagp_pci 		0x0 		
-    #define WIA_wagp_agp 		0x4 		
-    #define WIA_wiaddr_MASK 		0x7 		/* bits 3-31 */
-    #define WIA_wiaddr_SHIFT 		3 		
-
-    #define WIA2_wmode_MASK 		0xfffffffc 	/* bits 0-1 */
-    #define WIA2_wmode_suspend 		0x0 		/* val 0, shift 0 */
-    #define WIA2_wmode_resume 		0x1 		/* val 1, shift 0 */
-    #define WIA2_wmode_jump 		0x2 		/* val 2, shift 0 */
-    #define WIA2_wmode_start 		0x3 		/* val 3, shift 0 */
-    #define WIA2_wagp_MASK 		0xfffffffb 	/* bit 2 */
-    #define WIA2_wagp_pci 		0x0 		
-    #define WIA2_wagp_agp 		0x4 		
-    #define WIA2_wiaddr_MASK 		0x7 		/* bits 3-31 */
-    #define WIA2_wiaddr_SHIFT 		3 		
-
-    #define WIMA_wimemaddr_MASK 	0xffffff00 	/* bits 0-7 */
-    #define WIMA_wimemaddr_SHIFT 	0 		
-
-    #define WM_wucodecache_MASK 	0xfffffffe 	/* bit 0 */
-    #define WM_wucodecache_disable 	0x0 		
-    #define WM_wucodecache_enable 	0x1 		
-    #define WM_wmaster_MASK 		0xfffffffd 	/* bit 1 */
-    #define WM_wmaster_disable 		0x0 		
-    #define WM_wmaster_enable 		0x2 		
-    #define WM_wcacheflush_MASK 	0xfffffff7 	/* bit 3 */
-    #define WM_wcacheflush_disable 	0x0 		
-    #define WM_wcacheflush_enable 	0x8 		
-
-    #define WVS_wvrtxsz_MASK 		0xffffffc0 	/* bits 0-5 */
-    #define WVS_wvrtxsz_SHIFT 		0 		
-    #define WVS_primsz_MASK 		0xffffc0ff 	/* bits 8-13 */
-    #define WVS_primsz_SHIFT 		8 		
-
-    #define XYEA_x_end_MASK 		0xffff0000 	/* bits 0-15 */
-    #define XYEA_x_end_SHIFT 		0 		
-    #define XYEA_y_end_MASK 		0xffff 		/* bits 16-31 */
-    #define XYEA_y_end_SHIFT 		16 		
-
-    #define XYSA_x_start_MASK 		0xffff0000 	/* bits 0-15 */
-    #define XYSA_x_start_SHIFT 		0 		
-    #define XYSA_y_start_MASK 		0xffff 		/* bits 16-31 */
-    #define XYSA_y_start_SHIFT 		16 		
-
-    #define YA_ydst_MASK 		0xff800000 	/* bits 0-22 */
-    #define YA_ydst_SHIFT 		0 		
-    #define YA_sellin_MASK 		0x1fffffff 	/* bits 29-31 */
-    #define YA_sellin_SHIFT 		29 		
-
-    #define YDL_length_MASK 		0xffff0000 	/* bits 0-15 */
-    #define YDL_length_SHIFT 		0 		
-    #define YDL_yval_MASK 		0xffff 		/* bits 16-31 */
-    #define YDL_yval_SHIFT 		16 		
-
-    #define ZO_zorgmap_MASK 		0xfffffffe 	/* bit 0 */
-    #define ZO_zorgmap_fb 		0x0 		
-    #define ZO_zorgmap_sys 		0x1 		
-    #define ZO_zorgacc_MASK 		0xfffffffd 	/* bit 1 */
-    #define ZO_zorgacc_pci 		0x0 		
-    #define ZO_zorgacc_agp 		0x2 		
-    #define ZO_zorg_MASK 		0x3 		/* bits 2-31 */
-    #define ZO_zorg_SHIFT 		2 		
-
-
-
-
-/**************** (END) AUTOMATICLY GENERATED REGISTER FILE ******************/
-
-#endif 	/* _MGAREGS_H_ */
-
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/picker.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/picker.c
index ecdb2c15a..0bd8bfd57 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/picker.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/picker.c
@@ -9,6 +9,16 @@
 #define CONFIG_MODVERSIONS 0
 #endif
 
+#ifndef CONFIG_AGP_MODULE
+#define CONFIG_AGP_MODULE 0
+#endif
+
+#ifndef CONFIG_AGP
+#define CONFIG_AGP 0
+#endif
+
 SMP = CONFIG_SMP
 MODVERSIONS = CONFIG_MODVERSIONS
+AGP = CONFIG_AGP
+AGP_MODULE = CONFIG_AGP_MODULE
 RELEASE = UTS_RELEASE
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 54aba58c4..db98fd6a4 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
@@ -164,7 +164,10 @@ static int _drm_vm_info(char *buf, char **start, off_t offset, int len,
 {
 	drm_device_t *dev = (drm_device_t *)data;
 	drm_map_t    *map;
-	const char   *types[] = { "FB", "REG", "SHM" };
+				/* Hardcoded from _DRM_FRAME_BUFFER,
+                                   _DRM_REGISTERS, _DRM_SHM, and
+                                   _DRM_AGP. */
+	const char   *types[] = { "FB", "REG", "SHM", "AGP" };
 	const char   *type;
 	int	     i;
 
@@ -175,7 +178,7 @@ static int _drm_vm_info(char *buf, char **start, off_t offset, int len,
 		       "address mtrr\n\n");
 	for (i = 0; i < dev->map_count; i++) {
 		map = dev->maplist[i];
-		if (map->type < 0 || map->type > 2) type = "??";
+		if (map->type < 0 || map->type > 3) type = "??";
 		else				    type = types[map->type];
 		DRM_PROC_PRINT("%4d 0x%08lx 0x%08lx %4.4s  0x%02x 0x%08lx ",
 			       i,
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/tdfx_context.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/tdfx_context.c
index 842d6f5d5..74b107bd2 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/tdfx_context.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/tdfx_context.c
@@ -38,9 +38,7 @@ 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? */
+	return drm_ctxbitmap_next(dev);
 }
 
 int tdfx_context_switch(drm_device_t *dev, int old, int new)
@@ -137,6 +135,12 @@ int tdfx_addctx(struct inode *inode, struct file *filp, unsigned int cmd,
 		ctx.handle = tdfx_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;
 }
@@ -193,13 +197,13 @@ int tdfx_newctx(struct inode *inode, struct file *filp, unsigned int cmd,
 int tdfx_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);
-				/* This is currently a noop because we
-				   don't reuse context values.  Perhaps we
-				   should? */
-	
+	drm_ctxbitmap_free(dev, ctx.handle);
+
 	return 0;
 }
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/tdfx_drv.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/tdfx_drv.c
index 57c1c719d..fb7a997b4 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/tdfx_drv.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/tdfx_drv.c
@@ -85,6 +85,16 @@ static drm_ioctl_desc_t	      tdfx_ioctls[] = {
 	[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 },
+#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, 1},
+	[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_unbind,  1, 1},
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_UNBIND)]    = {drm_agp_bind,    1, 1},
+#endif
 };
 #define TDFX_IOCTL_COUNT DRM_ARRAY_SIZE(tdfx_ioctls)
 
@@ -228,7 +238,24 @@ static int tdfx_takedown(drm_device_t *dev)
 		}
 		dev->magiclist[i].head = dev->magiclist[i].tail = NULL;
 	}
-	
+#ifdef DRM_AGP
+				/* Clear AGP information */
+	if (dev->agp) {
+		drm_agp_mem_t *temp;
+		drm_agp_mem_t *temp_next;
+	   
+		temp = dev->agp->memory;
+		while(temp != NULL) {
+			temp_next = temp->next;
+			drm_free_agp(temp->memory, temp->pages);
+			drm_free(temp, sizeof(*temp), DRM_MEM_AGPLISTS);
+			temp = temp_next;
+		}
+		if(dev->agp->acquired) (*drm_agp.release)();
+		drm_free(dev->agp, sizeof(*dev->agp), DRM_MEM_AGPLISTS);
+		dev->agp = NULL;
+	}
+#endif
 				/* Clear vma list (only built for debugging) */
 	if (dev->vmalist) {
 		for (vma = dev->vmalist; vma; vma = vma_next) {
@@ -262,6 +289,10 @@ static int tdfx_takedown(drm_device_t *dev)
 					       - 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);
 		}
@@ -309,6 +340,16 @@ int tdfx_init(void)
 
 	drm_mem_init();
 	drm_proc_init(dev);
+#ifdef DRM_AGP
+	dev->agp    = drm_agp_init();
+#endif
+	if((retcode = drm_ctxbitmap_init(dev))) {
+		DRM_ERROR("Cannot allocate memory for context bitmap.\n");
+		drm_proc_cleanup();
+		misc_deregister(&tdfx_misc);
+		tdfx_takedown(dev);
+		return retcode;
+	}
 
 	DRM_INFO("Initialized %s %d.%d.%d %s on minor %d\n",
 		 TDFX_NAME,
@@ -335,6 +376,7 @@ void tdfx_cleanup(void)
 	} else {
 		DRM_INFO("Module unloaded\n");
 	}
+	drm_ctxbitmap_cleanup(dev);
 	tdfx_takedown(dev);
 }
 
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 85470ac52..389f2faee 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
@@ -250,9 +250,10 @@ int drm_mmap(struct file *filp, struct vm_area_struct *vma)
 	switch (map->type) {
 	case _DRM_FRAME_BUFFER:
 	case _DRM_REGISTERS:
+	case _DRM_AGP:
 		if (VM_OFFSET(vma) >= __pa(high_memory)) {
 #if defined(__i386__)
-			if (boot_cpu_data.x86 > 3) {
+			if (boot_cpu_data.x86 > 3 && map->type != _DRM_AGP) {
 				pgprot_val(vma->vm_page_prot) |= _PAGE_PCD;
 				pgprot_val(vma->vm_page_prot) &= ~_PAGE_PWT;
 			}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drm.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drm.c
index 3b0f98acd..8b97c7da9 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drm.c
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drm.c
@@ -418,7 +418,8 @@ int drmAddMap(int fd,
     return 0;
 }
 
-int drmAddBufs(int fd, int count, int size, int flags)
+int drmAddBufs(int fd, int count, int size, drmBufDescFlags flags,
+	       int agp_offset)
 {
     drm_buf_desc_t request;
     
@@ -427,6 +428,8 @@ int drmAddBufs(int fd, int count, int size, int flags)
     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;
 }
@@ -744,6 +747,143 @@ int drmDestroyDrawable(int fd, drmDrawable handle)
     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) {
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..a7b95fbf0
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmI810.c
@@ -0,0 +1,81 @@
+#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, unsigned long start, unsigned long end, 
+		    unsigned long size, int ring_map_idx, int buffer_map_idx, 
+		    int sarea_off)
+{
+   drm_i810_init_t init;
+   
+   memset(&init, 0, sizeof(drm_i810_init_t));
+   init.func = I810_INIT_DMA;
+   init.ring_map_idx = ring_map_idx;
+   init.buffer_map_idx = buffer_map_idx;
+   init.ring_start = start;
+   init.ring_end = end;
+   init.ring_size = size;
+   init.sarea_priv_offset = sarea_off;
+
+   if(ioctl(driSubFD, DRM_IOCTL_I810_INIT, &init)) {
+         return FALSE;
+      }
+   return TRUE;
+}
diff --git a/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmMga.c b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmMga.c
new file mode 100644
index 000000000..5957ffe33
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/linux/drm/xf86drmMga.c
@@ -0,0 +1,117 @@
+#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 "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/shared/drm/kernel/drm.h b/xc/programs/Xserver/hw/xfree86/os-support/shared/drm/kernel/drm.h
index 7b8e88265..ae4c65ca3 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
@@ -61,6 +61,19 @@ 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"
 
 typedef struct drm_version {
 	int    version_major;	  /* Major version			    */
@@ -101,7 +114,8 @@ typedef struct drm_control {
 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_SHM	  = 2,	  /* shared, cached			    */
+	_DRM_AGP          = 3	  /* AGP/GART                               */
 } drm_map_type_t;
 
 typedef enum drm_map_flags {
@@ -165,8 +179,11 @@ typedef struct drm_buf_desc {
 	int	      low_mark;	 /* Low water mark			     */
 	int	      high_mark; /* High water mark			     */
 	enum {
-		DRM_PAGE_ALIGN = 0x01  /* Align on page boundaries for DMA   */
+		_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 {
@@ -237,6 +254,38 @@ typedef struct drm_irq_busid {
 	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)	     _IOC_NR(n)
 #define DRM_IO(nr)	     _IO(DRM_IOCTL_BASE,nr)
@@ -276,4 +325,28 @@ typedef struct drm_irq_busid {
 #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_IOR( 0x32, drm_agp_mode_t)
+#define DRM_IOCTL_AGP_INFO    DRM_IOW( 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_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 )
+
+/* 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_FLUSH   DRM_IO ( 0x43)
+#define DRM_IOCTL_I810_GETAGE  DRM_IO ( 0x44)
+
 #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..0fd1141b5 100644
--- a/xc/programs/Xserver/hw/xfree86/os-support/xf86drm.h
+++ b/xc/programs/Xserver/hw/xfree86/os-support/xf86drm.h
@@ -65,7 +65,8 @@ 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 +96,11 @@ 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 +202,7 @@ 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 +297,9 @@ 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 +340,29 @@ 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..88bb58a3b
--- /dev/null
+++ b/xc/programs/Xserver/hw/xfree86/os-support/xf86drmI810.h
@@ -0,0 +1,29 @@
+
+/* 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 2
+
+/* 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
+
+Bool drmI810CleanupDma(int driSubFD);
+Bool drmI810InitDma(int driSubFD, unsigned long start, unsigned long end, 
+		    unsigned long size, int ring_map_idx, int buffer_map_idx, 
+		    int sarea_off);
+#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..9525a2710
--- /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 */
+
+/* 64 buffers of 16k each, total 1 meg.
+ */
+#define MGA_DMA_BUF_ORDER     14
+#define MGA_DMA_BUF_SZ        (1<<MGA_DMA_BUF_ORDER)
+#define MGA_DMA_BUF_NR        63
+
+/* 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;
+   int frontOffset;
+   int backOffset;
+   int depthOffset;
+   int textureOffset;
+   int textureSize;
+   int agpTextureSize;
+   int agpTextureOffset;
+   int cpp;
+   int stride;
+   int sgram;
+   int chipset;
+   drmMgaWarpIndex WarpIndex[MGA_MAX_WARP_PIPES];
+   int mAccess;
+} drmMgaInit;
+
+
+Bool drmMgaCleanupDma(int driSubFD);
+Bool drmMgaLockUpdate(int driSubFD, drmLockFlags flags);
+Bool drmMgaInitDma(int driSubFD, drmMgaInit *info);
+#endif
author	jhartmann <jhartmann>	2000-04-04 22:08:12 +0000
committer	jhartmann <jhartmann>	2000-04-04 22:08:12 +0000
commit	4fe5b7fbada2f73c1533e30f8abc5cf71e7c61c4 (patch)
tree	d526515e629abcae4e03e1c5c02b60243dc35f72 /xc
parent	d13b9c4ecba68ca9c63d94dbf48a395e3d39b847 (diff)