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

44 files changed, 5515 insertions, 4150 deletions

diff --git a/libdrm/xf86drm.c b/libdrm/xf86drm.c
index 3b0f98ac..8b97c7da 100644
--- a/libdrm/xf86drm.c
+++ b/libdrm/xf86drm.c
@@ -418,7 +418,8 @@ int drmAddMap(int fd,
     return 0;
 }
 
-int drmAddBufs(int fd, int count, int size, int flags)
+int drmAddBufs(int fd, int count, int size, drmBufDescFlags flags,
+	       int agp_offset)
 {
     drm_buf_desc_t request;
     
@@ -427,6 +428,8 @@ int drmAddBufs(int fd, int count, int size, int flags)
     request.low_mark  = 0;
     request.high_mark = 0;
     request.flags     = flags;
+    request.agp_start = agp_offset;
+   
     if (ioctl(fd, DRM_IOCTL_ADD_BUFS, &request)) return -errno;
     return request.count;
 }
@@ -744,6 +747,143 @@ int drmDestroyDrawable(int fd, drmDrawable handle)
     return 0;
 }
 
+int drmAgpAcquire(int fd)
+{
+    if (ioctl(fd, DRM_IOCTL_AGP_ACQUIRE, NULL)) return -errno;
+    return 0;
+}
+
+int drmAgpRelease(int fd)
+{
+    if (ioctl(fd, DRM_IOCTL_AGP_RELEASE, NULL)) return -errno;
+    return 0;
+}
+
+int drmAgpEnable(int fd, unsigned long mode)
+{
+    drm_agp_mode_t m;
+
+    m.mode = mode;
+    if (ioctl(fd, DRM_IOCTL_AGP_ENABLE, &m)) return -errno;
+    return 0;
+}
+
+int drmAgpAlloc(int fd, unsigned long size, unsigned long type,
+		unsigned long *address, unsigned long *handle)
+{
+    drm_agp_buffer_t b;
+    *handle = 0;
+    b.size   = size;
+    b.handle = 0;
+    b.type   = type;
+    if (ioctl(fd, DRM_IOCTL_AGP_ALLOC, &b)) return -errno;
+    if (address != 0UL) *address = b.physical;
+    *handle = b.handle;
+    return 0;
+}
+
+int drmAgpFree(int fd, unsigned long handle)
+{
+    drm_agp_buffer_t b;
+
+    b.size   = 0;
+    b.handle = handle;
+    if (ioctl(fd, DRM_IOCTL_AGP_FREE, &b)) return -errno;
+    return 0;
+}
+
+int drmAgpBind(int fd, unsigned long handle, unsigned long offset)
+{
+    drm_agp_binding_t b;
+
+    b.handle = handle;
+    b.offset = offset;
+    if (ioctl(fd, DRM_IOCTL_AGP_BIND, &b)) return -errno;
+    return 0;
+}
+
+int drmAgpUnbind(int fd, unsigned long handle)
+{
+    drm_agp_binding_t b;
+
+    b.handle = handle;
+    b.offset = 0;
+    if (ioctl(fd, DRM_IOCTL_AGP_UNBIND, &b)) return -errno;
+    return 0;
+}
+
+int drmAgpVersionMajor(int fd)
+{
+    drm_agp_info_t i;
+
+    if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) return -errno;
+    return i.agp_version_major;
+}
+
+int drmAgpVersionMinor(int fd)
+{
+    drm_agp_info_t i;
+
+    if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) return -errno;
+    return i.agp_version_minor;
+}
+
+unsigned long drmAgpGetMode(int fd)
+{
+    drm_agp_info_t i;
+
+    if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) return 0;
+    return i.mode;
+}
+
+unsigned long drmAgpBase(int fd)
+{
+    drm_agp_info_t i;
+
+    if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) return 0;
+    return i.aperture_base;
+}
+
+unsigned long drmAgpSize(int fd)
+{
+    drm_agp_info_t i;
+
+    if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) return 0;
+    return i.aperture_size;
+}
+
+unsigned long drmAgpMemoryUsed(int fd)
+{
+    drm_agp_info_t i;
+
+    if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) return 0;
+    return i.memory_used;
+}
+
+unsigned long drmAgpMemoryAvail(int fd)
+{
+    drm_agp_info_t i;
+
+    if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) return 0;
+    return i.memory_allowed;
+}
+
+unsigned int drmAgpVendorId(int fd)
+{
+    drm_agp_info_t i;
+
+    if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) return 0;
+    return i.id_vendor;
+}
+
+unsigned int drmAgpDeviceId(int fd)
+{
+    drm_agp_info_t i;
+
+    if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) return 0;
+    return i.id_device;
+}
+
 int drmError(int err, const char *label)
 {
     switch (err) {
diff --git a/linux-core/Makefile.kernel b/linux-core/Makefile.kernel
index 44d75c48..2ea6c721 100644
--- a/linux-core/Makefile.kernel
+++ b/linux-core/Makefile.kernel
@@ -14,7 +14,8 @@
 L_TARGET        := libdrm.a
 
 L_OBJS		:= init.o memory.o proc.o auth.o context.o drawable.o bufs.o \
-			lists.o lock.o ioctl.o fops.o vm.o dma.o
+			lists.o lock.o ioctl.o fops.o vm.o dma.o ctxbitmap.o \
+			agpsupport.o
 
 M_OBJS		:= 
 
@@ -26,6 +27,14 @@ ifdef CONFIG_DRM_TDFX
 M_OBJS          += tdfx.o
 endif
 
+ifdef CONFIG_DRM_MGA
+M_OBJS		+= mga.o
+endif
+
+ifdef CONFIG_DRM_R128
+M_OBJS		+= r128.o
+endif
+
 include $(TOPDIR)/Rules.make
 
 gamma.o: gamma_drv.o gamma_dma.o $(L_TARGET)
@@ -33,3 +42,10 @@ gamma.o: gamma_drv.o gamma_dma.o $(L_TARGET)
 
 tdfx.o: tdfx_drv.o tdfx_context.o $(L_TARGET)
 	$(LD) $(LD_RFLAG) -r -o $@ tdfx_drv.o tdfx_context.o -L. -ldrm
+
+i810.o: i810_drv.o i810_context.o $(L_TARGET)
+	$(LD) $(LD_RFLAG) -r -o $@ i810_drv.o i810_bufs.o i810_dma.o i810_context.o -L. -ldrm
+
+mga.o: mga_drv.o mga_context.o mga_dma.o mga_bufs.o $(L_TARGET)
+	$(LD) $(LD_RFLAG) -r -o $@ mga_drv.o mga_bufs.o mga_dma.o mga_context.o mga_state.o -L. -ldrm
+
diff --git a/linux-core/drmP.h b/linux-core/drmP.h
index 312fba36..f8e78eab 100644
--- a/linux-core/drmP.h
+++ b/linux-core/drmP.h
@@ -48,8 +48,12 @@
 #ifdef CONFIG_MTRR
 #include <asm/mtrr.h>
 #endif
+#ifdef DRM_AGP
+#include <linux/types.h>
+#include <linux/agp_backend.h>
+#endif
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,1,0)
-#include <asm/spinlock.h>
+#include <linux/tqueue.h>
 #include <linux/poll.h>
 #endif
 #include "drm.h"
@@ -69,21 +73,27 @@
 #define DRM_FLAG_DEBUG	  0x01
 #define DRM_FLAG_NOCTX	  0x02
 
-#define DRM_MEM_DMA	  0
-#define DRM_MEM_SAREA	  1
-#define DRM_MEM_DRIVER	  2
-#define DRM_MEM_MAGIC	  3
-#define DRM_MEM_IOCTLS	  4
-#define DRM_MEM_MAPS	  5
-#define DRM_MEM_VMAS	  6
-#define DRM_MEM_BUFS	  7
-#define DRM_MEM_SEGS	  8
-#define DRM_MEM_PAGES	  9
-#define DRM_MEM_FILES	 10
-#define DRM_MEM_QUEUES	 11
-#define DRM_MEM_CMDS	 12
-#define DRM_MEM_MAPPINGS 13
-#define DRM_MEM_BUFLISTS 14
+#define DRM_MEM_DMA	   0
+#define DRM_MEM_SAREA	   1
+#define DRM_MEM_DRIVER	   2
+#define DRM_MEM_MAGIC	   3
+#define DRM_MEM_IOCTLS	   4
+#define DRM_MEM_MAPS	   5
+#define DRM_MEM_VMAS	   6
+#define DRM_MEM_BUFS	   7
+#define DRM_MEM_SEGS	   8
+#define DRM_MEM_PAGES	   9
+#define DRM_MEM_FILES	  10
+#define DRM_MEM_QUEUES	  11
+#define DRM_MEM_CMDS	  12
+#define DRM_MEM_MAPPINGS  13
+#define DRM_MEM_BUFLISTS  14
+#define DRM_MEM_AGPLISTS  15
+#define DRM_MEM_TOTALAGP  16
+#define DRM_MEM_BOUNDAGP  17
+#define DRM_MEM_CTXBITMAP 18
+
+#define DRM_MAX_CTXBITMAP (PAGE_SIZE * 8)
 
 				/* Backward compatibility section */
 				/* _PAGE_WT changed to _PAGE_PWT in 2.2.6 */
@@ -235,6 +245,7 @@ typedef struct drm_buf {
 	int		  used;	       /* Amount of buffer in use (for DMA)  */
 	unsigned long	  offset;      /* Byte offset (used internally)	     */
 	void		  *address;    /* Address of buffer		     */
+	unsigned long	  bus_address; /* Bus address of buffer		     */
 	struct drm_buf	  *next;       /* Kernel-only: used for free list    */
 	__volatile__ int  waiting;     /* On kernel DMA queue		     */
 	__volatile__ int  pending;     /* On hardware DMA queue		     */
@@ -250,6 +261,11 @@ typedef struct drm_buf {
 		DRM_LIST_PRIO	 = 4,
 		DRM_LIST_RECLAIM = 5
 	}		  list;	       /* Which list we're on		     */
+
+
+	void *dev_private;
+	int dev_priv_size;
+
 #if DRM_DMA_HISTOGRAM
 	cycles_t	  time_queued;	   /* Queued to kernel DMA queue     */
 	cycles_t	  time_dispatched; /* Dispatched to hardware	     */
@@ -376,6 +392,9 @@ typedef struct drm_device_dma {
 	int		  page_count;
 	unsigned long	  *pagelist;
 	unsigned long	  byte_count;
+	enum {
+	   _DRM_DMA_USE_AGP = 0x01
+	} flags;
 
 				/* DMA support */
 	drm_buf_t	  *this_buffer;	/* Buffer being sent		   */
@@ -384,6 +403,41 @@ typedef struct drm_device_dma {
 	wait_queue_head_t waiting;	/* Processes waiting on free bufs  */
 } drm_device_dma_t;
 
+#ifdef DRM_AGP
+typedef struct drm_agp_mem {
+	unsigned long      handle;
+	agp_memory         *memory;
+	unsigned long      bound; /* address */
+	int                pages;
+	struct drm_agp_mem *prev;
+	struct drm_agp_mem *next;
+} drm_agp_mem_t;
+
+typedef struct drm_agp_head {
+	agp_kern_info      agp_info;
+	const char         *chipset;
+	drm_agp_mem_t      *memory;
+	unsigned long      mode;
+	int                enabled;
+	int                acquired;
+	unsigned long      base;
+   	int 		   agp_mtrr;
+} drm_agp_head_t;
+
+typedef struct {
+	void       (*free_memory)(agp_memory *);
+	agp_memory *(*allocate_memory)(size_t, u32);
+	int        (*bind_memory)(agp_memory *, off_t);
+	int        (*unbind_memory)(agp_memory *);
+	void       (*enable)(u32);
+	int        (*acquire)(void);
+	void       (*release)(void);
+	void       (*copy_info)(agp_kern_info *);
+} drm_agp_func_t;
+
+extern drm_agp_func_t drm_agp;
+#endif
+
 typedef struct drm_device {
 	const char	  *name;	/* Simple driver name		   */
 	char		  *unique;	/* Unique identifier: e.g., busid  */
@@ -462,6 +516,12 @@ typedef struct drm_device {
 	struct fasync_struct *buf_async;/* Processes waiting for SIGIO	   */
 	wait_queue_head_t buf_readers;	/* Processes waiting to read	   */
 	wait_queue_head_t buf_writers;	/* Processes waiting to ctx switch */
+	
+#ifdef DRM_AGP
+	drm_agp_head_t    *agp;
+#endif
+	unsigned long     *ctx_bitmap;
+	void		  *dev_private;
 } drm_device_t;
 
 
@@ -533,6 +593,14 @@ extern void	     drm_free_pages(unsigned long address, int order,
 extern void	     *drm_ioremap(unsigned long offset, unsigned long size);
 extern void	     drm_ioremapfree(void *pt, unsigned long size);
 
+#ifdef DRM_AGP
+extern agp_memory    *drm_alloc_agp(int pages, u32 type);
+extern int           drm_free_agp(agp_memory *handle, int pages);
+extern int           drm_bind_agp(agp_memory *handle, unsigned int start);
+extern int           drm_unbind_agp(agp_memory *handle);
+#endif
+
+
 				/* Buffer management support (bufs.c) */
 extern int	     drm_order(unsigned long size);
 extern int	     drm_addmap(struct inode *inode, struct file *filp,
@@ -642,5 +710,32 @@ extern int	     drm_flush_unblock(drm_device_t *dev, int context,
 				       drm_lock_flags_t flags);
 extern int	     drm_flush_block_and_flush(drm_device_t *dev, int context,
 					       drm_lock_flags_t flags);
+
+				/* Context Bitmap support (ctxbitmap.c) */
+extern int	     drm_ctxbitmap_init(drm_device_t *dev);
+extern void	     drm_ctxbitmap_cleanup(drm_device_t *dev);
+extern int	     drm_ctxbitmap_next(drm_device_t *dev);
+extern void	     drm_ctxbitmap_free(drm_device_t *dev, int ctx_handle);
+
+#ifdef DRM_AGP
+				/* AGP/GART support (agpsupport.c) */
+extern drm_agp_head_t *drm_agp_init(void);
+extern int            drm_agp_acquire(struct inode *inode, struct file *filp,
+				      unsigned int cmd, unsigned long arg);
+extern int            drm_agp_release(struct inode *inode, struct file *filp,
+				      unsigned int cmd, unsigned long arg);
+extern int            drm_agp_enable(struct inode *inode, struct file *filp,
+				     unsigned int cmd, unsigned long arg);
+extern int            drm_agp_info(struct inode *inode, struct file *filp,
+				   unsigned int cmd, unsigned long arg);
+extern int            drm_agp_alloc(struct inode *inode, struct file *filp,
+				    unsigned int cmd, unsigned long arg);
+extern int            drm_agp_free(struct inode *inode, struct file *filp,
+				   unsigned int cmd, unsigned long arg);
+extern int            drm_agp_unbind(struct inode *inode, struct file *filp,
+				     unsigned int cmd, unsigned long arg);
+extern int            drm_agp_bind(struct inode *inode, struct file *filp,
+				   unsigned int cmd, unsigned long arg);
+#endif
 #endif
 #endif
diff --git a/linux-core/i810_dma.c b/linux-core/i810_dma.c
index 09959b65..30fda5b8 100644
--- a/linux-core/i810_dma.c
+++ b/linux-core/i810_dma.c
@@ -25,8 +25,9 @@
  *
  * Authors: Rickard E. (Rik) Faith <faith@precisioninsight.com>
  *	    Jeff Hartmann <jhartmann@precisioninsight.com>
+ *          Keith Whitwell <keithw@precisioninsight.com>
  *
- * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_dma.c,v 1.1 2000/02/11 17:26:04 dawes Exp $
+ * $XFree86$
  *
  */
 
@@ -36,6 +37,9 @@
 
 #include <linux/interrupt.h>	/* For task queue support */
 
+#define I810_BUF_FREE		1
+#define I810_BUF_USED		0
+
 #define I810_REG(reg)		2
 #define I810_BASE(reg)		((unsigned long) \
 				dev->maplist[I810_REG(reg)]->handle)
@@ -43,544 +47,457 @@
 #define I810_DEREF(reg)		*(__volatile__ int *)I810_ADDR(reg)
 #define I810_READ(reg)		I810_DEREF(reg)
 #define I810_WRITE(reg,val) 	do { I810_DEREF(reg) = val; } while (0)
-
-void i810_dma_init(drm_device_t *dev)
+#define I810_DEREF16(reg)	*(__volatile__ u16 *)I810_ADDR(reg)
+#define I810_READ16(reg)	I810_DEREF16(reg)
+#define I810_WRITE16(reg,val)	do { I810_DEREF16(reg) = val; } while (0)
+
+#define RING_LOCALS	unsigned int outring, ringmask; volatile char *virt;
+
+#define BEGIN_LP_RING(n) do {				\
+	if (I810_VERBOSE)				\
+		DRM_DEBUG("BEGIN_LP_RING(%d) in %s\n",	\
+			  n, __FUNCTION__);		\
+	if (dev_priv->ring.space < n*4) 		\
+		i810_wait_ring(dev, n*4);		\
+	dev_priv->ring.space -= n*4;			\
+	outring = dev_priv->ring.tail;			\
+	ringmask = dev_priv->ring.tail_mask;		\
+	virt = dev_priv->ring.virtual_start;		\
+} while (0)
+
+#define ADVANCE_LP_RING() do {					\
+	if (I810_VERBOSE) DRM_DEBUG("ADVANCE_LP_RING\n");	\
+	dev_priv->ring.tail = outring;				\
+	I810_WRITE(LP_RING + RING_TAIL, outring);		\
+} while(0)
+
+#define OUT_RING(n) do {						\
+	if (I810_VERBOSE) DRM_DEBUG("   OUT_RING %x\n", (int)(n));	\
+	*(volatile unsigned int *)(virt + outring) = n;			\
+	outring += 4;							\
+	outring &= ringmask;						\
+} while (0);
+
+static inline void i810_print_status_page(drm_device_t *dev)
 {
-        printk(KERN_INFO "i810_dma_init\n");
+   	drm_device_dma_t *dma = dev->dma;
+      	drm_i810_private_t *dev_priv = dev->dev_private;
+	u32 *temp = (u32 *)dev_priv->hw_status_page;
+   	int i;
+
+   	DRM_DEBUG(  "hw_status: Interrupt Status : %x\n", temp[0]);
+   	DRM_DEBUG(  "hw_status: LpRing Head ptr : %x\n", temp[1]);
+   	DRM_DEBUG(  "hw_status: IRing Head ptr : %x\n", temp[2]);
+      	DRM_DEBUG(  "hw_status: Reserved : %x\n", temp[3]);
+   	DRM_DEBUG(  "hw_status: Driver Counter : %d\n", temp[5]);
+   	for(i = 6; i < dma->buf_count + 6; i++) {
+	   	DRM_DEBUG(  "buffer status idx : %d used: %d\n", i - 6, temp[i]);
+	}
 }
 
-void i810_dma_cleanup(drm_device_t *dev)
+static drm_buf_t *i810_freelist_get(drm_device_t *dev)
 {
-   printk(KERN_INFO "i810_dma_cleanup\n");
+   	drm_device_dma_t *dma = dev->dma;
+	int		 i;
+   	int 		 used;
+   
+	/* Linear search might not be the best solution */
+
+   	for (i = 0; i < dma->buf_count; i++) {
+	   	drm_buf_t *buf = dma->buflist[ i ];
+	   	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+		/* In use is already a pointer */
+	   	used = cmpxchg(buf_priv->in_use, I810_BUF_FREE, 
+			       I810_BUF_USED);
+	   	if(used == I810_BUF_FREE) {
+			return buf;
+		}
+	}
+   	return NULL;
 }
 
-static inline void i810_dma_dispatch(drm_device_t *dev, unsigned long address,
-				    unsigned long length)
+/* This should only be called if the buffer is not sent to the hardware
+ * yet, the hardware updates in use for us once its on the ring buffer.
+ */
+
+static int i810_freelist_put(drm_device_t *dev, drm_buf_t *buf)
 {
-   printk(KERN_INFO "i810_dma_dispatch\n");
+   	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+   	int used;
+   
+   	/* In use is already a pointer */
+   	used = cmpxchg(buf_priv->in_use, I810_BUF_USED, I810_BUF_FREE);
+   	if(used != I810_BUF_USED) {
+	   	DRM_ERROR("Freeing buffer thats not in use : %d\n", buf->idx);
+	   	return -EINVAL;
+	}
+   
+   	return 0;
 }
 
-static inline void i810_dma_quiescent(drm_device_t *dev)
+static int i810_dma_get_buffers(drm_device_t *dev, drm_dma_t *d)
 {
+	int		  i;
+	drm_buf_t	  *buf;
+
+	for (i = d->granted_count; i < d->request_count; i++) {
+		buf = i810_freelist_get(dev);
+		if (!buf) break;
+		buf->pid     = current->pid;
+		copy_to_user_ret(&d->request_indices[i],
+				 &buf->idx,
+				 sizeof(buf->idx),
+				 -EFAULT);
+		copy_to_user_ret(&d->request_sizes[i],
+				 &buf->total,
+				 sizeof(buf->total),
+				 -EFAULT);
+		++d->granted_count;
+	}
+	return 0;
 }
 
-static inline void i810_dma_ready(drm_device_t *dev)
+static unsigned long i810_alloc_page(drm_device_t *dev)
 {
-   i810_dma_quiescent(dev);
-   printk(KERN_INFO "i810_dma_ready\n");
+	unsigned long address;
+   
+	address = __get_free_page(GFP_KERNEL);
+	if(address == 0UL) 
+		return 0;
+	
+	atomic_inc(&mem_map[MAP_NR((void *) address)].count);
+	set_bit(PG_locked, &mem_map[MAP_NR((void *) address)].flags);
+   
+	return address;
 }
 
-static inline int i810_dma_is_ready(drm_device_t *dev)
+static void i810_free_page(drm_device_t *dev, unsigned long page)
 {
-
-   i810_dma_quiescent(dev);
-
-   printk(KERN_INFO "i810_dma_is_ready\n");
-   return 1;
+	if(page == 0UL) 
+		return;
+	
+	atomic_dec(&mem_map[MAP_NR((void *) page)].count);
+	clear_bit(PG_locked, &mem_map[MAP_NR((void *) page)].flags);
+	wake_up(&mem_map[MAP_NR((void *) page)].wait);
+	free_page(page);
+	return;
 }
 
-
-static void i810_dma_service(int irq, void *device, struct pt_regs *regs)
+static int i810_dma_cleanup(drm_device_t *dev)
 {
-	drm_device_t	 *dev = (drm_device_t *)device;
-	drm_device_dma_t *dma = dev->dma;
-	
-	atomic_inc(&dev->total_irq);
-	if (i810_dma_is_ready(dev)) {
-				/* Free previous buffer */
-		if (test_and_set_bit(0, &dev->dma_flag)) {
-			atomic_inc(&dma->total_missed_free);
-			return;
+	if(dev->dev_private) {
+	   	drm_i810_private_t *dev_priv = 
+	     		(drm_i810_private_t *) dev->dev_private;
+	   
+	   	if(dev_priv->ring.virtual_start) {
+		   	drm_ioremapfree((void *) dev_priv->ring.virtual_start,
+					dev_priv->ring.Size);
 		}
-		if (dma->this_buffer) {
-			drm_free_buffer(dev, dma->this_buffer);
-			dma->this_buffer = NULL;
+	   	if(dev_priv->hw_status_page != 0UL) {
+		   	i810_free_page(dev, dev_priv->hw_status_page);
+		   	/* Need to rewrite hardware status page */
+		   	I810_WRITE(0x02080, 0x1ffff000);
 		}
-		clear_bit(0, &dev->dma_flag);
-
-				/* Dispatch new buffer */
-		queue_task(&dev->tq, &tq_immediate);
-		mark_bh(IMMEDIATE_BH);
+	   	drm_free(dev->dev_private, sizeof(drm_i810_private_t), 
+			 DRM_MEM_DRIVER);
+	   	dev->dev_private = NULL;
 	}
+   	return 0;
 }
 
-/* Only called by i810_dma_schedule. */
-static int i810_do_dma(drm_device_t *dev, int locked)
+static int i810_wait_ring(drm_device_t *dev, int n)
 {
-	unsigned long	 address;
-	unsigned long	 length;
-	drm_buf_t	 *buf;
-	int		 retcode = 0;
-	drm_device_dma_t *dma = dev->dma;
-#if DRM_DMA_HISTOGRAM
-	cycles_t	 dma_start, dma_stop;
-#endif
-
-	if (test_and_set_bit(0, &dev->dma_flag)) {
-		atomic_inc(&dma->total_missed_dma);
-		return -EBUSY;
-	}
-	
-#if DRM_DMA_HISTOGRAM
-	dma_start = get_cycles();
-#endif
-
-	if (!dma->next_buffer) {
-		DRM_ERROR("No next_buffer\n");
-		clear_bit(0, &dev->dma_flag);
-		return -EINVAL;
-	}
-
-	buf	= dma->next_buffer;
-	address = (unsigned long)buf->bus_address;
-	length	= buf->used;
+   	drm_i810_private_t *dev_priv = dev->dev_private;
+   	drm_i810_ring_buffer_t *ring = &(dev_priv->ring);
+   	int iters = 0;
+   	unsigned long end;
+
+	end = jiffies + (HZ*3);
+   	while (ring->space < n) {
+	   	int i;
 	
-
-	DRM_DEBUG("context %d, buffer %d (%ld bytes)\n",
-		  buf->context, buf->idx, length);
-
-	if (buf->list == DRM_LIST_RECLAIM) {
-		drm_clear_next_buffer(dev);
-		drm_free_buffer(dev, buf);
-		clear_bit(0, &dev->dma_flag);
-		return -EINVAL;
-	}
-
-	if (!length) {
-		DRM_ERROR("0 length buffer\n");
-		drm_clear_next_buffer(dev);
-		drm_free_buffer(dev, buf);
-		clear_bit(0, &dev->dma_flag);
-		return 0;
-	}
-	
-	if (!i810_dma_is_ready(dev)) {
-		clear_bit(0, &dev->dma_flag);
-		return -EBUSY;
-	}
-
-	if (buf->while_locked) {
-		if (!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
-			DRM_ERROR("Dispatching buffer %d from pid %d"
-				  " \"while locked\", but no lock held\n",
-				  buf->idx, buf->pid);
-		}
-	} else {
-		if (!locked && !drm_lock_take(&dev->lock.hw_lock->lock,
-					      DRM_KERNEL_CONTEXT)) {
-			atomic_inc(&dma->total_missed_lock);
-			clear_bit(0, &dev->dma_flag);
-			return -EBUSY;
+	   	ring->head = I810_READ(LP_RING + RING_HEAD) & HEAD_ADDR;
+	   	ring->space = ring->head - (ring->tail+8);
+	   
+	   	if (ring->space < 0) ring->space += ring->Size;
+	   
+	   	iters++;
+		if((signed)(end - jiffies) <= 0) {
+		   	DRM_ERROR("space: %d wanted %d\n", ring->space, n);
+		   	DRM_ERROR("lockup\n");
+		   	goto out_wait_ring;
 		}
-	}
-
-	if (dev->last_context != buf->context
-	    && !(dev->queuelist[buf->context]->flags
-		 & _DRM_CONTEXT_PRESERVED)) {
-				/* PRE: dev->last_context != buf->context */
-		if (drm_context_switch(dev, dev->last_context, buf->context)) {
-			drm_clear_next_buffer(dev);
-			drm_free_buffer(dev, buf);
-		}
-		retcode = -EBUSY;
-		goto cleanup;
-			
-				/* POST: we will wait for the context
-				   switch and will dispatch on a later call
-				   when dev->last_context == buf->context.
-				   NOTE WE HOLD THE LOCK THROUGHOUT THIS
-				   TIME! */
-	}
-
-	drm_clear_next_buffer(dev);
-	buf->pending	 = 1;
-	buf->waiting	 = 0;
-	buf->list	 = DRM_LIST_PEND;
-#if DRM_DMA_HISTOGRAM
-	buf->time_dispatched = get_cycles();
-#endif
 
-	i810_dma_dispatch(dev, address, length);
-	drm_free_buffer(dev, dma->this_buffer);
-	dma->this_buffer = buf;
-
-	atomic_add(length, &dma->total_bytes);
-	atomic_inc(&dma->total_dmas);
-
-	if (!buf->while_locked && !dev->context_flag && !locked) {
-		if (drm_lock_free(dev, &dev->lock.hw_lock->lock,
-				  DRM_KERNEL_CONTEXT)) {
-			DRM_ERROR("\n");
-		}
+	   	for (i = 0 ; i < 2000 ; i++) ;
 	}
-cleanup:
-
-	clear_bit(0, &dev->dma_flag);
 
-#if DRM_DMA_HISTOGRAM
-	dma_stop = get_cycles();
-	atomic_inc(&dev->histo.dma[drm_histogram_slot(dma_stop - dma_start)]);
-#endif
-
-	return retcode;
-}
-
-static void i810_dma_schedule_timer_wrapper(unsigned long dev)
-{
-	i810_dma_schedule((drm_device_t *)dev, 0);
+out_wait_ring:   
+   	return iters;
 }
 
-static void i810_dma_schedule_tq_wrapper(void *dev)
+static void i810_kernel_lost_context(drm_device_t *dev)
 {
-	i810_dma_schedule(dev, 0);
+      	drm_i810_private_t *dev_priv = dev->dev_private;
+   	drm_i810_ring_buffer_t *ring = &(dev_priv->ring);
+      
+   	ring->head = I810_READ(LP_RING + RING_HEAD) & HEAD_ADDR;
+     	ring->tail = I810_READ(LP_RING + RING_TAIL);
+     	ring->space = ring->head - (ring->tail+8);
+     	if (ring->space < 0) ring->space += ring->Size;
 }
 
-int i810_dma_schedule(drm_device_t *dev, int locked)
+static int i810_freelist_init(drm_device_t *dev)
 {
-	int		 next;
-	drm_queue_t	 *q;
-	drm_buf_t	 *buf;
-	int		 retcode   = 0;
-	int		 processed = 0;
-	int		 missed;
-	int		 expire	   = 20;
-	drm_device_dma_t *dma	   = dev->dma;
-#if DRM_DMA_HISTOGRAM
-	cycles_t	 schedule_start;
-#endif
-
-	if (test_and_set_bit(0, &dev->interrupt_flag)) {
-				/* Not reentrant */
-		atomic_inc(&dma->total_missed_sched);
-		return -EBUSY;
-	}
-	missed = atomic_read(&dma->total_missed_sched);
-
-#if DRM_DMA_HISTOGRAM
-	schedule_start = get_cycles();
-#endif
-
-again:
-	if (dev->context_flag) {
-		clear_bit(0, &dev->interrupt_flag);
-		return -EBUSY;
-	}
-	if (dma->next_buffer) {
-				/* Unsent buffer that was previously
-				   selected, but that couldn't be sent
-				   because the lock could not be obtained
-				   or the DMA engine wasn't ready.  Try
-				   again. */
-		atomic_inc(&dma->total_tried);
-		if (!(retcode = i810_do_dma(dev, locked))) {
-			atomic_inc(&dma->total_hit);
-			++processed;
-		}
-	} else {
-		do {
-			next = drm_select_queue(dev,
-					     i810_dma_schedule_timer_wrapper);
-			if (next >= 0) {
-				q   = dev->queuelist[next];
-				buf = drm_waitlist_get(&q->waitlist);
-				dma->next_buffer = buf;
-				dma->next_queue	 = q;
-				if (buf && buf->list == DRM_LIST_RECLAIM) {
-					drm_clear_next_buffer(dev);
-					drm_free_buffer(dev, buf);
-				}
-			}
-		} while (next >= 0 && !dma->next_buffer);
-		if (dma->next_buffer) {
-			if (!(retcode = i810_do_dma(dev, locked))) {
-				++processed;
-			}
-		}
+      	drm_device_dma_t *dma = dev->dma;
+   	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
+   	u8 *hw_status = (u8 *)dev_priv->hw_status_page;
+   	int i;
+   	int my_idx = 24;
+   
+   	if(dma->buf_count > 1019) {
+	   	/* Not enough space in the status page for the freelist */
+	   	return -EINVAL;
 	}
 
-	if (--expire) {
-		if (missed != atomic_read(&dma->total_missed_sched)) {
-			atomic_inc(&dma->total_lost);
-			if (i810_dma_is_ready(dev)) goto again;
-		}
-		if (processed && i810_dma_is_ready(dev)) {
-			atomic_inc(&dma->total_lost);
-			processed = 0;
-			goto again;
-		}
+   	for (i = 0; i < dma->buf_count; i++) {
+	   	drm_buf_t *buf = dma->buflist[ i ];
+	   	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+	   
+	   	buf_priv->in_use = hw_status + my_idx;
+	   	DRM_DEBUG("buf_priv->in_use : %p\n", buf_priv->in_use);
+	   	*buf_priv->in_use = I810_BUF_FREE;
+	   	buf_priv->my_use_idx = my_idx;
+	   	my_idx += 4;
 	}
-	
-	clear_bit(0, &dev->interrupt_flag);
-	
-#if DRM_DMA_HISTOGRAM
-	atomic_inc(&dev->histo.schedule[drm_histogram_slot(get_cycles()
-							   - schedule_start)]);
-#endif
-	return retcode;
+	return 0;
 }
 
-static int i810_dma_priority(drm_device_t *dev, drm_dma_t *d)
+static int i810_dma_initialize(drm_device_t *dev, 
+			       drm_i810_private_t *dev_priv,
+			       drm_i810_init_t *init)
 {
-	unsigned long	  address;
-	unsigned long	  length;
-	int		  must_free = 0;
-	int		  retcode   = 0;
-	int		  i;
-	int		  idx;
-	drm_buf_t	  *buf;
-	drm_buf_t	  *last_buf = NULL;
-	drm_device_dma_t  *dma	    = dev->dma;
-	DECLARE_WAITQUEUE(entry, current);
-
-				/* Turn off interrupt handling */
-	while (test_and_set_bit(0, &dev->interrupt_flag)) {
-		schedule();
-		if (signal_pending(current)) return -EINTR;
-	}
-	if (!(d->flags & _DRM_DMA_WHILE_LOCKED)) {
-		while (!drm_lock_take(&dev->lock.hw_lock->lock,
-				      DRM_KERNEL_CONTEXT)) {
-			schedule();
-			if (signal_pending(current)) {
-				clear_bit(0, &dev->interrupt_flag);
-				return -EINTR;
-			}
-		}
-		++must_free;
-	}
-	atomic_inc(&dma->total_prio);
-
-	for (i = 0; i < d->send_count; i++) {
-		idx = d->send_indices[i];
-		if (idx < 0 || idx >= dma->buf_count) {
-			DRM_ERROR("Index %d (of %d max)\n",
-				  d->send_indices[i], dma->buf_count - 1);
-			continue;
-		}
-		buf = dma->buflist[ idx ];
-		if (buf->pid != current->pid) {
-			DRM_ERROR("Process %d using buffer owned by %d\n",
-				  current->pid, buf->pid);
-			retcode = -EINVAL;
-			goto cleanup;
-		}
-		if (buf->list != DRM_LIST_NONE) {
-			DRM_ERROR("Process %d using %d's buffer on list %d\n",
-				  current->pid, buf->pid, buf->list);
-			retcode = -EINVAL;
-			goto cleanup;
-		}
-				/* This isn't a race condition on
-				   buf->list, since our concern is the
-				   buffer reclaim during the time the
-				   process closes the /dev/drm? handle, so
-				   it can't also be doing DMA. */
-		buf->list	  = DRM_LIST_PRIO;
-		buf->used	  = d->send_sizes[i];
-		buf->context	  = d->context;
-		buf->while_locked = d->flags & _DRM_DMA_WHILE_LOCKED;
-		address		  = (unsigned long)buf->address;
-		length		  = buf->used;
-		if (!length) {
-			DRM_ERROR("0 length buffer\n");
-		}
-		if (buf->pending) {
-			DRM_ERROR("Sending pending buffer:"
-				  " buffer %d, offset %d\n",
-				  d->send_indices[i], i);
-			retcode = -EINVAL;
-			goto cleanup;
-		}
-		if (buf->waiting) {
-			DRM_ERROR("Sending waiting buffer:"
-				  " buffer %d, offset %d\n",
-				  d->send_indices[i], i);
-			retcode = -EINVAL;
-			goto cleanup;
-		}
-		buf->pending = 1;
-		
-		if (dev->last_context != buf->context
-		    && !(dev->queuelist[buf->context]->flags
-			 & _DRM_CONTEXT_PRESERVED)) {
-			add_wait_queue(&dev->context_wait, &entry);
-			current->state = TASK_INTERRUPTIBLE;
-				/* PRE: dev->last_context != buf->context */
-			drm_context_switch(dev, dev->last_context,
-					   buf->context);
-				/* POST: we will wait for the context
-				   switch and will dispatch on a later call
-				   when dev->last_context == buf->context.
-				   NOTE WE HOLD THE LOCK THROUGHOUT THIS
-				   TIME! */
-			schedule();
-			current->state = TASK_RUNNING;
-			remove_wait_queue(&dev->context_wait, &entry);
-			if (signal_pending(current)) {
-				retcode = -EINTR;
-				goto cleanup;
-			}
-			if (dev->last_context != buf->context) {
-				DRM_ERROR("Context mismatch: %d %d\n",
-					  dev->last_context,
-					  buf->context);
-			}
-		}
-
-#if DRM_DMA_HISTOGRAM
-		buf->time_queued     = get_cycles();
-		buf->time_dispatched = buf->time_queued;
-#endif
-		i810_dma_dispatch(dev, address, length);
-	        if (drm_lock_free(dev, &dev->lock.hw_lock->lock,
-				  DRM_KERNEL_CONTEXT)) {
-		   DRM_ERROR("\n");
-		}
-
-		atomic_add(length, &dma->total_bytes);
-		atomic_inc(&dma->total_dmas);
-		
-		if (last_buf) {
-			drm_free_buffer(dev, last_buf);
-		}
-		last_buf = buf;
-	}
+	drm_map_t *sarea_map;
 
+   	dev->dev_private = (void *) dev_priv;
+   	memset(dev_priv, 0, sizeof(drm_i810_private_t));
 
-cleanup:
-	if (last_buf) {
-		i810_dma_ready(dev);
-		drm_free_buffer(dev, last_buf);
+   	if (init->ring_map_idx >= dev->map_count ||
+	    init->buffer_map_idx >= dev->map_count) {
+	   	i810_dma_cleanup(dev);
+	   	DRM_ERROR("ring_map or buffer_map are invalid\n");
+	   	return -EINVAL;
 	}
-	
-	if (must_free && !dev->context_flag) {
-		if (drm_lock_free(dev, &dev->lock.hw_lock->lock,
-				  DRM_KERNEL_CONTEXT)) {
-			DRM_ERROR("\n");
-		}
+   
+   	dev_priv->ring_map_idx = init->ring_map_idx;
+   	dev_priv->buffer_map_idx = init->buffer_map_idx;
+	sarea_map = dev->maplist[0];
+	dev_priv->sarea_priv = (drm_i810_sarea_t *) 
+		((u8 *)sarea_map->handle + 
+		 init->sarea_priv_offset);
+
+   	atomic_set(&dev_priv->flush_done, 0);
+	init_waitqueue_head(&dev_priv->flush_queue);
+   	
+   	dev_priv->ring.Start = init->ring_start;
+   	dev_priv->ring.End = init->ring_end;
+   	dev_priv->ring.Size = init->ring_size;
+   	dev_priv->ring.virtual_start = drm_ioremap(dev->agp->base + 
+						   init->ring_start, 
+						   init->ring_size);
+   	dev_priv->ring.tail_mask = dev_priv->ring.Size - 1;
+   
+   	if (dev_priv->ring.virtual_start == NULL) {
+	   	i810_dma_cleanup(dev);
+	   	DRM_ERROR("can not ioremap virtual address for"
+			  " ring buffer\n");
+	   	return -ENOMEM;
 	}
-	clear_bit(0, &dev->interrupt_flag);
-	return retcode;
+   
+   	/* Program Hardware Status Page */
+   	dev_priv->hw_status_page = i810_alloc_page(dev);
+   	memset((void *) dev_priv->hw_status_page, 0, PAGE_SIZE);
+   	if(dev_priv->hw_status_page == 0UL) {
+		i810_dma_cleanup(dev);
+		DRM_ERROR("Can not allocate hardware status page\n");
+		return -ENOMEM;
+	}
+   	DRM_DEBUG("hw status page @ %lx\n", dev_priv->hw_status_page);
+   
+   	I810_WRITE(0x02080, virt_to_bus((void *)dev_priv->hw_status_page));
+   	DRM_DEBUG("Enabled hardware status page\n");
+   
+   	/* Now we need to init our freelist */
+   	if(i810_freelist_init(dev) != 0) {
+	   	i810_dma_cleanup(dev);
+	   	DRM_ERROR("Not enough space in the status page for"
+			  " the freelist\n");
+	   	return -ENOMEM;
+	}
+   	return 0;
 }
 
-static int i810_dma_send_buffers(drm_device_t *dev, drm_dma_t *d)
+int i810_dma_init(struct inode *inode, struct file *filp,
+		  unsigned int cmd, unsigned long arg)
 {
-	DECLARE_WAITQUEUE(entry, current);
-	drm_buf_t	  *last_buf = NULL;
-	int		  retcode   = 0;
-	drm_device_dma_t  *dma	    = dev->dma;
-
-	if (d->flags & _DRM_DMA_BLOCK) {
-		last_buf = dma->buflist[d->send_indices[d->send_count-1]];
-		add_wait_queue(&last_buf->dma_wait, &entry);
-	}
+   	drm_file_t *priv = filp->private_data;
+   	drm_device_t *dev = priv->dev;
+   	drm_i810_private_t *dev_priv;
+   	drm_i810_init_t init;
+   	int retcode = 0;
 	
-	if ((retcode = drm_dma_enqueue(dev, d))) {
-		if (d->flags & _DRM_DMA_BLOCK)
-			remove_wait_queue(&last_buf->dma_wait, &entry);
-		return retcode;
-	}
-	
-	i810_dma_schedule(dev, 0);
+   	copy_from_user_ret(&init, (drm_i810_init_t *)arg, 
+			   sizeof(init), -EFAULT);
 	
-	if (d->flags & _DRM_DMA_BLOCK) {
-		DRM_DEBUG("%d waiting\n", current->pid);
-		current->state = TASK_INTERRUPTIBLE;
-		for (;;) {
-			if (!last_buf->waiting
-			    && !last_buf->pending)
-				break; /* finished */
-			schedule();
-			if (signal_pending(current)) {
-				retcode = -EINTR; /* Can't restart */
-				break;
-			}
-		}
-		current->state = TASK_RUNNING;
-		DRM_DEBUG("%d running\n", current->pid);
-		remove_wait_queue(&last_buf->dma_wait, &entry);
-		if (!retcode
-		    || (last_buf->list==DRM_LIST_PEND && !last_buf->pending)) {
-			if (!waitqueue_active(&last_buf->dma_wait)) {
-				drm_free_buffer(dev, last_buf);
-			}
-		}
-		if (retcode) {
-			DRM_ERROR("ctx%d w%d p%d c%d i%d l%d %d/%d\n",
-				  d->context,
-				  last_buf->waiting,
-				  last_buf->pending,
-				  DRM_WAITCOUNT(dev, d->context),
-				  last_buf->idx,
-				  last_buf->list,
-				  last_buf->pid,
-				  current->pid);
-		}
+   	switch(init.func) {
+	 	case I810_INIT_DMA:
+	   		dev_priv = drm_alloc(sizeof(drm_i810_private_t), 
+					     DRM_MEM_DRIVER);
+	   		if(dev_priv == NULL) return -ENOMEM;
+	   		retcode = i810_dma_initialize(dev, dev_priv, &init);
+	   	break;
+	 	case I810_CLEANUP_DMA:
+	   		retcode = i810_dma_cleanup(dev);
+	   	break;
+	 	default:
+	   		retcode = -EINVAL;
+	   	break;
 	}
-	return retcode;
+   
+   	return retcode;
 }
 
-int i810_dma(struct inode *inode, struct file *filp, unsigned int cmd,
-	      unsigned long arg)
+static void i810_dma_dispatch_general(drm_device_t *dev, drm_buf_t *buf,
+				      int used )
 {
-	drm_file_t	  *priv	    = filp->private_data;
-	drm_device_t	  *dev	    = priv->dev;
-	drm_device_dma_t  *dma	    = dev->dma;
-	int		  retcode   = 0;
-	drm_dma_t	  d;
-
-        printk("i810_dma start\n");
-   	copy_from_user_ret(&d, (drm_dma_t *)arg, sizeof(d), -EFAULT);
-	DRM_DEBUG("%d %d: %d send, %d req\n",
-		  current->pid, d.context, d.send_count, d.request_count);
+      	drm_i810_private_t *dev_priv = dev->dev_private;
+   	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+	unsigned long address = (unsigned long)buf->bus_address;
+	unsigned long start = address - dev->agp->base;
+   	RING_LOCALS;
+
+   	dev_priv->counter++;
+   	DRM_DEBUG(  "dispatch counter : %ld\n", dev_priv->counter);
+   	DRM_DEBUG(  "i810_dma_dispatch\n");
+   	DRM_DEBUG(  "start : 0x%lx\n", start);
+	DRM_DEBUG(  "used : 0x%x\n", used);
+   	DRM_DEBUG(  "start + used - 4 : 0x%lx\n", start + used - 4);
+   	i810_kernel_lost_context(dev);
+
+   	BEGIN_LP_RING(10);
+   	OUT_RING( CMD_OP_BATCH_BUFFER );
+   	OUT_RING( start | BB1_PROTECTED );
+   	OUT_RING( start + used - 4 );
+      	OUT_RING( CMD_STORE_DWORD_IDX );
+   	OUT_RING( 20 );
+   	OUT_RING( dev_priv->counter );
+   	OUT_RING( CMD_STORE_DWORD_IDX );
+   	OUT_RING( buf_priv->my_use_idx );
+   	OUT_RING( I810_BUF_FREE );   
+      	OUT_RING( CMD_REPORT_HEAD );
+      	ADVANCE_LP_RING();
+}
 
-	if (d.context == DRM_KERNEL_CONTEXT || d.context >= dev->queue_slots) {
-		DRM_ERROR("Process %d using context %d\n",
-			  current->pid, d.context);
-		return -EINVAL;
-	}
+static void i810_dma_dispatch_vertex(drm_device_t *dev, 
+				     drm_buf_t *buf,
+				     int discard,
+				     int used)
+{
+   	drm_i810_private_t *dev_priv = dev->dev_private;
+	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+   	drm_i810_sarea_t *sarea_priv = dev_priv->sarea_priv;
+   	drm_clip_rect_t *box = sarea_priv->boxes;
+   	int nbox = sarea_priv->nbox;
+	unsigned long address = (unsigned long)buf->bus_address;
+	unsigned long start = address - dev->agp->base;     
+	int i = 0;
+   	RING_LOCALS;
 
-	if (d.send_count < 0 || d.send_count > dma->buf_count) {
-		DRM_ERROR("Process %d trying to send %d buffers (of %d max)\n",
-			  current->pid, d.send_count, dma->buf_count);
-		return -EINVAL;
-	}
-	if (d.request_count < 0 || d.request_count > dma->buf_count) {
-		DRM_ERROR("Process %d trying to get %d buffers (of %d max)\n",
-			  current->pid, d.request_count, dma->buf_count);
-		return -EINVAL;
+   
+   	if (nbox > I810_NR_SAREA_CLIPRECTS) 
+		nbox = I810_NR_SAREA_CLIPRECTS;
+   
+   	DRM_DEBUG("dispatch vertex addr 0x%lx, used 0x%x nbox %d\n", 
+		  address, used, nbox);
+
+   	dev_priv->counter++;
+   	DRM_DEBUG(  "dispatch counter : %ld\n", dev_priv->counter);
+   	DRM_DEBUG(  "i810_dma_dispatch\n");
+   	DRM_DEBUG(  "start : %lx\n", start);
+	DRM_DEBUG(  "used : %d\n", used);
+   	DRM_DEBUG(  "start + used - 4 : %ld\n", start + used - 4);
+   	i810_kernel_lost_context(dev);
+
+	if (used) {
+		do {
+			if (i < nbox) {
+				BEGIN_LP_RING(4);
+				OUT_RING( GFX_OP_SCISSOR | SC_UPDATE_SCISSOR | 
+					  SC_ENABLE );
+				OUT_RING( GFX_OP_SCISSOR_INFO );
+				OUT_RING( box[i].x1 | (box[i].y1 << 16) );
+				OUT_RING( (box[i].x2-1) | ((box[i].y2-1)<<16) );
+				ADVANCE_LP_RING();
+			}
+			
+			BEGIN_LP_RING(4);
+			OUT_RING( CMD_OP_BATCH_BUFFER );
+			OUT_RING( start | BB1_PROTECTED );
+			OUT_RING( start + used - 4 );
+			OUT_RING( 0 );
+			ADVANCE_LP_RING();
+			
+		} while (++i < nbox);
 	}
 
-	if (d.send_count) {
-#if 0
-		if (d.flags & _DRM_DMA_PRIORITY)
-			retcode = i810_dma_priority(dev, &d);
-		else 
-			retcode = i810_dma_send_buffers(dev, &d);
-#endif
-	   printk("i810_dma priority\n");
+	BEGIN_LP_RING(10);
+	OUT_RING( CMD_STORE_DWORD_IDX );
+	OUT_RING( 20 );
+	OUT_RING( dev_priv->counter );
+	OUT_RING( 0 );
 
-	   retcode = i810_dma_priority(dev, &d);
+	if (discard) {
+		OUT_RING( CMD_STORE_DWORD_IDX );
+		OUT_RING( buf_priv->my_use_idx );
+		OUT_RING( I810_BUF_FREE );
+		OUT_RING( 0 );
 	}
 
-	d.granted_count = 0;
+      	OUT_RING( CMD_REPORT_HEAD );
+	OUT_RING( 0 );
+   	ADVANCE_LP_RING();
+}
 
-	if (!retcode && d.request_count) {
-		retcode = drm_dma_get_buffers(dev, &d);
-	}
 
-	DRM_DEBUG("%d returning, granted = %d\n",
-		  current->pid, d.granted_count);
-	copy_to_user_ret((drm_dma_t *)arg, &d, sizeof(d), -EFAULT);
+/* Interrupts are only for flushing */
+static void i810_dma_service(int irq, void *device, struct pt_regs *regs)
+{
+	drm_device_t	 *dev = (drm_device_t *)device;
+   	u16 temp;
+   
+	atomic_inc(&dev->total_irq);
+      	temp = I810_READ16(I810REG_INT_IDENTITY_R);
+   	temp = temp & ~(0x6000);
+   	if(temp != 0) I810_WRITE16(I810REG_INT_IDENTITY_R, 
+				   temp); /* Clear all interrupts */
+   
+   	queue_task(&dev->tq, &tq_immediate);
+   	mark_bh(IMMEDIATE_BH);
+}
 
-   	printk("i810_dma end (granted)\n");
-	return retcode;
+static void i810_dma_task_queue(void *device)
+{
+	drm_device_t *dev = (drm_device_t *) device;
+      	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
+
+   	atomic_set(&dev_priv->flush_done, 1);
+   	wake_up_interruptible(&dev_priv->flush_queue);
 }
 
 int i810_irq_install(drm_device_t *dev, int irq)
 {
 	int retcode;
-
+	u16 temp;
+   
 	if (!irq)     return -EINVAL;
 	
 	down(&dev->struct_sem);
@@ -591,6 +508,7 @@ int i810_irq_install(drm_device_t *dev, int irq)
 	dev->irq = irq;
 	up(&dev->struct_sem);
 	
+   	DRM_DEBUG(  "Interrupt Install : %d\n", irq);
 	DRM_DEBUG("%d\n", irq);
 
 	dev->context_flag     = 0;
@@ -603,12 +521,21 @@ int i810_irq_install(drm_device_t *dev, int irq)
 
 	dev->tq.next	      = NULL;
 	dev->tq.sync	      = 0;
-	dev->tq.routine	      = i810_dma_schedule_tq_wrapper;
+	dev->tq.routine	      = i810_dma_task_queue;
 	dev->tq.data	      = dev;
 
-
 				/* Before installing handler */
-	/* TODO */	
+   	temp = I810_READ16(I810REG_HWSTAM);
+   	temp = temp & 0x6000;
+   	I810_WRITE16(I810REG_HWSTAM, temp);
+   	
+      	temp = I810_READ16(I810REG_INT_MASK_R);
+   	temp = temp & 0x6000;
+   	I810_WRITE16(I810REG_INT_MASK_R, temp); /* Unmask interrupts */
+   	temp = I810_READ16(I810REG_INT_ENABLE_R);
+   	temp = temp & 0x6000;
+      	I810_WRITE16(I810REG_INT_ENABLE_R, temp); /* Disable all interrupts */
+
 				/* Install handler */
 	if ((retcode = request_irq(dev->irq,
 				   i810_dma_service,
@@ -620,15 +547,18 @@ int i810_irq_install(drm_device_t *dev, int irq)
 		up(&dev->struct_sem);
 		return retcode;
 	}
-
-				/* After installing handler */
-	/* TODO */
+   	temp = I810_READ16(I810REG_INT_ENABLE_R);
+   	temp = temp & 0x6000;
+   	temp = temp | 0x0003;
+   	I810_WRITE16(I810REG_INT_ENABLE_R, 
+		     temp); /* Enable bp & user interrupts */
 	return 0;
 }
 
 int i810_irq_uninstall(drm_device_t *dev)
 {
 	int irq;
+   	u16 temp;
 
 	down(&dev->struct_sem);
 	irq	 = dev->irq;
@@ -636,16 +566,25 @@ int i810_irq_uninstall(drm_device_t *dev)
 	up(&dev->struct_sem);
 	
 	if (!irq) return -EINVAL;
-	
+
+   	DRM_DEBUG(  "Interrupt UnInstall: %d\n", irq);	
 	DRM_DEBUG("%d\n", irq);
-	
-	/* TODO : Disable interrupts */
+   
+   	temp = I810_READ16(I810REG_INT_IDENTITY_R);
+   	temp = temp & ~(0x6000);
+   	if(temp != 0) I810_WRITE16(I810REG_INT_IDENTITY_R, 
+				   temp); /* Clear all interrupts */
+   
+   	temp = I810_READ16(I810REG_INT_ENABLE_R);
+   	temp = temp & 0x6000;
+   	I810_WRITE16(I810REG_INT_ENABLE_R, 
+		     temp);                     /* Disable all interrupts */
+
    	free_irq(irq, dev);
 
 	return 0;
 }
 
-
 int i810_control(struct inode *inode, struct file *filp, unsigned int cmd,
 		  unsigned long arg)
 {
@@ -654,8 +593,7 @@ int i810_control(struct inode *inode, struct file *filp, unsigned int cmd,
 	drm_control_t	ctl;
 	int		retcode;
    
-   	printk(KERN_INFO "i810_control\n");
-   	i810_dma_init(dev);
+   	DRM_DEBUG(  "i810_control\n");
 
 	copy_from_user_ret(&ctl, (drm_control_t *)arg, sizeof(ctl), -EFAULT);
 	
@@ -674,20 +612,134 @@ int i810_control(struct inode *inode, struct file *filp, unsigned int cmd,
 	return 0;
 }
 
+static inline void i810_dma_emit_flush(drm_device_t *dev)
+{
+   	drm_i810_private_t *dev_priv = dev->dev_private;
+   	RING_LOCALS;
+
+   	i810_kernel_lost_context(dev);
+   	BEGIN_LP_RING(2);
+      	OUT_RING( CMD_REPORT_HEAD );
+   	OUT_RING( GFX_OP_USER_INTERRUPT );
+      	ADVANCE_LP_RING();
+}
+
+static inline void i810_dma_quiescent_emit(drm_device_t *dev)
+{
+      	drm_i810_private_t *dev_priv = dev->dev_private;
+   	RING_LOCALS;
+
+  	i810_kernel_lost_context(dev);
+   	BEGIN_LP_RING(4);
+
+   	OUT_RING( INST_PARSER_CLIENT | INST_OP_FLUSH | INST_FLUSH_MAP_CACHE );
+   	OUT_RING( CMD_REPORT_HEAD );
+   	OUT_RING( GFX_OP_USER_INTERRUPT );
+   	OUT_RING( 0 );
+   	ADVANCE_LP_RING();
+}
+
+static void i810_dma_quiescent(drm_device_t *dev)
+{
+      	DECLARE_WAITQUEUE(entry, current);
+  	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
+	unsigned long end;      
+
+   	if(dev_priv == NULL) {
+	   	return;
+	}
+      	atomic_set(&dev_priv->flush_done, 0);
+   	current->state = TASK_INTERRUPTIBLE;
+   	add_wait_queue(&dev_priv->flush_queue, &entry);
+   	end = jiffies + (HZ*3);
+   
+   	for (;;) {
+	      	i810_dma_quiescent_emit(dev);
+	   	if (atomic_read(&dev_priv->flush_done) == 1) break;
+		if((signed)(end - jiffies) <= 0) {
+		   	DRM_ERROR("lockup\n");
+		   	break;
+		}	   
+	      	schedule_timeout(HZ*3);
+	      	if (signal_pending(current)) {
+		   	break;
+		}
+	}
+   
+   	current->state = TASK_RUNNING;
+   	remove_wait_queue(&dev_priv->flush_queue, &entry);
+   
+   	return;
+}
+
+static int i810_flush_queue(drm_device_t *dev)
+{
+   	DECLARE_WAITQUEUE(entry, current);
+  	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
+	unsigned long end;
+   	int ret = 0;      
+
+   	if(dev_priv == NULL) {
+	   	return 0;
+	}
+      	atomic_set(&dev_priv->flush_done, 0);
+   	current->state = TASK_INTERRUPTIBLE;
+   	add_wait_queue(&dev_priv->flush_queue, &entry);
+   	end = jiffies + (HZ*3);
+   	for (;;) {
+	      	i810_dma_emit_flush(dev);
+	   	if (atomic_read(&dev_priv->flush_done) == 1) break;
+		if((signed)(end - jiffies) <= 0) {
+		   	DRM_ERROR("lockup\n");
+		   	break;
+		}	   
+	      	schedule_timeout(HZ*3);
+	      	if (signal_pending(current)) {
+		   	ret = -EINTR; /* Can't restart */
+		   	break;
+		}
+	}
+   
+   	current->state = TASK_RUNNING;
+   	remove_wait_queue(&dev_priv->flush_queue, &entry);
+   
+   	return ret;
+}
+
+/* Must be called with the lock held */
+void i810_reclaim_buffers(drm_device_t *dev, pid_t pid)
+{
+	drm_device_dma_t *dma = dev->dma;
+	int		 i;
+
+	if (!dma) return;
+      	if(dev->dev_private == NULL) return;
+
+        i810_flush_queue(dev);
+
+	for (i = 0; i < dma->buf_count; i++) {
+	   	drm_buf_t *buf = dma->buflist[ i ];
+	   	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+
+		if (buf->pid == pid) {
+		   	/* Only buffers that need to get reclaimed ever 
+			 * get set to free */
+		   	if(buf_priv == NULL) return;
+		   	cmpxchg(buf_priv->in_use, 
+				I810_BUF_USED, I810_BUF_FREE);
+		}
+	}
+}
+
 int i810_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 	       unsigned long arg)
 {
 	drm_file_t	  *priv	  = filp->private_data;
 	drm_device_t	  *dev	  = priv->dev;
+
 	DECLARE_WAITQUEUE(entry, current);
 	int		  ret	= 0;
 	drm_lock_t	  lock;
-	drm_queue_t	  *q;
-#if DRM_DMA_HISTOGRAM
-	cycles_t	  start;
-
-	dev->lck_start = start = get_cycles();
-#endif
 
 	copy_from_user_ret(&lock, (drm_lock_t *)arg, sizeof(lock), -EFAULT);
 
@@ -696,16 +748,20 @@ int i810_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 			  current->pid, lock.context);
 		return -EINVAL;
 	}
+   
+   	DRM_DEBUG("%d (pid %d) requests lock (0x%08x), flags = 0x%08x\n",
+		  lock.context, current->pid, dev->lock.hw_lock->lock,
+		  lock.flags);
 
-	if (lock.context < 0 || lock.context >= dev->queue_count) {
+	if (lock.context < 0) {
 		return -EINVAL;
 	}
-	q = dev->queuelist[lock.context];
-	
-	ret = drm_flush_block_and_flush(dev, lock.context, lock.flags);
+	/* Only one queue:
+	 */
 
 	if (!ret) {
-		if (_DRM_LOCKING_CONTEXT(dev->lock.hw_lock->lock)
+#if 0
+	   	if (_DRM_LOCKING_CONTEXT(dev->lock.hw_lock->lock)
 		    != lock.context) {
 			long j = jiffies - dev->lock.lock_time;
 
@@ -716,6 +772,7 @@ int i810_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 				schedule_timeout(j);
 			}
 		}
+#endif
 		add_wait_queue(&dev->lock.lock_queue, &entry);
 		for (;;) {
 			if (!dev->lock.hw_lock) {
@@ -728,13 +785,13 @@ int i810_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 				dev->lock.pid	    = current->pid;
 				dev->lock.lock_time = jiffies;
 				atomic_inc(&dev->total_locks);
-				atomic_inc(&q->total_locks);
 				break;	/* Got lock */
 			}
 			
 				/* Contention */
 			atomic_inc(&dev->total_sleeps);
 			current->state = TASK_INTERRUPTIBLE;
+		   	DRM_DEBUG("Calling lock schedule\n");
 			schedule();
 			if (signal_pending(current)) {
 				ret = -ERESTARTSYS;
@@ -744,19 +801,184 @@ int i810_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 		current->state = TASK_RUNNING;
 		remove_wait_queue(&dev->lock.lock_queue, &entry);
 	}
-
-	drm_flush_unblock(dev, lock.context, lock.flags); /* cleanup phase */
 	
 	if (!ret) {
-		if (lock.flags & _DRM_LOCK_READY)
-			i810_dma_ready(dev);
-		if (lock.flags & _DRM_LOCK_QUIESCENT)
-			i810_dma_quiescent(dev);
+		if (lock.flags & _DRM_LOCK_QUIESCENT) {
+		   DRM_DEBUG("_DRM_LOCK_QUIESCENT\n");
+		   DRM_DEBUG("fred\n");
+		   i810_dma_quiescent(dev);
+		}
 	}
+	DRM_DEBUG("%d %s\n", lock.context, ret ? "interrupted" : "has lock");
+	return ret;
+}
 
-#if DRM_DMA_HISTOGRAM
-	atomic_inc(&dev->histo.lacq[drm_histogram_slot(get_cycles() - start)]);
-#endif
+int i810_flush_ioctl(struct inode *inode, struct file *filp, 
+		     unsigned int cmd, unsigned long arg)
+{
+   	drm_file_t	  *priv	  = filp->private_data;
+   	drm_device_t	  *dev	  = priv->dev;
+   
+   	DRM_DEBUG("i810_flush_ioctl\n");
+   	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("i810_flush_ioctl called without lock held\n");
+		return -EINVAL;
+	}
+
+   	i810_flush_queue(dev);
+   	return 0;
+}
+
+static int i810DmaGeneral(drm_device_t *dev, drm_i810_general_t *args)
+{
+	drm_device_dma_t *dma = dev->dma;
+   	drm_buf_t *buf = dma->buflist[ args->idx ];
+   	
+	if (!args->used) {
+	   	i810_freelist_put(dev, buf);
+	} else {  
+		i810_dma_dispatch_general( dev, buf, args->used );
+		atomic_add(args->used, &dma->total_bytes);
+		atomic_inc(&dma->total_dmas);
+	}
+	return 0; 
+}
+
+static int i810DmaVertex(drm_device_t *dev, drm_i810_vertex_t *args)
+{
+	drm_device_dma_t *dma = dev->dma;
+	drm_buf_t *buf = dma->buflist[ args->idx ];
+   	i810_dma_dispatch_vertex( dev, buf, args->discard, args->used );
+   	atomic_add(args->used, &dma->total_bytes);
+	atomic_inc(&dma->total_dmas);
+   	return 0;
+}
+
+int i810_dma_general(struct inode *inode, struct file *filp,
+		     unsigned int cmd, unsigned long arg)
+{
+	drm_file_t *priv = filp->private_data;
+	drm_device_t *dev = priv->dev;
+	drm_i810_general_t general;
+      	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
+      	u32 *hw_status = (u32 *)dev_priv->hw_status_page;
+   	drm_i810_sarea_t *sarea_priv = (drm_i810_sarea_t *) 
+     					dev_priv->sarea_priv; 
+
+	int retcode = 0;
 	
-	return ret;
+	copy_from_user_ret(&general, (drm_i810_general_t *)arg, sizeof(general),
+			   -EFAULT);
+
+	DRM_DEBUG("i810 dma general idx %d used %d\n",
+		  general.idx, general.used);
+   
+   	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("i810_dma_general called without lock held\n");
+		return -EINVAL;
+	}
+
+	retcode = i810DmaGeneral(dev, &general);
+	sarea_priv->last_enqueue = dev_priv->counter-1;
+   	sarea_priv->last_dispatch = (int) hw_status[5];
+   
+	return retcode;
+}
+
+int i810_dma_vertex(struct inode *inode, struct file *filp,
+	       unsigned int cmd, unsigned long arg)
+{
+	drm_file_t *priv = filp->private_data;
+	drm_device_t *dev = priv->dev;
+   	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
+      	u32 *hw_status = (u32 *)dev_priv->hw_status_page;
+   	drm_i810_sarea_t *sarea_priv = (drm_i810_sarea_t *) 
+     					dev_priv->sarea_priv; 
+	drm_i810_vertex_t vertex;
+	int retcode = 0;
+
+	copy_from_user_ret(&vertex, (drm_i810_vertex_t *)arg, sizeof(vertex),
+			   -EFAULT);
+   	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("i810_dma_vertex called without lock held\n");
+		return -EINVAL;
+	}
+
+	DRM_DEBUG("i810 dma vertex, idx %d used %d discard %d\n",
+		  vertex.idx, vertex.used, vertex.discard);
+
+	retcode = i810DmaVertex(dev, &vertex);
+	sarea_priv->last_enqueue = dev_priv->counter-1;
+   	sarea_priv->last_dispatch = (int) hw_status[5];
+   
+	return retcode;
+
+}
+
+int i810_getage(struct inode *inode, struct file *filp, unsigned int cmd,
+		unsigned long arg)
+{
+   	drm_file_t	  *priv	    = filp->private_data;
+	drm_device_t	  *dev	    = priv->dev;
+   	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
+      	u32 *hw_status = (u32 *)dev_priv->hw_status_page;
+   	drm_i810_sarea_t *sarea_priv = (drm_i810_sarea_t *) 
+     					dev_priv->sarea_priv; 
+
+      	sarea_priv->last_dispatch = (int) hw_status[5];
+	return 0;
+}
+
+int i810_dma(struct inode *inode, struct file *filp, unsigned int cmd,
+	    unsigned long arg)
+{
+	drm_file_t	  *priv	    = filp->private_data;
+	drm_device_t	  *dev	    = priv->dev;
+	drm_device_dma_t  *dma	    = dev->dma;
+	int		  retcode   = 0;
+	drm_dma_t	  d;
+   	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
+   	u32 *hw_status = (u32 *)dev_priv->hw_status_page;
+   	drm_i810_sarea_t *sarea_priv = (drm_i810_sarea_t *) 
+     					dev_priv->sarea_priv; 
+
+
+   	copy_from_user_ret(&d, (drm_dma_t *)arg, sizeof(d), -EFAULT);
+	DRM_DEBUG("%d %d: %d send, %d req\n",
+		  current->pid, d.context, d.send_count, d.request_count);
+   
+	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("i810_dma called without lock held\n");
+		return -EINVAL;
+	}
+
+	/* Please don't send us buffers.
+	 */
+	if (d.send_count != 0) {
+		DRM_ERROR("Process %d trying to send %d buffers via drmDMA\n",
+			  current->pid, d.send_count);
+		return -EINVAL;
+	}
+	
+	/* We'll send you buffers.
+	 */
+	if (d.request_count < 0 || d.request_count > dma->buf_count) {
+		DRM_ERROR("Process %d trying to get %d buffers (of %d max)\n",
+			  current->pid, d.request_count, dma->buf_count);
+		return -EINVAL;
+	}
+	
+	d.granted_count = 0;
+
+	if (!retcode && d.request_count) {
+		retcode = i810_dma_get_buffers(dev, &d);
+	}
+
+	DRM_DEBUG("i810_dma: %d returning, granted = %d\n",
+		  current->pid, d.granted_count);
+
+	copy_to_user_ret((drm_dma_t *)arg, &d, sizeof(d), -EFAULT);   
+   	sarea_priv->last_dispatch = (int) hw_status[5];
+
+	return retcode;
 }
diff --git a/linux-core/i810_drm.h b/linux-core/i810_drm.h
new file mode 100644
index 00000000..0754874c
--- /dev/null
+++ b/linux-core/i810_drm.h
@@ -0,0 +1,93 @@
+#ifndef _I810_DRM_H_
+#define _I810_DRM_H_
+
+/* WARNING: These defines must be the same as what the Xserver uses.
+ * if you change them, you must change the defines in the Xserver.
+ */
+
+/* Might one day want to support the client-side ringbuffer code again.
+ */
+#ifndef _I810_DEFINES_
+#define _I810_DEFINES_
+
+#define I810_USE_BATCH			1
+#define I810_DMA_BUF_ORDER		12
+#define I810_DMA_BUF_SZ 		(1<<I810_DMA_BUF_ORDER)
+#define I810_DMA_BUF_NR 		256
+#define I810_NR_SAREA_CLIPRECTS 	2
+
+/* Each region is a minimum of 64k, and there are at most 64 of them.
+ */
+
+#define I810_NR_TEX_REGIONS 64
+#define I810_LOG_MIN_TEX_REGION_SIZE 16
+#endif
+
+typedef struct _drm_i810_init {
+   	enum {
+	   	I810_INIT_DMA = 0x01,
+	   	I810_CLEANUP_DMA = 0x02
+	} func;
+   	int ring_map_idx;
+      	int buffer_map_idx;
+	int sarea_priv_offset;
+   	unsigned long ring_start;
+   	unsigned long ring_end;
+   	unsigned long ring_size;
+} drm_i810_init_t;
+
+/* Warning: If you change the SAREA structure you must change the Xserver
+ * structure as well */
+
+typedef struct _drm_i810_tex_region {
+	unsigned char next, prev; /* indices to form a circular LRU  */
+	unsigned char in_use;	/* owned by a client, or free? */
+	int age;		/* tracked by clients to update local LRU's */
+} drm_i810_tex_region_t;
+
+typedef struct _drm_i810_sarea {
+	unsigned int nbox;
+	drm_clip_rect_t boxes[I810_NR_SAREA_CLIPRECTS];
+
+	/* Maintain an LRU of contiguous regions of texture space.  If
+	 * you think you own a region of texture memory, and it has an
+	 * age different to the one you set, then you are mistaken and
+	 * it has been stolen by another client.  If global texAge
+	 * hasn't changed, there is no need to walk the list.
+	 *
+	 * These regions can be used as a proxy for the fine-grained
+	 * texture information of other clients - by maintaining them
+	 * in the same lru which is used to age their own textures,
+	 * clients have an approximate lru for the whole of global
+	 * texture space, and can make informed decisions as to which
+	 * areas to kick out.  There is no need to choose whether to
+	 * kick out your own texture or someone else's - simply eject
+	 * them all in LRU order.  
+	 */
+   
+	drm_i810_tex_region_t texList[I810_NR_TEX_REGIONS+1]; 
+				/* Last elt is sentinal */
+        int texAge;		/* last time texture was uploaded */
+        int last_enqueue;	/* last time a buffer was enqueued */
+	int last_dispatch;	/* age of the most recently dispatched buffer */
+	int last_quiescent;     /*  */
+	int ctxOwner;		/* last context to upload state */
+} drm_i810_sarea_t;
+
+typedef struct _drm_i810_general {
+   	int idx;
+	int used;
+} drm_i810_general_t;
+
+/* These may be placeholders if we have more cliprects than
+ * I810_NR_SAREA_CLIPRECTS.  In that case, the client sets discard to
+ * false, indicating that the buffer will be dispatched again with a
+ * new set of cliprects.
+ */
+typedef struct _drm_i810_vertex {
+   	int idx;		/* buffer index */
+	int used;		/* nr bytes in use */
+	int discard;		/* client is finished with the buffer? */
+} drm_i810_vertex_t;
+
+#endif /* _I810_DRM_H_ */
diff --git a/linux-core/i810_drv.c b/linux-core/i810_drv.c
index f33153a3..d3b35c49 100644
--- a/linux-core/i810_drv.c
+++ b/linux-core/i810_drv.c
@@ -33,11 +33,13 @@
 #define EXPORT_SYMTAB
 #include "drmP.h"
 #include "i810_drv.h"
+
+
 EXPORT_SYMBOL(i810_init);
 EXPORT_SYMBOL(i810_cleanup);
 
 #define I810_NAME	 "i810"
-#define I810_DESC	 "Matrox g200/g400"
+#define I810_DESC	 "Intel I810"
 #define I810_DATE	 "19991213"
 #define I810_MAJOR	 0
 #define I810_MINOR	 0
@@ -54,6 +56,7 @@ static struct file_operations i810_fops = {
 	mmap:	 drm_mmap,
 	read:	 drm_read,
 	fasync:	 drm_fasync,
+      	poll:	 drm_poll,
 };
 
 static struct miscdevice      i810_misc = {
@@ -80,13 +83,13 @@ static drm_ioctl_desc_t	      i810_ioctls[] = {
 	[DRM_IOCTL_NR(DRM_IOCTL_MAP_BUFS)]    = { i810_mapbufs,	  1, 0 },
 	[DRM_IOCTL_NR(DRM_IOCTL_FREE_BUFS)]   = { i810_freebufs,  1, 0 },
 
-	[DRM_IOCTL_NR(DRM_IOCTL_ADD_CTX)]     = { drm_addctx,	  1, 1 },
-	[DRM_IOCTL_NR(DRM_IOCTL_RM_CTX)]      = { drm_rmctx,	  1, 1 },
-	[DRM_IOCTL_NR(DRM_IOCTL_MOD_CTX)]     = { drm_modctx,	  1, 1 },
-	[DRM_IOCTL_NR(DRM_IOCTL_GET_CTX)]     = { drm_getctx,	  1, 0 },
-	[DRM_IOCTL_NR(DRM_IOCTL_SWITCH_CTX)]  = { drm_switchctx,  1, 1 },
-	[DRM_IOCTL_NR(DRM_IOCTL_NEW_CTX)]     = { drm_newctx,	  1, 1 },
-	[DRM_IOCTL_NR(DRM_IOCTL_RES_CTX)]     = { drm_resctx,	  1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_ADD_CTX)]     = { i810_addctx,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_RM_CTX)]      = { i810_rmctx,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_MOD_CTX)]     = { i810_modctx,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_GET_CTX)]     = { i810_getctx,	  1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_SWITCH_CTX)]  = { i810_switchctx,  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_NEW_CTX)]     = { i810_newctx,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_RES_CTX)]     = { i810_resctx,	  1, 0 },
 	[DRM_IOCTL_NR(DRM_IOCTL_ADD_DRAW)]    = { drm_adddraw,	  1, 1 },
 	[DRM_IOCTL_NR(DRM_IOCTL_RM_DRAW)]     = { drm_rmdraw,	  1, 1 },
 
@@ -104,6 +107,11 @@ static drm_ioctl_desc_t	      i810_ioctls[] = {
 	[DRM_IOCTL_NR(DRM_IOCTL_AGP_FREE)]    = { drm_agp_free,    1, 1 },
 	[DRM_IOCTL_NR(DRM_IOCTL_AGP_BIND)]    = { drm_agp_bind,    1, 1 },
 	[DRM_IOCTL_NR(DRM_IOCTL_AGP_UNBIND)]  = { drm_agp_unbind,  1, 1 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_I810_INIT)]   = { i810_dma_init,   1, 1 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_I810_VERTEX)] = { i810_dma_vertex, 1, 0 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_I810_DMA)]    = { i810_dma_general,1, 0 },
+      	[DRM_IOCTL_NR(DRM_IOCTL_I810_FLUSH)]  = { i810_flush_ioctl,1, 0 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_I810_GETAGE)] = { i810_getage,     1, 0 },
 };
 
 #define I810_IOCTL_COUNT DRM_ARRAY_SIZE(i810_ioctls)
@@ -121,7 +129,7 @@ MODULE_PARM(i810, "s");
 
 int init_module(void)
 {
-	printk("doing i810_init()\n");
+	DRM_DEBUG("doing i810_init()\n");
 	return i810_init();
 }
 
@@ -364,7 +372,7 @@ int i810_init(void)
 #ifdef MODULE
 	drm_parse_options(i810);
 #endif
-	printk("doing misc_register\n");
+	DRM_DEBUG("doing misc_register\n");
 	if ((retcode = misc_register(&i810_misc))) {
 		DRM_ERROR("Cannot register \"%s\"\n", I810_NAME);
 		return retcode;
@@ -372,13 +380,22 @@ int i810_init(void)
 	dev->device = MKDEV(MISC_MAJOR, i810_misc.minor);
 	dev->name   = I810_NAME;
 
-   	printk("doing mem init\n");
+   	DRM_DEBUG("doing mem init\n");
 	drm_mem_init();
-	printk("doing proc init\n");
+	DRM_DEBUG("doing proc init\n");
 	drm_proc_init(dev);
-	printk("doing agp init\n");
+	DRM_DEBUG("doing agp init\n");
 	dev->agp    = drm_agp_init();
-	printk("doing ctxbitmap init\n");
+   	if(dev->agp == NULL) {
+	   	DRM_INFO("The i810 drm module requires the agpgart module"
+			 " to function correctly\nPlease load the agpgart"
+			 " module before you load the i810 module\n");
+	   	drm_proc_cleanup();
+	   	misc_deregister(&i810_misc);
+	   	i810_takedown(dev);
+	   	return -ENOMEM;
+	}
+	DRM_DEBUG("doing ctxbitmap init\n");
 	if((retcode = drm_ctxbitmap_init(dev))) {
 		DRM_ERROR("Cannot allocate memory for context bitmap.\n");
 		drm_proc_cleanup();
@@ -386,10 +403,6 @@ int i810_init(void)
 		i810_takedown(dev);
 		return retcode;
 	}
-#if 0
-	printk("doing i810_dma_init\n");
-	i810_dma_init(dev);
-#endif
 
 	DRM_INFO("Initialized %s %d.%d.%d %s on minor %d\n",
 		 I810_NAME,
@@ -417,7 +430,6 @@ void i810_cleanup(void)
 		DRM_INFO("Module unloaded\n");
 	}
 	drm_ctxbitmap_cleanup(dev);
-	i810_dma_cleanup(dev);
 	i810_takedown(dev);
 	if (dev->agp) {
 		drm_free(dev->agp, sizeof(*dev->agp), DRM_MEM_AGPLISTS);
@@ -484,24 +496,82 @@ int i810_release(struct inode *inode, struct file *filp)
 	drm_device_t  *dev    = priv->dev;
 	int	      retcode = 0;
 
-	DRM_DEBUG("open_count = %d\n", dev->open_count);
-	if (!(retcode = drm_release(inode, filp))) {
-		MOD_DEC_USE_COUNT;
-		atomic_inc(&dev->total_close);
-		spin_lock(&dev->count_lock);
-		if (!--dev->open_count) {
-			if (atomic_read(&dev->ioctl_count) || dev->blocked) {
-				DRM_ERROR("Device busy: %d %d\n",
-					  atomic_read(&dev->ioctl_count),
-					  dev->blocked);
-				spin_unlock(&dev->count_lock);
-				return -EBUSY;
+	DRM_DEBUG("pid = %d, device = 0x%x, open_count = %d\n",
+		  current->pid, dev->device, dev->open_count);
+
+	if (_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)
+	    && dev->lock.pid == current->pid) {
+	      	i810_reclaim_buffers(dev, priv->pid);
+		DRM_ERROR("Process %d dead, freeing lock for context %d\n",
+			  current->pid,
+			  _DRM_LOCKING_CONTEXT(dev->lock.hw_lock->lock));
+		drm_lock_free(dev,
+			      &dev->lock.hw_lock->lock,
+			      _DRM_LOCKING_CONTEXT(dev->lock.hw_lock->lock));
+		
+				/* FIXME: may require heavy-handed reset of
+                                   hardware at this point, possibly
+                                   processed via a callback to the X
+                                   server. */
+	} else {
+	   	/* The lock is required to reclaim buffers */
+	   	DECLARE_WAITQUEUE(entry, current);
+	   	add_wait_queue(&dev->lock.lock_queue, &entry);
+		for (;;) {
+			if (!dev->lock.hw_lock) {
+				/* Device has been unregistered */
+				retcode = -EINTR;
+				break;
+			}
+			if (drm_lock_take(&dev->lock.hw_lock->lock,
+					  DRM_KERNEL_CONTEXT)) {
+				dev->lock.pid	    = priv->pid;
+				dev->lock.lock_time = jiffies;
+				atomic_inc(&dev->total_locks);
+				break;	/* Got lock */
+			}			
+				/* Contention */
+			atomic_inc(&dev->total_sleeps);
+			current->state = TASK_INTERRUPTIBLE;
+			schedule();
+			if (signal_pending(current)) {
+				retcode = -ERESTARTSYS;
+				break;
 			}
-			spin_unlock(&dev->count_lock);
-			return i810_takedown(dev);
 		}
-		spin_unlock(&dev->count_lock);
+		current->state = TASK_RUNNING;
+		remove_wait_queue(&dev->lock.lock_queue, &entry);
+	   	if(!retcode) {
+		   	i810_reclaim_buffers(dev, priv->pid);
+		   	drm_lock_free(dev, &dev->lock.hw_lock->lock,
+				      DRM_KERNEL_CONTEXT);
+		}
+	}
+	drm_fasync(-1, filp, 0);
+
+	down(&dev->struct_sem);
+	if (priv->prev) priv->prev->next = priv->next;
+	else		dev->file_first	 = priv->next;
+	if (priv->next) priv->next->prev = priv->prev;
+	else		dev->file_last	 = priv->prev;
+	up(&dev->struct_sem);
+	
+	drm_free(priv, sizeof(*priv), DRM_MEM_FILES);
+   	MOD_DEC_USE_COUNT;
+   	atomic_inc(&dev->total_close);
+   	spin_lock(&dev->count_lock);
+   	if (!--dev->open_count) {
+	   	if (atomic_read(&dev->ioctl_count) || dev->blocked) {
+		   	DRM_ERROR("Device busy: %d %d\n",
+				  atomic_read(&dev->ioctl_count),
+				  dev->blocked);
+		   	spin_unlock(&dev->count_lock);
+		   	return -EBUSY;
+		}
+	   	spin_unlock(&dev->count_lock);
+	   	return i810_takedown(dev);
 	}
+   	spin_unlock(&dev->count_lock);
 	return retcode;
 }
 
@@ -566,8 +636,7 @@ int i810_unlock(struct inode *inode, struct file *filp, unsigned int cmd,
 	atomic_inc(&dev->total_unlocks);
 	if (_DRM_LOCK_IS_CONT(dev->lock.hw_lock->lock))
 		atomic_inc(&dev->total_contends);
-	drm_lock_transfer(dev, &dev->lock.hw_lock->lock, DRM_KERNEL_CONTEXT);
-	i810_dma_schedule(dev, 1);
+   	drm_lock_transfer(dev, &dev->lock.hw_lock->lock, DRM_KERNEL_CONTEXT);
 	if (!dev->context_flag) {
 		if (drm_lock_free(dev, &dev->lock.hw_lock->lock,
 				  DRM_KERNEL_CONTEXT)) {
diff --git a/linux-core/i810_drv.h b/linux-core/i810_drv.h
index 0f5f42bb..1badd36b 100644
--- a/linux-core/i810_drv.h
+++ b/linux-core/i810_drv.h
@@ -32,6 +32,31 @@
 #ifndef _I810_DRV_H_
 #define _I810_DRV_H_
 
+typedef struct _drm_i810_ring_buffer{
+	int tail_mask;
+	unsigned long Start;
+	unsigned long End;
+	unsigned long Size;
+	u8 *virtual_start;
+	int head;
+	int tail;
+	int space;
+} drm_i810_ring_buffer_t;
+
+typedef struct drm_i810_private {
+   	int ring_map_idx;
+   	int buffer_map_idx;
+
+   	drm_i810_ring_buffer_t ring;
+	drm_i810_sarea_t *sarea_priv;
+
+      	unsigned long hw_status_page;
+   	unsigned long counter;
+
+   	atomic_t flush_done;
+   	wait_queue_head_t flush_queue;	/* Processes waiting until flush    */
+} drm_i810_private_t;
+
 				/* i810_drv.c */
 extern int  i810_init(void);
 extern void i810_cleanup(void);
@@ -54,8 +79,13 @@ extern int  i810_control(struct inode *inode, struct file *filp,
 			  unsigned int cmd, unsigned long arg);
 extern int  i810_lock(struct inode *inode, struct file *filp,
 		       unsigned int cmd, unsigned long arg);
-extern void i810_dma_init(drm_device_t *dev);
-extern void i810_dma_cleanup(drm_device_t *dev);
+extern int  i810_dma_init(struct inode *inode, struct file *filp,
+			  unsigned int cmd, unsigned long arg);
+extern int  i810_flush_ioctl(struct inode *inode, struct file *filp,
+			     unsigned int cmd, unsigned long arg);
+extern void i810_reclaim_buffers(drm_device_t *dev, pid_t pid);
+extern int  i810_getage(struct inode *inode, struct file *filp, unsigned int cmd,
+			unsigned long arg);
 
 
 				/* i810_bufs.c */
@@ -72,5 +102,103 @@ extern int  i810_mapbufs(struct inode *inode, struct file *filp,
 extern int  i810_addmap(struct inode *inode, struct file *filp,
 		       unsigned int cmd, unsigned long arg);
 
+				/* i810_context.c */
+extern int  i810_resctx(struct inode *inode, struct file *filp,
+		       unsigned int cmd, unsigned long arg);
+extern int  i810_addctx(struct inode *inode, struct file *filp,
+		       unsigned int cmd, unsigned long arg);
+extern int  i810_modctx(struct inode *inode, struct file *filp,
+		       unsigned int cmd, unsigned long arg);
+extern int  i810_getctx(struct inode *inode, struct file *filp,
+		       unsigned int cmd, unsigned long arg);
+extern int  i810_switchctx(struct inode *inode, struct file *filp,
+			  unsigned int cmd, unsigned long arg);
+extern int  i810_newctx(struct inode *inode, struct file *filp,
+		       unsigned int cmd, unsigned long arg);
+extern int  i810_rmctx(struct inode *inode, struct file *filp,
+		      unsigned int cmd, unsigned long arg);
+
+extern int  i810_context_switch(drm_device_t *dev, int old, int new);
+extern int  i810_context_switch_complete(drm_device_t *dev, int new);
+
+
+
+
+/* Copy the outstanding cliprects for every I810_DMA_VERTEX buffer.
+ * This can be fixed by emitting directly to the ringbuffer in the
+ * 'vertex_dma' ioctl.  
+*/
+typedef struct {
+   	u32 *in_use;
+   	int my_use_idx;
+} drm_i810_buf_priv_t;
+
+
+#define I810_DMA_GENERAL 0
+#define I810_DMA_VERTEX  1
+#define I810_DMA_DISCARD 2	/* not used */
+
+#define I810_VERBOSE 0
+
+
+int i810_dma_vertex(struct inode *inode, struct file *filp,
+		    unsigned int cmd, unsigned long arg);
+
+int i810_dma_general(struct inode *inode, struct file *filp,
+		     unsigned int cmd, unsigned long arg);
+
+
+#define GFX_OP_USER_INTERRUPT 		((0<<29)|(2<<23))
+#define GFX_OP_BREAKPOINT_INTERRUPT	((0<<29)|(1<<23))
+#define CMD_REPORT_HEAD			(7<<23)
+#define CMD_STORE_DWORD_IDX		((0x21<<23) | 0x1)
+#define CMD_OP_BATCH_BUFFER  ((0x0<<29)|(0x30<<23)|0x1)
+
+#define INST_PARSER_CLIENT   0x00000000
+#define INST_OP_FLUSH        0x02000000
+#define INST_FLUSH_MAP_CACHE 0x00000001
+
+
+#define BB1_START_ADDR_MASK   (~0x7)
+#define BB1_PROTECTED         (1<<0)
+#define BB1_UNPROTECTED       (0<<0)
+#define BB2_END_ADDR_MASK     (~0x7)
+
+#define I810REG_HWSTAM		0x02098
+#define I810REG_INT_IDENTITY_R	0x020a4
+#define I810REG_INT_MASK_R 	0x020a8
+#define I810REG_INT_ENABLE_R	0x020a0
+
+#define LP_RING     		0x2030
+#define HP_RING     		0x2040
+#define RING_TAIL      		0x00
+#define TAIL_ADDR		0x000FFFF8
+#define RING_HEAD      		0x04
+#define HEAD_WRAP_COUNT     	0xFFE00000
+#define HEAD_WRAP_ONE       	0x00200000
+#define HEAD_ADDR           	0x001FFFFC
+#define RING_START     		0x08
+#define START_ADDR          	0x00FFFFF8
+#define RING_LEN       		0x0C
+#define RING_NR_PAGES       	0x000FF000 
+#define RING_REPORT_MASK    	0x00000006
+#define RING_REPORT_64K     	0x00000002
+#define RING_REPORT_128K    	0x00000004
+#define RING_NO_REPORT      	0x00000000
+#define RING_VALID_MASK     	0x00000001
+#define RING_VALID          	0x00000001
+#define RING_INVALID        	0x00000000
+
+#define GFX_OP_SCISSOR         ((0x3<<29)|(0x1c<<24)|(0x10<<19))
+#define SC_UPDATE_SCISSOR       (0x1<<1)
+#define SC_ENABLE_MASK          (0x1<<0)
+#define SC_ENABLE               (0x1<<0)
+
+#define GFX_OP_SCISSOR_INFO    ((0x3<<29)|(0x1d<<24)|(0x81<<16)|(0x1))
+#define SCI_YMIN_MASK      (0xffff<<16)
+#define SCI_XMIN_MASK      (0xffff<<0)
+#define SCI_YMAX_MASK      (0xffff<<16)
+#define SCI_XMAX_MASK      (0xffff<<0)
 
 #endif
+
diff --git a/linux-core/mga_drv.c b/linux-core/mga_drv.c
index 8bc25617..48041354 100644
--- a/linux-core/mga_drv.c
+++ b/linux-core/mga_drv.c
@@ -54,6 +54,7 @@ static struct file_operations mga_fops = {
 	mmap:	 drm_mmap,
 	read:	 drm_read,
 	fasync:	 drm_fasync,
+   	poll:	 drm_poll,
 };
 
 static struct miscdevice      mga_misc = {
@@ -105,9 +106,11 @@ static drm_ioctl_desc_t	      mga_ioctls[] = {
 	[DRM_IOCTL_NR(DRM_IOCTL_AGP_BIND)]    = { drm_agp_bind,    1, 1 },
 	[DRM_IOCTL_NR(DRM_IOCTL_AGP_UNBIND)]  = { drm_agp_unbind,  1, 1 },
    	[DRM_IOCTL_NR(DRM_IOCTL_MGA_INIT)]    = { mga_dma_init,    1, 1 },
-   	[DRM_IOCTL_NR(DRM_IOCTL_MGA_SWAP)]    = { mga_clear_bufs,  1, 1 },
-   	[DRM_IOCTL_NR(DRM_IOCTL_MGA_CLEAR)]   = { mga_swap_bufs,   1, 1 },
-   	[DRM_IOCTL_NR(DRM_IOCTL_MGA_ILOAD)]   = { mga_iload,       1, 1 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_MGA_SWAP)]    = { mga_swap_bufs,   1, 0 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_MGA_CLEAR)]   = { mga_clear_bufs,  1, 0 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_MGA_ILOAD)]   = { mga_iload,       1, 0 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_MGA_VERTEX)]  = { mga_vertex,      1, 0 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_MGA_FLUSH)]   = { mga_flush_ioctl, 1, 0 },
 };
 
 #define MGA_IOCTL_COUNT DRM_ARRAY_SIZE(mga_ioctls)
@@ -380,6 +383,21 @@ int mga_init(void)
 	drm_proc_init(dev);
 	DRM_DEBUG("doing agp init\n");
 	dev->agp    = drm_agp_init();
+      	if(dev->agp == NULL) {
+	   	DRM_DEBUG("The mga drm module requires the agpgart module"
+		          " to function correctly\nPlease load the agpgart"
+		          " module before you load the mga module\n");
+	   	drm_proc_cleanup();
+	   	misc_deregister(&mga_misc);
+	   	mga_takedown(dev);
+	   	return -ENOMEM;
+	}
+#ifdef CONFIG_MTRR
+   	dev->agp->agp_mtrr = mtrr_add(dev->agp->agp_info.aper_base,
+				      dev->agp->agp_info.aper_size * 1024 * 1024,
+				      MTRR_TYPE_WRCOMB,
+				      1);
+#endif
 	DRM_DEBUG("doing ctxbitmap init\n");
 	if((retcode = drm_ctxbitmap_init(dev))) {
 		DRM_ERROR("Cannot allocate memory for context bitmap.\n");
@@ -416,6 +434,16 @@ void mga_cleanup(void)
 	}
 	drm_ctxbitmap_cleanup(dev);
 	mga_dma_cleanup(dev);
+#ifdef CONFIG_MTRR
+   	if(dev->agp && dev->agp->agp_mtrr) {
+	   	int retval;
+	   	retval = mtrr_del(dev->agp->agp_mtrr, 
+				  dev->agp->agp_info.aper_base,
+				  dev->agp->agp_info.aper_size * 1024*1024);
+	   	DRM_DEBUG("mtrr_del = %d\n", retval);
+	}
+#endif
+
 	mga_takedown(dev);
 	if (dev->agp) {
 		drm_free(dev->agp, sizeof(*dev->agp), DRM_MEM_AGPLISTS);
diff --git a/linux-core/tdfx_drv.c b/linux-core/tdfx_drv.c
index 57c1c719..fb7a997b 100644
--- a/linux-core/tdfx_drv.c
+++ b/linux-core/tdfx_drv.c
@@ -85,6 +85,16 @@ static drm_ioctl_desc_t	      tdfx_ioctls[] = {
 	[DRM_IOCTL_NR(DRM_IOCTL_LOCK)]	     = { tdfx_lock,	  1, 0 },
 	[DRM_IOCTL_NR(DRM_IOCTL_UNLOCK)]     = { tdfx_unlock,	  1, 0 },
 	[DRM_IOCTL_NR(DRM_IOCTL_FINISH)]     = { drm_finish,	  1, 0 },
+#ifdef DRM_AGP
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_ACQUIRE)]   = {drm_agp_acquire, 1, 1},
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_RELEASE)]   = {drm_agp_release, 1, 1},
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_ENABLE)]    = {drm_agp_enable,  1, 1},
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_INFO)]      = {drm_agp_info,    1, 1},
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_ALLOC)]     = {drm_agp_alloc,   1, 1},
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_FREE)]      = {drm_agp_free,    1, 1},
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_BIND)]      = {drm_agp_unbind,  1, 1},
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_UNBIND)]    = {drm_agp_bind,    1, 1},
+#endif
 };
 #define TDFX_IOCTL_COUNT DRM_ARRAY_SIZE(tdfx_ioctls)
 
@@ -228,7 +238,24 @@ static int tdfx_takedown(drm_device_t *dev)
 		}
 		dev->magiclist[i].head = dev->magiclist[i].tail = NULL;
 	}
-	
+#ifdef DRM_AGP
+				/* Clear AGP information */
+	if (dev->agp) {
+		drm_agp_mem_t *temp;
+		drm_agp_mem_t *temp_next;
+	   
+		temp = dev->agp->memory;
+		while(temp != NULL) {
+			temp_next = temp->next;
+			drm_free_agp(temp->memory, temp->pages);
+			drm_free(temp, sizeof(*temp), DRM_MEM_AGPLISTS);
+			temp = temp_next;
+		}
+		if(dev->agp->acquired) (*drm_agp.release)();
+		drm_free(dev->agp, sizeof(*dev->agp), DRM_MEM_AGPLISTS);
+		dev->agp = NULL;
+	}
+#endif
 				/* Clear vma list (only built for debugging) */
 	if (dev->vmalist) {
 		for (vma = dev->vmalist; vma; vma = vma_next) {
@@ -262,6 +289,10 @@ static int tdfx_takedown(drm_device_t *dev)
 					       - PAGE_SHIFT,
 					       DRM_MEM_SAREA);
 				break;
+			case _DRM_AGP:
+				/* Do nothing here, because this is all
+                                   handled in the AGP/GART driver. */
+				break;
 			}
 			drm_free(map, sizeof(*map), DRM_MEM_MAPS);
 		}
@@ -309,6 +340,16 @@ int tdfx_init(void)
 
 	drm_mem_init();
 	drm_proc_init(dev);
+#ifdef DRM_AGP
+	dev->agp    = drm_agp_init();
+#endif
+	if((retcode = drm_ctxbitmap_init(dev))) {
+		DRM_ERROR("Cannot allocate memory for context bitmap.\n");
+		drm_proc_cleanup();
+		misc_deregister(&tdfx_misc);
+		tdfx_takedown(dev);
+		return retcode;
+	}
 
 	DRM_INFO("Initialized %s %d.%d.%d %s on minor %d\n",
 		 TDFX_NAME,
@@ -335,6 +376,7 @@ void tdfx_cleanup(void)
 	} else {
 		DRM_INFO("Module unloaded\n");
 	}
+	drm_ctxbitmap_cleanup(dev);
 	tdfx_takedown(dev);
 }
 
diff --git a/linux/Makefile.kernel b/linux/Makefile.kernel
index 44d75c48..2ea6c721 100644
--- a/linux/Makefile.kernel
+++ b/linux/Makefile.kernel
@@ -14,7 +14,8 @@
 L_TARGET        := libdrm.a
 
 L_OBJS		:= init.o memory.o proc.o auth.o context.o drawable.o bufs.o \
-			lists.o lock.o ioctl.o fops.o vm.o dma.o
+			lists.o lock.o ioctl.o fops.o vm.o dma.o ctxbitmap.o \
+			agpsupport.o
 
 M_OBJS		:= 
 
@@ -26,6 +27,14 @@ ifdef CONFIG_DRM_TDFX
 M_OBJS          += tdfx.o
 endif
 
+ifdef CONFIG_DRM_MGA
+M_OBJS		+= mga.o
+endif
+
+ifdef CONFIG_DRM_R128
+M_OBJS		+= r128.o
+endif
+
 include $(TOPDIR)/Rules.make
 
 gamma.o: gamma_drv.o gamma_dma.o $(L_TARGET)
@@ -33,3 +42,10 @@ gamma.o: gamma_drv.o gamma_dma.o $(L_TARGET)
 
 tdfx.o: tdfx_drv.o tdfx_context.o $(L_TARGET)
 	$(LD) $(LD_RFLAG) -r -o $@ tdfx_drv.o tdfx_context.o -L. -ldrm
+
+i810.o: i810_drv.o i810_context.o $(L_TARGET)
+	$(LD) $(LD_RFLAG) -r -o $@ i810_drv.o i810_bufs.o i810_dma.o i810_context.o -L. -ldrm
+
+mga.o: mga_drv.o mga_context.o mga_dma.o mga_bufs.o $(L_TARGET)
+	$(LD) $(LD_RFLAG) -r -o $@ mga_drv.o mga_bufs.o mga_dma.o mga_context.o mga_state.o -L. -ldrm
+
diff --git a/linux/Makefile.linux b/linux/Makefile.linux
index 8a41f880..1db7a81f 100644
--- a/linux/Makefile.linux
+++ b/linux/Makefile.linux
@@ -31,12 +31,14 @@
 
 # *** Setup
 
+# **** End of SMP/MODVERSIONS detection
+
 MODS=           gamma.o tdfx.o
 LIBS=           libdrm.a
 PROGS=		drmstat
 
 DRMOBJS=	init.o memory.o proc.o auth.o context.o drawable.o bufs.o \
-		lists.o lock.o ioctl.o fops.o vm.o dma.o
+		lists.o lock.o ioctl.o fops.o vm.o dma.o ctxbitmap.o
 DRMHEADERS=	drm.h drmP.h
 
 GAMMAOBJS=      gamma_drv.o gamma_dma.o
@@ -48,6 +50,8 @@ TDFXHEADERS=    tdfx_drv.h $(DRMHEADERS)
 PROGOBJS=       drmstat.po xf86drm.po xf86drmHash.po xf86drmRandom.po sigio.po
 PROGHEADERS=    xf86drm.h $(DRMHEADERS)
 
+INC=		/usr/include
+
 CFLAGS=		-O2 $(WARNINGS)
 WARNINGS=	-Wall -Wwrite-strings -Wpointer-arith -Wcast-align \
 		-Wstrict-prototypes -Wshadow -Wnested-externs \
@@ -102,7 +106,27 @@ SMP := $(shell gcc -E -nostdinc -I$(TREE) picker.c 2>/dev/null \
 	| grep -s 'SMP = ' | cut -d' ' -f3)
 MODVERSIONS := $(shell gcc -E -I $(TREE) picker.c 2>/dev/null \
 	| grep -s 'MODVERSIONS = ' | cut -d' ' -f3)
-all::;@echo KERNEL HEADERS IN $(TREE): SMP=${SMP} MODVERSIONS=${MODVERSIONS}
+AGP := $(shell gcc -E -nostdinc -I$(TREE) picker.c 2>/dev/null \
+	| grep -s 'AGP = ' | cut -d' ' -f3)
+ifeq ($(AGP),0)
+AGP := $(shell gcc -E -nostdinc -I$(TREE) picker.c 2>/dev/null \
+	| grep -s 'AGP_MODULE = ' | cut -d' ' -f3)
+endif
+
+ifeq ($(AGP),1)
+MODCFLAGS += -DDRM_AGP
+DRMOBJS += agpsupport.o
+MODS += mga.o i810.o
+
+MGAOBJS=	mga_drv.o mga_dma.o mga_bufs.o mga_state.o mga_context.o
+MGAHEADERS=	mga_drv.h $(DRMHEADERS)
+
+I810OBJS=	i810_drv.o i810_dma.o i810_bufs.o i810_context.o
+I810HEADERS=	i810_drv.h $(DRMHEADERS)
+endif
+
+all::;@echo KERNEL HEADERS IN $(TREE): SMP=${SMP} MODVERSIONS=${MODVERSIONS} \
+	AGP=${AGP}
 all:: $(LIBS) $(MODS) $(PROGS)
 endif
 
@@ -116,6 +140,7 @@ ifeq ($(MODVERSIONS),1)
 MODCFLAGS += -DMODVERSIONS -include $(TREE)/linux/modversions.h
 endif
 
+
 # **** End of configuration
 
 libdrm.a: $(DRMOBJS)
@@ -128,6 +153,14 @@ gamma.o: $(GAMMAOBJS) $(LIBS)
 tdfx.o: $(TDFXOBJS) $(LIBS)
 	$(LD) -r $^ -o $@
 
+ifeq ($(AGP),1)
+mga.o: $(MGAOBJS) $(LIBS)
+	$(LD) -r $^ -o $@
+
+i810.o: $(I810OBJS) $(LIBS)
+	$(LD) -r $^ -o $@
+endif
+
 drmstat: $(PROGOBJS)
 	$(CC) $(PRGCFLAGS) $^ $(PRGLIBS) -o $@
 
@@ -149,8 +182,11 @@ ChangeLog:
 $(DRMOBJS): $(DRMHEADERS)
 $(GAMMAOBJS): $(GAMMAHEADERS)
 $(TDFXOBJS): $(TDFXHEADERS)
+ifeq ($(AGP),1)
+$(MGAOBJS): $(MGAHEADERS)
+$(I810OBJS): $(I810HEADERS)
+endif
 $(PROGOBJS): $(PROGHEADERS)
 
 clean:
-	rm -f *.o *.po *~ core $(PROGS)
-
+	rm -f *.o *.a *.po *~ core $(PROGS)
diff --git a/linux/agpsupport.c b/linux/agpsupport.c
index fb58154d..c2da9ec7 100644
--- a/linux/agpsupport.c
+++ b/linux/agpsupport.c
@@ -159,6 +159,8 @@ int drm_agp_alloc(struct inode *inode, struct file *filp, unsigned int cmd,
 			   -EFAULT);
 	if (!(entry = drm_alloc(sizeof(*entry), DRM_MEM_AGPLISTS)))
 		return -ENOMEM;
+   
+   	memset(entry, 0, sizeof(*entry));
 
 	pages = (request.size + PAGE_SIZE - 1) / PAGE_SIZE;
 	type = (u32) request.type;
@@ -254,8 +256,10 @@ int drm_agp_free(struct inode *inode, struct file *filp, unsigned int cmd,
 	if (!(entry = drm_agp_lookup_entry(dev, request.handle)))
 		return -EINVAL;
 	if (entry->bound) drm_unbind_agp(entry->memory);
-	entry->prev->next = entry->next;
-	entry->next->prev = entry->prev;
+   
+	if (entry->prev) entry->prev->next = entry->next;
+	else             dev->agp->memory  = entry->next;
+	if (entry->next) entry->next->prev = entry->prev;
 	drm_free_agp(entry->memory, entry->pages);
 	drm_free(entry, sizeof(*entry), DRM_MEM_AGPLISTS);
 	return 0;
@@ -269,15 +273,12 @@ drm_agp_head_t *drm_agp_init(void)
 
 	for (fill = &drm_agp_fill[0]; fill->name; fill++) {
 		char *n  = (char *)fill->name;
-#if 0
-		*fill->f = (drm_agp_func_u)get_module_symbol(NULL, n);
-#endif
 		*fill->f = (drm_agp_func_u)get_module_symbol(NULL, n);
-		printk("%s resolves to 0x%08lx\n", n, (*fill->f).address);
+		DRM_DEBUG("%s resolves to 0x%08lx\n", n, (*fill->f).address);
 		if (!(*fill->f).address) agp_available = 0;
 	}
    
-	printk("agp_available = %d\n", agp_available);
+	DRM_DEBUG("agp_available = %d\n", agp_available);
 
 	if (agp_available) {
 		if (!(head = drm_alloc(sizeof(*head), DRM_MEM_AGPLISTS)))
diff --git a/linux/bufs.c b/linux/bufs.c
index 85244c7d..7902f0cc 100644
--- a/linux/bufs.c
+++ b/linux/bufs.c
@@ -102,6 +102,11 @@ int drm_addmap(struct inode *inode, struct file *filp, unsigned int cmd,
 			dev->lock.hw_lock = map->handle; /* Pointer to lock */
 		}
 		break;
+#ifdef DRM_AGP
+	case _DRM_AGP:
+		map->offset = map->offset + dev->agp->base;
+		break;
+#endif
 	default:
 		drm_free(map, sizeof(*map), DRM_MEM_MAPS);
 		return -EINVAL;
@@ -172,7 +177,7 @@ int drm_addbufs(struct inode *inode, struct file *filp, unsigned int cmd,
 	if (order < DRM_MIN_ORDER || order > DRM_MAX_ORDER) return -EINVAL;
 	if (dev->queue_count) return -EBUSY; /* Not while in use */
 
-	alignment  = (request.flags & DRM_PAGE_ALIGN) ? PAGE_ALIGN(size) :size;
+	alignment  = (request.flags & _DRM_PAGE_ALIGN) ? PAGE_ALIGN(size):size;
 	page_order = order - PAGE_SHIFT > 0 ? order - PAGE_SHIFT : 0;
 	total	   = PAGE_SIZE << page_order;
 
diff --git a/linux/ctxbitmap.c b/linux/ctxbitmap.c
index f67209d4..781984f1 100644
--- a/linux/ctxbitmap.c
+++ b/linux/ctxbitmap.c
@@ -53,7 +53,7 @@ int drm_ctxbitmap_next(drm_device_t *dev)
 	bit = find_first_zero_bit(dev->ctx_bitmap, DRM_MAX_CTXBITMAP);
 	if (bit < DRM_MAX_CTXBITMAP) {
 		set_bit(bit, dev->ctx_bitmap);
-	   	printk("drm_ctxbitmap_next bit : %d\n", bit);
+	   	DRM_DEBUG("drm_ctxbitmap_next bit : %d\n", bit);
 		return bit;
 	}
 	return -1;
@@ -64,15 +64,15 @@ int drm_ctxbitmap_init(drm_device_t *dev)
 	int i;
    	int temp;
 
-	dev->ctx_bitmap = (unsigned long *) drm_alloc(PAGE_SIZE * 4, 
+	dev->ctx_bitmap = (unsigned long *) drm_alloc(PAGE_SIZE, 
 						      DRM_MEM_CTXBITMAP);
 	if(dev->ctx_bitmap == NULL) {
 		return -ENOMEM;
 	}
-	memset((void *) dev->ctx_bitmap, 0, PAGE_SIZE * 4);
+	memset((void *) dev->ctx_bitmap, 0, PAGE_SIZE);
 	for(i = 0; i < DRM_RESERVED_CONTEXTS; i++) {
 		temp = drm_ctxbitmap_next(dev);
-	   	printk("drm_ctxbitmap_init : %d\n", temp);
+	   	DRM_DEBUG("drm_ctxbitmap_init : %d\n", temp);
 	}
 
 	return 0;
@@ -80,7 +80,7 @@ int drm_ctxbitmap_init(drm_device_t *dev)
 
 void drm_ctxbitmap_cleanup(drm_device_t *dev)
 {
-	drm_free((void *)dev->ctx_bitmap, PAGE_SIZE * 4,
+	drm_free((void *)dev->ctx_bitmap, PAGE_SIZE,
 		 DRM_MEM_CTXBITMAP);
 }
 
diff --git a/linux/dma.c b/linux/dma.c
index 8291e52e..37ab674f 100644
--- a/linux/dma.c
+++ b/linux/dma.c
@@ -63,15 +63,24 @@ void drm_dma_takedown(drm_device_t *dev)
 					       dma->bufs[i].page_order,
 					       DRM_MEM_DMA);
 			}
-			drm_free(dma->bufs[i].buflist,
-				 dma->buf_count
-				 * sizeof(*dma->bufs[0].buflist),
-				 DRM_MEM_BUFS);
 			drm_free(dma->bufs[i].seglist,
-				 dma->buf_count
+				 dma->bufs[i].seg_count
 				 * sizeof(*dma->bufs[0].seglist),
 				 DRM_MEM_SEGS);
-			drm_freelist_destroy(&dma->bufs[i].freelist);
+		}
+	   	if(dma->bufs[i].buf_count) {
+		   	for(j = 0; j < dma->bufs[i].buf_count; j++) {
+			   if(dma->bufs[i].buflist[j].dev_private) {
+			      drm_free(dma->bufs[i].buflist[j].dev_private,
+				       dma->bufs[i].buflist[j].dev_priv_size,
+				       DRM_MEM_BUFS);
+			   }
+			}
+		   	drm_free(dma->bufs[i].buflist,
+				 dma->bufs[i].buf_count *
+				 sizeof(*dma->bufs[0].buflist),
+				 DRM_MEM_BUFS);
+		   	drm_freelist_destroy(&dma->bufs[i].freelist);
 		}
 	}
 	
diff --git a/linux/drm.h b/linux/drm.h
index 7b8e8826..ae4c65ca 100644
--- a/linux/drm.h
+++ b/linux/drm.h
@@ -61,6 +61,19 @@ typedef unsigned int  drm_context_t;
 typedef unsigned int  drm_drawable_t;
 typedef unsigned int  drm_magic_t;
 
+/* Warning: If you change this structure, make sure you change 
+ * XF86DRIClipRectRec in the server as well */
+
+typedef struct drm_clip_rect {
+           unsigned short x1;
+           unsigned short y1;
+           unsigned short x2;
+           unsigned short y2;
+} drm_clip_rect_t;
+
+/* Seperate include files for the i810/mga specific structures */
+#include "mga_drm.h"
+#include "i810_drm.h"
 
 typedef struct drm_version {
 	int    version_major;	  /* Major version			    */
@@ -101,7 +114,8 @@ typedef struct drm_control {
 typedef enum drm_map_type {
 	_DRM_FRAME_BUFFER = 0,	  /* WC (no caching), no core dump	    */
 	_DRM_REGISTERS	  = 1,	  /* no caching, no core dump		    */
-	_DRM_SHM	  = 2	  /* shared, cached			    */
+	_DRM_SHM	  = 2,	  /* shared, cached			    */
+	_DRM_AGP          = 3	  /* AGP/GART                               */
 } drm_map_type_t;
 
 typedef enum drm_map_flags {
@@ -165,8 +179,11 @@ typedef struct drm_buf_desc {
 	int	      low_mark;	 /* Low water mark			     */
 	int	      high_mark; /* High water mark			     */
 	enum {
-		DRM_PAGE_ALIGN = 0x01  /* Align on page boundaries for DMA   */
+		_DRM_PAGE_ALIGN = 0x01, /* Align on page boundaries for DMA  */
+		_DRM_AGP_BUFFER = 0x02  /* Buffer is in agp space            */
 	}	      flags;
+	unsigned long agp_start; /* Start address of where the agp buffers
+				  * are in the agp aperture */
 } drm_buf_desc_t;
 
 typedef struct drm_buf_info {
@@ -237,6 +254,38 @@ typedef struct drm_irq_busid {
 	int funcnum;
 } drm_irq_busid_t;
 
+typedef struct drm_agp_mode {
+	unsigned long mode;
+} drm_agp_mode_t;
+
+				/* For drm_agp_alloc -- allocated a buffer */
+typedef struct drm_agp_buffer {
+	unsigned long size;	/* In bytes -- will round to page boundary */
+	unsigned long handle;	/* Used for BIND/UNBIND ioctls */
+	unsigned long type;     /* Type of memory to allocate  */
+        unsigned long physical; /* Physical used by i810       */
+} drm_agp_buffer_t;
+
+				/* For drm_agp_bind */
+typedef struct drm_agp_binding {
+	unsigned long handle;   /* From drm_agp_buffer */
+	unsigned long offset;	/* In bytes -- will round to page boundary */
+} drm_agp_binding_t;
+
+typedef struct drm_agp_info {
+	int            agp_version_major;
+	int            agp_version_minor;
+	unsigned long  mode;
+	unsigned long  aperture_base;  /* physical address */
+	unsigned long  aperture_size;  /* bytes */
+	unsigned long  memory_allowed; /* bytes */
+	unsigned long  memory_used;
+
+				/* PCI information */
+	unsigned short id_vendor;
+	unsigned short id_device;
+} drm_agp_info_t;
+
 #define DRM_IOCTL_BASE	     'd'
 #define DRM_IOCTL_NR(n)	     _IOC_NR(n)
 #define DRM_IO(nr)	     _IO(DRM_IOCTL_BASE,nr)
@@ -276,4 +325,28 @@ typedef struct drm_irq_busid {
 #define DRM_IOCTL_UNLOCK     DRM_IOW( 0x2b, drm_lock_t)
 #define DRM_IOCTL_FINISH     DRM_IOW( 0x2c, drm_lock_t)
 
+#define DRM_IOCTL_AGP_ACQUIRE DRM_IO(  0x30)
+#define DRM_IOCTL_AGP_RELEASE DRM_IO(  0x31)
+#define DRM_IOCTL_AGP_ENABLE  DRM_IOR( 0x32, drm_agp_mode_t)
+#define DRM_IOCTL_AGP_INFO    DRM_IOW( 0x33, drm_agp_info_t)
+#define DRM_IOCTL_AGP_ALLOC   DRM_IOWR(0x34, drm_agp_buffer_t)
+#define DRM_IOCTL_AGP_FREE    DRM_IOW( 0x35, drm_agp_buffer_t)
+#define DRM_IOCTL_AGP_BIND    DRM_IOWR(0x36, drm_agp_binding_t)
+#define DRM_IOCTL_AGP_UNBIND  DRM_IOW( 0x37, drm_agp_binding_t)
+
+/* Mga specific ioctls */
+#define DRM_IOCTL_MGA_INIT    DRM_IOW( 0x40, drm_mga_init_t)
+#define DRM_IOCTL_MGA_SWAP    DRM_IOW( 0x41, drm_mga_swap_t)
+#define DRM_IOCTL_MGA_CLEAR   DRM_IOW( 0x42, drm_mga_clear_t)
+#define DRM_IOCTL_MGA_ILOAD   DRM_IOW( 0x43, drm_mga_iload_t)
+#define DRM_IOCTL_MGA_VERTEX  DRM_IOW( 0x44, drm_mga_vertex_t)
+#define DRM_IOCTL_MGA_FLUSH   DRM_IOW( 0x45, drm_lock_t )
+
+/* I810 specific ioctls */
+#define DRM_IOCTL_I810_INIT    DRM_IOW( 0x40, drm_i810_init_t)
+#define DRM_IOCTL_I810_VERTEX  DRM_IOW( 0x41, drm_i810_vertex_t)
+#define DRM_IOCTL_I810_DMA     DRM_IOW( 0x42, drm_i810_general_t)
+#define DRM_IOCTL_I810_FLUSH   DRM_IO ( 0x43)
+#define DRM_IOCTL_I810_GETAGE  DRM_IO ( 0x44)
+
 #endif
diff --git a/linux/drmP.h b/linux/drmP.h
index 312fba36..f8e78eab 100644
--- a/linux/drmP.h
+++ b/linux/drmP.h
@@ -48,8 +48,12 @@
 #ifdef CONFIG_MTRR
 #include <asm/mtrr.h>
 #endif
+#ifdef DRM_AGP
+#include <linux/types.h>
+#include <linux/agp_backend.h>
+#endif
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,1,0)
-#include <asm/spinlock.h>
+#include <linux/tqueue.h>
 #include <linux/poll.h>
 #endif
 #include "drm.h"
@@ -69,21 +73,27 @@
 #define DRM_FLAG_DEBUG	  0x01
 #define DRM_FLAG_NOCTX	  0x02
 
-#define DRM_MEM_DMA	  0
-#define DRM_MEM_SAREA	  1
-#define DRM_MEM_DRIVER	  2
-#define DRM_MEM_MAGIC	  3
-#define DRM_MEM_IOCTLS	  4
-#define DRM_MEM_MAPS	  5
-#define DRM_MEM_VMAS	  6
-#define DRM_MEM_BUFS	  7
-#define DRM_MEM_SEGS	  8
-#define DRM_MEM_PAGES	  9
-#define DRM_MEM_FILES	 10
-#define DRM_MEM_QUEUES	 11
-#define DRM_MEM_CMDS	 12
-#define DRM_MEM_MAPPINGS 13
-#define DRM_MEM_BUFLISTS 14
+#define DRM_MEM_DMA	   0
+#define DRM_MEM_SAREA	   1
+#define DRM_MEM_DRIVER	   2
+#define DRM_MEM_MAGIC	   3
+#define DRM_MEM_IOCTLS	   4
+#define DRM_MEM_MAPS	   5
+#define DRM_MEM_VMAS	   6
+#define DRM_MEM_BUFS	   7
+#define DRM_MEM_SEGS	   8
+#define DRM_MEM_PAGES	   9
+#define DRM_MEM_FILES	  10
+#define DRM_MEM_QUEUES	  11
+#define DRM_MEM_CMDS	  12
+#define DRM_MEM_MAPPINGS  13
+#define DRM_MEM_BUFLISTS  14
+#define DRM_MEM_AGPLISTS  15
+#define DRM_MEM_TOTALAGP  16
+#define DRM_MEM_BOUNDAGP  17
+#define DRM_MEM_CTXBITMAP 18
+
+#define DRM_MAX_CTXBITMAP (PAGE_SIZE * 8)
 
 				/* Backward compatibility section */
 				/* _PAGE_WT changed to _PAGE_PWT in 2.2.6 */
@@ -235,6 +245,7 @@ typedef struct drm_buf {
 	int		  used;	       /* Amount of buffer in use (for DMA)  */
 	unsigned long	  offset;      /* Byte offset (used internally)	     */
 	void		  *address;    /* Address of buffer		     */
+	unsigned long	  bus_address; /* Bus address of buffer		     */
 	struct drm_buf	  *next;       /* Kernel-only: used for free list    */
 	__volatile__ int  waiting;     /* On kernel DMA queue		     */
 	__volatile__ int  pending;     /* On hardware DMA queue		     */
@@ -250,6 +261,11 @@ typedef struct drm_buf {
 		DRM_LIST_PRIO	 = 4,
 		DRM_LIST_RECLAIM = 5
 	}		  list;	       /* Which list we're on		     */
+
+
+	void *dev_private;
+	int dev_priv_size;
+
 #if DRM_DMA_HISTOGRAM
 	cycles_t	  time_queued;	   /* Queued to kernel DMA queue     */
 	cycles_t	  time_dispatched; /* Dispatched to hardware	     */
@@ -376,6 +392,9 @@ typedef struct drm_device_dma {
 	int		  page_count;
 	unsigned long	  *pagelist;
 	unsigned long	  byte_count;
+	enum {
+	   _DRM_DMA_USE_AGP = 0x01
+	} flags;
 
 				/* DMA support */
 	drm_buf_t	  *this_buffer;	/* Buffer being sent		   */
@@ -384,6 +403,41 @@ typedef struct drm_device_dma {
 	wait_queue_head_t waiting;	/* Processes waiting on free bufs  */
 } drm_device_dma_t;
 
+#ifdef DRM_AGP
+typedef struct drm_agp_mem {
+	unsigned long      handle;
+	agp_memory         *memory;
+	unsigned long      bound; /* address */
+	int                pages;
+	struct drm_agp_mem *prev;
+	struct drm_agp_mem *next;
+} drm_agp_mem_t;
+
+typedef struct drm_agp_head {
+	agp_kern_info      agp_info;
+	const char         *chipset;
+	drm_agp_mem_t      *memory;
+	unsigned long      mode;
+	int                enabled;
+	int                acquired;
+	unsigned long      base;
+   	int 		   agp_mtrr;
+} drm_agp_head_t;
+
+typedef struct {
+	void       (*free_memory)(agp_memory *);
+	agp_memory *(*allocate_memory)(size_t, u32);
+	int        (*bind_memory)(agp_memory *, off_t);
+	int        (*unbind_memory)(agp_memory *);
+	void       (*enable)(u32);
+	int        (*acquire)(void);
+	void       (*release)(void);
+	void       (*copy_info)(agp_kern_info *);
+} drm_agp_func_t;
+
+extern drm_agp_func_t drm_agp;
+#endif
+
 typedef struct drm_device {
 	const char	  *name;	/* Simple driver name		   */
 	char		  *unique;	/* Unique identifier: e.g., busid  */
@@ -462,6 +516,12 @@ typedef struct drm_device {
 	struct fasync_struct *buf_async;/* Processes waiting for SIGIO	   */
 	wait_queue_head_t buf_readers;	/* Processes waiting to read	   */
 	wait_queue_head_t buf_writers;	/* Processes waiting to ctx switch */
+	
+#ifdef DRM_AGP
+	drm_agp_head_t    *agp;
+#endif
+	unsigned long     *ctx_bitmap;
+	void		  *dev_private;
 } drm_device_t;
 
 
@@ -533,6 +593,14 @@ extern void	     drm_free_pages(unsigned long address, int order,
 extern void	     *drm_ioremap(unsigned long offset, unsigned long size);
 extern void	     drm_ioremapfree(void *pt, unsigned long size);
 
+#ifdef DRM_AGP
+extern agp_memory    *drm_alloc_agp(int pages, u32 type);
+extern int           drm_free_agp(agp_memory *handle, int pages);
+extern int           drm_bind_agp(agp_memory *handle, unsigned int start);
+extern int           drm_unbind_agp(agp_memory *handle);
+#endif
+
+
 				/* Buffer management support (bufs.c) */
 extern int	     drm_order(unsigned long size);
 extern int	     drm_addmap(struct inode *inode, struct file *filp,
@@ -642,5 +710,32 @@ extern int	     drm_flush_unblock(drm_device_t *dev, int context,
 				       drm_lock_flags_t flags);
 extern int	     drm_flush_block_and_flush(drm_device_t *dev, int context,
 					       drm_lock_flags_t flags);
+
+				/* Context Bitmap support (ctxbitmap.c) */
+extern int	     drm_ctxbitmap_init(drm_device_t *dev);
+extern void	     drm_ctxbitmap_cleanup(drm_device_t *dev);
+extern int	     drm_ctxbitmap_next(drm_device_t *dev);
+extern void	     drm_ctxbitmap_free(drm_device_t *dev, int ctx_handle);
+
+#ifdef DRM_AGP
+				/* AGP/GART support (agpsupport.c) */
+extern drm_agp_head_t *drm_agp_init(void);
+extern int            drm_agp_acquire(struct inode *inode, struct file *filp,
+				      unsigned int cmd, unsigned long arg);
+extern int            drm_agp_release(struct inode *inode, struct file *filp,
+				      unsigned int cmd, unsigned long arg);
+extern int            drm_agp_enable(struct inode *inode, struct file *filp,
+				     unsigned int cmd, unsigned long arg);
+extern int            drm_agp_info(struct inode *inode, struct file *filp,
+				   unsigned int cmd, unsigned long arg);
+extern int            drm_agp_alloc(struct inode *inode, struct file *filp,
+				    unsigned int cmd, unsigned long arg);
+extern int            drm_agp_free(struct inode *inode, struct file *filp,
+				   unsigned int cmd, unsigned long arg);
+extern int            drm_agp_unbind(struct inode *inode, struct file *filp,
+				     unsigned int cmd, unsigned long arg);
+extern int            drm_agp_bind(struct inode *inode, struct file *filp,
+				   unsigned int cmd, unsigned long arg);
+#endif
 #endif
 #endif
diff --git a/linux/gamma_drv.c b/linux/gamma_drv.c
index ce970835..a3c2bd4a 100644
--- a/linux/gamma_drv.c
+++ b/linux/gamma_drv.c
@@ -271,6 +271,10 @@ static int gamma_takedown(drm_device_t *dev)
 					       - PAGE_SHIFT,
 					       DRM_MEM_SAREA);
 				break;
+			case _DRM_AGP:
+				/* Do nothing here, because this is all
+                                   handled in the AGP/GART driver. */
+				break;
 			}
 			drm_free(map, sizeof(*map), DRM_MEM_MAPS);
 		}
diff --git a/linux/i810_bufs.c b/linux/i810_bufs.c
index 49455b43..c025e400 100644
--- a/linux/i810_bufs.c
+++ b/linux/i810_bufs.c
@@ -32,152 +32,28 @@
 
 #define __NO_VERSION__
 #include "drmP.h"
+#include "i810_drv.h"
 #include "linux/un.h"
 
 int i810_addbufs_agp(struct inode *inode, struct file *filp, unsigned int cmd,
 		    unsigned long arg)
 {
-   drm_file_t *priv = filp->private_data;
-   drm_device_t *dev = priv->dev;
-   drm_device_dma_t *dma = dev->dma;
-   drm_buf_desc_t request;
-   drm_buf_entry_t *entry;
-   drm_buf_t *buf;
-   unsigned long offset;
-   unsigned long agp_offset;
-   int count;
-   int order;
-   int size;
-   int alignment;
-   int page_order;
-   int total;
-   int byte_count;
-   int i;
-
-   if (!dma) return -EINVAL;
-
-   copy_from_user_ret(&request,
-		      (drm_buf_desc_t *)arg,
-		      sizeof(request),
-		      -EFAULT);
-
-   count = request.count;
-   order = drm_order(request.size);
-   size	= 1 << order;
-   agp_offset = request.agp_start;
-   alignment  = (request.flags & _DRM_PAGE_ALIGN) ? PAGE_ALIGN(size) :size;
-   page_order = order - PAGE_SHIFT > 0 ? order - PAGE_SHIFT : 0;
-   total = PAGE_SIZE << page_order;
-   byte_count = 0;
-   
-   if (order < DRM_MIN_ORDER || order > DRM_MAX_ORDER) return -EINVAL;
-   if (dev->queue_count) return -EBUSY; /* Not while in use */
-   spin_lock(&dev->count_lock);
-   if (dev->buf_use) {
-      spin_unlock(&dev->count_lock);
-      return -EBUSY;
-   }
-   atomic_inc(&dev->buf_alloc);
-   spin_unlock(&dev->count_lock);
-   
-   down(&dev->struct_sem);
-   entry = &dma->bufs[order];
-   if (entry->buf_count) {
-      up(&dev->struct_sem);
-      atomic_dec(&dev->buf_alloc);
-      return -ENOMEM; /* May only call once for each order */
-   }
-   
-   entry->buflist = drm_alloc(count * sizeof(*entry->buflist),
-			      DRM_MEM_BUFS);
-   if (!entry->buflist) {
-      up(&dev->struct_sem);
-      atomic_dec(&dev->buf_alloc);
-      return -ENOMEM;
-   }
-   memset(entry->buflist, 0, count * sizeof(*entry->buflist));
-   
-   entry->buf_size   = size;
-   entry->page_order = page_order;
-   
-   while(entry->buf_count < count) {
-      for(offset = 0; offset + size <= total && entry->buf_count < count;
-	  offset += alignment, ++entry->buf_count) {
-	 buf = &entry->buflist[entry->buf_count];
-	 buf->idx = dma->buf_count + entry->buf_count;
-	 buf->total = alignment;
-	 buf->order = order;
-	 buf->used = 0;
-	 buf->offset = agp_offset - dev->agp->base + offset;/* ?? */
-	 buf->bus_address = agp_offset + offset;
-	 buf->address = agp_offset + offset + dev->agp->base;
-	 buf->next = NULL;
-	 buf->waiting = 0;
-	 buf->pending = 0;
-	 init_waitqueue_head(&buf->dma_wait);
-	 buf->pid = 0;
-#if DRM_DMA_HISTOGRAM
-	 buf->time_queued = 0;
-	 buf->time_dispatched = 0;
-	 buf->time_completed = 0;
-	 buf->time_freed = 0;
-#endif
-	 DRM_DEBUG("buffer %d @ %p\n",
-		   entry->buf_count, buf->address);
-      }
-      byte_count += PAGE_SIZE << page_order;
-   }
-   
-   dma->buflist = drm_realloc(dma->buflist,
-			      dma->buf_count * sizeof(*dma->buflist),
-			      (dma->buf_count + entry->buf_count)
-			      * sizeof(*dma->buflist),
-			      DRM_MEM_BUFS);
-   for (i = dma->buf_count; i < dma->buf_count + entry->buf_count; i++)
-     dma->buflist[i] = &entry->buflist[i - dma->buf_count];
-   
-   dma->buf_count  += entry->buf_count;
-   dma->byte_count += PAGE_SIZE * (entry->seg_count << page_order);
-   
-   drm_freelist_create(&entry->freelist, entry->buf_count);
-   for (i = 0; i < entry->buf_count; i++) {
-      drm_freelist_put(dev, &entry->freelist, &entry->buflist[i]);
-   }
-   
-   up(&dev->struct_sem);
-   
-   request.count = entry->buf_count;
-   request.size  = size;
-   
-   copy_to_user_ret((drm_buf_desc_t *)arg,
-		    &request,
-		    sizeof(request),
-		    -EFAULT);
-   
-   atomic_dec(&dev->buf_alloc);
-   return 0;
-}
-
-int i810_addbufs_pci(struct inode *inode, struct file *filp, unsigned int cmd,
-		    unsigned long arg)
-{
-   	drm_file_t	 *priv	 = filp->private_data;
-	drm_device_t	 *dev	 = priv->dev;
-	drm_device_dma_t *dma	 = dev->dma;
-	drm_buf_desc_t	 request;
-	int		 count;
-	int		 order;
-	int		 size;
-	int		 total;
-	int		 page_order;
-	drm_buf_entry_t	 *entry;
-	unsigned long	 page;
-	drm_buf_t	 *buf;
-	int		 alignment;
-	unsigned long	 offset;
-	int		 i;
-	int		 byte_count;
-	int		 page_count;
+	drm_file_t *priv = filp->private_data;
+	drm_device_t *dev = priv->dev;
+	drm_device_dma_t *dma = dev->dma;
+	drm_buf_desc_t request;
+	drm_buf_entry_t *entry;
+	drm_buf_t *buf;
+	unsigned long offset;
+	unsigned long agp_offset;
+	int count;
+	int order;
+	int size;
+	int alignment;
+	int page_order;
+	int total;
+	int byte_count;
+	int i;
 
 	if (!dma) return -EINVAL;
 
@@ -186,20 +62,17 @@ int i810_addbufs_pci(struct inode *inode, struct file *filp, unsigned int cmd,
 			   sizeof(request),
 			   -EFAULT);
 
-	count	   = request.count;
-	order	   = drm_order(request.size);
-	size	   = 1 << order;
-	
-	DRM_DEBUG("count = %d, size = %d (%d), order = %d, queue_count = %d\n",
-		  request.count, request.size, size, order, dev->queue_count);
-
-	if (order < DRM_MIN_ORDER || order > DRM_MAX_ORDER) return -EINVAL;
-	if (dev->queue_count) return -EBUSY; /* Not while in use */
-
+	count = request.count;
+	order = drm_order(request.size);
+	size	= 1 << order;
+	agp_offset = request.agp_start;
 	alignment  = (request.flags & _DRM_PAGE_ALIGN) ? PAGE_ALIGN(size) :size;
 	page_order = order - PAGE_SHIFT > 0 ? order - PAGE_SHIFT : 0;
-	total	   = PAGE_SIZE << page_order;
-
+	total = PAGE_SIZE << page_order;
+	byte_count = 0;
+   
+	if (order < DRM_MIN_ORDER || order > DRM_MAX_ORDER) return -EINVAL;
+	if (dev->queue_count) return -EBUSY; /* Not while in use */
 	spin_lock(&dev->count_lock);
 	if (dev->buf_use) {
 		spin_unlock(&dev->count_lock);
@@ -207,15 +80,15 @@ int i810_addbufs_pci(struct inode *inode, struct file *filp, unsigned int cmd,
 	}
 	atomic_inc(&dev->buf_alloc);
 	spin_unlock(&dev->count_lock);
-	
+   
 	down(&dev->struct_sem);
 	entry = &dma->bufs[order];
 	if (entry->buf_count) {
 		up(&dev->struct_sem);
 		atomic_dec(&dev->buf_alloc);
-		return -ENOMEM;	/* May only call once for each order */
+		return -ENOMEM; /* May only call once for each order */
 	}
-	
+   
 	entry->buflist = drm_alloc(count * sizeof(*entry->buflist),
 				   DRM_MEM_BUFS);
 	if (!entry->buflist) {
@@ -224,69 +97,44 @@ int i810_addbufs_pci(struct inode *inode, struct file *filp, unsigned int cmd,
 		return -ENOMEM;
 	}
 	memset(entry->buflist, 0, count * sizeof(*entry->buflist));
-
-	entry->seglist = drm_alloc(count * sizeof(*entry->seglist),
-				   DRM_MEM_SEGS);
-	if (!entry->seglist) {
-		drm_free(entry->buflist,
-			 count * sizeof(*entry->buflist),
-			 DRM_MEM_BUFS);
-		up(&dev->struct_sem);
-		atomic_dec(&dev->buf_alloc);
-		return -ENOMEM;
-	}
-	memset(entry->seglist, 0, count * sizeof(*entry->seglist));
-
-	dma->pagelist = drm_realloc(dma->pagelist,
-				    dma->page_count * sizeof(*dma->pagelist),
-				    (dma->page_count + (count << page_order))
-				    * sizeof(*dma->pagelist),
-				    DRM_MEM_PAGES);
-	DRM_DEBUG("pagelist: %d entries\n",
-		  dma->page_count + (count << page_order));
-
-
-	entry->buf_size	  = size;
+   
+	entry->buf_size   = size;
 	entry->page_order = page_order;
-	byte_count	  = 0;
-	page_count	  = 0;
-	while (entry->buf_count < count) {
-		if (!(page = drm_alloc_pages(page_order, DRM_MEM_DMA))) break;
-		entry->seglist[entry->seg_count++] = page;
-		for (i = 0; i < (1 << page_order); i++) {
-			DRM_DEBUG("page %d @ 0x%08lx\n",
-				  dma->page_count + page_count,
-				  page + PAGE_SIZE * i);
-			dma->pagelist[dma->page_count + page_count++]
-				= page + PAGE_SIZE * i;
-		}
-		for (offset = 0;
-		     offset + size <= total && entry->buf_count < count;
-		     offset += alignment, ++entry->buf_count) {
-			buf	     = &entry->buflist[entry->buf_count];
-			buf->idx     = dma->buf_count + entry->buf_count;
-			buf->total   = alignment;
-			buf->order   = order;
-			buf->used    = 0;
-			buf->offset  = (dma->byte_count + byte_count + offset);
-			buf->address = (void *)(page + offset);
-			buf->next    = NULL;
-			buf->waiting = 0;
-			buf->pending = 0;
-			init_waitqueue_head(&buf->dma_wait);
-			buf->pid     = 0;
+	offset = 0;
+   
+	while(entry->buf_count < count) {
+		buf = &entry->buflist[entry->buf_count];
+		buf->idx = dma->buf_count + entry->buf_count;
+		buf->total = alignment;
+		buf->order = order;
+		buf->used = 0;
+		buf->offset = offset;
+		buf->bus_address = dev->agp->base + agp_offset + offset;
+		buf->address = (void *)(agp_offset + offset + dev->agp->base);
+		buf->next = NULL;
+		buf->waiting = 0;
+		buf->pending = 0;
+		init_waitqueue_head(&buf->dma_wait);
+		buf->pid = 0;
+
+		buf->dev_private = drm_alloc(sizeof(drm_i810_buf_priv_t), 
+					     DRM_MEM_BUFS);
+		buf->dev_priv_size = sizeof(drm_i810_buf_priv_t);
+
 #if DRM_DMA_HISTOGRAM
-			buf->time_queued     = 0;
-			buf->time_dispatched = 0;
-			buf->time_completed  = 0;
-			buf->time_freed	     = 0;
+		buf->time_queued = 0;
+		buf->time_dispatched = 0;
+		buf->time_completed = 0;
+		buf->time_freed = 0;
 #endif
-			DRM_DEBUG("buffer %d @ %p\n",
-				  entry->buf_count, buf->address);
-		}
+		offset = offset + alignment;
+		entry->buf_count++;
 		byte_count += PAGE_SIZE << page_order;
+      
+		DRM_DEBUG("buffer %d @ %p\n",
+			  entry->buf_count, buf->address);
 	}
-
+   
 	dma->buflist = drm_realloc(dma->buflist,
 				   dma->buf_count * sizeof(*dma->buflist),
 				   (dma->buf_count + entry->buf_count)
@@ -294,45 +142,43 @@ int i810_addbufs_pci(struct inode *inode, struct file *filp, unsigned int cmd,
 				   DRM_MEM_BUFS);
 	for (i = dma->buf_count; i < dma->buf_count + entry->buf_count; i++)
 		dma->buflist[i] = &entry->buflist[i - dma->buf_count];
-
-	dma->buf_count	+= entry->buf_count;
-	dma->seg_count	+= entry->seg_count;
-	dma->page_count += entry->seg_count << page_order;
-	dma->byte_count += PAGE_SIZE * (entry->seg_count << page_order);
-	
+   
+	dma->buf_count  += entry->buf_count;
+	dma->byte_count += byte_count;
 	drm_freelist_create(&entry->freelist, entry->buf_count);
 	for (i = 0; i < entry->buf_count; i++) {
 		drm_freelist_put(dev, &entry->freelist, &entry->buflist[i]);
 	}
-	
+   
 	up(&dev->struct_sem);
-
+   
 	request.count = entry->buf_count;
 	request.size  = size;
-
+   
 	copy_to_user_ret((drm_buf_desc_t *)arg,
 			 &request,
 			 sizeof(request),
 			 -EFAULT);
-	
+   
 	atomic_dec(&dev->buf_alloc);
+	dma->flags = _DRM_DMA_USE_AGP;
 	return 0;
 }
 
 int i810_addbufs(struct inode *inode, struct file *filp, unsigned int cmd,
 		unsigned long arg)
 {
-   drm_buf_desc_t	 request;
+	drm_buf_desc_t	 request;
 
-   copy_from_user_ret(&request,
-		      (drm_buf_desc_t *)arg,
-		      sizeof(request),
-		      -EFAULT);
+	copy_from_user_ret(&request,
+			   (drm_buf_desc_t *)arg,
+			   sizeof(request),
+			   -EFAULT);
 
-   if(request.flags & _DRM_AGP_BUFFER)
-     return i810_addbufs_agp(inode, filp, cmd, arg);
-   else
-     return i810_addbufs_pci(inode, filp, cmd, arg);
+	if(request.flags & _DRM_AGP_BUFFER)
+		return i810_addbufs_agp(inode, filp, cmd, arg);
+	else
+		return -EINVAL;
 }
 
 int i810_infobufs(struct inode *inode, struct file *filp, unsigned int cmd,
@@ -487,7 +333,7 @@ int i810_freebufs(struct inode *inode, struct file *filp, unsigned int cmd,
 }
 
 int i810_mapbufs(struct inode *inode, struct file *filp, unsigned int cmd,
-		unsigned long arg)
+		 unsigned long arg)
 {
 	drm_file_t	 *priv	 = filp->private_data;
 	drm_device_t	 *dev	 = priv->dev;
@@ -506,6 +352,7 @@ int i810_mapbufs(struct inode *inode, struct file *filp, unsigned int cmd,
 	spin_lock(&dev->count_lock);
 	if (atomic_read(&dev->buf_alloc)) {
 		spin_unlock(&dev->count_lock);
+		DRM_DEBUG("Busy\n");
 		return -EBUSY;
 	}
 	++dev->buf_use;		/* Can't allocate more after this call */
@@ -515,34 +362,46 @@ int i810_mapbufs(struct inode *inode, struct file *filp, unsigned int cmd,
 			   (drm_buf_map_t *)arg,
 			   sizeof(request),
 			   -EFAULT);
-
-	if (request.count >= dma->buf_count) {
-	   if(dma->flags & _DRM_DMA_USE_AGP) {
-	      /* This is an ugly vicious hack */
-	      drm_map_t *map = NULL;
-	      for(i = 0; i < dev->map_count; i++) {
-		 map = dev->maplist[i];
-		 if(map->type == _DRM_AGP) break;
-	      }
-	      if (i >= dev->map_count || !map) {
-		 retcode = -EINVAL;
-		 goto done;
-	      }
+   	DRM_DEBUG("dma->flags : %lx\n", dma->flags);
+   	if (request.count >= dma->buf_count) {
+		if(dma->flags & _DRM_DMA_USE_AGP) {
+			drm_i810_private_t *dev_priv = 
+				(drm_i810_private_t *)dev->dev_private;
+			drm_map_t *map = NULL;
 	      
-	      virtual = do_mmap(filp, 0, map->size, PROT_READ|PROT_WRITE,
-				MAP_SHARED, (unsigned long)map->handle);
-	   }
-	   else {
-	      virtual = do_mmap(filp, 0, dma->byte_count,
-				PROT_READ|PROT_WRITE, MAP_SHARED, 0);
-	   }
+			map = dev->maplist[dev_priv->buffer_map_idx];
+			if (!map) {
+				DRM_DEBUG("map is null\n");
+				retcode = -EINVAL;
+				goto done;
+			}
+			DRM_DEBUG("map->offset : %lx\n", map->offset);
+			DRM_DEBUG("map->size : %lx\n", map->size);
+			DRM_DEBUG("map->type : %d\n", map->type);
+			DRM_DEBUG("map->flags : %x\n", map->flags);
+			DRM_DEBUG("map->handle : %lx\n", map->handle);
+			DRM_DEBUG("map->mtrr : %d\n", map->mtrr);
+			down(&current->mm->mmap_sem);
+			virtual = do_mmap(filp, 0, map->size, 
+					  PROT_READ|PROT_WRITE,
+					  MAP_SHARED, 
+					  (unsigned long)map->offset);
+
+			up(&current->mm->mmap_sem);
+		} else {
+			down(&current->mm->mmap_sem);
+			virtual = do_mmap(filp, 0, dma->byte_count,
+					  PROT_READ|PROT_WRITE, MAP_SHARED, 0);
+			up(&current->mm->mmap_sem);
+		}
 		if (virtual > -1024UL) {
-				/* Real error */
+			/* Real error */
+			DRM_DEBUG("mmap error\n");
 			retcode = (signed long)virtual;
 			goto done;
 		}
 		request.virtual = (void *)virtual;
-
+	   
 		for (i = 0; i < dma->buf_count; i++) {
 			if (copy_to_user(&request.list[i].idx,
 					 &dma->buflist[i]->idx,
@@ -571,14 +430,15 @@ int i810_mapbufs(struct inode *inode, struct file *filp, unsigned int cmd,
 			}
 		}
 	}
-done:
+ done:
 	request.count = dma->buf_count;
 	DRM_DEBUG("%d buffers, retcode = %d\n", request.count, retcode);
-
+   
 	copy_to_user_ret((drm_buf_map_t *)arg,
 			 &request,
 			 sizeof(request),
 			 -EFAULT);
 
+	DRM_DEBUG("retcode : %d\n", retcode);
 	return retcode;
 }
diff --git a/linux/i810_context.c b/linux/i810_context.c
new file mode 100644
index 00000000..5503edf2
--- /dev/null
+++ b/linux/i810_context.c
@@ -0,0 +1,205 @@
+/* i810_context.c -- IOCTLs for i810 contexts -*- linux-c -*-
+ * Created: Mon Dec 13 09:51:35 1999 by faith@precisioninsight.com
+ *
+ * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ * 
+ * Author: Rickard E. (Rik) Faith <faith@precisioninsight.com>
+ *
+ * $XFree86$
+ *
+ */
+
+#include <linux/sched.h>
+
+#define __NO_VERSION__
+#include "drmP.h"
+#include "i810_drv.h"
+
+static int i810_alloc_queue(drm_device_t *dev)
+{
+   	int temp = drm_ctxbitmap_next(dev);
+   	DRM_DEBUG("i810_alloc_queue: %d\n", temp);
+	return temp;
+}
+
+int i810_context_switch(drm_device_t *dev, int old, int new)
+{
+        char        buf[64];
+
+        atomic_inc(&dev->total_ctx);
+
+        if (test_and_set_bit(0, &dev->context_flag)) {
+                DRM_ERROR("Reentering -- FIXME\n");
+                return -EBUSY;
+        }
+
+#if DRM_DMA_HISTOGRAM
+        dev->ctx_start = get_cycles();
+#endif
+        
+        DRM_DEBUG("Context switch from %d to %d\n", old, new);
+
+        if (new == dev->last_context) {
+                clear_bit(0, &dev->context_flag);
+                return 0;
+        }
+        
+        if (drm_flags & DRM_FLAG_NOCTX) {
+                i810_context_switch_complete(dev, new);
+        } else {
+                sprintf(buf, "C %d %d\n", old, new);
+                drm_write_string(dev, buf);
+        }
+        
+        return 0;
+}
+
+int i810_context_switch_complete(drm_device_t *dev, int new)
+{
+        dev->last_context = new;  /* PRE/POST: This is the _only_ writer. */
+        dev->last_switch  = jiffies;
+        
+        if (!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+                DRM_ERROR("Lock isn't held after context switch\n");
+        }
+
+				/* If a context switch is ever initiated
+                                   when the kernel holds the lock, release
+                                   that lock here. */
+#if DRM_DMA_HISTOGRAM
+        atomic_inc(&dev->histo.ctx[drm_histogram_slot(get_cycles()
+                                                      - dev->ctx_start)]);
+                   
+#endif
+        clear_bit(0, &dev->context_flag);
+        wake_up(&dev->context_wait);
+        
+        return 0;
+}
+
+int i810_resctx(struct inode *inode, struct file *filp, unsigned int cmd,
+	       unsigned long arg)
+{
+	drm_ctx_res_t	res;
+	drm_ctx_t	ctx;
+	int		i;
+
+	DRM_DEBUG("%d\n", DRM_RESERVED_CONTEXTS);
+	copy_from_user_ret(&res, (drm_ctx_res_t *)arg, sizeof(res), -EFAULT);
+	if (res.count >= DRM_RESERVED_CONTEXTS) {
+		memset(&ctx, 0, sizeof(ctx));
+		for (i = 0; i < DRM_RESERVED_CONTEXTS; i++) {
+			ctx.handle = i;
+			copy_to_user_ret(&res.contexts[i],
+					 &i,
+					 sizeof(i),
+					 -EFAULT);
+		}
+	}
+	res.count = DRM_RESERVED_CONTEXTS;
+	copy_to_user_ret((drm_ctx_res_t *)arg, &res, sizeof(res), -EFAULT);
+	return 0;
+}
+
+int i810_addctx(struct inode *inode, struct file *filp, unsigned int cmd,
+	       unsigned long arg)
+{
+	drm_file_t	*priv	= filp->private_data;
+	drm_device_t	*dev	= priv->dev;
+	drm_ctx_t	ctx;
+
+	copy_from_user_ret(&ctx, (drm_ctx_t *)arg, sizeof(ctx), -EFAULT);
+	if ((ctx.handle = i810_alloc_queue(dev)) == DRM_KERNEL_CONTEXT) {
+				/* Skip kernel's context and get a new one. */
+		ctx.handle = i810_alloc_queue(dev);
+	}
+        if (ctx.handle == -1) {
+		DRM_DEBUG("Not enough free contexts.\n");
+				/* Should this return -EBUSY instead? */
+		return -ENOMEM;
+	}
+	DRM_DEBUG("%d\n", ctx.handle);
+	copy_to_user_ret((drm_ctx_t *)arg, &ctx, sizeof(ctx), -EFAULT);
+	return 0;
+}
+
+int i810_modctx(struct inode *inode, struct file *filp, unsigned int cmd,
+	unsigned long arg)
+{
+   	/* This does nothing for the i810 */
+	return 0;
+}
+
+int i810_getctx(struct inode *inode, struct file *filp, unsigned int cmd,
+	unsigned long arg)
+{
+	drm_ctx_t ctx;
+
+	copy_from_user_ret(&ctx, (drm_ctx_t*)arg, sizeof(ctx), -EFAULT);
+	/* This is 0, because we don't hanlde any context flags */
+	ctx.flags = 0;
+	copy_to_user_ret((drm_ctx_t*)arg, &ctx, sizeof(ctx), -EFAULT);
+	return 0;
+}
+
+int i810_switchctx(struct inode *inode, struct file *filp, unsigned int cmd,
+		   unsigned long arg)
+{
+	drm_file_t	*priv	= filp->private_data;
+	drm_device_t	*dev	= priv->dev;
+	drm_ctx_t	ctx;
+
+	copy_from_user_ret(&ctx, (drm_ctx_t *)arg, sizeof(ctx), -EFAULT);
+	DRM_DEBUG("%d\n", ctx.handle);
+	return i810_context_switch(dev, dev->last_context, ctx.handle);
+}
+
+int i810_newctx(struct inode *inode, struct file *filp, unsigned int cmd,
+		unsigned long arg)
+{
+	drm_file_t	*priv	= filp->private_data;
+	drm_device_t	*dev	= priv->dev;
+	drm_ctx_t	ctx;
+
+	copy_from_user_ret(&ctx, (drm_ctx_t *)arg, sizeof(ctx), -EFAULT);
+	DRM_DEBUG("%d\n", ctx.handle);
+	i810_context_switch_complete(dev, ctx.handle);
+
+	return 0;
+}
+
+int i810_rmctx(struct inode *inode, struct file *filp, unsigned int cmd,
+	      unsigned long arg)
+{
+	drm_file_t	*priv	= filp->private_data;
+	drm_device_t	*dev	= priv->dev;
+	drm_ctx_t	ctx;
+
+	copy_from_user_ret(&ctx, (drm_ctx_t *)arg, sizeof(ctx), -EFAULT);
+	DRM_DEBUG("%d\n", ctx.handle);
+   	if(ctx.handle != DRM_KERNEL_CONTEXT) {
+	   	drm_ctxbitmap_free(dev, ctx.handle);
+	}
+	
+	return 0;
+}
diff --git a/linux/i810_dma.c b/linux/i810_dma.c
index 09959b65..30fda5b8 100644
--- a/linux/i810_dma.c
+++ b/linux/i810_dma.c
@@ -25,8 +25,9 @@
  *
  * Authors: Rickard E. (Rik) Faith <faith@precisioninsight.com>
  *	    Jeff Hartmann <jhartmann@precisioninsight.com>
+ *          Keith Whitwell <keithw@precisioninsight.com>
  *
- * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/i810_dma.c,v 1.1 2000/02/11 17:26:04 dawes Exp $
+ * $XFree86$
  *
  */
 
@@ -36,6 +37,9 @@
 
 #include <linux/interrupt.h>	/* For task queue support */
 
+#define I810_BUF_FREE		1
+#define I810_BUF_USED		0
+
 #define I810_REG(reg)		2
 #define I810_BASE(reg)		((unsigned long) \
 				dev->maplist[I810_REG(reg)]->handle)
@@ -43,544 +47,457 @@
 #define I810_DEREF(reg)		*(__volatile__ int *)I810_ADDR(reg)
 #define I810_READ(reg)		I810_DEREF(reg)
 #define I810_WRITE(reg,val) 	do { I810_DEREF(reg) = val; } while (0)
-
-void i810_dma_init(drm_device_t *dev)
+#define I810_DEREF16(reg)	*(__volatile__ u16 *)I810_ADDR(reg)
+#define I810_READ16(reg)	I810_DEREF16(reg)
+#define I810_WRITE16(reg,val)	do { I810_DEREF16(reg) = val; } while (0)
+
+#define RING_LOCALS	unsigned int outring, ringmask; volatile char *virt;
+
+#define BEGIN_LP_RING(n) do {				\
+	if (I810_VERBOSE)				\
+		DRM_DEBUG("BEGIN_LP_RING(%d) in %s\n",	\
+			  n, __FUNCTION__);		\
+	if (dev_priv->ring.space < n*4) 		\
+		i810_wait_ring(dev, n*4);		\
+	dev_priv->ring.space -= n*4;			\
+	outring = dev_priv->ring.tail;			\
+	ringmask = dev_priv->ring.tail_mask;		\
+	virt = dev_priv->ring.virtual_start;		\
+} while (0)
+
+#define ADVANCE_LP_RING() do {					\
+	if (I810_VERBOSE) DRM_DEBUG("ADVANCE_LP_RING\n");	\
+	dev_priv->ring.tail = outring;				\
+	I810_WRITE(LP_RING + RING_TAIL, outring);		\
+} while(0)
+
+#define OUT_RING(n) do {						\
+	if (I810_VERBOSE) DRM_DEBUG("   OUT_RING %x\n", (int)(n));	\
+	*(volatile unsigned int *)(virt + outring) = n;			\
+	outring += 4;							\
+	outring &= ringmask;						\
+} while (0);
+
+static inline void i810_print_status_page(drm_device_t *dev)
 {
-        printk(KERN_INFO "i810_dma_init\n");
+   	drm_device_dma_t *dma = dev->dma;
+      	drm_i810_private_t *dev_priv = dev->dev_private;
+	u32 *temp = (u32 *)dev_priv->hw_status_page;
+   	int i;
+
+   	DRM_DEBUG(  "hw_status: Interrupt Status : %x\n", temp[0]);
+   	DRM_DEBUG(  "hw_status: LpRing Head ptr : %x\n", temp[1]);
+   	DRM_DEBUG(  "hw_status: IRing Head ptr : %x\n", temp[2]);
+      	DRM_DEBUG(  "hw_status: Reserved : %x\n", temp[3]);
+   	DRM_DEBUG(  "hw_status: Driver Counter : %d\n", temp[5]);
+   	for(i = 6; i < dma->buf_count + 6; i++) {
+	   	DRM_DEBUG(  "buffer status idx : %d used: %d\n", i - 6, temp[i]);
+	}
 }
 
-void i810_dma_cleanup(drm_device_t *dev)
+static drm_buf_t *i810_freelist_get(drm_device_t *dev)
 {
-   printk(KERN_INFO "i810_dma_cleanup\n");
+   	drm_device_dma_t *dma = dev->dma;
+	int		 i;
+   	int 		 used;
+   
+	/* Linear search might not be the best solution */
+
+   	for (i = 0; i < dma->buf_count; i++) {
+	   	drm_buf_t *buf = dma->buflist[ i ];
+	   	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+		/* In use is already a pointer */
+	   	used = cmpxchg(buf_priv->in_use, I810_BUF_FREE, 
+			       I810_BUF_USED);
+	   	if(used == I810_BUF_FREE) {
+			return buf;
+		}
+	}
+   	return NULL;
 }
 
-static inline void i810_dma_dispatch(drm_device_t *dev, unsigned long address,
-				    unsigned long length)
+/* This should only be called if the buffer is not sent to the hardware
+ * yet, the hardware updates in use for us once its on the ring buffer.
+ */
+
+static int i810_freelist_put(drm_device_t *dev, drm_buf_t *buf)
 {
-   printk(KERN_INFO "i810_dma_dispatch\n");
+   	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+   	int used;
+   
+   	/* In use is already a pointer */
+   	used = cmpxchg(buf_priv->in_use, I810_BUF_USED, I810_BUF_FREE);
+   	if(used != I810_BUF_USED) {
+	   	DRM_ERROR("Freeing buffer thats not in use : %d\n", buf->idx);
+	   	return -EINVAL;
+	}
+   
+   	return 0;
 }
 
-static inline void i810_dma_quiescent(drm_device_t *dev)
+static int i810_dma_get_buffers(drm_device_t *dev, drm_dma_t *d)
 {
+	int		  i;
+	drm_buf_t	  *buf;
+
+	for (i = d->granted_count; i < d->request_count; i++) {
+		buf = i810_freelist_get(dev);
+		if (!buf) break;
+		buf->pid     = current->pid;
+		copy_to_user_ret(&d->request_indices[i],
+				 &buf->idx,
+				 sizeof(buf->idx),
+				 -EFAULT);
+		copy_to_user_ret(&d->request_sizes[i],
+				 &buf->total,
+				 sizeof(buf->total),
+				 -EFAULT);
+		++d->granted_count;
+	}
+	return 0;
 }
 
-static inline void i810_dma_ready(drm_device_t *dev)
+static unsigned long i810_alloc_page(drm_device_t *dev)
 {
-   i810_dma_quiescent(dev);
-   printk(KERN_INFO "i810_dma_ready\n");
+	unsigned long address;
+   
+	address = __get_free_page(GFP_KERNEL);
+	if(address == 0UL) 
+		return 0;
+	
+	atomic_inc(&mem_map[MAP_NR((void *) address)].count);
+	set_bit(PG_locked, &mem_map[MAP_NR((void *) address)].flags);
+   
+	return address;
 }
 
-static inline int i810_dma_is_ready(drm_device_t *dev)
+static void i810_free_page(drm_device_t *dev, unsigned long page)
 {
-
-   i810_dma_quiescent(dev);
-
-   printk(KERN_INFO "i810_dma_is_ready\n");
-   return 1;
+	if(page == 0UL) 
+		return;
+	
+	atomic_dec(&mem_map[MAP_NR((void *) page)].count);
+	clear_bit(PG_locked, &mem_map[MAP_NR((void *) page)].flags);
+	wake_up(&mem_map[MAP_NR((void *) page)].wait);
+	free_page(page);
+	return;
 }
 
-
-static void i810_dma_service(int irq, void *device, struct pt_regs *regs)
+static int i810_dma_cleanup(drm_device_t *dev)
 {
-	drm_device_t	 *dev = (drm_device_t *)device;
-	drm_device_dma_t *dma = dev->dma;
-	
-	atomic_inc(&dev->total_irq);
-	if (i810_dma_is_ready(dev)) {
-				/* Free previous buffer */
-		if (test_and_set_bit(0, &dev->dma_flag)) {
-			atomic_inc(&dma->total_missed_free);
-			return;
+	if(dev->dev_private) {
+	   	drm_i810_private_t *dev_priv = 
+	     		(drm_i810_private_t *) dev->dev_private;
+	   
+	   	if(dev_priv->ring.virtual_start) {
+		   	drm_ioremapfree((void *) dev_priv->ring.virtual_start,
+					dev_priv->ring.Size);
 		}
-		if (dma->this_buffer) {
-			drm_free_buffer(dev, dma->this_buffer);
-			dma->this_buffer = NULL;
+	   	if(dev_priv->hw_status_page != 0UL) {
+		   	i810_free_page(dev, dev_priv->hw_status_page);
+		   	/* Need to rewrite hardware status page */
+		   	I810_WRITE(0x02080, 0x1ffff000);
 		}
-		clear_bit(0, &dev->dma_flag);
-
-				/* Dispatch new buffer */
-		queue_task(&dev->tq, &tq_immediate);
-		mark_bh(IMMEDIATE_BH);
+	   	drm_free(dev->dev_private, sizeof(drm_i810_private_t), 
+			 DRM_MEM_DRIVER);
+	   	dev->dev_private = NULL;
 	}
+   	return 0;
 }
 
-/* Only called by i810_dma_schedule. */
-static int i810_do_dma(drm_device_t *dev, int locked)
+static int i810_wait_ring(drm_device_t *dev, int n)
 {
-	unsigned long	 address;
-	unsigned long	 length;
-	drm_buf_t	 *buf;
-	int		 retcode = 0;
-	drm_device_dma_t *dma = dev->dma;
-#if DRM_DMA_HISTOGRAM
-	cycles_t	 dma_start, dma_stop;
-#endif
-
-	if (test_and_set_bit(0, &dev->dma_flag)) {
-		atomic_inc(&dma->total_missed_dma);
-		return -EBUSY;
-	}
-	
-#if DRM_DMA_HISTOGRAM
-	dma_start = get_cycles();
-#endif
-
-	if (!dma->next_buffer) {
-		DRM_ERROR("No next_buffer\n");
-		clear_bit(0, &dev->dma_flag);
-		return -EINVAL;
-	}
-
-	buf	= dma->next_buffer;
-	address = (unsigned long)buf->bus_address;
-	length	= buf->used;
+   	drm_i810_private_t *dev_priv = dev->dev_private;
+   	drm_i810_ring_buffer_t *ring = &(dev_priv->ring);
+   	int iters = 0;
+   	unsigned long end;
+
+	end = jiffies + (HZ*3);
+   	while (ring->space < n) {
+	   	int i;
 	
-
-	DRM_DEBUG("context %d, buffer %d (%ld bytes)\n",
-		  buf->context, buf->idx, length);
-
-	if (buf->list == DRM_LIST_RECLAIM) {
-		drm_clear_next_buffer(dev);
-		drm_free_buffer(dev, buf);
-		clear_bit(0, &dev->dma_flag);
-		return -EINVAL;
-	}
-
-	if (!length) {
-		DRM_ERROR("0 length buffer\n");
-		drm_clear_next_buffer(dev);
-		drm_free_buffer(dev, buf);
-		clear_bit(0, &dev->dma_flag);
-		return 0;
-	}
-	
-	if (!i810_dma_is_ready(dev)) {
-		clear_bit(0, &dev->dma_flag);
-		return -EBUSY;
-	}
-
-	if (buf->while_locked) {
-		if (!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
-			DRM_ERROR("Dispatching buffer %d from pid %d"
-				  " \"while locked\", but no lock held\n",
-				  buf->idx, buf->pid);
-		}
-	} else {
-		if (!locked && !drm_lock_take(&dev->lock.hw_lock->lock,
-					      DRM_KERNEL_CONTEXT)) {
-			atomic_inc(&dma->total_missed_lock);
-			clear_bit(0, &dev->dma_flag);
-			return -EBUSY;
+	   	ring->head = I810_READ(LP_RING + RING_HEAD) & HEAD_ADDR;
+	   	ring->space = ring->head - (ring->tail+8);
+	   
+	   	if (ring->space < 0) ring->space += ring->Size;
+	   
+	   	iters++;
+		if((signed)(end - jiffies) <= 0) {
+		   	DRM_ERROR("space: %d wanted %d\n", ring->space, n);
+		   	DRM_ERROR("lockup\n");
+		   	goto out_wait_ring;
 		}
-	}
-
-	if (dev->last_context != buf->context
-	    && !(dev->queuelist[buf->context]->flags
-		 & _DRM_CONTEXT_PRESERVED)) {
-				/* PRE: dev->last_context != buf->context */
-		if (drm_context_switch(dev, dev->last_context, buf->context)) {
-			drm_clear_next_buffer(dev);
-			drm_free_buffer(dev, buf);
-		}
-		retcode = -EBUSY;
-		goto cleanup;
-			
-				/* POST: we will wait for the context
-				   switch and will dispatch on a later call
-				   when dev->last_context == buf->context.
-				   NOTE WE HOLD THE LOCK THROUGHOUT THIS
-				   TIME! */
-	}
-
-	drm_clear_next_buffer(dev);
-	buf->pending	 = 1;
-	buf->waiting	 = 0;
-	buf->list	 = DRM_LIST_PEND;
-#if DRM_DMA_HISTOGRAM
-	buf->time_dispatched = get_cycles();
-#endif
 
-	i810_dma_dispatch(dev, address, length);
-	drm_free_buffer(dev, dma->this_buffer);
-	dma->this_buffer = buf;
-
-	atomic_add(length, &dma->total_bytes);
-	atomic_inc(&dma->total_dmas);
-
-	if (!buf->while_locked && !dev->context_flag && !locked) {
-		if (drm_lock_free(dev, &dev->lock.hw_lock->lock,
-				  DRM_KERNEL_CONTEXT)) {
-			DRM_ERROR("\n");
-		}
+	   	for (i = 0 ; i < 2000 ; i++) ;
 	}
-cleanup:
-
-	clear_bit(0, &dev->dma_flag);
 
-#if DRM_DMA_HISTOGRAM
-	dma_stop = get_cycles();
-	atomic_inc(&dev->histo.dma[drm_histogram_slot(dma_stop - dma_start)]);
-#endif
-
-	return retcode;
-}
-
-static void i810_dma_schedule_timer_wrapper(unsigned long dev)
-{
-	i810_dma_schedule((drm_device_t *)dev, 0);
+out_wait_ring:   
+   	return iters;
 }
 
-static void i810_dma_schedule_tq_wrapper(void *dev)
+static void i810_kernel_lost_context(drm_device_t *dev)
 {
-	i810_dma_schedule(dev, 0);
+      	drm_i810_private_t *dev_priv = dev->dev_private;
+   	drm_i810_ring_buffer_t *ring = &(dev_priv->ring);
+      
+   	ring->head = I810_READ(LP_RING + RING_HEAD) & HEAD_ADDR;
+     	ring->tail = I810_READ(LP_RING + RING_TAIL);
+     	ring->space = ring->head - (ring->tail+8);
+     	if (ring->space < 0) ring->space += ring->Size;
 }
 
-int i810_dma_schedule(drm_device_t *dev, int locked)
+static int i810_freelist_init(drm_device_t *dev)
 {
-	int		 next;
-	drm_queue_t	 *q;
-	drm_buf_t	 *buf;
-	int		 retcode   = 0;
-	int		 processed = 0;
-	int		 missed;
-	int		 expire	   = 20;
-	drm_device_dma_t *dma	   = dev->dma;
-#if DRM_DMA_HISTOGRAM
-	cycles_t	 schedule_start;
-#endif
-
-	if (test_and_set_bit(0, &dev->interrupt_flag)) {
-				/* Not reentrant */
-		atomic_inc(&dma->total_missed_sched);
-		return -EBUSY;
-	}
-	missed = atomic_read(&dma->total_missed_sched);
-
-#if DRM_DMA_HISTOGRAM
-	schedule_start = get_cycles();
-#endif
-
-again:
-	if (dev->context_flag) {
-		clear_bit(0, &dev->interrupt_flag);
-		return -EBUSY;
-	}
-	if (dma->next_buffer) {
-				/* Unsent buffer that was previously
-				   selected, but that couldn't be sent
-				   because the lock could not be obtained
-				   or the DMA engine wasn't ready.  Try
-				   again. */
-		atomic_inc(&dma->total_tried);
-		if (!(retcode = i810_do_dma(dev, locked))) {
-			atomic_inc(&dma->total_hit);
-			++processed;
-		}
-	} else {
-		do {
-			next = drm_select_queue(dev,
-					     i810_dma_schedule_timer_wrapper);
-			if (next >= 0) {
-				q   = dev->queuelist[next];
-				buf = drm_waitlist_get(&q->waitlist);
-				dma->next_buffer = buf;
-				dma->next_queue	 = q;
-				if (buf && buf->list == DRM_LIST_RECLAIM) {
-					drm_clear_next_buffer(dev);
-					drm_free_buffer(dev, buf);
-				}
-			}
-		} while (next >= 0 && !dma->next_buffer);
-		if (dma->next_buffer) {
-			if (!(retcode = i810_do_dma(dev, locked))) {
-				++processed;
-			}
-		}
+      	drm_device_dma_t *dma = dev->dma;
+   	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
+   	u8 *hw_status = (u8 *)dev_priv->hw_status_page;
+   	int i;
+   	int my_idx = 24;
+   
+   	if(dma->buf_count > 1019) {
+	   	/* Not enough space in the status page for the freelist */
+	   	return -EINVAL;
 	}
 
-	if (--expire) {
-		if (missed != atomic_read(&dma->total_missed_sched)) {
-			atomic_inc(&dma->total_lost);
-			if (i810_dma_is_ready(dev)) goto again;
-		}
-		if (processed && i810_dma_is_ready(dev)) {
-			atomic_inc(&dma->total_lost);
-			processed = 0;
-			goto again;
-		}
+   	for (i = 0; i < dma->buf_count; i++) {
+	   	drm_buf_t *buf = dma->buflist[ i ];
+	   	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+	   
+	   	buf_priv->in_use = hw_status + my_idx;
+	   	DRM_DEBUG("buf_priv->in_use : %p\n", buf_priv->in_use);
+	   	*buf_priv->in_use = I810_BUF_FREE;
+	   	buf_priv->my_use_idx = my_idx;
+	   	my_idx += 4;
 	}
-	
-	clear_bit(0, &dev->interrupt_flag);
-	
-#if DRM_DMA_HISTOGRAM
-	atomic_inc(&dev->histo.schedule[drm_histogram_slot(get_cycles()
-							   - schedule_start)]);
-#endif
-	return retcode;
+	return 0;
 }
 
-static int i810_dma_priority(drm_device_t *dev, drm_dma_t *d)
+static int i810_dma_initialize(drm_device_t *dev, 
+			       drm_i810_private_t *dev_priv,
+			       drm_i810_init_t *init)
 {
-	unsigned long	  address;
-	unsigned long	  length;
-	int		  must_free = 0;
-	int		  retcode   = 0;
-	int		  i;
-	int		  idx;
-	drm_buf_t	  *buf;
-	drm_buf_t	  *last_buf = NULL;
-	drm_device_dma_t  *dma	    = dev->dma;
-	DECLARE_WAITQUEUE(entry, current);
-
-				/* Turn off interrupt handling */
-	while (test_and_set_bit(0, &dev->interrupt_flag)) {
-		schedule();
-		if (signal_pending(current)) return -EINTR;
-	}
-	if (!(d->flags & _DRM_DMA_WHILE_LOCKED)) {
-		while (!drm_lock_take(&dev->lock.hw_lock->lock,
-				      DRM_KERNEL_CONTEXT)) {
-			schedule();
-			if (signal_pending(current)) {
-				clear_bit(0, &dev->interrupt_flag);
-				return -EINTR;
-			}
-		}
-		++must_free;
-	}
-	atomic_inc(&dma->total_prio);
-
-	for (i = 0; i < d->send_count; i++) {
-		idx = d->send_indices[i];
-		if (idx < 0 || idx >= dma->buf_count) {
-			DRM_ERROR("Index %d (of %d max)\n",
-				  d->send_indices[i], dma->buf_count - 1);
-			continue;
-		}
-		buf = dma->buflist[ idx ];
-		if (buf->pid != current->pid) {
-			DRM_ERROR("Process %d using buffer owned by %d\n",
-				  current->pid, buf->pid);
-			retcode = -EINVAL;
-			goto cleanup;
-		}
-		if (buf->list != DRM_LIST_NONE) {
-			DRM_ERROR("Process %d using %d's buffer on list %d\n",
-				  current->pid, buf->pid, buf->list);
-			retcode = -EINVAL;
-			goto cleanup;
-		}
-				/* This isn't a race condition on
-				   buf->list, since our concern is the
-				   buffer reclaim during the time the
-				   process closes the /dev/drm? handle, so
-				   it can't also be doing DMA. */
-		buf->list	  = DRM_LIST_PRIO;
-		buf->used	  = d->send_sizes[i];
-		buf->context	  = d->context;
-		buf->while_locked = d->flags & _DRM_DMA_WHILE_LOCKED;
-		address		  = (unsigned long)buf->address;
-		length		  = buf->used;
-		if (!length) {
-			DRM_ERROR("0 length buffer\n");
-		}
-		if (buf->pending) {
-			DRM_ERROR("Sending pending buffer:"
-				  " buffer %d, offset %d\n",
-				  d->send_indices[i], i);
-			retcode = -EINVAL;
-			goto cleanup;
-		}
-		if (buf->waiting) {
-			DRM_ERROR("Sending waiting buffer:"
-				  " buffer %d, offset %d\n",
-				  d->send_indices[i], i);
-			retcode = -EINVAL;
-			goto cleanup;
-		}
-		buf->pending = 1;
-		
-		if (dev->last_context != buf->context
-		    && !(dev->queuelist[buf->context]->flags
-			 & _DRM_CONTEXT_PRESERVED)) {
-			add_wait_queue(&dev->context_wait, &entry);
-			current->state = TASK_INTERRUPTIBLE;
-				/* PRE: dev->last_context != buf->context */
-			drm_context_switch(dev, dev->last_context,
-					   buf->context);
-				/* POST: we will wait for the context
-				   switch and will dispatch on a later call
-				   when dev->last_context == buf->context.
-				   NOTE WE HOLD THE LOCK THROUGHOUT THIS
-				   TIME! */
-			schedule();
-			current->state = TASK_RUNNING;
-			remove_wait_queue(&dev->context_wait, &entry);
-			if (signal_pending(current)) {
-				retcode = -EINTR;
-				goto cleanup;
-			}
-			if (dev->last_context != buf->context) {
-				DRM_ERROR("Context mismatch: %d %d\n",
-					  dev->last_context,
-					  buf->context);
-			}
-		}
-
-#if DRM_DMA_HISTOGRAM
-		buf->time_queued     = get_cycles();
-		buf->time_dispatched = buf->time_queued;
-#endif
-		i810_dma_dispatch(dev, address, length);
-	        if (drm_lock_free(dev, &dev->lock.hw_lock->lock,
-				  DRM_KERNEL_CONTEXT)) {
-		   DRM_ERROR("\n");
-		}
-
-		atomic_add(length, &dma->total_bytes);
-		atomic_inc(&dma->total_dmas);
-		
-		if (last_buf) {
-			drm_free_buffer(dev, last_buf);
-		}
-		last_buf = buf;
-	}
+	drm_map_t *sarea_map;
 
+   	dev->dev_private = (void *) dev_priv;
+   	memset(dev_priv, 0, sizeof(drm_i810_private_t));
 
-cleanup:
-	if (last_buf) {
-		i810_dma_ready(dev);
-		drm_free_buffer(dev, last_buf);
+   	if (init->ring_map_idx >= dev->map_count ||
+	    init->buffer_map_idx >= dev->map_count) {
+	   	i810_dma_cleanup(dev);
+	   	DRM_ERROR("ring_map or buffer_map are invalid\n");
+	   	return -EINVAL;
 	}
-	
-	if (must_free && !dev->context_flag) {
-		if (drm_lock_free(dev, &dev->lock.hw_lock->lock,
-				  DRM_KERNEL_CONTEXT)) {
-			DRM_ERROR("\n");
-		}
+   
+   	dev_priv->ring_map_idx = init->ring_map_idx;
+   	dev_priv->buffer_map_idx = init->buffer_map_idx;
+	sarea_map = dev->maplist[0];
+	dev_priv->sarea_priv = (drm_i810_sarea_t *) 
+		((u8 *)sarea_map->handle + 
+		 init->sarea_priv_offset);
+
+   	atomic_set(&dev_priv->flush_done, 0);
+	init_waitqueue_head(&dev_priv->flush_queue);
+   	
+   	dev_priv->ring.Start = init->ring_start;
+   	dev_priv->ring.End = init->ring_end;
+   	dev_priv->ring.Size = init->ring_size;
+   	dev_priv->ring.virtual_start = drm_ioremap(dev->agp->base + 
+						   init->ring_start, 
+						   init->ring_size);
+   	dev_priv->ring.tail_mask = dev_priv->ring.Size - 1;
+   
+   	if (dev_priv->ring.virtual_start == NULL) {
+	   	i810_dma_cleanup(dev);
+	   	DRM_ERROR("can not ioremap virtual address for"
+			  " ring buffer\n");
+	   	return -ENOMEM;
 	}
-	clear_bit(0, &dev->interrupt_flag);
-	return retcode;
+   
+   	/* Program Hardware Status Page */
+   	dev_priv->hw_status_page = i810_alloc_page(dev);
+   	memset((void *) dev_priv->hw_status_page, 0, PAGE_SIZE);
+   	if(dev_priv->hw_status_page == 0UL) {
+		i810_dma_cleanup(dev);
+		DRM_ERROR("Can not allocate hardware status page\n");
+		return -ENOMEM;
+	}
+   	DRM_DEBUG("hw status page @ %lx\n", dev_priv->hw_status_page);
+   
+   	I810_WRITE(0x02080, virt_to_bus((void *)dev_priv->hw_status_page));
+   	DRM_DEBUG("Enabled hardware status page\n");
+   
+   	/* Now we need to init our freelist */
+   	if(i810_freelist_init(dev) != 0) {
+	   	i810_dma_cleanup(dev);
+	   	DRM_ERROR("Not enough space in the status page for"
+			  " the freelist\n");
+	   	return -ENOMEM;
+	}
+   	return 0;
 }
 
-static int i810_dma_send_buffers(drm_device_t *dev, drm_dma_t *d)
+int i810_dma_init(struct inode *inode, struct file *filp,
+		  unsigned int cmd, unsigned long arg)
 {
-	DECLARE_WAITQUEUE(entry, current);
-	drm_buf_t	  *last_buf = NULL;
-	int		  retcode   = 0;
-	drm_device_dma_t  *dma	    = dev->dma;
-
-	if (d->flags & _DRM_DMA_BLOCK) {
-		last_buf = dma->buflist[d->send_indices[d->send_count-1]];
-		add_wait_queue(&last_buf->dma_wait, &entry);
-	}
+   	drm_file_t *priv = filp->private_data;
+   	drm_device_t *dev = priv->dev;
+   	drm_i810_private_t *dev_priv;
+   	drm_i810_init_t init;
+   	int retcode = 0;
 	
-	if ((retcode = drm_dma_enqueue(dev, d))) {
-		if (d->flags & _DRM_DMA_BLOCK)
-			remove_wait_queue(&last_buf->dma_wait, &entry);
-		return retcode;
-	}
-	
-	i810_dma_schedule(dev, 0);
+   	copy_from_user_ret(&init, (drm_i810_init_t *)arg, 
+			   sizeof(init), -EFAULT);
 	
-	if (d->flags & _DRM_DMA_BLOCK) {
-		DRM_DEBUG("%d waiting\n", current->pid);
-		current->state = TASK_INTERRUPTIBLE;
-		for (;;) {
-			if (!last_buf->waiting
-			    && !last_buf->pending)
-				break; /* finished */
-			schedule();
-			if (signal_pending(current)) {
-				retcode = -EINTR; /* Can't restart */
-				break;
-			}
-		}
-		current->state = TASK_RUNNING;
-		DRM_DEBUG("%d running\n", current->pid);
-		remove_wait_queue(&last_buf->dma_wait, &entry);
-		if (!retcode
-		    || (last_buf->list==DRM_LIST_PEND && !last_buf->pending)) {
-			if (!waitqueue_active(&last_buf->dma_wait)) {
-				drm_free_buffer(dev, last_buf);
-			}
-		}
-		if (retcode) {
-			DRM_ERROR("ctx%d w%d p%d c%d i%d l%d %d/%d\n",
-				  d->context,
-				  last_buf->waiting,
-				  last_buf->pending,
-				  DRM_WAITCOUNT(dev, d->context),
-				  last_buf->idx,
-				  last_buf->list,
-				  last_buf->pid,
-				  current->pid);
-		}
+   	switch(init.func) {
+	 	case I810_INIT_DMA:
+	   		dev_priv = drm_alloc(sizeof(drm_i810_private_t), 
+					     DRM_MEM_DRIVER);
+	   		if(dev_priv == NULL) return -ENOMEM;
+	   		retcode = i810_dma_initialize(dev, dev_priv, &init);
+	   	break;
+	 	case I810_CLEANUP_DMA:
+	   		retcode = i810_dma_cleanup(dev);
+	   	break;
+	 	default:
+	   		retcode = -EINVAL;
+	   	break;
 	}
-	return retcode;
+   
+   	return retcode;
 }
 
-int i810_dma(struct inode *inode, struct file *filp, unsigned int cmd,
-	      unsigned long arg)
+static void i810_dma_dispatch_general(drm_device_t *dev, drm_buf_t *buf,
+				      int used )
 {
-	drm_file_t	  *priv	    = filp->private_data;
-	drm_device_t	  *dev	    = priv->dev;
-	drm_device_dma_t  *dma	    = dev->dma;
-	int		  retcode   = 0;
-	drm_dma_t	  d;
-
-        printk("i810_dma start\n");
-   	copy_from_user_ret(&d, (drm_dma_t *)arg, sizeof(d), -EFAULT);
-	DRM_DEBUG("%d %d: %d send, %d req\n",
-		  current->pid, d.context, d.send_count, d.request_count);
+      	drm_i810_private_t *dev_priv = dev->dev_private;
+   	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+	unsigned long address = (unsigned long)buf->bus_address;
+	unsigned long start = address - dev->agp->base;
+   	RING_LOCALS;
+
+   	dev_priv->counter++;
+   	DRM_DEBUG(  "dispatch counter : %ld\n", dev_priv->counter);
+   	DRM_DEBUG(  "i810_dma_dispatch\n");
+   	DRM_DEBUG(  "start : 0x%lx\n", start);
+	DRM_DEBUG(  "used : 0x%x\n", used);
+   	DRM_DEBUG(  "start + used - 4 : 0x%lx\n", start + used - 4);
+   	i810_kernel_lost_context(dev);
+
+   	BEGIN_LP_RING(10);
+   	OUT_RING( CMD_OP_BATCH_BUFFER );
+   	OUT_RING( start | BB1_PROTECTED );
+   	OUT_RING( start + used - 4 );
+      	OUT_RING( CMD_STORE_DWORD_IDX );
+   	OUT_RING( 20 );
+   	OUT_RING( dev_priv->counter );
+   	OUT_RING( CMD_STORE_DWORD_IDX );
+   	OUT_RING( buf_priv->my_use_idx );
+   	OUT_RING( I810_BUF_FREE );   
+      	OUT_RING( CMD_REPORT_HEAD );
+      	ADVANCE_LP_RING();
+}
 
-	if (d.context == DRM_KERNEL_CONTEXT || d.context >= dev->queue_slots) {
-		DRM_ERROR("Process %d using context %d\n",
-			  current->pid, d.context);
-		return -EINVAL;
-	}
+static void i810_dma_dispatch_vertex(drm_device_t *dev, 
+				     drm_buf_t *buf,
+				     int discard,
+				     int used)
+{
+   	drm_i810_private_t *dev_priv = dev->dev_private;
+	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+   	drm_i810_sarea_t *sarea_priv = dev_priv->sarea_priv;
+   	drm_clip_rect_t *box = sarea_priv->boxes;
+   	int nbox = sarea_priv->nbox;
+	unsigned long address = (unsigned long)buf->bus_address;
+	unsigned long start = address - dev->agp->base;     
+	int i = 0;
+   	RING_LOCALS;
 
-	if (d.send_count < 0 || d.send_count > dma->buf_count) {
-		DRM_ERROR("Process %d trying to send %d buffers (of %d max)\n",
-			  current->pid, d.send_count, dma->buf_count);
-		return -EINVAL;
-	}
-	if (d.request_count < 0 || d.request_count > dma->buf_count) {
-		DRM_ERROR("Process %d trying to get %d buffers (of %d max)\n",
-			  current->pid, d.request_count, dma->buf_count);
-		return -EINVAL;
+   
+   	if (nbox > I810_NR_SAREA_CLIPRECTS) 
+		nbox = I810_NR_SAREA_CLIPRECTS;
+   
+   	DRM_DEBUG("dispatch vertex addr 0x%lx, used 0x%x nbox %d\n", 
+		  address, used, nbox);
+
+   	dev_priv->counter++;
+   	DRM_DEBUG(  "dispatch counter : %ld\n", dev_priv->counter);
+   	DRM_DEBUG(  "i810_dma_dispatch\n");
+   	DRM_DEBUG(  "start : %lx\n", start);
+	DRM_DEBUG(  "used : %d\n", used);
+   	DRM_DEBUG(  "start + used - 4 : %ld\n", start + used - 4);
+   	i810_kernel_lost_context(dev);
+
+	if (used) {
+		do {
+			if (i < nbox) {
+				BEGIN_LP_RING(4);
+				OUT_RING( GFX_OP_SCISSOR | SC_UPDATE_SCISSOR | 
+					  SC_ENABLE );
+				OUT_RING( GFX_OP_SCISSOR_INFO );
+				OUT_RING( box[i].x1 | (box[i].y1 << 16) );
+				OUT_RING( (box[i].x2-1) | ((box[i].y2-1)<<16) );
+				ADVANCE_LP_RING();
+			}
+			
+			BEGIN_LP_RING(4);
+			OUT_RING( CMD_OP_BATCH_BUFFER );
+			OUT_RING( start | BB1_PROTECTED );
+			OUT_RING( start + used - 4 );
+			OUT_RING( 0 );
+			ADVANCE_LP_RING();
+			
+		} while (++i < nbox);
 	}
 
-	if (d.send_count) {
-#if 0
-		if (d.flags & _DRM_DMA_PRIORITY)
-			retcode = i810_dma_priority(dev, &d);
-		else 
-			retcode = i810_dma_send_buffers(dev, &d);
-#endif
-	   printk("i810_dma priority\n");
+	BEGIN_LP_RING(10);
+	OUT_RING( CMD_STORE_DWORD_IDX );
+	OUT_RING( 20 );
+	OUT_RING( dev_priv->counter );
+	OUT_RING( 0 );
 
-	   retcode = i810_dma_priority(dev, &d);
+	if (discard) {
+		OUT_RING( CMD_STORE_DWORD_IDX );
+		OUT_RING( buf_priv->my_use_idx );
+		OUT_RING( I810_BUF_FREE );
+		OUT_RING( 0 );
 	}
 
-	d.granted_count = 0;
+      	OUT_RING( CMD_REPORT_HEAD );
+	OUT_RING( 0 );
+   	ADVANCE_LP_RING();
+}
 
-	if (!retcode && d.request_count) {
-		retcode = drm_dma_get_buffers(dev, &d);
-	}
 
-	DRM_DEBUG("%d returning, granted = %d\n",
-		  current->pid, d.granted_count);
-	copy_to_user_ret((drm_dma_t *)arg, &d, sizeof(d), -EFAULT);
+/* Interrupts are only for flushing */
+static void i810_dma_service(int irq, void *device, struct pt_regs *regs)
+{
+	drm_device_t	 *dev = (drm_device_t *)device;
+   	u16 temp;
+   
+	atomic_inc(&dev->total_irq);
+      	temp = I810_READ16(I810REG_INT_IDENTITY_R);
+   	temp = temp & ~(0x6000);
+   	if(temp != 0) I810_WRITE16(I810REG_INT_IDENTITY_R, 
+				   temp); /* Clear all interrupts */
+   
+   	queue_task(&dev->tq, &tq_immediate);
+   	mark_bh(IMMEDIATE_BH);
+}
 
-   	printk("i810_dma end (granted)\n");
-	return retcode;
+static void i810_dma_task_queue(void *device)
+{
+	drm_device_t *dev = (drm_device_t *) device;
+      	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
+
+   	atomic_set(&dev_priv->flush_done, 1);
+   	wake_up_interruptible(&dev_priv->flush_queue);
 }
 
 int i810_irq_install(drm_device_t *dev, int irq)
 {
 	int retcode;
-
+	u16 temp;
+   
 	if (!irq)     return -EINVAL;
 	
 	down(&dev->struct_sem);
@@ -591,6 +508,7 @@ int i810_irq_install(drm_device_t *dev, int irq)
 	dev->irq = irq;
 	up(&dev->struct_sem);
 	
+   	DRM_DEBUG(  "Interrupt Install : %d\n", irq);
 	DRM_DEBUG("%d\n", irq);
 
 	dev->context_flag     = 0;
@@ -603,12 +521,21 @@ int i810_irq_install(drm_device_t *dev, int irq)
 
 	dev->tq.next	      = NULL;
 	dev->tq.sync	      = 0;
-	dev->tq.routine	      = i810_dma_schedule_tq_wrapper;
+	dev->tq.routine	      = i810_dma_task_queue;
 	dev->tq.data	      = dev;
 
-
 				/* Before installing handler */
-	/* TODO */	
+   	temp = I810_READ16(I810REG_HWSTAM);
+   	temp = temp & 0x6000;
+   	I810_WRITE16(I810REG_HWSTAM, temp);
+   	
+      	temp = I810_READ16(I810REG_INT_MASK_R);
+   	temp = temp & 0x6000;
+   	I810_WRITE16(I810REG_INT_MASK_R, temp); /* Unmask interrupts */
+   	temp = I810_READ16(I810REG_INT_ENABLE_R);
+   	temp = temp & 0x6000;
+      	I810_WRITE16(I810REG_INT_ENABLE_R, temp); /* Disable all interrupts */
+
 				/* Install handler */
 	if ((retcode = request_irq(dev->irq,
 				   i810_dma_service,
@@ -620,15 +547,18 @@ int i810_irq_install(drm_device_t *dev, int irq)
 		up(&dev->struct_sem);
 		return retcode;
 	}
-
-				/* After installing handler */
-	/* TODO */
+   	temp = I810_READ16(I810REG_INT_ENABLE_R);
+   	temp = temp & 0x6000;
+   	temp = temp | 0x0003;
+   	I810_WRITE16(I810REG_INT_ENABLE_R, 
+		     temp); /* Enable bp & user interrupts */
 	return 0;
 }
 
 int i810_irq_uninstall(drm_device_t *dev)
 {
 	int irq;
+   	u16 temp;
 
 	down(&dev->struct_sem);
 	irq	 = dev->irq;
@@ -636,16 +566,25 @@ int i810_irq_uninstall(drm_device_t *dev)
 	up(&dev->struct_sem);
 	
 	if (!irq) return -EINVAL;
-	
+
+   	DRM_DEBUG(  "Interrupt UnInstall: %d\n", irq);	
 	DRM_DEBUG("%d\n", irq);
-	
-	/* TODO : Disable interrupts */
+   
+   	temp = I810_READ16(I810REG_INT_IDENTITY_R);
+   	temp = temp & ~(0x6000);
+   	if(temp != 0) I810_WRITE16(I810REG_INT_IDENTITY_R, 
+				   temp); /* Clear all interrupts */
+   
+   	temp = I810_READ16(I810REG_INT_ENABLE_R);
+   	temp = temp & 0x6000;
+   	I810_WRITE16(I810REG_INT_ENABLE_R, 
+		     temp);                     /* Disable all interrupts */
+
    	free_irq(irq, dev);
 
 	return 0;
 }
 
-
 int i810_control(struct inode *inode, struct file *filp, unsigned int cmd,
 		  unsigned long arg)
 {
@@ -654,8 +593,7 @@ int i810_control(struct inode *inode, struct file *filp, unsigned int cmd,
 	drm_control_t	ctl;
 	int		retcode;
    
-   	printk(KERN_INFO "i810_control\n");
-   	i810_dma_init(dev);
+   	DRM_DEBUG(  "i810_control\n");
 
 	copy_from_user_ret(&ctl, (drm_control_t *)arg, sizeof(ctl), -EFAULT);
 	
@@ -674,20 +612,134 @@ int i810_control(struct inode *inode, struct file *filp, unsigned int cmd,
 	return 0;
 }
 
+static inline void i810_dma_emit_flush(drm_device_t *dev)
+{
+   	drm_i810_private_t *dev_priv = dev->dev_private;
+   	RING_LOCALS;
+
+   	i810_kernel_lost_context(dev);
+   	BEGIN_LP_RING(2);
+      	OUT_RING( CMD_REPORT_HEAD );
+   	OUT_RING( GFX_OP_USER_INTERRUPT );
+      	ADVANCE_LP_RING();
+}
+
+static inline void i810_dma_quiescent_emit(drm_device_t *dev)
+{
+      	drm_i810_private_t *dev_priv = dev->dev_private;
+   	RING_LOCALS;
+
+  	i810_kernel_lost_context(dev);
+   	BEGIN_LP_RING(4);
+
+   	OUT_RING( INST_PARSER_CLIENT | INST_OP_FLUSH | INST_FLUSH_MAP_CACHE );
+   	OUT_RING( CMD_REPORT_HEAD );
+   	OUT_RING( GFX_OP_USER_INTERRUPT );
+   	OUT_RING( 0 );
+   	ADVANCE_LP_RING();
+}
+
+static void i810_dma_quiescent(drm_device_t *dev)
+{
+      	DECLARE_WAITQUEUE(entry, current);
+  	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
+	unsigned long end;      
+
+   	if(dev_priv == NULL) {
+	   	return;
+	}
+      	atomic_set(&dev_priv->flush_done, 0);
+   	current->state = TASK_INTERRUPTIBLE;
+   	add_wait_queue(&dev_priv->flush_queue, &entry);
+   	end = jiffies + (HZ*3);
+   
+   	for (;;) {
+	      	i810_dma_quiescent_emit(dev);
+	   	if (atomic_read(&dev_priv->flush_done) == 1) break;
+		if((signed)(end - jiffies) <= 0) {
+		   	DRM_ERROR("lockup\n");
+		   	break;
+		}	   
+	      	schedule_timeout(HZ*3);
+	      	if (signal_pending(current)) {
+		   	break;
+		}
+	}
+   
+   	current->state = TASK_RUNNING;
+   	remove_wait_queue(&dev_priv->flush_queue, &entry);
+   
+   	return;
+}
+
+static int i810_flush_queue(drm_device_t *dev)
+{
+   	DECLARE_WAITQUEUE(entry, current);
+  	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
+	unsigned long end;
+   	int ret = 0;      
+
+   	if(dev_priv == NULL) {
+	   	return 0;
+	}
+      	atomic_set(&dev_priv->flush_done, 0);
+   	current->state = TASK_INTERRUPTIBLE;
+   	add_wait_queue(&dev_priv->flush_queue, &entry);
+   	end = jiffies + (HZ*3);
+   	for (;;) {
+	      	i810_dma_emit_flush(dev);
+	   	if (atomic_read(&dev_priv->flush_done) == 1) break;
+		if((signed)(end - jiffies) <= 0) {
+		   	DRM_ERROR("lockup\n");
+		   	break;
+		}	   
+	      	schedule_timeout(HZ*3);
+	      	if (signal_pending(current)) {
+		   	ret = -EINTR; /* Can't restart */
+		   	break;
+		}
+	}
+   
+   	current->state = TASK_RUNNING;
+   	remove_wait_queue(&dev_priv->flush_queue, &entry);
+   
+   	return ret;
+}
+
+/* Must be called with the lock held */
+void i810_reclaim_buffers(drm_device_t *dev, pid_t pid)
+{
+	drm_device_dma_t *dma = dev->dma;
+	int		 i;
+
+	if (!dma) return;
+      	if(dev->dev_private == NULL) return;
+
+        i810_flush_queue(dev);
+
+	for (i = 0; i < dma->buf_count; i++) {
+	   	drm_buf_t *buf = dma->buflist[ i ];
+	   	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+
+		if (buf->pid == pid) {
+		   	/* Only buffers that need to get reclaimed ever 
+			 * get set to free */
+		   	if(buf_priv == NULL) return;
+		   	cmpxchg(buf_priv->in_use, 
+				I810_BUF_USED, I810_BUF_FREE);
+		}
+	}
+}
+
 int i810_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 	       unsigned long arg)
 {
 	drm_file_t	  *priv	  = filp->private_data;
 	drm_device_t	  *dev	  = priv->dev;
+
 	DECLARE_WAITQUEUE(entry, current);
 	int		  ret	= 0;
 	drm_lock_t	  lock;
-	drm_queue_t	  *q;
-#if DRM_DMA_HISTOGRAM
-	cycles_t	  start;
-
-	dev->lck_start = start = get_cycles();
-#endif
 
 	copy_from_user_ret(&lock, (drm_lock_t *)arg, sizeof(lock), -EFAULT);
 
@@ -696,16 +748,20 @@ int i810_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 			  current->pid, lock.context);
 		return -EINVAL;
 	}
+   
+   	DRM_DEBUG("%d (pid %d) requests lock (0x%08x), flags = 0x%08x\n",
+		  lock.context, current->pid, dev->lock.hw_lock->lock,
+		  lock.flags);
 
-	if (lock.context < 0 || lock.context >= dev->queue_count) {
+	if (lock.context < 0) {
 		return -EINVAL;
 	}
-	q = dev->queuelist[lock.context];
-	
-	ret = drm_flush_block_and_flush(dev, lock.context, lock.flags);
+	/* Only one queue:
+	 */
 
 	if (!ret) {
-		if (_DRM_LOCKING_CONTEXT(dev->lock.hw_lock->lock)
+#if 0
+	   	if (_DRM_LOCKING_CONTEXT(dev->lock.hw_lock->lock)
 		    != lock.context) {
 			long j = jiffies - dev->lock.lock_time;
 
@@ -716,6 +772,7 @@ int i810_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 				schedule_timeout(j);
 			}
 		}
+#endif
 		add_wait_queue(&dev->lock.lock_queue, &entry);
 		for (;;) {
 			if (!dev->lock.hw_lock) {
@@ -728,13 +785,13 @@ int i810_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 				dev->lock.pid	    = current->pid;
 				dev->lock.lock_time = jiffies;
 				atomic_inc(&dev->total_locks);
-				atomic_inc(&q->total_locks);
 				break;	/* Got lock */
 			}
 			
 				/* Contention */
 			atomic_inc(&dev->total_sleeps);
 			current->state = TASK_INTERRUPTIBLE;
+		   	DRM_DEBUG("Calling lock schedule\n");
 			schedule();
 			if (signal_pending(current)) {
 				ret = -ERESTARTSYS;
@@ -744,19 +801,184 @@ int i810_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 		current->state = TASK_RUNNING;
 		remove_wait_queue(&dev->lock.lock_queue, &entry);
 	}
-
-	drm_flush_unblock(dev, lock.context, lock.flags); /* cleanup phase */
 	
 	if (!ret) {
-		if (lock.flags & _DRM_LOCK_READY)
-			i810_dma_ready(dev);
-		if (lock.flags & _DRM_LOCK_QUIESCENT)
-			i810_dma_quiescent(dev);
+		if (lock.flags & _DRM_LOCK_QUIESCENT) {
+		   DRM_DEBUG("_DRM_LOCK_QUIESCENT\n");
+		   DRM_DEBUG("fred\n");
+		   i810_dma_quiescent(dev);
+		}
 	}
+	DRM_DEBUG("%d %s\n", lock.context, ret ? "interrupted" : "has lock");
+	return ret;
+}
 
-#if DRM_DMA_HISTOGRAM
-	atomic_inc(&dev->histo.lacq[drm_histogram_slot(get_cycles() - start)]);
-#endif
+int i810_flush_ioctl(struct inode *inode, struct file *filp, 
+		     unsigned int cmd, unsigned long arg)
+{
+   	drm_file_t	  *priv	  = filp->private_data;
+   	drm_device_t	  *dev	  = priv->dev;
+   
+   	DRM_DEBUG("i810_flush_ioctl\n");
+   	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("i810_flush_ioctl called without lock held\n");
+		return -EINVAL;
+	}
+
+   	i810_flush_queue(dev);
+   	return 0;
+}
+
+static int i810DmaGeneral(drm_device_t *dev, drm_i810_general_t *args)
+{
+	drm_device_dma_t *dma = dev->dma;
+   	drm_buf_t *buf = dma->buflist[ args->idx ];
+   	
+	if (!args->used) {
+	   	i810_freelist_put(dev, buf);
+	} else {  
+		i810_dma_dispatch_general( dev, buf, args->used );
+		atomic_add(args->used, &dma->total_bytes);
+		atomic_inc(&dma->total_dmas);
+	}
+	return 0; 
+}
+
+static int i810DmaVertex(drm_device_t *dev, drm_i810_vertex_t *args)
+{
+	drm_device_dma_t *dma = dev->dma;
+	drm_buf_t *buf = dma->buflist[ args->idx ];
+   	i810_dma_dispatch_vertex( dev, buf, args->discard, args->used );
+   	atomic_add(args->used, &dma->total_bytes);
+	atomic_inc(&dma->total_dmas);
+   	return 0;
+}
+
+int i810_dma_general(struct inode *inode, struct file *filp,
+		     unsigned int cmd, unsigned long arg)
+{
+	drm_file_t *priv = filp->private_data;
+	drm_device_t *dev = priv->dev;
+	drm_i810_general_t general;
+      	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
+      	u32 *hw_status = (u32 *)dev_priv->hw_status_page;
+   	drm_i810_sarea_t *sarea_priv = (drm_i810_sarea_t *) 
+     					dev_priv->sarea_priv; 
+
+	int retcode = 0;
 	
-	return ret;
+	copy_from_user_ret(&general, (drm_i810_general_t *)arg, sizeof(general),
+			   -EFAULT);
+
+	DRM_DEBUG("i810 dma general idx %d used %d\n",
+		  general.idx, general.used);
+   
+   	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("i810_dma_general called without lock held\n");
+		return -EINVAL;
+	}
+
+	retcode = i810DmaGeneral(dev, &general);
+	sarea_priv->last_enqueue = dev_priv->counter-1;
+   	sarea_priv->last_dispatch = (int) hw_status[5];
+   
+	return retcode;
+}
+
+int i810_dma_vertex(struct inode *inode, struct file *filp,
+	       unsigned int cmd, unsigned long arg)
+{
+	drm_file_t *priv = filp->private_data;
+	drm_device_t *dev = priv->dev;
+   	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
+      	u32 *hw_status = (u32 *)dev_priv->hw_status_page;
+   	drm_i810_sarea_t *sarea_priv = (drm_i810_sarea_t *) 
+     					dev_priv->sarea_priv; 
+	drm_i810_vertex_t vertex;
+	int retcode = 0;
+
+	copy_from_user_ret(&vertex, (drm_i810_vertex_t *)arg, sizeof(vertex),
+			   -EFAULT);
+   	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("i810_dma_vertex called without lock held\n");
+		return -EINVAL;
+	}
+
+	DRM_DEBUG("i810 dma vertex, idx %d used %d discard %d\n",
+		  vertex.idx, vertex.used, vertex.discard);
+
+	retcode = i810DmaVertex(dev, &vertex);
+	sarea_priv->last_enqueue = dev_priv->counter-1;
+   	sarea_priv->last_dispatch = (int) hw_status[5];
+   
+	return retcode;
+
+}
+
+int i810_getage(struct inode *inode, struct file *filp, unsigned int cmd,
+		unsigned long arg)
+{
+   	drm_file_t	  *priv	    = filp->private_data;
+	drm_device_t	  *dev	    = priv->dev;
+   	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
+      	u32 *hw_status = (u32 *)dev_priv->hw_status_page;
+   	drm_i810_sarea_t *sarea_priv = (drm_i810_sarea_t *) 
+     					dev_priv->sarea_priv; 
+
+      	sarea_priv->last_dispatch = (int) hw_status[5];
+	return 0;
+}
+
+int i810_dma(struct inode *inode, struct file *filp, unsigned int cmd,
+	    unsigned long arg)
+{
+	drm_file_t	  *priv	    = filp->private_data;
+	drm_device_t	  *dev	    = priv->dev;
+	drm_device_dma_t  *dma	    = dev->dma;
+	int		  retcode   = 0;
+	drm_dma_t	  d;
+   	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
+   	u32 *hw_status = (u32 *)dev_priv->hw_status_page;
+   	drm_i810_sarea_t *sarea_priv = (drm_i810_sarea_t *) 
+     					dev_priv->sarea_priv; 
+
+
+   	copy_from_user_ret(&d, (drm_dma_t *)arg, sizeof(d), -EFAULT);
+	DRM_DEBUG("%d %d: %d send, %d req\n",
+		  current->pid, d.context, d.send_count, d.request_count);
+   
+	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("i810_dma called without lock held\n");
+		return -EINVAL;
+	}
+
+	/* Please don't send us buffers.
+	 */
+	if (d.send_count != 0) {
+		DRM_ERROR("Process %d trying to send %d buffers via drmDMA\n",
+			  current->pid, d.send_count);
+		return -EINVAL;
+	}
+	
+	/* We'll send you buffers.
+	 */
+	if (d.request_count < 0 || d.request_count > dma->buf_count) {
+		DRM_ERROR("Process %d trying to get %d buffers (of %d max)\n",
+			  current->pid, d.request_count, dma->buf_count);
+		return -EINVAL;
+	}
+	
+	d.granted_count = 0;
+
+	if (!retcode && d.request_count) {
+		retcode = i810_dma_get_buffers(dev, &d);
+	}
+
+	DRM_DEBUG("i810_dma: %d returning, granted = %d\n",
+		  current->pid, d.granted_count);
+
+	copy_to_user_ret((drm_dma_t *)arg, &d, sizeof(d), -EFAULT);   
+   	sarea_priv->last_dispatch = (int) hw_status[5];
+
+	return retcode;
 }
diff --git a/linux/i810_drm.h b/linux/i810_drm.h
new file mode 100644
index 00000000..0754874c
--- /dev/null
+++ b/linux/i810_drm.h
@@ -0,0 +1,93 @@
+#ifndef _I810_DRM_H_
+#define _I810_DRM_H_
+
+/* WARNING: These defines must be the same as what the Xserver uses.
+ * if you change them, you must change the defines in the Xserver.
+ */
+
+/* Might one day want to support the client-side ringbuffer code again.
+ */
+#ifndef _I810_DEFINES_
+#define _I810_DEFINES_
+
+#define I810_USE_BATCH			1
+#define I810_DMA_BUF_ORDER		12
+#define I810_DMA_BUF_SZ 		(1<<I810_DMA_BUF_ORDER)
+#define I810_DMA_BUF_NR 		256
+#define I810_NR_SAREA_CLIPRECTS 	2
+
+/* Each region is a minimum of 64k, and there are at most 64 of them.
+ */
+
+#define I810_NR_TEX_REGIONS 64
+#define I810_LOG_MIN_TEX_REGION_SIZE 16
+#endif
+
+typedef struct _drm_i810_init {
+   	enum {
+	   	I810_INIT_DMA = 0x01,
+	   	I810_CLEANUP_DMA = 0x02
+	} func;
+   	int ring_map_idx;
+      	int buffer_map_idx;
+	int sarea_priv_offset;
+   	unsigned long ring_start;
+   	unsigned long ring_end;
+   	unsigned long ring_size;
+} drm_i810_init_t;
+
+/* Warning: If you change the SAREA structure you must change the Xserver
+ * structure as well */
+
+typedef struct _drm_i810_tex_region {
+	unsigned char next, prev; /* indices to form a circular LRU  */
+	unsigned char in_use;	/* owned by a client, or free? */
+	int age;		/* tracked by clients to update local LRU's */
+} drm_i810_tex_region_t;
+
+typedef struct _drm_i810_sarea {
+	unsigned int nbox;
+	drm_clip_rect_t boxes[I810_NR_SAREA_CLIPRECTS];
+
+	/* Maintain an LRU of contiguous regions of texture space.  If
+	 * you think you own a region of texture memory, and it has an
+	 * age different to the one you set, then you are mistaken and
+	 * it has been stolen by another client.  If global texAge
+	 * hasn't changed, there is no need to walk the list.
+	 *
+	 * These regions can be used as a proxy for the fine-grained
+	 * texture information of other clients - by maintaining them
+	 * in the same lru which is used to age their own textures,
+	 * clients have an approximate lru for the whole of global
+	 * texture space, and can make informed decisions as to which
+	 * areas to kick out.  There is no need to choose whether to
+	 * kick out your own texture or someone else's - simply eject
+	 * them all in LRU order.  
+	 */
+   
+	drm_i810_tex_region_t texList[I810_NR_TEX_REGIONS+1]; 
+				/* Last elt is sentinal */
+        int texAge;		/* last time texture was uploaded */
+        int last_enqueue;	/* last time a buffer was enqueued */
+	int last_dispatch;	/* age of the most recently dispatched buffer */
+	int last_quiescent;     /*  */
+	int ctxOwner;		/* last context to upload state */
+} drm_i810_sarea_t;
+
+typedef struct _drm_i810_general {
+   	int idx;
+	int used;
+} drm_i810_general_t;
+
+/* These may be placeholders if we have more cliprects than
+ * I810_NR_SAREA_CLIPRECTS.  In that case, the client sets discard to
+ * false, indicating that the buffer will be dispatched again with a
+ * new set of cliprects.
+ */
+typedef struct _drm_i810_vertex {
+   	int idx;		/* buffer index */
+	int used;		/* nr bytes in use */
+	int discard;		/* client is finished with the buffer? */
+} drm_i810_vertex_t;
+
+#endif /* _I810_DRM_H_ */
diff --git a/linux/i810_drv.c b/linux/i810_drv.c
index f33153a3..d3b35c49 100644
--- a/linux/i810_drv.c
+++ b/linux/i810_drv.c
@@ -33,11 +33,13 @@
 #define EXPORT_SYMTAB
 #include "drmP.h"
 #include "i810_drv.h"
+
+
 EXPORT_SYMBOL(i810_init);
 EXPORT_SYMBOL(i810_cleanup);
 
 #define I810_NAME	 "i810"
-#define I810_DESC	 "Matrox g200/g400"
+#define I810_DESC	 "Intel I810"
 #define I810_DATE	 "19991213"
 #define I810_MAJOR	 0
 #define I810_MINOR	 0
@@ -54,6 +56,7 @@ static struct file_operations i810_fops = {
 	mmap:	 drm_mmap,
 	read:	 drm_read,
 	fasync:	 drm_fasync,
+      	poll:	 drm_poll,
 };
 
 static struct miscdevice      i810_misc = {
@@ -80,13 +83,13 @@ static drm_ioctl_desc_t	      i810_ioctls[] = {
 	[DRM_IOCTL_NR(DRM_IOCTL_MAP_BUFS)]    = { i810_mapbufs,	  1, 0 },
 	[DRM_IOCTL_NR(DRM_IOCTL_FREE_BUFS)]   = { i810_freebufs,  1, 0 },
 
-	[DRM_IOCTL_NR(DRM_IOCTL_ADD_CTX)]     = { drm_addctx,	  1, 1 },
-	[DRM_IOCTL_NR(DRM_IOCTL_RM_CTX)]      = { drm_rmctx,	  1, 1 },
-	[DRM_IOCTL_NR(DRM_IOCTL_MOD_CTX)]     = { drm_modctx,	  1, 1 },
-	[DRM_IOCTL_NR(DRM_IOCTL_GET_CTX)]     = { drm_getctx,	  1, 0 },
-	[DRM_IOCTL_NR(DRM_IOCTL_SWITCH_CTX)]  = { drm_switchctx,  1, 1 },
-	[DRM_IOCTL_NR(DRM_IOCTL_NEW_CTX)]     = { drm_newctx,	  1, 1 },
-	[DRM_IOCTL_NR(DRM_IOCTL_RES_CTX)]     = { drm_resctx,	  1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_ADD_CTX)]     = { i810_addctx,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_RM_CTX)]      = { i810_rmctx,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_MOD_CTX)]     = { i810_modctx,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_GET_CTX)]     = { i810_getctx,	  1, 0 },
+	[DRM_IOCTL_NR(DRM_IOCTL_SWITCH_CTX)]  = { i810_switchctx,  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_NEW_CTX)]     = { i810_newctx,	  1, 1 },
+	[DRM_IOCTL_NR(DRM_IOCTL_RES_CTX)]     = { i810_resctx,	  1, 0 },
 	[DRM_IOCTL_NR(DRM_IOCTL_ADD_DRAW)]    = { drm_adddraw,	  1, 1 },
 	[DRM_IOCTL_NR(DRM_IOCTL_RM_DRAW)]     = { drm_rmdraw,	  1, 1 },
 
@@ -104,6 +107,11 @@ static drm_ioctl_desc_t	      i810_ioctls[] = {
 	[DRM_IOCTL_NR(DRM_IOCTL_AGP_FREE)]    = { drm_agp_free,    1, 1 },
 	[DRM_IOCTL_NR(DRM_IOCTL_AGP_BIND)]    = { drm_agp_bind,    1, 1 },
 	[DRM_IOCTL_NR(DRM_IOCTL_AGP_UNBIND)]  = { drm_agp_unbind,  1, 1 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_I810_INIT)]   = { i810_dma_init,   1, 1 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_I810_VERTEX)] = { i810_dma_vertex, 1, 0 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_I810_DMA)]    = { i810_dma_general,1, 0 },
+      	[DRM_IOCTL_NR(DRM_IOCTL_I810_FLUSH)]  = { i810_flush_ioctl,1, 0 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_I810_GETAGE)] = { i810_getage,     1, 0 },
 };
 
 #define I810_IOCTL_COUNT DRM_ARRAY_SIZE(i810_ioctls)
@@ -121,7 +129,7 @@ MODULE_PARM(i810, "s");
 
 int init_module(void)
 {
-	printk("doing i810_init()\n");
+	DRM_DEBUG("doing i810_init()\n");
 	return i810_init();
 }
 
@@ -364,7 +372,7 @@ int i810_init(void)
 #ifdef MODULE
 	drm_parse_options(i810);
 #endif
-	printk("doing misc_register\n");
+	DRM_DEBUG("doing misc_register\n");
 	if ((retcode = misc_register(&i810_misc))) {
 		DRM_ERROR("Cannot register \"%s\"\n", I810_NAME);
 		return retcode;
@@ -372,13 +380,22 @@ int i810_init(void)
 	dev->device = MKDEV(MISC_MAJOR, i810_misc.minor);
 	dev->name   = I810_NAME;
 
-   	printk("doing mem init\n");
+   	DRM_DEBUG("doing mem init\n");
 	drm_mem_init();
-	printk("doing proc init\n");
+	DRM_DEBUG("doing proc init\n");
 	drm_proc_init(dev);
-	printk("doing agp init\n");
+	DRM_DEBUG("doing agp init\n");
 	dev->agp    = drm_agp_init();
-	printk("doing ctxbitmap init\n");
+   	if(dev->agp == NULL) {
+	   	DRM_INFO("The i810 drm module requires the agpgart module"
+			 " to function correctly\nPlease load the agpgart"
+			 " module before you load the i810 module\n");
+	   	drm_proc_cleanup();
+	   	misc_deregister(&i810_misc);
+	   	i810_takedown(dev);
+	   	return -ENOMEM;
+	}
+	DRM_DEBUG("doing ctxbitmap init\n");
 	if((retcode = drm_ctxbitmap_init(dev))) {
 		DRM_ERROR("Cannot allocate memory for context bitmap.\n");
 		drm_proc_cleanup();
@@ -386,10 +403,6 @@ int i810_init(void)
 		i810_takedown(dev);
 		return retcode;
 	}
-#if 0
-	printk("doing i810_dma_init\n");
-	i810_dma_init(dev);
-#endif
 
 	DRM_INFO("Initialized %s %d.%d.%d %s on minor %d\n",
 		 I810_NAME,
@@ -417,7 +430,6 @@ void i810_cleanup(void)
 		DRM_INFO("Module unloaded\n");
 	}
 	drm_ctxbitmap_cleanup(dev);
-	i810_dma_cleanup(dev);
 	i810_takedown(dev);
 	if (dev->agp) {
 		drm_free(dev->agp, sizeof(*dev->agp), DRM_MEM_AGPLISTS);
@@ -484,24 +496,82 @@ int i810_release(struct inode *inode, struct file *filp)
 	drm_device_t  *dev    = priv->dev;
 	int	      retcode = 0;
 
-	DRM_DEBUG("open_count = %d\n", dev->open_count);
-	if (!(retcode = drm_release(inode, filp))) {
-		MOD_DEC_USE_COUNT;
-		atomic_inc(&dev->total_close);
-		spin_lock(&dev->count_lock);
-		if (!--dev->open_count) {
-			if (atomic_read(&dev->ioctl_count) || dev->blocked) {
-				DRM_ERROR("Device busy: %d %d\n",
-					  atomic_read(&dev->ioctl_count),
-					  dev->blocked);
-				spin_unlock(&dev->count_lock);
-				return -EBUSY;
+	DRM_DEBUG("pid = %d, device = 0x%x, open_count = %d\n",
+		  current->pid, dev->device, dev->open_count);
+
+	if (_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)
+	    && dev->lock.pid == current->pid) {
+	      	i810_reclaim_buffers(dev, priv->pid);
+		DRM_ERROR("Process %d dead, freeing lock for context %d\n",
+			  current->pid,
+			  _DRM_LOCKING_CONTEXT(dev->lock.hw_lock->lock));
+		drm_lock_free(dev,
+			      &dev->lock.hw_lock->lock,
+			      _DRM_LOCKING_CONTEXT(dev->lock.hw_lock->lock));
+		
+				/* FIXME: may require heavy-handed reset of
+                                   hardware at this point, possibly
+                                   processed via a callback to the X
+                                   server. */
+	} else {
+	   	/* The lock is required to reclaim buffers */
+	   	DECLARE_WAITQUEUE(entry, current);
+	   	add_wait_queue(&dev->lock.lock_queue, &entry);
+		for (;;) {
+			if (!dev->lock.hw_lock) {
+				/* Device has been unregistered */
+				retcode = -EINTR;
+				break;
+			}
+			if (drm_lock_take(&dev->lock.hw_lock->lock,
+					  DRM_KERNEL_CONTEXT)) {
+				dev->lock.pid	    = priv->pid;
+				dev->lock.lock_time = jiffies;
+				atomic_inc(&dev->total_locks);
+				break;	/* Got lock */
+			}			
+				/* Contention */
+			atomic_inc(&dev->total_sleeps);
+			current->state = TASK_INTERRUPTIBLE;
+			schedule();
+			if (signal_pending(current)) {
+				retcode = -ERESTARTSYS;
+				break;
 			}
-			spin_unlock(&dev->count_lock);
-			return i810_takedown(dev);
 		}
-		spin_unlock(&dev->count_lock);
+		current->state = TASK_RUNNING;
+		remove_wait_queue(&dev->lock.lock_queue, &entry);
+	   	if(!retcode) {
+		   	i810_reclaim_buffers(dev, priv->pid);
+		   	drm_lock_free(dev, &dev->lock.hw_lock->lock,
+				      DRM_KERNEL_CONTEXT);
+		}
+	}
+	drm_fasync(-1, filp, 0);
+
+	down(&dev->struct_sem);
+	if (priv->prev) priv->prev->next = priv->next;
+	else		dev->file_first	 = priv->next;
+	if (priv->next) priv->next->prev = priv->prev;
+	else		dev->file_last	 = priv->prev;
+	up(&dev->struct_sem);
+	
+	drm_free(priv, sizeof(*priv), DRM_MEM_FILES);
+   	MOD_DEC_USE_COUNT;
+   	atomic_inc(&dev->total_close);
+   	spin_lock(&dev->count_lock);
+   	if (!--dev->open_count) {
+	   	if (atomic_read(&dev->ioctl_count) || dev->blocked) {
+		   	DRM_ERROR("Device busy: %d %d\n",
+				  atomic_read(&dev->ioctl_count),
+				  dev->blocked);
+		   	spin_unlock(&dev->count_lock);
+		   	return -EBUSY;
+		}
+	   	spin_unlock(&dev->count_lock);
+	   	return i810_takedown(dev);
 	}
+   	spin_unlock(&dev->count_lock);
 	return retcode;
 }
 
@@ -566,8 +636,7 @@ int i810_unlock(struct inode *inode, struct file *filp, unsigned int cmd,
 	atomic_inc(&dev->total_unlocks);
 	if (_DRM_LOCK_IS_CONT(dev->lock.hw_lock->lock))
 		atomic_inc(&dev->total_contends);
-	drm_lock_transfer(dev, &dev->lock.hw_lock->lock, DRM_KERNEL_CONTEXT);
-	i810_dma_schedule(dev, 1);
+   	drm_lock_transfer(dev, &dev->lock.hw_lock->lock, DRM_KERNEL_CONTEXT);
 	if (!dev->context_flag) {
 		if (drm_lock_free(dev, &dev->lock.hw_lock->lock,
 				  DRM_KERNEL_CONTEXT)) {
diff --git a/linux/i810_drv.h b/linux/i810_drv.h
index 0f5f42bb..1badd36b 100644
--- a/linux/i810_drv.h
+++ b/linux/i810_drv.h
@@ -32,6 +32,31 @@
 #ifndef _I810_DRV_H_
 #define _I810_DRV_H_
 
+typedef struct _drm_i810_ring_buffer{
+	int tail_mask;
+	unsigned long Start;
+	unsigned long End;
+	unsigned long Size;
+	u8 *virtual_start;
+	int head;
+	int tail;
+	int space;
+} drm_i810_ring_buffer_t;
+
+typedef struct drm_i810_private {
+   	int ring_map_idx;
+   	int buffer_map_idx;
+
+   	drm_i810_ring_buffer_t ring;
+	drm_i810_sarea_t *sarea_priv;
+
+      	unsigned long hw_status_page;
+   	unsigned long counter;
+
+   	atomic_t flush_done;
+   	wait_queue_head_t flush_queue;	/* Processes waiting until flush    */
+} drm_i810_private_t;
+
 				/* i810_drv.c */
 extern int  i810_init(void);
 extern void i810_cleanup(void);
@@ -54,8 +79,13 @@ extern int  i810_control(struct inode *inode, struct file *filp,
 			  unsigned int cmd, unsigned long arg);
 extern int  i810_lock(struct inode *inode, struct file *filp,
 		       unsigned int cmd, unsigned long arg);
-extern void i810_dma_init(drm_device_t *dev);
-extern void i810_dma_cleanup(drm_device_t *dev);
+extern int  i810_dma_init(struct inode *inode, struct file *filp,
+			  unsigned int cmd, unsigned long arg);
+extern int  i810_flush_ioctl(struct inode *inode, struct file *filp,
+			     unsigned int cmd, unsigned long arg);
+extern void i810_reclaim_buffers(drm_device_t *dev, pid_t pid);
+extern int  i810_getage(struct inode *inode, struct file *filp, unsigned int cmd,
+			unsigned long arg);
 
 
 				/* i810_bufs.c */
@@ -72,5 +102,103 @@ extern int  i810_mapbufs(struct inode *inode, struct file *filp,
 extern int  i810_addmap(struct inode *inode, struct file *filp,
 		       unsigned int cmd, unsigned long arg);
 
+				/* i810_context.c */
+extern int  i810_resctx(struct inode *inode, struct file *filp,
+		       unsigned int cmd, unsigned long arg);
+extern int  i810_addctx(struct inode *inode, struct file *filp,
+		       unsigned int cmd, unsigned long arg);
+extern int  i810_modctx(struct inode *inode, struct file *filp,
+		       unsigned int cmd, unsigned long arg);
+extern int  i810_getctx(struct inode *inode, struct file *filp,
+		       unsigned int cmd, unsigned long arg);
+extern int  i810_switchctx(struct inode *inode, struct file *filp,
+			  unsigned int cmd, unsigned long arg);
+extern int  i810_newctx(struct inode *inode, struct file *filp,
+		       unsigned int cmd, unsigned long arg);
+extern int  i810_rmctx(struct inode *inode, struct file *filp,
+		      unsigned int cmd, unsigned long arg);
+
+extern int  i810_context_switch(drm_device_t *dev, int old, int new);
+extern int  i810_context_switch_complete(drm_device_t *dev, int new);
+
+
+
+
+/* Copy the outstanding cliprects for every I810_DMA_VERTEX buffer.
+ * This can be fixed by emitting directly to the ringbuffer in the
+ * 'vertex_dma' ioctl.  
+*/
+typedef struct {
+   	u32 *in_use;
+   	int my_use_idx;
+} drm_i810_buf_priv_t;
+
+
+#define I810_DMA_GENERAL 0
+#define I810_DMA_VERTEX  1
+#define I810_DMA_DISCARD 2	/* not used */
+
+#define I810_VERBOSE 0
+
+
+int i810_dma_vertex(struct inode *inode, struct file *filp,
+		    unsigned int cmd, unsigned long arg);
+
+int i810_dma_general(struct inode *inode, struct file *filp,
+		     unsigned int cmd, unsigned long arg);
+
+
+#define GFX_OP_USER_INTERRUPT 		((0<<29)|(2<<23))
+#define GFX_OP_BREAKPOINT_INTERRUPT	((0<<29)|(1<<23))
+#define CMD_REPORT_HEAD			(7<<23)
+#define CMD_STORE_DWORD_IDX		((0x21<<23) | 0x1)
+#define CMD_OP_BATCH_BUFFER  ((0x0<<29)|(0x30<<23)|0x1)
+
+#define INST_PARSER_CLIENT   0x00000000
+#define INST_OP_FLUSH        0x02000000
+#define INST_FLUSH_MAP_CACHE 0x00000001
+
+
+#define BB1_START_ADDR_MASK   (~0x7)
+#define BB1_PROTECTED         (1<<0)
+#define BB1_UNPROTECTED       (0<<0)
+#define BB2_END_ADDR_MASK     (~0x7)
+
+#define I810REG_HWSTAM		0x02098
+#define I810REG_INT_IDENTITY_R	0x020a4
+#define I810REG_INT_MASK_R 	0x020a8
+#define I810REG_INT_ENABLE_R	0x020a0
+
+#define LP_RING     		0x2030
+#define HP_RING     		0x2040
+#define RING_TAIL      		0x00
+#define TAIL_ADDR		0x000FFFF8
+#define RING_HEAD      		0x04
+#define HEAD_WRAP_COUNT     	0xFFE00000
+#define HEAD_WRAP_ONE       	0x00200000
+#define HEAD_ADDR           	0x001FFFFC
+#define RING_START     		0x08
+#define START_ADDR          	0x00FFFFF8
+#define RING_LEN       		0x0C
+#define RING_NR_PAGES       	0x000FF000 
+#define RING_REPORT_MASK    	0x00000006
+#define RING_REPORT_64K     	0x00000002
+#define RING_REPORT_128K    	0x00000004
+#define RING_NO_REPORT      	0x00000000
+#define RING_VALID_MASK     	0x00000001
+#define RING_VALID          	0x00000001
+#define RING_INVALID        	0x00000000
+
+#define GFX_OP_SCISSOR         ((0x3<<29)|(0x1c<<24)|(0x10<<19))
+#define SC_UPDATE_SCISSOR       (0x1<<1)
+#define SC_ENABLE_MASK          (0x1<<0)
+#define SC_ENABLE               (0x1<<0)
+
+#define GFX_OP_SCISSOR_INFO    ((0x3<<29)|(0x1d<<24)|(0x81<<16)|(0x1))
+#define SCI_YMIN_MASK      (0xffff<<16)
+#define SCI_XMIN_MASK      (0xffff<<0)
+#define SCI_YMAX_MASK      (0xffff<<16)
+#define SCI_XMAX_MASK      (0xffff<<0)
 
 #endif
+
diff --git a/linux/memory.c b/linux/memory.c
index a8a81abd..559ac739 100644
--- a/linux/memory.c
+++ b/linux/memory.c
@@ -44,21 +44,25 @@ static spinlock_t	  drm_mem_lock	    = SPIN_LOCK_UNLOCKED;
 static unsigned long	  drm_ram_available = 0; /* In pages */
 static unsigned long	  drm_ram_used	    = 0;
 static drm_mem_stats_t	  drm_mem_stats[]   = {
-	[DRM_MEM_DMA]	   = { "dmabufs"  },
-	[DRM_MEM_SAREA]	   = { "sareas"	  },
-	[DRM_MEM_DRIVER]   = { "driver"	  },
-	[DRM_MEM_MAGIC]	   = { "magic"	  },
-	[DRM_MEM_IOCTLS]   = { "ioctltab" },
-	[DRM_MEM_MAPS]	   = { "maplist"  },
-	[DRM_MEM_VMAS]	   = { "vmalist"  },
-	[DRM_MEM_BUFS]	   = { "buflist"  },
-	[DRM_MEM_SEGS]	   = { "seglist"  },
-	[DRM_MEM_PAGES]	   = { "pagelist" },
-	[DRM_MEM_FILES]	   = { "files"	  },
-	[DRM_MEM_QUEUES]   = { "queues"	  },
-	[DRM_MEM_CMDS]	   = { "commands" },
-	[DRM_MEM_MAPPINGS] = { "mappings" },
-	[DRM_MEM_BUFLISTS] = { "buflists" },
+	[DRM_MEM_DMA]	    = { "dmabufs"  },
+	[DRM_MEM_SAREA]	    = { "sareas"   },
+	[DRM_MEM_DRIVER]    = { "driver"   },
+	[DRM_MEM_MAGIC]	    = { "magic"	   },
+	[DRM_MEM_IOCTLS]    = { "ioctltab" },
+	[DRM_MEM_MAPS]	    = { "maplist"  },
+	[DRM_MEM_VMAS]	    = { "vmalist"  },
+	[DRM_MEM_BUFS]	    = { "buflist"  },
+	[DRM_MEM_SEGS]	    = { "seglist"  },
+	[DRM_MEM_PAGES]	    = { "pagelist" },
+	[DRM_MEM_FILES]	    = { "files"	   },
+	[DRM_MEM_QUEUES]    = { "queues"   },
+	[DRM_MEM_CMDS]	    = { "commands" },
+	[DRM_MEM_MAPPINGS]  = { "mappings" },
+	[DRM_MEM_BUFLISTS]  = { "buflists" },
+	[DRM_MEM_AGPLISTS]  = { "agplist"  },
+	[DRM_MEM_TOTALAGP]  = { "totalagp" },
+	[DRM_MEM_BOUNDAGP]  = { "boundagp" },
+	[DRM_MEM_CTXBITMAP] = { "ctxbitmap"},
 	{ NULL, 0, }		/* Last entry must be null */
 };
 
@@ -324,3 +328,120 @@ void drm_ioremapfree(void *pt, unsigned long size)
 			      free_count, alloc_count);
 	}
 }
+
+#ifdef DRM_AGP
+agp_memory *drm_alloc_agp(int pages, u32 type)
+{
+	agp_memory *handle;
+
+	if (!pages) {
+		DRM_MEM_ERROR(DRM_MEM_TOTALAGP, "Allocating 0 pages\n");
+		return NULL;
+	}
+	
+	if (drm_agp.allocate_memory) {
+		if ((handle = (*drm_agp.allocate_memory)(pages,
+							 type))) {
+			spin_lock(&drm_mem_lock);
+			++drm_mem_stats[DRM_MEM_TOTALAGP].succeed_count;
+			drm_mem_stats[DRM_MEM_TOTALAGP].bytes_allocated
+				+= pages << PAGE_SHIFT;
+			spin_unlock(&drm_mem_lock);
+			return handle;
+		}
+	}
+	spin_lock(&drm_mem_lock);
+	++drm_mem_stats[DRM_MEM_TOTALAGP].fail_count;
+	spin_unlock(&drm_mem_lock);
+	return NULL;
+}
+
+int drm_free_agp(agp_memory *handle, int pages)
+{
+	int           alloc_count;
+	int           free_count;
+	int           retval = -EINVAL;
+
+	if (!handle) {
+		DRM_MEM_ERROR(DRM_MEM_TOTALAGP,
+			      "Attempt to free NULL AGP handle\n");
+		return retval;;
+	}
+	
+	if (drm_agp.free_memory) {
+		(*drm_agp.free_memory)(handle);
+		spin_lock(&drm_mem_lock);
+		free_count  = ++drm_mem_stats[DRM_MEM_TOTALAGP].free_count;
+		alloc_count =   drm_mem_stats[DRM_MEM_TOTALAGP].succeed_count;
+		drm_mem_stats[DRM_MEM_TOTALAGP].bytes_freed
+			+= pages << PAGE_SHIFT;
+		spin_unlock(&drm_mem_lock);
+		if (free_count > alloc_count) {
+			DRM_MEM_ERROR(DRM_MEM_TOTALAGP,
+				      "Excess frees: %d frees, %d allocs\n",
+				      free_count, alloc_count);
+		}
+		return 0;
+	}
+	return retval;
+}
+
+int drm_bind_agp(agp_memory *handle, unsigned int start)
+{
+	int retcode = -EINVAL;
+
+   DRM_DEBUG("drm_bind_agp called\n");
+	if (!handle) {
+		DRM_MEM_ERROR(DRM_MEM_BOUNDAGP,
+			      "Attempt to bind NULL AGP handle\n");
+		return retcode;
+	}
+
+   DRM_DEBUG("drm_agp.bind_memory : %p\n", drm_agp.bind_memory);
+	if (drm_agp.bind_memory) {
+		if (!(retcode = (*drm_agp.bind_memory)(handle, start))) {
+			spin_lock(&drm_mem_lock);
+			++drm_mem_stats[DRM_MEM_BOUNDAGP].succeed_count;
+			drm_mem_stats[DRM_MEM_BOUNDAGP].bytes_allocated
+				+= handle->page_count << PAGE_SHIFT;
+			spin_unlock(&drm_mem_lock);
+		   DRM_DEBUG("drm_agp.bind_memory: retcode %d\n", retcode);
+			return retcode;
+		}
+	}
+	spin_lock(&drm_mem_lock);
+	++drm_mem_stats[DRM_MEM_BOUNDAGP].fail_count;
+	spin_unlock(&drm_mem_lock);
+	return retcode;
+}
+
+int drm_unbind_agp(agp_memory *handle)
+{
+	int alloc_count;
+	int free_count;
+	int retcode = -EINVAL;
+	
+	if (!handle) {
+		DRM_MEM_ERROR(DRM_MEM_BOUNDAGP,
+			      "Attempt to unbind NULL AGP handle\n");
+		return retcode;
+	}
+
+	if (drm_agp.unbind_memory) {
+		int c = handle->page_count;
+		if ((retcode = (*drm_agp.unbind_memory)(handle)))
+			return retcode;
+		spin_lock(&drm_mem_lock);
+		free_count  = ++drm_mem_stats[DRM_MEM_BOUNDAGP].free_count;
+		alloc_count = drm_mem_stats[DRM_MEM_BOUNDAGP].succeed_count;
+		drm_mem_stats[DRM_MEM_BOUNDAGP].bytes_freed += c << PAGE_SHIFT;
+		spin_unlock(&drm_mem_lock);
+		if (free_count > alloc_count) {
+			DRM_MEM_ERROR(DRM_MEM_BOUNDAGP,
+				      "Excess frees: %d frees, %d allocs\n",
+				      free_count, alloc_count);
+		}
+	}
+	return retcode;
+}
+#endif
diff --git a/linux/mga_bufs.c b/linux/mga_bufs.c
index 34f1112a..3ce428b2 100644
--- a/linux/mga_bufs.c
+++ b/linux/mga_bufs.c
@@ -33,169 +33,168 @@
 #define __NO_VERSION__
 #include "drmP.h"
 #include "mga_drv.h"
-#include "mga_dma.h"
 #include "linux/un.h"
 
 
 int mga_addbufs_agp(struct inode *inode, struct file *filp, unsigned int cmd,
 		    unsigned long arg)
 {
-   drm_file_t *priv = filp->private_data;
-   drm_device_t *dev = priv->dev;
-   drm_device_dma_t *dma = dev->dma;
-   drm_buf_desc_t request;
-   drm_buf_entry_t *entry;
-   drm_buf_t *buf;
-   unsigned long offset;
-   unsigned long agp_offset;
-   int count;
-   int order;
-   int size;
-   int alignment;
-   int page_order;
-   int total;
-   int byte_count;
-   int i;
-
-   if (!dma) return -EINVAL;
-
-   copy_from_user_ret(&request,
-		      (drm_buf_desc_t *)arg,
-		      sizeof(request),
-		      -EFAULT);
-
-   count = request.count;
-   order = drm_order(request.size);
-   size	= 1 << order;
-   agp_offset = request.agp_start;
-   alignment  = (request.flags & _DRM_PAGE_ALIGN) ? PAGE_ALIGN(size) :size;
-   page_order = order - PAGE_SHIFT > 0 ? order - PAGE_SHIFT : 0;
-   total = PAGE_SIZE << page_order;
-   byte_count = 0;
-
-   DRM_DEBUG("count: %d\n", count);
-   DRM_DEBUG("order: %d\n", order);
-   DRM_DEBUG("size: %d\n", size);
-   DRM_DEBUG("agp_offset: %d\n", agp_offset);
-   DRM_DEBUG("alignment: %d\n", alignment);
-   DRM_DEBUG("page_order: %d\n", page_order);
-   DRM_DEBUG("total: %d\n", total);
-   DRM_DEBUG("byte_count: %d\n", byte_count);
-
-   if (order < DRM_MIN_ORDER || order > DRM_MAX_ORDER) return -EINVAL;
-   if (dev->queue_count) return -EBUSY; /* Not while in use */
-   spin_lock(&dev->count_lock);
-   if (dev->buf_use) {
-      spin_unlock(&dev->count_lock);
-      return -EBUSY;
-   }
-   atomic_inc(&dev->buf_alloc);
-   spin_unlock(&dev->count_lock);
+	drm_file_t *priv = filp->private_data;
+	drm_device_t *dev = priv->dev;
+	drm_device_dma_t *dma = dev->dma;
+	drm_buf_desc_t request;
+	drm_buf_entry_t *entry;
+	drm_buf_t *buf;
+	unsigned long offset;
+	unsigned long agp_offset;
+	int count;
+	int order;
+	int size;
+	int alignment;
+	int page_order;
+	int total;
+	int byte_count;
+	int i;
+
+	if (!dma) return -EINVAL;
+
+	copy_from_user_ret(&request,
+			   (drm_buf_desc_t *)arg,
+			   sizeof(request),
+			   -EFAULT);
+
+	count = request.count;
+	order = drm_order(request.size);
+	size	= 1 << order;
+	agp_offset = request.agp_start;
+	alignment  = (request.flags & _DRM_PAGE_ALIGN) ? PAGE_ALIGN(size) :size;
+	page_order = order - PAGE_SHIFT > 0 ? order - PAGE_SHIFT : 0;
+	total = PAGE_SIZE << page_order;
+	byte_count = 0;
+
+	DRM_DEBUG("count: %d\n", count);
+	DRM_DEBUG("order: %d\n", order);
+	DRM_DEBUG("size: %d\n", size);
+	DRM_DEBUG("agp_offset: %ld\n", agp_offset);
+	DRM_DEBUG("alignment: %d\n", alignment);
+	DRM_DEBUG("page_order: %d\n", page_order);
+	DRM_DEBUG("total: %d\n", total);
+	DRM_DEBUG("byte_count: %d\n", byte_count);
+
+	if (order < DRM_MIN_ORDER || order > DRM_MAX_ORDER) return -EINVAL;
+	if (dev->queue_count) return -EBUSY; /* Not while in use */
+	spin_lock(&dev->count_lock);
+	if (dev->buf_use) {
+		spin_unlock(&dev->count_lock);
+		return -EBUSY;
+	}
+	atomic_inc(&dev->buf_alloc);
+	spin_unlock(&dev->count_lock);
    
-   down(&dev->struct_sem);
-   entry = &dma->bufs[order];
-   if (entry->buf_count) {
-      up(&dev->struct_sem);
-      atomic_dec(&dev->buf_alloc);
-      return -ENOMEM; /* May only call once for each order */
-   }
+	down(&dev->struct_sem);
+	entry = &dma->bufs[order];
+	if (entry->buf_count) {
+		up(&dev->struct_sem);
+		atomic_dec(&dev->buf_alloc);
+		return -ENOMEM; /* May only call once for each order */
+	}
    
-   entry->buflist = drm_alloc(count * sizeof(*entry->buflist),
-			      DRM_MEM_BUFS);
-   if (!entry->buflist) {
-      up(&dev->struct_sem);
-      atomic_dec(&dev->buf_alloc);
-      return -ENOMEM;
-   }
-   memset(entry->buflist, 0, count * sizeof(*entry->buflist));
+	entry->buflist = drm_alloc(count * sizeof(*entry->buflist),
+				   DRM_MEM_BUFS);
+	if (!entry->buflist) {
+		up(&dev->struct_sem);
+		atomic_dec(&dev->buf_alloc);
+		return -ENOMEM;
+	}
+	memset(entry->buflist, 0, count * sizeof(*entry->buflist));
    
-   entry->buf_size   = size;
-   entry->page_order = page_order;
-   offset = 0;
+	entry->buf_size   = size;
+	entry->page_order = page_order;
+	offset = 0;
 
    
-   while(entry->buf_count < count) {
-      buf = &entry->buflist[entry->buf_count];
-      buf->idx = dma->buf_count + entry->buf_count;
-      buf->total = alignment;
-      buf->order = order;
-      buf->used = 0;
-
-      DRM_DEBUG("offset : %d\n", offset);
-
-      buf->offset = offset; /* Hrm */
-      buf->bus_address = dev->agp->base + agp_offset + offset;
-      buf->address = (void *)(agp_offset + offset + dev->agp->base);
-      buf->next = NULL;
-      buf->waiting = 0;
-      buf->pending = 0;
-      init_waitqueue_head(&buf->dma_wait);
-      buf->pid = 0;
-
-      buf->dev_private = drm_alloc(sizeof(drm_mga_buf_priv_t), DRM_MEM_BUFS);
-      buf->dev_priv_size = sizeof(drm_mga_buf_priv_t);
+	while(entry->buf_count < count) {
+		buf = &entry->buflist[entry->buf_count];
+		buf->idx = dma->buf_count + entry->buf_count;
+		buf->total = alignment;
+		buf->order = order;
+		buf->used = 0;
+
+		DRM_DEBUG("offset : %ld\n", offset);
+
+		buf->offset = offset; /* Hrm */
+		buf->bus_address = dev->agp->base + agp_offset + offset;
+		buf->address = (void *)(agp_offset + offset + dev->agp->base);
+		buf->next = NULL;
+		buf->waiting = 0;
+		buf->pending = 0;
+		init_waitqueue_head(&buf->dma_wait);
+		buf->pid = 0;
+
+		buf->dev_private = drm_alloc(sizeof(drm_mga_buf_priv_t), DRM_MEM_BUFS);
+		buf->dev_priv_size = sizeof(drm_mga_buf_priv_t);
 
 #if DRM_DMA_HISTOGRAM
-      buf->time_queued = 0;
-      buf->time_dispatched = 0;
-      buf->time_completed = 0;
-      buf->time_freed = 0;
+		buf->time_queued = 0;
+		buf->time_dispatched = 0;
+		buf->time_completed = 0;
+		buf->time_freed = 0;
 #endif
-      offset = offset + alignment;
-      entry->buf_count++;
-      byte_count += PAGE_SIZE << page_order;
+		offset = offset + alignment;
+		entry->buf_count++;
+		byte_count += PAGE_SIZE << page_order;
       
-      DRM_DEBUG("buffer %d @ %p\n",
-	     entry->buf_count, buf->address);
-   }
+		DRM_DEBUG("buffer %d @ %p\n",
+			  entry->buf_count, buf->address);
+	}
    
-   dma->buflist = drm_realloc(dma->buflist,
-			      dma->buf_count * sizeof(*dma->buflist),
-			      (dma->buf_count + entry->buf_count)
-			      * sizeof(*dma->buflist),
-			      DRM_MEM_BUFS);
-   for (i = dma->buf_count; i < dma->buf_count + entry->buf_count; i++)
-     dma->buflist[i] = &entry->buflist[i - dma->buf_count];
+	dma->buflist = drm_realloc(dma->buflist,
+				   dma->buf_count * sizeof(*dma->buflist),
+				   (dma->buf_count + entry->buf_count)
+				   * sizeof(*dma->buflist),
+				   DRM_MEM_BUFS);
+	for (i = dma->buf_count; i < dma->buf_count + entry->buf_count; i++)
+		dma->buflist[i] = &entry->buflist[i - dma->buf_count];
    
-   dma->buf_count  += entry->buf_count;
+	dma->buf_count  += entry->buf_count;
 
-   DRM_DEBUG("dma->buf_count : %d\n", dma->buf_count);
+	DRM_DEBUG("dma->buf_count : %d\n", dma->buf_count);
 
-   dma->byte_count += byte_count;
+	dma->byte_count += byte_count;
 
-   DRM_DEBUG("entry->buf_count : %d\n", entry->buf_count);
+	DRM_DEBUG("entry->buf_count : %d\n", entry->buf_count);
 
-   drm_freelist_create(&entry->freelist, entry->buf_count);
-   for (i = 0; i < entry->buf_count; i++) {
-      drm_freelist_put(dev, &entry->freelist, &entry->buflist[i]);
-   }
+	drm_freelist_create(&entry->freelist, entry->buf_count);
+	for (i = 0; i < entry->buf_count; i++) {
+		drm_freelist_put(dev, &entry->freelist, &entry->buflist[i]);
+	}
    
-   up(&dev->struct_sem);
+	up(&dev->struct_sem);
    
-   request.count = entry->buf_count;
-   request.size  = size;
+	request.count = entry->buf_count;
+	request.size  = size;
    
-   copy_to_user_ret((drm_buf_desc_t *)arg,
-		    &request,
-		    sizeof(request),
-		    -EFAULT);
+	copy_to_user_ret((drm_buf_desc_t *)arg,
+			 &request,
+			 sizeof(request),
+			 -EFAULT);
    
-   atomic_dec(&dev->buf_alloc);
+	atomic_dec(&dev->buf_alloc);
 
-   DRM_DEBUG("count: %d\n", count);
-   DRM_DEBUG("order: %d\n", order);
-   DRM_DEBUG("size: %d\n", size);
-   DRM_DEBUG("agp_offset: %d\n", agp_offset);
-   DRM_DEBUG("alignment: %d\n", alignment);
-   DRM_DEBUG("page_order: %d\n", page_order);
-   DRM_DEBUG("total: %d\n", total);
-   DRM_DEBUG("byte_count: %d\n", byte_count);
+	DRM_DEBUG("count: %d\n", count);
+	DRM_DEBUG("order: %d\n", order);
+	DRM_DEBUG("size: %d\n", size);
+	DRM_DEBUG("agp_offset: %ld\n", agp_offset);
+	DRM_DEBUG("alignment: %d\n", alignment);
+	DRM_DEBUG("page_order: %d\n", page_order);
+	DRM_DEBUG("total: %d\n", total);
+	DRM_DEBUG("byte_count: %d\n", byte_count);
 
-   dma->flags = _DRM_DMA_USE_AGP;
+	dma->flags = _DRM_DMA_USE_AGP;
 
-   DRM_DEBUG("dma->flags : %lx\n", dma->flags);
+	DRM_DEBUG("dma->flags : %x\n", dma->flags);
 
-   return 0;
+	return 0;
 }
 
 int mga_addbufs_pci(struct inode *inode, struct file *filp, unsigned int cmd,
@@ -362,17 +361,17 @@ int mga_addbufs_pci(struct inode *inode, struct file *filp, unsigned int cmd,
 int mga_addbufs(struct inode *inode, struct file *filp, unsigned int cmd,
 		unsigned long arg)
 {
-   drm_buf_desc_t	 request;
+	drm_buf_desc_t	 request;
 
-   copy_from_user_ret(&request,
-		      (drm_buf_desc_t *)arg,
-		      sizeof(request),
-		      -EFAULT);
+	copy_from_user_ret(&request,
+			   (drm_buf_desc_t *)arg,
+			   sizeof(request),
+			   -EFAULT);
 
-   if(request.flags & _DRM_AGP_BUFFER)
-     return mga_addbufs_agp(inode, filp, cmd, arg);
-   else
-     return mga_addbufs_pci(inode, filp, cmd, arg);
+	if(request.flags & _DRM_AGP_BUFFER)
+		return mga_addbufs_agp(inode, filp, cmd, arg);
+	else
+		return mga_addbufs_pci(inode, filp, cmd, arg);
 }
 
 int mga_infobufs(struct inode *inode, struct file *filp, unsigned int cmd,
@@ -546,7 +545,7 @@ int mga_mapbufs(struct inode *inode, struct file *filp, unsigned int cmd,
 	spin_lock(&dev->count_lock);
 	if (atomic_read(&dev->buf_alloc)) {
 		spin_unlock(&dev->count_lock);
-	   DRM_DEBUG("Buzy\n");
+		DRM_DEBUG("Busy\n");
 		return -EBUSY;
 	}
 	++dev->buf_use;		/* Can't allocate more after this call */
@@ -558,79 +557,84 @@ int mga_mapbufs(struct inode *inode, struct file *filp, unsigned int cmd,
 			   -EFAULT);
 
 	DRM_DEBUG("mga_mapbufs\n");
-   	DRM_DEBUG("dma->flags : %lx\n", dma->flags);
+   	DRM_DEBUG("dma->flags : %x\n", dma->flags);
    
-   if (request.count >= dma->buf_count) {
-      if(dma->flags & _DRM_DMA_USE_AGP) {
-	 drm_mga_private_t *dev_priv = dev->dev_private;
-	 drm_map_t *map = NULL;
+	if (request.count >= dma->buf_count) {
+		if(dma->flags & _DRM_DMA_USE_AGP) {
+			drm_mga_private_t *dev_priv = dev->dev_private;
+			drm_map_t *map = NULL;
 	 
-	 map = dev->maplist[dev_priv->buffer_map_idx];
-	 if (!map) {
-	    DRM_DEBUG("map is null\n");
-	    retcode = -EINVAL;
-	    goto done;
-	 }
-
-	 DRM_DEBUG("map->offset : %lx\n", map->offset);
-	 DRM_DEBUG("map->size : %lx\n", map->size);
-	 DRM_DEBUG("map->type : %d\n", map->type);
-	 DRM_DEBUG("map->flags : %x\n", map->flags);
-	 DRM_DEBUG("map->handle : %lx\n", map->handle);
-	 DRM_DEBUG("map->mtrr : %d\n", map->mtrr);
-
-	 virtual = do_mmap(filp, 0, map->size, PROT_READ|PROT_WRITE,
-			   MAP_SHARED, (unsigned long)map->offset);
-      } else {
-	      virtual = do_mmap(filp, 0, dma->byte_count,
-				PROT_READ|PROT_WRITE, MAP_SHARED, 0);
-      }
-      if (virtual > -1024UL) {
-	 /* Real error */
-	 DRM_DEBUG("mmap error\n");
-	 retcode = (signed long)virtual;
-	 goto done;
-      }
-      request.virtual = (void *)virtual;
+			map = dev->maplist[dev_priv->buffer_map_idx];
+			if (!map) {
+				DRM_DEBUG("map is null\n");
+				retcode = -EINVAL;
+				goto done;
+			}
+
+			DRM_DEBUG("map->offset : %lx\n", map->offset);
+			DRM_DEBUG("map->size : %lx\n", map->size);
+			DRM_DEBUG("map->type : %d\n", map->type);
+			DRM_DEBUG("map->flags : %x\n", map->flags);
+			DRM_DEBUG("map->handle : %p\n", map->handle);
+			DRM_DEBUG("map->mtrr : %d\n", map->mtrr);
+			down(&current->mm->mmap_sem);
+			virtual = do_mmap(filp, 0, map->size, 
+					  PROT_READ|PROT_WRITE,
+					  MAP_SHARED, 
+					  (unsigned long)map->offset);
+			up(&current->mm->mmap_sem);
+		} else {
+			down(&current->mm->mmap_sem);
+			virtual = do_mmap(filp, 0, dma->byte_count,
+					  PROT_READ|PROT_WRITE, MAP_SHARED, 0);
+			up(&current->mm->mmap_sem);
+		}
+		if (virtual > -1024UL) {
+			/* Real error */
+			DRM_DEBUG("mmap error\n");
+			retcode = (signed long)virtual;
+			goto done;
+		}
+		request.virtual = (void *)virtual;
       
-      for (i = 0; i < dma->buf_count; i++) {
-	 if (copy_to_user(&request.list[i].idx,
-			  &dma->buflist[i]->idx,
-			  sizeof(request.list[0].idx))) {
-	    retcode = -EFAULT;
-	    goto done;
-	 }
-	 if (copy_to_user(&request.list[i].total,
-			  &dma->buflist[i]->total,
-			  sizeof(request.list[0].total))) {
-	    retcode = -EFAULT;
-	    goto done;
-	 }
-	 if (copy_to_user(&request.list[i].used,
-			  &zero,
-			  sizeof(zero))) {
-	    retcode = -EFAULT;
-	    goto done;
-	 }
-	 address = virtual + dma->buflist[i]->offset;
-	 if (copy_to_user(&request.list[i].address,
-			  &address,
-			  sizeof(address))) {
-	    retcode = -EFAULT;
-	    goto done;
-	 }
-      }
-   }
-   done:
-   request.count = dma->buf_count;
-   DRM_DEBUG("%d buffers, retcode = %d\n", request.count, retcode);
+		for (i = 0; i < dma->buf_count; i++) {
+			if (copy_to_user(&request.list[i].idx,
+					 &dma->buflist[i]->idx,
+					 sizeof(request.list[0].idx))) {
+				retcode = -EFAULT;
+				goto done;
+			}
+			if (copy_to_user(&request.list[i].total,
+					 &dma->buflist[i]->total,
+					 sizeof(request.list[0].total))) {
+				retcode = -EFAULT;
+				goto done;
+			}
+			if (copy_to_user(&request.list[i].used,
+					 &zero,
+					 sizeof(zero))) {
+				retcode = -EFAULT;
+				goto done;
+			}
+			address = virtual + dma->buflist[i]->offset;
+			if (copy_to_user(&request.list[i].address,
+					 &address,
+					 sizeof(address))) {
+				retcode = -EFAULT;
+				goto done;
+			}
+		}
+	}
+ done:
+	request.count = dma->buf_count;
+	DRM_DEBUG("%d buffers, retcode = %d\n", request.count, retcode);
    
-   copy_to_user_ret((drm_buf_map_t *)arg,
-		    &request,
-		    sizeof(request),
-		    -EFAULT);
+	copy_to_user_ret((drm_buf_map_t *)arg,
+			 &request,
+			 sizeof(request),
+			 -EFAULT);
 
-   DRM_DEBUG("retcode : %d\n", retcode);
+	DRM_DEBUG("retcode : %d\n", retcode);
 
-   return retcode;
+	return retcode;
 }
diff --git a/linux/mga_clear.c b/linux/mga_clear.c
deleted file mode 100644
index 05895415..00000000
--- a/linux/mga_clear.c
+++ /dev/null
@@ -1,417 +0,0 @@
-/* mga_state.c -- State support for mga g200/g400 -*- linux-c -*-
- *
- * Created: February 2000 by keithw@precisioninsight.com
- *
- * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- * 
- * The above copyright notice and this permission notice (including the next
- * paragraph) shall be included in all copies or substantial portions of the
- * Software.
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
- * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
- * DEALINGS IN THE SOFTWARE.
- *
- * Authors: 
- *           Keith Whitwell <keithw@precisioninsight.com>
- *
- */
- 
-#define __NO_VERSION__
-#include "drmP.h"
-#include "mga_drv.h"
-#include "mgareg_flags.h"
-#include "mga_dma.h"
-#include "mga_state.h"
-
-#define MGA_CLEAR_CMD (DC_opcod_trap | DC_arzero_enable | 		\
-		       DC_sgnzero_enable | DC_shftzero_enable | 	\
-		       (0xC << DC_bop_SHIFT) | DC_clipdis_enable | 	\
-		       DC_solid_enable | DC_transc_enable)
-	  
-
-#define MGA_COPY_CMD (DC_opcod_bitblt | DC_atype_rpl | DC_linear_xy |	\
-		      DC_solid_disable | DC_arzero_disable | 		\
-		      DC_sgnzero_enable | DC_shftzero_enable | 		\
-		      (0xC << DC_bop_SHIFT) | DC_bltmod_bfcol | 	\
-		      DC_pattern_disable | DC_transc_disable | 		\
-		      DC_clipdis_enable)				\
-
-
-
-/* Build and queue a TT_GENERAL secondary buffer to do the clears.
- * With Jeff's ringbuffer idea, it might make sense if there are only
- * one or two cliprects to emit straight to the primary buffer.
- */
-static int mgaClearBuffers(drm_device_t *dev,
-			   int clear_color,
-			   int clear_depth,
-			   int flags)
-{
-	int cmd, i;	
-	drm_device_dma_t *dma = dev->dma;
-	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;   
-	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
-	xf86drmClipRectRec *pbox = sarea_priv->boxes;
-	int nbox = sarea_priv->nbox;
-	drm_buf_t *buf;
-	drm_dma_t d;
-	int order = 10;		/* ??? what orders do we have ???*/
-	DMALOCALS;
-
-
-	if (!nbox) 
-		return -EINVAL;
-
-	if ( dev_priv->sgram ) 
-		cmd = MGA_CLEAR_CMD | DC_atype_blk;
-	else
-		cmd = MGA_CLEAR_CMD | DC_atype_rstr;
-	    
-	buf = drm_freelist_get(&dma->bufs[order].freelist, _DRM_DMA_WAIT);
-
-
-	DMAGETPTR( buf );
-
-	for (i = 0 ; i < nbox ; i++) {
-		unsigned int height = pbox[i].y2 - pbox[i].y1;
-		
-		/* Is it necessary to be this paranoid?  I don't think so.
-		if (pbox[i].x1 > dev_priv->width) continue;
-		if (pbox[i].y1 > dev_priv->height) continue;
-		if (pbox[i].x2 > dev_priv->width) continue;
-		if (pbox[i].y2 > dev_priv->height) continue;
-		if (pbox[i].x2 <= pbox[i].x1) continue;
-		if (pbox[i].y2 <= pbox[i].x1) continue;
-		 */
-
-		DMAOUTREG(MGAREG_YDSTLEN, (pbox[i].y1<<16)|height);
-		DMAOUTREG(MGAREG_FXBNDRY, (pbox[i].x2<<16)|pbox[i].x1);
-
-		if ( flags & MGA_CLEAR_FRONT ) {	    
-			DMAOUTREG(MGAREG_FCOL, clear_color);
-			DMAOUTREG(MGAREG_DSTORG, dev_priv->frontOrg);
-			DMAOUTREG(MGAREG_DWGCTL+MGAREG_MGA_EXEC, cmd );
-		}
-
-		if ( flags & MGA_CLEAR_BACK ) {
-			DMAOUTREG(MGAREG_FCOL, clear_color);
-			DMAOUTREG(MGAREG_DSTORG, dev_priv->backOrg);
-			DMAOUTREG(MGAREG_DWGCTL+MGAREG_MGA_EXEC, cmd );
-		}
-
-		if ( flags & MGA_CLEAR_DEPTH ) 
-		{
-			DMAOUTREG(MGAREG_FCOL, clear_depth);
-			DMAOUTREG(MGAREG_DSTORG, dev_priv->depthOrg);
-			DMAOUTREG(MGAREG_DWGCTL+MGAREG_MGA_EXEC, cmd );
-		}
-	}
-
-	DMAADVANCE( buf );
-
-	/* Make sure we restore the 3D state next time.
-	 */
-	sarea_priv->dirty |= MGASAREA_NEW_CONTEXT;
-
-	((drm_mga_buf_priv_t *)buf->dev_private)->dma_type = MGA_DMA_GENERAL;
-
-	d.context = DRM_KERNEL_CONTEXT;
-	d.send_count = 1;
-	d.send_indices = &buf->idx;
-	d.send_sizes = &buf->used;
-   	d.flags = 0;
-	d.request_count = 0;
-	d.request_size = 0;
-	d.request_indices = NULL;
-	d.request_sizes = NULL;
-	d.granted_count = 0;	   
-
-      	atomic_inc(&dev_priv->pending_bufs);
-   	if((drm_dma_enqueue(dev, &d)) != 0) 
-     		atomic_dec(&dev_priv->pending_bufs);
-	mga_dma_schedule(dev, 1);
-   	return 0;
-}
-
-int mgaSwapBuffers(drm_device_t *dev, int flags) 
-{
-	drm_device_dma_t *dma = dev->dma;
-	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
-	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
-	xf86drmClipRectRec *pbox = sarea_priv->boxes;
-	int nbox = sarea_priv->nbox;
-	drm_buf_t *buf;
-	drm_dma_t d;
-	int order = 10;		/* ??? */
-	int i;
-	DMALOCALS;	
-
-	if (!nbox) 
-		return -EINVAL;
-
-	buf = drm_freelist_get(&dma->bufs[order].freelist, _DRM_DMA_WAIT);
-
-	DMAGETPTR(buf);
-
-	DMAOUTREG(MGAREG_DSTORG, dev_priv->frontOrg);
-	DMAOUTREG(MGAREG_MACCESS, dev_priv->mAccess);
-	DMAOUTREG(MGAREG_SRCORG, dev_priv->backOrg);
-	DMAOUTREG(MGAREG_AR5, dev_priv->stride); /* unnecessary? */
-	DMAOUTREG(MGAREG_DWGCTL, MGA_COPY_CMD); 
-	     
-	for (i = 0 ; i < nbox; i++) {
-		unsigned int h = pbox[i].y2 - pbox[i].y1;
-		unsigned int start = pbox[i].y1 * dev_priv->stride;
-
-		/*
-		if (pbox[i].x1 > dev_priv->width) continue;
-		if (pbox[i].y1 > dev_priv->height) continue;
-		if (pbox[i].x2 > dev_priv->width) continue;
-		if (pbox[i].y2 > dev_priv->height) continue;
-		if (pbox[i].x2 <= pbox[i].x1) continue;
-		if (pbox[i].y2 <= pbox[i].x1) continue;		
-		*/
-
-		DMAOUTREG(MGAREG_AR0, start + pbox[i].x2 - 1);
-		DMAOUTREG(MGAREG_AR3, start + pbox[i].x1);		
-		DMAOUTREG(MGAREG_FXBNDRY, pbox[i].x1|((pbox[i].x2 - 1)<<16));
-		DMAOUTREG(MGAREG_YDSTLEN+MGAREG_MGA_EXEC, (pbox[i].y1<<16)|h);
-	}
-  
-	DMAOUTREG(MGAREG_SRCORG, 0);
-	DMAADVANCE( buf );
-
-	/* Make sure we restore the 3D state next time.
-	 */
-	sarea_priv->dirty |= MGASAREA_NEW_CONTEXT;
-
-	((drm_mga_buf_priv_t *)buf->dev_private)->dma_type = MGA_DMA_GENERAL;
-
-	d.context = DRM_KERNEL_CONTEXT;
-	d.send_count = 1;
-	d.send_indices = &buf->idx;
-	d.send_sizes = &buf->used;
-   	d.flags = 0;
-	d.request_count = 0;
-	d.request_size = 0;
-	d.request_indices = NULL;
-	d.request_sizes = NULL;
-	d.granted_count = 0;	 
-
-   	atomic_inc(&dev_priv->pending_bufs);
-      	if((drm_dma_enqueue(dev, &d)) != 0) 
-     		atomic_dec(&dev_priv->pending_bufs);
-	mga_dma_schedule(dev, 1);
-   	return 0;
-}
-
-
-static int mgaIload(drm_device_t *dev, drm_mga_iload_t *args)
-{
-	drm_device_dma_t *dma = dev->dma;
-   	drm_buf_t *buf = dma->buflist[ args->idx ];
-	drm_mga_buf_priv_t *buf_priv = (drm_mga_buf_priv_t *)buf->dev_private;
-   	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
-	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
-	drm_dma_t d;
-   	int pixperdword;
-   	
-   	buf_priv->dma_type = MGA_DMA_ILOAD;
-   	buf_priv->boxes[0].y1 = args->texture.y1;
-   	buf_priv->boxes[0].y2 = args->texture.y2;
-   	buf_priv->boxes[0].x1 = args->texture.x1;
-   	buf_priv->boxes[0].x2 = args->texture.x2;
-   	buf_priv->ContextState[MGA_CTXREG_DSTORG] = args->destOrg;
-   	buf_priv->ContextState[MGA_CTXREG_MACCESS] = args->mAccess;
-   	buf_priv->ServerState[MGA_2DREG_PITCH] = args->pitch;
-	buf_priv->nbox = 1;   
-   	sarea_priv->dirty |= (MGASAREA_NEW_CONTEXT | MGASAREA_NEW_2D);
-   	switch((args->mAccess & 0x00000003)) {
-	 	case 0:
-	   		pixperdword = 4;
-	   	break;
-	 	case 1:
-	   		pixperdword = 2;
-	   	break;
-	 	case 2:
-	   		pixperdword = 1;
-	  	break;
-		default:
-	   		DRM_ERROR("Invalid maccess value passed"
-				  " to mgaIload\n");
-	   	return -EINVAL;
-	}
-   	buf->used = ((args->texture.y2 - args->texture.y1) *
-		     (args->texture.x2 - args->texture.x1) /
-		     pixperdword);
-   	DRM_DEBUG("buf->used : %d\n", buf->used);
-	d.context = DRM_KERNEL_CONTEXT;
-	d.send_count = 1;
-	d.send_indices = &buf->idx;
-	d.send_sizes = &buf->used;
-	d.flags = 0;
-	d.request_count = 0;
-	d.request_size = 0;
-	d.request_indices = NULL;
-	d.request_sizes = NULL;
-	d.granted_count = 0;	 
-   
-      	atomic_inc(&dev_priv->pending_bufs);
-      	if((drm_dma_enqueue(dev, &d)) != 0) 
-     		atomic_dec(&dev_priv->pending_bufs);
-	mga_dma_schedule(dev, 1);
-
-	return 0; 
-}
-
-
-/* Necessary?  Not necessary??
- */
-static int check_lock(void)
-{
-	return 1;
-}
-
-int mga_clear_bufs(struct inode *inode, struct file *filp,
-		   unsigned int cmd, unsigned long arg)
-{
-	drm_file_t *priv = filp->private_data;
-	drm_device_t *dev = priv->dev;
-	drm_mga_clear_t clear;
-	int retcode;
-   
-	copy_from_user_ret(&clear, (drm_mga_clear_t *)arg,
-			   sizeof(clear), -EFAULT);
-   
-/*  	if (!check_lock( dev )) */
-/*  		return -EIEIO; */
-		
-	retcode = mgaClearBuffers(dev, clear.clear_color,
-				  clear.clear_depth,
-				  clear.flags);
-   
-	return retcode;
-}
-
-int mga_swap_bufs(struct inode *inode, struct file *filp,
-		  unsigned int cmd, unsigned long arg)
-{
-	drm_file_t *priv = filp->private_data;
-	drm_device_t *dev = priv->dev;
-	drm_mga_swap_t swap;
-	int retcode = 0;
-
-/*  	if (!check_lock( dev )) */
-/*  		return -EIEIO; */
-   
-	copy_from_user_ret(&swap, (drm_mga_swap_t *)arg,
-			   sizeof(swap), -EFAULT);
-   
-	retcode = mgaSwapBuffers(dev, swap.flags);
-   
-	return retcode;
-}
-
-int mga_iload(struct inode *inode, struct file *filp,
-	      unsigned int cmd, unsigned long arg)
-{
-	drm_file_t *priv = filp->private_data;
-	drm_device_t *dev = priv->dev;
-	drm_mga_iload_t iload;
-	int retcode = 0;
-
-/*  	if (!check_lock( dev )) */
-/*  		return -EIEIO; */
-
-	copy_from_user_ret(&iload, (drm_mga_iload_t *)arg, 
-			   sizeof(iload), -EFAULT);
-   
-	retcode = mgaIload(dev, &iload);
-   
-	return retcode;
-}
-
-
-int mga_dma(struct inode *inode, struct file *filp, unsigned int cmd,
-	    unsigned long arg)
-{
-	drm_file_t	  *priv	    = filp->private_data;
-	drm_device_t	  *dev	    = priv->dev;
-	drm_device_dma_t  *dma	    = dev->dma;
-	int		  retcode   = 0;
-	drm_dma_t	  d;
-
-   	copy_from_user_ret(&d, (drm_dma_t *)arg, sizeof(d), -EFAULT);
-	DRM_DEBUG("%d %d: %d send, %d req\n",
-		  current->pid, d.context, d.send_count, d.request_count);
-
-	/* Per-context queues are unworkable if you are trying to do
-	 * state management from the client.
-	 */
-	d.context = DRM_KERNEL_CONTEXT;
-	d.flags &= ~_DRM_DMA_WHILE_LOCKED;
-
-	/* Maybe multiple buffers is useful for iload...
-	 * But this ioctl is only for *despatching* vertex data...
-	 */
-	if (d.send_count < 0 || d.send_count > 1) {
-		DRM_ERROR("Process %d trying to send %d buffers (max 1)\n",
-			  current->pid, d.send_count);
-		return -EINVAL;
-	}
-
-	
-	/* But it *is* used to request buffers for all types of dma:
-	 */
-	if (d.request_count < 0 || d.request_count > dma->buf_count) {
-		DRM_ERROR("Process %d trying to get %d buffers (of %d max)\n",
-			  current->pid, d.request_count, dma->buf_count);
-		return -EINVAL;
-	}
-
-	if (d.send_count) {
-		int idx = d.send_indices[0];
-		drm_mga_buf_priv_t *buf_priv = dma->buflist[ idx ]->dev_private;
-		drm_mga_private_t *dev_priv = dev->dev_private;
-
-		buf_priv->dma_type = MGA_DMA_VERTEX;
-
-/*         	if (!check_lock( dev )) */
-/*  		        return -EIEIO; */
-
-		/* Snapshot the relevent bits of the sarea... 
-		 */
-		mgaCopyAndVerifyState( dev_priv, buf_priv );
-
-	      	atomic_inc(&dev_priv->pending_bufs);
-		retcode = drm_dma_enqueue(dev, &d);
-	      	if(retcode != 0) 
-     			atomic_dec(&dev_priv->pending_bufs);
-		mga_dma_schedule(dev, 1);
-	}
-	
-	d.granted_count = 0;
-
-	if (!retcode && d.request_count) {
-		retcode = drm_dma_get_buffers(dev, &d);
-	}
-
-	DRM_DEBUG("%d returning, granted = %d\n",
-		  current->pid, d.granted_count);
-	copy_to_user_ret((drm_dma_t *)arg, &d, sizeof(d), -EFAULT);
-
-	return retcode;
-}
diff --git a/linux/mga_context.c b/linux/mga_context.c
index fb0d336f..2459b35b 100644
--- a/linux/mga_context.c
+++ b/linux/mga_context.c
@@ -38,7 +38,7 @@
 static int mga_alloc_queue(drm_device_t *dev)
 {
    	int temp = drm_ctxbitmap_next(dev);
-   	printk("mga_alloc_queue: %d\n", temp);
+   	DRM_DEBUG("mga_alloc_queue: %d\n", temp);
 	return temp;
 }
 
@@ -57,7 +57,7 @@ int mga_context_switch(drm_device_t *dev, int old, int new)
         dev->ctx_start = get_cycles();
 #endif
         
-        printk("Context switch from %d to %d\n", old, new);
+        DRM_DEBUG("Context switch from %d to %d\n", old, new);
 
         if (new == dev->last_context) {
                 clear_bit(0, &dev->context_flag);
@@ -104,7 +104,7 @@ int mga_resctx(struct inode *inode, struct file *filp, unsigned int cmd,
 	drm_ctx_t	ctx;
 	int		i;
 
-	printk("%d\n", DRM_RESERVED_CONTEXTS);
+	DRM_DEBUG("%d\n", DRM_RESERVED_CONTEXTS);
 	copy_from_user_ret(&res, (drm_ctx_res_t *)arg, sizeof(res), -EFAULT);
 	if (res.count >= DRM_RESERVED_CONTEXTS) {
 		memset(&ctx, 0, sizeof(ctx));
@@ -134,11 +134,11 @@ int mga_addctx(struct inode *inode, struct file *filp, unsigned int cmd,
 		ctx.handle = mga_alloc_queue(dev);
 	}
         if (ctx.handle == -1) {
-		printk("Not enough free contexts.\n");
+		DRM_DEBUG("Not enough free contexts.\n");
 				/* Should this return -EBUSY instead? */
 		return -ENOMEM;
 	}
-	printk("%d\n", ctx.handle);
+	DRM_DEBUG("%d\n", ctx.handle);
 	copy_to_user_ret((drm_ctx_t *)arg, &ctx, sizeof(ctx), -EFAULT);
 	return 0;
 }
@@ -170,7 +170,7 @@ int mga_switchctx(struct inode *inode, struct file *filp, unsigned int cmd,
 	drm_ctx_t	ctx;
 
 	copy_from_user_ret(&ctx, (drm_ctx_t *)arg, sizeof(ctx), -EFAULT);
-	printk("%d\n", ctx.handle);
+	DRM_DEBUG("%d\n", ctx.handle);
 	return mga_context_switch(dev, dev->last_context, ctx.handle);
 }
 
@@ -182,7 +182,7 @@ int mga_newctx(struct inode *inode, struct file *filp, unsigned int cmd,
 	drm_ctx_t	ctx;
 
 	copy_from_user_ret(&ctx, (drm_ctx_t *)arg, sizeof(ctx), -EFAULT);
-	printk("%d\n", ctx.handle);
+	DRM_DEBUG("%d\n", ctx.handle);
 	mga_context_switch_complete(dev, ctx.handle);
 
 	return 0;
@@ -194,51 +194,11 @@ int mga_rmctx(struct inode *inode, struct file *filp, unsigned int cmd,
 	drm_file_t	*priv	= filp->private_data;
 	drm_device_t	*dev	= priv->dev;
 	drm_ctx_t	ctx;
-	drm_queue_t	*q;
-	drm_buf_t	*buf;
 
 	copy_from_user_ret(&ctx, (drm_ctx_t *)arg, sizeof(ctx), -EFAULT);
-	printk("%d\n", ctx.handle);
-	if(ctx.handle == DRM_KERNEL_CONTEXT) {
-	   q = dev->queuelist[ctx.handle];
-	   atomic_inc(&q->use_count);
-	   if (atomic_read(&q->use_count) == 1) {
-	      /* No longer in use */
-	      atomic_dec(&q->use_count);
-	      return -EINVAL;
-	   }
-	   atomic_inc(&q->finalization); /* Mark queue in finalization state */
-	   atomic_sub(2, &q->use_count); 
-	   /* Mark queue as unused (pending finalization) */
-	   
-	   while (test_and_set_bit(0, &dev->interrupt_flag)) {
-	      printk("Calling schedule from rmctx\n");
-	      schedule();
-	      if (signal_pending(current)) {
-		 clear_bit(0, &dev->interrupt_flag);
-		 return -EINTR;
-	      }
-	   }
-	   
-	   /* Remove queued buffers */
-	   while ((buf = drm_waitlist_get(&q->waitlist))) {
-	      drm_free_buffer(dev, buf);
-	   }
-	   clear_bit(0, &dev->interrupt_flag);
-	   
-	   /* Wakeup blocked processes */
-	   wake_up_interruptible(&q->read_queue);
-	   wake_up_interruptible(&q->write_queue);
-	   wake_up_interruptible(&q->flush_queue);
-	   
-	   /* Finalization over.  Queue is made
-	    available when both use_count and
-	    finalization become 0, which won't
-	    happen until all the waiting processes
-	    stop waiting. */
-	   atomic_dec(&q->finalization);
-	} else {
-	   drm_ctxbitmap_free(dev, ctx.handle);
+	DRM_DEBUG("%d\n", ctx.handle);
+      	if(ctx.handle != DRM_KERNEL_CONTEXT) {
+	   	drm_ctxbitmap_free(dev, ctx.handle);
 	}
 	
 	return 0;
diff --git a/linux/mga_dma.c b/linux/mga_dma.c
index 2e24e5b4..59cbad6b 100644
--- a/linux/mga_dma.c
+++ b/linux/mga_dma.c
@@ -25,17 +25,15 @@
  *
  * Authors: Rickard E. (Rik) Faith <faith@precisioninsight.com>
  *	    Jeff Hartmann <jhartmann@precisioninsight.com>
+ *	    Keith Whitwell <keithw@precisioninsight.com>
  *
- * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_dma.c,v 1.1 2000/02/11 17:26:07 dawes Exp $
+ * $XFree86$
  *
  */
 
 #define __NO_VERSION__
 #include "drmP.h"
 #include "mga_drv.h"
-#include "mgareg_flags.h"
-#include "mga_dma.h"
-#include "mga_state.h"
 
 #include <linux/interrupt.h>	/* For task queue support */
 
@@ -48,643 +46,599 @@
 #define MGA_WRITE(reg,val) 	do { MGA_DEREF(reg) = val; } while (0)
 
 #define PDEA_pagpxfer_enable 	     0x2
-#define MGA_SYNC_TAG         	     0x423f4200
 
-typedef enum {
-	TT_GENERAL,
-	TT_BLIT,
-	TT_VECTOR,
-	TT_VERTEX
-} transferType_t;
+static int mga_flush_queue(drm_device_t *dev);
 
+static unsigned long mga_alloc_page(drm_device_t *dev)
+{
+	unsigned long address;
+   
+	address = __get_free_page(GFP_KERNEL);
+	if(address == 0UL) {
+		return 0;
+	}
+	atomic_inc(&mem_map[MAP_NR((void *) address)].count);
+	set_bit(PG_locked, &mem_map[MAP_NR((void *) address)].flags);
+   
+	return address;
+}
+
+static void mga_free_page(drm_device_t *dev, unsigned long page)
+{
+	if(page == 0UL) {
+		return;
+	}
+	atomic_dec(&mem_map[MAP_NR((void *) page)].count);
+	clear_bit(PG_locked, &mem_map[MAP_NR((void *) page)].flags);
+	wake_up(&mem_map[MAP_NR((void *) page)].wait);
+	free_page(page);
+	return;
+}
 
 static void mga_delay(void)
 {
    	return;
 }
 
-int mga_dma_cleanup(drm_device_t *dev)
+static void mga_flush_write_combine(void)
 {
-	if(dev->dev_private) {
-		drm_mga_private_t *dev_priv = 
-			(drm_mga_private_t *) dev->dev_private;
-      
-		if(dev_priv->ioremap) {
-			int temp = (dev_priv->warp_ucode_size + 
-				    dev_priv->primary_size + 
-				    PAGE_SIZE - 1) / PAGE_SIZE * PAGE_SIZE;
-
-			drm_ioremapfree((void *) dev_priv->ioremap, temp);
-		}
-
-		drm_free(dev->dev_private, sizeof(drm_mga_private_t), 
-			 DRM_MEM_DRIVER);
-		dev->dev_private = NULL;
-	}
-
-	return 0;
+   	int xchangeDummy;
+   	__asm__ volatile(" push %%eax ; xchg %%eax, %0 ; pop %%eax" : : "m" (xchangeDummy));
+   	__asm__ volatile(" push %%eax ; push %%ebx ; push %%ecx ; push %%edx ;"
+			 " movl $0,%%eax ; cpuid ; pop %%edx ; pop %%ecx ; pop %%ebx ;"
+			 " pop %%eax" : /* no outputs */ :  /* no inputs */ );
 }
 
-static int mga_alloc_kernel_queue(drm_device_t *dev)
+/* These are two age tags that will never be sent to
+ * the hardware */
+#define MGA_BUF_USED 	0xffffffff
+#define MGA_BUF_FREE	0
+
+static void mga_freelist_debug(drm_mga_freelist_t *item)
 {
-	drm_queue_t *queue = NULL;
-				/* Allocate a new queue */
-	down(&dev->struct_sem);
-	
-   	if(dev->queue_count != 0) {
-	   /* Reseting the kernel context here is not
-	    * a race, since it can only happen when that
-	    * queue is empty.
-	    */
-	   queue = dev->queuelist[DRM_KERNEL_CONTEXT];
-	   printk("Kernel queue already allocated\n");
+   	if(item->buf != NULL) {
+     		DRM_DEBUG("buf index : %d\n", item->buf->idx);
 	} else {
-	   queue = drm_alloc(sizeof(*queue), DRM_MEM_QUEUES);
-	   if(!queue) {
-	      up(&dev->struct_sem);
-	      printk("out of memory\n");
-	      return -ENOMEM;
-	   }
-	   ++dev->queue_count;
-	   dev->queuelist = drm_alloc(sizeof(*dev->queuelist), 
-				      DRM_MEM_QUEUES);
-	   if(!dev->queuelist) {
-	      up(&dev->struct_sem);
-	      drm_free(queue, sizeof(*queue), DRM_MEM_QUEUES);
-	      printk("out of memory\n");
-	      return -ENOMEM;
-	   }  
+	   	DRM_DEBUG("Freelist head\n");
 	}
-	   
-	memset(queue, 0, sizeof(*queue));
-	atomic_set(&queue->use_count, 1);
-   	atomic_set(&queue->finalization,  0);
-	atomic_set(&queue->block_count,   0);
-	atomic_set(&queue->block_read,    0);
-	atomic_set(&queue->block_write,   0);
-	atomic_set(&queue->total_queued,  0);
-	atomic_set(&queue->total_flushed, 0);
-	atomic_set(&queue->total_locks,   0);
-
-	init_waitqueue_head(&queue->write_queue);
-	init_waitqueue_head(&queue->read_queue);
-	init_waitqueue_head(&queue->flush_queue);
-
-	queue->flags = 0;
-
-	drm_waitlist_create(&queue->waitlist, dev->dma->buf_count);
-   
-   	dev->queue_slots = 1;
-   	dev->queuelist[DRM_KERNEL_CONTEXT] = queue;
-   	dev->queue_count--;
-	
-	up(&dev->struct_sem);
-	printk("%d (new)\n", dev->queue_count - 1);
-	return DRM_KERNEL_CONTEXT;
+   	DRM_DEBUG("item->age : %x\n", item->age);
+   	DRM_DEBUG("item->next : %p\n", item->next);
+   	DRM_DEBUG("item->prev : %p\n", item->prev);
 }
 
-static int mga_dma_initialize(drm_device_t *dev, drm_mga_init_t *init) {
-	drm_mga_private_t *dev_priv;
-	drm_map_t *prim_map = NULL;
-	drm_map_t *sarea_map = NULL;
-	int temp;
-
-
-	dev_priv = drm_alloc(sizeof(drm_mga_private_t), DRM_MEM_DRIVER);
-	if(dev_priv == NULL) return -ENOMEM;
-	dev->dev_private = (void *) dev_priv;
-
-	printk("dev_private\n");
-
-	memset(dev_priv, 0, sizeof(drm_mga_private_t));
-      	atomic_set(&dev_priv->pending_bufs, 0);
-
-	if((init->reserved_map_idx >= dev->map_count) ||
-	   (init->buffer_map_idx >= dev->map_count)) {
-		mga_dma_cleanup(dev);
-		printk("reserved_map or buffer_map are invalid\n");
-		return -EINVAL;
-	}
+static int mga_freelist_init(drm_device_t *dev)
+{
+      	drm_device_dma_t *dma = dev->dma;
+   	drm_buf_t *buf;
+   	drm_mga_buf_priv_t *buf_priv;
+      	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
+   	drm_mga_freelist_t *item;
+   	int i;
+   
+   	dev_priv->head = drm_alloc(sizeof(drm_mga_freelist_t), DRM_MEM_DRIVER);
+	if(dev_priv->head == NULL) return -ENOMEM;
+   	memset(dev_priv->head, 0, sizeof(drm_mga_freelist_t));
+   	dev_priv->head->age = MGA_BUF_USED;
    
-	if(mga_alloc_kernel_queue(dev) != DRM_KERNEL_CONTEXT) {
-	   mga_dma_cleanup(dev);
-	   DRM_ERROR("Kernel context queue not present\n");
+   	for (i = 0; i < dma->buf_count; i++) {
+	   	buf = dma->buflist[ i ];
+	        buf_priv = buf->dev_private;
+		item = drm_alloc(sizeof(drm_mga_freelist_t),
+				 DRM_MEM_DRIVER);
+	   	if(item == NULL) return -ENOMEM;
+	   	memset(item, 0, sizeof(drm_mga_freelist_t));
+	  	item->age = MGA_BUF_FREE;
+	   	item->prev = dev_priv->head;
+	   	item->next = dev_priv->head->next;
+	   	if(dev_priv->head->next != NULL)
+			dev_priv->head->next->prev = item;
+	   	if(item->next == NULL) dev_priv->tail = item;
+	   	item->buf = buf;
+	   	buf_priv->my_freelist = item;
+		buf_priv->discard = 0;
+	   	dev_priv->head->next = item;
 	}
-
-	dev_priv->reserved_map_idx = init->reserved_map_idx;
-	dev_priv->buffer_map_idx = init->buffer_map_idx;
-	sarea_map = dev->maplist[0];
-	dev_priv->sarea_priv = (drm_mga_sarea_t *) 
-		((u8 *)sarea_map->handle + 
-		 init->sarea_priv_offset);
-	printk("sarea_priv\n");
-
-	/* Scale primary size to the next page */
-	dev_priv->primary_size = ((init->primary_size + PAGE_SIZE - 1) / 
-				  PAGE_SIZE) * PAGE_SIZE;
-	dev_priv->warp_ucode_size = init->warp_ucode_size;
-	dev_priv->chipset = init->chipset;
-	dev_priv->fbOffset = init->fbOffset;
-	dev_priv->backOffset = init->backOffset;
-	dev_priv->depthOffset = init->depthOffset;
-	dev_priv->textureOffset = init->textureOffset;
-	dev_priv->textureSize = init->textureSize;
-	dev_priv->cpp = init->cpp;
-	dev_priv->sgram = init->sgram;
-	dev_priv->stride = init->stride;
-
-	dev_priv->frontOrg = init->frontOrg;
-	dev_priv->backOrg = init->backOrg;
-	dev_priv->depthOrg = init->depthOrg;
-	dev_priv->mAccess = init->mAccess;
-	
    
-	printk("memcpy\n");
-	memcpy(&dev_priv->WarpIndex, &init->WarpIndex, 
-	       sizeof(mgaWarpIndex) * MGA_MAX_WARP_PIPES);
-	printk("memcpy done\n");
-	prim_map = dev->maplist[init->reserved_map_idx];
-	dev_priv->prim_phys_head = dev->agp->base + init->reserved_map_agpstart;
-	temp = init->warp_ucode_size + dev_priv->primary_size;
-	temp = ((temp + PAGE_SIZE - 1) / 
-		PAGE_SIZE) * PAGE_SIZE;
-	printk("temp : %x\n", temp);
-	printk("dev->agp->base: %lx\n", dev->agp->base);
-	printk("init->reserved_map_agpstart: %x\n", init->reserved_map_agpstart);
-
-
-	dev_priv->ioremap = drm_ioremap(dev->agp->base + init->reserved_map_agpstart, 
-					temp);
-	if(dev_priv->ioremap == NULL) {
-		printk("Ioremap failed\n");
-		mga_dma_cleanup(dev);
-		return -ENOMEM;
+   	item = dev_priv->head;
+   	while(item) {
+		mga_freelist_debug(item);
+		item = item->next;
 	}
-
-
-
-	dev_priv->prim_head = (u32 *)dev_priv->ioremap;
-	printk("dev_priv->prim_head : %p\n", dev_priv->prim_head);
-	dev_priv->current_dma_ptr = dev_priv->prim_head;
-	dev_priv->prim_num_dwords = 0;
-	dev_priv->prim_max_dwords = dev_priv->primary_size / 4;
+   	DRM_DEBUG("Head\n");
+   	mga_freelist_debug(dev_priv->head);
+   	DRM_DEBUG("Tail\n");
+   	mga_freelist_debug(dev_priv->tail);
    
-	printk("dma initialization\n");
-
-	/* Private is now filled in, initialize the hardware */
-	{
-		PRIMLOCALS;
-		PRIMRESET( dev_priv );
-		PRIMGETPTR( dev_priv );
-		PRIMOUTREG(MGAREG_DMAPAD, 0);
-		PRIMOUTREG(MGAREG_DMAPAD, 0);
-		PRIMOUTREG(MGAREG_DWGSYNC, 0);
-		PRIMOUTREG(MGAREG_SOFTRAP, 0);
-		PRIMADVANCE( dev_priv );
+   	return 0;
+}
 
-		/* Poll for the first buffer to insure that
-		 * the status register will be correct
-		 */
-	   	printk("phys_head : %lx\n", phys_head);
+static void mga_freelist_cleanup(drm_device_t *dev)
+{
+      	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
+   	drm_mga_freelist_t *item;
+   	drm_mga_freelist_t *prev;
+
+   	item = dev_priv->head;
+   	while(item) {
+	   	prev = item;
+	   	item = item->next;
+	   	drm_free(prev, sizeof(drm_mga_freelist_t), DRM_MEM_DRIVER);
+	}
    
-		MGA_WRITE(MGAREG_DWGSYNC, MGA_SYNC_TAG);
+   	dev_priv->head = dev_priv->tail = NULL;
+}
 
-		while(MGA_READ(MGAREG_DWGSYNC) != MGA_SYNC_TAG) {
-			int i;
-			for(i = 0 ; i < 4096; i++) mga_delay();
+/* Frees dispatch lock */
+static inline void mga_dma_quiescent(drm_device_t *dev)
+{
+	drm_device_dma_t  *dma      = dev->dma;
+	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
+	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	__volatile__ unsigned int *status =
+		(__volatile__ unsigned int *)dev_priv->status_page;
+   	unsigned long end;
+	int i;
+
+	end = jiffies + (HZ*3);
+    	while(1) {
+		if(!test_and_set_bit(0, &dev_priv->dispatch_lock)) {
+			break;
 		}
-
-	   	MGA_WRITE(MGAREG_PRIMADDRESS, phys_head | TT_GENERAL);
-
-		MGA_WRITE(MGAREG_PRIMEND, ((phys_head + num_dwords * 4) | 
-					   PDEA_pagpxfer_enable));
-
-		while(MGA_READ(MGAREG_DWGSYNC) == MGA_SYNC_TAG) {
-			int i;
-			for(i = 0; i < 4096; i++) mga_delay();
+	   	if((signed)(end - jiffies) <= 0) {
+			DRM_ERROR("irqs: %d wanted %d\n", 
+				  atomic_read(&dev->total_irq), 
+				  atomic_read(&dma->total_lost));
+			DRM_ERROR("lockup\n"); 
+			goto out_nolock;
 		}
-
+		for (i = 0 ; i < 2000 ; i++) mga_delay();
 	}
-   
-	printk("dma init was successful\n");
-	return 0;
+	end = jiffies + (HZ*3);
+    	DRM_DEBUG("quiescent status : %x\n", MGA_READ(MGAREG_STATUS));
+    	while((MGA_READ(MGAREG_STATUS) & 0x00030001) != 0x00020000) {
+		if((signed)(end - jiffies) <= 0) {
+			DRM_ERROR("irqs: %d wanted %d\n", 
+				  atomic_read(&dev->total_irq), 
+				  atomic_read(&dma->total_lost));
+			DRM_ERROR("lockup\n"); 
+			goto out_status;
+		}
+		for (i = 0 ; i < 2000 ; i++) mga_delay();	  
+	}
+    	DRM_DEBUG("status[1] : %x last_sync_tag : %x\n", status[1],
+		  dev_priv->last_sync_tag);
+    	sarea_priv->dirty |= MGA_DMA_FLUSH;
+
+out_status:
+    	clear_bit(0, &dev_priv->dispatch_lock);
+out_nolock:
 }
 
-int mga_dma_init(struct inode *inode, struct file *filp,
-		 unsigned int cmd, unsigned long arg)
+#define FREELIST_INITIAL (MGA_DMA_BUF_NR * 2)
+#define FREELIST_COMPARE(age) ((age >> 2))
+
+unsigned int mga_create_sync_tag(drm_device_t *dev)
 {
-	drm_file_t *priv = filp->private_data;
-	drm_device_t *dev = priv->dev;
-	drm_mga_init_t init;
+      	drm_mga_private_t *dev_priv = 
+     		(drm_mga_private_t *) dev->dev_private;
+   	unsigned int temp;
+      	drm_buf_t *buf;
+   	drm_mga_buf_priv_t *buf_priv;
+      	drm_device_dma_t *dma = dev->dma;
+   	int i;
    
-	copy_from_user_ret(&init, (drm_mga_init_t *)arg, sizeof(init), -EFAULT);
+   	dev_priv->sync_tag++;
    
-	switch(init.func) {
-	case MGA_INIT_DMA:
-		return mga_dma_initialize(dev, &init);
-	case MGA_CLEANUP_DMA:
-		return mga_dma_cleanup(dev);
+   	if(dev_priv->sync_tag < FREELIST_INITIAL) {
+	   	dev_priv->sync_tag = FREELIST_INITIAL;
 	}
+   	if(dev_priv->sync_tag > 0x3fffffff) {
+		mga_flush_queue(dev);
+	   	mga_dma_quiescent(dev);
+	    
+	   	for (i = 0; i < dma->buf_count; i++) {
+			buf = dma->buflist[ i ];
+			buf_priv = buf->dev_private;
+			buf_priv->my_freelist->age = MGA_BUF_FREE;
+		}
+	   	
+	   	dev_priv->sync_tag = FREELIST_INITIAL;
+	}
+   	temp = dev_priv->sync_tag << 2;
 
-	return -EINVAL;
-}
-
-#define MGA_ILOAD_CMD (DC_opcod_iload | DC_atype_rpl |          	\
-		       DC_linear_linear | DC_bltmod_bfcol |       	\
-		       (0xC << DC_bop_SHIFT) | DC_sgnzero_enable |	\
-		       DC_shftzero_enable | DC_clipdis_enable)
+	dev_priv->sarea_priv->last_enqueue = temp;
 
-static void __mga_iload_small(drm_device_t *dev,
-			      drm_buf_t *buf,
-			      int use_agp) 
-{
-   	drm_mga_private_t *dev_priv = dev->dev_private;
-   	drm_mga_buf_priv_t *buf_priv = buf->dev_private;
-      	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
-   	unsigned long address = (unsigned long)buf->bus_address;
-	int length = buf->used;
-	int y1 = buf_priv->boxes[0].y1;
-	int x1 = buf_priv->boxes[0].x1;
-	int y2 = buf_priv->boxes[0].y2;
-	int x2 = buf_priv->boxes[0].x2;
-   	int dstorg = buf_priv->ContextState[MGA_CTXREG_DSTORG];
-   	int maccess = buf_priv->ContextState[MGA_CTXREG_MACCESS];
-   	PRIMLOCALS;
-
-	PRIMRESET(dev_priv);		
-   	PRIMGETPTR(dev_priv);
-   
-   	PRIMOUTREG(MGAREG_DSTORG, dstorg | use_agp);
-   	PRIMOUTREG(MGAREG_MACCESS, maccess);
-   	PRIMOUTREG(MGAREG_PITCH, (1 << 15));
-   	PRIMOUTREG(MGAREG_YDST, y1 * (x2 - x1));   
-   	PRIMOUTREG(MGAREG_LEN, 1);
-   	PRIMOUTREG(MGAREG_FXBNDRY, ((x2 - x1) * (y2 - y1) - 1) << 16);
-   	PRIMOUTREG(MGAREG_AR0, (x2 - x1) * (y2 - y1) - 1);
-   	PRIMOUTREG(MGAREG_AR3, 0);
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_DWGCTL+MGAREG_MGA_EXEC, MGA_ILOAD_CMD);
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_SECADDRESS, address | TT_BLIT);
-   	PRIMOUTREG(MGAREG_SECEND, (address + length) | use_agp);
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_DWGSYNC, 0);
-   	PRIMOUTREG(MGAREG_SOFTRAP, 0);
-   	PRIMADVANCE(dev_priv);
-#if 0
-   	/* For now we need to set this in the ioctl */
-	sarea_priv->dirty |= MGASAREA_NEW_CONTEXT;
-#endif
-   	MGA_WRITE(MGAREG_DWGSYNC, MGA_SYNC_TAG);
-   	while(MGA_READ(MGAREG_DWGSYNC) != MGA_SYNC_TAG) ;
-
-      	MGA_WRITE(MGAREG_PRIMADDRESS, dev_priv->prim_phys_head | TT_GENERAL);
-	MGA_WRITE(MGAREG_PRIMEND, (phys_head + num_dwords * 4) | use_agp);   
+   	DRM_DEBUG("sync_tag : %x\n", temp);
+   	return temp;
 }
 
-static void __mga_iload_xy(drm_device_t *dev,
-			   drm_buf_t *buf,
-			   int use_agp) 
+/* Least recently used :
+ * These operations are not atomic b/c they are protected by the 
+ * hardware lock */
+
+drm_buf_t *mga_freelist_get(drm_device_t *dev)
 {
-      	drm_mga_private_t *dev_priv = dev->dev_private;
-   	drm_mga_buf_priv_t *buf_priv = buf->dev_private;
-      	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
-	unsigned long address = (unsigned long)buf->bus_address;
-   	int length = buf->used;
-	int y1 = buf_priv->boxes[0].y1;
-	int x1 = buf_priv->boxes[0].x1;
-	int y2 = buf_priv->boxes[0].y2;
-	int x2 = buf_priv->boxes[0].x2;
-   	int dstorg = buf_priv->ContextState[MGA_CTXREG_DSTORG];
-   	int maccess = buf_priv->ContextState[MGA_CTXREG_MACCESS];
-   	int pitch = buf_priv->ServerState[MGA_2DREG_PITCH];
-   	int width, height;
-   	int texperdword = 0;
-   	PRIMLOCALS;
-   
-   	width = (x2 - x1);
-   	height = (y2 - y1);
-   	switch((maccess & 0x00000003)) {
-	 	case 0:
-	   	texperdword = 4;
-      		break;
-	 	case 1:
-      		texperdword = 2;
-	   	break;
-    		case 2:
-      		texperdword = 1;
-      		break;
-    		default:
-      		DRM_ERROR("Invalid maccess value passed to __mga_iload_xy\n");
-	   	return;  
-	}
-   
-   	x2 = x1 + width;
-   	x2 = (x2 + (texperdword - 1)) & ~(texperdword - 1);
-   	x1 = (x1 + (texperdword - 1)) & ~(texperdword - 1);
-   	width = x2 - x1;
-   
-	PRIMRESET(dev_priv);		
-   	PRIMGETPTR(dev_priv);
-   	PRIMOUTREG(MGAREG_DSTORG, dstorg | use_agp);
-   	PRIMOUTREG(MGAREG_MACCESS, maccess);
-   	PRIMOUTREG(MGAREG_PITCH, pitch);
-   	PRIMOUTREG(MGAREG_YDSTLEN, (y1 << 16) | height);
-   
-   	PRIMOUTREG(MGAREG_FXBNDRY, ((x1+width-1) << 16) | x1);
-   	PRIMOUTREG(MGAREG_AR0, width * height - 1);
-      	PRIMOUTREG(MGAREG_AR3, 0 );
-   	PRIMOUTREG(MGAREG_DWGCTL+MGAREG_MGA_EXEC, MGA_ILOAD_CMD);
-	   	   
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_SECADDRESS, address | TT_BLIT);
-   	PRIMOUTREG(MGAREG_SECEND, (address + length) | use_agp);
-	   
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_DMAPAD, 0);
-   	PRIMOUTREG(MGAREG_DWGSYNC, 0);
-   	PRIMOUTREG(MGAREG_SOFTRAP, 0);
-   	PRIMADVANCE(dev_priv);
-#if 0
-   	/* For now we need to set this in the ioctl */
-	sarea_priv->dirty |= MGASAREA_NEW_CONTEXT;
-#endif
-   	MGA_WRITE(MGAREG_DWGSYNC, MGA_SYNC_TAG);
-   	while(MGA_READ(MGAREG_DWGSYNC) != MGA_SYNC_TAG) ;
-
-      	MGA_WRITE(MGAREG_PRIMADDRESS, dev_priv->prim_phys_head | TT_GENERAL);
-	MGA_WRITE(MGAREG_PRIMEND, (phys_head + num_dwords * 4) | use_agp);
+   	drm_mga_private_t *dev_priv = 
+     		(drm_mga_private_t *) dev->dev_private;
+   	__volatile__ unsigned int *status = 
+     		(__volatile__ unsigned int *)dev_priv->status_page;
+	drm_mga_freelist_t *prev;
+   	drm_mga_freelist_t *next;
+   
+   	if((dev_priv->tail->age >> 2) <= FREELIST_COMPARE(status[1])) {
+		prev = dev_priv->tail->prev;
+	   	next = dev_priv->tail;
+	   	prev->next = NULL;
+	   	next->prev = next->next = NULL;
+	   	dev_priv->tail = prev;
+	   	next->age = MGA_BUF_USED;
+	   	return next->buf;
+	} 
+
+   	return NULL;
 }
 
-static void mga_dma_dispatch_iload(drm_device_t *dev, drm_buf_t *buf)
+int mga_freelist_put(drm_device_t *dev, drm_buf_t *buf)
 {
+      	drm_mga_private_t *dev_priv = 
+     		(drm_mga_private_t *) dev->dev_private;
    	drm_mga_buf_priv_t *buf_priv = buf->dev_private;
-
-	int use_agp = PDEA_pagpxfer_enable;
-	int x1 = buf_priv->boxes[0].x1;
-	int x2 = buf_priv->boxes[0].x2;
-   
-   	if((x2 - x1) < 32) {
-	   	printk("using iload small\n");
-	   	__mga_iload_small(dev, buf, use_agp);
+	drm_mga_freelist_t *prev;
+   	drm_mga_freelist_t *head;
+   	drm_mga_freelist_t *next;
+
+   	if(buf_priv->my_freelist->age == MGA_BUF_USED) {
+		/* Discarded buffer, put it on the tail */
+		next = buf_priv->my_freelist;
+		next->age = MGA_BUF_FREE;
+		prev = dev_priv->tail;
+		prev->next = next;
+		next->prev = prev;
+		next->next = NULL;
+		dev_priv->tail = next;
+		DRM_DEBUG("Discarded\n");
 	} else {
-	   	printk("using iload xy\n");
-	   	__mga_iload_xy(dev, buf, use_agp); 
-	}   
+		/* Normally aged buffer, put it on the head + 1,
+		 * as the real head is a sentinal element
+		 */
+		next = buf_priv->my_freelist;
+		head = dev_priv->head;
+		prev = head->next;
+		head->next = next;
+		prev->prev = next;
+		next->prev = head;
+		next->next = prev;
+	}
+   
+   	return 0;
 }
 
-static void mga_dma_dispatch_vertex(drm_device_t *dev, drm_buf_t *buf)
+static void mga_print_all_primary(drm_device_t *dev)
 {
-   	drm_mga_private_t *dev_priv = dev->dev_private;
-	drm_mga_buf_priv_t *buf_priv = buf->dev_private;
-	unsigned long address = (unsigned long)buf->bus_address;
-	int length = buf->used;
-	int use_agp = PDEA_pagpxfer_enable;
-	int i, count;
-   	PRIMLOCALS;
-
-	PRIMRESET(dev_priv);
-	
-	count = buf_priv->nbox;
-	if (count == 0) 
-		count = 1;
-
-	mgaEmitState( dev_priv, buf_priv );
-
-	for (i = 0 ; i < count ; i++) {		
-		if (i < buf_priv->nbox)
-			mgaEmitClipRect( dev_priv, &buf_priv->boxes[i] );
-
-		PRIMGETPTR(dev_priv);
-		PRIMOUTREG( MGAREG_DMAPAD, 0);
-		PRIMOUTREG( MGAREG_DMAPAD, 0);
-		PRIMOUTREG( MGAREG_SECADDRESS, address | TT_VERTEX);
-		PRIMOUTREG( MGAREG_SECEND, (address + length) | use_agp);
-
-		PRIMOUTREG( MGAREG_DMAPAD, 0);
-		PRIMOUTREG( MGAREG_DMAPAD, 0);
-		PRIMOUTREG( MGAREG_DWGSYNC, 0);
-		PRIMOUTREG( MGAREG_SOFTRAP, 0);
-		PRIMADVANCE(dev_priv);
+      	drm_mga_private_t *dev_priv = dev->dev_private;
+   	drm_mga_prim_buf_t *prim;
+   	int i;
+
+   	DRM_DEBUG("Full list of primarys\n");
+     	for(i = 0; i < MGA_NUM_PRIM_BUFS; i++) {
+		prim = dev_priv->prim_bufs[i];
+		DRM_DEBUG("index : %d num_dwords : %d "
+			  "max_dwords : %d phy_head : %x\n",
+			  prim->idx, prim->num_dwords, 
+			  prim->max_dwords, prim->phys_head);
+		DRM_DEBUG("sec_used : %d swap_pending : %x "
+			  "in_use : %x force_fire : %d\n",
+			  prim->sec_used, prim->swap_pending, 
+			  prim->in_use, atomic_read(&prim->force_fire));
+		DRM_DEBUG("needs_overflow : %d\n", 
+			  atomic_read(&prim->needs_overflow));
 	}
-
-	PRIMGETPTR( dev_priv );
-
-   	MGA_WRITE(MGAREG_DWGSYNC, MGA_SYNC_TAG);
-   	while(MGA_READ(MGAREG_DWGSYNC) != MGA_SYNC_TAG) ;
-
-      	MGA_WRITE(MGAREG_PRIMADDRESS, dev_priv->prim_phys_head | TT_GENERAL);
-	MGA_WRITE(MGAREG_PRIMEND, (phys_head + num_dwords * 4) | use_agp);
+   
+   	DRM_DEBUG("current_idx : %d, next_idx : %d, last_idx : %d\n",
+		  dev_priv->next_prim->idx, dev_priv->last_prim->idx,
+		  dev_priv->current_prim->idx);
 }
 
-
-/* Used internally for the small buffers generated from client state
- * information. 
- */
-static void mga_dma_dispatch_general(drm_device_t *dev, drm_buf_t *buf)
+static int mga_init_primary_bufs(drm_device_t *dev, drm_mga_init_t *init)
 {
    	drm_mga_private_t *dev_priv = dev->dev_private;
-	unsigned long address = (unsigned long)buf->bus_address;
-	int length = buf->used;
-	int use_agp = PDEA_pagpxfer_enable;
-   	PRIMLOCALS;
-
-	PRIMRESET(dev_priv);
-   	PRIMGETPTR(dev_priv);
-
-	PRIMOUTREG( MGAREG_DMAPAD, 0);
-	PRIMOUTREG( MGAREG_DMAPAD, 0);
-	PRIMOUTREG( MGAREG_SECADDRESS, address | TT_GENERAL);
-	PRIMOUTREG( MGAREG_SECEND, (address + length) | use_agp);
-
-	PRIMOUTREG( MGAREG_DMAPAD, 0);
-	PRIMOUTREG( MGAREG_DMAPAD, 0);
-      	PRIMOUTREG( MGAREG_DWGSYNC, 0);
-   	PRIMOUTREG( MGAREG_SOFTRAP, 0);
-   	PRIMADVANCE(dev_priv);
-
-   	MGA_WRITE(MGAREG_DWGSYNC, MGA_SYNC_TAG);
-   	while(MGA_READ(MGAREG_DWGSYNC) != MGA_SYNC_TAG) ;
-
-      	MGA_WRITE(MGAREG_PRIMADDRESS, dev_priv->prim_phys_head | TT_GENERAL);
-	MGA_WRITE(MGAREG_PRIMEND, (phys_head + num_dwords * 4) | use_agp);
-}
+	drm_mga_prim_buf_t *prim_buffer;
+   	int i, temp, size_of_buf;
+   	int offset = init->reserved_map_agpstart;
 
-/* Frees dispatch lock */
-static inline void mga_dma_quiescent(drm_device_t *dev)
-{
-   	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
-
-   	while(1) {
-	   atomic_inc(&dev_priv->dispatch_lock);
-	   if(atomic_read(&dev_priv->dispatch_lock) == 1) {
-	      break;
-	   } else {
-	      atomic_dec(&dev_priv->dispatch_lock);
-	   }
-	}
-	while((MGA_READ(MGAREG_STATUS) & 0x00020001) != 0x00020000) ;
-#if 0
-   MGA_WRITE(MGAREG_DWGSYNC, MGA_SYNC_TAG);
-#endif
-   	while(MGA_READ(MGAREG_DWGSYNC) == MGA_SYNC_TAG) ;
-   	MGA_WRITE(MGAREG_DWGSYNC, MGA_SYNC_TAG);
-	while(MGA_READ(MGAREG_DWGSYNC) != MGA_SYNC_TAG) ;
-   	atomic_dec(&dev_priv->dispatch_lock);
+   	DRM_DEBUG("mga_init_primary_bufs\n");
+   	dev_priv->primary_size = ((init->primary_size + PAGE_SIZE - 1) / 
+				  PAGE_SIZE) * PAGE_SIZE;
+   	DRM_DEBUG("primary_size\n");
+   	size_of_buf = dev_priv->primary_size / MGA_NUM_PRIM_BUFS;
+	dev_priv->warp_ucode_size = init->warp_ucode_size;
+   	dev_priv->prim_bufs = drm_alloc(sizeof(drm_mga_prim_buf_t *) * 
+					(MGA_NUM_PRIM_BUFS + 1), 
+					DRM_MEM_DRIVER);
+   	if(dev_priv->prim_bufs == NULL) {
+		DRM_ERROR("Unable to allocate memory for prim_buf\n");
+		return -ENOMEM;
+	}
+   	DRM_DEBUG("memset\n");
+   	memset(dev_priv->prim_bufs, 
+	       0, sizeof(drm_mga_prim_buf_t *) * (MGA_NUM_PRIM_BUFS + 1));
+   
+   	temp = init->warp_ucode_size + dev_priv->primary_size;
+	temp = ((temp + PAGE_SIZE - 1) / PAGE_SIZE) * PAGE_SIZE;
+	   
+   	DRM_DEBUG("temp : %x\n", temp);
+   	DRM_DEBUG("dev->agp->base: %lx\n", dev->agp->base);
+   	DRM_DEBUG("init->reserved_map_agpstart: %x\n", 
+		  init->reserved_map_agpstart);
+   	DRM_DEBUG("ioremap\n");
+	dev_priv->ioremap = drm_ioremap(dev->agp->base + offset, 
+					temp);
+	if(dev_priv->ioremap == NULL) {
+		DRM_DEBUG("Ioremap failed\n");
+		return -ENOMEM;
+	}
+   	init_waitqueue_head(&dev_priv->wait_queue);
+   
+   	for(i = 0; i < MGA_NUM_PRIM_BUFS; i++) {
+	   	DRM_DEBUG("For loop\n");
+	   	prim_buffer = drm_alloc(sizeof(drm_mga_prim_buf_t), 
+					DRM_MEM_DRIVER);
+	   	if(prim_buffer == NULL) return -ENOMEM;
+	   	DRM_DEBUG("memset\n");
+	   	memset(prim_buffer, 0, sizeof(drm_mga_prim_buf_t));
+	   	prim_buffer->phys_head = offset + dev->agp->base;
+	   	prim_buffer->current_dma_ptr = 
+			prim_buffer->head = 
+			(u32 *) (dev_priv->ioremap + 
+				 offset - 
+				 init->reserved_map_agpstart);
+	   	prim_buffer->num_dwords = 0;
+	   	prim_buffer->max_dwords = size_of_buf / sizeof(u32);
+	   	prim_buffer->max_dwords -= 5; /* Leave room for the softrap */
+	   	prim_buffer->sec_used = 0;
+	   	prim_buffer->idx = i;
+	   	offset = offset + size_of_buf;
+	   	dev_priv->prim_bufs[i] = prim_buffer;
+	   	DRM_DEBUG("Looping\n");
+	}
+	dev_priv->current_prim_idx = 0;
+        dev_priv->next_prim = 
+		dev_priv->last_prim = 
+		dev_priv->current_prim =
+        	dev_priv->prim_bufs[0];
+   	set_bit(0, &dev_priv->current_prim->in_use);
+	DRM_DEBUG("init done\n");
+   	return 0;
 }
 
-/* Keeps dispatch lock held */
-
-static inline int mga_dma_is_ready(drm_device_t *dev)
+void mga_fire_primary(drm_device_t *dev, drm_mga_prim_buf_t *prim)
 {
-	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
+       	drm_mga_private_t *dev_priv = dev->dev_private;
+      	drm_device_dma_t  *dma	    = dev->dma;
+       	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+ 	int use_agp = PDEA_pagpxfer_enable;
+	unsigned long end;
+   	int i;
+   	int next_idx;
+       	PRIMLOCALS;
+   
+    	DRM_DEBUG("mga_fire_primary\n");
+    	dev_priv->last_sync_tag = mga_create_sync_tag(dev); 
+   	dev_priv->last_prim = prim;
    
-	atomic_inc(&dev_priv->dispatch_lock);
-	if(atomic_read(&dev_priv->dispatch_lock) == 1) {
-		/* We got the lock */
-		return 1;
+ 	/* We never check for overflow, b/c there is always room */
+    	PRIMPTR(prim);
+   	if(num_dwords <= 0) {
+		DRM_DEBUG("num_dwords == 0 when dispatched\n");
+		goto out_prim_wait;
+	}
+ 	PRIMOUTREG( MGAREG_DMAPAD, 0);
+ 	PRIMOUTREG( MGAREG_DMAPAD, 0);
+       	PRIMOUTREG( MGAREG_DWGSYNC, dev_priv->last_sync_tag);
+   	PRIMOUTREG( MGAREG_SOFTRAP, 0);
+    	PRIMFINISH(prim);
+
+	end = jiffies + (HZ*3);
+    	if(sarea_priv->dirty & MGA_DMA_FLUSH) {
+		DRM_DEBUG("Dma top flush\n");	   
+		while((MGA_READ(MGAREG_STATUS) & 0x00030001) != 0x00020000) {
+			if((signed)(end - jiffies) <= 0) {
+				DRM_ERROR("irqs: %d wanted %d\n", 
+					  atomic_read(&dev->total_irq), 
+					  atomic_read(&dma->total_lost));
+				DRM_ERROR("lockup in fire primary "
+					  "(Dma Top Flush)\n");
+				goto out_prim_wait;
+			}
+	      
+			for (i = 0 ; i < 4096 ; i++) mga_delay();
+		}
+		sarea_priv->dirty &= ~(MGA_DMA_FLUSH);
 	} else {
-		atomic_dec(&dev_priv->dispatch_lock);
-		return 0;
+		DRM_DEBUG("Status wait\n");
+		while((MGA_READ(MGAREG_STATUS) & 0x00020001) != 0x00020000) {
+			if((signed)(end - jiffies) <= 0) {
+				DRM_ERROR("irqs: %d wanted %d\n", 
+					  atomic_read(&dev->total_irq), 
+					  atomic_read(&dma->total_lost));
+				DRM_ERROR("lockup in fire primary "
+					  "(Status Wait)\n");
+				goto out_prim_wait;
+			}
+	   
+			for (i = 0 ; i < 4096 ; i++) mga_delay();
+		}
 	}
+	
+   	mga_flush_write_combine();
+    	atomic_inc(&dev_priv->pending_bufs);
+	atomic_inc(&dma->total_lost);
+       	MGA_WRITE(MGAREG_PRIMADDRESS, phys_head | TT_GENERAL);
+ 	MGA_WRITE(MGAREG_PRIMEND, (phys_head + num_dwords * 4) | use_agp);
+   	prim->num_dwords = 0;
+    
+   	next_idx = prim->idx + 1;
+    	if(next_idx >= MGA_NUM_PRIM_BUFS) 
+		next_idx = 0;
+
+    	dev_priv->next_prim = dev_priv->prim_bufs[next_idx];
+	return;
+
+ out_prim_wait:
+	prim->num_dwords = 0;
+	prim->sec_used = 0;
+	clear_bit(0, &prim->in_use);
+   	wake_up_interruptible(&dev_priv->wait_queue);
+	clear_bit(0, &prim->swap_pending);
+	clear_bit(0, &dev_priv->dispatch_lock);
+	atomic_dec(&dev_priv->pending_bufs);
 }
 
-static inline int mga_dma_is_ready_no_hold(drm_device_t *dev)
+int mga_advance_primary(drm_device_t *dev)
 {
-	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
+   	DECLARE_WAITQUEUE(entry, current);
+   	drm_mga_private_t *dev_priv = dev->dev_private;
+   	drm_mga_prim_buf_t *prim_buffer;
+   	drm_device_dma_t  *dma      = dev->dma;
+   	int next_prim_idx;
+   	int ret = 0;
    
-	atomic_inc(&dev_priv->dispatch_lock);
-	if(atomic_read(&dev_priv->dispatch_lock) == 1) {
-		/* We got the lock, but free it */
-		atomic_dec(&dev_priv->dispatch_lock);
-		return 1;
-	} else {
-		atomic_dec(&dev_priv->dispatch_lock);
-		return 0;
+   	/* This needs to reset the primary buffer if available,
+	 * we should collect stats on how many times it bites
+	 * it's tail */
+   
+   	next_prim_idx = dev_priv->current_prim_idx + 1;
+   	if(next_prim_idx >= MGA_NUM_PRIM_BUFS)
+     		next_prim_idx = 0;
+   	prim_buffer = dev_priv->prim_bufs[next_prim_idx];
+   	atomic_set(&dev_priv->in_wait, 1);
+   
+      	/* In use is cleared in interrupt handler */
+   
+   	if(test_and_set_bit(0, &prim_buffer->in_use)) {
+	   	add_wait_queue(&dev_priv->wait_queue, &entry);
+	   	for (;;) {
+		   	current->state = TASK_INTERRUPTIBLE;
+		   	mga_dma_schedule(dev, 0);
+		   	if(!test_and_set_bit(0, &prim_buffer->in_use)) break;
+		   	atomic_inc(&dev->total_sleeps);
+		   	atomic_inc(&dma->total_missed_sched);
+		   	mga_print_all_primary(dev);
+		   	DRM_DEBUG("Schedule in advance\n");
+		   	/* Three second delay */
+		   	schedule_timeout(HZ*3);
+		   	if (signal_pending(current)) {
+			   	ret = -ERESTARTSYS;
+			   	break;
+			}
+		}
+	   	current->state = TASK_RUNNING;
+	   	remove_wait_queue(&dev_priv->wait_queue, &entry);
+	   	if(ret) return ret;
 	}
+   	atomic_set(&dev_priv->in_wait, 0);
+   	/* This primary buffer is now free to use */
+   	prim_buffer->current_dma_ptr = prim_buffer->head;
+   	prim_buffer->num_dwords = 0;
+   	prim_buffer->sec_used = 0;
+   	atomic_set(&prim_buffer->needs_overflow, 0);
+   	dev_priv->current_prim = prim_buffer;
+   	dev_priv->current_prim_idx = next_prim_idx;
+   	DRM_DEBUG("Primarys at advance\n");
+   	mga_print_all_primary(dev);
+   	return 0;
 }
 
-static void mga_dma_service(int irq, void *device, struct pt_regs *regs)
+/* More dynamic performance decisions */
+static inline int mga_decide_to_fire(drm_device_t *dev)
 {
-	drm_device_t	 *dev = (drm_device_t *)device;
-	drm_device_dma_t *dma = dev->dma;
    	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
+      	drm_device_dma_t  *dma	    = dev->dma;
+	
+   	if(atomic_read(&dev_priv->next_prim->force_fire))
+	{
+	   	atomic_inc(&dma->total_prio);
+	   	return 1;
+	}
 
-	atomic_dec(&dev_priv->dispatch_lock);   	
-	atomic_inc(&dev->total_irq);
-      	MGA_WRITE(MGAREG_ICLEAR, 0xfa7);
-
-				/* Free previous buffer */
-   	if (test_and_set_bit(0, &dev->dma_flag)) {
-	   	atomic_inc(&dma->total_missed_free);
-	   	return;
+	if (atomic_read(&dev_priv->in_flush) && dev_priv->next_prim->num_dwords)
+	{
+	   	atomic_inc(&dma->total_prio);
+	   	return 1;
 	}
-   	if (dma->this_buffer) {
-	   	drm_free_buffer(dev, dma->this_buffer);
-	   	dma->this_buffer = NULL;
+   
+   	if(atomic_read(&dev_priv->pending_bufs) <= MGA_NUM_PRIM_BUFS - 1) {
+		if(test_bit(0, &dev_priv->next_prim->swap_pending)) {
+			atomic_inc(&dma->total_dmas);
+			return 1;
+		}
 	}
-   	clear_bit(0, &dev->dma_flag);
 
-				/* Dispatch new buffer */
-   	queue_task(&dev->tq, &tq_immediate);
-   	mark_bh(IMMEDIATE_BH);
+   	if(atomic_read(&dev_priv->pending_bufs) <= MGA_NUM_PRIM_BUFS / 2) {
+		if(dev_priv->next_prim->sec_used >= MGA_DMA_BUF_NR / 8) {
+			atomic_inc(&dma->total_hit);
+			return 1;
+		}
+	}
 
+   	if(atomic_read(&dev_priv->pending_bufs) >= MGA_NUM_PRIM_BUFS / 2) {
+		if(dev_priv->next_prim->sec_used >= MGA_DMA_BUF_NR / 4) {
+			atomic_inc(&dma->total_missed_free);
+			return 1;
+		}
+	}
+
+   	atomic_inc(&dma->total_tried);
+   	return 0;
 }
 
-/* Only called by mga_dma_schedule. */
-static int mga_do_dma(drm_device_t *dev, int locked)
+int mga_dma_schedule(drm_device_t *dev, int locked)
 {
-	drm_buf_t	 *buf;
-	int		 retcode = 0;
-	drm_device_dma_t *dma = dev->dma;
       	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
-	drm_mga_buf_priv_t *buf_priv;
+      	drm_device_dma_t  *dma	    = dev->dma;
 
-   	printk("mga_do_dma\n");
-	if (test_and_set_bit(0, &dev->dma_flag)) {
+   	if (test_and_set_bit(0, &dev->dma_flag)) {
 		atomic_inc(&dma->total_missed_dma);
 		return -EBUSY;
 	}
-	
-	if (!dma->next_buffer) {
-		DRM_ERROR("No next_buffer\n");
-		clear_bit(0, &dev->dma_flag);
-		return -EINVAL;
-	}
-
-	buf	= dma->next_buffer;
-
-	printk("context %d, buffer %d\n", buf->context, buf->idx);
-
-	if (buf->list == DRM_LIST_RECLAIM) {
-		drm_clear_next_buffer(dev);
-		drm_free_buffer(dev, buf);
-		clear_bit(0, &dev->dma_flag);
-		return -EINVAL;
-	}
+   
+       	DRM_DEBUG("mga_dma_schedule\n");
 
-	if (!buf->used) {
-		DRM_ERROR("0 length buffer\n");
-		drm_clear_next_buffer(dev);
-		drm_free_buffer(dev, buf);
-		clear_bit(0, &dev->dma_flag);
-		return 0;
-	}
-	
-	if (mga_dma_is_ready(dev) == 0) {
-		clear_bit(0, &dev->dma_flag);
-		return -EBUSY;
+   	if(atomic_read(&dev_priv->in_flush) || 
+	   atomic_read(&dev_priv->in_wait)) {
+		locked = 1;
 	}
    
-	/* Always hold the hardware lock while dispatching.
-	 */
-	if (!locked && !drm_lock_take(&dev->lock.hw_lock->lock,
-				      DRM_KERNEL_CONTEXT)) {
-		atomic_inc(&dma->total_missed_lock);
-		clear_bit(0, &dev->dma_flag);
-		atomic_dec(&dev_priv->dispatch_lock);
-		return -EBUSY;
+   	if (!locked && 
+	    !drm_lock_take(&dev->lock.hw_lock->lock, DRM_KERNEL_CONTEXT)) {
+	   	atomic_inc(&dma->total_missed_lock);
+	   	clear_bit(0, &dev->dma_flag);
+	   	return -EBUSY;
 	}
+   	DRM_DEBUG("I'm locked\n");
 
-   	dma->next_queue	 = dev->queuelist[DRM_KERNEL_CONTEXT];
-	drm_clear_next_buffer(dev);
-	buf->pending	 = 1;
-	buf->waiting	 = 0;
-	buf->list	 = DRM_LIST_PEND;
-
-	buf_priv = buf->dev_private;
-
-   	printk("dispatch!\n");
-	switch (buf_priv->dma_type) {
-	case MGA_DMA_GENERAL:
-		mga_dma_dispatch_general(dev, buf);
-		break;
-	case MGA_DMA_VERTEX:
-		mga_dma_dispatch_vertex(dev, buf);
-		break;
-/*  	case MGA_DMA_SETUP: */
-/*  		mga_dma_dispatch_setup(dev, address, length); */
-/*  		break; */
-	case MGA_DMA_ILOAD:
-		mga_dma_dispatch_iload(dev, buf);
-		break;
-	default:
-		printk("bad buffer type %x in dispatch\n", buf_priv->dma_type);
-		break;
+   
+   	if(!test_and_set_bit(0, &dev_priv->dispatch_lock)) {
+	   	/* Fire dma buffer */
+	   	if(mga_decide_to_fire(dev)) {
+		   	DRM_DEBUG("mga_fire_primary\n");
+		   	DRM_DEBUG("idx :%d\n", dev_priv->next_prim->idx);
+		   	atomic_set(&dev_priv->next_prim->force_fire, 0);
+		   	if(dev_priv->current_prim == dev_priv->next_prim &&
+			   dev_priv->next_prim->num_dwords != 0) {
+				/* Schedule overflow for a later time */
+				atomic_set(
+					&dev_priv->current_prim->needs_overflow,
+					1);
+			}
+		   	mga_fire_primary(dev, dev_priv->next_prim);
+		} else {
+			clear_bit(0, &dev_priv->dispatch_lock);
+		}
+	} else {
+		DRM_DEBUG("I can't get the dispatch lock\n");
 	}
-   	atomic_dec(&dev_priv->pending_bufs);
-
-	drm_free_buffer(dev, dma->this_buffer);
-	dma->this_buffer = buf;
-
-	atomic_add(buf->used, &dma->total_bytes);
-	atomic_inc(&dma->total_dmas);
-
+   	
 	if (!locked) {
 		if (drm_lock_free(dev, &dev->lock.hw_lock->lock,
 				  DRM_KERNEL_CONTEXT)) {
@@ -692,90 +646,251 @@ static int mga_do_dma(drm_device_t *dev, int locked)
 		}
 	}
 
-	clear_bit(0, &dev->dma_flag);
-   
-   if(!atomic_read(&dev_priv->pending_bufs)) {
-      wake_up_interruptible(&dev->queuelist[DRM_KERNEL_CONTEXT]->flush_queue);
-   }
-   
-#if 0
-   wake_up_interruptible(&dev->lock.lock_queue);
-#endif
+   	clear_bit(0, &dev->dma_flag);
    
-	/* We hold the dispatch lock until the interrupt handler
-	 * frees it
-	 */
-	return retcode;
+      	if(atomic_read(&dev_priv->in_flush) == 1 &&
+	   dev_priv->next_prim->num_dwords == 0) {
+	   	/* Everything is on the hardware */
+	   	DRM_DEBUG("Primarys at Flush\n");
+	   	mga_print_all_primary(dev);
+	   	atomic_set(&dev_priv->in_flush, 0);
+	   	wake_up_interruptible(&dev_priv->flush_queue);
+	}
+
+	return 0;
 }
 
-static void mga_dma_schedule_timer_wrapper(unsigned long dev)
+static void mga_dma_service(int irq, void *device, struct pt_regs *regs)
 {
-	mga_dma_schedule((drm_device_t *)dev, 0);
+    	drm_device_t	 *dev = (drm_device_t *)device;
+    	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
+    	drm_mga_prim_buf_t *last_prim_buffer;
+    	__volatile__ unsigned int *status = 
+      		(__volatile__ unsigned int *)dev_priv->status_page;
+    
+    	atomic_inc(&dev->total_irq);
+      	MGA_WRITE(MGAREG_ICLEAR, 0x00000001);
+   	last_prim_buffer = dev_priv->last_prim;
+    	last_prim_buffer->num_dwords = 0;
+    	last_prim_buffer->sec_used = 0;
+      	clear_bit(0, &last_prim_buffer->in_use);
+   	wake_up_interruptible(&dev_priv->wait_queue);
+      	clear_bit(0, &last_prim_buffer->swap_pending);
+      	clear_bit(0, &dev_priv->dispatch_lock);
+      	atomic_dec(&dev_priv->pending_bufs);
+   	dev_priv->sarea_priv->last_dispatch = status[1];
+   	queue_task(&dev->tq, &tq_immediate);
+   	mark_bh(IMMEDIATE_BH);
 }
 
-static void mga_dma_schedule_tq_wrapper(void *dev)
+static void mga_dma_task_queue(void *device)
 {
+   	drm_device_t *dev = (drm_device_t *) device;
+
 	mga_dma_schedule(dev, 0);
 }
 
-int mga_dma_schedule(drm_device_t *dev, int locked)
+int mga_dma_cleanup(drm_device_t *dev)
 {
-	drm_queue_t	 *q;
-	drm_buf_t	 *buf;
-	int		 retcode   = 0;
-	int		 processed = 0;
-	int		 missed;
-	int		 expire	   = 20;
-	drm_device_dma_t *dma	   = dev->dma;
-
-      	printk("mga_dma_schedule\n");
-
-	if (test_and_set_bit(0, &dev->interrupt_flag)) {
-				/* Not reentrant */
-		atomic_inc(&dma->total_missed_sched);
-		return -EBUSY;
-	}
-	missed = atomic_read(&dma->total_missed_sched);
+	if(dev->dev_private) {
+		drm_mga_private_t *dev_priv = 
+			(drm_mga_private_t *) dev->dev_private;
+      
+		if(dev_priv->ioremap) {
+			int temp = (dev_priv->warp_ucode_size + 
+				    dev_priv->primary_size + 
+				    PAGE_SIZE - 1) / PAGE_SIZE * PAGE_SIZE;
 
-again:
-	/* There is only one queue:
-	 */
-	if (!dma->next_buffer && DRM_WAITCOUNT(dev, DRM_KERNEL_CONTEXT)) {
-		q   = dev->queuelist[DRM_KERNEL_CONTEXT];
-		buf = drm_waitlist_get(&q->waitlist);
-		dma->next_buffer = buf;
-		dma->next_queue	 = q;
-		if (buf && buf->list == DRM_LIST_RECLAIM) {
-			drm_clear_next_buffer(dev);
-			drm_free_buffer(dev, buf);
+			drm_ioremapfree((void *) dev_priv->ioremap, temp);
+		}
+	   	if(dev_priv->status_page != NULL) {
+		   	iounmap(dev_priv->status_page);
+		}
+	   	if(dev_priv->real_status_page != 0UL) {
+		   	mga_free_page(dev, dev_priv->real_status_page);
+		}
+	   	if(dev_priv->prim_bufs != NULL) {
+		   	int i;
+		   	for(i = 0; i < MGA_NUM_PRIM_BUFS; i++) {
+			   	if(dev_priv->prim_bufs[i] != NULL) {
+			     		drm_free(dev_priv->prim_bufs[i],
+						 sizeof(drm_mga_prim_buf_t),
+						 DRM_MEM_DRIVER);
+				}
+			}
+		   	drm_free(dev_priv->prim_bufs, sizeof(void *) *
+				 (MGA_NUM_PRIM_BUFS + 1), 
+				 DRM_MEM_DRIVER);
+		}
+		if(dev_priv->head != NULL) {
+		   	mga_freelist_cleanup(dev);
 		}
+
+
+		drm_free(dev->dev_private, sizeof(drm_mga_private_t), 
+			 DRM_MEM_DRIVER);
+		dev->dev_private = NULL;
 	}
 
-	if (dma->next_buffer) {
-		if (!(retcode = mga_do_dma(dev, locked))) 
-			++processed;
+	return 0;
+}
+
+static int mga_dma_initialize(drm_device_t *dev, drm_mga_init_t *init) {
+	drm_mga_private_t *dev_priv;
+	drm_map_t *sarea_map = NULL;
+	int i;
+
+	dev_priv = drm_alloc(sizeof(drm_mga_private_t), DRM_MEM_DRIVER);
+	if(dev_priv == NULL) return -ENOMEM;
+	dev->dev_private = (void *) dev_priv;
+
+	DRM_DEBUG("dev_private\n");
+
+	memset(dev_priv, 0, sizeof(drm_mga_private_t));
+      	atomic_set(&dev_priv->in_flush, 0);
+
+	if((init->reserved_map_idx >= dev->map_count) ||
+	   (init->buffer_map_idx >= dev->map_count)) {
+		mga_dma_cleanup(dev);
+		DRM_DEBUG("reserved_map or buffer_map are invalid\n");
+		return -EINVAL;
 	}
+   
+	dev_priv->reserved_map_idx = init->reserved_map_idx;
+	dev_priv->buffer_map_idx = init->buffer_map_idx;
+	sarea_map = dev->maplist[0];
+	dev_priv->sarea_priv = (drm_mga_sarea_t *) 
+		((u8 *)sarea_map->handle + 
+		 init->sarea_priv_offset);
+	DRM_DEBUG("sarea_priv\n");
 
-	/* Try again if we succesfully dispatched a buffer, or if someone 
-	 * tried to schedule while we were working.
-	 */
-	if (--expire) {
-		if (missed != atomic_read(&dma->total_missed_sched)) {
-			atomic_inc(&dma->total_lost);
-		   	if (mga_dma_is_ready_no_hold(dev)) 
-				goto again;
-		}
+	/* Scale primary size to the next page */
+	dev_priv->chipset = init->chipset;
+	dev_priv->frontOffset = init->frontOffset;
+	dev_priv->backOffset = init->backOffset;
+	dev_priv->depthOffset = init->depthOffset;
+	dev_priv->textureOffset = init->textureOffset;
+	dev_priv->textureSize = init->textureSize;
+	dev_priv->cpp = init->cpp;
+	dev_priv->sgram = init->sgram;
+	dev_priv->stride = init->stride;
 
-		if (processed && mga_dma_is_ready_no_hold(dev)) {
-			atomic_inc(&dma->total_lost);
-			processed = 0;
-			goto again;
-		}
+	dev_priv->mAccess = init->mAccess;
+   	init_waitqueue_head(&dev_priv->flush_queue);
+	dev_priv->WarpPipe = -1;
+
+   	DRM_DEBUG("chipset: %d ucode_size: %d backOffset: %x depthOffset: %x\n",
+		  dev_priv->chipset, dev_priv->warp_ucode_size, 
+		  dev_priv->backOffset, dev_priv->depthOffset);
+   	DRM_DEBUG("cpp: %d sgram: %d stride: %d maccess: %x\n",
+		  dev_priv->cpp, dev_priv->sgram, dev_priv->stride, 
+		  dev_priv->mAccess);
+   
+	memcpy(&dev_priv->WarpIndex, &init->WarpIndex, 
+	       sizeof(drm_mga_warp_index_t) * MGA_MAX_WARP_PIPES);
+
+   	for (i = 0 ; i < MGA_MAX_WARP_PIPES ; i++) 
+		DRM_DEBUG("warp pipe %d: installed: %d phys: %lx size: %x\n",
+			  i, 
+			  dev_priv->WarpIndex[i].installed,
+			  dev_priv->WarpIndex[i].phys_addr,
+			  dev_priv->WarpIndex[i].size);
+
+   	DRM_DEBUG("Doing init prim buffers\n");
+   	if(mga_init_primary_bufs(dev, init) != 0) {
+		DRM_ERROR("Can not initialize primary buffers\n");
+		mga_dma_cleanup(dev);
+		return -ENOMEM;
 	}
-	
-	clear_bit(0, &dev->interrupt_flag);
-	
-	return retcode;
+   	DRM_DEBUG("Done with init prim buffers\n");
+   	dev_priv->real_status_page = mga_alloc_page(dev);
+      	if(dev_priv->real_status_page == 0UL) {
+		mga_dma_cleanup(dev);
+		DRM_ERROR("Can not allocate status page\n");
+		return -ENOMEM;
+	}
+   	DRM_DEBUG("Status page at %lx\n", dev_priv->real_status_page);
+
+   	dev_priv->status_page = 
+		ioremap_nocache(virt_to_bus((void *)dev_priv->real_status_page),
+				PAGE_SIZE);
+
+   	if(dev_priv->status_page == NULL) {
+		mga_dma_cleanup(dev);
+		DRM_ERROR("Can not remap status page\n");
+		return -ENOMEM;
+	}
+
+   	DRM_DEBUG("Status page remapped to %p\n", dev_priv->status_page);
+   	/* Write status page when secend or softrap occurs */
+   	MGA_WRITE(MGAREG_PRIMPTR, 
+		  virt_to_bus((void *)dev_priv->real_status_page) | 0x00000003);
+   
+   	dev_priv->device = pci_find_device(0x102b, 0x0525, NULL);
+   	if(dev_priv->device == NULL) {
+		DRM_ERROR("Could not find pci device for card\n");
+		mga_dma_cleanup(dev);
+		return -EINVAL;
+	}
+   
+   	DRM_DEBUG("dma initialization\n");
+
+	/* Private is now filled in, initialize the hardware */
+	{
+	   	__volatile__ unsigned int *status = 
+			(unsigned int *)dev_priv->status_page;
+		PRIMLOCALS;
+		PRIMGETPTR( dev_priv );
+	   
+	   	dev_priv->last_sync_tag = mga_create_sync_tag(dev);
+	   
+		PRIMOUTREG(MGAREG_DMAPAD, 0);
+		PRIMOUTREG(MGAREG_DMAPAD, 0);
+		PRIMOUTREG(MGAREG_DWGSYNC, dev_priv->last_sync_tag);
+		PRIMOUTREG(MGAREG_SOFTRAP, 0);
+		/* Poll for the first buffer to insure that
+		 * the status register will be correct
+		 */
+	   	DRM_DEBUG("phys_head : %lx\n", (unsigned long)phys_head);
+   		status[1] = 0;
+	   
+		mga_flush_write_combine();
+	   	MGA_WRITE(MGAREG_PRIMADDRESS, phys_head | TT_GENERAL);
+
+		MGA_WRITE(MGAREG_PRIMEND, ((phys_head + num_dwords * 4) | 
+					   PDEA_pagpxfer_enable));
+	   
+	   	while(MGA_READ(MGAREG_DWGSYNC) != dev_priv->last_sync_tag) ;
+	   	DRM_DEBUG("status[0] after initialization : %x\n", status[0]);
+	   	DRM_DEBUG("status[1] after initialization : %x\n", status[1]);
+	}
+
+	if(mga_freelist_init(dev) != 0) {
+	   	DRM_ERROR("Could not initialize freelist\n");
+	   	mga_dma_cleanup(dev);
+	   	return -ENOMEM;
+	}
+   	DRM_DEBUG("dma init was successful\n");
+	return 0;
+}
+
+int mga_dma_init(struct inode *inode, struct file *filp,
+		 unsigned int cmd, unsigned long arg)
+{
+	drm_file_t *priv = filp->private_data;
+	drm_device_t *dev = priv->dev;
+	drm_mga_init_t init;
+   
+	copy_from_user_ret(&init, (drm_mga_init_t *)arg, sizeof(init), -EFAULT);
+   
+	switch(init.func) {
+	case MGA_INIT_DMA:
+		return mga_dma_initialize(dev, &init);
+	case MGA_CLEANUP_DMA:
+		return mga_dma_cleanup(dev);
+	}
+
+	return -EINVAL;
 }
 
 int mga_irq_install(drm_device_t *dev, int irq)
@@ -792,7 +907,7 @@ int mga_irq_install(drm_device_t *dev, int irq)
 	dev->irq = irq;
 	up(&dev->struct_sem);
 	
-	printk("install irq handler %d\n", irq);
+	DRM_DEBUG("install irq handler %d\n", irq);
 
 	dev->context_flag     = 0;
 	dev->interrupt_flag   = 0;
@@ -802,13 +917,11 @@ int mga_irq_install(drm_device_t *dev, int irq)
 	dev->dma->this_buffer = NULL;
 	dev->tq.next	      = NULL;
 	dev->tq.sync	      = 0;
-	dev->tq.routine	      = mga_dma_schedule_tq_wrapper;
+	dev->tq.routine	      = mga_dma_task_queue;
 	dev->tq.data	      = dev;
 
 				/* Before installing handler */
-	MGA_WRITE(MGAREG_ICLEAR, 0xfa7);
 	MGA_WRITE(MGAREG_IEN, 0);
-
    				/* Install handler */
 	if ((retcode = request_irq(dev->irq,
 				   mga_dma_service,
@@ -820,11 +933,9 @@ int mga_irq_install(drm_device_t *dev, int irq)
 		up(&dev->struct_sem);
 		return retcode;
 	}
-
 				/* After installing handler */
-   	MGA_WRITE(MGAREG_ICLEAR, 0xfa7);
+   	MGA_WRITE(MGAREG_ICLEAR, 0x00000001);
 	MGA_WRITE(MGAREG_IEN, 0x00000001);
-
 	return 0;
 }
 
@@ -838,19 +949,13 @@ int mga_irq_uninstall(drm_device_t *dev)
 	up(&dev->struct_sem);
 	
 	if (!irq) return -EINVAL;
-	
-	printk("remove irq handler %d\n", irq);
-
-      	MGA_WRITE(MGAREG_ICLEAR, 0xfa7);
+   	DRM_DEBUG("remove irq handler %d\n", irq);
+      	MGA_WRITE(MGAREG_ICLEAR, 0x00000001);
 	MGA_WRITE(MGAREG_IEN, 0);
-   	MGA_WRITE(MGAREG_ICLEAR, 0xfa7);
-
 	free_irq(irq, dev);
-
 	return 0;
 }
 
-
 int mga_control(struct inode *inode, struct file *filp, unsigned int cmd,
 		  unsigned long arg)
 {
@@ -870,36 +975,64 @@ int mga_control(struct inode *inode, struct file *filp, unsigned int cmd,
 	}
 }
 
-int mga_flush_queue(drm_device_t *dev)
+static int mga_flush_queue(drm_device_t *dev)
 {
    	DECLARE_WAITQUEUE(entry, current);
-	drm_queue_t	  *q = dev->queuelist[DRM_KERNEL_CONTEXT];
-   	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
+  	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
    	int ret = 0;
-   	
-   	printk("mga_flush_queue\n");
-   	if(atomic_read(&dev_priv->pending_bufs) != 0) {
-	   current->state = TASK_INTERRUPTIBLE;
-	   add_wait_queue(&q->flush_queue, &entry);
-	   for (;;) {
-	      	if (!atomic_read(&dev_priv->pending_bufs)) break;
-	      	printk("Calling schedule from flush_queue : %d\n",
-		       atomic_read(&dev_priv->pending_bufs));
-	      	mga_dma_schedule(dev, 1);
-	      	schedule();
-	      	if (signal_pending(current)) {
-		   	ret = -EINTR; /* Can't restart */
-		   	break;
-		}
-	   }
-	   printk("Exited out of schedule from flush_queue\n");
-	   current->state = TASK_RUNNING;
-	   remove_wait_queue(&q->flush_queue, &entry);
+
+   	if(dev_priv == NULL) {
+	   	return 0;
 	}
    
+   	if(dev_priv->next_prim->num_dwords != 0) {
+	   	atomic_set(&dev_priv->in_flush, 1);
+   		current->state = TASK_INTERRUPTIBLE;
+   		add_wait_queue(&dev_priv->flush_queue, &entry);
+   		for (;;) {
+		   	mga_dma_schedule(dev, 0);
+	   		if (atomic_read(&dev_priv->in_flush) == 0) 
+				break;
+		   	atomic_inc(&dev->total_sleeps);
+		   	DRM_DEBUG("Schedule in flush_queue\n");
+	      		schedule_timeout(HZ*3);
+	      		if (signal_pending(current)) {
+		   		ret = -EINTR; /* Can't restart */
+		   		break;
+			}
+		}
+   		current->state = TASK_RUNNING;
+   		remove_wait_queue(&dev_priv->flush_queue, &entry);
+	}
+   	atomic_set(&dev_priv->in_flush, 0);
    	return ret;
 }
 
+/* Must be called with the lock held */
+void mga_reclaim_buffers(drm_device_t *dev, pid_t pid)
+{
+	drm_device_dma_t *dma = dev->dma;
+	int		 i;
+
+	if (!dma) return;
+      	if(dev->dev_private == NULL) return;
+
+        mga_flush_queue(dev);
+
+	for (i = 0; i < dma->buf_count; i++) {
+	   	drm_buf_t *buf = dma->buflist[ i ];
+	   	drm_mga_buf_priv_t *buf_priv = buf->dev_private;
+
+		if (buf->pid == pid) {
+		   	if(buf_priv == NULL) return;
+		   	/* Only buffers that need to get reclaimed ever 
+			 * get set to free */
+			if(buf_priv->my_freelist->age == MGA_BUF_USED) 
+		     		buf_priv->my_freelist->age = MGA_BUF_FREE;
+		}
+	}
+}
+
 int mga_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 	       unsigned long arg)
 {
@@ -916,16 +1049,15 @@ int mga_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 			  current->pid, lock.context);
 		return -EINVAL;
 	}
-
-   	printk("%d (pid %d) requests lock (0x%08x), flags = 0x%08x\n",
-		  lock.context, current->pid, dev->lock.hw_lock->lock,
-		  lock.flags);
-
+   
+   	DRM_DEBUG("%d (pid %d) requests lock (0x%08x), flags = 0x%08x\n",
+	       lock.context, current->pid, dev->lock.hw_lock->lock,
+	       lock.flags);
 
 	if (lock.context < 0) {
 		return -EINVAL;
 	}
-
+   
 	/* Only one queue:
 	 */
 
@@ -948,8 +1080,7 @@ int mga_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 				/* Contention */
 			atomic_inc(&dev->total_sleeps);
 			current->state = TASK_INTERRUPTIBLE;
-		   	current->policy |= SCHED_YIELD;
-		   	printk("Calling lock schedule\n");
+		   	DRM_DEBUG("Calling lock schedule\n");
 			schedule();
 			if (signal_pending(current)) {
 				ret = -ERESTARTSYS;
@@ -962,17 +1093,60 @@ int mga_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 	
 	if (!ret) {
 		if (lock.flags & _DRM_LOCK_QUIESCENT) {
-		   	printk("_DRM_LOCK_QUIESCENT\n");
-		   	ret = mga_flush_queue(dev);
-		   	if(ret != 0) {
-			   drm_lock_free(dev, &dev->lock.hw_lock->lock,
-					 lock.context);
-			} else {
-			   mga_dma_quiescent(dev);
-			}
+		   DRM_DEBUG("_DRM_LOCK_QUIESCENT\n");
+		   mga_flush_queue(dev);
+		   mga_dma_quiescent(dev);
 		}
 	}
-	printk("%d %s\n", lock.context, ret ? "interrupted" : "has lock");
+   
+	DRM_DEBUG("%d %s\n", lock.context, ret ? "interrupted" : "has lock");
 	return ret;
 }
 		
+int mga_flush_ioctl(struct inode *inode, struct file *filp, 
+		    unsigned int cmd, unsigned long arg)
+{
+       	drm_file_t	  *priv	  = filp->private_data;
+    	drm_device_t	  *dev	  = priv->dev;
+	drm_lock_t	  lock;
+      	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
+   	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+      	__volatile__ unsigned int *status = 
+     		(__volatile__ unsigned int *)dev_priv->status_page;
+	int		 i;
+ 
+	copy_from_user_ret(&lock, (drm_lock_t *)arg, sizeof(lock), -EFAULT);
+
+	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("mga_flush_ioctl called without lock held\n");
+		return -EINVAL;
+	}
+
+   	if(lock.flags & _DRM_LOCK_FLUSH || lock.flags & _DRM_LOCK_FLUSH_ALL) {
+		mga_flush_queue(dev);
+
+		if((MGA_READ(MGAREG_STATUS) & 0x00030001) == 0x00020000 &&
+		   status[1] != dev_priv->last_sync_tag) 
+		{
+			DRM_DEBUG("Reseting hardware status\n");
+			MGA_WRITE(MGAREG_DWGSYNC, dev_priv->last_sync_tag);
+
+			while(MGA_READ(MGAREG_DWGSYNC) != 
+			      dev_priv->last_sync_tag) 
+			{
+				for(i = 0; i < 4096; i++) mga_delay();
+			}
+
+			status[1] = 
+				sarea_priv->last_dispatch = 
+				dev_priv->last_sync_tag;
+		} else {
+			sarea_priv->last_dispatch = status[1];
+		}
+	}
+   	if(lock.flags & _DRM_LOCK_QUIESCENT) {
+		mga_dma_quiescent(dev);
+	}
+
+    	return 0;
+}
diff --git a/linux/mga_dma.h b/linux/mga_dma.h
deleted file mode 100644
index 7f1c5795..00000000
--- a/linux/mga_dma.h
+++ /dev/null
@@ -1,112 +0,0 @@
-#ifndef MGA_DMA_H
-#define MGA_DMA_H
-
-#include "mga_drm_public.h"
-
-
-/* Isn't this fun.  This has to be fixed asap by emitting primary
- * dma commands in the 'do_dma' ioctl.
- */
-typedef struct {
-	int dma_type;
-
-   	unsigned int ContextState[MGA_CTX_SETUP_SIZE];
-   	unsigned int ServerState[MGA_2D_SETUP_SIZE];
-   	unsigned int TexState[2][MGA_TEX_SETUP_SIZE];
-   	unsigned int WarpPipe;
-   	unsigned int dirty;
-
-   	unsigned int nbox;
-   	xf86drmClipRectRec boxes[MGA_NR_SAREA_CLIPRECTS];
-} drm_mga_buf_priv_t;
-
-
-#define MGA_DMA_GENERAL 0
-#define MGA_DMA_VERTEX  1
-#define MGA_DMA_SETUP   2
-#define MGA_DMA_ILOAD   3
-
-
-#define DWGREG0 	0x1c00
-#define DWGREG0_END 	0x1dff
-#define DWGREG1		0x2c00
-#define DWGREG1_END	0x2dff
-
-#define ISREG0(r)	(r >= DWGREG0 && r <= DWGREG0_END)
-#define ADRINDEX0(r)	(u8)((r - DWGREG0) >> 2)
-#define ADRINDEX1(r)	(u8)(((r - DWGREG1) >> 2) | 0x80)
-#define ADRINDEX(r)	(ISREG0(r) ? ADRINDEX0(r) : ADRINDEX1(r)) 
-
-
-/* Macros for inserting commands into a secondary dma buffer.
- */
-
-#define DMALOCALS	u8 tempIndex[4]; u32 *dma_ptr; \
-			int outcount, num_dwords;
-
-#define	DMAGETPTR(buf) do {			\
-  dma_ptr = (u32 *)((u8 *)buf->address + buf->used);	\
-  outcount = 0;					\
-  num_dwords = buf->used / 4;			\
-} while(0)
-
-#define DMAADVANCE(buf)	do {			\
-  buf->used = num_dwords * 4;			\
-} while(0)
-
-#define DMAOUTREG(reg, val) do {		\
-  tempIndex[outcount]=ADRINDEX(reg);		\
-  dma_ptr[++outcount] = val;			\
-  if (outcount == 4) {				\
-     outcount = 0;				\
-     dma_ptr[0] = *(u32 *)tempIndex;		\
-     dma_ptr+=5;				\
-     num_dwords += 5;				\
-  }						\
-}while (0)
-
-
-
-#define VERBO 0
-
-
-/* Primary buffer versions of above -- pretty similar really.
- */
-#define PRIMLOCALS	u8 tempIndex[4]; u32 *dma_ptr; u32 phys_head; \
-			int outcount, num_dwords
-
-#define PRIMRESET(dev_priv) do {				\
-	dev_priv->prim_num_dwords = 0;				\
-   	dev_priv->current_dma_ptr = dev_priv->prim_head;	\
-} while (0)
-	
-#define	PRIMGETPTR(dev_priv) do {		\
-	dma_ptr = dev_priv->current_dma_ptr;	\
-	phys_head = dev_priv->prim_phys_head;	\
-	num_dwords = dev_priv->prim_num_dwords;	\
-	outcount = 0;				\
-} while (0)
-
-#define PRIMADVANCE(dev_priv)	do {		\
-	dev_priv->prim_num_dwords = num_dwords;	\
-	dev_priv->current_dma_ptr = dma_ptr;	\
-} while (0)
-
-#define PRIMOUTREG(reg, val) do {					\
-	tempIndex[outcount]=ADRINDEX(reg);				\
-	dma_ptr[1+outcount] = val;					\
-	if( ++outcount == 4) {						\
-		outcount = 0;						\
-		dma_ptr[0] = *(u32 *)tempIndex;				\
-		dma_ptr+=5;						\
-		num_dwords += 5;					\
-	}								\
-	if (VERBO) 							\
-		printk(KERN_INFO 					\
-		       "OUT %x val %x dma_ptr %p nr_dwords %d\n",	\
-		       outcount, ADRINDEX(reg), dma_ptr, 		\
-		       num_dwords);   					\
-}while (0)
-
-
-#endif
diff --git a/linux/mga_drm_public.h b/linux/mga_drm.h
index 8b21df11..12a858e7 100644
--- a/linux/mga_drm_public.h
+++ b/linux/mga_drm.h
@@ -1,4 +1,4 @@
-/* mga_drm_public.h -- Public header for the Matrox g200/g400 driver -*- linux-c -*-
+/* mga_drm.h -- Public header for the Matrox g200/g400 driver -*- linux-c -*-
  * Created: Tue Jan 25 01:50:01 1999 by jhartmann@precisioninsight.com
  *
  * Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
@@ -26,12 +26,17 @@
  * Authors: Jeff Hartmann <jhartmann@precisioninsight.com>
  *          Keith Whitwell <keithw@precisioninsight.com>
  *
- * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drm_public.h,v 1.1 2000/02/11 17:26:07 dawes Exp $
+ * $XFree86$
  */
 
-#ifndef _MGA_DRM_PUBLIC_H_
-#define _MGA_DRM_PUBLIC_H_
+#ifndef _MGA_DRM_H_
+#define _MGA_DRM_H_
 
+/* WARNING: If you change any of these defines, make sure to change the
+ * defines in the Xserver file (xf86drmMga.h)
+ */
+#ifndef _MGA_DEFINES_
+#define _MGA_DEFINES_
 #define MGA_F  0x1		/* fog */
 #define MGA_A  0x2		/* alpha */
 #define MGA_S  0x4		/* specular */
@@ -54,66 +59,18 @@
 #define MGA_WARP_T2GZSA         (MGA_T2|MGA_S|MGA_A)
 #define MGA_WARP_T2GZSAF        (MGA_T2|MGA_S|MGA_F|MGA_A)
 
-
 #define MGA_MAX_G400_PIPES 16
 #define MGA_MAX_G200_PIPES  8	/* no multitex */
 
-
 #define MGA_MAX_WARP_PIPES MGA_MAX_G400_PIPES
 
 #define MGA_CARD_TYPE_G200 1
 #define MGA_CARD_TYPE_G400 2
+#define MGA_FRONT   0x1
+#define MGA_BACK    0x2
+#define MGA_DEPTH   0x4
 
-
-typedef struct _drm_mga_warp_index {
-   	int installed;
-   	unsigned long phys_addr;
-   	int size;
-} mgaWarpIndex;
-
-typedef struct drm_mga_init {
-   	enum { 
-	   	MGA_INIT_DMA = 0x01,
-	       	MGA_CLEANUP_DMA = 0x02
-	} func;
-   	int reserved_map_agpstart;
-   	int reserved_map_idx;
-   	int buffer_map_idx;
-   	int sarea_priv_offset;
-   	int primary_size;
-   	int warp_ucode_size;
-   	int fbOffset;
-   	int backOffset;
-   	int depthOffset;
-   	int textureOffset;
-   	int textureSize;
-   	int cpp;
-   	int stride;
-   	int sgram;
-	int chipset;
-   	mgaWarpIndex WarpIndex[MGA_MAX_WARP_PIPES];
-
-	/* Redundant?
-	 */
-	int frontOrg;
-	int backOrg;
-	int depthOrg;
-	int mAccess;
-} drm_mga_init_t;
-
-typedef struct _xf86drmClipRectRec {
-   	unsigned short x1;
-   	unsigned short y1;
-   	unsigned short x2;
-   	unsigned short y2;
-} xf86drmClipRectRec;
-
-#define MGA_CLEAR_FRONT 0x1
-#define MGA_CLEAR_BACK  0x2
-#define MGA_CLEAR_DEPTH 0x4
-
-
-/* Each context has a state:
+/* 3d state excluding texture units:
  */
 #define MGA_CTXREG_DSTORG   0	/* validated */
 #define MGA_CTXREG_MACCESS  1	
@@ -124,7 +81,8 @@ typedef struct _xf86drmClipRectRec {
 #define MGA_CTXREG_WFLAG     6
 #define MGA_CTXREG_TDUAL0    7
 #define MGA_CTXREG_TDUAL1    8
-#define MGA_CTX_SETUP_SIZE   9
+#define MGA_CTXREG_FCOL      9
+#define MGA_CTX_SETUP_SIZE   10
 
 /* 2d state
  */
@@ -146,33 +104,89 @@ typedef struct _xf86drmClipRectRec {
 #define MGA_TEXREG_HEIGHT     10
 #define MGA_TEX_SETUP_SIZE    11
 
-
 /* What needs to be changed for the current vertex dma buffer?
  */
-#define MGASAREA_NEW_CONTEXT    0x1
-#define MGASAREA_NEW_TEX0       0x2
-#define MGASAREA_NEW_TEX1       0x4
-#define MGASAREA_NEW_PIPE       0x8
-#define MGASAREA_NEW_2D 	0x10
-
+#define MGA_UPLOAD_CTX        0x1
+#define MGA_UPLOAD_TEX0       0x2
+#define MGA_UPLOAD_TEX1       0x4
+#define MGA_UPLOAD_PIPE       0x8
+#define MGA_UPLOAD_TEX0IMAGE  0x10
+#define MGA_UPLOAD_TEX1IMAGE  0x20
+#define MGA_UPLOAD_2D 	      0x40
+#define MGA_WAIT_AGE          0x80 /* handled client-side */
+#define MGA_UPLOAD_CLIPRECTS  0x100 /* handled client-side */
+#define MGA_DMA_FLUSH	      0x200 /* set when someone gets the lock
+                                       quiescent */
+
+/* 64 buffers of 16k each, total 1 meg.
+ */
+#define MGA_DMA_BUF_ORDER     14
+#define MGA_DMA_BUF_SZ        (1<<MGA_DMA_BUF_ORDER)
+#define MGA_DMA_BUF_NR        63
 
-/* Keep this small for testing
+/* Keep these small for testing.
  */
-#define MGA_NR_SAREA_CLIPRECTS 2
+#define MGA_NR_SAREA_CLIPRECTS 8
 
-/* Upto 128 regions.  Minimum region size of 256k.
+/* 2 heaps (1 for card, 1 for agp), each divided into upto 128
+ * regions, subject to a minimum region size of (1<<16) == 64k. 
+ *
+ * Clients may subdivide regions internally, but when sharing between
+ * clients, the region size is the minimum granularity. 
  */
-#define MGA_NR_TEX_REGIONS 128
-#define MGA_MIN_LOG_TEX_GRANULARITY 18
 
-typedef struct {
-   unsigned char next, prev;	
-   unsigned char in_use;	
-   int age;			
-} mgaTexRegion;
+#define MGA_CARD_HEAP 0
+#define MGA_AGP_HEAP  1
+#define MGA_NR_TEX_HEAPS 2
+#define MGA_NR_TEX_REGIONS 16
+#define MGA_LOG_MIN_TEX_REGION_SIZE 16
+#endif
+
+typedef struct _drm_mga_warp_index {
+   	int installed;
+   	unsigned long phys_addr;
+   	int size;
+} drm_mga_warp_index_t;
+
+typedef struct drm_mga_init {
+   	enum { 
+	   	MGA_INIT_DMA = 0x01,
+	       	MGA_CLEANUP_DMA = 0x02
+	} func;
+   	int reserved_map_agpstart;
+   	int reserved_map_idx;
+   	int buffer_map_idx;
+   	int sarea_priv_offset;
+   	int primary_size;
+   	int warp_ucode_size;
+   	int frontOffset;
+   	int backOffset;
+   	int depthOffset;
+   	int textureOffset;
+   	int textureSize;
+        int agpTextureOffset;
+        int agpTextureSize;
+   	int cpp;
+   	int stride;
+   	int sgram;
+	int chipset;
+   	drm_mga_warp_index_t WarpIndex[MGA_MAX_WARP_PIPES];
+	int mAccess;
+} drm_mga_init_t;
+
+/* Warning: if you change the sarea structure, you must change the Xserver
+ * structures as well */
+
+typedef struct _drm_mga_tex_region {
+	unsigned char next, prev;	
+	unsigned char in_use;	
+	int age;			
+} drm_mga_tex_region_t;
 
-typedef struct 
-{
+typedef struct _drm_mga_sarea {
+	/* The channel for communication of state information to the kernel
+	 * on firing a vertex dma buffer.
+	 */
    	unsigned int ContextState[MGA_CTX_SETUP_SIZE];
    	unsigned int ServerState[MGA_2D_SETUP_SIZE];
    	unsigned int TexState[2][MGA_TEX_SETUP_SIZE];
@@ -180,39 +194,68 @@ typedef struct
    	unsigned int dirty;
 
    	unsigned int nbox;
-   	xf86drmClipRectRec boxes[MGA_NR_SAREA_CLIPRECTS];
+   	drm_clip_rect_t boxes[MGA_NR_SAREA_CLIPRECTS];
 
-	/* kernel doesn't touch from here down */
-   	int ctxOwner;		
-	mgaTexRegion texList[MGA_NR_TEX_REGIONS+1];
-	int texAge;	                            
-} drm_mga_sarea_t;	
+
+	/* Information about the most recently used 3d drawable.  The
+	 * client fills in the req_* fields, the server fills in the 
+	 * exported_ fields and puts the cliprects into boxes, above.
+	 *
+	 * The client clears the exported_drawable field before
+	 * clobbering the boxes data.
+	 */
+        unsigned int req_drawable;	 /* the X drawable id */
+	unsigned int req_draw_buffer;	 /* MGA_FRONT or MGA_BACK */
+
+        unsigned int exported_drawable;	 
+	unsigned int exported_index; 
+        unsigned int exported_stamp;	
+        unsigned int exported_buffers;	 
+        unsigned int exported_nfront;
+        unsigned int exported_nback;
+	int exported_back_x, exported_front_x, exported_w;	
+	int exported_back_y, exported_front_y, exported_h;
+   	drm_clip_rect_t exported_boxes[MGA_NR_SAREA_CLIPRECTS];
+   
+	/* Counters for aging textures and for client-side throttling.
+	 */
+        int last_enqueue;	/* last time a buffer was enqueued */
+	int last_dispatch;	/* age of the most recently dispatched buffer */
+	int last_quiescent;     /*  */
 
 
+	/* LRU lists for texture memory in agp space and on the card
+	 */
+	drm_mga_tex_region_t texList[MGA_NR_TEX_HEAPS][MGA_NR_TEX_REGIONS+1];
+	unsigned int texAge[MGA_NR_TEX_HEAPS];
+	
+	/* Mechanism to validate card state.
+	 */
+   	int ctxOwner;
+} drm_mga_sarea_t;	
+
 /* Device specific ioctls:
  */
-typedef struct {
+typedef struct _drm_mga_clear {
 	int clear_color;
 	int clear_depth;
 	int flags;
 } drm_mga_clear_t;
 
-
-typedef struct {
-	int flags;		/* not actually used? */
+typedef struct _drm_mga_swap {
+   	int dummy;
 } drm_mga_swap_t;
 
-typedef struct {
+typedef struct _drm_mga_iload {
+	int idx;
+	int length;
 	unsigned int destOrg;
-	unsigned int mAccess;
-   	unsigned int pitch;
-	xf86drmClipRectRec texture;
-   	int idx;
 } drm_mga_iload_t;
 
+typedef struct _drm_mga_vertex {
+   	int idx;		/* buffer to queue */
+	int used;		/* bytes in use */
+	int discard;		/* client finished with buffer?  */
+} drm_mga_vertex_t;
 
-#define DRM_IOCTL_MGA_INIT    DRM_IOW( 0x40, drm_mga_init_t)
-#define DRM_IOCTL_MGA_SWAP    DRM_IOW( 0x41, drm_mga_swap_t)
-#define DRM_IOCTL_MGA_CLEAR   DRM_IOW( 0x42, drm_mga_clear_t)
-#define DRM_IOCTL_MGA_ILOAD   DRM_IOW( 0x43, drm_mga_iload_t)
 #endif
diff --git a/linux/mga_drv.c b/linux/mga_drv.c
index 8bc25617..48041354 100644
--- a/linux/mga_drv.c
+++ b/linux/mga_drv.c
@@ -54,6 +54,7 @@ static struct file_operations mga_fops = {
 	mmap:	 drm_mmap,
 	read:	 drm_read,
 	fasync:	 drm_fasync,
+   	poll:	 drm_poll,
 };
 
 static struct miscdevice      mga_misc = {
@@ -105,9 +106,11 @@ static drm_ioctl_desc_t	      mga_ioctls[] = {
 	[DRM_IOCTL_NR(DRM_IOCTL_AGP_BIND)]    = { drm_agp_bind,    1, 1 },
 	[DRM_IOCTL_NR(DRM_IOCTL_AGP_UNBIND)]  = { drm_agp_unbind,  1, 1 },
    	[DRM_IOCTL_NR(DRM_IOCTL_MGA_INIT)]    = { mga_dma_init,    1, 1 },
-   	[DRM_IOCTL_NR(DRM_IOCTL_MGA_SWAP)]    = { mga_clear_bufs,  1, 1 },
-   	[DRM_IOCTL_NR(DRM_IOCTL_MGA_CLEAR)]   = { mga_swap_bufs,   1, 1 },
-   	[DRM_IOCTL_NR(DRM_IOCTL_MGA_ILOAD)]   = { mga_iload,       1, 1 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_MGA_SWAP)]    = { mga_swap_bufs,   1, 0 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_MGA_CLEAR)]   = { mga_clear_bufs,  1, 0 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_MGA_ILOAD)]   = { mga_iload,       1, 0 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_MGA_VERTEX)]  = { mga_vertex,      1, 0 },
+   	[DRM_IOCTL_NR(DRM_IOCTL_MGA_FLUSH)]   = { mga_flush_ioctl, 1, 0 },
 };
 
 #define MGA_IOCTL_COUNT DRM_ARRAY_SIZE(mga_ioctls)
@@ -380,6 +383,21 @@ int mga_init(void)
 	drm_proc_init(dev);
 	DRM_DEBUG("doing agp init\n");
 	dev->agp    = drm_agp_init();
+      	if(dev->agp == NULL) {
+	   	DRM_DEBUG("The mga drm module requires the agpgart module"
+		          " to function correctly\nPlease load the agpgart"
+		          " module before you load the mga module\n");
+	   	drm_proc_cleanup();
+	   	misc_deregister(&mga_misc);
+	   	mga_takedown(dev);
+	   	return -ENOMEM;
+	}
+#ifdef CONFIG_MTRR
+   	dev->agp->agp_mtrr = mtrr_add(dev->agp->agp_info.aper_base,
+				      dev->agp->agp_info.aper_size * 1024 * 1024,
+				      MTRR_TYPE_WRCOMB,
+				      1);
+#endif
 	DRM_DEBUG("doing ctxbitmap init\n");
 	if((retcode = drm_ctxbitmap_init(dev))) {
 		DRM_ERROR("Cannot allocate memory for context bitmap.\n");
@@ -416,6 +434,16 @@ void mga_cleanup(void)
 	}
 	drm_ctxbitmap_cleanup(dev);
 	mga_dma_cleanup(dev);
+#ifdef CONFIG_MTRR
+   	if(dev->agp && dev->agp->agp_mtrr) {
+	   	int retval;
+	   	retval = mtrr_del(dev->agp->agp_mtrr, 
+				  dev->agp->agp_info.aper_base,
+				  dev->agp->agp_info.aper_size * 1024*1024);
+	   	DRM_DEBUG("mtrr_del = %d\n", retval);
+	}
+#endif
+
 	mga_takedown(dev);
 	if (dev->agp) {
 		drm_free(dev->agp, sizeof(*dev->agp), DRM_MEM_AGPLISTS);
diff --git a/linux/mga_drv.h b/linux/mga_drv.h
index bc7808b0..43deb613 100644
--- a/linux/mga_drv.h
+++ b/linux/mga_drv.h
@@ -26,12 +26,32 @@
  * Authors: Rickard E. (Rik) Faith <faith@precisioninsight.com>
  * 	    Jeff Hartmann <jhartmann@precisioninsight.com>
  *
- * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_drv.h,v 1.1 2000/02/11 17:26:08 dawes Exp $
+ * $XFree86$
  */
 
 #ifndef _MGA_DRV_H_
 #define _MGA_DRV_H_
-#include "mga_drm_public.h"
+
+typedef struct {
+   	unsigned int num_dwords;
+   	unsigned int max_dwords;
+   	u32 *current_dma_ptr;
+   	u32 *head;
+   	u32 phys_head;
+   	int sec_used;
+   	int idx;
+   	int swap_pending;
+   	u32 in_use;
+   	atomic_t force_fire;
+   	atomic_t needs_overflow;
+} drm_mga_prim_buf_t;
+
+typedef struct _drm_mga_freelist {
+   	unsigned int age;
+   	drm_buf_t *buf;
+   	struct _drm_mga_freelist *next;
+   	struct _drm_mga_freelist *prev;
+} drm_mga_freelist_t;
 
 typedef struct _drm_mga_private {
    	int reserved_map_idx;
@@ -40,7 +60,7 @@ typedef struct _drm_mga_private {
    	int primary_size;
    	int warp_ucode_size;
    	int chipset;
-   	int fbOffset;
+   	int frontOffset;
    	int backOffset;
    	int depthOffset;
    	int textureOffset;
@@ -49,25 +69,32 @@ typedef struct _drm_mga_private {
    	int stride;
    	int sgram;
 	int use_agp;
-   	mgaWarpIndex WarpIndex[MGA_MAX_G400_PIPES];
+   	drm_mga_warp_index_t WarpIndex[MGA_MAX_G400_PIPES];
+	unsigned int WarpPipe;
    	__volatile__ unsigned long softrap_age;
-   	atomic_t dispatch_lock;
+   	u32 dispatch_lock;
+   	atomic_t in_flush;
+   	atomic_t in_wait;
    	atomic_t pending_bufs;
-   	void *ioremap;
-   	u32 *prim_head;
-   	u32 *current_dma_ptr;
-   	u32 prim_phys_head;
-   	int prim_num_dwords;
-   	int prim_max_dwords;
-
+   	unsigned int last_sync_tag;
+   	unsigned int sync_tag;
+   	void *status_page;
+   	unsigned long real_status_page;
+   	u8 *ioremap;
+   	drm_mga_prim_buf_t **prim_bufs;
+   	drm_mga_prim_buf_t *next_prim;
+   	drm_mga_prim_buf_t *last_prim;
+   	drm_mga_prim_buf_t *current_prim;
+   	int current_prim_idx;
+   	struct pci_dev *device;
+   	drm_mga_freelist_t *head;
+   	drm_mga_freelist_t *tail;
+   	wait_queue_head_t flush_queue;	/* Processes waiting until flush    */
+      	wait_queue_head_t wait_queue;	/* Processes waiting until interrupt */
 
 	/* Some validated register values:
 	 */	
-	u32 frontOrg;
-	u32 backOrg;
-	u32 depthOrg;
 	u32 mAccess;
-
 } drm_mga_private_t;
 
 				/* mga_drv.c */
@@ -92,16 +119,18 @@ extern int  mga_control(struct inode *inode, struct file *filp,
 			  unsigned int cmd, unsigned long arg);
 extern int  mga_lock(struct inode *inode, struct file *filp,
 		       unsigned int cmd, unsigned long arg);
-#if 0
-extern void mga_dma_init(drm_device_t *dev);
-extern void mga_dma_cleanup(drm_device_t *dev);
 
-#endif
+/* mga_dma_init does init and release */
 extern int mga_dma_init(struct inode *inode, struct file *filp,
 			unsigned int cmd, unsigned long arg);
 extern int mga_dma_cleanup(drm_device_t *dev);
+extern int mga_flush_ioctl(struct inode *inode, struct file *filp,
+			   unsigned int cmd, unsigned long arg);
 
-/* mga_dma_init does init and release */
+extern unsigned int mga_create_sync_tag(drm_device_t *dev);
+extern drm_buf_t *mga_freelist_get(drm_device_t *dev);
+extern int mga_freelist_put(drm_device_t *dev, drm_buf_t *buf);
+extern int mga_advance_primary(drm_device_t *dev);
 
 
 				/* mga_bufs.c */
@@ -124,6 +153,8 @@ extern int  mga_swap_bufs(struct inode *inode, struct file *filp,
 			  unsigned int cmd, unsigned long arg);
 extern int  mga_iload(struct inode *inode, struct file *filp,
 		      unsigned int cmd, unsigned long arg);
+extern int  mga_vertex(struct inode *inode, struct file *filp,
+		      unsigned int cmd, unsigned long arg);
 				/* mga_context.c */
 extern int  mga_resctx(struct inode *inode, struct file *filp,
 		       unsigned int cmd, unsigned long arg);
@@ -144,13 +175,115 @@ extern int  mga_context_switch(drm_device_t *dev, int old, int new);
 extern int  mga_context_switch_complete(drm_device_t *dev, int new);
 
 
+typedef enum {
+	TT_GENERAL,
+	TT_BLIT,
+	TT_VECTOR,
+	TT_VERTEX
+} transferType_t;
+
+typedef struct {
+   	drm_mga_freelist_t *my_freelist;
+	int discard;
+} drm_mga_buf_priv_t;
+
+#define DWGREG0 	0x1c00
+#define DWGREG0_END 	0x1dff
+#define DWGREG1		0x2c00
+#define DWGREG1_END	0x2dff
+
+#define ISREG0(r)	(r >= DWGREG0 && r <= DWGREG0_END)
+#define ADRINDEX0(r)	(u8)((r - DWGREG0) >> 2)
+#define ADRINDEX1(r)	(u8)(((r - DWGREG1) >> 2) | 0x80)
+#define ADRINDEX(r)	(ISREG0(r) ? ADRINDEX0(r) : ADRINDEX1(r)) 
+
+#define MGA_VERBOSE 0
+#define MGA_NUM_PRIM_BUFS 	8
+
+#define PRIMLOCALS	u8 tempIndex[4]; u32 *dma_ptr; u32 phys_head; \
+			int outcount, num_dwords
+
+#define PRIM_OVERFLOW(dev, dev_priv, length) do {			\
+	drm_mga_prim_buf_t *tmp_buf =					\
+		dev_priv->prim_bufs[dev_priv->current_prim_idx];	\
+	if( tmp_buf->max_dwords - tmp_buf->num_dwords < length ||	\
+	    tmp_buf->sec_used > MGA_DMA_BUF_NR/2) {			\
+		atomic_set(&tmp_buf->force_fire, 1);			\
+		mga_advance_primary(dev);				\
+		mga_dma_schedule(dev, 1);				\
+	} else if( atomic_read(&tmp_buf->needs_overflow)) {		\
+		mga_advance_primary(dev);				\
+		mga_dma_schedule(dev, 1);				\
+	}								\
+} while(0)
+
+#define PRIMGETPTR(dev_priv) do {					\
+	drm_mga_prim_buf_t *tmp_buf =					\
+		dev_priv->prim_bufs[dev_priv->current_prim_idx];	\
+	if(MGA_VERBOSE)							\
+		DRM_DEBUG("PRIMGETPTR in %s\n", __FUNCTION__);		\
+	dma_ptr = tmp_buf->current_dma_ptr;				\
+	num_dwords = tmp_buf->num_dwords;				\
+	phys_head = tmp_buf->phys_head;					\
+	outcount = 0;							\
+} while(0)
+
+#define PRIMPTR(prim_buf) do {					\
+	if(MGA_VERBOSE)						\
+		DRM_DEBUG("PRIMPTR in %s\n", __FUNCTION__);	\
+	dma_ptr = prim_buf->current_dma_ptr;			\
+	num_dwords = prim_buf->num_dwords;			\
+	phys_head = prim_buf->phys_head;			\
+	outcount = 0;						\
+} while(0)
+
+#define PRIMFINISH(prim_buf) do {				\
+	if (MGA_VERBOSE) {					\
+		DRM_DEBUG( "PRIMFINISH in %s\n", __FUNCTION__);	\
+                if (outcount & 3)				\
+                      DRM_DEBUG(" --- truncation\n");	        \
+        }							\
+	prim_buf->num_dwords = num_dwords;			\
+	prim_buf->current_dma_ptr = dma_ptr;			\
+} while(0)
+
+#define PRIMADVANCE(dev_priv)	do {				\
+drm_mga_prim_buf_t *tmp_buf = 					\
+	dev_priv->prim_bufs[dev_priv->current_prim_idx];	\
+	if (MGA_VERBOSE) {					\
+		DRM_DEBUG("PRIMADVANCE in %s\n", __FUNCTION__);	\
+                if (outcount & 3)				\
+                      DRM_DEBUG(" --- truncation\n");	\
+        }							\
+	tmp_buf->num_dwords = num_dwords;      			\
+	tmp_buf->current_dma_ptr = dma_ptr;    			\
+} while (0)
+
+#define PRIMUPDATE(dev_priv)	do {					\
+	drm_mga_prim_buf_t *tmp_buf =					\
+		dev_priv->prim_bufs[dev_priv->current_prim_idx];	\
+	tmp_buf->sec_used++;						\
+} while (0)
+
+#define PRIMOUTREG(reg, val) do {					\
+	tempIndex[outcount]=ADRINDEX(reg);				\
+	dma_ptr[1+outcount] = val;					\
+	if (MGA_VERBOSE)						\
+		DRM_DEBUG("   PRIMOUT %d: 0x%x -- 0x%x\n",		\
+		       num_dwords + 1 + outcount, ADRINDEX(reg), val);	\
+	if( ++outcount == 4) {						\
+		outcount = 0;						\
+		dma_ptr[0] = *(u32 *)tempIndex;				\
+		dma_ptr+=5;						\
+		num_dwords += 5;					\
+	}								\
+}while (0)
+
+/* A reduced set of the mga registers.
+ */
 
 #define MGAREG_MGA_EXEC 			0x0100
-#define MGAREG_AGP_PLL				0x1e4c
 #define MGAREG_ALPHACTRL 			0x2c7c
-#define MGAREG_ALPHASTART 			0x2c70
-#define MGAREG_ALPHAXINC 			0x2c74
-#define MGAREG_ALPHAYINC 			0x2c78
 #define MGAREG_AR0 				0x1c60
 #define MGAREG_AR1 				0x1c64
 #define MGAREG_AR2 				0x1c68
@@ -158,39 +291,16 @@ extern int  mga_context_switch_complete(drm_device_t *dev, int new);
 #define MGAREG_AR4 				0x1c70
 #define MGAREG_AR5 				0x1c74
 #define MGAREG_AR6 				0x1c78
-#define MGAREG_BCOL 				0x1c20
 #define MGAREG_CXBNDRY				0x1c80
 #define MGAREG_CXLEFT 				0x1ca0
 #define MGAREG_CXRIGHT				0x1ca4
 #define MGAREG_DMAPAD 				0x1c54
-#define MGAREG_DR0_Z32LSB 			0x2c50
-#define MGAREG_DR0_Z32MSB 			0x2c54
-#define MGAREG_DR2_Z32LSB 			0x2c60
-#define MGAREG_DR2_Z32MSB 			0x2c64
-#define MGAREG_DR3_Z32LSB 			0x2c68
-#define MGAREG_DR3_Z32MSB 			0x2c6c
-#define MGAREG_DR0 				0x1cc0
-#define MGAREG_DR2 				0x1cc8
-#define MGAREG_DR3 				0x1ccc
-#define MGAREG_DR4 				0x1cd0
-#define MGAREG_DR6 				0x1cd8
-#define MGAREG_DR7 				0x1cdc
-#define MGAREG_DR8 				0x1ce0
-#define MGAREG_DR10 				0x1ce8
-#define MGAREG_DR11 				0x1cec
-#define MGAREG_DR12 				0x1cf0
-#define MGAREG_DR14 				0x1cf8
-#define MGAREG_DR15 				0x1cfc
 #define MGAREG_DSTORG 				0x2cb8
-#define MGAREG_DWG_INDIR_WT 			0x1e80
 #define MGAREG_DWGCTL 				0x1c00
 #define MGAREG_DWGSYNC				0x2c4c
 #define MGAREG_FCOL 				0x1c24
 #define MGAREG_FIFOSTATUS 			0x1e10
 #define MGAREG_FOGCOL 				0x1cf4
-#define MGAREG_FOGSTART 			0x1cc4
-#define MGAREG_FOGXINC				0x1cd4
-#define MGAREG_FOGYINC				0x1ce4
 #define MGAREG_FXBNDRY				0x1c84
 #define MGAREG_FXLEFT 				0x1ca8
 #define MGAREG_FXRIGHT				0x1cac
@@ -198,44 +308,22 @@ extern int  mga_context_switch_complete(drm_device_t *dev, int new);
 #define MGAREG_IEN 				0x1e1c
 #define MGAREG_LEN 				0x1c5c
 #define MGAREG_MACCESS				0x1c04
-#define MGAREG_MCTLWTST 			0x1c08
-#define MGAREG_MEMRDBK				0x1e44
-#define MGAREG_OPMODE 				0x1e54
-#define MGAREG_PAT0 				0x1c10
-#define MGAREG_PAT1 				0x1c14
 #define MGAREG_PITCH 				0x1c8c
 #define MGAREG_PLNWT 				0x1c1c
 #define MGAREG_PRIMADDRESS 			0x1e58
 #define MGAREG_PRIMEND				0x1e5c
 #define MGAREG_PRIMPTR				0x1e50
-#define MGAREG_RST 				0x1e40
 #define MGAREG_SECADDRESS 			0x2c40
 #define MGAREG_SECEND 				0x2c44
 #define MGAREG_SETUPADDRESS 			0x2cd0
 #define MGAREG_SETUPEND 			0x2cd4
-#define MGAREG_SGN 				0x1c58
-#define MGAREG_SHIFT 				0x1c50
 #define MGAREG_SOFTRAP				0x2c48
-#define MGAREG_SPECBSTART 			0x2c98
-#define MGAREG_SPECBXINC 			0x2c9c
-#define MGAREG_SPECBYINC 			0x2ca0
-#define MGAREG_SPECGSTART 			0x2c8c
-#define MGAREG_SPECGXINC 			0x2c90
-#define MGAREG_SPECGYINC 			0x2c94
-#define MGAREG_SPECRSTART 			0x2c80
-#define MGAREG_SPECRXINC 			0x2c84
-#define MGAREG_SPECRYINC 			0x2c88
-#define MGAREG_SRC0 				0x1c30
-#define MGAREG_SRC1 				0x1c34
-#define MGAREG_SRC2 				0x1c38
-#define MGAREG_SRC3 				0x1c3c
 #define MGAREG_SRCORG 				0x2cb4
 #define MGAREG_STATUS 				0x1e14
 #define MGAREG_STENCIL				0x2cc8
 #define MGAREG_STENCILCTL 			0x2ccc
 #define MGAREG_TDUALSTAGE0 			0x2cf8
 #define MGAREG_TDUALSTAGE1 			0x2cfc
-#define MGAREG_TEST0 				0x1e48
 #define MGAREG_TEXBORDERCOL 			0x2c5c
 #define MGAREG_TEXCTL 				0x2c30
 #define MGAREG_TEXCTL2				0x2c3c
@@ -249,18 +337,6 @@ extern int  mga_context_switch_complete(drm_device_t *dev, int new);
 #define MGAREG_TEXTRANS 			0x2c34
 #define MGAREG_TEXTRANSHIGH 			0x2c38
 #define MGAREG_TEXWIDTH 			0x2c28
-#define MGAREG_TMR0 				0x2c00
-#define MGAREG_TMR1 				0x2c04
-#define MGAREG_TMR2 				0x2c08
-#define MGAREG_TMR3 				0x2c0c
-#define MGAREG_TMR4 				0x2c10
-#define MGAREG_TMR5 				0x2c14
-#define MGAREG_TMR6 				0x2c18
-#define MGAREG_TMR7 				0x2c1c
-#define MGAREG_TMR8 				0x2c20
-#define MGAREG_VBIADDR0 			0x3e08
-#define MGAREG_VBIADDR1 			0x3e0c
-#define MGAREG_VCOUNT 				0x1e20
 #define MGAREG_WACCEPTSEQ 			0x1dd4
 #define MGAREG_WCODEADDR 			0x1e6c
 #define MGAREG_WFLAG 				0x1dc4
@@ -270,18 +346,8 @@ extern int  mga_context_switch_complete(drm_device_t *dev, int new);
 #define MGAREG_WGETMSB				0x1dc8
 #define MGAREG_WIADDR 				0x1dc0
 #define MGAREG_WIADDR2				0x1dd8
-#define MGAREG_WIADDRNB 			0x1e60
-#define MGAREG_WIADDRNB1 			0x1e04
-#define MGAREG_WIADDRNB2 			0x1e00
-#define MGAREG_WIMEMADDR 			0x1e68
-#define MGAREG_WIMEMDATA 			0x2000
-#define MGAREG_WIMEMDATA1 			0x2100
 #define MGAREG_WMISC 				0x1e70
-#define MGAREG_WR 				0x2d00
 #define MGAREG_WVRTXSZ				0x1dcc
-#define MGAREG_XDST 				0x1cb0
-#define MGAREG_XYEND 				0x1c44
-#define MGAREG_XYSTRT 				0x1c40
 #define MGAREG_YBOT 				0x1c9c
 #define MGAREG_YDST 				0x1c90
 #define MGAREG_YDSTLEN				0x1c88
@@ -289,4 +355,40 @@ extern int  mga_context_switch_complete(drm_device_t *dev, int new);
 #define MGAREG_YTOP 				0x1c98
 #define MGAREG_ZORG 				0x1c0c
 
+#define DC_atype_rstr				0x10
+#define DC_atype_blk				0x40
+#define PDEA_pagpxfer_enable			0x2
+#define WIA_wmode_suspend			0x0
+#define WIA_wmode_start 			0x3
+#define WIA_wagp_agp				0x4
+#define DC_opcod_trap				0x4
+#define DC_arzero_enable			0x1000
+#define DC_sgnzero_enable			0x2000
+#define DC_shftzero_enable			0x4000
+#define DC_bop_SHIFT				16
+#define DC_clipdis_enable			0x80000000
+#define DC_solid_enable				0x800
+#define DC_transc_enable			0x40000000
+#define DC_opcod_bitblt 			0x8
+#define DC_atype_rpl				0x0
+#define DC_linear_xy				0x0
+#define DC_solid_disable			0x0
+#define DC_arzero_disable			0x0
+#define DC_bltmod_bfcol				0x4000000
+#define DC_pattern_disable			0x0
+#define DC_transc_disable			0x0
+
+#define MGA_CLEAR_CMD (DC_opcod_trap | DC_arzero_enable | 		\
+		       DC_sgnzero_enable | DC_shftzero_enable | 	\
+		       (0xC << DC_bop_SHIFT) | DC_clipdis_enable | 	\
+		       DC_solid_enable | DC_transc_enable)
+	  
+
+#define MGA_COPY_CMD (DC_opcod_bitblt | DC_atype_rpl | DC_linear_xy |	\
+		      DC_solid_disable | DC_arzero_disable | 		\
+		      DC_sgnzero_enable | DC_shftzero_enable | 		\
+		      (0xC << DC_bop_SHIFT) | DC_bltmod_bfcol | 	\
+		      DC_pattern_disable | DC_transc_disable | 		\
+		      DC_clipdis_enable)				\
+
 #endif
diff --git a/linux/mga_state.c b/linux/mga_state.c
index d09881ba..a70f86d2 100644
--- a/linux/mga_state.c
+++ b/linux/mga_state.c
@@ -26,189 +26,217 @@
  * Authors: Jeff Hartmann <jhartmann@precisioninsight.com>
  * 	    Keith Whitwell <keithw@precisioninsight.com>
  *
- * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/linux/drm/kernel/mga_state.c,v 1.1 2000/02/11 17:26:08 dawes Exp $
+ * $XFree86$
  *
  */
  
 #define __NO_VERSION__
 #include "drmP.h"
 #include "mga_drv.h"
-#include "mgareg_flags.h"
-#include "mga_dma.h"
-#include "mga_state.h"
 #include "drm.h"
 
-void mgaEmitClipRect( drm_mga_private_t *dev_priv, xf86drmClipRectRec *box )
+static void mgaEmitClipRect( drm_mga_private_t *dev_priv, 
+			     drm_clip_rect_t *box )
 {
+   	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	unsigned int *regs = sarea_priv->ContextState;
 	PRIMLOCALS;
 
+   	/* This takes 10 dwords */
 	PRIMGETPTR( dev_priv );
+   
+   	/* Force reset of dwgctl (eliminates clip disable) */
+   	PRIMOUTREG( MGAREG_DMAPAD, 0 );
+	PRIMOUTREG( MGAREG_DWGSYNC, dev_priv->last_sync_tag - 1 );
+	PRIMOUTREG( MGAREG_DWGSYNC, dev_priv->last_sync_tag - 1 );
+   	PRIMOUTREG( MGAREG_DWGCTL, regs[MGA_CTXREG_DWGCTL] );
 
-	/* The G400 seems to have an issue with the second WARP not
-	 * stalling clipper register writes.  This bothers me, but the only
-	 * way I could get it to never clip the last triangle under any
-	 * circumstances is by inserting TWO dwgsync commands.
-	 */
- 	if (dev_priv->chipset == MGA_CARD_TYPE_G400) { 
-		PRIMOUTREG( MGAREG_DWGSYNC, 0 );
-		PRIMOUTREG( MGAREG_DWGSYNC, 0 );
-	}
-
+   	PRIMOUTREG( MGAREG_DMAPAD, 0 );
 	PRIMOUTREG( MGAREG_CXBNDRY, ((box->x2)<<16)|(box->x1) );
-	PRIMOUTREG( MGAREG_YTOP, box->y1 * dev_priv->stride );
-	PRIMOUTREG( MGAREG_YBOT, box->y2 * dev_priv->stride );
+	PRIMOUTREG( MGAREG_YTOP, box->y1 * dev_priv->stride/2 );
+	PRIMOUTREG( MGAREG_YBOT, box->y2 * dev_priv->stride/2 );
+
 	PRIMADVANCE( dev_priv );
 }
 
-
-static void mgaEmitContext(drm_mga_private_t *dev_priv, 
-			   drm_mga_buf_priv_t *buf_priv)
+static void mgaEmitContext(drm_mga_private_t *dev_priv )
 {
-	unsigned int *regs = buf_priv->ContextState;
+      	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	unsigned int *regs = sarea_priv->ContextState;
 	PRIMLOCALS;
 	
+   	/* This takes a max of 15 dwords */
 	PRIMGETPTR( dev_priv );
+
 	PRIMOUTREG( MGAREG_DSTORG, regs[MGA_CTXREG_DSTORG] );
 	PRIMOUTREG( MGAREG_MACCESS, regs[MGA_CTXREG_MACCESS] );
 	PRIMOUTREG( MGAREG_PLNWT, regs[MGA_CTXREG_PLNWT] );
 	PRIMOUTREG( MGAREG_DWGCTL, regs[MGA_CTXREG_DWGCTL] );
+
 	PRIMOUTREG( MGAREG_ALPHACTRL, regs[MGA_CTXREG_ALPHACTRL] );
 	PRIMOUTREG( MGAREG_FOGCOL, regs[MGA_CTXREG_FOGCOLOR] );
 	PRIMOUTREG( MGAREG_WFLAG, regs[MGA_CTXREG_WFLAG] );
+	PRIMOUTREG( MGAREG_ZORG, dev_priv->depthOffset ); /* invarient */
 
  	if (dev_priv->chipset == MGA_CARD_TYPE_G400) { 
 		PRIMOUTREG( MGAREG_WFLAG1, regs[MGA_CTXREG_WFLAG] );
 		PRIMOUTREG( MGAREG_TDUALSTAGE0, regs[MGA_CTXREG_TDUAL0] );
 		PRIMOUTREG( MGAREG_TDUALSTAGE1, regs[MGA_CTXREG_TDUAL1] );      
-	}   
+		PRIMOUTREG( MGAREG_FCOL, regs[MGA_CTXREG_FCOL] );
+	} else {
+	   	PRIMOUTREG( MGAREG_FCOL, regs[MGA_CTXREG_FCOL] );
+	   	PRIMOUTREG( MGAREG_DMAPAD, 0);
+	   	PRIMOUTREG( MGAREG_DMAPAD, 0);
+	   	PRIMOUTREG( MGAREG_DMAPAD, 0);
+	}
    
 	PRIMADVANCE( dev_priv );
 }
 
-static void mgaG200EmitTex( drm_mga_private_t *dev_priv, 
-			    drm_mga_buf_priv_t *buf_priv )
+static void mgaG200EmitTex( drm_mga_private_t *dev_priv )
 {
-	unsigned int *regs = buf_priv->TexState[0];
+      	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	unsigned int *regs = sarea_priv->TexState[0];
 	PRIMLOCALS;
 
 	PRIMGETPTR( dev_priv );
+   
+   	/* This takes 20 dwords */
+
 	PRIMOUTREG(MGAREG_TEXCTL2, regs[MGA_TEXREG_CTL2] );
 	PRIMOUTREG(MGAREG_TEXCTL, regs[MGA_TEXREG_CTL] );
 	PRIMOUTREG(MGAREG_TEXFILTER, regs[MGA_TEXREG_FILTER] );
 	PRIMOUTREG(MGAREG_TEXBORDERCOL, regs[MGA_TEXREG_BORDERCOL] );
+
 	PRIMOUTREG(MGAREG_TEXORG, regs[MGA_TEXREG_ORG] );
 	PRIMOUTREG(MGAREG_TEXORG1, regs[MGA_TEXREG_ORG1] );
 	PRIMOUTREG(MGAREG_TEXORG2, regs[MGA_TEXREG_ORG2] );
 	PRIMOUTREG(MGAREG_TEXORG3, regs[MGA_TEXREG_ORG3] );
+
 	PRIMOUTREG(MGAREG_TEXORG4, regs[MGA_TEXREG_ORG4] );		
 	PRIMOUTREG(MGAREG_TEXWIDTH, regs[MGA_TEXREG_WIDTH] );
-	PRIMOUTREG(MGAREG_TEXHEIGHT, regs[MGA_TEXREG_HEIGHT] );
-   
+	PRIMOUTREG(MGAREG_TEXHEIGHT, regs[MGA_TEXREG_HEIGHT] );   
 	PRIMOUTREG(0x2d00 + 24*4, regs[MGA_TEXREG_WIDTH] );
+
 	PRIMOUTREG(0x2d00 + 34*4, regs[MGA_TEXREG_HEIGHT] );
+	PRIMOUTREG( MGAREG_TEXTRANS, 0xffff );
+	PRIMOUTREG( MGAREG_TEXTRANSHIGH, 0xffff );
+	PRIMOUTREG( MGAREG_DMAPAD, 0 );
 
 	PRIMADVANCE( dev_priv );  
 }
 
-static void mgaG400EmitTex0( drm_mga_private_t *dev_priv, 
-			     drm_mga_buf_priv_t *buf_priv )
+static void mgaG400EmitTex0( drm_mga_private_t *dev_priv )
 {
-	unsigned int *regs = buf_priv->TexState[0];
-	int multitex = buf_priv->WarpPipe & MGA_T2;
-
+      	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	unsigned int *regs = sarea_priv->TexState[0];
+	int multitex = sarea_priv->WarpPipe & MGA_T2;
 	PRIMLOCALS;
+   
 	PRIMGETPTR( dev_priv );
+   
+   	/* This takes a max of 30 dwords */
 
 	PRIMOUTREG(MGAREG_TEXCTL2, regs[MGA_TEXREG_CTL2] );
 	PRIMOUTREG(MGAREG_TEXCTL, regs[MGA_TEXREG_CTL] );
 	PRIMOUTREG(MGAREG_TEXFILTER, regs[MGA_TEXREG_FILTER] );
 	PRIMOUTREG(MGAREG_TEXBORDERCOL, regs[MGA_TEXREG_BORDERCOL] );
+
 	PRIMOUTREG(MGAREG_TEXORG, regs[MGA_TEXREG_ORG] );
 	PRIMOUTREG(MGAREG_TEXORG1, regs[MGA_TEXREG_ORG1] );
 	PRIMOUTREG(MGAREG_TEXORG2, regs[MGA_TEXREG_ORG2] );
 	PRIMOUTREG(MGAREG_TEXORG3, regs[MGA_TEXREG_ORG3] );
+
 	PRIMOUTREG(MGAREG_TEXORG4, regs[MGA_TEXREG_ORG4] );		
 	PRIMOUTREG(MGAREG_TEXWIDTH, regs[MGA_TEXREG_WIDTH] );
-	PRIMOUTREG(MGAREG_TEXHEIGHT, regs[MGA_TEXREG_HEIGHT] );
-   
+	PRIMOUTREG(MGAREG_TEXHEIGHT, regs[MGA_TEXREG_HEIGHT] );   
 	PRIMOUTREG(0x2d00 + 49*4, 0);
+
 	PRIMOUTREG(0x2d00 + 57*4, 0);
 	PRIMOUTREG(0x2d00 + 53*4, 0);
 	PRIMOUTREG(0x2d00 + 61*4, 0);
+	PRIMOUTREG( MGAREG_DMAPAD, 0 );
 
 	if (!multitex) {
 		PRIMOUTREG(0x2d00 + 52*4, 0x40 ); 
 		PRIMOUTREG(0x2d00 + 60*4, 0x40 );
+		PRIMOUTREG( MGAREG_DMAPAD, 0 );
+		PRIMOUTREG( MGAREG_DMAPAD, 0 );
 	} 
 
-	PRIMOUTREG(0x2d00 + 54*4, regs[MGA_TEXREG_WIDTH] | 0x40 );
-	PRIMOUTREG(0x2d00 + 62*4, regs[MGA_TEXREG_HEIGHT] | 0x40 );
+	PRIMOUTREG( 0x2d00 + 54*4, regs[MGA_TEXREG_WIDTH] | 0x40 );
+	PRIMOUTREG( 0x2d00 + 62*4, regs[MGA_TEXREG_HEIGHT] | 0x40 );
+	PRIMOUTREG( MGAREG_TEXTRANS, 0xffff );
+	PRIMOUTREG( MGAREG_TEXTRANSHIGH, 0xffff );
 
 	PRIMADVANCE( dev_priv );  
 }
 
 #define TMC_map1_enable 		0x80000000 	
 
-
-static void mgaG400EmitTex1( drm_mga_private_t *dev_priv, 
-			     drm_mga_buf_priv_t *buf_priv )
+static void mgaG400EmitTex1( drm_mga_private_t *dev_priv )
 {
-	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+      	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
 	unsigned int *regs = sarea_priv->TexState[1];
-
 	PRIMLOCALS;
+   
 	PRIMGETPTR(dev_priv);
 
+   	/* This takes 25 dwords */
 	PRIMOUTREG(MGAREG_TEXCTL2, regs[MGA_TEXREG_CTL2] | TMC_map1_enable);
 	PRIMOUTREG(MGAREG_TEXCTL, regs[MGA_TEXREG_CTL] );
 	PRIMOUTREG(MGAREG_TEXFILTER, regs[MGA_TEXREG_FILTER] );
 	PRIMOUTREG(MGAREG_TEXBORDERCOL, regs[MGA_TEXREG_BORDERCOL] );
+
 	PRIMOUTREG(MGAREG_TEXORG, regs[MGA_TEXREG_ORG] );
 	PRIMOUTREG(MGAREG_TEXORG1, regs[MGA_TEXREG_ORG1] );
 	PRIMOUTREG(MGAREG_TEXORG2, regs[MGA_TEXREG_ORG2] );
 	PRIMOUTREG(MGAREG_TEXORG3, regs[MGA_TEXREG_ORG3] );
+
 	PRIMOUTREG(MGAREG_TEXORG4, regs[MGA_TEXREG_ORG4] );		
 	PRIMOUTREG(MGAREG_TEXWIDTH, regs[MGA_TEXREG_WIDTH] );
-	PRIMOUTREG(MGAREG_TEXHEIGHT, regs[MGA_TEXREG_HEIGHT] );
-   
+	PRIMOUTREG(MGAREG_TEXHEIGHT, regs[MGA_TEXREG_HEIGHT] );   
 	PRIMOUTREG(0x2d00 + 49*4, 0);
+
 	PRIMOUTREG(0x2d00 + 57*4, 0);
 	PRIMOUTREG(0x2d00 + 53*4, 0);
 	PRIMOUTREG(0x2d00 + 61*4, 0);
-
 	PRIMOUTREG(0x2d00 + 52*4, regs[MGA_TEXREG_WIDTH] | 0x40 ); 
-	PRIMOUTREG(0x2d00 + 60*4, regs[MGA_TEXREG_HEIGHT] | 0x40 );
 
+	PRIMOUTREG(0x2d00 + 60*4, regs[MGA_TEXREG_HEIGHT] | 0x40 );
+	PRIMOUTREG( MGAREG_TEXTRANS, 0xffff );
+	PRIMOUTREG( MGAREG_TEXTRANSHIGH, 0xffff );
 	PRIMOUTREG(MGAREG_TEXCTL2, regs[MGA_TEXREG_CTL2] );
 
-	PRIMADVANCE( dev_priv );     
+	PRIMADVANCE( dev_priv );
 }
 
-
-static void mgaG400EmitPipe(drm_mga_private_t *dev_priv, 
-			    drm_mga_buf_priv_t *buf_priv)
+static void mgaG400EmitPipe(drm_mga_private_t *dev_priv )
 {
-	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+      	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
 	unsigned int pipe = sarea_priv->WarpPipe;
 	float fParam = 12800.0f;
 	PRIMLOCALS;
    
 	PRIMGETPTR(dev_priv);
-	PRIMOUTREG(MGAREG_WIADDR2, WIA_wmode_suspend);
+
+   	/* This takes 25 dwords */
    
 	/* Establish vertex size.  
 	 */
 	if (pipe & MGA_T2) {
+		PRIMOUTREG(MGAREG_WIADDR2, WIA_wmode_suspend);
 		PRIMOUTREG(MGAREG_WVRTXSZ, 0x00001e09);
 		PRIMOUTREG(MGAREG_WACCEPTSEQ, 0x1e000000);
+		PRIMOUTREG(MGAREG_WFLAG, 0);
 	} else {
+		PRIMOUTREG(MGAREG_WIADDR2, WIA_wmode_suspend);
 		PRIMOUTREG(MGAREG_WVRTXSZ, 0x00001807);
 		PRIMOUTREG(MGAREG_WACCEPTSEQ, 0x18000000);
-	}
-   
-	PRIMOUTREG(MGAREG_WFLAG, 0);
-	PRIMOUTREG(MGAREG_WFLAG1, 0);
-   
+		PRIMOUTREG(MGAREG_WFLAG, 0);
+	}   
+
+	PRIMOUTREG(MGAREG_WFLAG1, 0);   
 	PRIMOUTREG(0x2d00 + 56*4, *((u32 *)(&fParam)));
 	PRIMOUTREG(MGAREG_DMAPAD, 0);
 	PRIMOUTREG(MGAREG_DMAPAD, 0);
@@ -227,19 +255,21 @@ static void mgaG400EmitPipe(drm_mga_private_t *dev_priv,
 	PRIMOUTREG(MGAREG_DMAPAD, 0xffffffff);
 	PRIMOUTREG(MGAREG_DMAPAD, 0xffffffff);
 	PRIMOUTREG(MGAREG_DMAPAD, 0xffffffff);
-	PRIMOUTREG(MGAREG_WIADDR2, (dev_priv->WarpIndex[pipe].phys_addr | 
-				   WIA_wmode_start | WIA_wagp_agp));
+	PRIMOUTREG(MGAREG_WIADDR2, (__u32)(dev_priv->WarpIndex[pipe].phys_addr |
+					   WIA_wmode_start | WIA_wagp_agp));
 	PRIMADVANCE(dev_priv);
 }
 
-static void mgaG200EmitPipe( drm_mga_private_t *dev_priv,
-			     drm_mga_buf_priv_t *buf_priv )
+static void mgaG200EmitPipe( drm_mga_private_t *dev_priv )
 {
-	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+      	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
 	unsigned int pipe = sarea_priv->WarpPipe;
 	PRIMLOCALS;
 
 	PRIMGETPTR(dev_priv);
+   
+   	/* This takes 15 dwords */
+
 	PRIMOUTREG(MGAREG_WIADDR, WIA_wmode_suspend);
 	PRIMOUTREG(MGAREG_WVRTXSZ, 7);
 	PRIMOUTREG(MGAREG_WFLAG, 0);
@@ -254,109 +284,629 @@ static void mgaG200EmitPipe( drm_mga_private_t *dev_priv,
 	PRIMOUTREG(MGAREG_DMAPAD, 0xffffffff);
 	PRIMOUTREG(MGAREG_DMAPAD, 0xffffffff);
 	PRIMOUTREG(MGAREG_DMAPAD, 0xffffffff);
-	PRIMOUTREG(MGAREG_WIADDR, (dev_priv->WarpIndex[pipe].phys_addr | 
-				  WIA_wmode_start | WIA_wagp_agp));
+	PRIMOUTREG(MGAREG_WIADDR, (__u32)(dev_priv->WarpIndex[pipe].phys_addr | 
+					  WIA_wmode_start | WIA_wagp_agp));
 
 	PRIMADVANCE(dev_priv);
 }
 
-void mgaEmitState( drm_mga_private_t *dev_priv, drm_mga_buf_priv_t *buf_priv )
+static void mgaEmitState( drm_mga_private_t *dev_priv )
 {
-	unsigned int dirty = buf_priv->dirty;
+      	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	unsigned int dirty = sarea_priv->dirty;
 
 	if (dev_priv->chipset == MGA_CARD_TYPE_G400) {	   
-		if (dirty & MGASAREA_NEW_CONTEXT)
-			mgaEmitContext( dev_priv, buf_priv );
-
-		if (dirty & MGASAREA_NEW_TEX1)
-			mgaG400EmitTex1( dev_priv, buf_priv );
-	   
-		if (dirty & MGASAREA_NEW_TEX0)
-			mgaG400EmitTex0( dev_priv, buf_priv );
-
-		if (dirty & MGASAREA_NEW_PIPE) 
-			mgaG400EmitPipe( dev_priv, buf_priv );
+		int multitex = sarea_priv->WarpPipe & MGA_T2;
+
+		if (sarea_priv->WarpPipe != dev_priv->WarpPipe) { 
+			mgaG400EmitPipe( dev_priv );
+			dev_priv->WarpPipe = sarea_priv->WarpPipe;
+		}
+
+		if (dirty & MGA_UPLOAD_CTX) {
+			mgaEmitContext( dev_priv );
+		   	sarea_priv->dirty &= ~MGA_UPLOAD_CTX;
+		}
+
+		if (dirty & MGA_UPLOAD_TEX0) {
+			mgaG400EmitTex0( dev_priv );
+		   	sarea_priv->dirty &= ~MGA_UPLOAD_TEX0;
+		}
+
+		if ((dirty & MGA_UPLOAD_TEX1) && multitex) {
+			mgaG400EmitTex1( dev_priv );
+		   	sarea_priv->dirty &= ~MGA_UPLOAD_TEX1;
+		}
 	} else {
-		if (dirty & MGASAREA_NEW_CONTEXT)
-			mgaEmitContext( dev_priv, buf_priv );
-
-		if (dirty & MGASAREA_NEW_TEX0)
-			mgaG200EmitTex( dev_priv, buf_priv );
-
-		if (dirty & MGASAREA_NEW_PIPE) 
-			mgaG200EmitPipe( dev_priv, buf_priv );
-	}	  
+		if (sarea_priv->WarpPipe != dev_priv->WarpPipe) { 
+			mgaG200EmitPipe( dev_priv );
+			dev_priv->WarpPipe = sarea_priv->WarpPipe;
+		}
+
+		if (dirty & MGA_UPLOAD_CTX) {
+			mgaEmitContext( dev_priv );
+		   	sarea_priv->dirty &= ~MGA_UPLOAD_CTX;
+		}
+
+		if (dirty & MGA_UPLOAD_TEX0) {
+			mgaG200EmitTex( dev_priv );
+		   	sarea_priv->dirty &= ~MGA_UPLOAD_TEX0;
+		}
+	}
 }
 
 
-
 /* Disallow all write destinations except the front and backbuffer.
  */
-static int mgaCopyContext(drm_mga_private_t *dev_priv, 
-			  drm_mga_buf_priv_t *buf_priv)
+static int mgaVerifyContext(drm_mga_private_t *dev_priv )
 {
 	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
 	unsigned int *regs = sarea_priv->ContextState;
-	
-	if (regs[MGA_CTXREG_DSTORG] != dev_priv->frontOrg &&
-	    regs[MGA_CTXREG_DSTORG] != dev_priv->backOrg)
+
+	if (regs[MGA_CTXREG_DSTORG] != dev_priv->frontOffset &&
+	    regs[MGA_CTXREG_DSTORG] != dev_priv->backOffset) {
+		DRM_DEBUG("BAD DSTORG: %x (front %x, back %x)\n\n", 
+		       regs[MGA_CTXREG_DSTORG], dev_priv->frontOffset,
+		       dev_priv->backOffset);
+  	        regs[MGA_CTXREG_DSTORG] = 0;
 		return -1;
+	}
 
-	memcpy(buf_priv->ContextState, sarea_priv->ContextState,
-	       sizeof(buf_priv->ContextState));
 	return 0;
 }
 
-
 /* Disallow texture reads from PCI space.
  */
-static int mgaCopyTex(drm_mga_private_t *dev_priv, 
-		      drm_mga_buf_priv_t *buf_priv,
+static int mgaVerifyTex(drm_mga_private_t *dev_priv, 
 		      int unit)
 {
 	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
 
-	if ((sarea_priv->TexState[unit][MGA_TEXREG_ORG] & 0x3) == 0x1)
+	if ((sarea_priv->TexState[unit][MGA_TEXREG_ORG] & 0x3) == 0x1) {
+		DRM_DEBUG("BAD TEXREG_ORG: %x, unit %d\n", 
+		       sarea_priv->TexState[unit][MGA_TEXREG_ORG],
+		       unit);
+		sarea_priv->TexState[unit][MGA_TEXREG_ORG] = 0;
 		return -1;
-
-	memcpy(buf_priv->TexState[unit], sarea_priv->TexState[unit],
-	       sizeof(buf_priv->TexState[0]));
+	} 
 
 	return 0;
 }
 
-
-int mgaCopyAndVerifyState( drm_mga_private_t *dev_priv, 
-			   drm_mga_buf_priv_t *buf_priv )
+static int mgaVerifyState( drm_mga_private_t *dev_priv )
 {
 	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
-	unsigned int dirty = sarea_priv->dirty ;
+	unsigned int dirty = sarea_priv->dirty;
 	int rv = 0;
 
-	buf_priv->dirty = sarea_priv->dirty;
-	buf_priv->WarpPipe = sarea_priv->WarpPipe;
+   	if (sarea_priv->nbox > MGA_NR_SAREA_CLIPRECTS)
+     		sarea_priv->nbox = MGA_NR_SAREA_CLIPRECTS;
 
-	if (dirty & MGASAREA_NEW_CONTEXT)
-		rv |= mgaCopyContext( dev_priv, buf_priv );
+	if (dirty & MGA_UPLOAD_CTX)
+		rv |= mgaVerifyContext( dev_priv );
 
-	if (dirty & MGASAREA_NEW_TEX0)
-		rv |= mgaCopyTex( dev_priv, buf_priv, 0 );
+	if (dirty & MGA_UPLOAD_TEX0)
+		rv |= mgaVerifyTex( dev_priv, 0 );
 
 	if (dev_priv->chipset == MGA_CARD_TYPE_G400) 
 	{	   
-		if (dirty & MGASAREA_NEW_TEX1)
-			rv |= mgaCopyTex( dev_priv, buf_priv, 1 );
+		if (dirty & MGA_UPLOAD_TEX1)
+			rv |= mgaVerifyTex( dev_priv, 1 );
 	   
-		if (dirty & MGASAREA_NEW_PIPE) 
-			rv |= (buf_priv->WarpPipe > MGA_MAX_G400_PIPES); 
+		if (dirty & MGA_UPLOAD_PIPE) 
+			rv |= (sarea_priv->WarpPipe > MGA_MAX_G400_PIPES); 
 	} 
 	else 
 	{
-		if (dirty & MGASAREA_NEW_PIPE) 
-			rv |= (buf_priv->WarpPipe > MGA_MAX_G200_PIPES); 
+		if (dirty & MGA_UPLOAD_PIPE) 
+			rv |= (sarea_priv->WarpPipe > MGA_MAX_G200_PIPES); 
 	}	  
 
 	return rv == 0;
 }
 
+static int mgaVerifyIload( drm_mga_private_t *dev_priv, 
+			   unsigned long bus_address, 
+			   unsigned int dstOrg, int length )
+{
+   	if(dstOrg < dev_priv->textureOffset ||
+	   dstOrg + length > 
+	   (dev_priv->textureOffset + dev_priv->textureSize)) {
+	   	return -EINVAL;
+	}
+   	if(length % 64) {
+	   	return -EINVAL;
+	}
+   	return 0;
+}
+
+/* This copies a 64 byte aligned agp region to the frambuffer
+ * with a standard blit, the ioctl needs to do checking */
+
+static void mga_dma_dispatch_tex_blit( drm_device_t *dev, 
+				       unsigned long bus_address, 
+				       int length, 
+				       unsigned int destOrg )
+{
+   	drm_mga_private_t *dev_priv = dev->dev_private;
+   	int use_agp = PDEA_pagpxfer_enable | 0x00000001;
+   	u16 y2;
+    	PRIMLOCALS;
+    
+        y2 = length / 64;
+ 
+	PRIM_OVERFLOW(dev, dev_priv, 30);
+	PRIMGETPTR( dev_priv );
+
+      	dev_priv->last_sync_tag = mga_create_sync_tag(dev);
+
+	PRIMOUTREG( MGAREG_DSTORG, destOrg);
+	PRIMOUTREG( MGAREG_MACCESS, 0x00000000);
+   	DRM_DEBUG("srcorg : %lx\n", bus_address | use_agp);
+	PRIMOUTREG( MGAREG_SRCORG, (u32) bus_address | use_agp);
+	PRIMOUTREG( MGAREG_AR5, 64);  
+
+	PRIMOUTREG( MGAREG_PITCH, 64);
+	PRIMOUTREG( MGAREG_DMAPAD, 0);
+	PRIMOUTREG( MGAREG_DMAPAD, 0);
+	PRIMOUTREG( MGAREG_DWGCTL, MGA_COPY_CMD); 
+
+   	PRIMOUTREG(MGAREG_AR0, 63);
+   	PRIMOUTREG(MGAREG_AR3, 0);		
+   	PRIMOUTREG(MGAREG_FXBNDRY, (63 << 16));
+   	PRIMOUTREG(MGAREG_YDSTLEN+MGAREG_MGA_EXEC, y2);
+  
+	PRIMOUTREG( MGAREG_SRCORG, 0);
+	PRIMOUTREG( MGAREG_PITCH, dev_priv->stride / dev_priv->cpp);
+      	PRIMOUTREG( MGAREG_DMAPAD, 0);
+      	PRIMOUTREG( MGAREG_DWGSYNC, dev_priv->last_sync_tag);
+    	PRIMADVANCE(dev_priv);
+}
+
+static void mga_dma_dispatch_vertex(drm_device_t *dev, 
+				    drm_buf_t *buf)
+{
+   	drm_mga_private_t *dev_priv = dev->dev_private;
+	drm_mga_buf_priv_t *buf_priv = buf->dev_private;
+   	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	unsigned long address = (unsigned long)buf->bus_address;
+	int length = buf->used;	
+	int use_agp = PDEA_pagpxfer_enable;
+	int i = 0;
+   	int primary_needed;
+      	PRIMLOCALS;
+
+	DRM_DEBUG("dispatch vertex %d addr 0x%lx, "
+		  "length 0x%x nbox %d dirty %x\n", 
+		  buf->idx, address, length, 
+		  sarea_priv->nbox, sarea_priv->dirty);
+
+
+   	dev_priv->last_sync_tag = mga_create_sync_tag(dev);
+
+   	if (buf_priv->discard) { 
+		buf_priv->my_freelist->age = dev_priv->last_sync_tag;
+	   	mga_freelist_put(dev, buf);
+	}
+
+
+	/* WARNING: if you change any of the state functions verify
+	 * these numbers (Overestimating this doesn't hurt).  
+	 */
+	primary_needed = (25+15+30+25+ 
+			  10 + 
+			  15 * MGA_NR_SAREA_CLIPRECTS);
+
+
+   	PRIM_OVERFLOW(dev, dev_priv, primary_needed);
+   	mgaEmitState( dev_priv );
+
+	if (buf->used) {
+		do {
+			if (i < sarea_priv->nbox) {
+				DRM_DEBUG("idx %d Emit box %d/%d:"
+					  "%d,%d - %d,%d\n", 
+					  buf->idx,
+					  i, sarea_priv->nbox,
+					  sarea_priv->boxes[i].x1, 
+					  sarea_priv->boxes[i].y1,
+					  sarea_priv->boxes[i].x2, 
+					  sarea_priv->boxes[i].y2);
+				
+				mgaEmitClipRect( dev_priv, 
+						 &sarea_priv->boxes[i] );
+			}
+			
+			PRIMGETPTR(dev_priv);
+			PRIMOUTREG( MGAREG_DMAPAD, 0);
+			PRIMOUTREG( MGAREG_DMAPAD, 0);
+			PRIMOUTREG( MGAREG_SECADDRESS, 
+				    ((__u32)address) | TT_VERTEX);
+			PRIMOUTREG( MGAREG_SECEND, 
+				    (((__u32)(address + length)) | 
+				     use_agp));
+			PRIMADVANCE( dev_priv );	       
+		}  while (++i < sarea_priv->nbox);
+	}
+
+	PRIMGETPTR( dev_priv );
+   	PRIMOUTREG(MGAREG_DMAPAD, 0);
+   	PRIMOUTREG(MGAREG_DMAPAD, 0);
+   	PRIMOUTREG(MGAREG_DMAPAD, 0);
+   	PRIMOUTREG(MGAREG_DWGSYNC, dev_priv->last_sync_tag);   
+	PRIMADVANCE( dev_priv );
+}
+
 
+static void mga_dma_dispatch_clear( drm_device_t *dev, int flags, 
+				    unsigned int clear_color,
+				    unsigned int clear_zval )
+{
+   	drm_mga_private_t *dev_priv = dev->dev_private;
+      	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	unsigned int *regs = sarea_priv->ContextState;
+	int nbox = sarea_priv->nbox;
+	drm_clip_rect_t *pbox = sarea_priv->boxes;
+	unsigned int cmd;
+	int i;
+   	int primary_needed;
+   	PRIMLOCALS;
+
+	if ( dev_priv->sgram ) 
+		cmd = MGA_CLEAR_CMD | DC_atype_blk;
+	else
+		cmd = MGA_CLEAR_CMD | DC_atype_rstr;
+   
+   	primary_needed = nbox * 70;
+   	if(primary_needed == 0) primary_needed = 70;
+	PRIM_OVERFLOW(dev, dev_priv, primary_needed);
+	PRIMGETPTR( dev_priv );
+      	dev_priv->last_sync_tag = mga_create_sync_tag(dev);
+   
+	for (i = 0 ; i < nbox ; i++) {
+		unsigned int height = pbox[i].y2 - pbox[i].y1;
+
+	   	DRM_DEBUG("dispatch clear %d,%d-%d,%d flags %x!\n",
+			  pbox[i].x1, pbox[i].y1, pbox[i].x2, 
+			  pbox[i].y2, flags);
+
+	   	if ( flags & MGA_FRONT ) {	    
+		   	DRM_DEBUG("clear front\n");
+			PRIMOUTREG( MGAREG_DMAPAD, 0);
+			PRIMOUTREG( MGAREG_DMAPAD, 0);
+			PRIMOUTREG(MGAREG_YDSTLEN, (pbox[i].y1<<16)|height);
+			PRIMOUTREG(MGAREG_FXBNDRY, (pbox[i].x2<<16)|pbox[i].x1);
+
+			PRIMOUTREG( MGAREG_DMAPAD, 0);
+			PRIMOUTREG(MGAREG_FCOL, clear_color);
+			PRIMOUTREG(MGAREG_DSTORG, dev_priv->frontOffset);
+			PRIMOUTREG(MGAREG_DWGCTL+MGAREG_MGA_EXEC, cmd );
+		}
+
+		if ( flags & MGA_BACK ) {
+			DRM_DEBUG("clear back\n");
+			PRIMOUTREG( MGAREG_DMAPAD, 0);
+			PRIMOUTREG( MGAREG_DMAPAD, 0);
+			PRIMOUTREG(MGAREG_YDSTLEN, (pbox[i].y1<<16)|height);
+			PRIMOUTREG(MGAREG_FXBNDRY, (pbox[i].x2<<16)|pbox[i].x1);
+
+			PRIMOUTREG( MGAREG_DMAPAD, 0);
+			PRIMOUTREG(MGAREG_FCOL, clear_color);
+			PRIMOUTREG(MGAREG_DSTORG, dev_priv->backOffset);
+			PRIMOUTREG(MGAREG_DWGCTL+MGAREG_MGA_EXEC, cmd );
+		}
+
+		if ( flags & MGA_DEPTH ) {
+			DRM_DEBUG("clear depth\n");
+			PRIMOUTREG( MGAREG_DMAPAD, 0);
+			PRIMOUTREG( MGAREG_DMAPAD, 0);
+			PRIMOUTREG(MGAREG_YDSTLEN, (pbox[i].y1<<16)|height);
+			PRIMOUTREG(MGAREG_FXBNDRY, (pbox[i].x2<<16)|pbox[i].x1);
+
+			PRIMOUTREG( MGAREG_DMAPAD, 0);
+			PRIMOUTREG(MGAREG_FCOL, clear_zval);
+			PRIMOUTREG(MGAREG_DSTORG, dev_priv->depthOffset);
+			PRIMOUTREG(MGAREG_DWGCTL+MGAREG_MGA_EXEC, cmd );
+		}
+	}
+
+	/* Force reset of DWGCTL */
+   	PRIMOUTREG( MGAREG_DMAPAD, 0);   
+	PRIMOUTREG( MGAREG_DMAPAD, 0);   
+      	PRIMOUTREG( MGAREG_DMAPAD, 0);
+      	PRIMOUTREG( MGAREG_DWGCTL, regs[MGA_CTXREG_DWGCTL] );
+
+	PRIMOUTREG( MGAREG_DMAPAD, 0);
+	PRIMOUTREG( MGAREG_DMAPAD, 0);
+      	PRIMOUTREG( MGAREG_DMAPAD, 0);
+   	PRIMOUTREG( MGAREG_DWGSYNC, dev_priv->last_sync_tag);
+	PRIMADVANCE(dev_priv);
+}
+
+static void mga_dma_dispatch_swap( drm_device_t *dev )
+{
+   	drm_mga_private_t *dev_priv = dev->dev_private;
+      	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	unsigned int *regs = sarea_priv->ContextState;
+	int nbox = sarea_priv->nbox;
+	drm_clip_rect_t *pbox = sarea_priv->boxes;
+	int i;
+   	int primary_needed;
+   	PRIMLOCALS;
+
+   	primary_needed = nbox * 5;
+   	primary_needed += 60;
+	PRIM_OVERFLOW(dev, dev_priv, primary_needed);
+	PRIMGETPTR( dev_priv );
+   
+      	dev_priv->last_sync_tag = mga_create_sync_tag(dev);
+
+	PRIMOUTREG(MGAREG_DSTORG, dev_priv->frontOffset);
+	PRIMOUTREG(MGAREG_MACCESS, dev_priv->mAccess);
+	PRIMOUTREG(MGAREG_SRCORG, dev_priv->backOffset);
+	PRIMOUTREG(MGAREG_AR5, dev_priv->stride/2);  
+
+	PRIMOUTREG( MGAREG_DMAPAD, 0);
+	PRIMOUTREG( MGAREG_DMAPAD, 0);
+	PRIMOUTREG( MGAREG_DMAPAD, 0);
+	PRIMOUTREG(MGAREG_DWGCTL, MGA_COPY_CMD); 
+	     
+	for (i = 0 ; i < nbox; i++) {
+		unsigned int h = pbox[i].y2 - pbox[i].y1;
+		unsigned int start = pbox[i].y1 * dev_priv->stride/2;
+		
+	   	DRM_DEBUG("dispatch swap %d,%d-%d,%d!\n",
+			  pbox[i].x1, pbox[i].y1,
+			  pbox[i].x2, pbox[i].y2);
+
+		PRIMOUTREG(MGAREG_AR0, start + pbox[i].x2 - 1);
+		PRIMOUTREG(MGAREG_AR3, start + pbox[i].x1);		
+		PRIMOUTREG(MGAREG_FXBNDRY, pbox[i].x1|((pbox[i].x2 - 1)<<16));
+		PRIMOUTREG(MGAREG_YDSTLEN+MGAREG_MGA_EXEC, (pbox[i].y1<<16)|h);
+	}
+
+   	/* Force reset of DWGCTL */
+   	PRIMOUTREG( MGAREG_DMAPAD, 0);
+      	PRIMOUTREG( MGAREG_DMAPAD, 0);
+   	PRIMOUTREG( MGAREG_DMAPAD, 0);
+      	PRIMOUTREG( MGAREG_DWGCTL, regs[MGA_CTXREG_DWGCTL] );
+
+	PRIMOUTREG( MGAREG_SRCORG, 0);
+	PRIMOUTREG( MGAREG_DMAPAD, 0);
+      	PRIMOUTREG( MGAREG_DMAPAD, 0);
+      	PRIMOUTREG( MGAREG_DWGSYNC, dev_priv->last_sync_tag);
+   	PRIMADVANCE(dev_priv);
+}
+
+int mga_clear_bufs(struct inode *inode, struct file *filp,
+		   unsigned int cmd, unsigned long arg)
+{
+	drm_file_t *priv = filp->private_data;
+	drm_device_t *dev = priv->dev;
+   	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
+   	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+      	__volatile__ unsigned int *status = 
+     		(__volatile__ unsigned int *)dev_priv->status_page;
+	drm_mga_clear_t clear;
+
+   	copy_from_user_ret(&clear, (drm_mga_clear_t *)arg, sizeof(clear), 
+			   -EFAULT);
+
+	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("mga_clear_bufs called without lock held\n");
+		return -EINVAL;
+	}
+   
+   	if (sarea_priv->nbox > MGA_NR_SAREA_CLIPRECTS)
+     		sarea_priv->nbox = MGA_NR_SAREA_CLIPRECTS;
+
+	/* Make sure we restore the 3D state next time.
+	 */
+	dev_priv->sarea_priv->dirty |= MGA_UPLOAD_CTX;
+	mga_dma_dispatch_clear( dev, clear.flags, 
+			        clear.clear_color, 
+			        clear.clear_depth );
+      	PRIMUPDATE(dev_priv);
+	mga_dma_schedule(dev, 1);
+   	sarea_priv->last_dispatch = status[1];
+   	return 0;
+}
+
+int mga_swap_bufs(struct inode *inode, struct file *filp,
+		  unsigned int cmd, unsigned long arg)
+{
+	drm_file_t *priv = filp->private_data;
+	drm_device_t *dev = priv->dev;
+   	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
+      	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+      	__volatile__ unsigned int *status = 
+     		(__volatile__ unsigned int *)dev_priv->status_page;
+   
+	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("mga_swap_bufs called without lock held\n");
+		return -EINVAL;
+	}
+
+      	if (sarea_priv->nbox > MGA_NR_SAREA_CLIPRECTS)
+     		sarea_priv->nbox = MGA_NR_SAREA_CLIPRECTS;
+
+	/* Make sure we restore the 3D state next time.
+	 */
+	dev_priv->sarea_priv->dirty |= MGA_UPLOAD_CTX;
+	mga_dma_dispatch_swap( dev );
+      	PRIMUPDATE(dev_priv);
+   	set_bit(0, &dev_priv->current_prim->swap_pending);
+   	dev_priv->current_prim->swap_pending = 1;
+	mga_dma_schedule(dev, 1);
+      	sarea_priv->last_dispatch = status[1];
+   	return 0;
+}
+
+int mga_iload(struct inode *inode, struct file *filp,
+ 	      unsigned int cmd, unsigned long arg)
+{
+ 	drm_file_t *priv = filp->private_data;
+ 	drm_device_t *dev = priv->dev;
+   	drm_device_dma_t  *dma	    = dev->dma;
+       	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
+       	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+       	__volatile__ unsigned int *status = 
+      		(__volatile__ unsigned int *)dev_priv->status_page;
+   	drm_buf_t *buf;
+   	drm_mga_buf_priv_t *buf_priv;
+ 	drm_mga_iload_t iload;
+	unsigned long bus_address;
+   
+   	DRM_DEBUG("Starting Iload\n");
+ 	copy_from_user_ret(&iload, (drm_mga_iload_t *)arg, sizeof(iload),
+ 			   -EFAULT);
+
+	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("mga_iload called without lock held\n");
+		return -EINVAL;
+	}
+
+   	buf = dma->buflist[ iload.idx ];
+	buf_priv = buf->dev_private;
+   	bus_address = buf->bus_address;
+   	DRM_DEBUG("bus_address %lx, length %d, destorg : %x\n",
+	       bus_address, iload.length, iload.destOrg);
+   
+	if(mgaVerifyIload(dev_priv, 
+			  bus_address,
+			  iload.destOrg,  
+			  iload.length)) {
+	      	mga_freelist_put(dev, buf);
+	   	return -EINVAL;
+	}
+    
+   	sarea_priv->dirty |= MGA_UPLOAD_CTX;
+
+   	mga_dma_dispatch_tex_blit(dev, bus_address, iload.length, 
+				  iload.destOrg);
+   	buf_priv->my_freelist->age = dev_priv->last_sync_tag;
+	buf_priv->discard = 1;
+    	mga_freelist_put(dev, buf);
+	mga_dma_schedule(dev, 1);
+    	sarea_priv->last_dispatch = status[1];
+ 	return 0; 
+}
+
+int mga_vertex(struct inode *inode, struct file *filp,
+	       unsigned int cmd, unsigned long arg)
+{
+	drm_file_t *priv = filp->private_data;
+	drm_device_t *dev = priv->dev;
+      	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
+   	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+      	__volatile__ unsigned int *status = 
+     		(__volatile__ unsigned int *)dev_priv->status_page;
+   	drm_device_dma_t *dma = dev->dma;
+	drm_buf_t *buf;
+      	drm_mga_buf_priv_t *buf_priv;
+	drm_mga_vertex_t vertex;
+
+	copy_from_user_ret(&vertex, (drm_mga_vertex_t *)arg, sizeof(vertex),
+			   -EFAULT);
+   
+	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("mga_vertex called without lock held\n");
+		return -EINVAL;
+	}
+
+	DRM_DEBUG("mga_vertex\n");
+
+	buf = dma->buflist[ vertex.idx ];
+   	buf_priv = buf->dev_private;
+
+	buf->used = vertex.used;
+	buf_priv->discard = vertex.discard;
+   
+	if (!mgaVerifyState(dev_priv)) {
+	   if (vertex.discard) { 
+	     	buf_priv->my_freelist->age = dev_priv->last_sync_tag;
+	      	mga_freelist_put(dev, buf);
+	   }
+	   return -EINVAL;
+	}
+
+	mga_dma_dispatch_vertex(dev, buf);
+
+   	PRIMUPDATE(dev_priv);
+	mga_dma_schedule(dev, 1);
+      	sarea_priv->last_dispatch = status[1];
+   	return 0;
+}
+
+static int mga_dma_get_buffers(drm_device_t *dev, drm_dma_t *d)
+{
+	int i;
+	drm_buf_t *buf;
+
+	for (i = d->granted_count; i < d->request_count; i++) {
+		buf = mga_freelist_get(dev);
+		if (!buf) break;
+		buf->pid = current->pid;
+		copy_to_user_ret(&d->request_indices[i],
+				 &buf->idx,
+				 sizeof(buf->idx),
+				 -EFAULT);
+		copy_to_user_ret(&d->request_sizes[i],
+				 &buf->total,
+				 sizeof(buf->total),
+				 -EFAULT);
+		++d->granted_count;
+	}
+	return 0;
+}
+
+int mga_dma(struct inode *inode, struct file *filp, unsigned int cmd,
+	    unsigned long arg)
+{
+	drm_file_t	  *priv	    = filp->private_data;
+	drm_device_t	  *dev	    = priv->dev;
+	drm_device_dma_t  *dma	    = dev->dma;
+   	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
+   	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
+      	__volatile__ unsigned int *status = 
+     		(__volatile__ unsigned int *)dev_priv->status_page;
+	int		  retcode   = 0;
+	drm_dma_t	  d;
+
+   	copy_from_user_ret(&d, (drm_dma_t *)arg, sizeof(d), -EFAULT);
+	DRM_DEBUG("%d %d: %d send, %d req\n",
+		  current->pid, d.context, d.send_count, d.request_count);
+
+	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("mga_dma called without lock held\n");
+		return -EINVAL;
+	}
+
+	/* Please don't send us buffers.
+	 */
+	if (d.send_count != 0) {
+		DRM_ERROR("Process %d trying to send %d buffers via drmDMA\n",
+			  current->pid, d.send_count);
+		return -EINVAL;
+	}
+	
+	/* We'll send you buffers.
+	 */
+	if (d.request_count < 0 || d.request_count > dma->buf_count) {
+		DRM_ERROR("Process %d trying to get %d buffers (of %d max)\n",
+			  current->pid, d.request_count, dma->buf_count);
+		return -EINVAL;
+	}
+	
+	d.granted_count = 0;
+
+	if (d.request_count) {
+		retcode = mga_dma_get_buffers(dev, &d);
+	}
+
+	DRM_DEBUG("%d returning, granted = %d\n",
+		  current->pid, d.granted_count);
+	copy_to_user_ret((drm_dma_t *)arg, &d, sizeof(d), -EFAULT);
+      	sarea_priv->last_dispatch = status[1];
+	return retcode;
+}
diff --git a/linux/mga_state.h b/linux/mga_state.h
deleted file mode 100644
index e7b952e0..00000000
--- a/linux/mga_state.h
+++ /dev/null
@@ -1,13 +0,0 @@
-#ifndef MGA_STATE_H
-#define MGA_STATE_H
-
-#include "mga_drv.h"
-
-int mgaCopyAndVerifyState( drm_mga_private_t *dev_priv, 
-			   drm_mga_buf_priv_t *buf_priv );
-
-void mgaEmitClipRect( drm_mga_private_t *dev_priv, xf86drmClipRectRec *box );
-
-void mgaEmitState( drm_mga_private_t *dev_priv, drm_mga_buf_priv_t *buf_priv );
-
-#endif
diff --git a/linux/mgareg_flags.h b/linux/mgareg_flags.h
deleted file mode 100644
index 901f1831..00000000
--- a/linux/mgareg_flags.h
+++ /dev/null
@@ -1,930 +0,0 @@
-/* author: stephen crowley, crow@debian.org */
-
-/*
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included
- * in all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * STEPHEN CROWLEY, OR ANY OTHER CONTRIBUTORS BE LIABLE FOR ANY CLAIM, 
- * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
- * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE 
- * OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- */
-
-#ifndef _MGAREGS_H_
-#define _MGAREGS_H_
-
-/*************** (START) AUTOMATICLY GENERATED REGISTER FILE *****************/
-/*
- * Generated on Sat Nov 20 21:25:36 CST 1999
- */
-
-
-
-/*
- * Power Graphic Mode Memory Space Registers
- */
-
-    #define AGP_PLL_agp2xpllen_MASK 	0xfffffffe 	/* bit 0 */
-    #define AGP_PLL_agp2xpllen_disable 	0x0 		
-    #define AGP_PLL_agp2xpllen_enable 	0x1 		
-
-    #define AC_src_MASK 		0xfffffff0 	/* bits 0-3 */
-    #define AC_src_zero 		0x0 		/* val 0, shift 0 */
-    #define AC_src_one 			0x1 		/* val 1, shift 0 */
-    #define AC_src_dst_color 		0x2 		/* val 2, shift 0 */
-    #define AC_src_om_dst_color 	0x3 		/* val 3, shift 0 */
-    #define AC_src_src_alpha 		0x4 		/* val 4, shift 0 */
-    #define AC_src_om_src_alpha 	0x5 		/* val 5, shift 0 */
-    #define AC_src_dst_alpha 		0x6 		/* val 6, shift 0 */
-    #define AC_src_om_dst_alpha 	0x7 		/* val 7, shift 0 */
-    #define AC_src_src_alpha_sat 	0x8 		/* val 8, shift 0 */
-    #define AC_dst_MASK 		0xffffff0f 	/* bits 4-7 */
-    #define AC_dst_zero 		0x0 		/* val 0, shift 4 */
-    #define AC_dst_one 			0x10 		/* val 1, shift 4 */
-    #define AC_dst_src_color 		0x20 		/* val 2, shift 4 */
-    #define AC_dst_om_src_color 	0x30 		/* val 3, shift 4 */
-    #define AC_dst_src_alpha 		0x40 		/* val 4, shift 4 */
-    #define AC_dst_om_src_alpha 	0x50 		/* val 5, shift 4 */
-    #define AC_dst_dst_alpha 		0x60 		/* val 6, shift 4 */
-    #define AC_dst_om_dst_alpha 	0x70 		/* val 7, shift 4 */
-    #define AC_amode_MASK 		0xfffffcff 	/* bits 8-9 */
-    #define AC_amode_FCOL 		0x0 		/* val 0, shift 8 */
-    #define AC_amode_alpha_channel 	0x100 		/* val 1, shift 8 */
-    #define AC_amode_video_alpha 	0x200 		/* val 2, shift 8 */
-    #define AC_amode_RSVD 		0x300 		/* val 3, shift 8 */
-    #define AC_astipple_MASK 		0xfffff7ff 	/* bit 11 */
-    #define AC_astipple_disable 	0x0 		
-    #define AC_astipple_enable 		0x800 		
-    #define AC_aten_MASK 		0xffffefff 	/* bit 12 */
-    #define AC_aten_disable 		0x0 		
-    #define AC_aten_enable 		0x1000 		
-    #define AC_atmode_MASK 		0xffff1fff 	/* bits 13-15 */
-    #define AC_atmode_noacmp 		0x0 		/* val 0, shift 13 */
-    #define AC_atmode_ae 		0x4000 		/* val 2, shift 13 */
-    #define AC_atmode_ane 		0x6000 		/* val 3, shift 13 */
-    #define AC_atmode_alt 		0x8000 		/* val 4, shift 13 */
-    #define AC_atmode_alte 		0xa000 		/* val 5, shift 13 */
-    #define AC_atmode_agt 		0xc000 		/* val 6, shift 13 */
-    #define AC_atmode_agte 		0xe000 		/* val 7, shift 13 */
-    #define AC_atref_MASK 		0xff00ffff 	/* bits 16-23 */
-    #define AC_atref_SHIFT 		16 		
-    #define AC_alphasel_MASK 		0xfcffffff 	/* bits 24-25 */
-    #define AC_alphasel_fromtex 	0x0 		/* val 0, shift 24 */
-    #define AC_alphasel_diffused 	0x1000000 	/* val 1, shift 24 */
-    #define AC_alphasel_modulated 	0x2000000 	/* val 2, shift 24 */
-    #define AC_alphasel_trans 		0x3000000 	/* val 3, shift 24 */
-
-    #define AR0_ar0_MASK 		0xfffc0000 	/* bits 0-17 */
-    #define AR0_ar0_SHIFT 		0 		
-
-    #define AR1_ar1_MASK 		0xff000000 	/* bits 0-23 */
-    #define AR1_ar1_SHIFT 		0 		
-
-    #define AR2_ar2_MASK 		0xfffc0000 	/* bits 0-17 */
-    #define AR2_ar2_SHIFT 		0 		
-
-    #define AR3_ar3_MASK 		0xff000000 	/* bits 0-23 */
-    #define AR3_ar3_SHIFT 		0 		
-    #define AR3_spage_MASK 		0xf8ffffff 	/* bits 24-26 */
-    #define AR3_spage_SHIFT 		24 		
-
-    #define AR4_ar4_MASK 		0xfffc0000 	/* bits 0-17 */
-    #define AR4_ar4_SHIFT 		0 		
-
-    #define AR5_ar5_MASK 		0xfffc0000 	/* bits 0-17 */
-    #define AR5_ar5_SHIFT 		0 		
-
-    #define AR6_ar6_MASK 		0xfffc0000 	/* bits 0-17 */
-    #define AR6_ar6_SHIFT 		0 		
-
-    #define BC_besen_MASK 		0xfffffffe 	/* bit 0 */
-    #define BC_besen_disable 		0x0 		
-    #define BC_besen_enable 		0x1 		
-    #define BC_besv1srcstp_MASK 	0xffffffbf 	/* bit 6 */
-    #define BC_besv1srcstp_even 	0x0 		
-    #define BC_besv1srcstp_odd 		0x40 		
-    #define BC_besv2srcstp_MASK 	0xfffffeff 	/* bit 8 */
-    #define BC_besv2srcstp_disable 	0x0 		
-    #define BC_besv2srcstp_enable 	0x100 		
-    #define BC_beshfen_MASK 		0xfffffbff 	/* bit 10 */
-    #define BC_beshfen_disable 		0x0 		
-    #define BC_beshfen_enable 		0x400 		
-    #define BC_besvfen_MASK 		0xfffff7ff 	/* bit 11 */
-    #define BC_besvfen_disable 		0x0 		
-    #define BC_besvfen_enable 		0x800 		
-    #define BC_beshfixc_MASK 		0xffffefff 	/* bit 12 */
-    #define BC_beshfixc_weight 		0x0 		
-    #define BC_beshfixc_coeff 		0x1000 		
-    #define BC_bescups_MASK 		0xfffeffff 	/* bit 16 */
-    #define BC_bescups_disable 		0x0 		
-    #define BC_bescups_enable 		0x10000 	
-    #define BC_bes420pl_MASK 		0xfffdffff 	/* bit 17 */
-    #define BC_bes420pl_422 		0x0 		
-    #define BC_bes420pl_420 		0x20000 	
-    #define BC_besdith_MASK 		0xfffbffff 	/* bit 18 */
-    #define BC_besdith_disable 		0x0 		
-    #define BC_besdith_enable 		0x40000 	
-    #define BC_beshmir_MASK 		0xfff7ffff 	/* bit 19 */
-    #define BC_beshmir_disable 		0x0 		
-    #define BC_beshmir_enable 		0x80000 	
-    #define BC_besbwen_MASK 		0xffefffff 	/* bit 20 */
-    #define BC_besbwen_color 		0x0 		
-    #define BC_besbwen_bw 		0x100000 	
-    #define BC_besblank_MASK 		0xffdfffff 	/* bit 21 */
-    #define BC_besblank_disable 	0x0 		
-    #define BC_besblank_enable 		0x200000 	
-    #define BC_besfselm_MASK 		0xfeffffff 	/* bit 24 */
-    #define BC_besfselm_soft 		0x0 		
-    #define BC_besfselm_hard 		0x1000000 	
-    #define BC_besfsel_MASK 		0xf9ffffff 	/* bits 25-26 */
-    #define BC_besfsel_a1 		0x0 		/* val 0, shift 25 */
-    #define BC_besfsel_a2 		0x2000000 	/* val 1, shift 25 */
-    #define BC_besfsel_b1 		0x4000000 	/* val 2, shift 25 */
-    #define BC_besfsel_b2 		0x6000000 	/* val 3, shift 25 */
-
-    #define BGC_beshzoom_MASK 		0xfffffffe 	/* bit 0 */
-    #define BGC_beshzoom_disable 	0x0 		
-    #define BGC_beshzoom_enable 	0x1 		
-    #define BGC_beshzoomf_MASK 		0xfffffffd 	/* bit 1 */
-    #define BGC_beshzoomf_disable 	0x0 		
-    #define BGC_beshzoomf_enable 	0x2 		
-    #define BGC_bescorder_MASK 		0xfffffff7 	/* bit 3 */
-    #define BGC_bescorder_even 		0x0 		
-    #define BGC_bescorder_odd 		0x8 		
-    #define BGC_besreghup_MASK 		0xffffffef 	/* bit 4 */
-    #define BGC_besreghup_disable 	0x0 		
-    #define BGC_besreghup_enable 	0x10 		
-    #define BGC_besvcnt_MASK 		0xf000ffff 	/* bits 16-27 */
-    #define BGC_besvcnt_SHIFT 		16 		
-
-    #define BHC_besright_MASK 		0xfffff800 	/* bits 0-10 */
-    #define BHC_besright_SHIFT 		0 		
-    #define BHC_besleft_MASK 		0xf800ffff 	/* bits 16-26 */
-    #define BHC_besleft_SHIFT 		16 		
-
-    #define BHISF_beshiscal_MASK 	0xffe00003 	/* bits 2-20 */
-    #define BHISF_beshiscal_SHIFT 	2 		
-
-    #define BHSE_beshsrcend_MASK 	0xfc000003 	/* bits 2-25 */
-    #define BHSE_beshsrcend_SHIFT 	2 		
-
-    #define BHSL_beshsrclst_MASK 	0xfc00ffff 	/* bits 16-25 */
-    #define BHSL_beshsrclst_SHIFT 	16 		
-
-    #define BHSS_beshsrcst_MASK 	0xfc000003 	/* bits 2-25 */
-    #define BHSS_beshsrcst_SHIFT 	2 		
-
-    #define BP_bespitch_MASK 		0xfffff000 	/* bits 0-11 */
-    #define BP_bespitch_SHIFT 		0 		
-
-    #define BS_besstat_MASK 		0xfffffffc 	/* bits 0-1 */
-    #define BS_besstat_a1 		0x0 		/* val 0, shift 0 */
-    #define BS_besstat_a2 		0x1 		/* val 1, shift 0 */
-    #define BS_besstat_b1 		0x2 		/* val 2, shift 0 */
-    #define BS_besstat_b2 		0x3 		/* val 3, shift 0 */
-
-    #define BSF_besv1srclast_MASK 	0xfffffc00 	/* bits 0-9 */
-    #define BSF_besv1srclast_SHIFT 	0 		
-
-    #define BSF_besv2srclst_MASK 	0xfffffc00 	/* bits 0-9 */
-    #define BSF_besv2srclst_SHIFT 	0 		
-
-    #define BSF_besv1wght_MASK 		0xffff0003 	/* bits 2-15 */
-    #define BSF_besv1wght_SHIFT 	2 		
-    #define BSF_besv1wghts_MASK 	0xfffeffff 	/* bit 16 */
-    #define BSF_besv1wghts_disable 	0x0 		
-    #define BSF_besv1wghts_enable 	0x10000 	
-
-    #define BSF_besv2wght_MASK 		0xffff0003 	/* bits 2-15 */
-    #define BSF_besv2wght_SHIFT 	2 		
-    #define BSF_besv2wghts_MASK 	0xfffeffff 	/* bit 16 */
-    #define BSF_besv2wghts_disable 	0x0 		
-    #define BSF_besv2wghts_enable 	0x10000 	
-
-    #define BVC_besbot_MASK 		0xfffff800 	/* bits 0-10 */
-    #define BVC_besbot_SHIFT 		0 		
-    #define BVC_bestop_MASK 		0xf800ffff 	/* bits 16-26 */
-    #define BVC_bestop_SHIFT 		16 		
-
-    #define BVISF_besviscal_MASK 	0xffe00003 	/* bits 2-20 */
-    #define BVISF_besviscal_SHIFT 	2 		
-
-    #define CXB_cxleft_MASK 		0xfffff000 	/* bits 0-11 */
-    #define CXB_cxleft_SHIFT 		0 		
-    #define CXB_cxright_MASK 		0xf000ffff 	/* bits 16-27 */
-    #define CXB_cxright_SHIFT 		16 		
-
-    #define DO_dstmap_MASK 		0xfffffffe 	/* bit 0 */
-    #define DO_dstmap_fb 		0x0 		
-    #define DO_dstmap_sys 		0x1 		
-    #define DO_dstacc_MASK 		0xfffffffd 	/* bit 1 */
-    #define DO_dstacc_pci 		0x0 		
-    #define DO_dstacc_agp 		0x2 		
-    #define DO_dstorg_MASK 		0x7 		/* bits 3-31 */
-    #define DO_dstorg_SHIFT 		3 		
-
-    #define DC_opcod_MASK 		0xfffffff0 	/* bits 0-3 */
-    #define DC_opcod_line_open 		0x0 		/* val 0, shift 0 */
-    #define DC_opcod_autoline_open 	0x1 		/* val 1, shift 0 */
-    #define DC_opcod_line_close 	0x2 		/* val 2, shift 0 */
-    #define DC_opcod_autoline_close 	0x3 		/* val 3, shift 0 */
-    #define DC_opcod_trap 		0x4 		/* val 4, shift 0 */
-    #define DC_opcod_texture_trap 	0x6 		/* val 6, shift 0 */
-    #define DC_opcod_bitblt 		0x8 		/* val 8, shift 0 */
-    #define DC_opcod_iload 		0x9 		/* val 9, shift 0 */
-    #define DC_atype_MASK 		0xffffff8f 	/* bits 4-6 */
-    #define DC_atype_rpl 		0x0 		/* val 0, shift 4 */
-    #define DC_atype_rstr 		0x10 		/* val 1, shift 4 */
-    #define DC_atype_zi 		0x30 		/* val 3, shift 4 */
-    #define DC_atype_blk 		0x40 		/* val 4, shift 4 */
-    #define DC_atype_i 			0x70 		/* val 7, shift 4 */
-    #define DC_linear_MASK 		0xffffff7f 	/* bit 7 */
-    #define DC_linear_xy 		0x0 		
-    #define DC_linear_linear 		0x80 		
-    #define DC_zmode_MASK 		0xfffff8ff 	/* bits 8-10 */
-    #define DC_zmode_nozcmp 		0x0 		/* val 0, shift 8 */
-    #define DC_zmode_ze 		0x200 		/* val 2, shift 8 */
-    #define DC_zmode_zne 		0x300 		/* val 3, shift 8 */
-    #define DC_zmode_zlt 		0x400 		/* val 4, shift 8 */
-    #define DC_zmode_zlte 		0x500 		/* val 5, shift 8 */
-    #define DC_zmode_zgt 		0x600 		/* val 6, shift 8 */
-    #define DC_zmode_zgte 		0x700 		/* val 7, shift 8 */
-    #define DC_solid_MASK 		0xfffff7ff 	/* bit 11 */
-    #define DC_solid_disable 		0x0 		
-    #define DC_solid_enable 		0x800 		
-    #define DC_arzero_MASK 		0xffffefff 	/* bit 12 */
-    #define DC_arzero_disable 		0x0 		
-    #define DC_arzero_enable 		0x1000 		
-    #define DC_sgnzero_MASK 		0xffffdfff 	/* bit 13 */
-    #define DC_sgnzero_disable 		0x0 		
-    #define DC_sgnzero_enable 		0x2000 		
-    #define DC_shftzero_MASK 		0xffffbfff 	/* bit 14 */
-    #define DC_shftzero_disable 	0x0 		
-    #define DC_shftzero_enable 		0x4000 		
-    #define DC_bop_MASK 		0xfff0ffff 	/* bits 16-19 */
-    #define DC_bop_SHIFT 		16 		
-    #define DC_trans_MASK 		0xff0fffff 	/* bits 20-23 */
-    #define DC_trans_SHIFT 		20 		
-    #define DC_bltmod_MASK 		0xe1ffffff 	/* bits 25-28 */
-    #define DC_bltmod_bmonolef 		0x0 		/* val 0, shift 25 */
-    #define DC_bltmod_bmonowf 		0x8000000 	/* val 4, shift 25 */
-    #define DC_bltmod_bplan 		0x2000000 	/* val 1, shift 25 */
-    #define DC_bltmod_bfcol 		0x4000000 	/* val 2, shift 25 */
-    #define DC_bltmod_bu32bgr 		0x6000000 	/* val 3, shift 25 */
-    #define DC_bltmod_bu32rgb 		0xe000000 	/* val 7, shift 25 */
-    #define DC_bltmod_bu24bgr 		0x16000000 	/* val 11, shift 25 */
-    #define DC_bltmod_bu24rgb 		0x1e000000 	/* val 15, shift 25 */
-    #define DC_pattern_MASK 		0xdfffffff 	/* bit 29 */
-    #define DC_pattern_disable 		0x0 		
-    #define DC_pattern_enable 		0x20000000 	
-    #define DC_transc_MASK 		0xbfffffff 	/* bit 30 */
-    #define DC_transc_disable 		0x0 		
-    #define DC_transc_enable 		0x40000000 	
-    #define DC_clipdis_MASK 		0x7fffffff 	/* bit 31 */
-    #define DC_clipdis_disable 		0x0 		
-    #define DC_clipdis_enable 		0x80000000 	
-
-    #define DS_dwgsyncaddr_MASK 	0x3 		/* bits 2-31 */
-    #define DS_dwgsyncaddr_SHIFT 	2 		
-
-    #define FS_fifocount_MASK 		0xffffff80 	/* bits 0-6 */
-    #define FS_fifocount_SHIFT 		0 		
-    #define FS_bfull_MASK 		0xfffffeff 	/* bit 8 */
-    #define FS_bfull_disable 		0x0 		
-    #define FS_bfull_enable 		0x100 		
-    #define FS_bempty_MASK 		0xfffffdff 	/* bit 9 */
-    #define FS_bempty_disable 		0x0 		
-    #define FS_bempty_enable 		0x200 		
-
-    #define XA_fxleft_MASK 		0xffff0000 	/* bits 0-15 */
-    #define XA_fxleft_SHIFT 		0 		
-    #define XA_fxright_MASK 		0xffff 		/* bits 16-31 */
-    #define XA_fxright_SHIFT 		16 		
-
-    #define IC_softrapiclr_MASK 	0xfffffffe 	/* bit 0 */
-    #define IC_softrapiclr_disable 	0x0 		
-    #define IC_softrapiclr_enable 	0x1 		
-    #define IC_pickiclr_MASK 		0xfffffffb 	/* bit 2 */
-    #define IC_pickiclr_disable 	0x0 		
-    #define IC_pickiclr_enable 		0x4 		
-    #define IC_vlineiclr_MASK 		0xffffffdf 	/* bit 5 */
-    #define IC_vlineiclr_disable 	0x0 		
-    #define IC_vlineiclr_enable 	0x20 		
-    #define IC_wiclr_MASK 		0xffffff7f 	/* bit 7 */
-    #define IC_wiclr_disable 		0x0 		
-    #define IC_wiclr_enable 		0x80 		
-    #define IC_wciclr_MASK 		0xfffffeff 	/* bit 8 */
-    #define IC_wciclr_disable 		0x0 		
-    #define IC_wciclr_enable 		0x100 		
-
-    #define IE_softrapien_MASK 		0xfffffffe 	/* bit 0 */
-    #define IE_softrapien_disable 	0x0 		
-    #define IE_softrapien_enable 	0x1 		
-    #define IE_pickien_MASK 		0xfffffffb 	/* bit 2 */
-    #define IE_pickien_disable 		0x0 		
-    #define IE_pickien_enable 		0x4 		
-    #define IE_vlineien_MASK 		0xffffffdf 	/* bit 5 */
-    #define IE_vlineien_disable 	0x0 		
-    #define IE_vlineien_enable 		0x20 		
-    #define IE_extien_MASK 		0xffffffbf 	/* bit 6 */
-    #define IE_extien_disable 		0x0 		
-    #define IE_extien_enable 		0x40 		
-    #define IE_wien_MASK 		0xffffff7f 	/* bit 7 */
-    #define IE_wien_disable 		0x0 		
-    #define IE_wien_enable 		0x80 		
-    #define IE_wcien_MASK 		0xfffffeff 	/* bit 8 */
-    #define IE_wcien_disable 		0x0 		
-    #define IE_wcien_enable 		0x100 		
-
-    #define MA_pwidth_MASK 		0xfffffffc 	/* bits 0-1 */
-    #define MA_pwidth_8 		0x0 		/* val 0, shift 0 */
-    #define MA_pwidth_16 		0x1 		/* val 1, shift 0 */
-    #define MA_pwidth_32 		0x2 		/* val 2, shift 0 */
-    #define MA_pwidth_24 		0x3 		/* val 3, shift 0 */
-    #define MA_zwidth_MASK 		0xffffffe7 	/* bits 3-4 */
-    #define MA_zwidth_16 		0x0 		/* val 0, shift 3 */
-    #define MA_zwidth_32 		0x8 		/* val 1, shift 3 */
-    #define MA_zwidth_15 		0x10 		/* val 2, shift 3 */
-    #define MA_zwidth_24 		0x18 		/* val 3, shift 3 */
-    #define MA_memreset_MASK 		0xffff7fff 	/* bit 15 */
-    #define MA_memreset_disable 	0x0 		
-    #define MA_memreset_enable 		0x8000 		
-    #define MA_fogen_MASK 		0xfbffffff 	/* bit 26 */
-    #define MA_fogen_disable 		0x0 		
-    #define MA_fogen_enable 		0x4000000 	
-    #define MA_tlutload_MASK 		0xdfffffff 	/* bit 29 */
-    #define MA_tlutload_disable 	0x0 		
-    #define MA_tlutload_enable 		0x20000000 	
-    #define MA_nodither_MASK 		0xbfffffff 	/* bit 30 */
-    #define MA_nodither_disable 	0x0 		
-    #define MA_nodither_enable 		0x40000000 	
-    #define MA_dit555_MASK 		0x7fffffff 	/* bit 31 */
-    #define MA_dit555_disable 		0x0 		
-    #define MA_dit555_enable 		0x80000000 	
-
-    #define MCWS_casltncy_MASK 		0xfffffff8 	/* bits 0-2 */
-    #define MCWS_casltncy_SHIFT 	0 		
-    #define MCWS_rrddelay_MASK 		0xffffffcf 	/* bits 4-5 */
-    #define MCWS_rcddelay_MASK 		0xfffffe7f 	/* bits 7-8 */
-    #define MCWS_rasmin_MASK 		0xffffe3ff 	/* bits 10-12 */
-    #define MCWS_rasmin_SHIFT 		10 		
-    #define MCWS_rpdelay_MASK 		0xffff3fff 	/* bits 14-15 */
-    #define MCWS_wrdelay_MASK 		0xfff3ffff 	/* bits 18-19 */
-    #define MCWS_rddelay_MASK 		0xffdfffff 	/* bit 21 */
-    #define MCWS_rddelay_disable 	0x0 		
-    #define MCWS_rddelay_enable 	0x200000 	
-    #define MCWS_smrdelay_MASK 		0xfe7fffff 	/* bits 23-24 */
-    #define MCWS_bwcdelay_MASK 		0xf3ffffff 	/* bits 26-27 */
-    #define MCWS_bpldelay_MASK 		0x1fffffff 	/* bits 29-31 */
-    #define MCWS_bpldelay_SHIFT 	29 		
-
-    #define MRB_mclkbrd0_MASK 		0xfffffff0 	/* bits 0-3 */
-    #define MRB_mclkbrd0_SHIFT 		0 		
-    #define MRB_mclkbrd1_MASK 		0xfffffe1f 	/* bits 5-8 */
-    #define MRB_mclkbrd1_SHIFT 		5 		
-    #define MRB_strmfctl_MASK 		0xff3fffff 	/* bits 22-23 */
-    #define MRB_mrsopcod_MASK 		0xe1ffffff 	/* bits 25-28 */
-    #define MRB_mrsopcod_SHIFT 		25 		
-
-    #define OM_dmamod_MASK 		0xfffffff3 	/* bits 2-3 */
-    #define OM_dmamod_general 		0x0 		/* val 0, shift 2 */
-    #define OM_dmamod_blit 		0x4 		/* val 1, shift 2 */
-    #define OM_dmamod_vector 		0x8 		/* val 2, shift 2 */
-    #define OM_dmamod_vertex 		0xc 		/* val 3, shift 2 */
-    #define OM_dmadatasiz_MASK 		0xfffffcff 	/* bits 8-9 */
-    #define OM_dmadatasiz_8 		0x0 		/* val 0, shift 8 */
-    #define OM_dmadatasiz_16 		0x100 		/* val 1, shift 8 */
-    #define OM_dmadatasiz_32 		0x200 		/* val 2, shift 8 */
-    #define OM_dirdatasiz_MASK 		0xfffcffff 	/* bits 16-17 */
-    #define OM_dirdatasiz_8 		0x0 		/* val 0, shift 16 */
-    #define OM_dirdatasiz_16 		0x10000 	/* val 1, shift 16 */
-    #define OM_dirdatasiz_32 		0x20000 	/* val 2, shift 16 */
-
-    #define P_iy_MASK 			0xffffe000 	/* bits 0-12 */
-    #define P_iy_SHIFT 			0 		
-    #define P_ylin_MASK 		0xffff7fff 	/* bit 15 */
-    #define P_ylin_disable 		0x0 		
-    #define P_ylin_enable 		0x8000 		
-
-    #define PDCA_primod_MASK 		0xfffffffc 	/* bits 0-1 */
-    #define PDCA_primod_general 	0x0 		/* val 0, shift 0 */
-    #define PDCA_primod_blit 		0x1 		/* val 1, shift 0 */
-    #define PDCA_primod_vector 		0x2 		/* val 2, shift 0 */
-    #define PDCA_primod_vertex 		0x3 		/* val 3, shift 0 */
-    #define PDCA_primaddress_MASK 	0x3 		/* bits 2-31 */
-    #define PDCA_primaddress_SHIFT 	2 		
-
-    #define PDEA_primnostart_MASK 	0xfffffffe 	/* bit 0 */
-    #define PDEA_primnostart_disable 	0x0 		
-    #define PDEA_primnostart_enable 	0x1 		
-    #define PDEA_pagpxfer_MASK 		0xfffffffd 	/* bit 1 */
-    #define PDEA_pagpxfer_disable 	0x0 		
-    #define PDEA_pagpxfer_enable 	0x2 		
-    #define PDEA_primend_MASK 		0x3 		/* bits 2-31 */
-    #define PDEA_primend_SHIFT 		2 		
-
-    #define PLS_primptren0_MASK 	0xfffffffe 	/* bit 0 */
-    #define PLS_primptren0_disable 	0x0 		
-    #define PLS_primptren0_enable 	0x1 		
-    #define PLS_primptren1_MASK 	0xfffffffd 	/* bit 1 */
-    #define PLS_primptren1_disable 	0x0 		
-    #define PLS_primptren1_enable 	0x2 		
-    #define PLS_primptr_MASK 		0x7 		/* bits 3-31 */
-    #define PLS_primptr_SHIFT 		3 		
-
-    #define R_softreset_MASK 		0xfffffffe 	/* bit 0 */
-    #define R_softreset_disable 	0x0 		
-    #define R_softreset_enable 		0x1 		
-    #define R_softextrst_MASK 		0xfffffffd 	/* bit 1 */
-    #define R_softextrst_disable 	0x0 		
-    #define R_softextrst_enable 	0x2 		
-
-    #define SDCA_secmod_MASK 		0xfffffffc 	/* bits 0-1 */
-    #define SDCA_secmod_general 	0x0 		/* val 0, shift 0 */
-    #define SDCA_secmod_blit 		0x1 		/* val 1, shift 0 */
-    #define SDCA_secmod_vector 		0x2 		/* val 2, shift 0 */
-    #define SDCA_secmod_vertex 		0x3 		/* val 3, shift 0 */
-    #define SDCA_secaddress_MASK 	0x3 		/* bits 2-31 */
-    #define SDCA_secaddress_SHIFT 	2 		
-
-    #define SDEA_sagpxfer_MASK 		0xfffffffd 	/* bit 1 */
-    #define SDEA_sagpxfer_disable 	0x0 		
-    #define SDEA_sagpxfer_enable 	0x2 		
-    #define SDEA_secend_MASK 		0x3 		/* bits 2-31 */
-    #define SDEA_secend_SHIFT 		2 		
-
-    #define SETDCA_setupmod_MASK 	0xfffffffc 	/* bits 0-1 */
-    #define SETDCA_setupmod_vertlist 	0x0 		/* val 0, shift 0 */
-    #define SETDCA_setupaddress_MASK 	0x3 		/* bits 2-31 */
-    #define SETDCA_setupaddress_SHIFT 	2 		
-
-    #define SETDEA_setupagpxfer_MASK 	0xfffffffd 	/* bit 1 */
-    #define SETDEA_setupagpxfer_disable 0x0 		
-    #define SETDEA_setupagpxfer_enable 	0x2 		
-    #define SETDEA_setupend_MASK 	0x3 		/* bits 2-31 */
-    #define SETDEA_setupend_SHIFT 	2 		
-
-    #define S_sdydxl_MASK 		0xfffffffe 	/* bit 0 */
-    #define S_sdydxl_y 			0x0 		
-    #define S_sdydxl_x 			0x1 		
-    #define S_scanleft_MASK 		0xfffffffe 	/* bit 0 */
-    #define S_scanleft_disable 		0x0 		
-    #define S_scanleft_enable 		0x1 		
-    #define S_sdxl_MASK 		0xfffffffd 	/* bit 1 */
-    #define S_sdxl_pos 			0x0 		
-    #define S_sdxl_neg 			0x2 		
-    #define S_sdy_MASK 			0xfffffffb 	/* bit 2 */
-    #define S_sdy_pos 			0x0 		
-    #define S_sdy_neg 			0x4 		
-    #define S_sdxr_MASK 		0xffffffdf 	/* bit 5 */
-    #define S_sdxr_pos 			0x0 		
-    #define S_sdxr_neg 			0x20 		
-    #define S_brkleft_MASK 		0xfffffeff 	/* bit 8 */
-    #define S_brkleft_disable 		0x0 		
-    #define S_brkleft_enable 		0x100 		
-    #define S_errorinit_MASK 		0x7fffffff 	/* bit 31 */
-    #define S_errorinit_disable 	0x0 		
-    #define S_errorinit_enable 		0x80000000 	
-
-    #define FSC_x_off_MASK 		0xfffffff0 	/* bits 0-3 */
-    #define FSC_x_off_SHIFT 		0 		
-    #define FSC_funcnt_MASK 		0xffffff80 	/* bits 0-6 */
-    #define FSC_funcnt_SHIFT 		0 		
-    #define FSC_y_off_MASK 		0xffffff8f 	/* bits 4-6 */
-    #define FSC_y_off_SHIFT 		4 		
-    #define FSC_funoff_MASK 		0xffc0ffff 	/* bits 16-21 */
-    #define FSC_funoff_SHIFT 		16 		
-    #define FSC_stylelen_MASK 		0xffc0ffff 	/* bits 16-21 */
-    #define FSC_stylelen_SHIFT 		16 		
-
-
-    #define STH_softraphand_MASK 	0x3 		/* bits 2-31 */
-    #define STH_softraphand_SHIFT 	2 		
-
-    #define SO_srcmap_MASK 		0xfffffffe 	/* bit 0 */
-    #define SO_srcmap_fb 		0x0 		
-    #define SO_srcmap_sys 		0x1 		
-    #define SO_srcacc_MASK 		0xfffffffd 	/* bit 1 */
-    #define SO_srcacc_pci 		0x0 		
-    #define SO_srcacc_agp 		0x2 		
-    #define SO_srcorg_MASK 		0x7 		/* bits 3-31 */
-    #define SO_srcorg_SHIFT 		3 		
-
-    #define STAT_softrapen_MASK 	0xfffffffe 	/* bit 0 */
-    #define STAT_softrapen_disable 	0x0 		
-    #define STAT_softrapen_enable 	0x1 		
-    #define STAT_pickpen_MASK 		0xfffffffb 	/* bit 2 */
-    #define STAT_pickpen_disable 	0x0 		
-    #define STAT_pickpen_enable 	0x4 		
-    #define STAT_vsyncsts_MASK 		0xfffffff7 	/* bit 3 */
-    #define STAT_vsyncsts_disable 	0x0 		
-    #define STAT_vsyncsts_enable 	0x8 		
-    #define STAT_vsyncpen_MASK 		0xffffffef 	/* bit 4 */
-    #define STAT_vsyncpen_disable 	0x0 		
-    #define STAT_vsyncpen_enable 	0x10 		
-    #define STAT_vlinepen_MASK 		0xffffffdf 	/* bit 5 */
-    #define STAT_vlinepen_disable 	0x0 		
-    #define STAT_vlinepen_enable 	0x20 		
-    #define STAT_extpen_MASK 		0xffffffbf 	/* bit 6 */
-    #define STAT_extpen_disable 	0x0 		
-    #define STAT_extpen_enable 		0x40 		
-    #define STAT_wpen_MASK 		0xffffff7f 	/* bit 7 */
-    #define STAT_wpen_disable 		0x0 		
-    #define STAT_wpen_enable 		0x80 		
-    #define STAT_wcpen_MASK 		0xfffffeff 	/* bit 8 */
-    #define STAT_wcpen_disable 		0x0 		
-    #define STAT_wcpen_enable 		0x100 		
-    #define STAT_dwgengsts_MASK 	0xfffeffff 	/* bit 16 */
-    #define STAT_dwgengsts_disable 	0x0 		
-    #define STAT_dwgengsts_enable 	0x10000 	
-    #define STAT_endprdmasts_MASK 	0xfffdffff 	/* bit 17 */
-    #define STAT_endprdmasts_disable 	0x0 		
-    #define STAT_endprdmasts_enable 	0x20000 	
-    #define STAT_wbusy_MASK 		0xfffbffff 	/* bit 18 */
-    #define STAT_wbusy_disable 		0x0 		
-    #define STAT_wbusy_enable 		0x40000 	
-    #define STAT_swflag_MASK 		0xfffffff 	/* bits 28-31 */
-    #define STAT_swflag_SHIFT 		28 		
-
-    #define S_sref_MASK 		0xffffff00 	/* bits 0-7 */
-    #define S_sref_SHIFT 		0 		
-    #define S_smsk_MASK 		0xffff00ff 	/* bits 8-15 */
-    #define S_smsk_SHIFT 		8 		
-    #define S_swtmsk_MASK 		0xff00ffff 	/* bits 16-23 */
-    #define S_swtmsk_SHIFT 		16 		
-
-    #define SC_smode_MASK 		0xfffffff8 	/* bits 0-2 */
-    #define SC_smode_salways 		0x0 		/* val 0, shift 0 */
-    #define SC_smode_snever 		0x1 		/* val 1, shift 0 */
-    #define SC_smode_se 		0x2 		/* val 2, shift 0 */
-    #define SC_smode_sne 		0x3 		/* val 3, shift 0 */
-    #define SC_smode_slt 		0x4 		/* val 4, shift 0 */
-    #define SC_smode_slte 		0x5 		/* val 5, shift 0 */
-    #define SC_smode_sgt 		0x6 		/* val 6, shift 0 */
-    #define SC_smode_sgte 		0x7 		/* val 7, shift 0 */
-    #define SC_sfailop_MASK 		0xffffffc7 	/* bits 3-5 */
-    #define SC_sfailop_keep 		0x0 		/* val 0, shift 3 */
-    #define SC_sfailop_zero 		0x8 		/* val 1, shift 3 */
-    #define SC_sfailop_replace 		0x10 		/* val 2, shift 3 */
-    #define SC_sfailop_incrsat 		0x18 		/* val 3, shift 3 */
-    #define SC_sfailop_decrsat 		0x20 		/* val 4, shift 3 */
-    #define SC_sfailop_invert 		0x28 		/* val 5, shift 3 */
-    #define SC_sfailop_incr 		0x30 		/* val 6, shift 3 */
-    #define SC_sfailop_decr 		0x38 		/* val 7, shift 3 */
-    #define SC_szfailop_MASK 		0xfffffe3f 	/* bits 6-8 */
-    #define SC_szfailop_keep 		0x0 		/* val 0, shift 6 */
-    #define SC_szfailop_zero 		0x40 		/* val 1, shift 6 */
-    #define SC_szfailop_replace 	0x80 		/* val 2, shift 6 */
-    #define SC_szfailop_incrsat 	0xc0 		/* val 3, shift 6 */
-    #define SC_szfailop_decrsat 	0x100 		/* val 4, shift 6 */
-    #define SC_szfailop_invert 		0x140 		/* val 5, shift 6 */
-    #define SC_szfailop_incr 		0x180 		/* val 6, shift 6 */
-    #define SC_szfailop_decr 		0x1c0 		/* val 7, shift 6 */
-    #define SC_szpassop_MASK 		0xfffff1ff 	/* bits 9-11 */
-    #define SC_szpassop_keep 		0x0 		/* val 0, shift 9 */
-    #define SC_szpassop_zero 		0x200 		/* val 1, shift 9 */
-    #define SC_szpassop_replace 	0x400 		/* val 2, shift 9 */
-    #define SC_szpassop_incrsat 	0x600 		/* val 3, shift 9 */
-    #define SC_szpassop_decrsat 	0x800 		/* val 4, shift 9 */
-    #define SC_szpassop_invert 		0xa00 		/* val 5, shift 9 */
-    #define SC_szpassop_incr 		0xc00 		/* val 6, shift 9 */
-    #define SC_szpassop_decr 		0xe00 		/* val 7, shift 9 */
-
-    #define TD1_color1arg2selMASK 	0xfffffffc 	/* bits 0-1 */
-    #define TD1_color1alphaselMASK 	0xffffffe3 	/* bits 2-4 */
-    #define TD1_color1alphaselSHIFT 	2 		
-    #define TD1_color1arg1alphaMASK 	0xffffffdf 	/* bit 5 */
-    #define TD1_color1arg1alphadisable 	0x0 		
-    #define TD1_color1arg1alphaenable 	0x20 		
-    #define TD1_color1arg1invMASK 	0xffffffbf 	/* bit 6 */
-    #define TD1_color1arg1invdisable 	0x0 		
-    #define TD1_color1arg1invenable 	0x40 		
-    #define TD1_color1arg2alphaMASK 	0xffffff7f 	/* bit 7 */
-    #define TD1_color1arg2alphadisable 	0x0 		
-    #define TD1_color1arg2alphaenable 	0x80 		
-    #define TD1_color1arg2invMASK 	0xfffffeff 	/* bit 8 */
-    #define TD1_color1arg2invdisable 	0x0 		
-    #define TD1_color1arg2invenable 	0x100 		
-    #define TD1_color1alpha1invMASK 	0xfffffdff 	/* bit 9 */
-    #define TD1_color1alpha1invdisable 	0x0 		
-    #define TD1_color1alpha1invenable 	0x200 		
-    #define TD1_color1alpha2invMASK 	0xfffffbff 	/* bit 10 */
-    #define TD1_color1alpha2invdisable 	0x0 		
-    #define TD1_color1alpha2invenable 	0x400 		
-    #define TD1_color1selMASK 		0xff9fffff 	/* bits 21-22 */
-    #define TD1_color1selarg1 		0x0 		/* val 0, shift 21 */
-    #define TD1_color1selarg2 		0x200000 	/* val 1, shift 21 */
-    #define TD1_color1seladd 		0x400000 	/* val 2, shift 21 */
-    #define TD1_color1selmul 		0x600000 	/* val 3, shift 21 */
-    #define TD1_alpha1selMASK 		0x3fffffff 	/* bits 30-31 */
-    #define TD1_alpha1selarg1 		0x0 		/* val 0, shift 30 */
-    #define TD1_alpha1selarg2 		0x40000000 	/* val 1, shift 30 */
-    #define TD1_alpha1seladd 		0x80000000 	/* val 2, shift 30 */
-    #define TD1_alpha1selmul 		0xc0000000 	/* val 3, shift 30 */
-
-    #define TST_ramtsten_MASK 		0xfffffffe 	/* bit 0 */
-    #define TST_ramtsten_disable 	0x0 		
-    #define TST_ramtsten_enable 	0x1 		
-    #define TST_ramtstdone_MASK 	0xfffffffd 	/* bit 1 */
-    #define TST_ramtstdone_disable 	0x0 		
-    #define TST_ramtstdone_enable 	0x2 		
-    #define TST_wramtstpass_MASK 	0xfffffffb 	/* bit 2 */
-    #define TST_wramtstpass_disable 	0x0 		
-    #define TST_wramtstpass_enable 	0x4 		
-    #define TST_tcachetstpass_MASK 	0xfffffff7 	/* bit 3 */
-    #define TST_tcachetstpass_disable 	0x0 		
-    #define TST_tcachetstpass_enable 	0x8 		
-    #define TST_tluttstpass_MASK 	0xffffffef 	/* bit 4 */
-    #define TST_tluttstpass_disable 	0x0 		
-    #define TST_tluttstpass_enable 	0x10 		
-    #define TST_luttstpass_MASK 	0xffffffdf 	/* bit 5 */
-    #define TST_luttstpass_disable 	0x0 		
-    #define TST_luttstpass_enable 	0x20 		
-    #define TST_besramtstpass_MASK 	0xffffffbf 	/* bit 6 */
-    #define TST_besramtstpass_disable 	0x0 		
-    #define TST_besramtstpass_enable 	0x40 		
-    #define TST_ringen_MASK 		0xfffffeff 	/* bit 8 */
-    #define TST_ringen_disable 		0x0 		
-    #define TST_ringen_enable 		0x100 		
-    #define TST_apllbyp_MASK 		0xfffffdff 	/* bit 9 */
-    #define TST_apllbyp_disable 	0x0 		
-    #define TST_apllbyp_enable 		0x200 		
-    #define TST_hiten_MASK 		0xfffffbff 	/* bit 10 */
-    #define TST_hiten_disable 		0x0 		
-    #define TST_hiten_enable 		0x400 		
-    #define TST_tmode_MASK 		0xffffc7ff 	/* bits 11-13 */
-    #define TST_tmode_SHIFT 		11 		
-    #define TST_tclksel_MASK 		0xfffe3fff 	/* bits 14-16 */
-    #define TST_tclksel_SHIFT 		14 		
-    #define TST_ringcnten_MASK 		0xfffdffff 	/* bit 17 */
-    #define TST_ringcnten_disable 	0x0 		
-    #define TST_ringcnten_enable 	0x20000 	
-    #define TST_ringcnt_MASK 		0xc003ffff 	/* bits 18-29 */
-    #define TST_ringcnt_SHIFT 		18 		
-    #define TST_ringcntclksl_MASK 	0xbfffffff 	/* bit 30 */
-    #define TST_ringcntclksl_disable 	0x0 		
-    #define TST_ringcntclksl_enable 	0x40000000 	
-    #define TST_biosboot_MASK 		0x7fffffff 	/* bit 31 */
-    #define TST_biosboot_disable 	0x0 		
-    #define TST_biosboot_enable 	0x80000000 	
-
-    #define TMC_tformat_MASK 		0xfffffff0 	/* bits 0-3 */
-    #define TMC_tformat_tw4 		0x0 		/* val 0, shift 0 */
-    #define TMC_tformat_tw8 		0x1 		/* val 1, shift 0 */
-    #define TMC_tformat_tw15 		0x2 		/* val 2, shift 0 */
-    #define TMC_tformat_tw16 		0x3 		/* val 3, shift 0 */
-    #define TMC_tformat_tw12 		0x4 		/* val 4, shift 0 */
-    #define TMC_tformat_tw32 		0x6 		/* val 6, shift 0 */
-    #define TMC_tformat_tw422 		0xa 		/* val 10, shift 0 */
-    #define TMC_tpitchlin_MASK 		0xfffffeff 	/* bit 8 */
-    #define TMC_tpitchlin_disable 	0x0 		
-    #define TMC_tpitchlin_enable 	0x100 		
-    #define TMC_tpitchext_MASK 		0xfff001ff 	/* bits 9-19 */
-    #define TMC_tpitchext_SHIFT 	9 		
-    #define TMC_tpitch_MASK 		0xfff8ffff 	/* bits 16-18 */
-    #define TMC_tpitch_SHIFT 		16 		
-    #define TMC_owalpha_MASK 		0xffbfffff 	/* bit 22 */
-    #define TMC_owalpha_disable 	0x0 		
-    #define TMC_owalpha_enable 		0x400000 	
-    #define TMC_azeroextend_MASK 	0xff7fffff 	/* bit 23 */
-    #define TMC_azeroextend_disable 	0x0 		
-    #define TMC_azeroextend_enable 	0x800000 	
-    #define TMC_decalckey_MASK 		0xfeffffff 	/* bit 24 */
-    #define TMC_decalckey_disable 	0x0 		
-    #define TMC_decalckey_enable 	0x1000000 	
-    #define TMC_takey_MASK 		0xfdffffff 	/* bit 25 */
-    #define TMC_takey_0 		0x0 		
-    #define TMC_takey_1 		0x2000000 	
-    #define TMC_tamask_MASK 		0xfbffffff 	/* bit 26 */
-    #define TMC_tamask_0 		0x0 		
-    #define TMC_tamask_1 		0x4000000 	
-    #define TMC_clampv_MASK 		0xf7ffffff 	/* bit 27 */
-    #define TMC_clampv_disable 		0x0 		
-    #define TMC_clampv_enable 		0x8000000 	
-    #define TMC_clampu_MASK 		0xefffffff 	/* bit 28 */
-    #define TMC_clampu_disable 		0x0 		
-    #define TMC_clampu_enable 		0x10000000 	
-    #define TMC_tmodulate_MASK 		0xdfffffff 	/* bit 29 */
-    #define TMC_tmodulate_disable 	0x0 		
-    #define TMC_tmodulate_enable 	0x20000000 	
-    #define TMC_strans_MASK 		0xbfffffff 	/* bit 30 */
-    #define TMC_strans_disable 		0x0 		
-    #define TMC_strans_enable 		0x40000000 	
-    #define TMC_itrans_MASK 		0x7fffffff 	/* bit 31 */
-    #define TMC_itrans_disable 		0x0 		
-    #define TMC_itrans_enable 		0x80000000 	
-
-    #define TMC_decalblend_MASK 	0xfffffffe 	/* bit 0 */
-    #define TMC_decalblend_disable 	0x0 		
-    #define TMC_decalblend_enable 	0x1 		
-    #define TMC_idecal_MASK 		0xfffffffd 	/* bit 1 */
-    #define TMC_idecal_disable 		0x0 		
-    #define TMC_idecal_enable 		0x2 		
-    #define TMC_decaldis_MASK 		0xfffffffb 	/* bit 2 */
-    #define TMC_decaldis_disable 	0x0 		
-    #define TMC_decaldis_enable 	0x4 		
-    #define TMC_ckstransdis_MASK 	0xffffffef 	/* bit 4 */
-    #define TMC_ckstransdis_disable 	0x0 		
-    #define TMC_ckstransdis_enable 	0x10 		
-    #define TMC_borderen_MASK 		0xffffffdf 	/* bit 5 */
-    #define TMC_borderen_disable 	0x0 		
-    #define TMC_borderen_enable 	0x20 		
-    #define TMC_specen_MASK 		0xffffffbf 	/* bit 6 */
-    #define TMC_specen_disable 		0x0 		
-    #define TMC_specen_enable 		0x40 		
-
-    #define TF_minfilter_MASK 		0xfffffff0 	/* bits 0-3 */
-    #define TF_minfilter_nrst 		0x0 		/* val 0, shift 0 */
-    #define TF_minfilter_bilin 		0x2 		/* val 2, shift 0 */
-    #define TF_minfilter_cnst 		0x3 		/* val 3, shift 0 */
-    #define TF_minfilter_mm1s 		0x8 		/* val 8, shift 0 */
-    #define TF_minfilter_mm2s 		0x9 		/* val 9, shift 0 */
-    #define TF_minfilter_mm4s 		0xa 		/* val 10, shift 0 */
-    #define TF_minfilter_mm8s 		0xc 		/* val 12, shift 0 */
-    #define TF_magfilter_MASK 		0xffffff0f 	/* bits 4-7 */
-    #define TF_magfilter_nrst 		0x0 		/* val 0, shift 4 */
-    #define TF_magfilter_bilin 		0x20 		/* val 2, shift 4 */
-    #define TF_magfilter_cnst 		0x30 		/* val 3, shift 4 */
-    #define TF_avgstride_MASK 		0xfff7ffff 	/* bit 19 */
-    #define TF_avgstride_disable 	0x0 		
-    #define TF_avgstride_enable 	0x80000 	
-    #define TF_filteralpha_MASK 	0xffefffff 	/* bit 20 */
-    #define TF_filteralpha_disable 	0x0 		
-    #define TF_filteralpha_enable 	0x100000 	
-    #define TF_fthres_MASK 		0xe01fffff 	/* bits 21-28 */
-    #define TF_fthres_SHIFT 		21 		
-    #define TF_mapnb_MASK 		0x1fffffff 	/* bits 29-31 */
-    #define TF_mapnb_SHIFT 		29 		
-
-    #define TH_th_MASK 			0xffffffc0 	/* bits 0-5 */
-    #define TH_th_SHIFT 		0 		
-    #define TH_rfh_MASK 		0xffff81ff 	/* bits 9-14 */
-    #define TH_rfh_SHIFT 		9 		
-    #define TH_thmask_MASK 		0xe003ffff 	/* bits 18-28 */
-    #define TH_thmask_SHIFT 		18 		
-
-    #define TO_texorgmap_MASK 		0xfffffffe 	/* bit 0 */
-    #define TO_texorgmap_fb 		0x0 		
-    #define TO_texorgmap_sys 		0x1 		
-    #define TO_texorgacc_MASK 		0xfffffffd 	/* bit 1 */
-    #define TO_texorgacc_pci 		0x0 		
-    #define TO_texorgacc_agp 		0x2 		
-    #define TO_texorg_MASK 		0x1f 		/* bits 5-31 */
-    #define TO_texorg_SHIFT 		5 		
-
-    #define TT_tckey_MASK 		0xffff0000 	/* bits 0-15 */
-    #define TT_tckey_SHIFT 		0 		
-    #define TT_tkmask_MASK 		0xffff 		/* bits 16-31 */
-    #define TT_tkmask_SHIFT 		16 		
-
-    #define TT_tckeyh_MASK 		0xffff0000 	/* bits 0-15 */
-    #define TT_tckeyh_SHIFT 		0 		
-    #define TT_tkmaskh_MASK 		0xffff 		/* bits 16-31 */
-    #define TT_tkmaskh_SHIFT 		16 		
-
-    #define TW_tw_MASK 			0xffffffc0 	/* bits 0-5 */
-    #define TW_tw_SHIFT 		0 		
-    #define TW_rfw_MASK 		0xffff81ff 	/* bits 9-14 */
-    #define TW_rfw_SHIFT 		9 		
-    #define TW_twmask_MASK 		0xe003ffff 	/* bits 18-28 */
-    #define TW_twmask_SHIFT 		18 		
-
-    #define WAS_seqdst0_MASK 		0xffffffc0 	/* bits 0-5 */
-    #define WAS_seqdst0_SHIFT 		0 		
-    #define WAS_seqdst1_MASK 		0xfffff03f 	/* bits 6-11 */
-    #define WAS_seqdst1_SHIFT 		6 		
-    #define WAS_seqdst2_MASK 		0xfffc0fff 	/* bits 12-17 */
-    #define WAS_seqdst2_SHIFT 		12 		
-    #define WAS_seqdst3_MASK 		0xff03ffff 	/* bits 18-23 */
-    #define WAS_seqdst3_SHIFT 		18 		
-    #define WAS_seqlen_MASK 		0xfcffffff 	/* bits 24-25 */
-    #define WAS_wfirsttag_MASK 		0xfbffffff 	/* bit 26 */
-    #define WAS_wfirsttag_disable 	0x0 		
-    #define WAS_wfirsttag_enable 	0x4000000 	
-    #define WAS_wsametag_MASK 		0xf7ffffff 	/* bit 27 */
-    #define WAS_wsametag_disable 	0x0 		
-    #define WAS_wsametag_enable 	0x8000000 	
-    #define WAS_seqoff_MASK 		0xefffffff 	/* bit 28 */
-    #define WAS_seqoff_disable 		0x0 		
-    #define WAS_seqoff_enable 		0x10000000 	
-
-    #define WMA_wcodeaddr_MASK 		0xff 		/* bits 8-31 */
-    #define WMA_wcodeaddr_SHIFT 	8 		
-
-    #define WF_walustsflag_MASK 	0xffffff00 	/* bits 0-7 */
-    #define WF_walustsflag_SHIFT 	0 		
-    #define WF_walucfgflag_MASK 	0xffff00ff 	/* bits 8-15 */
-    #define WF_walucfgflag_SHIFT 	8 		
-    #define WF_wprgflag_MASK 		0xffff 		/* bits 16-31 */
-    #define WF_wprgflag_SHIFT 		16 		
-
-    #define WF1_walustsflag1_MASK 	0xffffff00 	/* bits 0-7 */
-    #define WF1_walustsflag1_SHIFT 	0 		
-    #define WF1_walucfgflag1_MASK 	0xffff00ff 	/* bits 8-15 */
-    #define WF1_walucfgflag1_SHIFT 	8 		
-    #define WF1_wprgflag1_MASK 		0xffff 		/* bits 16-31 */
-    #define WF1_wprgflag1_SHIFT 	16 		
-
-    #define WGV_wgetmsbmin_MASK 	0xffffffe0 	/* bits 0-4 */
-    #define WGV_wgetmsbmin_SHIFT 	0 		
-    #define WGV_wgetmsbmax_MASK 	0xffffe0ff 	/* bits 8-12 */
-    #define WGV_wgetmsbmax_SHIFT 	8 		
-    #define WGV_wbrklefttop_MASK 	0xfffeffff 	/* bit 16 */
-    #define WGV_wbrklefttop_disable 	0x0 		
-    #define WGV_wbrklefttop_enable 	0x10000 	
-    #define WGV_wfastcrop_MASK 		0xfffdffff 	/* bit 17 */
-    #define WGV_wfastcrop_disable 	0x0 		
-    #define WGV_wfastcrop_enable 	0x20000 	
-    #define WGV_wcentersnap_MASK 	0xfffbffff 	/* bit 18 */
-    #define WGV_wcentersnap_disable 	0x0 		
-    #define WGV_wcentersnap_enable 	0x40000 	
-    #define WGV_wbrkrighttop_MASK 	0xfff7ffff 	/* bit 19 */
-    #define WGV_wbrkrighttop_disable 	0x0 		
-    #define WGV_wbrkrighttop_enable 	0x80000 	
-
-    #define WIA_wmode_MASK 		0xfffffffc 	/* bits 0-1 */
-    #define WIA_wmode_suspend 		0x0 		/* val 0, shift 0 */
-    #define WIA_wmode_resume 		0x1 		/* val 1, shift 0 */
-    #define WIA_wmode_jump 		0x2 		/* val 2, shift 0 */
-    #define WIA_wmode_start 		0x3 		/* val 3, shift 0 */
-    #define WIA_wagp_MASK 		0xfffffffb 	/* bit 2 */
-    #define WIA_wagp_pci 		0x0 		
-    #define WIA_wagp_agp 		0x4 		
-    #define WIA_wiaddr_MASK 		0x7 		/* bits 3-31 */
-    #define WIA_wiaddr_SHIFT 		3 		
-
-    #define WIA2_wmode_MASK 		0xfffffffc 	/* bits 0-1 */
-    #define WIA2_wmode_suspend 		0x0 		/* val 0, shift 0 */
-    #define WIA2_wmode_resume 		0x1 		/* val 1, shift 0 */
-    #define WIA2_wmode_jump 		0x2 		/* val 2, shift 0 */
-    #define WIA2_wmode_start 		0x3 		/* val 3, shift 0 */
-    #define WIA2_wagp_MASK 		0xfffffffb 	/* bit 2 */
-    #define WIA2_wagp_pci 		0x0 		
-    #define WIA2_wagp_agp 		0x4 		
-    #define WIA2_wiaddr_MASK 		0x7 		/* bits 3-31 */
-    #define WIA2_wiaddr_SHIFT 		3 		
-
-    #define WIMA_wimemaddr_MASK 	0xffffff00 	/* bits 0-7 */
-    #define WIMA_wimemaddr_SHIFT 	0 		
-
-    #define WM_wucodecache_MASK 	0xfffffffe 	/* bit 0 */
-    #define WM_wucodecache_disable 	0x0 		
-    #define WM_wucodecache_enable 	0x1 		
-    #define WM_wmaster_MASK 		0xfffffffd 	/* bit 1 */
-    #define WM_wmaster_disable 		0x0 		
-    #define WM_wmaster_enable 		0x2 		
-    #define WM_wcacheflush_MASK 	0xfffffff7 	/* bit 3 */
-    #define WM_wcacheflush_disable 	0x0 		
-    #define WM_wcacheflush_enable 	0x8 		
-
-    #define WVS_wvrtxsz_MASK 		0xffffffc0 	/* bits 0-5 */
-    #define WVS_wvrtxsz_SHIFT 		0 		
-    #define WVS_primsz_MASK 		0xffffc0ff 	/* bits 8-13 */
-    #define WVS_primsz_SHIFT 		8 		
-
-    #define XYEA_x_end_MASK 		0xffff0000 	/* bits 0-15 */
-    #define XYEA_x_end_SHIFT 		0 		
-    #define XYEA_y_end_MASK 		0xffff 		/* bits 16-31 */
-    #define XYEA_y_end_SHIFT 		16 		
-
-    #define XYSA_x_start_MASK 		0xffff0000 	/* bits 0-15 */
-    #define XYSA_x_start_SHIFT 		0 		
-    #define XYSA_y_start_MASK 		0xffff 		/* bits 16-31 */
-    #define XYSA_y_start_SHIFT 		16 		
-
-    #define YA_ydst_MASK 		0xff800000 	/* bits 0-22 */
-    #define YA_ydst_SHIFT 		0 		
-    #define YA_sellin_MASK 		0x1fffffff 	/* bits 29-31 */
-    #define YA_sellin_SHIFT 		29 		
-
-    #define YDL_length_MASK 		0xffff0000 	/* bits 0-15 */
-    #define YDL_length_SHIFT 		0 		
-    #define YDL_yval_MASK 		0xffff 		/* bits 16-31 */
-    #define YDL_yval_SHIFT 		16 		
-
-    #define ZO_zorgmap_MASK 		0xfffffffe 	/* bit 0 */
-    #define ZO_zorgmap_fb 		0x0 		
-    #define ZO_zorgmap_sys 		0x1 		
-    #define ZO_zorgacc_MASK 		0xfffffffd 	/* bit 1 */
-    #define ZO_zorgacc_pci 		0x0 		
-    #define ZO_zorgacc_agp 		0x2 		
-    #define ZO_zorg_MASK 		0x3 		/* bits 2-31 */
-    #define ZO_zorg_SHIFT 		2 		
-
-
-
-
-/**************** (END) AUTOMATICLY GENERATED REGISTER FILE ******************/
-
-#endif 	/* _MGAREGS_H_ */
-
diff --git a/linux/picker.c b/linux/picker.c
index ecdb2c15..0bd8bfd5 100644
--- a/linux/picker.c
+++ b/linux/picker.c
@@ -9,6 +9,16 @@
 #define CONFIG_MODVERSIONS 0
 #endif
 
+#ifndef CONFIG_AGP_MODULE
+#define CONFIG_AGP_MODULE 0
+#endif
+
+#ifndef CONFIG_AGP
+#define CONFIG_AGP 0
+#endif
+
 SMP = CONFIG_SMP
 MODVERSIONS = CONFIG_MODVERSIONS
+AGP = CONFIG_AGP
+AGP_MODULE = CONFIG_AGP_MODULE
 RELEASE = UTS_RELEASE
diff --git a/linux/proc.c b/linux/proc.c
index 54aba58c..db98fd6a 100644
--- a/linux/proc.c
+++ b/linux/proc.c
@@ -164,7 +164,10 @@ static int _drm_vm_info(char *buf, char **start, off_t offset, int len,
 {
 	drm_device_t *dev = (drm_device_t *)data;
 	drm_map_t    *map;
-	const char   *types[] = { "FB", "REG", "SHM" };
+				/* Hardcoded from _DRM_FRAME_BUFFER,
+                                   _DRM_REGISTERS, _DRM_SHM, and
+                                   _DRM_AGP. */
+	const char   *types[] = { "FB", "REG", "SHM", "AGP" };
 	const char   *type;
 	int	     i;
 
@@ -175,7 +178,7 @@ static int _drm_vm_info(char *buf, char **start, off_t offset, int len,
 		       "address mtrr\n\n");
 	for (i = 0; i < dev->map_count; i++) {
 		map = dev->maplist[i];
-		if (map->type < 0 || map->type > 2) type = "??";
+		if (map->type < 0 || map->type > 3) type = "??";
 		else				    type = types[map->type];
 		DRM_PROC_PRINT("%4d 0x%08lx 0x%08lx %4.4s  0x%02x 0x%08lx ",
 			       i,
diff --git a/linux/tdfx_context.c b/linux/tdfx_context.c
index 842d6f5d..74b107bd 100644
--- a/linux/tdfx_context.c
+++ b/linux/tdfx_context.c
@@ -38,9 +38,7 @@ extern drm_ctx_t tdfx_res_ctx;
 
 static int tdfx_alloc_queue(drm_device_t *dev)
 {
-	static int context = 0;
-
-	return ++context;	/* Should this reuse contexts in the future? */
+	return drm_ctxbitmap_next(dev);
 }
 
 int tdfx_context_switch(drm_device_t *dev, int old, int new)
@@ -137,6 +135,12 @@ int tdfx_addctx(struct inode *inode, struct file *filp, unsigned int cmd,
 		ctx.handle = tdfx_alloc_queue(dev);
 	}
 	DRM_DEBUG("%d\n", ctx.handle);
+	if (ctx.handle == -1) {
+		DRM_DEBUG("Not enough free contexts.\n");
+				/* Should this return -EBUSY instead? */
+		return -ENOMEM;
+	}
+   
 	copy_to_user_ret((drm_ctx_t *)arg, &ctx, sizeof(ctx), -EFAULT);
 	return 0;
 }
@@ -193,13 +197,13 @@ int tdfx_newctx(struct inode *inode, struct file *filp, unsigned int cmd,
 int tdfx_rmctx(struct inode *inode, struct file *filp, unsigned int cmd,
 	       unsigned long arg)
 {
+	drm_file_t      *priv   = filp->private_data;
+	drm_device_t    *dev    = priv->dev;
 	drm_ctx_t	ctx;
 
 	copy_from_user_ret(&ctx, (drm_ctx_t *)arg, sizeof(ctx), -EFAULT);
 	DRM_DEBUG("%d\n", ctx.handle);
-				/* This is currently a noop because we
-				   don't reuse context values.  Perhaps we
-				   should? */
-	
+	drm_ctxbitmap_free(dev, ctx.handle);
+
 	return 0;
 }
diff --git a/linux/tdfx_drv.c b/linux/tdfx_drv.c
index 57c1c719..fb7a997b 100644
--- a/linux/tdfx_drv.c
+++ b/linux/tdfx_drv.c
@@ -85,6 +85,16 @@ static drm_ioctl_desc_t	      tdfx_ioctls[] = {
 	[DRM_IOCTL_NR(DRM_IOCTL_LOCK)]	     = { tdfx_lock,	  1, 0 },
 	[DRM_IOCTL_NR(DRM_IOCTL_UNLOCK)]     = { tdfx_unlock,	  1, 0 },
 	[DRM_IOCTL_NR(DRM_IOCTL_FINISH)]     = { drm_finish,	  1, 0 },
+#ifdef DRM_AGP
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_ACQUIRE)]   = {drm_agp_acquire, 1, 1},
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_RELEASE)]   = {drm_agp_release, 1, 1},
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_ENABLE)]    = {drm_agp_enable,  1, 1},
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_INFO)]      = {drm_agp_info,    1, 1},
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_ALLOC)]     = {drm_agp_alloc,   1, 1},
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_FREE)]      = {drm_agp_free,    1, 1},
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_BIND)]      = {drm_agp_unbind,  1, 1},
+	[DRM_IOCTL_NR(DRM_IOCTL_AGP_UNBIND)]    = {drm_agp_bind,    1, 1},
+#endif
 };
 #define TDFX_IOCTL_COUNT DRM_ARRAY_SIZE(tdfx_ioctls)
 
@@ -228,7 +238,24 @@ static int tdfx_takedown(drm_device_t *dev)
 		}
 		dev->magiclist[i].head = dev->magiclist[i].tail = NULL;
 	}
-	
+#ifdef DRM_AGP
+				/* Clear AGP information */
+	if (dev->agp) {
+		drm_agp_mem_t *temp;
+		drm_agp_mem_t *temp_next;
+	   
+		temp = dev->agp->memory;
+		while(temp != NULL) {
+			temp_next = temp->next;
+			drm_free_agp(temp->memory, temp->pages);
+			drm_free(temp, sizeof(*temp), DRM_MEM_AGPLISTS);
+			temp = temp_next;
+		}
+		if(dev->agp->acquired) (*drm_agp.release)();
+		drm_free(dev->agp, sizeof(*dev->agp), DRM_MEM_AGPLISTS);
+		dev->agp = NULL;
+	}
+#endif
 				/* Clear vma list (only built for debugging) */
 	if (dev->vmalist) {
 		for (vma = dev->vmalist; vma; vma = vma_next) {
@@ -262,6 +289,10 @@ static int tdfx_takedown(drm_device_t *dev)
 					       - PAGE_SHIFT,
 					       DRM_MEM_SAREA);
 				break;
+			case _DRM_AGP:
+				/* Do nothing here, because this is all
+                                   handled in the AGP/GART driver. */
+				break;
 			}
 			drm_free(map, sizeof(*map), DRM_MEM_MAPS);
 		}
@@ -309,6 +340,16 @@ int tdfx_init(void)
 
 	drm_mem_init();
 	drm_proc_init(dev);
+#ifdef DRM_AGP
+	dev->agp    = drm_agp_init();
+#endif
+	if((retcode = drm_ctxbitmap_init(dev))) {
+		DRM_ERROR("Cannot allocate memory for context bitmap.\n");
+		drm_proc_cleanup();
+		misc_deregister(&tdfx_misc);
+		tdfx_takedown(dev);
+		return retcode;
+	}
 
 	DRM_INFO("Initialized %s %d.%d.%d %s on minor %d\n",
 		 TDFX_NAME,
@@ -335,6 +376,7 @@ void tdfx_cleanup(void)
 	} else {
 		DRM_INFO("Module unloaded\n");
 	}
+	drm_ctxbitmap_cleanup(dev);
 	tdfx_takedown(dev);
 }
 
diff --git a/linux/vm.c b/linux/vm.c
index 85470ac5..389f2fae 100644
--- a/linux/vm.c
+++ b/linux/vm.c
@@ -250,9 +250,10 @@ int drm_mmap(struct file *filp, struct vm_area_struct *vma)
 	switch (map->type) {
 	case _DRM_FRAME_BUFFER:
 	case _DRM_REGISTERS:
+	case _DRM_AGP:
 		if (VM_OFFSET(vma) >= __pa(high_memory)) {
 #if defined(__i386__)
-			if (boot_cpu_data.x86 > 3) {
+			if (boot_cpu_data.x86 > 3 && map->type != _DRM_AGP) {
 				pgprot_val(vma->vm_page_prot) |= _PAGE_PCD;
 				pgprot_val(vma->vm_page_prot) &= ~_PAGE_PWT;
 			}
diff --git a/shared-core/drm.h b/shared-core/drm.h
index 7b8e8826..ae4c65ca 100644
--- a/shared-core/drm.h
+++ b/shared-core/drm.h
@@ -61,6 +61,19 @@ typedef unsigned int  drm_context_t;
 typedef unsigned int  drm_drawable_t;
 typedef unsigned int  drm_magic_t;
 
+/* Warning: If you change this structure, make sure you change 
+ * XF86DRIClipRectRec in the server as well */
+
+typedef struct drm_clip_rect {
+           unsigned short x1;
+           unsigned short y1;
+           unsigned short x2;
+           unsigned short y2;
+} drm_clip_rect_t;
+
+/* Seperate include files for the i810/mga specific structures */
+#include "mga_drm.h"
+#include "i810_drm.h"
 
 typedef struct drm_version {
 	int    version_major;	  /* Major version			    */
@@ -101,7 +114,8 @@ typedef struct drm_control {
 typedef enum drm_map_type {
 	_DRM_FRAME_BUFFER = 0,	  /* WC (no caching), no core dump	    */
 	_DRM_REGISTERS	  = 1,	  /* no caching, no core dump		    */
-	_DRM_SHM	  = 2	  /* shared, cached			    */
+	_DRM_SHM	  = 2,	  /* shared, cached			    */
+	_DRM_AGP          = 3	  /* AGP/GART                               */
 } drm_map_type_t;
 
 typedef enum drm_map_flags {
@@ -165,8 +179,11 @@ typedef struct drm_buf_desc {
 	int	      low_mark;	 /* Low water mark			     */
 	int	      high_mark; /* High water mark			     */
 	enum {
-		DRM_PAGE_ALIGN = 0x01  /* Align on page boundaries for DMA   */
+		_DRM_PAGE_ALIGN = 0x01, /* Align on page boundaries for DMA  */
+		_DRM_AGP_BUFFER = 0x02  /* Buffer is in agp space            */
 	}	      flags;
+	unsigned long agp_start; /* Start address of where the agp buffers
+				  * are in the agp aperture */
 } drm_buf_desc_t;
 
 typedef struct drm_buf_info {
@@ -237,6 +254,38 @@ typedef struct drm_irq_busid {
 	int funcnum;
 } drm_irq_busid_t;
 
+typedef struct drm_agp_mode {
+	unsigned long mode;
+} drm_agp_mode_t;
+
+				/* For drm_agp_alloc -- allocated a buffer */
+typedef struct drm_agp_buffer {
+	unsigned long size;	/* In bytes -- will round to page boundary */
+	unsigned long handle;	/* Used for BIND/UNBIND ioctls */
+	unsigned long type;     /* Type of memory to allocate  */
+        unsigned long physical; /* Physical used by i810       */
+} drm_agp_buffer_t;
+
+				/* For drm_agp_bind */
+typedef struct drm_agp_binding {
+	unsigned long handle;   /* From drm_agp_buffer */
+	unsigned long offset;	/* In bytes -- will round to page boundary */
+} drm_agp_binding_t;
+
+typedef struct drm_agp_info {
+	int            agp_version_major;
+	int            agp_version_minor;
+	unsigned long  mode;
+	unsigned long  aperture_base;  /* physical address */
+	unsigned long  aperture_size;  /* bytes */
+	unsigned long  memory_allowed; /* bytes */
+	unsigned long  memory_used;
+
+				/* PCI information */
+	unsigned short id_vendor;
+	unsigned short id_device;
+} drm_agp_info_t;
+
 #define DRM_IOCTL_BASE	     'd'
 #define DRM_IOCTL_NR(n)	     _IOC_NR(n)
 #define DRM_IO(nr)	     _IO(DRM_IOCTL_BASE,nr)
@@ -276,4 +325,28 @@ typedef struct drm_irq_busid {
 #define DRM_IOCTL_UNLOCK     DRM_IOW( 0x2b, drm_lock_t)
 #define DRM_IOCTL_FINISH     DRM_IOW( 0x2c, drm_lock_t)
 
+#define DRM_IOCTL_AGP_ACQUIRE DRM_IO(  0x30)
+#define DRM_IOCTL_AGP_RELEASE DRM_IO(  0x31)
+#define DRM_IOCTL_AGP_ENABLE  DRM_IOR( 0x32, drm_agp_mode_t)
+#define DRM_IOCTL_AGP_INFO    DRM_IOW( 0x33, drm_agp_info_t)
+#define DRM_IOCTL_AGP_ALLOC   DRM_IOWR(0x34, drm_agp_buffer_t)
+#define DRM_IOCTL_AGP_FREE    DRM_IOW( 0x35, drm_agp_buffer_t)
+#define DRM_IOCTL_AGP_BIND    DRM_IOWR(0x36, drm_agp_binding_t)
+#define DRM_IOCTL_AGP_UNBIND  DRM_IOW( 0x37, drm_agp_binding_t)
+
+/* Mga specific ioctls */
+#define DRM_IOCTL_MGA_INIT    DRM_IOW( 0x40, drm_mga_init_t)
+#define DRM_IOCTL_MGA_SWAP    DRM_IOW( 0x41, drm_mga_swap_t)
+#define DRM_IOCTL_MGA_CLEAR   DRM_IOW( 0x42, drm_mga_clear_t)
+#define DRM_IOCTL_MGA_ILOAD   DRM_IOW( 0x43, drm_mga_iload_t)
+#define DRM_IOCTL_MGA_VERTEX  DRM_IOW( 0x44, drm_mga_vertex_t)
+#define DRM_IOCTL_MGA_FLUSH   DRM_IOW( 0x45, drm_lock_t )
+
+/* I810 specific ioctls */
+#define DRM_IOCTL_I810_INIT    DRM_IOW( 0x40, drm_i810_init_t)
+#define DRM_IOCTL_I810_VERTEX  DRM_IOW( 0x41, drm_i810_vertex_t)
+#define DRM_IOCTL_I810_DMA     DRM_IOW( 0x42, drm_i810_general_t)
+#define DRM_IOCTL_I810_FLUSH   DRM_IO ( 0x43)
+#define DRM_IOCTL_I810_GETAGE  DRM_IO ( 0x44)
+
 #endif
diff --git a/shared/drm.h b/shared/drm.h
index 7b8e8826..ae4c65ca 100644
--- a/shared/drm.h
+++ b/shared/drm.h
@@ -61,6 +61,19 @@ typedef unsigned int  drm_context_t;
 typedef unsigned int  drm_drawable_t;
 typedef unsigned int  drm_magic_t;
 
+/* Warning: If you change this structure, make sure you change 
+ * XF86DRIClipRectRec in the server as well */
+
+typedef struct drm_clip_rect {
+           unsigned short x1;
+           unsigned short y1;
+           unsigned short x2;
+           unsigned short y2;
+} drm_clip_rect_t;
+
+/* Seperate include files for the i810/mga specific structures */
+#include "mga_drm.h"
+#include "i810_drm.h"
 
 typedef struct drm_version {
 	int    version_major;	  /* Major version			    */
@@ -101,7 +114,8 @@ typedef struct drm_control {
 typedef enum drm_map_type {
 	_DRM_FRAME_BUFFER = 0,	  /* WC (no caching), no core dump	    */
 	_DRM_REGISTERS	  = 1,	  /* no caching, no core dump		    */
-	_DRM_SHM	  = 2	  /* shared, cached			    */
+	_DRM_SHM	  = 2,	  /* shared, cached			    */
+	_DRM_AGP          = 3	  /* AGP/GART                               */
 } drm_map_type_t;
 
 typedef enum drm_map_flags {
@@ -165,8 +179,11 @@ typedef struct drm_buf_desc {
 	int	      low_mark;	 /* Low water mark			     */
 	int	      high_mark; /* High water mark			     */
 	enum {
-		DRM_PAGE_ALIGN = 0x01  /* Align on page boundaries for DMA   */
+		_DRM_PAGE_ALIGN = 0x01, /* Align on page boundaries for DMA  */
+		_DRM_AGP_BUFFER = 0x02  /* Buffer is in agp space            */
 	}	      flags;
+	unsigned long agp_start; /* Start address of where the agp buffers
+				  * are in the agp aperture */
 } drm_buf_desc_t;
 
 typedef struct drm_buf_info {
@@ -237,6 +254,38 @@ typedef struct drm_irq_busid {
 	int funcnum;
 } drm_irq_busid_t;
 
+typedef struct drm_agp_mode {
+	unsigned long mode;
+} drm_agp_mode_t;
+
+				/* For drm_agp_alloc -- allocated a buffer */
+typedef struct drm_agp_buffer {
+	unsigned long size;	/* In bytes -- will round to page boundary */
+	unsigned long handle;	/* Used for BIND/UNBIND ioctls */
+	unsigned long type;     /* Type of memory to allocate  */
+        unsigned long physical; /* Physical used by i810       */
+} drm_agp_buffer_t;
+
+				/* For drm_agp_bind */
+typedef struct drm_agp_binding {
+	unsigned long handle;   /* From drm_agp_buffer */
+	unsigned long offset;	/* In bytes -- will round to page boundary */
+} drm_agp_binding_t;
+
+typedef struct drm_agp_info {
+	int            agp_version_major;
+	int            agp_version_minor;
+	unsigned long  mode;
+	unsigned long  aperture_base;  /* physical address */
+	unsigned long  aperture_size;  /* bytes */
+	unsigned long  memory_allowed; /* bytes */
+	unsigned long  memory_used;
+
+				/* PCI information */
+	unsigned short id_vendor;
+	unsigned short id_device;
+} drm_agp_info_t;
+
 #define DRM_IOCTL_BASE	     'd'
 #define DRM_IOCTL_NR(n)	     _IOC_NR(n)
 #define DRM_IO(nr)	     _IO(DRM_IOCTL_BASE,nr)
@@ -276,4 +325,28 @@ typedef struct drm_irq_busid {
 #define DRM_IOCTL_UNLOCK     DRM_IOW( 0x2b, drm_lock_t)
 #define DRM_IOCTL_FINISH     DRM_IOW( 0x2c, drm_lock_t)
 
+#define DRM_IOCTL_AGP_ACQUIRE DRM_IO(  0x30)
+#define DRM_IOCTL_AGP_RELEASE DRM_IO(  0x31)
+#define DRM_IOCTL_AGP_ENABLE  DRM_IOR( 0x32, drm_agp_mode_t)
+#define DRM_IOCTL_AGP_INFO    DRM_IOW( 0x33, drm_agp_info_t)
+#define DRM_IOCTL_AGP_ALLOC   DRM_IOWR(0x34, drm_agp_buffer_t)
+#define DRM_IOCTL_AGP_FREE    DRM_IOW( 0x35, drm_agp_buffer_t)
+#define DRM_IOCTL_AGP_BIND    DRM_IOWR(0x36, drm_agp_binding_t)
+#define DRM_IOCTL_AGP_UNBIND  DRM_IOW( 0x37, drm_agp_binding_t)
+
+/* Mga specific ioctls */
+#define DRM_IOCTL_MGA_INIT    DRM_IOW( 0x40, drm_mga_init_t)
+#define DRM_IOCTL_MGA_SWAP    DRM_IOW( 0x41, drm_mga_swap_t)
+#define DRM_IOCTL_MGA_CLEAR   DRM_IOW( 0x42, drm_mga_clear_t)
+#define DRM_IOCTL_MGA_ILOAD   DRM_IOW( 0x43, drm_mga_iload_t)
+#define DRM_IOCTL_MGA_VERTEX  DRM_IOW( 0x44, drm_mga_vertex_t)
+#define DRM_IOCTL_MGA_FLUSH   DRM_IOW( 0x45, drm_lock_t )
+
+/* I810 specific ioctls */
+#define DRM_IOCTL_I810_INIT    DRM_IOW( 0x40, drm_i810_init_t)
+#define DRM_IOCTL_I810_VERTEX  DRM_IOW( 0x41, drm_i810_vertex_t)
+#define DRM_IOCTL_I810_DMA     DRM_IOW( 0x42, drm_i810_general_t)
+#define DRM_IOCTL_I810_FLUSH   DRM_IO ( 0x43)
+#define DRM_IOCTL_I810_GETAGE  DRM_IO ( 0x44)
+
 #endif
diff --git a/tests/drmstat.c b/tests/drmstat.c
index 8fe6f707..a96a38b2 100644
--- a/tests/drmstat.c
+++ b/tests/drmstat.c
@@ -187,7 +187,7 @@ int main(int argc, char **argv)
 	case 'b':
 	    count   = strtoul(optarg, &pt, 0);
 	    size    = strtoul(pt+1, NULL, 0);
-	    if ((r = drmAddBufs(fd, count, size, 0)) < 0) {
+	    if ((r = drmAddBufs(fd, count, size, 0, 65536)) < 0) {
 		drmError(r, argv[0]);
 		return 1;
 	    }