Add mach64 to the trunk

18 files changed, 7465 insertions, 6 deletions

diff --git a/linux-core/Kconfig b/linux-core/Kconfig
index e45e723b..0dd10b8d 100644
--- a/linux-core/Kconfig
+++ b/linux-core/Kconfig
@@ -88,3 +88,15 @@ config DRM_VIA
 	  Choose this option if you have a Via unichrome or compatible video 
           chipset. If M is selected the module will be called via.
 
+config DRM_MACH64
+	tristate "ATI Rage Pro (Mach64)"
+	depends on DRM
+	help
+	  Choose this option if you have an ATI Rage Pro (mach64 chipset)
+	  graphics card.  Example cards include:  3D Rage Pro, Xpert 98,
+	  3D Rage LT Pro, 3D Rage XL/XC, and 3D Rage Mobility (P/M, M1).
+	  Cards earlier than ATI Rage Pro (e.g. Rage II) are not supported.
+	  If M is selected, the module will be called mach64.  AGP support for
+	  this card is strongly suggested (unless you have a PCI version).
+
+
diff --git a/linux-core/Makefile.kernel b/linux-core/Makefile.kernel
index bcb0f92c..3f467384 100644
--- a/linux-core/Makefile.kernel
+++ b/linux-core/Makefile.kernel
@@ -18,6 +18,7 @@ sis-objs    := sis_drv.o sis_ds.o sis_mm.o
 ffb-objs    := ffb_drv.o ffb_context.o
 savage-objs := savage_drv.o savage_dma.o
 via-objs    := via_irq.o via_drv.o via_ds.o via_map.o via_mm.o 
+mach64-objs := mach64_drv.o mach64_dma.o mach64_irq.o mach64_state.o
 
 # Kernel version checks
 
@@ -33,7 +34,7 @@ endif
 ifeq ($(BELOW25),y)
 O_TARGET	:= drm.o
 list-multi	:= gamma.o tdfx.o r128.o mga.o i810.o i830.o ffb.o radeon.o \
-                   savage.o via.o
+                   savage.o via.o mach64.o
 obj-m		:=
 obj-n		:=
 obj-		:=
@@ -51,6 +52,7 @@ obj-$(CONFIG_DRM_SIS)   += sis.o
 obj-$(CONFIG_DRM_FFB)   += ffb.o
 obj-$(CONFIG_DRM_SAVAGE)+= savage.o
 obj-$(CONFIG_DRM_VIA)   += via.o
+obj-$(CONFIG_DRM_MACH64)+= mach64.o
 
 ifeq ($(BELOW2552),y)
 include $(TOPDIR)/Rules.make
@@ -90,5 +92,8 @@ savage.o: $(savage-objs) $(lib)
 via.o: $(via-objs) $(lib)
 	$(LD) -r -o $@ $(via-objs) $(lib)
 
+mach64.o: $(mach64-objs) $(lib)
+	$(LD) -r -o $@ $(mach64-objs) $(lib)
+
 endif
 
diff --git a/linux-core/drmP.h b/linux-core/drmP.h
index c61bac1d..09556b89 100644
--- a/linux-core/drmP.h
+++ b/linux-core/drmP.h
@@ -1057,6 +1057,12 @@ extern int            DRM(ati_pcigart_cleanup)(drm_device_t *dev,
 					       unsigned long addr,
 					       dma_addr_t bus_addr);
 
+extern void	      *DRM(pci_alloc)(drm_device_t *dev, size_t size, 
+					size_t align, dma_addr_t maxaddr,
+					dma_addr_t *busaddr);
+extern void	      DRM(pci_free)(drm_device_t *dev, size_t size, 
+					void *vaddr, dma_addr_t busaddr);
+
 /*@}*/
 
 #endif /* __KERNEL__ */
diff --git a/linux/Kconfig b/linux/Kconfig
index e45e723b..0dd10b8d 100644
--- a/linux/Kconfig
+++ b/linux/Kconfig
@@ -88,3 +88,15 @@ config DRM_VIA
 	  Choose this option if you have a Via unichrome or compatible video 
           chipset. If M is selected the module will be called via.
 
+config DRM_MACH64
+	tristate "ATI Rage Pro (Mach64)"
+	depends on DRM
+	help
+	  Choose this option if you have an ATI Rage Pro (mach64 chipset)
+	  graphics card.  Example cards include:  3D Rage Pro, Xpert 98,
+	  3D Rage LT Pro, 3D Rage XL/XC, and 3D Rage Mobility (P/M, M1).
+	  Cards earlier than ATI Rage Pro (e.g. Rage II) are not supported.
+	  If M is selected, the module will be called mach64.  AGP support for
+	  this card is strongly suggested (unless you have a PCI version).
+
+
diff --git a/linux/Makefile.kernel b/linux/Makefile.kernel
index bcb0f92c..3f467384 100644
--- a/linux/Makefile.kernel
+++ b/linux/Makefile.kernel
@@ -18,6 +18,7 @@ sis-objs    := sis_drv.o sis_ds.o sis_mm.o
 ffb-objs    := ffb_drv.o ffb_context.o
 savage-objs := savage_drv.o savage_dma.o
 via-objs    := via_irq.o via_drv.o via_ds.o via_map.o via_mm.o 
+mach64-objs := mach64_drv.o mach64_dma.o mach64_irq.o mach64_state.o
 
 # Kernel version checks
 
@@ -33,7 +34,7 @@ endif
 ifeq ($(BELOW25),y)
 O_TARGET	:= drm.o
 list-multi	:= gamma.o tdfx.o r128.o mga.o i810.o i830.o ffb.o radeon.o \
-                   savage.o via.o
+                   savage.o via.o mach64.o
 obj-m		:=
 obj-n		:=
 obj-		:=
@@ -51,6 +52,7 @@ obj-$(CONFIG_DRM_SIS)   += sis.o
 obj-$(CONFIG_DRM_FFB)   += ffb.o
 obj-$(CONFIG_DRM_SAVAGE)+= savage.o
 obj-$(CONFIG_DRM_VIA)   += via.o
+obj-$(CONFIG_DRM_MACH64)+= mach64.o
 
 ifeq ($(BELOW2552),y)
 include $(TOPDIR)/Rules.make
@@ -90,5 +92,8 @@ savage.o: $(savage-objs) $(lib)
 via.o: $(via-objs) $(lib)
 	$(LD) -r -o $@ $(via-objs) $(lib)
 
+mach64.o: $(mach64-objs) $(lib)
+	$(LD) -r -o $@ $(mach64-objs) $(lib)
+
 endif
 
diff --git a/linux/Makefile.linux b/linux/Makefile.linux
index 9634bc3e..229219fb 100644
--- a/linux/Makefile.linux
+++ b/linux/Makefile.linux
@@ -44,7 +44,7 @@ endif
 MACHINE := $(shell uname -m)
 
 # Modules for all architectures
-MODULE_LIST := gamma.o tdfx.o r128.o radeon.o mga.o sis.o savage.o via.o
+MODULE_LIST := gamma.o tdfx.o r128.o radeon.o mga.o sis.o savage.o via.o mach64.o
 
 # Modules only for ix86 architectures
 ifneq (,$(findstring 86,$(MACHINE)))
@@ -89,9 +89,13 @@ VIAHEADERS	= via_drm.h via_drv.h via.h via_mm.h via_ds.h \
 		$(DRMHEADERS) $(DRMTEMPLATES)
 VIASHARED	= via_drm.h via_drv.h via.h via_mm.h via_ds.h via_drv.c \
 		via_ds.c via_irq.c via_map.c via_mm.c
+MACH64HEADERS	= mach64.h mach64_drv.h mach64_drm.h $(DRMHEADERS) \
+		$(DRMTEMPLATES)
+MACH64SHARED 	= mach64.h mach64_drv.h mach64_drm.h mach64_dma.c \
+		mach64_irq.c mach64_state.c
 
 SHAREDSRC = $(DRMSHARED) $(MGASHARED) $(R128SHARED) $(RADEONSHARED) \
-	$(SISSHARED) $(TDFXSHARED) $(VIASHARED)
+	$(SISSHARED) $(TDFXSHARED) $(VIASHARED) $(MACH64SHARED)
 
 PROGS = dristat drmstat
 
@@ -322,7 +326,7 @@ CONFIG_DRM_SIS := n
 CONFIG_DRM_FFB := n
 CONFIG_DRM_SAVAGE := n
 CONFIG_DRM_VIA := n
-
+CONFIG_DRM_MACH64 := n
 
 # Enable module builds for the modules requested/supported.
 
@@ -344,6 +348,10 @@ endif
 ifneq (,$(findstring via,$(DRM_MODULES)))
 CONFIG_DRM_VIA := m
 endif
+ifneq (,$(findstring mach64,$(DRM_MODULES)))
+CONFIG_DRM_MACH64 := m
+endif
+
 
 
 # These require AGP support
@@ -377,7 +385,7 @@ $(sis-objs):	$(SISHEADERS)
 $(ffb-objs):	$(FFBHEADERS)
 $(savage-objs): $(SAVAGEHEADERS)
 $(via-objs):	$(VIAHEADERS)
-
+$(mach64-objs): $(MACH64HEADERS)
 
 endif
 
diff --git a/linux/drmP.h b/linux/drmP.h
index c61bac1d..09556b89 100644
--- a/linux/drmP.h
+++ b/linux/drmP.h
@@ -1057,6 +1057,12 @@ extern int            DRM(ati_pcigart_cleanup)(drm_device_t *dev,
 					       unsigned long addr,
 					       dma_addr_t bus_addr);
 
+extern void	      *DRM(pci_alloc)(drm_device_t *dev, size_t size, 
+					size_t align, dma_addr_t maxaddr,
+					dma_addr_t *busaddr);
+extern void	      DRM(pci_free)(drm_device_t *dev, size_t size, 
+					void *vaddr, dma_addr_t busaddr);
+
 /*@}*/
 
 #endif /* __KERNEL__ */
diff --git a/shared-core/mach64_dma.c b/shared-core/mach64_dma.c
new file mode 100644
index 00000000..1bbffc35
--- /dev/null
+++ b/shared-core/mach64_dma.c
@@ -0,0 +1,1325 @@
+/* mach64_dma.c -- DMA support for mach64 (Rage Pro) driver -*- linux-c -*-
+ * Created: Sun Dec 03 19:20:26 2000 by gareth@valinux.com
+ *
+ * Copyright 2000 Gareth Hughes
+ * Copyright 2002 Frank C. Earl
+ * Copyright 2002-2003 Leif Delgass
+ * 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
+ * THE COPYRIGHT OWNER(S) 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:
+ *   Gareth Hughes <gareth@valinux.com>
+ *   Frank C. Earl <fearl@airmail.net>
+ *   Leif Delgass <ldelgass@retinalburn.net>
+ *   José Fonseca <j_r_fonseca@yahoo.co.uk>
+ */
+
+#include "mach64.h"
+#include "drmP.h"
+#include "drm.h"
+#include "mach64_drm.h"
+#include "mach64_drv.h"
+
+
+/* ================================================================
+ * Engine, FIFO control
+ */
+
+int mach64_do_wait_for_fifo( drm_mach64_private_t *dev_priv, int entries )
+{
+	int slots = 0, i;
+
+	for ( i = 0 ; i < dev_priv->usec_timeout ; i++ ) {
+		slots = (MACH64_READ( MACH64_FIFO_STAT ) &
+			 MACH64_FIFO_SLOT_MASK);
+		if ( slots <= (0x8000 >> entries) ) return 0;
+		DRM_UDELAY( 1 );
+	}
+
+	DRM_INFO( "%s failed! slots=%d entries=%d\n", __FUNCTION__, slots, entries );
+	return DRM_ERR(EBUSY);
+}
+
+int mach64_do_wait_for_idle( drm_mach64_private_t *dev_priv )
+{
+	int i, ret;
+
+	ret = mach64_do_wait_for_fifo( dev_priv, 16 );
+	if ( ret < 0 ) return ret;
+
+	for ( i = 0 ; i < dev_priv->usec_timeout ; i++ ) {
+		if ( !(MACH64_READ( MACH64_GUI_STAT ) & MACH64_GUI_ACTIVE) ) {
+			return 0;
+		}
+		DRM_UDELAY( 1 );
+	}
+
+	DRM_INFO( "%s failed! GUI_STAT=0x%08x\n", __FUNCTION__, 
+		   MACH64_READ( MACH64_GUI_STAT ) );
+	mach64_dump_ring_info( dev_priv );
+	return DRM_ERR(EBUSY);
+}
+
+int mach64_wait_ring( drm_mach64_private_t *dev_priv, int n )
+{
+	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;
+	int i;
+
+	for ( i = 0 ; i < dev_priv->usec_timeout ; i++ ) {
+		mach64_update_ring_snapshot( dev_priv );
+		if ( ring->space >= n ) {
+			if (i > 0) {
+				DRM_DEBUG( "%s: %d usecs\n", __FUNCTION__, i );
+			}
+			return 0;
+		}
+		DRM_UDELAY( 1 );
+	}
+
+	/* FIXME: This is being ignored... */
+	DRM_ERROR( "failed!\n" );
+	mach64_dump_ring_info( dev_priv );
+	return DRM_ERR(EBUSY);
+}
+
+/* Wait until all DMA requests have been processed... */
+static int mach64_ring_idle( drm_mach64_private_t *dev_priv )
+{
+	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;
+	u32 head;
+	int i;
+
+	head = ring->head;
+	i = 0;
+	while ( i < dev_priv->usec_timeout ) {
+		mach64_update_ring_snapshot( dev_priv );
+		if ( ring->head == ring->tail && 
+		     !(MACH64_READ(MACH64_GUI_STAT) & MACH64_GUI_ACTIVE) ) {
+			if (i > 0) {
+				DRM_DEBUG( "%s: %d usecs\n", __FUNCTION__, i );
+			}
+			return 0;
+		} 
+		if ( ring->head == head ) {
+			++i;
+		} else {
+			head = ring->head;
+			i = 0;
+		}
+		DRM_UDELAY( 1 );
+	}
+
+	DRM_INFO( "%s failed! GUI_STAT=0x%08x\n", __FUNCTION__, 
+		   MACH64_READ( MACH64_GUI_STAT ) );
+	mach64_dump_ring_info( dev_priv );
+	return DRM_ERR(EBUSY);
+}
+
+static void mach64_ring_reset( drm_mach64_private_t *dev_priv )
+{
+	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;
+
+	mach64_do_release_used_buffers( dev_priv );
+	ring->head_addr = ring->start_addr;
+	ring->head = ring->tail = 0;
+	ring->space = ring->size;
+	
+	MACH64_WRITE( MACH64_BM_GUI_TABLE_CMD, 
+		      ring->head_addr | MACH64_CIRCULAR_BUF_SIZE_16KB );
+
+	dev_priv->ring_running = 0;
+}
+
+int mach64_do_dma_flush( drm_mach64_private_t *dev_priv )
+{
+	/* FIXME: It's not necessary to wait for idle when flushing 
+	 * we just need to ensure the ring will be completely processed
+	 * in finite time without another ioctl
+	 */
+	return mach64_ring_idle( dev_priv );
+}
+
+int mach64_do_dma_idle( drm_mach64_private_t *dev_priv )
+{
+	int ret;
+
+	/* wait for completion */
+	if ( (ret = mach64_ring_idle( dev_priv )) < 0 ) {
+		DRM_ERROR( "%s failed BM_GUI_TABLE=0x%08x tail: %u\n", __FUNCTION__, 
+			   MACH64_READ(MACH64_BM_GUI_TABLE), dev_priv->ring.tail );
+		return ret;
+	}
+
+	mach64_ring_stop( dev_priv );
+
+	/* clean up after pass */
+	mach64_do_release_used_buffers( dev_priv );
+	return 0;
+}
+
+/* Reset the engine.  This will stop the DMA if it is running.
+ */
+int mach64_do_engine_reset( drm_mach64_private_t *dev_priv )
+{
+	u32 tmp;
+
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+	/* Kill off any outstanding DMA transfers.
+	 */
+	tmp = MACH64_READ( MACH64_BUS_CNTL );
+	MACH64_WRITE( MACH64_BUS_CNTL,
+				  tmp | MACH64_BUS_MASTER_DIS );
+	
+	/* Reset the GUI engine (high to low transition).
+	 */
+	tmp = MACH64_READ( MACH64_GEN_TEST_CNTL );
+	MACH64_WRITE( MACH64_GEN_TEST_CNTL,
+				  tmp & ~MACH64_GUI_ENGINE_ENABLE );
+	/* Enable the GUI engine
+	 */
+	tmp = MACH64_READ( MACH64_GEN_TEST_CNTL );
+	MACH64_WRITE( MACH64_GEN_TEST_CNTL,
+				  tmp | MACH64_GUI_ENGINE_ENABLE );
+
+	/* ensure engine is not locked up by clearing any FIFO or HOST errors
+	*/
+	tmp = MACH64_READ( MACH64_BUS_CNTL );
+	MACH64_WRITE( MACH64_BUS_CNTL, tmp | 0x00a00000 );
+
+	/* Once GUI engine is restored, disable bus mastering */
+	MACH64_WRITE( MACH64_SRC_CNTL, 0 );
+
+	/* Reset descriptor ring */
+	mach64_ring_reset( dev_priv );
+	
+	return 0;
+}
+
+/* ================================================================
+ * Debugging output
+ */
+
+void mach64_dump_engine_info( drm_mach64_private_t *dev_priv )
+{
+	DRM_INFO( "\n" );
+	if ( !dev_priv->is_pci )
+	{
+		DRM_INFO( "           AGP_BASE = 0x%08x\n", MACH64_READ( MACH64_AGP_BASE ) );
+		DRM_INFO( "           AGP_CNTL = 0x%08x\n", MACH64_READ( MACH64_AGP_CNTL ) );
+	}
+	DRM_INFO( "     ALPHA_TST_CNTL = 0x%08x\n", MACH64_READ( MACH64_ALPHA_TST_CNTL ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "         BM_COMMAND = 0x%08x\n", MACH64_READ( MACH64_BM_COMMAND ) );
+	DRM_INFO( "BM_FRAME_BUF_OFFSET = 0x%08x\n", MACH64_READ( MACH64_BM_FRAME_BUF_OFFSET ) );
+	DRM_INFO( "       BM_GUI_TABLE = 0x%08x\n", MACH64_READ( MACH64_BM_GUI_TABLE ) );
+	DRM_INFO( "          BM_STATUS = 0x%08x\n", MACH64_READ( MACH64_BM_STATUS ) );
+	DRM_INFO( " BM_SYSTEM_MEM_ADDR = 0x%08x\n", MACH64_READ( MACH64_BM_SYSTEM_MEM_ADDR ) );
+	DRM_INFO( "    BM_SYSTEM_TABLE = 0x%08x\n", MACH64_READ( MACH64_BM_SYSTEM_TABLE ) );
+	DRM_INFO( "           BUS_CNTL = 0x%08x\n", MACH64_READ( MACH64_BUS_CNTL ) );
+	DRM_INFO( "\n" );
+	/* DRM_INFO( "         CLOCK_CNTL = 0x%08x\n", MACH64_READ( MACH64_CLOCK_CNTL ) ); */
+	DRM_INFO( "        CLR_CMP_CLR = 0x%08x\n", MACH64_READ( MACH64_CLR_CMP_CLR ) );
+	DRM_INFO( "       CLR_CMP_CNTL = 0x%08x\n", MACH64_READ( MACH64_CLR_CMP_CNTL ) );
+	/* DRM_INFO( "        CLR_CMP_MSK = 0x%08x\n", MACH64_READ( MACH64_CLR_CMP_MSK ) ); */
+	DRM_INFO( "     CONFIG_CHIP_ID = 0x%08x\n", MACH64_READ( MACH64_CONFIG_CHIP_ID ) );
+	DRM_INFO( "        CONFIG_CNTL = 0x%08x\n", MACH64_READ( MACH64_CONFIG_CNTL ) );
+	DRM_INFO( "       CONFIG_STAT0 = 0x%08x\n", MACH64_READ( MACH64_CONFIG_STAT0 ) );
+	DRM_INFO( "       CONFIG_STAT1 = 0x%08x\n", MACH64_READ( MACH64_CONFIG_STAT1 ) );
+	DRM_INFO( "       CONFIG_STAT2 = 0x%08x\n", MACH64_READ( MACH64_CONFIG_STAT2 ) );
+	DRM_INFO( "            CRC_SIG = 0x%08x\n", MACH64_READ( MACH64_CRC_SIG ) );
+	DRM_INFO( "  CUSTOM_MACRO_CNTL = 0x%08x\n", MACH64_READ( MACH64_CUSTOM_MACRO_CNTL ) );
+	DRM_INFO( "\n" );
+	/* DRM_INFO( "           DAC_CNTL = 0x%08x\n", MACH64_READ( MACH64_DAC_CNTL ) ); */
+	/* DRM_INFO( "           DAC_REGS = 0x%08x\n", MACH64_READ( MACH64_DAC_REGS ) ); */
+	DRM_INFO( "        DP_BKGD_CLR = 0x%08x\n", MACH64_READ( MACH64_DP_BKGD_CLR ) );
+	DRM_INFO( "        DP_FRGD_CLR = 0x%08x\n", MACH64_READ( MACH64_DP_FRGD_CLR ) );
+	DRM_INFO( "             DP_MIX = 0x%08x\n", MACH64_READ( MACH64_DP_MIX ) );
+	DRM_INFO( "       DP_PIX_WIDTH = 0x%08x\n", MACH64_READ( MACH64_DP_PIX_WIDTH ) );
+	DRM_INFO( "             DP_SRC = 0x%08x\n", MACH64_READ( MACH64_DP_SRC ) );
+	DRM_INFO( "      DP_WRITE_MASK = 0x%08x\n", MACH64_READ( MACH64_DP_WRITE_MASK ) );
+	DRM_INFO( "         DSP_CONFIG = 0x%08x\n", MACH64_READ( MACH64_DSP_CONFIG ) );
+	DRM_INFO( "         DSP_ON_OFF = 0x%08x\n", MACH64_READ( MACH64_DSP_ON_OFF ) );
+	DRM_INFO( "           DST_CNTL = 0x%08x\n", MACH64_READ( MACH64_DST_CNTL ) );
+	DRM_INFO( "      DST_OFF_PITCH = 0x%08x\n", MACH64_READ( MACH64_DST_OFF_PITCH ) );
+	DRM_INFO( "\n" );
+	/* DRM_INFO( "       EXT_DAC_REGS = 0x%08x\n", MACH64_READ( MACH64_EXT_DAC_REGS ) ); */
+	DRM_INFO( "       EXT_MEM_CNTL = 0x%08x\n", MACH64_READ( MACH64_EXT_MEM_CNTL ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "          FIFO_STAT = 0x%08x\n", MACH64_READ( MACH64_FIFO_STAT ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "      GEN_TEST_CNTL = 0x%08x\n", MACH64_READ( MACH64_GEN_TEST_CNTL ) );
+	/* DRM_INFO( "              GP_IO = 0x%08x\n", MACH64_READ( MACH64_GP_IO ) ); */
+	DRM_INFO( "   GUI_CMDFIFO_DATA = 0x%08x\n", MACH64_READ( MACH64_GUI_CMDFIFO_DATA ) );
+	DRM_INFO( "  GUI_CMDFIFO_DEBUG = 0x%08x\n", MACH64_READ( MACH64_GUI_CMDFIFO_DEBUG ) );
+	DRM_INFO( "           GUI_CNTL = 0x%08x\n", MACH64_READ( MACH64_GUI_CNTL ) );
+	DRM_INFO( "           GUI_STAT = 0x%08x\n", MACH64_READ( MACH64_GUI_STAT ) );
+	DRM_INFO( "      GUI_TRAJ_CNTL = 0x%08x\n", MACH64_READ( MACH64_GUI_TRAJ_CNTL ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "          HOST_CNTL = 0x%08x\n", MACH64_READ( MACH64_HOST_CNTL ) );
+	DRM_INFO( "           HW_DEBUG = 0x%08x\n", MACH64_READ( MACH64_HW_DEBUG ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "    MEM_ADDR_CONFIG = 0x%08x\n", MACH64_READ( MACH64_MEM_ADDR_CONFIG ) );
+	DRM_INFO( "       MEM_BUF_CNTL = 0x%08x\n", MACH64_READ( MACH64_MEM_BUF_CNTL ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "           PAT_REG0 = 0x%08x\n", MACH64_READ( MACH64_PAT_REG0 ) );
+	DRM_INFO( "           PAT_REG1 = 0x%08x\n", MACH64_READ( MACH64_PAT_REG1 ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "            SC_LEFT = 0x%08x\n", MACH64_READ( MACH64_SC_LEFT ) );
+	DRM_INFO( "           SC_RIGHT = 0x%08x\n", MACH64_READ( MACH64_SC_RIGHT ) );
+	DRM_INFO( "             SC_TOP = 0x%08x\n", MACH64_READ( MACH64_SC_TOP ) );
+	DRM_INFO( "          SC_BOTTOM = 0x%08x\n", MACH64_READ( MACH64_SC_BOTTOM ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "      SCALE_3D_CNTL = 0x%08x\n", MACH64_READ( MACH64_SCALE_3D_CNTL ) );
+	DRM_INFO( "       SCRATCH_REG0 = 0x%08x\n", MACH64_READ( MACH64_SCRATCH_REG0 ) );
+	DRM_INFO( "       SCRATCH_REG1 = 0x%08x\n", MACH64_READ( MACH64_SCRATCH_REG1 ) );
+	DRM_INFO( "         SETUP_CNTL = 0x%08x\n", MACH64_READ( MACH64_SETUP_CNTL ) );
+	DRM_INFO( "           SRC_CNTL = 0x%08x\n", MACH64_READ( MACH64_SRC_CNTL ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "           TEX_CNTL = 0x%08x\n", MACH64_READ( MACH64_TEX_CNTL ) );
+	DRM_INFO( "     TEX_SIZE_PITCH = 0x%08x\n", MACH64_READ( MACH64_TEX_SIZE_PITCH ) );
+	DRM_INFO( "       TIMER_CONFIG = 0x%08x\n", MACH64_READ( MACH64_TIMER_CONFIG ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "             Z_CNTL = 0x%08x\n", MACH64_READ( MACH64_Z_CNTL ) );
+	DRM_INFO( "        Z_OFF_PITCH = 0x%08x\n", MACH64_READ( MACH64_Z_OFF_PITCH ) );
+	DRM_INFO( "\n" );
+}
+
+#define MACH64_DUMP_CONTEXT	3
+
+static void mach64_dump_buf_info( drm_mach64_private_t *dev_priv, drm_buf_t *buf)
+{
+	u32 addr = GETBUFADDR( buf );
+	u32 used = buf->used >> 2;
+	u32 sys_addr = MACH64_READ( MACH64_BM_SYSTEM_MEM_ADDR );
+	u32 *p = GETBUFPTR( buf );
+	int skipped = 0;
+
+	DRM_INFO( "buffer contents:\n" );
+	
+	while ( used ) {
+		u32 reg, count;
+
+		reg = le32_to_cpu(*p++);
+		if( addr <= GETBUFADDR( buf ) + MACH64_DUMP_CONTEXT * 4 ||
+		    (addr >= sys_addr - MACH64_DUMP_CONTEXT * 4 &&
+		    addr <= sys_addr + MACH64_DUMP_CONTEXT * 4) ||
+		    addr >= GETBUFADDR( buf ) + buf->used - MACH64_DUMP_CONTEXT * 4) {
+			DRM_INFO("%08x:  0x%08x\n", addr, reg);
+		}
+		addr += 4;
+		used--;
+		
+		count = (reg >> 16) + 1;
+		reg = reg & 0xffff;
+		reg = MMSELECT( reg );
+		while ( count && used ) {
+			if( addr <= GETBUFADDR( buf ) + MACH64_DUMP_CONTEXT * 4 ||
+			    (addr >= sys_addr - MACH64_DUMP_CONTEXT * 4 &&
+			    addr <= sys_addr + MACH64_DUMP_CONTEXT * 4) ||
+			    addr >= GETBUFADDR( buf ) + buf->used - MACH64_DUMP_CONTEXT * 4) {
+				DRM_INFO("%08x:    0x%04x = 0x%08x\n", addr, reg, le32_to_cpu(*p));
+				skipped = 0;
+			} else {
+				if( !skipped ) {
+					DRM_INFO( "  ...\n" );
+					skipped = 1;
+				}
+			}
+			p++;
+			addr += 4;
+			used--;
+			
+			reg += 4;
+			count--;
+		}
+	}
+	
+	DRM_INFO( "\n" );
+}
+
+void mach64_dump_ring_info( drm_mach64_private_t *dev_priv )
+{
+	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;
+	int i, skipped;
+	
+	DRM_INFO( "\n" );
+	
+	DRM_INFO("ring contents:\n");
+	DRM_INFO("  head_addr: 0x%08x head: %u tail: %u\n\n", 
+		 ring->head_addr, ring->head, ring->tail );
+	
+	skipped = 0;
+	for ( i = 0; i < ring->size / sizeof(u32); i += 4) {
+		if( i <= MACH64_DUMP_CONTEXT * 4 || 
+		    i >= ring->size / sizeof(u32) - MACH64_DUMP_CONTEXT * 4 ||
+		    (i >= ring->tail - MACH64_DUMP_CONTEXT * 4 &&
+		    i <= ring->tail + MACH64_DUMP_CONTEXT * 4) ||
+		    (i >= ring->head - MACH64_DUMP_CONTEXT * 4 &&
+		    i <= ring->head + MACH64_DUMP_CONTEXT * 4)) {
+			DRM_INFO( "  0x%08x:  0x%08x 0x%08x 0x%08x 0x%08x%s%s\n",
+				ring->start_addr + i * sizeof(u32),
+				le32_to_cpu(((u32*) ring->start)[i + 0]),
+				le32_to_cpu(((u32*) ring->start)[i + 1]),
+				le32_to_cpu(((u32*) ring->start)[i + 2]), 
+				le32_to_cpu(((u32*) ring->start)[i + 3]),
+				i == ring->head ? " (head)" : "",
+				i == ring->tail ? " (tail)" : ""
+			);
+			skipped = 0;
+		} else {
+			if( !skipped ) {
+				DRM_INFO( "  ...\n" );
+				skipped = 1;
+			}
+		}
+	}
+
+	DRM_INFO( "\n" );
+	
+	if( ring->head >= 0 && ring->head < ring->size / sizeof(u32) ) {
+		struct list_head *ptr;
+		u32 addr = le32_to_cpu(((u32 *)ring->start)[ring->head + 1]);
+
+		list_for_each(ptr, &dev_priv->pending) {
+			drm_mach64_freelist_t *entry = 
+				list_entry(ptr, drm_mach64_freelist_t, list);
+			drm_buf_t *buf = entry->buf;
+
+			u32 buf_addr = GETBUFADDR( buf );
+			
+			if ( buf_addr <= addr && addr < buf_addr + buf->used ) {
+				mach64_dump_buf_info ( dev_priv, buf );
+			}
+		}
+	}
+
+	DRM_INFO( "\n" );
+	DRM_INFO( "       BM_GUI_TABLE = 0x%08x\n", MACH64_READ( MACH64_BM_GUI_TABLE ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "BM_FRAME_BUF_OFFSET = 0x%08x\n", MACH64_READ( MACH64_BM_FRAME_BUF_OFFSET ) );
+	DRM_INFO( " BM_SYSTEM_MEM_ADDR = 0x%08x\n", MACH64_READ( MACH64_BM_SYSTEM_MEM_ADDR ) );
+	DRM_INFO( "         BM_COMMAND = 0x%08x\n", MACH64_READ( MACH64_BM_COMMAND ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "          BM_STATUS = 0x%08x\n", MACH64_READ( MACH64_BM_STATUS ) );
+	DRM_INFO( "           BUS_CNTL = 0x%08x\n", MACH64_READ( MACH64_BUS_CNTL ) );
+	DRM_INFO( "          FIFO_STAT = 0x%08x\n", MACH64_READ( MACH64_FIFO_STAT ) );
+	DRM_INFO( "           GUI_STAT = 0x%08x\n", MACH64_READ( MACH64_GUI_STAT ) );
+	DRM_INFO( "           SRC_CNTL = 0x%08x\n", MACH64_READ( MACH64_SRC_CNTL ) );
+}
+
+
+/* ================================================================
+ * DMA test and initialization
+ */
+
+static int mach64_bm_dma_test( drm_device_t *dev )
+{
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	dma_addr_t data_handle;
+	void *cpu_addr_data;
+	u32 data_addr;
+	u32 *table, *data;
+	u32 expected[2];
+	u32 src_cntl, pat_reg0, pat_reg1;
+	int i, count, failed;
+
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+	table = (u32 *) dev_priv->ring.start;
+
+	/* FIXME: get a dma buffer from the freelist here */
+	DRM_DEBUG( "Allocating data memory ...\n" );
+	cpu_addr_data = DRM(pci_alloc)( dev, 0x1000, 0x1000, 0xfffffffful, &data_handle );
+	if (!cpu_addr_data || !data_handle) {
+		DRM_INFO( "data-memory allocation failed!\n" );
+		return DRM_ERR(ENOMEM);
+	} else {
+		data = (u32 *) cpu_addr_data;
+		data_addr = (u32) data_handle;
+	}
+
+	/* Save the X server's value for SRC_CNTL and restore it
+	 * in case our test fails.  This prevents the X server
+	 * from disabling it's cache for this register
+	 */
+	src_cntl = MACH64_READ( MACH64_SRC_CNTL );
+	pat_reg0 = MACH64_READ( MACH64_PAT_REG0 );
+	pat_reg1 = MACH64_READ( MACH64_PAT_REG1 );
+
+	mach64_do_wait_for_fifo( dev_priv, 3 );
+
+	MACH64_WRITE( MACH64_SRC_CNTL, 0 );
+	MACH64_WRITE( MACH64_PAT_REG0, 0x11111111 );
+	MACH64_WRITE( MACH64_PAT_REG1, 0x11111111 );
+
+	mach64_do_wait_for_idle( dev_priv );
+
+	for (i=0; i < 2; i++) {
+		u32 reg;
+		reg = MACH64_READ( (MACH64_PAT_REG0 + i*4) );
+		DRM_DEBUG( "(Before DMA Transfer) reg %d = 0x%08x\n", i, reg );
+		if ( reg != 0x11111111 ) {
+			DRM_INFO( "Error initializing test registers\n" );
+			DRM_INFO( "resetting engine ...\n");
+			mach64_do_engine_reset( dev_priv );
+			DRM_INFO( "freeing data buffer memory.\n" );
+			DRM(pci_free)( dev, 0x1000, cpu_addr_data, data_handle );
+			return DRM_ERR(EIO);
+		}
+	}
+
+	/* fill up a buffer with sets of 2 consecutive writes starting with PAT_REG0 */
+	count = 0;
+
+	data[count++] = cpu_to_le32(DMAREG(MACH64_PAT_REG0) | (1 << 16));
+	data[count++] = expected[0] = 0x22222222;
+	data[count++] = expected[1] = 0xaaaaaaaa;
+
+	while (count < 1020) {
+		data[count++] = cpu_to_le32(DMAREG(MACH64_PAT_REG0) | (1 << 16));
+		data[count++] = 0x22222222;
+		data[count++] = 0xaaaaaaaa;
+	}
+	data[count++] = cpu_to_le32(DMAREG(MACH64_SRC_CNTL) | (0 << 16));
+	data[count++] = 0;
+
+	DRM_DEBUG( "Preparing table ...\n" );
+	table[MACH64_DMA_FRAME_BUF_OFFSET] = cpu_to_le32(MACH64_BM_ADDR + 
+							 MACH64_APERTURE_OFFSET);
+	table[MACH64_DMA_SYS_MEM_ADDR]     = cpu_to_le32(data_addr);
+	table[MACH64_DMA_COMMAND]          = cpu_to_le32(count * sizeof( u32 ) 
+							 | MACH64_DMA_HOLD_OFFSET 
+							 | MACH64_DMA_EOL);
+	table[MACH64_DMA_RESERVED]         = 0;
+
+	DRM_DEBUG( "table[0] = 0x%08x\n", table[0] );
+	DRM_DEBUG( "table[1] = 0x%08x\n", table[1] );
+	DRM_DEBUG( "table[2] = 0x%08x\n", table[2] );
+	DRM_DEBUG( "table[3] = 0x%08x\n", table[3] );
+
+	for ( i = 0 ; i < 6 ; i++ ) {
+		DRM_DEBUG( " data[%d] = 0x%08x\n", i, data[i] );
+	}
+	DRM_DEBUG( " ...\n" );
+	for ( i = count-5 ; i < count ; i++ ) {
+		DRM_DEBUG( " data[%d] = 0x%08x\n", i, data[i] );
+	}
+
+	DRM_MEMORYBARRIER();
+
+	DRM_DEBUG( "waiting for idle...\n" );
+	if ( ( i = mach64_do_wait_for_idle( dev_priv ) ) ) {
+		DRM_INFO( "mach64_do_wait_for_idle failed (result=%d)\n", i);
+		DRM_INFO( "resetting engine ...\n");
+		mach64_do_engine_reset( dev_priv );
+		mach64_do_wait_for_fifo( dev_priv, 3 );
+		MACH64_WRITE( MACH64_SRC_CNTL, src_cntl );
+		MACH64_WRITE( MACH64_PAT_REG0, pat_reg0 );
+		MACH64_WRITE( MACH64_PAT_REG1, pat_reg1 );
+		DRM_INFO( "freeing data buffer memory.\n" );
+		DRM(pci_free)( dev, 0x1000, cpu_addr_data, data_handle );
+		return i;
+	}
+	DRM_DEBUG( "waiting for idle...done\n" );
+
+	DRM_DEBUG( "BUS_CNTL = 0x%08x\n", MACH64_READ( MACH64_BUS_CNTL ) );
+	DRM_DEBUG( "SRC_CNTL = 0x%08x\n", MACH64_READ( MACH64_SRC_CNTL ) );
+	DRM_DEBUG( "\n" );
+	DRM_DEBUG( "data bus addr = 0x%08x\n", data_addr );
+	DRM_DEBUG( "table bus addr = 0x%08x\n", dev_priv->ring.start_addr );
+
+	DRM_DEBUG( "starting DMA transfer...\n" );
+	MACH64_WRITE( MACH64_BM_GUI_TABLE_CMD,
+			  dev_priv->ring.start_addr |
+			  MACH64_CIRCULAR_BUF_SIZE_16KB );
+
+	MACH64_WRITE( MACH64_SRC_CNTL, 
+		      MACH64_SRC_BM_ENABLE | MACH64_SRC_BM_SYNC |
+		      MACH64_SRC_BM_OP_SYSTEM_TO_REG );
+
+	/* Kick off the transfer */
+	DRM_DEBUG( "starting DMA transfer... done.\n" );
+	MACH64_WRITE( MACH64_DST_HEIGHT_WIDTH, 0 );
+
+	DRM_DEBUG( "waiting for idle...\n" );
+
+	if ( ( i = mach64_do_wait_for_idle( dev_priv ) ) ) {
+		/* engine locked up, dump register state and reset */
+		DRM_INFO( "mach64_do_wait_for_idle failed (result=%d)\n", i);
+		mach64_dump_engine_info( dev_priv );
+		DRM_INFO( "resetting engine ...\n");
+		mach64_do_engine_reset( dev_priv );
+		mach64_do_wait_for_fifo( dev_priv, 3 );
+		MACH64_WRITE( MACH64_SRC_CNTL, src_cntl );
+		MACH64_WRITE( MACH64_PAT_REG0, pat_reg0 );
+		MACH64_WRITE( MACH64_PAT_REG1, pat_reg1 );
+		DRM_INFO( "freeing data buffer memory.\n" );
+		DRM(pci_free)( dev, 0x1000, cpu_addr_data, data_handle );
+		return i;
+	}
+
+	DRM_DEBUG( "waiting for idle...done\n" );
+
+	/* restore SRC_CNTL */
+	mach64_do_wait_for_fifo( dev_priv, 1 );
+	MACH64_WRITE( MACH64_SRC_CNTL, src_cntl );
+
+	failed = 0;
+
+	/* Check register values to see if the GUI master operation succeeded */
+	for ( i = 0; i < 2; i++ ) {
+		u32 reg;
+		reg = MACH64_READ( (MACH64_PAT_REG0 + i*4) );
+		DRM_DEBUG( "(After DMA Transfer) reg %d = 0x%08x\n", i, reg );
+		if (reg != expected[i]) {
+			failed = -1;
+		}
+	}
+
+	/* restore pattern registers */
+	mach64_do_wait_for_fifo( dev_priv, 2 );
+	MACH64_WRITE( MACH64_PAT_REG0, pat_reg0 );
+	MACH64_WRITE( MACH64_PAT_REG1, pat_reg1 );
+
+	DRM_DEBUG( "freeing data buffer memory.\n" );
+	DRM(pci_free)( dev, 0x1000, cpu_addr_data, data_handle );
+	DRM_DEBUG( "returning ...\n" );
+
+	return failed;
+}
+
+
+static int mach64_do_dma_init( drm_device_t *dev, drm_mach64_init_t *init )
+{
+	drm_mach64_private_t *dev_priv;
+	u32 tmp;
+	int i, ret;
+
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+	dev_priv = DRM(alloc)( sizeof(drm_mach64_private_t), DRM_MEM_DRIVER );
+	if ( dev_priv == NULL )
+		return DRM_ERR(ENOMEM);
+	
+	memset( dev_priv, 0, sizeof(drm_mach64_private_t) );
+
+	dev_priv->is_pci	= init->is_pci;
+
+	dev_priv->fb_bpp	= init->fb_bpp;
+	dev_priv->front_offset  = init->front_offset;
+	dev_priv->front_pitch   = init->front_pitch;
+	dev_priv->back_offset   = init->back_offset;
+	dev_priv->back_pitch    = init->back_pitch;
+
+	dev_priv->depth_bpp     = init->depth_bpp;
+	dev_priv->depth_offset  = init->depth_offset;
+	dev_priv->depth_pitch   = init->depth_pitch;
+
+	dev_priv->front_offset_pitch	= (((dev_priv->front_pitch/8) << 22) |
+					   (dev_priv->front_offset >> 3));
+	dev_priv->back_offset_pitch	= (((dev_priv->back_pitch/8) << 22) |
+					   (dev_priv->back_offset >> 3));
+	dev_priv->depth_offset_pitch	= (((dev_priv->depth_pitch/8) << 22) |
+					   (dev_priv->depth_offset >> 3));
+
+	dev_priv->usec_timeout		= 1000000;
+
+	/* Set up the freelist, placeholder list and pending list */
+	INIT_LIST_HEAD(&dev_priv->free_list);
+	INIT_LIST_HEAD(&dev_priv->placeholders);
+	INIT_LIST_HEAD(&dev_priv->pending);
+
+	DRM_GETSAREA();
+
+	if (!dev_priv->sarea) {
+		DRM_ERROR("can not find sarea!\n");
+		dev->dev_private = (void *)dev_priv;
+	   	mach64_do_cleanup_dma(dev);
+	   	return DRM_ERR(EINVAL);
+	}
+	DRM_FIND_MAP( dev_priv->fb, init->fb_offset );
+	if (!dev_priv->fb) {
+		DRM_ERROR("can not find frame buffer map!\n");
+		dev->dev_private = (void *)dev_priv;
+	   	mach64_do_cleanup_dma(dev);
+	   	return DRM_ERR(EINVAL);
+	}
+	DRM_FIND_MAP( dev_priv->mmio, init->mmio_offset );
+	if (!dev_priv->mmio) {
+		DRM_ERROR("can not find mmio map!\n");
+		dev->dev_private = (void *)dev_priv;
+	   	mach64_do_cleanup_dma(dev);
+	   	return DRM_ERR(EINVAL);
+	}
+
+	dev_priv->sarea_priv = (drm_mach64_sarea_t *)
+		((u8 *)dev_priv->sarea->handle +
+		 init->sarea_priv_offset);
+
+	if( !dev_priv->is_pci ) {
+		DRM_FIND_MAP( dev_priv->ring_map, init->ring_offset );
+		if ( !dev_priv->ring_map ) {
+			DRM_ERROR( "can not find ring map!\n" );
+			dev->dev_private = (void *)dev_priv;
+			mach64_do_cleanup_dma(dev);
+			return -EINVAL;
+		}
+		DRM_IOREMAP( dev_priv->ring_map, dev );
+		if ( !dev_priv->ring_map->handle ) {
+		        DRM_ERROR( "can not ioremap virtual address for"
+                                   " descriptor ring\n" );
+			dev->dev_private = (void *) dev_priv;
+			mach64_do_cleanup_dma( dev );
+			return -ENOMEM;
+                }
+	        DRM_FIND_MAP( dev_priv->buffers, init->buffers_offset );
+		if ( !dev_priv->buffers ) {
+			DRM_ERROR( "can not find dma buffer map!\n" );
+			dev->dev_private = (void *)dev_priv;
+			mach64_do_cleanup_dma( dev );
+			return DRM_ERR(EINVAL);
+		}
+		DRM_IOREMAP( dev_priv->buffers, dev );
+		if ( !dev_priv->buffers->handle ) {
+			DRM_ERROR( "can not ioremap virtual address for"
+				   " dma buffer\n" );
+			dev->dev_private = (void *) dev_priv;
+			mach64_do_cleanup_dma( dev );
+			return DRM_ERR(ENOMEM);
+		}
+		DRM_FIND_MAP( dev_priv->agp_textures,
+			      init->agp_textures_offset );
+		if (!dev_priv->agp_textures) {
+			DRM_ERROR( "can not find agp texture region!\n" );
+			dev->dev_private = (void *)dev_priv;
+			mach64_do_cleanup_dma( dev );
+			return DRM_ERR(EINVAL);
+		}
+	}
+
+	dev->dev_private = (void *) dev_priv;
+
+	dev_priv->driver_mode = init->dma_mode;
+
+	/* changing the FIFO size from the default causes problems with DMA */
+	tmp = MACH64_READ( MACH64_GUI_CNTL );
+	if ( (tmp & MACH64_CMDFIFO_SIZE_MASK) != MACH64_CMDFIFO_SIZE_128 ) {
+		DRM_INFO( "Setting FIFO size to 128 entries\n");
+		/* FIFO must be empty to change the FIFO depth */
+		if ((ret=mach64_do_wait_for_idle( dev_priv ))) {
+			DRM_ERROR("wait for idle failed before changing FIFO depth!\n");
+			mach64_do_cleanup_dma( dev );
+			return ret;
+		}
+		MACH64_WRITE( MACH64_GUI_CNTL, ( ( tmp & ~MACH64_CMDFIFO_SIZE_MASK ) \
+						 | MACH64_CMDFIFO_SIZE_128 ) );
+		/* need to read GUI_STAT for proper sync according to docs */
+		if ((ret=mach64_do_wait_for_idle( dev_priv ))) {
+			DRM_ERROR("wait for idle failed when changing FIFO depth!\n");
+			mach64_do_cleanup_dma( dev );
+			return ret;
+		}
+	}
+
+	/* allocate descriptor memory from pci pool */
+	DRM_DEBUG( "Allocating dma descriptor ring\n" );
+	dev_priv->ring.size = 0x4000; /* 16KB */
+
+	if ( dev_priv->is_pci ) {
+		dev_priv->ring.start = DRM(pci_alloc)( dev, dev_priv->ring.size, 
+						       dev_priv->ring.size, 0xfffffffful,
+						       &dev_priv->ring.handle );
+
+		if (!dev_priv->ring.start || !dev_priv->ring.handle) {
+			DRM_ERROR( "Allocating dma descriptor ring failed\n");
+			return DRM_ERR(ENOMEM);
+		} else {
+			dev_priv->ring.start_addr = (u32) dev_priv->ring.handle;
+		}
+        } else {
+		dev_priv->ring.start = dev_priv->ring_map->handle;
+		dev_priv->ring.start_addr = (u32) dev_priv->ring_map->offset;
+	}
+
+	memset( dev_priv->ring.start, 0, dev_priv->ring.size );
+	DRM_INFO( "descriptor ring: cpu addr 0x%08x, bus addr: 0x%08x\n", 
+		  (u32) dev_priv->ring.start, dev_priv->ring.start_addr );
+
+	ret = 0;
+	if ( dev_priv->driver_mode != MACH64_MODE_MMIO ) {
+
+		/* enable block 1 registers and bus mastering */
+		MACH64_WRITE( MACH64_BUS_CNTL, 
+			      ( ( MACH64_READ(MACH64_BUS_CNTL) 
+				  | MACH64_BUS_EXT_REG_EN ) 
+				& ~MACH64_BUS_MASTER_DIS ) );
+
+		/* try a DMA GUI-mastering pass and fall back to MMIO if it fails */
+		DRM_DEBUG( "Starting DMA test...\n");
+		if ( (ret=mach64_bm_dma_test( dev )) ) {
+			dev_priv->driver_mode = MACH64_MODE_MMIO;
+		}
+	}
+
+	switch (dev_priv->driver_mode) {
+	case MACH64_MODE_MMIO:
+		MACH64_WRITE( MACH64_BUS_CNTL, ( MACH64_READ(MACH64_BUS_CNTL) 
+						 | MACH64_BUS_EXT_REG_EN
+						 | MACH64_BUS_MASTER_DIS ) );
+		if ( init->dma_mode == MACH64_MODE_MMIO )
+			DRM_INFO( "Forcing pseudo-DMA mode\n" );
+		else
+			DRM_INFO( "DMA test failed (ret=%d), using pseudo-DMA mode\n", ret );
+		break;
+	case MACH64_MODE_DMA_SYNC:
+		DRM_INFO( "DMA test succeeded, using synchronous DMA mode\n");
+		break;
+	case MACH64_MODE_DMA_ASYNC:
+	default:
+		DRM_INFO( "DMA test succeeded, using asynchronous DMA mode\n");
+	}
+
+	dev_priv->ring_running = 0;
+
+	/* setup offsets for physical address of table start and end */
+	dev_priv->ring.head_addr = dev_priv->ring.start_addr;
+	dev_priv->ring.head = dev_priv->ring.tail = 0;
+	dev_priv->ring.tail_mask = (dev_priv->ring.size / sizeof(u32)) - 1;
+	dev_priv->ring.space = dev_priv->ring.size;
+
+	/* setup physical address and size of descriptor table */
+	mach64_do_wait_for_fifo( dev_priv, 1 );
+	MACH64_WRITE( MACH64_BM_GUI_TABLE_CMD, 
+		      ( dev_priv->ring.head_addr | MACH64_CIRCULAR_BUF_SIZE_16KB ) );
+
+	/* init frame counter */
+	dev_priv->sarea_priv->frames_queued = 0;
+	for (i = 0; i < MACH64_MAX_QUEUED_FRAMES; i++) {
+		dev_priv->frame_ofs[i] = ~0; /* All ones indicates placeholder */
+	}
+
+	/* Allocate the DMA buffer freelist */
+	if ( (ret=mach64_init_freelist( dev )) ) {
+		DRM_ERROR("Freelist allocation failed\n");
+		mach64_do_cleanup_dma( dev );
+		return ret;
+	}
+
+	return 0;
+}
+
+/* ===================================================================
+ * MMIO Pseudo-DMA (intended primarily for debugging, not performance)
+ */
+
+int mach64_do_dispatch_pseudo_dma( drm_mach64_private_t *dev_priv )
+{
+	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;
+	volatile u32 *ring_read;
+	struct list_head *ptr;
+	drm_mach64_freelist_t *entry;
+	drm_buf_t *buf = NULL;
+	u32 *buf_ptr;
+	u32 used, reg, target;
+	int fifo, count, found, ret, no_idle_wait;
+
+	fifo = count = reg = no_idle_wait = 0;
+	target = MACH64_BM_ADDR;
+
+	if ( (ret=mach64_do_wait_for_idle( dev_priv )) < 0) {
+		DRM_INFO( "%s: idle failed before pseudo-dma dispatch, resetting engine\n", 
+			  __FUNCTION__);
+		mach64_dump_engine_info( dev_priv );
+		mach64_do_engine_reset( dev_priv );
+		return ret;
+	}
+
+	ring_read = (u32 *) ring->start;
+
+	while ( ring->tail != ring->head ) {
+		u32 buf_addr, new_target, offset; 
+		u32 bytes, remaining, head, eol;
+
+		head = ring->head;
+
+		new_target = le32_to_cpu( ring_read[head++] ) - MACH64_APERTURE_OFFSET;
+		buf_addr   = le32_to_cpu( ring_read[head++] );
+		eol        = le32_to_cpu( ring_read[head]   ) & MACH64_DMA_EOL;
+		bytes      = le32_to_cpu( ring_read[head++] )
+			& ~(MACH64_DMA_HOLD_OFFSET | MACH64_DMA_EOL);
+		head++;
+		head &= ring->tail_mask;
+
+		/* can't wait for idle between a blit setup descriptor 
+		 * and a HOSTDATA descriptor or the engine will lock
+		 */
+		if (new_target == MACH64_BM_HOSTDATA && target == MACH64_BM_ADDR)
+			no_idle_wait = 1;
+
+		target = new_target;
+
+		found = 0;
+		offset = 0;
+		list_for_each(ptr, &dev_priv->pending)
+		{
+			entry = list_entry(ptr, drm_mach64_freelist_t, list);
+			buf = entry->buf;
+			offset = buf_addr - GETBUFADDR( buf );
+			if (offset >= 0 && offset < MACH64_BUFFER_SIZE) {
+				found = 1;
+				break;
+			}
+		}
+
+		if (!found || buf == NULL) {
+			DRM_ERROR("Couldn't find pending buffer: head: %u tail: %u buf_addr: 0x%08x %s\n",
+				  head, ring->tail, buf_addr, (eol ? "eol" : ""));
+			mach64_dump_ring_info( dev_priv );
+			mach64_do_engine_reset( dev_priv );
+			return DRM_ERR(EINVAL);
+		}
+
+		/* Hand feed the buffer to the card via MMIO, waiting for the fifo 
+		 * every 16 writes 
+		 */
+		DRM_DEBUG("target: (0x%08x) %s\n", target, 
+				 (target == MACH64_BM_HOSTDATA ? "BM_HOSTDATA" : "BM_ADDR"));
+		DRM_DEBUG("offset: %u bytes: %u used: %u\n", offset, bytes, buf->used);
+
+		remaining = (buf->used - offset) >> 2; /* dwords remaining in buffer */
+		used = bytes >> 2; /* dwords in buffer for this descriptor */
+		buf_ptr = (u32 *)((char *)GETBUFPTR( buf ) + offset);
+
+		while ( used ) {
+			
+			if (count == 0) {
+				if (target == MACH64_BM_HOSTDATA) {
+					reg = DMAREG(MACH64_HOST_DATA0);
+					count = (remaining > 16) ? 16 : remaining;
+					fifo = 0;
+				} else {
+					reg = le32_to_cpu(*buf_ptr++);
+					used--;
+					count = (reg >> 16) + 1;
+				}
+
+				reg = reg & 0xffff;
+				reg = MMSELECT( reg );
+			}
+			while ( count && used ) {
+				if ( !fifo ) {
+					if (no_idle_wait) {
+						if ( (ret=mach64_do_wait_for_fifo( dev_priv, 16 )) < 0 ) {
+							no_idle_wait = 0;
+							return ret;
+						}
+					} else {
+						if ( (ret=mach64_do_wait_for_idle( dev_priv )) < 0 ) {
+							return ret;
+						}
+					}
+					fifo = 16;
+				}
+				--fifo;
+				MACH64_WRITE(reg, le32_to_cpu(*buf_ptr++));
+				used--; remaining--;
+					
+				reg += 4;
+				count--;
+			}
+		}
+		ring->head = head;
+		ring->head_addr = ring->start_addr + (ring->head * sizeof(u32));
+		ring->space += (4 * sizeof(u32));
+	}
+
+	if ( (ret=mach64_do_wait_for_idle( dev_priv )) < 0 ) {
+		return ret;
+	}
+	MACH64_WRITE( MACH64_BM_GUI_TABLE_CMD, 
+		      ring->head_addr | MACH64_CIRCULAR_BUF_SIZE_16KB );
+
+	DRM_DEBUG( "%s completed\n", __FUNCTION__ );
+	return 0;
+}
+
+/* ================================================================
+ * DMA cleanup
+ */
+
+int mach64_do_cleanup_dma( drm_device_t *dev )
+{
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+#if _HAVE_DMA_IRQ
+	/* Make sure interrupts are disabled here because the uninstall ioctl
+	 * may not have been called from userspace and after dev_private
+	 * is freed, it's too late.
+	 */
+	if ( dev->irq ) DRM(irq_uninstall)(dev);
+#endif
+
+	if ( dev->dev_private ) {
+		drm_mach64_private_t *dev_priv = dev->dev_private;
+
+		if ( dev_priv->is_pci ) {
+			if ( (dev_priv->ring.start != NULL) && dev_priv->ring.handle ) {
+				DRM(pci_free)( dev, dev_priv->ring.size, 
+					       dev_priv->ring.start, dev_priv->ring.handle );
+			}
+		} else {
+			if ( dev_priv->ring_map )
+				DRM_IOREMAPFREE( dev_priv->ring_map, dev );
+		}
+
+		if ( dev_priv->buffers )
+			DRM_IOREMAPFREE( dev_priv->buffers, dev );
+
+		mach64_destroy_freelist( dev );
+
+		DRM(free)( dev_priv, sizeof(drm_mach64_private_t),
+			   DRM_MEM_DRIVER );
+		dev->dev_private = NULL;
+	}
+
+	return 0;
+}
+
+/* ================================================================
+ * IOCTL handlers
+ */
+
+int mach64_dma_init( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_mach64_init_t init;
+		
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+	LOCK_TEST_WITH_RETURN( dev, filp );
+
+	DRM_COPY_FROM_USER_IOCTL( init, (drm_mach64_init_t *)data, 
+	    sizeof(init) );
+
+	switch ( init.func ) {
+	case DRM_MACH64_INIT_DMA:
+		return mach64_do_dma_init( dev, &init );
+	case DRM_MACH64_CLEANUP_DMA:
+		return mach64_do_cleanup_dma( dev );
+	}
+		
+	return DRM_ERR(EINVAL);
+}
+
+int mach64_dma_idle( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+	LOCK_TEST_WITH_RETURN( dev, filp );
+
+	return mach64_do_dma_idle( dev_priv );
+}
+
+int mach64_dma_flush( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+	LOCK_TEST_WITH_RETURN( dev, filp );
+
+	return mach64_do_dma_flush( dev_priv );
+}
+
+int mach64_engine_reset( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+	LOCK_TEST_WITH_RETURN( dev, filp );
+
+	return mach64_do_engine_reset( dev_priv );
+}
+
+
+/* ================================================================
+ * Freelist management
+ */
+
+int mach64_init_freelist( drm_device_t *dev )
+{
+	drm_device_dma_t *dma = dev->dma;
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	drm_mach64_freelist_t *entry;
+	struct list_head *ptr;
+	int i;
+
+	DRM_DEBUG("%s: adding %d buffers to freelist\n", __FUNCTION__, dma->buf_count);
+
+	for ( i = 0 ; i < dma->buf_count ; i++ ) {
+		if ((entry = 
+		     (drm_mach64_freelist_t *) DRM(alloc)(sizeof(drm_mach64_freelist_t), 
+							  DRM_MEM_BUFLISTS)) == NULL)
+			return DRM_ERR(ENOMEM);
+		memset( entry, 0, sizeof(drm_mach64_freelist_t) );
+		entry->buf = dma->buflist[i];
+		ptr = &entry->list;
+		list_add_tail(ptr, &dev_priv->free_list);
+	}
+
+	return 0;
+}
+
+void mach64_destroy_freelist( drm_device_t *dev )
+{
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	drm_mach64_freelist_t *entry;
+	struct list_head *ptr;
+	struct list_head *tmp;
+
+	DRM_DEBUG("%s\n", __FUNCTION__);
+
+	list_for_each_safe(ptr, tmp, &dev_priv->pending)
+	{
+		list_del(ptr);
+		entry = list_entry(ptr, drm_mach64_freelist_t, list);
+		DRM(free)(entry, sizeof(*entry), DRM_MEM_BUFLISTS);
+	}
+	list_for_each_safe(ptr, tmp, &dev_priv->placeholders)
+	{
+		list_del(ptr);
+		entry = list_entry(ptr, drm_mach64_freelist_t, list);
+		DRM(free)(entry, sizeof(*entry), DRM_MEM_BUFLISTS);
+	}
+
+	list_for_each_safe(ptr, tmp, &dev_priv->free_list)
+	{
+		list_del(ptr);
+		entry = list_entry(ptr, drm_mach64_freelist_t, list);
+		DRM(free)(entry, sizeof(*entry), DRM_MEM_BUFLISTS);
+	}
+}
+
+/* IMPORTANT: This function should only be called when the engine is idle or locked up, 
+ * as it assumes all buffers in the pending list have been completed by the hardware.
+ */
+int mach64_do_release_used_buffers( drm_mach64_private_t *dev_priv )
+{
+	struct list_head *ptr;
+	struct list_head *tmp;
+	drm_mach64_freelist_t *entry;
+	int i;
+
+	if ( list_empty(&dev_priv->pending) )
+		return 0;
+
+	/* Iterate the pending list and move all buffers into the freelist... */
+	i = 0;
+	list_for_each_safe(ptr, tmp, &dev_priv->pending)
+	{
+		entry = list_entry(ptr, drm_mach64_freelist_t, list);
+		if (entry->discard) {
+			entry->buf->pending = 0;
+			list_del(ptr);
+			list_add_tail(ptr, &dev_priv->free_list);
+			i++;
+		}
+	}
+
+	DRM_DEBUG( "%s: released %d buffers from pending list\n", __FUNCTION__, i );
+
+        return 0;
+}
+
+drm_buf_t *mach64_freelist_get( drm_mach64_private_t *dev_priv )
+{
+	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;
+	drm_mach64_freelist_t *entry;
+	struct list_head *ptr;
+	struct list_head *tmp;
+	int t;
+
+	if ( list_empty(&dev_priv->free_list) ) {
+		u32 head, tail, ofs;
+
+		if ( list_empty( &dev_priv->pending ) ) {
+			DRM_ERROR( "Couldn't get buffer - pending and free lists empty\n" );
+			t = 0;
+			list_for_each( ptr, &dev_priv->placeholders ) {
+				t++;
+			}
+			DRM_INFO( "Placeholders: %d\n", t );
+			return NULL;
+		}
+
+		tail = ring->tail;
+		for ( t = 0 ; t < dev_priv->usec_timeout ; t++ ) {
+			mach64_ring_tick( dev_priv, ring );
+			head = ring->head;
+
+			if ( head == tail ) {
+#if MACH64_EXTRA_CHECKING
+				if ( MACH64_READ(MACH64_GUI_STAT) & MACH64_GUI_ACTIVE ) {
+					DRM_ERROR( "Empty ring with non-idle engine!\n" );
+					mach64_dump_ring_info( dev_priv );
+					return NULL;
+				}
+#endif
+				/* last pass is complete, so release everything */
+				mach64_do_release_used_buffers( dev_priv );
+				DRM_DEBUG( "%s: idle engine, freed all buffers.\n", __FUNCTION__ );
+				if ( list_empty(&dev_priv->free_list) ) {
+					DRM_ERROR( "Freelist empty with idle engine\n" );
+					return NULL;
+				}
+				goto _freelist_entry_found;
+			}
+			/* Look for a completed buffer and bail out of the loop 
+			 * as soon as we find one -- don't waste time trying
+			 * to free extra bufs here, leave that to do_release_used_buffers
+			 */
+			list_for_each_safe(ptr, tmp, &dev_priv->pending) {
+				entry = list_entry(ptr, drm_mach64_freelist_t, list);
+				ofs = entry->ring_ofs;
+				if ( entry->discard &&
+				     ((head < tail && (ofs < head || ofs >= tail)) ||
+				      (head > tail && (ofs < head && ofs >= tail))) ) {
+#if MACH64_EXTRA_CHECKING
+					int i;
+					
+					for ( i = head ; i != tail ; i = (i + 4) & ring->tail_mask ) {
+						u32 o1 = le32_to_cpu(((u32 *)ring->start)[i + 1]);
+						u32 o2 = GETBUFADDR( entry->buf );
+
+						if ( o1 == o2 ) {
+							DRM_ERROR ( "Attempting to free used buffer: "
+								    "i=%d  buf=0x%08x\n", i, o1 );
+							mach64_dump_ring_info( dev_priv );
+							return NULL;
+						}
+					}
+#endif
+					/* found a processed buffer */
+					entry->buf->pending = 0;
+					list_del(ptr);
+					entry->buf->used = 0;
+					list_add_tail(ptr, &dev_priv->placeholders);
+					DRM_DEBUG( "%s: freed processed buffer (head=%d tail=%d "
+						   "buf ring ofs=%d).\n", __FUNCTION__, head, tail, ofs );
+					return entry->buf;
+				}
+			}
+			DRM_UDELAY( 1 );
+		}
+		mach64_dump_ring_info( dev_priv );
+		DRM_ERROR( "timeout waiting for buffers: ring head_addr: 0x%08x head: %d tail: %d\n", 
+			   ring->head_addr, ring->head, ring->tail );
+		return NULL;
+	}
+
+_freelist_entry_found:
+	ptr = dev_priv->free_list.next;
+	list_del(ptr);
+	entry = list_entry(ptr, drm_mach64_freelist_t, list);
+	entry->buf->used = 0;
+	list_add_tail(ptr, &dev_priv->placeholders);
+	return entry->buf;
+}
+
+/* ================================================================
+ * DMA buffer request and submission IOCTL handler
+ */
+
+static int mach64_dma_get_buffers( DRMFILE filp, drm_device_t *dev, drm_dma_t *d )
+{
+	int i;
+	drm_buf_t *buf;
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+
+	for ( i = d->granted_count ; i < d->request_count ; i++ ) {
+		buf = mach64_freelist_get( dev_priv );
+#if MACH64_EXTRA_CHECKING
+		if ( !buf ) return DRM_ERR(EFAULT);
+#else
+		if ( !buf ) return DRM_ERR(EAGAIN);
+#endif
+
+		buf->filp = filp;
+
+		if ( DRM_COPY_TO_USER( &d->request_indices[i], &buf->idx,
+				   sizeof(buf->idx) ) )
+			return DRM_ERR(EFAULT);
+		if ( DRM_COPY_TO_USER( &d->request_sizes[i], &buf->total,
+				   sizeof(buf->total) ) )
+			return DRM_ERR(EFAULT);
+
+		d->granted_count++;
+	}
+	return 0;
+}
+
+int mach64_dma_buffers( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_device_dma_t *dma = dev->dma;
+	drm_dma_t d;
+        int ret = 0;
+
+        LOCK_TEST_WITH_RETURN( dev, filp );
+
+	DRM_COPY_FROM_USER_IOCTL( d, (drm_dma_t *)data, sizeof(d) );
+
+        /* Please don't send us buffers.
+	 */
+        if ( d.send_count != 0 ) 
+        {
+		DRM_ERROR( "Process %d trying to send %d buffers via drmDMA\n",
+			   DRM_CURRENTPID, d.send_count );
+		return DRM_ERR(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",
+			   DRM_CURRENTPID, d.request_count, dma->buf_count );
+		ret = DRM_ERR(EINVAL);
+	}
+        
+	d.granted_count = 0;
+
+	if ( d.request_count ) 
+	{
+		ret = mach64_dma_get_buffers( filp, dev, &d );
+	}
+
+	DRM_COPY_TO_USER_IOCTL( (drm_dma_t *)data, d, sizeof(d) );
+
+        return ret;
+}
+
diff --git a/shared-core/mach64_drm.h b/shared-core/mach64_drm.h
new file mode 100644
index 00000000..9d748a37
--- /dev/null
+++ b/shared-core/mach64_drm.h
@@ -0,0 +1,258 @@
+/* mach64_drm.h -- Public header for the mach64 driver -*- linux-c -*-
+ * Created: Thu Nov 30 20:04:32 2000 by gareth@valinux.com
+ *
+ * Copyright 2000 Gareth Hughes
+ * Copyright 2002 Frank C. Earl
+ * Copyright 2002-2003 Leif Delgass
+ * 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
+ * THE COPYRIGHT OWNER(S) 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:
+ *    Gareth Hughes <gareth@valinux.com>
+ *    Frank C. Earl <fearl@airmail.net>
+ *    Leif Delgass <ldelgass@retinalburn.net>
+ */
+
+#ifndef __MACH64_DRM_H__
+#define __MACH64_DRM_H__
+
+
+/* WARNING: If you change any of these defines, make sure to change the
+ * defines in the Xserver file (mach64_sarea.h)
+ */
+#ifndef __MACH64_SAREA_DEFINES__
+#define __MACH64_SAREA_DEFINES__
+
+/* What needs to be changed for the current vertex buffer?
+ * GH: We're going to be pedantic about this.  We want the card to do as
+ * little as possible, so let's avoid having it fetch a whole bunch of
+ * register values that don't change all that often, if at all.
+ */
+#define MACH64_UPLOAD_DST_OFF_PITCH	0x0001
+#define MACH64_UPLOAD_Z_OFF_PITCH	0x0002
+#define MACH64_UPLOAD_Z_ALPHA_CNTL	0x0004
+#define MACH64_UPLOAD_SCALE_3D_CNTL	0x0008
+#define MACH64_UPLOAD_DP_FOG_CLR	0x0010
+#define MACH64_UPLOAD_DP_WRITE_MASK	0x0020
+#define MACH64_UPLOAD_DP_PIX_WIDTH	0x0040
+#define MACH64_UPLOAD_SETUP_CNTL	0x0080
+#define MACH64_UPLOAD_MISC		0x0100
+#define MACH64_UPLOAD_TEXTURE		0x0200
+#define MACH64_UPLOAD_TEX0IMAGE		0x0400
+#define MACH64_UPLOAD_TEX1IMAGE		0x0800
+#define MACH64_UPLOAD_CLIPRECTS		0x1000 /* handled client-side */
+#define MACH64_UPLOAD_CONTEXT		0x00ff
+#define MACH64_UPLOAD_ALL		0x1fff
+
+/* DMA buffer size
+ */
+#define MACH64_BUFFER_SIZE		16384
+
+/* Max number of swaps allowed on the ring
+ * before the client must wait
+ */
+#define MACH64_MAX_QUEUED_FRAMES        3
+
+/* Byte offsets for host blit buffer data
+ */
+#define MACH64_HOSTDATA_BLIT_OFFSET	104
+
+/* Keep these small for testing.
+ */
+#define MACH64_NR_SAREA_CLIPRECTS	8
+
+
+#define MACH64_CARD_HEAP		0
+#define MACH64_AGP_HEAP			1
+#define MACH64_NR_TEX_HEAPS		2
+#define MACH64_NR_TEX_REGIONS		64
+#define MACH64_LOG_TEX_GRANULARITY	16
+
+#define MACH64_TEX_MAXLEVELS		1
+
+#define MACH64_NR_CONTEXT_REGS		15
+#define MACH64_NR_TEXTURE_REGS		4
+
+#endif /* __MACH64_SAREA_DEFINES__ */
+
+typedef struct {
+	unsigned int dst_off_pitch;
+
+	unsigned int z_off_pitch;
+	unsigned int z_cntl;
+	unsigned int alpha_tst_cntl;
+
+	unsigned int scale_3d_cntl;
+
+	unsigned int sc_left_right;
+	unsigned int sc_top_bottom;
+
+	unsigned int dp_fog_clr;
+	unsigned int dp_write_mask;
+	unsigned int dp_pix_width;
+	unsigned int dp_mix;
+	unsigned int dp_src;
+
+	unsigned int clr_cmp_cntl;
+	unsigned int gui_traj_cntl;
+
+	unsigned int setup_cntl;
+
+	unsigned int tex_size_pitch;
+	unsigned int tex_cntl;
+	unsigned int secondary_tex_off;
+	unsigned int tex_offset;
+} drm_mach64_context_regs_t;
+
+typedef struct drm_mach64_sarea {
+	/* The channel for communication of state information to the kernel
+	 * on firing a vertex dma buffer.
+	 */
+	drm_mach64_context_regs_t context_state;
+	unsigned int dirty;
+	unsigned int vertsize;
+
+	/* The current cliprects, or a subset thereof.
+	 */
+	drm_clip_rect_t boxes[MACH64_NR_SAREA_CLIPRECTS];
+	unsigned int nbox;
+
+	/* Counters for client-side throttling of rendering clients.
+	 */
+	unsigned int frames_queued;
+
+	/* Texture memory LRU.
+	 */
+	drm_tex_region_t tex_list[MACH64_NR_TEX_HEAPS][MACH64_NR_TEX_REGIONS+1];
+	unsigned int tex_age[MACH64_NR_TEX_HEAPS];
+	int ctx_owner;
+} drm_mach64_sarea_t;
+
+
+/* WARNING: If you change any of these defines, make sure to change the
+ * defines in the Xserver file (mach64_common.h)
+ */
+
+/* Mach64 specific ioctls 
+ * The device specific ioctl range is 0x40 to 0x79.
+ */
+
+#define DRM_MACH64_INIT           0x00
+#define DRM_MACH64_IDLE           0x01
+#define DRM_MACH64_RESET          0x02
+#define DRM_MACH64_SWAP           0x03
+#define DRM_MACH64_CLEAR          0x04
+#define DRM_MACH64_VERTEX         0x05
+#define DRM_MACH64_BLIT           0x06
+#define DRM_MACH64_FLUSH          0x07
+#define DRM_MACH64_GETPARAM       0x08
+
+#define DRM_IOCTL_MACH64_INIT           DRM_IOW( DRM_COMMAND_BASE + DRM_MACH64_INIT, drm_mach64_init_t)
+#define DRM_IOCTL_MACH64_IDLE           DRM_IO(  DRM_COMMAND_BASE + DRM_MACH64_IDLE )
+#define DRM_IOCTL_MACH64_RESET          DRM_IO(  DRM_COMMAND_BASE + DRM_MACH64_RESET )
+#define DRM_IOCTL_MACH64_SWAP           DRM_IO(  DRM_COMMAND_BASE + DRM_MACH64_SWAP )
+#define DRM_IOCTL_MACH64_CLEAR          DRM_IOW( DRM_COMMAND_BASE + DRM_MACH64_CLEAR, drm_mach64_clear_t)
+#define DRM_IOCTL_MACH64_VERTEX         DRM_IOW( DRM_COMMAND_BASE + DRM_MACH64_VERTEX, drm_mach64_vertex_t)
+#define DRM_IOCTL_MACH64_BLIT           DRM_IOW( DRM_COMMAND_BASE + DRM_MACH64_BLIT, drm_mach64_blit_t)
+#define DRM_IOCTL_MACH64_FLUSH          DRM_IO(  DRM_COMMAND_BASE + DRM_MACH64_FLUSH )
+#define DRM_IOCTL_MACH64_GETPARAM       DRM_IOWR( DRM_COMMAND_BASE + DRM_MACH64_GETPARAM, drm_mach64_getparam_t)
+
+/* Buffer flags for clears
+ */
+#define MACH64_FRONT			0x1
+#define MACH64_BACK			0x2
+#define MACH64_DEPTH			0x4
+
+/* Primitive types for vertex buffers
+ */
+#define MACH64_PRIM_POINTS		0x00000000
+#define MACH64_PRIM_LINES		0x00000001
+#define MACH64_PRIM_LINE_LOOP		0x00000002
+#define MACH64_PRIM_LINE_STRIP		0x00000003
+#define MACH64_PRIM_TRIANGLES		0x00000004
+#define MACH64_PRIM_TRIANGLE_STRIP	0x00000005
+#define MACH64_PRIM_TRIANGLE_FAN	0x00000006
+#define MACH64_PRIM_QUADS		0x00000007
+#define MACH64_PRIM_QUAD_STRIP		0x00000008
+#define MACH64_PRIM_POLYGON		0x00000009
+
+
+typedef enum _drm_mach64_dma_mode_t {
+   MACH64_MODE_DMA_ASYNC,
+   MACH64_MODE_DMA_SYNC,
+   MACH64_MODE_MMIO
+} drm_mach64_dma_mode_t;
+
+typedef struct drm_mach64_init {
+	enum {
+		DRM_MACH64_INIT_DMA = 0x01,
+		DRM_MACH64_CLEANUP_DMA = 0x02
+	} func;
+
+	unsigned long sarea_priv_offset;
+	int is_pci;
+	drm_mach64_dma_mode_t dma_mode;
+
+	unsigned int fb_bpp;
+	unsigned int front_offset, front_pitch;
+	unsigned int back_offset, back_pitch;
+
+	unsigned int depth_bpp;
+	unsigned int depth_offset, depth_pitch;
+
+	unsigned long fb_offset;
+	unsigned long mmio_offset;
+	unsigned long ring_offset;
+	unsigned long buffers_offset;
+	unsigned long agp_textures_offset;
+} drm_mach64_init_t;
+
+typedef struct drm_mach64_clear {
+	unsigned int flags;
+	int x, y, w, h;
+	unsigned int clear_color;
+	unsigned int clear_depth;
+} drm_mach64_clear_t;
+
+typedef struct drm_mach64_vertex {
+	int prim;
+	void *buf;			/* Address of vertex buffer */
+	unsigned long used;		/* Number of bytes in buffer */
+	int discard;			/* Client finished with buffer? */
+} drm_mach64_vertex_t;
+
+typedef struct drm_mach64_blit {
+	int idx;
+	int pitch;
+	int offset;
+	int format;
+	unsigned short x, y;
+	unsigned short width, height;
+} drm_mach64_blit_t;
+
+typedef struct drm_mach64_getparam {
+	enum {
+		MACH64_PARAM_FRAMES_QUEUED = 0x01,
+		MACH64_PARAM_IRQ_NR = 0x02
+	} param;
+	void *value;
+} drm_mach64_getparam_t;
+
+#endif
diff --git a/shared-core/mach64_drv.h b/shared-core/mach64_drv.h
new file mode 100644
index 00000000..4331cde5
--- /dev/null
+++ b/shared-core/mach64_drv.h
@@ -0,0 +1,1034 @@
+/* mach64_drv.h -- Private header for mach64 driver -*- linux-c -*-
+ * Created: Fri Nov 24 22:07:58 2000 by gareth@valinux.com
+ *
+ * Copyright 2000 Gareth Hughes
+ * Copyright 2002 Frank C. Earl
+ * Copyright 2002-2003 Leif Delgass
+ * 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
+ * THE COPYRIGHT OWNER(S) 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:
+ *    Gareth Hughes <gareth@valinux.com>
+ *    Frank C. Earl <fearl@airmail.net>
+ *    Leif Delgass <ldelgass@retinalburn.net>
+ *    José Fonseca <j_r_fonseca@yahoo.co.uk>
+ */
+
+#ifndef __MACH64_DRV_H__
+#define __MACH64_DRV_H__
+
+
+/* FIXME: remove these when not needed */
+/* Development driver options */
+#define MACH64_EXTRA_CHECKING     0 /* Extra sanity checks for DMA/freelist management */
+#define MACH64_VERBOSE		  0 /* Verbose debugging output */
+
+typedef struct drm_mach64_freelist {
+	struct list_head  list;  /* List pointers for free_list, placeholders, or pending list */
+   	drm_buf_t *buf;          /* Pointer to the buffer */
+	int discard;             /* This flag is set when we're done (re)using a buffer */
+	u32 ring_ofs;            /* dword offset in ring of last descriptor for this buffer */
+} drm_mach64_freelist_t;
+
+typedef struct drm_mach64_descriptor_ring {
+	dma_addr_t handle;       /* handle (bus address) of ring returned by pci_alloc_consistent() */
+	void *start;             /* write pointer (cpu address) to start of descriptor ring */
+	u32 start_addr;          /* bus address of beginning of descriptor ring */
+       	int size;                /* size of ring in bytes */
+
+	u32 head_addr;           /* bus address of descriptor ring head */
+	u32 head;                /* dword offset of descriptor ring head */
+	u32 tail;                /* dword offset of descriptor ring tail */
+	u32 tail_mask;           /* mask used to wrap ring */
+	int space;               /* number of free bytes in ring */
+} drm_mach64_descriptor_ring_t;
+
+typedef struct drm_mach64_private {
+	drm_mach64_sarea_t *sarea_priv;
+
+	int is_pci;
+	drm_mach64_dma_mode_t driver_mode;       /* Async DMA, sync DMA, or MMIO */
+
+	int usec_timeout;                        /* Timeout for the wait functions */
+
+	drm_mach64_descriptor_ring_t ring;       /* DMA descriptor table (ring buffer) */
+	int ring_running;                        /* Is bus mastering is enabled */
+
+	struct list_head free_list;              /* Free-list head */
+	struct list_head placeholders;           /* Placeholder list for buffers held by clients */
+	struct list_head pending;                /* Buffers pending completion */
+
+	u32 frame_ofs[MACH64_MAX_QUEUED_FRAMES]; /* dword ring offsets of most recent frame swaps */
+
+	unsigned int fb_bpp;
+	unsigned int front_offset, front_pitch;
+	unsigned int back_offset, back_pitch;
+
+	unsigned int depth_bpp;
+	unsigned int depth_offset, depth_pitch;
+
+	u32 front_offset_pitch;
+	u32 back_offset_pitch;
+	u32 depth_offset_pitch;
+
+	drm_local_map_t *sarea;
+	drm_local_map_t *fb;
+	drm_local_map_t *mmio;
+	drm_local_map_t *ring_map;
+	drm_local_map_t *buffers;
+	drm_local_map_t *agp_textures;
+} drm_mach64_private_t;
+
+				/* mach64_dma.c */
+extern int mach64_dma_init( DRM_IOCTL_ARGS );
+extern int mach64_dma_idle( DRM_IOCTL_ARGS );
+extern int mach64_dma_flush( DRM_IOCTL_ARGS );
+extern int mach64_engine_reset( DRM_IOCTL_ARGS );
+extern int mach64_dma_buffers( DRM_IOCTL_ARGS );
+
+extern int mach64_init_freelist( drm_device_t *dev );
+extern void mach64_destroy_freelist( drm_device_t *dev );
+extern drm_buf_t *mach64_freelist_get( drm_mach64_private_t *dev_priv );
+
+extern int mach64_do_wait_for_fifo( drm_mach64_private_t *dev_priv,
+				    int entries );
+extern int mach64_do_wait_for_idle( drm_mach64_private_t *dev_priv );
+extern int mach64_wait_ring( drm_mach64_private_t *dev_priv, int n );
+extern int mach64_do_dispatch_pseudo_dma( drm_mach64_private_t *dev_priv );
+extern int mach64_do_release_used_buffers( drm_mach64_private_t *dev_priv );
+extern void mach64_dump_engine_info( drm_mach64_private_t *dev_priv );
+extern void mach64_dump_ring_info( drm_mach64_private_t *dev_priv );
+extern int mach64_do_engine_reset( drm_mach64_private_t *dev_priv );
+
+extern int mach64_do_dma_idle( drm_mach64_private_t *dev_priv );
+extern int mach64_do_dma_flush( drm_mach64_private_t *dev_priv );
+extern int mach64_do_cleanup_dma( drm_device_t *dev );
+
+				/* mach64_state.c */
+extern int mach64_dma_clear( DRM_IOCTL_ARGS );
+extern int mach64_dma_swap( DRM_IOCTL_ARGS );
+extern int mach64_dma_vertex( DRM_IOCTL_ARGS );
+extern int mach64_dma_blit( DRM_IOCTL_ARGS );
+extern int mach64_get_param( DRM_IOCTL_ARGS );
+
+/* ================================================================
+ * Registers
+ */
+
+#define MACH64_AGP_BASE				0x0148
+#define MACH64_AGP_CNTL				0x014c
+#define MACH64_ALPHA_TST_CNTL			0x0550
+
+
+#define MACH64_DSP_CONFIG 			0x0420
+#define MACH64_DSP_ON_OFF 			0x0424
+#define MACH64_EXT_MEM_CNTL 			0x04ac
+#define MACH64_GEN_TEST_CNTL 			0x04d0
+#define MACH64_HW_DEBUG 			0x047c
+#define MACH64_MEM_ADDR_CONFIG 			0x0434
+#define MACH64_MEM_BUF_CNTL 			0x042c
+#define MACH64_MEM_CNTL 			0x04b0
+
+
+#define MACH64_BM_ADDR				0x0648
+#define MACH64_BM_COMMAND			0x0188
+#define MACH64_BM_DATA				0x0648
+#define MACH64_BM_FRAME_BUF_OFFSET		0x0180
+#define MACH64_BM_GUI_TABLE			0x01b8
+#define MACH64_BM_GUI_TABLE_CMD			0x064c
+#	define MACH64_CIRCULAR_BUF_SIZE_16KB		(0 << 0)
+#	define MACH64_CIRCULAR_BUF_SIZE_32KB		(1 << 0)
+#	define MACH64_CIRCULAR_BUF_SIZE_64KB		(2 << 0)
+#	define MACH64_CIRCULAR_BUF_SIZE_128KB		(3 << 0)
+#	define MACH64_LAST_DESCRIPTOR			(1 << 31)
+#define MACH64_BM_HOSTDATA			0x0644
+#define MACH64_BM_STATUS			0x018c
+#define MACH64_BM_SYSTEM_MEM_ADDR		0x0184
+#define MACH64_BM_SYSTEM_TABLE			0x01bc
+#define MACH64_BUS_CNTL				0x04a0
+#	define MACH64_BUS_MSTR_RESET			(1 << 1)
+#	define MACH64_BUS_APER_REG_DIS			(1 << 4)
+#	define MACH64_BUS_FLUSH_BUF			(1 << 2)
+#	define MACH64_BUS_MASTER_DIS			(1 << 6)
+#	define MACH64_BUS_EXT_REG_EN			(1 << 27)
+
+#define MACH64_CLR_CMP_CLR			0x0700
+#define MACH64_CLR_CMP_CNTL			0x0708
+#define MACH64_CLR_CMP_MASK			0x0704
+#define MACH64_CONFIG_CHIP_ID 			0x04e0
+#define MACH64_CONFIG_CNTL 			0x04dc
+#define MACH64_CONFIG_STAT0 			0x04e4
+#define MACH64_CONFIG_STAT1 			0x0494
+#define MACH64_CONFIG_STAT2 			0x0498
+#define MACH64_CONTEXT_LOAD_CNTL		0x072c
+#define MACH64_CONTEXT_MASK			0x0720
+#define MACH64_COMPOSITE_SHADOW_ID		0x0798
+#define MACH64_CRC_SIG 				0x04e8
+#define MACH64_CUSTOM_MACRO_CNTL 		0x04d4
+
+#define MACH64_DP_BKGD_CLR			0x06c0
+#define MACH64_DP_FOG_CLR			0x06c4
+#define MACH64_DP_FGRD_BKGD_CLR			0x06e0
+#define MACH64_DP_FRGD_CLR			0x06c4
+#define MACH64_DP_FGRD_CLR_MIX			0x06dc
+
+#define MACH64_DP_MIX				0x06d4
+#	define BKGD_MIX_NOT_D				(0 << 0)
+#	define BKGD_MIX_ZERO				(1 << 0)
+#	define BKGD_MIX_ONE				(2 << 0)
+#	define MACH64_BKGD_MIX_D			(3 << 0)
+#	define BKGD_MIX_NOT_S				(4 << 0)
+#	define BKGD_MIX_D_XOR_S				(5 << 0)
+#	define BKGD_MIX_NOT_D_XOR_S			(6 << 0)
+#	define MACH64_BKGD_MIX_S			(7 << 0)
+#	define BKGD_MIX_NOT_D_OR_NOT_S			(8 << 0)
+#	define BKGD_MIX_D_OR_NOT_S			(9 << 0)
+#	define BKGD_MIX_NOT_D_OR_S			(10 << 0)
+#	define BKGD_MIX_D_OR_S				(11 << 0)
+#	define BKGD_MIX_D_AND_S				(12 << 0)
+#	define BKGD_MIX_NOT_D_AND_S			(13 << 0)
+#	define BKGD_MIX_D_AND_NOT_S			(14 << 0)
+#	define BKGD_MIX_NOT_D_AND_NOT_S			(15 << 0)
+#	define BKGD_MIX_D_PLUS_S_DIV2			(23 << 0)
+#	define FRGD_MIX_NOT_D				(0 << 16)
+#	define FRGD_MIX_ZERO				(1 << 16)
+#	define FRGD_MIX_ONE				(2 << 16)
+#	define FRGD_MIX_D				(3 << 16)
+#	define FRGD_MIX_NOT_S				(4 << 16)
+#	define FRGD_MIX_D_XOR_S				(5 << 16)
+#	define FRGD_MIX_NOT_D_XOR_S			(6 << 16)
+#	define MACH64_FRGD_MIX_S			(7 << 16)
+#	define FRGD_MIX_NOT_D_OR_NOT_S			(8 << 16)
+#	define FRGD_MIX_D_OR_NOT_S			(9 << 16)
+#	define FRGD_MIX_NOT_D_OR_S			(10 << 16)
+#	define FRGD_MIX_D_OR_S				(11 << 16)
+#	define FRGD_MIX_D_AND_S				(12 << 16)
+#	define FRGD_MIX_NOT_D_AND_S			(13 << 16)
+#	define FRGD_MIX_D_AND_NOT_S			(14 << 16)
+#	define FRGD_MIX_NOT_D_AND_NOT_S			(15 << 16)
+#	define FRGD_MIX_D_PLUS_S_DIV2			(23 << 16)
+
+#define MACH64_DP_PIX_WIDTH			0x06d0
+#	define MACH64_HOST_TRIPLE_ENABLE		(1 << 13)
+#	define MACH64_BYTE_ORDER_MSB_TO_LSB		(0 << 24)
+#	define MACH64_BYTE_ORDER_LSB_TO_MSB		(1 << 24)
+
+#define MACH64_DP_SRC				0x06d8
+#	define MACH64_BKGD_SRC_BKGD_CLR			(0 << 0)
+#	define MACH64_BKGD_SRC_FRGD_CLR			(1 << 0)
+#	define MACH64_BKGD_SRC_HOST			(2 << 0)
+#	define MACH64_BKGD_SRC_BLIT			(3 << 0)
+#	define MACH64_BKGD_SRC_PATTERN			(4 << 0)
+#	define MACH64_BKGD_SRC_3D			(5 << 0)
+#	define MACH64_FRGD_SRC_BKGD_CLR			(0 << 8)
+#	define MACH64_FRGD_SRC_FRGD_CLR			(1 << 8)
+#	define MACH64_FRGD_SRC_HOST			(2 << 8)
+#	define MACH64_FRGD_SRC_BLIT			(3 << 8)
+#	define MACH64_FRGD_SRC_PATTERN			(4 << 8)
+#	define MACH64_FRGD_SRC_3D			(5 << 8)
+#	define MACH64_MONO_SRC_ONE			(0 << 16)
+#	define MACH64_MONO_SRC_PATTERN			(1 << 16)
+#	define MACH64_MONO_SRC_HOST			(2 << 16)
+#	define MACH64_MONO_SRC_BLIT			(3 << 16)
+
+#define MACH64_DP_WRITE_MASK			0x06c8
+
+#define MACH64_DST_CNTL				0x0530
+#	define MACH64_DST_X_RIGHT_TO_LEFT		(0 << 0)
+#	define MACH64_DST_X_LEFT_TO_RIGHT		(1 << 0)
+#	define MACH64_DST_Y_BOTTOM_TO_TOP		(0 << 1)
+#	define MACH64_DST_Y_TOP_TO_BOTTOM		(1 << 1)
+#	define MACH64_DST_X_MAJOR			(0 << 2)
+#	define MACH64_DST_Y_MAJOR			(1 << 2)
+#	define MACH64_DST_X_TILE			(1 << 3)
+#	define MACH64_DST_Y_TILE			(1 << 4)
+#	define MACH64_DST_LAST_PEL			(1 << 5)
+#	define MACH64_DST_POLYGON_ENABLE		(1 << 6)
+#	define MACH64_DST_24_ROTATION_ENABLE		(1 << 7)
+
+#define MACH64_DST_HEIGHT_WIDTH			0x0518
+#define MACH64_DST_OFF_PITCH			0x0500
+#define MACH64_DST_WIDTH_HEIGHT			0x06ec
+#define MACH64_DST_X_Y				0x06e8
+#define MACH64_DST_Y_X				0x050c
+
+#define MACH64_FIFO_STAT			0x0710
+#	define MACH64_FIFO_SLOT_MASK			0x0000ffff
+#	define MACH64_FIFO_ERR				(1 << 31)
+
+#define MACH64_GEN_TEST_CNTL			0x04d0
+#	define MACH64_GUI_ENGINE_ENABLE			(1 << 8)
+#define MACH64_GUI_CMDFIFO_DEBUG		0x0170
+#define MACH64_GUI_CMDFIFO_DATA			0x0174
+#define MACH64_GUI_CNTL				0x0178
+#       define MACH64_CMDFIFO_SIZE_MASK                 0x00000003ul
+#       define MACH64_CMDFIFO_SIZE_192                  0x00000000ul
+#       define MACH64_CMDFIFO_SIZE_128                  0x00000001ul
+#       define MACH64_CMDFIFO_SIZE_64                   0x00000002ul
+#define MACH64_GUI_STAT				0x0738
+#	define MACH64_GUI_ACTIVE			(1 << 0)
+#define MACH64_GUI_TRAJ_CNTL			0x0730
+
+#define MACH64_HOST_CNTL			0x0640
+#define MACH64_HOST_DATA0			0x0600
+
+#define MACH64_ONE_OVER_AREA			0x029c
+#define MACH64_ONE_OVER_AREA_UC			0x0300
+
+#define MACH64_PAT_REG0				0x0680
+#define MACH64_PAT_REG1				0x0684
+
+#define MACH64_SC_LEFT                          0x06a0
+#define MACH64_SC_RIGHT                         0x06a4
+#define MACH64_SC_LEFT_RIGHT                    0x06a8
+#define MACH64_SC_TOP                           0x06ac
+#define MACH64_SC_BOTTOM                        0x06b0
+#define MACH64_SC_TOP_BOTTOM                    0x06b4
+
+#define MACH64_SCALE_3D_CNTL			0x05fc
+#define MACH64_SCRATCH_REG0			0x0480
+#define MACH64_SCRATCH_REG1			0x0484
+#define MACH64_SECONDARY_TEX_OFF		0x0778
+#define MACH64_SETUP_CNTL			0x0304
+#define MACH64_SRC_CNTL				0x05b4
+#	define MACH64_SRC_BM_ENABLE			(1 << 8)
+#	define MACH64_SRC_BM_SYNC			(1 << 9)
+#	define MACH64_SRC_BM_OP_FRAME_TO_SYSTEM		(0 << 10)
+#	define MACH64_SRC_BM_OP_SYSTEM_TO_FRAME		(1 << 10)
+#	define MACH64_SRC_BM_OP_REG_TO_SYSTEM		(2 << 10)
+#	define MACH64_SRC_BM_OP_SYSTEM_TO_REG		(3 << 10)
+#define MACH64_SRC_HEIGHT1			0x0594
+#define MACH64_SRC_HEIGHT2			0x05ac
+#define MACH64_SRC_HEIGHT1_WIDTH1		0x0598
+#define MACH64_SRC_HEIGHT2_WIDTH2		0x05b0
+#define MACH64_SRC_OFF_PITCH			0x0580
+#define MACH64_SRC_WIDTH1			0x0590
+#define MACH64_SRC_Y_X				0x058c
+
+#define MACH64_TEX_0_OFF			0x05c0
+#define MACH64_TEX_CNTL				0x0774
+#define MACH64_TEX_SIZE_PITCH			0x0770
+#define MACH64_TIMER_CONFIG 			0x0428
+
+#define MACH64_VERTEX_1_ARGB			0x0254
+#define MACH64_VERTEX_1_S			0x0240
+#define MACH64_VERTEX_1_SECONDARY_S		0x0328
+#define MACH64_VERTEX_1_SECONDARY_T		0x032c
+#define MACH64_VERTEX_1_SECONDARY_W		0x0330
+#define MACH64_VERTEX_1_SPEC_ARGB		0x024c
+#define MACH64_VERTEX_1_T			0x0244
+#define MACH64_VERTEX_1_W			0x0248
+#define MACH64_VERTEX_1_X_Y			0x0258
+#define MACH64_VERTEX_1_Z			0x0250
+#define MACH64_VERTEX_2_ARGB			0x0274
+#define MACH64_VERTEX_2_S			0x0260
+#define MACH64_VERTEX_2_SECONDARY_S		0x0334
+#define MACH64_VERTEX_2_SECONDARY_T		0x0338
+#define MACH64_VERTEX_2_SECONDARY_W		0x033c
+#define MACH64_VERTEX_2_SPEC_ARGB		0x026c
+#define MACH64_VERTEX_2_T			0x0264
+#define MACH64_VERTEX_2_W			0x0268
+#define MACH64_VERTEX_2_X_Y			0x0278
+#define MACH64_VERTEX_2_Z			0x0270
+#define MACH64_VERTEX_3_ARGB			0x0294
+#define MACH64_VERTEX_3_S			0x0280
+#define MACH64_VERTEX_3_SECONDARY_S		0x02a0
+#define MACH64_VERTEX_3_SECONDARY_T		0x02a4
+#define MACH64_VERTEX_3_SECONDARY_W		0x02a8
+#define MACH64_VERTEX_3_SPEC_ARGB		0x028c
+#define MACH64_VERTEX_3_T			0x0284
+#define MACH64_VERTEX_3_W			0x0288
+#define MACH64_VERTEX_3_X_Y			0x0298
+#define MACH64_VERTEX_3_Z			0x0290
+
+#define MACH64_Z_CNTL				0x054c
+#define MACH64_Z_OFF_PITCH			0x0548
+
+#define MACH64_CRTC_VLINE_CRNT_VLINE		0x0410
+#	define MACH64_CRTC_VLINE_MASK		        0x000007ff
+#	define MACH64_CRTC_CRNT_VLINE_MASK		0x07ff0000
+#define MACH64_CRTC_OFF_PITCH			0x0414
+#define MACH64_CRTC_INT_CNTL			0x0418
+#	define MACH64_CRTC_VBLANK			(1 << 0)
+#	define MACH64_CRTC_VBLANK_INT_EN		(1 << 1)
+#	define MACH64_CRTC_VBLANK_INT			(1 << 2)
+#	define MACH64_CRTC_VLINE_INT_EN			(1 << 3)
+#	define MACH64_CRTC_VLINE_INT			(1 << 4)
+#	define MACH64_CRTC_VLINE_SYNC			(1 << 5) /* 0=even, 1=odd */
+#	define MACH64_CRTC_FRAME			(1 << 6) /* 0=even, 1=odd */
+#	define MACH64_CRTC_SNAPSHOT_INT_EN		(1 << 7)
+#	define MACH64_CRTC_SNAPSHOT_INT			(1 << 8)
+#	define MACH64_CRTC_I2C_INT_EN			(1 << 9)
+#	define MACH64_CRTC_I2C_INT			(1 << 10)
+#	define MACH64_CRTC2_VBLANK			(1 << 11) /* LT Pro */
+#	define MACH64_CRTC2_VBLANK_INT_EN		(1 << 12) /* LT Pro */
+#	define MACH64_CRTC2_VBLANK_INT			(1 << 13) /* LT Pro */
+#	define MACH64_CRTC2_VLINE_INT_EN		(1 << 14) /* LT Pro */
+#	define MACH64_CRTC2_VLINE_INT			(1 << 15) /* LT Pro */
+#	define MACH64_CRTC_CAPBUF0_INT_EN		(1 << 16)
+#	define MACH64_CRTC_CAPBUF0_INT			(1 << 17)
+#	define MACH64_CRTC_CAPBUF1_INT_EN		(1 << 18)
+#	define MACH64_CRTC_CAPBUF1_INT			(1 << 19)
+#	define MACH64_CRTC_OVERLAY_EOF_INT_EN		(1 << 20)
+#	define MACH64_CRTC_OVERLAY_EOF_INT		(1 << 21)
+#	define MACH64_CRTC_ONESHOT_CAP_INT_EN		(1 << 22)
+#	define MACH64_CRTC_ONESHOT_CAP_INT		(1 << 23)
+#	define MACH64_CRTC_BUSMASTER_EOL_INT_EN		(1 << 24)
+#	define MACH64_CRTC_BUSMASTER_EOL_INT		(1 << 25)
+#	define MACH64_CRTC_GP_INT_EN			(1 << 26)
+#	define MACH64_CRTC_GP_INT			(1 << 27)
+#	define MACH64_CRTC2_VLINE_SYNC			(1 << 28) /* LT Pro */  /* 0=even, 1=odd */
+#	define MACH64_CRTC_SNAPSHOT2_INT_EN		(1 << 29) /* LT Pro */
+#	define MACH64_CRTC_SNAPSHOT2_INT		(1 << 30) /* LT Pro */
+#	define MACH64_CRTC_VBLANK2_INT			(1 << 31)
+#	define MACH64_CRTC_INT_ENS				\
+		(						\
+			MACH64_CRTC_VBLANK_INT_EN |		\
+			MACH64_CRTC_VLINE_INT_EN |		\
+			MACH64_CRTC_SNAPSHOT_INT_EN |		\
+			MACH64_CRTC_I2C_INT_EN |		\
+			MACH64_CRTC2_VBLANK_INT_EN |		\
+			MACH64_CRTC2_VLINE_INT_EN |		\
+			MACH64_CRTC_CAPBUF0_INT_EN |		\
+			MACH64_CRTC_CAPBUF1_INT_EN |		\
+			MACH64_CRTC_OVERLAY_EOF_INT_EN |	\
+			MACH64_CRTC_ONESHOT_CAP_INT_EN |	\
+			MACH64_CRTC_BUSMASTER_EOL_INT_EN |	\
+			MACH64_CRTC_GP_INT_EN |			\
+			MACH64_CRTC_SNAPSHOT2_INT_EN |		\
+			0					\
+		)
+#	define MACH64_CRTC_INT_ACKS			\
+		(					\
+			MACH64_CRTC_VBLANK_INT |	\
+			MACH64_CRTC_VLINE_INT |		\
+			MACH64_CRTC_SNAPSHOT_INT |	\
+			MACH64_CRTC_I2C_INT |		\
+			MACH64_CRTC2_VBLANK_INT |	\
+			MACH64_CRTC2_VLINE_INT |	\
+			MACH64_CRTC_CAPBUF0_INT |	\
+			MACH64_CRTC_CAPBUF1_INT |	\
+			MACH64_CRTC_OVERLAY_EOF_INT |	\
+			MACH64_CRTC_ONESHOT_CAP_INT |	\
+			MACH64_CRTC_BUSMASTER_EOL_INT |	\
+			MACH64_CRTC_GP_INT |		\
+			MACH64_CRTC_SNAPSHOT2_INT |	\
+			MACH64_CRTC_VBLANK2_INT |	\
+			0				\
+		)
+
+#define MACH64_DATATYPE_CI8				2
+#define MACH64_DATATYPE_ARGB1555			3
+#define MACH64_DATATYPE_RGB565				4
+#define MACH64_DATATYPE_ARGB8888			6
+#define MACH64_DATATYPE_RGB332				7
+#define MACH64_DATATYPE_Y8				8
+#define MACH64_DATATYPE_RGB8				9
+#define MACH64_DATATYPE_VYUY422				11
+#define MACH64_DATATYPE_YVYU422				12
+#define MACH64_DATATYPE_AYUV444				14
+#define MACH64_DATATYPE_ARGB4444			15
+
+#define MACH64_READ(reg)	DRM_READ32(dev_priv->mmio, (reg) )
+#define MACH64_WRITE(reg,val)	DRM_WRITE32(dev_priv->mmio, (reg), (val) )
+
+
+#define DWMREG0		0x0400
+#define DWMREG0_END	0x07ff
+#define DWMREG1		0x0000
+#define DWMREG1_END	0x03ff
+
+#define ISREG0(r)	(((r) >= DWMREG0) && ((r) <= DWMREG0_END))
+#define DMAREG0(r)	(((r) - DWMREG0) >> 2)
+#define DMAREG1(r)	((((r) - DWMREG1) >> 2 ) | 0x0100)
+#define DMAREG(r)	(ISREG0(r) ? DMAREG0(r) : DMAREG1(r))
+
+#define MMREG0		0x0000
+#define MMREG0_END	0x00ff
+
+#define ISMMREG0(r)	(((r) >= MMREG0) && ((r) <= MMREG0_END))
+#define MMSELECT0(r)	(((r) << 2) + DWMREG0)
+#define MMSELECT1(r)	(((((r) & 0xff) << 2) + DWMREG1))
+#define MMSELECT(r)	(ISMMREG0(r) ? MMSELECT0(r) : MMSELECT1(r))
+
+/* ================================================================
+ * DMA constants
+ */
+
+/* DMA descriptor field indices:
+ * The descriptor fields are loaded into the read-only 
+ * BM_* system bus master registers during a bus-master operation
+ */
+#define MACH64_DMA_FRAME_BUF_OFFSET	0        /* BM_FRAME_BUF_OFFSET */
+#define MACH64_DMA_SYS_MEM_ADDR		1        /* BM_SYSTEM_MEM_ADDR */
+#define MACH64_DMA_COMMAND		2        /* BM_COMMAND */
+#define MACH64_DMA_RESERVED		3        /* BM_STATUS */
+
+/* BM_COMMAND descriptor field flags */
+#define MACH64_DMA_HOLD_OFFSET		(1<<30)  /* Don't increment DMA_FRAME_BUF_OFFSET */
+#define MACH64_DMA_EOL			(1<<31)  /* End of descriptor list flag */
+
+#define MACH64_DMA_CHUNKSIZE	        0x1000   /* 4kB per DMA descriptor */
+#define MACH64_APERTURE_OFFSET	        0x7ff800 /* frame-buffer offset for gui-masters */
+
+
+/* ================================================================
+ * Misc helper macros
+ */
+
+static __inline__ void mach64_set_dma_eol( volatile u32 * addr )
+{
+#if defined(__i386__)
+	int nr = 31;
+	
+	/* Taken from include/asm-i386/bitops.h linux header */
+        __asm__ __volatile__( "lock;"
+                "btsl %1,%0"
+                :"=m" (*addr)
+                :"Ir" (nr));
+#elif defined(__powerpc__)
+	u32 old;
+	u32 mask = cpu_to_le32( MACH64_DMA_EOL );
+
+	/* Taken from the include/asm-ppc/bitops.h linux header */
+	__asm__ __volatile__("\n\
+1:	lwarx	%0,0,%3 \n\
+	or	%0,%0,%2 \n\
+	stwcx.	%0,0,%3 \n\
+	bne-	1b"
+	: "=&r" (old), "=m" (*addr)
+	: "r" (mask), "r" (addr), "m" (*addr)
+	: "cc");
+#elif defined(__alpha__)
+	u32 temp;
+	u32 mask = MACH64_DMA_EOL;
+
+	/* Taken from the include/asm-alpha/bitops.h linux header */
+	__asm__ __volatile__(
+	"1:	ldl_l %0,%3\n"
+	"	bis %0,%2,%0\n"
+	"	stl_c %0,%1\n"
+	"	beq %0,2f\n"
+	".subsection 2\n"
+	"2:	br 1b\n"
+	".previous"
+	:"=&r" (temp), "=m" (*addr)
+	:"Ir" (mask), "m" (*addr));
+#else
+	u32 mask = cpu_to_le32( MACH64_DMA_EOL );
+
+	*addr |= mask;
+#endif
+}
+
+static __inline__ void mach64_clear_dma_eol( volatile u32 * addr )
+{
+#if defined(__i386__)
+	int nr = 31;
+	
+	/* Taken from include/asm-i386/bitops.h linux header */
+        __asm__ __volatile__( "lock;"
+                "btrl %1,%0"
+                :"=m" (*addr)
+                :"Ir" (nr));
+#elif defined(__powerpc__)
+	u32 old;
+	u32 mask = cpu_to_le32( MACH64_DMA_EOL );
+
+	/* Taken from the include/asm-ppc/bitops.h linux header */
+	__asm__ __volatile__("\n\
+1:	lwarx	%0,0,%3 \n\
+	andc	%0,%0,%2 \n\
+	stwcx.	%0,0,%3 \n\
+	bne-	1b"
+	: "=&r" (old), "=m" (*addr)
+	: "r" (mask), "r" (addr), "m" (*addr)
+	: "cc");
+#elif defined(__alpha__)
+	u32 temp;
+	u32 mask = ~MACH64_DMA_EOL;
+
+	/* Taken from the include/asm-alpha/bitops.h linux header */
+	__asm__ __volatile__(
+	"1:	ldl_l %0,%3\n"
+	"	and %0,%2,%0\n"
+	"	stl_c %0,%1\n"
+	"	beq %0,2f\n"
+	".subsection 2\n"
+	"2:	br 1b\n"
+	".previous"
+	:"=&r" (temp), "=m" (*addr)
+	:"Ir" (mask), "m" (*addr));
+#else
+	u32 mask = cpu_to_le32( ~MACH64_DMA_EOL );
+
+	*addr &= mask;
+#endif
+}
+
+static __inline__ void mach64_ring_start( drm_mach64_private_t *dev_priv )
+{
+	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;
+	
+	DRM_DEBUG( "%s: head_addr: 0x%08x head: %d tail: %d space: %d\n",
+		   __FUNCTION__, 
+		   ring->head_addr, ring->head, ring->tail, ring->space );
+
+	if ( mach64_do_wait_for_idle( dev_priv ) < 0 ) {
+		mach64_do_engine_reset( dev_priv );
+	}
+
+	if (dev_priv->driver_mode != MACH64_MODE_MMIO ) {
+		/* enable bus mastering and block 1 registers */
+		MACH64_WRITE( MACH64_BUS_CNTL, 
+			      ( MACH64_READ(MACH64_BUS_CNTL) & 	~MACH64_BUS_MASTER_DIS ) 
+			      | MACH64_BUS_EXT_REG_EN );
+		mach64_do_wait_for_idle( dev_priv );
+	}
+	
+	/* reset descriptor table ring head */
+	MACH64_WRITE( MACH64_BM_GUI_TABLE_CMD, 
+		      ring->head_addr | MACH64_CIRCULAR_BUF_SIZE_16KB );
+	
+	dev_priv->ring_running = 1;
+}
+
+static __inline__ void mach64_ring_resume( drm_mach64_private_t *dev_priv, 
+					   drm_mach64_descriptor_ring_t *ring )
+{
+	DRM_DEBUG( "%s: head_addr: 0x%08x head: %d tail: %d space: %d\n",
+		   __FUNCTION__, 
+		   ring->head_addr, ring->head, ring->tail, ring->space );
+
+	/* reset descriptor table ring head */
+	MACH64_WRITE( MACH64_BM_GUI_TABLE_CMD, 
+		      ring->head_addr | MACH64_CIRCULAR_BUF_SIZE_16KB );
+
+	if ( dev_priv->driver_mode == MACH64_MODE_MMIO ) {
+		mach64_do_dispatch_pseudo_dma( dev_priv );
+	} else {
+		/* enable GUI bus mastering, and sync the bus master to the GUI */
+		MACH64_WRITE( MACH64_SRC_CNTL, 
+			      MACH64_SRC_BM_ENABLE | MACH64_SRC_BM_SYNC |
+			      MACH64_SRC_BM_OP_SYSTEM_TO_REG );
+
+		/* kick off the transfer */
+		MACH64_WRITE( MACH64_DST_HEIGHT_WIDTH, 0 );
+		if ( dev_priv->driver_mode == MACH64_MODE_DMA_SYNC ) {
+			if ( (mach64_do_wait_for_idle( dev_priv )) < 0 ) {
+				DRM_ERROR( "%s: idle failed, resetting engine\n", 
+					   __FUNCTION__);
+				mach64_dump_engine_info( dev_priv );
+				mach64_do_engine_reset( dev_priv );
+				return;
+			}
+			mach64_do_release_used_buffers( dev_priv );
+		}
+	}
+}
+
+static __inline__ void mach64_ring_tick( drm_mach64_private_t *dev_priv, 
+					 drm_mach64_descriptor_ring_t *ring )
+{
+	DRM_DEBUG( "%s: head_addr: 0x%08x head: %d tail: %d space: %d\n",
+		   __FUNCTION__, 
+		   ring->head_addr, ring->head, ring->tail, ring->space );
+
+	if ( !dev_priv->ring_running ) {
+		mach64_ring_start( dev_priv );
+		
+		if ( ring->head != ring->tail ) {
+			mach64_ring_resume( dev_priv, ring );
+		}
+	} else {
+		/* GUI_ACTIVE must be read before BM_GUI_TABLE to 
+		 * correctly determine the ring head 
+		 */
+		int gui_active = MACH64_READ(MACH64_GUI_STAT) & MACH64_GUI_ACTIVE;
+		
+		ring->head_addr = MACH64_READ(MACH64_BM_GUI_TABLE) & 0xfffffff0;
+		
+		if ( gui_active ) {
+			/* If not idle, BM_GUI_TABLE points one descriptor 
+			 * past the current head 
+			 */
+			if ( ring->head_addr == ring->start_addr ) {
+				ring->head_addr += ring->size;
+			}
+			ring->head_addr -= 4 * sizeof(u32);
+		}
+
+		if( ring->head_addr < ring->start_addr || 
+		    ring->head_addr >= ring->start_addr + ring->size ) {
+			DRM_ERROR( "bad ring head address: 0x%08x\n", ring->head_addr );
+			mach64_dump_ring_info( dev_priv );
+			mach64_do_engine_reset( dev_priv );
+			return;
+		}
+	
+		ring->head = (ring->head_addr - ring->start_addr) / sizeof(u32);
+		
+		if ( !gui_active && ring->head != ring->tail ) {
+			mach64_ring_resume( dev_priv, ring );
+		}
+	}
+}
+
+static __inline__ void mach64_ring_stop( drm_mach64_private_t *dev_priv )
+{
+	DRM_DEBUG( "%s: head_addr: 0x%08x head: %d tail: %d space: %d\n",
+		   __FUNCTION__, 
+		   dev_priv->ring.head_addr, dev_priv->ring.head, 
+		   dev_priv->ring.tail, dev_priv->ring.space );
+
+	/* restore previous SRC_CNTL to disable busmastering */
+	mach64_do_wait_for_fifo( dev_priv, 1 );
+	MACH64_WRITE( MACH64_SRC_CNTL, 0 );
+
+	/* disable busmastering but keep the block 1 registers enabled */ 
+	mach64_do_wait_for_idle( dev_priv );
+	MACH64_WRITE( MACH64_BUS_CNTL, MACH64_READ( MACH64_BUS_CNTL ) 
+		      | MACH64_BUS_MASTER_DIS | MACH64_BUS_EXT_REG_EN );
+		
+	dev_priv->ring_running = 0;
+}
+
+static __inline__ void
+mach64_update_ring_snapshot( drm_mach64_private_t *dev_priv )
+{
+	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;
+
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+	
+	mach64_ring_tick( dev_priv, ring );
+
+	ring->space = (ring->head - ring->tail) * sizeof(u32);
+	if ( ring->space <= 0 ) {
+		ring->space += ring->size;
+	}
+}
+
+/* ================================================================
+ * DMA descriptor ring macros
+ */
+
+#define RING_LOCALS									\
+	int _ring_tail, _ring_write; unsigned int _ring_mask; volatile u32 *_ring
+
+#define RING_WRITE_OFS  _ring_write
+
+#define BEGIN_RING( n ) 								\
+do {											\
+	if ( MACH64_VERBOSE ) {								\
+		DRM_INFO( "BEGIN_RING( %d ) in %s\n",					\
+			   (n), __FUNCTION__ );						\
+	}										\
+	if ( dev_priv->ring.space <= (n) * sizeof(u32) ) {				\
+		int ret;								\
+		if ((ret=mach64_wait_ring( dev_priv, (n) * sizeof(u32))) < 0 ) {	\
+			DRM_ERROR( "wait_ring failed, resetting engine\n");		\
+			mach64_dump_engine_info( dev_priv );				\
+			mach64_do_engine_reset( dev_priv );				\
+			return ret;							\
+		}									\
+	}										\
+	dev_priv->ring.space -= (n) * sizeof(u32);					\
+	_ring = (u32 *) dev_priv->ring.start;						\
+	_ring_tail = _ring_write = dev_priv->ring.tail;					\
+	_ring_mask = dev_priv->ring.tail_mask;						\
+} while (0)
+
+#define OUT_RING( x )						\
+do {								\
+	if ( MACH64_VERBOSE ) {					\
+		DRM_INFO( "   OUT_RING( 0x%08x ) at 0x%x\n",	\
+			   (unsigned int)(x), _ring_write );	\
+	}							\
+	_ring[_ring_write++] = cpu_to_le32( x );		\
+	_ring_write &= _ring_mask;				\
+} while (0)
+
+#define ADVANCE_RING() 							\
+do {									\
+	if ( MACH64_VERBOSE ) {						\
+		DRM_INFO( "ADVANCE_RING() wr=0x%06x tail=0x%06x\n",	\
+			  _ring_write, _ring_tail );			\
+	}								\
+	DRM_MEMORYBARRIER();						\
+	mach64_clear_dma_eol( &_ring[(_ring_tail - 2) & _ring_mask] );	\
+	DRM_MEMORYBARRIER();						\
+	dev_priv->ring.tail = _ring_write;				\
+	mach64_ring_tick( dev_priv, &(dev_priv)->ring );		\
+} while (0)
+
+
+/* ================================================================
+ * DMA macros
+ */
+
+#define DMALOCALS				\
+	drm_mach64_freelist_t *_entry = NULL;	\
+	drm_buf_t *_buf = NULL; 		\
+	u32 *_buf_wptr; int _outcount
+
+#define GETBUFPTR( __buf )						\
+((dev_priv->is_pci) ? 							\
+	((u32 *)(__buf)->address) : 					\
+	((u32 *)((char *)dev_priv->buffers->handle + (__buf)->offset)))
+
+#define GETBUFADDR( __buf ) ((u32)(__buf)->bus_address)
+
+#define GETRINGOFFSET() (_entry->ring_ofs)
+
+static __inline__ int mach64_find_pending_buf_entry ( drm_mach64_private_t *dev_priv, 
+						      drm_mach64_freelist_t **entry, 
+						      drm_buf_t *buf )
+{
+	struct list_head *ptr;
+#if MACH64_EXTRA_CHECKING
+	if (list_empty(&dev_priv->pending)) {
+		DRM_ERROR("Empty pending list in %s\n", __FUNCTION__);
+		return DRM_ERR(EINVAL);
+	}
+#endif
+	ptr = dev_priv->pending.prev;
+	*entry = list_entry(ptr, drm_mach64_freelist_t, list);
+	while ((*entry)->buf != buf) {
+		if (ptr == &dev_priv->pending) {
+			return DRM_ERR(EFAULT);
+		}
+		ptr = ptr->prev;
+		*entry = list_entry(ptr, drm_mach64_freelist_t, list);
+	}
+	return 0;
+}
+
+#define DMASETPTR( _p ) 			\
+do {						\
+	_buf = (_p);				\
+	_outcount = 0;				\
+	_buf_wptr = GETBUFPTR( _buf );		\
+} while(0)
+
+/* FIXME: use a private set of smaller buffers for state emits, clears, and swaps? */
+#define DMAGETPTR( filp, dev_priv, n )					\
+do {									\
+	if ( MACH64_VERBOSE ) {						\
+		DRM_INFO( "DMAGETPTR( %d ) in %s\n",			\
+			  n, __FUNCTION__ );				\
+	}								\
+	_buf = mach64_freelist_get( dev_priv );				\
+	if (_buf == NULL) {						\
+		DRM_ERROR("%s: couldn't get buffer in DMAGETPTR\n",	\
+			   __FUNCTION__ );				\
+		return DRM_ERR(EAGAIN);					\
+	}								\
+	if (_buf->pending) {						\
+	        DRM_ERROR("%s: pending buf in DMAGETPTR\n",		\
+			   __FUNCTION__ );				\
+		return DRM_ERR(EFAULT);					\
+	}								\
+	_buf->filp = filp;						\
+	_outcount = 0;							\
+									\
+        _buf_wptr = GETBUFPTR( _buf );					\
+} while (0)
+
+#define DMAOUTREG( reg, val )					\
+do {								\
+	if ( MACH64_VERBOSE ) {					\
+		DRM_INFO( "   DMAOUTREG( 0x%x = 0x%08x )\n",	\
+			  reg, val );				\
+	}							\
+	_buf_wptr[_outcount++] = cpu_to_le32(DMAREG(reg));	\
+	_buf_wptr[_outcount++] = cpu_to_le32((val));		\
+	_buf->used += 8;					\
+} while (0)
+
+#define DMAADVANCE( dev_priv, _discard )						     \
+do {											     \
+	struct list_head *ptr;								     \
+	RING_LOCALS;									     \
+											     \
+	if ( MACH64_VERBOSE ) {								     \
+		DRM_INFO( "DMAADVANCE() in %s\n", __FUNCTION__ );			     \
+	}										     \
+											     \
+	if (_buf->used <= 0) {								     \
+		DRM_ERROR( "DMAADVANCE() in %s: sending empty buf %d\n",		     \
+				   __FUNCTION__, _buf->idx );				     \
+		return DRM_ERR(EFAULT);							     \
+	}										     \
+	if (_buf->pending) {								     \
+                /* This is a resued buffer, so we need to find it in the pending list */     \
+		int ret;								     \
+		if ( (ret=mach64_find_pending_buf_entry(dev_priv, &_entry, _buf)) ) {	     \
+			DRM_ERROR( "DMAADVANCE() in %s: couldn't find pending buf %d\n",     \
+				   __FUNCTION__, _buf->idx );				     \
+			return ret;							     \
+		}									     \
+		if (_entry->discard) {							     \
+			DRM_ERROR( "DMAADVANCE() in %s: sending discarded pending buf %d\n", \
+				   __FUNCTION__, _buf->idx );				     \
+			return DRM_ERR(EFAULT);						     \
+		}									     \
+     	} else {									     \
+		if (list_empty(&dev_priv->placeholders)) {				     \
+			DRM_ERROR( "DMAADVANCE() in %s: empty placeholder list\n",	     \
+			   	__FUNCTION__ );						     \
+			return DRM_ERR(EFAULT);						     \
+		}									     \
+		ptr = dev_priv->placeholders.next;					     \
+		list_del(ptr);								     \
+		_entry = list_entry(ptr, drm_mach64_freelist_t, list);			     \
+		_buf->pending = 1;							     \
+		_entry->buf = _buf;							     \
+		list_add_tail(ptr, &dev_priv->pending);					     \
+	}										     \
+	_entry->discard = (_discard);							     \
+	ADD_BUF_TO_RING( dev_priv );							     \
+} while (0)
+
+#define DMADISCARDBUF()									\
+do {											\
+	if (_entry == NULL) {								\
+		int ret;								\
+		if ( (ret=mach64_find_pending_buf_entry(dev_priv, &_entry, _buf)) ) {	\
+			DRM_ERROR( "%s: couldn't find pending buf %d\n",		\
+				   __FUNCTION__, _buf->idx );				\
+			return ret;							\
+		}									\
+	}										\
+	_entry->discard = 1;								\
+} while(0)
+
+#define ADD_BUF_TO_RING( dev_priv )							\
+do {											\
+	int bytes, pages, remainder;							\
+	u32 address, page;								\
+	int i;										\
+											\
+	bytes = _buf->used;								\
+	address = GETBUFADDR( _buf );							\
+											\
+	pages = (bytes + MACH64_DMA_CHUNKSIZE - 1) / MACH64_DMA_CHUNKSIZE;		\
+											\
+	BEGIN_RING( pages * 4 );							\
+											\
+	for ( i = 0 ; i < pages-1 ; i++ ) {						\
+		page = address + i * MACH64_DMA_CHUNKSIZE;				\
+		OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_ADDR );			\
+		OUT_RING( page );							\
+		OUT_RING( MACH64_DMA_CHUNKSIZE | MACH64_DMA_HOLD_OFFSET );		\
+		OUT_RING( 0 );								\
+	}										\
+											\
+	/* generate the final descriptor for any remaining commands in this buffer */	\
+	page = address + i * MACH64_DMA_CHUNKSIZE;					\
+	remainder = bytes - i * MACH64_DMA_CHUNKSIZE;					\
+											\
+	/* Save dword offset of last descriptor for this buffer.			\
+	 * This is needed to check for completion of the buffer in freelist_get		\
+	 */										\
+	_entry->ring_ofs = RING_WRITE_OFS;						\
+											\
+	OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_ADDR );				\
+	OUT_RING( page );								\
+	OUT_RING( remainder | MACH64_DMA_HOLD_OFFSET | MACH64_DMA_EOL );		\
+	OUT_RING( 0 );									\
+											\
+	ADVANCE_RING();									\
+} while(0)
+
+#define DMAADVANCEHOSTDATA( dev_priv )							\
+do {											\
+	struct list_head *ptr;								\
+	RING_LOCALS;									\
+											\
+	if ( MACH64_VERBOSE ) {								\
+		DRM_INFO( "DMAADVANCEHOSTDATA() in %s\n", __FUNCTION__ );		\
+	}										\
+											\
+	if (_buf->used <= 0) {								\
+		DRM_ERROR( "DMAADVANCEHOSTDATA() in %s: sending empty buf %d\n",	\
+				   __FUNCTION__, _buf->idx );				\
+		return DRM_ERR(EFAULT);							\
+	}										\
+	if (list_empty(&dev_priv->placeholders)) {					\
+		DRM_ERROR( "%s: empty placeholder list in DMAADVANCEHOSTDATA()\n",	\
+			   __FUNCTION__ );						\
+		return DRM_ERR(EFAULT);							\
+	}										\
+											\
+        ptr = dev_priv->placeholders.next;						\
+	list_del(ptr);									\
+	_entry = list_entry(ptr, drm_mach64_freelist_t, list);				\
+	_entry->buf = _buf;								\
+	_entry->buf->pending = 1;							\
+	list_add_tail(ptr, &dev_priv->pending);						\
+	_entry->discard = 1;								\
+	ADD_HOSTDATA_BUF_TO_RING( dev_priv );						\
+} while (0)
+
+#define ADD_HOSTDATA_BUF_TO_RING( dev_priv )						 \
+do {											 \
+	int bytes, pages, remainder;							 \
+	u32 address, page;								 \
+	int i;										 \
+											 \
+	bytes = _buf->used - MACH64_HOSTDATA_BLIT_OFFSET;				 \
+	pages = (bytes + MACH64_DMA_CHUNKSIZE - 1) / MACH64_DMA_CHUNKSIZE;		 \
+	address = GETBUFADDR( _buf );							 \
+											 \
+	BEGIN_RING( 4 + pages * 4 );							 \
+											 \
+	OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_ADDR );				 \
+	OUT_RING( address );								 \
+	OUT_RING( MACH64_HOSTDATA_BLIT_OFFSET | MACH64_DMA_HOLD_OFFSET );		 \
+	OUT_RING( 0 );									 \
+											 \
+	address += MACH64_HOSTDATA_BLIT_OFFSET;						 \
+											 \
+	for ( i = 0 ; i < pages-1 ; i++ ) {						 \
+		page = address + i * MACH64_DMA_CHUNKSIZE;				 \
+		OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_HOSTDATA );		 \
+		OUT_RING( page );							 \
+		OUT_RING( MACH64_DMA_CHUNKSIZE | MACH64_DMA_HOLD_OFFSET );		 \
+		OUT_RING( 0 );								 \
+	}										 \
+											 \
+	/* generate the final descriptor for any remaining commands in this buffer */	 \
+	page = address + i * MACH64_DMA_CHUNKSIZE;					 \
+	remainder = bytes - i * MACH64_DMA_CHUNKSIZE;					 \
+											 \
+	/* Save dword offset of last descriptor for this buffer.			 \
+	 * This is needed to check for completion of the buffer in freelist_get		 \
+	 */										 \
+	_entry->ring_ofs = RING_WRITE_OFS;						 \
+											 \
+	OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_HOSTDATA );			 \
+	OUT_RING( page );								 \
+	OUT_RING( remainder | MACH64_DMA_HOLD_OFFSET | MACH64_DMA_EOL );		 \
+	OUT_RING( 0 );									 \
+											 \
+	ADVANCE_RING();									 \
+} while(0)
+
+#endif /* __MACH64_DRV_H__ */
diff --git a/shared-core/mach64_irq.c b/shared-core/mach64_irq.c
new file mode 100644
index 00000000..9acd6269
--- /dev/null
+++ b/shared-core/mach64_irq.c
@@ -0,0 +1,130 @@
+/* mach64_irq.c -- IRQ handling for ATI Mach64 -*- linux-c -*-
+ * Created: Tue Feb 25, 2003 by Leif Delgass, based on radeon_irq.c/r128_irq.c
+ *
+ * Copyright (C) The Weather Channel, Inc.  2002.
+ * Copyright 2003 Leif Delgass
+ * All Rights Reserved.
+ *
+ * The Weather Channel (TM) funded Tungsten Graphics to develop the
+ * initial release of the Radeon 8500 driver under the XFree86 license.
+ * This notice must be preserved.
+ *
+ * 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
+ * THE COPYRIGHT OWNER(S) 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 <keith@tungstengraphics.com>
+ *    Eric Anholt <anholt@FreeBSD.org>
+ *    Leif Delgass <ldelgass@retinalburn.net>
+ */
+
+#include "mach64.h"
+#include "drmP.h"
+#include "drm.h"
+#include "mach64_drm.h"
+#include "mach64_drv.h"
+
+irqreturn_t DRM(irq_handler)( DRM_IRQ_ARGS )
+{
+	drm_device_t *dev = (drm_device_t *) arg;
+	drm_mach64_private_t *dev_priv = 
+	   (drm_mach64_private_t *)dev->dev_private;
+	int status;
+
+	status = MACH64_READ( MACH64_CRTC_INT_CNTL );
+
+	/* VBLANK interrupt */
+	if (status & MACH64_CRTC_VBLANK_INT) {
+	    /* Mask off all interrupt ack bits before setting the ack bit, since
+	     * there may be other handlers outside the DRM.  
+	     *
+	     * NOTE: On mach64, you need to keep the enable bits set when doing
+	     * the ack, despite what the docs say about not acking and enabling
+	     * in a single write.
+	     */
+	    MACH64_WRITE( MACH64_CRTC_INT_CNTL, (status & ~MACH64_CRTC_INT_ACKS)
+			  | MACH64_CRTC_VBLANK_INT );
+
+	    atomic_inc(&dev->vbl_received);
+	    DRM_WAKEUP(&dev->vbl_queue);
+	    DRM(vbl_send_signals)( dev );
+            return IRQ_HANDLED;
+	}
+	return IRQ_NONE;
+}
+
+int DRM(vblank_wait)(drm_device_t *dev, unsigned int *sequence)
+{
+	unsigned int cur_vblank;
+	int ret = 0;
+
+	/* Assume that the user has missed the current sequence number
+	 * by about a day rather than she wants to wait for years
+	 * using vertical blanks... 
+	 */
+	DRM_WAIT_ON( ret, dev->vbl_queue, 3*DRM_HZ, 
+		     ( ( ( cur_vblank = atomic_read(&dev->vbl_received ) )
+			 - *sequence ) <= (1<<23) ) );
+
+	*sequence = cur_vblank;
+
+	return ret;
+}
+
+/* drm_dma.h hooks
+*/
+void DRM(driver_irq_preinstall)( drm_device_t *dev ) {
+	drm_mach64_private_t *dev_priv =
+		(drm_mach64_private_t *)dev->dev_private;
+
+	u32 status = MACH64_READ( MACH64_CRTC_INT_CNTL );
+
+	DRM_DEBUG("before install CRTC_INT_CTNL: 0x%08x\n", status );
+
+ 	/* Disable and clear VBLANK interrupt */
+	MACH64_WRITE( MACH64_CRTC_INT_CNTL, (status & ~MACH64_CRTC_VBLANK_INT_EN) 
+		      | MACH64_CRTC_VBLANK_INT );
+}
+
+void DRM(driver_irq_postinstall)( drm_device_t *dev ) {
+	drm_mach64_private_t *dev_priv =
+		(drm_mach64_private_t *)dev->dev_private;
+
+	/* Turn on VBLANK interrupt */
+   	MACH64_WRITE( MACH64_CRTC_INT_CNTL, MACH64_READ( MACH64_CRTC_INT_CNTL )
+		      | MACH64_CRTC_VBLANK_INT_EN );
+
+	DRM_DEBUG("after install CRTC_INT_CTNL: 0x%08x\n", MACH64_READ( MACH64_CRTC_INT_CNTL ));
+
+}
+
+void DRM(driver_irq_uninstall)( drm_device_t *dev ) {
+	drm_mach64_private_t *dev_priv =
+		(drm_mach64_private_t *)dev->dev_private;
+	if ( !dev_priv )
+		return;
+
+	/* Disable and clear VBLANK interrupt */
+	MACH64_WRITE( MACH64_CRTC_INT_CNTL, 
+		      (MACH64_READ( MACH64_CRTC_INT_CNTL ) & ~MACH64_CRTC_VBLANK_INT_EN) 
+		      | MACH64_CRTC_VBLANK_INT );
+
+	DRM_DEBUG("after uninstall CRTC_INT_CTNL: 0x%08x\n", 
+		  MACH64_READ( MACH64_CRTC_INT_CNTL ));
+}
diff --git a/shared-core/mach64_state.c b/shared-core/mach64_state.c
new file mode 100644
index 00000000..eb3722cb
--- /dev/null
+++ b/shared-core/mach64_state.c
@@ -0,0 +1,897 @@
+/* mach64_state.c -- State support for mach64 (Rage Pro) driver -*- linux-c -*-
+ * Created: Sun Dec 03 19:20:26 2000 by gareth@valinux.com
+ *
+ * Copyright 2000 Gareth Hughes
+ * Copyright 2002-2003 Leif Delgass
+ * 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
+ * THE COPYRIGHT OWNER(S) 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:
+ *    Gareth Hughes <gareth@valinux.com>
+ *    Leif Delgass <ldelgass@retinalburn.net>
+ *    José Fonseca <j_r_fonseca@yahoo.co.uk>
+ */
+
+#define __NO_VERSION__
+#include "mach64.h"
+#include "drmP.h"
+#include "drm.h"
+#include "mach64_drm.h"
+#include "mach64_drv.h"
+
+
+/* ================================================================
+ * DMA hardware state programming functions
+ */
+
+static void mach64_print_dirty( const char *msg, unsigned int flags )
+{
+	DRM_DEBUG( "%s: (0x%x) %s%s%s%s%s%s%s%s%s%s%s%s\n",
+		msg,
+		flags,
+		(flags & MACH64_UPLOAD_DST_OFF_PITCH) ? "dst_off_pitch, " : "",
+		(flags & MACH64_UPLOAD_Z_ALPHA_CNTL)  ? "z_alpha_cntl, " : "",
+		(flags & MACH64_UPLOAD_SCALE_3D_CNTL) ? "scale_3d_cntl, " : "",
+		(flags & MACH64_UPLOAD_DP_FOG_CLR)    ? "dp_fog_clr, " : "",
+		(flags & MACH64_UPLOAD_DP_WRITE_MASK) ? "dp_write_mask, " : "",
+		(flags & MACH64_UPLOAD_DP_PIX_WIDTH)  ? "dp_pix_width, " : "",
+		(flags & MACH64_UPLOAD_SETUP_CNTL)    ? "setup_cntl, " : "",
+		(flags & MACH64_UPLOAD_MISC)          ? "misc, " : "",
+		(flags & MACH64_UPLOAD_TEXTURE)       ? "texture, " : "",
+		(flags & MACH64_UPLOAD_TEX0IMAGE)     ? "tex0 image, " : "",
+		(flags & MACH64_UPLOAD_TEX1IMAGE)     ? "tex1 image, " : "",
+		(flags & MACH64_UPLOAD_CLIPRECTS)     ? "cliprects, " : "" );
+}
+
+/* Mach64 doesn't have hardware cliprects, just one hardware scissor,
+ * so the GL scissor is intersected with each cliprect here
+ */
+/* This function returns 0 on success, 1 for no intersection, and
+ * negative for an error
+ */
+static int mach64_emit_cliprect( DRMFILE filp, drm_mach64_private_t *dev_priv,
+				 drm_clip_rect_t *box )
+{
+	u32 sc_left_right, sc_top_bottom;
+	drm_clip_rect_t scissor;
+	drm_mach64_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	drm_mach64_context_regs_t *regs = &sarea_priv->context_state;
+	DMALOCALS;
+
+	DRM_DEBUG( "%s: box=%p\n", __FUNCTION__, box );
+
+	/* Get GL scissor */
+	/* FIXME: store scissor in SAREA as a cliprect instead of in
+	 * hardware format, or do intersection client-side
+	 */
+	scissor.x1 = regs->sc_left_right & 0xffff;
+	scissor.x2 = (regs->sc_left_right & 0xffff0000) >> 16;
+	scissor.y1 = regs->sc_top_bottom & 0xffff;
+	scissor.y2 = (regs->sc_top_bottom & 0xffff0000) >> 16;
+
+	/* Intersect GL scissor with cliprect */
+	if ( box->x1 > scissor.x1 ) scissor.x1 = box->x1;
+	if ( box->y1 > scissor.y1 ) scissor.y1 = box->y1;
+	if ( box->x2 < scissor.x2 ) scissor.x2 = box->x2;
+	if ( box->y2 < scissor.y2 ) scissor.y2 = box->y2;
+	 /* positive return means skip */
+	if ( scissor.x1 >= scissor.x2 ) return 1;
+	if ( scissor.y1 >= scissor.y2 ) return 1;
+
+	DMAGETPTR( filp, dev_priv, 2 ); /* returns on failure to get buffer */
+
+	sc_left_right = ( (scissor.x1 << 0) | (scissor.x2 << 16) );
+	sc_top_bottom = ( (scissor.y1 << 0) | (scissor.y2 << 16) );
+
+	DMAOUTREG( MACH64_SC_LEFT_RIGHT, sc_left_right );
+	DMAOUTREG( MACH64_SC_TOP_BOTTOM, sc_top_bottom );
+
+	DMAADVANCE( dev_priv, 1 );
+
+	return 0;
+}
+
+static __inline__ int mach64_emit_state( DRMFILE filp, drm_mach64_private_t *dev_priv )
+{
+	drm_mach64_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	drm_mach64_context_regs_t *regs = &sarea_priv->context_state;
+	unsigned int dirty = sarea_priv->dirty;
+	u32 offset = ((regs->tex_size_pitch & 0xf0) >> 2);
+	DMALOCALS;
+
+	if ( MACH64_VERBOSE ) {
+		mach64_print_dirty( __FUNCTION__, dirty );
+	} else {
+		DRM_DEBUG( "%s: dirty=0x%08x\n", __FUNCTION__, dirty );
+	}
+
+	DMAGETPTR( filp, dev_priv, 17 ); /* returns on failure to get buffer */
+
+	if ( dirty & MACH64_UPLOAD_MISC ) {
+		DMAOUTREG( MACH64_DP_MIX, regs->dp_mix );
+		DMAOUTREG( MACH64_DP_SRC, regs->dp_src );
+		DMAOUTREG( MACH64_CLR_CMP_CNTL, regs->clr_cmp_cntl );
+		DMAOUTREG( MACH64_GUI_TRAJ_CNTL, regs->gui_traj_cntl );
+		sarea_priv->dirty &= ~MACH64_UPLOAD_MISC;
+	}
+
+	if ( dirty & MACH64_UPLOAD_DST_OFF_PITCH ) {
+		DMAOUTREG( MACH64_DST_OFF_PITCH, regs->dst_off_pitch );
+		sarea_priv->dirty &= ~MACH64_UPLOAD_DST_OFF_PITCH;
+	}
+	if ( dirty & MACH64_UPLOAD_Z_OFF_PITCH ) {
+		DMAOUTREG( MACH64_Z_OFF_PITCH, regs->z_off_pitch );
+		sarea_priv->dirty &= ~MACH64_UPLOAD_Z_OFF_PITCH;
+	}
+	if ( dirty & MACH64_UPLOAD_Z_ALPHA_CNTL ) {
+		DMAOUTREG( MACH64_Z_CNTL, regs->z_cntl );
+		DMAOUTREG( MACH64_ALPHA_TST_CNTL, regs->alpha_tst_cntl );
+		sarea_priv->dirty &= ~MACH64_UPLOAD_Z_ALPHA_CNTL;
+	}
+	if ( dirty & MACH64_UPLOAD_SCALE_3D_CNTL ) {
+		DMAOUTREG( MACH64_SCALE_3D_CNTL, regs->scale_3d_cntl );
+		sarea_priv->dirty &= ~MACH64_UPLOAD_SCALE_3D_CNTL;
+	}
+	if ( dirty & MACH64_UPLOAD_DP_FOG_CLR ) {
+		DMAOUTREG( MACH64_DP_FOG_CLR, regs->dp_fog_clr );
+		sarea_priv->dirty &= ~MACH64_UPLOAD_DP_FOG_CLR;
+	}
+	if ( dirty & MACH64_UPLOAD_DP_WRITE_MASK ) {
+		DMAOUTREG( MACH64_DP_WRITE_MASK, regs->dp_write_mask );
+		sarea_priv->dirty &= ~MACH64_UPLOAD_DP_WRITE_MASK;
+	}
+	if ( dirty & MACH64_UPLOAD_DP_PIX_WIDTH ) {
+		DMAOUTREG( MACH64_DP_PIX_WIDTH, regs->dp_pix_width );
+		sarea_priv->dirty &= ~MACH64_UPLOAD_DP_PIX_WIDTH;
+	}
+	if ( dirty & MACH64_UPLOAD_SETUP_CNTL ) {
+		DMAOUTREG( MACH64_SETUP_CNTL, regs->setup_cntl );
+		sarea_priv->dirty &= ~MACH64_UPLOAD_SETUP_CNTL;
+	}
+
+	if ( dirty & MACH64_UPLOAD_TEXTURE ) {
+		DMAOUTREG( MACH64_TEX_SIZE_PITCH, regs->tex_size_pitch );
+		DMAOUTREG( MACH64_TEX_CNTL, regs->tex_cntl );
+		DMAOUTREG( MACH64_SECONDARY_TEX_OFF, regs->secondary_tex_off );
+		DMAOUTREG( MACH64_TEX_0_OFF + offset, regs->tex_offset );
+		sarea_priv->dirty &= ~MACH64_UPLOAD_TEXTURE;
+	}
+
+	DMAADVANCE( dev_priv, 1 );
+
+	sarea_priv->dirty &= MACH64_UPLOAD_CLIPRECTS;
+
+	return 0;
+
+}
+
+
+/* ================================================================
+ * DMA command dispatch functions
+ */
+
+static int mach64_dma_dispatch_clear( DRMFILE filp, drm_device_t *dev,
+				      unsigned int flags,
+				      int cx, int cy, int cw, int ch,
+				      unsigned int clear_color,
+				      unsigned int clear_depth )
+{
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	drm_mach64_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	drm_mach64_context_regs_t *ctx = &sarea_priv->context_state;
+	int nbox = sarea_priv->nbox;
+	drm_clip_rect_t *pbox = sarea_priv->boxes;
+	u32 fb_bpp, depth_bpp;
+	int i;
+	DMALOCALS;
+
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+	switch ( dev_priv->fb_bpp ) {
+	case 16:
+		fb_bpp = MACH64_DATATYPE_RGB565;
+		break;
+	case 32:
+		fb_bpp = MACH64_DATATYPE_ARGB8888;
+		break;
+	default:
+		return DRM_ERR(EINVAL);
+	}
+	switch ( dev_priv->depth_bpp ) {
+	case 16:
+		depth_bpp = MACH64_DATATYPE_RGB565;
+		break;
+	case 24:
+	case 32:
+		depth_bpp = MACH64_DATATYPE_ARGB8888;
+		break;
+	default:
+		return DRM_ERR(EINVAL);
+	}
+
+	if ( !nbox ) 
+		return 0;
+
+	DMAGETPTR( filp, dev_priv, nbox * 31 ); /* returns on failure to get buffer */
+
+	for ( i = 0 ; i < nbox ; i++ ) {
+		int x = pbox[i].x1;
+		int y = pbox[i].y1;
+		int w = pbox[i].x2 - x;
+		int h = pbox[i].y2 - y;
+		
+		DRM_DEBUG( "dispatch clear %d,%d-%d,%d flags 0x%x\n",
+			   pbox[i].x1, pbox[i].y1,
+			   pbox[i].x2, pbox[i].y2, flags );
+		
+		if ( flags & (MACH64_FRONT | MACH64_BACK) ) {
+			/* Setup for color buffer clears
+			 */
+			
+			DMAOUTREG( MACH64_Z_CNTL, 0 );
+			DMAOUTREG( MACH64_SCALE_3D_CNTL, 0 );
+			
+			DMAOUTREG( MACH64_SC_LEFT_RIGHT, ctx->sc_left_right );
+			DMAOUTREG( MACH64_SC_TOP_BOTTOM, ctx->sc_top_bottom );
+
+			DMAOUTREG( MACH64_CLR_CMP_CNTL, 0 );
+			DMAOUTREG( MACH64_GUI_TRAJ_CNTL,
+				   (MACH64_DST_X_LEFT_TO_RIGHT |
+				    MACH64_DST_Y_TOP_TO_BOTTOM) );
+			
+			DMAOUTREG( MACH64_DP_PIX_WIDTH, ((fb_bpp << 0) |
+							 (fb_bpp << 4) |
+							 (fb_bpp << 8) |
+							 (fb_bpp << 16) |
+							 (fb_bpp << 28)) );
+			
+			DMAOUTREG( MACH64_DP_FRGD_CLR, clear_color );
+			DMAOUTREG( MACH64_DP_WRITE_MASK, ctx->dp_write_mask );
+			DMAOUTREG( MACH64_DP_MIX, (MACH64_BKGD_MIX_D |
+						   MACH64_FRGD_MIX_S) );
+			DMAOUTREG( MACH64_DP_SRC, (MACH64_BKGD_SRC_FRGD_CLR |
+						   MACH64_FRGD_SRC_FRGD_CLR |
+						   MACH64_MONO_SRC_ONE) );
+			
+						
+		}
+
+		if ( flags & MACH64_FRONT ) {
+						
+			DMAOUTREG( MACH64_DST_OFF_PITCH,
+				   dev_priv->front_offset_pitch );
+			DMAOUTREG( MACH64_DST_X_Y,
+				   (y << 16) | x );
+			DMAOUTREG( MACH64_DST_WIDTH_HEIGHT,
+				   (h << 16) | w );
+			
+		}
+		
+		if ( flags & MACH64_BACK ) {
+						
+			DMAOUTREG( MACH64_DST_OFF_PITCH,
+				   dev_priv->back_offset_pitch );
+			DMAOUTREG( MACH64_DST_X_Y,
+				   (y << 16) | x );
+			DMAOUTREG( MACH64_DST_WIDTH_HEIGHT,
+				   (h << 16) | w );
+			
+		}
+				
+		if ( flags & MACH64_DEPTH ) {
+			/* Setup for depth buffer clear
+			 */
+			DMAOUTREG( MACH64_Z_CNTL, 0 );
+			DMAOUTREG( MACH64_SCALE_3D_CNTL, 0 );
+			
+			DMAOUTREG( MACH64_SC_LEFT_RIGHT, ctx->sc_left_right );
+			DMAOUTREG( MACH64_SC_TOP_BOTTOM, ctx->sc_top_bottom );
+
+			DMAOUTREG( MACH64_CLR_CMP_CNTL, 0 );
+			DMAOUTREG( MACH64_GUI_TRAJ_CNTL,
+				   (MACH64_DST_X_LEFT_TO_RIGHT |
+				    MACH64_DST_Y_TOP_TO_BOTTOM) );
+			
+			DMAOUTREG( MACH64_DP_PIX_WIDTH, ((depth_bpp << 0) |
+							 (depth_bpp << 4) |
+							 (depth_bpp << 8) |
+							 (depth_bpp << 16) |
+							 (depth_bpp << 28)) );
+			
+			DMAOUTREG( MACH64_DP_FRGD_CLR, clear_depth );
+			DMAOUTREG( MACH64_DP_WRITE_MASK, 0xffffffff );
+			DMAOUTREG( MACH64_DP_MIX, (MACH64_BKGD_MIX_D |
+						   MACH64_FRGD_MIX_S) );
+			DMAOUTREG( MACH64_DP_SRC, (MACH64_BKGD_SRC_FRGD_CLR |
+						   MACH64_FRGD_SRC_FRGD_CLR |
+						   MACH64_MONO_SRC_ONE) );
+			
+			DMAOUTREG( MACH64_DST_OFF_PITCH,
+				   dev_priv->depth_offset_pitch );
+			DMAOUTREG( MACH64_DST_X_Y,
+				   (y << 16) | x );
+			DMAOUTREG( MACH64_DST_WIDTH_HEIGHT,
+				   (h << 16) | w );
+		}
+	}
+
+	DMAADVANCE( dev_priv, 1 );
+
+	return 0;
+}
+
+static int mach64_dma_dispatch_swap( DRMFILE filp, drm_device_t *dev )
+{
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	drm_mach64_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	int nbox = sarea_priv->nbox;
+	drm_clip_rect_t *pbox = sarea_priv->boxes;
+	u32 fb_bpp;
+	int i;
+	DMALOCALS;
+
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+	switch ( dev_priv->fb_bpp ) {
+	case 16:
+		fb_bpp = MACH64_DATATYPE_RGB565;
+		break;
+	case 32:
+	default:
+		fb_bpp = MACH64_DATATYPE_ARGB8888;
+		break;
+	}
+
+	if ( !nbox ) 
+		return 0;
+
+	DMAGETPTR( filp, dev_priv, 13 + nbox * 4 ); /* returns on failure to get buffer */
+
+	DMAOUTREG( MACH64_Z_CNTL, 0 );
+	DMAOUTREG( MACH64_SCALE_3D_CNTL, 0 );
+
+	DMAOUTREG( MACH64_SC_LEFT_RIGHT, 0 | ( 8191 << 16 ) ); /* no scissor */
+	DMAOUTREG( MACH64_SC_TOP_BOTTOM, 0 | ( 16383 << 16 ) );
+
+	DMAOUTREG( MACH64_CLR_CMP_CNTL, 0 );
+	DMAOUTREG( MACH64_GUI_TRAJ_CNTL, (MACH64_DST_X_LEFT_TO_RIGHT |
+					  MACH64_DST_Y_TOP_TO_BOTTOM) );
+
+	DMAOUTREG( MACH64_DP_PIX_WIDTH, ((fb_bpp << 0) |
+					 (fb_bpp << 4) |
+					 (fb_bpp << 8) |
+					 (fb_bpp << 16) |
+					 (fb_bpp << 28)) );
+
+	DMAOUTREG( MACH64_DP_WRITE_MASK, 0xffffffff );
+	DMAOUTREG( MACH64_DP_MIX, (MACH64_BKGD_MIX_D |
+				   MACH64_FRGD_MIX_S) );
+	DMAOUTREG( MACH64_DP_SRC, (MACH64_BKGD_SRC_BKGD_CLR |
+				   MACH64_FRGD_SRC_BLIT |
+				   MACH64_MONO_SRC_ONE) );
+
+	DMAOUTREG( MACH64_SRC_OFF_PITCH, dev_priv->back_offset_pitch );
+	DMAOUTREG( MACH64_DST_OFF_PITCH, dev_priv->front_offset_pitch );
+
+	for ( i = 0 ; i < nbox ; i++ ) {
+		int x = pbox[i].x1;
+		int y = pbox[i].y1;
+		int w = pbox[i].x2 - x;
+		int h = pbox[i].y2 - y;
+
+		DRM_DEBUG( "dispatch swap %d,%d-%d,%d\n",
+			  pbox[i].x1, pbox[i].y1,
+			  pbox[i].x2, pbox[i].y2 );
+
+		DMAOUTREG( MACH64_SRC_WIDTH1, w );
+		DMAOUTREG( MACH64_SRC_Y_X, (x << 16) | y );
+		DMAOUTREG( MACH64_DST_Y_X, (x << 16) | y );
+		DMAOUTREG( MACH64_DST_WIDTH_HEIGHT, (h << 16) | w );
+
+	}
+
+	DMAADVANCE( dev_priv, 1 );
+
+	if (dev_priv->driver_mode == MACH64_MODE_DMA_ASYNC) {
+		for (i = 0; i < MACH64_MAX_QUEUED_FRAMES - 1; i++) {
+			dev_priv->frame_ofs[i] = dev_priv->frame_ofs[i+1];
+		}
+		dev_priv->frame_ofs[i] = GETRINGOFFSET();
+
+		dev_priv->sarea_priv->frames_queued++;
+	}
+
+	return 0;
+}
+
+static int mach64_do_get_frames_queued( drm_mach64_private_t *dev_priv )
+{
+	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;
+	drm_mach64_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	int i, start;
+	u32 head, tail, ofs;
+
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+	if (sarea_priv->frames_queued == 0)
+		return 0;
+
+	tail = ring->tail;
+	mach64_ring_tick( dev_priv, ring );
+	head = ring->head;
+
+	start = ( MACH64_MAX_QUEUED_FRAMES - 
+		  DRM_MIN(MACH64_MAX_QUEUED_FRAMES, sarea_priv->frames_queued) );
+
+	if ( head == tail ) {
+		sarea_priv->frames_queued = 0;
+		for (i = start; i < MACH64_MAX_QUEUED_FRAMES; i++) {
+			dev_priv->frame_ofs[i] = ~0;
+		}
+		return 0;
+	}
+
+	for ( i = start; i < MACH64_MAX_QUEUED_FRAMES; i++ ) {
+		ofs = dev_priv->frame_ofs[i];
+		DRM_DEBUG( "frame_ofs[%d] ofs: %d\n", i, ofs );
+		if ( ofs == ~0 ||
+		     ( head < tail && (ofs < head || ofs >= tail) ) ||
+		     ( head > tail && (ofs < head && ofs >= tail) ) ) {
+			sarea_priv->frames_queued = (MACH64_MAX_QUEUED_FRAMES - 1) - i;
+			dev_priv->frame_ofs[i] = ~0;
+		}
+	}
+
+	return sarea_priv->frames_queued;
+}
+
+/* Copy and verify a client submited buffer.
+ * FIXME: Make an assembly optimized version
+ */
+static __inline__ int copy_and_verify_from_user( u32 *to, const u32 *from, unsigned long bytes )
+{
+	unsigned long n = bytes; /* dwords remaining in buffer */
+
+	if ( DRM_VERIFYAREA_READ( from, n ) ) {
+		DRM_ERROR( "%s: verify_area\n", __FUNCTION__ );
+		return DRM_ERR(EFAULT);
+	}
+
+	n >>= 2;
+
+	while ( n > 1 ) {
+		u32 data, reg, count;
+
+		if ( DRM_GET_USER_UNCHECKED( data, from++ ) ) {
+			DRM_ERROR( "%s: get_user\n", __FUNCTION__ );
+			return DRM_ERR(EFAULT);
+		}
+
+		n--;
+
+		reg = le32_to_cpu(data);
+		count = (reg >> 16) + 1;
+		if( count <= n ) {
+			n -= count;
+			reg &= 0xffff;
+
+			/* This is an exact match of Mach64's Setup Engine registers,
+			 * excluding SETUP_CNTL (1_C1).
+			 */
+			if( (reg >= 0x0190 && reg <  0x01c1) || 
+			    (reg >= 0x01ca && reg <= 0x01cf) ) {
+				*to++ = data;
+				if ( DRM_COPY_FROM_USER_UNCHECKED( to, from, count << 2 ) ) {
+					DRM_ERROR( "%s: copy_from_user\n", __FUNCTION__ );
+					return DRM_ERR(EFAULT);
+				}
+				to += count;
+			} else {
+				DRM_ERROR( "%s: Got bad command: 0x%04x\n", __FUNCTION__, reg );
+				return DRM_ERR(EACCES);
+			}
+
+			from += count;
+		} else {
+			DRM_ERROR( "%s: Got bad command count(=%u) dwords remaining=%lu\n", 
+				   __FUNCTION__, count, n );
+			return DRM_ERR(EINVAL);
+		}
+	}
+
+	if (n == 0)
+		return 0;
+	else {
+		DRM_ERROR( "%s: Bad buf->used(=%lu)\n", __FUNCTION__, bytes );
+		return DRM_ERR(EINVAL);
+	}
+}
+
+static int mach64_dma_dispatch_vertex( DRMFILE filp, drm_device_t *dev, int prim, void *buf, 
+				       unsigned long used, int discard )
+{
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	drm_mach64_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	drm_buf_t *copy_buf;
+	int done = 0;
+	int verify_ret = 0;
+	DMALOCALS;
+
+	DRM_DEBUG( "%s: buf=%p used=%lu nbox=%d\n",
+		   __FUNCTION__, buf, used, sarea_priv->nbox );
+
+	if ( used ) {
+		int ret = 0;
+		int i = 0;
+
+		copy_buf = mach64_freelist_get( dev_priv );
+		if (copy_buf == NULL) {
+			DRM_ERROR("%s: couldn't get buffer in DMAGETPTR\n",
+				   __FUNCTION__ );
+			return DRM_ERR(EAGAIN);
+		}
+
+		if ( (verify_ret = 
+		      copy_and_verify_from_user( GETBUFPTR( copy_buf ), buf, used )) == 0 ) {
+
+			copy_buf->used = used;
+
+			DMASETPTR( copy_buf );
+
+			if ( sarea_priv->dirty & ~MACH64_UPLOAD_CLIPRECTS ) {
+				ret = mach64_emit_state( filp, dev_priv );
+				if (ret < 0) return ret;
+			}
+ 
+			do {
+				/* Emit the next cliprect */
+				if ( i < sarea_priv->nbox ) {
+					ret = mach64_emit_cliprect(filp, dev_priv, 
+								   &sarea_priv->boxes[i]);
+					if ( ret < 0 ) {
+						/* failed to get buffer */
+						return ret;
+					} else if ( ret != 0 ) {
+						/* null intersection with scissor */
+						continue;
+					}
+				}
+				if ((i >= sarea_priv->nbox - 1))
+					done = 1;
+
+				/* Add the buffer to the DMA queue */
+				DMAADVANCE( dev_priv, done );
+
+			} while ( ++i < sarea_priv->nbox );
+		}
+
+		if (copy_buf->pending && !done) {
+			DMADISCARDBUF();
+		} else if (!done) {
+			/* This buffer wasn't used (no cliprects or verify failed), so place it back
+			 * on the free list
+			 */
+			struct list_head *ptr;
+			drm_mach64_freelist_t *entry;
+#if MACH64_EXTRA_CHECKING
+			list_for_each(ptr, &dev_priv->pending) {
+				entry = list_entry(ptr, drm_mach64_freelist_t, list);
+				if (copy_buf == entry->buf) {
+					DRM_ERROR( "%s: Trying to release a pending buf\n",
+						   __FUNCTION__ );
+					return DRM_ERR(EFAULT);
+				}
+			}
+#endif
+			ptr = dev_priv->placeholders.next;
+			entry = list_entry(ptr, drm_mach64_freelist_t, list);
+			copy_buf->pending = 0;
+			copy_buf->used = 0;
+			entry->buf = copy_buf;
+			entry->discard = 1;
+			list_del(ptr);
+			list_add_tail(ptr, &dev_priv->free_list);
+		}
+	}
+
+	sarea_priv->dirty &= ~MACH64_UPLOAD_CLIPRECTS;
+	sarea_priv->nbox = 0;
+
+	return verify_ret;
+}
+
+
+static int mach64_dma_dispatch_blit( DRMFILE filp, drm_device_t *dev,
+				     drm_mach64_blit_t *blit )
+{
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	drm_device_dma_t *dma = dev->dma;
+	int dword_shift, dwords;
+	drm_buf_t *buf;
+	DMALOCALS;
+
+	/* The compiler won't optimize away a division by a variable,
+	 * even if the only legal values are powers of two.  Thus, we'll
+	 * use a shift instead.
+	 */
+	switch ( blit->format ) {
+	case MACH64_DATATYPE_ARGB8888:
+		dword_shift = 0;
+		break;
+	case MACH64_DATATYPE_ARGB1555:
+	case MACH64_DATATYPE_RGB565:
+	case MACH64_DATATYPE_VYUY422:
+	case MACH64_DATATYPE_YVYU422:
+	case MACH64_DATATYPE_ARGB4444:
+		dword_shift = 1;
+		break;
+	case MACH64_DATATYPE_CI8:
+	case MACH64_DATATYPE_RGB8:
+		dword_shift = 2;
+		break;
+	default:
+		DRM_ERROR( "invalid blit format %d\n", blit->format );
+		return DRM_ERR(EINVAL);
+	}
+
+	/* Dispatch the blit buffer.
+	 */
+	buf = dma->buflist[blit->idx];
+	
+	if ( buf->filp != filp ) {
+		DRM_ERROR( "process %d (filp %p) using buffer with filp %p\n",
+			   DRM_CURRENTPID, filp, buf->filp );
+		return DRM_ERR(EINVAL);
+	}
+
+	if ( buf->pending ) {
+		DRM_ERROR( "sending pending buffer %d\n", blit->idx );
+		return DRM_ERR(EINVAL);
+	}
+
+	/* Set buf->used to the bytes of blit data based on the blit dimensions 
+	 * and verify the size.  When the setup is emitted to the buffer with 
+	 * the DMA* macros below, buf->used is incremented to include the bytes 
+	 * used for setup as well as the blit data.
+	 */
+	dwords =  (blit->width * blit->height) >> dword_shift;
+	buf->used = dwords << 2;
+	if ( buf->used <= 0 || 
+	     buf->used > MACH64_BUFFER_SIZE - MACH64_HOSTDATA_BLIT_OFFSET ) {
+		DRM_ERROR( "Invalid blit size: %d bytes\n", buf->used );
+		return DRM_ERR(EINVAL);
+	}
+
+	/* FIXME: Use a last buffer flag and reduce the state emitted for subsequent,
+	 * continuation buffers? 
+	 */
+
+	/* Blit via BM_HOSTDATA (gui-master) - like HOST_DATA[0-15], but doesn't require
+	 * a register command every 16 dwords.  State setup is added at the start of the 
+	 * buffer -- the client leaves space for this based on MACH64_HOSTDATA_BLIT_OFFSET
+	 */
+	DMASETPTR( buf );
+
+	DMAOUTREG( MACH64_Z_CNTL, 0 );
+	DMAOUTREG( MACH64_SCALE_3D_CNTL, 0 );
+
+	DMAOUTREG( MACH64_SC_LEFT_RIGHT, 0 | ( 8191 << 16 ) );	/* no scissor */
+	DMAOUTREG( MACH64_SC_TOP_BOTTOM, 0 | ( 16383 << 16 )  );
+
+	DMAOUTREG( MACH64_CLR_CMP_CNTL, 0 );			/* disable */
+	DMAOUTREG( MACH64_GUI_TRAJ_CNTL, 
+		   MACH64_DST_X_LEFT_TO_RIGHT 
+		   | MACH64_DST_Y_TOP_TO_BOTTOM );
+
+	DMAOUTREG( MACH64_DP_PIX_WIDTH,
+		   ( blit->format << 0 )			/* dst pix width */
+		   | ( blit->format << 4 )			/* composite pix width */
+		   | ( blit->format << 8 )			/* src pix width */
+		   | ( blit->format << 16 )			/* host data pix width */
+		   | ( blit->format << 28 )			/* scaler/3D pix width */
+		   );
+
+	DMAOUTREG( MACH64_DP_WRITE_MASK, 0xffffffff );		/* enable all planes */
+	DMAOUTREG( MACH64_DP_MIX, 
+		   MACH64_BKGD_MIX_D 
+		   | MACH64_FRGD_MIX_S );
+	DMAOUTREG( MACH64_DP_SRC, 
+		   MACH64_BKGD_SRC_BKGD_CLR 
+		   | MACH64_FRGD_SRC_HOST 
+		   | MACH64_MONO_SRC_ONE );
+
+	DMAOUTREG( MACH64_DST_OFF_PITCH, (blit->pitch << 22) | (blit->offset >> 3) );
+	DMAOUTREG( MACH64_DST_X_Y, (blit->y << 16) | blit->x );
+	DMAOUTREG( MACH64_DST_WIDTH_HEIGHT, (blit->height << 16) | blit->width );
+
+	DRM_DEBUG( "%s: %d bytes\n", __FUNCTION__, buf->used );
+
+	/* Add the buffer to the queue */
+	DMAADVANCEHOSTDATA( dev_priv );
+
+	return 0;
+}
+
+/* ================================================================
+ * IOCTL functions
+ */
+
+int mach64_dma_clear( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	drm_mach64_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	drm_mach64_clear_t clear;
+	int ret;
+
+	DRM_DEBUG( "%s: pid=%d\n", __FUNCTION__, DRM_CURRENTPID  );
+
+	LOCK_TEST_WITH_RETURN( dev, filp );
+	
+	DRM_COPY_FROM_USER_IOCTL( clear, (drm_mach64_clear_t *)data,
+			     sizeof(clear) );
+	
+	if ( sarea_priv->nbox > MACH64_NR_SAREA_CLIPRECTS )
+		sarea_priv->nbox = MACH64_NR_SAREA_CLIPRECTS;
+
+	ret = mach64_dma_dispatch_clear( filp, dev, clear.flags,
+					 clear.x, clear.y, clear.w, clear.h,
+					 clear.clear_color, clear.clear_depth );
+
+	/* Make sure we restore the 3D state next time.
+	 */
+	sarea_priv->dirty |= (MACH64_UPLOAD_CONTEXT |
+			      MACH64_UPLOAD_MISC);
+	return ret;
+}
+
+int mach64_dma_swap( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	drm_mach64_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	int ret;
+
+	DRM_DEBUG( "%s: pid=%d\n", __FUNCTION__, DRM_CURRENTPID );
+
+	LOCK_TEST_WITH_RETURN( dev, filp );
+	
+	if ( sarea_priv->nbox > MACH64_NR_SAREA_CLIPRECTS )
+		sarea_priv->nbox = MACH64_NR_SAREA_CLIPRECTS;
+
+	ret = mach64_dma_dispatch_swap( filp, dev );
+
+	/* Make sure we restore the 3D state next time.
+	 */
+	sarea_priv->dirty |= (MACH64_UPLOAD_CONTEXT |
+			      MACH64_UPLOAD_MISC);
+	return ret;
+}
+
+int mach64_dma_vertex( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	drm_mach64_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	drm_mach64_vertex_t vertex;
+
+	LOCK_TEST_WITH_RETURN( dev, filp );
+
+	if ( !dev_priv ) {
+		DRM_ERROR( "%s called with no initialization\n", __FUNCTION__ );
+		return DRM_ERR(EINVAL);
+	}
+
+	DRM_COPY_FROM_USER_IOCTL( vertex, (drm_mach64_vertex_t *)data,
+			     sizeof(vertex) );
+
+	DRM_DEBUG( "%s: pid=%d buf=%p used=%lu discard=%d\n",
+		   __FUNCTION__, DRM_CURRENTPID,
+		   vertex.buf, vertex.used, vertex.discard );
+
+	if ( vertex.prim < 0 ||
+	     vertex.prim > MACH64_PRIM_POLYGON ) {
+		DRM_ERROR( "buffer prim %d\n", vertex.prim );
+		return DRM_ERR(EINVAL);
+	}
+
+	if ( vertex.used > MACH64_BUFFER_SIZE || (vertex.used & 3) != 0) {
+		DRM_ERROR( "Invalid vertex buffer size: %lu bytes\n", vertex.used );
+		return DRM_ERR(EINVAL);
+	}
+
+	if ( sarea_priv->nbox > MACH64_NR_SAREA_CLIPRECTS )
+		sarea_priv->nbox = MACH64_NR_SAREA_CLIPRECTS;
+
+	return mach64_dma_dispatch_vertex( filp, dev, vertex.prim, vertex.buf, 
+					   vertex.used, vertex.discard );
+}
+
+int mach64_dma_blit( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_device_dma_t *dma = dev->dma;
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	drm_mach64_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	drm_mach64_blit_t blit;
+	int ret;
+
+	LOCK_TEST_WITH_RETURN( dev, filp );
+
+	DRM_COPY_FROM_USER_IOCTL( blit, (drm_mach64_blit_t *)data,
+			     sizeof(blit) );
+
+	DRM_DEBUG( "%s: pid=%d index=%d\n",
+		   __FUNCTION__, DRM_CURRENTPID, blit.idx );
+
+	if ( blit.idx < 0 || blit.idx >= dma->buf_count ) {
+		DRM_ERROR( "buffer index %d (of %d max)\n",
+			   blit.idx, dma->buf_count - 1 );
+		return DRM_ERR(EINVAL);
+	}
+
+	ret = mach64_dma_dispatch_blit( filp, dev, &blit );
+
+	/* Make sure we restore the 3D state next time.
+	 */
+	sarea_priv->dirty |= (MACH64_UPLOAD_CONTEXT |
+			      MACH64_UPLOAD_MISC | 
+			      MACH64_UPLOAD_CLIPRECTS);
+
+	return ret;
+}
+
+int mach64_get_param( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	drm_mach64_getparam_t param;
+	int value;
+
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+	if ( !dev_priv ) {
+		DRM_ERROR( "%s called with no initialization\n", __FUNCTION__ );
+		return DRM_ERR(EINVAL);
+	}
+
+	DRM_COPY_FROM_USER_IOCTL( param, (drm_mach64_getparam_t *)data,
+	    sizeof(param) );
+
+	switch ( param.param ) {
+	case MACH64_PARAM_FRAMES_QUEUED:
+		/* Needs lock since it calls mach64_ring_tick() */
+		LOCK_TEST_WITH_RETURN( dev, filp );
+		value = mach64_do_get_frames_queued( dev_priv );
+		break;
+	case MACH64_PARAM_IRQ_NR:
+		value = dev->irq;
+		break;
+	default:
+		return DRM_ERR(EINVAL);
+	}
+
+	if ( DRM_COPY_TO_USER( param.value, &value, sizeof(int) ) ) {
+		DRM_ERROR( "copy_to_user\n" );
+		return DRM_ERR(EFAULT);
+	}
+
+	return 0;
+}
diff --git a/shared/mach64.h b/shared/mach64.h
new file mode 100644
index 00000000..d081e07a
--- /dev/null
+++ b/shared/mach64.h
@@ -0,0 +1,117 @@
+/* mach64.h -- ATI Mach 64 DRM template customization -*- linux-c -*-
+ * Created: Wed Feb 14 16:07:10 2001 by gareth@valinux.com
+ *
+ * Copyright 2000 VA Linux Systems, Inc., Sunnyvale, California.
+ * Copyright 2002-2003 Leif Delgass
+ * 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
+ * THE COPYRIGHT OWNER(S) 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:
+ *    Gareth Hughes <gareth@valinux.com>
+ *    Leif Delgass <ldelgass@retinalburn.net>
+ */
+
+#ifndef __MACH64_H__
+#define __MACH64_H__
+
+/* This remains constant for all DRM template files.
+ */
+#define DRM(x) mach64_##x
+
+/* General customization:
+ */
+#define __HAVE_AGP		1
+#define __MUST_HAVE_AGP		0
+#define __HAVE_MTRR		1
+#define __HAVE_CTX_BITMAP	1
+#define __HAVE_PCI_DMA		1
+
+#define DRIVER_AUTHOR		"Gareth Hughes, Leif Delgass, José Fonseca"
+
+#define DRIVER_NAME		"mach64"
+#define DRIVER_DESC		"DRM module for the ATI Rage Pro"
+#define DRIVER_DATE		"20020904"
+
+#define DRIVER_MAJOR		1
+#define DRIVER_MINOR		0
+#define DRIVER_PATCHLEVEL	0
+
+/* Interface history:
+ *
+ * 1.0 - Initial mach64 DRM
+ *
+ */
+#define DRIVER_IOCTLS									\
+	[DRM_IOCTL_NR(DRM_IOCTL_DMA)]	          = { mach64_dma_buffers,    1, 0 },	\
+   	[DRM_IOCTL_NR(DRM_IOCTL_MACH64_INIT)]     = { mach64_dma_init,       1, 1 },	\
+   	[DRM_IOCTL_NR(DRM_IOCTL_MACH64_CLEAR)]    = { mach64_dma_clear,      1, 0 },	\
+   	[DRM_IOCTL_NR(DRM_IOCTL_MACH64_SWAP)]     = { mach64_dma_swap,       1, 0 },	\
+   	[DRM_IOCTL_NR(DRM_IOCTL_MACH64_IDLE)]     = { mach64_dma_idle,       1, 0 },	\
+   	[DRM_IOCTL_NR(DRM_IOCTL_MACH64_RESET)]    = { mach64_engine_reset,   1, 0 },	\
+   	[DRM_IOCTL_NR(DRM_IOCTL_MACH64_VERTEX)]   = { mach64_dma_vertex,     1, 0 },	\
+   	[DRM_IOCTL_NR(DRM_IOCTL_MACH64_BLIT)]     = { mach64_dma_blit,       1, 0 },	\
+   	[DRM_IOCTL_NR(DRM_IOCTL_MACH64_FLUSH)]    = { mach64_dma_flush,      1, 0 },    \
+   	[DRM_IOCTL_NR(DRM_IOCTL_MACH64_GETPARAM)] = { mach64_get_param,      1, 0 }
+
+#define DRIVER_PCI_IDS                                                  \
+	{0x1002, 0x4749, 0, "3D Rage Pro"},                             \
+	{0x1002, 0x4750, 0, "3D Rage Pro 215GP"},                       \
+	{0x1002, 0x4751, 0, "3D Rage Pro 215GQ"},                       \
+	{0x1002, 0x4742, 0, "3D Rage Pro AGP 1X/2X"},                   \
+	{0x1002, 0x4744, 0, "3D Rage Pro AGP 1X"},                      \
+	{0x1002, 0x4c49, 0, "3D Rage LT Pro"},                          \
+	{0x1002, 0x4c50, 0, "3D Rage LT Pro"},                          \
+	{0x1002, 0x4c51, 0, "3D Rage LT Pro"},                          \
+	{0x1002, 0x4c42, 0, "3D Rage LT Pro AGP-133"},                  \
+	{0x1002, 0x4c44, 0, "3D Rage LT Pro AGP-66"},                   \
+	{0x1002, 0x474c, 0, "Rage XC"},                                 \
+	{0x1002, 0x474f, 0, "Rage XL"},                                 \
+	{0x1002, 0x4752, 0, "Rage XL"},                                 \
+	{0x1002, 0x4753, 0, "Rage XC"},                                 \
+	{0x1002, 0x474d, 0, "Rage XL AGP 2X"},                          \
+	{0x1002, 0x474e, 0, "Rage XC AGP"},                             \
+	{0x1002, 0x4c52, 0, "Rage Mobility P/M"},                       \
+	{0x1002, 0x4c53, 0, "Rage Mobility L"},                         \
+	{0x1002, 0x4c4d, 0, "Rage Mobility P/M AGP 2X"},                \
+	{0x1002, 0x4c4e, 0, "Rage Mobility L AGP 2X"},                  \
+	{0, 0, 0, NULL}
+
+/* Driver customization:
+ */
+#define DRIVER_PRETAKEDOWN() do {					\
+	mach64_do_cleanup_dma( dev );					\
+} while (0)
+
+/* DMA customization:
+ */
+#define __HAVE_DMA		1
+#define __HAVE_DMA_IRQ          1
+#define __HAVE_VBL_IRQ		1
+#define __HAVE_SHARED_IRQ       1
+#define __HAVE_IRQ		1
+
+/* Buffer customization:
+ */
+
+#define DRIVER_AGP_BUFFERS_MAP( dev )					\
+	((drm_mach64_private_t *)((dev)->dev_private))->buffers
+
+#endif
diff --git a/shared/mach64_dma.c b/shared/mach64_dma.c
new file mode 100644
index 00000000..1bbffc35
--- /dev/null
+++ b/shared/mach64_dma.c
@@ -0,0 +1,1325 @@
+/* mach64_dma.c -- DMA support for mach64 (Rage Pro) driver -*- linux-c -*-
+ * Created: Sun Dec 03 19:20:26 2000 by gareth@valinux.com
+ *
+ * Copyright 2000 Gareth Hughes
+ * Copyright 2002 Frank C. Earl
+ * Copyright 2002-2003 Leif Delgass
+ * 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
+ * THE COPYRIGHT OWNER(S) 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:
+ *   Gareth Hughes <gareth@valinux.com>
+ *   Frank C. Earl <fearl@airmail.net>
+ *   Leif Delgass <ldelgass@retinalburn.net>
+ *   José Fonseca <j_r_fonseca@yahoo.co.uk>
+ */
+
+#include "mach64.h"
+#include "drmP.h"
+#include "drm.h"
+#include "mach64_drm.h"
+#include "mach64_drv.h"
+
+
+/* ================================================================
+ * Engine, FIFO control
+ */
+
+int mach64_do_wait_for_fifo( drm_mach64_private_t *dev_priv, int entries )
+{
+	int slots = 0, i;
+
+	for ( i = 0 ; i < dev_priv->usec_timeout ; i++ ) {
+		slots = (MACH64_READ( MACH64_FIFO_STAT ) &
+			 MACH64_FIFO_SLOT_MASK);
+		if ( slots <= (0x8000 >> entries) ) return 0;
+		DRM_UDELAY( 1 );
+	}
+
+	DRM_INFO( "%s failed! slots=%d entries=%d\n", __FUNCTION__, slots, entries );
+	return DRM_ERR(EBUSY);
+}
+
+int mach64_do_wait_for_idle( drm_mach64_private_t *dev_priv )
+{
+	int i, ret;
+
+	ret = mach64_do_wait_for_fifo( dev_priv, 16 );
+	if ( ret < 0 ) return ret;
+
+	for ( i = 0 ; i < dev_priv->usec_timeout ; i++ ) {
+		if ( !(MACH64_READ( MACH64_GUI_STAT ) & MACH64_GUI_ACTIVE) ) {
+			return 0;
+		}
+		DRM_UDELAY( 1 );
+	}
+
+	DRM_INFO( "%s failed! GUI_STAT=0x%08x\n", __FUNCTION__, 
+		   MACH64_READ( MACH64_GUI_STAT ) );
+	mach64_dump_ring_info( dev_priv );
+	return DRM_ERR(EBUSY);
+}
+
+int mach64_wait_ring( drm_mach64_private_t *dev_priv, int n )
+{
+	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;
+	int i;
+
+	for ( i = 0 ; i < dev_priv->usec_timeout ; i++ ) {
+		mach64_update_ring_snapshot( dev_priv );
+		if ( ring->space >= n ) {
+			if (i > 0) {
+				DRM_DEBUG( "%s: %d usecs\n", __FUNCTION__, i );
+			}
+			return 0;
+		}
+		DRM_UDELAY( 1 );
+	}
+
+	/* FIXME: This is being ignored... */
+	DRM_ERROR( "failed!\n" );
+	mach64_dump_ring_info( dev_priv );
+	return DRM_ERR(EBUSY);
+}
+
+/* Wait until all DMA requests have been processed... */
+static int mach64_ring_idle( drm_mach64_private_t *dev_priv )
+{
+	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;
+	u32 head;
+	int i;
+
+	head = ring->head;
+	i = 0;
+	while ( i < dev_priv->usec_timeout ) {
+		mach64_update_ring_snapshot( dev_priv );
+		if ( ring->head == ring->tail && 
+		     !(MACH64_READ(MACH64_GUI_STAT) & MACH64_GUI_ACTIVE) ) {
+			if (i > 0) {
+				DRM_DEBUG( "%s: %d usecs\n", __FUNCTION__, i );
+			}
+			return 0;
+		} 
+		if ( ring->head == head ) {
+			++i;
+		} else {
+			head = ring->head;
+			i = 0;
+		}
+		DRM_UDELAY( 1 );
+	}
+
+	DRM_INFO( "%s failed! GUI_STAT=0x%08x\n", __FUNCTION__, 
+		   MACH64_READ( MACH64_GUI_STAT ) );
+	mach64_dump_ring_info( dev_priv );
+	return DRM_ERR(EBUSY);
+}
+
+static void mach64_ring_reset( drm_mach64_private_t *dev_priv )
+{
+	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;
+
+	mach64_do_release_used_buffers( dev_priv );
+	ring->head_addr = ring->start_addr;
+	ring->head = ring->tail = 0;
+	ring->space = ring->size;
+	
+	MACH64_WRITE( MACH64_BM_GUI_TABLE_CMD, 
+		      ring->head_addr | MACH64_CIRCULAR_BUF_SIZE_16KB );
+
+	dev_priv->ring_running = 0;
+}
+
+int mach64_do_dma_flush( drm_mach64_private_t *dev_priv )
+{
+	/* FIXME: It's not necessary to wait for idle when flushing 
+	 * we just need to ensure the ring will be completely processed
+	 * in finite time without another ioctl
+	 */
+	return mach64_ring_idle( dev_priv );
+}
+
+int mach64_do_dma_idle( drm_mach64_private_t *dev_priv )
+{
+	int ret;
+
+	/* wait for completion */
+	if ( (ret = mach64_ring_idle( dev_priv )) < 0 ) {
+		DRM_ERROR( "%s failed BM_GUI_TABLE=0x%08x tail: %u\n", __FUNCTION__, 
+			   MACH64_READ(MACH64_BM_GUI_TABLE), dev_priv->ring.tail );
+		return ret;
+	}
+
+	mach64_ring_stop( dev_priv );
+
+	/* clean up after pass */
+	mach64_do_release_used_buffers( dev_priv );
+	return 0;
+}
+
+/* Reset the engine.  This will stop the DMA if it is running.
+ */
+int mach64_do_engine_reset( drm_mach64_private_t *dev_priv )
+{
+	u32 tmp;
+
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+	/* Kill off any outstanding DMA transfers.
+	 */
+	tmp = MACH64_READ( MACH64_BUS_CNTL );
+	MACH64_WRITE( MACH64_BUS_CNTL,
+				  tmp | MACH64_BUS_MASTER_DIS );
+	
+	/* Reset the GUI engine (high to low transition).
+	 */
+	tmp = MACH64_READ( MACH64_GEN_TEST_CNTL );
+	MACH64_WRITE( MACH64_GEN_TEST_CNTL,
+				  tmp & ~MACH64_GUI_ENGINE_ENABLE );
+	/* Enable the GUI engine
+	 */
+	tmp = MACH64_READ( MACH64_GEN_TEST_CNTL );
+	MACH64_WRITE( MACH64_GEN_TEST_CNTL,
+				  tmp | MACH64_GUI_ENGINE_ENABLE );
+
+	/* ensure engine is not locked up by clearing any FIFO or HOST errors
+	*/
+	tmp = MACH64_READ( MACH64_BUS_CNTL );
+	MACH64_WRITE( MACH64_BUS_CNTL, tmp | 0x00a00000 );
+
+	/* Once GUI engine is restored, disable bus mastering */
+	MACH64_WRITE( MACH64_SRC_CNTL, 0 );
+
+	/* Reset descriptor ring */
+	mach64_ring_reset( dev_priv );
+	
+	return 0;
+}
+
+/* ================================================================
+ * Debugging output
+ */
+
+void mach64_dump_engine_info( drm_mach64_private_t *dev_priv )
+{
+	DRM_INFO( "\n" );
+	if ( !dev_priv->is_pci )
+	{
+		DRM_INFO( "           AGP_BASE = 0x%08x\n", MACH64_READ( MACH64_AGP_BASE ) );
+		DRM_INFO( "           AGP_CNTL = 0x%08x\n", MACH64_READ( MACH64_AGP_CNTL ) );
+	}
+	DRM_INFO( "     ALPHA_TST_CNTL = 0x%08x\n", MACH64_READ( MACH64_ALPHA_TST_CNTL ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "         BM_COMMAND = 0x%08x\n", MACH64_READ( MACH64_BM_COMMAND ) );
+	DRM_INFO( "BM_FRAME_BUF_OFFSET = 0x%08x\n", MACH64_READ( MACH64_BM_FRAME_BUF_OFFSET ) );
+	DRM_INFO( "       BM_GUI_TABLE = 0x%08x\n", MACH64_READ( MACH64_BM_GUI_TABLE ) );
+	DRM_INFO( "          BM_STATUS = 0x%08x\n", MACH64_READ( MACH64_BM_STATUS ) );
+	DRM_INFO( " BM_SYSTEM_MEM_ADDR = 0x%08x\n", MACH64_READ( MACH64_BM_SYSTEM_MEM_ADDR ) );
+	DRM_INFO( "    BM_SYSTEM_TABLE = 0x%08x\n", MACH64_READ( MACH64_BM_SYSTEM_TABLE ) );
+	DRM_INFO( "           BUS_CNTL = 0x%08x\n", MACH64_READ( MACH64_BUS_CNTL ) );
+	DRM_INFO( "\n" );
+	/* DRM_INFO( "         CLOCK_CNTL = 0x%08x\n", MACH64_READ( MACH64_CLOCK_CNTL ) ); */
+	DRM_INFO( "        CLR_CMP_CLR = 0x%08x\n", MACH64_READ( MACH64_CLR_CMP_CLR ) );
+	DRM_INFO( "       CLR_CMP_CNTL = 0x%08x\n", MACH64_READ( MACH64_CLR_CMP_CNTL ) );
+	/* DRM_INFO( "        CLR_CMP_MSK = 0x%08x\n", MACH64_READ( MACH64_CLR_CMP_MSK ) ); */
+	DRM_INFO( "     CONFIG_CHIP_ID = 0x%08x\n", MACH64_READ( MACH64_CONFIG_CHIP_ID ) );
+	DRM_INFO( "        CONFIG_CNTL = 0x%08x\n", MACH64_READ( MACH64_CONFIG_CNTL ) );
+	DRM_INFO( "       CONFIG_STAT0 = 0x%08x\n", MACH64_READ( MACH64_CONFIG_STAT0 ) );
+	DRM_INFO( "       CONFIG_STAT1 = 0x%08x\n", MACH64_READ( MACH64_CONFIG_STAT1 ) );
+	DRM_INFO( "       CONFIG_STAT2 = 0x%08x\n", MACH64_READ( MACH64_CONFIG_STAT2 ) );
+	DRM_INFO( "            CRC_SIG = 0x%08x\n", MACH64_READ( MACH64_CRC_SIG ) );
+	DRM_INFO( "  CUSTOM_MACRO_CNTL = 0x%08x\n", MACH64_READ( MACH64_CUSTOM_MACRO_CNTL ) );
+	DRM_INFO( "\n" );
+	/* DRM_INFO( "           DAC_CNTL = 0x%08x\n", MACH64_READ( MACH64_DAC_CNTL ) ); */
+	/* DRM_INFO( "           DAC_REGS = 0x%08x\n", MACH64_READ( MACH64_DAC_REGS ) ); */
+	DRM_INFO( "        DP_BKGD_CLR = 0x%08x\n", MACH64_READ( MACH64_DP_BKGD_CLR ) );
+	DRM_INFO( "        DP_FRGD_CLR = 0x%08x\n", MACH64_READ( MACH64_DP_FRGD_CLR ) );
+	DRM_INFO( "             DP_MIX = 0x%08x\n", MACH64_READ( MACH64_DP_MIX ) );
+	DRM_INFO( "       DP_PIX_WIDTH = 0x%08x\n", MACH64_READ( MACH64_DP_PIX_WIDTH ) );
+	DRM_INFO( "             DP_SRC = 0x%08x\n", MACH64_READ( MACH64_DP_SRC ) );
+	DRM_INFO( "      DP_WRITE_MASK = 0x%08x\n", MACH64_READ( MACH64_DP_WRITE_MASK ) );
+	DRM_INFO( "         DSP_CONFIG = 0x%08x\n", MACH64_READ( MACH64_DSP_CONFIG ) );
+	DRM_INFO( "         DSP_ON_OFF = 0x%08x\n", MACH64_READ( MACH64_DSP_ON_OFF ) );
+	DRM_INFO( "           DST_CNTL = 0x%08x\n", MACH64_READ( MACH64_DST_CNTL ) );
+	DRM_INFO( "      DST_OFF_PITCH = 0x%08x\n", MACH64_READ( MACH64_DST_OFF_PITCH ) );
+	DRM_INFO( "\n" );
+	/* DRM_INFO( "       EXT_DAC_REGS = 0x%08x\n", MACH64_READ( MACH64_EXT_DAC_REGS ) ); */
+	DRM_INFO( "       EXT_MEM_CNTL = 0x%08x\n", MACH64_READ( MACH64_EXT_MEM_CNTL ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "          FIFO_STAT = 0x%08x\n", MACH64_READ( MACH64_FIFO_STAT ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "      GEN_TEST_CNTL = 0x%08x\n", MACH64_READ( MACH64_GEN_TEST_CNTL ) );
+	/* DRM_INFO( "              GP_IO = 0x%08x\n", MACH64_READ( MACH64_GP_IO ) ); */
+	DRM_INFO( "   GUI_CMDFIFO_DATA = 0x%08x\n", MACH64_READ( MACH64_GUI_CMDFIFO_DATA ) );
+	DRM_INFO( "  GUI_CMDFIFO_DEBUG = 0x%08x\n", MACH64_READ( MACH64_GUI_CMDFIFO_DEBUG ) );
+	DRM_INFO( "           GUI_CNTL = 0x%08x\n", MACH64_READ( MACH64_GUI_CNTL ) );
+	DRM_INFO( "           GUI_STAT = 0x%08x\n", MACH64_READ( MACH64_GUI_STAT ) );
+	DRM_INFO( "      GUI_TRAJ_CNTL = 0x%08x\n", MACH64_READ( MACH64_GUI_TRAJ_CNTL ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "          HOST_CNTL = 0x%08x\n", MACH64_READ( MACH64_HOST_CNTL ) );
+	DRM_INFO( "           HW_DEBUG = 0x%08x\n", MACH64_READ( MACH64_HW_DEBUG ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "    MEM_ADDR_CONFIG = 0x%08x\n", MACH64_READ( MACH64_MEM_ADDR_CONFIG ) );
+	DRM_INFO( "       MEM_BUF_CNTL = 0x%08x\n", MACH64_READ( MACH64_MEM_BUF_CNTL ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "           PAT_REG0 = 0x%08x\n", MACH64_READ( MACH64_PAT_REG0 ) );
+	DRM_INFO( "           PAT_REG1 = 0x%08x\n", MACH64_READ( MACH64_PAT_REG1 ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "            SC_LEFT = 0x%08x\n", MACH64_READ( MACH64_SC_LEFT ) );
+	DRM_INFO( "           SC_RIGHT = 0x%08x\n", MACH64_READ( MACH64_SC_RIGHT ) );
+	DRM_INFO( "             SC_TOP = 0x%08x\n", MACH64_READ( MACH64_SC_TOP ) );
+	DRM_INFO( "          SC_BOTTOM = 0x%08x\n", MACH64_READ( MACH64_SC_BOTTOM ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "      SCALE_3D_CNTL = 0x%08x\n", MACH64_READ( MACH64_SCALE_3D_CNTL ) );
+	DRM_INFO( "       SCRATCH_REG0 = 0x%08x\n", MACH64_READ( MACH64_SCRATCH_REG0 ) );
+	DRM_INFO( "       SCRATCH_REG1 = 0x%08x\n", MACH64_READ( MACH64_SCRATCH_REG1 ) );
+	DRM_INFO( "         SETUP_CNTL = 0x%08x\n", MACH64_READ( MACH64_SETUP_CNTL ) );
+	DRM_INFO( "           SRC_CNTL = 0x%08x\n", MACH64_READ( MACH64_SRC_CNTL ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "           TEX_CNTL = 0x%08x\n", MACH64_READ( MACH64_TEX_CNTL ) );
+	DRM_INFO( "     TEX_SIZE_PITCH = 0x%08x\n", MACH64_READ( MACH64_TEX_SIZE_PITCH ) );
+	DRM_INFO( "       TIMER_CONFIG = 0x%08x\n", MACH64_READ( MACH64_TIMER_CONFIG ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "             Z_CNTL = 0x%08x\n", MACH64_READ( MACH64_Z_CNTL ) );
+	DRM_INFO( "        Z_OFF_PITCH = 0x%08x\n", MACH64_READ( MACH64_Z_OFF_PITCH ) );
+	DRM_INFO( "\n" );
+}
+
+#define MACH64_DUMP_CONTEXT	3
+
+static void mach64_dump_buf_info( drm_mach64_private_t *dev_priv, drm_buf_t *buf)
+{
+	u32 addr = GETBUFADDR( buf );
+	u32 used = buf->used >> 2;
+	u32 sys_addr = MACH64_READ( MACH64_BM_SYSTEM_MEM_ADDR );
+	u32 *p = GETBUFPTR( buf );
+	int skipped = 0;
+
+	DRM_INFO( "buffer contents:\n" );
+	
+	while ( used ) {
+		u32 reg, count;
+
+		reg = le32_to_cpu(*p++);
+		if( addr <= GETBUFADDR( buf ) + MACH64_DUMP_CONTEXT * 4 ||
+		    (addr >= sys_addr - MACH64_DUMP_CONTEXT * 4 &&
+		    addr <= sys_addr + MACH64_DUMP_CONTEXT * 4) ||
+		    addr >= GETBUFADDR( buf ) + buf->used - MACH64_DUMP_CONTEXT * 4) {
+			DRM_INFO("%08x:  0x%08x\n", addr, reg);
+		}
+		addr += 4;
+		used--;
+		
+		count = (reg >> 16) + 1;
+		reg = reg & 0xffff;
+		reg = MMSELECT( reg );
+		while ( count && used ) {
+			if( addr <= GETBUFADDR( buf ) + MACH64_DUMP_CONTEXT * 4 ||
+			    (addr >= sys_addr - MACH64_DUMP_CONTEXT * 4 &&
+			    addr <= sys_addr + MACH64_DUMP_CONTEXT * 4) ||
+			    addr >= GETBUFADDR( buf ) + buf->used - MACH64_DUMP_CONTEXT * 4) {
+				DRM_INFO("%08x:    0x%04x = 0x%08x\n", addr, reg, le32_to_cpu(*p));
+				skipped = 0;
+			} else {
+				if( !skipped ) {
+					DRM_INFO( "  ...\n" );
+					skipped = 1;
+				}
+			}
+			p++;
+			addr += 4;
+			used--;
+			
+			reg += 4;
+			count--;
+		}
+	}
+	
+	DRM_INFO( "\n" );
+}
+
+void mach64_dump_ring_info( drm_mach64_private_t *dev_priv )
+{
+	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;
+	int i, skipped;
+	
+	DRM_INFO( "\n" );
+	
+	DRM_INFO("ring contents:\n");
+	DRM_INFO("  head_addr: 0x%08x head: %u tail: %u\n\n", 
+		 ring->head_addr, ring->head, ring->tail );
+	
+	skipped = 0;
+	for ( i = 0; i < ring->size / sizeof(u32); i += 4) {
+		if( i <= MACH64_DUMP_CONTEXT * 4 || 
+		    i >= ring->size / sizeof(u32) - MACH64_DUMP_CONTEXT * 4 ||
+		    (i >= ring->tail - MACH64_DUMP_CONTEXT * 4 &&
+		    i <= ring->tail + MACH64_DUMP_CONTEXT * 4) ||
+		    (i >= ring->head - MACH64_DUMP_CONTEXT * 4 &&
+		    i <= ring->head + MACH64_DUMP_CONTEXT * 4)) {
+			DRM_INFO( "  0x%08x:  0x%08x 0x%08x 0x%08x 0x%08x%s%s\n",
+				ring->start_addr + i * sizeof(u32),
+				le32_to_cpu(((u32*) ring->start)[i + 0]),
+				le32_to_cpu(((u32*) ring->start)[i + 1]),
+				le32_to_cpu(((u32*) ring->start)[i + 2]), 
+				le32_to_cpu(((u32*) ring->start)[i + 3]),
+				i == ring->head ? " (head)" : "",
+				i == ring->tail ? " (tail)" : ""
+			);
+			skipped = 0;
+		} else {
+			if( !skipped ) {
+				DRM_INFO( "  ...\n" );
+				skipped = 1;
+			}
+		}
+	}
+
+	DRM_INFO( "\n" );
+	
+	if( ring->head >= 0 && ring->head < ring->size / sizeof(u32) ) {
+		struct list_head *ptr;
+		u32 addr = le32_to_cpu(((u32 *)ring->start)[ring->head + 1]);
+
+		list_for_each(ptr, &dev_priv->pending) {
+			drm_mach64_freelist_t *entry = 
+				list_entry(ptr, drm_mach64_freelist_t, list);
+			drm_buf_t *buf = entry->buf;
+
+			u32 buf_addr = GETBUFADDR( buf );
+			
+			if ( buf_addr <= addr && addr < buf_addr + buf->used ) {
+				mach64_dump_buf_info ( dev_priv, buf );
+			}
+		}
+	}
+
+	DRM_INFO( "\n" );
+	DRM_INFO( "       BM_GUI_TABLE = 0x%08x\n", MACH64_READ( MACH64_BM_GUI_TABLE ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "BM_FRAME_BUF_OFFSET = 0x%08x\n", MACH64_READ( MACH64_BM_FRAME_BUF_OFFSET ) );
+	DRM_INFO( " BM_SYSTEM_MEM_ADDR = 0x%08x\n", MACH64_READ( MACH64_BM_SYSTEM_MEM_ADDR ) );
+	DRM_INFO( "         BM_COMMAND = 0x%08x\n", MACH64_READ( MACH64_BM_COMMAND ) );
+	DRM_INFO( "\n" );
+	DRM_INFO( "          BM_STATUS = 0x%08x\n", MACH64_READ( MACH64_BM_STATUS ) );
+	DRM_INFO( "           BUS_CNTL = 0x%08x\n", MACH64_READ( MACH64_BUS_CNTL ) );
+	DRM_INFO( "          FIFO_STAT = 0x%08x\n", MACH64_READ( MACH64_FIFO_STAT ) );
+	DRM_INFO( "           GUI_STAT = 0x%08x\n", MACH64_READ( MACH64_GUI_STAT ) );
+	DRM_INFO( "           SRC_CNTL = 0x%08x\n", MACH64_READ( MACH64_SRC_CNTL ) );
+}
+
+
+/* ================================================================
+ * DMA test and initialization
+ */
+
+static int mach64_bm_dma_test( drm_device_t *dev )
+{
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	dma_addr_t data_handle;
+	void *cpu_addr_data;
+	u32 data_addr;
+	u32 *table, *data;
+	u32 expected[2];
+	u32 src_cntl, pat_reg0, pat_reg1;
+	int i, count, failed;
+
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+	table = (u32 *) dev_priv->ring.start;
+
+	/* FIXME: get a dma buffer from the freelist here */
+	DRM_DEBUG( "Allocating data memory ...\n" );
+	cpu_addr_data = DRM(pci_alloc)( dev, 0x1000, 0x1000, 0xfffffffful, &data_handle );
+	if (!cpu_addr_data || !data_handle) {
+		DRM_INFO( "data-memory allocation failed!\n" );
+		return DRM_ERR(ENOMEM);
+	} else {
+		data = (u32 *) cpu_addr_data;
+		data_addr = (u32) data_handle;
+	}
+
+	/* Save the X server's value for SRC_CNTL and restore it
+	 * in case our test fails.  This prevents the X server
+	 * from disabling it's cache for this register
+	 */
+	src_cntl = MACH64_READ( MACH64_SRC_CNTL );
+	pat_reg0 = MACH64_READ( MACH64_PAT_REG0 );
+	pat_reg1 = MACH64_READ( MACH64_PAT_REG1 );
+
+	mach64_do_wait_for_fifo( dev_priv, 3 );
+
+	MACH64_WRITE( MACH64_SRC_CNTL, 0 );
+	MACH64_WRITE( MACH64_PAT_REG0, 0x11111111 );
+	MACH64_WRITE( MACH64_PAT_REG1, 0x11111111 );
+
+	mach64_do_wait_for_idle( dev_priv );
+
+	for (i=0; i < 2; i++) {
+		u32 reg;
+		reg = MACH64_READ( (MACH64_PAT_REG0 + i*4) );
+		DRM_DEBUG( "(Before DMA Transfer) reg %d = 0x%08x\n", i, reg );
+		if ( reg != 0x11111111 ) {
+			DRM_INFO( "Error initializing test registers\n" );
+			DRM_INFO( "resetting engine ...\n");
+			mach64_do_engine_reset( dev_priv );
+			DRM_INFO( "freeing data buffer memory.\n" );
+			DRM(pci_free)( dev, 0x1000, cpu_addr_data, data_handle );
+			return DRM_ERR(EIO);
+		}
+	}
+
+	/* fill up a buffer with sets of 2 consecutive writes starting with PAT_REG0 */
+	count = 0;
+
+	data[count++] = cpu_to_le32(DMAREG(MACH64_PAT_REG0) | (1 << 16));
+	data[count++] = expected[0] = 0x22222222;
+	data[count++] = expected[1] = 0xaaaaaaaa;
+
+	while (count < 1020) {
+		data[count++] = cpu_to_le32(DMAREG(MACH64_PAT_REG0) | (1 << 16));
+		data[count++] = 0x22222222;
+		data[count++] = 0xaaaaaaaa;
+	}
+	data[count++] = cpu_to_le32(DMAREG(MACH64_SRC_CNTL) | (0 << 16));
+	data[count++] = 0;
+
+	DRM_DEBUG( "Preparing table ...\n" );
+	table[MACH64_DMA_FRAME_BUF_OFFSET] = cpu_to_le32(MACH64_BM_ADDR + 
+							 MACH64_APERTURE_OFFSET);
+	table[MACH64_DMA_SYS_MEM_ADDR]     = cpu_to_le32(data_addr);
+	table[MACH64_DMA_COMMAND]          = cpu_to_le32(count * sizeof( u32 ) 
+							 | MACH64_DMA_HOLD_OFFSET 
+							 | MACH64_DMA_EOL);
+	table[MACH64_DMA_RESERVED]         = 0;
+
+	DRM_DEBUG( "table[0] = 0x%08x\n", table[0] );
+	DRM_DEBUG( "table[1] = 0x%08x\n", table[1] );
+	DRM_DEBUG( "table[2] = 0x%08x\n", table[2] );
+	DRM_DEBUG( "table[3] = 0x%08x\n", table[3] );
+
+	for ( i = 0 ; i < 6 ; i++ ) {
+		DRM_DEBUG( " data[%d] = 0x%08x\n", i, data[i] );
+	}
+	DRM_DEBUG( " ...\n" );
+	for ( i = count-5 ; i < count ; i++ ) {
+		DRM_DEBUG( " data[%d] = 0x%08x\n", i, data[i] );
+	}
+
+	DRM_MEMORYBARRIER();
+
+	DRM_DEBUG( "waiting for idle...\n" );
+	if ( ( i = mach64_do_wait_for_idle( dev_priv ) ) ) {
+		DRM_INFO( "mach64_do_wait_for_idle failed (result=%d)\n", i);
+		DRM_INFO( "resetting engine ...\n");
+		mach64_do_engine_reset( dev_priv );
+		mach64_do_wait_for_fifo( dev_priv, 3 );
+		MACH64_WRITE( MACH64_SRC_CNTL, src_cntl );
+		MACH64_WRITE( MACH64_PAT_REG0, pat_reg0 );
+		MACH64_WRITE( MACH64_PAT_REG1, pat_reg1 );
+		DRM_INFO( "freeing data buffer memory.\n" );
+		DRM(pci_free)( dev, 0x1000, cpu_addr_data, data_handle );
+		return i;
+	}
+	DRM_DEBUG( "waiting for idle...done\n" );
+
+	DRM_DEBUG( "BUS_CNTL = 0x%08x\n", MACH64_READ( MACH64_BUS_CNTL ) );
+	DRM_DEBUG( "SRC_CNTL = 0x%08x\n", MACH64_READ( MACH64_SRC_CNTL ) );
+	DRM_DEBUG( "\n" );
+	DRM_DEBUG( "data bus addr = 0x%08x\n", data_addr );
+	DRM_DEBUG( "table bus addr = 0x%08x\n", dev_priv->ring.start_addr );
+
+	DRM_DEBUG( "starting DMA transfer...\n" );
+	MACH64_WRITE( MACH64_BM_GUI_TABLE_CMD,
+			  dev_priv->ring.start_addr |
+			  MACH64_CIRCULAR_BUF_SIZE_16KB );
+
+	MACH64_WRITE( MACH64_SRC_CNTL, 
+		      MACH64_SRC_BM_ENABLE | MACH64_SRC_BM_SYNC |
+		      MACH64_SRC_BM_OP_SYSTEM_TO_REG );
+
+	/* Kick off the transfer */
+	DRM_DEBUG( "starting DMA transfer... done.\n" );
+	MACH64_WRITE( MACH64_DST_HEIGHT_WIDTH, 0 );
+
+	DRM_DEBUG( "waiting for idle...\n" );
+
+	if ( ( i = mach64_do_wait_for_idle( dev_priv ) ) ) {
+		/* engine locked up, dump register state and reset */
+		DRM_INFO( "mach64_do_wait_for_idle failed (result=%d)\n", i);
+		mach64_dump_engine_info( dev_priv );
+		DRM_INFO( "resetting engine ...\n");
+		mach64_do_engine_reset( dev_priv );
+		mach64_do_wait_for_fifo( dev_priv, 3 );
+		MACH64_WRITE( MACH64_SRC_CNTL, src_cntl );
+		MACH64_WRITE( MACH64_PAT_REG0, pat_reg0 );
+		MACH64_WRITE( MACH64_PAT_REG1, pat_reg1 );
+		DRM_INFO( "freeing data buffer memory.\n" );
+		DRM(pci_free)( dev, 0x1000, cpu_addr_data, data_handle );
+		return i;
+	}
+
+	DRM_DEBUG( "waiting for idle...done\n" );
+
+	/* restore SRC_CNTL */
+	mach64_do_wait_for_fifo( dev_priv, 1 );
+	MACH64_WRITE( MACH64_SRC_CNTL, src_cntl );
+
+	failed = 0;
+
+	/* Check register values to see if the GUI master operation succeeded */
+	for ( i = 0; i < 2; i++ ) {
+		u32 reg;
+		reg = MACH64_READ( (MACH64_PAT_REG0 + i*4) );
+		DRM_DEBUG( "(After DMA Transfer) reg %d = 0x%08x\n", i, reg );
+		if (reg != expected[i]) {
+			failed = -1;
+		}
+	}
+
+	/* restore pattern registers */
+	mach64_do_wait_for_fifo( dev_priv, 2 );
+	MACH64_WRITE( MACH64_PAT_REG0, pat_reg0 );
+	MACH64_WRITE( MACH64_PAT_REG1, pat_reg1 );
+
+	DRM_DEBUG( "freeing data buffer memory.\n" );
+	DRM(pci_free)( dev, 0x1000, cpu_addr_data, data_handle );
+	DRM_DEBUG( "returning ...\n" );
+
+	return failed;
+}
+
+
+static int mach64_do_dma_init( drm_device_t *dev, drm_mach64_init_t *init )
+{
+	drm_mach64_private_t *dev_priv;
+	u32 tmp;
+	int i, ret;
+
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+	dev_priv = DRM(alloc)( sizeof(drm_mach64_private_t), DRM_MEM_DRIVER );
+	if ( dev_priv == NULL )
+		return DRM_ERR(ENOMEM);
+	
+	memset( dev_priv, 0, sizeof(drm_mach64_private_t) );
+
+	dev_priv->is_pci	= init->is_pci;
+
+	dev_priv->fb_bpp	= init->fb_bpp;
+	dev_priv->front_offset  = init->front_offset;
+	dev_priv->front_pitch   = init->front_pitch;
+	dev_priv->back_offset   = init->back_offset;
+	dev_priv->back_pitch    = init->back_pitch;
+
+	dev_priv->depth_bpp     = init->depth_bpp;
+	dev_priv->depth_offset  = init->depth_offset;
+	dev_priv->depth_pitch   = init->depth_pitch;
+
+	dev_priv->front_offset_pitch	= (((dev_priv->front_pitch/8) << 22) |
+					   (dev_priv->front_offset >> 3));
+	dev_priv->back_offset_pitch	= (((dev_priv->back_pitch/8) << 22) |
+					   (dev_priv->back_offset >> 3));
+	dev_priv->depth_offset_pitch	= (((dev_priv->depth_pitch/8) << 22) |
+					   (dev_priv->depth_offset >> 3));
+
+	dev_priv->usec_timeout		= 1000000;
+
+	/* Set up the freelist, placeholder list and pending list */
+	INIT_LIST_HEAD(&dev_priv->free_list);
+	INIT_LIST_HEAD(&dev_priv->placeholders);
+	INIT_LIST_HEAD(&dev_priv->pending);
+
+	DRM_GETSAREA();
+
+	if (!dev_priv->sarea) {
+		DRM_ERROR("can not find sarea!\n");
+		dev->dev_private = (void *)dev_priv;
+	   	mach64_do_cleanup_dma(dev);
+	   	return DRM_ERR(EINVAL);
+	}
+	DRM_FIND_MAP( dev_priv->fb, init->fb_offset );
+	if (!dev_priv->fb) {
+		DRM_ERROR("can not find frame buffer map!\n");
+		dev->dev_private = (void *)dev_priv;
+	   	mach64_do_cleanup_dma(dev);
+	   	return DRM_ERR(EINVAL);
+	}
+	DRM_FIND_MAP( dev_priv->mmio, init->mmio_offset );
+	if (!dev_priv->mmio) {
+		DRM_ERROR("can not find mmio map!\n");
+		dev->dev_private = (void *)dev_priv;
+	   	mach64_do_cleanup_dma(dev);
+	   	return DRM_ERR(EINVAL);
+	}
+
+	dev_priv->sarea_priv = (drm_mach64_sarea_t *)
+		((u8 *)dev_priv->sarea->handle +
+		 init->sarea_priv_offset);
+
+	if( !dev_priv->is_pci ) {
+		DRM_FIND_MAP( dev_priv->ring_map, init->ring_offset );
+		if ( !dev_priv->ring_map ) {
+			DRM_ERROR( "can not find ring map!\n" );
+			dev->dev_private = (void *)dev_priv;
+			mach64_do_cleanup_dma(dev);
+			return -EINVAL;
+		}
+		DRM_IOREMAP( dev_priv->ring_map, dev );
+		if ( !dev_priv->ring_map->handle ) {
+		        DRM_ERROR( "can not ioremap virtual address for"
+                                   " descriptor ring\n" );
+			dev->dev_private = (void *) dev_priv;
+			mach64_do_cleanup_dma( dev );
+			return -ENOMEM;
+                }
+	        DRM_FIND_MAP( dev_priv->buffers, init->buffers_offset );
+		if ( !dev_priv->buffers ) {
+			DRM_ERROR( "can not find dma buffer map!\n" );
+			dev->dev_private = (void *)dev_priv;
+			mach64_do_cleanup_dma( dev );
+			return DRM_ERR(EINVAL);
+		}
+		DRM_IOREMAP( dev_priv->buffers, dev );
+		if ( !dev_priv->buffers->handle ) {
+			DRM_ERROR( "can not ioremap virtual address for"
+				   " dma buffer\n" );
+			dev->dev_private = (void *) dev_priv;
+			mach64_do_cleanup_dma( dev );
+			return DRM_ERR(ENOMEM);
+		}
+		DRM_FIND_MAP( dev_priv->agp_textures,
+			      init->agp_textures_offset );
+		if (!dev_priv->agp_textures) {
+			DRM_ERROR( "can not find agp texture region!\n" );
+			dev->dev_private = (void *)dev_priv;
+			mach64_do_cleanup_dma( dev );
+			return DRM_ERR(EINVAL);
+		}
+	}
+
+	dev->dev_private = (void *) dev_priv;
+
+	dev_priv->driver_mode = init->dma_mode;
+
+	/* changing the FIFO size from the default causes problems with DMA */
+	tmp = MACH64_READ( MACH64_GUI_CNTL );
+	if ( (tmp & MACH64_CMDFIFO_SIZE_MASK) != MACH64_CMDFIFO_SIZE_128 ) {
+		DRM_INFO( "Setting FIFO size to 128 entries\n");
+		/* FIFO must be empty to change the FIFO depth */
+		if ((ret=mach64_do_wait_for_idle( dev_priv ))) {
+			DRM_ERROR("wait for idle failed before changing FIFO depth!\n");
+			mach64_do_cleanup_dma( dev );
+			return ret;
+		}
+		MACH64_WRITE( MACH64_GUI_CNTL, ( ( tmp & ~MACH64_CMDFIFO_SIZE_MASK ) \
+						 | MACH64_CMDFIFO_SIZE_128 ) );
+		/* need to read GUI_STAT for proper sync according to docs */
+		if ((ret=mach64_do_wait_for_idle( dev_priv ))) {
+			DRM_ERROR("wait for idle failed when changing FIFO depth!\n");
+			mach64_do_cleanup_dma( dev );
+			return ret;
+		}
+	}
+
+	/* allocate descriptor memory from pci pool */
+	DRM_DEBUG( "Allocating dma descriptor ring\n" );
+	dev_priv->ring.size = 0x4000; /* 16KB */
+
+	if ( dev_priv->is_pci ) {
+		dev_priv->ring.start = DRM(pci_alloc)( dev, dev_priv->ring.size, 
+						       dev_priv->ring.size, 0xfffffffful,
+						       &dev_priv->ring.handle );
+
+		if (!dev_priv->ring.start || !dev_priv->ring.handle) {
+			DRM_ERROR( "Allocating dma descriptor ring failed\n");
+			return DRM_ERR(ENOMEM);
+		} else {
+			dev_priv->ring.start_addr = (u32) dev_priv->ring.handle;
+		}
+        } else {
+		dev_priv->ring.start = dev_priv->ring_map->handle;
+		dev_priv->ring.start_addr = (u32) dev_priv->ring_map->offset;
+	}
+
+	memset( dev_priv->ring.start, 0, dev_priv->ring.size );
+	DRM_INFO( "descriptor ring: cpu addr 0x%08x, bus addr: 0x%08x\n", 
+		  (u32) dev_priv->ring.start, dev_priv->ring.start_addr );
+
+	ret = 0;
+	if ( dev_priv->driver_mode != MACH64_MODE_MMIO ) {
+
+		/* enable block 1 registers and bus mastering */
+		MACH64_WRITE( MACH64_BUS_CNTL, 
+			      ( ( MACH64_READ(MACH64_BUS_CNTL) 
+				  | MACH64_BUS_EXT_REG_EN ) 
+				& ~MACH64_BUS_MASTER_DIS ) );
+
+		/* try a DMA GUI-mastering pass and fall back to MMIO if it fails */
+		DRM_DEBUG( "Starting DMA test...\n");
+		if ( (ret=mach64_bm_dma_test( dev )) ) {
+			dev_priv->driver_mode = MACH64_MODE_MMIO;
+		}
+	}
+
+	switch (dev_priv->driver_mode) {
+	case MACH64_MODE_MMIO:
+		MACH64_WRITE( MACH64_BUS_CNTL, ( MACH64_READ(MACH64_BUS_CNTL) 
+						 | MACH64_BUS_EXT_REG_EN
+						 | MACH64_BUS_MASTER_DIS ) );
+		if ( init->dma_mode == MACH64_MODE_MMIO )
+			DRM_INFO( "Forcing pseudo-DMA mode\n" );
+		else
+			DRM_INFO( "DMA test failed (ret=%d), using pseudo-DMA mode\n", ret );
+		break;
+	case MACH64_MODE_DMA_SYNC:
+		DRM_INFO( "DMA test succeeded, using synchronous DMA mode\n");
+		break;
+	case MACH64_MODE_DMA_ASYNC:
+	default:
+		DRM_INFO( "DMA test succeeded, using asynchronous DMA mode\n");
+	}
+
+	dev_priv->ring_running = 0;
+
+	/* setup offsets for physical address of table start and end */
+	dev_priv->ring.head_addr = dev_priv->ring.start_addr;
+	dev_priv->ring.head = dev_priv->ring.tail = 0;
+	dev_priv->ring.tail_mask = (dev_priv->ring.size / sizeof(u32)) - 1;
+	dev_priv->ring.space = dev_priv->ring.size;
+
+	/* setup physical address and size of descriptor table */
+	mach64_do_wait_for_fifo( dev_priv, 1 );
+	MACH64_WRITE( MACH64_BM_GUI_TABLE_CMD, 
+		      ( dev_priv->ring.head_addr | MACH64_CIRCULAR_BUF_SIZE_16KB ) );
+
+	/* init frame counter */
+	dev_priv->sarea_priv->frames_queued = 0;
+	for (i = 0; i < MACH64_MAX_QUEUED_FRAMES; i++) {
+		dev_priv->frame_ofs[i] = ~0; /* All ones indicates placeholder */
+	}
+
+	/* Allocate the DMA buffer freelist */
+	if ( (ret=mach64_init_freelist( dev )) ) {
+		DRM_ERROR("Freelist allocation failed\n");
+		mach64_do_cleanup_dma( dev );
+		return ret;
+	}
+
+	return 0;
+}
+
+/* ===================================================================
+ * MMIO Pseudo-DMA (intended primarily for debugging, not performance)
+ */
+
+int mach64_do_dispatch_pseudo_dma( drm_mach64_private_t *dev_priv )
+{
+	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;
+	volatile u32 *ring_read;
+	struct list_head *ptr;
+	drm_mach64_freelist_t *entry;
+	drm_buf_t *buf = NULL;
+	u32 *buf_ptr;
+	u32 used, reg, target;
+	int fifo, count, found, ret, no_idle_wait;
+
+	fifo = count = reg = no_idle_wait = 0;
+	target = MACH64_BM_ADDR;
+
+	if ( (ret=mach64_do_wait_for_idle( dev_priv )) < 0) {
+		DRM_INFO( "%s: idle failed before pseudo-dma dispatch, resetting engine\n", 
+			  __FUNCTION__);
+		mach64_dump_engine_info( dev_priv );
+		mach64_do_engine_reset( dev_priv );
+		return ret;
+	}
+
+	ring_read = (u32 *) ring->start;
+
+	while ( ring->tail != ring->head ) {
+		u32 buf_addr, new_target, offset; 
+		u32 bytes, remaining, head, eol;
+
+		head = ring->head;
+
+		new_target = le32_to_cpu( ring_read[head++] ) - MACH64_APERTURE_OFFSET;
+		buf_addr   = le32_to_cpu( ring_read[head++] );
+		eol        = le32_to_cpu( ring_read[head]   ) & MACH64_DMA_EOL;
+		bytes      = le32_to_cpu( ring_read[head++] )
+			& ~(MACH64_DMA_HOLD_OFFSET | MACH64_DMA_EOL);
+		head++;
+		head &= ring->tail_mask;
+
+		/* can't wait for idle between a blit setup descriptor 
+		 * and a HOSTDATA descriptor or the engine will lock
+		 */
+		if (new_target == MACH64_BM_HOSTDATA && target == MACH64_BM_ADDR)
+			no_idle_wait = 1;
+
+		target = new_target;
+
+		found = 0;
+		offset = 0;
+		list_for_each(ptr, &dev_priv->pending)
+		{
+			entry = list_entry(ptr, drm_mach64_freelist_t, list);
+			buf = entry->buf;
+			offset = buf_addr - GETBUFADDR( buf );
+			if (offset >= 0 && offset < MACH64_BUFFER_SIZE) {
+				found = 1;
+				break;
+			}
+		}
+
+		if (!found || buf == NULL) {
+			DRM_ERROR("Couldn't find pending buffer: head: %u tail: %u buf_addr: 0x%08x %s\n",
+				  head, ring->tail, buf_addr, (eol ? "eol" : ""));
+			mach64_dump_ring_info( dev_priv );
+			mach64_do_engine_reset( dev_priv );
+			return DRM_ERR(EINVAL);
+		}
+
+		/* Hand feed the buffer to the card via MMIO, waiting for the fifo 
+		 * every 16 writes 
+		 */
+		DRM_DEBUG("target: (0x%08x) %s\n", target, 
+				 (target == MACH64_BM_HOSTDATA ? "BM_HOSTDATA" : "BM_ADDR"));
+		DRM_DEBUG("offset: %u bytes: %u used: %u\n", offset, bytes, buf->used);
+
+		remaining = (buf->used - offset) >> 2; /* dwords remaining in buffer */
+		used = bytes >> 2; /* dwords in buffer for this descriptor */
+		buf_ptr = (u32 *)((char *)GETBUFPTR( buf ) + offset);
+
+		while ( used ) {
+			
+			if (count == 0) {
+				if (target == MACH64_BM_HOSTDATA) {
+					reg = DMAREG(MACH64_HOST_DATA0);
+					count = (remaining > 16) ? 16 : remaining;
+					fifo = 0;
+				} else {
+					reg = le32_to_cpu(*buf_ptr++);
+					used--;
+					count = (reg >> 16) + 1;
+				}
+
+				reg = reg & 0xffff;
+				reg = MMSELECT( reg );
+			}
+			while ( count && used ) {
+				if ( !fifo ) {
+					if (no_idle_wait) {
+						if ( (ret=mach64_do_wait_for_fifo( dev_priv, 16 )) < 0 ) {
+							no_idle_wait = 0;
+							return ret;
+						}
+					} else {
+						if ( (ret=mach64_do_wait_for_idle( dev_priv )) < 0 ) {
+							return ret;
+						}
+					}
+					fifo = 16;
+				}
+				--fifo;
+				MACH64_WRITE(reg, le32_to_cpu(*buf_ptr++));
+				used--; remaining--;
+					
+				reg += 4;
+				count--;
+			}
+		}
+		ring->head = head;
+		ring->head_addr = ring->start_addr + (ring->head * sizeof(u32));
+		ring->space += (4 * sizeof(u32));
+	}
+
+	if ( (ret=mach64_do_wait_for_idle( dev_priv )) < 0 ) {
+		return ret;
+	}
+	MACH64_WRITE( MACH64_BM_GUI_TABLE_CMD, 
+		      ring->head_addr | MACH64_CIRCULAR_BUF_SIZE_16KB );
+
+	DRM_DEBUG( "%s completed\n", __FUNCTION__ );
+	return 0;
+}
+
+/* ================================================================
+ * DMA cleanup
+ */
+
+int mach64_do_cleanup_dma( drm_device_t *dev )
+{
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+#if _HAVE_DMA_IRQ
+	/* Make sure interrupts are disabled here because the uninstall ioctl
+	 * may not have been called from userspace and after dev_private
+	 * is freed, it's too late.
+	 */
+	if ( dev->irq ) DRM(irq_uninstall)(dev);
+#endif
+
+	if ( dev->dev_private ) {
+		drm_mach64_private_t *dev_priv = dev->dev_private;
+
+		if ( dev_priv->is_pci ) {
+			if ( (dev_priv->ring.start != NULL) && dev_priv->ring.handle ) {
+				DRM(pci_free)( dev, dev_priv->ring.size, 
+					       dev_priv->ring.start, dev_priv->ring.handle );
+			}
+		} else {
+			if ( dev_priv->ring_map )
+				DRM_IOREMAPFREE( dev_priv->ring_map, dev );
+		}
+
+		if ( dev_priv->buffers )
+			DRM_IOREMAPFREE( dev_priv->buffers, dev );
+
+		mach64_destroy_freelist( dev );
+
+		DRM(free)( dev_priv, sizeof(drm_mach64_private_t),
+			   DRM_MEM_DRIVER );
+		dev->dev_private = NULL;
+	}
+
+	return 0;
+}
+
+/* ================================================================
+ * IOCTL handlers
+ */
+
+int mach64_dma_init( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_mach64_init_t init;
+		
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+	LOCK_TEST_WITH_RETURN( dev, filp );
+
+	DRM_COPY_FROM_USER_IOCTL( init, (drm_mach64_init_t *)data, 
+	    sizeof(init) );
+
+	switch ( init.func ) {
+	case DRM_MACH64_INIT_DMA:
+		return mach64_do_dma_init( dev, &init );
+	case DRM_MACH64_CLEANUP_DMA:
+		return mach64_do_cleanup_dma( dev );
+	}
+		
+	return DRM_ERR(EINVAL);
+}
+
+int mach64_dma_idle( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+	LOCK_TEST_WITH_RETURN( dev, filp );
+
+	return mach64_do_dma_idle( dev_priv );
+}
+
+int mach64_dma_flush( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+	LOCK_TEST_WITH_RETURN( dev, filp );
+
+	return mach64_do_dma_flush( dev_priv );
+}
+
+int mach64_engine_reset( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+	LOCK_TEST_WITH_RETURN( dev, filp );
+
+	return mach64_do_engine_reset( dev_priv );
+}
+
+
+/* ================================================================
+ * Freelist management
+ */
+
+int mach64_init_freelist( drm_device_t *dev )
+{
+	drm_device_dma_t *dma = dev->dma;
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	drm_mach64_freelist_t *entry;
+	struct list_head *ptr;
+	int i;
+
+	DRM_DEBUG("%s: adding %d buffers to freelist\n", __FUNCTION__, dma->buf_count);
+
+	for ( i = 0 ; i < dma->buf_count ; i++ ) {
+		if ((entry = 
+		     (drm_mach64_freelist_t *) DRM(alloc)(sizeof(drm_mach64_freelist_t), 
+							  DRM_MEM_BUFLISTS)) == NULL)
+			return DRM_ERR(ENOMEM);
+		memset( entry, 0, sizeof(drm_mach64_freelist_t) );
+		entry->buf = dma->buflist[i];
+		ptr = &entry->list;
+		list_add_tail(ptr, &dev_priv->free_list);
+	}
+
+	return 0;
+}
+
+void mach64_destroy_freelist( drm_device_t *dev )
+{
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	drm_mach64_freelist_t *entry;
+	struct list_head *ptr;
+	struct list_head *tmp;
+
+	DRM_DEBUG("%s\n", __FUNCTION__);
+
+	list_for_each_safe(ptr, tmp, &dev_priv->pending)
+	{
+		list_del(ptr);
+		entry = list_entry(ptr, drm_mach64_freelist_t, list);
+		DRM(free)(entry, sizeof(*entry), DRM_MEM_BUFLISTS);
+	}
+	list_for_each_safe(ptr, tmp, &dev_priv->placeholders)
+	{
+		list_del(ptr);
+		entry = list_entry(ptr, drm_mach64_freelist_t, list);
+		DRM(free)(entry, sizeof(*entry), DRM_MEM_BUFLISTS);
+	}
+
+	list_for_each_safe(ptr, tmp, &dev_priv->free_list)
+	{
+		list_del(ptr);
+		entry = list_entry(ptr, drm_mach64_freelist_t, list);
+		DRM(free)(entry, sizeof(*entry), DRM_MEM_BUFLISTS);
+	}
+}
+
+/* IMPORTANT: This function should only be called when the engine is idle or locked up, 
+ * as it assumes all buffers in the pending list have been completed by the hardware.
+ */
+int mach64_do_release_used_buffers( drm_mach64_private_t *dev_priv )
+{
+	struct list_head *ptr;
+	struct list_head *tmp;
+	drm_mach64_freelist_t *entry;
+	int i;
+
+	if ( list_empty(&dev_priv->pending) )
+		return 0;
+
+	/* Iterate the pending list and move all buffers into the freelist... */
+	i = 0;
+	list_for_each_safe(ptr, tmp, &dev_priv->pending)
+	{
+		entry = list_entry(ptr, drm_mach64_freelist_t, list);
+		if (entry->discard) {
+			entry->buf->pending = 0;
+			list_del(ptr);
+			list_add_tail(ptr, &dev_priv->free_list);
+			i++;
+		}
+	}
+
+	DRM_DEBUG( "%s: released %d buffers from pending list\n", __FUNCTION__, i );
+
+        return 0;
+}
+
+drm_buf_t *mach64_freelist_get( drm_mach64_private_t *dev_priv )
+{
+	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;
+	drm_mach64_freelist_t *entry;
+	struct list_head *ptr;
+	struct list_head *tmp;
+	int t;
+
+	if ( list_empty(&dev_priv->free_list) ) {
+		u32 head, tail, ofs;
+
+		if ( list_empty( &dev_priv->pending ) ) {
+			DRM_ERROR( "Couldn't get buffer - pending and free lists empty\n" );
+			t = 0;
+			list_for_each( ptr, &dev_priv->placeholders ) {
+				t++;
+			}
+			DRM_INFO( "Placeholders: %d\n", t );
+			return NULL;
+		}
+
+		tail = ring->tail;
+		for ( t = 0 ; t < dev_priv->usec_timeout ; t++ ) {
+			mach64_ring_tick( dev_priv, ring );
+			head = ring->head;
+
+			if ( head == tail ) {
+#if MACH64_EXTRA_CHECKING
+				if ( MACH64_READ(MACH64_GUI_STAT) & MACH64_GUI_ACTIVE ) {
+					DRM_ERROR( "Empty ring with non-idle engine!\n" );
+					mach64_dump_ring_info( dev_priv );
+					return NULL;
+				}
+#endif
+				/* last pass is complete, so release everything */
+				mach64_do_release_used_buffers( dev_priv );
+				DRM_DEBUG( "%s: idle engine, freed all buffers.\n", __FUNCTION__ );
+				if ( list_empty(&dev_priv->free_list) ) {
+					DRM_ERROR( "Freelist empty with idle engine\n" );
+					return NULL;
+				}
+				goto _freelist_entry_found;
+			}
+			/* Look for a completed buffer and bail out of the loop 
+			 * as soon as we find one -- don't waste time trying
+			 * to free extra bufs here, leave that to do_release_used_buffers
+			 */
+			list_for_each_safe(ptr, tmp, &dev_priv->pending) {
+				entry = list_entry(ptr, drm_mach64_freelist_t, list);
+				ofs = entry->ring_ofs;
+				if ( entry->discard &&
+				     ((head < tail && (ofs < head || ofs >= tail)) ||
+				      (head > tail && (ofs < head && ofs >= tail))) ) {
+#if MACH64_EXTRA_CHECKING
+					int i;
+					
+					for ( i = head ; i != tail ; i = (i + 4) & ring->tail_mask ) {
+						u32 o1 = le32_to_cpu(((u32 *)ring->start)[i + 1]);
+						u32 o2 = GETBUFADDR( entry->buf );
+
+						if ( o1 == o2 ) {
+							DRM_ERROR ( "Attempting to free used buffer: "
+								    "i=%d  buf=0x%08x\n", i, o1 );
+							mach64_dump_ring_info( dev_priv );
+							return NULL;
+						}
+					}
+#endif
+					/* found a processed buffer */
+					entry->buf->pending = 0;
+					list_del(ptr);
+					entry->buf->used = 0;
+					list_add_tail(ptr, &dev_priv->placeholders);
+					DRM_DEBUG( "%s: freed processed buffer (head=%d tail=%d "
+						   "buf ring ofs=%d).\n", __FUNCTION__, head, tail, ofs );
+					return entry->buf;
+				}
+			}
+			DRM_UDELAY( 1 );
+		}
+		mach64_dump_ring_info( dev_priv );
+		DRM_ERROR( "timeout waiting for buffers: ring head_addr: 0x%08x head: %d tail: %d\n", 
+			   ring->head_addr, ring->head, ring->tail );
+		return NULL;
+	}
+
+_freelist_entry_found:
+	ptr = dev_priv->free_list.next;
+	list_del(ptr);
+	entry = list_entry(ptr, drm_mach64_freelist_t, list);
+	entry->buf->used = 0;
+	list_add_tail(ptr, &dev_priv->placeholders);
+	return entry->buf;
+}
+
+/* ================================================================
+ * DMA buffer request and submission IOCTL handler
+ */
+
+static int mach64_dma_get_buffers( DRMFILE filp, drm_device_t *dev, drm_dma_t *d )
+{
+	int i;
+	drm_buf_t *buf;
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+
+	for ( i = d->granted_count ; i < d->request_count ; i++ ) {
+		buf = mach64_freelist_get( dev_priv );
+#if MACH64_EXTRA_CHECKING
+		if ( !buf ) return DRM_ERR(EFAULT);
+#else
+		if ( !buf ) return DRM_ERR(EAGAIN);
+#endif
+
+		buf->filp = filp;
+
+		if ( DRM_COPY_TO_USER( &d->request_indices[i], &buf->idx,
+				   sizeof(buf->idx) ) )
+			return DRM_ERR(EFAULT);
+		if ( DRM_COPY_TO_USER( &d->request_sizes[i], &buf->total,
+				   sizeof(buf->total) ) )
+			return DRM_ERR(EFAULT);
+
+		d->granted_count++;
+	}
+	return 0;
+}
+
+int mach64_dma_buffers( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_device_dma_t *dma = dev->dma;
+	drm_dma_t d;
+        int ret = 0;
+
+        LOCK_TEST_WITH_RETURN( dev, filp );
+
+	DRM_COPY_FROM_USER_IOCTL( d, (drm_dma_t *)data, sizeof(d) );
+
+        /* Please don't send us buffers.
+	 */
+        if ( d.send_count != 0 ) 
+        {
+		DRM_ERROR( "Process %d trying to send %d buffers via drmDMA\n",
+			   DRM_CURRENTPID, d.send_count );
+		return DRM_ERR(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",
+			   DRM_CURRENTPID, d.request_count, dma->buf_count );
+		ret = DRM_ERR(EINVAL);
+	}
+        
+	d.granted_count = 0;
+
+	if ( d.request_count ) 
+	{
+		ret = mach64_dma_get_buffers( filp, dev, &d );
+	}
+
+	DRM_COPY_TO_USER_IOCTL( (drm_dma_t *)data, d, sizeof(d) );
+
+        return ret;
+}
+
diff --git a/shared/mach64_drm.h b/shared/mach64_drm.h
new file mode 100644
index 00000000..9d748a37
--- /dev/null
+++ b/shared/mach64_drm.h
@@ -0,0 +1,258 @@
+/* mach64_drm.h -- Public header for the mach64 driver -*- linux-c -*-
+ * Created: Thu Nov 30 20:04:32 2000 by gareth@valinux.com
+ *
+ * Copyright 2000 Gareth Hughes
+ * Copyright 2002 Frank C. Earl
+ * Copyright 2002-2003 Leif Delgass
+ * 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
+ * THE COPYRIGHT OWNER(S) 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:
+ *    Gareth Hughes <gareth@valinux.com>
+ *    Frank C. Earl <fearl@airmail.net>
+ *    Leif Delgass <ldelgass@retinalburn.net>
+ */
+
+#ifndef __MACH64_DRM_H__
+#define __MACH64_DRM_H__
+
+
+/* WARNING: If you change any of these defines, make sure to change the
+ * defines in the Xserver file (mach64_sarea.h)
+ */
+#ifndef __MACH64_SAREA_DEFINES__
+#define __MACH64_SAREA_DEFINES__
+
+/* What needs to be changed for the current vertex buffer?
+ * GH: We're going to be pedantic about this.  We want the card to do as
+ * little as possible, so let's avoid having it fetch a whole bunch of
+ * register values that don't change all that often, if at all.
+ */
+#define MACH64_UPLOAD_DST_OFF_PITCH	0x0001
+#define MACH64_UPLOAD_Z_OFF_PITCH	0x0002
+#define MACH64_UPLOAD_Z_ALPHA_CNTL	0x0004
+#define MACH64_UPLOAD_SCALE_3D_CNTL	0x0008
+#define MACH64_UPLOAD_DP_FOG_CLR	0x0010
+#define MACH64_UPLOAD_DP_WRITE_MASK	0x0020
+#define MACH64_UPLOAD_DP_PIX_WIDTH	0x0040
+#define MACH64_UPLOAD_SETUP_CNTL	0x0080
+#define MACH64_UPLOAD_MISC		0x0100
+#define MACH64_UPLOAD_TEXTURE		0x0200
+#define MACH64_UPLOAD_TEX0IMAGE		0x0400
+#define MACH64_UPLOAD_TEX1IMAGE		0x0800
+#define MACH64_UPLOAD_CLIPRECTS		0x1000 /* handled client-side */
+#define MACH64_UPLOAD_CONTEXT		0x00ff
+#define MACH64_UPLOAD_ALL		0x1fff
+
+/* DMA buffer size
+ */
+#define MACH64_BUFFER_SIZE		16384
+
+/* Max number of swaps allowed on the ring
+ * before the client must wait
+ */
+#define MACH64_MAX_QUEUED_FRAMES        3
+
+/* Byte offsets for host blit buffer data
+ */
+#define MACH64_HOSTDATA_BLIT_OFFSET	104
+
+/* Keep these small for testing.
+ */
+#define MACH64_NR_SAREA_CLIPRECTS	8
+
+
+#define MACH64_CARD_HEAP		0
+#define MACH64_AGP_HEAP			1
+#define MACH64_NR_TEX_HEAPS		2
+#define MACH64_NR_TEX_REGIONS		64
+#define MACH64_LOG_TEX_GRANULARITY	16
+
+#define MACH64_TEX_MAXLEVELS		1
+
+#define MACH64_NR_CONTEXT_REGS		15
+#define MACH64_NR_TEXTURE_REGS		4
+
+#endif /* __MACH64_SAREA_DEFINES__ */
+
+typedef struct {
+	unsigned int dst_off_pitch;
+
+	unsigned int z_off_pitch;
+	unsigned int z_cntl;
+	unsigned int alpha_tst_cntl;
+
+	unsigned int scale_3d_cntl;
+
+	unsigned int sc_left_right;
+	unsigned int sc_top_bottom;
+
+	unsigned int dp_fog_clr;
+	unsigned int dp_write_mask;
+	unsigned int dp_pix_width;
+	unsigned int dp_mix;
+	unsigned int dp_src;
+
+	unsigned int clr_cmp_cntl;
+	unsigned int gui_traj_cntl;
+
+	unsigned int setup_cntl;
+
+	unsigned int tex_size_pitch;
+	unsigned int tex_cntl;
+	unsigned int secondary_tex_off;
+	unsigned int tex_offset;
+} drm_mach64_context_regs_t;
+
+typedef struct drm_mach64_sarea {
+	/* The channel for communication of state information to the kernel
+	 * on firing a vertex dma buffer.
+	 */
+	drm_mach64_context_regs_t context_state;
+	unsigned int dirty;
+	unsigned int vertsize;
+
+	/* The current cliprects, or a subset thereof.
+	 */
+	drm_clip_rect_t boxes[MACH64_NR_SAREA_CLIPRECTS];
+	unsigned int nbox;
+
+	/* Counters for client-side throttling of rendering clients.
+	 */
+	unsigned int frames_queued;
+
+	/* Texture memory LRU.
+	 */
+	drm_tex_region_t tex_list[MACH64_NR_TEX_HEAPS][MACH64_NR_TEX_REGIONS+1];
+	unsigned int tex_age[MACH64_NR_TEX_HEAPS];
+	int ctx_owner;
+} drm_mach64_sarea_t;
+
+
+/* WARNING: If you change any of these defines, make sure to change the
+ * defines in the Xserver file (mach64_common.h)
+ */
+
+/* Mach64 specific ioctls 
+ * The device specific ioctl range is 0x40 to 0x79.
+ */
+
+#define DRM_MACH64_INIT           0x00
+#define DRM_MACH64_IDLE           0x01
+#define DRM_MACH64_RESET          0x02
+#define DRM_MACH64_SWAP           0x03
+#define DRM_MACH64_CLEAR          0x04
+#define DRM_MACH64_VERTEX         0x05
+#define DRM_MACH64_BLIT           0x06
+#define DRM_MACH64_FLUSH          0x07
+#define DRM_MACH64_GETPARAM       0x08
+
+#define DRM_IOCTL_MACH64_INIT           DRM_IOW( DRM_COMMAND_BASE + DRM_MACH64_INIT, drm_mach64_init_t)
+#define DRM_IOCTL_MACH64_IDLE           DRM_IO(  DRM_COMMAND_BASE + DRM_MACH64_IDLE )
+#define DRM_IOCTL_MACH64_RESET          DRM_IO(  DRM_COMMAND_BASE + DRM_MACH64_RESET )
+#define DRM_IOCTL_MACH64_SWAP           DRM_IO(  DRM_COMMAND_BASE + DRM_MACH64_SWAP )
+#define DRM_IOCTL_MACH64_CLEAR          DRM_IOW( DRM_COMMAND_BASE + DRM_MACH64_CLEAR, drm_mach64_clear_t)
+#define DRM_IOCTL_MACH64_VERTEX         DRM_IOW( DRM_COMMAND_BASE + DRM_MACH64_VERTEX, drm_mach64_vertex_t)
+#define DRM_IOCTL_MACH64_BLIT           DRM_IOW( DRM_COMMAND_BASE + DRM_MACH64_BLIT, drm_mach64_blit_t)
+#define DRM_IOCTL_MACH64_FLUSH          DRM_IO(  DRM_COMMAND_BASE + DRM_MACH64_FLUSH )
+#define DRM_IOCTL_MACH64_GETPARAM       DRM_IOWR( DRM_COMMAND_BASE + DRM_MACH64_GETPARAM, drm_mach64_getparam_t)
+
+/* Buffer flags for clears
+ */
+#define MACH64_FRONT			0x1
+#define MACH64_BACK			0x2
+#define MACH64_DEPTH			0x4
+
+/* Primitive types for vertex buffers
+ */
+#define MACH64_PRIM_POINTS		0x00000000
+#define MACH64_PRIM_LINES		0x00000001
+#define MACH64_PRIM_LINE_LOOP		0x00000002
+#define MACH64_PRIM_LINE_STRIP		0x00000003
+#define MACH64_PRIM_TRIANGLES		0x00000004
+#define MACH64_PRIM_TRIANGLE_STRIP	0x00000005
+#define MACH64_PRIM_TRIANGLE_FAN	0x00000006
+#define MACH64_PRIM_QUADS		0x00000007
+#define MACH64_PRIM_QUAD_STRIP		0x00000008
+#define MACH64_PRIM_POLYGON		0x00000009
+
+
+typedef enum _drm_mach64_dma_mode_t {
+   MACH64_MODE_DMA_ASYNC,
+   MACH64_MODE_DMA_SYNC,
+   MACH64_MODE_MMIO
+} drm_mach64_dma_mode_t;
+
+typedef struct drm_mach64_init {
+	enum {
+		DRM_MACH64_INIT_DMA = 0x01,
+		DRM_MACH64_CLEANUP_DMA = 0x02
+	} func;
+
+	unsigned long sarea_priv_offset;
+	int is_pci;
+	drm_mach64_dma_mode_t dma_mode;
+
+	unsigned int fb_bpp;
+	unsigned int front_offset, front_pitch;
+	unsigned int back_offset, back_pitch;
+
+	unsigned int depth_bpp;
+	unsigned int depth_offset, depth_pitch;
+
+	unsigned long fb_offset;
+	unsigned long mmio_offset;
+	unsigned long ring_offset;
+	unsigned long buffers_offset;
+	unsigned long agp_textures_offset;
+} drm_mach64_init_t;
+
+typedef struct drm_mach64_clear {
+	unsigned int flags;
+	int x, y, w, h;
+	unsigned int clear_color;
+	unsigned int clear_depth;
+} drm_mach64_clear_t;
+
+typedef struct drm_mach64_vertex {
+	int prim;
+	void *buf;			/* Address of vertex buffer */
+	unsigned long used;		/* Number of bytes in buffer */
+	int discard;			/* Client finished with buffer? */
+} drm_mach64_vertex_t;
+
+typedef struct drm_mach64_blit {
+	int idx;
+	int pitch;
+	int offset;
+	int format;
+	unsigned short x, y;
+	unsigned short width, height;
+} drm_mach64_blit_t;
+
+typedef struct drm_mach64_getparam {
+	enum {
+		MACH64_PARAM_FRAMES_QUEUED = 0x01,
+		MACH64_PARAM_IRQ_NR = 0x02
+	} param;
+	void *value;
+} drm_mach64_getparam_t;
+
+#endif
diff --git a/shared/mach64_drv.h b/shared/mach64_drv.h
new file mode 100644
index 00000000..4331cde5
--- /dev/null
+++ b/shared/mach64_drv.h
@@ -0,0 +1,1034 @@
+/* mach64_drv.h -- Private header for mach64 driver -*- linux-c -*-
+ * Created: Fri Nov 24 22:07:58 2000 by gareth@valinux.com
+ *
+ * Copyright 2000 Gareth Hughes
+ * Copyright 2002 Frank C. Earl
+ * Copyright 2002-2003 Leif Delgass
+ * 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
+ * THE COPYRIGHT OWNER(S) 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:
+ *    Gareth Hughes <gareth@valinux.com>
+ *    Frank C. Earl <fearl@airmail.net>
+ *    Leif Delgass <ldelgass@retinalburn.net>
+ *    José Fonseca <j_r_fonseca@yahoo.co.uk>
+ */
+
+#ifndef __MACH64_DRV_H__
+#define __MACH64_DRV_H__
+
+
+/* FIXME: remove these when not needed */
+/* Development driver options */
+#define MACH64_EXTRA_CHECKING     0 /* Extra sanity checks for DMA/freelist management */
+#define MACH64_VERBOSE		  0 /* Verbose debugging output */
+
+typedef struct drm_mach64_freelist {
+	struct list_head  list;  /* List pointers for free_list, placeholders, or pending list */
+   	drm_buf_t *buf;          /* Pointer to the buffer */
+	int discard;             /* This flag is set when we're done (re)using a buffer */
+	u32 ring_ofs;            /* dword offset in ring of last descriptor for this buffer */
+} drm_mach64_freelist_t;
+
+typedef struct drm_mach64_descriptor_ring {
+	dma_addr_t handle;       /* handle (bus address) of ring returned by pci_alloc_consistent() */
+	void *start;             /* write pointer (cpu address) to start of descriptor ring */
+	u32 start_addr;          /* bus address of beginning of descriptor ring */
+       	int size;                /* size of ring in bytes */
+
+	u32 head_addr;           /* bus address of descriptor ring head */
+	u32 head;                /* dword offset of descriptor ring head */
+	u32 tail;                /* dword offset of descriptor ring tail */
+	u32 tail_mask;           /* mask used to wrap ring */
+	int space;               /* number of free bytes in ring */
+} drm_mach64_descriptor_ring_t;
+
+typedef struct drm_mach64_private {
+	drm_mach64_sarea_t *sarea_priv;
+
+	int is_pci;
+	drm_mach64_dma_mode_t driver_mode;       /* Async DMA, sync DMA, or MMIO */
+
+	int usec_timeout;                        /* Timeout for the wait functions */
+
+	drm_mach64_descriptor_ring_t ring;       /* DMA descriptor table (ring buffer) */
+	int ring_running;                        /* Is bus mastering is enabled */
+
+	struct list_head free_list;              /* Free-list head */
+	struct list_head placeholders;           /* Placeholder list for buffers held by clients */
+	struct list_head pending;                /* Buffers pending completion */
+
+	u32 frame_ofs[MACH64_MAX_QUEUED_FRAMES]; /* dword ring offsets of most recent frame swaps */
+
+	unsigned int fb_bpp;
+	unsigned int front_offset, front_pitch;
+	unsigned int back_offset, back_pitch;
+
+	unsigned int depth_bpp;
+	unsigned int depth_offset, depth_pitch;
+
+	u32 front_offset_pitch;
+	u32 back_offset_pitch;
+	u32 depth_offset_pitch;
+
+	drm_local_map_t *sarea;
+	drm_local_map_t *fb;
+	drm_local_map_t *mmio;
+	drm_local_map_t *ring_map;
+	drm_local_map_t *buffers;
+	drm_local_map_t *agp_textures;
+} drm_mach64_private_t;
+
+				/* mach64_dma.c */
+extern int mach64_dma_init( DRM_IOCTL_ARGS );
+extern int mach64_dma_idle( DRM_IOCTL_ARGS );
+extern int mach64_dma_flush( DRM_IOCTL_ARGS );
+extern int mach64_engine_reset( DRM_IOCTL_ARGS );
+extern int mach64_dma_buffers( DRM_IOCTL_ARGS );
+
+extern int mach64_init_freelist( drm_device_t *dev );
+extern void mach64_destroy_freelist( drm_device_t *dev );
+extern drm_buf_t *mach64_freelist_get( drm_mach64_private_t *dev_priv );
+
+extern int mach64_do_wait_for_fifo( drm_mach64_private_t *dev_priv,
+				    int entries );
+extern int mach64_do_wait_for_idle( drm_mach64_private_t *dev_priv );
+extern int mach64_wait_ring( drm_mach64_private_t *dev_priv, int n );
+extern int mach64_do_dispatch_pseudo_dma( drm_mach64_private_t *dev_priv );
+extern int mach64_do_release_used_buffers( drm_mach64_private_t *dev_priv );
+extern void mach64_dump_engine_info( drm_mach64_private_t *dev_priv );
+extern void mach64_dump_ring_info( drm_mach64_private_t *dev_priv );
+extern int mach64_do_engine_reset( drm_mach64_private_t *dev_priv );
+
+extern int mach64_do_dma_idle( drm_mach64_private_t *dev_priv );
+extern int mach64_do_dma_flush( drm_mach64_private_t *dev_priv );
+extern int mach64_do_cleanup_dma( drm_device_t *dev );
+
+				/* mach64_state.c */
+extern int mach64_dma_clear( DRM_IOCTL_ARGS );
+extern int mach64_dma_swap( DRM_IOCTL_ARGS );
+extern int mach64_dma_vertex( DRM_IOCTL_ARGS );
+extern int mach64_dma_blit( DRM_IOCTL_ARGS );
+extern int mach64_get_param( DRM_IOCTL_ARGS );
+
+/* ================================================================
+ * Registers
+ */
+
+#define MACH64_AGP_BASE				0x0148
+#define MACH64_AGP_CNTL				0x014c
+#define MACH64_ALPHA_TST_CNTL			0x0550
+
+
+#define MACH64_DSP_CONFIG 			0x0420
+#define MACH64_DSP_ON_OFF 			0x0424
+#define MACH64_EXT_MEM_CNTL 			0x04ac
+#define MACH64_GEN_TEST_CNTL 			0x04d0
+#define MACH64_HW_DEBUG 			0x047c
+#define MACH64_MEM_ADDR_CONFIG 			0x0434
+#define MACH64_MEM_BUF_CNTL 			0x042c
+#define MACH64_MEM_CNTL 			0x04b0
+
+
+#define MACH64_BM_ADDR				0x0648
+#define MACH64_BM_COMMAND			0x0188
+#define MACH64_BM_DATA				0x0648
+#define MACH64_BM_FRAME_BUF_OFFSET		0x0180
+#define MACH64_BM_GUI_TABLE			0x01b8
+#define MACH64_BM_GUI_TABLE_CMD			0x064c
+#	define MACH64_CIRCULAR_BUF_SIZE_16KB		(0 << 0)
+#	define MACH64_CIRCULAR_BUF_SIZE_32KB		(1 << 0)
+#	define MACH64_CIRCULAR_BUF_SIZE_64KB		(2 << 0)
+#	define MACH64_CIRCULAR_BUF_SIZE_128KB		(3 << 0)
+#	define MACH64_LAST_DESCRIPTOR			(1 << 31)
+#define MACH64_BM_HOSTDATA			0x0644
+#define MACH64_BM_STATUS			0x018c
+#define MACH64_BM_SYSTEM_MEM_ADDR		0x0184
+#define MACH64_BM_SYSTEM_TABLE			0x01bc
+#define MACH64_BUS_CNTL				0x04a0
+#	define MACH64_BUS_MSTR_RESET			(1 << 1)
+#	define MACH64_BUS_APER_REG_DIS			(1 << 4)
+#	define MACH64_BUS_FLUSH_BUF			(1 << 2)
+#	define MACH64_BUS_MASTER_DIS			(1 << 6)
+#	define MACH64_BUS_EXT_REG_EN			(1 << 27)
+
+#define MACH64_CLR_CMP_CLR			0x0700
+#define MACH64_CLR_CMP_CNTL			0x0708
+#define MACH64_CLR_CMP_MASK			0x0704
+#define MACH64_CONFIG_CHIP_ID 			0x04e0
+#define MACH64_CONFIG_CNTL 			0x04dc
+#define MACH64_CONFIG_STAT0 			0x04e4
+#define MACH64_CONFIG_STAT1 			0x0494
+#define MACH64_CONFIG_STAT2 			0x0498
+#define MACH64_CONTEXT_LOAD_CNTL		0x072c
+#define MACH64_CONTEXT_MASK			0x0720
+#define MACH64_COMPOSITE_SHADOW_ID		0x0798
+#define MACH64_CRC_SIG 				0x04e8
+#define MACH64_CUSTOM_MACRO_CNTL 		0x04d4
+
+#define MACH64_DP_BKGD_CLR			0x06c0
+#define MACH64_DP_FOG_CLR			0x06c4
+#define MACH64_DP_FGRD_BKGD_CLR			0x06e0
+#define MACH64_DP_FRGD_CLR			0x06c4
+#define MACH64_DP_FGRD_CLR_MIX			0x06dc
+
+#define MACH64_DP_MIX				0x06d4
+#	define BKGD_MIX_NOT_D				(0 << 0)
+#	define BKGD_MIX_ZERO				(1 << 0)
+#	define BKGD_MIX_ONE				(2 << 0)
+#	define MACH64_BKGD_MIX_D			(3 << 0)
+#	define BKGD_MIX_NOT_S				(4 << 0)
+#	define BKGD_MIX_D_XOR_S				(5 << 0)
+#	define BKGD_MIX_NOT_D_XOR_S			(6 << 0)
+#	define MACH64_BKGD_MIX_S			(7 << 0)
+#	define BKGD_MIX_NOT_D_OR_NOT_S			(8 << 0)
+#	define BKGD_MIX_D_OR_NOT_S			(9 << 0)
+#	define BKGD_MIX_NOT_D_OR_S			(10 << 0)
+#	define BKGD_MIX_D_OR_S				(11 << 0)
+#	define BKGD_MIX_D_AND_S				(12 << 0)
+#	define BKGD_MIX_NOT_D_AND_S			(13 << 0)
+#	define BKGD_MIX_D_AND_NOT_S			(14 << 0)
+#	define BKGD_MIX_NOT_D_AND_NOT_S			(15 << 0)
+#	define BKGD_MIX_D_PLUS_S_DIV2			(23 << 0)
+#	define FRGD_MIX_NOT_D				(0 << 16)
+#	define FRGD_MIX_ZERO				(1 << 16)
+#	define FRGD_MIX_ONE				(2 << 16)
+#	define FRGD_MIX_D				(3 << 16)
+#	define FRGD_MIX_NOT_S				(4 << 16)
+#	define FRGD_MIX_D_XOR_S				(5 << 16)
+#	define FRGD_MIX_NOT_D_XOR_S			(6 << 16)
+#	define MACH64_FRGD_MIX_S			(7 << 16)
+#	define FRGD_MIX_NOT_D_OR_NOT_S			(8 << 16)
+#	define FRGD_MIX_D_OR_NOT_S			(9 << 16)
+#	define FRGD_MIX_NOT_D_OR_S			(10 << 16)
+#	define FRGD_MIX_D_OR_S				(11 << 16)
+#	define FRGD_MIX_D_AND_S				(12 << 16)
+#	define FRGD_MIX_NOT_D_AND_S			(13 << 16)
+#	define FRGD_MIX_D_AND_NOT_S			(14 << 16)
+#	define FRGD_MIX_NOT_D_AND_NOT_S			(15 << 16)
+#	define FRGD_MIX_D_PLUS_S_DIV2			(23 << 16)
+
+#define MACH64_DP_PIX_WIDTH			0x06d0
+#	define MACH64_HOST_TRIPLE_ENABLE		(1 << 13)
+#	define MACH64_BYTE_ORDER_MSB_TO_LSB		(0 << 24)
+#	define MACH64_BYTE_ORDER_LSB_TO_MSB		(1 << 24)
+
+#define MACH64_DP_SRC				0x06d8
+#	define MACH64_BKGD_SRC_BKGD_CLR			(0 << 0)
+#	define MACH64_BKGD_SRC_FRGD_CLR			(1 << 0)
+#	define MACH64_BKGD_SRC_HOST			(2 << 0)
+#	define MACH64_BKGD_SRC_BLIT			(3 << 0)
+#	define MACH64_BKGD_SRC_PATTERN			(4 << 0)
+#	define MACH64_BKGD_SRC_3D			(5 << 0)
+#	define MACH64_FRGD_SRC_BKGD_CLR			(0 << 8)
+#	define MACH64_FRGD_SRC_FRGD_CLR			(1 << 8)
+#	define MACH64_FRGD_SRC_HOST			(2 << 8)
+#	define MACH64_FRGD_SRC_BLIT			(3 << 8)
+#	define MACH64_FRGD_SRC_PATTERN			(4 << 8)
+#	define MACH64_FRGD_SRC_3D			(5 << 8)
+#	define MACH64_MONO_SRC_ONE			(0 << 16)
+#	define MACH64_MONO_SRC_PATTERN			(1 << 16)
+#	define MACH64_MONO_SRC_HOST			(2 << 16)
+#	define MACH64_MONO_SRC_BLIT			(3 << 16)
+
+#define MACH64_DP_WRITE_MASK			0x06c8
+
+#define MACH64_DST_CNTL				0x0530
+#	define MACH64_DST_X_RIGHT_TO_LEFT		(0 << 0)
+#	define MACH64_DST_X_LEFT_TO_RIGHT		(1 << 0)
+#	define MACH64_DST_Y_BOTTOM_TO_TOP		(0 << 1)
+#	define MACH64_DST_Y_TOP_TO_BOTTOM		(1 << 1)
+#	define MACH64_DST_X_MAJOR			(0 << 2)
+#	define MACH64_DST_Y_MAJOR			(1 << 2)
+#	define MACH64_DST_X_TILE			(1 << 3)
+#	define MACH64_DST_Y_TILE			(1 << 4)
+#	define MACH64_DST_LAST_PEL			(1 << 5)
+#	define MACH64_DST_POLYGON_ENABLE		(1 << 6)
+#	define MACH64_DST_24_ROTATION_ENABLE		(1 << 7)
+
+#define MACH64_DST_HEIGHT_WIDTH			0x0518
+#define MACH64_DST_OFF_PITCH			0x0500
+#define MACH64_DST_WIDTH_HEIGHT			0x06ec
+#define MACH64_DST_X_Y				0x06e8
+#define MACH64_DST_Y_X				0x050c
+
+#define MACH64_FIFO_STAT			0x0710
+#	define MACH64_FIFO_SLOT_MASK			0x0000ffff
+#	define MACH64_FIFO_ERR				(1 << 31)
+
+#define MACH64_GEN_TEST_CNTL			0x04d0
+#	define MACH64_GUI_ENGINE_ENABLE			(1 << 8)
+#define MACH64_GUI_CMDFIFO_DEBUG		0x0170
+#define MACH64_GUI_CMDFIFO_DATA			0x0174
+#define MACH64_GUI_CNTL				0x0178
+#       define MACH64_CMDFIFO_SIZE_MASK                 0x00000003ul
+#       define MACH64_CMDFIFO_SIZE_192                  0x00000000ul
+#       define MACH64_CMDFIFO_SIZE_128                  0x00000001ul
+#       define MACH64_CMDFIFO_SIZE_64                   0x00000002ul
+#define MACH64_GUI_STAT				0x0738
+#	define MACH64_GUI_ACTIVE			(1 << 0)
+#define MACH64_GUI_TRAJ_CNTL			0x0730
+
+#define MACH64_HOST_CNTL			0x0640
+#define MACH64_HOST_DATA0			0x0600
+
+#define MACH64_ONE_OVER_AREA			0x029c
+#define MACH64_ONE_OVER_AREA_UC			0x0300
+
+#define MACH64_PAT_REG0				0x0680
+#define MACH64_PAT_REG1				0x0684
+
+#define MACH64_SC_LEFT                          0x06a0
+#define MACH64_SC_RIGHT                         0x06a4
+#define MACH64_SC_LEFT_RIGHT                    0x06a8
+#define MACH64_SC_TOP                           0x06ac
+#define MACH64_SC_BOTTOM                        0x06b0
+#define MACH64_SC_TOP_BOTTOM                    0x06b4
+
+#define MACH64_SCALE_3D_CNTL			0x05fc
+#define MACH64_SCRATCH_REG0			0x0480
+#define MACH64_SCRATCH_REG1			0x0484
+#define MACH64_SECONDARY_TEX_OFF		0x0778
+#define MACH64_SETUP_CNTL			0x0304
+#define MACH64_SRC_CNTL				0x05b4
+#	define MACH64_SRC_BM_ENABLE			(1 << 8)
+#	define MACH64_SRC_BM_SYNC			(1 << 9)
+#	define MACH64_SRC_BM_OP_FRAME_TO_SYSTEM		(0 << 10)
+#	define MACH64_SRC_BM_OP_SYSTEM_TO_FRAME		(1 << 10)
+#	define MACH64_SRC_BM_OP_REG_TO_SYSTEM		(2 << 10)
+#	define MACH64_SRC_BM_OP_SYSTEM_TO_REG		(3 << 10)
+#define MACH64_SRC_HEIGHT1			0x0594
+#define MACH64_SRC_HEIGHT2			0x05ac
+#define MACH64_SRC_HEIGHT1_WIDTH1		0x0598
+#define MACH64_SRC_HEIGHT2_WIDTH2		0x05b0
+#define MACH64_SRC_OFF_PITCH			0x0580
+#define MACH64_SRC_WIDTH1			0x0590
+#define MACH64_SRC_Y_X				0x058c
+
+#define MACH64_TEX_0_OFF			0x05c0
+#define MACH64_TEX_CNTL				0x0774
+#define MACH64_TEX_SIZE_PITCH			0x0770
+#define MACH64_TIMER_CONFIG 			0x0428
+
+#define MACH64_VERTEX_1_ARGB			0x0254
+#define MACH64_VERTEX_1_S			0x0240
+#define MACH64_VERTEX_1_SECONDARY_S		0x0328
+#define MACH64_VERTEX_1_SECONDARY_T		0x032c
+#define MACH64_VERTEX_1_SECONDARY_W		0x0330
+#define MACH64_VERTEX_1_SPEC_ARGB		0x024c
+#define MACH64_VERTEX_1_T			0x0244
+#define MACH64_VERTEX_1_W			0x0248
+#define MACH64_VERTEX_1_X_Y			0x0258
+#define MACH64_VERTEX_1_Z			0x0250
+#define MACH64_VERTEX_2_ARGB			0x0274
+#define MACH64_VERTEX_2_S			0x0260
+#define MACH64_VERTEX_2_SECONDARY_S		0x0334
+#define MACH64_VERTEX_2_SECONDARY_T		0x0338
+#define MACH64_VERTEX_2_SECONDARY_W		0x033c
+#define MACH64_VERTEX_2_SPEC_ARGB		0x026c
+#define MACH64_VERTEX_2_T			0x0264
+#define MACH64_VERTEX_2_W			0x0268
+#define MACH64_VERTEX_2_X_Y			0x0278
+#define MACH64_VERTEX_2_Z			0x0270
+#define MACH64_VERTEX_3_ARGB			0x0294
+#define MACH64_VERTEX_3_S			0x0280
+#define MACH64_VERTEX_3_SECONDARY_S		0x02a0
+#define MACH64_VERTEX_3_SECONDARY_T		0x02a4
+#define MACH64_VERTEX_3_SECONDARY_W		0x02a8
+#define MACH64_VERTEX_3_SPEC_ARGB		0x028c
+#define MACH64_VERTEX_3_T			0x0284
+#define MACH64_VERTEX_3_W			0x0288
+#define MACH64_VERTEX_3_X_Y			0x0298
+#define MACH64_VERTEX_3_Z			0x0290
+
+#define MACH64_Z_CNTL				0x054c
+#define MACH64_Z_OFF_PITCH			0x0548
+
+#define MACH64_CRTC_VLINE_CRNT_VLINE		0x0410
+#	define MACH64_CRTC_VLINE_MASK		        0x000007ff
+#	define MACH64_CRTC_CRNT_VLINE_MASK		0x07ff0000
+#define MACH64_CRTC_OFF_PITCH			0x0414
+#define MACH64_CRTC_INT_CNTL			0x0418
+#	define MACH64_CRTC_VBLANK			(1 << 0)
+#	define MACH64_CRTC_VBLANK_INT_EN		(1 << 1)
+#	define MACH64_CRTC_VBLANK_INT			(1 << 2)
+#	define MACH64_CRTC_VLINE_INT_EN			(1 << 3)
+#	define MACH64_CRTC_VLINE_INT			(1 << 4)
+#	define MACH64_CRTC_VLINE_SYNC			(1 << 5) /* 0=even, 1=odd */
+#	define MACH64_CRTC_FRAME			(1 << 6) /* 0=even, 1=odd */
+#	define MACH64_CRTC_SNAPSHOT_INT_EN		(1 << 7)
+#	define MACH64_CRTC_SNAPSHOT_INT			(1 << 8)
+#	define MACH64_CRTC_I2C_INT_EN			(1 << 9)
+#	define MACH64_CRTC_I2C_INT			(1 << 10)
+#	define MACH64_CRTC2_VBLANK			(1 << 11) /* LT Pro */
+#	define MACH64_CRTC2_VBLANK_INT_EN		(1 << 12) /* LT Pro */
+#	define MACH64_CRTC2_VBLANK_INT			(1 << 13) /* LT Pro */
+#	define MACH64_CRTC2_VLINE_INT_EN		(1 << 14) /* LT Pro */
+#	define MACH64_CRTC2_VLINE_INT			(1 << 15) /* LT Pro */
+#	define MACH64_CRTC_CAPBUF0_INT_EN		(1 << 16)
+#	define MACH64_CRTC_CAPBUF0_INT			(1 << 17)
+#	define MACH64_CRTC_CAPBUF1_INT_EN		(1 << 18)
+#	define MACH64_CRTC_CAPBUF1_INT			(1 << 19)
+#	define MACH64_CRTC_OVERLAY_EOF_INT_EN		(1 << 20)
+#	define MACH64_CRTC_OVERLAY_EOF_INT		(1 << 21)
+#	define MACH64_CRTC_ONESHOT_CAP_INT_EN		(1 << 22)
+#	define MACH64_CRTC_ONESHOT_CAP_INT		(1 << 23)
+#	define MACH64_CRTC_BUSMASTER_EOL_INT_EN		(1 << 24)
+#	define MACH64_CRTC_BUSMASTER_EOL_INT		(1 << 25)
+#	define MACH64_CRTC_GP_INT_EN			(1 << 26)
+#	define MACH64_CRTC_GP_INT			(1 << 27)
+#	define MACH64_CRTC2_VLINE_SYNC			(1 << 28) /* LT Pro */  /* 0=even, 1=odd */
+#	define MACH64_CRTC_SNAPSHOT2_INT_EN		(1 << 29) /* LT Pro */
+#	define MACH64_CRTC_SNAPSHOT2_INT		(1 << 30) /* LT Pro */
+#	define MACH64_CRTC_VBLANK2_INT			(1 << 31)
+#	define MACH64_CRTC_INT_ENS				\
+		(						\
+			MACH64_CRTC_VBLANK_INT_EN |		\
+			MACH64_CRTC_VLINE_INT_EN |		\
+			MACH64_CRTC_SNAPSHOT_INT_EN |		\
+			MACH64_CRTC_I2C_INT_EN |		\
+			MACH64_CRTC2_VBLANK_INT_EN |		\
+			MACH64_CRTC2_VLINE_INT_EN |		\
+			MACH64_CRTC_CAPBUF0_INT_EN |		\
+			MACH64_CRTC_CAPBUF1_INT_EN |		\
+			MACH64_CRTC_OVERLAY_EOF_INT_EN |	\
+			MACH64_CRTC_ONESHOT_CAP_INT_EN |	\
+			MACH64_CRTC_BUSMASTER_EOL_INT_EN |	\
+			MACH64_CRTC_GP_INT_EN |			\
+			MACH64_CRTC_SNAPSHOT2_INT_EN |		\
+			0					\
+		)
+#	define MACH64_CRTC_INT_ACKS			\
+		(					\
+			MACH64_CRTC_VBLANK_INT |	\
+			MACH64_CRTC_VLINE_INT |		\
+			MACH64_CRTC_SNAPSHOT_INT |	\
+			MACH64_CRTC_I2C_INT |		\
+			MACH64_CRTC2_VBLANK_INT |	\
+			MACH64_CRTC2_VLINE_INT |	\
+			MACH64_CRTC_CAPBUF0_INT |	\
+			MACH64_CRTC_CAPBUF1_INT |	\
+			MACH64_CRTC_OVERLAY_EOF_INT |	\
+			MACH64_CRTC_ONESHOT_CAP_INT |	\
+			MACH64_CRTC_BUSMASTER_EOL_INT |	\
+			MACH64_CRTC_GP_INT |		\
+			MACH64_CRTC_SNAPSHOT2_INT |	\
+			MACH64_CRTC_VBLANK2_INT |	\
+			0				\
+		)
+
+#define MACH64_DATATYPE_CI8				2
+#define MACH64_DATATYPE_ARGB1555			3
+#define MACH64_DATATYPE_RGB565				4
+#define MACH64_DATATYPE_ARGB8888			6
+#define MACH64_DATATYPE_RGB332				7
+#define MACH64_DATATYPE_Y8				8
+#define MACH64_DATATYPE_RGB8				9
+#define MACH64_DATATYPE_VYUY422				11
+#define MACH64_DATATYPE_YVYU422				12
+#define MACH64_DATATYPE_AYUV444				14
+#define MACH64_DATATYPE_ARGB4444			15
+
+#define MACH64_READ(reg)	DRM_READ32(dev_priv->mmio, (reg) )
+#define MACH64_WRITE(reg,val)	DRM_WRITE32(dev_priv->mmio, (reg), (val) )
+
+
+#define DWMREG0		0x0400
+#define DWMREG0_END	0x07ff
+#define DWMREG1		0x0000
+#define DWMREG1_END	0x03ff
+
+#define ISREG0(r)	(((r) >= DWMREG0) && ((r) <= DWMREG0_END))
+#define DMAREG0(r)	(((r) - DWMREG0) >> 2)
+#define DMAREG1(r)	((((r) - DWMREG1) >> 2 ) | 0x0100)
+#define DMAREG(r)	(ISREG0(r) ? DMAREG0(r) : DMAREG1(r))
+
+#define MMREG0		0x0000
+#define MMREG0_END	0x00ff
+
+#define ISMMREG0(r)	(((r) >= MMREG0) && ((r) <= MMREG0_END))
+#define MMSELECT0(r)	(((r) << 2) + DWMREG0)
+#define MMSELECT1(r)	(((((r) & 0xff) << 2) + DWMREG1))
+#define MMSELECT(r)	(ISMMREG0(r) ? MMSELECT0(r) : MMSELECT1(r))
+
+/* ================================================================
+ * DMA constants
+ */
+
+/* DMA descriptor field indices:
+ * The descriptor fields are loaded into the read-only 
+ * BM_* system bus master registers during a bus-master operation
+ */
+#define MACH64_DMA_FRAME_BUF_OFFSET	0        /* BM_FRAME_BUF_OFFSET */
+#define MACH64_DMA_SYS_MEM_ADDR		1        /* BM_SYSTEM_MEM_ADDR */
+#define MACH64_DMA_COMMAND		2        /* BM_COMMAND */
+#define MACH64_DMA_RESERVED		3        /* BM_STATUS */
+
+/* BM_COMMAND descriptor field flags */
+#define MACH64_DMA_HOLD_OFFSET		(1<<30)  /* Don't increment DMA_FRAME_BUF_OFFSET */
+#define MACH64_DMA_EOL			(1<<31)  /* End of descriptor list flag */
+
+#define MACH64_DMA_CHUNKSIZE	        0x1000   /* 4kB per DMA descriptor */
+#define MACH64_APERTURE_OFFSET	        0x7ff800 /* frame-buffer offset for gui-masters */
+
+
+/* ================================================================
+ * Misc helper macros
+ */
+
+static __inline__ void mach64_set_dma_eol( volatile u32 * addr )
+{
+#if defined(__i386__)
+	int nr = 31;
+	
+	/* Taken from include/asm-i386/bitops.h linux header */
+        __asm__ __volatile__( "lock;"
+                "btsl %1,%0"
+                :"=m" (*addr)
+                :"Ir" (nr));
+#elif defined(__powerpc__)
+	u32 old;
+	u32 mask = cpu_to_le32( MACH64_DMA_EOL );
+
+	/* Taken from the include/asm-ppc/bitops.h linux header */
+	__asm__ __volatile__("\n\
+1:	lwarx	%0,0,%3 \n\
+	or	%0,%0,%2 \n\
+	stwcx.	%0,0,%3 \n\
+	bne-	1b"
+	: "=&r" (old), "=m" (*addr)
+	: "r" (mask), "r" (addr), "m" (*addr)
+	: "cc");
+#elif defined(__alpha__)
+	u32 temp;
+	u32 mask = MACH64_DMA_EOL;
+
+	/* Taken from the include/asm-alpha/bitops.h linux header */
+	__asm__ __volatile__(
+	"1:	ldl_l %0,%3\n"
+	"	bis %0,%2,%0\n"
+	"	stl_c %0,%1\n"
+	"	beq %0,2f\n"
+	".subsection 2\n"
+	"2:	br 1b\n"
+	".previous"
+	:"=&r" (temp), "=m" (*addr)
+	:"Ir" (mask), "m" (*addr));
+#else
+	u32 mask = cpu_to_le32( MACH64_DMA_EOL );
+
+	*addr |= mask;
+#endif
+}
+
+static __inline__ void mach64_clear_dma_eol( volatile u32 * addr )
+{
+#if defined(__i386__)
+	int nr = 31;
+	
+	/* Taken from include/asm-i386/bitops.h linux header */
+        __asm__ __volatile__( "lock;"
+                "btrl %1,%0"
+                :"=m" (*addr)
+                :"Ir" (nr));
+#elif defined(__powerpc__)
+	u32 old;
+	u32 mask = cpu_to_le32( MACH64_DMA_EOL );
+
+	/* Taken from the include/asm-ppc/bitops.h linux header */
+	__asm__ __volatile__("\n\
+1:	lwarx	%0,0,%3 \n\
+	andc	%0,%0,%2 \n\
+	stwcx.	%0,0,%3 \n\
+	bne-	1b"
+	: "=&r" (old), "=m" (*addr)
+	: "r" (mask), "r" (addr), "m" (*addr)
+	: "cc");
+#elif defined(__alpha__)
+	u32 temp;
+	u32 mask = ~MACH64_DMA_EOL;
+
+	/* Taken from the include/asm-alpha/bitops.h linux header */
+	__asm__ __volatile__(
+	"1:	ldl_l %0,%3\n"
+	"	and %0,%2,%0\n"
+	"	stl_c %0,%1\n"
+	"	beq %0,2f\n"
+	".subsection 2\n"
+	"2:	br 1b\n"
+	".previous"
+	:"=&r" (temp), "=m" (*addr)
+	:"Ir" (mask), "m" (*addr));
+#else
+	u32 mask = cpu_to_le32( ~MACH64_DMA_EOL );
+
+	*addr &= mask;
+#endif
+}
+
+static __inline__ void mach64_ring_start( drm_mach64_private_t *dev_priv )
+{
+	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;
+	
+	DRM_DEBUG( "%s: head_addr: 0x%08x head: %d tail: %d space: %d\n",
+		   __FUNCTION__, 
+		   ring->head_addr, ring->head, ring->tail, ring->space );
+
+	if ( mach64_do_wait_for_idle( dev_priv ) < 0 ) {
+		mach64_do_engine_reset( dev_priv );
+	}
+
+	if (dev_priv->driver_mode != MACH64_MODE_MMIO ) {
+		/* enable bus mastering and block 1 registers */
+		MACH64_WRITE( MACH64_BUS_CNTL, 
+			      ( MACH64_READ(MACH64_BUS_CNTL) & 	~MACH64_BUS_MASTER_DIS ) 
+			      | MACH64_BUS_EXT_REG_EN );
+		mach64_do_wait_for_idle( dev_priv );
+	}
+	
+	/* reset descriptor table ring head */
+	MACH64_WRITE( MACH64_BM_GUI_TABLE_CMD, 
+		      ring->head_addr | MACH64_CIRCULAR_BUF_SIZE_16KB );
+	
+	dev_priv->ring_running = 1;
+}
+
+static __inline__ void mach64_ring_resume( drm_mach64_private_t *dev_priv, 
+					   drm_mach64_descriptor_ring_t *ring )
+{
+	DRM_DEBUG( "%s: head_addr: 0x%08x head: %d tail: %d space: %d\n",
+		   __FUNCTION__, 
+		   ring->head_addr, ring->head, ring->tail, ring->space );
+
+	/* reset descriptor table ring head */
+	MACH64_WRITE( MACH64_BM_GUI_TABLE_CMD, 
+		      ring->head_addr | MACH64_CIRCULAR_BUF_SIZE_16KB );
+
+	if ( dev_priv->driver_mode == MACH64_MODE_MMIO ) {
+		mach64_do_dispatch_pseudo_dma( dev_priv );
+	} else {
+		/* enable GUI bus mastering, and sync the bus master to the GUI */
+		MACH64_WRITE( MACH64_SRC_CNTL, 
+			      MACH64_SRC_BM_ENABLE | MACH64_SRC_BM_SYNC |
+			      MACH64_SRC_BM_OP_SYSTEM_TO_REG );
+
+		/* kick off the transfer */
+		MACH64_WRITE( MACH64_DST_HEIGHT_WIDTH, 0 );
+		if ( dev_priv->driver_mode == MACH64_MODE_DMA_SYNC ) {
+			if ( (mach64_do_wait_for_idle( dev_priv )) < 0 ) {
+				DRM_ERROR( "%s: idle failed, resetting engine\n", 
+					   __FUNCTION__);
+				mach64_dump_engine_info( dev_priv );
+				mach64_do_engine_reset( dev_priv );
+				return;
+			}
+			mach64_do_release_used_buffers( dev_priv );
+		}
+	}
+}
+
+static __inline__ void mach64_ring_tick( drm_mach64_private_t *dev_priv, 
+					 drm_mach64_descriptor_ring_t *ring )
+{
+	DRM_DEBUG( "%s: head_addr: 0x%08x head: %d tail: %d space: %d\n",
+		   __FUNCTION__, 
+		   ring->head_addr, ring->head, ring->tail, ring->space );
+
+	if ( !dev_priv->ring_running ) {
+		mach64_ring_start( dev_priv );
+		
+		if ( ring->head != ring->tail ) {
+			mach64_ring_resume( dev_priv, ring );
+		}
+	} else {
+		/* GUI_ACTIVE must be read before BM_GUI_TABLE to 
+		 * correctly determine the ring head 
+		 */
+		int gui_active = MACH64_READ(MACH64_GUI_STAT) & MACH64_GUI_ACTIVE;
+		
+		ring->head_addr = MACH64_READ(MACH64_BM_GUI_TABLE) & 0xfffffff0;
+		
+		if ( gui_active ) {
+			/* If not idle, BM_GUI_TABLE points one descriptor 
+			 * past the current head 
+			 */
+			if ( ring->head_addr == ring->start_addr ) {
+				ring->head_addr += ring->size;
+			}
+			ring->head_addr -= 4 * sizeof(u32);
+		}
+
+		if( ring->head_addr < ring->start_addr || 
+		    ring->head_addr >= ring->start_addr + ring->size ) {
+			DRM_ERROR( "bad ring head address: 0x%08x\n", ring->head_addr );
+			mach64_dump_ring_info( dev_priv );
+			mach64_do_engine_reset( dev_priv );
+			return;
+		}
+	
+		ring->head = (ring->head_addr - ring->start_addr) / sizeof(u32);
+		
+		if ( !gui_active && ring->head != ring->tail ) {
+			mach64_ring_resume( dev_priv, ring );
+		}
+	}
+}
+
+static __inline__ void mach64_ring_stop( drm_mach64_private_t *dev_priv )
+{
+	DRM_DEBUG( "%s: head_addr: 0x%08x head: %d tail: %d space: %d\n",
+		   __FUNCTION__, 
+		   dev_priv->ring.head_addr, dev_priv->ring.head, 
+		   dev_priv->ring.tail, dev_priv->ring.space );
+
+	/* restore previous SRC_CNTL to disable busmastering */
+	mach64_do_wait_for_fifo( dev_priv, 1 );
+	MACH64_WRITE( MACH64_SRC_CNTL, 0 );
+
+	/* disable busmastering but keep the block 1 registers enabled */ 
+	mach64_do_wait_for_idle( dev_priv );
+	MACH64_WRITE( MACH64_BUS_CNTL, MACH64_READ( MACH64_BUS_CNTL ) 
+		      | MACH64_BUS_MASTER_DIS | MACH64_BUS_EXT_REG_EN );
+		
+	dev_priv->ring_running = 0;
+}
+
+static __inline__ void
+mach64_update_ring_snapshot( drm_mach64_private_t *dev_priv )
+{
+	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;
+
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+	
+	mach64_ring_tick( dev_priv, ring );
+
+	ring->space = (ring->head - ring->tail) * sizeof(u32);
+	if ( ring->space <= 0 ) {
+		ring->space += ring->size;
+	}
+}
+
+/* ================================================================
+ * DMA descriptor ring macros
+ */
+
+#define RING_LOCALS									\
+	int _ring_tail, _ring_write; unsigned int _ring_mask; volatile u32 *_ring
+
+#define RING_WRITE_OFS  _ring_write
+
+#define BEGIN_RING( n ) 								\
+do {											\
+	if ( MACH64_VERBOSE ) {								\
+		DRM_INFO( "BEGIN_RING( %d ) in %s\n",					\
+			   (n), __FUNCTION__ );						\
+	}										\
+	if ( dev_priv->ring.space <= (n) * sizeof(u32) ) {				\
+		int ret;								\
+		if ((ret=mach64_wait_ring( dev_priv, (n) * sizeof(u32))) < 0 ) {	\
+			DRM_ERROR( "wait_ring failed, resetting engine\n");		\
+			mach64_dump_engine_info( dev_priv );				\
+			mach64_do_engine_reset( dev_priv );				\
+			return ret;							\
+		}									\
+	}										\
+	dev_priv->ring.space -= (n) * sizeof(u32);					\
+	_ring = (u32 *) dev_priv->ring.start;						\
+	_ring_tail = _ring_write = dev_priv->ring.tail;					\
+	_ring_mask = dev_priv->ring.tail_mask;						\
+} while (0)
+
+#define OUT_RING( x )						\
+do {								\
+	if ( MACH64_VERBOSE ) {					\
+		DRM_INFO( "   OUT_RING( 0x%08x ) at 0x%x\n",	\
+			   (unsigned int)(x), _ring_write );	\
+	}							\
+	_ring[_ring_write++] = cpu_to_le32( x );		\
+	_ring_write &= _ring_mask;				\
+} while (0)
+
+#define ADVANCE_RING() 							\
+do {									\
+	if ( MACH64_VERBOSE ) {						\
+		DRM_INFO( "ADVANCE_RING() wr=0x%06x tail=0x%06x\n",	\
+			  _ring_write, _ring_tail );			\
+	}								\
+	DRM_MEMORYBARRIER();						\
+	mach64_clear_dma_eol( &_ring[(_ring_tail - 2) & _ring_mask] );	\
+	DRM_MEMORYBARRIER();						\
+	dev_priv->ring.tail = _ring_write;				\
+	mach64_ring_tick( dev_priv, &(dev_priv)->ring );		\
+} while (0)
+
+
+/* ================================================================
+ * DMA macros
+ */
+
+#define DMALOCALS				\
+	drm_mach64_freelist_t *_entry = NULL;	\
+	drm_buf_t *_buf = NULL; 		\
+	u32 *_buf_wptr; int _outcount
+
+#define GETBUFPTR( __buf )						\
+((dev_priv->is_pci) ? 							\
+	((u32 *)(__buf)->address) : 					\
+	((u32 *)((char *)dev_priv->buffers->handle + (__buf)->offset)))
+
+#define GETBUFADDR( __buf ) ((u32)(__buf)->bus_address)
+
+#define GETRINGOFFSET() (_entry->ring_ofs)
+
+static __inline__ int mach64_find_pending_buf_entry ( drm_mach64_private_t *dev_priv, 
+						      drm_mach64_freelist_t **entry, 
+						      drm_buf_t *buf )
+{
+	struct list_head *ptr;
+#if MACH64_EXTRA_CHECKING
+	if (list_empty(&dev_priv->pending)) {
+		DRM_ERROR("Empty pending list in %s\n", __FUNCTION__);
+		return DRM_ERR(EINVAL);
+	}
+#endif
+	ptr = dev_priv->pending.prev;
+	*entry = list_entry(ptr, drm_mach64_freelist_t, list);
+	while ((*entry)->buf != buf) {
+		if (ptr == &dev_priv->pending) {
+			return DRM_ERR(EFAULT);
+		}
+		ptr = ptr->prev;
+		*entry = list_entry(ptr, drm_mach64_freelist_t, list);
+	}
+	return 0;
+}
+
+#define DMASETPTR( _p ) 			\
+do {						\
+	_buf = (_p);				\
+	_outcount = 0;				\
+	_buf_wptr = GETBUFPTR( _buf );		\
+} while(0)
+
+/* FIXME: use a private set of smaller buffers for state emits, clears, and swaps? */
+#define DMAGETPTR( filp, dev_priv, n )					\
+do {									\
+	if ( MACH64_VERBOSE ) {						\
+		DRM_INFO( "DMAGETPTR( %d ) in %s\n",			\
+			  n, __FUNCTION__ );				\
+	}								\
+	_buf = mach64_freelist_get( dev_priv );				\
+	if (_buf == NULL) {						\
+		DRM_ERROR("%s: couldn't get buffer in DMAGETPTR\n",	\
+			   __FUNCTION__ );				\
+		return DRM_ERR(EAGAIN);					\
+	}								\
+	if (_buf->pending) {						\
+	        DRM_ERROR("%s: pending buf in DMAGETPTR\n",		\
+			   __FUNCTION__ );				\
+		return DRM_ERR(EFAULT);					\
+	}								\
+	_buf->filp = filp;						\
+	_outcount = 0;							\
+									\
+        _buf_wptr = GETBUFPTR( _buf );					\
+} while (0)
+
+#define DMAOUTREG( reg, val )					\
+do {								\
+	if ( MACH64_VERBOSE ) {					\
+		DRM_INFO( "   DMAOUTREG( 0x%x = 0x%08x )\n",	\
+			  reg, val );				\
+	}							\
+	_buf_wptr[_outcount++] = cpu_to_le32(DMAREG(reg));	\
+	_buf_wptr[_outcount++] = cpu_to_le32((val));		\
+	_buf->used += 8;					\
+} while (0)
+
+#define DMAADVANCE( dev_priv, _discard )						     \
+do {											     \
+	struct list_head *ptr;								     \
+	RING_LOCALS;									     \
+											     \
+	if ( MACH64_VERBOSE ) {								     \
+		DRM_INFO( "DMAADVANCE() in %s\n", __FUNCTION__ );			     \
+	}										     \
+											     \
+	if (_buf->used <= 0) {								     \
+		DRM_ERROR( "DMAADVANCE() in %s: sending empty buf %d\n",		     \
+				   __FUNCTION__, _buf->idx );				     \
+		return DRM_ERR(EFAULT);							     \
+	}										     \
+	if (_buf->pending) {								     \
+                /* This is a resued buffer, so we need to find it in the pending list */     \
+		int ret;								     \
+		if ( (ret=mach64_find_pending_buf_entry(dev_priv, &_entry, _buf)) ) {	     \
+			DRM_ERROR( "DMAADVANCE() in %s: couldn't find pending buf %d\n",     \
+				   __FUNCTION__, _buf->idx );				     \
+			return ret;							     \
+		}									     \
+		if (_entry->discard) {							     \
+			DRM_ERROR( "DMAADVANCE() in %s: sending discarded pending buf %d\n", \
+				   __FUNCTION__, _buf->idx );				     \
+			return DRM_ERR(EFAULT);						     \
+		}									     \
+     	} else {									     \
+		if (list_empty(&dev_priv->placeholders)) {				     \
+			DRM_ERROR( "DMAADVANCE() in %s: empty placeholder list\n",	     \
+			   	__FUNCTION__ );						     \
+			return DRM_ERR(EFAULT);						     \
+		}									     \
+		ptr = dev_priv->placeholders.next;					     \
+		list_del(ptr);								     \
+		_entry = list_entry(ptr, drm_mach64_freelist_t, list);			     \
+		_buf->pending = 1;							     \
+		_entry->buf = _buf;							     \
+		list_add_tail(ptr, &dev_priv->pending);					     \
+	}										     \
+	_entry->discard = (_discard);							     \
+	ADD_BUF_TO_RING( dev_priv );							     \
+} while (0)
+
+#define DMADISCARDBUF()									\
+do {											\
+	if (_entry == NULL) {								\
+		int ret;								\
+		if ( (ret=mach64_find_pending_buf_entry(dev_priv, &_entry, _buf)) ) {	\
+			DRM_ERROR( "%s: couldn't find pending buf %d\n",		\
+				   __FUNCTION__, _buf->idx );				\
+			return ret;							\
+		}									\
+	}										\
+	_entry->discard = 1;								\
+} while(0)
+
+#define ADD_BUF_TO_RING( dev_priv )							\
+do {											\
+	int bytes, pages, remainder;							\
+	u32 address, page;								\
+	int i;										\
+											\
+	bytes = _buf->used;								\
+	address = GETBUFADDR( _buf );							\
+											\
+	pages = (bytes + MACH64_DMA_CHUNKSIZE - 1) / MACH64_DMA_CHUNKSIZE;		\
+											\
+	BEGIN_RING( pages * 4 );							\
+											\
+	for ( i = 0 ; i < pages-1 ; i++ ) {						\
+		page = address + i * MACH64_DMA_CHUNKSIZE;				\
+		OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_ADDR );			\
+		OUT_RING( page );							\
+		OUT_RING( MACH64_DMA_CHUNKSIZE | MACH64_DMA_HOLD_OFFSET );		\
+		OUT_RING( 0 );								\
+	}										\
+											\
+	/* generate the final descriptor for any remaining commands in this buffer */	\
+	page = address + i * MACH64_DMA_CHUNKSIZE;					\
+	remainder = bytes - i * MACH64_DMA_CHUNKSIZE;					\
+											\
+	/* Save dword offset of last descriptor for this buffer.			\
+	 * This is needed to check for completion of the buffer in freelist_get		\
+	 */										\
+	_entry->ring_ofs = RING_WRITE_OFS;						\
+											\
+	OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_ADDR );				\
+	OUT_RING( page );								\
+	OUT_RING( remainder | MACH64_DMA_HOLD_OFFSET | MACH64_DMA_EOL );		\
+	OUT_RING( 0 );									\
+											\
+	ADVANCE_RING();									\
+} while(0)
+
+#define DMAADVANCEHOSTDATA( dev_priv )							\
+do {											\
+	struct list_head *ptr;								\
+	RING_LOCALS;									\
+											\
+	if ( MACH64_VERBOSE ) {								\
+		DRM_INFO( "DMAADVANCEHOSTDATA() in %s\n", __FUNCTION__ );		\
+	}										\
+											\
+	if (_buf->used <= 0) {								\
+		DRM_ERROR( "DMAADVANCEHOSTDATA() in %s: sending empty buf %d\n",	\
+				   __FUNCTION__, _buf->idx );				\
+		return DRM_ERR(EFAULT);							\
+	}										\
+	if (list_empty(&dev_priv->placeholders)) {					\
+		DRM_ERROR( "%s: empty placeholder list in DMAADVANCEHOSTDATA()\n",	\
+			   __FUNCTION__ );						\
+		return DRM_ERR(EFAULT);							\
+	}										\
+											\
+        ptr = dev_priv->placeholders.next;						\
+	list_del(ptr);									\
+	_entry = list_entry(ptr, drm_mach64_freelist_t, list);				\
+	_entry->buf = _buf;								\
+	_entry->buf->pending = 1;							\
+	list_add_tail(ptr, &dev_priv->pending);						\
+	_entry->discard = 1;								\
+	ADD_HOSTDATA_BUF_TO_RING( dev_priv );						\
+} while (0)
+
+#define ADD_HOSTDATA_BUF_TO_RING( dev_priv )						 \
+do {											 \
+	int bytes, pages, remainder;							 \
+	u32 address, page;								 \
+	int i;										 \
+											 \
+	bytes = _buf->used - MACH64_HOSTDATA_BLIT_OFFSET;				 \
+	pages = (bytes + MACH64_DMA_CHUNKSIZE - 1) / MACH64_DMA_CHUNKSIZE;		 \
+	address = GETBUFADDR( _buf );							 \
+											 \
+	BEGIN_RING( 4 + pages * 4 );							 \
+											 \
+	OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_ADDR );				 \
+	OUT_RING( address );								 \
+	OUT_RING( MACH64_HOSTDATA_BLIT_OFFSET | MACH64_DMA_HOLD_OFFSET );		 \
+	OUT_RING( 0 );									 \
+											 \
+	address += MACH64_HOSTDATA_BLIT_OFFSET;						 \
+											 \
+	for ( i = 0 ; i < pages-1 ; i++ ) {						 \
+		page = address + i * MACH64_DMA_CHUNKSIZE;				 \
+		OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_HOSTDATA );		 \
+		OUT_RING( page );							 \
+		OUT_RING( MACH64_DMA_CHUNKSIZE | MACH64_DMA_HOLD_OFFSET );		 \
+		OUT_RING( 0 );								 \
+	}										 \
+											 \
+	/* generate the final descriptor for any remaining commands in this buffer */	 \
+	page = address + i * MACH64_DMA_CHUNKSIZE;					 \
+	remainder = bytes - i * MACH64_DMA_CHUNKSIZE;					 \
+											 \
+	/* Save dword offset of last descriptor for this buffer.			 \
+	 * This is needed to check for completion of the buffer in freelist_get		 \
+	 */										 \
+	_entry->ring_ofs = RING_WRITE_OFS;						 \
+											 \
+	OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_HOSTDATA );			 \
+	OUT_RING( page );								 \
+	OUT_RING( remainder | MACH64_DMA_HOLD_OFFSET | MACH64_DMA_EOL );		 \
+	OUT_RING( 0 );									 \
+											 \
+	ADVANCE_RING();									 \
+} while(0)
+
+#endif /* __MACH64_DRV_H__ */
diff --git a/shared/mach64_irq.c b/shared/mach64_irq.c
new file mode 100644
index 00000000..9acd6269
--- /dev/null
+++ b/shared/mach64_irq.c
@@ -0,0 +1,130 @@
+/* mach64_irq.c -- IRQ handling for ATI Mach64 -*- linux-c -*-
+ * Created: Tue Feb 25, 2003 by Leif Delgass, based on radeon_irq.c/r128_irq.c
+ *
+ * Copyright (C) The Weather Channel, Inc.  2002.
+ * Copyright 2003 Leif Delgass
+ * All Rights Reserved.
+ *
+ * The Weather Channel (TM) funded Tungsten Graphics to develop the
+ * initial release of the Radeon 8500 driver under the XFree86 license.
+ * This notice must be preserved.
+ *
+ * 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
+ * THE COPYRIGHT OWNER(S) 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 <keith@tungstengraphics.com>
+ *    Eric Anholt <anholt@FreeBSD.org>
+ *    Leif Delgass <ldelgass@retinalburn.net>
+ */
+
+#include "mach64.h"
+#include "drmP.h"
+#include "drm.h"
+#include "mach64_drm.h"
+#include "mach64_drv.h"
+
+irqreturn_t DRM(irq_handler)( DRM_IRQ_ARGS )
+{
+	drm_device_t *dev = (drm_device_t *) arg;
+	drm_mach64_private_t *dev_priv = 
+	   (drm_mach64_private_t *)dev->dev_private;
+	int status;
+
+	status = MACH64_READ( MACH64_CRTC_INT_CNTL );
+
+	/* VBLANK interrupt */
+	if (status & MACH64_CRTC_VBLANK_INT) {
+	    /* Mask off all interrupt ack bits before setting the ack bit, since
+	     * there may be other handlers outside the DRM.  
+	     *
+	     * NOTE: On mach64, you need to keep the enable bits set when doing
+	     * the ack, despite what the docs say about not acking and enabling
+	     * in a single write.
+	     */
+	    MACH64_WRITE( MACH64_CRTC_INT_CNTL, (status & ~MACH64_CRTC_INT_ACKS)
+			  | MACH64_CRTC_VBLANK_INT );
+
+	    atomic_inc(&dev->vbl_received);
+	    DRM_WAKEUP(&dev->vbl_queue);
+	    DRM(vbl_send_signals)( dev );
+            return IRQ_HANDLED;
+	}
+	return IRQ_NONE;
+}
+
+int DRM(vblank_wait)(drm_device_t *dev, unsigned int *sequence)
+{
+	unsigned int cur_vblank;
+	int ret = 0;
+
+	/* Assume that the user has missed the current sequence number
+	 * by about a day rather than she wants to wait for years
+	 * using vertical blanks... 
+	 */
+	DRM_WAIT_ON( ret, dev->vbl_queue, 3*DRM_HZ, 
+		     ( ( ( cur_vblank = atomic_read(&dev->vbl_received ) )
+			 - *sequence ) <= (1<<23) ) );
+
+	*sequence = cur_vblank;
+
+	return ret;
+}
+
+/* drm_dma.h hooks
+*/
+void DRM(driver_irq_preinstall)( drm_device_t *dev ) {
+	drm_mach64_private_t *dev_priv =
+		(drm_mach64_private_t *)dev->dev_private;
+
+	u32 status = MACH64_READ( MACH64_CRTC_INT_CNTL );
+
+	DRM_DEBUG("before install CRTC_INT_CTNL: 0x%08x\n", status );
+
+ 	/* Disable and clear VBLANK interrupt */
+	MACH64_WRITE( MACH64_CRTC_INT_CNTL, (status & ~MACH64_CRTC_VBLANK_INT_EN) 
+		      | MACH64_CRTC_VBLANK_INT );
+}
+
+void DRM(driver_irq_postinstall)( drm_device_t *dev ) {
+	drm_mach64_private_t *dev_priv =
+		(drm_mach64_private_t *)dev->dev_private;
+
+	/* Turn on VBLANK interrupt */
+   	MACH64_WRITE( MACH64_CRTC_INT_CNTL, MACH64_READ( MACH64_CRTC_INT_CNTL )
+		      | MACH64_CRTC_VBLANK_INT_EN );
+
+	DRM_DEBUG("after install CRTC_INT_CTNL: 0x%08x\n", MACH64_READ( MACH64_CRTC_INT_CNTL ));
+
+}
+
+void DRM(driver_irq_uninstall)( drm_device_t *dev ) {
+	drm_mach64_private_t *dev_priv =
+		(drm_mach64_private_t *)dev->dev_private;
+	if ( !dev_priv )
+		return;
+
+	/* Disable and clear VBLANK interrupt */
+	MACH64_WRITE( MACH64_CRTC_INT_CNTL, 
+		      (MACH64_READ( MACH64_CRTC_INT_CNTL ) & ~MACH64_CRTC_VBLANK_INT_EN) 
+		      | MACH64_CRTC_VBLANK_INT );
+
+	DRM_DEBUG("after uninstall CRTC_INT_CTNL: 0x%08x\n", 
+		  MACH64_READ( MACH64_CRTC_INT_CNTL ));
+}
diff --git a/shared/mach64_state.c b/shared/mach64_state.c
new file mode 100644
index 00000000..eb3722cb
--- /dev/null
+++ b/shared/mach64_state.c
@@ -0,0 +1,897 @@
+/* mach64_state.c -- State support for mach64 (Rage Pro) driver -*- linux-c -*-
+ * Created: Sun Dec 03 19:20:26 2000 by gareth@valinux.com
+ *
+ * Copyright 2000 Gareth Hughes
+ * Copyright 2002-2003 Leif Delgass
+ * 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
+ * THE COPYRIGHT OWNER(S) 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:
+ *    Gareth Hughes <gareth@valinux.com>
+ *    Leif Delgass <ldelgass@retinalburn.net>
+ *    José Fonseca <j_r_fonseca@yahoo.co.uk>
+ */
+
+#define __NO_VERSION__
+#include "mach64.h"
+#include "drmP.h"
+#include "drm.h"
+#include "mach64_drm.h"
+#include "mach64_drv.h"
+
+
+/* ================================================================
+ * DMA hardware state programming functions
+ */
+
+static void mach64_print_dirty( const char *msg, unsigned int flags )
+{
+	DRM_DEBUG( "%s: (0x%x) %s%s%s%s%s%s%s%s%s%s%s%s\n",
+		msg,
+		flags,
+		(flags & MACH64_UPLOAD_DST_OFF_PITCH) ? "dst_off_pitch, " : "",
+		(flags & MACH64_UPLOAD_Z_ALPHA_CNTL)  ? "z_alpha_cntl, " : "",
+		(flags & MACH64_UPLOAD_SCALE_3D_CNTL) ? "scale_3d_cntl, " : "",
+		(flags & MACH64_UPLOAD_DP_FOG_CLR)    ? "dp_fog_clr, " : "",
+		(flags & MACH64_UPLOAD_DP_WRITE_MASK) ? "dp_write_mask, " : "",
+		(flags & MACH64_UPLOAD_DP_PIX_WIDTH)  ? "dp_pix_width, " : "",
+		(flags & MACH64_UPLOAD_SETUP_CNTL)    ? "setup_cntl, " : "",
+		(flags & MACH64_UPLOAD_MISC)          ? "misc, " : "",
+		(flags & MACH64_UPLOAD_TEXTURE)       ? "texture, " : "",
+		(flags & MACH64_UPLOAD_TEX0IMAGE)     ? "tex0 image, " : "",
+		(flags & MACH64_UPLOAD_TEX1IMAGE)     ? "tex1 image, " : "",
+		(flags & MACH64_UPLOAD_CLIPRECTS)     ? "cliprects, " : "" );
+}
+
+/* Mach64 doesn't have hardware cliprects, just one hardware scissor,
+ * so the GL scissor is intersected with each cliprect here
+ */
+/* This function returns 0 on success, 1 for no intersection, and
+ * negative for an error
+ */
+static int mach64_emit_cliprect( DRMFILE filp, drm_mach64_private_t *dev_priv,
+				 drm_clip_rect_t *box )
+{
+	u32 sc_left_right, sc_top_bottom;
+	drm_clip_rect_t scissor;
+	drm_mach64_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	drm_mach64_context_regs_t *regs = &sarea_priv->context_state;
+	DMALOCALS;
+
+	DRM_DEBUG( "%s: box=%p\n", __FUNCTION__, box );
+
+	/* Get GL scissor */
+	/* FIXME: store scissor in SAREA as a cliprect instead of in
+	 * hardware format, or do intersection client-side
+	 */
+	scissor.x1 = regs->sc_left_right & 0xffff;
+	scissor.x2 = (regs->sc_left_right & 0xffff0000) >> 16;
+	scissor.y1 = regs->sc_top_bottom & 0xffff;
+	scissor.y2 = (regs->sc_top_bottom & 0xffff0000) >> 16;
+
+	/* Intersect GL scissor with cliprect */
+	if ( box->x1 > scissor.x1 ) scissor.x1 = box->x1;
+	if ( box->y1 > scissor.y1 ) scissor.y1 = box->y1;
+	if ( box->x2 < scissor.x2 ) scissor.x2 = box->x2;
+	if ( box->y2 < scissor.y2 ) scissor.y2 = box->y2;
+	 /* positive return means skip */
+	if ( scissor.x1 >= scissor.x2 ) return 1;
+	if ( scissor.y1 >= scissor.y2 ) return 1;
+
+	DMAGETPTR( filp, dev_priv, 2 ); /* returns on failure to get buffer */
+
+	sc_left_right = ( (scissor.x1 << 0) | (scissor.x2 << 16) );
+	sc_top_bottom = ( (scissor.y1 << 0) | (scissor.y2 << 16) );
+
+	DMAOUTREG( MACH64_SC_LEFT_RIGHT, sc_left_right );
+	DMAOUTREG( MACH64_SC_TOP_BOTTOM, sc_top_bottom );
+
+	DMAADVANCE( dev_priv, 1 );
+
+	return 0;
+}
+
+static __inline__ int mach64_emit_state( DRMFILE filp, drm_mach64_private_t *dev_priv )
+{
+	drm_mach64_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	drm_mach64_context_regs_t *regs = &sarea_priv->context_state;
+	unsigned int dirty = sarea_priv->dirty;
+	u32 offset = ((regs->tex_size_pitch & 0xf0) >> 2);
+	DMALOCALS;
+
+	if ( MACH64_VERBOSE ) {
+		mach64_print_dirty( __FUNCTION__, dirty );
+	} else {
+		DRM_DEBUG( "%s: dirty=0x%08x\n", __FUNCTION__, dirty );
+	}
+
+	DMAGETPTR( filp, dev_priv, 17 ); /* returns on failure to get buffer */
+
+	if ( dirty & MACH64_UPLOAD_MISC ) {
+		DMAOUTREG( MACH64_DP_MIX, regs->dp_mix );
+		DMAOUTREG( MACH64_DP_SRC, regs->dp_src );
+		DMAOUTREG( MACH64_CLR_CMP_CNTL, regs->clr_cmp_cntl );
+		DMAOUTREG( MACH64_GUI_TRAJ_CNTL, regs->gui_traj_cntl );
+		sarea_priv->dirty &= ~MACH64_UPLOAD_MISC;
+	}
+
+	if ( dirty & MACH64_UPLOAD_DST_OFF_PITCH ) {
+		DMAOUTREG( MACH64_DST_OFF_PITCH, regs->dst_off_pitch );
+		sarea_priv->dirty &= ~MACH64_UPLOAD_DST_OFF_PITCH;
+	}
+	if ( dirty & MACH64_UPLOAD_Z_OFF_PITCH ) {
+		DMAOUTREG( MACH64_Z_OFF_PITCH, regs->z_off_pitch );
+		sarea_priv->dirty &= ~MACH64_UPLOAD_Z_OFF_PITCH;
+	}
+	if ( dirty & MACH64_UPLOAD_Z_ALPHA_CNTL ) {
+		DMAOUTREG( MACH64_Z_CNTL, regs->z_cntl );
+		DMAOUTREG( MACH64_ALPHA_TST_CNTL, regs->alpha_tst_cntl );
+		sarea_priv->dirty &= ~MACH64_UPLOAD_Z_ALPHA_CNTL;
+	}
+	if ( dirty & MACH64_UPLOAD_SCALE_3D_CNTL ) {
+		DMAOUTREG( MACH64_SCALE_3D_CNTL, regs->scale_3d_cntl );
+		sarea_priv->dirty &= ~MACH64_UPLOAD_SCALE_3D_CNTL;
+	}
+	if ( dirty & MACH64_UPLOAD_DP_FOG_CLR ) {
+		DMAOUTREG( MACH64_DP_FOG_CLR, regs->dp_fog_clr );
+		sarea_priv->dirty &= ~MACH64_UPLOAD_DP_FOG_CLR;
+	}
+	if ( dirty & MACH64_UPLOAD_DP_WRITE_MASK ) {
+		DMAOUTREG( MACH64_DP_WRITE_MASK, regs->dp_write_mask );
+		sarea_priv->dirty &= ~MACH64_UPLOAD_DP_WRITE_MASK;
+	}
+	if ( dirty & MACH64_UPLOAD_DP_PIX_WIDTH ) {
+		DMAOUTREG( MACH64_DP_PIX_WIDTH, regs->dp_pix_width );
+		sarea_priv->dirty &= ~MACH64_UPLOAD_DP_PIX_WIDTH;
+	}
+	if ( dirty & MACH64_UPLOAD_SETUP_CNTL ) {
+		DMAOUTREG( MACH64_SETUP_CNTL, regs->setup_cntl );
+		sarea_priv->dirty &= ~MACH64_UPLOAD_SETUP_CNTL;
+	}
+
+	if ( dirty & MACH64_UPLOAD_TEXTURE ) {
+		DMAOUTREG( MACH64_TEX_SIZE_PITCH, regs->tex_size_pitch );
+		DMAOUTREG( MACH64_TEX_CNTL, regs->tex_cntl );
+		DMAOUTREG( MACH64_SECONDARY_TEX_OFF, regs->secondary_tex_off );
+		DMAOUTREG( MACH64_TEX_0_OFF + offset, regs->tex_offset );
+		sarea_priv->dirty &= ~MACH64_UPLOAD_TEXTURE;
+	}
+
+	DMAADVANCE( dev_priv, 1 );
+
+	sarea_priv->dirty &= MACH64_UPLOAD_CLIPRECTS;
+
+	return 0;
+
+}
+
+
+/* ================================================================
+ * DMA command dispatch functions
+ */
+
+static int mach64_dma_dispatch_clear( DRMFILE filp, drm_device_t *dev,
+				      unsigned int flags,
+				      int cx, int cy, int cw, int ch,
+				      unsigned int clear_color,
+				      unsigned int clear_depth )
+{
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	drm_mach64_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	drm_mach64_context_regs_t *ctx = &sarea_priv->context_state;
+	int nbox = sarea_priv->nbox;
+	drm_clip_rect_t *pbox = sarea_priv->boxes;
+	u32 fb_bpp, depth_bpp;
+	int i;
+	DMALOCALS;
+
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+	switch ( dev_priv->fb_bpp ) {
+	case 16:
+		fb_bpp = MACH64_DATATYPE_RGB565;
+		break;
+	case 32:
+		fb_bpp = MACH64_DATATYPE_ARGB8888;
+		break;
+	default:
+		return DRM_ERR(EINVAL);
+	}
+	switch ( dev_priv->depth_bpp ) {
+	case 16:
+		depth_bpp = MACH64_DATATYPE_RGB565;
+		break;
+	case 24:
+	case 32:
+		depth_bpp = MACH64_DATATYPE_ARGB8888;
+		break;
+	default:
+		return DRM_ERR(EINVAL);
+	}
+
+	if ( !nbox ) 
+		return 0;
+
+	DMAGETPTR( filp, dev_priv, nbox * 31 ); /* returns on failure to get buffer */
+
+	for ( i = 0 ; i < nbox ; i++ ) {
+		int x = pbox[i].x1;
+		int y = pbox[i].y1;
+		int w = pbox[i].x2 - x;
+		int h = pbox[i].y2 - y;
+		
+		DRM_DEBUG( "dispatch clear %d,%d-%d,%d flags 0x%x\n",
+			   pbox[i].x1, pbox[i].y1,
+			   pbox[i].x2, pbox[i].y2, flags );
+		
+		if ( flags & (MACH64_FRONT | MACH64_BACK) ) {
+			/* Setup for color buffer clears
+			 */
+			
+			DMAOUTREG( MACH64_Z_CNTL, 0 );
+			DMAOUTREG( MACH64_SCALE_3D_CNTL, 0 );
+			
+			DMAOUTREG( MACH64_SC_LEFT_RIGHT, ctx->sc_left_right );
+			DMAOUTREG( MACH64_SC_TOP_BOTTOM, ctx->sc_top_bottom );
+
+			DMAOUTREG( MACH64_CLR_CMP_CNTL, 0 );
+			DMAOUTREG( MACH64_GUI_TRAJ_CNTL,
+				   (MACH64_DST_X_LEFT_TO_RIGHT |
+				    MACH64_DST_Y_TOP_TO_BOTTOM) );
+			
+			DMAOUTREG( MACH64_DP_PIX_WIDTH, ((fb_bpp << 0) |
+							 (fb_bpp << 4) |
+							 (fb_bpp << 8) |
+							 (fb_bpp << 16) |
+							 (fb_bpp << 28)) );
+			
+			DMAOUTREG( MACH64_DP_FRGD_CLR, clear_color );
+			DMAOUTREG( MACH64_DP_WRITE_MASK, ctx->dp_write_mask );
+			DMAOUTREG( MACH64_DP_MIX, (MACH64_BKGD_MIX_D |
+						   MACH64_FRGD_MIX_S) );
+			DMAOUTREG( MACH64_DP_SRC, (MACH64_BKGD_SRC_FRGD_CLR |
+						   MACH64_FRGD_SRC_FRGD_CLR |
+						   MACH64_MONO_SRC_ONE) );
+			
+						
+		}
+
+		if ( flags & MACH64_FRONT ) {
+						
+			DMAOUTREG( MACH64_DST_OFF_PITCH,
+				   dev_priv->front_offset_pitch );
+			DMAOUTREG( MACH64_DST_X_Y,
+				   (y << 16) | x );
+			DMAOUTREG( MACH64_DST_WIDTH_HEIGHT,
+				   (h << 16) | w );
+			
+		}
+		
+		if ( flags & MACH64_BACK ) {
+						
+			DMAOUTREG( MACH64_DST_OFF_PITCH,
+				   dev_priv->back_offset_pitch );
+			DMAOUTREG( MACH64_DST_X_Y,
+				   (y << 16) | x );
+			DMAOUTREG( MACH64_DST_WIDTH_HEIGHT,
+				   (h << 16) | w );
+			
+		}
+				
+		if ( flags & MACH64_DEPTH ) {
+			/* Setup for depth buffer clear
+			 */
+			DMAOUTREG( MACH64_Z_CNTL, 0 );
+			DMAOUTREG( MACH64_SCALE_3D_CNTL, 0 );
+			
+			DMAOUTREG( MACH64_SC_LEFT_RIGHT, ctx->sc_left_right );
+			DMAOUTREG( MACH64_SC_TOP_BOTTOM, ctx->sc_top_bottom );
+
+			DMAOUTREG( MACH64_CLR_CMP_CNTL, 0 );
+			DMAOUTREG( MACH64_GUI_TRAJ_CNTL,
+				   (MACH64_DST_X_LEFT_TO_RIGHT |
+				    MACH64_DST_Y_TOP_TO_BOTTOM) );
+			
+			DMAOUTREG( MACH64_DP_PIX_WIDTH, ((depth_bpp << 0) |
+							 (depth_bpp << 4) |
+							 (depth_bpp << 8) |
+							 (depth_bpp << 16) |
+							 (depth_bpp << 28)) );
+			
+			DMAOUTREG( MACH64_DP_FRGD_CLR, clear_depth );
+			DMAOUTREG( MACH64_DP_WRITE_MASK, 0xffffffff );
+			DMAOUTREG( MACH64_DP_MIX, (MACH64_BKGD_MIX_D |
+						   MACH64_FRGD_MIX_S) );
+			DMAOUTREG( MACH64_DP_SRC, (MACH64_BKGD_SRC_FRGD_CLR |
+						   MACH64_FRGD_SRC_FRGD_CLR |
+						   MACH64_MONO_SRC_ONE) );
+			
+			DMAOUTREG( MACH64_DST_OFF_PITCH,
+				   dev_priv->depth_offset_pitch );
+			DMAOUTREG( MACH64_DST_X_Y,
+				   (y << 16) | x );
+			DMAOUTREG( MACH64_DST_WIDTH_HEIGHT,
+				   (h << 16) | w );
+		}
+	}
+
+	DMAADVANCE( dev_priv, 1 );
+
+	return 0;
+}
+
+static int mach64_dma_dispatch_swap( DRMFILE filp, drm_device_t *dev )
+{
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	drm_mach64_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	int nbox = sarea_priv->nbox;
+	drm_clip_rect_t *pbox = sarea_priv->boxes;
+	u32 fb_bpp;
+	int i;
+	DMALOCALS;
+
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+	switch ( dev_priv->fb_bpp ) {
+	case 16:
+		fb_bpp = MACH64_DATATYPE_RGB565;
+		break;
+	case 32:
+	default:
+		fb_bpp = MACH64_DATATYPE_ARGB8888;
+		break;
+	}
+
+	if ( !nbox ) 
+		return 0;
+
+	DMAGETPTR( filp, dev_priv, 13 + nbox * 4 ); /* returns on failure to get buffer */
+
+	DMAOUTREG( MACH64_Z_CNTL, 0 );
+	DMAOUTREG( MACH64_SCALE_3D_CNTL, 0 );
+
+	DMAOUTREG( MACH64_SC_LEFT_RIGHT, 0 | ( 8191 << 16 ) ); /* no scissor */
+	DMAOUTREG( MACH64_SC_TOP_BOTTOM, 0 | ( 16383 << 16 ) );
+
+	DMAOUTREG( MACH64_CLR_CMP_CNTL, 0 );
+	DMAOUTREG( MACH64_GUI_TRAJ_CNTL, (MACH64_DST_X_LEFT_TO_RIGHT |
+					  MACH64_DST_Y_TOP_TO_BOTTOM) );
+
+	DMAOUTREG( MACH64_DP_PIX_WIDTH, ((fb_bpp << 0) |
+					 (fb_bpp << 4) |
+					 (fb_bpp << 8) |
+					 (fb_bpp << 16) |
+					 (fb_bpp << 28)) );
+
+	DMAOUTREG( MACH64_DP_WRITE_MASK, 0xffffffff );
+	DMAOUTREG( MACH64_DP_MIX, (MACH64_BKGD_MIX_D |
+				   MACH64_FRGD_MIX_S) );
+	DMAOUTREG( MACH64_DP_SRC, (MACH64_BKGD_SRC_BKGD_CLR |
+				   MACH64_FRGD_SRC_BLIT |
+				   MACH64_MONO_SRC_ONE) );
+
+	DMAOUTREG( MACH64_SRC_OFF_PITCH, dev_priv->back_offset_pitch );
+	DMAOUTREG( MACH64_DST_OFF_PITCH, dev_priv->front_offset_pitch );
+
+	for ( i = 0 ; i < nbox ; i++ ) {
+		int x = pbox[i].x1;
+		int y = pbox[i].y1;
+		int w = pbox[i].x2 - x;
+		int h = pbox[i].y2 - y;
+
+		DRM_DEBUG( "dispatch swap %d,%d-%d,%d\n",
+			  pbox[i].x1, pbox[i].y1,
+			  pbox[i].x2, pbox[i].y2 );
+
+		DMAOUTREG( MACH64_SRC_WIDTH1, w );
+		DMAOUTREG( MACH64_SRC_Y_X, (x << 16) | y );
+		DMAOUTREG( MACH64_DST_Y_X, (x << 16) | y );
+		DMAOUTREG( MACH64_DST_WIDTH_HEIGHT, (h << 16) | w );
+
+	}
+
+	DMAADVANCE( dev_priv, 1 );
+
+	if (dev_priv->driver_mode == MACH64_MODE_DMA_ASYNC) {
+		for (i = 0; i < MACH64_MAX_QUEUED_FRAMES - 1; i++) {
+			dev_priv->frame_ofs[i] = dev_priv->frame_ofs[i+1];
+		}
+		dev_priv->frame_ofs[i] = GETRINGOFFSET();
+
+		dev_priv->sarea_priv->frames_queued++;
+	}
+
+	return 0;
+}
+
+static int mach64_do_get_frames_queued( drm_mach64_private_t *dev_priv )
+{
+	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;
+	drm_mach64_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	int i, start;
+	u32 head, tail, ofs;
+
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+	if (sarea_priv->frames_queued == 0)
+		return 0;
+
+	tail = ring->tail;
+	mach64_ring_tick( dev_priv, ring );
+	head = ring->head;
+
+	start = ( MACH64_MAX_QUEUED_FRAMES - 
+		  DRM_MIN(MACH64_MAX_QUEUED_FRAMES, sarea_priv->frames_queued) );
+
+	if ( head == tail ) {
+		sarea_priv->frames_queued = 0;
+		for (i = start; i < MACH64_MAX_QUEUED_FRAMES; i++) {
+			dev_priv->frame_ofs[i] = ~0;
+		}
+		return 0;
+	}
+
+	for ( i = start; i < MACH64_MAX_QUEUED_FRAMES; i++ ) {
+		ofs = dev_priv->frame_ofs[i];
+		DRM_DEBUG( "frame_ofs[%d] ofs: %d\n", i, ofs );
+		if ( ofs == ~0 ||
+		     ( head < tail && (ofs < head || ofs >= tail) ) ||
+		     ( head > tail && (ofs < head && ofs >= tail) ) ) {
+			sarea_priv->frames_queued = (MACH64_MAX_QUEUED_FRAMES - 1) - i;
+			dev_priv->frame_ofs[i] = ~0;
+		}
+	}
+
+	return sarea_priv->frames_queued;
+}
+
+/* Copy and verify a client submited buffer.
+ * FIXME: Make an assembly optimized version
+ */
+static __inline__ int copy_and_verify_from_user( u32 *to, const u32 *from, unsigned long bytes )
+{
+	unsigned long n = bytes; /* dwords remaining in buffer */
+
+	if ( DRM_VERIFYAREA_READ( from, n ) ) {
+		DRM_ERROR( "%s: verify_area\n", __FUNCTION__ );
+		return DRM_ERR(EFAULT);
+	}
+
+	n >>= 2;
+
+	while ( n > 1 ) {
+		u32 data, reg, count;
+
+		if ( DRM_GET_USER_UNCHECKED( data, from++ ) ) {
+			DRM_ERROR( "%s: get_user\n", __FUNCTION__ );
+			return DRM_ERR(EFAULT);
+		}
+
+		n--;
+
+		reg = le32_to_cpu(data);
+		count = (reg >> 16) + 1;
+		if( count <= n ) {
+			n -= count;
+			reg &= 0xffff;
+
+			/* This is an exact match of Mach64's Setup Engine registers,
+			 * excluding SETUP_CNTL (1_C1).
+			 */
+			if( (reg >= 0x0190 && reg <  0x01c1) || 
+			    (reg >= 0x01ca && reg <= 0x01cf) ) {
+				*to++ = data;
+				if ( DRM_COPY_FROM_USER_UNCHECKED( to, from, count << 2 ) ) {
+					DRM_ERROR( "%s: copy_from_user\n", __FUNCTION__ );
+					return DRM_ERR(EFAULT);
+				}
+				to += count;
+			} else {
+				DRM_ERROR( "%s: Got bad command: 0x%04x\n", __FUNCTION__, reg );
+				return DRM_ERR(EACCES);
+			}
+
+			from += count;
+		} else {
+			DRM_ERROR( "%s: Got bad command count(=%u) dwords remaining=%lu\n", 
+				   __FUNCTION__, count, n );
+			return DRM_ERR(EINVAL);
+		}
+	}
+
+	if (n == 0)
+		return 0;
+	else {
+		DRM_ERROR( "%s: Bad buf->used(=%lu)\n", __FUNCTION__, bytes );
+		return DRM_ERR(EINVAL);
+	}
+}
+
+static int mach64_dma_dispatch_vertex( DRMFILE filp, drm_device_t *dev, int prim, void *buf, 
+				       unsigned long used, int discard )
+{
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	drm_mach64_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	drm_buf_t *copy_buf;
+	int done = 0;
+	int verify_ret = 0;
+	DMALOCALS;
+
+	DRM_DEBUG( "%s: buf=%p used=%lu nbox=%d\n",
+		   __FUNCTION__, buf, used, sarea_priv->nbox );
+
+	if ( used ) {
+		int ret = 0;
+		int i = 0;
+
+		copy_buf = mach64_freelist_get( dev_priv );
+		if (copy_buf == NULL) {
+			DRM_ERROR("%s: couldn't get buffer in DMAGETPTR\n",
+				   __FUNCTION__ );
+			return DRM_ERR(EAGAIN);
+		}
+
+		if ( (verify_ret = 
+		      copy_and_verify_from_user( GETBUFPTR( copy_buf ), buf, used )) == 0 ) {
+
+			copy_buf->used = used;
+
+			DMASETPTR( copy_buf );
+
+			if ( sarea_priv->dirty & ~MACH64_UPLOAD_CLIPRECTS ) {
+				ret = mach64_emit_state( filp, dev_priv );
+				if (ret < 0) return ret;
+			}
+ 
+			do {
+				/* Emit the next cliprect */
+				if ( i < sarea_priv->nbox ) {
+					ret = mach64_emit_cliprect(filp, dev_priv, 
+								   &sarea_priv->boxes[i]);
+					if ( ret < 0 ) {
+						/* failed to get buffer */
+						return ret;
+					} else if ( ret != 0 ) {
+						/* null intersection with scissor */
+						continue;
+					}
+				}
+				if ((i >= sarea_priv->nbox - 1))
+					done = 1;
+
+				/* Add the buffer to the DMA queue */
+				DMAADVANCE( dev_priv, done );
+
+			} while ( ++i < sarea_priv->nbox );
+		}
+
+		if (copy_buf->pending && !done) {
+			DMADISCARDBUF();
+		} else if (!done) {
+			/* This buffer wasn't used (no cliprects or verify failed), so place it back
+			 * on the free list
+			 */
+			struct list_head *ptr;
+			drm_mach64_freelist_t *entry;
+#if MACH64_EXTRA_CHECKING
+			list_for_each(ptr, &dev_priv->pending) {
+				entry = list_entry(ptr, drm_mach64_freelist_t, list);
+				if (copy_buf == entry->buf) {
+					DRM_ERROR( "%s: Trying to release a pending buf\n",
+						   __FUNCTION__ );
+					return DRM_ERR(EFAULT);
+				}
+			}
+#endif
+			ptr = dev_priv->placeholders.next;
+			entry = list_entry(ptr, drm_mach64_freelist_t, list);
+			copy_buf->pending = 0;
+			copy_buf->used = 0;
+			entry->buf = copy_buf;
+			entry->discard = 1;
+			list_del(ptr);
+			list_add_tail(ptr, &dev_priv->free_list);
+		}
+	}
+
+	sarea_priv->dirty &= ~MACH64_UPLOAD_CLIPRECTS;
+	sarea_priv->nbox = 0;
+
+	return verify_ret;
+}
+
+
+static int mach64_dma_dispatch_blit( DRMFILE filp, drm_device_t *dev,
+				     drm_mach64_blit_t *blit )
+{
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	drm_device_dma_t *dma = dev->dma;
+	int dword_shift, dwords;
+	drm_buf_t *buf;
+	DMALOCALS;
+
+	/* The compiler won't optimize away a division by a variable,
+	 * even if the only legal values are powers of two.  Thus, we'll
+	 * use a shift instead.
+	 */
+	switch ( blit->format ) {
+	case MACH64_DATATYPE_ARGB8888:
+		dword_shift = 0;
+		break;
+	case MACH64_DATATYPE_ARGB1555:
+	case MACH64_DATATYPE_RGB565:
+	case MACH64_DATATYPE_VYUY422:
+	case MACH64_DATATYPE_YVYU422:
+	case MACH64_DATATYPE_ARGB4444:
+		dword_shift = 1;
+		break;
+	case MACH64_DATATYPE_CI8:
+	case MACH64_DATATYPE_RGB8:
+		dword_shift = 2;
+		break;
+	default:
+		DRM_ERROR( "invalid blit format %d\n", blit->format );
+		return DRM_ERR(EINVAL);
+	}
+
+	/* Dispatch the blit buffer.
+	 */
+	buf = dma->buflist[blit->idx];
+	
+	if ( buf->filp != filp ) {
+		DRM_ERROR( "process %d (filp %p) using buffer with filp %p\n",
+			   DRM_CURRENTPID, filp, buf->filp );
+		return DRM_ERR(EINVAL);
+	}
+
+	if ( buf->pending ) {
+		DRM_ERROR( "sending pending buffer %d\n", blit->idx );
+		return DRM_ERR(EINVAL);
+	}
+
+	/* Set buf->used to the bytes of blit data based on the blit dimensions 
+	 * and verify the size.  When the setup is emitted to the buffer with 
+	 * the DMA* macros below, buf->used is incremented to include the bytes 
+	 * used for setup as well as the blit data.
+	 */
+	dwords =  (blit->width * blit->height) >> dword_shift;
+	buf->used = dwords << 2;
+	if ( buf->used <= 0 || 
+	     buf->used > MACH64_BUFFER_SIZE - MACH64_HOSTDATA_BLIT_OFFSET ) {
+		DRM_ERROR( "Invalid blit size: %d bytes\n", buf->used );
+		return DRM_ERR(EINVAL);
+	}
+
+	/* FIXME: Use a last buffer flag and reduce the state emitted for subsequent,
+	 * continuation buffers? 
+	 */
+
+	/* Blit via BM_HOSTDATA (gui-master) - like HOST_DATA[0-15], but doesn't require
+	 * a register command every 16 dwords.  State setup is added at the start of the 
+	 * buffer -- the client leaves space for this based on MACH64_HOSTDATA_BLIT_OFFSET
+	 */
+	DMASETPTR( buf );
+
+	DMAOUTREG( MACH64_Z_CNTL, 0 );
+	DMAOUTREG( MACH64_SCALE_3D_CNTL, 0 );
+
+	DMAOUTREG( MACH64_SC_LEFT_RIGHT, 0 | ( 8191 << 16 ) );	/* no scissor */
+	DMAOUTREG( MACH64_SC_TOP_BOTTOM, 0 | ( 16383 << 16 )  );
+
+	DMAOUTREG( MACH64_CLR_CMP_CNTL, 0 );			/* disable */
+	DMAOUTREG( MACH64_GUI_TRAJ_CNTL, 
+		   MACH64_DST_X_LEFT_TO_RIGHT 
+		   | MACH64_DST_Y_TOP_TO_BOTTOM );
+
+	DMAOUTREG( MACH64_DP_PIX_WIDTH,
+		   ( blit->format << 0 )			/* dst pix width */
+		   | ( blit->format << 4 )			/* composite pix width */
+		   | ( blit->format << 8 )			/* src pix width */
+		   | ( blit->format << 16 )			/* host data pix width */
+		   | ( blit->format << 28 )			/* scaler/3D pix width */
+		   );
+
+	DMAOUTREG( MACH64_DP_WRITE_MASK, 0xffffffff );		/* enable all planes */
+	DMAOUTREG( MACH64_DP_MIX, 
+		   MACH64_BKGD_MIX_D 
+		   | MACH64_FRGD_MIX_S );
+	DMAOUTREG( MACH64_DP_SRC, 
+		   MACH64_BKGD_SRC_BKGD_CLR 
+		   | MACH64_FRGD_SRC_HOST 
+		   | MACH64_MONO_SRC_ONE );
+
+	DMAOUTREG( MACH64_DST_OFF_PITCH, (blit->pitch << 22) | (blit->offset >> 3) );
+	DMAOUTREG( MACH64_DST_X_Y, (blit->y << 16) | blit->x );
+	DMAOUTREG( MACH64_DST_WIDTH_HEIGHT, (blit->height << 16) | blit->width );
+
+	DRM_DEBUG( "%s: %d bytes\n", __FUNCTION__, buf->used );
+
+	/* Add the buffer to the queue */
+	DMAADVANCEHOSTDATA( dev_priv );
+
+	return 0;
+}
+
+/* ================================================================
+ * IOCTL functions
+ */
+
+int mach64_dma_clear( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	drm_mach64_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	drm_mach64_clear_t clear;
+	int ret;
+
+	DRM_DEBUG( "%s: pid=%d\n", __FUNCTION__, DRM_CURRENTPID  );
+
+	LOCK_TEST_WITH_RETURN( dev, filp );
+	
+	DRM_COPY_FROM_USER_IOCTL( clear, (drm_mach64_clear_t *)data,
+			     sizeof(clear) );
+	
+	if ( sarea_priv->nbox > MACH64_NR_SAREA_CLIPRECTS )
+		sarea_priv->nbox = MACH64_NR_SAREA_CLIPRECTS;
+
+	ret = mach64_dma_dispatch_clear( filp, dev, clear.flags,
+					 clear.x, clear.y, clear.w, clear.h,
+					 clear.clear_color, clear.clear_depth );
+
+	/* Make sure we restore the 3D state next time.
+	 */
+	sarea_priv->dirty |= (MACH64_UPLOAD_CONTEXT |
+			      MACH64_UPLOAD_MISC);
+	return ret;
+}
+
+int mach64_dma_swap( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	drm_mach64_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	int ret;
+
+	DRM_DEBUG( "%s: pid=%d\n", __FUNCTION__, DRM_CURRENTPID );
+
+	LOCK_TEST_WITH_RETURN( dev, filp );
+	
+	if ( sarea_priv->nbox > MACH64_NR_SAREA_CLIPRECTS )
+		sarea_priv->nbox = MACH64_NR_SAREA_CLIPRECTS;
+
+	ret = mach64_dma_dispatch_swap( filp, dev );
+
+	/* Make sure we restore the 3D state next time.
+	 */
+	sarea_priv->dirty |= (MACH64_UPLOAD_CONTEXT |
+			      MACH64_UPLOAD_MISC);
+	return ret;
+}
+
+int mach64_dma_vertex( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	drm_mach64_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	drm_mach64_vertex_t vertex;
+
+	LOCK_TEST_WITH_RETURN( dev, filp );
+
+	if ( !dev_priv ) {
+		DRM_ERROR( "%s called with no initialization\n", __FUNCTION__ );
+		return DRM_ERR(EINVAL);
+	}
+
+	DRM_COPY_FROM_USER_IOCTL( vertex, (drm_mach64_vertex_t *)data,
+			     sizeof(vertex) );
+
+	DRM_DEBUG( "%s: pid=%d buf=%p used=%lu discard=%d\n",
+		   __FUNCTION__, DRM_CURRENTPID,
+		   vertex.buf, vertex.used, vertex.discard );
+
+	if ( vertex.prim < 0 ||
+	     vertex.prim > MACH64_PRIM_POLYGON ) {
+		DRM_ERROR( "buffer prim %d\n", vertex.prim );
+		return DRM_ERR(EINVAL);
+	}
+
+	if ( vertex.used > MACH64_BUFFER_SIZE || (vertex.used & 3) != 0) {
+		DRM_ERROR( "Invalid vertex buffer size: %lu bytes\n", vertex.used );
+		return DRM_ERR(EINVAL);
+	}
+
+	if ( sarea_priv->nbox > MACH64_NR_SAREA_CLIPRECTS )
+		sarea_priv->nbox = MACH64_NR_SAREA_CLIPRECTS;
+
+	return mach64_dma_dispatch_vertex( filp, dev, vertex.prim, vertex.buf, 
+					   vertex.used, vertex.discard );
+}
+
+int mach64_dma_blit( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_device_dma_t *dma = dev->dma;
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	drm_mach64_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	drm_mach64_blit_t blit;
+	int ret;
+
+	LOCK_TEST_WITH_RETURN( dev, filp );
+
+	DRM_COPY_FROM_USER_IOCTL( blit, (drm_mach64_blit_t *)data,
+			     sizeof(blit) );
+
+	DRM_DEBUG( "%s: pid=%d index=%d\n",
+		   __FUNCTION__, DRM_CURRENTPID, blit.idx );
+
+	if ( blit.idx < 0 || blit.idx >= dma->buf_count ) {
+		DRM_ERROR( "buffer index %d (of %d max)\n",
+			   blit.idx, dma->buf_count - 1 );
+		return DRM_ERR(EINVAL);
+	}
+
+	ret = mach64_dma_dispatch_blit( filp, dev, &blit );
+
+	/* Make sure we restore the 3D state next time.
+	 */
+	sarea_priv->dirty |= (MACH64_UPLOAD_CONTEXT |
+			      MACH64_UPLOAD_MISC | 
+			      MACH64_UPLOAD_CLIPRECTS);
+
+	return ret;
+}
+
+int mach64_get_param( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_mach64_private_t *dev_priv = dev->dev_private;
+	drm_mach64_getparam_t param;
+	int value;
+
+	DRM_DEBUG( "%s\n", __FUNCTION__ );
+
+	if ( !dev_priv ) {
+		DRM_ERROR( "%s called with no initialization\n", __FUNCTION__ );
+		return DRM_ERR(EINVAL);
+	}
+
+	DRM_COPY_FROM_USER_IOCTL( param, (drm_mach64_getparam_t *)data,
+	    sizeof(param) );
+
+	switch ( param.param ) {
+	case MACH64_PARAM_FRAMES_QUEUED:
+		/* Needs lock since it calls mach64_ring_tick() */
+		LOCK_TEST_WITH_RETURN( dev, filp );
+		value = mach64_do_get_frames_queued( dev_priv );
+		break;
+	case MACH64_PARAM_IRQ_NR:
+		value = dev->irq;
+		break;
+	default:
+		return DRM_ERR(EINVAL);
+	}
+
+	if ( DRM_COPY_TO_USER( param.value, &value, sizeof(int) ) ) {
+		DRM_ERROR( "copy_to_user\n" );
+		return DRM_ERR(EFAULT);
+	}
+
+	return 0;
+}