Merge refs/heads/drm-latest from master.kernel.org:/pub/scm/linux/kernel/git/airlied/drm-2.6

57 files changed, 7407 insertions, 3243 deletions

diff --git a/drivers/char/drm/Kconfig b/drivers/char/drm/Kconfig
index 123417e43040..56ace9d5e2ae 100644
--- a/drivers/char/drm/Kconfig
+++ b/drivers/char/drm/Kconfig
@@ -23,13 +23,6 @@ config DRM_TDFX
 	  Choose this option if you have a 3dfx Banshee or Voodoo3 (or later),
 	  graphics card.  If M is selected, the module will be called tdfx.
 
-config DRM_GAMMA
-	tristate "3dlabs GMX 2000"
-	depends on DRM && BROKEN
-	help
-	  This is the old gamma driver, please tell me if it might actually
-	  work.
-
 config DRM_R128
 	tristate "ATI Rage 128"
 	depends on DRM && PCI
@@ -82,7 +75,7 @@ endchoice
 
 config DRM_MGA
 	tristate "Matrox g200/g400"
-	depends on DRM && AGP
+	depends on DRM
 	help
 	  Choose this option if you have a Matrox G200, G400 or G450 graphics
 	  card.  If M is selected, the module will be called mga.  AGP
@@ -103,3 +96,10 @@ 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_SAVAGE
+	tristate "Savage video cards"
+	depends on DRM
+	help
+	  Choose this option if you have a Savage3D/4/SuperSavage/Pro/Twister
+	  chipset. If M is selected the module will be called savage.
+
diff --git a/drivers/char/drm/Makefile b/drivers/char/drm/Makefile
index ddd941045b1f..e41060c76226 100644
--- a/drivers/char/drm/Makefile
+++ b/drivers/char/drm/Makefile
@@ -8,16 +8,16 @@ drm-objs    :=	drm_auth.o drm_bufs.o drm_context.o drm_dma.o drm_drawable.o \
 		drm_agpsupport.o drm_scatter.o ati_pcigart.o drm_pci.o \
 		drm_sysfs.o
 
-gamma-objs  := gamma_drv.o gamma_dma.o
 tdfx-objs   := tdfx_drv.o
 r128-objs   := r128_drv.o r128_cce.o r128_state.o r128_irq.o
 mga-objs    := mga_drv.o mga_dma.o mga_state.o mga_warp.o mga_irq.o
 i810-objs   := i810_drv.o i810_dma.o
 i830-objs   := i830_drv.o i830_dma.o i830_irq.o
 i915-objs   := i915_drv.o i915_dma.o i915_irq.o i915_mem.o
-radeon-objs := radeon_drv.o radeon_cp.o radeon_state.o radeon_mem.o radeon_irq.o
+radeon-objs := radeon_drv.o radeon_cp.o radeon_state.o radeon_mem.o radeon_irq.o r300_cmdbuf.o
 ffb-objs    := ffb_drv.o ffb_context.o
 sis-objs    := sis_drv.o sis_ds.o sis_mm.o
+savage-objs := savage_drv.o savage_bci.o savage_state.o
 via-objs    := via_irq.o via_drv.o via_ds.o via_map.o via_mm.o via_dma.o via_verifier.o via_video.o
 
 ifeq ($(CONFIG_COMPAT),y)
@@ -29,7 +29,6 @@ i915-objs   += i915_ioc32.o
 endif
 
 obj-$(CONFIG_DRM)	+= drm.o
-obj-$(CONFIG_DRM_GAMMA) += gamma.o
 obj-$(CONFIG_DRM_TDFX)	+= tdfx.o
 obj-$(CONFIG_DRM_R128)	+= r128.o
 obj-$(CONFIG_DRM_RADEON)+= radeon.o
@@ -39,5 +38,7 @@ obj-$(CONFIG_DRM_I830)	+= i830.o
 obj-$(CONFIG_DRM_I915)  += i915.o
 obj-$(CONFIG_DRM_FFB)   += ffb.o
 obj-$(CONFIG_DRM_SIS)   += sis.o
+obj-$(CONFIG_DRM_SAVAGE)+= savage.o
 obj-$(CONFIG_DRM_VIA)	+=via.o
 
+
diff --git a/drivers/char/drm/drm.h b/drivers/char/drm/drm.h
index e8371dd87fbc..fc6598a81acd 100644
--- a/drivers/char/drm/drm.h
+++ b/drivers/char/drm/drm.h
@@ -98,7 +98,7 @@
 #define _DRM_LOCKING_CONTEXT(lock) ((lock) & ~(_DRM_LOCK_HELD|_DRM_LOCK_CONT))
 
 
-typedef unsigned long drm_handle_t;
+typedef unsigned int  drm_handle_t;
 typedef unsigned int  drm_context_t;
 typedef unsigned int  drm_drawable_t;
 typedef unsigned int  drm_magic_t;
@@ -209,7 +209,8 @@ typedef enum drm_map_type {
 	_DRM_REGISTERS	    = 1,  /**< no caching, no core dump */
 	_DRM_SHM	    = 2,  /**< shared, cached */
 	_DRM_AGP            = 3,  /**< AGP/GART */
-	_DRM_SCATTER_GATHER = 4	  /**< Scatter/gather memory for PCI DMA */
+	_DRM_SCATTER_GATHER = 4,  /**< Scatter/gather memory for PCI DMA */
+	_DRM_CONSISTENT     = 5,  /**< Consistent memory for PCI DMA */
 } drm_map_type_t;
 
 
@@ -368,7 +369,8 @@ typedef struct drm_buf_desc {
 	enum {
 		_DRM_PAGE_ALIGN = 0x01, /**< Align on page boundaries for DMA */
 		_DRM_AGP_BUFFER = 0x02, /**< Buffer is in AGP space */
-		_DRM_SG_BUFFER  = 0x04  /**< Scatter/gather memory buffer */
+		_DRM_SG_BUFFER  = 0x04, /**< Scatter/gather memory buffer */
+		_DRM_FB_BUFFER  = 0x08  /**< Buffer is in frame buffer */
 	}	      flags;
 	unsigned long agp_start; /**< 
 				  * Start address of where the AGP buffers are
diff --git a/drivers/char/drm/drmP.h b/drivers/char/drm/drmP.h
index 5df09cc8c6db..6f98701dfe15 100644
--- a/drivers/char/drm/drmP.h
+++ b/drivers/char/drm/drmP.h
@@ -53,7 +53,6 @@
 #include <linux/init.h>
 #include <linux/file.h>
 #include <linux/pci.h>
-#include <linux/version.h>
 #include <linux/jiffies.h>
 #include <linux/smp_lock.h>	/* For (un)lock_kernel */
 #include <linux/mm.h>
@@ -96,6 +95,7 @@
 #define DRIVER_IRQ_SHARED  0x80
 #define DRIVER_IRQ_VBL     0x100
 #define DRIVER_DMA_QUEUE   0x200
+#define DRIVER_FB_DMA      0x400
 
 /***********************************************************************/
 /** \name Begin the DRM... */
@@ -160,36 +160,7 @@
 #define pte_unmap(pte)
 #endif
 
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,4,19)
-static inline struct page * vmalloc_to_page(void * vmalloc_addr)
-{
-	unsigned long addr = (unsigned long) vmalloc_addr;
-	struct page *page = NULL;
-	pgd_t *pgd = pgd_offset_k(addr);
-	pmd_t *pmd;
-	pte_t *ptep, pte;
-  
-	if (!pgd_none(*pgd)) {
-		pmd = pmd_offset(pgd, addr);
-		if (!pmd_none(*pmd)) {
-			preempt_disable();
-			ptep = pte_offset_map(pmd, addr);
-			pte = *ptep;
-			if (pte_present(pte))
-				page = pte_page(pte);
-			pte_unmap(ptep);
-			preempt_enable();
-		}
-	}
-	return page;
-}
-#endif
-
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
-#define DRM_RPR_ARG(vma)
-#else
 #define DRM_RPR_ARG(vma) vma,
-#endif
 
 #define VM_OFFSET(vma) ((vma)->vm_pgoff << PAGE_SHIFT)
 
@@ -474,7 +445,8 @@ typedef struct drm_device_dma {
 	unsigned long	  byte_count;
 	enum {
 		_DRM_DMA_USE_AGP = 0x01,
-		_DRM_DMA_USE_SG  = 0x02
+		_DRM_DMA_USE_SG  = 0x02,
+		_DRM_DMA_USE_FB  = 0x04
 	} flags;
 
 } drm_device_dma_t;
@@ -525,12 +497,19 @@ typedef struct drm_sigdata {
 	drm_hw_lock_t *lock;
 } drm_sigdata_t;
 
+typedef struct drm_dma_handle {
+	dma_addr_t busaddr;
+	void *vaddr;
+	size_t size;
+} drm_dma_handle_t;
+
 /**
  * Mappings list
  */
 typedef struct drm_map_list {
 	struct list_head	head;	/**< list head */
 	drm_map_t		*map;	/**< mapping */
+	unsigned int user_token;
 } drm_map_list_t;
 
 typedef drm_map_t drm_local_map_t;
@@ -578,7 +557,22 @@ struct drm_driver {
  	int (*kernel_context_switch)(struct drm_device *dev, int old, int new);
 	void (*kernel_context_switch_unlock)(struct drm_device *dev, drm_lock_t *lock);
 	int (*vblank_wait)(struct drm_device *dev, unsigned int *sequence);
+	
+	/**
+	 * Called by \c drm_device_is_agp.  Typically used to determine if a
+	 * card is really attached to AGP or not.
+	 *
+	 * \param dev  DRM device handle
+	 *
+	 * \returns
+	 * One of three values is returned depending on whether or not the
+	 * card is absolutely \b not AGP (return of 0), absolutely \b is AGP
+	 * (return of 1), or may or may not be AGP (return of 2).
+	 */
+	int (*device_is_agp) (struct drm_device * dev);
+
 	/* these have to be filled in */
+  
  	int (*postinit)(struct drm_device *, unsigned long flags);
 	irqreturn_t (*irq_handler)( DRM_IRQ_ARGS );
  	void (*irq_preinstall)(struct drm_device *dev);
@@ -722,12 +716,8 @@ typedef struct drm_device {
 	int               pci_slot;	/**< PCI slot number */
 	int               pci_func;	/**< PCI function number */
 #ifdef __alpha__
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,4,3)
-	struct pci_controler *hose;
-#else
 	struct pci_controller *hose;
 #endif
-#endif
 	drm_sg_mem_t      *sg;  /**< Scatter gather memory */
 	unsigned long     *ctx_bitmap;	/**< context bitmap */
 	void		  *dev_private; /**< device private data */
@@ -736,6 +726,7 @@ typedef struct drm_device {
 
 	struct            drm_driver *driver;
 	drm_local_map_t   *agp_buffer_map;
+	unsigned int agp_buffer_token;
 	drm_head_t primary;		/**< primary screen head */
 } drm_device_t;
 
@@ -806,7 +797,7 @@ extern void	     *drm_ioremap_nocache(unsigned long offset, unsigned long size,
 					   drm_device_t *dev);
 extern void	     drm_ioremapfree(void *pt, unsigned long size, drm_device_t *dev);
 
-extern DRM_AGP_MEM   *drm_alloc_agp(struct agp_bridge_data *bridge, int pages, u32 type);
+extern DRM_AGP_MEM   *drm_alloc_agp(drm_device_t *dev, int pages, u32 type);
 extern int           drm_free_agp(DRM_AGP_MEM *handle, int pages);
 extern int           drm_bind_agp(DRM_AGP_MEM *handle, unsigned int start);
 extern int           drm_unbind_agp(DRM_AGP_MEM *handle);
@@ -881,11 +872,19 @@ extern int	     drm_lock_free(drm_device_t *dev,
 				    unsigned int context);
 
 				/* Buffer management support (drm_bufs.h) */
+extern int drm_addbufs_agp(drm_device_t *dev, drm_buf_desc_t *request);
+extern int drm_addbufs_pci(drm_device_t *dev, drm_buf_desc_t *request);
+extern int drm_addmap(drm_device_t *dev, unsigned int offset,
+		      unsigned int size, drm_map_type_t type,
+		      drm_map_flags_t flags, drm_local_map_t **map_ptr);
+extern int drm_addmap_ioctl(struct inode *inode, struct file *filp,
+			    unsigned int cmd, unsigned long arg);
+extern int drm_rmmap(drm_device_t *dev, drm_local_map_t *map);
+extern int drm_rmmap_locked(drm_device_t *dev, drm_local_map_t *map);
+extern int drm_rmmap_ioctl(struct inode *inode, struct file *filp,
+			   unsigned int cmd, unsigned long arg);
+
 extern int	     drm_order( unsigned long size );
-extern int	     drm_addmap( struct inode *inode, struct file *filp,
-				  unsigned int cmd, unsigned long arg );
-extern int	     drm_rmmap( struct inode *inode, struct file *filp,
-				 unsigned int cmd, unsigned long arg );
 extern int	     drm_addbufs( struct inode *inode, struct file *filp,
 				   unsigned int cmd, unsigned long arg );
 extern int	     drm_infobufs( struct inode *inode, struct file *filp,
@@ -896,6 +895,10 @@ extern int	     drm_freebufs( struct inode *inode, struct file *filp,
 				    unsigned int cmd, unsigned long arg );
 extern int	     drm_mapbufs( struct inode *inode, struct file *filp,
 				   unsigned int cmd, unsigned long arg );
+extern unsigned long drm_get_resource_start(drm_device_t *dev,
+					    unsigned int resource);
+extern unsigned long drm_get_resource_len(drm_device_t *dev,
+					  unsigned int resource);
 
 				/* DMA support (drm_dma.h) */
 extern int	     drm_dma_setup(drm_device_t *dev);
@@ -919,15 +922,18 @@ extern void          drm_vbl_send_signals( drm_device_t *dev );
 
 				/* AGP/GART support (drm_agpsupport.h) */
 extern drm_agp_head_t *drm_agp_init(drm_device_t *dev);
-extern int            drm_agp_acquire(struct inode *inode, struct file *filp,
-				       unsigned int cmd, unsigned long arg);
-extern void           drm_agp_do_release(drm_device_t *dev);
-extern int            drm_agp_release(struct inode *inode, struct file *filp,
-				       unsigned int cmd, unsigned long arg);
-extern int            drm_agp_enable(struct inode *inode, struct file *filp,
-				      unsigned int cmd, unsigned long arg);
-extern int            drm_agp_info(struct inode *inode, struct file *filp,
-				    unsigned int cmd, unsigned long arg);
+extern int drm_agp_acquire(drm_device_t * dev);
+extern int drm_agp_acquire_ioctl(struct inode *inode, struct file *filp,
+			   unsigned int cmd, unsigned long arg);
+extern int drm_agp_release(drm_device_t *dev);
+extern int drm_agp_release_ioctl(struct inode *inode, struct file *filp,
+			   unsigned int cmd, unsigned long arg);
+extern int drm_agp_enable(drm_device_t *dev, drm_agp_mode_t mode);
+extern int drm_agp_enable_ioctl(struct inode *inode, struct file *filp,
+			  unsigned int cmd, unsigned long arg);
+extern int drm_agp_info(drm_device_t * dev, drm_agp_info_t *info);
+extern int drm_agp_info_ioctl(struct inode *inode, struct file *filp,
+			unsigned int cmd, unsigned long arg);
 extern int            drm_agp_alloc(struct inode *inode, struct file *filp,
 				     unsigned int cmd, unsigned long arg);
 extern int            drm_agp_free(struct inode *inode, struct file *filp,
@@ -976,12 +982,10 @@ 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);
+extern drm_dma_handle_t *drm_pci_alloc(drm_device_t *dev, size_t size,
+				       size_t align, dma_addr_t maxaddr);
+extern void __drm_pci_free(drm_device_t *dev, drm_dma_handle_t *dmah);
+extern void drm_pci_free(drm_device_t *dev, drm_dma_handle_t *dmah);
 
 			       /* sysfs support (drm_sysfs.c) */
 struct drm_sysfs_class;
@@ -1012,17 +1016,26 @@ static __inline__ void drm_core_ioremapfree(struct drm_map *map, struct drm_devi
 		drm_ioremapfree( map->handle, map->size, dev );
 }
 
-static __inline__ struct drm_map *drm_core_findmap(struct drm_device *dev, unsigned long offset)
+static __inline__ struct drm_map *drm_core_findmap(struct drm_device *dev, unsigned int token)
 {
-	struct list_head *_list;
-	list_for_each( _list, &dev->maplist->head ) {
-		drm_map_list_t *_entry = list_entry( _list, drm_map_list_t, head );
-		if ( _entry->map &&
-		     _entry->map->offset == offset ) {
+	drm_map_list_t *_entry;
+	list_for_each_entry(_entry, &dev->maplist->head, head)
+		if (_entry->user_token == token)
 			return _entry->map;
+	return NULL;
+}
+
+static __inline__ int drm_device_is_agp(drm_device_t *dev)
+{
+	if ( dev->driver->device_is_agp != NULL ) {
+		int err = (*dev->driver->device_is_agp)( dev );
+	
+		if (err != 2) {
+			return err;
 		}
 	}
-	return NULL;
+
+	return pci_find_capability(dev->pdev, PCI_CAP_ID_AGP);
 }
 
 static __inline__ void drm_core_dropmap(struct drm_map *map)
diff --git a/drivers/char/drm/drm_agpsupport.c b/drivers/char/drm/drm_agpsupport.c
index 8d94c0b5fa44..8c215adcb4b2 100644
--- a/drivers/char/drm/drm_agpsupport.c
+++ b/drivers/char/drm/drm_agpsupport.c
@@ -37,7 +37,7 @@
 #if __OS_HAS_AGP
 
 /**
- * AGP information ioctl.
+ * Get AGP information.
  *
  * \param inode device inode.
  * \param filp file pointer.
@@ -48,51 +48,56 @@
  * Verifies the AGP device has been initialized and acquired and fills in the
  * drm_agp_info structure with the information in drm_agp_head::agp_info.
  */
-int drm_agp_info(struct inode *inode, struct file *filp,
-		  unsigned int cmd, unsigned long arg)
+int drm_agp_info(drm_device_t *dev, drm_agp_info_t *info)
 {
-	drm_file_t	 *priv	 = filp->private_data;
-	drm_device_t	 *dev	 = priv->head->dev;
 	DRM_AGP_KERN     *kern;
-	drm_agp_info_t   info;
 
 	if (!dev->agp || !dev->agp->acquired)
 		return -EINVAL;
 
 	kern                   = &dev->agp->agp_info;
-	info.agp_version_major = kern->version.major;
-	info.agp_version_minor = kern->version.minor;
-	info.mode              = kern->mode;
-	info.aperture_base     = kern->aper_base;
-	info.aperture_size     = kern->aper_size * 1024 * 1024;
-	info.memory_allowed    = kern->max_memory << PAGE_SHIFT;
-	info.memory_used       = kern->current_memory << PAGE_SHIFT;
-	info.id_vendor         = kern->device->vendor;
-	info.id_device         = kern->device->device;
-
-	if (copy_to_user((drm_agp_info_t __user *)arg, &info, sizeof(info)))
+	info->agp_version_major = kern->version.major;
+	info->agp_version_minor = kern->version.minor;
+	info->mode              = kern->mode;
+	info->aperture_base     = kern->aper_base;
+	info->aperture_size     = kern->aper_size * 1024 * 1024;
+	info->memory_allowed    = kern->max_memory << PAGE_SHIFT;
+	info->memory_used       = kern->current_memory << PAGE_SHIFT;
+	info->id_vendor         = kern->device->vendor;
+	info->id_device         = kern->device->device;
+
+	return 0;
+}
+EXPORT_SYMBOL(drm_agp_info);
+
+int drm_agp_info_ioctl(struct inode *inode, struct file *filp,
+		 unsigned int cmd, unsigned long arg)
+{
+	drm_file_t *priv = filp->private_data;
+	drm_device_t *dev = priv->head->dev;
+	drm_agp_info_t info;
+	int err;
+
+	err = drm_agp_info(dev, &info);
+	if (err)
+		return err;
+	
+	if (copy_to_user((drm_agp_info_t __user *) arg, &info, sizeof(info)))
 		return -EFAULT;
 	return 0;
 }
 
 /**
- * Acquire the AGP device (ioctl).
+ * Acquire the AGP device.
  *
- * \param inode device inode.
- * \param filp file pointer.
- * \param cmd command.
- * \param arg user argument.
+ * \param dev DRM device that is to acquire AGP
  * \return zero on success or a negative number on failure. 
  *
  * Verifies the AGP device hasn't been acquired before and calls
- * agp_acquire().
+ * \c agp_backend_acquire.
  */
-int drm_agp_acquire(struct inode *inode, struct file *filp,
-		     unsigned int cmd, unsigned long arg)
+int drm_agp_acquire(drm_device_t *dev)
 {
-	drm_file_t	 *priv	 = filp->private_data;
-	drm_device_t	 *dev	 = priv->head->dev;
-
 	if (!dev->agp)
 		return -ENODEV;
 	if (dev->agp->acquired)
@@ -102,9 +107,10 @@ int drm_agp_acquire(struct inode *inode, struct file *filp,
 	dev->agp->acquired = 1;
 	return 0;
 }
+EXPORT_SYMBOL(drm_agp_acquire);
 
 /**
- * Release the AGP device (ioctl).
+ * Acquire the AGP device (ioctl).
  *
  * \param inode device inode.
  * \param filp file pointer.
@@ -112,63 +118,80 @@ int drm_agp_acquire(struct inode *inode, struct file *filp,
  * \param arg user argument.
  * \return zero on success or a negative number on failure.
  *
- * Verifies the AGP device has been acquired and calls agp_backend_release().
+ * Verifies the AGP device hasn't been acquired before and calls
+ * \c agp_backend_acquire.
  */
-int drm_agp_release(struct inode *inode, struct file *filp,
-		     unsigned int cmd, unsigned long arg)
+int drm_agp_acquire_ioctl(struct inode *inode, struct file *filp,
+			  unsigned int cmd, unsigned long arg)
 {
-	drm_file_t	 *priv	 = filp->private_data;
-	drm_device_t	 *dev	 = priv->head->dev;
+	drm_file_t *priv = filp->private_data;
+	
+	return drm_agp_acquire( (drm_device_t *) priv->head->dev );
+}
 
+/**
+ * Release the AGP device.
+ *
+ * \param dev DRM device that is to release AGP
+ * \return zero on success or a negative number on failure.
+ *
+ * Verifies the AGP device has been acquired and calls \c agp_backend_release.
+ */
+int drm_agp_release(drm_device_t *dev)
+{
 	if (!dev->agp || !dev->agp->acquired)
 		return -EINVAL;
 	agp_backend_release(dev->agp->bridge);
 	dev->agp->acquired = 0;
 	return 0;
-
 }
+EXPORT_SYMBOL(drm_agp_release);
 
-/**
- * Release the AGP device.
- *
- * Calls agp_backend_release().
- */
-void drm_agp_do_release(drm_device_t *dev)
+int drm_agp_release_ioctl(struct inode *inode, struct file *filp,
+			  unsigned int cmd, unsigned long arg)
 {
-  agp_backend_release(dev->agp->bridge);
+	drm_file_t *priv = filp->private_data;
+	drm_device_t *dev = priv->head->dev;
+	
+	return drm_agp_release(dev);
 }
 
 /**
  * Enable the AGP bus.
  * 
- * \param inode device inode.
- * \param filp file pointer.
- * \param cmd command.
- * \param arg pointer to a drm_agp_mode structure.
+ * \param dev DRM device that has previously acquired AGP.
+ * \param mode Requested AGP mode.
  * \return zero on success or a negative number on failure.
  *
  * Verifies the AGP device has been acquired but not enabled, and calls
- * agp_enable().
+ * \c agp_enable.
  */
-int drm_agp_enable(struct inode *inode, struct file *filp,
-		    unsigned int cmd, unsigned long arg)
+int drm_agp_enable(drm_device_t *dev, drm_agp_mode_t mode)
 {
-	drm_file_t	 *priv	 = filp->private_data;
-	drm_device_t	 *dev	 = priv->head->dev;
-	drm_agp_mode_t   mode;
-
 	if (!dev->agp || !dev->agp->acquired)
 		return -EINVAL;
 
-	if (copy_from_user(&mode, (drm_agp_mode_t __user *)arg, sizeof(mode)))
-		return -EFAULT;
-
 	dev->agp->mode    = mode.mode;
 	agp_enable(dev->agp->bridge, mode.mode);
 	dev->agp->base    = dev->agp->agp_info.aper_base;
 	dev->agp->enabled = 1;
 	return 0;
 }
+EXPORT_SYMBOL(drm_agp_enable);
+
+int drm_agp_enable_ioctl(struct inode *inode, struct file *filp,
+		   unsigned int cmd, unsigned long arg)
+{
+	drm_file_t *priv = filp->private_data;
+	drm_device_t *dev = priv->head->dev;
+	drm_agp_mode_t mode;
+
+
+	if (copy_from_user(&mode, (drm_agp_mode_t __user *) arg, sizeof(mode)))
+		return -EFAULT;
+
+	return drm_agp_enable(dev, mode);
+}
 
 /**
  * Allocate AGP memory.
@@ -206,7 +229,7 @@ int drm_agp_alloc(struct inode *inode, struct file *filp,
 	pages = (request.size + PAGE_SIZE - 1) / PAGE_SIZE;
 	type = (u32) request.type;
 
-	if (!(memory = drm_alloc_agp(dev->agp->bridge, pages, type))) {
+	if (!(memory = drm_alloc_agp(dev, pages, type))) {
 		drm_free(entry, sizeof(*entry), DRM_MEM_AGPLISTS);
 		return -ENOMEM;
 	}
@@ -403,13 +426,8 @@ drm_agp_head_t *drm_agp_init(drm_device_t *dev)
 		return NULL;
 	}
 	head->memory = NULL;
-#if LINUX_VERSION_CODE <= 0x020408
-	head->cant_use_aperture = 0;
-	head->page_mask = ~(0xfff);
-#else
 	head->cant_use_aperture = head->agp_info.cant_use_aperture;
 	head->page_mask = head->agp_info.page_mask;
-#endif
 
 	return head;
 }
@@ -436,6 +454,7 @@ int drm_agp_bind_memory(DRM_AGP_MEM *handle, off_t start)
 		return -EINVAL;
 	return agp_bind_memory(handle, start);
 }
+EXPORT_SYMBOL(drm_agp_bind_memory);
 
 /** Calls agp_unbind_memory() */
 int drm_agp_unbind_memory(DRM_AGP_MEM *handle)
diff --git a/drivers/char/drm/drm_bufs.c b/drivers/char/drm/drm_bufs.c
index 4c6191d231b8..e0743ebbe4bd 100644
--- a/drivers/char/drm/drm_bufs.c
+++ b/drivers/char/drm/drm_bufs.c
@@ -36,37 +36,69 @@
 #include <linux/vmalloc.h>
 #include "drmP.h"
 
-/**
- * Compute size order.  Returns the exponent of the smaller power of two which
- * is greater or equal to given number.
- * 
- * \param size size.
- * \return order.
- *
- * \todo Can be made faster.
- */
-int drm_order( unsigned long size )
+unsigned long drm_get_resource_start(drm_device_t *dev, unsigned int resource)
 {
-	int order;
-	unsigned long tmp;
+	return pci_resource_start(dev->pdev, resource);
+}
+EXPORT_SYMBOL(drm_get_resource_start);
 
-	for (order = 0, tmp = size >> 1; tmp; tmp >>= 1, order++)
-		;
+unsigned long drm_get_resource_len(drm_device_t *dev, unsigned int resource)
+{
+	return pci_resource_len(dev->pdev, resource);
+}
+EXPORT_SYMBOL(drm_get_resource_len);
 
-	if (size & (size - 1))
-		++order;
+static drm_local_map_t *drm_find_matching_map(drm_device_t *dev,
+					      drm_local_map_t *map)
+{
+	struct list_head *list;
 
-	return order;
+	list_for_each(list, &dev->maplist->head) {
+		drm_map_list_t *entry = list_entry(list, drm_map_list_t, head);
+		if (entry->map && map->type == entry->map->type &&
+		    entry->map->offset == map->offset) {
+			return entry->map;
+		}
+	}
+
+	return NULL;
 }
-EXPORT_SYMBOL(drm_order);
 
-#ifdef CONFIG_COMPAT
 /*
- * Used to allocate 32-bit handles for _DRM_SHM regions
- * The 0x10000000 value is chosen to be out of the way of
- * FB/register and GART physical addresses.
+ * Used to allocate 32-bit handles for mappings.
  */
-static unsigned int map32_handle = 0x10000000;
+#define START_RANGE 0x10000000
+#define END_RANGE 0x40000000
+
+#ifdef _LP64
+static __inline__ unsigned int HandleID(unsigned long lhandle, drm_device_t *dev) 
+{
+	static unsigned int map32_handle = START_RANGE;
+	unsigned int hash;
+
+	if (lhandle & 0xffffffff00000000) {
+		hash = map32_handle;
+		map32_handle += PAGE_SIZE;
+		if (map32_handle > END_RANGE)
+			map32_handle = START_RANGE;
+	} else 
+		hash = lhandle;
+
+	while (1) {
+		drm_map_list_t *_entry;
+		list_for_each_entry(_entry, &dev->maplist->head,head) {
+			if (_entry->user_token == hash)
+				break;
+		}
+		if (&_entry->head == &dev->maplist->head)
+			return hash;
+
+		hash += PAGE_SIZE;
+		map32_handle += PAGE_SIZE;
+	}
+}
+#else
+# define HandleID(x,dev) (unsigned int)(x)
 #endif
 
 /**
@@ -82,25 +114,23 @@ static unsigned int map32_handle = 0x10000000;
  * type.  Adds the map to the map list drm_device::maplist. Adds MTRR's where
  * applicable and if supported by the kernel.
  */
-int drm_addmap( struct inode *inode, struct file *filp,
-		 unsigned int cmd, unsigned long arg )
+int drm_addmap(drm_device_t * dev, unsigned int offset,
+	       unsigned int size, drm_map_type_t type,
+	       drm_map_flags_t flags, drm_local_map_t ** map_ptr)
 {
-	drm_file_t *priv = filp->private_data;
-	drm_device_t *dev = priv->head->dev;
 	drm_map_t *map;
-	drm_map_t __user *argp = (void __user *)arg;
 	drm_map_list_t *list;
-
-	if ( !(filp->f_mode & 3) ) return -EACCES; /* Require read/write */
+	drm_dma_handle_t *dmah;
+	drm_local_map_t *found_map;
 
 	map = drm_alloc( sizeof(*map), DRM_MEM_MAPS );
 	if ( !map )
 		return -ENOMEM;
 
-	if ( copy_from_user( map, argp, sizeof(*map) ) ) {
-		drm_free( map, sizeof(*map), DRM_MEM_MAPS );
-		return -EFAULT;
-	}
+	map->offset = offset;
+	map->size = size;
+	map->flags = flags;
+	map->type = type;
 
 	/* Only allow shared memory to be removable since we only keep enough
 	 * book keeping information about shared memory to allow for removal
@@ -122,7 +152,7 @@ int drm_addmap( struct inode *inode, struct file *filp,
 	switch ( map->type ) {
 	case _DRM_REGISTERS:
 	case _DRM_FRAME_BUFFER:
-#if !defined(__sparc__) && !defined(__alpha__) && !defined(__ia64__)
+#if !defined(__sparc__) && !defined(__alpha__) && !defined(__ia64__) && !defined(__powerpc64__) && !defined(__x86_64__)
 		if ( map->offset + map->size < map->offset ||
 		     map->offset < virt_to_phys(high_memory) ) {
 			drm_free( map, sizeof(*map), DRM_MEM_MAPS );
@@ -132,6 +162,24 @@ int drm_addmap( struct inode *inode, struct file *filp,
 #ifdef __alpha__
 		map->offset += dev->hose->mem_space->start;
 #endif
+		/* Some drivers preinitialize some maps, without the X Server
+		 * needing to be aware of it.  Therefore, we just return success
+		 * when the server tries to create a duplicate map.
+		 */
+		found_map = drm_find_matching_map(dev, map);
+		if (found_map != NULL) {
+			if (found_map->size != map->size) {
+				DRM_DEBUG("Matching maps of type %d with "
+				   "mismatched sizes, (%ld vs %ld)\n",
+				    map->type, map->size, found_map->size);
+				found_map->size = map->size;
+			}
+
+			drm_free(map, sizeof(*map), DRM_MEM_MAPS);
+			*map_ptr = found_map;
+			return 0;
+		}
+
 		if (drm_core_has_MTRR(dev)) {
 			if ( map->type == _DRM_FRAME_BUFFER ||
 			     (map->flags & _DRM_WRITE_COMBINING) ) {
@@ -178,9 +226,22 @@ int drm_addmap( struct inode *inode, struct file *filp,
 			drm_free(map, sizeof(*map), DRM_MEM_MAPS);
 			return -EINVAL;
 		}
-		map->offset += dev->sg->handle;
+		map->offset += (unsigned long)dev->sg->virtual;
+		break;
+	case _DRM_CONSISTENT: 
+		/* dma_addr_t is 64bit on i386 with CONFIG_HIGHMEM64G,
+		 * As we're limiting the address to 2^32-1 (or less),
+		 * casting it down to 32 bits is no problem, but we
+		 * need to point to a 64bit variable first. */
+		dmah = drm_pci_alloc(dev, map->size, map->size, 0xffffffffUL);
+		if (!dmah) {
+			drm_free(map, sizeof(*map), DRM_MEM_MAPS);
+			return -ENOMEM;
+		}
+		map->handle = dmah->vaddr;
+		map->offset = (unsigned long)dmah->busaddr;
+		kfree(dmah);
 		break;
-
 	default:
 		drm_free( map, sizeof(*map), DRM_MEM_MAPS );
 		return -EINVAL;
@@ -196,17 +257,56 @@ int drm_addmap( struct inode *inode, struct file *filp,
 
 	down(&dev->struct_sem);
 	list_add(&list->head, &dev->maplist->head);
-#ifdef CONFIG_COMPAT
-	/* Assign a 32-bit handle for _DRM_SHM mappings */
+	/* Assign a 32-bit handle */
 	/* We do it here so that dev->struct_sem protects the increment */
-	if (map->type == _DRM_SHM)
-		map->offset = map32_handle += PAGE_SIZE;
-#endif
+	list->user_token = HandleID(map->type==_DRM_SHM
+				    ? (unsigned long)map->handle
+				    : map->offset, dev);
  	up(&dev->struct_sem);
 
-	if ( copy_to_user( argp, map, sizeof(*map) ) )
+	*map_ptr = map;
+	return 0;
+}
+EXPORT_SYMBOL(drm_addmap);
+
+int drm_addmap_ioctl(struct inode *inode, struct file *filp,
+		     unsigned int cmd, unsigned long arg)
+{
+	drm_file_t *priv = filp->private_data;
+	drm_device_t *dev = priv->head->dev;
+	drm_map_t map;
+	drm_map_t *map_ptr;
+	drm_map_t __user *argp = (void __user *)arg;
+	int err;
+	unsigned long handle = 0;
+
+	if (!(filp->f_mode & 3))
+		return -EACCES;	/* Require read/write */
+
+	if (copy_from_user(& map, argp, sizeof(map))) {
+		return -EFAULT;
+	}
+
+	err = drm_addmap(dev, map.offset, map.size, map.type, map.flags,
+			 &map_ptr);
+
+	if (err) {
+		return err;
+	}
+
+	{
+		drm_map_list_t *_entry;
+		list_for_each_entry(_entry, &dev->maplist->head, head) {
+			if (_entry->map == map_ptr)
+				handle = _entry->user_token;
+		}
+		if (!handle)
+			return -EFAULT;
+	}
+
+	if (copy_to_user(argp, map_ptr, sizeof(*map_ptr)))
 		return -EFAULT;
-	if (copy_to_user(&argp->handle, &map->offset, sizeof(map->offset)))
+	if (put_user(handle, &argp->handle))
 		return -EFAULT;
 	return 0;
 }
@@ -226,81 +326,138 @@ int drm_addmap( struct inode *inode, struct file *filp,
  * its being used, and free any associate resource (such as MTRR's) if it's not
  * being on use.
  *
- * \sa addmap().
+ * \sa drm_addmap
  */
-int drm_rmmap(struct inode *inode, struct file *filp,
-	       unsigned int cmd, unsigned long arg)
+int drm_rmmap_locked(drm_device_t *dev, drm_local_map_t *map)
 {
-	drm_file_t	*priv	= filp->private_data;
-	drm_device_t	*dev	= priv->head->dev;
 	struct list_head *list;
 	drm_map_list_t *r_list = NULL;
-	drm_vma_entry_t *pt, *prev;
-	drm_map_t *map;
+	drm_dma_handle_t dmah;
+
+	/* Find the list entry for the map and remove it */
+	list_for_each(list, &dev->maplist->head) {
+		r_list = list_entry(list, drm_map_list_t, head);
+
+		if (r_list->map == map) {
+			list_del(list);
+			drm_free(list, sizeof(*list), DRM_MEM_MAPS);
+			break;
+		}
+	}
+
+	/* List has wrapped around to the head pointer, or it's empty and we
+	 * didn't find anything.
+	 */
+	if (list == (&dev->maplist->head)) {
+		return -EINVAL;
+	}
+
+	switch (map->type) {
+	case _DRM_REGISTERS:
+		drm_ioremapfree(map->handle, map->size, dev);
+		/* FALLTHROUGH */
+	case _DRM_FRAME_BUFFER:
+		if (drm_core_has_MTRR(dev) && map->mtrr >= 0) {
+			int retcode;
+			retcode = mtrr_del(map->mtrr, map->offset,
+					   map->size);
+			DRM_DEBUG ("mtrr_del=%d\n", retcode);
+		}
+		break;
+	case _DRM_SHM:
+		vfree(map->handle);
+		break;
+	case _DRM_AGP:
+	case _DRM_SCATTER_GATHER:
+		break;
+	case _DRM_CONSISTENT:
+		dmah.vaddr = map->handle;
+		dmah.busaddr = map->offset;
+		dmah.size = map->size;
+		__drm_pci_free(dev, &dmah);
+		break;
+	}
+	drm_free(map, sizeof(*map), DRM_MEM_MAPS);
+
+	return 0;
+}
+EXPORT_SYMBOL(drm_rmmap_locked);
+
+int drm_rmmap(drm_device_t *dev, drm_local_map_t *map)
+{
+	int ret;
+
+	down(&dev->struct_sem);
+	ret = drm_rmmap_locked(dev, map);
+	up(&dev->struct_sem);
+
+	return ret;
+}
+EXPORT_SYMBOL(drm_rmmap);
+
+/* The rmmap ioctl appears to be unnecessary.  All mappings are torn down on
+ * the last close of the device, and this is necessary for cleanup when things
+ * exit uncleanly.  Therefore, having userland manually remove mappings seems
+ * like a pointless exercise since they're going away anyway.
+ *
+ * One use case might be after addmap is allowed for normal users for SHM and
+ * gets used by drivers that the server doesn't need to care about.  This seems
+ * unlikely.
+ */
+int drm_rmmap_ioctl(struct inode *inode, struct file *filp,
+		    unsigned int cmd, unsigned long arg)
+{
+	drm_file_t *priv = filp->private_data;
+	drm_device_t *dev = priv->head->dev;
 	drm_map_t request;
-	int found_maps = 0;
+	drm_local_map_t *map = NULL;
+	struct list_head *list;
+	int ret;
 
-	if (copy_from_user(&request, (drm_map_t __user *)arg,
-			   sizeof(request))) {
+	if (copy_from_user(&request, (drm_map_t __user *)arg, sizeof(request))) {
 		return -EFAULT;
 	}
 
 	down(&dev->struct_sem);
-	list = &dev->maplist->head;
 	list_for_each(list, &dev->maplist->head) {
-		r_list = list_entry(list, drm_map_list_t, head);
+		drm_map_list_t *r_list = list_entry(list, drm_map_list_t, head);
 
-		if(r_list->map &&
-		   r_list->map->offset == (unsigned long) request.handle &&
-		   r_list->map->flags & _DRM_REMOVABLE) break;
+		if (r_list->map &&
+		    r_list->user_token == (unsigned long) request.handle &&
+		    r_list->map->flags & _DRM_REMOVABLE) {
+			map = r_list->map;
+			break;
+		}
 	}
 
 	/* List has wrapped around to the head pointer, or its empty we didn't
 	 * find anything.
 	 */
-	if(list == (&dev->maplist->head)) {
+	if (list == (&dev->maplist->head)) {
 		up(&dev->struct_sem);
 		return -EINVAL;
 	}
-	map = r_list->map;
-	list_del(list);
-	drm_free(list, sizeof(*list), DRM_MEM_MAPS);
 
-	for (pt = dev->vmalist, prev = NULL; pt; prev = pt, pt = pt->next) {
-		if (pt->vma->vm_private_data == map) found_maps++;
-	}
+	if (!map)
+		return -EINVAL;
 
-	if(!found_maps) {
-		switch (map->type) {
-		case _DRM_REGISTERS:
-		case _DRM_FRAME_BUFFER:
-		  if (drm_core_has_MTRR(dev)) {
-				if (map->mtrr >= 0) {
-					int retcode;
-					retcode = mtrr_del(map->mtrr,
-							   map->offset,
-							   map->size);
-					DRM_DEBUG("mtrr_del = %d\n", retcode);
-				}
-			}
-			drm_ioremapfree(map->handle, map->size, dev);
-			break;
-		case _DRM_SHM:
-			vfree(map->handle);
-			break;
-		case _DRM_AGP:
-		case _DRM_SCATTER_GATHER:
-			break;
-		}
-		drm_free(map, sizeof(*map), DRM_MEM_MAPS);
+	/* Register and framebuffer maps are permanent */
+	if ((map->type == _DRM_REGISTERS) || (map->type == _DRM_FRAME_BUFFER)) {
+		up(&dev->struct_sem);
+		return 0;
 	}
+
+	ret = drm_rmmap_locked(dev, map);
+
 	up(&dev->struct_sem);
-	return 0;
+
+	return ret;
 }
 
 /**
  * Cleanup after an error on one of the addbufs() functions.
  *
+ * \param dev DRM device.
  * \param entry buffer entry where the error occurred.
  *
  * Frees any pages and buffers associated with the given entry.
@@ -344,25 +501,19 @@ static void drm_cleanup_buf_error(drm_device_t *dev, drm_buf_entry_t *entry)
 
 #if __OS_HAS_AGP
 /**
- * Add AGP buffers for DMA transfers (ioctl).
+ * Add AGP buffers for DMA transfers.
  *
- * \param inode device inode.
- * \param filp file pointer.
- * \param cmd command.
- * \param arg pointer to a drm_buf_desc_t request.
+ * \param dev drm_device_t to which the buffers are to be added.
+ * \param request pointer to a drm_buf_desc_t describing the request.
  * \return zero on success or a negative number on failure.
  * 
  * After some sanity checks creates a drm_buf structure for each buffer and
  * reallocates the buffer list of the same size order to accommodate the new
  * buffers.
  */
-static int drm_addbufs_agp( struct inode *inode, struct file *filp,
-			    unsigned int cmd, unsigned long arg )
+int drm_addbufs_agp(drm_device_t *dev, drm_buf_desc_t *request)
 {
-	drm_file_t *priv = filp->private_data;
-	drm_device_t *dev = priv->head->dev;
 	drm_device_dma_t *dma = dev->dma;
-	drm_buf_desc_t request;
 	drm_buf_entry_t *entry;
 	drm_buf_t *buf;
 	unsigned long offset;
@@ -376,25 +527,20 @@ static int drm_addbufs_agp( struct inode *inode, struct file *filp,
 	int byte_count;
 	int i;
 	drm_buf_t **temp_buflist;
-	drm_buf_desc_t __user *argp = (void __user *)arg;
 
 	if ( !dma ) return -EINVAL;
 
-	if ( copy_from_user( &request, argp,
-			     sizeof(request) ) )
-		return -EFAULT;
-
-	count = request.count;
-	order = drm_order( request.size );
+	count = request->count;
+	order = drm_order(request->size);
 	size = 1 << order;
 
-	alignment  = (request.flags & _DRM_PAGE_ALIGN)
+	alignment  = (request->flags & _DRM_PAGE_ALIGN)
 		? PAGE_ALIGN(size) : size;
 	page_order = order - PAGE_SHIFT > 0 ? order - PAGE_SHIFT : 0;
 	total = PAGE_SIZE << page_order;
 
 	byte_count = 0;
-	agp_offset = dev->agp->base + request.agp_start;
+	agp_offset = dev->agp->base + request->agp_start;
 
 	DRM_DEBUG( "count:      %d\n",  count );
 	DRM_DEBUG( "order:      %d\n",  order );
@@ -508,26 +654,20 @@ static int drm_addbufs_agp( struct inode *inode, struct file *filp,
 
 	up( &dev->struct_sem );
 
-	request.count = entry->buf_count;
-	request.size = size;
-
-	if ( copy_to_user( argp, &request, sizeof(request) ) )
-		return -EFAULT;
+	request->count = entry->buf_count;
+	request->size = size;
 
 	dma->flags = _DRM_DMA_USE_AGP;
 
 	atomic_dec( &dev->buf_alloc );
 	return 0;
 }
+EXPORT_SYMBOL(drm_addbufs_agp);
 #endif /* __OS_HAS_AGP */
 
-static int drm_addbufs_pci( struct inode *inode, struct file *filp,
-			    unsigned int cmd, unsigned long arg )
+int drm_addbufs_pci(drm_device_t *dev, drm_buf_desc_t *request)
 {
-   	drm_file_t *priv = filp->private_data;
-	drm_device_t *dev = priv->head->dev;
 	drm_device_dma_t *dma = dev->dma;
-	drm_buf_desc_t request;
 	int count;
 	int order;
 	int size;
@@ -543,26 +683,22 @@ static int drm_addbufs_pci( struct inode *inode, struct file *filp,
 	int page_count;
 	unsigned long *temp_pagelist;
 	drm_buf_t **temp_buflist;
-	drm_buf_desc_t __user *argp = (void __user *)arg;
 
 	if (!drm_core_check_feature(dev, DRIVER_PCI_DMA)) return -EINVAL;
 	if ( !dma ) return -EINVAL;
 
-	if ( copy_from_user( &request, argp, sizeof(request) ) )
-		return -EFAULT;
-
-	count = request.count;
-	order = drm_order( request.size );
+	count = request->count;
+	order = drm_order(request->size);
 	size = 1 << order;
 
 	DRM_DEBUG( "count=%d, size=%d (%d), order=%d, queue_count=%d\n",
-		   request.count, request.size, size,
+		   request->count, request->size, size,
 		   order, dev->queue_count );
 
 	if ( order < DRM_MIN_ORDER || order > DRM_MAX_ORDER ) return -EINVAL;
 	if ( dev->queue_count ) return -EBUSY; /* Not while in use */
 
-	alignment = (request.flags & _DRM_PAGE_ALIGN)
+	alignment = (request->flags & _DRM_PAGE_ALIGN)
 		? PAGE_ALIGN(size) : size;
 	page_order = order - PAGE_SHIFT > 0 ? order - PAGE_SHIFT : 0;
 	total = PAGE_SIZE << page_order;
@@ -740,25 +876,18 @@ static int drm_addbufs_pci( struct inode *inode, struct file *filp,
 
 	up( &dev->struct_sem );
 
-	request.count = entry->buf_count;
-	request.size = size;
-
-	if ( copy_to_user( argp, &request, sizeof(request) ) )
-		return -EFAULT;
+	request->count = entry->buf_count;
+	request->size = size;
 
 	atomic_dec( &dev->buf_alloc );
 	return 0;
 
 }
+EXPORT_SYMBOL(drm_addbufs_pci);
 
-static int drm_addbufs_sg( struct inode *inode, struct file *filp,
-			   unsigned int cmd, unsigned long arg )
+static int drm_addbufs_sg(drm_device_t *dev, drm_buf_desc_t *request)
 {
-	drm_file_t *priv = filp->private_data;
-	drm_device_t *dev = priv->head->dev;
 	drm_device_dma_t *dma = dev->dma;
-	drm_buf_desc_t __user *argp = (void __user *)arg;
-	drm_buf_desc_t request;
 	drm_buf_entry_t *entry;
 	drm_buf_t *buf;
 	unsigned long offset;
@@ -777,20 +906,17 @@ static int drm_addbufs_sg( struct inode *inode, struct file *filp,
 	
 	if ( !dma ) return -EINVAL;
 
-	if ( copy_from_user( &request, argp, sizeof(request) ) )
-		return -EFAULT;
-
-	count = request.count;
-	order = drm_order( request.size );
+	count = request->count;
+	order = drm_order(request->size);
 	size = 1 << order;
 
-	alignment  = (request.flags & _DRM_PAGE_ALIGN)
+	alignment  = (request->flags & _DRM_PAGE_ALIGN)
 			? PAGE_ALIGN(size) : size;
 	page_order = order - PAGE_SHIFT > 0 ? order - PAGE_SHIFT : 0;
 	total = PAGE_SIZE << page_order;
 
 	byte_count = 0;
-	agp_offset = request.agp_start;
+	agp_offset = request->agp_start;
 
 	DRM_DEBUG( "count:      %d\n",  count );
 	DRM_DEBUG( "order:      %d\n",  order );
@@ -848,7 +974,8 @@ static int drm_addbufs_sg( struct inode *inode, struct file *filp,
 
 		buf->offset  = (dma->byte_count + offset);
 		buf->bus_address = agp_offset + offset;
-		buf->address = (void *)(agp_offset + offset + dev->sg->handle);
+		buf->address = (void *)(agp_offset + offset 
+					+ (unsigned long)dev->sg->virtual);
 		buf->next    = NULL;
 		buf->waiting = 0;
 		buf->pending = 0;
@@ -905,11 +1032,8 @@ static int drm_addbufs_sg( struct inode *inode, struct file *filp,
 
 	up( &dev->struct_sem );
 
-	request.count = entry->buf_count;
-	request.size = size;
-
-	if ( copy_to_user( argp, &request, sizeof(request) ) )
-		return -EFAULT;
+	request->count = entry->buf_count;
+	request->size = size;
 
 	dma->flags = _DRM_DMA_USE_SG;
 
@@ -917,6 +1041,161 @@ static int drm_addbufs_sg( struct inode *inode, struct file *filp,
 	return 0;
 }
 
+int drm_addbufs_fb(drm_device_t *dev, drm_buf_desc_t *request)
+{
+	drm_device_dma_t *dma = dev->dma;
+	drm_buf_entry_t *entry;
+	drm_buf_t *buf;
+	unsigned long offset;
+	unsigned long agp_offset;
+	int count;
+	int order;
+	int size;
+	int alignment;
+	int page_order;
+	int total;
+	int byte_count;
+	int i;
+	drm_buf_t **temp_buflist;
+
+	if (!drm_core_check_feature(dev, DRIVER_FB_DMA))
+		return -EINVAL;
+    
+	if (!dma)
+		return -EINVAL;
+
+	count = request->count;
+	order = drm_order(request->size);
+	size = 1 << order;
+
+	alignment = (request->flags & _DRM_PAGE_ALIGN)
+	    ? PAGE_ALIGN(size) : size;
+	page_order = order - PAGE_SHIFT > 0 ? order - PAGE_SHIFT : 0;
+	total = PAGE_SIZE << page_order;
+
+	byte_count = 0;
+	agp_offset = request->agp_start;
+
+	DRM_DEBUG("count:      %d\n", count);
+	DRM_DEBUG("order:      %d\n", order);
+	DRM_DEBUG("size:       %d\n", size);
+	DRM_DEBUG("agp_offset: %lu\n", agp_offset);
+	DRM_DEBUG("alignment:  %d\n", alignment);
+	DRM_DEBUG("page_order: %d\n", page_order);
+	DRM_DEBUG("total:      %d\n", total);
+
+	if (order < DRM_MIN_ORDER || order > DRM_MAX_ORDER)
+		return -EINVAL;
+	if (dev->queue_count)
+		return -EBUSY;	/* Not while in use */
+
+	spin_lock(&dev->count_lock);
+	if (dev->buf_use) {
+		spin_unlock(&dev->count_lock);
+		return -EBUSY;
+	}
+	atomic_inc(&dev->buf_alloc);
+	spin_unlock(&dev->count_lock);
+
+	down(&dev->struct_sem);
+	entry = &dma->bufs[order];
+	if (entry->buf_count) {
+		up(&dev->struct_sem);
+		atomic_dec(&dev->buf_alloc);
+		return -ENOMEM;	/* May only call once for each order */
+	}
+
+	if (count < 0 || count > 4096) {
+		up(&dev->struct_sem);
+		atomic_dec(&dev->buf_alloc);
+		return -EINVAL;
+	}
+
+	entry->buflist = drm_alloc(count * sizeof(*entry->buflist),
+				   DRM_MEM_BUFS);
+	if (!entry->buflist) {
+		up(&dev->struct_sem);
+		atomic_dec(&dev->buf_alloc);
+		return -ENOMEM;
+	}
+	memset(entry->buflist, 0, count * sizeof(*entry->buflist));
+
+	entry->buf_size = size;
+	entry->page_order = page_order;
+
+	offset = 0;
+
+	while (entry->buf_count < count) {
+		buf = &entry->buflist[entry->buf_count];
+		buf->idx = dma->buf_count + entry->buf_count;
+		buf->total = alignment;
+		buf->order = order;
+		buf->used = 0;
+
+		buf->offset = (dma->byte_count + offset);
+		buf->bus_address = agp_offset + offset;
+		buf->address = (void *)(agp_offset + offset);
+		buf->next = NULL;
+		buf->waiting = 0;
+		buf->pending = 0;
+		init_waitqueue_head(&buf->dma_wait);
+		buf->filp = NULL;
+
+		buf->dev_priv_size = dev->driver->dev_priv_size;
+		buf->dev_private = drm_alloc(buf->dev_priv_size, DRM_MEM_BUFS);
+		if (!buf->dev_private) {
+			/* Set count correctly so we free the proper amount. */
+			entry->buf_count = count;
+			drm_cleanup_buf_error(dev, entry);
+			up(&dev->struct_sem);
+			atomic_dec(&dev->buf_alloc);
+			return -ENOMEM;
+		}
+		memset(buf->dev_private, 0, buf->dev_priv_size);
+
+		DRM_DEBUG("buffer %d @ %p\n", entry->buf_count, buf->address);
+
+		offset += alignment;
+		entry->buf_count++;
+		byte_count += PAGE_SIZE << page_order;
+	}
+
+	DRM_DEBUG("byte_count: %d\n", byte_count);
+
+	temp_buflist = drm_realloc(dma->buflist,
+				   dma->buf_count * sizeof(*dma->buflist),
+				   (dma->buf_count + entry->buf_count)
+				   * sizeof(*dma->buflist), DRM_MEM_BUFS);
+	if (!temp_buflist) {
+		/* Free the entry because it isn't valid */
+		drm_cleanup_buf_error(dev, entry);
+		up(&dev->struct_sem);
+		atomic_dec(&dev->buf_alloc);
+		return -ENOMEM;
+	}
+	dma->buflist = temp_buflist;
+
+	for (i = 0; i < entry->buf_count; i++) {
+		dma->buflist[i + dma->buf_count] = &entry->buflist[i];
+	}
+
+	dma->buf_count += entry->buf_count;
+	dma->byte_count += byte_count;
+
+	DRM_DEBUG("dma->buf_count : %d\n", dma->buf_count);
+	DRM_DEBUG("entry->buf_count : %d\n", entry->buf_count);
+
+	up(&dev->struct_sem);
+
+	request->count = entry->buf_count;
+	request->size = size;
+
+	dma->flags = _DRM_DMA_USE_FB;
+
+	atomic_dec(&dev->buf_alloc);
+	return 0;
+}
+
 /**
  * Add buffers for DMA transfers (ioctl).
  *
@@ -937,6 +1216,7 @@ int drm_addbufs( struct inode *inode, struct file *filp,
 	drm_buf_desc_t request;
 	drm_file_t *priv = filp->private_data;
 	drm_device_t *dev = priv->head->dev;
+	int ret;
 	
 	if (!drm_core_check_feature(dev, DRIVER_HAVE_DMA))
 		return -EINVAL;
@@ -947,13 +1227,23 @@ int drm_addbufs( struct inode *inode, struct file *filp,
 
 #if __OS_HAS_AGP
 	if ( request.flags & _DRM_AGP_BUFFER )
-		return drm_addbufs_agp( inode, filp, cmd, arg );
+		ret=drm_addbufs_agp(dev, &request);
 	else
 #endif
 	if ( request.flags & _DRM_SG_BUFFER )
-		return drm_addbufs_sg( inode, filp, cmd, arg );
+		ret=drm_addbufs_sg(dev, &request);
+	else if ( request.flags & _DRM_FB_BUFFER)
+		ret=drm_addbufs_fb(dev, &request);
 	else
-		return drm_addbufs_pci( inode, filp, cmd, arg );
+		ret=drm_addbufs_pci(dev, &request);
+
+	if (ret==0) {
+		if (copy_to_user((void __user *)arg, &request,
+				 sizeof(request))) {
+			ret = -EFAULT;
+		}
+	}
+	return ret;
 }
 
 
@@ -1196,43 +1486,31 @@ int drm_mapbufs( struct inode *inode, struct file *filp,
 		return -EFAULT;
 
 	if ( request.count >= dma->buf_count ) {
-		if ((drm_core_has_AGP(dev) && (dma->flags & _DRM_DMA_USE_AGP)) ||
-		    (drm_core_check_feature(dev, DRIVER_SG) && (dma->flags & _DRM_DMA_USE_SG)) ) {
+		if ((drm_core_has_AGP(dev) && (dma->flags & _DRM_DMA_USE_AGP))
+		    || (drm_core_check_feature(dev, DRIVER_SG) 
+			&& (dma->flags & _DRM_DMA_USE_SG))
+		    || (drm_core_check_feature(dev, DRIVER_FB_DMA)
+			&& (dma->flags & _DRM_DMA_USE_FB))) {
 			drm_map_t *map = dev->agp_buffer_map;
+			unsigned long token = dev->agp_buffer_token;
 
 			if ( !map ) {
 				retcode = -EINVAL;
 				goto done;
 			}
 
-#if LINUX_VERSION_CODE <= 0x020402
-			down( &current->mm->mmap_sem );
-#else
 			down_write( &current->mm->mmap_sem );
-#endif
 			virtual = do_mmap( filp, 0, map->size,
 					   PROT_READ | PROT_WRITE,
 					   MAP_SHARED,
-					   (unsigned long)map->offset );
-#if LINUX_VERSION_CODE <= 0x020402
-			up( &current->mm->mmap_sem );
-#else
+					   token );
 			up_write( &current->mm->mmap_sem );
-#endif
 		} else {
-#if LINUX_VERSION_CODE <= 0x020402
-			down( &current->mm->mmap_sem );
-#else
 			down_write( &current->mm->mmap_sem );
-#endif
 			virtual = do_mmap( filp, 0, dma->byte_count,
 					   PROT_READ | PROT_WRITE,
 					   MAP_SHARED, 0 );
-#if LINUX_VERSION_CODE <= 0x020402
-			up( &current->mm->mmap_sem );
-#else
 			up_write( &current->mm->mmap_sem );
-#endif
 		}
 		if ( virtual > -1024UL ) {
 			/* Real error */
@@ -1279,3 +1557,26 @@ int drm_mapbufs( struct inode *inode, struct file *filp,
 	return retcode;
 }
 
+/**
+ * Compute size order.  Returns the exponent of the smaller power of two which
+ * is greater or equal to given number.
+ * 
+ * \param size size.
+ * \return order.
+ *
+ * \todo Can be made faster.
+ */
+int drm_order( unsigned long size )
+{
+	int order;
+	unsigned long tmp;
+
+	for (order = 0, tmp = size >> 1; tmp; tmp >>= 1, order++)
+		;
+
+	if (size & (size - 1))
+		++order;
+
+	return order;
+}
+EXPORT_SYMBOL(drm_order);
diff --git a/drivers/char/drm/drm_context.c b/drivers/char/drm/drm_context.c
index a7cfabd1ca2e..f515567e5b6f 100644
--- a/drivers/char/drm/drm_context.c
+++ b/drivers/char/drm/drm_context.c
@@ -212,6 +212,7 @@ int drm_getsareactx(struct inode *inode, struct file *filp,
 	drm_ctx_priv_map_t __user *argp = (void __user *)arg;
 	drm_ctx_priv_map_t request;
 	drm_map_t *map;
+	drm_map_list_t *_entry;
 
 	if (copy_from_user(&request, argp, sizeof(request)))
 		return -EFAULT;
@@ -225,7 +226,17 @@ int drm_getsareactx(struct inode *inode, struct file *filp,
 	map = dev->context_sareas[request.ctx_id];
 	up(&dev->struct_sem);
 
-	request.handle = (void *) map->offset;
+	request.handle = 0;
+	list_for_each_entry(_entry, &dev->maplist->head,head) {
+		if (_entry->map == map) {
+			request.handle = (void *)(unsigned long)_entry->user_token;
+			break;
+		}
+	}
+	if (request.handle == 0)
+		return -EINVAL;
+
+
 	if (copy_to_user(argp, &request, sizeof(request)))
 		return -EFAULT;
 	return 0;
@@ -262,7 +273,7 @@ int drm_setsareactx(struct inode *inode, struct file *filp,
 	list_for_each(list, &dev->maplist->head) {
 		r_list = list_entry(list, drm_map_list_t, head);
 		if (r_list->map
-		    && r_list->map->offset == (unsigned long) request.handle)
+		    && r_list->user_token == (unsigned long) request.handle)
 			goto found;
 	}
 bad:
@@ -369,7 +380,7 @@ int drm_resctx( struct inode *inode, struct file *filp,
 		for ( i = 0 ; i < DRM_RESERVED_CONTEXTS ; i++ ) {
 			ctx.handle = i;
 			if ( copy_to_user( &res.contexts[i],
-					   &i, sizeof(i) ) )
+					   &ctx, sizeof(ctx) ) )
 				return -EFAULT;
 		}
 	}
diff --git a/drivers/char/drm/drm_drv.c b/drivers/char/drm/drm_drv.c
index 3333c250c4d9..6ba48f346fcf 100644
--- a/drivers/char/drm/drm_drv.c
+++ b/drivers/char/drm/drm_drv.c
@@ -70,8 +70,8 @@ static drm_ioctl_desc_t		  drm_ioctls[] = {
 	[DRM_IOCTL_NR(DRM_IOCTL_UNBLOCK)]       = { drm_noop,        1, 1 },
 	[DRM_IOCTL_NR(DRM_IOCTL_AUTH_MAGIC)]    = { drm_authmagic,   1, 1 },
 
-	[DRM_IOCTL_NR(DRM_IOCTL_ADD_MAP)]       = { drm_addmap,      1, 1 },
-	[DRM_IOCTL_NR(DRM_IOCTL_RM_MAP)]        = { drm_rmmap,       1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_ADD_MAP)]       = { drm_addmap_ioctl,1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_RM_MAP)]        = { drm_rmmap_ioctl, 1, 0 },
 
 	[DRM_IOCTL_NR(DRM_IOCTL_SET_SAREA_CTX)] = { drm_setsareactx, 1, 1 },
 	[DRM_IOCTL_NR(DRM_IOCTL_GET_SAREA_CTX)] = { drm_getsareactx, 1, 0 },
@@ -102,10 +102,10 @@ static drm_ioctl_desc_t		  drm_ioctls[] = {
 	[DRM_IOCTL_NR(DRM_IOCTL_CONTROL)]       = { drm_control,     1, 1 },
 
 #if __OS_HAS_AGP
-	[DRM_IOCTL_NR(DRM_IOCTL_AGP_ACQUIRE)]   = { drm_agp_acquire, 1, 1 },
-	[DRM_IOCTL_NR(DRM_IOCTL_AGP_RELEASE)]   = { drm_agp_release, 1, 1 },
-	[DRM_IOCTL_NR(DRM_IOCTL_AGP_ENABLE)]    = { drm_agp_enable,  1, 1 },
-	[DRM_IOCTL_NR(DRM_IOCTL_AGP_INFO)]      = { drm_agp_info,    1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_ACQUIRE)]   = { drm_agp_acquire_ioctl, 1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_RELEASE)]   = { drm_agp_release_ioctl, 1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_ENABLE)]    = { drm_agp_enable_ioctl, 1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_INFO)]      = { drm_agp_info_ioctl, 1, 0 },
 	[DRM_IOCTL_NR(DRM_IOCTL_AGP_ALLOC)]     = { drm_agp_alloc,   1, 1 },
 	[DRM_IOCTL_NR(DRM_IOCTL_AGP_FREE)]      = { drm_agp_free,    1, 1 },
 	[DRM_IOCTL_NR(DRM_IOCTL_AGP_BIND)]      = { drm_agp_bind,    1, 1 },
@@ -127,14 +127,12 @@ static drm_ioctl_desc_t		  drm_ioctls[] = {
  *
  * Frees every resource in \p dev.
  *
- * \sa drm_device and setup().
+ * \sa drm_device
  */
 int drm_takedown( drm_device_t *dev )
 {
 	drm_magic_entry_t *pt, *next;
-	drm_map_t *map;
 	drm_map_list_t *r_list;
-	struct list_head *list, *list_next;
 	drm_vma_entry_t *vma, *vma_next;
 	int i;
 
@@ -142,6 +140,7 @@ int drm_takedown( drm_device_t *dev )
 
 	if (dev->driver->pretakedown)
 	  dev->driver->pretakedown(dev);
+	DRM_DEBUG("driver pretakedown completed\n");
 
 	if (dev->unique) {
 		drm_free(dev->unique, strlen(dev->unique) + 1, DRM_MEM_DRIVER);
@@ -178,11 +177,16 @@ int drm_takedown( drm_device_t *dev )
 		}
 		dev->agp->memory = NULL;
 
-		if ( dev->agp->acquired ) drm_agp_do_release(dev);
+		if (dev->agp->acquired)
+		  drm_agp_release(dev);
 
 		dev->agp->acquired = 0;
 		dev->agp->enabled  = 0;
 	}
+	if (drm_core_check_feature(dev, DRIVER_SG) && dev->sg) {
+		drm_sg_cleanup(dev->sg);
+		dev->sg = NULL;
+	}
 
 				/* Clear vma list (only built for debugging) */
 	if ( dev->vmalist ) {
@@ -194,48 +198,11 @@ int drm_takedown( drm_device_t *dev )
 	}
 
 	if( dev->maplist ) {
-		list_for_each_safe( list, list_next, &dev->maplist->head ) {
-			r_list = (drm_map_list_t *)list;
-
-			if ( ( map = r_list->map ) ) {
-				switch ( map->type ) {
-				case _DRM_REGISTERS:
-				case _DRM_FRAME_BUFFER:
-					if (drm_core_has_MTRR(dev)) {
-						if ( map->mtrr >= 0 ) {
-							int retcode;
-							retcode = mtrr_del( map->mtrr,
-									    map->offset,
-									    map->size );
-							DRM_DEBUG( "mtrr_del=%d\n", retcode );
-						}
-					}
-					drm_ioremapfree( map->handle, map->size, dev );
-					break;
-				case _DRM_SHM:
-					vfree(map->handle);
-					break;
-
-				case _DRM_AGP:
-					/* Do nothing here, because this is all
-					 * handled in the AGP/GART driver.
-					 */
-					break;
-				case _DRM_SCATTER_GATHER:
-					/* Handle it */
-					if (drm_core_check_feature(dev, DRIVER_SG) && dev->sg) {
-						drm_sg_cleanup(dev->sg);
-						dev->sg = NULL;
-					}
-					break;
-				}
-				drm_free(map, sizeof(*map), DRM_MEM_MAPS);
-			}
-			list_del( list );
-			drm_free(r_list, sizeof(*r_list), DRM_MEM_MAPS);
- 		}
-		drm_free(dev->maplist, sizeof(*dev->maplist), DRM_MEM_MAPS);
-		dev->maplist = NULL;
+		while (!list_empty(&dev->maplist->head)) {
+			struct list_head *list = dev->maplist->head.next;
+			r_list = list_entry(list, drm_map_list_t, head);
+			drm_rmmap_locked(dev, r_list->map);
+		}
  	}
 
 	if (drm_core_check_feature(dev, DRIVER_DMA_QUEUE) && dev->queuelist ) {
@@ -264,6 +231,7 @@ int drm_takedown( drm_device_t *dev )
 	}
 	up( &dev->struct_sem );
 
+	DRM_DEBUG("takedown completed\n");
 	return 0;
 }
 
@@ -312,7 +280,7 @@ EXPORT_SYMBOL(drm_init);
  *
  * Cleans up all DRM device, calling takedown().
  * 
- * \sa drm_init().
+ * \sa drm_init
  */
 static void drm_cleanup( drm_device_t *dev )
 {
@@ -325,6 +293,11 @@ static void drm_cleanup( drm_device_t *dev )
 
 	drm_takedown( dev );	
 
+	if (dev->maplist) {
+		drm_free(dev->maplist, sizeof(*dev->maplist), DRM_MEM_MAPS);
+		dev->maplist = NULL;
+	}
+
 	drm_ctxbitmap_cleanup( dev );
 	
 	if (drm_core_has_MTRR(dev) && drm_core_has_AGP(dev) &&
diff --git a/drivers/char/drm/drm_fops.c b/drivers/char/drm/drm_fops.c
index 10e64fde8d78..a1f4e9cd64ed 100644
--- a/drivers/char/drm/drm_fops.c
+++ b/drivers/char/drm/drm_fops.c
@@ -71,12 +71,6 @@ static int drm_setup( drm_device_t *dev )
 		dev->magiclist[i].tail = NULL;
 	}
 
-	dev->maplist = drm_alloc(sizeof(*dev->maplist),
-				  DRM_MEM_MAPS);
-	if(dev->maplist == NULL) return -ENOMEM;
-	memset(dev->maplist, 0, sizeof(*dev->maplist));
-	INIT_LIST_HEAD(&dev->maplist->head);
-
 	dev->ctxlist = drm_alloc(sizeof(*dev->ctxlist),
 				  DRM_MEM_CTXLIST);
 	if(dev->ctxlist == NULL) return -ENOMEM;
diff --git a/drivers/char/drm/drm_ioctl.c b/drivers/char/drm/drm_ioctl.c
index 39afda0ccabe..d2ed3ba5aca9 100644
--- a/drivers/char/drm/drm_ioctl.c
+++ b/drivers/char/drm/drm_ioctl.c
@@ -208,7 +208,7 @@ int drm_getmap( struct inode *inode, struct file *filp,
 	map.size   = r_list->map->size;
 	map.type   = r_list->map->type;
 	map.flags  = r_list->map->flags;
-	map.handle = r_list->map->handle;
+	map.handle = (void *)(unsigned long) r_list->user_token;
 	map.mtrr   = r_list->map->mtrr;
 	up(&dev->struct_sem);
 
diff --git a/drivers/char/drm/drm_memory.c b/drivers/char/drm/drm_memory.c
index ace3d42f4407..ff483fb418aa 100644
--- a/drivers/char/drm/drm_memory.c
+++ b/drivers/char/drm/drm_memory.c
@@ -142,27 +142,31 @@ void drm_free_pages(unsigned long address, int order, int area)
 
 #if __OS_HAS_AGP
 /** Wrapper around agp_allocate_memory() */
-DRM_AGP_MEM *drm_alloc_agp(struct agp_bridge_data *bridge, int pages, u32 type)
+DRM_AGP_MEM *drm_alloc_agp(drm_device_t *dev, int pages, u32 type)
 {
-	return drm_agp_allocate_memory(bridge, pages, type);
+	return drm_agp_allocate_memory(dev->agp->bridge, pages, type);
 }
+EXPORT_SYMBOL(drm_alloc_agp);
 
 /** Wrapper around agp_free_memory() */
 int drm_free_agp(DRM_AGP_MEM *handle, int pages)
 {
 	return drm_agp_free_memory(handle) ? 0 : -EINVAL;
 }
+EXPORT_SYMBOL(drm_free_agp);
 
 /** Wrapper around agp_bind_memory() */
 int drm_bind_agp(DRM_AGP_MEM *handle, unsigned int start)
 {
 	return drm_agp_bind_memory(handle, start);
 }
+EXPORT_SYMBOL(drm_bind_agp);
 
 /** Wrapper around agp_unbind_memory() */
 int drm_unbind_agp(DRM_AGP_MEM *handle)
 {
 	return drm_agp_unbind_memory(handle);
 }
+EXPORT_SYMBOL(drm_unbind_agp);
 #endif /* agp */
 #endif /* debug_memory */
diff --git a/drivers/char/drm/drm_pci.c b/drivers/char/drm/drm_pci.c
index 192e8762571c..09ed712c1a7f 100644
--- a/drivers/char/drm/drm_pci.c
+++ b/drivers/char/drm/drm_pci.c
@@ -46,11 +46,11 @@
 /**
  * \brief Allocate a PCI consistent memory block, for DMA.
  */
-void *drm_pci_alloc(drm_device_t * dev, size_t size, size_t align,
-		    dma_addr_t maxaddr, dma_addr_t * busaddr)
+drm_dma_handle_t *drm_pci_alloc(drm_device_t * dev, size_t size, size_t align,
+				dma_addr_t maxaddr)
 {
-	void *address;
-#if DRM_DEBUG_MEMORY
+	drm_dma_handle_t *dmah;
+#ifdef DRM_DEBUG_MEMORY
 	int area = DRM_MEM_DMA;
 
 	spin_lock(&drm_mem_lock);
@@ -74,13 +74,19 @@ void *drm_pci_alloc(drm_device_t * dev, size_t size, size_t align,
 		return NULL;
 	}
 
-	address = pci_alloc_consistent(dev->pdev, size, busaddr);
+	dmah = kmalloc(sizeof(drm_dma_handle_t), GFP_KERNEL);
+	if (!dmah)
+		return NULL;
+	
+	dmah->size = size;
+	dmah->vaddr = pci_alloc_consistent(dev->pdev, size, &dmah->busaddr);
 
-#if DRM_DEBUG_MEMORY
-	if (address == NULL) {
+#ifdef DRM_DEBUG_MEMORY
+	if (dmah->vaddr == NULL) {
 		spin_lock(&drm_mem_lock);
 		++drm_mem_stats[area].fail_count;
 		spin_unlock(&drm_mem_lock);
+		kfree(dmah);
 		return NULL;
 	}
 
@@ -90,37 +96,42 @@ void *drm_pci_alloc(drm_device_t * dev, size_t size, size_t align,
 	drm_ram_used += size;
 	spin_unlock(&drm_mem_lock);
 #else
-	if (address == NULL)
+	if (dmah->vaddr == NULL) {
+		kfree(dmah);
 		return NULL;
+	}
 #endif
 
-	memset(address, 0, size);
+	memset(dmah->vaddr, 0, size);
 
-	return address;
+	return dmah;
 }
 EXPORT_SYMBOL(drm_pci_alloc);
 
 /**
- * \brief Free a PCI consistent memory block.
+ * \brief Free a PCI consistent memory block with freeing its descriptor.
+ *
+ * This function is for internal use in the Linux-specific DRM core code.
  */
 void
-drm_pci_free(drm_device_t * dev, size_t size, void *vaddr, dma_addr_t busaddr)
+__drm_pci_free(drm_device_t * dev, drm_dma_handle_t *dmah)
 {
-#if DRM_DEBUG_MEMORY
+#ifdef DRM_DEBUG_MEMORY
 	int area = DRM_MEM_DMA;
 	int alloc_count;
 	int free_count;
 #endif
 
-	if (!vaddr) {
-#if DRM_DEBUG_MEMORY
+	if (!dmah->vaddr) {
+#ifdef DRM_DEBUG_MEMORY
 		DRM_MEM_ERROR(area, "Attempt to free address 0\n");
 #endif
 	} else {
-		pci_free_consistent(dev->pdev, size, vaddr, busaddr);
+		pci_free_consistent(dev->pdev, dmah->size, dmah->vaddr,
+				    dmah->busaddr);
 	}
 
-#if DRM_DEBUG_MEMORY
+#ifdef DRM_DEBUG_MEMORY
 	spin_lock(&drm_mem_lock);
 	free_count = ++drm_mem_stats[area].free_count;
 	alloc_count = drm_mem_stats[area].succeed_count;
@@ -135,6 +146,16 @@ drm_pci_free(drm_device_t * dev, size_t size, void *vaddr, dma_addr_t busaddr)
 #endif
 
 }
+
+/**
+ * \brief Free a PCI consistent memory block
+ */
+void
+drm_pci_free(drm_device_t *dev, drm_dma_handle_t *dmah)
+{
+	__drm_pci_free(dev, dmah);
+	kfree(dmah);
+}
 EXPORT_SYMBOL(drm_pci_free);
 
 /*@}*/
diff --git a/drivers/char/drm/drm_pciids.h b/drivers/char/drm/drm_pciids.h
index 70ca4fa55c9d..58b1747cd440 100644
--- a/drivers/char/drm/drm_pciids.h
+++ b/drivers/char/drm/drm_pciids.h
@@ -25,6 +25,8 @@
 	{0x1002, 0x4965, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R250}, \
 	{0x1002, 0x4966, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R250}, \
 	{0x1002, 0x4967, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R250}, \
+	{0x1002, 0x4A49, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R420}, \
+	{0x1002, 0x4A4B, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R420}, \
 	{0x1002, 0x4C57, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RV200|CHIP_IS_MOBILITY}, \
 	{0x1002, 0x4C58, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RV200|CHIP_IS_MOBILITY}, \
 	{0x1002, 0x4C59, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RV100|CHIP_IS_MOBILITY}, \
@@ -33,7 +35,17 @@
 	{0x1002, 0x4C65, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R250|CHIP_IS_MOBILITY}, \
 	{0x1002, 0x4C66, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R250|CHIP_IS_MOBILITY}, \
 	{0x1002, 0x4C67, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R250|CHIP_IS_MOBILITY}, \
+	{0x1002, 0x4E44, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R300}, \
+	{0x1002, 0x4E45, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R300}, \
+	{0x1002, 0x4E46, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RV350}, \
+	{0x1002, 0x4E47, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R300}, \
+	{0x1002, 0x4E48, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R350}, \
+	{0x1002, 0x4E49, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R350}, \
+	{0x1002, 0x4E4A, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RV350}, \
+	{0x1002, 0x4E4B, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R350}, \
 	{0x1002, 0x4E50, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RV350|CHIP_IS_MOBILITY}, \
+	{0x1002, 0x4E51, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RV350|CHIP_IS_MOBILITY}, \
+	{0x1002, 0x4E54, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RV350|CHIP_IS_MOBILITY}, \
 	{0x1002, 0x5144, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R100|CHIP_SINGLE_CRTC}, \
 	{0x1002, 0x5145, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R100|CHIP_SINGLE_CRTC}, \
 	{0x1002, 0x5146, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R100|CHIP_SINGLE_CRTC}, \
@@ -56,6 +68,7 @@
 	{0x1002, 0x516A, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R200}, \
 	{0x1002, 0x516B, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R200}, \
 	{0x1002, 0x516C, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R200}, \
+	{0x1002, 0x5460, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RV350}, \
 	{0x1002, 0x5834, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RS300|CHIP_IS_IGP}, \
 	{0x1002, 0x5835, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RS300|CHIP_IS_IGP|CHIP_IS_MOBILITY}, \
 	{0x1002, 0x5836, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RS300|CHIP_IS_IGP}, \
@@ -116,9 +129,10 @@
 	{0, 0, 0}
 
 #define mga_PCI_IDS \
-	{0x102b, 0x0521, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x102b, 0x0525, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x102b, 0x2527, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
+	{0x102b, 0x0520, PCI_ANY_ID, PCI_ANY_ID, 0, 0, MGA_CARD_TYPE_G200}, \
+	{0x102b, 0x0521, PCI_ANY_ID, PCI_ANY_ID, 0, 0, MGA_CARD_TYPE_G200}, \
+	{0x102b, 0x0525, PCI_ANY_ID, PCI_ANY_ID, 0, 0, MGA_CARD_TYPE_G400}, \
+	{0x102b, 0x2527, PCI_ANY_ID, PCI_ANY_ID, 0, 0, MGA_CARD_TYPE_G550}, \
 	{0, 0, 0}
 
 #define mach64_PCI_IDS \
@@ -162,9 +176,10 @@
 
 #define viadrv_PCI_IDS \
 	{0x1106, 0x3022, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
+	{0x1106, 0x3118, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
 	{0x1106, 0x3122, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
 	{0x1106, 0x7205, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x1106, 0x7204, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
+	{0x1106, 0x3108, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
 	{0, 0, 0}
 
 #define i810_PCI_IDS \
@@ -181,33 +196,30 @@
 	{0x8086, 0x2572, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
 	{0, 0, 0}
 
-#define gamma_PCI_IDS \
-	{0x3d3d, 0x0008, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0, 0, 0}
-
 #define savage_PCI_IDS \
-	{0x5333, 0x8a22, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x5333, 0x8a23, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x5333, 0x8c10, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x5333, 0x8c11, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x5333, 0x8c12, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x5333, 0x8c13, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x5333, 0x8c20, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x5333, 0x8c21, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x5333, 0x8c22, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x5333, 0x8c24, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x5333, 0x8c26, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x5333, 0x8c2a, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x5333, 0x8c2b, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x5333, 0x8c2c, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x5333, 0x8c2d, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x5333, 0x8c2e, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x5333, 0x8c2f, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x5333, 0x8a25, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x5333, 0x8a26, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x5333, 0x8d01, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x5333, 0x8d02, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x5333, 0x8d04, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
+	{0x5333, 0x8a20, PCI_ANY_ID, PCI_ANY_ID, 0, 0, S3_SAVAGE3D}, \
+	{0x5333, 0x8a21, PCI_ANY_ID, PCI_ANY_ID, 0, 0, S3_SAVAGE3D}, \
+	{0x5333, 0x8a22, PCI_ANY_ID, PCI_ANY_ID, 0, 0, S3_SAVAGE4}, \
+	{0x5333, 0x8a23, PCI_ANY_ID, PCI_ANY_ID, 0, 0, S3_SAVAGE4}, \
+	{0x5333, 0x8c10, PCI_ANY_ID, PCI_ANY_ID, 0, 0, S3_SAVAGE_MX}, \
+	{0x5333, 0x8c11, PCI_ANY_ID, PCI_ANY_ID, 0, 0, S3_SAVAGE_MX}, \
+	{0x5333, 0x8c12, PCI_ANY_ID, PCI_ANY_ID, 0, 0, S3_SAVAGE_MX}, \
+	{0x5333, 0x8c13, PCI_ANY_ID, PCI_ANY_ID, 0, 0, S3_SAVAGE_MX}, \
+	{0x5333, 0x8c22, PCI_ANY_ID, PCI_ANY_ID, 0, 0, S3_SUPERSAVAGE}, \
+	{0x5333, 0x8c24, PCI_ANY_ID, PCI_ANY_ID, 0, 0, S3_SUPERSAVAGE}, \
+	{0x5333, 0x8c26, PCI_ANY_ID, PCI_ANY_ID, 0, 0, S3_SUPERSAVAGE}, \
+	{0x5333, 0x8c2a, PCI_ANY_ID, PCI_ANY_ID, 0, 0, S3_SUPERSAVAGE}, \
+	{0x5333, 0x8c2b, PCI_ANY_ID, PCI_ANY_ID, 0, 0, S3_SUPERSAVAGE}, \
+	{0x5333, 0x8c2c, PCI_ANY_ID, PCI_ANY_ID, 0, 0, S3_SUPERSAVAGE}, \
+	{0x5333, 0x8c2d, PCI_ANY_ID, PCI_ANY_ID, 0, 0, S3_SUPERSAVAGE}, \
+	{0x5333, 0x8c2e, PCI_ANY_ID, PCI_ANY_ID, 0, 0, S3_SUPERSAVAGE}, \
+	{0x5333, 0x8c2f, PCI_ANY_ID, PCI_ANY_ID, 0, 0, S3_SUPERSAVAGE}, \
+	{0x5333, 0x8a25, PCI_ANY_ID, PCI_ANY_ID, 0, 0, S3_PROSAVAGE}, \
+	{0x5333, 0x8a26, PCI_ANY_ID, PCI_ANY_ID, 0, 0, S3_PROSAVAGE}, \
+	{0x5333, 0x8d01, PCI_ANY_ID, PCI_ANY_ID, 0, 0, S3_TWISTER}, \
+	{0x5333, 0x8d02, PCI_ANY_ID, PCI_ANY_ID, 0, 0, S3_TWISTER}, \
+	{0x5333, 0x8d03, PCI_ANY_ID, PCI_ANY_ID, 0, 0, S3_PROSAVAGEDDR}, \
+	{0x5333, 0x8d04, PCI_ANY_ID, PCI_ANY_ID, 0, 0, S3_PROSAVAGEDDR}, \
 	{0, 0, 0}
 
 #define ffb_PCI_IDS \
@@ -223,10 +235,3 @@
 	{0x8086, 0x2772, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
 	{0, 0, 0}
 
-#define viadrv_PCI_IDS \
-	{0x1106, 0x3022, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x1106, 0x3122, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x1106, 0x7205, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0x1106, 0x7204, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0}, \
-	{0, 0, 0}
-
diff --git a/drivers/char/drm/drm_proc.c b/drivers/char/drm/drm_proc.c
index 4774087d2e9e..32d2bb99462c 100644
--- a/drivers/char/drm/drm_proc.c
+++ b/drivers/char/drm/drm_proc.c
@@ -210,8 +210,8 @@ static int drm__vm_info(char *buf, char **start, off_t offset, int request,
 
 				/* Hardcoded from _DRM_FRAME_BUFFER,
                                    _DRM_REGISTERS, _DRM_SHM, _DRM_AGP, and
-                                   _DRM_SCATTER_GATHER. */
-	const char   *types[] = { "FB", "REG", "SHM", "AGP", "SG" };
+                                   _DRM_SCATTER_GATHER and _DRM_CONSISTENT */
+	const char   *types[] = { "FB", "REG", "SHM", "AGP", "SG", "PCI" };
 	const char   *type;
 	int	     i;
 
@@ -229,16 +229,19 @@ static int drm__vm_info(char *buf, char **start, off_t offset, int request,
 	if (dev->maplist != NULL) list_for_each(list, &dev->maplist->head) {
 		r_list = list_entry(list, drm_map_list_t, head);
 		map = r_list->map;
-		if(!map) continue;
-		if (map->type < 0 || map->type > 4) type = "??";
-		else				    type = types[map->type];
-		DRM_PROC_PRINT("%4d 0x%08lx 0x%08lx %4.4s  0x%02x 0x%08lx ",
+		if(!map)
+			continue;
+		if (map->type < 0 || map->type > 5)
+			type = "??";
+		else	
+			type = types[map->type];
+		DRM_PROC_PRINT("%4d 0x%08lx 0x%08lx %4.4s  0x%02x 0x%08x ",
 			       i,
 			       map->offset,
 			       map->size,
 			       type,
 			       map->flags,
-			       (unsigned long)map->handle);
+			       r_list->user_token);
 		if (map->mtrr < 0) {
 			DRM_PROC_PRINT("none\n");
 		} else {
diff --git a/drivers/char/drm/drm_scatter.c b/drivers/char/drm/drm_scatter.c
index 54fddb6ea2d1..ed267d49bc6a 100644
--- a/drivers/char/drm/drm_scatter.c
+++ b/drivers/char/drm/drm_scatter.c
@@ -61,6 +61,12 @@ void drm_sg_cleanup( drm_sg_mem_t *entry )
 		   DRM_MEM_SGLISTS );
 }
 
+#ifdef _LP64
+# define ScatterHandle(x) (unsigned int)((x >> 32) + (x & ((1L << 32) - 1)))
+#else
+# define ScatterHandle(x) (unsigned int)(x)
+#endif
+
 int drm_sg_alloc( struct inode *inode, struct file *filp,
 		   unsigned int cmd, unsigned long arg )
 {
@@ -133,12 +139,13 @@ int drm_sg_alloc( struct inode *inode, struct file *filp,
 	 */
 	memset( entry->virtual, 0, pages << PAGE_SHIFT );
 
-	entry->handle = (unsigned long)entry->virtual;
+	entry->handle = ScatterHandle((unsigned long)entry->virtual);
 
 	DRM_DEBUG( "sg alloc handle  = %08lx\n", entry->handle );
 	DRM_DEBUG( "sg alloc virtual = %p\n", entry->virtual );
 
-	for ( i = entry->handle, j = 0 ; j < pages ; i += PAGE_SIZE, j++ ) {
+	for (i = (unsigned long)entry->virtual, j = 0; j < pages; 
+		i += PAGE_SIZE, j++) {
 		entry->pagelist[j] = vmalloc_to_page((void *)i);
 		if (!entry->pagelist[j])
 			goto failed;
diff --git a/drivers/char/drm/drm_stub.c b/drivers/char/drm/drm_stub.c
index 48829a1a086a..95a976c96eb8 100644
--- a/drivers/char/drm/drm_stub.c
+++ b/drivers/char/drm/drm_stub.c
@@ -75,6 +75,11 @@ static int drm_fill_in_dev(drm_device_t *dev, struct pci_dev *pdev, const struct
 	dev->pci_func = PCI_FUNC(pdev->devfn);
 	dev->irq = pdev->irq;
 
+	dev->maplist = drm_calloc(1, sizeof(*dev->maplist), DRM_MEM_MAPS);
+	if (dev->maplist == NULL)
+		return -ENOMEM;
+	INIT_LIST_HEAD(&dev->maplist->head);
+
 	/* the DRM has 6 basic counters */
 	dev->counters = 6;
 	dev->types[0]  = _DRM_STAT_LOCK;
@@ -91,7 +96,8 @@ static int drm_fill_in_dev(drm_device_t *dev, struct pci_dev *pdev, const struct
 			goto error_out_unreg;
 
 	if (drm_core_has_AGP(dev)) {
-		dev->agp = drm_agp_init(dev);
+		if (drm_device_is_agp(dev))
+			dev->agp = drm_agp_init(dev);
 		if (drm_core_check_feature(dev, DRIVER_REQUIRE_AGP) && (dev->agp == NULL)) {
 			DRM_ERROR( "Cannot initialize the agpgart module.\n" );
 			retcode = -EINVAL;
diff --git a/drivers/char/drm/drm_vm.c b/drivers/char/drm/drm_vm.c
index 621220f3f372..ced4215e2275 100644
--- a/drivers/char/drm/drm_vm.c
+++ b/drivers/char/drm/drm_vm.c
@@ -73,12 +73,13 @@ static __inline__ struct page *drm_do_vm_nopage(struct vm_area_struct *vma,
 		r_list = list_entry(list, drm_map_list_t, head);
 		map = r_list->map;
 		if (!map) continue;
-		if (map->offset == VM_OFFSET(vma)) break;
+		if (r_list->user_token == VM_OFFSET(vma))
+			break;
 	}
 
 	if (map && map->type == _DRM_AGP) {
 		unsigned long offset = address - vma->vm_start;
-		unsigned long baddr = VM_OFFSET(vma) + offset;
+		unsigned long baddr = map->offset + offset;
 		struct drm_agp_mem *agpmem;
 		struct page *page;
 
@@ -210,6 +211,8 @@ static void drm_vm_shm_close(struct vm_area_struct *vma)
 		}
 
 		if(!found_maps) {
+			drm_dma_handle_t dmah;
+
 			switch (map->type) {
 			case _DRM_REGISTERS:
 			case _DRM_FRAME_BUFFER:
@@ -228,6 +231,12 @@ static void drm_vm_shm_close(struct vm_area_struct *vma)
 			case _DRM_AGP:
 			case _DRM_SCATTER_GATHER:
 				break;
+			case _DRM_CONSISTENT:
+				dmah.vaddr = map->handle;
+				dmah.busaddr = map->offset;
+				dmah.size = map->size;
+				__drm_pci_free(dev, &dmah);
+				break;
 			}
 			drm_free(map, sizeof(*map), DRM_MEM_MAPS);
 		}
@@ -296,7 +305,7 @@ static __inline__ struct page *drm_do_vm_sg_nopage(struct vm_area_struct *vma,
 
 
 	offset = address - vma->vm_start;
-	map_offset = map->offset - dev->sg->handle;
+	map_offset = map->offset - (unsigned long)dev->sg->virtual;
 	page_offset = (offset >> PAGE_SHIFT) + (map_offset >> PAGE_SHIFT);
 	page = entry->pagelist[page_offset];
 	get_page(page);
@@ -305,8 +314,6 @@ static __inline__ struct page *drm_do_vm_sg_nopage(struct vm_area_struct *vma,
 }
 
 
-#if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,0)
-
 static struct page *drm_vm_nopage(struct vm_area_struct *vma,
 				   unsigned long address,
 				   int *type) {
@@ -335,35 +342,6 @@ static struct page *drm_vm_sg_nopage(struct vm_area_struct *vma,
 	return drm_do_vm_sg_nopage(vma, address);
 }
 
-#else	/* LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,0) */
-
-static struct page *drm_vm_nopage(struct vm_area_struct *vma,
-				   unsigned long address,
-				   int unused) {
-	return drm_do_vm_nopage(vma, address);
-}
-
-static struct page *drm_vm_shm_nopage(struct vm_area_struct *vma,
-				       unsigned long address,
-				       int unused) {
-	return drm_do_vm_shm_nopage(vma, address);
-}
-
-static struct page *drm_vm_dma_nopage(struct vm_area_struct *vma,
-				       unsigned long address,
-				       int unused) {
-	return drm_do_vm_dma_nopage(vma, address);
-}
-
-static struct page *drm_vm_sg_nopage(struct vm_area_struct *vma,
-				      unsigned long address,
-				      int unused) {
-	return drm_do_vm_sg_nopage(vma, address);
-}
-
-#endif
-
-
 /** AGP virtual memory operations */
 static struct vm_operations_struct   drm_vm_ops = {
 	.nopage = drm_vm_nopage,
@@ -487,11 +465,7 @@ static int drm_mmap_dma(struct file *filp, struct vm_area_struct *vma)
 
 	vma->vm_ops   = &drm_vm_dma_ops;
 
-#if LINUX_VERSION_CODE <= 0x02040e /* KERNEL_VERSION(2,4,14) */
-	vma->vm_flags |= VM_LOCKED | VM_SHM; /* Don't swap */
-#else
 	vma->vm_flags |= VM_RESERVED; /* Don't swap */
-#endif
 
 	vma->vm_file  =	 filp;	/* Needed for drm_vm_open() */
 	drm_vm_open(vma);
@@ -560,13 +534,12 @@ int drm_mmap(struct file *filp, struct vm_area_struct *vma)
 				   for performance, even if the list was a
 				   bit longer. */
 	list_for_each(list, &dev->maplist->head) {
-		unsigned long off;
 
 		r_list = list_entry(list, drm_map_list_t, head);
 		map = r_list->map;
 		if (!map) continue;
-		off = dev->driver->get_map_ofs(map);
-		if (off == VM_OFFSET(vma)) break;
+		if (r_list->user_token == VM_OFFSET(vma))
+			break;
 	}
 
 	if (!map || ((map->flags&_DRM_RESTRICTED) && !capable(CAP_SYS_ADMIN)))
@@ -605,17 +578,17 @@ int drm_mmap(struct file *filp, struct vm_area_struct *vma)
                 /* fall through to _DRM_FRAME_BUFFER... */        
 	case _DRM_FRAME_BUFFER:
 	case _DRM_REGISTERS:
-		if (VM_OFFSET(vma) >= __pa(high_memory)) {
 #if defined(__i386__) || defined(__x86_64__)
-			if (boot_cpu_data.x86 > 3 && map->type != _DRM_AGP) {
-				pgprot_val(vma->vm_page_prot) |= _PAGE_PCD;
-				pgprot_val(vma->vm_page_prot) &= ~_PAGE_PWT;
-			}
+		if (boot_cpu_data.x86 > 3 && map->type != _DRM_AGP) {
+			pgprot_val(vma->vm_page_prot) |= _PAGE_PCD;
+			pgprot_val(vma->vm_page_prot) &= ~_PAGE_PWT;
+		}
 #elif defined(__powerpc__)
-			pgprot_val(vma->vm_page_prot) |= _PAGE_NO_CACHE | _PAGE_GUARDED;
+		pgprot_val(vma->vm_page_prot) |= _PAGE_NO_CACHE;
+		if (map->type == _DRM_REGISTERS)
+			pgprot_val(vma->vm_page_prot) |= _PAGE_GUARDED;
 #endif
-			vma->vm_flags |= VM_IO;	/* not in core dump */
-		}
+		vma->vm_flags |= VM_IO;	/* not in core dump */
 #if defined(__ia64__)
 		if (efi_range_is_wc(vma->vm_start, vma->vm_end -
 				    vma->vm_start))
@@ -628,12 +601,12 @@ int drm_mmap(struct file *filp, struct vm_area_struct *vma)
 		offset = dev->driver->get_reg_ofs(dev);
 #ifdef __sparc__
 		if (io_remap_pfn_range(DRM_RPR_ARG(vma) vma->vm_start,
-					(VM_OFFSET(vma) + offset) >> PAGE_SHIFT,
+					(map->offset + offset) >> PAGE_SHIFT,
 					vma->vm_end - vma->vm_start,
 					vma->vm_page_prot))
 #else
 		if (io_remap_pfn_range(vma, vma->vm_start,
-				     (VM_OFFSET(vma) + offset) >> PAGE_SHIFT,
+				     (map->offset + offset) >> PAGE_SHIFT,
 				     vma->vm_end - vma->vm_start,
 				     vma->vm_page_prot))
 #endif
@@ -641,37 +614,28 @@ int drm_mmap(struct file *filp, struct vm_area_struct *vma)
 		DRM_DEBUG("   Type = %d; start = 0x%lx, end = 0x%lx,"
 			  " offset = 0x%lx\n",
 			  map->type,
-			  vma->vm_start, vma->vm_end, VM_OFFSET(vma) + offset);
+			  vma->vm_start, vma->vm_end, map->offset + offset);
 		vma->vm_ops = &drm_vm_ops;
 		break;
 	case _DRM_SHM:
+	case _DRM_CONSISTENT:
+		/* Consistent memory is really like shared memory. It's only
+		 * allocate in a different way */
 		vma->vm_ops = &drm_vm_shm_ops;
 		vma->vm_private_data = (void *)map;
 				/* Don't let this area swap.  Change when
 				   DRM_KERNEL advisory is supported. */
-#if LINUX_VERSION_CODE <= 0x02040e /* KERNEL_VERSION(2,4,14) */
-		vma->vm_flags |= VM_LOCKED;
-#else
 		vma->vm_flags |= VM_RESERVED;
-#endif
 		break;
 	case _DRM_SCATTER_GATHER:
 		vma->vm_ops = &drm_vm_sg_ops;
 		vma->vm_private_data = (void *)map;
-#if LINUX_VERSION_CODE <= 0x02040e /* KERNEL_VERSION(2,4,14) */
-		vma->vm_flags |= VM_LOCKED;
-#else
 		vma->vm_flags |= VM_RESERVED;
-#endif
                 break;
 	default:
 		return -EINVAL;	/* This should never happen. */
 	}
-#if LINUX_VERSION_CODE <= 0x02040e /* KERNEL_VERSION(2,4,14) */
-	vma->vm_flags |= VM_LOCKED | VM_SHM; /* Don't swap */
-#else
 	vma->vm_flags |= VM_RESERVED; /* Don't swap */
-#endif
 
 	vma->vm_file  =	 filp;	/* Needed for drm_vm_open() */
 	drm_vm_open(vma);
diff --git a/drivers/char/drm/ffb_drv.c b/drivers/char/drm/ffb_drv.c
index ec614fff8f04..1bd0d55ee0f0 100644
--- a/drivers/char/drm/ffb_drv.c
+++ b/drivers/char/drm/ffb_drv.c
@@ -152,14 +152,11 @@ static drm_map_t *ffb_find_map(struct file *filp, unsigned long off)
 		return NULL;
 
 	list_for_each(list, &dev->maplist->head) {
-		unsigned long uoff;
-
 		r_list = (drm_map_list_t *)list;
 		map = r_list->map;
 		if (!map)
 			continue;
-		uoff = (map->offset & 0xffffffff);
-		if (uoff == off)
+		if (r_list->user_token == off)
 			return map;
 	}
 
diff --git a/drivers/char/drm/gamma_context.h b/drivers/char/drm/gamma_context.h
deleted file mode 100644
index d11b507f87ee..000000000000
--- a/drivers/char/drm/gamma_context.h
+++ /dev/null
@@ -1,492 +0,0 @@
-/* drm_context.h -- IOCTLs for generic contexts -*- linux-c -*-
- * Created: Fri Nov 24 18:31:37 2000 by gareth@valinux.com
- *
- * Copyright 1999, 2000 Precision Insight, Inc., Cedar Park, Texas.
- * Copyright 2000 VA Linux Systems, Inc., Sunnyvale, California.
- * 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
- * VA LINUX SYSTEMS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
- * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
- * OTHER DEALINGS IN THE SOFTWARE.
- *
- * Authors:
- *    Rickard E. (Rik) Faith <faith@valinux.com>
- *    Gareth Hughes <gareth@valinux.com>
- * ChangeLog:
- *  2001-11-16	Torsten Duwe <duwe@caldera.de>
- *		added context constructor/destructor hooks,
- *		needed by SiS driver's memory management.
- */
-
-/* ================================================================
- * Old-style context support -- only used by gamma.  
- */
-
-
-/* The drm_read and drm_write_string code (especially that which manages
-   the circular buffer), is based on Alessandro Rubini's LINUX DEVICE
-   DRIVERS (Cambridge: O'Reilly, 1998), pages 111-113. */
-
-ssize_t gamma_fops_read(struct file *filp, char __user *buf, size_t count, loff_t *off)
-{
-	drm_file_t    *priv   = filp->private_data;
-	drm_device_t  *dev    = priv->dev;
-	int	      left;
-	int	      avail;
-	int	      send;
-	int	      cur;
-
-	DRM_DEBUG("%p, %p\n", dev->buf_rp, dev->buf_wp);
-
-	while (dev->buf_rp == dev->buf_wp) {
-		DRM_DEBUG("  sleeping\n");
-		if (filp->f_flags & O_NONBLOCK) {
-			return -EAGAIN;
-		}
-		interruptible_sleep_on(&dev->buf_readers);
-		if (signal_pending(current)) {
-			DRM_DEBUG("  interrupted\n");
-			return -ERESTARTSYS;
-		}
-		DRM_DEBUG("  awake\n");
-	}
-
-	left  = (dev->buf_rp + DRM_BSZ - dev->buf_wp) % DRM_BSZ;
-	avail = DRM_BSZ - left;
-	send  = DRM_MIN(avail, count);
-
-	while (send) {
-		if (dev->buf_wp > dev->buf_rp) {
-			cur = DRM_MIN(send, dev->buf_wp - dev->buf_rp);
-		} else {
-			cur = DRM_MIN(send, dev->buf_end - dev->buf_rp);
-		}
-		if (copy_to_user(buf, dev->buf_rp, cur))
-			return -EFAULT;
-		dev->buf_rp += cur;
-		if (dev->buf_rp == dev->buf_end) dev->buf_rp = dev->buf;
-		send -= cur;
-	}
-
-	wake_up_interruptible(&dev->buf_writers);
-	return DRM_MIN(avail, count);
-}
-
-
-/* In an incredibly convoluted setup, the kernel module actually calls
- * back into the X server to perform context switches on behalf of the
- * 3d clients.
- */
-int DRM(write_string)(drm_device_t *dev, const char *s)
-{
-	int left   = (dev->buf_rp + DRM_BSZ - dev->buf_wp) % DRM_BSZ;
-	int send   = strlen(s);
-	int count;
-
-	DRM_DEBUG("%d left, %d to send (%p, %p)\n",
-		  left, send, dev->buf_rp, dev->buf_wp);
-
-	if (left == 1 || dev->buf_wp != dev->buf_rp) {
-		DRM_ERROR("Buffer not empty (%d left, wp = %p, rp = %p)\n",
-			  left,
-			  dev->buf_wp,
-			  dev->buf_rp);
-	}
-
-	while (send) {
-		if (dev->buf_wp >= dev->buf_rp) {
-			count = DRM_MIN(send, dev->buf_end - dev->buf_wp);
-			if (count == left) --count; /* Leave a hole */
-		} else {
-			count = DRM_MIN(send, dev->buf_rp - dev->buf_wp - 1);
-		}
-		strncpy(dev->buf_wp, s, count);
-		dev->buf_wp += count;
-		if (dev->buf_wp == dev->buf_end) dev->buf_wp = dev->buf;
-		send -= count;
-	}
-
-	if (dev->buf_async) kill_fasync(&dev->buf_async, SIGIO, POLL_IN);
-
-	DRM_DEBUG("waking\n");
-	wake_up_interruptible(&dev->buf_readers);
-	return 0;
-}
-
-unsigned int gamma_fops_poll(struct file *filp, struct poll_table_struct *wait)
-{
-	drm_file_t   *priv = filp->private_data;
-	drm_device_t *dev  = priv->dev;
-
-	poll_wait(filp, &dev->buf_readers, wait);
-	if (dev->buf_wp != dev->buf_rp) return POLLIN | POLLRDNORM;
-	return 0;
-}
-
-int DRM(context_switch)(drm_device_t *dev, int old, int new)
-{
-	char	    buf[64];
-	drm_queue_t *q;
-
-	if (test_and_set_bit(0, &dev->context_flag)) {
-		DRM_ERROR("Reentering -- FIXME\n");
-		return -EBUSY;
-	}
-
-	DRM_DEBUG("Context switch from %d to %d\n", old, new);
-
-	if (new >= dev->queue_count) {
-		clear_bit(0, &dev->context_flag);
-		return -EINVAL;
-	}
-
-	if (new == dev->last_context) {
-		clear_bit(0, &dev->context_flag);
-		return 0;
-	}
-
-	q = dev->queuelist[new];
-	atomic_inc(&q->use_count);
-	if (atomic_read(&q->use_count) == 1) {
-		atomic_dec(&q->use_count);
-		clear_bit(0, &dev->context_flag);
-		return -EINVAL;
-	}
-
-	/* This causes the X server to wake up & do a bunch of hardware
-	 * interaction to actually effect the context switch.
-	 */
-	sprintf(buf, "C %d %d\n", old, new);
-	DRM(write_string)(dev, buf);
-
-	atomic_dec(&q->use_count);
-
-	return 0;
-}
-
-int DRM(context_switch_complete)(drm_device_t *dev, int new)
-{
-	drm_device_dma_t *dma = dev->dma;
-
-	dev->last_context = new;  /* PRE/POST: This is the _only_ writer. */
-	dev->last_switch  = jiffies;
-
-	if (!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
-		DRM_ERROR("Lock isn't held after context switch\n");
-	}
-
-	if (!dma || !(dma->next_buffer && dma->next_buffer->while_locked)) {
-		if (DRM(lock_free)(dev, &dev->lock.hw_lock->lock,
-				  DRM_KERNEL_CONTEXT)) {
-			DRM_ERROR("Cannot free lock\n");
-		}
-	}
-
-	clear_bit(0, &dev->context_flag);
-	wake_up_interruptible(&dev->context_wait);
-
-	return 0;
-}
-
-static int DRM(init_queue)(drm_device_t *dev, drm_queue_t *q, drm_ctx_t *ctx)
-{
-	DRM_DEBUG("\n");
-
-	if (atomic_read(&q->use_count) != 1
-	    || atomic_read(&q->finalization)
-	    || atomic_read(&q->block_count)) {
-		DRM_ERROR("New queue is already in use: u%d f%d b%d\n",
-			  atomic_read(&q->use_count),
-			  atomic_read(&q->finalization),
-			  atomic_read(&q->block_count));
-	}
-
-	atomic_set(&q->finalization,  0);
-	atomic_set(&q->block_count,   0);
-	atomic_set(&q->block_read,    0);
-	atomic_set(&q->block_write,   0);
-	atomic_set(&q->total_queued,  0);
-	atomic_set(&q->total_flushed, 0);
-	atomic_set(&q->total_locks,   0);
-
-	init_waitqueue_head(&q->write_queue);
-	init_waitqueue_head(&q->read_queue);
-	init_waitqueue_head(&q->flush_queue);
-
-	q->flags = ctx->flags;
-
-	DRM(waitlist_create)(&q->waitlist, dev->dma->buf_count);
-
-	return 0;
-}
-
-
-/* drm_alloc_queue:
-PRE: 1) dev->queuelist[0..dev->queue_count] is allocated and will not
-	disappear (so all deallocation must be done after IOCTLs are off)
-     2) dev->queue_count < dev->queue_slots
-     3) dev->queuelist[i].use_count == 0 and
-	dev->queuelist[i].finalization == 0 if i not in use
-POST: 1) dev->queuelist[i].use_count == 1
-      2) dev->queue_count < dev->queue_slots */
-
-static int DRM(alloc_queue)(drm_device_t *dev)
-{
-	int	    i;
-	drm_queue_t *queue;
-	int	    oldslots;
-	int	    newslots;
-				/* Check for a free queue */
-	for (i = 0; i < dev->queue_count; i++) {
-		atomic_inc(&dev->queuelist[i]->use_count);
-		if (atomic_read(&dev->queuelist[i]->use_count) == 1
-		    && !atomic_read(&dev->queuelist[i]->finalization)) {
-			DRM_DEBUG("%d (free)\n", i);
-			return i;
-		}
-		atomic_dec(&dev->queuelist[i]->use_count);
-	}
-				/* Allocate a new queue */
-	down(&dev->struct_sem);
-
-	queue = DRM(alloc)(sizeof(*queue), DRM_MEM_QUEUES);
-	memset(queue, 0, sizeof(*queue));
-	atomic_set(&queue->use_count, 1);
-
-	++dev->queue_count;
-	if (dev->queue_count >= dev->queue_slots) {
-		oldslots = dev->queue_slots * sizeof(*dev->queuelist);
-		if (!dev->queue_slots) dev->queue_slots = 1;
-		dev->queue_slots *= 2;
-		newslots = dev->queue_slots * sizeof(*dev->queuelist);
-
-		dev->queuelist = DRM(realloc)(dev->queuelist,
-					      oldslots,
-					      newslots,
-					      DRM_MEM_QUEUES);
-		if (!dev->queuelist) {
-			up(&dev->struct_sem);
-			DRM_DEBUG("out of memory\n");
-			return -ENOMEM;
-		}
-	}
-	dev->queuelist[dev->queue_count-1] = queue;
-
-	up(&dev->struct_sem);
-	DRM_DEBUG("%d (new)\n", dev->queue_count - 1);
-	return dev->queue_count - 1;
-}
-
-int DRM(resctx)(struct inode *inode, struct file *filp,
-		unsigned int cmd, unsigned long arg)
-{
-	drm_ctx_res_t __user *argp = (void __user *)arg;
-	drm_ctx_res_t	res;
-	drm_ctx_t	ctx;
-	int		i;
-
-	DRM_DEBUG("%d\n", DRM_RESERVED_CONTEXTS);
-	if (copy_from_user(&res, argp, sizeof(res)))
-		return -EFAULT;
-	if (res.count >= DRM_RESERVED_CONTEXTS) {
-		memset(&ctx, 0, sizeof(ctx));
-		for (i = 0; i < DRM_RESERVED_CONTEXTS; i++) {
-			ctx.handle = i;
-			if (copy_to_user(&res.contexts[i],
-					 &i,
-					 sizeof(i)))
-				return -EFAULT;
-		}
-	}
-	res.count = DRM_RESERVED_CONTEXTS;
-	if (copy_to_user(argp, &res, sizeof(res)))
-		return -EFAULT;
-	return 0;
-}
-
-int DRM(addctx)(struct inode *inode, struct file *filp,
-		unsigned int cmd, unsigned long arg)
-{
-	drm_file_t	*priv	= filp->private_data;
-	drm_device_t	*dev	= priv->dev;
-	drm_ctx_t	ctx;
-	drm_ctx_t	__user *argp = (void __user *)arg;
-
-	if (copy_from_user(&ctx, argp, sizeof(ctx)))
-		return -EFAULT;
-	if ((ctx.handle = DRM(alloc_queue)(dev)) == DRM_KERNEL_CONTEXT) {
-				/* Init kernel's context and get a new one. */
-		DRM(init_queue)(dev, dev->queuelist[ctx.handle], &ctx);
-		ctx.handle = DRM(alloc_queue)(dev);
-	}
-	DRM(init_queue)(dev, dev->queuelist[ctx.handle], &ctx);
-	DRM_DEBUG("%d\n", ctx.handle);
-	if (copy_to_user(argp, &ctx, sizeof(ctx)))
-		return -EFAULT;
-	return 0;
-}
-
-int DRM(modctx)(struct inode *inode, struct file *filp,
-		unsigned int cmd, unsigned long arg)
-{
-	drm_file_t	*priv	= filp->private_data;
-	drm_device_t	*dev	= priv->dev;
-	drm_ctx_t	ctx;
-	drm_queue_t	*q;
-
-	if (copy_from_user(&ctx, (drm_ctx_t __user *)arg, sizeof(ctx)))
-		return -EFAULT;
-
-	DRM_DEBUG("%d\n", ctx.handle);
-
-	if (ctx.handle < 0 || ctx.handle >= dev->queue_count) return -EINVAL;
-	q = dev->queuelist[ctx.handle];
-
-	atomic_inc(&q->use_count);
-	if (atomic_read(&q->use_count) == 1) {
-				/* No longer in use */
-		atomic_dec(&q->use_count);
-		return -EINVAL;
-	}
-
-	if (DRM_BUFCOUNT(&q->waitlist)) {
-		atomic_dec(&q->use_count);
-		return -EBUSY;
-	}
-
-	q->flags = ctx.flags;
-
-	atomic_dec(&q->use_count);
-	return 0;
-}
-
-int DRM(getctx)(struct inode *inode, struct file *filp,
-		unsigned int cmd, unsigned long arg)
-{
-	drm_file_t	*priv	= filp->private_data;
-	drm_device_t	*dev	= priv->dev;
-	drm_ctx_t	__user *argp = (void __user *)arg;
-	drm_ctx_t	ctx;
-	drm_queue_t	*q;
-
-	if (copy_from_user(&ctx, argp, sizeof(ctx)))
-		return -EFAULT;
-
-	DRM_DEBUG("%d\n", ctx.handle);
-
-	if (ctx.handle >= dev->queue_count) return -EINVAL;
-	q = dev->queuelist[ctx.handle];
-
-	atomic_inc(&q->use_count);
-	if (atomic_read(&q->use_count) == 1) {
-				/* No longer in use */
-		atomic_dec(&q->use_count);
-		return -EINVAL;
-	}
-
-	ctx.flags = q->flags;
-	atomic_dec(&q->use_count);
-
-	if (copy_to_user(argp, &ctx, sizeof(ctx)))
-		return -EFAULT;
-
-	return 0;
-}
-
-int DRM(switchctx)(struct inode *inode, struct file *filp,
-		   unsigned int cmd, unsigned long arg)
-{
-	drm_file_t	*priv	= filp->private_data;
-	drm_device_t	*dev	= priv->dev;
-	drm_ctx_t	ctx;
-
-	if (copy_from_user(&ctx, (drm_ctx_t __user *)arg, sizeof(ctx)))
-		return -EFAULT;
-	DRM_DEBUG("%d\n", ctx.handle);
-	return DRM(context_switch)(dev, dev->last_context, ctx.handle);
-}
-
-int DRM(newctx)(struct inode *inode, struct file *filp,
-		unsigned int cmd, unsigned long arg)
-{
-	drm_file_t	*priv	= filp->private_data;
-	drm_device_t	*dev	= priv->dev;
-	drm_ctx_t	ctx;
-
-	if (copy_from_user(&ctx, (drm_ctx_t __user *)arg, sizeof(ctx)))
-		return -EFAULT;
-	DRM_DEBUG("%d\n", ctx.handle);
-	DRM(context_switch_complete)(dev, ctx.handle);
-
-	return 0;
-}
-
-int DRM(rmctx)(struct inode *inode, struct file *filp,
-	       unsigned int cmd, unsigned long arg)
-{
-	drm_file_t	*priv	= filp->private_data;
-	drm_device_t	*dev	= priv->dev;
-	drm_ctx_t	ctx;
-	drm_queue_t	*q;
-	drm_buf_t	*buf;
-
-	if (copy_from_user(&ctx, (drm_ctx_t __user *)arg, sizeof(ctx)))
-		return -EFAULT;
-	DRM_DEBUG("%d\n", ctx.handle);
-
-	if (ctx.handle >= dev->queue_count) return -EINVAL;
-	q = dev->queuelist[ctx.handle];
-
-	atomic_inc(&q->use_count);
-	if (atomic_read(&q->use_count) == 1) {
-				/* No longer in use */
-		atomic_dec(&q->use_count);
-		return -EINVAL;
-	}
-
-	atomic_inc(&q->finalization); /* Mark queue in finalization state */
-	atomic_sub(2, &q->use_count); /* Mark queue as unused (pending
-					 finalization) */
-
-	while (test_and_set_bit(0, &dev->interrupt_flag)) {
-		schedule();
-		if (signal_pending(current)) {
-			clear_bit(0, &dev->interrupt_flag);
-			return -EINTR;
-		}
-	}
-				/* Remove queued buffers */
-	while ((buf = DRM(waitlist_get)(&q->waitlist))) {
-		DRM(free_buffer)(dev, buf);
-	}
-	clear_bit(0, &dev->interrupt_flag);
-
-				/* Wakeup blocked processes */
-	wake_up_interruptible(&q->read_queue);
-	wake_up_interruptible(&q->write_queue);
-	wake_up_interruptible(&q->flush_queue);
-
-				/* Finalization over.  Queue is made
-				   available when both use_count and
-				   finalization become 0, which won't
-				   happen until all the waiting processes
-				   stop waiting. */
-	atomic_dec(&q->finalization);
-	return 0;
-}
-
diff --git a/drivers/char/drm/gamma_dma.c b/drivers/char/drm/gamma_dma.c
deleted file mode 100644
index e486fb8d31e9..000000000000
--- a/drivers/char/drm/gamma_dma.c
+++ /dev/null
@@ -1,946 +0,0 @@
-/* gamma_dma.c -- DMA support for GMX 2000 -*- linux-c -*-
- * Created: Fri Mar 19 14:30:16 1999 by faith@precisioninsight.com
- *
- * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
- * Copyright 2000 VA Linux Systems, Inc., Sunnyvale, California.
- * All Rights Reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice (including the next
- * paragraph) shall be included in all copies or substantial portions of the
- * Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
- * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
- * DEALINGS IN THE SOFTWARE.
- *
- * Authors:
- *    Rickard E. (Rik) Faith <faith@valinux.com>
- *
- */
-
-#include "gamma.h"
-#include "drmP.h"
-#include "drm.h"
-#include "gamma_drm.h"
-#include "gamma_drv.h"
-
-#include <linux/interrupt.h>	/* For task queue support */
-#include <linux/delay.h>
-
-static inline void gamma_dma_dispatch(drm_device_t *dev, unsigned long address,
-				      unsigned long length)
-{
-	drm_gamma_private_t *dev_priv =
-				(drm_gamma_private_t *)dev->dev_private;
-	mb();
-	while ( GAMMA_READ(GAMMA_INFIFOSPACE) < 2)
-		cpu_relax();
-
-	GAMMA_WRITE(GAMMA_DMAADDRESS, address);
-
-	while (GAMMA_READ(GAMMA_GCOMMANDSTATUS) != 4)
-		cpu_relax();
-
-	GAMMA_WRITE(GAMMA_DMACOUNT, length / 4);
-}
-
-void gamma_dma_quiescent_single(drm_device_t *dev)
-{
-	drm_gamma_private_t *dev_priv =
-				(drm_gamma_private_t *)dev->dev_private;
-	while (GAMMA_READ(GAMMA_DMACOUNT))
-		cpu_relax();
-
-	while (GAMMA_READ(GAMMA_INFIFOSPACE) < 2)
-		cpu_relax();
-
-	GAMMA_WRITE(GAMMA_FILTERMODE, 1 << 10);
-	GAMMA_WRITE(GAMMA_SYNC, 0);
-
-	do {
-		while (!GAMMA_READ(GAMMA_OUTFIFOWORDS))
-			cpu_relax();
-	} while (GAMMA_READ(GAMMA_OUTPUTFIFO) != GAMMA_SYNC_TAG);
-}
-
-void gamma_dma_quiescent_dual(drm_device_t *dev)
-{
-	drm_gamma_private_t *dev_priv =
-				(drm_gamma_private_t *)dev->dev_private;
-	while (GAMMA_READ(GAMMA_DMACOUNT))
-		cpu_relax();
-
-	while (GAMMA_READ(GAMMA_INFIFOSPACE) < 3)
-		cpu_relax();
-
-	GAMMA_WRITE(GAMMA_BROADCASTMASK, 3);
-	GAMMA_WRITE(GAMMA_FILTERMODE, 1 << 10);
-	GAMMA_WRITE(GAMMA_SYNC, 0);
-
-	/* Read from first MX */
-	do {
-		while (!GAMMA_READ(GAMMA_OUTFIFOWORDS))
-			cpu_relax();
-	} while (GAMMA_READ(GAMMA_OUTPUTFIFO) != GAMMA_SYNC_TAG);
-
-	/* Read from second MX */
-	do {
-		while (!GAMMA_READ(GAMMA_OUTFIFOWORDS + 0x10000))
-			cpu_relax();
-	} while (GAMMA_READ(GAMMA_OUTPUTFIFO + 0x10000) != GAMMA_SYNC_TAG);
-}
-
-void gamma_dma_ready(drm_device_t *dev)
-{
-	drm_gamma_private_t *dev_priv =
-				(drm_gamma_private_t *)dev->dev_private;
-	while (GAMMA_READ(GAMMA_DMACOUNT))
-		cpu_relax();
-}
-
-static inline int gamma_dma_is_ready(drm_device_t *dev)
-{
-	drm_gamma_private_t *dev_priv =
-				(drm_gamma_private_t *)dev->dev_private;
-	return (!GAMMA_READ(GAMMA_DMACOUNT));
-}
-
-irqreturn_t gamma_driver_irq_handler( DRM_IRQ_ARGS )
-{
-	drm_device_t	 *dev = (drm_device_t *)arg;
-	drm_device_dma_t *dma = dev->dma;
-	drm_gamma_private_t *dev_priv =
-				(drm_gamma_private_t *)dev->dev_private;
-
-	/* FIXME: should check whether we're actually interested in the interrupt? */
-	atomic_inc(&dev->counts[6]); /* _DRM_STAT_IRQ */
-
-	while (GAMMA_READ(GAMMA_INFIFOSPACE) < 3)
-		cpu_relax();
-
-	GAMMA_WRITE(GAMMA_GDELAYTIMER, 0xc350/2); /* 0x05S */
-	GAMMA_WRITE(GAMMA_GCOMMANDINTFLAGS, 8);
-	GAMMA_WRITE(GAMMA_GINTFLAGS, 0x2001);
-	if (gamma_dma_is_ready(dev)) {
-				/* Free previous buffer */
-		if (test_and_set_bit(0, &dev->dma_flag))
-			return IRQ_HANDLED;
-		if (dma->this_buffer) {
-			gamma_free_buffer(dev, dma->this_buffer);
-			dma->this_buffer = NULL;
-		}
-		clear_bit(0, &dev->dma_flag);
-
-		/* Dispatch new buffer */
-		schedule_work(&dev->work);
-	}
-	return IRQ_HANDLED;
-}
-
-/* Only called by gamma_dma_schedule. */
-static int gamma_do_dma(drm_device_t *dev, int locked)
-{
-	unsigned long	 address;
-	unsigned long	 length;
-	drm_buf_t	 *buf;
-	int		 retcode = 0;
-	drm_device_dma_t *dma = dev->dma;
-
-	if (test_and_set_bit(0, &dev->dma_flag)) return -EBUSY;
-
-
-	if (!dma->next_buffer) {
-		DRM_ERROR("No next_buffer\n");
-		clear_bit(0, &dev->dma_flag);
-		return -EINVAL;
-	}
-
-	buf	= dma->next_buffer;
-	/* WE NOW ARE ON LOGICAL PAGES!! - using page table setup in dma_init */
-	/* So we pass the buffer index value into the physical page offset */
-	address = buf->idx << 12;
-	length	= buf->used;
-
-	DRM_DEBUG("context %d, buffer %d (%ld bytes)\n",
-		  buf->context, buf->idx, length);
-
-	if (buf->list == DRM_LIST_RECLAIM) {
-		gamma_clear_next_buffer(dev);
-		gamma_free_buffer(dev, buf);
-		clear_bit(0, &dev->dma_flag);
-		return -EINVAL;
-	}
-
-	if (!length) {
-		DRM_ERROR("0 length buffer\n");
-		gamma_clear_next_buffer(dev);
-		gamma_free_buffer(dev, buf);
-		clear_bit(0, &dev->dma_flag);
-		return 0;
-	}
-
-	if (!gamma_dma_is_ready(dev)) {
-		clear_bit(0, &dev->dma_flag);
-		return -EBUSY;
-	}
-
-	if (buf->while_locked) {
-		if (!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
-			DRM_ERROR("Dispatching buffer %d from pid %d"
-				  " \"while locked\", but no lock held\n",
-				  buf->idx, current->pid);
-		}
-	} else {
-		if (!locked && !gamma_lock_take(&dev->lock.hw_lock->lock,
-					      DRM_KERNEL_CONTEXT)) {
-			clear_bit(0, &dev->dma_flag);
-			return -EBUSY;
-		}
-	}
-
-	if (dev->last_context != buf->context
-	    && !(dev->queuelist[buf->context]->flags
-		 & _DRM_CONTEXT_PRESERVED)) {
-				/* PRE: dev->last_context != buf->context */
-		if (DRM(context_switch)(dev, dev->last_context,
-					buf->context)) {
-			DRM(clear_next_buffer)(dev);
-			DRM(free_buffer)(dev, buf);
-		}
-		retcode = -EBUSY;
-		goto cleanup;
-
-				/* POST: we will wait for the context
-				   switch and will dispatch on a later call
-				   when dev->last_context == buf->context.
-				   NOTE WE HOLD THE LOCK THROUGHOUT THIS
-				   TIME! */
-	}
-
-	gamma_clear_next_buffer(dev);
-	buf->pending	 = 1;
-	buf->waiting	 = 0;
-	buf->list	 = DRM_LIST_PEND;
-
-	/* WE NOW ARE ON LOGICAL PAGES!!! - overriding address */
-	address = buf->idx << 12;
-
-	gamma_dma_dispatch(dev, address, length);
-	gamma_free_buffer(dev, dma->this_buffer);
-	dma->this_buffer = buf;
-
-	atomic_inc(&dev->counts[7]); /* _DRM_STAT_DMA */
-	atomic_add(length, &dev->counts[8]); /* _DRM_STAT_PRIMARY */
-
-	if (!buf->while_locked && !dev->context_flag && !locked) {
-		if (gamma_lock_free(dev, &dev->lock.hw_lock->lock,
-				  DRM_KERNEL_CONTEXT)) {
-			DRM_ERROR("\n");
-		}
-	}
-cleanup:
-
-	clear_bit(0, &dev->dma_flag);
-
-
-	return retcode;
-}
-
-static void gamma_dma_timer_bh(unsigned long dev)
-{
-	gamma_dma_schedule((drm_device_t *)dev, 0);
-}
-
-void gamma_irq_immediate_bh(void *dev)
-{
-	gamma_dma_schedule(dev, 0);
-}
-
-int gamma_dma_schedule(drm_device_t *dev, int locked)
-{
-	int		 next;
-	drm_queue_t	 *q;
-	drm_buf_t	 *buf;
-	int		 retcode   = 0;
-	int		 processed = 0;
-	int		 missed;
-	int		 expire	   = 20;
-	drm_device_dma_t *dma	   = dev->dma;
-
-	if (test_and_set_bit(0, &dev->interrupt_flag)) {
-				/* Not reentrant */
-		atomic_inc(&dev->counts[10]); /* _DRM_STAT_MISSED */
-		return -EBUSY;
-	}
-	missed = atomic_read(&dev->counts[10]);
-
-
-again:
-	if (dev->context_flag) {
-		clear_bit(0, &dev->interrupt_flag);
-		return -EBUSY;
-	}
-	if (dma->next_buffer) {
-				/* Unsent buffer that was previously
-				   selected, but that couldn't be sent
-				   because the lock could not be obtained
-				   or the DMA engine wasn't ready.  Try
-				   again. */
-		if (!(retcode = gamma_do_dma(dev, locked))) ++processed;
-	} else {
-		do {
-			next = gamma_select_queue(dev, gamma_dma_timer_bh);
-			if (next >= 0) {
-				q   = dev->queuelist[next];
-				buf = gamma_waitlist_get(&q->waitlist);
-				dma->next_buffer = buf;
-				dma->next_queue	 = q;
-				if (buf && buf->list == DRM_LIST_RECLAIM) {
-					gamma_clear_next_buffer(dev);
-					gamma_free_buffer(dev, buf);
-				}
-			}
-		} while (next >= 0 && !dma->next_buffer);
-		if (dma->next_buffer) {
-			if (!(retcode = gamma_do_dma(dev, locked))) {
-				++processed;
-			}
-		}
-	}
-
-	if (--expire) {
-		if (missed != atomic_read(&dev->counts[10])) {
-			if (gamma_dma_is_ready(dev)) goto again;
-		}
-		if (processed && gamma_dma_is_ready(dev)) {
-			processed = 0;
-			goto again;
-		}
-	}
-
-	clear_bit(0, &dev->interrupt_flag);
-
-	return retcode;
-}
-
-static int gamma_dma_priority(struct file *filp, 
-			      drm_device_t *dev, drm_dma_t *d)
-{
-	unsigned long	  address;
-	unsigned long	  length;
-	int		  must_free = 0;
-	int		  retcode   = 0;
-	int		  i;
-	int		  idx;
-	drm_buf_t	  *buf;
-	drm_buf_t	  *last_buf = NULL;
-	drm_device_dma_t  *dma	    = dev->dma;
-	int		  *send_indices = NULL;
-	int		  *send_sizes = NULL;
-
-	DECLARE_WAITQUEUE(entry, current);
-
-				/* Turn off interrupt handling */
-	while (test_and_set_bit(0, &dev->interrupt_flag)) {
-		schedule();
-		if (signal_pending(current)) return -EINTR;
-	}
-	if (!(d->flags & _DRM_DMA_WHILE_LOCKED)) {
-		while (!gamma_lock_take(&dev->lock.hw_lock->lock,
-				      DRM_KERNEL_CONTEXT)) {
-			schedule();
-			if (signal_pending(current)) {
-				clear_bit(0, &dev->interrupt_flag);
-				return -EINTR;
-			}
-		}
-		++must_free;
-	}
-
-	send_indices = DRM(alloc)(d->send_count * sizeof(*send_indices),
-				  DRM_MEM_DRIVER);
-	if (send_indices == NULL)
-		return -ENOMEM;
-	if (copy_from_user(send_indices, d->send_indices, 
-			   d->send_count * sizeof(*send_indices))) {
-		retcode = -EFAULT;
-                goto cleanup;
-	}
-	
-	send_sizes = DRM(alloc)(d->send_count * sizeof(*send_sizes),
-				DRM_MEM_DRIVER);
-	if (send_sizes == NULL)
-		return -ENOMEM;
-	if (copy_from_user(send_sizes, d->send_sizes, 
-			   d->send_count * sizeof(*send_sizes))) {
-		retcode = -EFAULT;
-                goto cleanup;
-	}
-
-	for (i = 0; i < d->send_count; i++) {
-		idx = send_indices[i];
-		if (idx < 0 || idx >= dma->buf_count) {
-			DRM_ERROR("Index %d (of %d max)\n",
-				  send_indices[i], dma->buf_count - 1);
-			continue;
-		}
-		buf = dma->buflist[ idx ];
-		if (buf->filp != filp) {
-			DRM_ERROR("Process %d using buffer not owned\n",
-				  current->pid);
-			retcode = -EINVAL;
-			goto cleanup;
-		}
-		if (buf->list != DRM_LIST_NONE) {
-			DRM_ERROR("Process %d using buffer on list %d\n",
-				  current->pid, buf->list);
-			retcode = -EINVAL;
-			goto cleanup;
-		}
-				/* This isn't a race condition on
-				   buf->list, since our concern is the
-				   buffer reclaim during the time the
-				   process closes the /dev/drm? handle, so
-				   it can't also be doing DMA. */
-		buf->list	  = DRM_LIST_PRIO;
-		buf->used	  = send_sizes[i];
-		buf->context	  = d->context;
-		buf->while_locked = d->flags & _DRM_DMA_WHILE_LOCKED;
-		address		  = (unsigned long)buf->address;
-		length		  = buf->used;
-		if (!length) {
-			DRM_ERROR("0 length buffer\n");
-		}
-		if (buf->pending) {
-			DRM_ERROR("Sending pending buffer:"
-				  " buffer %d, offset %d\n",
-				  send_indices[i], i);
-			retcode = -EINVAL;
-			goto cleanup;
-		}
-		if (buf->waiting) {
-			DRM_ERROR("Sending waiting buffer:"
-				  " buffer %d, offset %d\n",
-				  send_indices[i], i);
-			retcode = -EINVAL;
-			goto cleanup;
-		}
-		buf->pending = 1;
-
-		if (dev->last_context != buf->context
-		    && !(dev->queuelist[buf->context]->flags
-			 & _DRM_CONTEXT_PRESERVED)) {
-			add_wait_queue(&dev->context_wait, &entry);
-			current->state = TASK_INTERRUPTIBLE;
-				/* PRE: dev->last_context != buf->context */
-			DRM(context_switch)(dev, dev->last_context,
-					    buf->context);
-				/* POST: we will wait for the context
-				   switch and will dispatch on a later call
-				   when dev->last_context == buf->context.
-				   NOTE WE HOLD THE LOCK THROUGHOUT THIS
-				   TIME! */
-			schedule();
-			current->state = TASK_RUNNING;
-			remove_wait_queue(&dev->context_wait, &entry);
-			if (signal_pending(current)) {
-				retcode = -EINTR;
-				goto cleanup;
-			}
-			if (dev->last_context != buf->context) {
-				DRM_ERROR("Context mismatch: %d %d\n",
-					  dev->last_context,
-					  buf->context);
-			}
-		}
-
-		gamma_dma_dispatch(dev, address, length);
-		atomic_inc(&dev->counts[9]); /* _DRM_STAT_SPECIAL */
-		atomic_add(length, &dev->counts[8]); /* _DRM_STAT_PRIMARY */
-
-		if (last_buf) {
-			gamma_free_buffer(dev, last_buf);
-		}
-		last_buf = buf;
-	}
-
-
-cleanup:
-	if (last_buf) {
-		gamma_dma_ready(dev);
-		gamma_free_buffer(dev, last_buf);
-	}
-	if (send_indices)
-		DRM(free)(send_indices, d->send_count * sizeof(*send_indices), 
-			  DRM_MEM_DRIVER);
-	if (send_sizes)
-		DRM(free)(send_sizes, d->send_count * sizeof(*send_sizes), 
-			  DRM_MEM_DRIVER);
-
-	if (must_free && !dev->context_flag) {
-		if (gamma_lock_free(dev, &dev->lock.hw_lock->lock,
-				  DRM_KERNEL_CONTEXT)) {
-			DRM_ERROR("\n");
-		}
-	}
-	clear_bit(0, &dev->interrupt_flag);
-	return retcode;
-}
-
-static int gamma_dma_send_buffers(struct file *filp,
-				  drm_device_t *dev, drm_dma_t *d)
-{
-	DECLARE_WAITQUEUE(entry, current);
-	drm_buf_t	  *last_buf = NULL;
-	int		  retcode   = 0;
-	drm_device_dma_t  *dma	    = dev->dma;
-	int               send_index;
-
-	if (get_user(send_index, &d->send_indices[d->send_count-1]))
-		return -EFAULT;
-
-	if (d->flags & _DRM_DMA_BLOCK) {
-		last_buf = dma->buflist[send_index];
-		add_wait_queue(&last_buf->dma_wait, &entry);
-	}
-
-	if ((retcode = gamma_dma_enqueue(filp, d))) {
-		if (d->flags & _DRM_DMA_BLOCK)
-			remove_wait_queue(&last_buf->dma_wait, &entry);
-		return retcode;
-	}
-
-	gamma_dma_schedule(dev, 0);
-
-	if (d->flags & _DRM_DMA_BLOCK) {
-		DRM_DEBUG("%d waiting\n", current->pid);
-		for (;;) {
-			current->state = TASK_INTERRUPTIBLE;
-			if (!last_buf->waiting && !last_buf->pending)
-				break; /* finished */
-			schedule();
-			if (signal_pending(current)) {
-				retcode = -EINTR; /* Can't restart */
-				break;
-			}
-		}
-		current->state = TASK_RUNNING;
-		DRM_DEBUG("%d running\n", current->pid);
-		remove_wait_queue(&last_buf->dma_wait, &entry);
-		if (!retcode
-		    || (last_buf->list==DRM_LIST_PEND && !last_buf->pending)) {
-			if (!waitqueue_active(&last_buf->dma_wait)) {
-				gamma_free_buffer(dev, last_buf);
-			}
-		}
-		if (retcode) {
-			DRM_ERROR("ctx%d w%d p%d c%ld i%d l%d pid:%d\n",
-				  d->context,
-				  last_buf->waiting,
-				  last_buf->pending,
-				  (long)DRM_WAITCOUNT(dev, d->context),
-				  last_buf->idx,
-				  last_buf->list,
-				  current->pid);
-		}
-	}
-	return retcode;
-}
-
-int gamma_dma(struct inode *inode, struct file *filp, unsigned int cmd,
-	      unsigned long arg)
-{
-	drm_file_t	  *priv	    = filp->private_data;
-	drm_device_t	  *dev	    = priv->dev;
-	drm_device_dma_t  *dma	    = dev->dma;
-	int		  retcode   = 0;
-	drm_dma_t	  __user *argp = (void __user *)arg;
-	drm_dma_t	  d;
-
-	if (copy_from_user(&d, argp, sizeof(d)))
-		return -EFAULT;
-
-	if (d.send_count < 0 || d.send_count > dma->buf_count) {
-		DRM_ERROR("Process %d trying to send %d buffers (of %d max)\n",
-			  current->pid, d.send_count, dma->buf_count);
-		return -EINVAL;
-	}
-
-	if (d.request_count < 0 || d.request_count > dma->buf_count) {
-		DRM_ERROR("Process %d trying to get %d buffers (of %d max)\n",
-			  current->pid, d.request_count, dma->buf_count);
-		return -EINVAL;
-	}
-
-	if (d.send_count) {
-		if (d.flags & _DRM_DMA_PRIORITY)
-			retcode = gamma_dma_priority(filp, dev, &d);
-		else
-			retcode = gamma_dma_send_buffers(filp, dev, &d);
-	}
-
-	d.granted_count = 0;
-
-	if (!retcode && d.request_count) {
-		retcode = gamma_dma_get_buffers(filp, &d);
-	}
-
-	DRM_DEBUG("%d returning, granted = %d\n",
-		  current->pid, d.granted_count);
-	if (copy_to_user(argp, &d, sizeof(d)))
-		return -EFAULT;
-
-	return retcode;
-}
-
-/* =============================================================
- * DMA initialization, cleanup
- */
-
-static int gamma_do_init_dma( drm_device_t *dev, drm_gamma_init_t *init )
-{
-	drm_gamma_private_t *dev_priv;
-	drm_device_dma_t    *dma = dev->dma;
-	drm_buf_t	    *buf;
-	int i;
-	struct list_head    *list;
-	unsigned long	    *pgt;
-
-	DRM_DEBUG( "%s\n", __FUNCTION__ );
-
-	dev_priv = DRM(alloc)( sizeof(drm_gamma_private_t),
-							DRM_MEM_DRIVER );
-	if ( !dev_priv )
-		return -ENOMEM;
-
-	dev->dev_private = (void *)dev_priv;
-
-	memset( dev_priv, 0, sizeof(drm_gamma_private_t) );
-
-	dev_priv->num_rast = init->num_rast;
-
-	list_for_each(list, &dev->maplist->head) {
-		drm_map_list_t *r_list = list_entry(list, drm_map_list_t, head);
-		if( r_list->map &&
-		    r_list->map->type == _DRM_SHM &&
-		    r_list->map->flags & _DRM_CONTAINS_LOCK ) {
-			dev_priv->sarea = r_list->map;
- 			break;
- 		}
- 	}
-	
-	dev_priv->mmio0 = drm_core_findmap(dev, init->mmio0);
-	dev_priv->mmio1 = drm_core_findmap(dev, init->mmio1);
-	dev_priv->mmio2 = drm_core_findmap(dev, init->mmio2);
-	dev_priv->mmio3 = drm_core_findmap(dev, init->mmio3);
-	
-	dev_priv->sarea_priv = (drm_gamma_sarea_t *)
-		((u8 *)dev_priv->sarea->handle +
-		 init->sarea_priv_offset);
-
-	if (init->pcimode) {
-		buf = dma->buflist[GLINT_DRI_BUF_COUNT];
-		pgt = buf->address;
-
- 		for (i = 0; i < GLINT_DRI_BUF_COUNT; i++) {
-			buf = dma->buflist[i];
-			*pgt = virt_to_phys((void*)buf->address) | 0x07;
-			pgt++;
-		}
-
-		buf = dma->buflist[GLINT_DRI_BUF_COUNT];
-	} else {
-		dev->agp_buffer_map = drm_core_findmap(dev, init->buffers_offset);
-		drm_core_ioremap( dev->agp_buffer_map, dev);
-
-		buf = dma->buflist[GLINT_DRI_BUF_COUNT];
-		pgt = buf->address;
-
- 		for (i = 0; i < GLINT_DRI_BUF_COUNT; i++) {
-			buf = dma->buflist[i];
-			*pgt = (unsigned long)buf->address + 0x07;
-			pgt++;
-		}
-
-		buf = dma->buflist[GLINT_DRI_BUF_COUNT];
-
-		while (GAMMA_READ(GAMMA_INFIFOSPACE) < 1);
-		GAMMA_WRITE( GAMMA_GDMACONTROL, 0xe);
-	}
-	while (GAMMA_READ(GAMMA_INFIFOSPACE) < 2);
-	GAMMA_WRITE( GAMMA_PAGETABLEADDR, virt_to_phys((void*)buf->address) );
-	GAMMA_WRITE( GAMMA_PAGETABLELENGTH, 2 );
-
-	return 0;
-}
-
-int gamma_do_cleanup_dma( drm_device_t *dev )
-{
-	DRM_DEBUG( "%s\n", __FUNCTION__ );
-
-	/* 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 (drm_core_check_feature(dev, DRIVER_HAVE_IRQ))
-		if ( dev->irq_enabled ) 
-			DRM(irq_uninstall)(dev);
-
-	if ( dev->dev_private ) {
-
-		if ( dev->agp_buffer_map != NULL )
-			drm_core_ioremapfree( dev->agp_buffer_map, dev );
-
-		DRM(free)( dev->dev_private, sizeof(drm_gamma_private_t),
-			   DRM_MEM_DRIVER );
-		dev->dev_private = NULL;
-	}
-
-	return 0;
-}
-
-int gamma_dma_init( struct inode *inode, struct file *filp,
-		  unsigned int cmd, unsigned long arg )
-{
-	drm_file_t *priv = filp->private_data;
-	drm_device_t *dev = priv->dev;
-	drm_gamma_init_t init;
-
-	LOCK_TEST_WITH_RETURN( dev, filp );
-
-	if ( copy_from_user( &init, (drm_gamma_init_t __user *)arg, sizeof(init) ) )
-		return -EFAULT;
-
-	switch ( init.func ) {
-	case GAMMA_INIT_DMA:
-		return gamma_do_init_dma( dev, &init );
-	case GAMMA_CLEANUP_DMA:
-		return gamma_do_cleanup_dma( dev );
-	}
-
-	return -EINVAL;
-}
-
-static int gamma_do_copy_dma( drm_device_t *dev, drm_gamma_copy_t *copy )
-{
-	drm_device_dma_t    *dma = dev->dma;
-	unsigned int        *screenbuf;
-
-	DRM_DEBUG( "%s\n", __FUNCTION__ );
-
-	/* We've DRM_RESTRICTED this DMA buffer */
-
-	screenbuf = dma->buflist[ GLINT_DRI_BUF_COUNT + 1 ]->address;
-
-#if 0
-	*buffer++ = 0x180;	/* Tag (FilterMode) */
-	*buffer++ = 0x200;	/* Allow FBColor through */
-	*buffer++ = 0x53B;	/* Tag */
-	*buffer++ = copy->Pitch;
-	*buffer++ = 0x53A;	/* Tag */
-	*buffer++ = copy->SrcAddress;
-	*buffer++ = 0x539;	/* Tag */
-	*buffer++ = copy->WidthHeight; /* Initiates transfer */
-	*buffer++ = 0x53C;	/* Tag - DMAOutputAddress */
-	*buffer++ = virt_to_phys((void*)screenbuf);
-	*buffer++ = 0x53D;	/* Tag - DMAOutputCount */
-	*buffer++ = copy->Count; /* Reads HostOutFifo BLOCKS until ..*/
-
-	/* Data now sitting in dma->buflist[ GLINT_DRI_BUF_COUNT + 1 ] */
-	/* Now put it back to the screen */
-
-	*buffer++ = 0x180;	/* Tag (FilterMode) */
-	*buffer++ = 0x400;	/* Allow Sync through */
-	*buffer++ = 0x538;	/* Tag - DMARectangleReadTarget */
-	*buffer++ = 0x155;	/* FBSourceData | count */
-	*buffer++ = 0x537;	/* Tag */
-	*buffer++ = copy->Pitch;
-	*buffer++ = 0x536;	/* Tag */
-	*buffer++ = copy->DstAddress;
-	*buffer++ = 0x535;	/* Tag */
-	*buffer++ = copy->WidthHeight; /* Initiates transfer */
-	*buffer++ = 0x530;	/* Tag - DMAAddr */
-	*buffer++ = virt_to_phys((void*)screenbuf);
-	*buffer++ = 0x531;
-	*buffer++ = copy->Count; /* initiates DMA transfer of color data */
-#endif
-
-	/* need to dispatch it now */
-
-	return 0;
-}
-
-int gamma_dma_copy( struct inode *inode, struct file *filp,
-		  unsigned int cmd, unsigned long arg )
-{
-	drm_file_t *priv = filp->private_data;
-	drm_device_t *dev = priv->dev;
-	drm_gamma_copy_t copy;
-
-	if ( copy_from_user( &copy, (drm_gamma_copy_t __user *)arg, sizeof(copy) ) )
-		return -EFAULT;
-
-	return gamma_do_copy_dma( dev, &copy );
-}
-
-/* =============================================================
- * Per Context SAREA Support
- */
-
-int gamma_getsareactx(struct inode *inode, struct file *filp,
-		     unsigned int cmd, unsigned long arg)
-{
-	drm_file_t	*priv	= filp->private_data;
-	drm_device_t	*dev	= priv->dev;
-	drm_ctx_priv_map_t __user *argp = (void __user *)arg;
-	drm_ctx_priv_map_t request;
-	drm_map_t *map;
-
-	if (copy_from_user(&request, argp, sizeof(request)))
-		return -EFAULT;
-
-	down(&dev->struct_sem);
-	if ((int)request.ctx_id >= dev->max_context) {
-		up(&dev->struct_sem);
-		return -EINVAL;
-	}
-
-	map = dev->context_sareas[request.ctx_id];
-	up(&dev->struct_sem);
-
-	request.handle = map->handle;
-	if (copy_to_user(argp, &request, sizeof(request)))
-		return -EFAULT;
-	return 0;
-}
-
-int gamma_setsareactx(struct inode *inode, struct file *filp,
-		     unsigned int cmd, unsigned long arg)
-{
-	drm_file_t	*priv	= filp->private_data;
-	drm_device_t	*dev	= priv->dev;
-	drm_ctx_priv_map_t request;
-	drm_map_t *map = NULL;
-	drm_map_list_t *r_list;
-	struct list_head *list;
-
-	if (copy_from_user(&request,
-			   (drm_ctx_priv_map_t __user *)arg,
-			   sizeof(request)))
-		return -EFAULT;
-
-	down(&dev->struct_sem);
-	r_list = NULL;
-	list_for_each(list, &dev->maplist->head) {
-		r_list = list_entry(list, drm_map_list_t, head);
-		if(r_list->map &&
-		   r_list->map->handle == request.handle) break;
-	}
-	if (list == &(dev->maplist->head)) {
-		up(&dev->struct_sem);
-		return -EINVAL;
-	}
-	map = r_list->map;
-	up(&dev->struct_sem);
-
-	if (!map) return -EINVAL;
-
-	down(&dev->struct_sem);
-	if ((int)request.ctx_id >= dev->max_context) {
-		up(&dev->struct_sem);
-		return -EINVAL;
-	}
-	dev->context_sareas[request.ctx_id] = map;
-	up(&dev->struct_sem);
-	return 0;
-}
-
-void gamma_driver_irq_preinstall( drm_device_t *dev ) {
-	drm_gamma_private_t *dev_priv =
-				(drm_gamma_private_t *)dev->dev_private;
-
-	while(GAMMA_READ(GAMMA_INFIFOSPACE) < 2)
-		cpu_relax();
-
-	GAMMA_WRITE( GAMMA_GCOMMANDMODE,	0x00000004 );
-	GAMMA_WRITE( GAMMA_GDMACONTROL,		0x00000000 );
-}
-
-void gamma_driver_irq_postinstall( drm_device_t *dev ) {
-	drm_gamma_private_t *dev_priv =
-				(drm_gamma_private_t *)dev->dev_private;
-
-	while(GAMMA_READ(GAMMA_INFIFOSPACE) < 3)
-		cpu_relax();
-
-	GAMMA_WRITE( GAMMA_GINTENABLE,		0x00002001 );
-	GAMMA_WRITE( GAMMA_COMMANDINTENABLE,	0x00000008 );
-	GAMMA_WRITE( GAMMA_GDELAYTIMER,		0x00039090 );
-}
-
-void gamma_driver_irq_uninstall( drm_device_t *dev ) {
-	drm_gamma_private_t *dev_priv =
-				(drm_gamma_private_t *)dev->dev_private;
-	if (!dev_priv)
-		return;
-
-	while(GAMMA_READ(GAMMA_INFIFOSPACE) < 3)
-		cpu_relax();
-
-	GAMMA_WRITE( GAMMA_GDELAYTIMER,		0x00000000 );
-	GAMMA_WRITE( GAMMA_COMMANDINTENABLE,	0x00000000 );
-	GAMMA_WRITE( GAMMA_GINTENABLE,		0x00000000 );
-}
-
-extern drm_ioctl_desc_t DRM(ioctls)[];
-
-static int gamma_driver_preinit(drm_device_t *dev)
-{
-	/* reset the finish ioctl */
-	DRM(ioctls)[DRM_IOCTL_NR(DRM_IOCTL_FINISH)].func = DRM(finish);
-	return 0;
-}
-
-static void gamma_driver_pretakedown(drm_device_t *dev)
-{
-	gamma_do_cleanup_dma(dev);
-}
-
-static void gamma_driver_dma_ready(drm_device_t *dev)
-{
-	gamma_dma_ready(dev);
-}
-
-static int gamma_driver_dma_quiescent(drm_device_t *dev)
-{
-	drm_gamma_private_t *dev_priv =	(
-		drm_gamma_private_t *)dev->dev_private;
-	if (dev_priv->num_rast == 2)
-		gamma_dma_quiescent_dual(dev);
-	else gamma_dma_quiescent_single(dev);
-	return 0;
-}
-
-void gamma_driver_register_fns(drm_device_t *dev)
-{
-	dev->driver_features = DRIVER_USE_AGP | DRIVER_USE_MTRR | DRIVER_PCI_DMA | DRIVER_HAVE_DMA | DRIVER_HAVE_IRQ;
-	DRM(fops).read = gamma_fops_read;
-	DRM(fops).poll = gamma_fops_poll;
-	dev->driver.preinit = gamma_driver_preinit;
-	dev->driver.pretakedown = gamma_driver_pretakedown;
-	dev->driver.dma_ready = gamma_driver_dma_ready;
-	dev->driver.dma_quiescent = gamma_driver_dma_quiescent;
-	dev->driver.dma_flush_block_and_flush = gamma_flush_block_and_flush;
-	dev->driver.dma_flush_unblock = gamma_flush_unblock;
-}
diff --git a/drivers/char/drm/gamma_drm.h b/drivers/char/drm/gamma_drm.h
deleted file mode 100644
index 20819ded0e15..000000000000
--- a/drivers/char/drm/gamma_drm.h
+++ /dev/null
@@ -1,90 +0,0 @@
-#ifndef _GAMMA_DRM_H_
-#define _GAMMA_DRM_H_
-
-typedef struct _drm_gamma_tex_region {
-	unsigned char next, prev; /* indices to form a circular LRU  */
-	unsigned char in_use;	/* owned by a client, or free? */
-	int age;		/* tracked by clients to update local LRU's */
-} drm_gamma_tex_region_t;
-
-typedef struct {
-	unsigned int	GDeltaMode;
-	unsigned int	GDepthMode;
-	unsigned int	GGeometryMode;
-	unsigned int	GTransformMode;
-} drm_gamma_context_regs_t;
-
-typedef struct _drm_gamma_sarea {
-   	drm_gamma_context_regs_t context_state;
-
-	unsigned int dirty;
-
-
-	/* Maintain an LRU of contiguous regions of texture space.  If
-	 * you think you own a region of texture memory, and it has an
-	 * age different to the one you set, then you are mistaken and
-	 * it has been stolen by another client.  If global texAge
-	 * hasn't changed, there is no need to walk the list.
-	 *
-	 * These regions can be used as a proxy for the fine-grained
-	 * texture information of other clients - by maintaining them
-	 * in the same lru which is used to age their own textures,
-	 * clients have an approximate lru for the whole of global
-	 * texture space, and can make informed decisions as to which
-	 * areas to kick out.  There is no need to choose whether to
-	 * kick out your own texture or someone else's - simply eject
-	 * them all in LRU order.  
-	 */
-   
-#define GAMMA_NR_TEX_REGIONS 64
-	drm_gamma_tex_region_t texList[GAMMA_NR_TEX_REGIONS+1]; 
-				/* Last elt is sentinal */
-        int texAge;		/* last time texture was uploaded */
-        int last_enqueue;	/* last time a buffer was enqueued */
-	int last_dispatch;	/* age of the most recently dispatched buffer */
-	int last_quiescent;     /*  */
-	int ctxOwner;		/* last context to upload state */
-
-	int vertex_prim;
-} drm_gamma_sarea_t;
-
-/* WARNING: If you change any of these defines, make sure to change the
- * defines in the Xserver file (xf86drmGamma.h)
- */
-
-/* Gamma specific ioctls
- * The device specific ioctl range is 0x40 to 0x79.
- */
-#define DRM_IOCTL_GAMMA_INIT		DRM_IOW( 0x40, drm_gamma_init_t)
-#define DRM_IOCTL_GAMMA_COPY		DRM_IOW( 0x41, drm_gamma_copy_t)
-
-typedef struct drm_gamma_copy {
-	unsigned int	DMAOutputAddress;
-	unsigned int	DMAOutputCount;
-	unsigned int	DMAReadGLINTSource;
-	unsigned int	DMARectangleWriteAddress;
-	unsigned int	DMARectangleWriteLinePitch;
-	unsigned int	DMARectangleWrite;
-	unsigned int	DMARectangleReadAddress;
-	unsigned int	DMARectangleReadLinePitch;
-	unsigned int	DMARectangleRead;
-	unsigned int	DMARectangleReadTarget;
-} drm_gamma_copy_t;
-
-typedef struct drm_gamma_init {
-   	enum {
-	   	GAMMA_INIT_DMA    = 0x01,
-	       	GAMMA_CLEANUP_DMA = 0x02
-	} func;
-
-   	int sarea_priv_offset;
-	int pcimode;
-	unsigned int mmio0;
-	unsigned int mmio1;
-	unsigned int mmio2;
-	unsigned int mmio3;
-	unsigned int buffers_offset;
-	int num_rast;
-} drm_gamma_init_t;
-
-#endif /* _GAMMA_DRM_H_ */
diff --git a/drivers/char/drm/gamma_drv.c b/drivers/char/drm/gamma_drv.c
deleted file mode 100644
index e7e64b62792a..000000000000
--- a/drivers/char/drm/gamma_drv.c
+++ /dev/null
@@ -1,59 +0,0 @@
-/* gamma.c -- 3dlabs GMX 2000 driver -*- linux-c -*-
- * Created: Mon Jan  4 08:58:31 1999 by faith@precisioninsight.com
- *
- * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
- * Copyright 2000 VA Linux Systems, Inc., Sunnyvale, California.
- * All Rights Reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice (including the next
- * paragraph) shall be included in all copies or substantial portions of the
- * Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
- * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
- * DEALINGS IN THE SOFTWARE.
- *
- * Authors:
- *    Rickard E. (Rik) Faith <faith@valinux.com>
- *    Gareth Hughes <gareth@valinux.com>
- */
-
-#include <linux/config.h>
-#include "gamma.h"
-#include "drmP.h"
-#include "drm.h"
-#include "gamma_drm.h"
-#include "gamma_drv.h"
-
-#include "drm_auth.h"
-#include "drm_agpsupport.h"
-#include "drm_bufs.h"
-#include "gamma_context.h"	/* NOTE! */
-#include "drm_dma.h"
-#include "gamma_old_dma.h"	/* NOTE */
-#include "drm_drawable.h"
-#include "drm_drv.h"
-
-#include "drm_fops.h"
-#include "drm_init.h"
-#include "drm_ioctl.h"
-#include "drm_irq.h"
-#include "gamma_lists.h"        /* NOTE */
-#include "drm_lock.h"
-#include "gamma_lock.h"		/* NOTE */
-#include "drm_memory.h"
-#include "drm_proc.h"
-#include "drm_vm.h"
-#include "drm_stub.h"
-#include "drm_scatter.h"
diff --git a/drivers/char/drm/gamma_drv.h b/drivers/char/drm/gamma_drv.h
deleted file mode 100644
index 146fcc6253cd..000000000000
--- a/drivers/char/drm/gamma_drv.h
+++ /dev/null
@@ -1,147 +0,0 @@
-/* gamma_drv.h -- Private header for 3dlabs GMX 2000 driver -*- linux-c -*-
- * Created: Mon Jan  4 10:05:05 1999 by faith@precisioninsight.com
- *
- * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
- * Copyright 2000 VA Linux Systems, Inc., Sunnyvale, California.
- * All rights reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice (including the next
- * paragraph) shall be included in all copies or substantial portions of the
- * Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
- * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
- * DEALINGS IN THE SOFTWARE.
- *
- * Authors:
- *    Rickard E. (Rik) Faith <faith@valinux.com>
- *
- */
-
-#ifndef _GAMMA_DRV_H_
-#define _GAMMA_DRV_H_
-
-typedef struct drm_gamma_private {
-	drm_gamma_sarea_t *sarea_priv;
-	drm_map_t *sarea;
-	drm_map_t *mmio0;
-	drm_map_t *mmio1;
-	drm_map_t *mmio2;
-	drm_map_t *mmio3;
-	int num_rast;
-} drm_gamma_private_t;
-
-				/* gamma_dma.c */
-extern int gamma_dma_init( struct inode *inode, struct file *filp,
-			 unsigned int cmd, unsigned long arg );
-extern int gamma_dma_copy( struct inode *inode, struct file *filp,
-			 unsigned int cmd, unsigned long arg );
-
-extern int gamma_do_cleanup_dma( drm_device_t *dev );
-extern void gamma_dma_ready(drm_device_t *dev);
-extern void gamma_dma_quiescent_single(drm_device_t *dev);
-extern void gamma_dma_quiescent_dual(drm_device_t *dev);
-
-				/* gamma_dma.c */
-extern int  gamma_dma_schedule(drm_device_t *dev, int locked);
-extern int  gamma_dma(struct inode *inode, struct file *filp,
-		      unsigned int cmd, unsigned long arg);
-extern int  gamma_find_devices(void);
-extern int  gamma_found(void);
-
-/* Gamma-specific code pulled from drm_fops.h:
- */
-extern int	     DRM(finish)(struct inode *inode, struct file *filp,
-				 unsigned int cmd, unsigned long arg);
-extern int	     DRM(flush_unblock)(drm_device_t *dev, int context,
-					drm_lock_flags_t flags);
-extern int	     DRM(flush_block_and_flush)(drm_device_t *dev, int context,
-						drm_lock_flags_t flags);
-
-/* Gamma-specific code pulled from drm_dma.h:
- */
-extern void	     DRM(clear_next_buffer)(drm_device_t *dev);
-extern int	     DRM(select_queue)(drm_device_t *dev,
-				       void (*wrapper)(unsigned long));
-extern int	     DRM(dma_enqueue)(struct file *filp, drm_dma_t *dma);
-extern int	     DRM(dma_get_buffers)(struct file *filp, drm_dma_t *dma);
-
-
-/* Gamma-specific code pulled from drm_lists.h (now renamed gamma_lists.h):
- */
-extern int	     DRM(waitlist_create)(drm_waitlist_t *bl, int count);
-extern int	     DRM(waitlist_destroy)(drm_waitlist_t *bl);
-extern int	     DRM(waitlist_put)(drm_waitlist_t *bl, drm_buf_t *buf);
-extern drm_buf_t     *DRM(waitlist_get)(drm_waitlist_t *bl);
-extern int	     DRM(freelist_create)(drm_freelist_t *bl, int count);
-extern int	     DRM(freelist_destroy)(drm_freelist_t *bl);
-extern int	     DRM(freelist_put)(drm_device_t *dev, drm_freelist_t *bl,
-				       drm_buf_t *buf);
-extern drm_buf_t     *DRM(freelist_get)(drm_freelist_t *bl, int block);
-
-/* externs for gamma changes to the ops */
-extern struct file_operations DRM(fops);
-extern unsigned int gamma_fops_poll(struct file *filp, struct poll_table_struct *wait);
-extern ssize_t gamma_fops_read(struct file *filp, char __user *buf, size_t count, loff_t *off);
-
-
-#define GLINT_DRI_BUF_COUNT 256
-
-#define GAMMA_OFF(reg)						   \
-	((reg < 0x1000)						   \
-	 ? reg							   \
-	 : ((reg < 0x10000)					   \
-	    ? (reg - 0x1000)					   \
-	    : ((reg < 0x11000)					   \
-	       ? (reg - 0x10000)				   \
-	       : (reg - 0x11000))))
-
-#define GAMMA_BASE(reg)	 ((unsigned long)				     \
-			  ((reg < 0x1000)    ? dev_priv->mmio0->handle :     \
-			   ((reg < 0x10000)  ? dev_priv->mmio1->handle :     \
-			    ((reg < 0x11000) ? dev_priv->mmio2->handle :     \
-					       dev_priv->mmio3->handle))))
-#define GAMMA_ADDR(reg)	 (GAMMA_BASE(reg) + GAMMA_OFF(reg))
-#define GAMMA_DEREF(reg) *(__volatile__ int *)GAMMA_ADDR(reg)
-#define GAMMA_READ(reg)	 GAMMA_DEREF(reg)
-#define GAMMA_WRITE(reg,val) do { GAMMA_DEREF(reg) = val; } while (0)
-
-#define GAMMA_BROADCASTMASK    0x9378
-#define GAMMA_COMMANDINTENABLE 0x0c48
-#define GAMMA_DMAADDRESS       0x0028
-#define GAMMA_DMACOUNT	       0x0030
-#define GAMMA_FILTERMODE       0x8c00
-#define GAMMA_GCOMMANDINTFLAGS 0x0c50
-#define GAMMA_GCOMMANDMODE     0x0c40
-#define		GAMMA_QUEUED_DMA_MODE		1<<1
-#define GAMMA_GCOMMANDSTATUS   0x0c60
-#define GAMMA_GDELAYTIMER      0x0c38
-#define GAMMA_GDMACONTROL      0x0060
-#define 	GAMMA_USE_AGP			1<<1
-#define GAMMA_GINTENABLE       0x0808
-#define GAMMA_GINTFLAGS	       0x0810
-#define GAMMA_INFIFOSPACE      0x0018
-#define GAMMA_OUTFIFOWORDS     0x0020
-#define GAMMA_OUTPUTFIFO       0x2000
-#define GAMMA_SYNC	       0x8c40
-#define GAMMA_SYNC_TAG	       0x0188
-#define GAMMA_PAGETABLEADDR    0x0C00
-#define GAMMA_PAGETABLELENGTH  0x0C08
-
-#define GAMMA_PASSTHROUGH	0x1FE
-#define GAMMA_DMAADDRTAG	0x530
-#define GAMMA_DMACOUNTTAG	0x531
-#define GAMMA_COMMANDINTTAG	0x532
-
-#endif
diff --git a/drivers/char/drm/gamma_lists.h b/drivers/char/drm/gamma_lists.h
deleted file mode 100644
index 2d93f412b96b..000000000000
--- a/drivers/char/drm/gamma_lists.h
+++ /dev/null
@@ -1,215 +0,0 @@
-/* drm_lists.h -- Buffer list handling routines -*- linux-c -*-
- * Created: Mon Apr 19 20:54:22 1999 by faith@valinux.com
- *
- * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
- * Copyright 2000 VA Linux Systems, Inc., Sunnyvale, California.
- * 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
- * VA LINUX SYSTEMS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
- * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
- * OTHER DEALINGS IN THE SOFTWARE.
- *
- * Authors:
- *    Rickard E. (Rik) Faith <faith@valinux.com>
- *    Gareth Hughes <gareth@valinux.com>
- */
-
-#include "drmP.h"
-
-
-int DRM(waitlist_create)(drm_waitlist_t *bl, int count)
-{
-	if (bl->count) return -EINVAL;
-
-	bl->bufs       = DRM(alloc)((bl->count + 2) * sizeof(*bl->bufs),
-				    DRM_MEM_BUFLISTS);
-
-	if(!bl->bufs) return -ENOMEM;
-	memset(bl->bufs, 0, sizeof(*bl->bufs));
-	bl->count      = count;
-	bl->rp	       = bl->bufs;
-	bl->wp	       = bl->bufs;
-	bl->end	       = &bl->bufs[bl->count+1];
-	spin_lock_init(&bl->write_lock);
-	spin_lock_init(&bl->read_lock);
-	return 0;
-}
-
-int DRM(waitlist_destroy)(drm_waitlist_t *bl)
-{
-	if (bl->rp != bl->wp) return -EINVAL;
-	if (bl->bufs) DRM(free)(bl->bufs,
-				(bl->count + 2) * sizeof(*bl->bufs),
-				DRM_MEM_BUFLISTS);
-	bl->count = 0;
-	bl->bufs  = NULL;
-	bl->rp	  = NULL;
-	bl->wp	  = NULL;
-	bl->end	  = NULL;
-	return 0;
-}
-
-int DRM(waitlist_put)(drm_waitlist_t *bl, drm_buf_t *buf)
-{
-	int	      left;
-	unsigned long flags;
-
-	left = DRM_LEFTCOUNT(bl);
-	if (!left) {
-		DRM_ERROR("Overflow while adding buffer %d from filp %p\n",
-			  buf->idx, buf->filp);
-		return -EINVAL;
-	}
-	buf->list	 = DRM_LIST_WAIT;
-
-	spin_lock_irqsave(&bl->write_lock, flags);
-	*bl->wp = buf;
-	if (++bl->wp >= bl->end) bl->wp = bl->bufs;
-	spin_unlock_irqrestore(&bl->write_lock, flags);
-
-	return 0;
-}
-
-drm_buf_t *DRM(waitlist_get)(drm_waitlist_t *bl)
-{
-	drm_buf_t     *buf;
-	unsigned long flags;
-
-	spin_lock_irqsave(&bl->read_lock, flags);
-	buf = *bl->rp;
-	if (bl->rp == bl->wp) {
-		spin_unlock_irqrestore(&bl->read_lock, flags);
-		return NULL;
-	}
-	if (++bl->rp >= bl->end) bl->rp = bl->bufs;
-	spin_unlock_irqrestore(&bl->read_lock, flags);
-
-	return buf;
-}
-
-int DRM(freelist_create)(drm_freelist_t *bl, int count)
-{
-	atomic_set(&bl->count, 0);
-	bl->next      = NULL;
-	init_waitqueue_head(&bl->waiting);
-	bl->low_mark  = 0;
-	bl->high_mark = 0;
-	atomic_set(&bl->wfh,   0);
-	spin_lock_init(&bl->lock);
-	++bl->initialized;
-	return 0;
-}
-
-int DRM(freelist_destroy)(drm_freelist_t *bl)
-{
-	atomic_set(&bl->count, 0);
-	bl->next = NULL;
-	return 0;
-}
-
-int DRM(freelist_put)(drm_device_t *dev, drm_freelist_t *bl, drm_buf_t *buf)
-{
-	drm_device_dma_t *dma  = dev->dma;
-
-	if (!dma) {
-		DRM_ERROR("No DMA support\n");
-		return 1;
-	}
-
-	if (buf->waiting || buf->pending || buf->list == DRM_LIST_FREE) {
-		DRM_ERROR("Freed buffer %d: w%d, p%d, l%d\n",
-			  buf->idx, buf->waiting, buf->pending, buf->list);
-	}
-	if (!bl) return 1;
-	buf->list	= DRM_LIST_FREE;
-
-	spin_lock(&bl->lock);
-	buf->next	= bl->next;
-	bl->next	= buf;
-	spin_unlock(&bl->lock);
-
-	atomic_inc(&bl->count);
-	if (atomic_read(&bl->count) > dma->buf_count) {
-		DRM_ERROR("%d of %d buffers free after addition of %d\n",
-			  atomic_read(&bl->count), dma->buf_count, buf->idx);
-		return 1;
-	}
-				/* Check for high water mark */
-	if (atomic_read(&bl->wfh) && atomic_read(&bl->count)>=bl->high_mark) {
-		atomic_set(&bl->wfh, 0);
-		wake_up_interruptible(&bl->waiting);
-	}
-	return 0;
-}
-
-static drm_buf_t *DRM(freelist_try)(drm_freelist_t *bl)
-{
-	drm_buf_t	  *buf;
-
-	if (!bl) return NULL;
-
-				/* Get buffer */
-	spin_lock(&bl->lock);
-	if (!bl->next) {
-		spin_unlock(&bl->lock);
-		return NULL;
-	}
-	buf	  = bl->next;
-	bl->next  = bl->next->next;
-	spin_unlock(&bl->lock);
-
-	atomic_dec(&bl->count);
-	buf->next = NULL;
-	buf->list = DRM_LIST_NONE;
-	if (buf->waiting || buf->pending) {
-		DRM_ERROR("Free buffer %d: w%d, p%d, l%d\n",
-			  buf->idx, buf->waiting, buf->pending, buf->list);
-	}
-
-	return buf;
-}
-
-drm_buf_t *DRM(freelist_get)(drm_freelist_t *bl, int block)
-{
-	drm_buf_t	  *buf	= NULL;
-	DECLARE_WAITQUEUE(entry, current);
-
-	if (!bl || !bl->initialized) return NULL;
-
-				/* Check for low water mark */
-	if (atomic_read(&bl->count) <= bl->low_mark) /* Became low */
-		atomic_set(&bl->wfh, 1);
-	if (atomic_read(&bl->wfh)) {
-		if (block) {
-			add_wait_queue(&bl->waiting, &entry);
-			for (;;) {
-				current->state = TASK_INTERRUPTIBLE;
-				if (!atomic_read(&bl->wfh)
-				    && (buf = DRM(freelist_try)(bl))) break;
-				schedule();
-				if (signal_pending(current)) break;
-			}
-			current->state = TASK_RUNNING;
-			remove_wait_queue(&bl->waiting, &entry);
-		}
-		return buf;
-	}
-
-	return DRM(freelist_try)(bl);
-}
-
diff --git a/drivers/char/drm/gamma_lock.h b/drivers/char/drm/gamma_lock.h
deleted file mode 100644
index ddec67e4ed16..000000000000
--- a/drivers/char/drm/gamma_lock.h
+++ /dev/null
@@ -1,140 +0,0 @@
-/* lock.c -- IOCTLs for locking -*- linux-c -*-
- * Created: Tue Feb  2 08:37:54 1999 by faith@valinux.com
- *
- * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
- * Copyright 2000 VA Linux Systems, Inc., Sunnyvale, California.
- * 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
- * VA LINUX SYSTEMS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
- * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
- * OTHER DEALINGS IN THE SOFTWARE.
- *
- * Authors:
- *    Rickard E. (Rik) Faith <faith@valinux.com>
- *    Gareth Hughes <gareth@valinux.com>
- */
-
-
-/* Gamma-specific code extracted from drm_lock.h:
- */
-static int DRM(flush_queue)(drm_device_t *dev, int context)
-{
-	DECLARE_WAITQUEUE(entry, current);
-	int		  ret	= 0;
-	drm_queue_t	  *q	= dev->queuelist[context];
-
-	DRM_DEBUG("\n");
-
-	atomic_inc(&q->use_count);
-	if (atomic_read(&q->use_count) > 1) {
-		atomic_inc(&q->block_write);
-		add_wait_queue(&q->flush_queue, &entry);
-		atomic_inc(&q->block_count);
-		for (;;) {
-			current->state = TASK_INTERRUPTIBLE;
-			if (!DRM_BUFCOUNT(&q->waitlist)) break;
-			schedule();
-			if (signal_pending(current)) {
-				ret = -EINTR; /* Can't restart */
-				break;
-			}
-		}
-		atomic_dec(&q->block_count);
-		current->state = TASK_RUNNING;
-		remove_wait_queue(&q->flush_queue, &entry);
-	}
-	atomic_dec(&q->use_count);
-
-				/* NOTE: block_write is still incremented!
-				   Use drm_flush_unlock_queue to decrement. */
-	return ret;
-}
-
-static int DRM(flush_unblock_queue)(drm_device_t *dev, int context)
-{
-	drm_queue_t	  *q	= dev->queuelist[context];
-
-	DRM_DEBUG("\n");
-
-	atomic_inc(&q->use_count);
-	if (atomic_read(&q->use_count) > 1) {
-		if (atomic_read(&q->block_write)) {
-			atomic_dec(&q->block_write);
-			wake_up_interruptible(&q->write_queue);
-		}
-	}
-	atomic_dec(&q->use_count);
-	return 0;
-}
-
-int DRM(flush_block_and_flush)(drm_device_t *dev, int context,
-			       drm_lock_flags_t flags)
-{
-	int ret = 0;
-	int i;
-
-	DRM_DEBUG("\n");
-
-	if (flags & _DRM_LOCK_FLUSH) {
-		ret = DRM(flush_queue)(dev, DRM_KERNEL_CONTEXT);
-		if (!ret) ret = DRM(flush_queue)(dev, context);
-	}
-	if (flags & _DRM_LOCK_FLUSH_ALL) {
-		for (i = 0; !ret && i < dev->queue_count; i++) {
-			ret = DRM(flush_queue)(dev, i);
-		}
-	}
-	return ret;
-}
-
-int DRM(flush_unblock)(drm_device_t *dev, int context, drm_lock_flags_t flags)
-{
-	int ret = 0;
-	int i;
-
-	DRM_DEBUG("\n");
-
-	if (flags & _DRM_LOCK_FLUSH) {
-		ret = DRM(flush_unblock_queue)(dev, DRM_KERNEL_CONTEXT);
-		if (!ret) ret = DRM(flush_unblock_queue)(dev, context);
-	}
-	if (flags & _DRM_LOCK_FLUSH_ALL) {
-		for (i = 0; !ret && i < dev->queue_count; i++) {
-			ret = DRM(flush_unblock_queue)(dev, i);
-		}
-	}
-
-	return ret;
-}
-
-int DRM(finish)(struct inode *inode, struct file *filp, unsigned int cmd,
-		unsigned long arg)
-{
-	drm_file_t	  *priv	  = filp->private_data;
-	drm_device_t	  *dev	  = priv->dev;
-	int		  ret	  = 0;
-	drm_lock_t	  lock;
-
-	DRM_DEBUG("\n");
-
-	if (copy_from_user(&lock, (drm_lock_t __user *)arg, sizeof(lock)))
-		return -EFAULT;
-	ret = DRM(flush_block_and_flush)(dev, lock.context, lock.flags);
-	DRM(flush_unblock)(dev, lock.context, lock.flags);
-	return ret;
-}
diff --git a/drivers/char/drm/gamma_old_dma.h b/drivers/char/drm/gamma_old_dma.h
deleted file mode 100644
index abdd454aab9f..000000000000
--- a/drivers/char/drm/gamma_old_dma.h
+++ /dev/null
@@ -1,313 +0,0 @@
-/* drm_dma.c -- DMA IOCTL and function support -*- linux-c -*-
- * Created: Fri Mar 19 14:30:16 1999 by faith@valinux.com
- *
- * Copyright 1999, 2000 Precision Insight, Inc., Cedar Park, Texas.
- * Copyright 2000 VA Linux Systems, Inc., Sunnyvale, California.
- * 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
- * VA LINUX SYSTEMS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
- * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
- * OTHER DEALINGS IN THE SOFTWARE.
- *
- * Authors:
- *    Rickard E. (Rik) Faith <faith@valinux.com>
- *    Gareth Hughes <gareth@valinux.com>
- */
-
-
-/* Gamma-specific code pulled from drm_dma.h:
- */
-
-void DRM(clear_next_buffer)(drm_device_t *dev)
-{
-	drm_device_dma_t *dma = dev->dma;
-
-	dma->next_buffer = NULL;
-	if (dma->next_queue && !DRM_BUFCOUNT(&dma->next_queue->waitlist)) {
-		wake_up_interruptible(&dma->next_queue->flush_queue);
-	}
-	dma->next_queue	 = NULL;
-}
-
-int DRM(select_queue)(drm_device_t *dev, void (*wrapper)(unsigned long))
-{
-	int	   i;
-	int	   candidate = -1;
-	int	   j	     = jiffies;
-
-	if (!dev) {
-		DRM_ERROR("No device\n");
-		return -1;
-	}
-	if (!dev->queuelist || !dev->queuelist[DRM_KERNEL_CONTEXT]) {
-				/* This only happens between the time the
-				   interrupt is initialized and the time
-				   the queues are initialized. */
-		return -1;
-	}
-
-				/* Doing "while locked" DMA? */
-	if (DRM_WAITCOUNT(dev, DRM_KERNEL_CONTEXT)) {
-		return DRM_KERNEL_CONTEXT;
-	}
-
-				/* If there are buffers on the last_context
-				   queue, and we have not been executing
-				   this context very long, continue to
-				   execute this context. */
-	if (dev->last_switch <= j
-	    && dev->last_switch + DRM_TIME_SLICE > j
-	    && DRM_WAITCOUNT(dev, dev->last_context)) {
-		return dev->last_context;
-	}
-
-				/* Otherwise, find a candidate */
-	for (i = dev->last_checked + 1; i < dev->queue_count; i++) {
-		if (DRM_WAITCOUNT(dev, i)) {
-			candidate = dev->last_checked = i;
-			break;
-		}
-	}
-
-	if (candidate < 0) {
-		for (i = 0; i < dev->queue_count; i++) {
-			if (DRM_WAITCOUNT(dev, i)) {
-				candidate = dev->last_checked = i;
-				break;
-			}
-		}
-	}
-
-	if (wrapper
-	    && candidate >= 0
-	    && candidate != dev->last_context
-	    && dev->last_switch <= j
-	    && dev->last_switch + DRM_TIME_SLICE > j) {
-		if (dev->timer.expires != dev->last_switch + DRM_TIME_SLICE) {
-			del_timer(&dev->timer);
-			dev->timer.function = wrapper;
-			dev->timer.data	    = (unsigned long)dev;
-			dev->timer.expires  = dev->last_switch+DRM_TIME_SLICE;
-			add_timer(&dev->timer);
-		}
-		return -1;
-	}
-
-	return candidate;
-}
-
-
-int DRM(dma_enqueue)(struct file *filp, drm_dma_t *d)
-{
-	drm_file_t    *priv   = filp->private_data;
-	drm_device_t  *dev    = priv->dev;
-	int		  i;
-	drm_queue_t	  *q;
-	drm_buf_t	  *buf;
-	int		  idx;
-	int		  while_locked = 0;
-	drm_device_dma_t  *dma = dev->dma;
-	int		  *ind;
-	int		  err;
-	DECLARE_WAITQUEUE(entry, current);
-
-	DRM_DEBUG("%d\n", d->send_count);
-
-	if (d->flags & _DRM_DMA_WHILE_LOCKED) {
-		int context = dev->lock.hw_lock->lock;
-
-		if (!_DRM_LOCK_IS_HELD(context)) {
-			DRM_ERROR("No lock held during \"while locked\""
-				  " request\n");
-			return -EINVAL;
-		}
-		if (d->context != _DRM_LOCKING_CONTEXT(context)
-		    && _DRM_LOCKING_CONTEXT(context) != DRM_KERNEL_CONTEXT) {
-			DRM_ERROR("Lock held by %d while %d makes"
-				  " \"while locked\" request\n",
-				  _DRM_LOCKING_CONTEXT(context),
-				  d->context);
-			return -EINVAL;
-		}
-		q = dev->queuelist[DRM_KERNEL_CONTEXT];
-		while_locked = 1;
-	} else {
-		q = dev->queuelist[d->context];
-	}
-
-
-	atomic_inc(&q->use_count);
-	if (atomic_read(&q->block_write)) {
-		add_wait_queue(&q->write_queue, &entry);
-		atomic_inc(&q->block_count);
-		for (;;) {
-			current->state = TASK_INTERRUPTIBLE;
-			if (!atomic_read(&q->block_write)) break;
-			schedule();
-			if (signal_pending(current)) {
-				atomic_dec(&q->use_count);
-				remove_wait_queue(&q->write_queue, &entry);
-				return -EINTR;
-			}
-		}
-		atomic_dec(&q->block_count);
-		current->state = TASK_RUNNING;
-		remove_wait_queue(&q->write_queue, &entry);
-	}
-
-	ind = DRM(alloc)(d->send_count * sizeof(int), DRM_MEM_DRIVER);
-	if (!ind)
-		return -ENOMEM;
-
-	if (copy_from_user(ind, d->send_indices, d->send_count * sizeof(int))) {
-		err = -EFAULT;
-                goto out;
-	}
-
-	err = -EINVAL;
-	for (i = 0; i < d->send_count; i++) {
-		idx = ind[i];
-		if (idx < 0 || idx >= dma->buf_count) {
-			DRM_ERROR("Index %d (of %d max)\n",
-				  ind[i], dma->buf_count - 1);
-			goto out;
-		}
-		buf = dma->buflist[ idx ];
-		if (buf->filp != filp) {
-			DRM_ERROR("Process %d using buffer not owned\n",
-				  current->pid);
-			goto out;
-		}
-		if (buf->list != DRM_LIST_NONE) {
-			DRM_ERROR("Process %d using buffer %d on list %d\n",
-				  current->pid, buf->idx, buf->list);
-			goto out;
-		}
-		buf->used	  = ind[i];
-		buf->while_locked = while_locked;
-		buf->context	  = d->context;
-		if (!buf->used) {
-			DRM_ERROR("Queueing 0 length buffer\n");
-		}
-		if (buf->pending) {
-			DRM_ERROR("Queueing pending buffer:"
-				  " buffer %d, offset %d\n",
-				  ind[i], i);
-			goto out;
-		}
-		if (buf->waiting) {
-			DRM_ERROR("Queueing waiting buffer:"
-				  " buffer %d, offset %d\n",
-				  ind[i], i);
-			goto out;
-		}
-		buf->waiting = 1;
-		if (atomic_read(&q->use_count) == 1
-		    || atomic_read(&q->finalization)) {
-			DRM(free_buffer)(dev, buf);
-		} else {
-			DRM(waitlist_put)(&q->waitlist, buf);
-			atomic_inc(&q->total_queued);
-		}
-	}
-	atomic_dec(&q->use_count);
-
-	return 0;
-
-out:
-	DRM(free)(ind, d->send_count * sizeof(int), DRM_MEM_DRIVER);
-	atomic_dec(&q->use_count);
-	return err;
-}
-
-static int DRM(dma_get_buffers_of_order)(struct file *filp, drm_dma_t *d,
-					 int order)
-{
-	drm_file_t    *priv   = filp->private_data;
-	drm_device_t  *dev    = priv->dev;
-	int		  i;
-	drm_buf_t	  *buf;
-	drm_device_dma_t  *dma = dev->dma;
-
-	for (i = d->granted_count; i < d->request_count; i++) {
-		buf = DRM(freelist_get)(&dma->bufs[order].freelist,
-					d->flags & _DRM_DMA_WAIT);
-		if (!buf) break;
-		if (buf->pending || buf->waiting) {
-			DRM_ERROR("Free buffer %d in use: filp %p (w%d, p%d)\n",
-				  buf->idx,
-				  buf->filp,
-				  buf->waiting,
-				  buf->pending);
-		}
-		buf->filp     = filp;
-		if (copy_to_user(&d->request_indices[i],
-				 &buf->idx,
-				 sizeof(buf->idx)))
-			return -EFAULT;
-
-		if (copy_to_user(&d->request_sizes[i],
-				 &buf->total,
-				 sizeof(buf->total)))
-			return -EFAULT;
-
-		++d->granted_count;
-	}
-	return 0;
-}
-
-
-int DRM(dma_get_buffers)(struct file *filp, drm_dma_t *dma)
-{
-	int		  order;
-	int		  retcode = 0;
-	int		  tmp_order;
-
-	order = DRM(order)(dma->request_size);
-
-	dma->granted_count = 0;
-	retcode		   = DRM(dma_get_buffers_of_order)(filp, dma, order);
-
-	if (dma->granted_count < dma->request_count
-	    && (dma->flags & _DRM_DMA_SMALLER_OK)) {
-		for (tmp_order = order - 1;
-		     !retcode
-			     && dma->granted_count < dma->request_count
-			     && tmp_order >= DRM_MIN_ORDER;
-		     --tmp_order) {
-
-			retcode = DRM(dma_get_buffers_of_order)(filp, dma,
-								tmp_order);
-		}
-	}
-
-	if (dma->granted_count < dma->request_count
-	    && (dma->flags & _DRM_DMA_LARGER_OK)) {
-		for (tmp_order = order + 1;
-		     !retcode
-			     && dma->granted_count < dma->request_count
-			     && tmp_order <= DRM_MAX_ORDER;
-		     ++tmp_order) {
-
-			retcode = DRM(dma_get_buffers_of_order)(filp, dma,
-								tmp_order);
-		}
-	}
-	return 0;
-}
-
diff --git a/drivers/char/drm/i810_dma.c b/drivers/char/drm/i810_dma.c
index 18e0b7622893..2f1659b96fd1 100644
--- a/drivers/char/drm/i810_dma.c
+++ b/drivers/char/drm/i810_dma.c
@@ -45,11 +45,6 @@
 #define I810_BUF_UNMAPPED 0
 #define I810_BUF_MAPPED   1
 
-#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,4,2)
-#define down_write down
-#define up_write up
-#endif
-
 static drm_buf_t *i810_freelist_get(drm_device_t *dev)
 {
    	drm_device_dma_t *dma = dev->dma;
@@ -351,6 +346,7 @@ static int i810_dma_initialize(drm_device_t *dev,
 	   	DRM_ERROR("can not find mmio map!\n");
 	   	return -EINVAL;
 	}
+	dev->agp_buffer_token = init->buffers_offset;
 	dev->agp_buffer_map = drm_core_findmap(dev, init->buffers_offset);
 	if (!dev->agp_buffer_map) {
 		dev->dev_private = (void *)dev_priv;
@@ -1383,3 +1379,19 @@ drm_ioctl_desc_t i810_ioctls[] = {
 };
 
 int i810_max_ioctl = DRM_ARRAY_SIZE(i810_ioctls);
+
+/**
+ * Determine if the device really is AGP or not.
+ *
+ * All Intel graphics chipsets are treated as AGP, even if they are really
+ * PCI-e.
+ *
+ * \param dev   The device to be tested.
+ *
+ * \returns
+ * A value of 1 is always retured to indictate every i810 is AGP.
+ */
+int i810_driver_device_is_agp(drm_device_t * dev)
+{
+	return 1;
+}
diff --git a/drivers/char/drm/i810_drv.c b/drivers/char/drm/i810_drv.c
index ff51b3259af9..00609329d578 100644
--- a/drivers/char/drm/i810_drv.c
+++ b/drivers/char/drm/i810_drv.c
@@ -84,6 +84,7 @@ static struct drm_driver driver = {
 	.dev_priv_size = sizeof(drm_i810_buf_priv_t),
 	.pretakedown = i810_driver_pretakedown,
 	.prerelease = i810_driver_prerelease,
+	.device_is_agp = i810_driver_device_is_agp,
 	.release = i810_driver_release,
 	.dma_quiescent = i810_driver_dma_quiescent,
 	.reclaim_buffers = i810_reclaim_buffers,
diff --git a/drivers/char/drm/i810_drv.h b/drivers/char/drm/i810_drv.h
index 1b40538d1725..62ee4f58c59a 100644
--- a/drivers/char/drm/i810_drv.h
+++ b/drivers/char/drm/i810_drv.h
@@ -120,6 +120,7 @@ extern int i810_driver_dma_quiescent(drm_device_t *dev);
 extern void i810_driver_release(drm_device_t *dev, struct file *filp);
 extern void i810_driver_pretakedown(drm_device_t *dev);
 extern void i810_driver_prerelease(drm_device_t *dev, DRMFILE filp);
+extern int i810_driver_device_is_agp(drm_device_t * dev);
 
 #define I810_BASE(reg)		((unsigned long) \
 				dev_priv->mmio_map->handle)
diff --git a/drivers/char/drm/i830_dma.c b/drivers/char/drm/i830_dma.c
index dc7733035864..6f89d5796ef3 100644
--- a/drivers/char/drm/i830_dma.c
+++ b/drivers/char/drm/i830_dma.c
@@ -47,11 +47,6 @@
 #define I830_BUF_UNMAPPED 0
 #define I830_BUF_MAPPED   1
 
-#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,4,2)
-#define down_write down
-#define up_write up
-#endif
-
 static drm_buf_t *i830_freelist_get(drm_device_t *dev)
 {
    	drm_device_dma_t *dma = dev->dma;
@@ -358,6 +353,7 @@ static int i830_dma_initialize(drm_device_t *dev,
 		DRM_ERROR("can not find mmio map!\n");
 		return -EINVAL;
 	}
+	dev->agp_buffer_token = init->buffers_offset;
 	dev->agp_buffer_map = drm_core_findmap(dev, init->buffers_offset);
 	if(!dev->agp_buffer_map) {
 		dev->dev_private = (void *)dev_priv;
@@ -1586,3 +1582,19 @@ drm_ioctl_desc_t i830_ioctls[] = {
 };
 
 int i830_max_ioctl = DRM_ARRAY_SIZE(i830_ioctls);
+
+/**
+ * Determine if the device really is AGP or not.
+ *
+ * All Intel graphics chipsets are treated as AGP, even if they are really
+ * PCI-e.
+ *
+ * \param dev   The device to be tested.
+ *
+ * \returns
+ * A value of 1 is always retured to indictate every i8xx is AGP.
+ */
+int i830_driver_device_is_agp(drm_device_t * dev)
+{
+	return 1;
+}
diff --git a/drivers/char/drm/i830_drv.c b/drivers/char/drm/i830_drv.c
index bc36be76b8b2..0da9cd19919e 100644
--- a/drivers/char/drm/i830_drv.c
+++ b/drivers/char/drm/i830_drv.c
@@ -88,6 +88,7 @@ static struct drm_driver driver = {
 	.dev_priv_size = sizeof(drm_i830_buf_priv_t),
 	.pretakedown = i830_driver_pretakedown,
 	.prerelease = i830_driver_prerelease,
+	.device_is_agp = i830_driver_device_is_agp,
 	.release = i830_driver_release,
 	.dma_quiescent = i830_driver_dma_quiescent,
 	.reclaim_buffers = i830_reclaim_buffers,
diff --git a/drivers/char/drm/i830_drv.h b/drivers/char/drm/i830_drv.h
index df7746131dea..63f96a8b6a4a 100644
--- a/drivers/char/drm/i830_drv.h
+++ b/drivers/char/drm/i830_drv.h
@@ -137,6 +137,7 @@ extern void i830_driver_pretakedown(drm_device_t *dev);
 extern void i830_driver_release(drm_device_t *dev, struct file *filp);
 extern int i830_driver_dma_quiescent(drm_device_t *dev);
 extern void i830_driver_prerelease(drm_device_t *dev, DRMFILE filp);
+extern int i830_driver_device_is_agp(drm_device_t * dev);
 
 #define I830_BASE(reg)		((unsigned long) \
 				dev_priv->mmio_map->handle)
diff --git a/drivers/char/drm/i915_dma.c b/drivers/char/drm/i915_dma.c
index acf9e52a9507..34f552f90c4a 100644
--- a/drivers/char/drm/i915_dma.c
+++ b/drivers/char/drm/i915_dma.c
@@ -95,9 +95,8 @@ static int i915_dma_cleanup(drm_device_t * dev)
 			drm_core_ioremapfree( &dev_priv->ring.map, dev);
 		}
 
-		if (dev_priv->hw_status_page) {
-			drm_pci_free(dev, PAGE_SIZE, dev_priv->hw_status_page,
-				     dev_priv->dma_status_page);
+		if (dev_priv->status_page_dmah) {
+			drm_pci_free(dev, dev_priv->status_page_dmah);
 			/* Need to rewrite hardware status page */
 			I915_WRITE(0x02080, 0x1ffff000);
 		}
@@ -174,16 +173,18 @@ static int i915_initialize(drm_device_t * dev,
 	dev_priv->allow_batchbuffer = 1;
 
 	/* Program Hardware Status Page */
-	dev_priv->hw_status_page = drm_pci_alloc(dev, PAGE_SIZE, PAGE_SIZE,
-						 0xffffffff, 
-						 &dev_priv->dma_status_page);
+	dev_priv->status_page_dmah = drm_pci_alloc(dev, PAGE_SIZE, PAGE_SIZE,
+						   0xffffffff);
 
-	if (!dev_priv->hw_status_page) {
+	if (!dev_priv->status_page_dmah) {
 		dev->dev_private = (void *)dev_priv;
 		i915_dma_cleanup(dev);
 		DRM_ERROR("Can not allocate hardware status page\n");
 		return DRM_ERR(ENOMEM);
 	}
+	dev_priv->hw_status_page = dev_priv->status_page_dmah->vaddr;
+	dev_priv->dma_status_page = dev_priv->status_page_dmah->busaddr;
+
 	memset(dev_priv->hw_status_page, 0, PAGE_SIZE);
 	DRM_DEBUG("hw status page @ %p\n", dev_priv->hw_status_page);
 
@@ -731,3 +732,19 @@ drm_ioctl_desc_t i915_ioctls[] = {
 };
 
 int i915_max_ioctl = DRM_ARRAY_SIZE(i915_ioctls);
+
+/**
+ * Determine if the device really is AGP or not.
+ *
+ * All Intel graphics chipsets are treated as AGP, even if they are really
+ * PCI-e.
+ *
+ * \param dev   The device to be tested.
+ *
+ * \returns
+ * A value of 1 is always retured to indictate every i9x5 is AGP.
+ */
+int i915_driver_device_is_agp(drm_device_t * dev)
+{
+	return 1;
+}
diff --git a/drivers/char/drm/i915_drv.c b/drivers/char/drm/i915_drv.c
index 1f59d3fc79bc..106b9ec02213 100644
--- a/drivers/char/drm/i915_drv.c
+++ b/drivers/char/drm/i915_drv.c
@@ -79,6 +79,7 @@ static struct drm_driver driver = {
 				DRIVER_HAVE_IRQ | DRIVER_IRQ_SHARED,
 	.pretakedown = i915_driver_pretakedown,
 	.prerelease = i915_driver_prerelease,
+	.device_is_agp = i915_driver_device_is_agp,
 	.irq_preinstall = i915_driver_irq_preinstall,
 	.irq_postinstall = i915_driver_irq_postinstall,
 	.irq_uninstall = i915_driver_irq_uninstall,
diff --git a/drivers/char/drm/i915_drv.h b/drivers/char/drm/i915_drv.h
index 9c37d2367dd5..70ed4e68eac8 100644
--- a/drivers/char/drm/i915_drv.h
+++ b/drivers/char/drm/i915_drv.h
@@ -79,9 +79,10 @@ typedef struct drm_i915_private {
 	drm_i915_sarea_t *sarea_priv;
 	drm_i915_ring_buffer_t ring;
 
+	drm_dma_handle_t *status_page_dmah;
 	void *hw_status_page;
-	unsigned long counter;
 	dma_addr_t dma_status_page;
+	unsigned long counter;
 
 	int back_offset;
 	int front_offset;
@@ -102,6 +103,7 @@ typedef struct drm_i915_private {
 extern void i915_kernel_lost_context(drm_device_t * dev);
 extern void i915_driver_pretakedown(drm_device_t *dev);
 extern void i915_driver_prerelease(drm_device_t *dev, DRMFILE filp);
+extern int i915_driver_device_is_agp(drm_device_t *dev);
 
 /* i915_irq.c */
 extern int i915_irq_emit(DRM_IOCTL_ARGS);
diff --git a/drivers/char/drm/mga_dma.c b/drivers/char/drm/mga_dma.c
index 832eaf8a5068..567b425b784f 100644
--- a/drivers/char/drm/mga_dma.c
+++ b/drivers/char/drm/mga_dma.c
@@ -23,18 +23,21 @@
  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  * DEALINGS IN THE SOFTWARE.
- *
- * Authors:
- *    Rickard E. (Rik) Faith <faith@valinux.com>
- *    Jeff Hartmann <jhartmann@valinux.com>
- *    Keith Whitwell <keith@tungstengraphics.com>
- *
- * Rewritten by:
- *    Gareth Hughes <gareth@valinux.com>
+ */
+
+/**
+ * \file mga_dma.c
+ * DMA support for MGA G200 / G400.
+ * 
+ * \author Rickard E. (Rik) Faith <faith@valinux.com>
+ * \author Jeff Hartmann <jhartmann@valinux.com>
+ * \author Keith Whitwell <keith@tungstengraphics.com>
+ * \author Gareth Hughes <gareth@valinux.com>
  */
 
 #include "drmP.h"
 #include "drm.h"
+#include "drm_sarea.h"
 #include "mga_drm.h"
 #include "mga_drv.h"
 
@@ -148,7 +151,7 @@ void mga_do_dma_flush( drm_mga_private_t *dev_priv )
 	DRM_DEBUG( "  space = 0x%06x\n", primary->space );
 
 	mga_flush_write_combine();
-	MGA_WRITE( MGA_PRIMEND, tail | MGA_PAGPXFER );
+	MGA_WRITE(MGA_PRIMEND, tail | dev_priv->dma_access);
 
 	DRM_DEBUG( "done.\n" );
 }
@@ -190,7 +193,7 @@ void mga_do_dma_wrap_start( drm_mga_private_t *dev_priv )
 	DRM_DEBUG( "  space = 0x%06x\n", primary->space );
 
 	mga_flush_write_combine();
-	MGA_WRITE( MGA_PRIMEND, tail | MGA_PAGPXFER );
+	MGA_WRITE(MGA_PRIMEND, tail | dev_priv->dma_access);
 
 	set_bit( 0, &primary->wrapped );
 	DRM_DEBUG( "done.\n" );
@@ -396,23 +399,383 @@ int mga_freelist_put( drm_device_t *dev, drm_buf_t *buf )
  * DMA initialization, cleanup
  */
 
+
+int mga_driver_preinit(drm_device_t *dev, unsigned long flags)
+{
+	drm_mga_private_t * dev_priv;
+
+	dev_priv = drm_alloc(sizeof(drm_mga_private_t), DRM_MEM_DRIVER);
+	if (!dev_priv)
+		return DRM_ERR(ENOMEM);
+
+	dev->dev_private = (void *)dev_priv;
+	memset(dev_priv, 0, sizeof(drm_mga_private_t));
+
+	dev_priv->usec_timeout = MGA_DEFAULT_USEC_TIMEOUT;
+	dev_priv->chipset = flags;
+
+	return 0;
+}
+
+/**
+ * Bootstrap the driver for AGP DMA.
+ * 
+ * \todo
+ * Investigate whether there is any benifit to storing the WARP microcode in
+ * AGP memory.  If not, the microcode may as well always be put in PCI
+ * memory.
+ *
+ * \todo
+ * This routine needs to set dma_bs->agp_mode to the mode actually configured
+ * in the hardware.  Looking just at the Linux AGP driver code, I don't see
+ * an easy way to determine this.
+ *
+ * \sa mga_do_dma_bootstrap, mga_do_pci_dma_bootstrap
+ */
+static int mga_do_agp_dma_bootstrap(drm_device_t * dev,
+				    drm_mga_dma_bootstrap_t * dma_bs)
+{
+	drm_mga_private_t * const dev_priv = (drm_mga_private_t *) dev->dev_private;
+	const unsigned int warp_size = mga_warp_microcode_size(dev_priv);
+	int err;
+	unsigned  offset;
+	const unsigned secondary_size = dma_bs->secondary_bin_count
+		* dma_bs->secondary_bin_size;
+	const unsigned agp_size = (dma_bs->agp_size << 20);
+	drm_buf_desc_t req;
+	drm_agp_mode_t mode;
+	drm_agp_info_t info;
+
+	
+	/* Acquire AGP. */
+	err = drm_agp_acquire(dev);
+	if (err) {
+		DRM_ERROR("Unable to acquire AGP\n");
+		return err;
+	}
+
+	err = drm_agp_info(dev, &info);
+	if (err) {
+		DRM_ERROR("Unable to get AGP info\n");
+		return err;
+	}
+
+	mode.mode = (info.mode & ~0x07) | dma_bs->agp_mode;
+	err = drm_agp_enable(dev, mode);
+	if (err) {
+		DRM_ERROR("Unable to enable AGP (mode = 0x%lx)\n", mode.mode);
+		return err;
+	}
+
+
+	/* In addition to the usual AGP mode configuration, the G200 AGP cards
+	 * need to have the AGP mode "manually" set.
+	 */
+
+	if (dev_priv->chipset == MGA_CARD_TYPE_G200) {
+		if (mode.mode & 0x02) {
+			MGA_WRITE(MGA_AGP_PLL, MGA_AGP2XPLL_ENABLE);
+		}
+		else {
+			MGA_WRITE(MGA_AGP_PLL, MGA_AGP2XPLL_DISABLE);
+		}
+	}
+
+
+	/* Allocate and bind AGP memory. */
+	dev_priv->agp_pages = agp_size / PAGE_SIZE;
+	dev_priv->agp_mem = drm_alloc_agp( dev, dev_priv->agp_pages, 0 );
+	if (dev_priv->agp_mem == NULL) {
+		dev_priv->agp_pages = 0;
+		DRM_ERROR("Unable to allocate %uMB AGP memory\n",
+			  dma_bs->agp_size);
+		return DRM_ERR(ENOMEM);
+	}
+		
+	err = drm_bind_agp( dev_priv->agp_mem, 0 );
+	if (err) {
+		DRM_ERROR("Unable to bind AGP memory\n");
+		return err;
+	}
+
+	offset = 0;
+	err = drm_addmap( dev, offset, warp_size,
+			  _DRM_AGP, _DRM_READ_ONLY, & dev_priv->warp );
+	if (err) {
+		DRM_ERROR("Unable to map WARP microcode\n");
+		return err;
+	}
+
+	offset += warp_size;
+	err = drm_addmap( dev, offset, dma_bs->primary_size,
+			  _DRM_AGP, _DRM_READ_ONLY, & dev_priv->primary );
+	if (err) {
+		DRM_ERROR("Unable to map primary DMA region\n");
+		return err;
+	}
+
+	offset += dma_bs->primary_size;
+	err = drm_addmap( dev, offset, secondary_size,
+			  _DRM_AGP, 0, & dev->agp_buffer_map );
+	if (err) {
+		DRM_ERROR("Unable to map secondary DMA region\n");
+		return err;
+	}
+
+	(void) memset( &req, 0, sizeof(req) );
+	req.count = dma_bs->secondary_bin_count;
+	req.size = dma_bs->secondary_bin_size;
+	req.flags = _DRM_AGP_BUFFER;
+	req.agp_start = offset;
+
+	err = drm_addbufs_agp( dev, & req );
+	if (err) {
+		DRM_ERROR("Unable to add secondary DMA buffers\n");
+		return err;
+	}
+
+	offset += secondary_size;
+	err = drm_addmap( dev, offset, agp_size - offset,
+			  _DRM_AGP, 0, & dev_priv->agp_textures );
+	if (err) {
+		DRM_ERROR("Unable to map AGP texture region\n");
+		return err;
+	}
+
+	drm_core_ioremap(dev_priv->warp, dev);
+	drm_core_ioremap(dev_priv->primary, dev);
+	drm_core_ioremap(dev->agp_buffer_map, dev);
+
+	if (!dev_priv->warp->handle ||
+	    !dev_priv->primary->handle || !dev->agp_buffer_map->handle) {
+		DRM_ERROR("failed to ioremap agp regions! (%p, %p, %p)\n",
+			  dev_priv->warp->handle, dev_priv->primary->handle,
+			  dev->agp_buffer_map->handle);
+		return DRM_ERR(ENOMEM);
+	}
+
+	dev_priv->dma_access = MGA_PAGPXFER;
+	dev_priv->wagp_enable = MGA_WAGP_ENABLE;
+
+	DRM_INFO("Initialized card for AGP DMA.\n");
+	return 0;
+}
+
+/**
+ * Bootstrap the driver for PCI DMA.
+ * 
+ * \todo
+ * The algorithm for decreasing the size of the primary DMA buffer could be
+ * better.  The size should be rounded up to the nearest page size, then
+ * decrease the request size by a single page each pass through the loop.
+ *
+ * \todo
+ * Determine whether the maximum address passed to drm_pci_alloc is correct.
+ * The same goes for drm_addbufs_pci.
+ * 
+ * \sa mga_do_dma_bootstrap, mga_do_agp_dma_bootstrap
+ */
+static int mga_do_pci_dma_bootstrap(drm_device_t * dev,
+				    drm_mga_dma_bootstrap_t * dma_bs)
+{
+	drm_mga_private_t * const dev_priv = (drm_mga_private_t *) dev->dev_private;
+	const unsigned int warp_size = mga_warp_microcode_size(dev_priv);
+	unsigned int primary_size;
+	unsigned int bin_count;
+	int err;
+	drm_buf_desc_t req;
+
+	
+	if (dev->dma == NULL) {
+		DRM_ERROR("dev->dma is NULL\n");
+		return DRM_ERR(EFAULT);
+	}
+
+	/* The proper alignment is 0x100 for this mapping */
+	err = drm_addmap(dev, 0, warp_size, _DRM_CONSISTENT,
+			 _DRM_READ_ONLY, &dev_priv->warp);
+	if (err != 0) {
+		DRM_ERROR("Unable to create mapping for WARP microcode\n");
+		return err;
+	}
+
+	/* Other than the bottom two bits being used to encode other
+	 * information, there don't appear to be any restrictions on the
+	 * alignment of the primary or secondary DMA buffers.
+	 */
+
+	for ( primary_size = dma_bs->primary_size
+	      ; primary_size != 0
+	      ; primary_size >>= 1 ) {
+		/* The proper alignment for this mapping is 0x04 */
+		err = drm_addmap(dev, 0, primary_size, _DRM_CONSISTENT,
+				 _DRM_READ_ONLY, &dev_priv->primary);
+		if (!err)
+			break;
+	}
+
+	if (err != 0) {
+		DRM_ERROR("Unable to allocate primary DMA region\n");
+		return DRM_ERR(ENOMEM);
+	}
+
+	if (dev_priv->primary->size != dma_bs->primary_size) {
+		DRM_INFO("Primary DMA buffer size reduced from %u to %u.\n",
+			 dma_bs->primary_size, 
+			 (unsigned) dev_priv->primary->size);
+		dma_bs->primary_size = dev_priv->primary->size;
+	}
+
+	for ( bin_count = dma_bs->secondary_bin_count
+	      ; bin_count > 0 
+	      ; bin_count-- ) {
+		(void) memset( &req, 0, sizeof(req) );
+		req.count = bin_count;
+		req.size = dma_bs->secondary_bin_size;
+
+		err = drm_addbufs_pci( dev, & req );
+		if (!err) {
+			break;
+		}
+	}
+	
+	if (bin_count == 0) {
+		DRM_ERROR("Unable to add secondary DMA buffers\n");
+		return err;
+	}
+
+	if (bin_count != dma_bs->secondary_bin_count) {
+		DRM_INFO("Secondary PCI DMA buffer bin count reduced from %u "
+			 "to %u.\n", dma_bs->secondary_bin_count, bin_count);
+
+		dma_bs->secondary_bin_count = bin_count;
+	}
+
+	dev_priv->dma_access = 0;
+	dev_priv->wagp_enable = 0;
+
+	dma_bs->agp_mode = 0;
+
+	DRM_INFO("Initialized card for PCI DMA.\n");
+	return 0;
+}
+
+
+static int mga_do_dma_bootstrap(drm_device_t * dev,
+				drm_mga_dma_bootstrap_t * dma_bs)
+{
+	const int is_agp = (dma_bs->agp_mode != 0) && drm_device_is_agp(dev);
+	int err;
+	drm_mga_private_t * const dev_priv =
+		(drm_mga_private_t *) dev->dev_private;
+
+
+	dev_priv->used_new_dma_init = 1;
+
+	/* The first steps are the same for both PCI and AGP based DMA.  Map
+	 * the cards MMIO registers and map a status page.
+	 */
+	err = drm_addmap( dev, dev_priv->mmio_base, dev_priv->mmio_size,
+			  _DRM_REGISTERS, _DRM_READ_ONLY, & dev_priv->mmio );
+	if (err) {
+		DRM_ERROR("Unable to map MMIO region\n");
+		return err;
+	}
+
+
+	err = drm_addmap( dev, 0, SAREA_MAX, _DRM_SHM,
+			  _DRM_READ_ONLY | _DRM_LOCKED | _DRM_KERNEL,
+			  & dev_priv->status );
+	if (err) {
+		DRM_ERROR("Unable to map status region\n");
+		return err;
+	}
+
+
+	/* The DMA initialization procedure is slightly different for PCI and
+	 * AGP cards.  AGP cards just allocate a large block of AGP memory and
+	 * carve off portions of it for internal uses.  The remaining memory
+	 * is returned to user-mode to be used for AGP textures.
+	 */
+
+	if (is_agp) {
+		err = mga_do_agp_dma_bootstrap(dev, dma_bs);
+	}
+	
+	/* If we attempted to initialize the card for AGP DMA but failed,
+	 * clean-up any mess that may have been created.
+	 */
+
+	if (err) {
+		mga_do_cleanup_dma(dev);
+	}
+
+
+	/* Not only do we want to try and initialized PCI cards for PCI DMA,
+	 * but we also try to initialized AGP cards that could not be
+	 * initialized for AGP DMA.  This covers the case where we have an AGP
+	 * card in a system with an unsupported AGP chipset.  In that case the
+	 * card will be detected as AGP, but we won't be able to allocate any
+	 * AGP memory, etc.
+	 */
+
+	if (!is_agp || err) {
+		err = mga_do_pci_dma_bootstrap(dev, dma_bs);
+	}
+
+
+	return err;
+}
+
+int mga_dma_bootstrap(DRM_IOCTL_ARGS)
+{
+	DRM_DEVICE;
+	drm_mga_dma_bootstrap_t bootstrap;
+	int err;
+
+
+	DRM_COPY_FROM_USER_IOCTL(bootstrap,
+				 (drm_mga_dma_bootstrap_t __user *) data,
+				 sizeof(bootstrap));
+
+	err = mga_do_dma_bootstrap(dev, & bootstrap);
+	if (! err) {
+		static const int modes[] = { 0, 1, 2, 2, 4, 4, 4, 4 };
+		const drm_mga_private_t * const dev_priv = 
+			(drm_mga_private_t *) dev->dev_private;
+
+		if (dev_priv->agp_textures != NULL) {
+			bootstrap.texture_handle = dev_priv->agp_textures->offset;
+			bootstrap.texture_size = dev_priv->agp_textures->size;
+		}
+		else {
+			bootstrap.texture_handle = 0;
+			bootstrap.texture_size = 0;
+		}
+
+		bootstrap.agp_mode = modes[ bootstrap.agp_mode & 0x07 ];
+		if (DRM_COPY_TO_USER( (void __user *) data, & bootstrap,
+				     sizeof(bootstrap))) {
+			err = DRM_ERR(EFAULT);
+		}
+	}
+	else {
+		mga_do_cleanup_dma(dev);
+	}
+
+	return err;
+}
+
 static int mga_do_init_dma( drm_device_t *dev, drm_mga_init_t *init )
 {
 	drm_mga_private_t *dev_priv;
 	int ret;
 	DRM_DEBUG( "\n" );
 
-	dev_priv = drm_alloc( sizeof(drm_mga_private_t), DRM_MEM_DRIVER );
-	if ( !dev_priv )
-		return DRM_ERR(ENOMEM);
-
-	memset( dev_priv, 0, sizeof(drm_mga_private_t) );
 
-	dev_priv->chipset = init->chipset;
+	dev_priv = dev->dev_private;
 
-	dev_priv->usec_timeout = MGA_DEFAULT_USEC_TIMEOUT;
-
-	if ( init->sgram ) {
+	if (init->sgram) {
 		dev_priv->clear_cmd = MGA_DWGCTL_CLEAR | MGA_ATYPE_BLK;
 	} else {
 		dev_priv->clear_cmd = MGA_DWGCTL_CLEAR | MGA_ATYPE_RSTR;
@@ -436,88 +799,66 @@ static int mga_do_init_dma( drm_device_t *dev, drm_mga_init_t *init )
 
 	DRM_GETSAREA();
 
-	if(!dev_priv->sarea) {
-		DRM_ERROR( "failed to find sarea!\n" );
-		/* Assign dev_private so we can do cleanup. */
-		dev->dev_private = (void *)dev_priv;
-		mga_do_cleanup_dma( dev );
+	if (!dev_priv->sarea) {
+		DRM_ERROR("failed to find sarea!\n");
 		return DRM_ERR(EINVAL);
 	}
 
-	dev_priv->mmio = drm_core_findmap(dev, init->mmio_offset);
-	if(!dev_priv->mmio) {
-		DRM_ERROR( "failed to find mmio region!\n" );
-		/* Assign dev_private so we can do cleanup. */
-		dev->dev_private = (void *)dev_priv;
-		mga_do_cleanup_dma( dev );
-		return DRM_ERR(EINVAL);
-	}
-	dev_priv->status = drm_core_findmap(dev, init->status_offset);
-	if(!dev_priv->status) {
-		DRM_ERROR( "failed to find status page!\n" );
-		/* Assign dev_private so we can do cleanup. */
-		dev->dev_private = (void *)dev_priv;
-		mga_do_cleanup_dma( dev );
-		return DRM_ERR(EINVAL);
-	}
-	dev_priv->warp = drm_core_findmap(dev, init->warp_offset);
-	if(!dev_priv->warp) {
-		DRM_ERROR( "failed to find warp microcode region!\n" );
-		/* Assign dev_private so we can do cleanup. */
-		dev->dev_private = (void *)dev_priv;
-		mga_do_cleanup_dma( dev );
-		return DRM_ERR(EINVAL);
-	}
-	dev_priv->primary = drm_core_findmap(dev, init->primary_offset);
-	if(!dev_priv->primary) {
-		DRM_ERROR( "failed to find primary dma region!\n" );
-		/* Assign dev_private so we can do cleanup. */
-		dev->dev_private = (void *)dev_priv;
-		mga_do_cleanup_dma( dev );
-		return DRM_ERR(EINVAL);
-	}
-	dev->agp_buffer_map = drm_core_findmap(dev, init->buffers_offset);
-	if(!dev->agp_buffer_map) {
-		DRM_ERROR( "failed to find dma buffer region!\n" );
-		/* Assign dev_private so we can do cleanup. */
-		dev->dev_private = (void *)dev_priv;
-		mga_do_cleanup_dma( dev );
-		return DRM_ERR(EINVAL);
+	if (! dev_priv->used_new_dma_init) {
+		dev_priv->status = drm_core_findmap(dev, init->status_offset);
+		if (!dev_priv->status) {
+			DRM_ERROR("failed to find status page!\n");
+			return DRM_ERR(EINVAL);
+		}
+		dev_priv->mmio = drm_core_findmap(dev, init->mmio_offset);
+		if (!dev_priv->mmio) {
+			DRM_ERROR("failed to find mmio region!\n");
+			return DRM_ERR(EINVAL);
+		}
+		dev_priv->warp = drm_core_findmap(dev, init->warp_offset);
+		if (!dev_priv->warp) {
+			DRM_ERROR("failed to find warp microcode region!\n");
+			return DRM_ERR(EINVAL);
+		}
+		dev_priv->primary = drm_core_findmap(dev, init->primary_offset);
+		if (!dev_priv->primary) {
+			DRM_ERROR("failed to find primary dma region!\n");
+			return DRM_ERR(EINVAL);
+		}
+		dev->agp_buffer_token = init->buffers_offset;
+		dev->agp_buffer_map = drm_core_findmap(dev, init->buffers_offset);
+		if (!dev->agp_buffer_map) {
+			DRM_ERROR("failed to find dma buffer region!\n");
+			return DRM_ERR(EINVAL);
+		}
+
+		drm_core_ioremap(dev_priv->warp, dev);
+		drm_core_ioremap(dev_priv->primary, dev);
+		drm_core_ioremap(dev->agp_buffer_map, dev);
 	}
 
 	dev_priv->sarea_priv =
 		(drm_mga_sarea_t *)((u8 *)dev_priv->sarea->handle +
 				    init->sarea_priv_offset);
 
-	drm_core_ioremap( dev_priv->warp, dev );
-	drm_core_ioremap( dev_priv->primary, dev );
-	drm_core_ioremap( dev->agp_buffer_map, dev );
-
-	if(!dev_priv->warp->handle ||
-	   !dev_priv->primary->handle ||
-	   !dev->agp_buffer_map->handle ) {
-		DRM_ERROR( "failed to ioremap agp regions!\n" );
-		/* Assign dev_private so we can do cleanup. */
-		dev->dev_private = (void *)dev_priv;
-		mga_do_cleanup_dma( dev );
+	if (!dev_priv->warp->handle ||
+	    !dev_priv->primary->handle ||
+	    ((dev_priv->dma_access != 0) &&
+	     ((dev->agp_buffer_map == NULL) ||
+	      (dev->agp_buffer_map->handle == NULL)))) {
+		DRM_ERROR("failed to ioremap agp regions!\n");
 		return DRM_ERR(ENOMEM);
 	}
 
-	ret = mga_warp_install_microcode( dev_priv );
-	if ( ret < 0 ) {
-		DRM_ERROR( "failed to install WARP ucode!\n" );
-		/* Assign dev_private so we can do cleanup. */
-		dev->dev_private = (void *)dev_priv;
-		mga_do_cleanup_dma( dev );
+	ret = mga_warp_install_microcode(dev_priv);
+	if (ret < 0) {
+		DRM_ERROR("failed to install WARP ucode!\n");
 		return ret;
 	}
 
-	ret = mga_warp_init( dev_priv );
-	if ( ret < 0 ) {
-		DRM_ERROR( "failed to init WARP engine!\n" );
-		/* Assign dev_private so we can do cleanup. */
-		dev->dev_private = (void *)dev_priv;
-		mga_do_cleanup_dma( dev );
+	ret = mga_warp_init(dev_priv);
+	if (ret < 0) {
+		DRM_ERROR("failed to init WARP engine!\n");
 		return ret;
 	}
 
@@ -557,22 +898,18 @@ static int mga_do_init_dma( drm_device_t *dev, drm_mga_init_t *init )
 	dev_priv->sarea_priv->last_frame.head = 0;
 	dev_priv->sarea_priv->last_frame.wrap = 0;
 
-	if ( mga_freelist_init( dev, dev_priv ) < 0 ) {
-		DRM_ERROR( "could not initialize freelist\n" );
-		/* Assign dev_private so we can do cleanup. */
-		dev->dev_private = (void *)dev_priv;
-		mga_do_cleanup_dma( dev );
+	if (mga_freelist_init(dev, dev_priv) < 0) {
+		DRM_ERROR("could not initialize freelist\n");
 		return DRM_ERR(ENOMEM);
 	}
 
-	/* Make dev_private visable to others. */
-	dev->dev_private = (void *)dev_priv;
 	return 0;
 }
 
 static int mga_do_cleanup_dma( drm_device_t *dev )
 {
-	DRM_DEBUG( "\n" );
+	int err = 0;
+	DRM_DEBUG("\n");
 
 	/* Make sure interrupts are disabled here because the uninstall ioctl
 	 * may not have been called from userspace and after dev_private
@@ -583,20 +920,49 @@ static int mga_do_cleanup_dma( drm_device_t *dev )
 	if ( dev->dev_private ) {
 		drm_mga_private_t *dev_priv = dev->dev_private;
 
-		if ( dev_priv->warp != NULL )
-			drm_core_ioremapfree( dev_priv->warp, dev );
-		if ( dev_priv->primary != NULL )
-			drm_core_ioremapfree( dev_priv->primary, dev );
-		if ( dev->agp_buffer_map != NULL )
-			drm_core_ioremapfree( dev->agp_buffer_map, dev );
+		if ((dev_priv->warp != NULL) 
+		    && (dev_priv->mmio->type != _DRM_CONSISTENT))
+			drm_core_ioremapfree(dev_priv->warp, dev);
+
+		if ((dev_priv->primary != NULL) 
+		    && (dev_priv->primary->type != _DRM_CONSISTENT))
+			drm_core_ioremapfree(dev_priv->primary, dev);
 
-		if ( dev_priv->head != NULL ) {
-			mga_freelist_cleanup( dev );
+		if (dev->agp_buffer_map != NULL)
+			drm_core_ioremapfree(dev->agp_buffer_map, dev);
+
+		if (dev_priv->used_new_dma_init) {
+			if (dev_priv->agp_mem != NULL) {
+				dev_priv->agp_textures = NULL;
+				drm_unbind_agp(dev_priv->agp_mem);
+
+				drm_free_agp(dev_priv->agp_mem, dev_priv->agp_pages);
+				dev_priv->agp_pages = 0;
+				dev_priv->agp_mem = NULL;
+			}
+
+			if ((dev->agp != NULL) && dev->agp->acquired) {
+				err = drm_agp_release(dev);
+			}
+
+			dev_priv->used_new_dma_init = 0;
 		}
 
-		drm_free( dev->dev_private, sizeof(drm_mga_private_t),
-			   DRM_MEM_DRIVER );
-		dev->dev_private = NULL;
+		dev_priv->warp = NULL;
+		dev_priv->primary = NULL;
+		dev_priv->mmio = NULL;
+		dev_priv->status = NULL;
+		dev_priv->sarea = NULL;
+		dev_priv->sarea_priv = NULL;
+		dev->agp_buffer_map = NULL;
+
+		memset(&dev_priv->prim, 0, sizeof(dev_priv->prim));
+		dev_priv->warp_pipe = 0;
+		memset(dev_priv->warp_pipe_phys, 0, sizeof(dev_priv->warp_pipe_phys));
+
+		if (dev_priv->head != NULL) {
+			mga_freelist_cleanup(dev);
+		}
 	}
 
 	return 0;
@@ -606,14 +972,20 @@ int mga_dma_init( DRM_IOCTL_ARGS )
 {
 	DRM_DEVICE;
 	drm_mga_init_t init;
+	int err;
 
 	LOCK_TEST_WITH_RETURN( dev, filp );
 
-	DRM_COPY_FROM_USER_IOCTL( init, (drm_mga_init_t __user *)data, sizeof(init) );
+	DRM_COPY_FROM_USER_IOCTL(init, (drm_mga_init_t __user *) data,
+				 sizeof(init));
 
 	switch ( init.func ) {
 	case MGA_INIT_DMA:
-		return mga_do_init_dma( dev, &init );
+		err = mga_do_init_dma(dev, &init);
+		if (err) {
+			(void) mga_do_cleanup_dma(dev);
+		}
+		return err;
 	case MGA_CLEANUP_DMA:
 		return mga_do_cleanup_dma( dev );
 	}
@@ -742,7 +1114,21 @@ int mga_dma_buffers( DRM_IOCTL_ARGS )
 	return ret;
 }
 
-void mga_driver_pretakedown(drm_device_t *dev)
+/**
+ * Called just before the module is unloaded.
+ */
+int mga_driver_postcleanup(drm_device_t * dev)
+{
+	drm_free(dev->dev_private, sizeof(drm_mga_private_t), DRM_MEM_DRIVER);
+	dev->dev_private = NULL;
+
+	return 0;
+}
+
+/**
+ * Called when the last opener of the device is closed.
+ */
+void mga_driver_pretakedown(drm_device_t * dev)
 {
 	mga_do_cleanup_dma( dev );
 }
diff --git a/drivers/char/drm/mga_drm.h b/drivers/char/drm/mga_drm.h
index 521d4451d012..d20aab3bd57b 100644
--- a/drivers/char/drm/mga_drm.h
+++ b/drivers/char/drm/mga_drm.h
@@ -73,7 +73,8 @@
 
 #define MGA_CARD_TYPE_G200	1
 #define MGA_CARD_TYPE_G400	2
-
+#define MGA_CARD_TYPE_G450	3       /* not currently used */
+#define MGA_CARD_TYPE_G550	4
 
 #define MGA_FRONT		0x1
 #define MGA_BACK		0x2
@@ -225,10 +226,6 @@ typedef struct _drm_mga_sarea {
 } drm_mga_sarea_t;
 
 
-/* WARNING: If you change any of these defines, make sure to change the
- * defines in the Xserver file (xf86drmMga.h)
- */
-
 /* MGA specific ioctls
  * The device specific ioctl range is 0x40 to 0x79.
  */
@@ -243,6 +240,14 @@ typedef struct _drm_mga_sarea {
 #define DRM_MGA_BLIT     0x08
 #define DRM_MGA_GETPARAM 0x09
 
+/* 3.2:
+ * ioctls for operating on fences.
+ */
+#define DRM_MGA_SET_FENCE      0x0a
+#define DRM_MGA_WAIT_FENCE     0x0b
+#define DRM_MGA_DMA_BOOTSTRAP  0x0c
+
+
 #define DRM_IOCTL_MGA_INIT     DRM_IOW( DRM_COMMAND_BASE + DRM_MGA_INIT, drm_mga_init_t)
 #define DRM_IOCTL_MGA_FLUSH    DRM_IOW( DRM_COMMAND_BASE + DRM_MGA_FLUSH, drm_lock_t)
 #define DRM_IOCTL_MGA_RESET    DRM_IO(  DRM_COMMAND_BASE + DRM_MGA_RESET)
@@ -253,6 +258,9 @@ typedef struct _drm_mga_sarea {
 #define DRM_IOCTL_MGA_ILOAD    DRM_IOW( DRM_COMMAND_BASE + DRM_MGA_ILOAD, drm_mga_iload_t)
 #define DRM_IOCTL_MGA_BLIT     DRM_IOW( DRM_COMMAND_BASE + DRM_MGA_BLIT, drm_mga_blit_t)
 #define DRM_IOCTL_MGA_GETPARAM DRM_IOWR(DRM_COMMAND_BASE + DRM_MGA_GETPARAM, drm_mga_getparam_t)
+#define DRM_IOCTL_MGA_SET_FENCE     DRM_IOW( DRM_COMMAND_BASE + DRM_MGA_SET_FENCE, uint32_t)
+#define DRM_IOCTL_MGA_WAIT_FENCE    DRM_IOWR(DRM_COMMAND_BASE + DRM_MGA_WAIT_FENCE, uint32_t)
+#define DRM_IOCTL_MGA_DMA_BOOTSTRAP DRM_IOWR(DRM_COMMAND_BASE + DRM_MGA_DMA_BOOTSTRAP, drm_mga_dma_bootstrap_t)
 
 typedef struct _drm_mga_warp_index {
    	int installed;
@@ -291,12 +299,72 @@ typedef struct drm_mga_init {
 	unsigned long buffers_offset;
 } drm_mga_init_t;
 
-typedef struct drm_mga_fullscreen {
-	enum {
-		MGA_INIT_FULLSCREEN    = 0x01,
-		MGA_CLEANUP_FULLSCREEN = 0x02
-	} func;
-} drm_mga_fullscreen_t;
+typedef struct drm_mga_dma_bootstrap {
+	/**
+	 * \name AGP texture region
+	 * 
+	 * On return from the DRM_MGA_DMA_BOOTSTRAP ioctl, these fields will
+	 * be filled in with the actual AGP texture settings.
+	 * 
+	 * \warning
+	 * If these fields are non-zero, but dma_mga_dma_bootstrap::agp_mode
+	 * is zero, it means that PCI memory (most likely through the use of
+	 * an IOMMU) is being used for "AGP" textures.
+	 */
+	/*@{*/
+	unsigned long texture_handle; /**< Handle used to map AGP textures. */
+	uint32_t     texture_size;    /**< Size of the AGP texture region. */
+	/*@}*/
+
+
+	/**
+	 * Requested size of the primary DMA region.
+	 * 
+	 * On return from the DRM_MGA_DMA_BOOTSTRAP ioctl, this field will be
+	 * filled in with the actual AGP mode.  If AGP was not available
+	 */
+	uint32_t primary_size;
+
+
+	/**
+	 * Requested number of secondary DMA buffers.
+	 * 
+	 * On return from the DRM_MGA_DMA_BOOTSTRAP ioctl, this field will be
+	 * filled in with the actual number of secondary DMA buffers
+	 * allocated.  Particularly when PCI DMA is used, this may be
+	 * (subtantially) less than the number requested.
+	 */
+	uint32_t secondary_bin_count;
+	
+	
+	/**
+	 * Requested size of each secondary DMA buffer.
+	 * 
+	 * While the kernel \b is free to reduce
+	 * dma_mga_dma_bootstrap::secondary_bin_count, it is \b not allowed
+	 * to reduce dma_mga_dma_bootstrap::secondary_bin_size.
+	 */
+	uint32_t secondary_bin_size;
+
+
+	/**
+	 * Bit-wise mask of AGPSTAT2_* values.  Currently only \c AGPSTAT2_1X,
+	 * \c AGPSTAT2_2X, and \c AGPSTAT2_4X are supported.  If this value is
+	 * zero, it means that PCI DMA should be used, even if AGP is
+	 * possible.
+	 * 
+	 * On return from the DRM_MGA_DMA_BOOTSTRAP ioctl, this field will be
+	 * filled in with the actual AGP mode.  If AGP was not available
+	 * (i.e., PCI DMA was used), this value will be zero.
+	 */
+	uint32_t agp_mode;
+
+
+	/**
+	 * Desired AGP GART size, measured in megabytes.
+	 */
+	uint8_t agp_size;
+} drm_mga_dma_bootstrap_t;
 
 typedef struct drm_mga_clear {
 	unsigned int flags;
@@ -341,6 +409,14 @@ typedef struct _drm_mga_blit {
  */
 #define MGA_PARAM_IRQ_NR            1
 
+/* 3.2: Query the actual card type.  The DDX only distinguishes between
+ * G200 chips and non-G200 chips, which it calls G400.  It turns out that
+ * there are some very sublte differences between the G4x0 chips and the G550
+ * chips.  Using this parameter query, a client-side driver can detect the
+ * difference between a G4x0 and a G550.
+ */
+#define MGA_PARAM_CARD_TYPE         2
+
 typedef struct drm_mga_getparam {
 	int param;
 	void __user *value;
diff --git a/drivers/char/drm/mga_drv.c b/drivers/char/drm/mga_drv.c
index 844cca9cb29d..daabbba3b297 100644
--- a/drivers/char/drm/mga_drv.c
+++ b/drivers/char/drm/mga_drv.c
@@ -38,8 +38,15 @@
   
 #include "drm_pciids.h"
 
+static int mga_driver_device_is_agp(drm_device_t * dev);
 static int postinit( struct drm_device *dev, unsigned long flags )
 {
+	drm_mga_private_t * const dev_priv =
+		(drm_mga_private_t *) dev->dev_private;
+
+	dev_priv->mmio_base = pci_resource_start(dev->pdev, 1);
+	dev_priv->mmio_size = pci_resource_len(dev->pdev, 1);
+
 	dev->counters += 3;
 	dev->types[6] = _DRM_STAT_IRQ;
 	dev->types[7] = _DRM_STAT_PRIMARY;
@@ -79,8 +86,11 @@ extern int mga_max_ioctl;
 
 static struct drm_driver driver = {
 	.driver_features = DRIVER_USE_AGP | DRIVER_REQUIRE_AGP | DRIVER_USE_MTRR | DRIVER_HAVE_DMA | DRIVER_HAVE_IRQ | DRIVER_IRQ_SHARED | DRIVER_IRQ_VBL,
+	.preinit = mga_driver_preinit,
+	.postcleanup = mga_driver_postcleanup,
 	.pretakedown = mga_driver_pretakedown,
 	.dma_quiescent = mga_driver_dma_quiescent,
+	.device_is_agp = mga_driver_device_is_agp,
 	.vblank_wait = mga_driver_vblank_wait,
 	.irq_preinstall = mga_driver_irq_preinstall,
 	.irq_postinstall = mga_driver_irq_postinstall,
@@ -128,3 +138,38 @@ module_exit(mga_exit);
 MODULE_AUTHOR( DRIVER_AUTHOR );
 MODULE_DESCRIPTION( DRIVER_DESC );
 MODULE_LICENSE("GPL and additional rights");
+
+/**
+ * Determine if the device really is AGP or not.
+ *
+ * In addition to the usual tests performed by \c drm_device_is_agp, this
+ * function detects PCI G450 cards that appear to the system exactly like
+ * AGP G450 cards.
+ *
+ * \param dev   The device to be tested.
+ *
+ * \returns
+ * If the device is a PCI G450, zero is returned.  Otherwise 2 is returned.
+ */
+int mga_driver_device_is_agp(drm_device_t * dev)
+{
+	const struct pci_dev * const pdev = dev->pdev;
+
+
+	/* There are PCI versions of the G450.  These cards have the
+	 * same PCI ID as the AGP G450, but have an additional PCI-to-PCI
+	 * bridge chip.  We detect these cards, which are not currently
+	 * supported by this driver, by looking at the device ID of the
+	 * bus the "card" is on.  If vendor is 0x3388 (Hint Corp) and the
+	 * device is 0x0021 (HB6 Universal PCI-PCI bridge), we reject the
+	 * device.
+	 */
+	
+	if ( (pdev->device == 0x0525)
+	     && (pdev->bus->self->vendor == 0x3388)
+	     && (pdev->bus->self->device == 0x0021) ) {
+		return 0;
+	}
+
+	return 2;
+}
diff --git a/drivers/char/drm/mga_drv.h b/drivers/char/drm/mga_drv.h
index 9412e2816eb7..b22fdbd4f830 100644
--- a/drivers/char/drm/mga_drv.h
+++ b/drivers/char/drm/mga_drv.h
@@ -38,10 +38,10 @@
 
 #define DRIVER_NAME		"mga"
 #define DRIVER_DESC		"Matrox G200/G400"
-#define DRIVER_DATE		"20021029"
+#define DRIVER_DATE		"20050607"
 
 #define DRIVER_MAJOR		3
-#define DRIVER_MINOR		1
+#define DRIVER_MINOR		2
 #define DRIVER_PATCHLEVEL	0
 
 typedef struct drm_mga_primary_buffer {
@@ -87,9 +87,43 @@ typedef struct drm_mga_private {
 	int chipset;
 	int usec_timeout;
 
+	/**
+	 * If set, the new DMA initialization sequence was used.  This is
+	 * primarilly used to select how the driver should uninitialized its
+	 * internal DMA structures.
+	 */
+	int used_new_dma_init;
+
+	/**
+	 * If AGP memory is used for DMA buffers, this will be the value
+	 * \c MGA_PAGPXFER.  Otherwise, it will be zero (for a PCI transfer).
+	 */
+	u32 dma_access;
+
+	/**
+	 * If AGP memory is used for DMA buffers, this will be the value
+	 * \c MGA_WAGP_ENABLE.  Otherwise, it will be zero (for a PCI
+	 * transfer).
+	 */
+	u32 wagp_enable;
+
+	/**
+	 * \name MMIO region parameters.
+	 * 
+	 * \sa drm_mga_private_t::mmio
+	 */
+	/*@{*/
+	u32 mmio_base;             /**< Bus address of base of MMIO. */
+	u32 mmio_size;             /**< Size of the MMIO region. */
+	/*@}*/
+
 	u32 clear_cmd;
 	u32 maccess;
 
+	wait_queue_head_t fence_queue;
+	atomic_t last_fence_retired;
+	u32 next_fence_to_post;
+
 	unsigned int fb_cpp;
 	unsigned int front_offset;
 	unsigned int front_pitch;
@@ -108,35 +142,43 @@ typedef struct drm_mga_private {
 	drm_local_map_t *status;
 	drm_local_map_t *warp;
 	drm_local_map_t *primary;
-	drm_local_map_t *buffers;
 	drm_local_map_t *agp_textures;
+	
+	DRM_AGP_MEM *agp_mem;
+	unsigned int agp_pages;
 } drm_mga_private_t;
 
 				/* mga_dma.c */
-extern int mga_dma_init( DRM_IOCTL_ARGS );
-extern int mga_dma_flush( DRM_IOCTL_ARGS );
-extern int mga_dma_reset( DRM_IOCTL_ARGS );
-extern int mga_dma_buffers( DRM_IOCTL_ARGS );
-extern void mga_driver_pretakedown(drm_device_t *dev);
-extern int mga_driver_dma_quiescent(drm_device_t *dev);
-
-extern int mga_do_wait_for_idle( drm_mga_private_t *dev_priv );
-
-extern void mga_do_dma_flush( drm_mga_private_t *dev_priv );
-extern void mga_do_dma_wrap_start( drm_mga_private_t *dev_priv );
-extern void mga_do_dma_wrap_end( drm_mga_private_t *dev_priv );
+extern int mga_driver_preinit(drm_device_t * dev, unsigned long flags);
+extern int mga_dma_bootstrap(DRM_IOCTL_ARGS);
+extern int mga_dma_init(DRM_IOCTL_ARGS);
+extern int mga_dma_flush(DRM_IOCTL_ARGS);
+extern int mga_dma_reset(DRM_IOCTL_ARGS);
+extern int mga_dma_buffers(DRM_IOCTL_ARGS);
+extern int mga_driver_postcleanup(drm_device_t * dev);
+extern void mga_driver_pretakedown(drm_device_t * dev);
+extern int mga_driver_dma_quiescent(drm_device_t * dev);
+
+extern int mga_do_wait_for_idle(drm_mga_private_t * dev_priv);
+
+extern void mga_do_dma_flush(drm_mga_private_t * dev_priv);
+extern void mga_do_dma_wrap_start(drm_mga_private_t * dev_priv);
+extern void mga_do_dma_wrap_end(drm_mga_private_t * dev_priv);
 
 extern int mga_freelist_put( drm_device_t *dev, drm_buf_t *buf );
 
 				/* mga_warp.c */
-extern int mga_warp_install_microcode( drm_mga_private_t *dev_priv );
-extern int mga_warp_init( drm_mga_private_t *dev_priv );
-
-extern int mga_driver_vblank_wait(drm_device_t *dev, unsigned int *sequence);
-extern irqreturn_t mga_driver_irq_handler( DRM_IRQ_ARGS );
-extern void mga_driver_irq_preinstall( drm_device_t *dev );
-extern void mga_driver_irq_postinstall( drm_device_t *dev );
-extern void mga_driver_irq_uninstall( drm_device_t *dev );
+extern unsigned int mga_warp_microcode_size(const drm_mga_private_t * dev_priv);
+extern int mga_warp_install_microcode(drm_mga_private_t * dev_priv);
+extern int mga_warp_init(drm_mga_private_t * dev_priv);
+
+				/* mga_irq.c */
+extern int mga_driver_fence_wait(drm_device_t * dev, unsigned int *sequence);
+extern int mga_driver_vblank_wait(drm_device_t * dev, unsigned int *sequence);
+extern irqreturn_t mga_driver_irq_handler(DRM_IRQ_ARGS);
+extern void mga_driver_irq_preinstall(drm_device_t * dev);
+extern void mga_driver_irq_postinstall(drm_device_t * dev);
+extern void mga_driver_irq_uninstall(drm_device_t * dev);
 extern long mga_compat_ioctl(struct file *filp, unsigned int cmd,
 			     unsigned long arg);
 
@@ -527,6 +569,12 @@ do {									\
  */
 #define MGA_EXEC 			0x0100
 
+/* AGP PLL encoding (for G200 only).
+ */
+#define MGA_AGP_PLL 			0x1e4c
+#	define MGA_AGP2XPLL_DISABLE		(0 << 0)
+#	define MGA_AGP2XPLL_ENABLE		(1 << 0)
+
 /* Warp registers
  */
 #define MGA_WR0				0x2d00
diff --git a/drivers/char/drm/mga_ioc32.c b/drivers/char/drm/mga_ioc32.c
index bc745cfa2095..77d738e75a4d 100644
--- a/drivers/char/drm/mga_ioc32.c
+++ b/drivers/char/drm/mga_ioc32.c
@@ -129,9 +129,76 @@ static int compat_mga_getparam(struct file *file, unsigned int cmd,
 			 DRM_IOCTL_MGA_GETPARAM, (unsigned long)getparam);
 }
 
+typedef struct drm_mga_drm_bootstrap32 {
+	u32 texture_handle;
+	u32 texture_size;
+	u32 primary_size;
+	u32 secondary_bin_count;
+	u32 secondary_bin_size;
+	u32 agp_mode;
+	u8 agp_size;
+} drm_mga_dma_bootstrap32_t;
+
+static int compat_mga_dma_bootstrap(struct file *file, unsigned int cmd,
+				    unsigned long arg)
+{
+	drm_mga_dma_bootstrap32_t dma_bootstrap32;
+	drm_mga_dma_bootstrap_t __user *dma_bootstrap;
+	int err;
+
+	if (copy_from_user(&dma_bootstrap32, (void __user *)arg,
+			   sizeof(dma_bootstrap32)))
+		return -EFAULT;
+
+	dma_bootstrap = compat_alloc_user_space(sizeof(*dma_bootstrap));
+	if (!access_ok(VERIFY_WRITE, dma_bootstrap, sizeof(*dma_bootstrap))
+	    || __put_user(dma_bootstrap32.texture_handle,
+			  &dma_bootstrap->texture_handle)
+	    || __put_user(dma_bootstrap32.texture_size,
+			  &dma_bootstrap->texture_size)
+	    || __put_user(dma_bootstrap32.primary_size,
+			  &dma_bootstrap->primary_size)
+	    || __put_user(dma_bootstrap32.secondary_bin_count,
+			  &dma_bootstrap->secondary_bin_count)
+	    || __put_user(dma_bootstrap32.secondary_bin_size,
+			  &dma_bootstrap->secondary_bin_size)
+	    || __put_user(dma_bootstrap32.agp_mode, &dma_bootstrap->agp_mode)
+	    || __put_user(dma_bootstrap32.agp_size, &dma_bootstrap->agp_size))
+		return -EFAULT;
+
+	err = drm_ioctl(file->f_dentry->d_inode, file,
+			DRM_IOCTL_MGA_DMA_BOOTSTRAP,
+			(unsigned long)dma_bootstrap);
+	if (err)
+		return err;
+
+	if (__get_user(dma_bootstrap32.texture_handle,
+		       &dma_bootstrap->texture_handle)
+	    || __get_user(dma_bootstrap32.texture_size,
+			  &dma_bootstrap->texture_size)
+	    || __get_user(dma_bootstrap32.primary_size,
+			  &dma_bootstrap->primary_size)
+	    || __get_user(dma_bootstrap32.secondary_bin_count,
+			  &dma_bootstrap->secondary_bin_count)
+	    || __get_user(dma_bootstrap32.secondary_bin_size,
+			  &dma_bootstrap->secondary_bin_size)
+	    || __get_user(dma_bootstrap32.agp_mode,
+			  &dma_bootstrap->agp_mode)
+	    || __get_user(dma_bootstrap32.agp_size,
+			  &dma_bootstrap->agp_size))
+		return -EFAULT;
+
+	if (copy_to_user((void __user *)arg, &dma_bootstrap32,
+	    		 sizeof(dma_bootstrap32)))
+		return -EFAULT;
+
+	return 0;
+}
+
 drm_ioctl_compat_t *mga_compat_ioctls[] = {
 	[DRM_MGA_INIT] = compat_mga_init,
 	[DRM_MGA_GETPARAM] = compat_mga_getparam,
+	[DRM_MGA_DMA_BOOTSTRAP] = compat_mga_dma_bootstrap,
 };
 
 /**
diff --git a/drivers/char/drm/mga_irq.c b/drivers/char/drm/mga_irq.c
index bc0b6b5d43a6..52eaa4e788f9 100644
--- a/drivers/char/drm/mga_irq.c
+++ b/drivers/char/drm/mga_irq.c
@@ -41,15 +41,40 @@ irqreturn_t mga_driver_irq_handler( DRM_IRQ_ARGS )
 	drm_mga_private_t *dev_priv = 
 	   (drm_mga_private_t *)dev->dev_private;
 	int status;
+	int handled = 0;
+
+	status = MGA_READ(MGA_STATUS);
 
-	status = MGA_READ( MGA_STATUS );
-	
 	/* VBLANK interrupt */
 	if ( status & MGA_VLINEPEN ) {
 		MGA_WRITE( MGA_ICLEAR, MGA_VLINEICLR );
 		atomic_inc(&dev->vbl_received);
 		DRM_WAKEUP(&dev->vbl_queue);
-		drm_vbl_send_signals( dev );
+		drm_vbl_send_signals(dev);
+		handled = 1;
+	}
+
+	/* SOFTRAP interrupt */
+	if (status & MGA_SOFTRAPEN) {
+		const u32 prim_start = MGA_READ(MGA_PRIMADDRESS);
+		const u32 prim_end   = MGA_READ(MGA_PRIMEND);
+
+
+		MGA_WRITE(MGA_ICLEAR, MGA_SOFTRAPICLR);
+
+		/* In addition to clearing the interrupt-pending bit, we
+		 * have to write to MGA_PRIMEND to re-start the DMA operation.
+		 */
+		if ( (prim_start & ~0x03) != (prim_end & ~0x03) ) {
+			MGA_WRITE(MGA_PRIMEND, prim_end);
+		}
+
+		atomic_inc(&dev_priv->last_fence_retired);
+		DRM_WAKEUP(&dev_priv->fence_queue);
+		handled = 1;
+	}
+
+	if ( handled ) {
 		return IRQ_HANDLED;
 	}
 	return IRQ_NONE;
@@ -73,9 +98,28 @@ int mga_driver_vblank_wait(drm_device_t *dev, unsigned int *sequence)
 	return ret;
 }
 
-void mga_driver_irq_preinstall( drm_device_t *dev ) {
-  	drm_mga_private_t *dev_priv = 
-	   (drm_mga_private_t *)dev->dev_private;
+int mga_driver_fence_wait(drm_device_t * dev, unsigned int *sequence)
+{
+	drm_mga_private_t *dev_priv = (drm_mga_private_t *) dev->dev_private;
+	unsigned int cur_fence;
+	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 fences.
+	 */
+	DRM_WAIT_ON(ret, dev_priv->fence_queue, 3 * DRM_HZ,
+		    (((cur_fence = atomic_read(&dev_priv->last_fence_retired))
+		      - *sequence) <= (1 << 23)));
+
+	*sequence = cur_fence;
+
+	return ret;
+}
+
+void mga_driver_irq_preinstall(drm_device_t * dev)
+{
+	drm_mga_private_t *dev_priv = (drm_mga_private_t *) dev->dev_private;
 
 	/* Disable *all* interrupts */
       	MGA_WRITE( MGA_IEN, 0 );
@@ -83,12 +127,14 @@ void mga_driver_irq_preinstall( drm_device_t *dev ) {
    	MGA_WRITE( MGA_ICLEAR, ~0 );
 }
 
-void mga_driver_irq_postinstall( drm_device_t *dev ) {
-  	drm_mga_private_t *dev_priv = 
-	   (drm_mga_private_t *)dev->dev_private;
+void mga_driver_irq_postinstall(drm_device_t * dev)
+{
+	drm_mga_private_t *dev_priv = (drm_mga_private_t *) dev->dev_private;
+
+	DRM_INIT_WAITQUEUE( &dev_priv->fence_queue );
 
-	/* Turn on VBL interrupt */
-   	MGA_WRITE( MGA_IEN, MGA_VLINEIEN );
+	/* Turn on vertical blank interrupt and soft trap interrupt. */
+	MGA_WRITE(MGA_IEN, MGA_VLINEIEN | MGA_SOFTRAPEN);
 }
 
 void mga_driver_irq_uninstall( drm_device_t *dev ) {
@@ -98,5 +144,7 @@ void mga_driver_irq_uninstall( drm_device_t *dev ) {
 		return;
 
 	/* Disable *all* interrupts */
-	MGA_WRITE( MGA_IEN, 0 );
+	MGA_WRITE(MGA_IEN, 0);
+	
+	dev->irq_enabled = 0;
 }
diff --git a/drivers/char/drm/mga_state.c b/drivers/char/drm/mga_state.c
index 3c7a8f5ba501..05bbb4719376 100644
--- a/drivers/char/drm/mga_state.c
+++ b/drivers/char/drm/mga_state.c
@@ -53,16 +53,16 @@ static void mga_emit_clip_rect( drm_mga_private_t *dev_priv,
 
 	/* Force reset of DWGCTL on G400 (eliminates clip disable bit).
 	 */
-	if ( dev_priv->chipset == MGA_CARD_TYPE_G400 ) {
-		DMA_BLOCK( MGA_DWGCTL,		ctx->dwgctl,
-			   MGA_LEN + MGA_EXEC,	0x80000000,
-			   MGA_DWGCTL,		ctx->dwgctl,
-			   MGA_LEN + MGA_EXEC,	0x80000000 );
+	if (dev_priv->chipset == MGA_CARD_TYPE_G400) {
+		DMA_BLOCK(MGA_DWGCTL, ctx->dwgctl,
+			  MGA_LEN + MGA_EXEC, 0x80000000,
+			  MGA_DWGCTL, ctx->dwgctl,
+			  MGA_LEN + MGA_EXEC, 0x80000000);
 	}
-	DMA_BLOCK( MGA_DMAPAD,	0x00000000,
-		   MGA_CXBNDRY,	(box->x2 << 16) | box->x1,
-		   MGA_YTOP,	box->y1 * pitch,
-		   MGA_YBOT,	box->y2 * pitch );
+	DMA_BLOCK(MGA_DMAPAD, 0x00000000,
+		  MGA_CXBNDRY, ((box->x2 - 1) << 16) | box->x1,
+		  MGA_YTOP, box->y1 * pitch,
+		  MGA_YBOT, (box->y2 - 1) * pitch);
 
 	ADVANCE_DMA();
 }
@@ -260,12 +260,11 @@ static __inline__ void mga_g200_emit_pipe( drm_mga_private_t *dev_priv )
 
 	/* Padding required to to hardware bug.
 	 */
-	DMA_BLOCK( MGA_DMAPAD,	0xffffffff,
-		   MGA_DMAPAD,	0xffffffff,
-		   MGA_DMAPAD,	0xffffffff,
-		   MGA_WIADDR,	(dev_priv->warp_pipe_phys[pipe] |
-				 MGA_WMODE_START |
-				 MGA_WAGP_ENABLE) );
+	DMA_BLOCK(MGA_DMAPAD, 0xffffffff,
+		  MGA_DMAPAD, 0xffffffff,
+		  MGA_DMAPAD, 0xffffffff,
+		  MGA_WIADDR, (dev_priv->warp_pipe_phys[pipe] |
+			       MGA_WMODE_START | dev_priv->wagp_enable));
 
 	ADVANCE_DMA();
 }
@@ -342,12 +341,11 @@ static __inline__ void mga_g400_emit_pipe( drm_mga_private_t *dev_priv )
 		   MGA_WR60,	MGA_G400_WR_MAGIC );	/* tex1 height       */
 
 	/* Padding required to to hardware bug */
-	DMA_BLOCK( MGA_DMAPAD,	0xffffffff,
-		   MGA_DMAPAD,	0xffffffff,
-		   MGA_DMAPAD,	0xffffffff,
-		   MGA_WIADDR2,	(dev_priv->warp_pipe_phys[pipe] |
-				 MGA_WMODE_START |
-				 MGA_WAGP_ENABLE) );
+	DMA_BLOCK(MGA_DMAPAD, 0xffffffff,
+		  MGA_DMAPAD, 0xffffffff,
+		  MGA_DMAPAD, 0xffffffff,
+		  MGA_WIADDR2, (dev_priv->warp_pipe_phys[pipe] |
+				MGA_WMODE_START | dev_priv->wagp_enable));
 
 	ADVANCE_DMA();
 }
@@ -459,9 +457,9 @@ static int mga_verify_state( drm_mga_private_t *dev_priv )
 	if ( dirty & MGA_UPLOAD_TEX0 )
 		ret |= mga_verify_tex( dev_priv, 0 );
 
-	if ( dev_priv->chipset == MGA_CARD_TYPE_G400 ) {
-		if ( dirty & MGA_UPLOAD_TEX1 )
-			ret |= mga_verify_tex( dev_priv, 1 );
+	if (dev_priv->chipset >= MGA_CARD_TYPE_G400) {
+		if (dirty & MGA_UPLOAD_TEX1)
+			ret |= mga_verify_tex(dev_priv, 1);
 
 		if ( dirty & MGA_UPLOAD_PIPE )
 			ret |= ( sarea_priv->warp_pipe > MGA_MAX_G400_PIPES );
@@ -686,12 +684,12 @@ static void mga_dma_dispatch_vertex( drm_device_t *dev, drm_buf_t *buf )
 
 			BEGIN_DMA( 1 );
 
-			DMA_BLOCK( MGA_DMAPAD,		0x00000000,
-				   MGA_DMAPAD,		0x00000000,
-				   MGA_SECADDRESS,	(address |
-							 MGA_DMA_VERTEX),
-				   MGA_SECEND,		((address + length) |
-							 MGA_PAGPXFER) );
+			DMA_BLOCK(MGA_DMAPAD, 0x00000000,
+				  MGA_DMAPAD, 0x00000000,
+				  MGA_SECADDRESS, (address |
+						   MGA_DMA_VERTEX),
+				  MGA_SECEND, ((address + length) |
+					       dev_priv->dma_access));
 
 			ADVANCE_DMA();
 		} while ( ++i < sarea_priv->nbox );
@@ -733,11 +731,11 @@ static void mga_dma_dispatch_indices( drm_device_t *dev, drm_buf_t *buf,
 
 			BEGIN_DMA( 1 );
 
-			DMA_BLOCK( MGA_DMAPAD,		0x00000000,
-				   MGA_DMAPAD,		0x00000000,
-				   MGA_SETUPADDRESS,	address + start,
-				   MGA_SETUPEND,	((address + end) |
-							 MGA_PAGPXFER) );
+			DMA_BLOCK(MGA_DMAPAD, 0x00000000,
+				  MGA_DMAPAD, 0x00000000,
+				  MGA_SETUPADDRESS, address + start,
+				  MGA_SETUPEND, ((address + end) |
+						 dev_priv->dma_access));
 
 			ADVANCE_DMA();
 		} while ( ++i < sarea_priv->nbox );
@@ -764,7 +762,7 @@ static void mga_dma_dispatch_iload( drm_device_t *dev, drm_buf_t *buf,
 	drm_mga_private_t *dev_priv = dev->dev_private;
 	drm_mga_buf_priv_t *buf_priv = buf->dev_private;
 	drm_mga_context_regs_t *ctx = &dev_priv->sarea_priv->context_state;
-	u32 srcorg = buf->bus_address | MGA_SRCACC_AGP | MGA_SRCMAP_SYSMEM;
+	u32 srcorg = buf->bus_address | dev_priv->dma_access | MGA_SRCMAP_SYSMEM;
 	u32 y2;
 	DMA_LOCALS;
 	DRM_DEBUG( "buf=%d used=%d\n", buf->idx, buf->used );
@@ -1095,6 +1093,9 @@ static int mga_getparam( DRM_IOCTL_ARGS )
 	case MGA_PARAM_IRQ_NR:
 		value = dev->irq;
 		break;
+	case MGA_PARAM_CARD_TYPE:
+		value = dev_priv->chipset;
+		break;
 	default:
 		return DRM_ERR(EINVAL);
 	}
@@ -1107,17 +1108,82 @@ static int mga_getparam( DRM_IOCTL_ARGS )
 	return 0;
 }
 
+static int mga_set_fence(DRM_IOCTL_ARGS)
+{
+	DRM_DEVICE;
+	drm_mga_private_t *dev_priv = dev->dev_private;
+	u32 temp;
+	DMA_LOCALS;
+
+	if (!dev_priv) {
+		DRM_ERROR("%s called with no initialization\n", __FUNCTION__);
+		return DRM_ERR(EINVAL);
+	}
+
+	DRM_DEBUG("pid=%d\n", DRM_CURRENTPID);
+
+	/* I would normal do this assignment in the declaration of temp,
+	 * but dev_priv may be NULL.
+	 */
+
+	temp = dev_priv->next_fence_to_post;
+	dev_priv->next_fence_to_post++;
+
+	BEGIN_DMA(1);
+	DMA_BLOCK(MGA_DMAPAD, 0x00000000,
+		  MGA_DMAPAD, 0x00000000,
+		  MGA_DMAPAD, 0x00000000,
+		  MGA_SOFTRAP, 0x00000000);
+	ADVANCE_DMA();
+
+	if (DRM_COPY_TO_USER( (u32 __user *) data, & temp, sizeof(u32))) {
+		DRM_ERROR("copy_to_user\n");
+		return DRM_ERR(EFAULT);
+	}
+
+	return 0;
+}
+
+static int mga_wait_fence(DRM_IOCTL_ARGS)
+{
+	DRM_DEVICE;
+	drm_mga_private_t *dev_priv = dev->dev_private;
+	u32 fence;
+
+	if (!dev_priv) {
+		DRM_ERROR("%s called with no initialization\n", __FUNCTION__);
+		return DRM_ERR(EINVAL);
+	}
+
+	DRM_COPY_FROM_USER_IOCTL(fence, (u32 __user *) data, sizeof(u32));
+
+	DRM_DEBUG("pid=%d\n", DRM_CURRENTPID);
+
+	mga_driver_fence_wait(dev, & fence);
+
+	if (DRM_COPY_TO_USER( (u32 __user *) data, & fence, sizeof(u32))) {
+		DRM_ERROR("copy_to_user\n");
+		return DRM_ERR(EFAULT);
+	}
+
+	return 0;
+}
+
 drm_ioctl_desc_t mga_ioctls[] = {
-	[DRM_IOCTL_NR(DRM_MGA_INIT)]    = { mga_dma_init,    1, 1 },
-	[DRM_IOCTL_NR(DRM_MGA_FLUSH)]   = { mga_dma_flush,   1, 0 },
-	[DRM_IOCTL_NR(DRM_MGA_RESET)]   = { mga_dma_reset,   1, 0 },
-	[DRM_IOCTL_NR(DRM_MGA_SWAP)]    = { mga_dma_swap,    1, 0 },
-	[DRM_IOCTL_NR(DRM_MGA_CLEAR)]   = { mga_dma_clear,   1, 0 },
-	[DRM_IOCTL_NR(DRM_MGA_VERTEX)]  = { mga_dma_vertex,  1, 0 },
-	[DRM_IOCTL_NR(DRM_MGA_INDICES)] = { mga_dma_indices, 1, 0 },
-	[DRM_IOCTL_NR(DRM_MGA_ILOAD)]   = { mga_dma_iload,   1, 0 },
-	[DRM_IOCTL_NR(DRM_MGA_BLIT)]    = { mga_dma_blit,    1, 0 },
-	[DRM_IOCTL_NR(DRM_MGA_GETPARAM)]= { mga_getparam,    1, 0 },
+	[DRM_IOCTL_NR(DRM_MGA_INIT)] = {mga_dma_init, 1, 1},
+	[DRM_IOCTL_NR(DRM_MGA_FLUSH)] = {mga_dma_flush, 1, 0},
+	[DRM_IOCTL_NR(DRM_MGA_RESET)] = {mga_dma_reset, 1, 0},
+	[DRM_IOCTL_NR(DRM_MGA_SWAP)] = {mga_dma_swap, 1, 0},
+	[DRM_IOCTL_NR(DRM_MGA_CLEAR)] = {mga_dma_clear, 1, 0},
+	[DRM_IOCTL_NR(DRM_MGA_VERTEX)] = {mga_dma_vertex, 1, 0},
+	[DRM_IOCTL_NR(DRM_MGA_INDICES)] = {mga_dma_indices, 1, 0},
+	[DRM_IOCTL_NR(DRM_MGA_ILOAD)] = {mga_dma_iload, 1, 0},
+	[DRM_IOCTL_NR(DRM_MGA_BLIT)] = {mga_dma_blit, 1, 0},
+	[DRM_IOCTL_NR(DRM_MGA_GETPARAM)] = {mga_getparam, 1, 0},
+	[DRM_IOCTL_NR(DRM_MGA_SET_FENCE)] = {mga_set_fence, 1, 0},
+	[DRM_IOCTL_NR(DRM_MGA_WAIT_FENCE)] = {mga_wait_fence, 1, 0},
+	[DRM_IOCTL_NR(DRM_MGA_DMA_BOOTSTRAP)] = {mga_dma_bootstrap, 1, 1},
+
 };
 
 int mga_max_ioctl = DRM_ARRAY_SIZE(mga_ioctls);
diff --git a/drivers/char/drm/mga_warp.c b/drivers/char/drm/mga_warp.c
index 0a3a0cc700dc..55ccc8a0ac29 100644
--- a/drivers/char/drm/mga_warp.c
+++ b/drivers/char/drm/mga_warp.c
@@ -48,65 +48,52 @@ do {									\
 	vcbase += WARP_UCODE_SIZE( which );				\
 } while (0)
 
-
-static unsigned int mga_warp_g400_microcode_size( drm_mga_private_t *dev_priv )
-{
-	unsigned int size;
-
-	size = ( WARP_UCODE_SIZE( warp_g400_tgz ) +
-		 WARP_UCODE_SIZE( warp_g400_tgza ) +
-		 WARP_UCODE_SIZE( warp_g400_tgzaf ) +
-		 WARP_UCODE_SIZE( warp_g400_tgzf ) +
-		 WARP_UCODE_SIZE( warp_g400_tgzs ) +
-		 WARP_UCODE_SIZE( warp_g400_tgzsa ) +
-		 WARP_UCODE_SIZE( warp_g400_tgzsaf ) +
-		 WARP_UCODE_SIZE( warp_g400_tgzsf ) +
-		 WARP_UCODE_SIZE( warp_g400_t2gz ) +
-		 WARP_UCODE_SIZE( warp_g400_t2gza ) +
-		 WARP_UCODE_SIZE( warp_g400_t2gzaf ) +
-		 WARP_UCODE_SIZE( warp_g400_t2gzf ) +
-		 WARP_UCODE_SIZE( warp_g400_t2gzs ) +
-		 WARP_UCODE_SIZE( warp_g400_t2gzsa ) +
-		 WARP_UCODE_SIZE( warp_g400_t2gzsaf ) +
-		 WARP_UCODE_SIZE( warp_g400_t2gzsf ) );
-
-	size = PAGE_ALIGN( size );
-
-	DRM_DEBUG( "G400 ucode size = %d bytes\n", size );
-	return size;
-}
-
-static unsigned int mga_warp_g200_microcode_size( drm_mga_private_t *dev_priv )
+static const unsigned int mga_warp_g400_microcode_size =
+	       (WARP_UCODE_SIZE(warp_g400_tgz) +
+		WARP_UCODE_SIZE(warp_g400_tgza) +
+		WARP_UCODE_SIZE(warp_g400_tgzaf) +
+		WARP_UCODE_SIZE(warp_g400_tgzf) +
+		WARP_UCODE_SIZE(warp_g400_tgzs) +
+		WARP_UCODE_SIZE(warp_g400_tgzsa) +
+		WARP_UCODE_SIZE(warp_g400_tgzsaf) +
+		WARP_UCODE_SIZE(warp_g400_tgzsf) +
+		WARP_UCODE_SIZE(warp_g400_t2gz) +
+		WARP_UCODE_SIZE(warp_g400_t2gza) +
+		WARP_UCODE_SIZE(warp_g400_t2gzaf) +
+		WARP_UCODE_SIZE(warp_g400_t2gzf) +
+		WARP_UCODE_SIZE(warp_g400_t2gzs) +
+		WARP_UCODE_SIZE(warp_g400_t2gzsa) +
+		WARP_UCODE_SIZE(warp_g400_t2gzsaf) +
+		WARP_UCODE_SIZE(warp_g400_t2gzsf));
+
+static const unsigned int mga_warp_g200_microcode_size =
+	       (WARP_UCODE_SIZE(warp_g200_tgz) +
+		WARP_UCODE_SIZE(warp_g200_tgza) +
+		WARP_UCODE_SIZE(warp_g200_tgzaf) +
+		WARP_UCODE_SIZE(warp_g200_tgzf) +
+		WARP_UCODE_SIZE(warp_g200_tgzs) +
+		WARP_UCODE_SIZE(warp_g200_tgzsa) +
+		WARP_UCODE_SIZE(warp_g200_tgzsaf) +
+		WARP_UCODE_SIZE(warp_g200_tgzsf));
+
+
+unsigned int mga_warp_microcode_size(const drm_mga_private_t * dev_priv)
 {
-	unsigned int size;
-
-	size = ( WARP_UCODE_SIZE( warp_g200_tgz ) +
-		 WARP_UCODE_SIZE( warp_g200_tgza ) +
-		 WARP_UCODE_SIZE( warp_g200_tgzaf ) +
-		 WARP_UCODE_SIZE( warp_g200_tgzf ) +
-		 WARP_UCODE_SIZE( warp_g200_tgzs ) +
-		 WARP_UCODE_SIZE( warp_g200_tgzsa ) +
-		 WARP_UCODE_SIZE( warp_g200_tgzsaf ) +
-		 WARP_UCODE_SIZE( warp_g200_tgzsf ) );
-
-	size = PAGE_ALIGN( size );
-
-	DRM_DEBUG( "G200 ucode size = %d bytes\n", size );
-	return size;
+	switch (dev_priv->chipset) {
+	case MGA_CARD_TYPE_G400:
+	case MGA_CARD_TYPE_G550:
+		return PAGE_ALIGN(mga_warp_g400_microcode_size);
+	case MGA_CARD_TYPE_G200:
+		return PAGE_ALIGN(mga_warp_g200_microcode_size);
+	default:
+		return 0;
+	}
 }
 
 static int mga_warp_install_g400_microcode( drm_mga_private_t *dev_priv )
 {
 	unsigned char *vcbase = dev_priv->warp->handle;
 	unsigned long pcbase = dev_priv->warp->offset;
-	unsigned int size;
-
-	size = mga_warp_g400_microcode_size( dev_priv );
-	if ( size > dev_priv->warp->size ) {
-		DRM_ERROR( "microcode too large! (%u > %lu)\n",
-			   size, dev_priv->warp->size );
-		return DRM_ERR(ENOMEM);
-	}
 
 	memset( dev_priv->warp_pipe_phys, 0,
 		sizeof(dev_priv->warp_pipe_phys) );
@@ -136,35 +123,36 @@ static int mga_warp_install_g200_microcode( drm_mga_private_t *dev_priv )
 {
 	unsigned char *vcbase = dev_priv->warp->handle;
 	unsigned long pcbase = dev_priv->warp->offset;
-	unsigned int size;
-
-	size = mga_warp_g200_microcode_size( dev_priv );
-	if ( size > dev_priv->warp->size ) {
-		DRM_ERROR( "microcode too large! (%u > %lu)\n",
-			   size, dev_priv->warp->size );
-		return DRM_ERR(ENOMEM);
-	}
 
-	memset( dev_priv->warp_pipe_phys, 0,
-		sizeof(dev_priv->warp_pipe_phys) );
+	memset(dev_priv->warp_pipe_phys, 0, sizeof(dev_priv->warp_pipe_phys));
 
-	WARP_UCODE_INSTALL( warp_g200_tgz,	MGA_WARP_TGZ );
-	WARP_UCODE_INSTALL( warp_g200_tgzf,	MGA_WARP_TGZF );
-	WARP_UCODE_INSTALL( warp_g200_tgza,	MGA_WARP_TGZA );
-	WARP_UCODE_INSTALL( warp_g200_tgzaf,	MGA_WARP_TGZAF );
-	WARP_UCODE_INSTALL( warp_g200_tgzs,	MGA_WARP_TGZS );
-	WARP_UCODE_INSTALL( warp_g200_tgzsf,	MGA_WARP_TGZSF );
-	WARP_UCODE_INSTALL( warp_g200_tgzsa,	MGA_WARP_TGZSA );
-	WARP_UCODE_INSTALL( warp_g200_tgzsaf,	MGA_WARP_TGZSAF );
+	WARP_UCODE_INSTALL(warp_g200_tgz, MGA_WARP_TGZ);
+	WARP_UCODE_INSTALL(warp_g200_tgzf, MGA_WARP_TGZF);
+	WARP_UCODE_INSTALL(warp_g200_tgza, MGA_WARP_TGZA);
+	WARP_UCODE_INSTALL(warp_g200_tgzaf, MGA_WARP_TGZAF);
+	WARP_UCODE_INSTALL(warp_g200_tgzs, MGA_WARP_TGZS);
+	WARP_UCODE_INSTALL(warp_g200_tgzsf, MGA_WARP_TGZSF);
+	WARP_UCODE_INSTALL(warp_g200_tgzsa, MGA_WARP_TGZSA);
+	WARP_UCODE_INSTALL(warp_g200_tgzsaf, MGA_WARP_TGZSAF);
 
 	return 0;
 }
 
 int mga_warp_install_microcode(	drm_mga_private_t *dev_priv )
 {
-	switch ( dev_priv->chipset ) {
+	const unsigned int size = mga_warp_microcode_size(dev_priv);
+
+	DRM_DEBUG("MGA ucode size = %d bytes\n", size);
+	if (size > dev_priv->warp->size) {
+		DRM_ERROR("microcode too large! (%u > %lu)\n",
+			  size, dev_priv->warp->size);
+		return DRM_ERR(ENOMEM);
+	}
+
+	switch (dev_priv->chipset) {
 	case MGA_CARD_TYPE_G400:
-		return mga_warp_install_g400_microcode( dev_priv );
+	case MGA_CARD_TYPE_G550:
+		return mga_warp_install_g400_microcode(dev_priv);
 	case MGA_CARD_TYPE_G200:
 		return mga_warp_install_g200_microcode( dev_priv );
 	default:
@@ -182,10 +170,11 @@ int mga_warp_init( drm_mga_private_t *dev_priv )
 	 */
 	switch ( dev_priv->chipset ) {
 	case MGA_CARD_TYPE_G400:
-		MGA_WRITE( MGA_WIADDR2, MGA_WMODE_SUSPEND );
-		MGA_WRITE( MGA_WGETMSB, 0x00000E00 );
-		MGA_WRITE( MGA_WVRTXSZ, 0x00001807 );
-		MGA_WRITE( MGA_WACCEPTSEQ, 0x18000000 );
+	case MGA_CARD_TYPE_G550:
+		MGA_WRITE(MGA_WIADDR2, MGA_WMODE_SUSPEND);
+		MGA_WRITE(MGA_WGETMSB, 0x00000E00);
+		MGA_WRITE(MGA_WVRTXSZ, 0x00001807);
+		MGA_WRITE(MGA_WACCEPTSEQ, 0x18000000);
 		break;
 	case MGA_CARD_TYPE_G200:
 		MGA_WRITE( MGA_WIADDR, MGA_WMODE_SUSPEND );
diff --git a/drivers/char/drm/r128_cce.c b/drivers/char/drm/r128_cce.c
index 08ed8d01d9d9..895152206b31 100644
--- a/drivers/char/drm/r128_cce.c
+++ b/drivers/char/drm/r128_cce.c
@@ -326,7 +326,8 @@ static void r128_cce_init_ring_buffer( drm_device_t *dev,
 		ring_start = dev_priv->cce_ring->offset - dev->agp->base;
 	else
 #endif
-		ring_start = dev_priv->cce_ring->offset - dev->sg->handle;
+		ring_start = dev_priv->cce_ring->offset - 
+				(unsigned long)dev->sg->virtual;
 
 	R128_WRITE( R128_PM4_BUFFER_OFFSET, ring_start | R128_AGP_OFFSET );
 
@@ -487,6 +488,7 @@ static int r128_do_init_cce( drm_device_t *dev, drm_r128_init_t *init )
 		r128_do_cleanup_cce( dev );
 		return DRM_ERR(EINVAL);
 	}
+	dev->agp_buffer_token = init->buffers_offset;
 	dev->agp_buffer_map = drm_core_findmap(dev, init->buffers_offset);
 	if(!dev->agp_buffer_map) {
 		DRM_ERROR("could not find dma buffer region!\n");
@@ -537,7 +539,7 @@ static int r128_do_init_cce( drm_device_t *dev, drm_r128_init_t *init )
 		dev_priv->cce_buffers_offset = dev->agp->base;
 	else
 #endif
-		dev_priv->cce_buffers_offset = dev->sg->handle;
+		dev_priv->cce_buffers_offset = (unsigned long)dev->sg->virtual;
 
 	dev_priv->ring.start = (u32 *)dev_priv->cce_ring->handle;
 	dev_priv->ring.end = ((u32 *)dev_priv->cce_ring->handle
diff --git a/drivers/char/drm/r128_drm.h b/drivers/char/drm/r128_drm.h
index 0cba17d1e0ff..b616cd3ed2cd 100644
--- a/drivers/char/drm/r128_drm.h
+++ b/drivers/char/drm/r128_drm.h
@@ -215,7 +215,7 @@ typedef struct drm_r128_sarea {
 #define DRM_IOCTL_R128_INDIRECT   DRM_IOWR(DRM_COMMAND_BASE + DRM_R128_INDIRECT, drm_r128_indirect_t)
 #define DRM_IOCTL_R128_FULLSCREEN DRM_IOW( DRM_COMMAND_BASE + DRM_R128_FULLSCREEN, drm_r128_fullscreen_t)
 #define DRM_IOCTL_R128_CLEAR2     DRM_IOW( DRM_COMMAND_BASE + DRM_R128_CLEAR2, drm_r128_clear2_t)
-#define DRM_IOCTL_R128_GETPARAM   DRM_IOW( DRM_COMMAND_BASE + DRM_R128_GETPARAM, drm_r128_getparam_t)
+#define DRM_IOCTL_R128_GETPARAM   DRM_IOWR( DRM_COMMAND_BASE + DRM_R128_GETPARAM, drm_r128_getparam_t)
 #define DRM_IOCTL_R128_FLIP       DRM_IO(  DRM_COMMAND_BASE + DRM_R128_FLIP)
 
 typedef struct drm_r128_init {
diff --git a/drivers/char/drm/r300_cmdbuf.c b/drivers/char/drm/r300_cmdbuf.c
new file mode 100644
index 000000000000..623f1f460cb5
--- /dev/null
+++ b/drivers/char/drm/r300_cmdbuf.c
@@ -0,0 +1,801 @@
+/* r300_cmdbuf.c -- Command buffer emission for R300 -*- linux-c -*-
+ *
+ * Copyright (C) The Weather Channel, Inc.  2002.
+ * Copyright (C) 2004 Nicolai Haehnle.
+ * 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
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Nicolai Haehnle <prefect_@gmx.net>
+ */
+
+#include "drmP.h"
+#include "drm.h"
+#include "radeon_drm.h"
+#include "radeon_drv.h"
+#include "r300_reg.h"
+
+
+#define R300_SIMULTANEOUS_CLIPRECTS		4
+
+/* Values for R300_RE_CLIPRECT_CNTL depending on the number of cliprects
+ */
+static const int r300_cliprect_cntl[4] = {
+	0xAAAA,
+	0xEEEE,
+	0xFEFE,
+	0xFFFE
+};
+
+
+/**
+ * Emit up to R300_SIMULTANEOUS_CLIPRECTS cliprects from the given command
+ * buffer, starting with index n.
+ */
+static int r300_emit_cliprects(drm_radeon_private_t* dev_priv,
+			       drm_radeon_cmd_buffer_t* cmdbuf,
+			       int n)
+{
+	drm_clip_rect_t box;
+	int nr;
+	int i;
+	RING_LOCALS;
+
+	nr = cmdbuf->nbox - n;
+	if (nr > R300_SIMULTANEOUS_CLIPRECTS)
+		nr = R300_SIMULTANEOUS_CLIPRECTS;
+
+	DRM_DEBUG("%i cliprects\n", nr);
+
+	if (nr) {
+		BEGIN_RING(6 + nr*2);
+		OUT_RING( CP_PACKET0( R300_RE_CLIPRECT_TL_0, nr*2 - 1 ) );
+
+		for(i = 0; i < nr; ++i) {
+			if (DRM_COPY_FROM_USER_UNCHECKED(&box, &cmdbuf->boxes[n+i], sizeof(box))) {
+				DRM_ERROR("copy cliprect faulted\n");
+				return DRM_ERR(EFAULT);
+			}
+
+			box.x1 = (box.x1 + R300_CLIPRECT_OFFSET) & R300_CLIPRECT_MASK;
+			box.y1 = (box.y1 + R300_CLIPRECT_OFFSET) & R300_CLIPRECT_MASK;
+			box.x2 = (box.x2 + R300_CLIPRECT_OFFSET) & R300_CLIPRECT_MASK;
+			box.y2 = (box.y2 + R300_CLIPRECT_OFFSET) & R300_CLIPRECT_MASK;
+
+			OUT_RING((box.x1 << R300_CLIPRECT_X_SHIFT) |
+					(box.y1 << R300_CLIPRECT_Y_SHIFT));
+			OUT_RING((box.x2 << R300_CLIPRECT_X_SHIFT) |
+					(box.y2 << R300_CLIPRECT_Y_SHIFT));
+		}
+
+		OUT_RING_REG( R300_RE_CLIPRECT_CNTL, r300_cliprect_cntl[nr-1] );
+
+		/* TODO/SECURITY: Force scissors to a safe value, otherwise the
+		* client might be able to trample over memory.
+		* The impact should be very limited, but I'd rather be safe than
+		* sorry.
+		*/
+		OUT_RING( CP_PACKET0( R300_RE_SCISSORS_TL, 1 ) );
+		OUT_RING( 0 );
+		OUT_RING( R300_SCISSORS_X_MASK | R300_SCISSORS_Y_MASK );
+		ADVANCE_RING();
+		} else {
+		/* Why we allow zero cliprect rendering:
+		 * There are some commands in a command buffer that must be submitted
+		 * even when there are no cliprects, e.g. DMA buffer discard
+		 * or state setting (though state setting could be avoided by
+		 * simulating a loss of context).
+		 *
+		 * Now since the cmdbuf interface is so chaotic right now (and is
+		 * bound to remain that way for a bit until things settle down),
+		 * it is basically impossible to filter out the commands that are
+		 * necessary and those that aren't.
+		 *
+		 * So I choose the safe way and don't do any filtering at all;
+		 * instead, I simply set up the engine so that all rendering
+		 * can't produce any fragments.
+		 */
+		BEGIN_RING(2);
+		OUT_RING_REG( R300_RE_CLIPRECT_CNTL, 0 );
+		ADVANCE_RING();
+		}
+
+	return 0;
+}
+
+u8  r300_reg_flags[0x10000>>2];
+
+
+void r300_init_reg_flags(void)
+{
+	int i;
+	memset(r300_reg_flags, 0, 0x10000>>2);
+	#define ADD_RANGE_MARK(reg, count,mark) \
+		for(i=((reg)>>2);i<((reg)>>2)+(count);i++)\
+			r300_reg_flags[i]|=(mark);
+	
+	#define MARK_SAFE		1
+	#define MARK_CHECK_OFFSET	2
+	
+	#define ADD_RANGE(reg, count)	ADD_RANGE_MARK(reg, count, MARK_SAFE)
+
+	/* these match cmducs() command in r300_driver/r300/r300_cmdbuf.c */
+	ADD_RANGE(R300_SE_VPORT_XSCALE, 6);
+	ADD_RANGE(0x2080, 1);
+	ADD_RANGE(R300_SE_VTE_CNTL, 2);
+	ADD_RANGE(0x2134, 2);
+	ADD_RANGE(0x2140, 1);
+	ADD_RANGE(R300_VAP_INPUT_CNTL_0, 2);
+	ADD_RANGE(0x21DC, 1);
+	ADD_RANGE(0x221C, 1);
+	ADD_RANGE(0x2220, 4);
+	ADD_RANGE(0x2288, 1);
+	ADD_RANGE(R300_VAP_OUTPUT_VTX_FMT_0, 2);
+	ADD_RANGE(R300_VAP_PVS_CNTL_1, 3);
+	ADD_RANGE(R300_GB_ENABLE, 1);
+	ADD_RANGE(R300_GB_MSPOS0, 5);
+	ADD_RANGE(R300_TX_ENABLE, 1);
+	ADD_RANGE(0x4200, 4);
+	ADD_RANGE(0x4214, 1);
+	ADD_RANGE(R300_RE_POINTSIZE, 1);
+	ADD_RANGE(0x4230, 3);
+	ADD_RANGE(R300_RE_LINE_CNT, 1);
+	ADD_RANGE(0x4238, 1);
+	ADD_RANGE(0x4260, 3);
+	ADD_RANGE(0x4274, 4);
+	ADD_RANGE(0x4288, 5);
+	ADD_RANGE(0x42A0, 1);
+	ADD_RANGE(R300_RE_ZBIAS_T_FACTOR, 4);
+	ADD_RANGE(0x42B4, 1);
+	ADD_RANGE(R300_RE_CULL_CNTL, 1);
+	ADD_RANGE(0x42C0, 2);
+	ADD_RANGE(R300_RS_CNTL_0, 2);
+	ADD_RANGE(R300_RS_INTERP_0, 8);
+	ADD_RANGE(R300_RS_ROUTE_0, 8);
+	ADD_RANGE(0x43A4, 2);
+	ADD_RANGE(0x43E8, 1);
+	ADD_RANGE(R300_PFS_CNTL_0, 3);
+	ADD_RANGE(R300_PFS_NODE_0, 4);
+	ADD_RANGE(R300_PFS_TEXI_0, 64);
+	ADD_RANGE(0x46A4, 5);
+	ADD_RANGE(R300_PFS_INSTR0_0, 64);
+	ADD_RANGE(R300_PFS_INSTR1_0, 64);
+	ADD_RANGE(R300_PFS_INSTR2_0, 64);
+	ADD_RANGE(R300_PFS_INSTR3_0, 64);
+	ADD_RANGE(0x4BC0, 1);
+	ADD_RANGE(0x4BC8, 3);
+	ADD_RANGE(R300_PP_ALPHA_TEST, 2);
+	ADD_RANGE(0x4BD8, 1);
+	ADD_RANGE(R300_PFS_PARAM_0_X, 64);
+	ADD_RANGE(0x4E00, 1);
+	ADD_RANGE(R300_RB3D_CBLEND, 2);
+	ADD_RANGE(R300_RB3D_COLORMASK, 1);
+	ADD_RANGE(0x4E10, 3);
+	ADD_RANGE_MARK(R300_RB3D_COLOROFFSET0, 1, MARK_CHECK_OFFSET); /* check offset */
+	ADD_RANGE(R300_RB3D_COLORPITCH0, 1);
+	ADD_RANGE(0x4E50, 9);
+	ADD_RANGE(0x4E88, 1);
+	ADD_RANGE(0x4EA0, 2);
+	ADD_RANGE(R300_RB3D_ZSTENCIL_CNTL_0, 3);
+	ADD_RANGE(0x4F10, 4);
+	ADD_RANGE_MARK(R300_RB3D_DEPTHOFFSET, 1, MARK_CHECK_OFFSET); /* check offset */
+	ADD_RANGE(R300_RB3D_DEPTHPITCH, 1); 
+	ADD_RANGE(0x4F28, 1);
+	ADD_RANGE(0x4F30, 2);
+	ADD_RANGE(0x4F44, 1);
+	ADD_RANGE(0x4F54, 1);
+
+	ADD_RANGE(R300_TX_FILTER_0, 16);
+	ADD_RANGE(R300_TX_UNK1_0, 16);
+	ADD_RANGE(R300_TX_SIZE_0, 16);
+	ADD_RANGE(R300_TX_FORMAT_0, 16);
+		/* Texture offset is dangerous and needs more checking */
+	ADD_RANGE_MARK(R300_TX_OFFSET_0, 16, MARK_CHECK_OFFSET);
+	ADD_RANGE(R300_TX_UNK4_0, 16);
+	ADD_RANGE(R300_TX_BORDER_COLOR_0, 16);
+
+	/* Sporadic registers used as primitives are emitted */
+	ADD_RANGE(0x4f18, 1);
+	ADD_RANGE(R300_RB3D_DSTCACHE_CTLSTAT, 1);
+	ADD_RANGE(R300_VAP_INPUT_ROUTE_0_0, 8);
+	ADD_RANGE(R300_VAP_INPUT_ROUTE_1_0, 8);
+
+}
+
+static __inline__ int r300_check_range(unsigned  reg, int count)
+{
+	int i;
+	if(reg & ~0xffff)return -1;
+	for(i=(reg>>2);i<(reg>>2)+count;i++)
+		if(r300_reg_flags[i]!=MARK_SAFE)return 1;
+	return 0;
+}
+
+  /* we expect offsets passed to the framebuffer to be either within video memory or
+      within AGP space */
+static __inline__ int r300_check_offset(drm_radeon_private_t* dev_priv, u32 offset)
+{
+	/* we realy want to check against end of video aperture
+		but this value is not being kept. 
+		This code is correct for now (does the same thing as the
+		code that sets MC_FB_LOCATION) in radeon_cp.c */
+	if((offset>=dev_priv->fb_location) && 
+		(offset<dev_priv->gart_vm_start))return 0;
+	if((offset>=dev_priv->gart_vm_start) &&
+		 (offset<dev_priv->gart_vm_start+dev_priv->gart_size))return 0;
+	return 1;
+}
+
+static __inline__ int r300_emit_carefully_checked_packet0(drm_radeon_private_t* dev_priv,
+						drm_radeon_cmd_buffer_t* cmdbuf,
+						drm_r300_cmd_header_t header)
+{
+	int reg;
+	int sz;
+	int i;
+	int values[64];
+	RING_LOCALS;
+
+	sz = header.packet0.count;
+	reg = (header.packet0.reghi << 8) | header.packet0.reglo;
+	
+	if((sz>64)||(sz<0)){
+		DRM_ERROR("Cannot emit more than 64 values at a time (reg=%04x sz=%d)\n", reg, sz);
+		return DRM_ERR(EINVAL);
+		}
+	for(i=0;i<sz;i++){
+		values[i]=((int __user*)cmdbuf->buf)[i];
+		switch(r300_reg_flags[(reg>>2)+i]){
+		case MARK_SAFE:
+			break;
+		case MARK_CHECK_OFFSET:
+			if(r300_check_offset(dev_priv, (u32)values[i])){
+				DRM_ERROR("Offset failed range check (reg=%04x sz=%d)\n", reg, sz);
+				return DRM_ERR(EINVAL);
+				}
+			break;
+		default:
+			DRM_ERROR("Register %04x failed check as flag=%02x\n", reg+i*4, r300_reg_flags[(reg>>2)+i]);
+			return DRM_ERR(EINVAL);
+			}
+		}
+		
+	BEGIN_RING(1+sz);
+	OUT_RING( CP_PACKET0( reg, sz-1 ) );
+	OUT_RING_TABLE( values, sz );
+	ADVANCE_RING();
+
+	cmdbuf->buf += sz*4;
+	cmdbuf->bufsz -= sz*4;
+
+	return 0;
+}
+
+/**
+ * Emits a packet0 setting arbitrary registers.
+ * Called by r300_do_cp_cmdbuf.
+ *
+ * Note that checks are performed on contents and addresses of the registers
+ */
+static __inline__ int r300_emit_packet0(drm_radeon_private_t* dev_priv,
+						drm_radeon_cmd_buffer_t* cmdbuf,
+						drm_r300_cmd_header_t header)
+{
+	int reg;
+	int sz;
+	RING_LOCALS;
+
+	sz = header.packet0.count;
+	reg = (header.packet0.reghi << 8) | header.packet0.reglo;
+
+	if (!sz)
+		return 0;
+
+	if (sz*4 > cmdbuf->bufsz)
+		return DRM_ERR(EINVAL);
+		
+	if (reg+sz*4 >= 0x10000){
+		DRM_ERROR("No such registers in hardware reg=%04x sz=%d\n", reg, sz);
+		return DRM_ERR(EINVAL);
+		}
+
+	if(r300_check_range(reg, sz)){
+		/* go and check everything */
+		return r300_emit_carefully_checked_packet0(dev_priv, cmdbuf, header);
+		}
+	/* the rest of the data is safe to emit, whatever the values the user passed */
+
+	BEGIN_RING(1+sz);
+	OUT_RING( CP_PACKET0( reg, sz-1 ) );
+	OUT_RING_TABLE( (int __user*)cmdbuf->buf, sz );
+	ADVANCE_RING();
+
+	cmdbuf->buf += sz*4;
+	cmdbuf->bufsz -= sz*4;
+
+	return 0;
+}
+
+
+/**
+ * Uploads user-supplied vertex program instructions or parameters onto
+ * the graphics card.
+ * Called by r300_do_cp_cmdbuf.
+ */
+static __inline__ int r300_emit_vpu(drm_radeon_private_t* dev_priv,
+				    drm_radeon_cmd_buffer_t* cmdbuf,
+				    drm_r300_cmd_header_t header)
+{
+	int sz;
+	int addr;
+	RING_LOCALS;
+
+	sz = header.vpu.count;
+	addr = (header.vpu.adrhi << 8) | header.vpu.adrlo;
+
+	if (!sz)
+		return 0;
+	if (sz*16 > cmdbuf->bufsz)
+		return DRM_ERR(EINVAL);
+
+	BEGIN_RING(5+sz*4);
+	/* Wait for VAP to come to senses.. */
+	/* there is no need to emit it multiple times, (only once before VAP is programmed,
+	   but this optimization is for later */
+	OUT_RING_REG( R300_VAP_PVS_WAITIDLE, 0 );
+	OUT_RING_REG( R300_VAP_PVS_UPLOAD_ADDRESS, addr );
+	OUT_RING( CP_PACKET0_TABLE( R300_VAP_PVS_UPLOAD_DATA, sz*4 - 1 ) );
+	OUT_RING_TABLE( (int __user*)cmdbuf->buf, sz*4 );
+
+	ADVANCE_RING();
+
+	cmdbuf->buf += sz*16;
+	cmdbuf->bufsz -= sz*16;
+
+	return 0;
+}
+
+
+/**
+ * Emit a clear packet from userspace.
+ * Called by r300_emit_packet3.
+ */
+static __inline__ int r300_emit_clear(drm_radeon_private_t* dev_priv,
+				      drm_radeon_cmd_buffer_t* cmdbuf)
+{
+	RING_LOCALS;
+
+	if (8*4 > cmdbuf->bufsz)
+		return DRM_ERR(EINVAL);
+
+	BEGIN_RING(10);
+	OUT_RING( CP_PACKET3( R200_3D_DRAW_IMMD_2, 8 ) );
+	OUT_RING( R300_PRIM_TYPE_POINT|R300_PRIM_WALK_RING|
+	          (1<<R300_PRIM_NUM_VERTICES_SHIFT) );
+	OUT_RING_TABLE( (int __user*)cmdbuf->buf, 8 );
+	ADVANCE_RING();
+
+	cmdbuf->buf += 8*4;
+	cmdbuf->bufsz -= 8*4;
+
+	return 0;
+}
+
+static __inline__ int r300_emit_3d_load_vbpntr(drm_radeon_private_t* dev_priv,
+				      drm_radeon_cmd_buffer_t* cmdbuf,
+				      u32 header)
+{
+	int count, i,k;
+	#define MAX_ARRAY_PACKET  64
+	u32 payload[MAX_ARRAY_PACKET];
+	u32 narrays;
+	RING_LOCALS;
+
+	count=(header>>16) & 0x3fff;
+	
+	if((count+1)>MAX_ARRAY_PACKET){
+		DRM_ERROR("Too large payload in 3D_LOAD_VBPNTR (count=%d)\n", count);
+		return DRM_ERR(EINVAL);
+		}
+	memset(payload, 0, MAX_ARRAY_PACKET*4);
+	memcpy(payload, cmdbuf->buf+4, (count+1)*4);	
+	
+	/* carefully check packet contents */
+	
+	narrays=payload[0];
+	k=0;
+	i=1;
+	while((k<narrays) && (i<(count+1))){
+		i++; /* skip attribute field */
+		if(r300_check_offset(dev_priv, payload[i])){
+			DRM_ERROR("Offset failed range check (k=%d i=%d) while processing 3D_LOAD_VBPNTR packet.\n", k, i);
+			return DRM_ERR(EINVAL);
+			}
+		k++;
+		i++;
+		if(k==narrays)break;
+		/* have one more to process, they come in pairs */
+		if(r300_check_offset(dev_priv, payload[i])){
+			DRM_ERROR("Offset failed range check (k=%d i=%d) while processing 3D_LOAD_VBPNTR packet.\n", k, i);
+			return DRM_ERR(EINVAL);
+			}
+		k++;
+		i++;			
+		}
+	/* do the counts match what we expect ? */
+	if((k!=narrays) || (i!=(count+1))){
+		DRM_ERROR("Malformed 3D_LOAD_VBPNTR packet (k=%d i=%d narrays=%d count+1=%d).\n", k, i, narrays, count+1);
+		return DRM_ERR(EINVAL);
+		}
+
+	/* all clear, output packet */
+
+	BEGIN_RING(count+2);
+	OUT_RING(header);
+	OUT_RING_TABLE(payload, count+1);
+	ADVANCE_RING();
+
+	cmdbuf->buf += (count+2)*4;
+	cmdbuf->bufsz -= (count+2)*4;
+
+	return 0;
+}
+
+static __inline__ int r300_emit_raw_packet3(drm_radeon_private_t* dev_priv,
+				      drm_radeon_cmd_buffer_t* cmdbuf)
+{
+	u32 header;
+	int count;
+	RING_LOCALS;
+
+	if (4 > cmdbuf->bufsz)
+		return DRM_ERR(EINVAL);
+
+        /* Fixme !! This simply emits a packet without much checking.
+	   We need to be smarter. */
+
+	/* obtain first word - actual packet3 header */
+	header = *(u32 __user*)cmdbuf->buf;
+
+	/* Is it packet 3 ? */
+	if( (header>>30)!=0x3 ) {
+		DRM_ERROR("Not a packet3 header (0x%08x)\n", header);
+		return DRM_ERR(EINVAL);
+		}
+
+	count=(header>>16) & 0x3fff;
+
+	/* Check again now that we know how much data to expect */
+	if ((count+2)*4 > cmdbuf->bufsz){
+		DRM_ERROR("Expected packet3 of length %d but have only %d bytes left\n",
+			(count+2)*4, cmdbuf->bufsz);
+		return DRM_ERR(EINVAL);
+		}
+
+	/* Is it a packet type we know about ? */
+	switch(header & 0xff00){
+	case RADEON_3D_LOAD_VBPNTR: /* load vertex array pointers */
+		return r300_emit_3d_load_vbpntr(dev_priv, cmdbuf, header);
+
+	case RADEON_CP_3D_DRAW_IMMD_2: /* triggers drawing using in-packet vertex data */
+	case RADEON_CP_3D_DRAW_VBUF_2: /* triggers drawing of vertex buffers setup elsewhere */
+	case RADEON_CP_3D_DRAW_INDX_2: /* triggers drawing using indices to vertex buffer */
+	case RADEON_CP_INDX_BUFFER: /* DRAW_INDX_2 without INDX_BUFFER seems to lock up the gpu */
+	case RADEON_WAIT_FOR_IDLE:
+	case RADEON_CP_NOP:
+		/* these packets are safe */
+		break;
+	default:
+		DRM_ERROR("Unknown packet3 header (0x%08x)\n", header);
+		return DRM_ERR(EINVAL);
+		}
+
+
+	BEGIN_RING(count+2);
+	OUT_RING(header);
+	OUT_RING_TABLE( (int __user*)(cmdbuf->buf+4), count+1);
+	ADVANCE_RING();
+
+	cmdbuf->buf += (count+2)*4;
+	cmdbuf->bufsz -= (count+2)*4;
+
+	return 0;
+}
+
+
+/**
+ * Emit a rendering packet3 from userspace.
+ * Called by r300_do_cp_cmdbuf.
+ */
+static __inline__ int r300_emit_packet3(drm_radeon_private_t* dev_priv,
+					drm_radeon_cmd_buffer_t* cmdbuf,
+					drm_r300_cmd_header_t header)
+{
+	int n;
+	int ret;
+	char __user* orig_buf = cmdbuf->buf;
+	int orig_bufsz = cmdbuf->bufsz;
+
+	/* This is a do-while-loop so that we run the interior at least once,
+	 * even if cmdbuf->nbox is 0. Compare r300_emit_cliprects for rationale.
+	 */
+	n = 0;
+	do {
+		if (cmdbuf->nbox > R300_SIMULTANEOUS_CLIPRECTS) {
+			ret = r300_emit_cliprects(dev_priv, cmdbuf, n);
+			if (ret)
+				return ret;
+
+			cmdbuf->buf = orig_buf;
+			cmdbuf->bufsz = orig_bufsz;
+			}
+
+		switch(header.packet3.packet) {
+		case R300_CMD_PACKET3_CLEAR:
+			DRM_DEBUG("R300_CMD_PACKET3_CLEAR\n");
+			ret = r300_emit_clear(dev_priv, cmdbuf);
+			if (ret) {
+				DRM_ERROR("r300_emit_clear failed\n");
+				return ret;
+				}
+			break;
+
+		case R300_CMD_PACKET3_RAW:
+			DRM_DEBUG("R300_CMD_PACKET3_RAW\n");
+			ret = r300_emit_raw_packet3(dev_priv, cmdbuf);
+			if (ret) {
+				DRM_ERROR("r300_emit_raw_packet3 failed\n");
+				return ret;
+				}
+			break;
+
+		default:
+			DRM_ERROR("bad packet3 type %i at %p\n",
+				header.packet3.packet,
+				cmdbuf->buf - sizeof(header));
+			return DRM_ERR(EINVAL);
+			}
+
+		n += R300_SIMULTANEOUS_CLIPRECTS;
+	} while(n < cmdbuf->nbox);
+
+	return 0;
+}
+
+/* Some of the R300 chips seem to be extremely touchy about the two registers
+ * that are configured in r300_pacify.
+ * Among the worst offenders seems to be the R300 ND (0x4E44): When userspace
+ * sends a command buffer that contains only state setting commands and a
+ * vertex program/parameter upload sequence, this will eventually lead to a
+ * lockup, unless the sequence is bracketed by calls to r300_pacify.
+ * So we should take great care to *always* call r300_pacify before
+ * *anything* 3D related, and again afterwards. This is what the
+ * call bracket in r300_do_cp_cmdbuf is for.
+ */
+
+/**
+ * Emit the sequence to pacify R300.
+ */
+static __inline__ void r300_pacify(drm_radeon_private_t* dev_priv)
+{
+	RING_LOCALS;
+
+	BEGIN_RING(6);
+	OUT_RING( CP_PACKET0( R300_RB3D_DSTCACHE_CTLSTAT, 0 ) );
+	OUT_RING( 0xa );
+	OUT_RING( CP_PACKET0( 0x4f18, 0 ) );
+	OUT_RING( 0x3 );
+	OUT_RING( CP_PACKET3( RADEON_CP_NOP, 0 ) );
+	OUT_RING( 0x0 );
+	ADVANCE_RING();
+}
+
+
+/**
+ * Called by r300_do_cp_cmdbuf to update the internal buffer age and state.
+ * The actual age emit is done by r300_do_cp_cmdbuf, which is why you must
+ * be careful about how this function is called.
+ */
+static void r300_discard_buffer(drm_device_t * dev, drm_buf_t * buf)
+{
+	drm_radeon_private_t *dev_priv = dev->dev_private;
+	drm_radeon_buf_priv_t *buf_priv = buf->dev_private;
+
+	buf_priv->age = ++dev_priv->sarea_priv->last_dispatch;
+	buf->pending = 1;
+	buf->used = 0;
+}
+
+
+/**
+ * Parses and validates a user-supplied command buffer and emits appropriate
+ * commands on the DMA ring buffer.
+ * Called by the ioctl handler function radeon_cp_cmdbuf.
+ */
+int r300_do_cp_cmdbuf(drm_device_t* dev,
+			  DRMFILE filp,
+		      drm_file_t* filp_priv,
+		      drm_radeon_cmd_buffer_t* cmdbuf)
+{
+	drm_radeon_private_t *dev_priv = dev->dev_private;
+        drm_device_dma_t *dma = dev->dma;
+        drm_buf_t *buf = NULL;
+	int emit_dispatch_age = 0;
+	int ret = 0;
+
+	DRM_DEBUG("\n");
+
+	/* See the comment above r300_emit_begin3d for why this call must be here,
+	 * and what the cleanup gotos are for. */
+	r300_pacify(dev_priv);
+
+	if (cmdbuf->nbox <= R300_SIMULTANEOUS_CLIPRECTS) {
+		ret = r300_emit_cliprects(dev_priv, cmdbuf, 0);
+		if (ret)
+			goto cleanup;
+		}
+
+	while(cmdbuf->bufsz >= sizeof(drm_r300_cmd_header_t)) {
+		int idx;
+		drm_r300_cmd_header_t header;
+
+		header.u = *(unsigned int *)cmdbuf->buf;
+
+		cmdbuf->buf += sizeof(header);
+		cmdbuf->bufsz -= sizeof(header);
+
+		switch(header.header.cmd_type) {
+		case R300_CMD_PACKET0: 
+			DRM_DEBUG("R300_CMD_PACKET0\n");
+			ret = r300_emit_packet0(dev_priv, cmdbuf, header);
+			if (ret) {
+				DRM_ERROR("r300_emit_packet0 failed\n");
+				goto cleanup;
+				}
+			break;
+
+		case R300_CMD_VPU:
+			DRM_DEBUG("R300_CMD_VPU\n");
+			ret = r300_emit_vpu(dev_priv, cmdbuf, header);
+			if (ret) {
+				DRM_ERROR("r300_emit_vpu failed\n");
+				goto cleanup;
+				}
+			break;
+
+		case R300_CMD_PACKET3:
+			DRM_DEBUG("R300_CMD_PACKET3\n");
+			ret = r300_emit_packet3(dev_priv, cmdbuf, header);
+			if (ret) {
+				DRM_ERROR("r300_emit_packet3 failed\n");
+				goto cleanup;
+				}
+			break;
+
+		case R300_CMD_END3D:
+			DRM_DEBUG("R300_CMD_END3D\n");
+			/* TODO: 
+				Ideally userspace driver should not need to issue this call, 
+				i.e. the drm driver should issue it automatically and prevent
+				lockups.
+				
+				In practice, we do not understand why this call is needed and what
+				it does (except for some vague guesses that it has to do with cache
+				coherence) and so the user space driver does it. 
+				
+				Once we are sure which uses prevent lockups the code could be moved
+				into the kernel and the userspace driver will not
+				need to use this command.
+
+				Note that issuing this command does not hurt anything
+				except, possibly, performance */
+			r300_pacify(dev_priv);
+			break;
+
+		case R300_CMD_CP_DELAY:
+			/* simple enough, we can do it here */
+			DRM_DEBUG("R300_CMD_CP_DELAY\n");
+			{
+				int i;
+				RING_LOCALS;
+
+				BEGIN_RING(header.delay.count);
+				for(i=0;i<header.delay.count;i++)
+					OUT_RING(RADEON_CP_PACKET2);
+				ADVANCE_RING();
+			}
+			break;
+
+		case R300_CMD_DMA_DISCARD:
+			DRM_DEBUG("RADEON_CMD_DMA_DISCARD\n");
+            		idx = header.dma.buf_idx;
+            		if (idx < 0 || idx >= dma->buf_count) {
+                		DRM_ERROR("buffer index %d (of %d max)\n",
+                      			idx, dma->buf_count - 1);
+				ret = DRM_ERR(EINVAL);
+                		goto cleanup;
+            			}
+
+	                buf = dma->buflist[idx];
+            		if (buf->filp != filp || buf->pending) {
+                		DRM_ERROR("bad buffer %p %p %d\n",
+                      		buf->filp, filp, buf->pending);
+                		ret = DRM_ERR(EINVAL);
+				goto cleanup;
+            			}
+
+			emit_dispatch_age = 1;
+			r300_discard_buffer(dev, buf);
+            		break;
+
+		case R300_CMD_WAIT:
+			/* simple enough, we can do it here */
+			DRM_DEBUG("R300_CMD_WAIT\n");
+			if(header.wait.flags==0)break; /* nothing to do */
+
+			{
+				RING_LOCALS;
+
+				BEGIN_RING(2);
+				OUT_RING( CP_PACKET0( RADEON_WAIT_UNTIL, 0 ) );
+				OUT_RING( (header.wait.flags & 0xf)<<14 );
+				ADVANCE_RING();
+			}
+			break;
+
+		default:
+			DRM_ERROR("bad cmd_type %i at %p\n",
+			          header.header.cmd_type,
+				  cmdbuf->buf - sizeof(header));
+			ret = DRM_ERR(EINVAL);
+			goto cleanup;
+			}
+	}
+
+	DRM_DEBUG("END\n");
+
+cleanup:
+	r300_pacify(dev_priv);
+
+	/* We emit the vertex buffer age here, outside the pacifier "brackets"
+	 * for two reasons:
+	 *  (1) This may coalesce multiple age emissions into a single one and
+	 *  (2) more importantly, some chips lock up hard when scratch registers
+	 *      are written inside the pacifier bracket.
+	 */
+	if (emit_dispatch_age) {
+		RING_LOCALS;
+
+		/* Emit the vertex buffer age */
+		BEGIN_RING(2);
+		RADEON_DISPATCH_AGE(dev_priv->sarea_priv->last_dispatch);
+		ADVANCE_RING();
+		}
+
+	COMMIT_RING();
+
+	return ret;
+}
+
diff --git a/drivers/char/drm/r300_reg.h b/drivers/char/drm/r300_reg.h
new file mode 100644
index 000000000000..c3e7ca3dbe3d
--- /dev/null
+++ b/drivers/char/drm/r300_reg.h
@@ -0,0 +1,1412 @@
+/**************************************************************************
+
+Copyright (C) 2004-2005 Nicolai Haehnle et al.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+#ifndef _R300_REG_H
+#define _R300_REG_H
+
+#define R300_MC_INIT_MISC_LAT_TIMER	0x180
+#	define R300_MC_MISC__MC_CPR_INIT_LAT_SHIFT	0
+#	define R300_MC_MISC__MC_VF_INIT_LAT_SHIFT	4
+#	define R300_MC_MISC__MC_DISP0R_INIT_LAT_SHIFT	8
+#	define R300_MC_MISC__MC_DISP1R_INIT_LAT_SHIFT	12
+#	define R300_MC_MISC__MC_FIXED_INIT_LAT_SHIFT	16
+#	define R300_MC_MISC__MC_E2R_INIT_LAT_SHIFT	20
+#	define R300_MC_MISC__MC_SAME_PAGE_PRIO_SHIFT	24
+#	define R300_MC_MISC__MC_GLOBW_INIT_LAT_SHIFT	28
+
+
+#define R300_MC_INIT_GFX_LAT_TIMER	0x154
+#	define R300_MC_MISC__MC_G3D0R_INIT_LAT_SHIFT	0
+#	define R300_MC_MISC__MC_G3D1R_INIT_LAT_SHIFT	4
+#	define R300_MC_MISC__MC_G3D2R_INIT_LAT_SHIFT	8
+#	define R300_MC_MISC__MC_G3D3R_INIT_LAT_SHIFT	12
+#	define R300_MC_MISC__MC_TX0R_INIT_LAT_SHIFT	16
+#	define R300_MC_MISC__MC_TX1R_INIT_LAT_SHIFT	20
+#	define R300_MC_MISC__MC_GLOBR_INIT_LAT_SHIFT	24
+#	define R300_MC_MISC__MC_GLOBW_FULL_LAT_SHIFT	28
+
+/*
+This file contains registers and constants for the R300. They have been
+found mostly by examining command buffers captured using glxtest, as well
+as by extrapolating some known registers and constants from the R200.
+
+I am fairly certain that they are correct unless stated otherwise in comments.
+*/
+
+#define R300_SE_VPORT_XSCALE                0x1D98
+#define R300_SE_VPORT_XOFFSET               0x1D9C
+#define R300_SE_VPORT_YSCALE                0x1DA0
+#define R300_SE_VPORT_YOFFSET               0x1DA4
+#define R300_SE_VPORT_ZSCALE                0x1DA8
+#define R300_SE_VPORT_ZOFFSET               0x1DAC
+
+
+/* This register is written directly and also starts data section in many 3d CP_PACKET3's */
+#define R300_VAP_VF_CNTL	0x2084
+
+#	define	R300_VAP_VF_CNTL__PRIM_TYPE__SHIFT                       0
+#	define  R300_VAP_VF_CNTL__PRIM_NONE				 (0<<0)
+#	define  R300_VAP_VF_CNTL__PRIM_POINTS				 (1<<0)
+#	define  R300_VAP_VF_CNTL__PRIM_LINES				 (2<<0)
+#	define  R300_VAP_VF_CNTL__PRIM_LINE_STRIP			 (3<<0)
+#	define  R300_VAP_VF_CNTL__PRIM_TRIANGLES			 (4<<0)
+#	define  R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN			 (5<<0)
+#	define  R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP			 (6<<0)
+#	define  R300_VAP_VF_CNTL__PRIM_LINE_LOOP			 (12<<0)
+#	define  R300_VAP_VF_CNTL__PRIM_QUADS			 	 (13<<0)
+#	define  R300_VAP_VF_CNTL__PRIM_QUAD_STRIP			 (14<<0)
+#	define  R300_VAP_VF_CNTL__PRIM_POLYGON			 	 (15<<0)
+
+#	define	R300_VAP_VF_CNTL__PRIM_WALK__SHIFT                       4
+	/* State based - direct writes to registers trigger vertex generation */
+#	define	R300_VAP_VF_CNTL__PRIM_WALK_STATE_BASED                      (0<<4)
+#	define	R300_VAP_VF_CNTL__PRIM_WALK_INDICES                          (1<<4)
+#	define	R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST                      (2<<4)
+#	define	R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED                  (3<<4)
+
+		/* I don't think I saw these three used.. */
+#	define	R300_VAP_VF_CNTL__COLOR_ORDER__SHIFT                     6
+#	define	R300_VAP_VF_CNTL__TCL_OUTPUT_CTL_ENA__SHIFT              9
+#	define	R300_VAP_VF_CNTL__PROG_STREAM_ENA__SHIFT                 10
+
+		/* index size - when not set the indices are assumed to be 16 bit */
+#	define	R300_VAP_VF_CNTL__INDEX_SIZE_32bit                      (1<<11)
+                /* number of vertices */
+#	define	R300_VAP_VF_CNTL__NUM_VERTICES__SHIFT                    16
+
+/* BEGIN: Wild guesses */
+#define R300_VAP_OUTPUT_VTX_FMT_0           0x2090
+#       define R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT     (1<<0)
+#       define R300_VAP_OUTPUT_VTX_FMT_0__COLOR_PRESENT   (1<<1)
+#       define R300_VAP_OUTPUT_VTX_FMT_0__COLOR_1_PRESENT (1<<2) /* GUESS */
+#       define R300_VAP_OUTPUT_VTX_FMT_0__COLOR_2_PRESENT (1<<3) /* GUESS */
+#       define R300_VAP_OUTPUT_VTX_FMT_0__COLOR_3_PRESENT (1<<4) /* GUESS */
+#       define R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT (1<<16) /* GUESS */
+
+#define R300_VAP_OUTPUT_VTX_FMT_1           0x2094
+#       define R300_VAP_OUTPUT_VTX_FMT_1__TEX_0_COMP_CNT_SHIFT 0
+#       define R300_VAP_OUTPUT_VTX_FMT_1__TEX_1_COMP_CNT_SHIFT 3
+#       define R300_VAP_OUTPUT_VTX_FMT_1__TEX_2_COMP_CNT_SHIFT 6
+#       define R300_VAP_OUTPUT_VTX_FMT_1__TEX_3_COMP_CNT_SHIFT 9
+#       define R300_VAP_OUTPUT_VTX_FMT_1__TEX_4_COMP_CNT_SHIFT 12
+#       define R300_VAP_OUTPUT_VTX_FMT_1__TEX_5_COMP_CNT_SHIFT 15
+#       define R300_VAP_OUTPUT_VTX_FMT_1__TEX_6_COMP_CNT_SHIFT 18
+#       define R300_VAP_OUTPUT_VTX_FMT_1__TEX_7_COMP_CNT_SHIFT 21
+/* END */
+
+#define R300_SE_VTE_CNTL                  0x20b0
+#	define     R300_VPORT_X_SCALE_ENA                0x00000001
+#	define     R300_VPORT_X_OFFSET_ENA               0x00000002
+#	define     R300_VPORT_Y_SCALE_ENA                0x00000004
+#	define     R300_VPORT_Y_OFFSET_ENA               0x00000008
+#	define     R300_VPORT_Z_SCALE_ENA                0x00000010
+#	define     R300_VPORT_Z_OFFSET_ENA               0x00000020
+#	define     R300_VTX_XY_FMT                       0x00000100
+#	define     R300_VTX_Z_FMT                        0x00000200
+#	define     R300_VTX_W0_FMT                       0x00000400
+#	define     R300_VTX_W0_NORMALIZE                 0x00000800
+#	define     R300_VTX_ST_DENORMALIZED              0x00001000
+
+/* BEGIN: Vertex data assembly - lots of uncertainties */
+/* gap */
+/* Where do we get our vertex data?
+//
+// Vertex data either comes either from immediate mode registers or from
+// vertex arrays.
+// There appears to be no mixed mode (though we can force the pitch of
+// vertex arrays to 0, effectively reusing the same element over and over
+// again).
+//
+// Immediate mode is controlled by the INPUT_CNTL registers. I am not sure
+// if these registers influence vertex array processing.
+//
+// Vertex arrays are controlled via the 3D_LOAD_VBPNTR packet3.
+//
+// In both cases, vertex attributes are then passed through INPUT_ROUTE.
+
+// Beginning with INPUT_ROUTE_0_0 is a list of WORDs that route vertex data
+// into the vertex processor's input registers.
+// The first word routes the first input, the second word the second, etc.
+// The corresponding input is routed into the register with the given index.
+// The list is ended by a word with INPUT_ROUTE_END set.
+//
+// Always set COMPONENTS_4 in immediate mode. */
+
+#define R300_VAP_INPUT_ROUTE_0_0            0x2150
+#       define R300_INPUT_ROUTE_COMPONENTS_1     (0 << 0)
+#       define R300_INPUT_ROUTE_COMPONENTS_2     (1 << 0)
+#       define R300_INPUT_ROUTE_COMPONENTS_3     (2 << 0)
+#       define R300_INPUT_ROUTE_COMPONENTS_4     (3 << 0)
+#       define R300_INPUT_ROUTE_COMPONENTS_RGBA  (4 << 0) /* GUESS */
+#       define R300_VAP_INPUT_ROUTE_IDX_SHIFT    8
+#       define R300_VAP_INPUT_ROUTE_IDX_MASK     (31 << 8) /* GUESS */
+#       define R300_VAP_INPUT_ROUTE_END          (1 << 13)
+#       define R300_INPUT_ROUTE_IMMEDIATE_MODE   (0 << 14) /* GUESS */
+#       define R300_INPUT_ROUTE_FLOAT            (1 << 14) /* GUESS */
+#       define R300_INPUT_ROUTE_UNSIGNED_BYTE    (2 << 14) /* GUESS */
+#       define R300_INPUT_ROUTE_FLOAT_COLOR      (3 << 14) /* GUESS */
+#define R300_VAP_INPUT_ROUTE_0_1            0x2154
+#define R300_VAP_INPUT_ROUTE_0_2            0x2158
+#define R300_VAP_INPUT_ROUTE_0_3            0x215C
+#define R300_VAP_INPUT_ROUTE_0_4            0x2160
+#define R300_VAP_INPUT_ROUTE_0_5            0x2164
+#define R300_VAP_INPUT_ROUTE_0_6            0x2168
+#define R300_VAP_INPUT_ROUTE_0_7            0x216C
+
+/* gap */
+/* Notes:
+//  - always set up to produce at least two attributes:
+//    if vertex program uses only position, fglrx will set normal, too
+//  - INPUT_CNTL_0_COLOR and INPUT_CNTL_COLOR bits are always equal */
+#define R300_VAP_INPUT_CNTL_0               0x2180
+#       define R300_INPUT_CNTL_0_COLOR           0x00000001
+#define R300_VAP_INPUT_CNTL_1               0x2184
+#       define R300_INPUT_CNTL_POS               0x00000001
+#       define R300_INPUT_CNTL_NORMAL            0x00000002
+#       define R300_INPUT_CNTL_COLOR             0x00000004
+#       define R300_INPUT_CNTL_TC0               0x00000400
+#       define R300_INPUT_CNTL_TC1               0x00000800
+#       define R300_INPUT_CNTL_TC2               0x00001000 /* GUESS */
+#       define R300_INPUT_CNTL_TC3               0x00002000 /* GUESS */
+#       define R300_INPUT_CNTL_TC4               0x00004000 /* GUESS */
+#       define R300_INPUT_CNTL_TC5               0x00008000 /* GUESS */
+#       define R300_INPUT_CNTL_TC6               0x00010000 /* GUESS */
+#       define R300_INPUT_CNTL_TC7               0x00020000 /* GUESS */
+
+/* gap */
+/* Words parallel to INPUT_ROUTE_0; All words that are active in INPUT_ROUTE_0
+// are set to a swizzling bit pattern, other words are 0.
+//
+// In immediate mode, the pattern is always set to xyzw. In vertex array
+// mode, the swizzling pattern is e.g. used to set zw components in texture
+// coordinates with only tweo components. */
+#define R300_VAP_INPUT_ROUTE_1_0            0x21E0
+#       define R300_INPUT_ROUTE_SELECT_X    0
+#       define R300_INPUT_ROUTE_SELECT_Y    1
+#       define R300_INPUT_ROUTE_SELECT_Z    2
+#       define R300_INPUT_ROUTE_SELECT_W    3
+#       define R300_INPUT_ROUTE_SELECT_ZERO 4
+#       define R300_INPUT_ROUTE_SELECT_ONE  5
+#       define R300_INPUT_ROUTE_SELECT_MASK 7
+#       define R300_INPUT_ROUTE_X_SHIFT          0
+#       define R300_INPUT_ROUTE_Y_SHIFT          3
+#       define R300_INPUT_ROUTE_Z_SHIFT          6
+#       define R300_INPUT_ROUTE_W_SHIFT          9
+#       define R300_INPUT_ROUTE_ENABLE           (15 << 12)
+#define R300_VAP_INPUT_ROUTE_1_1            0x21E4
+#define R300_VAP_INPUT_ROUTE_1_2            0x21E8
+#define R300_VAP_INPUT_ROUTE_1_3            0x21EC
+#define R300_VAP_INPUT_ROUTE_1_4            0x21F0
+#define R300_VAP_INPUT_ROUTE_1_5            0x21F4
+#define R300_VAP_INPUT_ROUTE_1_6            0x21F8
+#define R300_VAP_INPUT_ROUTE_1_7            0x21FC
+
+/* END */
+
+/* gap */
+/* BEGIN: Upload vertex program and data
+// The programmable vertex shader unit has a memory bank of unknown size
+// that can be written to in 16 byte units by writing the address into
+// UPLOAD_ADDRESS, followed by data in UPLOAD_DATA (multiples of 4 DWORDs).
+//
+// Pointers into the memory bank are always in multiples of 16 bytes.
+//
+// The memory bank is divided into areas with fixed meaning.
+//
+// Starting at address UPLOAD_PROGRAM: Vertex program instructions.
+// Native limits reported by drivers from ATI suggest size 256 (i.e. 4KB),
+// whereas the difference between known addresses suggests size 512.
+//
+// Starting at address UPLOAD_PARAMETERS: Vertex program parameters.
+// Native reported limits and the VPI layout suggest size 256, whereas
+// difference between known addresses suggests size 512.
+//
+// At address UPLOAD_POINTSIZE is a vector (0, 0, ps, 0), where ps is the
+// floating point pointsize. The exact purpose of this state is uncertain,
+// as there is also the R300_RE_POINTSIZE register.
+//
+// Multiple vertex programs and parameter sets can be loaded at once,
+// which could explain the size discrepancy. */
+#define R300_VAP_PVS_UPLOAD_ADDRESS         0x2200
+#       define R300_PVS_UPLOAD_PROGRAM           0x00000000
+#       define R300_PVS_UPLOAD_PARAMETERS        0x00000200
+#       define R300_PVS_UPLOAD_POINTSIZE         0x00000406
+/* gap */
+#define R300_VAP_PVS_UPLOAD_DATA            0x2208
+/* END */
+
+/* gap */
+/* I do not know the purpose of this register. However, I do know that
+// it is set to 221C_CLEAR for clear operations and to 221C_NORMAL
+// for normal rendering. */
+#define R300_VAP_UNKNOWN_221C               0x221C
+#       define R300_221C_NORMAL                  0x00000000
+#       define R300_221C_CLEAR                   0x0001C000
+
+/* gap */
+/* Sometimes, END_OF_PKT and 0x2284=0 are the only commands sent between
+// rendering commands and overwriting vertex program parameters.
+// Therefore, I suspect writing zero to 0x2284 synchronizes the engine and
+// avoids bugs caused by still running shaders reading bad data from memory. */
+#define R300_VAP_PVS_WAITIDLE               0x2284 /* GUESS */
+
+/* Absolutely no clue what this register is about. */
+#define R300_VAP_UNKNOWN_2288               0x2288
+#       define R300_2288_R300                    0x00750000 /* -- nh */
+#       define R300_2288_RV350                   0x0000FFFF /* -- Vladimir */
+
+/* gap */
+/* Addresses are relative to the vertex program instruction area of the
+// memory bank. PROGRAM_END points to the last instruction of the active
+// program
+//
+// The meaning of the two UNKNOWN fields is obviously not known. However,
+// experiments so far have shown that both *must* point to an instruction
+// inside the vertex program, otherwise the GPU locks up.
+// fglrx usually sets CNTL_3_UNKNOWN to the end of the program and
+// CNTL_1_UNKNOWN points to instruction where last write to position takes place. 
+// Most likely this is used to ignore rest of the program in cases where group of verts arent visible.
+// For some reason this "section" is sometimes accepted other instruction that have
+// no relationship with position calculations. 
+*/
+#define R300_VAP_PVS_CNTL_1                 0x22D0
+#       define R300_PVS_CNTL_1_PROGRAM_START_SHIFT   0
+#       define R300_PVS_CNTL_1_POS_END_SHIFT         10
+#       define R300_PVS_CNTL_1_PROGRAM_END_SHIFT     20
+/* Addresses are relative the the vertex program parameters area. */
+#define R300_VAP_PVS_CNTL_2                 0x22D4
+#       define R300_PVS_CNTL_2_PARAM_OFFSET_SHIFT 0
+#       define R300_PVS_CNTL_2_PARAM_COUNT_SHIFT  16
+#define R300_VAP_PVS_CNTL_3	           0x22D8
+#       define R300_PVS_CNTL_3_PROGRAM_UNKNOWN_SHIFT 10
+#       define R300_PVS_CNTL_3_PROGRAM_UNKNOWN2_SHIFT 0
+
+/* The entire range from 0x2300 to 0x2AC inclusive seems to be used for
+// immediate vertices */
+#define R300_VAP_VTX_COLOR_R                0x2464
+#define R300_VAP_VTX_COLOR_G                0x2468
+#define R300_VAP_VTX_COLOR_B                0x246C
+#define R300_VAP_VTX_POS_0_X_1              0x2490 /* used for glVertex2*() */
+#define R300_VAP_VTX_POS_0_Y_1              0x2494
+#define R300_VAP_VTX_COLOR_PKD              0x249C /* RGBA */
+#define R300_VAP_VTX_POS_0_X_2              0x24A0 /* used for glVertex3*() */
+#define R300_VAP_VTX_POS_0_Y_2              0x24A4
+#define R300_VAP_VTX_POS_0_Z_2              0x24A8
+#define R300_VAP_VTX_END_OF_PKT             0x24AC /* write 0 to indicate end of packet? */
+
+/* gap */
+
+/* These are values from r300_reg/r300_reg.h - they are known to be correct
+   and are here so we can use one register file instead of several
+   - Vladimir */
+#define R300_GB_VAP_RASTER_VTX_FMT_0	0x4000
+#	define R300_GB_VAP_RASTER_VTX_FMT_0__POS_PRESENT	(1<<0)
+#	define R300_GB_VAP_RASTER_VTX_FMT_0__COLOR_0_PRESENT	(1<<1)
+#	define R300_GB_VAP_RASTER_VTX_FMT_0__COLOR_1_PRESENT	(1<<2)
+#	define R300_GB_VAP_RASTER_VTX_FMT_0__COLOR_2_PRESENT	(1<<3)
+#	define R300_GB_VAP_RASTER_VTX_FMT_0__COLOR_3_PRESENT	(1<<4)
+#	define R300_GB_VAP_RASTER_VTX_FMT_0__COLOR_SPACE	(0xf<<5)
+#	define R300_GB_VAP_RASTER_VTX_FMT_0__PT_SIZE_PRESENT	(0x1<<16)
+
+#define R300_GB_VAP_RASTER_VTX_FMT_1	0x4004
+	/* each of the following is 3 bits wide, specifies number
+	   of components */
+#	define R300_GB_VAP_RASTER_VTX_FMT_1__TEX_0_COMP_CNT_SHIFT	0
+#	define R300_GB_VAP_RASTER_VTX_FMT_1__TEX_1_COMP_CNT_SHIFT	3
+#	define R300_GB_VAP_RASTER_VTX_FMT_1__TEX_2_COMP_CNT_SHIFT	6
+#	define R300_GB_VAP_RASTER_VTX_FMT_1__TEX_3_COMP_CNT_SHIFT	9
+#	define R300_GB_VAP_RASTER_VTX_FMT_1__TEX_4_COMP_CNT_SHIFT	12
+#	define R300_GB_VAP_RASTER_VTX_FMT_1__TEX_5_COMP_CNT_SHIFT	15
+#	define R300_GB_VAP_RASTER_VTX_FMT_1__TEX_6_COMP_CNT_SHIFT	18
+#	define R300_GB_VAP_RASTER_VTX_FMT_1__TEX_7_COMP_CNT_SHIFT	21
+
+/* UNK30 seems to enables point to quad transformation on textures
+   (or something closely related to that).
+   This bit is rather fatal at the time being due to lackings at pixel shader side */
+#define R300_GB_ENABLE	0x4008
+#	define R300_GB_POINT_STUFF_ENABLE	(1<<0)
+#	define R300_GB_LINE_STUFF_ENABLE	(1<<1)
+#	define R300_GB_TRIANGLE_STUFF_ENABLE	(1<<2)
+#	define R300_GB_STENCIL_AUTO_ENABLE	(1<<4)
+#	define R300_GB_UNK30			(1<<30)
+	/* each of the following is 2 bits wide */
+#define R300_GB_TEX_REPLICATE	0
+#define R300_GB_TEX_ST		1
+#define R300_GB_TEX_STR		2
+#	define R300_GB_TEX0_SOURCE_SHIFT	16
+#	define R300_GB_TEX1_SOURCE_SHIFT	18
+#	define R300_GB_TEX2_SOURCE_SHIFT	20
+#	define R300_GB_TEX3_SOURCE_SHIFT	22
+#	define R300_GB_TEX4_SOURCE_SHIFT	24
+#	define R300_GB_TEX5_SOURCE_SHIFT	26
+#	define R300_GB_TEX6_SOURCE_SHIFT	28
+#	define R300_GB_TEX7_SOURCE_SHIFT	30
+
+/* MSPOS - positions for multisample antialiasing (?) */
+#define R300_GB_MSPOS0	0x4010
+	/* shifts - each of the fields is 4 bits */
+#	define R300_GB_MSPOS0__MS_X0_SHIFT	0
+#	define R300_GB_MSPOS0__MS_Y0_SHIFT	4
+#	define R300_GB_MSPOS0__MS_X1_SHIFT	8
+#	define R300_GB_MSPOS0__MS_Y1_SHIFT	12
+#	define R300_GB_MSPOS0__MS_X2_SHIFT	16
+#	define R300_GB_MSPOS0__MS_Y2_SHIFT	20
+#	define R300_GB_MSPOS0__MSBD0_Y		24
+#	define R300_GB_MSPOS0__MSBD0_X		28
+
+#define R300_GB_MSPOS1	0x4014
+#	define R300_GB_MSPOS1__MS_X3_SHIFT	0
+#	define R300_GB_MSPOS1__MS_Y3_SHIFT	4
+#	define R300_GB_MSPOS1__MS_X4_SHIFT	8
+#	define R300_GB_MSPOS1__MS_Y4_SHIFT	12
+#	define R300_GB_MSPOS1__MS_X5_SHIFT	16
+#	define R300_GB_MSPOS1__MS_Y5_SHIFT	20
+#	define R300_GB_MSPOS1__MSBD1		24
+
+
+#define R300_GB_TILE_CONFIG	0x4018
+#	define R300_GB_TILE_ENABLE	(1<<0)
+#	define R300_GB_TILE_PIPE_COUNT_RV300	0
+#	define R300_GB_TILE_PIPE_COUNT_R300	(3<<1)
+#	define R300_GB_TILE_PIPE_COUNT_R420	(7<<1)
+#	define R300_GB_TILE_SIZE_8		0
+#	define R300_GB_TILE_SIZE_16		(1<<4)
+#	define R300_GB_TILE_SIZE_32		(2<<4)
+#	define R300_GB_SUPER_SIZE_1		(0<<6)
+#	define R300_GB_SUPER_SIZE_2		(1<<6)
+#	define R300_GB_SUPER_SIZE_4		(2<<6)
+#	define R300_GB_SUPER_SIZE_8		(3<<6)
+#	define R300_GB_SUPER_SIZE_16		(4<<6)
+#	define R300_GB_SUPER_SIZE_32		(5<<6)
+#	define R300_GB_SUPER_SIZE_64		(6<<6)
+#	define R300_GB_SUPER_SIZE_128		(7<<6)
+#	define R300_GB_SUPER_X_SHIFT		9	/* 3 bits wide */
+#	define R300_GB_SUPER_Y_SHIFT		12	/* 3 bits wide */
+#	define R300_GB_SUPER_TILE_A		0
+#	define R300_GB_SUPER_TILE_B		(1<<15)
+#	define R300_GB_SUBPIXEL_1_12		0
+#	define R300_GB_SUBPIXEL_1_16		(1<<16)
+
+#define R300_GB_FIFO_SIZE	0x4024
+	/* each of the following is 2 bits wide */
+#define R300_GB_FIFO_SIZE_32	0
+#define R300_GB_FIFO_SIZE_64	1
+#define R300_GB_FIFO_SIZE_128	2
+#define R300_GB_FIFO_SIZE_256	3
+#	define R300_SC_IFIFO_SIZE_SHIFT	0
+#	define R300_SC_TZFIFO_SIZE_SHIFT	2
+#	define R300_SC_BFIFO_SIZE_SHIFT	4
+
+#	define R300_US_OFIFO_SIZE_SHIFT	12
+#	define R300_US_WFIFO_SIZE_SHIFT	14
+	/* the following use the same constants as above, but meaning is
+	   is times 2 (i.e. instead of 32 words it means 64 */
+#	define R300_RS_TFIFO_SIZE_SHIFT	6
+#	define R300_RS_CFIFO_SIZE_SHIFT	8
+#	define R300_US_RAM_SIZE_SHIFT		10
+	/* watermarks, 3 bits wide */
+#	define R300_RS_HIGHWATER_COL_SHIFT	16
+#	define R300_RS_HIGHWATER_TEX_SHIFT	19
+#	define R300_OFIFO_HIGHWATER_SHIFT	22	/* two bits only */
+#	define R300_CUBE_FIFO_HIGHWATER_COL_SHIFT	24
+
+#define R300_GB_SELECT	0x401C
+#	define R300_GB_FOG_SELECT_C0A		0
+#	define R300_GB_FOG_SELECT_C1A		1
+#	define R300_GB_FOG_SELECT_C2A		2
+#	define R300_GB_FOG_SELECT_C3A		3
+#	define R300_GB_FOG_SELECT_1_1_W	4
+#	define R300_GB_FOG_SELECT_Z		5
+#	define R300_GB_DEPTH_SELECT_Z		0
+#	define R300_GB_DEPTH_SELECT_1_1_W	(1<<3)
+#	define R300_GB_W_SELECT_1_W		0
+#	define R300_GB_W_SELECT_1		(1<<4)
+
+#define R300_GB_AA_CONFIG		0x4020
+#	define R300_AA_ENABLE			0x01
+#	define R300_AA_SUBSAMPLES_2		0
+#	define R300_AA_SUBSAMPLES_3		(1<<1)
+#	define R300_AA_SUBSAMPLES_4		(2<<1)
+#	define R300_AA_SUBSAMPLES_6		(3<<1)
+
+/* END */
+
+/* gap */
+/* The upper enable bits are guessed, based on fglrx reported limits. */
+#define R300_TX_ENABLE                      0x4104
+#       define R300_TX_ENABLE_0                  (1 << 0)
+#       define R300_TX_ENABLE_1                  (1 << 1)
+#       define R300_TX_ENABLE_2                  (1 << 2)
+#       define R300_TX_ENABLE_3                  (1 << 3)
+#       define R300_TX_ENABLE_4                  (1 << 4)
+#       define R300_TX_ENABLE_5                  (1 << 5)
+#       define R300_TX_ENABLE_6                  (1 << 6)
+#       define R300_TX_ENABLE_7                  (1 << 7)
+#       define R300_TX_ENABLE_8                  (1 << 8)
+#       define R300_TX_ENABLE_9                  (1 << 9)
+#       define R300_TX_ENABLE_10                 (1 << 10)
+#       define R300_TX_ENABLE_11                 (1 << 11)
+#       define R300_TX_ENABLE_12                 (1 << 12)
+#       define R300_TX_ENABLE_13                 (1 << 13)
+#       define R300_TX_ENABLE_14                 (1 << 14)
+#       define R300_TX_ENABLE_15                 (1 << 15)
+
+/* The pointsize is given in multiples of 6. The pointsize can be
+// enormous: Clear() renders a single point that fills the entire
+// framebuffer. */
+#define R300_RE_POINTSIZE                   0x421C
+#       define R300_POINTSIZE_Y_SHIFT            0
+#       define R300_POINTSIZE_Y_MASK             (0xFFFF << 0) /* GUESS */
+#       define R300_POINTSIZE_X_SHIFT            16
+#       define R300_POINTSIZE_X_MASK             (0xFFFF << 16) /* GUESS */
+#       define R300_POINTSIZE_MAX             (R300_POINTSIZE_Y_MASK / 6)
+
+/* The line width is given in multiples of 6.
+   In default mode lines are classified as vertical lines.
+   HO: horizontal
+   VE: vertical or horizontal
+   HO & VE: no classification
+*/
+#define R300_RE_LINE_CNT                      0x4234
+#       define R300_LINESIZE_SHIFT            0
+#       define R300_LINESIZE_MASK             (0xFFFF << 0) /* GUESS */
+#       define R300_LINESIZE_MAX             (R300_LINESIZE_MASK / 6)
+#       define R300_LINE_CNT_HO               (1 << 16)
+#       define R300_LINE_CNT_VE               (1 << 17)
+
+/* Some sort of scale or clamp value for texcoordless textures. */
+#define R300_RE_UNK4238                       0x4238
+
+#define R300_RE_SHADE_MODEL                   0x4278
+#	define R300_RE_SHADE_MODEL_SMOOTH     0x3aaaa
+#	define R300_RE_SHADE_MODEL_FLAT       0x39595
+
+/* Dangerous */
+#define R300_RE_POLYGON_MODE                  0x4288
+#	define R300_PM_ENABLED                (1 << 0)
+#	define R300_PM_FRONT_POINT            (0 << 0)
+#	define R300_PM_BACK_POINT             (0 << 0)
+#	define R300_PM_FRONT_LINE             (1 << 4)
+#	define R300_PM_FRONT_FILL             (1 << 5)
+#	define R300_PM_BACK_LINE              (1 << 7)
+#	define R300_PM_BACK_FILL              (1 << 8)
+
+/* Not sure why there are duplicate of factor and constant values. 
+   My best guess so far is that there are seperate zbiases for test and write. 
+   Ordering might be wrong.
+   Some of the tests indicate that fgl has a fallback implementation of zbias
+   via pixel shaders. */
+#define R300_RE_ZBIAS_T_FACTOR                0x42A4
+#define R300_RE_ZBIAS_T_CONSTANT              0x42A8
+#define R300_RE_ZBIAS_W_FACTOR                0x42AC
+#define R300_RE_ZBIAS_W_CONSTANT              0x42B0
+
+/* This register needs to be set to (1<<1) for RV350 to correctly
+   perform depth test (see --vb-triangles in r300_demo)
+   Don't know about other chips. - Vladimir
+   This is set to 3 when GL_POLYGON_OFFSET_FILL is on.
+   My guess is that there are two bits for each zbias primitive (FILL, LINE, POINT).
+   One to enable depth test and one for depth write.
+   Yet this doesnt explain why depth writes work ...
+    */
+#define R300_RE_OCCLUSION_CNTL		    0x42B4
+#	define R300_OCCLUSION_ON		(1<<1)
+
+#define R300_RE_CULL_CNTL                   0x42B8
+#       define R300_CULL_FRONT                   (1 << 0)
+#       define R300_CULL_BACK                    (1 << 1)
+#       define R300_FRONT_FACE_CCW               (0 << 2)
+#       define R300_FRONT_FACE_CW                (1 << 2)
+
+
+/* BEGIN: Rasterization / Interpolators - many guesses
+// 0_UNKNOWN_18 has always been set except for clear operations.
+// TC_CNT is the number of incoming texture coordinate sets (i.e. it depends
+// on the vertex program, *not* the fragment program) */
+#define R300_RS_CNTL_0                      0x4300
+#       define R300_RS_CNTL_TC_CNT_SHIFT         2
+#       define R300_RS_CNTL_TC_CNT_MASK          (7 << 2)
+#		define R300_RS_CNTL_CI_CNT_SHIFT         7 /* number of color interpolators used */
+#       define R300_RS_CNTL_0_UNKNOWN_18         (1 << 18)
+/* Guess: RS_CNTL_1 holds the index of the highest used RS_ROUTE_n register. */
+#define R300_RS_CNTL_1                      0x4304
+
+/* gap */
+/* Only used for texture coordinates.
+// Use the source field to route texture coordinate input from the vertex program
+// to the desired interpolator. Note that the source field is relative to the
+// outputs the vertex program *actually* writes. If a vertex program only writes
+// texcoord[1], this will be source index 0.
+// Set INTERP_USED on all interpolators that produce data used by the
+// fragment program. INTERP_USED looks like a swizzling mask, but
+// I haven't seen it used that way.
+//
+// Note: The _UNKNOWN constants are always set in their respective register.
+// I don't know if this is necessary. */
+#define R300_RS_INTERP_0                    0x4310
+#define R300_RS_INTERP_1                    0x4314
+#       define R300_RS_INTERP_1_UNKNOWN          0x40
+#define R300_RS_INTERP_2                    0x4318
+#       define R300_RS_INTERP_2_UNKNOWN          0x80
+#define R300_RS_INTERP_3                    0x431C
+#       define R300_RS_INTERP_3_UNKNOWN          0xC0
+#define R300_RS_INTERP_4                    0x4320
+#define R300_RS_INTERP_5                    0x4324
+#define R300_RS_INTERP_6                    0x4328
+#define R300_RS_INTERP_7                    0x432C
+#       define R300_RS_INTERP_SRC_SHIFT          2
+#       define R300_RS_INTERP_SRC_MASK           (7 << 2)
+#       define R300_RS_INTERP_USED               0x00D10000
+
+/* These DWORDs control how vertex data is routed into fragment program
+// registers, after interpolators. */
+#define R300_RS_ROUTE_0                     0x4330
+#define R300_RS_ROUTE_1                     0x4334
+#define R300_RS_ROUTE_2                     0x4338
+#define R300_RS_ROUTE_3                     0x433C /* GUESS */
+#define R300_RS_ROUTE_4                     0x4340 /* GUESS */
+#define R300_RS_ROUTE_5                     0x4344 /* GUESS */
+#define R300_RS_ROUTE_6                     0x4348 /* GUESS */
+#define R300_RS_ROUTE_7                     0x434C /* GUESS */
+#       define R300_RS_ROUTE_SOURCE_INTERP_0     0
+#       define R300_RS_ROUTE_SOURCE_INTERP_1     1
+#       define R300_RS_ROUTE_SOURCE_INTERP_2     2
+#       define R300_RS_ROUTE_SOURCE_INTERP_3     3
+#       define R300_RS_ROUTE_SOURCE_INTERP_4     4
+#       define R300_RS_ROUTE_SOURCE_INTERP_5     5 /* GUESS */
+#       define R300_RS_ROUTE_SOURCE_INTERP_6     6 /* GUESS */
+#       define R300_RS_ROUTE_SOURCE_INTERP_7     7 /* GUESS */
+#       define R300_RS_ROUTE_ENABLE              (1 << 3) /* GUESS */
+#       define R300_RS_ROUTE_DEST_SHIFT          6
+#       define R300_RS_ROUTE_DEST_MASK           (31 << 6) /* GUESS */
+
+/* Special handling for color: When the fragment program uses color,
+// the ROUTE_0_COLOR bit is set and ROUTE_0_COLOR_DEST contains the
+// color register index. */
+#       define R300_RS_ROUTE_0_COLOR             (1 << 14)
+#       define R300_RS_ROUTE_0_COLOR_DEST_SHIFT  17
+#       define R300_RS_ROUTE_0_COLOR_DEST_MASK   (31 << 17) /* GUESS */
+/* As above, but for secondary color */
+#		define R300_RS_ROUTE_1_COLOR1            (1 << 14)
+#		define R300_RS_ROUTE_1_COLOR1_DEST_SHIFT 17
+#		define R300_RS_ROUTE_1_COLOR1_DEST_MASK  (31 << 17)
+#		define R300_RS_ROUTE_1_UNKNOWN11         (1 << 11)
+/* END */
+
+/* BEGIN: Scissors and cliprects
+// There are four clipping rectangles. Their corner coordinates are inclusive.
+// Every pixel is assigned a number from 0 and 15 by setting bits 0-3 depending
+// on whether the pixel is inside cliprects 0-3, respectively. For example,
+// if a pixel is inside cliprects 0 and 1, but outside 2 and 3, it is assigned
+// the number 3 (binary 0011).
+// Iff the bit corresponding to the pixel's number in RE_CLIPRECT_CNTL is set,
+// the pixel is rasterized.
+//
+// In addition to this, there is a scissors rectangle. Only pixels inside the
+// scissors rectangle are drawn. (coordinates are inclusive)
+//
+// For some reason, the top-left corner of the framebuffer is at (1440, 1440)
+// for the purpose of clipping and scissors. */
+#define R300_RE_CLIPRECT_TL_0               0x43B0
+#define R300_RE_CLIPRECT_BR_0               0x43B4
+#define R300_RE_CLIPRECT_TL_1               0x43B8
+#define R300_RE_CLIPRECT_BR_1               0x43BC
+#define R300_RE_CLIPRECT_TL_2               0x43C0
+#define R300_RE_CLIPRECT_BR_2               0x43C4
+#define R300_RE_CLIPRECT_TL_3               0x43C8
+#define R300_RE_CLIPRECT_BR_3               0x43CC
+#       define R300_CLIPRECT_OFFSET              1440
+#       define R300_CLIPRECT_MASK                0x1FFF
+#       define R300_CLIPRECT_X_SHIFT             0
+#       define R300_CLIPRECT_X_MASK              (0x1FFF << 0)
+#       define R300_CLIPRECT_Y_SHIFT             13
+#       define R300_CLIPRECT_Y_MASK              (0x1FFF << 13)
+#define R300_RE_CLIPRECT_CNTL               0x43D0
+#       define R300_CLIP_OUT                     (1 << 0)
+#       define R300_CLIP_0                       (1 << 1)
+#       define R300_CLIP_1                       (1 << 2)
+#       define R300_CLIP_10                      (1 << 3)
+#       define R300_CLIP_2                       (1 << 4)
+#       define R300_CLIP_20                      (1 << 5)
+#       define R300_CLIP_21                      (1 << 6)
+#       define R300_CLIP_210                     (1 << 7)
+#       define R300_CLIP_3                       (1 << 8)
+#       define R300_CLIP_30                      (1 << 9)
+#       define R300_CLIP_31                      (1 << 10)
+#       define R300_CLIP_310                     (1 << 11)
+#       define R300_CLIP_32                      (1 << 12)
+#       define R300_CLIP_320                     (1 << 13)
+#       define R300_CLIP_321                     (1 << 14)
+#       define R300_CLIP_3210                    (1 << 15)
+
+/* gap */
+#define R300_RE_SCISSORS_TL                 0x43E0
+#define R300_RE_SCISSORS_BR                 0x43E4
+#       define R300_SCISSORS_OFFSET              1440
+#       define R300_SCISSORS_X_SHIFT             0
+#       define R300_SCISSORS_X_MASK              (0x1FFF << 0)
+#       define R300_SCISSORS_Y_SHIFT             13
+#       define R300_SCISSORS_Y_MASK              (0x1FFF << 13)
+/* END */
+
+/* BEGIN: Texture specification
+// The texture specification dwords are grouped by meaning and not by texture unit.
+// This means that e.g. the offset for texture image unit N is found in register
+// TX_OFFSET_0 + (4*N) */
+#define R300_TX_FILTER_0                    0x4400
+#       define R300_TX_REPEAT                    0
+#       define R300_TX_MIRRORED                  1
+#       define R300_TX_CLAMP                     4
+#       define R300_TX_CLAMP_TO_EDGE             2
+#       define R300_TX_CLAMP_TO_BORDER           6
+#       define R300_TX_WRAP_S_SHIFT              0
+#       define R300_TX_WRAP_S_MASK               (7 << 0)
+#       define R300_TX_WRAP_T_SHIFT              3
+#       define R300_TX_WRAP_T_MASK               (7 << 3)
+#       define R300_TX_WRAP_Q_SHIFT              6
+#       define R300_TX_WRAP_Q_MASK               (7 << 6)
+#       define R300_TX_MAG_FILTER_NEAREST        (1 << 9)
+#       define R300_TX_MAG_FILTER_LINEAR         (2 << 9)
+#       define R300_TX_MAG_FILTER_MASK           (3 << 9)
+#       define R300_TX_MIN_FILTER_NEAREST        (1 << 11)
+#       define R300_TX_MIN_FILTER_LINEAR         (2 << 11)
+#	define R300_TX_MIN_FILTER_NEAREST_MIP_NEAREST       (5  <<  11)
+#	define R300_TX_MIN_FILTER_NEAREST_MIP_LINEAR        (9  <<  11)
+#	define R300_TX_MIN_FILTER_LINEAR_MIP_NEAREST        (6  <<  11)
+#	define R300_TX_MIN_FILTER_LINEAR_MIP_LINEAR         (10 <<  11)
+
+/* NOTE: NEAREST doesnt seem to exist.
+   Im not seting MAG_FILTER_MASK and (3 << 11) on for all
+   anisotropy modes because that would void selected mag filter */
+#	define R300_TX_MIN_FILTER_ANISO_NEAREST             ((0 << 13) /*|R300_TX_MAG_FILTER_MASK|(3<<11)*/)
+#	define R300_TX_MIN_FILTER_ANISO_LINEAR              ((0 << 13) /*|R300_TX_MAG_FILTER_MASK|(3<<11)*/)
+#	define R300_TX_MIN_FILTER_ANISO_NEAREST_MIP_NEAREST ((1 << 13) /*|R300_TX_MAG_FILTER_MASK|(3<<11)*/)
+#	define R300_TX_MIN_FILTER_ANISO_NEAREST_MIP_LINEAR  ((2 << 13) /*|R300_TX_MAG_FILTER_MASK|(3<<11)*/)
+#       define R300_TX_MIN_FILTER_MASK           ( (15 << 11) | (3 << 13) )
+#	define R300_TX_MAX_ANISO_1_TO_1  (0 << 21)
+#	define R300_TX_MAX_ANISO_2_TO_1  (2 << 21)
+#	define R300_TX_MAX_ANISO_4_TO_1  (4 << 21)
+#	define R300_TX_MAX_ANISO_8_TO_1  (6 << 21)
+#	define R300_TX_MAX_ANISO_16_TO_1 (8 << 21)
+#	define R300_TX_MAX_ANISO_MASK    (14 << 21)
+
+#define R300_TX_UNK1_0                      0x4440
+#	define R300_LOD_BIAS_MASK	    0x1fff
+
+#define R300_TX_SIZE_0                      0x4480
+#       define R300_TX_WIDTHMASK_SHIFT           0
+#       define R300_TX_WIDTHMASK_MASK            (2047 << 0)
+#       define R300_TX_HEIGHTMASK_SHIFT          11
+#       define R300_TX_HEIGHTMASK_MASK           (2047 << 11)
+#       define R300_TX_UNK23                     (1 << 23)
+#       define R300_TX_SIZE_SHIFT                26 /* largest of width, height */
+#       define R300_TX_SIZE_MASK                 (15 << 26)
+#define R300_TX_FORMAT_0                    0x44C0
+	/* The interpretation of the format word by Wladimir van der Laan */
+	/* The X, Y, Z and W refer to the layout of the components.
+	   They are given meanings as R, G, B and Alpha by the swizzle
+	   specification */
+#	define R300_TX_FORMAT_X8		    0x0
+#	define R300_TX_FORMAT_X16		    0x1
+#	define R300_TX_FORMAT_Y4X4		    0x2
+#	define R300_TX_FORMAT_Y8X8		    0x3
+#	define R300_TX_FORMAT_Y16X16		    0x4
+#	define R300_TX_FORMAT_Z3Y3X2		    0x5
+#	define R300_TX_FORMAT_Z5Y6X5		    0x6
+#	define R300_TX_FORMAT_Z6Y5X5		    0x7
+#	define R300_TX_FORMAT_Z11Y11X10		    0x8
+#	define R300_TX_FORMAT_Z10Y11X11		    0x9
+#	define R300_TX_FORMAT_W4Z4Y4X4		    0xA
+#	define R300_TX_FORMAT_W1Z5Y5X5		    0xB
+#	define R300_TX_FORMAT_W8Z8Y8X8		    0xC
+#	define R300_TX_FORMAT_W2Z10Y10X10	    0xD
+#	define R300_TX_FORMAT_W16Z16Y16X16	    0xE
+#	define R300_TX_FORMAT_DXT1	    	    0xF
+#	define R300_TX_FORMAT_DXT3	    	    0x10
+#	define R300_TX_FORMAT_DXT5	    	    0x11
+#	define R300_TX_FORMAT_D3DMFT_CxV8U8	    0x12     /* no swizzle */
+#	define R300_TX_FORMAT_A8R8G8B8	    	    0x13     /* no swizzle */
+#	define R300_TX_FORMAT_B8G8_B8G8	    	    0x14     /* no swizzle */
+#	define R300_TX_FORMAT_G8R8_G8B8	    	    0x15     /* no swizzle */
+						  /* 0x16 - some 16 bit green format.. ?? */
+#	define R300_TX_FORMAT_UNK25		   (1 << 25) /* no swizzle */
+
+	/* gap */
+	/* Floating point formats */
+	/* Note - hardware supports both 16 and 32 bit floating point */
+#	define R300_TX_FORMAT_FL_I16	    	    0x18
+#	define R300_TX_FORMAT_FL_I16A16	    	    0x19
+#	define R300_TX_FORMAT_FL_R16G16B16A16	    0x1A
+#	define R300_TX_FORMAT_FL_I32	    	    0x1B
+#	define R300_TX_FORMAT_FL_I32A32	    	    0x1C
+#	define R300_TX_FORMAT_FL_R32G32B32A32	    0x1D
+	/* alpha modes, convenience mostly */
+	/* if you have alpha, pick constant appropriate to the
+	   number of channels (1 for I8, 2 for I8A8, 4 for R8G8B8A8, etc */
+# 	define R300_TX_FORMAT_ALPHA_1CH		    0x000
+# 	define R300_TX_FORMAT_ALPHA_2CH		    0x200
+# 	define R300_TX_FORMAT_ALPHA_4CH		    0x600
+# 	define R300_TX_FORMAT_ALPHA_NONE	    0xA00
+	/* Swizzling */
+	/* constants */
+#	define R300_TX_FORMAT_X		0
+#	define R300_TX_FORMAT_Y		1
+#	define R300_TX_FORMAT_Z		2
+#	define R300_TX_FORMAT_W		3
+#	define R300_TX_FORMAT_ZERO	4
+#	define R300_TX_FORMAT_ONE	5
+#	define R300_TX_FORMAT_CUT_Z	6		/* 2.0*Z, everything above 1.0 is set to 0.0 */
+#	define R300_TX_FORMAT_CUT_W	7		/* 2.0*W, everything above 1.0 is set to 0.0 */
+
+#	define R300_TX_FORMAT_B_SHIFT	18
+#	define R300_TX_FORMAT_G_SHIFT	15
+#	define R300_TX_FORMAT_R_SHIFT	12
+#	define R300_TX_FORMAT_A_SHIFT	9
+	/* Convenience macro to take care of layout and swizzling */
+#	define R300_EASY_TX_FORMAT(B, G, R, A, FMT)	(\
+	  ((R300_TX_FORMAT_##B)<<R300_TX_FORMAT_B_SHIFT) \
+	| ((R300_TX_FORMAT_##G)<<R300_TX_FORMAT_G_SHIFT) \
+	| ((R300_TX_FORMAT_##R)<<R300_TX_FORMAT_R_SHIFT) \
+	| ((R300_TX_FORMAT_##A)<<R300_TX_FORMAT_A_SHIFT) \
+	| (R300_TX_FORMAT_##FMT) \
+	  )
+	/* These can be ORed with result of R300_EASY_TX_FORMAT() */
+	/* We don't really know what they do. Take values from a constant color ? */
+#	define R300_TX_FORMAT_CONST_X		(1<<5)
+#	define R300_TX_FORMAT_CONST_Y		(2<<5)
+#	define R300_TX_FORMAT_CONST_Z		(4<<5)
+#	define R300_TX_FORMAT_CONST_W		(8<<5)
+
+#	define R300_TX_FORMAT_YUV_MODE		0x00800000
+
+#define R300_TX_OFFSET_0                    0x4540
+/* BEGIN: Guess from R200 */
+#       define R300_TXO_ENDIAN_NO_SWAP           (0 << 0)
+#       define R300_TXO_ENDIAN_BYTE_SWAP         (1 << 0)
+#       define R300_TXO_ENDIAN_WORD_SWAP         (2 << 0)
+#       define R300_TXO_ENDIAN_HALFDW_SWAP       (3 << 0)
+#       define R300_TXO_OFFSET_MASK              0xffffffe0
+#       define R300_TXO_OFFSET_SHIFT             5
+/* END */
+#define R300_TX_UNK4_0                      0x4580
+#define R300_TX_BORDER_COLOR_0              0x45C0 //ff00ff00 == { 0, 1.0, 0, 1.0 }
+
+/* END */
+
+/* BEGIN: Fragment program instruction set
+// Fragment programs are written directly into register space.
+// There are separate instruction streams for texture instructions and ALU
+// instructions.
+// In order to synchronize these streams, the program is divided into up
+// to 4 nodes. Each node begins with a number of TEX operations, followed
+// by a number of ALU operations.
+// The first node can have zero TEX ops, all subsequent nodes must have at least
+// one TEX ops.
+// All nodes must have at least one ALU op.
+//
+// The index of the last node is stored in PFS_CNTL_0: A value of 0 means
+// 1 node, a value of 3 means 4 nodes.
+// The total amount of instructions is defined in PFS_CNTL_2. The offsets are
+// offsets into the respective instruction streams, while *_END points to the
+// last instruction relative to this offset. */
+#define R300_PFS_CNTL_0                     0x4600
+#       define R300_PFS_CNTL_LAST_NODES_SHIFT    0
+#       define R300_PFS_CNTL_LAST_NODES_MASK     (3 << 0)
+#       define R300_PFS_CNTL_FIRST_NODE_HAS_TEX  (1 << 3)
+#define R300_PFS_CNTL_1                     0x4604
+/* There is an unshifted value here which has so far always been equal to the
+// index of the highest used temporary register. */
+#define R300_PFS_CNTL_2                     0x4608
+#       define R300_PFS_CNTL_ALU_OFFSET_SHIFT    0
+#       define R300_PFS_CNTL_ALU_OFFSET_MASK     (63 << 0)
+#       define R300_PFS_CNTL_ALU_END_SHIFT       6
+#       define R300_PFS_CNTL_ALU_END_MASK        (63 << 0)
+#       define R300_PFS_CNTL_TEX_OFFSET_SHIFT    12
+#       define R300_PFS_CNTL_TEX_OFFSET_MASK     (31 << 12) /* GUESS */
+#       define R300_PFS_CNTL_TEX_END_SHIFT       18
+#       define R300_PFS_CNTL_TEX_END_MASK        (31 << 18) /* GUESS */
+
+/* gap */
+/* Nodes are stored backwards. The last active node is always stored in
+// PFS_NODE_3.
+// Example: In a 2-node program, NODE_0 and NODE_1 are set to 0. The
+// first node is stored in NODE_2, the second node is stored in NODE_3.
+//
+// Offsets are relative to the master offset from PFS_CNTL_2.
+// LAST_NODE is set for the last node, and only for the last node. */
+#define R300_PFS_NODE_0                     0x4610
+#define R300_PFS_NODE_1                     0x4614
+#define R300_PFS_NODE_2                     0x4618
+#define R300_PFS_NODE_3                     0x461C
+#       define R300_PFS_NODE_ALU_OFFSET_SHIFT    0
+#       define R300_PFS_NODE_ALU_OFFSET_MASK     (63 << 0)
+#       define R300_PFS_NODE_ALU_END_SHIFT       6
+#       define R300_PFS_NODE_ALU_END_MASK        (63 << 6)
+#       define R300_PFS_NODE_TEX_OFFSET_SHIFT    12
+#       define R300_PFS_NODE_TEX_OFFSET_MASK     (31 << 12)
+#       define R300_PFS_NODE_TEX_END_SHIFT       17
+#       define R300_PFS_NODE_TEX_END_MASK        (31 << 17)
+#       define R300_PFS_NODE_LAST_NODE           (1 << 22)
+
+/* TEX
+// As far as I can tell, texture instructions cannot write into output
+// registers directly. A subsequent ALU instruction is always necessary,
+// even if it's just MAD o0, r0, 1, 0 */
+#define R300_PFS_TEXI_0                     0x4620
+#       define R300_FPITX_SRC_SHIFT              0
+#       define R300_FPITX_SRC_MASK               (31 << 0)
+#       define R300_FPITX_SRC_CONST              (1 << 5) /* GUESS */
+#       define R300_FPITX_DST_SHIFT              6
+#       define R300_FPITX_DST_MASK               (31 << 6)
+#       define R300_FPITX_IMAGE_SHIFT            11
+#       define R300_FPITX_IMAGE_MASK             (15 << 11) /* GUESS based on layout and native limits */
+/* Unsure if these are opcodes, or some kind of bitfield, but this is how
+ * they were set when I checked
+ */
+#		define R300_FPITX_OPCODE_SHIFT			15
+#			define R300_FPITX_OP_TEX			1
+#			define R300_FPITX_OP_TXP			3
+#			define R300_FPITX_OP_TXB			4
+
+/* ALU
+// The ALU instructions register blocks are enumerated according to the order
+// in which fglrx. I assume there is space for 64 instructions, since
+// each block has space for a maximum of 64 DWORDs, and this matches reported
+// native limits.
+//
+// The basic functional block seems to be one MAD for each color and alpha,
+// and an adder that adds all components after the MUL.
+//  - ADD, MUL, MAD etc.: use MAD with appropriate neutral operands
+//  - DP4: Use OUTC_DP4, OUTA_DP4
+//  - DP3: Use OUTC_DP3, OUTA_DP4, appropriate alpha operands
+//  - DPH: Use OUTC_DP4, OUTA_DP4, appropriate alpha operands
+//  - CMP: If ARG2 < 0, return ARG1, else return ARG0
+//  - FLR: use FRC+MAD
+//  - XPD: use MAD+MAD
+//  - SGE, SLT: use MAD+CMP
+//  - RSQ: use ABS modifier for argument
+//  - Use OUTC_REPL_ALPHA to write results of an alpha-only operation (e.g. RCP)
+//    into color register
+//  - apparently, there's no quick DST operation
+//  - fglrx set FPI2_UNKNOWN_31 on a "MAD fragment.color, tmp0, tmp1, tmp2"
+//  - fglrx set FPI2_UNKNOWN_31 on a "MAX r2, r1, c0"
+//  - fglrx once set FPI0_UNKNOWN_31 on a "FRC r1, r1"
+//
+// Operand selection
+// First stage selects three sources from the available registers and
+// constant parameters. This is defined in INSTR1 (color) and INSTR3 (alpha).
+// fglrx sorts the three source fields: Registers before constants,
+// lower indices before higher indices; I do not know whether this is necessary.
+// fglrx fills unused sources with "read constant 0"
+// According to specs, you cannot select more than two different constants.
+//
+// Second stage selects the operands from the sources. This is defined in
+// INSTR0 (color) and INSTR2 (alpha). You can also select the special constants
+// zero and one.
+// Swizzling and negation happens in this stage, as well.
+//
+// Important: Color and alpha seem to be mostly separate, i.e. their sources
+// selection appears to be fully independent (the register storage is probably
+// physically split into a color and an alpha section).
+// However (because of the apparent physical split), there is some interaction
+// WRT swizzling. If, for example, you want to load an R component into an
+// Alpha operand, this R component is taken from a *color* source, not from
+// an alpha source. The corresponding register doesn't even have to appear in
+// the alpha sources list. (I hope this alll makes sense to you)
+//
+// Destination selection
+// The destination register index is in FPI1 (color) and FPI3 (alpha) together
+// with enable bits.
+// There are separate enable bits for writing into temporary registers
+// (DSTC_REG_* /DSTA_REG) and and program output registers (DSTC_OUTPUT_* /DSTA_OUTPUT).
+// You can write to both at once, or not write at all (the same index
+// must be used for both).
+//
+// Note: There is a special form for LRP
+//  - Argument order is the same as in ARB_fragment_program.
+//  - Operation is MAD
+//  - ARG1 is set to ARGC_SRC1C_LRP/ARGC_SRC1A_LRP
+//  - Set FPI0/FPI2_SPECIAL_LRP
+// Arbitrary LRP (including support for swizzling) requires vanilla MAD+MAD */
+#define R300_PFS_INSTR1_0                   0x46C0
+#       define R300_FPI1_SRC0C_SHIFT             0
+#       define R300_FPI1_SRC0C_MASK              (31 << 0)
+#       define R300_FPI1_SRC0C_CONST             (1 << 5)
+#       define R300_FPI1_SRC1C_SHIFT             6
+#       define R300_FPI1_SRC1C_MASK              (31 << 6)
+#       define R300_FPI1_SRC1C_CONST             (1 << 11)
+#       define R300_FPI1_SRC2C_SHIFT             12
+#       define R300_FPI1_SRC2C_MASK              (31 << 12)
+#       define R300_FPI1_SRC2C_CONST             (1 << 17)
+#       define R300_FPI1_DSTC_SHIFT              18
+#       define R300_FPI1_DSTC_MASK               (31 << 18)
+#       define R300_FPI1_DSTC_REG_X              (1 << 23)
+#       define R300_FPI1_DSTC_REG_Y              (1 << 24)
+#       define R300_FPI1_DSTC_REG_Z              (1 << 25)
+#       define R300_FPI1_DSTC_OUTPUT_X           (1 << 26)
+#       define R300_FPI1_DSTC_OUTPUT_Y           (1 << 27)
+#       define R300_FPI1_DSTC_OUTPUT_Z           (1 << 28)
+
+#define R300_PFS_INSTR3_0                   0x47C0
+#       define R300_FPI3_SRC0A_SHIFT             0
+#       define R300_FPI3_SRC0A_MASK              (31 << 0)
+#       define R300_FPI3_SRC0A_CONST             (1 << 5)
+#       define R300_FPI3_SRC1A_SHIFT             6
+#       define R300_FPI3_SRC1A_MASK              (31 << 6)
+#       define R300_FPI3_SRC1A_CONST             (1 << 11)
+#       define R300_FPI3_SRC2A_SHIFT             12
+#       define R300_FPI3_SRC2A_MASK              (31 << 12)
+#       define R300_FPI3_SRC2A_CONST             (1 << 17)
+#       define R300_FPI3_DSTA_SHIFT              18
+#       define R300_FPI3_DSTA_MASK               (31 << 18)
+#       define R300_FPI3_DSTA_REG                (1 << 23)
+#       define R300_FPI3_DSTA_OUTPUT             (1 << 24)
+
+#define R300_PFS_INSTR0_0                   0x48C0
+#       define R300_FPI0_ARGC_SRC0C_XYZ          0
+#       define R300_FPI0_ARGC_SRC0C_XXX          1
+#       define R300_FPI0_ARGC_SRC0C_YYY          2
+#       define R300_FPI0_ARGC_SRC0C_ZZZ          3
+#       define R300_FPI0_ARGC_SRC1C_XYZ          4
+#       define R300_FPI0_ARGC_SRC1C_XXX          5
+#       define R300_FPI0_ARGC_SRC1C_YYY          6
+#       define R300_FPI0_ARGC_SRC1C_ZZZ          7
+#       define R300_FPI0_ARGC_SRC2C_XYZ          8
+#       define R300_FPI0_ARGC_SRC2C_XXX          9
+#       define R300_FPI0_ARGC_SRC2C_YYY          10
+#       define R300_FPI0_ARGC_SRC2C_ZZZ          11
+#       define R300_FPI0_ARGC_SRC0A              12
+#       define R300_FPI0_ARGC_SRC1A              13
+#       define R300_FPI0_ARGC_SRC2A              14
+#       define R300_FPI0_ARGC_SRC1C_LRP          15
+#       define R300_FPI0_ARGC_ZERO               20
+#       define R300_FPI0_ARGC_ONE                21
+#       define R300_FPI0_ARGC_HALF               22 /* GUESS */
+#       define R300_FPI0_ARGC_SRC0C_YZX          23
+#       define R300_FPI0_ARGC_SRC1C_YZX          24
+#       define R300_FPI0_ARGC_SRC2C_YZX          25
+#       define R300_FPI0_ARGC_SRC0C_ZXY          26
+#       define R300_FPI0_ARGC_SRC1C_ZXY          27
+#       define R300_FPI0_ARGC_SRC2C_ZXY          28
+#       define R300_FPI0_ARGC_SRC0CA_WZY         29
+#       define R300_FPI0_ARGC_SRC1CA_WZY         30
+#       define R300_FPI0_ARGC_SRC2CA_WZY         31
+
+#       define R300_FPI0_ARG0C_SHIFT             0
+#       define R300_FPI0_ARG0C_MASK              (31 << 0)
+#       define R300_FPI0_ARG0C_NEG               (1 << 5)
+#       define R300_FPI0_ARG0C_ABS               (1 << 6)
+#       define R300_FPI0_ARG1C_SHIFT             7
+#       define R300_FPI0_ARG1C_MASK              (31 << 7)
+#       define R300_FPI0_ARG1C_NEG               (1 << 12)
+#       define R300_FPI0_ARG1C_ABS               (1 << 13)
+#       define R300_FPI0_ARG2C_SHIFT             14
+#       define R300_FPI0_ARG2C_MASK              (31 << 14)
+#       define R300_FPI0_ARG2C_NEG               (1 << 19)
+#       define R300_FPI0_ARG2C_ABS               (1 << 20)
+#       define R300_FPI0_SPECIAL_LRP             (1 << 21)
+#       define R300_FPI0_OUTC_MAD                (0 << 23)
+#       define R300_FPI0_OUTC_DP3                (1 << 23)
+#       define R300_FPI0_OUTC_DP4                (2 << 23)
+#       define R300_FPI0_OUTC_MIN                (4 << 23)
+#       define R300_FPI0_OUTC_MAX                (5 << 23)
+#       define R300_FPI0_OUTC_CMP                (8 << 23)
+#       define R300_FPI0_OUTC_FRC                (9 << 23)
+#       define R300_FPI0_OUTC_REPL_ALPHA         (10 << 23)
+#       define R300_FPI0_OUTC_SAT                (1 << 30)
+#       define R300_FPI0_UNKNOWN_31              (1 << 31)
+
+#define R300_PFS_INSTR2_0                   0x49C0
+#       define R300_FPI2_ARGA_SRC0C_X            0
+#       define R300_FPI2_ARGA_SRC0C_Y            1
+#       define R300_FPI2_ARGA_SRC0C_Z            2
+#       define R300_FPI2_ARGA_SRC1C_X            3
+#       define R300_FPI2_ARGA_SRC1C_Y            4
+#       define R300_FPI2_ARGA_SRC1C_Z            5
+#       define R300_FPI2_ARGA_SRC2C_X            6
+#       define R300_FPI2_ARGA_SRC2C_Y            7
+#       define R300_FPI2_ARGA_SRC2C_Z            8
+#       define R300_FPI2_ARGA_SRC0A              9
+#       define R300_FPI2_ARGA_SRC1A              10
+#       define R300_FPI2_ARGA_SRC2A              11
+#       define R300_FPI2_ARGA_SRC1A_LRP          15
+#       define R300_FPI2_ARGA_ZERO               16
+#       define R300_FPI2_ARGA_ONE                17
+#       define R300_FPI2_ARGA_HALF               18 /* GUESS */
+
+#       define R300_FPI2_ARG0A_SHIFT             0
+#       define R300_FPI2_ARG0A_MASK              (31 << 0)
+#       define R300_FPI2_ARG0A_NEG               (1 << 5)
+#		define R300_FPI2_ARG0A_ABS				 (1 << 6) /* GUESS */
+#       define R300_FPI2_ARG1A_SHIFT             7
+#       define R300_FPI2_ARG1A_MASK              (31 << 7)
+#       define R300_FPI2_ARG1A_NEG               (1 << 12)
+#		define R300_FPI2_ARG1A_ABS				 (1 << 13) /* GUESS */
+#       define R300_FPI2_ARG2A_SHIFT             14
+#       define R300_FPI2_ARG2A_MASK              (31 << 14)
+#       define R300_FPI2_ARG2A_NEG               (1 << 19)
+#		define R300_FPI2_ARG2A_ABS				 (1 << 20) /* GUESS */
+#       define R300_FPI2_SPECIAL_LRP             (1 << 21)
+#       define R300_FPI2_OUTA_MAD                (0 << 23)
+#       define R300_FPI2_OUTA_DP4                (1 << 23)
+#       define R300_FPI2_OUTA_MIN                (2 << 23)
+#       define R300_FPI2_OUTA_MAX                (3 << 23)
+#       define R300_FPI2_OUTA_CMP                (6 << 23)
+#       define R300_FPI2_OUTA_FRC                (7 << 23)
+#       define R300_FPI2_OUTA_EX2                (8 << 23)
+#       define R300_FPI2_OUTA_LG2                (9 << 23)
+#       define R300_FPI2_OUTA_RCP                (10 << 23)
+#       define R300_FPI2_OUTA_RSQ                (11 << 23)
+#       define R300_FPI2_OUTA_SAT                (1 << 30)
+#       define R300_FPI2_UNKNOWN_31              (1 << 31)
+/* END */
+
+/* gap */
+#define R300_PP_ALPHA_TEST                  0x4BD4
+#       define R300_REF_ALPHA_MASK               0x000000ff
+#       define R300_ALPHA_TEST_FAIL              (0 << 8)
+#       define R300_ALPHA_TEST_LESS              (1 << 8)
+#       define R300_ALPHA_TEST_LEQUAL            (3 << 8)
+#       define R300_ALPHA_TEST_EQUAL             (2 << 8)
+#       define R300_ALPHA_TEST_GEQUAL            (6 << 8)
+#       define R300_ALPHA_TEST_GREATER           (4 << 8)
+#       define R300_ALPHA_TEST_NEQUAL            (5 << 8)
+#       define R300_ALPHA_TEST_PASS              (7 << 8)
+#       define R300_ALPHA_TEST_OP_MASK           (7 << 8)
+#       define R300_ALPHA_TEST_ENABLE            (1 << 11)
+
+/* gap */
+/* Fragment program parameters in 7.16 floating point */
+#define R300_PFS_PARAM_0_X                  0x4C00
+#define R300_PFS_PARAM_0_Y                  0x4C04
+#define R300_PFS_PARAM_0_Z                  0x4C08
+#define R300_PFS_PARAM_0_W                  0x4C0C
+/* GUESS: PARAM_31 is last, based on native limits reported by fglrx */
+#define R300_PFS_PARAM_31_X                 0x4DF0
+#define R300_PFS_PARAM_31_Y                 0x4DF4
+#define R300_PFS_PARAM_31_Z                 0x4DF8
+#define R300_PFS_PARAM_31_W                 0x4DFC
+
+/* Notes:
+// - AFAIK fglrx always sets BLEND_UNKNOWN when blending is used in the application
+// - AFAIK fglrx always sets BLEND_NO_SEPARATE when CBLEND and ABLEND are set to the same
+//   function (both registers are always set up completely in any case)
+// - Most blend flags are simply copied from R200 and not tested yet */
+#define R300_RB3D_CBLEND                    0x4E04
+#define R300_RB3D_ABLEND                    0x4E08
+ /* the following only appear in CBLEND */
+#       define R300_BLEND_ENABLE                     (1 << 0)
+#       define R300_BLEND_UNKNOWN                    (3 << 1)
+#       define R300_BLEND_NO_SEPARATE                (1 << 3)
+ /* the following are shared between CBLEND and ABLEND */
+#       define R300_FCN_MASK                         (3  << 12)
+#       define R300_COMB_FCN_ADD_CLAMP               (0  << 12)
+#       define R300_COMB_FCN_ADD_NOCLAMP             (1  << 12)
+#       define R300_COMB_FCN_SUB_CLAMP               (2  << 12)
+#       define R300_COMB_FCN_SUB_NOCLAMP             (3  << 12)
+#       define R300_SRC_BLEND_GL_ZERO                (32 << 16)
+#       define R300_SRC_BLEND_GL_ONE                 (33 << 16)
+#       define R300_SRC_BLEND_GL_SRC_COLOR           (34 << 16)
+#       define R300_SRC_BLEND_GL_ONE_MINUS_SRC_COLOR (35 << 16)
+#       define R300_SRC_BLEND_GL_DST_COLOR           (36 << 16)
+#       define R300_SRC_BLEND_GL_ONE_MINUS_DST_COLOR (37 << 16)
+#       define R300_SRC_BLEND_GL_SRC_ALPHA           (38 << 16)
+#       define R300_SRC_BLEND_GL_ONE_MINUS_SRC_ALPHA (39 << 16)
+#       define R300_SRC_BLEND_GL_DST_ALPHA           (40 << 16)
+#       define R300_SRC_BLEND_GL_ONE_MINUS_DST_ALPHA (41 << 16)
+#       define R300_SRC_BLEND_GL_SRC_ALPHA_SATURATE  (42 << 16)
+#       define R300_SRC_BLEND_MASK                   (63 << 16)
+#       define R300_DST_BLEND_GL_ZERO                (32 << 24)
+#       define R300_DST_BLEND_GL_ONE                 (33 << 24)
+#       define R300_DST_BLEND_GL_SRC_COLOR           (34 << 24)
+#       define R300_DST_BLEND_GL_ONE_MINUS_SRC_COLOR (35 << 24)
+#       define R300_DST_BLEND_GL_DST_COLOR           (36 << 24)
+#       define R300_DST_BLEND_GL_ONE_MINUS_DST_COLOR (37 << 24)
+#       define R300_DST_BLEND_GL_SRC_ALPHA           (38 << 24)
+#       define R300_DST_BLEND_GL_ONE_MINUS_SRC_ALPHA (39 << 24)
+#       define R300_DST_BLEND_GL_DST_ALPHA           (40 << 24)
+#       define R300_DST_BLEND_GL_ONE_MINUS_DST_ALPHA (41 << 24)
+#       define R300_DST_BLEND_MASK                   (63 << 24)
+#define R300_RB3D_COLORMASK                 0x4E0C
+#       define R300_COLORMASK0_B                 (1<<0)
+#       define R300_COLORMASK0_G                 (1<<1)
+#       define R300_COLORMASK0_R                 (1<<2)
+#       define R300_COLORMASK0_A                 (1<<3)
+
+/* gap */
+#define R300_RB3D_COLOROFFSET0              0x4E28
+#       define R300_COLOROFFSET_MASK             0xFFFFFFF0 /* GUESS */
+#define R300_RB3D_COLOROFFSET1              0x4E2C /* GUESS */
+#define R300_RB3D_COLOROFFSET2              0x4E30 /* GUESS */
+#define R300_RB3D_COLOROFFSET3              0x4E34 /* GUESS */
+/* gap */
+/* Bit 16: Larger tiles
+// Bit 17: 4x2 tiles
+// Bit 18: Extremely weird tile like, but some pixels duplicated? */
+#define R300_RB3D_COLORPITCH0               0x4E38
+#       define R300_COLORPITCH_MASK              0x00001FF8 /* GUESS */
+#       define R300_COLOR_TILE_ENABLE            (1 << 16) /* GUESS */
+#       define R300_COLOR_MICROTILE_ENABLE       (1 << 17) /* GUESS */
+#       define R300_COLOR_ENDIAN_NO_SWAP         (0 << 18) /* GUESS */
+#       define R300_COLOR_ENDIAN_WORD_SWAP       (1 << 18) /* GUESS */
+#       define R300_COLOR_ENDIAN_DWORD_SWAP      (2 << 18) /* GUESS */
+#       define R300_COLOR_FORMAT_RGB565          (2 << 22)
+#       define R300_COLOR_FORMAT_ARGB8888        (3 << 22)
+#define R300_RB3D_COLORPITCH1               0x4E3C /* GUESS */
+#define R300_RB3D_COLORPITCH2               0x4E40 /* GUESS */
+#define R300_RB3D_COLORPITCH3               0x4E44 /* GUESS */
+
+/* gap */
+/* Guess by Vladimir.
+// Set to 0A before 3D operations, set to 02 afterwards. */
+#define R300_RB3D_DSTCACHE_CTLSTAT          0x4E4C
+#       define R300_RB3D_DSTCACHE_02             0x00000002
+#       define R300_RB3D_DSTCACHE_0A             0x0000000A
+
+/* gap */
+/* There seems to be no "write only" setting, so use Z-test = ALWAYS for this. */
+/* Bit (1<<8) is the "test" bit. so plain write is 6  - vd */
+#define R300_RB3D_ZSTENCIL_CNTL_0                   0x4F00
+#       define R300_RB3D_Z_DISABLED_1            0x00000010 /* GUESS */
+#       define R300_RB3D_Z_DISABLED_2            0x00000014 /* GUESS */
+#       define R300_RB3D_Z_TEST                  0x00000012
+#       define R300_RB3D_Z_TEST_AND_WRITE        0x00000016
+#       define R300_RB3D_Z_WRITE_ONLY        	 0x00000006
+
+#       define R300_RB3D_Z_TEST                  0x00000012
+#       define R300_RB3D_Z_TEST_AND_WRITE        0x00000016
+#       define R300_RB3D_Z_WRITE_ONLY        	 0x00000006
+#	define R300_RB3D_STENCIL_ENABLE		 0x00000001
+
+#define R300_RB3D_ZSTENCIL_CNTL_1                   0x4F04
+		/* functions */
+#	define R300_ZS_NEVER			0
+#	define R300_ZS_LESS			1
+#	define R300_ZS_LEQUAL			2
+#	define R300_ZS_EQUAL			3
+#	define R300_ZS_GEQUAL			4
+#	define R300_ZS_GREATER			5
+#	define R300_ZS_NOTEQUAL			6
+#	define R300_ZS_ALWAYS			7
+#       define R300_ZS_MASK                     7
+		/* operations */
+#	define R300_ZS_KEEP			0
+#	define R300_ZS_ZERO			1
+#	define R300_ZS_REPLACE			2
+#	define R300_ZS_INCR			3
+#	define R300_ZS_DECR			4
+#	define R300_ZS_INVERT			5
+#	define R300_ZS_INCR_WRAP		6
+#	define R300_ZS_DECR_WRAP		7
+
+       /* front and back refer to operations done for front
+          and back faces, i.e. separate stencil function support */
+#	define R300_RB3D_ZS1_DEPTH_FUNC_SHIFT		0
+#	define R300_RB3D_ZS1_FRONT_FUNC_SHIFT		3
+#	define R300_RB3D_ZS1_FRONT_FAIL_OP_SHIFT	6
+#	define R300_RB3D_ZS1_FRONT_ZPASS_OP_SHIFT	9
+#	define R300_RB3D_ZS1_FRONT_ZFAIL_OP_SHIFT      12
+#	define R300_RB3D_ZS1_BACK_FUNC_SHIFT           15
+#	define R300_RB3D_ZS1_BACK_FAIL_OP_SHIFT        18
+#	define R300_RB3D_ZS1_BACK_ZPASS_OP_SHIFT       21
+#	define R300_RB3D_ZS1_BACK_ZFAIL_OP_SHIFT       24
+
+
+
+#define R300_RB3D_ZSTENCIL_CNTL_2                   0x4F08
+#	define R300_RB3D_ZS2_STENCIL_REF_SHIFT		0
+#	define R300_RB3D_ZS2_STENCIL_MASK		0xFF
+#	define R300_RB3D_ZS2_STENCIL_MASK_SHIFT	        8
+#	define R300_RB3D_ZS2_STENCIL_WRITE_MASK_SHIFT	16
+
+/* gap */
+
+#define R300_RB3D_ZSTENCIL_FORMAT                   0x4F10
+#	define R300_DEPTH_FORMAT_16BIT_INT_Z     (0 << 0)
+#	define R300_DEPTH_FORMAT_24BIT_INT_Z     (2 << 0)
+
+/* gap */
+#define R300_RB3D_DEPTHOFFSET               0x4F20
+#define R300_RB3D_DEPTHPITCH                0x4F24
+#       define R300_DEPTHPITCH_MASK              0x00001FF8 /* GUESS */
+#       define R300_DEPTH_TILE_ENABLE            (1 << 16) /* GUESS */
+#       define R300_DEPTH_MICROTILE_ENABLE       (1 << 17) /* GUESS */
+#       define R300_DEPTH_ENDIAN_NO_SWAP         (0 << 18) /* GUESS */
+#       define R300_DEPTH_ENDIAN_WORD_SWAP       (1 << 18) /* GUESS */
+#       define R300_DEPTH_ENDIAN_DWORD_SWAP      (2 << 18) /* GUESS */
+
+/* BEGIN: Vertex program instruction set
+// Every instruction is four dwords long:
+//  DWORD 0: output and opcode
+//  DWORD 1: first argument
+//  DWORD 2: second argument
+//  DWORD 3: third argument
+//
+// Notes:
+//  - ABS r, a is implemented as MAX r, a, -a
+//  - MOV is implemented as ADD to zero
+//  - XPD is implemented as MUL + MAD
+//  - FLR is implemented as FRC + ADD
+//  - apparently, fglrx tries to schedule instructions so that there is at least
+//    one instruction between the write to a temporary and the first read
+//    from said temporary; however, violations of this scheduling are allowed
+//  - register indices seem to be unrelated with OpenGL aliasing to conventional state
+//  - only one attribute and one parameter can be loaded at a time; however, the
+//    same attribute/parameter can be used for more than one argument
+//  - the second software argument for POW is the third hardware argument (no idea why)
+//  - MAD with only temporaries as input seems to use VPI_OUT_SELECT_MAD_2
+//
+// There is some magic surrounding LIT:
+//  The single argument is replicated across all three inputs, but swizzled:
+//   First argument: xyzy
+//   Second argument: xyzx
+//   Third argument: xyzw
+//  Whenever the result is used later in the fragment program, fglrx forces x and w
+//  to be 1.0 in the input selection; I don't know whether this is strictly necessary */
+#define R300_VPI_OUT_OP_DOT                     (1 << 0)
+#define R300_VPI_OUT_OP_MUL                     (2 << 0)
+#define R300_VPI_OUT_OP_ADD                     (3 << 0)
+#define R300_VPI_OUT_OP_MAD                     (4 << 0)
+#define R300_VPI_OUT_OP_DST                     (5 << 0)
+#define R300_VPI_OUT_OP_FRC                     (6 << 0)
+#define R300_VPI_OUT_OP_MAX                     (7 << 0)
+#define R300_VPI_OUT_OP_MIN                     (8 << 0)
+#define R300_VPI_OUT_OP_SGE                     (9 << 0)
+#define R300_VPI_OUT_OP_SLT                     (10 << 0)
+#define R300_VPI_OUT_OP_UNK12                   (12 << 0) /* Used in GL_POINT_DISTANCE_ATTENUATION_ARB, vector(scalar, vector) */
+#define R300_VPI_OUT_OP_EXP                     (65 << 0)
+#define R300_VPI_OUT_OP_LOG                     (66 << 0)
+#define R300_VPI_OUT_OP_UNK67                   (67 << 0) /* Used in fog computations, scalar(scalar) */
+#define R300_VPI_OUT_OP_LIT                     (68 << 0)
+#define R300_VPI_OUT_OP_POW                     (69 << 0)
+#define R300_VPI_OUT_OP_RCP                     (70 << 0)
+#define R300_VPI_OUT_OP_RSQ                     (72 << 0)
+#define R300_VPI_OUT_OP_UNK73                   (73 << 0) /* Used in GL_POINT_DISTANCE_ATTENUATION_ARB, scalar(scalar) */
+#define R300_VPI_OUT_OP_EX2                     (75 << 0)
+#define R300_VPI_OUT_OP_LG2                     (76 << 0)
+#define R300_VPI_OUT_OP_MAD_2                   (128 << 0)
+#define R300_VPI_OUT_OP_UNK129                  (129 << 0) /* all temps, vector(scalar, vector, vector) */
+
+#define R300_VPI_OUT_REG_CLASS_TEMPORARY        (0 << 8)
+#define R300_VPI_OUT_REG_CLASS_RESULT           (2 << 8)
+#define R300_VPI_OUT_REG_CLASS_MASK             (31 << 8)
+
+#define R300_VPI_OUT_REG_INDEX_SHIFT            13
+#define R300_VPI_OUT_REG_INDEX_MASK             (31 << 13) /* GUESS based on fglrx native limits */
+
+#define R300_VPI_OUT_WRITE_X                    (1 << 20)
+#define R300_VPI_OUT_WRITE_Y                    (1 << 21)
+#define R300_VPI_OUT_WRITE_Z                    (1 << 22)
+#define R300_VPI_OUT_WRITE_W                    (1 << 23)
+
+#define R300_VPI_IN_REG_CLASS_TEMPORARY         (0 << 0)
+#define R300_VPI_IN_REG_CLASS_ATTRIBUTE         (1 << 0)
+#define R300_VPI_IN_REG_CLASS_PARAMETER         (2 << 0)
+#define R300_VPI_IN_REG_CLASS_NONE              (9 << 0)
+#define R300_VPI_IN_REG_CLASS_MASK              (31 << 0) /* GUESS */
+
+#define R300_VPI_IN_REG_INDEX_SHIFT             5
+#define R300_VPI_IN_REG_INDEX_MASK              (255 << 5) /* GUESS based on fglrx native limits */
+
+/* The R300 can select components from the input register arbitrarily.
+// Use the following constants, shifted by the component shift you
+// want to select */
+#define R300_VPI_IN_SELECT_X    0
+#define R300_VPI_IN_SELECT_Y    1
+#define R300_VPI_IN_SELECT_Z    2
+#define R300_VPI_IN_SELECT_W    3
+#define R300_VPI_IN_SELECT_ZERO 4
+#define R300_VPI_IN_SELECT_ONE  5
+#define R300_VPI_IN_SELECT_MASK 7
+
+#define R300_VPI_IN_X_SHIFT                     13
+#define R300_VPI_IN_Y_SHIFT                     16
+#define R300_VPI_IN_Z_SHIFT                     19
+#define R300_VPI_IN_W_SHIFT                     22
+
+#define R300_VPI_IN_NEG_X                       (1 << 25)
+#define R300_VPI_IN_NEG_Y                       (1 << 26)
+#define R300_VPI_IN_NEG_Z                       (1 << 27)
+#define R300_VPI_IN_NEG_W                       (1 << 28)
+/* END */
+
+//BEGIN: Packet 3 commands
+
+// A primitive emission dword.
+#define R300_PRIM_TYPE_NONE                     (0 << 0)
+#define R300_PRIM_TYPE_POINT                    (1 << 0)
+#define R300_PRIM_TYPE_LINE                     (2 << 0)
+#define R300_PRIM_TYPE_LINE_STRIP               (3 << 0)
+#define R300_PRIM_TYPE_TRI_LIST                 (4 << 0)
+#define R300_PRIM_TYPE_TRI_FAN                  (5 << 0)
+#define R300_PRIM_TYPE_TRI_STRIP                (6 << 0)
+#define R300_PRIM_TYPE_TRI_TYPE2                (7 << 0)
+#define R300_PRIM_TYPE_RECT_LIST                (8 << 0)
+#define R300_PRIM_TYPE_3VRT_POINT_LIST          (9 << 0)
+#define R300_PRIM_TYPE_3VRT_LINE_LIST           (10 << 0)
+#define R300_PRIM_TYPE_POINT_SPRITES            (11 << 0) // GUESS (based on r200)
+#define R300_PRIM_TYPE_LINE_LOOP                (12 << 0)
+#define R300_PRIM_TYPE_QUADS                    (13 << 0)
+#define R300_PRIM_TYPE_QUAD_STRIP               (14 << 0)
+#define R300_PRIM_TYPE_POLYGON                  (15 << 0)
+#define R300_PRIM_TYPE_MASK                     0xF
+#define R300_PRIM_WALK_IND                      (1 << 4)
+#define R300_PRIM_WALK_LIST                     (2 << 4)
+#define R300_PRIM_WALK_RING                     (3 << 4)
+#define R300_PRIM_WALK_MASK                     (3 << 4)
+#define R300_PRIM_COLOR_ORDER_BGRA              (0 << 6) // GUESS (based on r200)
+#define R300_PRIM_COLOR_ORDER_RGBA              (1 << 6) // GUESS
+#define R300_PRIM_NUM_VERTICES_SHIFT            16
+
+// Draw a primitive from vertex data in arrays loaded via 3D_LOAD_VBPNTR.
+// Two parameter dwords:
+// 0. The first parameter appears to be always 0
+// 1. The second parameter is a standard primitive emission dword.
+#define R300_PACKET3_3D_DRAW_VBUF           0x00002800
+
+// Specify the full set of vertex arrays as (address, stride).
+// The first parameter is the number of vertex arrays specified.
+// The rest of the command is a variable length list of blocks, where
+// each block is three dwords long and specifies two arrays.
+// The first dword of a block is split into two words, the lower significant
+// word refers to the first array, the more significant word to the second
+// array in the block.
+// The low byte of each word contains the size of an array entry in dwords,
+// the high byte contains the stride of the array.
+// The second dword of a block contains the pointer to the first array,
+// the third dword of a block contains the pointer to the second array.
+// Note that if the total number of arrays is odd, the third dword of
+// the last block is omitted.
+#define R300_PACKET3_3D_LOAD_VBPNTR         0x00002F00
+
+#define R300_PACKET3_INDX_BUFFER            0x00003300
+#    define R300_EB_UNK1_SHIFT                      24
+#    define R300_EB_UNK1                    (0x80<<24)
+#    define R300_EB_UNK2                        0x0810
+#define R300_PACKET3_3D_DRAW_INDX_2         0x00003600
+
+//END
+
+#endif /* _R300_REG_H */
diff --git a/drivers/char/drm/radeon_cp.c b/drivers/char/drm/radeon_cp.c
index 20bcf872b348..6d9080a3ca7e 100644
--- a/drivers/char/drm/radeon_cp.c
+++ b/drivers/char/drm/radeon_cp.c
@@ -32,6 +32,7 @@
 #include "drm.h"
 #include "radeon_drm.h"
 #include "radeon_drv.h"
+#include "r300_reg.h"
 
 #define RADEON_FIFO_DEBUG	0
 
@@ -1151,6 +1152,8 @@ static void radeon_cp_init_ring_buffer( drm_device_t *dev,
 
 #if __OS_HAS_AGP
 	if ( !dev_priv->is_pci ) {
+		/* set RADEON_AGP_BASE here instead of relying on X from user space */
+		RADEON_WRITE(RADEON_AGP_BASE, (unsigned int)dev->agp->base);
 		RADEON_WRITE( RADEON_CP_RB_RPTR_ADDR,
 			      dev_priv->ring_rptr->offset
 			      - dev->agp->base
@@ -1407,6 +1410,7 @@ static int radeon_do_init_cp( drm_device_t *dev, drm_radeon_init_t *init )
 		radeon_do_cleanup_cp(dev);
 		return DRM_ERR(EINVAL);
 	}
+	dev->agp_buffer_token = init->buffers_offset;
 	dev->agp_buffer_map = drm_core_findmap(dev, init->buffers_offset);
 	if(!dev->agp_buffer_map) {
 		DRM_ERROR("could not find dma buffer region!\n");
@@ -1625,6 +1629,9 @@ int radeon_cp_init( DRM_IOCTL_ARGS )
 
 	DRM_COPY_FROM_USER_IOCTL( init, (drm_radeon_init_t __user *)data, sizeof(init) );
 
+	if(init.func == RADEON_INIT_R300_CP)
+		r300_init_reg_flags();
+
 	switch ( init.func ) {
 	case RADEON_INIT_CP:
 	case RADEON_INIT_R200_CP:
@@ -2039,15 +2046,43 @@ int radeon_driver_preinit(struct drm_device *dev, unsigned long flags)
 	case CHIP_RV200:
 	case CHIP_R200:
 	case CHIP_R300:
+	case CHIP_R420:
 		dev_priv->flags |= CHIP_HAS_HIERZ;
 		break;
 	default:
 	/* all other chips have no hierarchical z buffer */
 		break;
 	}
+
+	if (drm_device_is_agp(dev))
+		dev_priv->flags |= CHIP_IS_AGP;
+	
+	DRM_DEBUG("%s card detected\n",
+		  ((dev_priv->flags & CHIP_IS_AGP) ? "AGP" : "PCI"));
 	return ret;
 }
 
+int radeon_presetup(struct drm_device *dev)
+{
+	int ret;
+	drm_local_map_t *map;
+	drm_radeon_private_t *dev_priv = dev->dev_private;
+
+	ret = drm_addmap(dev, drm_get_resource_start(dev, 2),
+			 drm_get_resource_len(dev, 2), _DRM_REGISTERS,
+			 _DRM_READ_ONLY, &dev_priv->mmio);
+	if (ret != 0)
+		return ret;
+
+	ret = drm_addmap(dev, drm_get_resource_start(dev, 0),
+			 drm_get_resource_len(dev, 0), _DRM_FRAME_BUFFER,
+			 _DRM_WRITE_COMBINING, &map);
+	if (ret != 0)
+		return ret;
+
+	return 0;
+}
+
 int radeon_driver_postcleanup(struct drm_device *dev)
 {
 	drm_radeon_private_t *dev_priv = dev->dev_private;
diff --git a/drivers/char/drm/radeon_drm.h b/drivers/char/drm/radeon_drm.h
index c1e62d047989..3792798270a4 100644
--- a/drivers/char/drm/radeon_drm.h
+++ b/drivers/char/drm/radeon_drm.h
@@ -195,6 +195,52 @@ typedef union {
 #define RADEON_WAIT_2D  0x1
 #define RADEON_WAIT_3D  0x2
 
+/* Allowed parameters for R300_CMD_PACKET3
+ */
+#define R300_CMD_PACKET3_CLEAR		0
+#define R300_CMD_PACKET3_RAW		1
+
+/* Commands understood by cmd_buffer ioctl for R300.
+ * The interface has not been stabilized, so some of these may be removed
+ * and eventually reordered before stabilization.
+ */
+#define R300_CMD_PACKET0		1 
+#define R300_CMD_VPU			2 /* emit vertex program upload */
+#define R300_CMD_PACKET3		3 /* emit a packet3 */
+#define R300_CMD_END3D			4 /* emit sequence ending 3d rendering */
+#define R300_CMD_CP_DELAY		5
+#define R300_CMD_DMA_DISCARD		6
+#define R300_CMD_WAIT			7
+#	define R300_WAIT_2D  		0x1
+#	define R300_WAIT_3D  		0x2
+#	define R300_WAIT_2D_CLEAN  	0x3
+#	define R300_WAIT_3D_CLEAN  	0x4
+
+typedef union {
+	unsigned int u;
+	struct {
+		unsigned char cmd_type, pad0, pad1, pad2;
+	} header;
+	struct {
+		unsigned char cmd_type, count, reglo, reghi;
+	} packet0;
+	struct {
+		unsigned char cmd_type, count, adrlo, adrhi;
+	} vpu;
+	struct {
+		unsigned char cmd_type, packet, pad0, pad1;
+	} packet3;
+	struct {
+		unsigned char cmd_type, packet;
+		unsigned short count; /* amount of packet2 to emit */
+	} delay;
+	struct {
+		unsigned char cmd_type, buf_idx, pad0, pad1;
+	} dma;
+	struct {
+		unsigned char cmd_type, flags, pad0, pad1;	
+	} wait;
+} drm_r300_cmd_header_t;
 
 #define RADEON_FRONT			0x1
 #define RADEON_BACK			0x2
diff --git a/drivers/char/drm/radeon_drv.c b/drivers/char/drm/radeon_drv.c
index 18e4e5b0952f..e0682f64b400 100644
--- a/drivers/char/drm/radeon_drv.c
+++ b/drivers/char/drm/radeon_drv.c
@@ -76,6 +76,7 @@ static struct drm_driver driver = {
 	.driver_features = DRIVER_USE_AGP | DRIVER_USE_MTRR | DRIVER_PCI_DMA | DRIVER_SG | DRIVER_HAVE_IRQ | DRIVER_HAVE_DMA | DRIVER_IRQ_SHARED | DRIVER_IRQ_VBL,
 	.dev_priv_size = sizeof(drm_radeon_buf_priv_t),
 	.preinit = radeon_driver_preinit,
+	.presetup = radeon_presetup,
 	.postcleanup = radeon_driver_postcleanup,
 	.prerelease = radeon_driver_prerelease,
 	.pretakedown = radeon_driver_pretakedown,
diff --git a/drivers/char/drm/radeon_drv.h b/drivers/char/drm/radeon_drv.h
index 771aa80a5e8c..f12a963ede18 100644
--- a/drivers/char/drm/radeon_drv.h
+++ b/drivers/char/drm/radeon_drv.h
@@ -82,9 +82,10 @@
  *     - Add support for r100 cube maps
  * 1.16- Add R200_EMIT_PP_TRI_PERF_CNTL packet to support brilinear
  *       texture filtering on r200
+ * 1.17- Add initial support for R300 (3D).
  */
 #define DRIVER_MAJOR		1
-#define DRIVER_MINOR		16
+#define DRIVER_MINOR		17
 #define DRIVER_PATCHLEVEL	0
 
 #define GET_RING_HEAD(dev_priv)		DRM_READ32(  (dev_priv)->ring_rptr, 0 )
@@ -106,7 +107,9 @@ enum radeon_family {
 	CHIP_RV280,
 	CHIP_R300,
 	CHIP_RS300,
+	CHIP_R350,
 	CHIP_RV350,
+	CHIP_R420,
 	CHIP_LAST,
 };
 
@@ -290,6 +293,7 @@ extern int radeon_wait_ring( drm_radeon_private_t *dev_priv, int n );
 extern int radeon_do_cp_idle( drm_radeon_private_t *dev_priv );
 
 extern int radeon_driver_preinit(struct drm_device *dev, unsigned long flags);
+extern int radeon_presetup(struct drm_device *dev);
 extern int radeon_driver_postcleanup(struct drm_device *dev);
 
 extern int radeon_mem_alloc( DRM_IOCTL_ARGS );
@@ -320,6 +324,14 @@ extern int radeon_postcleanup( struct drm_device *dev );
 extern long radeon_compat_ioctl(struct file *filp, unsigned int cmd,
 				unsigned long arg);
 
+
+/* r300_cmdbuf.c */
+extern void r300_init_reg_flags(void);
+
+extern int r300_do_cp_cmdbuf(drm_device_t* dev, DRMFILE filp,
+			     drm_file_t* filp_priv,
+			     drm_radeon_cmd_buffer_t* cmdbuf);
+
 /* Flags for stats.boxes
  */
 #define RADEON_BOX_DMA_IDLE      0x1
@@ -357,6 +369,11 @@ extern long radeon_compat_ioctl(struct file *filp, unsigned int cmd,
 #define RADEON_CRTC2_OFFSET		0x0324
 #define RADEON_CRTC2_OFFSET_CNTL	0x0328
 
+#define RADEON_MPP_TB_CONFIG		0x01c0
+#define RADEON_MEM_CNTL			0x0140
+#define RADEON_MEM_SDRAM_MODE_REG	0x0158
+#define RADEON_AGP_BASE			0x0170
+
 #define RADEON_RB3D_COLOROFFSET		0x1c40
 #define RADEON_RB3D_COLORPITCH		0x1c48
 
@@ -651,16 +668,27 @@ extern long radeon_compat_ioctl(struct file *filp, unsigned int cmd,
 #define RADEON_CP_PACKET1		0x40000000
 #define RADEON_CP_PACKET2		0x80000000
 #define RADEON_CP_PACKET3		0xC0000000
+#       define RADEON_CP_NOP                    0x00001000
+#       define RADEON_CP_NEXT_CHAR              0x00001900
+#       define RADEON_CP_PLY_NEXTSCAN           0x00001D00
+#       define RADEON_CP_SET_SCISSORS           0x00001E00
+             /* GEN_INDX_PRIM is unsupported starting with R300 */
 #	define RADEON_3D_RNDR_GEN_INDX_PRIM	0x00002300
 #	define RADEON_WAIT_FOR_IDLE		0x00002600
 #	define RADEON_3D_DRAW_VBUF		0x00002800
 #	define RADEON_3D_DRAW_IMMD		0x00002900
 #	define RADEON_3D_DRAW_INDX		0x00002A00
+#       define RADEON_CP_LOAD_PALETTE           0x00002C00
 #	define RADEON_3D_LOAD_VBPNTR		0x00002F00
 #	define RADEON_MPEG_IDCT_MACROBLOCK	0x00003000
 #	define RADEON_MPEG_IDCT_MACROBLOCK_REV	0x00003100
 #	define RADEON_3D_CLEAR_ZMASK		0x00003200
+#	define RADEON_CP_INDX_BUFFER		0x00003300
+#       define RADEON_CP_3D_DRAW_VBUF_2         0x00003400
+#       define RADEON_CP_3D_DRAW_IMMD_2         0x00003500
+#       define RADEON_CP_3D_DRAW_INDX_2         0x00003600
 #	define RADEON_3D_CLEAR_HIZ		0x00003700
+#       define RADEON_CP_3D_CLEAR_CMASK         0x00003802
 #	define RADEON_CNTL_HOSTDATA_BLT		0x00009400
 #	define RADEON_CNTL_PAINT_MULTI		0x00009A00
 #	define RADEON_CNTL_BITBLT_MULTI		0x00009B00
diff --git a/drivers/char/drm/radeon_state.c b/drivers/char/drm/radeon_state.c
index 1f79e249146c..64a3e3a406ef 100644
--- a/drivers/char/drm/radeon_state.c
+++ b/drivers/char/drm/radeon_state.c
@@ -1493,7 +1493,7 @@ static void radeon_cp_dispatch_indices( drm_device_t *dev,
 
 }
 
-#define RADEON_MAX_TEXTURE_SIZE (RADEON_BUFFER_SIZE - 8 * sizeof(u32))
+#define RADEON_MAX_TEXTURE_SIZE RADEON_BUFFER_SIZE
 
 static int radeon_cp_dispatch_texture( DRMFILE filp,
 				       drm_device_t *dev,
@@ -1506,10 +1506,11 @@ static int radeon_cp_dispatch_texture( DRMFILE filp,
 	u32 format;
 	u32 *buffer;
 	const u8 __user *data;
-	int size, dwords, tex_width, blit_width;
+	int size, dwords, tex_width, blit_width, spitch;
 	u32 height;
 	int i;
 	u32 texpitch, microtile;
+	u32 offset;
 	RING_LOCALS;
 
 	DRM_GET_PRIV_WITH_RETURN( filp_priv, filp );
@@ -1530,17 +1531,6 @@ static int radeon_cp_dispatch_texture( DRMFILE filp,
 	RADEON_WAIT_UNTIL_IDLE();
 	ADVANCE_RING();
 
-#ifdef __BIG_ENDIAN
-	/* The Mesa texture functions provide the data in little endian as the
-	 * chip wants it, but we need to compensate for the fact that the CP
-	 * ring gets byte-swapped
-	 */
-	BEGIN_RING( 2 );
-	OUT_RING_REG( RADEON_RBBM_GUICNTL, RADEON_HOST_DATA_SWAP_32BIT );
-	ADVANCE_RING();
-#endif
-
-
 	/* 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.
@@ -1572,6 +1562,10 @@ static int radeon_cp_dispatch_texture( DRMFILE filp,
 		DRM_ERROR( "invalid texture format %d\n", tex->format );
 		return DRM_ERR(EINVAL);
 	}
+	spitch = blit_width >> 6;
+	if (spitch == 0 && image->height > 1)
+		return DRM_ERR(EINVAL);
+
 	texpitch = tex->pitch;
 	if ((texpitch << 22) & RADEON_DST_TILE_MICRO) {
 		microtile = 1;
@@ -1624,25 +1618,6 @@ static int radeon_cp_dispatch_texture( DRMFILE filp,
 		 */
 		buffer = (u32*)((char*)dev->agp_buffer_map->handle + buf->offset);
 		dwords = size / 4;
-		buffer[0] = CP_PACKET3( RADEON_CNTL_HOSTDATA_BLT, dwords + 6 );
-		buffer[1] = (RADEON_GMC_DST_PITCH_OFFSET_CNTL |
-			     RADEON_GMC_BRUSH_NONE |
-			     (format << 8) |
-			     RADEON_GMC_SRC_DATATYPE_COLOR |
-			     RADEON_ROP3_S |
-			     RADEON_DP_SRC_SOURCE_HOST_DATA |
-			     RADEON_GMC_CLR_CMP_CNTL_DIS |
-			     RADEON_GMC_WR_MSK_DIS);
-		
-		buffer[2] = (texpitch << 22) | (tex->offset >> 10);
-		buffer[3] = 0xffffffff;
-		buffer[4] = 0xffffffff;
-		buffer[5] = (image->y << 16) | image->x;
-		buffer[6] = (height << 16) | image->width;
-		buffer[7] = dwords;
-		buffer += 8;
-
-		
 
 		if (microtile) {
 			/* texture micro tiling in use, minimum texture width is thus 16 bytes.
@@ -1750,9 +1725,28 @@ static int radeon_cp_dispatch_texture( DRMFILE filp,
 		}
 
 		buf->filp = filp;
-		buf->used = (dwords + 8) * sizeof(u32);
-		radeon_cp_dispatch_indirect( dev, buf, 0, buf->used );
-		radeon_cp_discard_buffer( dev, buf );
+		buf->used = size;
+		offset = dev_priv->gart_buffers_offset + buf->offset;
+		BEGIN_RING(9);
+		OUT_RING(CP_PACKET3(RADEON_CNTL_BITBLT_MULTI, 5));
+		OUT_RING(RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
+			 RADEON_GMC_DST_PITCH_OFFSET_CNTL |
+			 RADEON_GMC_BRUSH_NONE |
+			 (format << 8) |
+			 RADEON_GMC_SRC_DATATYPE_COLOR |
+			 RADEON_ROP3_S |
+			 RADEON_DP_SRC_SOURCE_MEMORY |
+			 RADEON_GMC_CLR_CMP_CNTL_DIS |
+			 RADEON_GMC_WR_MSK_DIS );
+		OUT_RING((spitch << 22) | (offset >> 10));
+		OUT_RING((texpitch << 22) | (tex->offset >> 10));
+		OUT_RING(0);
+		OUT_RING((image->x << 16) | image->y);
+		OUT_RING((image->width << 16) | height);
+		RADEON_WAIT_UNTIL_2D_IDLE();
+		ADVANCE_RING();
+
+		radeon_cp_discard_buffer(dev, buf);
 
 		/* Update the input parameters for next time */
 		image->y += height;
@@ -2797,6 +2791,17 @@ static int radeon_cp_cmdbuf( DRM_IOCTL_ARGS )
 
 	orig_nbox = cmdbuf.nbox;
 
+	if(dev_priv->microcode_version == UCODE_R300) {
+		int temp;
+		temp=r300_do_cp_cmdbuf(dev, filp, filp_priv, &cmdbuf);
+	
+		if (orig_bufsz != 0)
+			drm_free(kbuf, orig_bufsz, DRM_MEM_DRIVER);
+	
+		return temp;
+	}
+
+	/* microcode_version != r300 */
 	while ( cmdbuf.bufsz >= sizeof(header) ) {
 
 		header.i = *(int *)cmdbuf.buf;
diff --git a/drivers/char/drm/savage_bci.c b/drivers/char/drm/savage_bci.c
new file mode 100644
index 000000000000..2fd40bac7c97
--- /dev/null
+++ b/drivers/char/drm/savage_bci.c
@@ -0,0 +1,1096 @@
+/* savage_bci.c -- BCI support for Savage
+ *
+ * Copyright 2004  Felix Kuehling
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sub license,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NON-INFRINGEMENT. IN NO EVENT SHALL FELIX KUEHLING BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF
+ * CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+#include "drmP.h"
+#include "savage_drm.h"
+#include "savage_drv.h"
+
+/* Need a long timeout for shadow status updates can take a while
+ * and so can waiting for events when the queue is full. */
+#define SAVAGE_DEFAULT_USEC_TIMEOUT	1000000 /* 1s */
+#define SAVAGE_EVENT_USEC_TIMEOUT	5000000 /* 5s */
+#define SAVAGE_FREELIST_DEBUG		0
+
+static int
+savage_bci_wait_fifo_shadow(drm_savage_private_t *dev_priv, unsigned int n)
+{
+	uint32_t mask = dev_priv->status_used_mask;
+	uint32_t threshold = dev_priv->bci_threshold_hi;
+	uint32_t status;
+	int i;
+
+#if SAVAGE_BCI_DEBUG
+	if (n > dev_priv->cob_size + SAVAGE_BCI_FIFO_SIZE - threshold)
+		DRM_ERROR("Trying to emit %d words "
+			  "(more than guaranteed space in COB)\n", n);
+#endif
+
+	for (i = 0; i < SAVAGE_DEFAULT_USEC_TIMEOUT; i++) {
+		DRM_MEMORYBARRIER();
+		status = dev_priv->status_ptr[0];
+		if ((status & mask) < threshold)
+			return 0;
+		DRM_UDELAY(1);
+	}
+
+#if SAVAGE_BCI_DEBUG
+	DRM_ERROR("failed!\n");
+	DRM_INFO("   status=0x%08x, threshold=0x%08x\n", status, threshold);
+#endif
+	return DRM_ERR(EBUSY);
+}
+
+static int
+savage_bci_wait_fifo_s3d(drm_savage_private_t *dev_priv, unsigned int n)
+{
+	uint32_t maxUsed = dev_priv->cob_size + SAVAGE_BCI_FIFO_SIZE - n;
+	uint32_t status;
+	int i;
+
+	for (i = 0; i < SAVAGE_DEFAULT_USEC_TIMEOUT; i++) {
+		status = SAVAGE_READ(SAVAGE_STATUS_WORD0);
+		if ((status & SAVAGE_FIFO_USED_MASK_S3D) <= maxUsed)
+			return 0;
+		DRM_UDELAY(1);
+	}
+
+#if SAVAGE_BCI_DEBUG
+	DRM_ERROR("failed!\n");
+	DRM_INFO("   status=0x%08x\n", status);
+#endif
+	return DRM_ERR(EBUSY);
+}
+
+static int
+savage_bci_wait_fifo_s4(drm_savage_private_t *dev_priv, unsigned int n)
+{
+	uint32_t maxUsed = dev_priv->cob_size + SAVAGE_BCI_FIFO_SIZE - n;
+	uint32_t status;
+	int i;
+
+	for (i = 0; i < SAVAGE_DEFAULT_USEC_TIMEOUT; i++) {
+		status = SAVAGE_READ(SAVAGE_ALT_STATUS_WORD0);
+		if ((status & SAVAGE_FIFO_USED_MASK_S4) <= maxUsed)
+			return 0;
+		DRM_UDELAY(1);
+	}
+
+#if SAVAGE_BCI_DEBUG
+	DRM_ERROR("failed!\n");
+	DRM_INFO("   status=0x%08x\n", status);
+#endif
+	return DRM_ERR(EBUSY);
+}
+
+/*
+ * Waiting for events.
+ *
+ * The BIOSresets the event tag to 0 on mode changes. Therefore we
+ * never emit 0 to the event tag. If we find a 0 event tag we know the
+ * BIOS stomped on it and return success assuming that the BIOS waited
+ * for engine idle.
+ *
+ * Note: if the Xserver uses the event tag it has to follow the same
+ * rule. Otherwise there may be glitches every 2^16 events.
+ */
+static int
+savage_bci_wait_event_shadow(drm_savage_private_t *dev_priv, uint16_t e)
+{
+	uint32_t status;
+	int i;
+
+	for (i = 0; i < SAVAGE_EVENT_USEC_TIMEOUT; i++) {
+		DRM_MEMORYBARRIER();
+		status = dev_priv->status_ptr[1];
+		if ((((status & 0xffff) - e) & 0xffff) <= 0x7fff ||
+		    (status & 0xffff) == 0)
+			return 0;
+		DRM_UDELAY(1);
+	}
+
+#if SAVAGE_BCI_DEBUG
+	DRM_ERROR("failed!\n");
+	DRM_INFO("   status=0x%08x, e=0x%04x\n", status, e);
+#endif
+
+	return DRM_ERR(EBUSY);
+}
+
+static int
+savage_bci_wait_event_reg(drm_savage_private_t *dev_priv, uint16_t e)
+{
+	uint32_t status;
+	int i;
+
+	for (i = 0; i < SAVAGE_EVENT_USEC_TIMEOUT; i++) {
+		status = SAVAGE_READ(SAVAGE_STATUS_WORD1);
+		if ((((status & 0xffff) - e) & 0xffff) <= 0x7fff ||
+		    (status & 0xffff) == 0)
+			return 0;
+		DRM_UDELAY(1);
+	}
+
+#if SAVAGE_BCI_DEBUG
+	DRM_ERROR("failed!\n");
+	DRM_INFO("   status=0x%08x, e=0x%04x\n", status, e);
+#endif
+
+	return DRM_ERR(EBUSY);
+}
+
+uint16_t savage_bci_emit_event(drm_savage_private_t *dev_priv,
+			       unsigned int flags)
+{
+	uint16_t count;
+	BCI_LOCALS;
+
+	if (dev_priv->status_ptr) {
+		/* coordinate with Xserver */
+		count = dev_priv->status_ptr[1023];
+		if (count < dev_priv->event_counter)
+			dev_priv->event_wrap++;
+	} else {
+		count = dev_priv->event_counter;
+	}
+	count = (count + 1) & 0xffff;
+	if (count == 0) {
+		count++; /* See the comment above savage_wait_event_*. */
+		dev_priv->event_wrap++;
+	}
+	dev_priv->event_counter = count;
+	if (dev_priv->status_ptr)
+		dev_priv->status_ptr[1023] = (uint32_t)count;
+
+	if ((flags & (SAVAGE_WAIT_2D | SAVAGE_WAIT_3D))) {
+		unsigned int wait_cmd = BCI_CMD_WAIT;
+		if ((flags & SAVAGE_WAIT_2D))
+			wait_cmd |= BCI_CMD_WAIT_2D;
+		if ((flags & SAVAGE_WAIT_3D))
+			wait_cmd |= BCI_CMD_WAIT_3D;
+		BEGIN_BCI(2);
+		BCI_WRITE(wait_cmd);
+	} else {
+		BEGIN_BCI(1);
+	}
+	BCI_WRITE(BCI_CMD_UPDATE_EVENT_TAG | (uint32_t)count);
+
+	return count;
+}
+
+/*
+ * Freelist management
+ */
+static int savage_freelist_init(drm_device_t *dev)
+{
+	drm_savage_private_t *dev_priv = dev->dev_private;
+	drm_device_dma_t *dma = dev->dma;
+	drm_buf_t *buf;
+	drm_savage_buf_priv_t *entry;
+	int i;
+	DRM_DEBUG("count=%d\n", dma->buf_count);
+
+	dev_priv->head.next = &dev_priv->tail;
+	dev_priv->head.prev = NULL;
+	dev_priv->head.buf = NULL;
+
+	dev_priv->tail.next = NULL;
+	dev_priv->tail.prev = &dev_priv->head;
+	dev_priv->tail.buf = NULL;
+
+	for (i = 0; i < dma->buf_count; i++) {
+		buf = dma->buflist[i];
+		entry = buf->dev_private;
+
+		SET_AGE(&entry->age, 0, 0);
+		entry->buf = buf;
+
+		entry->next = dev_priv->head.next;
+		entry->prev = &dev_priv->head;
+		dev_priv->head.next->prev = entry;
+		dev_priv->head.next = entry;
+	}
+
+	return 0;
+}
+
+static drm_buf_t *savage_freelist_get(drm_device_t *dev)
+{
+	drm_savage_private_t *dev_priv = dev->dev_private;
+	drm_savage_buf_priv_t *tail = dev_priv->tail.prev;
+	uint16_t event;
+	unsigned int wrap;
+	DRM_DEBUG("\n");
+
+	UPDATE_EVENT_COUNTER();
+	if (dev_priv->status_ptr)
+		event = dev_priv->status_ptr[1] & 0xffff;
+	else
+		event = SAVAGE_READ(SAVAGE_STATUS_WORD1) & 0xffff;
+	wrap = dev_priv->event_wrap;
+	if (event > dev_priv->event_counter)
+		wrap--; /* hardware hasn't passed the last wrap yet */
+
+	DRM_DEBUG("   tail=0x%04x %d\n", tail->age.event, tail->age.wrap);
+	DRM_DEBUG("   head=0x%04x %d\n", event, wrap);
+
+	if (tail->buf && (TEST_AGE(&tail->age, event, wrap) || event == 0)) {
+		drm_savage_buf_priv_t *next = tail->next;
+		drm_savage_buf_priv_t *prev = tail->prev;
+		prev->next = next;
+		next->prev = prev;
+		tail->next = tail->prev = NULL;
+		return tail->buf;
+	}
+
+	DRM_DEBUG("returning NULL, tail->buf=%p!\n", tail->buf);
+	return NULL;
+}
+
+void savage_freelist_put(drm_device_t *dev, drm_buf_t *buf)
+{
+	drm_savage_private_t *dev_priv = dev->dev_private;
+	drm_savage_buf_priv_t *entry = buf->dev_private, *prev, *next;
+
+	DRM_DEBUG("age=0x%04x wrap=%d\n", entry->age.event, entry->age.wrap);
+
+	if (entry->next != NULL || entry->prev != NULL) {
+		DRM_ERROR("entry already on freelist.\n");
+		return;
+	}
+
+	prev = &dev_priv->head;
+	next = prev->next;
+	prev->next = entry;
+	next->prev = entry;
+	entry->prev = prev;
+	entry->next = next;
+}
+
+/*
+ * Command DMA
+ */
+static int savage_dma_init(drm_savage_private_t *dev_priv)
+{
+	unsigned int i;
+
+	dev_priv->nr_dma_pages = dev_priv->cmd_dma->size /
+		(SAVAGE_DMA_PAGE_SIZE*4);
+	dev_priv->dma_pages = drm_alloc(sizeof(drm_savage_dma_page_t) *
+					dev_priv->nr_dma_pages,
+					DRM_MEM_DRIVER);
+	if (dev_priv->dma_pages == NULL)
+		return DRM_ERR(ENOMEM);
+
+	for (i = 0; i < dev_priv->nr_dma_pages; ++i) {
+		SET_AGE(&dev_priv->dma_pages[i].age, 0, 0);
+		dev_priv->dma_pages[i].used = 0;
+		dev_priv->dma_pages[i].flushed = 0;
+	}
+	SET_AGE(&dev_priv->last_dma_age, 0, 0);
+
+	dev_priv->first_dma_page = 0;
+	dev_priv->current_dma_page = 0;
+
+	return 0;
+}
+
+void savage_dma_reset(drm_savage_private_t *dev_priv)
+{
+	uint16_t event;
+	unsigned int wrap, i;
+	event = savage_bci_emit_event(dev_priv, 0);
+	wrap = dev_priv->event_wrap;
+	for (i = 0; i < dev_priv->nr_dma_pages; ++i) {
+		SET_AGE(&dev_priv->dma_pages[i].age, event, wrap);
+		dev_priv->dma_pages[i].used = 0;
+		dev_priv->dma_pages[i].flushed = 0;
+	}
+	SET_AGE(&dev_priv->last_dma_age, event, wrap);
+	dev_priv->first_dma_page = dev_priv->current_dma_page = 0;
+}
+
+void savage_dma_wait(drm_savage_private_t *dev_priv, unsigned int page)
+{
+	uint16_t event;
+	unsigned int wrap;
+
+	/* Faked DMA buffer pages don't age. */
+	if (dev_priv->cmd_dma == &dev_priv->fake_dma)
+		return;
+
+	UPDATE_EVENT_COUNTER();
+	if (dev_priv->status_ptr)
+		event = dev_priv->status_ptr[1] & 0xffff;
+	else
+		event = SAVAGE_READ(SAVAGE_STATUS_WORD1) & 0xffff;
+	wrap = dev_priv->event_wrap;
+	if (event > dev_priv->event_counter)
+		wrap--; /* hardware hasn't passed the last wrap yet */
+
+	if (dev_priv->dma_pages[page].age.wrap > wrap ||
+	    (dev_priv->dma_pages[page].age.wrap == wrap &&
+	     dev_priv->dma_pages[page].age.event > event)) {
+		if (dev_priv->wait_evnt(dev_priv,
+					dev_priv->dma_pages[page].age.event)
+		    < 0)
+			DRM_ERROR("wait_evnt failed!\n");
+	}
+}
+
+uint32_t *savage_dma_alloc(drm_savage_private_t *dev_priv, unsigned int n)
+{
+	unsigned int cur = dev_priv->current_dma_page;
+	unsigned int rest = SAVAGE_DMA_PAGE_SIZE -
+		dev_priv->dma_pages[cur].used;
+	unsigned int nr_pages = (n - rest + SAVAGE_DMA_PAGE_SIZE-1) /
+		SAVAGE_DMA_PAGE_SIZE;
+	uint32_t *dma_ptr;
+	unsigned int i;
+
+	DRM_DEBUG("cur=%u, cur->used=%u, n=%u, rest=%u, nr_pages=%u\n",
+		  cur, dev_priv->dma_pages[cur].used, n, rest, nr_pages);
+
+	if (cur + nr_pages < dev_priv->nr_dma_pages) {
+		dma_ptr = (uint32_t *)dev_priv->cmd_dma->handle +
+			cur*SAVAGE_DMA_PAGE_SIZE +
+			dev_priv->dma_pages[cur].used;
+		if (n < rest)
+			rest = n;
+		dev_priv->dma_pages[cur].used += rest;
+		n -= rest;
+		cur++;
+	} else {
+		dev_priv->dma_flush(dev_priv);
+		nr_pages = (n + SAVAGE_DMA_PAGE_SIZE-1) / SAVAGE_DMA_PAGE_SIZE;
+		for (i = cur; i < dev_priv->nr_dma_pages; ++i) {
+			dev_priv->dma_pages[i].age = dev_priv->last_dma_age;
+			dev_priv->dma_pages[i].used = 0;
+			dev_priv->dma_pages[i].flushed = 0;
+		}
+		dma_ptr = (uint32_t *)dev_priv->cmd_dma->handle;
+		dev_priv->first_dma_page = cur = 0;
+	}
+	for (i = cur; nr_pages > 0; ++i, --nr_pages) {
+#if SAVAGE_DMA_DEBUG
+		if (dev_priv->dma_pages[i].used) {
+			DRM_ERROR("unflushed page %u: used=%u\n",
+				  i, dev_priv->dma_pages[i].used);
+		}
+#endif
+		if (n > SAVAGE_DMA_PAGE_SIZE)
+			dev_priv->dma_pages[i].used = SAVAGE_DMA_PAGE_SIZE;
+		else
+			dev_priv->dma_pages[i].used = n;
+		n -= SAVAGE_DMA_PAGE_SIZE;
+	}
+	dev_priv->current_dma_page = --i;
+
+	DRM_DEBUG("cur=%u, cur->used=%u, n=%u\n",
+		  i, dev_priv->dma_pages[i].used, n);
+
+	savage_dma_wait(dev_priv, dev_priv->current_dma_page);
+
+	return dma_ptr;
+}
+
+static void savage_dma_flush(drm_savage_private_t *dev_priv)
+{
+	unsigned int first = dev_priv->first_dma_page;
+	unsigned int cur = dev_priv->current_dma_page;
+	uint16_t event;
+	unsigned int wrap, pad, align, len, i;
+	unsigned long phys_addr;
+	BCI_LOCALS;
+
+	if (first == cur &&
+	    dev_priv->dma_pages[cur].used == dev_priv->dma_pages[cur].flushed)
+		return;
+
+	/* pad length to multiples of 2 entries
+	 * align start of next DMA block to multiles of 8 entries */
+	pad = -dev_priv->dma_pages[cur].used & 1;
+	align = -(dev_priv->dma_pages[cur].used + pad) & 7;
+
+	DRM_DEBUG("first=%u, cur=%u, first->flushed=%u, cur->used=%u, "
+		  "pad=%u, align=%u\n",
+		  first, cur, dev_priv->dma_pages[first].flushed,
+		  dev_priv->dma_pages[cur].used, pad, align);
+
+	/* pad with noops */
+	if (pad) {
+		uint32_t *dma_ptr = (uint32_t *)dev_priv->cmd_dma->handle +
+			cur * SAVAGE_DMA_PAGE_SIZE +
+			dev_priv->dma_pages[cur].used;
+		dev_priv->dma_pages[cur].used += pad;
+		while(pad != 0) {
+			*dma_ptr++ = BCI_CMD_WAIT;
+			pad--;
+		}
+	}
+
+	DRM_MEMORYBARRIER();
+
+	/* do flush ... */
+	phys_addr = dev_priv->cmd_dma->offset +
+		(first * SAVAGE_DMA_PAGE_SIZE +
+		 dev_priv->dma_pages[first].flushed) * 4;
+	len = (cur - first) * SAVAGE_DMA_PAGE_SIZE +
+		dev_priv->dma_pages[cur].used -
+		dev_priv->dma_pages[first].flushed;
+
+	DRM_DEBUG("phys_addr=%lx, len=%u\n",
+		  phys_addr | dev_priv->dma_type, len);
+
+	BEGIN_BCI(3);
+	BCI_SET_REGISTERS(SAVAGE_DMABUFADDR, 1);
+	BCI_WRITE(phys_addr | dev_priv->dma_type);
+	BCI_DMA(len);
+
+	/* fix alignment of the start of the next block */
+	dev_priv->dma_pages[cur].used += align;
+
+	/* age DMA pages */
+	event = savage_bci_emit_event(dev_priv, 0);
+	wrap = dev_priv->event_wrap;
+	for (i = first; i < cur; ++i) {
+		SET_AGE(&dev_priv->dma_pages[i].age, event, wrap);
+		dev_priv->dma_pages[i].used = 0;
+		dev_priv->dma_pages[i].flushed = 0;
+	}
+	/* age the current page only when it's full */
+	if (dev_priv->dma_pages[cur].used == SAVAGE_DMA_PAGE_SIZE) {
+		SET_AGE(&dev_priv->dma_pages[cur].age, event, wrap);
+		dev_priv->dma_pages[cur].used = 0;
+		dev_priv->dma_pages[cur].flushed = 0;
+		/* advance to next page */
+		cur++;
+		if (cur == dev_priv->nr_dma_pages)
+			cur = 0;
+		dev_priv->first_dma_page = dev_priv->current_dma_page = cur;
+	} else {
+		dev_priv->first_dma_page = cur;
+		dev_priv->dma_pages[cur].flushed = dev_priv->dma_pages[i].used;
+	}
+	SET_AGE(&dev_priv->last_dma_age, event, wrap);
+
+	DRM_DEBUG("first=cur=%u, cur->used=%u, cur->flushed=%u\n", cur,
+		  dev_priv->dma_pages[cur].used,
+		  dev_priv->dma_pages[cur].flushed);
+}
+
+static void savage_fake_dma_flush(drm_savage_private_t *dev_priv)
+{
+	unsigned int i, j;
+	BCI_LOCALS;
+
+	if (dev_priv->first_dma_page == dev_priv->current_dma_page &&
+	    dev_priv->dma_pages[dev_priv->current_dma_page].used == 0)
+		return;
+
+	DRM_DEBUG("first=%u, cur=%u, cur->used=%u\n",
+		  dev_priv->first_dma_page, dev_priv->current_dma_page,
+		  dev_priv->dma_pages[dev_priv->current_dma_page].used);
+
+	for (i = dev_priv->first_dma_page;
+	     i <= dev_priv->current_dma_page && dev_priv->dma_pages[i].used;
+	     ++i) {
+		uint32_t *dma_ptr = (uint32_t *)dev_priv->cmd_dma->handle +
+			i * SAVAGE_DMA_PAGE_SIZE;
+#if SAVAGE_DMA_DEBUG
+		/* Sanity check: all pages except the last one must be full. */
+		if (i < dev_priv->current_dma_page &&
+		    dev_priv->dma_pages[i].used != SAVAGE_DMA_PAGE_SIZE) {
+			DRM_ERROR("partial DMA page %u: used=%u",
+				  i, dev_priv->dma_pages[i].used);
+		}
+#endif
+		BEGIN_BCI(dev_priv->dma_pages[i].used);
+		for (j = 0; j < dev_priv->dma_pages[i].used; ++j) {
+			BCI_WRITE(dma_ptr[j]);
+		}
+		dev_priv->dma_pages[i].used = 0;
+	}
+
+	/* reset to first page */
+	dev_priv->first_dma_page = dev_priv->current_dma_page = 0;
+}
+
+/*
+ * Initalize mappings. On Savage4 and SavageIX the alignment
+ * and size of the aperture is not suitable for automatic MTRR setup
+ * in drm_addmap. Therefore we do it manually before the maps are
+ * initialized. We also need to take care of deleting the MTRRs in
+ * postcleanup.
+ */
+int savage_preinit(drm_device_t *dev, unsigned long chipset)
+{
+	drm_savage_private_t *dev_priv;
+	unsigned long mmio_base, fb_base, fb_size, aperture_base;
+	/* fb_rsrc and aper_rsrc aren't really used currently, but still exist
+	 * in case we decide we need information on the BAR for BSD in the
+	 * future.
+	 */
+	unsigned int fb_rsrc, aper_rsrc;
+	int ret = 0;
+
+	dev_priv = drm_alloc(sizeof(drm_savage_private_t), DRM_MEM_DRIVER);
+	if (dev_priv == NULL)
+		return DRM_ERR(ENOMEM);
+
+	memset(dev_priv, 0, sizeof(drm_savage_private_t));
+	dev->dev_private = (void *)dev_priv;
+	dev_priv->chipset = (enum savage_family)chipset;
+
+	dev_priv->mtrr[0].handle = -1;
+	dev_priv->mtrr[1].handle = -1;
+	dev_priv->mtrr[2].handle = -1;
+	if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
+		fb_rsrc = 0;
+		fb_base = drm_get_resource_start(dev, 0);
+		fb_size = SAVAGE_FB_SIZE_S3;
+		mmio_base = fb_base + SAVAGE_FB_SIZE_S3;
+		aper_rsrc = 0;
+		aperture_base = fb_base + SAVAGE_APERTURE_OFFSET;
+		/* this should always be true */
+		if (drm_get_resource_len(dev, 0) == 0x08000000) {
+			/* Don't make MMIO write-cobining! We need 3
+			 * MTRRs. */
+			dev_priv->mtrr[0].base = fb_base;
+			dev_priv->mtrr[0].size = 0x01000000;
+			dev_priv->mtrr[0].handle = mtrr_add(
+				dev_priv->mtrr[0].base, dev_priv->mtrr[0].size,
+				MTRR_TYPE_WRCOMB, 1);
+			dev_priv->mtrr[1].base = fb_base+0x02000000;
+			dev_priv->mtrr[1].size = 0x02000000;
+			dev_priv->mtrr[1].handle = mtrr_add(
+				dev_priv->mtrr[1].base, dev_priv->mtrr[1].size,
+				MTRR_TYPE_WRCOMB, 1);
+			dev_priv->mtrr[2].base = fb_base+0x04000000;
+			dev_priv->mtrr[2].size = 0x04000000;
+			dev_priv->mtrr[2].handle = mtrr_add(
+				dev_priv->mtrr[2].base, dev_priv->mtrr[2].size,
+				MTRR_TYPE_WRCOMB, 1);
+		} else {
+			DRM_ERROR("strange pci_resource_len %08lx\n",
+				  drm_get_resource_len(dev, 0));
+		}
+	} else if (chipset != S3_SUPERSAVAGE && chipset != S3_SAVAGE2000) {
+		mmio_base = drm_get_resource_start(dev, 0);
+		fb_rsrc = 1;
+		fb_base = drm_get_resource_start(dev, 1);
+		fb_size = SAVAGE_FB_SIZE_S4;
+		aper_rsrc = 1;
+		aperture_base = fb_base + SAVAGE_APERTURE_OFFSET;
+		/* this should always be true */
+		if (drm_get_resource_len(dev, 1) == 0x08000000) {
+			/* Can use one MTRR to cover both fb and
+			 * aperture. */
+			dev_priv->mtrr[0].base = fb_base;
+			dev_priv->mtrr[0].size = 0x08000000;
+			dev_priv->mtrr[0].handle = mtrr_add(
+				dev_priv->mtrr[0].base, dev_priv->mtrr[0].size,
+				MTRR_TYPE_WRCOMB, 1);
+		} else {
+			DRM_ERROR("strange pci_resource_len %08lx\n",
+				  drm_get_resource_len(dev, 1));
+		}
+	} else {
+		mmio_base = drm_get_resource_start(dev, 0);
+		fb_rsrc = 1;
+		fb_base = drm_get_resource_start(dev, 1);
+		fb_size = drm_get_resource_len(dev, 1);
+		aper_rsrc = 2;
+		aperture_base = drm_get_resource_start(dev, 2);
+		/* Automatic MTRR setup will do the right thing. */
+	}
+
+	ret = drm_addmap(dev, mmio_base, SAVAGE_MMIO_SIZE, _DRM_REGISTERS,
+			 _DRM_READ_ONLY, &dev_priv->mmio);
+	if (ret)
+		return ret;
+
+	ret = drm_addmap(dev, fb_base, fb_size, _DRM_FRAME_BUFFER,
+			 _DRM_WRITE_COMBINING, &dev_priv->fb);
+	if (ret)
+		return ret;
+
+	ret = drm_addmap(dev, aperture_base, SAVAGE_APERTURE_SIZE,
+			 _DRM_FRAME_BUFFER, _DRM_WRITE_COMBINING,
+			 &dev_priv->aperture);
+	if (ret)
+		return ret;
+
+	return ret;
+}
+
+/*
+ * Delete MTRRs and free device-private data.
+ */
+int savage_postcleanup(drm_device_t *dev)
+{
+	drm_savage_private_t *dev_priv = dev->dev_private;
+	int i;
+
+	for (i = 0; i < 3; ++i)
+		if (dev_priv->mtrr[i].handle >= 0)
+			mtrr_del(dev_priv->mtrr[i].handle,
+				 dev_priv->mtrr[i].base,
+				 dev_priv->mtrr[i].size);
+
+	drm_free(dev_priv, sizeof(drm_savage_private_t), DRM_MEM_DRIVER);
+
+	return 0;
+}
+
+static int savage_do_init_bci(drm_device_t *dev, drm_savage_init_t *init)
+{
+	drm_savage_private_t *dev_priv = dev->dev_private;
+
+	if (init->fb_bpp != 16 && init->fb_bpp != 32) {
+		DRM_ERROR("invalid frame buffer bpp %d!\n", init->fb_bpp);
+		return DRM_ERR(EINVAL);
+	}
+	if (init->depth_bpp != 16 && init->depth_bpp != 32) {
+		DRM_ERROR("invalid depth buffer bpp %d!\n", init->fb_bpp);
+		return DRM_ERR(EINVAL);
+	}
+	if (init->dma_type != SAVAGE_DMA_AGP &&
+	    init->dma_type != SAVAGE_DMA_PCI) {
+		DRM_ERROR("invalid dma memory type %d!\n", init->dma_type);
+		return DRM_ERR(EINVAL);
+	}
+
+	dev_priv->cob_size = init->cob_size;
+	dev_priv->bci_threshold_lo = init->bci_threshold_lo;
+	dev_priv->bci_threshold_hi = init->bci_threshold_hi;
+	dev_priv->dma_type = init->dma_type;
+
+	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->texture_offset = init->texture_offset;
+	dev_priv->texture_size = init->texture_size;
+
+	DRM_GETSAREA();
+	if (!dev_priv->sarea) {
+		DRM_ERROR("could not find sarea!\n");
+		savage_do_cleanup_bci(dev);
+		return DRM_ERR(EINVAL);
+	}
+	if (init->status_offset != 0) {
+		dev_priv->status = drm_core_findmap(dev, init->status_offset);
+		if (!dev_priv->status) {
+			DRM_ERROR("could not find shadow status region!\n");
+			savage_do_cleanup_bci(dev);
+			return DRM_ERR(EINVAL);
+		}
+	} else {
+		dev_priv->status = NULL;
+	}
+	if (dev_priv->dma_type == SAVAGE_DMA_AGP && init->buffers_offset) {
+		dev->agp_buffer_map = drm_core_findmap(dev,
+						       init->buffers_offset);
+		if (!dev->agp_buffer_map) {
+			DRM_ERROR("could not find DMA buffer region!\n");
+			savage_do_cleanup_bci(dev);
+			return DRM_ERR(EINVAL);
+		}
+		drm_core_ioremap(dev->agp_buffer_map, dev);
+		if (!dev->agp_buffer_map) {
+			DRM_ERROR("failed to ioremap DMA buffer region!\n");
+			savage_do_cleanup_bci(dev);
+			return DRM_ERR(ENOMEM);
+		}
+	}
+	if (init->agp_textures_offset) {
+		dev_priv->agp_textures =
+			drm_core_findmap(dev, init->agp_textures_offset);
+		if (!dev_priv->agp_textures) {
+			DRM_ERROR("could not find agp texture region!\n");
+			savage_do_cleanup_bci(dev);
+			return DRM_ERR(EINVAL);
+		}
+	} else {
+		dev_priv->agp_textures = NULL;
+	}
+
+	if (init->cmd_dma_offset) {
+		if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
+			DRM_ERROR("command DMA not supported on "
+				  "Savage3D/MX/IX.\n");
+			savage_do_cleanup_bci(dev);
+			return DRM_ERR(EINVAL);
+		}
+		if (dev->dma && dev->dma->buflist) {
+			DRM_ERROR("command and vertex DMA not supported "
+				  "at the same time.\n");
+			savage_do_cleanup_bci(dev);
+			return DRM_ERR(EINVAL);
+		}
+		dev_priv->cmd_dma = drm_core_findmap(dev, init->cmd_dma_offset);
+		if (!dev_priv->cmd_dma) {
+			DRM_ERROR("could not find command DMA region!\n");
+			savage_do_cleanup_bci(dev);
+			return DRM_ERR(EINVAL);
+		}
+		if (dev_priv->dma_type == SAVAGE_DMA_AGP) {
+			if (dev_priv->cmd_dma->type != _DRM_AGP) {
+				DRM_ERROR("AGP command DMA region is not a "
+					  "_DRM_AGP map!\n");
+				savage_do_cleanup_bci(dev);
+				return DRM_ERR(EINVAL);
+			}
+			drm_core_ioremap(dev_priv->cmd_dma, dev);
+			if (!dev_priv->cmd_dma->handle) {
+				DRM_ERROR("failed to ioremap command "
+					  "DMA region!\n");
+				savage_do_cleanup_bci(dev);
+				return DRM_ERR(ENOMEM);
+			}
+		} else if (dev_priv->cmd_dma->type != _DRM_CONSISTENT) {
+			DRM_ERROR("PCI command DMA region is not a "
+				  "_DRM_CONSISTENT map!\n");
+			savage_do_cleanup_bci(dev);
+			return DRM_ERR(EINVAL);
+		}
+	} else {
+		dev_priv->cmd_dma = NULL;
+	}
+
+	dev_priv->dma_flush = savage_dma_flush;
+	if (!dev_priv->cmd_dma) {
+		DRM_DEBUG("falling back to faked command DMA.\n");
+		dev_priv->fake_dma.offset = 0;
+		dev_priv->fake_dma.size = SAVAGE_FAKE_DMA_SIZE;
+		dev_priv->fake_dma.type = _DRM_SHM;
+		dev_priv->fake_dma.handle = drm_alloc(SAVAGE_FAKE_DMA_SIZE,
+						      DRM_MEM_DRIVER);
+		if (!dev_priv->fake_dma.handle) {
+			DRM_ERROR("could not allocate faked DMA buffer!\n");
+			savage_do_cleanup_bci(dev);
+			return DRM_ERR(ENOMEM);
+		}
+		dev_priv->cmd_dma = &dev_priv->fake_dma;
+		dev_priv->dma_flush = savage_fake_dma_flush;
+	}
+
+	dev_priv->sarea_priv =
+		(drm_savage_sarea_t *)((uint8_t *)dev_priv->sarea->handle +
+				       init->sarea_priv_offset);
+
+	/* setup bitmap descriptors */
+	{
+		unsigned int color_tile_format;
+		unsigned int depth_tile_format;
+		unsigned int front_stride, back_stride, depth_stride;
+		if (dev_priv->chipset <= S3_SAVAGE4) {
+			color_tile_format = dev_priv->fb_bpp == 16 ?
+				SAVAGE_BD_TILE_16BPP : SAVAGE_BD_TILE_32BPP;
+			depth_tile_format = dev_priv->depth_bpp == 16 ?
+				SAVAGE_BD_TILE_16BPP : SAVAGE_BD_TILE_32BPP;
+		} else {
+			color_tile_format = SAVAGE_BD_TILE_DEST;
+			depth_tile_format = SAVAGE_BD_TILE_DEST;
+		}
+		front_stride = dev_priv->front_pitch / (dev_priv->fb_bpp/8);
+		back_stride  = dev_priv-> back_pitch / (dev_priv->fb_bpp/8);
+		depth_stride = dev_priv->depth_pitch / (dev_priv->depth_bpp/8);
+
+		dev_priv->front_bd = front_stride | SAVAGE_BD_BW_DISABLE |
+			(dev_priv->fb_bpp << SAVAGE_BD_BPP_SHIFT) |
+			(color_tile_format << SAVAGE_BD_TILE_SHIFT);
+
+		dev_priv-> back_bd =  back_stride | SAVAGE_BD_BW_DISABLE |
+			(dev_priv->fb_bpp << SAVAGE_BD_BPP_SHIFT) |
+			(color_tile_format << SAVAGE_BD_TILE_SHIFT);
+
+		dev_priv->depth_bd = depth_stride | SAVAGE_BD_BW_DISABLE |
+			(dev_priv->depth_bpp << SAVAGE_BD_BPP_SHIFT) |
+			(depth_tile_format << SAVAGE_BD_TILE_SHIFT);
+	}
+
+	/* setup status and bci ptr */
+	dev_priv->event_counter = 0;
+	dev_priv->event_wrap = 0;
+	dev_priv->bci_ptr = (volatile uint32_t *)
+	    ((uint8_t *)dev_priv->mmio->handle + SAVAGE_BCI_OFFSET);
+	if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
+		dev_priv->status_used_mask = SAVAGE_FIFO_USED_MASK_S3D;
+	} else {
+		dev_priv->status_used_mask = SAVAGE_FIFO_USED_MASK_S4;
+	}
+	if (dev_priv->status != NULL) {
+		dev_priv->status_ptr =
+			(volatile uint32_t *)dev_priv->status->handle;
+		dev_priv->wait_fifo = savage_bci_wait_fifo_shadow;
+		dev_priv->wait_evnt = savage_bci_wait_event_shadow;
+		dev_priv->status_ptr[1023] = dev_priv->event_counter;
+	} else {
+		dev_priv->status_ptr = NULL;
+		if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
+			dev_priv->wait_fifo = savage_bci_wait_fifo_s3d;
+		} else {
+			dev_priv->wait_fifo = savage_bci_wait_fifo_s4;
+		}
+		dev_priv->wait_evnt = savage_bci_wait_event_reg;
+	}
+
+	/* cliprect functions */
+	if (S3_SAVAGE3D_SERIES(dev_priv->chipset))
+		dev_priv->emit_clip_rect = savage_emit_clip_rect_s3d;
+	else
+		dev_priv->emit_clip_rect = savage_emit_clip_rect_s4;
+
+	if (savage_freelist_init(dev) < 0) {
+		DRM_ERROR("could not initialize freelist\n");
+		savage_do_cleanup_bci(dev);
+		return DRM_ERR(ENOMEM);
+	}
+
+	if (savage_dma_init(dev_priv) <  0) {
+		DRM_ERROR("could not initialize command DMA\n");
+		savage_do_cleanup_bci(dev);
+		return DRM_ERR(ENOMEM);
+	}
+
+	return 0;
+}
+
+int savage_do_cleanup_bci(drm_device_t *dev)
+{
+	drm_savage_private_t *dev_priv = dev->dev_private;
+
+	if (dev_priv->cmd_dma == &dev_priv->fake_dma) {
+		if (dev_priv->fake_dma.handle)
+			drm_free(dev_priv->fake_dma.handle,
+				 SAVAGE_FAKE_DMA_SIZE, DRM_MEM_DRIVER);
+	} else if (dev_priv->cmd_dma && dev_priv->cmd_dma->handle &&
+		   dev_priv->cmd_dma->type == _DRM_AGP &&
+		   dev_priv->dma_type == SAVAGE_DMA_AGP)
+		drm_core_ioremapfree(dev_priv->cmd_dma, dev);
+
+	if (dev_priv->dma_type == SAVAGE_DMA_AGP &&
+	    dev->agp_buffer_map && dev->agp_buffer_map->handle) {
+		drm_core_ioremapfree(dev->agp_buffer_map, dev);
+		/* make sure the next instance (which may be running
+		 * in PCI mode) doesn't try to use an old
+		 * agp_buffer_map. */
+		dev->agp_buffer_map = NULL;
+	}
+
+	if (dev_priv->dma_pages)
+		drm_free(dev_priv->dma_pages,
+			 sizeof(drm_savage_dma_page_t)*dev_priv->nr_dma_pages,
+			 DRM_MEM_DRIVER);
+
+	return 0;
+}
+
+static int savage_bci_init(DRM_IOCTL_ARGS)
+{
+	DRM_DEVICE;
+	drm_savage_init_t init;
+
+	LOCK_TEST_WITH_RETURN(dev, filp);
+
+	DRM_COPY_FROM_USER_IOCTL(init, (drm_savage_init_t __user *)data,
+				 sizeof(init));
+
+	switch (init.func) {
+	case SAVAGE_INIT_BCI:
+		return savage_do_init_bci(dev, &init);
+	case SAVAGE_CLEANUP_BCI:
+		return savage_do_cleanup_bci(dev);
+	}
+
+	return DRM_ERR(EINVAL);
+}
+
+static int savage_bci_event_emit(DRM_IOCTL_ARGS)
+{
+	DRM_DEVICE;
+	drm_savage_private_t *dev_priv = dev->dev_private;
+	drm_savage_event_emit_t event;
+
+	DRM_DEBUG("\n");
+
+	LOCK_TEST_WITH_RETURN(dev, filp);
+
+	DRM_COPY_FROM_USER_IOCTL(event, (drm_savage_event_emit_t __user *)data,
+				 sizeof(event));
+
+	event.count = savage_bci_emit_event(dev_priv, event.flags);
+	event.count |= dev_priv->event_wrap << 16;
+	DRM_COPY_TO_USER_IOCTL(&((drm_savage_event_emit_t __user *)data)->count,
+			       event.count, sizeof(event.count));
+	return 0;
+}
+
+static int savage_bci_event_wait(DRM_IOCTL_ARGS)
+{
+	DRM_DEVICE;
+	drm_savage_private_t *dev_priv = dev->dev_private;
+	drm_savage_event_wait_t event;
+	unsigned int event_e, hw_e;
+	unsigned int event_w, hw_w;
+
+	DRM_DEBUG("\n");
+
+	DRM_COPY_FROM_USER_IOCTL(event, (drm_savage_event_wait_t __user *)data,
+				 sizeof(event));
+
+	UPDATE_EVENT_COUNTER();
+	if (dev_priv->status_ptr)
+		hw_e = dev_priv->status_ptr[1] & 0xffff;
+	else
+		hw_e = SAVAGE_READ(SAVAGE_STATUS_WORD1) & 0xffff;
+	hw_w = dev_priv->event_wrap;
+	if (hw_e > dev_priv->event_counter)
+		hw_w--; /* hardware hasn't passed the last wrap yet */
+
+	event_e = event.count & 0xffff;
+	event_w = event.count >> 16;
+
+	/* Don't need to wait if
+	 * - event counter wrapped since the event was emitted or
+	 * - the hardware has advanced up to or over the event to wait for.
+	 */
+	if (event_w < hw_w || (event_w == hw_w && event_e <= hw_e) )
+		return 0;
+	else
+		return dev_priv->wait_evnt(dev_priv, event_e);
+}
+
+/*
+ * DMA buffer management
+ */
+
+static int savage_bci_get_buffers(DRMFILE filp, drm_device_t *dev, drm_dma_t *d)
+{
+	drm_buf_t *buf;
+	int i;
+
+	for (i = d->granted_count; i < d->request_count; i++) {
+		buf = savage_freelist_get(dev);
+		if (!buf)
+			return DRM_ERR(EAGAIN);
+
+		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 savage_bci_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 __user *)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);
+		return DRM_ERR(EINVAL);
+	}
+
+	d.granted_count = 0;
+
+	if (d.request_count) {
+		ret = savage_bci_get_buffers(filp, dev, &d);
+	}
+
+	DRM_COPY_TO_USER_IOCTL((drm_dma_t __user *)data, d, sizeof(d));
+
+	return ret;
+}
+
+void savage_reclaim_buffers(drm_device_t *dev, DRMFILE filp) {
+	drm_device_dma_t *dma = dev->dma;
+	drm_savage_private_t *dev_priv = dev->dev_private;
+	int i;
+
+	if (!dma)
+		return;
+	if (!dev_priv)
+		return;
+	if (!dma->buflist)
+		return;
+
+	/*i830_flush_queue(dev);*/
+
+	for (i = 0; i < dma->buf_count; i++) {
+		drm_buf_t *buf = dma->buflist[i];
+		drm_savage_buf_priv_t *buf_priv = buf->dev_private;
+
+		if (buf->filp == filp && buf_priv &&
+		    buf_priv->next == NULL && buf_priv->prev == NULL) {
+			uint16_t event;
+			DRM_DEBUG("reclaimed from client\n");
+			event = savage_bci_emit_event(dev_priv, SAVAGE_WAIT_3D);
+			SET_AGE(&buf_priv->age, event, dev_priv->event_wrap);
+			savage_freelist_put(dev, buf);
+		}
+	}
+
+	drm_core_reclaim_buffers(dev, filp);
+}
+
+
+drm_ioctl_desc_t savage_ioctls[] = {
+	[DRM_IOCTL_NR(DRM_SAVAGE_BCI_INIT)] = {savage_bci_init, 1, 1},
+	[DRM_IOCTL_NR(DRM_SAVAGE_BCI_CMDBUF)] = {savage_bci_cmdbuf, 1, 0},
+	[DRM_IOCTL_NR(DRM_SAVAGE_BCI_EVENT_EMIT)] = {savage_bci_event_emit, 1, 0},
+	[DRM_IOCTL_NR(DRM_SAVAGE_BCI_EVENT_WAIT)] = {savage_bci_event_wait, 1, 0},
+};
+
+int savage_max_ioctl = DRM_ARRAY_SIZE(savage_ioctls);
diff --git a/drivers/char/drm/savage_drm.h b/drivers/char/drm/savage_drm.h
new file mode 100644
index 000000000000..6526c9aa7589
--- /dev/null
+++ b/drivers/char/drm/savage_drm.h
@@ -0,0 +1,209 @@
+/* savage_drm.h -- Public header for the savage driver
+ *
+ * Copyright 2004  Felix Kuehling
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sub license,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NON-INFRINGEMENT. IN NO EVENT SHALL FELIX KUEHLING BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF
+ * CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#ifndef __SAVAGE_DRM_H__
+#define __SAVAGE_DRM_H__
+
+#ifndef __SAVAGE_SAREA_DEFINES__
+#define __SAVAGE_SAREA_DEFINES__
+
+/* 2 heaps (1 for card, 1 for agp), each divided into upto 128
+ * regions, subject to a minimum region size of (1<<16) == 64k.
+ *
+ * Clients may subdivide regions internally, but when sharing between
+ * clients, the region size is the minimum granularity.
+ */
+
+#define SAVAGE_CARD_HEAP		0
+#define SAVAGE_AGP_HEAP			1
+#define SAVAGE_NR_TEX_HEAPS		2
+#define SAVAGE_NR_TEX_REGIONS		16
+#define SAVAGE_LOG_MIN_TEX_REGION_SIZE	16
+
+#endif /* __SAVAGE_SAREA_DEFINES__ */
+
+typedef struct _drm_savage_sarea {
+	/* LRU lists for texture memory in agp space and on the card.
+	 */
+	drm_tex_region_t texList[SAVAGE_NR_TEX_HEAPS][SAVAGE_NR_TEX_REGIONS+1];
+	unsigned int texAge[SAVAGE_NR_TEX_HEAPS];
+
+	/* Mechanism to validate card state.
+	 */
+	int ctxOwner;
+} drm_savage_sarea_t, *drm_savage_sarea_ptr;
+
+/* Savage-specific ioctls
+ */
+#define DRM_SAVAGE_BCI_INIT		0x00
+#define DRM_SAVAGE_BCI_CMDBUF           0x01
+#define DRM_SAVAGE_BCI_EVENT_EMIT	0x02
+#define DRM_SAVAGE_BCI_EVENT_WAIT	0x03
+
+#define DRM_IOCTL_SAVAGE_INIT		DRM_IOW( DRM_COMMAND_BASE + DRM_SAVAGE_BCI_INIT, drm_savage_init_t)
+#define DRM_IOCTL_SAVAGE_CMDBUF		DRM_IOW( DRM_COMMAND_BASE + DRM_SAVAGE_BCI_CMDBUF, drm_savage_cmdbuf_t)
+#define DRM_IOCTL_SAVAGE_EVENT_EMIT	DRM_IOWR(DRM_COMMAND_BASE + DRM_SAVAGE_BCI_EVENT_EMIT, drm_savage_event_emit_t)
+#define DRM_IOCTL_SAVAGE_EVENT_WAIT	DRM_IOW( DRM_COMMAND_BASE + DRM_SAVAGE_BCI_EVENT_WAIT, drm_savage_event_wait_t)
+
+#define SAVAGE_DMA_PCI	1
+#define SAVAGE_DMA_AGP	3
+typedef struct drm_savage_init {
+	enum {
+		SAVAGE_INIT_BCI = 1,
+		SAVAGE_CLEANUP_BCI = 2
+	} func;
+	unsigned int sarea_priv_offset;
+
+	/* some parameters */
+	unsigned int cob_size;
+	unsigned int bci_threshold_lo, bci_threshold_hi;
+	unsigned int dma_type;
+
+	/* frame buffer layout */
+	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;
+
+	/* local textures */
+	unsigned int texture_offset;
+	unsigned int texture_size;
+
+	/* physical locations of non-permanent maps */
+	unsigned long status_offset;
+	unsigned long buffers_offset;
+	unsigned long agp_textures_offset;
+	unsigned long cmd_dma_offset;
+} drm_savage_init_t;
+
+typedef union drm_savage_cmd_header drm_savage_cmd_header_t;
+typedef struct drm_savage_cmdbuf {
+				/* command buffer in client's address space */
+	drm_savage_cmd_header_t __user *cmd_addr;
+	unsigned int size;	/* size of the command buffer in 64bit units */
+
+	unsigned int dma_idx;	/* DMA buffer index to use */
+	int discard;		/* discard DMA buffer when done */
+				/* vertex buffer in client's address space */
+	unsigned int __user *vb_addr;
+	unsigned int vb_size;	/* size of client vertex buffer in bytes */
+	unsigned int vb_stride;	/* stride of vertices in 32bit words */
+				/* boxes in client's address space */
+	drm_clip_rect_t __user *box_addr;
+	unsigned int nbox;	/* number of clipping boxes */
+} drm_savage_cmdbuf_t;
+
+#define SAVAGE_WAIT_2D  0x1 /* wait for 2D idle before updating event tag */
+#define SAVAGE_WAIT_3D  0x2 /* wait for 3D idle before updating event tag */
+#define SAVAGE_WAIT_IRQ 0x4 /* emit or wait for IRQ, not implemented yet */
+typedef struct drm_savage_event {
+	unsigned int count;
+	unsigned int flags;
+} drm_savage_event_emit_t, drm_savage_event_wait_t;
+
+/* Commands for the cmdbuf ioctl
+ */
+#define SAVAGE_CMD_STATE	0  /* a range of state registers */
+#define SAVAGE_CMD_DMA_PRIM	1  /* vertices from DMA buffer */
+#define SAVAGE_CMD_VB_PRIM	2  /* vertices from client vertex buffer */
+#define SAVAGE_CMD_DMA_IDX	3  /* indexed vertices from DMA buffer */
+#define SAVAGE_CMD_VB_IDX	4  /* indexed vertices client vertex buffer */
+#define SAVAGE_CMD_CLEAR	5  /* clear buffers */
+#define SAVAGE_CMD_SWAP		6  /* swap buffers */
+
+/* Primitive types
+*/
+#define SAVAGE_PRIM_TRILIST	0  /* triangle list */
+#define SAVAGE_PRIM_TRISTRIP	1  /* triangle strip */
+#define SAVAGE_PRIM_TRIFAN	2  /* triangle fan */
+#define SAVAGE_PRIM_TRILIST_201	3  /* reorder verts for correct flat
+				    * shading on s3d */
+
+/* Skip flags (vertex format)
+ */
+#define SAVAGE_SKIP_Z		0x01
+#define SAVAGE_SKIP_W		0x02
+#define SAVAGE_SKIP_C0		0x04
+#define SAVAGE_SKIP_C1		0x08
+#define SAVAGE_SKIP_S0		0x10
+#define SAVAGE_SKIP_T0		0x20
+#define SAVAGE_SKIP_ST0		0x30
+#define SAVAGE_SKIP_S1		0x40
+#define SAVAGE_SKIP_T1		0x80
+#define SAVAGE_SKIP_ST1		0xc0
+#define SAVAGE_SKIP_ALL_S3D	0x3f
+#define SAVAGE_SKIP_ALL_S4	0xff
+
+/* Buffer names for clear command
+ */
+#define SAVAGE_FRONT		0x1
+#define SAVAGE_BACK		0x2
+#define SAVAGE_DEPTH		0x4
+
+/* 64-bit command header
+ */
+union drm_savage_cmd_header {
+	struct {
+		unsigned char cmd;	/* command */
+		unsigned char pad0;
+		unsigned short pad1;
+		unsigned short pad2;
+		unsigned short pad3;
+	} cmd; /* generic */
+	struct {
+		unsigned char cmd;
+		unsigned char global;	/* need idle engine? */
+		unsigned short count;	/* number of consecutive registers */
+		unsigned short start;	/* first register */
+		unsigned short pad3;
+	} state; /* SAVAGE_CMD_STATE */
+	struct {
+		unsigned char cmd;
+		unsigned char prim;	/* primitive type */
+		unsigned short skip;	/* vertex format (skip flags) */
+		unsigned short count;	/* number of vertices */
+		unsigned short start;	/* first vertex in DMA/vertex buffer */
+	} prim; /* SAVAGE_CMD_DMA_PRIM, SAVAGE_CMD_VB_PRIM */
+	struct {
+		unsigned char cmd;
+		unsigned char prim;
+		unsigned short skip;
+		unsigned short count;	/* number of indices that follow */
+		unsigned short pad3;
+	} idx; /* SAVAGE_CMD_DMA_IDX, SAVAGE_CMD_VB_IDX */
+	struct {
+		unsigned char cmd;
+		unsigned char pad0;
+		unsigned short pad1;
+		unsigned int flags;
+	} clear0; /* SAVAGE_CMD_CLEAR */
+	struct {
+		unsigned int mask;
+		unsigned int value;
+	} clear1; /* SAVAGE_CMD_CLEAR data */
+};
+
+#endif
diff --git a/drivers/char/drm/savage_drv.c b/drivers/char/drm/savage_drv.c
new file mode 100644
index 000000000000..ac8d270427ca
--- /dev/null
+++ b/drivers/char/drm/savage_drv.c
@@ -0,0 +1,112 @@
+/* savage_drv.c -- Savage driver for Linux
+ *
+ * Copyright 2004  Felix Kuehling
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sub license,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NON-INFRINGEMENT. IN NO EVENT SHALL FELIX KUEHLING BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF
+ * CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#include <linux/config.h>
+#include "drmP.h"
+#include "savage_drm.h"
+#include "savage_drv.h"
+
+#include "drm_pciids.h"
+
+static int postinit( struct drm_device *dev, unsigned long flags )
+{
+	DRM_INFO( "Initialized %s %d.%d.%d %s on minor %d: %s\n",
+		DRIVER_NAME,
+		DRIVER_MAJOR,
+		DRIVER_MINOR,
+		DRIVER_PATCHLEVEL,
+		DRIVER_DATE,
+		dev->primary.minor,
+		pci_pretty_name(dev->pdev)
+		);
+	return 0;
+}
+
+static int version( drm_version_t *version )
+{
+	int len;
+
+	version->version_major = DRIVER_MAJOR;
+	version->version_minor = DRIVER_MINOR;
+	version->version_patchlevel = DRIVER_PATCHLEVEL;
+	DRM_COPY( version->name, DRIVER_NAME );
+	DRM_COPY( version->date, DRIVER_DATE );
+	DRM_COPY( version->desc, DRIVER_DESC );
+	return 0;
+}
+
+static struct pci_device_id pciidlist[] = {
+	savage_PCI_IDS
+};
+
+extern drm_ioctl_desc_t savage_ioctls[];
+extern int savage_max_ioctl;
+
+static struct drm_driver driver = {
+	.driver_features =
+	    DRIVER_USE_AGP | DRIVER_USE_MTRR |
+	    DRIVER_HAVE_DMA | DRIVER_PCI_DMA,
+	.dev_priv_size = sizeof(drm_savage_buf_priv_t),
+	.preinit = savage_preinit,
+	.postinit = postinit,
+	.postcleanup = savage_postcleanup,
+	.reclaim_buffers = savage_reclaim_buffers,
+	.get_map_ofs = drm_core_get_map_ofs,
+	.get_reg_ofs = drm_core_get_reg_ofs,
+	.version = version,
+	.ioctls = savage_ioctls,
+	.dma_ioctl = savage_bci_buffers,
+	.fops = {
+		.owner   = THIS_MODULE,
+		.open	 = drm_open,
+		.release = drm_release,
+		.ioctl	 = drm_ioctl,
+		.mmap	 = drm_mmap,
+		.poll = drm_poll,
+		.fasync  = drm_fasync,
+	},
+	.pci_driver = {
+		.name          = DRIVER_NAME,
+		.id_table      = pciidlist,
+	}
+};
+
+static int __init savage_init(void)
+{
+	driver.num_ioctls = savage_max_ioctl;
+	return drm_init(&driver);
+}
+
+static void __exit savage_exit(void)
+{
+	drm_exit(&driver);
+}
+
+module_init(savage_init);
+module_exit(savage_exit);
+
+MODULE_AUTHOR( DRIVER_AUTHOR );
+MODULE_DESCRIPTION( DRIVER_DESC );
+MODULE_LICENSE("GPL and additional rights");
diff --git a/drivers/char/drm/savage_drv.h b/drivers/char/drm/savage_drv.h
new file mode 100644
index 000000000000..a45434944658
--- /dev/null
+++ b/drivers/char/drm/savage_drv.h
@@ -0,0 +1,579 @@
+/* savage_drv.h -- Private header for the savage driver
+ *
+ * Copyright 2004  Felix Kuehling
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sub license,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NON-INFRINGEMENT. IN NO EVENT SHALL FELIX KUEHLING BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF
+ * CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#ifndef __SAVAGE_DRV_H__
+#define __SAVAGE_DRV_H__
+
+#define DRIVER_AUTHOR	"Felix Kuehling"
+
+#define DRIVER_NAME	"savage"
+#define DRIVER_DESC	"Savage3D/MX/IX, Savage4, SuperSavage, Twister, ProSavage[DDR]"
+#define DRIVER_DATE	"20050313"
+
+#define DRIVER_MAJOR		2
+#define DRIVER_MINOR		4
+#define DRIVER_PATCHLEVEL	1
+/* Interface history:
+ *
+ * 1.x   The DRM driver from the VIA/S3 code drop, basically a dummy
+ * 2.0   The first real DRM
+ * 2.1   Scissors registers managed by the DRM, 3D operations clipped by
+ *       cliprects of the cmdbuf ioctl
+ * 2.2   Implemented SAVAGE_CMD_DMA_IDX and SAVAGE_CMD_VB_IDX
+ * 2.3   Event counters used by BCI_EVENT_EMIT/WAIT ioctls are now 32 bits
+ *       wide and thus very long lived (unlikely to ever wrap). The size
+ *       in the struct was 32 bits before, but only 16 bits were used
+ * 2.4   Implemented command DMA. Now drm_savage_init_t.cmd_dma_offset is
+ *       actually used
+ */
+
+typedef struct drm_savage_age {
+	uint16_t event;
+	unsigned int wrap;
+} drm_savage_age_t;
+
+typedef struct drm_savage_buf_priv {
+	struct drm_savage_buf_priv *next;
+	struct drm_savage_buf_priv *prev;
+	drm_savage_age_t age;
+	drm_buf_t *buf;
+} drm_savage_buf_priv_t;
+
+typedef struct drm_savage_dma_page {
+	drm_savage_age_t age;
+	unsigned int used, flushed;
+} drm_savage_dma_page_t;
+#define SAVAGE_DMA_PAGE_SIZE 1024 /* in dwords */
+/* Fake DMA buffer size in bytes. 4 pages. Allows a maximum command
+ * size of 16kbytes or 4k entries. Minimum requirement would be
+ * 10kbytes for 255 40-byte vertices in one drawing command. */
+#define SAVAGE_FAKE_DMA_SIZE (SAVAGE_DMA_PAGE_SIZE*4*4)
+
+/* interesting bits of hardware state that are saved in dev_priv */
+typedef union {
+	struct drm_savage_common_state {
+		uint32_t vbaddr;
+	} common;
+	struct {
+		unsigned char pad[sizeof(struct drm_savage_common_state)];
+		uint32_t texctrl, texaddr;
+		uint32_t scstart, new_scstart;
+		uint32_t scend, new_scend;
+	} s3d;
+	struct {
+		unsigned char pad[sizeof(struct drm_savage_common_state)];
+		uint32_t texdescr, texaddr0, texaddr1;
+		uint32_t drawctrl0, new_drawctrl0;
+		uint32_t drawctrl1, new_drawctrl1;
+	} s4;
+} drm_savage_state_t;
+
+/* these chip tags should match the ones in the 2D driver in savage_regs.h. */
+enum savage_family {
+	S3_UNKNOWN = 0,
+	S3_SAVAGE3D,
+	S3_SAVAGE_MX,
+	S3_SAVAGE4,
+	S3_PROSAVAGE,
+	S3_TWISTER,
+	S3_PROSAVAGEDDR,
+	S3_SUPERSAVAGE,
+	S3_SAVAGE2000,
+	S3_LAST
+};
+
+#define S3_SAVAGE3D_SERIES(chip)  ((chip>=S3_SAVAGE3D) && (chip<=S3_SAVAGE_MX))
+
+#define S3_SAVAGE4_SERIES(chip)  ((chip==S3_SAVAGE4)            \
+                                  || (chip==S3_PROSAVAGE)       \
+                                  || (chip==S3_TWISTER)         \
+                                  || (chip==S3_PROSAVAGEDDR))
+
+#define	S3_SAVAGE_MOBILE_SERIES(chip)	((chip==S3_SAVAGE_MX) || (chip==S3_SUPERSAVAGE))
+
+#define S3_SAVAGE_SERIES(chip)    ((chip>=S3_SAVAGE3D) && (chip<=S3_SAVAGE2000))
+
+#define S3_MOBILE_TWISTER_SERIES(chip)   ((chip==S3_TWISTER)    \
+                                          ||(chip==S3_PROSAVAGEDDR))
+
+/* flags */
+#define SAVAGE_IS_AGP 1
+
+typedef struct drm_savage_private {
+	drm_savage_sarea_t *sarea_priv;
+
+	drm_savage_buf_priv_t head, tail;
+
+	/* who am I? */
+	enum savage_family chipset;
+
+	unsigned int cob_size;
+	unsigned int bci_threshold_lo, bci_threshold_hi;
+	unsigned int dma_type;
+
+	/* frame buffer layout */
+	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;
+
+	/* bitmap descriptors for swap and clear */
+	unsigned int front_bd, back_bd, depth_bd;
+
+	/* local textures */
+	unsigned int texture_offset;
+	unsigned int texture_size;
+
+	/* memory regions in physical memory */
+	drm_local_map_t *sarea;
+	drm_local_map_t *mmio;
+	drm_local_map_t *fb;
+	drm_local_map_t *aperture;
+	drm_local_map_t *status;
+	drm_local_map_t *agp_textures;
+	drm_local_map_t *cmd_dma;
+	drm_local_map_t fake_dma;
+
+	struct {
+		int handle;
+		unsigned long base, size;
+	} mtrr[3];
+
+	/* BCI and status-related stuff */
+	volatile uint32_t *status_ptr, *bci_ptr;
+	uint32_t status_used_mask;
+	uint16_t event_counter;
+	unsigned int event_wrap;
+
+	/* Savage4 command DMA */
+	drm_savage_dma_page_t *dma_pages;
+	unsigned int nr_dma_pages, first_dma_page, current_dma_page;
+	drm_savage_age_t last_dma_age;
+
+	/* saved hw state for global/local check on S3D */
+	uint32_t hw_draw_ctrl, hw_zbuf_ctrl;
+	/* and for scissors (global, so don't emit if not changed) */
+	uint32_t hw_scissors_start, hw_scissors_end;
+
+	drm_savage_state_t state;
+
+	/* after emitting a wait cmd Savage3D needs 63 nops before next DMA */
+	unsigned int waiting;
+
+	/* config/hardware-dependent function pointers */
+	int (*wait_fifo)(struct drm_savage_private *dev_priv, unsigned int n);
+	int (*wait_evnt)(struct drm_savage_private *dev_priv, uint16_t e);
+	/* Err, there is a macro wait_event in include/linux/wait.h.
+	 * Avoid unwanted macro expansion. */
+	void (*emit_clip_rect)(struct drm_savage_private *dev_priv,
+			       drm_clip_rect_t *pbox);
+	void (*dma_flush)(struct drm_savage_private *dev_priv);
+} drm_savage_private_t;
+
+/* ioctls */
+extern int savage_bci_cmdbuf(DRM_IOCTL_ARGS);
+extern int savage_bci_buffers(DRM_IOCTL_ARGS);
+
+/* BCI functions */
+extern uint16_t savage_bci_emit_event(drm_savage_private_t *dev_priv,
+				      unsigned int flags);
+extern void savage_freelist_put(drm_device_t *dev, drm_buf_t *buf);
+extern void savage_dma_reset(drm_savage_private_t *dev_priv);
+extern void savage_dma_wait(drm_savage_private_t *dev_priv, unsigned int page);
+extern uint32_t *savage_dma_alloc(drm_savage_private_t *dev_priv,
+				  unsigned int n);
+extern int savage_preinit(drm_device_t *dev, unsigned long chipset);
+extern int savage_postcleanup(drm_device_t *dev);
+extern int savage_do_cleanup_bci(drm_device_t *dev);
+extern void savage_reclaim_buffers(drm_device_t *dev, DRMFILE filp);
+
+/* state functions */
+extern void savage_emit_clip_rect_s3d(drm_savage_private_t *dev_priv,
+				      drm_clip_rect_t *pbox);
+extern void savage_emit_clip_rect_s4(drm_savage_private_t *dev_priv,
+				     drm_clip_rect_t *pbox);
+
+#define SAVAGE_FB_SIZE_S3	0x01000000	/*  16MB */
+#define SAVAGE_FB_SIZE_S4	0x02000000	/*  32MB */
+#define SAVAGE_MMIO_SIZE        0x00080000	/* 512kB */
+#define SAVAGE_APERTURE_OFFSET  0x02000000	/*  32MB */
+#define SAVAGE_APERTURE_SIZE    0x05000000	/* 5 tiled surfaces, 16MB each */
+
+#define SAVAGE_BCI_OFFSET       0x00010000      /* offset of the BCI region
+						 * inside the MMIO region */
+#define SAVAGE_BCI_FIFO_SIZE	32		/* number of entries in on-chip
+						 * BCI FIFO */
+
+/*
+ * MMIO registers
+ */
+#define SAVAGE_STATUS_WORD0		0x48C00
+#define SAVAGE_STATUS_WORD1		0x48C04
+#define SAVAGE_ALT_STATUS_WORD0 	0x48C60
+
+#define SAVAGE_FIFO_USED_MASK_S3D	0x0001ffff
+#define SAVAGE_FIFO_USED_MASK_S4	0x001fffff
+
+/* Copied from savage_bci.h in the 2D driver with some renaming. */
+
+/* Bitmap descriptors */
+#define SAVAGE_BD_STRIDE_SHIFT 0
+#define SAVAGE_BD_BPP_SHIFT   16
+#define SAVAGE_BD_TILE_SHIFT  24
+#define SAVAGE_BD_BW_DISABLE  (1<<28)
+/* common: */
+#define	SAVAGE_BD_TILE_LINEAR		0
+/* savage4, MX, IX, 3D */
+#define	SAVAGE_BD_TILE_16BPP		2
+#define	SAVAGE_BD_TILE_32BPP		3
+/* twister, prosavage, DDR, supersavage, 2000 */
+#define	SAVAGE_BD_TILE_DEST		1
+#define	SAVAGE_BD_TILE_TEXTURE		2
+/* GBD - BCI enable */
+/* savage4, MX, IX, 3D */
+#define SAVAGE_GBD_BCI_ENABLE                    8
+/* twister, prosavage, DDR, supersavage, 2000 */
+#define SAVAGE_GBD_BCI_ENABLE_TWISTER            0
+
+#define SAVAGE_GBD_BIG_ENDIAN                    4
+#define SAVAGE_GBD_LITTLE_ENDIAN                 0
+#define SAVAGE_GBD_64                            1
+
+/*  Global Bitmap Descriptor */
+#define SAVAGE_BCI_GLB_BD_LOW             0x8168
+#define SAVAGE_BCI_GLB_BD_HIGH            0x816C
+
+/*
+ * BCI registers
+ */
+/* Savage4/Twister/ProSavage 3D registers */
+#define SAVAGE_DRAWLOCALCTRL_S4		0x1e
+#define SAVAGE_TEXPALADDR_S4		0x1f
+#define SAVAGE_TEXCTRL0_S4		0x20
+#define SAVAGE_TEXCTRL1_S4		0x21
+#define SAVAGE_TEXADDR0_S4		0x22
+#define SAVAGE_TEXADDR1_S4		0x23
+#define SAVAGE_TEXBLEND0_S4		0x24
+#define SAVAGE_TEXBLEND1_S4		0x25
+#define SAVAGE_TEXXPRCLR_S4		0x26 /* never used */
+#define SAVAGE_TEXDESCR_S4		0x27
+#define SAVAGE_FOGTABLE_S4		0x28
+#define SAVAGE_FOGCTRL_S4		0x30
+#define SAVAGE_STENCILCTRL_S4		0x31
+#define SAVAGE_ZBUFCTRL_S4		0x32
+#define SAVAGE_ZBUFOFF_S4		0x33
+#define SAVAGE_DESTCTRL_S4		0x34
+#define SAVAGE_DRAWCTRL0_S4		0x35
+#define SAVAGE_DRAWCTRL1_S4		0x36
+#define SAVAGE_ZWATERMARK_S4		0x37
+#define SAVAGE_DESTTEXRWWATERMARK_S4	0x38
+#define SAVAGE_TEXBLENDCOLOR_S4		0x39
+/* Savage3D/MX/IX 3D registers */
+#define SAVAGE_TEXPALADDR_S3D		0x18
+#define SAVAGE_TEXXPRCLR_S3D		0x19 /* never used */
+#define SAVAGE_TEXADDR_S3D		0x1A
+#define SAVAGE_TEXDESCR_S3D		0x1B
+#define SAVAGE_TEXCTRL_S3D		0x1C
+#define SAVAGE_FOGTABLE_S3D		0x20
+#define SAVAGE_FOGCTRL_S3D		0x30
+#define SAVAGE_DRAWCTRL_S3D		0x31
+#define SAVAGE_ZBUFCTRL_S3D		0x32
+#define SAVAGE_ZBUFOFF_S3D		0x33
+#define SAVAGE_DESTCTRL_S3D		0x34
+#define SAVAGE_SCSTART_S3D		0x35
+#define SAVAGE_SCEND_S3D		0x36
+#define SAVAGE_ZWATERMARK_S3D		0x37 
+#define SAVAGE_DESTTEXRWWATERMARK_S3D	0x38
+/* common stuff */
+#define SAVAGE_VERTBUFADDR		0x3e
+#define SAVAGE_BITPLANEWTMASK		0xd7
+#define SAVAGE_DMABUFADDR		0x51
+
+/* texture enable bits (needed for tex addr checking) */
+#define SAVAGE_TEXCTRL_TEXEN_MASK	0x00010000 /* S3D */
+#define SAVAGE_TEXDESCR_TEX0EN_MASK	0x02000000 /* S4 */
+#define SAVAGE_TEXDESCR_TEX1EN_MASK	0x04000000 /* S4 */
+
+/* Global fields in Savage4/Twister/ProSavage 3D registers:
+ *
+ * All texture registers and DrawLocalCtrl are local. All other
+ * registers are global. */
+
+/* Global fields in Savage3D/MX/IX 3D registers:
+ *
+ * All texture registers are local. DrawCtrl and ZBufCtrl are
+ * partially local. All other registers are global.
+ *
+ * DrawCtrl global fields: cullMode, alphaTestCmpFunc, alphaTestEn, alphaRefVal
+ * ZBufCtrl global fields: zCmpFunc, zBufEn
+ */
+#define SAVAGE_DRAWCTRL_S3D_GLOBAL	0x03f3c00c
+#define SAVAGE_ZBUFCTRL_S3D_GLOBAL	0x00000027
+
+/* Masks for scissor bits (drawCtrl[01] on s4, scissorStart/End on s3d)
+ */
+#define SAVAGE_SCISSOR_MASK_S4		0x00fff7ff
+#define SAVAGE_SCISSOR_MASK_S3D		0x07ff07ff
+
+/*
+ * BCI commands
+ */
+#define BCI_CMD_NOP                  0x40000000
+#define BCI_CMD_RECT                 0x48000000
+#define BCI_CMD_RECT_XP              0x01000000
+#define BCI_CMD_RECT_YP              0x02000000
+#define BCI_CMD_SCANLINE             0x50000000
+#define BCI_CMD_LINE                 0x5C000000
+#define BCI_CMD_LINE_LAST_PIXEL      0x58000000
+#define BCI_CMD_BYTE_TEXT            0x63000000
+#define BCI_CMD_NT_BYTE_TEXT         0x67000000
+#define BCI_CMD_BIT_TEXT             0x6C000000
+#define BCI_CMD_GET_ROP(cmd)         (((cmd) >> 16) & 0xFF)
+#define BCI_CMD_SET_ROP(cmd, rop)    ((cmd) |= ((rop & 0xFF) << 16))
+#define BCI_CMD_SEND_COLOR           0x00008000
+
+#define BCI_CMD_CLIP_NONE            0x00000000
+#define BCI_CMD_CLIP_CURRENT         0x00002000
+#define BCI_CMD_CLIP_LR              0x00004000
+#define BCI_CMD_CLIP_NEW             0x00006000
+
+#define BCI_CMD_DEST_GBD             0x00000000
+#define BCI_CMD_DEST_PBD             0x00000800
+#define BCI_CMD_DEST_PBD_NEW         0x00000C00
+#define BCI_CMD_DEST_SBD             0x00001000
+#define BCI_CMD_DEST_SBD_NEW         0x00001400
+
+#define BCI_CMD_SRC_TRANSPARENT      0x00000200
+#define BCI_CMD_SRC_SOLID            0x00000000
+#define BCI_CMD_SRC_GBD              0x00000020
+#define BCI_CMD_SRC_COLOR            0x00000040
+#define BCI_CMD_SRC_MONO             0x00000060
+#define BCI_CMD_SRC_PBD_COLOR        0x00000080
+#define BCI_CMD_SRC_PBD_MONO         0x000000A0
+#define BCI_CMD_SRC_PBD_COLOR_NEW    0x000000C0
+#define BCI_CMD_SRC_PBD_MONO_NEW     0x000000E0
+#define BCI_CMD_SRC_SBD_COLOR        0x00000100
+#define BCI_CMD_SRC_SBD_MONO         0x00000120
+#define BCI_CMD_SRC_SBD_COLOR_NEW    0x00000140
+#define BCI_CMD_SRC_SBD_MONO_NEW     0x00000160
+
+#define BCI_CMD_PAT_TRANSPARENT      0x00000010
+#define BCI_CMD_PAT_NONE             0x00000000
+#define BCI_CMD_PAT_COLOR            0x00000002
+#define BCI_CMD_PAT_MONO             0x00000003
+#define BCI_CMD_PAT_PBD_COLOR        0x00000004
+#define BCI_CMD_PAT_PBD_MONO         0x00000005
+#define BCI_CMD_PAT_PBD_COLOR_NEW    0x00000006
+#define BCI_CMD_PAT_PBD_MONO_NEW     0x00000007
+#define BCI_CMD_PAT_SBD_COLOR        0x00000008
+#define BCI_CMD_PAT_SBD_MONO         0x00000009
+#define BCI_CMD_PAT_SBD_COLOR_NEW    0x0000000A
+#define BCI_CMD_PAT_SBD_MONO_NEW     0x0000000B
+
+#define BCI_BD_BW_DISABLE            0x10000000
+#define BCI_BD_TILE_MASK             0x03000000
+#define BCI_BD_TILE_NONE             0x00000000
+#define BCI_BD_TILE_16               0x02000000
+#define BCI_BD_TILE_32               0x03000000
+#define BCI_BD_GET_BPP(bd)           (((bd) >> 16) & 0xFF)
+#define BCI_BD_SET_BPP(bd, bpp)      ((bd) |= (((bpp) & 0xFF) << 16))
+#define BCI_BD_GET_STRIDE(bd)        ((bd) & 0xFFFF)
+#define BCI_BD_SET_STRIDE(bd, st)    ((bd) |= ((st) & 0xFFFF))
+
+#define BCI_CMD_SET_REGISTER            0x96000000
+
+#define BCI_CMD_WAIT                    0xC0000000
+#define BCI_CMD_WAIT_3D                 0x00010000
+#define BCI_CMD_WAIT_2D                 0x00020000
+
+#define BCI_CMD_UPDATE_EVENT_TAG        0x98000000
+
+#define BCI_CMD_DRAW_PRIM               0x80000000
+#define BCI_CMD_DRAW_INDEXED_PRIM       0x88000000
+#define BCI_CMD_DRAW_CONT               0x01000000
+#define BCI_CMD_DRAW_TRILIST            0x00000000
+#define BCI_CMD_DRAW_TRISTRIP           0x02000000
+#define BCI_CMD_DRAW_TRIFAN             0x04000000
+#define BCI_CMD_DRAW_SKIPFLAGS          0x000000ff
+#define BCI_CMD_DRAW_NO_Z		0x00000001
+#define BCI_CMD_DRAW_NO_W		0x00000002
+#define BCI_CMD_DRAW_NO_CD		0x00000004
+#define BCI_CMD_DRAW_NO_CS		0x00000008
+#define BCI_CMD_DRAW_NO_U0		0x00000010
+#define BCI_CMD_DRAW_NO_V0		0x00000020
+#define BCI_CMD_DRAW_NO_UV0		0x00000030
+#define BCI_CMD_DRAW_NO_U1		0x00000040
+#define BCI_CMD_DRAW_NO_V1		0x00000080
+#define BCI_CMD_DRAW_NO_UV1		0x000000c0
+
+#define BCI_CMD_DMA			0xa8000000
+
+#define BCI_W_H(w, h)                ((((h) << 16) | (w)) & 0x0FFF0FFF)
+#define BCI_X_Y(x, y)                ((((y) << 16) | (x)) & 0x0FFF0FFF)
+#define BCI_X_W(x, y)                ((((w) << 16) | (x)) & 0x0FFF0FFF)
+#define BCI_CLIP_LR(l, r)            ((((r) << 16) | (l)) & 0x0FFF0FFF)
+#define BCI_CLIP_TL(t, l)            ((((t) << 16) | (l)) & 0x0FFF0FFF)
+#define BCI_CLIP_BR(b, r)            ((((b) << 16) | (r)) & 0x0FFF0FFF)
+
+#define BCI_LINE_X_Y(x, y)           (((y) << 16) | ((x) & 0xFFFF))
+#define BCI_LINE_STEPS(diag, axi)    (((axi) << 16) | ((diag) & 0xFFFF))
+#define BCI_LINE_MISC(maj, ym, xp, yp, err) \
+	(((maj) & 0x1FFF) | \
+	((ym) ? 1<<13 : 0) | \
+	((xp) ? 1<<14 : 0) | \
+	((yp) ? 1<<15 : 0) | \
+	((err) << 16))
+
+/*
+ * common commands
+ */
+#define BCI_SET_REGISTERS( first, n )			\
+	BCI_WRITE(BCI_CMD_SET_REGISTER |		\
+		  ((uint32_t)(n) & 0xff) << 16 |	\
+		  ((uint32_t)(first) & 0xffff))
+#define DMA_SET_REGISTERS( first, n )			\
+	DMA_WRITE(BCI_CMD_SET_REGISTER |		\
+		  ((uint32_t)(n) & 0xff) << 16 |	\
+		  ((uint32_t)(first) & 0xffff))
+
+#define BCI_DRAW_PRIMITIVE(n, type, skip)         \
+        BCI_WRITE(BCI_CMD_DRAW_PRIM | (type) | (skip) | \
+		  ((n) << 16))
+#define DMA_DRAW_PRIMITIVE(n, type, skip)         \
+        DMA_WRITE(BCI_CMD_DRAW_PRIM | (type) | (skip) | \
+		  ((n) << 16))
+
+#define BCI_DRAW_INDICES_S3D(n, type, i0)         \
+        BCI_WRITE(BCI_CMD_DRAW_INDEXED_PRIM | (type) |  \
+		  ((n) << 16) | (i0))
+
+#define BCI_DRAW_INDICES_S4(n, type, skip)        \
+        BCI_WRITE(BCI_CMD_DRAW_INDEXED_PRIM | (type) |  \
+                  (skip) | ((n) << 16))
+
+#define BCI_DMA(n)	\
+	BCI_WRITE(BCI_CMD_DMA | (((n) >> 1) - 1))
+
+/*
+ * access to MMIO
+ */
+#define SAVAGE_READ(reg)	DRM_READ32(  dev_priv->mmio, (reg) )
+#define SAVAGE_WRITE(reg)	DRM_WRITE32( dev_priv->mmio, (reg) )
+
+/*
+ * access to the burst command interface (BCI)
+ */
+#define SAVAGE_BCI_DEBUG 1
+
+#define BCI_LOCALS    volatile uint32_t *bci_ptr;
+
+#define BEGIN_BCI( n ) do {			\
+	dev_priv->wait_fifo(dev_priv, (n));	\
+	bci_ptr = dev_priv->bci_ptr;		\
+} while(0)
+
+#define BCI_WRITE( val ) *bci_ptr++ = (uint32_t)(val)
+
+#define BCI_COPY_FROM_USER(src,n) do {				\
+    unsigned int i;						\
+    for (i = 0; i < n; ++i) {					\
+	uint32_t val;						\
+	DRM_GET_USER_UNCHECKED(val, &((uint32_t*)(src))[i]);	\
+	BCI_WRITE(val);						\
+    }								\
+} while(0)
+
+/*
+ * command DMA support
+ */
+#define SAVAGE_DMA_DEBUG 1
+
+#define DMA_LOCALS   uint32_t *dma_ptr;
+
+#define BEGIN_DMA( n ) do {						\
+	unsigned int cur = dev_priv->current_dma_page;			\
+	unsigned int rest = SAVAGE_DMA_PAGE_SIZE -			\
+		dev_priv->dma_pages[cur].used;				\
+	if ((n) > rest) {						\
+		dma_ptr = savage_dma_alloc(dev_priv, (n));		\
+	} else { /* fast path for small allocations */			\
+		dma_ptr = (uint32_t *)dev_priv->cmd_dma->handle +	\
+			cur * SAVAGE_DMA_PAGE_SIZE +			\
+			dev_priv->dma_pages[cur].used;			\
+		if (dev_priv->dma_pages[cur].used == 0)			\
+			savage_dma_wait(dev_priv, cur);			\
+		dev_priv->dma_pages[cur].used += (n);			\
+	}								\
+} while(0)
+
+#define DMA_WRITE( val ) *dma_ptr++ = (uint32_t)(val)
+
+#define DMA_COPY_FROM_USER(src,n) do {				\
+	DRM_COPY_FROM_USER_UNCHECKED(dma_ptr, (src), (n)*4);	\
+	dma_ptr += n;						\
+} while(0)
+
+#if SAVAGE_DMA_DEBUG
+#define DMA_COMMIT() do {						\
+	unsigned int cur = dev_priv->current_dma_page;			\
+	uint32_t *expected = (uint32_t *)dev_priv->cmd_dma->handle +	\
+			cur * SAVAGE_DMA_PAGE_SIZE +			\
+			dev_priv->dma_pages[cur].used;			\
+	if (dma_ptr != expected) {					\
+		DRM_ERROR("DMA allocation and use don't match: "	\
+			  "%p != %p\n", expected, dma_ptr);		\
+		savage_dma_reset(dev_priv);				\
+	}								\
+} while(0)
+#else
+#define DMA_COMMIT() do {/* nothing */} while(0)
+#endif
+
+#define DMA_FLUSH() dev_priv->dma_flush(dev_priv)
+
+/* Buffer aging via event tag
+ */
+
+#define UPDATE_EVENT_COUNTER( ) do {			\
+	if (dev_priv->status_ptr) {			\
+		uint16_t count;				\
+		/* coordinate with Xserver */		\
+		count = dev_priv->status_ptr[1023];	\
+		if (count < dev_priv->event_counter)	\
+			dev_priv->event_wrap++;		\
+		dev_priv->event_counter = count;	\
+	}						\
+} while(0)
+
+#define SET_AGE( age, e, w ) do {	\
+	(age)->event = e;		\
+	(age)->wrap = w;		\
+} while(0)
+
+#define TEST_AGE( age, e, w )				\
+	( (age)->wrap < (w) || ( (age)->wrap == (w) && (age)->event <= (e) ) )
+
+#endif /* __SAVAGE_DRV_H__ */
diff --git a/drivers/char/drm/savage_state.c b/drivers/char/drm/savage_state.c
new file mode 100644
index 000000000000..475695a00083
--- /dev/null
+++ b/drivers/char/drm/savage_state.c
@@ -0,0 +1,1146 @@
+/* savage_state.c -- State and drawing support for Savage
+ *
+ * Copyright 2004  Felix Kuehling
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sub license,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NON-INFRINGEMENT. IN NO EVENT SHALL FELIX KUEHLING BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF
+ * CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+#include "drmP.h"
+#include "savage_drm.h"
+#include "savage_drv.h"
+
+void savage_emit_clip_rect_s3d(drm_savage_private_t *dev_priv,
+			       drm_clip_rect_t *pbox)
+{
+	uint32_t scstart = dev_priv->state.s3d.new_scstart;
+	uint32_t scend   = dev_priv->state.s3d.new_scend;
+	scstart = (scstart & ~SAVAGE_SCISSOR_MASK_S3D) |
+		((uint32_t)pbox->x1 & 0x000007ff) | 
+		(((uint32_t)pbox->y1 << 16) & 0x07ff0000);
+	scend   = (scend   & ~SAVAGE_SCISSOR_MASK_S3D) |
+		(((uint32_t)pbox->x2-1) & 0x000007ff) |
+		((((uint32_t)pbox->y2-1) << 16) & 0x07ff0000);
+	if (scstart != dev_priv->state.s3d.scstart ||
+	    scend   != dev_priv->state.s3d.scend) {
+		DMA_LOCALS;
+		BEGIN_DMA(4);
+		DMA_WRITE(BCI_CMD_WAIT|BCI_CMD_WAIT_3D);
+		DMA_SET_REGISTERS(SAVAGE_SCSTART_S3D, 2);
+		DMA_WRITE(scstart);
+		DMA_WRITE(scend);
+		dev_priv->state.s3d.scstart = scstart;
+		dev_priv->state.s3d.scend   = scend;
+		dev_priv->waiting = 1;
+		DMA_COMMIT();
+	}
+}
+
+void savage_emit_clip_rect_s4(drm_savage_private_t *dev_priv,
+			      drm_clip_rect_t *pbox)
+{
+	uint32_t drawctrl0 = dev_priv->state.s4.new_drawctrl0;
+	uint32_t drawctrl1 = dev_priv->state.s4.new_drawctrl1;
+	drawctrl0 = (drawctrl0 & ~SAVAGE_SCISSOR_MASK_S4) |
+		((uint32_t)pbox->x1 & 0x000007ff) |
+		(((uint32_t)pbox->y1 << 12) & 0x00fff000);
+	drawctrl1 = (drawctrl1 & ~SAVAGE_SCISSOR_MASK_S4) |
+		(((uint32_t)pbox->x2-1) & 0x000007ff) |
+		((((uint32_t)pbox->y2-1) << 12) & 0x00fff000);
+	if (drawctrl0 != dev_priv->state.s4.drawctrl0 ||
+	    drawctrl1 != dev_priv->state.s4.drawctrl1) {
+		DMA_LOCALS;
+		BEGIN_DMA(4);
+		DMA_WRITE(BCI_CMD_WAIT|BCI_CMD_WAIT_3D);
+		DMA_SET_REGISTERS(SAVAGE_DRAWCTRL0_S4, 2);
+		DMA_WRITE(drawctrl0);
+		DMA_WRITE(drawctrl1);
+		dev_priv->state.s4.drawctrl0 = drawctrl0;
+		dev_priv->state.s4.drawctrl1 = drawctrl1;
+		dev_priv->waiting = 1;
+		DMA_COMMIT();
+	}
+}
+
+static int savage_verify_texaddr(drm_savage_private_t *dev_priv, int unit,
+				 uint32_t addr)
+{
+	if ((addr & 6) != 2) { /* reserved bits */
+		DRM_ERROR("bad texAddr%d %08x (reserved bits)\n", unit, addr);
+		return DRM_ERR(EINVAL);
+	}
+	if (!(addr & 1)) { /* local */
+		addr &= ~7;
+		if (addr <  dev_priv->texture_offset ||
+		    addr >= dev_priv->texture_offset+dev_priv->texture_size) {
+			DRM_ERROR("bad texAddr%d %08x (local addr out of range)\n",
+				  unit, addr);
+			return DRM_ERR(EINVAL);
+		}
+	} else { /* AGP */
+		if (!dev_priv->agp_textures) {
+			DRM_ERROR("bad texAddr%d %08x (AGP not available)\n",
+				  unit, addr);
+			return DRM_ERR(EINVAL);
+		}
+		addr &= ~7;
+		if (addr < dev_priv->agp_textures->offset ||
+		    addr >= (dev_priv->agp_textures->offset +
+			     dev_priv->agp_textures->size)) {
+			DRM_ERROR("bad texAddr%d %08x (AGP addr out of range)\n",
+				  unit, addr);
+			return DRM_ERR(EINVAL);
+		}
+	}
+	return 0;
+}
+
+#define SAVE_STATE(reg,where)			\
+	if(start <= reg && start+count > reg)	\
+		DRM_GET_USER_UNCHECKED(dev_priv->state.where, &regs[reg-start])
+#define SAVE_STATE_MASK(reg,where,mask) do {			\
+	if(start <= reg && start+count > reg) {			\
+		uint32_t tmp;					\
+		DRM_GET_USER_UNCHECKED(tmp, &regs[reg-start]);	\
+		dev_priv->state.where = (tmp & (mask)) |	\
+			(dev_priv->state.where & ~(mask));	\
+	}							\
+} while (0)
+static int savage_verify_state_s3d(drm_savage_private_t *dev_priv,
+				   unsigned int start, unsigned int count,
+				   const uint32_t __user *regs)
+{
+	if (start < SAVAGE_TEXPALADDR_S3D ||
+	    start+count-1 > SAVAGE_DESTTEXRWWATERMARK_S3D) {
+		DRM_ERROR("invalid register range (0x%04x-0x%04x)\n",
+			  start, start+count-1);
+		return DRM_ERR(EINVAL);
+	}
+
+	SAVE_STATE_MASK(SAVAGE_SCSTART_S3D, s3d.new_scstart,
+			~SAVAGE_SCISSOR_MASK_S3D);
+	SAVE_STATE_MASK(SAVAGE_SCEND_S3D, s3d.new_scend,
+			~SAVAGE_SCISSOR_MASK_S3D);
+
+	/* if any texture regs were changed ... */
+	if (start <= SAVAGE_TEXCTRL_S3D &&
+	    start+count > SAVAGE_TEXPALADDR_S3D) {
+		/* ... check texture state */
+		SAVE_STATE(SAVAGE_TEXCTRL_S3D, s3d.texctrl);
+		SAVE_STATE(SAVAGE_TEXADDR_S3D, s3d.texaddr);
+		if (dev_priv->state.s3d.texctrl & SAVAGE_TEXCTRL_TEXEN_MASK)
+			return savage_verify_texaddr(
+				dev_priv, 0, dev_priv->state.s3d.texaddr);
+	}
+
+	return 0;
+}
+
+static int savage_verify_state_s4(drm_savage_private_t *dev_priv,
+				  unsigned int start, unsigned int count,
+				  const uint32_t __user *regs)
+{
+	int ret = 0;
+
+	if (start < SAVAGE_DRAWLOCALCTRL_S4 ||
+	    start+count-1 > SAVAGE_TEXBLENDCOLOR_S4) {
+		DRM_ERROR("invalid register range (0x%04x-0x%04x)\n",
+			  start, start+count-1);
+		return DRM_ERR(EINVAL);
+	}
+
+	SAVE_STATE_MASK(SAVAGE_DRAWCTRL0_S4, s4.new_drawctrl0,
+			~SAVAGE_SCISSOR_MASK_S4);
+	SAVE_STATE_MASK(SAVAGE_DRAWCTRL1_S4, s4.new_drawctrl1,
+			~SAVAGE_SCISSOR_MASK_S4);
+
+	/* if any texture regs were changed ... */
+	if (start <= SAVAGE_TEXDESCR_S4 &&
+	    start+count > SAVAGE_TEXPALADDR_S4) {
+		/* ... check texture state */
+		SAVE_STATE(SAVAGE_TEXDESCR_S4, s4.texdescr);
+		SAVE_STATE(SAVAGE_TEXADDR0_S4, s4.texaddr0);
+		SAVE_STATE(SAVAGE_TEXADDR1_S4, s4.texaddr1);
+		if (dev_priv->state.s4.texdescr & SAVAGE_TEXDESCR_TEX0EN_MASK)
+			ret |= savage_verify_texaddr(
+				dev_priv, 0, dev_priv->state.s4.texaddr0);
+		if (dev_priv->state.s4.texdescr & SAVAGE_TEXDESCR_TEX1EN_MASK)
+			ret |= savage_verify_texaddr(
+				dev_priv, 1, dev_priv->state.s4.texaddr1);
+	}
+
+	return ret;
+}
+#undef SAVE_STATE
+#undef SAVE_STATE_MASK
+
+static int savage_dispatch_state(drm_savage_private_t *dev_priv,
+				 const drm_savage_cmd_header_t *cmd_header,
+				 const uint32_t __user *regs)
+{
+	unsigned int count = cmd_header->state.count;
+	unsigned int start = cmd_header->state.start;
+	unsigned int count2 = 0;
+	unsigned int bci_size;
+	int ret;
+	DMA_LOCALS;
+
+	if (!count)
+		return 0;
+
+	if (DRM_VERIFYAREA_READ(regs, count*4))
+		return DRM_ERR(EFAULT);
+
+	if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
+		ret = savage_verify_state_s3d(dev_priv, start, count, regs);
+		if (ret != 0)
+			return ret;
+		/* scissor regs are emitted in savage_dispatch_draw */
+		if (start < SAVAGE_SCSTART_S3D) {
+			if (start+count > SAVAGE_SCEND_S3D+1)
+				count2 = count - (SAVAGE_SCEND_S3D+1 - start);
+			if (start+count > SAVAGE_SCSTART_S3D)
+				count = SAVAGE_SCSTART_S3D - start;
+		} else if (start <= SAVAGE_SCEND_S3D) {
+			if (start+count > SAVAGE_SCEND_S3D+1) {
+				count -= SAVAGE_SCEND_S3D+1 - start;
+				start = SAVAGE_SCEND_S3D+1;
+			} else
+				return 0;
+		}
+	} else {
+		ret = savage_verify_state_s4(dev_priv, start, count, regs);
+		if (ret != 0)
+			return ret;
+		/* scissor regs are emitted in savage_dispatch_draw */
+		if (start < SAVAGE_DRAWCTRL0_S4) {
+			if (start+count > SAVAGE_DRAWCTRL1_S4+1)
+				count2 = count - (SAVAGE_DRAWCTRL1_S4+1 - start);
+			if (start+count > SAVAGE_DRAWCTRL0_S4)
+				count = SAVAGE_DRAWCTRL0_S4 - start;
+		} else if (start <= SAVAGE_DRAWCTRL1_S4) {
+			if (start+count > SAVAGE_DRAWCTRL1_S4+1) {
+				count -= SAVAGE_DRAWCTRL1_S4+1 - start;
+				start = SAVAGE_DRAWCTRL1_S4+1;
+			} else
+				return 0;
+		}
+	}
+
+	bci_size = count + (count+254)/255 + count2 + (count2+254)/255;
+
+	if (cmd_header->state.global) {
+		BEGIN_DMA(bci_size+1);
+		DMA_WRITE(BCI_CMD_WAIT | BCI_CMD_WAIT_3D);
+		dev_priv->waiting = 1;
+	} else {
+		BEGIN_DMA(bci_size);
+	}
+
+	do {
+		while (count > 0) {
+			unsigned int n = count < 255 ? count : 255;
+			DMA_SET_REGISTERS(start, n);
+			DMA_COPY_FROM_USER(regs, n);
+			count -= n;
+			start += n;
+			regs += n;
+		}
+		start += 2;
+		regs += 2;
+		count = count2;
+		count2 = 0;
+	} while (count);
+
+	DMA_COMMIT();
+
+	return 0;
+}
+
+static int savage_dispatch_dma_prim(drm_savage_private_t *dev_priv,
+				    const drm_savage_cmd_header_t *cmd_header,
+				    const drm_buf_t *dmabuf)
+{
+	unsigned char reorder = 0;
+	unsigned int prim = cmd_header->prim.prim;
+	unsigned int skip = cmd_header->prim.skip;
+	unsigned int n = cmd_header->prim.count;
+	unsigned int start = cmd_header->prim.start;
+	unsigned int i;
+	BCI_LOCALS;
+
+	if (!dmabuf) {
+	    DRM_ERROR("called without dma buffers!\n");
+	    return DRM_ERR(EINVAL);
+	}
+
+	if (!n)
+		return 0;
+
+	switch (prim) {
+	case SAVAGE_PRIM_TRILIST_201:
+		reorder = 1;
+		prim = SAVAGE_PRIM_TRILIST;
+	case SAVAGE_PRIM_TRILIST:
+		if (n % 3 != 0) {
+			DRM_ERROR("wrong number of vertices %u in TRILIST\n",
+				  n);
+			return DRM_ERR(EINVAL);
+		}
+		break;
+	case SAVAGE_PRIM_TRISTRIP:
+	case SAVAGE_PRIM_TRIFAN:
+		if (n < 3) {
+			DRM_ERROR("wrong number of vertices %u in TRIFAN/STRIP\n",
+				  n);
+			return DRM_ERR(EINVAL);
+		}
+		break;
+	default:
+		DRM_ERROR("invalid primitive type %u\n", prim);
+		return DRM_ERR(EINVAL);
+	}
+
+	if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
+		if (skip != 0) {
+			DRM_ERROR("invalid skip flags 0x%04x for DMA\n",
+				  skip);
+			return DRM_ERR(EINVAL);
+		}
+	} else {
+		unsigned int size = 10 - (skip & 1) - (skip >> 1 & 1) -
+			(skip >> 2 & 1) - (skip >> 3 & 1) - (skip >> 4 & 1) -
+			(skip >> 5 & 1) - (skip >> 6 & 1) - (skip >> 7 & 1);
+		if (skip > SAVAGE_SKIP_ALL_S4 || size != 8) {
+			DRM_ERROR("invalid skip flags 0x%04x for DMA\n",
+				  skip);
+			return DRM_ERR(EINVAL);
+		}
+		if (reorder) {
+			DRM_ERROR("TRILIST_201 used on Savage4 hardware\n");
+			return DRM_ERR(EINVAL);
+		}
+	}
+
+	if (start + n > dmabuf->total/32) {
+		DRM_ERROR("vertex indices (%u-%u) out of range (0-%u)\n",
+			  start, start + n - 1, dmabuf->total/32);
+		return DRM_ERR(EINVAL);
+	}
+
+	/* Vertex DMA doesn't work with command DMA at the same time,
+	 * so we use BCI_... to submit commands here. Flush buffered
+	 * faked DMA first. */
+	DMA_FLUSH();
+
+	if (dmabuf->bus_address != dev_priv->state.common.vbaddr) {
+		BEGIN_BCI(2);
+		BCI_SET_REGISTERS(SAVAGE_VERTBUFADDR, 1);
+		BCI_WRITE(dmabuf->bus_address | dev_priv->dma_type);
+		dev_priv->state.common.vbaddr = dmabuf->bus_address;
+	}
+	if (S3_SAVAGE3D_SERIES(dev_priv->chipset) && dev_priv->waiting) {
+		/* Workaround for what looks like a hardware bug. If a
+		 * WAIT_3D_IDLE was emitted some time before the
+		 * indexed drawing command then the engine will lock
+		 * up. There are two known workarounds:
+		 * WAIT_IDLE_EMPTY or emit at least 63 NOPs. */
+		BEGIN_BCI(63);
+		for (i = 0; i < 63; ++i)
+			BCI_WRITE(BCI_CMD_WAIT);
+		dev_priv->waiting = 0;
+	}
+
+	prim <<= 25;
+	while (n != 0) {
+		/* Can emit up to 255 indices (85 triangles) at once. */
+		unsigned int count = n > 255 ? 255 : n;
+		if (reorder) {
+			/* Need to reorder indices for correct flat
+			 * shading while preserving the clock sense
+			 * for correct culling. Only on Savage3D. */
+			int reorder[3] = {-1, -1, -1};
+			reorder[start%3] = 2;
+
+			BEGIN_BCI((count+1+1)/2);
+			BCI_DRAW_INDICES_S3D(count, prim, start+2);
+
+			for (i = start+1; i+1 < start+count; i += 2)
+				BCI_WRITE((i + reorder[i % 3]) |
+					  ((i+1 + reorder[(i+1) % 3]) << 16));
+			if (i < start+count)
+				BCI_WRITE(i + reorder[i%3]);
+		} else if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
+			BEGIN_BCI((count+1+1)/2);
+			BCI_DRAW_INDICES_S3D(count, prim, start);
+
+			for (i = start+1; i+1 < start+count; i += 2)
+				BCI_WRITE(i | ((i+1) << 16));
+			if (i < start+count)
+				BCI_WRITE(i);
+		} else {
+			BEGIN_BCI((count+2+1)/2);
+			BCI_DRAW_INDICES_S4(count, prim, skip);
+
+			for (i = start; i+1 < start+count; i += 2)
+				BCI_WRITE(i | ((i+1) << 16));
+			if (i < start+count)
+				BCI_WRITE(i);
+		}
+
+		start += count;
+		n -= count;
+
+		prim |= BCI_CMD_DRAW_CONT;
+	}
+
+	return 0;
+}
+
+static int savage_dispatch_vb_prim(drm_savage_private_t *dev_priv,
+				   const drm_savage_cmd_header_t *cmd_header,
+				   const uint32_t __user *vtxbuf,
+				   unsigned int vb_size,
+				   unsigned int vb_stride)
+{
+	unsigned char reorder = 0;
+	unsigned int prim = cmd_header->prim.prim;
+	unsigned int skip = cmd_header->prim.skip;
+	unsigned int n = cmd_header->prim.count;
+	unsigned int start = cmd_header->prim.start;
+	unsigned int vtx_size;
+	unsigned int i;
+	DMA_LOCALS;
+
+	if (!n)
+		return 0;
+
+	switch (prim) {
+	case SAVAGE_PRIM_TRILIST_201:
+		reorder = 1;
+		prim = SAVAGE_PRIM_TRILIST;
+	case SAVAGE_PRIM_TRILIST:
+		if (n % 3 != 0) {
+			DRM_ERROR("wrong number of vertices %u in TRILIST\n",
+				  n);
+			return DRM_ERR(EINVAL);
+		}
+		break;
+	case SAVAGE_PRIM_TRISTRIP:
+	case SAVAGE_PRIM_TRIFAN:
+		if (n < 3) {
+			DRM_ERROR("wrong number of vertices %u in TRIFAN/STRIP\n",
+				  n);
+			return DRM_ERR(EINVAL);
+		}
+		break;
+	default:
+		DRM_ERROR("invalid primitive type %u\n", prim);
+		return DRM_ERR(EINVAL);
+	}
+
+	if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
+		if (skip > SAVAGE_SKIP_ALL_S3D) {
+			DRM_ERROR("invalid skip flags 0x%04x\n", skip);
+			return DRM_ERR(EINVAL);
+		}
+		vtx_size = 8; /* full vertex */
+	} else {
+		if (skip > SAVAGE_SKIP_ALL_S4) {
+			DRM_ERROR("invalid skip flags 0x%04x\n", skip);
+			return DRM_ERR(EINVAL);
+		}
+		vtx_size = 10; /* full vertex */
+	}
+
+	vtx_size -= (skip & 1) + (skip >> 1 & 1) +
+		(skip >> 2 & 1) + (skip >> 3 & 1) + (skip >> 4 & 1) +
+		(skip >> 5 & 1) + (skip >> 6 & 1) + (skip >> 7 & 1);
+
+	if (vtx_size > vb_stride) {
+		DRM_ERROR("vertex size greater than vb stride (%u > %u)\n",
+			  vtx_size, vb_stride);
+		return DRM_ERR(EINVAL);
+	}
+
+	if (start + n > vb_size / (vb_stride*4)) {
+		DRM_ERROR("vertex indices (%u-%u) out of range (0-%u)\n",
+			  start, start + n - 1, vb_size / (vb_stride*4));
+		return DRM_ERR(EINVAL);
+	}
+
+	prim <<= 25;
+	while (n != 0) {
+		/* Can emit up to 255 vertices (85 triangles) at once. */
+		unsigned int count = n > 255 ? 255 : n;
+		if (reorder) {
+			/* Need to reorder vertices for correct flat
+			 * shading while preserving the clock sense
+			 * for correct culling. Only on Savage3D. */
+			int reorder[3] = {-1, -1, -1};
+			reorder[start%3] = 2;
+
+			BEGIN_DMA(count*vtx_size+1);
+			DMA_DRAW_PRIMITIVE(count, prim, skip);
+
+			for (i = start; i < start+count; ++i) {
+				unsigned int j = i + reorder[i % 3];
+				DMA_COPY_FROM_USER(&vtxbuf[vb_stride*j],
+						   vtx_size);
+			}
+
+			DMA_COMMIT();
+		} else {
+			BEGIN_DMA(count*vtx_size+1);
+			DMA_DRAW_PRIMITIVE(count, prim, skip);
+
+			if (vb_stride == vtx_size) {
+				DMA_COPY_FROM_USER(&vtxbuf[vb_stride*start],
+						   vtx_size*count);
+			} else {
+				for (i = start; i < start+count; ++i) {
+					DMA_COPY_FROM_USER(
+						&vtxbuf[vb_stride*i],
+						vtx_size);
+				}
+			}
+
+			DMA_COMMIT();
+		}
+
+		start += count;
+		n -= count;
+
+		prim |= BCI_CMD_DRAW_CONT;
+	}
+
+	return 0;
+}
+
+static int savage_dispatch_dma_idx(drm_savage_private_t *dev_priv,
+				   const drm_savage_cmd_header_t *cmd_header,
+				   const uint16_t __user *usr_idx,
+				   const drm_buf_t *dmabuf)
+{
+	unsigned char reorder = 0;
+	unsigned int prim = cmd_header->idx.prim;
+	unsigned int skip = cmd_header->idx.skip;
+	unsigned int n = cmd_header->idx.count;
+	unsigned int i;
+	BCI_LOCALS;
+
+	if (!dmabuf) {
+	    DRM_ERROR("called without dma buffers!\n");
+	    return DRM_ERR(EINVAL);
+	}
+
+	if (!n)
+		return 0;
+
+	switch (prim) {
+	case SAVAGE_PRIM_TRILIST_201:
+		reorder = 1;
+		prim = SAVAGE_PRIM_TRILIST;
+	case SAVAGE_PRIM_TRILIST:
+		if (n % 3 != 0) {
+			DRM_ERROR("wrong number of indices %u in TRILIST\n",
+				  n);
+			return DRM_ERR(EINVAL);
+		}
+		break;
+	case SAVAGE_PRIM_TRISTRIP:
+	case SAVAGE_PRIM_TRIFAN:
+		if (n < 3) {
+			DRM_ERROR("wrong number of indices %u in TRIFAN/STRIP\n",
+				  n);
+			return DRM_ERR(EINVAL);
+		}
+		break;
+	default:
+		DRM_ERROR("invalid primitive type %u\n", prim);
+		return DRM_ERR(EINVAL);
+	}
+
+	if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
+		if (skip != 0) {
+			DRM_ERROR("invalid skip flags 0x%04x for DMA\n",
+				  skip);
+			return DRM_ERR(EINVAL);
+		}
+	} else {
+		unsigned int size = 10 - (skip & 1) - (skip >> 1 & 1) -
+			(skip >> 2 & 1) - (skip >> 3 & 1) - (skip >> 4 & 1) -
+			(skip >> 5 & 1) - (skip >> 6 & 1) - (skip >> 7 & 1);
+		if (skip > SAVAGE_SKIP_ALL_S4 || size != 8) {
+			DRM_ERROR("invalid skip flags 0x%04x for DMA\n",
+				  skip);
+			return DRM_ERR(EINVAL);
+		}
+		if (reorder) {
+			DRM_ERROR("TRILIST_201 used on Savage4 hardware\n");
+			return DRM_ERR(EINVAL);
+		}
+	}
+
+	/* Vertex DMA doesn't work with command DMA at the same time,
+	 * so we use BCI_... to submit commands here. Flush buffered
+	 * faked DMA first. */
+	DMA_FLUSH();
+
+	if (dmabuf->bus_address != dev_priv->state.common.vbaddr) {
+		BEGIN_BCI(2);
+		BCI_SET_REGISTERS(SAVAGE_VERTBUFADDR, 1);
+		BCI_WRITE(dmabuf->bus_address | dev_priv->dma_type);
+		dev_priv->state.common.vbaddr = dmabuf->bus_address;
+	}
+	if (S3_SAVAGE3D_SERIES(dev_priv->chipset) && dev_priv->waiting) {
+		/* Workaround for what looks like a hardware bug. If a
+		 * WAIT_3D_IDLE was emitted some time before the
+		 * indexed drawing command then the engine will lock
+		 * up. There are two known workarounds:
+		 * WAIT_IDLE_EMPTY or emit at least 63 NOPs. */
+		BEGIN_BCI(63);
+		for (i = 0; i < 63; ++i)
+			BCI_WRITE(BCI_CMD_WAIT);
+		dev_priv->waiting = 0;
+	}
+
+	prim <<= 25;
+	while (n != 0) {
+		/* Can emit up to 255 indices (85 triangles) at once. */
+		unsigned int count = n > 255 ? 255 : n;
+		/* Is it ok to allocate 510 bytes on the stack in an ioctl? */
+		uint16_t idx[255];
+
+		/* Copy and check indices */
+		DRM_COPY_FROM_USER_UNCHECKED(idx, usr_idx, count*2);
+		for (i = 0; i < count; ++i) {
+			if (idx[i] > dmabuf->total/32) {
+				DRM_ERROR("idx[%u]=%u out of range (0-%u)\n",
+					  i, idx[i], dmabuf->total/32);
+				return DRM_ERR(EINVAL);
+			}
+		}
+
+		if (reorder) {
+			/* Need to reorder indices for correct flat
+			 * shading while preserving the clock sense
+			 * for correct culling. Only on Savage3D. */
+			int reorder[3] = {2, -1, -1};
+
+			BEGIN_BCI((count+1+1)/2);
+			BCI_DRAW_INDICES_S3D(count, prim, idx[2]);
+
+			for (i = 1; i+1 < count; i += 2)
+				BCI_WRITE(idx[i + reorder[i % 3]] |
+					  (idx[i+1 + reorder[(i+1) % 3]] << 16));
+			if (i < count)
+				BCI_WRITE(idx[i + reorder[i%3]]);
+		} else if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
+			BEGIN_BCI((count+1+1)/2);
+			BCI_DRAW_INDICES_S3D(count, prim, idx[0]);
+
+			for (i = 1; i+1 < count; i += 2)
+				BCI_WRITE(idx[i] | (idx[i+1] << 16));
+			if (i < count)
+				BCI_WRITE(idx[i]);
+		} else {
+			BEGIN_BCI((count+2+1)/2);
+			BCI_DRAW_INDICES_S4(count, prim, skip);
+
+			for (i = 0; i+1 < count; i += 2)
+				BCI_WRITE(idx[i] | (idx[i+1] << 16));
+			if (i < count)
+				BCI_WRITE(idx[i]);
+		}
+
+		usr_idx += count;
+		n -= count;
+
+		prim |= BCI_CMD_DRAW_CONT;
+	}
+
+	return 0;
+}
+
+static int savage_dispatch_vb_idx(drm_savage_private_t *dev_priv,
+				  const drm_savage_cmd_header_t *cmd_header,
+				  const uint16_t __user *usr_idx,
+				  const uint32_t __user *vtxbuf,
+				  unsigned int vb_size,
+				  unsigned int vb_stride)
+{
+	unsigned char reorder = 0;
+	unsigned int prim = cmd_header->idx.prim;
+	unsigned int skip = cmd_header->idx.skip;
+	unsigned int n = cmd_header->idx.count;
+	unsigned int vtx_size;
+	unsigned int i;
+	DMA_LOCALS;
+
+	if (!n)
+		return 0;
+
+	switch (prim) {
+	case SAVAGE_PRIM_TRILIST_201:
+		reorder = 1;
+		prim = SAVAGE_PRIM_TRILIST;
+	case SAVAGE_PRIM_TRILIST:
+		if (n % 3 != 0) {
+			DRM_ERROR("wrong number of indices %u in TRILIST\n",
+				  n);
+			return DRM_ERR(EINVAL);
+		}
+		break;
+	case SAVAGE_PRIM_TRISTRIP:
+	case SAVAGE_PRIM_TRIFAN:
+		if (n < 3) {
+			DRM_ERROR("wrong number of indices %u in TRIFAN/STRIP\n",
+				  n);
+			return DRM_ERR(EINVAL);
+		}
+		break;
+	default:
+		DRM_ERROR("invalid primitive type %u\n", prim);
+		return DRM_ERR(EINVAL);
+	}
+
+	if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
+		if (skip > SAVAGE_SKIP_ALL_S3D) {
+			DRM_ERROR("invalid skip flags 0x%04x\n", skip);
+			return DRM_ERR(EINVAL);
+		}
+		vtx_size = 8; /* full vertex */
+	} else {
+		if (skip > SAVAGE_SKIP_ALL_S4) {
+			DRM_ERROR("invalid skip flags 0x%04x\n", skip);
+			return DRM_ERR(EINVAL);
+		}
+		vtx_size = 10; /* full vertex */
+	}
+
+	vtx_size -= (skip & 1) + (skip >> 1 & 1) +
+		(skip >> 2 & 1) + (skip >> 3 & 1) + (skip >> 4 & 1) +
+		(skip >> 5 & 1) + (skip >> 6 & 1) + (skip >> 7 & 1);
+
+	if (vtx_size > vb_stride) {
+		DRM_ERROR("vertex size greater than vb stride (%u > %u)\n",
+			  vtx_size, vb_stride);
+		return DRM_ERR(EINVAL);
+	}
+
+	prim <<= 25;
+	while (n != 0) {
+		/* Can emit up to 255 vertices (85 triangles) at once. */
+		unsigned int count = n > 255 ? 255 : n;
+		/* Is it ok to allocate 510 bytes on the stack in an ioctl? */
+		uint16_t idx[255];
+
+		/* Copy and check indices */
+		DRM_COPY_FROM_USER_UNCHECKED(idx, usr_idx, count*2);
+		for (i = 0; i < count; ++i) {
+			if (idx[i] > vb_size / (vb_stride*4)) {
+				DRM_ERROR("idx[%u]=%u out of range (0-%u)\n",
+					  i, idx[i],  vb_size / (vb_stride*4));
+				return DRM_ERR(EINVAL);
+			}
+		}
+
+		if (reorder) {
+			/* Need to reorder vertices for correct flat
+			 * shading while preserving the clock sense
+			 * for correct culling. Only on Savage3D. */
+			int reorder[3] = {2, -1, -1};
+
+			BEGIN_DMA(count*vtx_size+1);
+			DMA_DRAW_PRIMITIVE(count, prim, skip);
+
+			for (i = 0; i < count; ++i) {
+				unsigned int j = idx[i + reorder[i % 3]];
+				DMA_COPY_FROM_USER(&vtxbuf[vb_stride*j],
+						   vtx_size);
+			}
+
+			DMA_COMMIT();
+		} else {
+			BEGIN_DMA(count*vtx_size+1);
+			DMA_DRAW_PRIMITIVE(count, prim, skip);
+
+			for (i = 0; i < count; ++i) {
+				unsigned int j = idx[i];
+				DMA_COPY_FROM_USER(&vtxbuf[vb_stride*j],
+						   vtx_size);
+			}
+
+			DMA_COMMIT();
+		}
+
+		usr_idx += count;
+		n -= count;
+
+		prim |= BCI_CMD_DRAW_CONT;
+	}
+
+	return 0;
+}
+
+static int savage_dispatch_clear(drm_savage_private_t *dev_priv,
+				 const drm_savage_cmd_header_t *cmd_header,
+				 const drm_savage_cmd_header_t __user *data,
+				 unsigned int nbox,
+				 const drm_clip_rect_t __user *usr_boxes)
+{
+	unsigned int flags = cmd_header->clear0.flags, mask, value;
+	unsigned int clear_cmd;
+	unsigned int i, nbufs;
+	DMA_LOCALS;
+
+	if (nbox == 0)
+		return 0;
+
+	DRM_GET_USER_UNCHECKED(mask, &((const drm_savage_cmd_header_t*)data)
+			       ->clear1.mask);
+	DRM_GET_USER_UNCHECKED(value, &((const drm_savage_cmd_header_t*)data)
+			       ->clear1.value);
+
+	clear_cmd = BCI_CMD_RECT | BCI_CMD_RECT_XP | BCI_CMD_RECT_YP |
+		BCI_CMD_SEND_COLOR | BCI_CMD_DEST_PBD_NEW;
+	BCI_CMD_SET_ROP(clear_cmd,0xCC);
+
+	nbufs = ((flags & SAVAGE_FRONT) ? 1 : 0) +
+		((flags & SAVAGE_BACK) ? 1 : 0) +
+		((flags & SAVAGE_DEPTH) ? 1 : 0);
+	if (nbufs == 0)
+		return 0;
+
+	if (mask != 0xffffffff) {
+		/* set mask */
+		BEGIN_DMA(2);
+		DMA_SET_REGISTERS(SAVAGE_BITPLANEWTMASK, 1);
+		DMA_WRITE(mask);
+		DMA_COMMIT();
+	}
+	for (i = 0; i < nbox; ++i) {
+		drm_clip_rect_t box;
+		unsigned int x, y, w, h;
+		unsigned int buf;
+		DRM_COPY_FROM_USER_UNCHECKED(&box, &usr_boxes[i], sizeof(box));
+		x = box.x1, y = box.y1;
+		w = box.x2 - box.x1;
+		h = box.y2 - box.y1;
+		BEGIN_DMA(nbufs*6);
+		for (buf = SAVAGE_FRONT; buf <= SAVAGE_DEPTH; buf <<= 1) {
+			if (!(flags & buf))
+				continue;
+			DMA_WRITE(clear_cmd);
+			switch(buf) {
+			case SAVAGE_FRONT:
+				DMA_WRITE(dev_priv->front_offset);
+				DMA_WRITE(dev_priv->front_bd);
+				break;
+			case SAVAGE_BACK:
+				DMA_WRITE(dev_priv->back_offset);
+				DMA_WRITE(dev_priv->back_bd);
+				break;
+			case SAVAGE_DEPTH:
+				DMA_WRITE(dev_priv->depth_offset);
+				DMA_WRITE(dev_priv->depth_bd);
+				break;
+			}
+			DMA_WRITE(value);
+			DMA_WRITE(BCI_X_Y(x, y));
+			DMA_WRITE(BCI_W_H(w, h));
+		}
+		DMA_COMMIT();
+	}
+	if (mask != 0xffffffff) {
+		/* reset mask */
+		BEGIN_DMA(2);
+		DMA_SET_REGISTERS(SAVAGE_BITPLANEWTMASK, 1);
+		DMA_WRITE(0xffffffff);
+		DMA_COMMIT();
+	}
+
+	return 0;
+}
+
+static int savage_dispatch_swap(drm_savage_private_t *dev_priv,
+				unsigned int nbox,
+				const drm_clip_rect_t __user *usr_boxes)
+{
+	unsigned int swap_cmd;
+	unsigned int i;
+	DMA_LOCALS;
+
+	if (nbox == 0)
+		return 0;
+
+	swap_cmd = BCI_CMD_RECT | BCI_CMD_RECT_XP | BCI_CMD_RECT_YP |
+		BCI_CMD_SRC_PBD_COLOR_NEW | BCI_CMD_DEST_GBD;
+	BCI_CMD_SET_ROP(swap_cmd,0xCC);
+
+	for (i = 0; i < nbox; ++i) {
+		drm_clip_rect_t box;
+		DRM_COPY_FROM_USER_UNCHECKED(&box, &usr_boxes[i], sizeof(box));
+
+		BEGIN_DMA(6);
+		DMA_WRITE(swap_cmd);
+		DMA_WRITE(dev_priv->back_offset);
+		DMA_WRITE(dev_priv->back_bd);
+		DMA_WRITE(BCI_X_Y(box.x1, box.y1));
+		DMA_WRITE(BCI_X_Y(box.x1, box.y1));
+		DMA_WRITE(BCI_W_H(box.x2-box.x1, box.y2-box.y1));
+		DMA_COMMIT();
+	}
+
+	return 0;
+}
+
+static int savage_dispatch_draw(drm_savage_private_t *dev_priv,
+				const drm_savage_cmd_header_t __user *start,
+				const drm_savage_cmd_header_t __user *end,
+				const drm_buf_t *dmabuf,
+				const unsigned int __user *usr_vtxbuf,
+				unsigned int vb_size, unsigned int vb_stride,
+				unsigned int nbox,
+				const drm_clip_rect_t __user *usr_boxes)
+{
+	unsigned int i, j;
+	int ret;
+
+	for (i = 0; i < nbox; ++i) {
+		drm_clip_rect_t box;
+		const drm_savage_cmd_header_t __user *usr_cmdbuf;
+		DRM_COPY_FROM_USER_UNCHECKED(&box, &usr_boxes[i], sizeof(box));
+		dev_priv->emit_clip_rect(dev_priv, &box);
+
+		usr_cmdbuf = start;
+		while (usr_cmdbuf < end) {
+			drm_savage_cmd_header_t cmd_header;
+			DRM_COPY_FROM_USER_UNCHECKED(&cmd_header, usr_cmdbuf,
+						     sizeof(cmd_header));
+			usr_cmdbuf++;
+			switch (cmd_header.cmd.cmd) {
+			case SAVAGE_CMD_DMA_PRIM:
+				ret = savage_dispatch_dma_prim(
+					dev_priv, &cmd_header, dmabuf);
+				break;
+			case SAVAGE_CMD_VB_PRIM:
+				ret = savage_dispatch_vb_prim(
+					dev_priv, &cmd_header,
+					(const uint32_t __user *)usr_vtxbuf,
+					vb_size, vb_stride);
+				break;
+			case SAVAGE_CMD_DMA_IDX:
+				j = (cmd_header.idx.count + 3) / 4;
+				/* j was check in savage_bci_cmdbuf */
+				ret = savage_dispatch_dma_idx(
+					dev_priv, &cmd_header,
+					(const uint16_t __user *)usr_cmdbuf,
+					dmabuf);
+				usr_cmdbuf += j;
+				break;
+			case SAVAGE_CMD_VB_IDX:
+				j = (cmd_header.idx.count + 3) / 4;
+				/* j was check in savage_bci_cmdbuf */
+				ret = savage_dispatch_vb_idx(
+					dev_priv, &cmd_header,
+					(const uint16_t __user *)usr_cmdbuf,
+					(const uint32_t __user *)usr_vtxbuf,
+					vb_size, vb_stride);
+				usr_cmdbuf += j;
+				break;
+			default:
+				/* What's the best return code? EFAULT? */
+				DRM_ERROR("IMPLEMENTATION ERROR: "
+					  "non-drawing-command %d\n",
+					  cmd_header.cmd.cmd);
+				return DRM_ERR(EINVAL);
+			}
+
+			if (ret != 0)
+				return ret;
+		}
+	}
+
+	return 0;
+}
+
+int savage_bci_cmdbuf(DRM_IOCTL_ARGS)
+{
+	DRM_DEVICE;
+	drm_savage_private_t *dev_priv = dev->dev_private;
+	drm_device_dma_t *dma = dev->dma;
+	drm_buf_t *dmabuf;
+	drm_savage_cmdbuf_t cmdbuf;
+	drm_savage_cmd_header_t __user *usr_cmdbuf;
+	drm_savage_cmd_header_t __user *first_draw_cmd;
+	unsigned int __user *usr_vtxbuf;
+	drm_clip_rect_t __user *usr_boxes;
+	unsigned int i, j;
+	int ret = 0;
+
+	DRM_DEBUG("\n");
+	
+	LOCK_TEST_WITH_RETURN(dev, filp);
+
+	DRM_COPY_FROM_USER_IOCTL(cmdbuf, (drm_savage_cmdbuf_t __user *)data,
+				 sizeof(cmdbuf));
+
+	if (dma && dma->buflist) {
+		if (cmdbuf.dma_idx > dma->buf_count) {
+			DRM_ERROR("vertex buffer index %u out of range (0-%u)\n",
+				  cmdbuf.dma_idx, dma->buf_count-1);
+			return DRM_ERR(EINVAL);
+		}
+		dmabuf = dma->buflist[cmdbuf.dma_idx];
+	} else {
+		dmabuf = NULL;
+	}
+
+	usr_cmdbuf = (drm_savage_cmd_header_t __user *)cmdbuf.cmd_addr;
+	usr_vtxbuf = (unsigned int __user *)cmdbuf.vb_addr;
+	usr_boxes = (drm_clip_rect_t __user *)cmdbuf.box_addr;
+	if ((cmdbuf.size && DRM_VERIFYAREA_READ(usr_cmdbuf, cmdbuf.size*8)) ||
+	    (cmdbuf.vb_size && DRM_VERIFYAREA_READ(
+		    usr_vtxbuf, cmdbuf.vb_size)) ||
+	    (cmdbuf.nbox && DRM_VERIFYAREA_READ(
+		    usr_boxes, cmdbuf.nbox*sizeof(drm_clip_rect_t))))
+		return DRM_ERR(EFAULT);
+
+	/* Make sure writes to DMA buffers are finished before sending
+	 * DMA commands to the graphics hardware. */
+	DRM_MEMORYBARRIER();
+
+	/* Coming from user space. Don't know if the Xserver has
+	 * emitted wait commands. Assuming the worst. */
+	dev_priv->waiting = 1;
+
+	i = 0;
+	first_draw_cmd = NULL;
+	while (i < cmdbuf.size) {
+		drm_savage_cmd_header_t cmd_header;
+		DRM_COPY_FROM_USER_UNCHECKED(&cmd_header, usr_cmdbuf,
+					     sizeof(cmd_header));
+		usr_cmdbuf++;
+		i++;
+
+		/* Group drawing commands with same state to minimize
+		 * iterations over clip rects. */
+		j = 0;
+		switch (cmd_header.cmd.cmd) {
+		case SAVAGE_CMD_DMA_IDX:
+		case SAVAGE_CMD_VB_IDX:
+			j = (cmd_header.idx.count + 3) / 4;
+			if (i + j > cmdbuf.size) {
+				DRM_ERROR("indexed drawing command extends "
+					  "beyond end of command buffer\n");
+				DMA_FLUSH();
+				return DRM_ERR(EINVAL);
+			}
+			/* fall through */
+		case SAVAGE_CMD_DMA_PRIM:
+		case SAVAGE_CMD_VB_PRIM:
+			if (!first_draw_cmd)
+				first_draw_cmd = usr_cmdbuf-1;
+			usr_cmdbuf += j;
+			i += j;
+			break;
+		default:
+			if (first_draw_cmd) {
+				ret = savage_dispatch_draw (
+					dev_priv, first_draw_cmd, usr_cmdbuf-1,
+					dmabuf, usr_vtxbuf, cmdbuf.vb_size,
+					cmdbuf.vb_stride,
+					cmdbuf.nbox, usr_boxes);
+				if (ret != 0)
+					return ret;
+				first_draw_cmd = NULL;
+			}
+		}
+		if (first_draw_cmd)
+			continue;
+
+		switch (cmd_header.cmd.cmd) {
+		case SAVAGE_CMD_STATE:
+			j = (cmd_header.state.count + 1) / 2;
+			if (i + j > cmdbuf.size) {
+				DRM_ERROR("command SAVAGE_CMD_STATE extends "
+					  "beyond end of command buffer\n");
+				DMA_FLUSH();
+				return DRM_ERR(EINVAL);
+			}
+			ret = savage_dispatch_state(
+				dev_priv, &cmd_header,
+				(uint32_t __user *)usr_cmdbuf);
+			usr_cmdbuf += j;
+			i += j;
+			break;
+		case SAVAGE_CMD_CLEAR:
+			if (i + 1 > cmdbuf.size) {
+				DRM_ERROR("command SAVAGE_CMD_CLEAR extends "
+					  "beyond end of command buffer\n");
+				DMA_FLUSH();
+				return DRM_ERR(EINVAL);
+			}
+			ret = savage_dispatch_clear(dev_priv, &cmd_header,
+						    usr_cmdbuf,
+						    cmdbuf.nbox, usr_boxes);
+			usr_cmdbuf++;
+			i++;
+			break;
+		case SAVAGE_CMD_SWAP:
+			ret = savage_dispatch_swap(dev_priv,
+						   cmdbuf.nbox, usr_boxes);
+			break;
+		default:
+			DRM_ERROR("invalid command 0x%x\n", cmd_header.cmd.cmd);
+			DMA_FLUSH();
+			return DRM_ERR(EINVAL);
+		}
+
+		if (ret != 0) {
+			DMA_FLUSH();
+			return ret;
+		}
+	}
+
+	if (first_draw_cmd) {
+		ret = savage_dispatch_draw (
+			dev_priv, first_draw_cmd, usr_cmdbuf, dmabuf,
+			usr_vtxbuf, cmdbuf.vb_size, cmdbuf.vb_stride,
+			cmdbuf.nbox, usr_boxes);
+		if (ret != 0) {
+			DMA_FLUSH();
+			return ret;
+		}
+	}
+
+	DMA_FLUSH();
+
+	if (dmabuf && cmdbuf.discard) {
+		drm_savage_buf_priv_t *buf_priv = dmabuf->dev_private;
+		uint16_t event;
+		event = savage_bci_emit_event(dev_priv, SAVAGE_WAIT_3D);
+		SET_AGE(&buf_priv->age, event, dev_priv->event_wrap);
+		savage_freelist_put(dev, dmabuf);
+	}
+
+	return 0;
+}