Revert "Merge branch 'drm-gem'"

This reverts commit 53675e5c05c0598b7ea206d5c27dbcae786a2c03.

Conflicts:

	src/mesa/drivers/dri/i965/brw_wm_surface_state.c

76 files changed, 4275 insertions, 1736 deletions

diff --git a/src/mesa/drivers/dri/Makefile.template b/src/mesa/drivers/dri/Makefile.template
index 072d16ce8b..95eeecc7c4 100644
--- a/src/mesa/drivers/dri/Makefile.template
+++ b/src/mesa/drivers/dri/Makefile.template
@@ -10,6 +10,11 @@ COMMON_SOURCES = \
         ../common/xmlconfig.c \
         ../common/drirenderbuffer.c 
 
+COMMON_BM_SOURCES = \
+	../common/dri_bufmgr.c \
+	../common/dri_bufmgr_fake.c
+
+
 ifeq ($(WINDOW_SYSTEM),dri)
 WINOBJ=
 WINLIB=
diff --git a/src/mesa/drivers/dri/common/dri_bufmgr.c b/src/mesa/drivers/dri/common/dri_bufmgr.c
new file mode 100644
index 0000000000..8b5cb7628e
--- /dev/null
+++ b/src/mesa/drivers/dri/common/dri_bufmgr.c
@@ -0,0 +1,160 @@
+/*
+ * Copyright © 2007 Intel Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS 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:
+ *    Eric Anholt <eric@anholt.net>
+ *
+ */
+
+#include <string.h>
+#include <stdlib.h>
+#include <assert.h>
+#include "mtypes.h"
+#include "dri_bufmgr.h"
+
+/** @file dri_bufmgr.c
+ *
+ * Convenience functions for buffer management methods.
+ */
+
+dri_bo *
+dri_bo_alloc(dri_bufmgr *bufmgr, const char *name, unsigned long size,
+	     unsigned int alignment, uint64_t location_mask)
+{
+   return bufmgr->bo_alloc(bufmgr, name, size, alignment, location_mask);
+}
+
+dri_bo *
+dri_bo_alloc_static(dri_bufmgr *bufmgr, const char *name, unsigned long offset,
+		    unsigned long size, void *virtual,
+		    uint64_t location_mask)
+{
+   return bufmgr->bo_alloc_static(bufmgr, name, offset, size, virtual,
+				  location_mask);
+}
+
+void
+dri_bo_reference(dri_bo *bo)
+{
+   bo->bufmgr->bo_reference(bo);
+}
+
+void
+dri_bo_unreference(dri_bo *bo)
+{
+   if (bo == NULL)
+      return;
+
+   bo->bufmgr->bo_unreference(bo);
+}
+
+int
+dri_bo_map(dri_bo *buf, GLboolean write_enable)
+{
+   return buf->bufmgr->bo_map(buf, write_enable);
+}
+
+int
+dri_bo_unmap(dri_bo *buf)
+{
+   return buf->bufmgr->bo_unmap(buf);
+}
+
+void
+dri_fence_wait(dri_fence *fence)
+{
+   fence->bufmgr->fence_wait(fence);
+}
+
+void
+dri_fence_reference(dri_fence *fence)
+{
+   fence->bufmgr->fence_reference(fence);
+}
+
+void
+dri_fence_unreference(dri_fence *fence)
+{
+   if (fence == NULL)
+      return;
+
+   fence->bufmgr->fence_unreference(fence);
+}
+
+void
+dri_bo_subdata(dri_bo *bo, unsigned long offset,
+	       unsigned long size, const void *data)
+{
+   if (size == 0 || data == NULL)
+      return;
+
+   dri_bo_map(bo, GL_TRUE);
+   memcpy((unsigned char *)bo->virtual + offset, data, size);
+   dri_bo_unmap(bo);
+}
+
+void
+dri_bo_get_subdata(dri_bo *bo, unsigned long offset,
+		   unsigned long size, void *data)
+{
+   if (size == 0 || data == NULL)
+      return;
+
+   dri_bo_map(bo, GL_FALSE);
+   memcpy(data, (unsigned char *)bo->virtual + offset, size);
+   dri_bo_unmap(bo);
+}
+
+void
+dri_bufmgr_destroy(dri_bufmgr *bufmgr)
+{
+   bufmgr->destroy(bufmgr);
+}
+
+
+int dri_emit_reloc(dri_bo *reloc_buf, uint64_t flags, GLuint delta,
+		    GLuint offset, dri_bo *target_buf)
+{
+   return reloc_buf->bufmgr->emit_reloc(reloc_buf, flags, delta, offset, target_buf);
+}
+
+void *dri_process_relocs(dri_bo *batch_buf, GLuint *count)
+{
+   return batch_buf->bufmgr->process_relocs(batch_buf, count);
+}
+
+void dri_post_submit(dri_bo *batch_buf, dri_fence **last_fence)
+{
+   batch_buf->bufmgr->post_submit(batch_buf, last_fence);
+}
+
+void
+dri_bufmgr_set_debug(dri_bufmgr *bufmgr, GLboolean enable_debug)
+{
+   bufmgr->debug = enable_debug;
+}
+
+int
+dri_bufmgr_check_aperture_space(dri_bo *bo)
+{
+    return bo->bufmgr->check_aperture_space(bo);
+}
diff --git a/src/mesa/drivers/dri/common/dri_bufmgr.h b/src/mesa/drivers/dri/common/dri_bufmgr.h
new file mode 100644
index 0000000000..0a726dc108
--- /dev/null
+++ b/src/mesa/drivers/dri/common/dri_bufmgr.h
@@ -0,0 +1,260 @@
+/**************************************************************************
+ * 
+ * Copyright © 2007 Intel Corporation
+ * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS 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.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * 
+ **************************************************************************/
+/*
+ * Authors: Thomas Hellström <thomas-at-tungstengraphics-dot-com>
+ *          Keith Whitwell <keithw-at-tungstengraphics-dot-com>
+ *	    Eric Anholt <eric@anholt.net>
+ */
+
+#ifndef _DRI_BUFMGR_H_
+#define _DRI_BUFMGR_H_
+#include <xf86drm.h>
+
+typedef struct _dri_bufmgr dri_bufmgr;
+typedef struct _dri_bo dri_bo;
+typedef struct _dri_fence dri_fence;
+
+struct _dri_bo {
+   /** Size in bytes of the buffer object. */
+   unsigned long size;
+   /**
+    * Card virtual address (offset from the beginning of the aperture) for the
+    * object.  Only valid while validated.
+    */
+   unsigned long offset;
+   /**
+    * Virtual address for accessing the buffer data.  Only valid while mapped.
+    */
+   void *virtual;
+   /** Buffer manager context associated with this buffer object */
+   dri_bufmgr *bufmgr;
+};
+
+struct _dri_fence {
+   /**
+    * This is an ORed mask of DRM_BO_FLAG_READ, DRM_BO_FLAG_WRITE, and
+    * DRM_FLAG_EXE indicating the operations associated with this fence.
+    *
+    * It is constant for the life of the fence object.
+    */
+   unsigned int type;
+   /** Buffer manager context associated with this fence */
+   dri_bufmgr *bufmgr;
+};
+
+/**
+ * Context for a buffer manager instance.
+ *
+ * Contains public methods followed by private storage for the buffer manager.
+ */
+struct _dri_bufmgr {
+   /**
+    * Allocate a buffer object.
+    *
+    * Buffer objects are not necessarily initially mapped into CPU virtual
+    * address space or graphics device aperture.  They must be mapped using
+    * bo_map() to be used by the CPU, and validated for use using bo_validate()
+    * to be used from the graphics device.
+    */
+   dri_bo *(*bo_alloc)(dri_bufmgr *bufmgr_ctx, const char *name,
+		       unsigned long size, unsigned int alignment,
+		       uint64_t location_mask);
+
+   /**
+    * Allocates a buffer object for a static allocation.
+    *
+    * Static allocations are ones such as the front buffer that are offered by
+    * the X Server, which are never evicted and never moved.
+    */
+   dri_bo *(*bo_alloc_static)(dri_bufmgr *bufmgr_ctx, const char *name,
+			      unsigned long offset, unsigned long size,
+			      void *virtual, uint64_t location_mask);
+
+   /** Takes a reference on a buffer object */
+   void (*bo_reference)(dri_bo *bo);
+
+   /**
+    * Releases a reference on a buffer object, freeing the data if
+    * rerefences remain.
+    */
+   void (*bo_unreference)(dri_bo *bo);
+
+   /**
+    * Maps the buffer into userspace.
+    *
+    * This function will block waiting for any existing fence on the buffer to
+    * clear, first.  The resulting mapping is available at buf->virtual.
+\    */
+   int (*bo_map)(dri_bo *buf, GLboolean write_enable);
+
+   /** Reduces the refcount on the userspace mapping of the buffer object. */
+   int (*bo_unmap)(dri_bo *buf);
+
+   /** Takes a reference on a fence object */
+   void (*fence_reference)(dri_fence *fence);
+
+   /**
+    * Releases a reference on a fence object, freeing the data if
+    * rerefences remain.
+    */
+   void (*fence_unreference)(dri_fence *fence);
+
+   /**
+    * Blocks until the given fence is signaled.
+    */
+   void (*fence_wait)(dri_fence *fence);
+
+   /**
+    * Tears down the buffer manager instance.
+    */
+   void (*destroy)(dri_bufmgr *bufmgr);
+
+   /**
+    * Add relocation entry in reloc_buf, which will be updated with the
+    * target buffer's real offset on on command submission.
+    *
+    * Relocations remain in place for the lifetime of the buffer object.
+    *
+    * \param reloc_buf Buffer to write the relocation into.
+    * \param flags BO flags to be used in validating the target buffer.
+    *	     Applicable flags include:
+    *	     - DRM_BO_FLAG_READ: The buffer will be read in the process of
+    *	       command execution.
+    *	     - DRM_BO_FLAG_WRITE: The buffer will be written in the process of
+    *	       command execution.
+    *	     - DRM_BO_FLAG_MEM_TT: The buffer should be validated in TT memory.
+    *	     - DRM_BO_FLAG_MEM_VRAM: The buffer should be validated in video
+    *	       memory.
+    * \param delta Constant value to be added to the relocation target's offset.
+    * \param offset Byte offset within batch_buf of the relocated pointer.
+    * \param target Buffer whose offset should be written into the relocation
+    *	     entry.
+    */
+   int (*emit_reloc)(dri_bo *reloc_buf, uint64_t flags, GLuint delta,
+		      GLuint offset, dri_bo *target);
+
+   /**
+    * Processes the relocations, either in userland or by converting the list
+    * for use in batchbuffer submission.
+    *
+    * Kernel-based implementations will return a pointer to the arguments
+    * to be handed with batchbuffer submission to the kernel.  The userland
+    * implementation performs the buffer validation and emits relocations
+    * into them the appopriate order.
+    *
+    * \param batch_buf buffer at the root of the tree of relocations
+    * \param count returns the number of buffers validated.
+    * \return relocation record for use in command submission.
+    * */
+   void *(*process_relocs)(dri_bo *batch_buf, GLuint *count);
+
+   void (*post_submit)(dri_bo *batch_buf, dri_fence **fence);
+
+   int (*check_aperture_space)(dri_bo *bo);
+   GLboolean debug; /**< Enables verbose debugging printouts */
+};
+
+dri_bo *dri_bo_alloc(dri_bufmgr *bufmgr, const char *name, unsigned long size,
+		     unsigned int alignment, uint64_t location_mask);
+dri_bo *dri_bo_alloc_static(dri_bufmgr *bufmgr, const char *name,
+			    unsigned long offset, unsigned long size,
+			    void *virtual, uint64_t location_mask);
+void dri_bo_reference(dri_bo *bo);
+void dri_bo_unreference(dri_bo *bo);
+int dri_bo_map(dri_bo *buf, GLboolean write_enable);
+int dri_bo_unmap(dri_bo *buf);
+void dri_fence_wait(dri_fence *fence);
+void dri_fence_reference(dri_fence *fence);
+void dri_fence_unreference(dri_fence *fence);
+
+void dri_bo_subdata(dri_bo *bo, unsigned long offset,
+		    unsigned long size, const void *data);
+void dri_bo_get_subdata(dri_bo *bo, unsigned long offset,
+			unsigned long size, void *data);
+
+void dri_bufmgr_fake_contended_lock_take(dri_bufmgr *bufmgr);
+dri_bufmgr *dri_bufmgr_fake_init(unsigned long low_offset, void *low_virtual,
+				 unsigned long size,
+				 unsigned int (*fence_emit)(void *private),
+				 int (*fence_wait)(void *private,
+						   unsigned int cookie),
+				 void *driver_priv);
+void dri_bufmgr_set_debug(dri_bufmgr *bufmgr, GLboolean enable_debug);
+void dri_bo_fake_disable_backing_store(dri_bo *bo,
+				       void (*invalidate_cb)(dri_bo *bo,
+							     void *ptr),
+				       void *ptr);
+void dri_bufmgr_destroy(dri_bufmgr *bufmgr);
+
+int dri_emit_reloc(dri_bo *reloc_buf, uint64_t flags, GLuint delta,
+		   GLuint offset, dri_bo *target_buf);
+void *dri_process_relocs(dri_bo *batch_buf, uint32_t *count);
+void dri_post_process_relocs(dri_bo *batch_buf);
+void dri_post_submit(dri_bo *batch_buf, dri_fence **last_fence);
+int dri_bufmgr_check_aperture_space(dri_bo *bo);
+
+#ifndef TTM_API
+/* reuse some TTM API */
+
+#define DRM_BO_MEM_LOCAL 0
+#define DRM_BO_MEM_TT 1
+#define DRM_BO_MEM_VRAM 2
+#define DRM_BO_MEM_PRIV0 3
+#define DRM_BO_MEM_PRIV1 4
+#define DRM_BO_MEM_PRIV2 5
+#define DRM_BO_MEM_PRIV3 6
+#define DRM_BO_MEM_PRIV4 7
+
+#define DRM_BO_FLAG_READ        (1ULL << 0)
+#define DRM_BO_FLAG_WRITE       (1ULL << 1)
+#define DRM_BO_FLAG_EXE         (1ULL << 2)
+#define DRM_BO_MASK_ACCESS	(DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE | DRM_BO_FLAG_EXE)
+#define DRM_BO_FLAG_NO_EVICT    (1ULL << 4)
+
+#define DRM_BO_FLAG_MAPPABLE    (1ULL << 5)
+#define DRM_BO_FLAG_SHAREABLE   (1ULL << 6)
+
+#define DRM_BO_FLAG_CACHED      (1ULL << 7)
+
+#define DRM_BO_FLAG_NO_MOVE     (1ULL << 8)
+#define DRM_BO_FLAG_CACHED_MAPPED    (1ULL << 19)
+#define DRM_BO_FLAG_FORCE_CACHING  (1ULL << 13)
+#define DRM_BO_FLAG_FORCE_MAPPABLE (1ULL << 14)
+#define DRM_BO_FLAG_TILE           (1ULL << 15)
+
+#define DRM_BO_FLAG_MEM_LOCAL  (1ULL << 24)
+#define DRM_BO_FLAG_MEM_TT     (1ULL << 25)
+#define DRM_BO_FLAG_MEM_VRAM   (1ULL << 26)
+
+#define DRM_BO_MASK_MEM         0x00000000FF000000ULL
+
+#define DRM_FENCE_TYPE_EXE                 0x00000001
+#endif
+
+#endif
diff --git a/src/mesa/drivers/dri/common/dri_bufmgr_fake.c b/src/mesa/drivers/dri/common/dri_bufmgr_fake.c
new file mode 100644
index 0000000000..9bf3f3437c
--- /dev/null
+++ b/src/mesa/drivers/dri/common/dri_bufmgr_fake.c
@@ -0,0 +1,1245 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS 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.
+ * 
+ **************************************************************************/
+
+/* Originally a fake version of the buffer manager so that we can
+ * prototype the changes in a driver fairly quickly, has been fleshed
+ * out to a fully functional interim solution.
+ *
+ * Basically wraps the old style memory management in the new
+ * programming interface, but is more expressive and avoids many of
+ * the bugs in the old texture manager.
+ */
+#include "mtypes.h"
+#include "dri_bufmgr.h"
+#include "drm.h"
+
+#include "simple_list.h"
+#include "mm.h"
+#include "imports.h"
+
+#define DBG(...) do {					\
+   if (bufmgr_fake->bufmgr.debug)			\
+      _mesa_printf(__VA_ARGS__);			\
+} while (0)
+
+/* Internal flags:
+ */
+#define BM_NO_BACKING_STORE			0x00000001
+#define BM_NO_FENCE_SUBDATA			0x00000002
+#define BM_PINNED				0x00000004
+
+/* Wrapper around mm.c's mem_block, which understands that you must
+ * wait for fences to expire before memory can be freed.  This is
+ * specific to our use of memcpy for uploads - an upload that was
+ * processed through the command queue wouldn't need to care about
+ * fences.
+ */
+#define MAX_RELOCS 4096
+
+struct fake_buffer_reloc
+{
+   /** Buffer object that the relocation points at. */
+   dri_bo *target_buf;
+   /** Offset of the relocation entry within reloc_buf. */
+   GLuint offset;
+   /** Cached value of the offset when we last performed this relocation. */
+   GLuint last_target_offset;
+   /** Value added to target_buf's offset to get the relocation entry. */
+   GLuint delta;
+   /** Flags to validate the target buffer under. */
+   uint64_t validate_flags;
+};
+
+struct block {
+   struct block *next, *prev;
+   struct mem_block *mem;	/* BM_MEM_AGP */
+
+   /**
+    * Marks that the block is currently in the aperture and has yet to be
+    * fenced.
+    */
+   unsigned on_hardware:1;
+   /**
+    * Marks that the block is currently fenced (being used by rendering) and
+    * can't be freed until @fence is passed.
+    */
+   unsigned fenced:1;
+
+   /** Fence cookie for the block. */
+   unsigned fence; /* Split to read_fence, write_fence */
+
+   dri_bo *bo;
+   void *virtual;
+};
+
+typedef struct _bufmgr_fake {
+   dri_bufmgr bufmgr;
+
+   unsigned long low_offset;
+   unsigned long size;
+   void *virtual;
+
+   struct mem_block *heap;
+   struct block lru;		/* only allocated, non-fence-pending blocks here */
+
+   unsigned buf_nr;		/* for generating ids */
+
+   struct block on_hardware;	/* after bmValidateBuffers */
+   struct block fenced;		/* after bmFenceBuffers (mi_flush, emit irq, write dword) */
+                                /* then to bufmgr->lru or free() */
+
+   unsigned int last_fence;
+
+   unsigned fail:1;
+   unsigned need_fence:1;
+   GLboolean thrashing;
+
+   /**
+    * Driver callback to emit a fence, returning the cookie.
+    *
+    * Currently, this also requires that a write flush be emitted before
+    * emitting the fence, but this should change.
+    */
+   unsigned int (*fence_emit)(void *private);
+   /** Driver callback to wait for a fence cookie to have passed. */
+   int (*fence_wait)(void *private, unsigned int fence_cookie);
+   /** Driver-supplied argument to driver callbacks */
+   void *driver_priv;
+
+   GLboolean debug;
+
+   GLboolean performed_rendering;
+
+   /* keep track of the current total size of objects we have relocs for */
+   unsigned long current_total_size;
+} dri_bufmgr_fake;
+
+typedef struct _dri_bo_fake {
+   dri_bo bo;
+
+   unsigned id;			/* debug only */
+   const char *name;
+
+   unsigned dirty:1;
+   unsigned size_accounted:1; /*this buffers size has been accounted against the aperture */
+   unsigned card_dirty:1; /* has the card written to this buffer - we make need to copy it back */
+   unsigned int refcount;
+   /* Flags may consist of any of the DRM_BO flags, plus
+    * DRM_BO_NO_BACKING_STORE and BM_NO_FENCE_SUBDATA, which are the first two
+    * driver private flags.
+    */
+   uint64_t flags;
+   unsigned int alignment;
+   GLboolean is_static, validated;
+   unsigned int map_count;
+
+   /* Flags for the buffer to be validated with in command submission */
+   uint64_t validate_flags;
+
+   /** relocation list */
+   struct fake_buffer_reloc *relocs;
+   GLuint nr_relocs;
+
+   struct block *block;
+   void *backing_store;
+   void (*invalidate_cb)(dri_bo *bo, void *ptr);
+   void *invalidate_ptr;
+} dri_bo_fake;
+
+typedef struct _dri_fence_fake {
+   dri_fence fence;
+
+   const char *name;
+   unsigned int refcount;
+   unsigned int fence_cookie;
+   GLboolean flushed;
+} dri_fence_fake;
+
+static int clear_fenced(dri_bufmgr_fake *bufmgr_fake,
+			unsigned int fence_cookie);
+
+static int dri_fake_check_aperture_space(dri_bo *bo);
+
+#define MAXFENCE 0x7fffffff
+
+static GLboolean FENCE_LTE( unsigned a, unsigned b )
+{
+   if (a == b)
+      return GL_TRUE;
+
+   if (a < b && b - a < (1<<24))
+      return GL_TRUE;
+
+   if (a > b && MAXFENCE - a + b < (1<<24))
+      return GL_TRUE;
+
+   return GL_FALSE;
+}
+
+static unsigned int
+_fence_emit_internal(dri_bufmgr_fake *bufmgr_fake)
+{
+   bufmgr_fake->last_fence = bufmgr_fake->fence_emit(bufmgr_fake->driver_priv);
+   return bufmgr_fake->last_fence;
+}
+
+static void
+_fence_wait_internal(dri_bufmgr_fake *bufmgr_fake, unsigned int cookie)
+{
+   int ret;
+
+   ret = bufmgr_fake->fence_wait(bufmgr_fake->driver_priv, cookie);
+   if (ret != 0) {
+      _mesa_printf("%s:%d: Error %d waiting for fence.\n",
+		   __FILE__, __LINE__);
+      abort();
+   }
+   clear_fenced(bufmgr_fake, cookie);
+}
+
+static GLboolean
+_fence_test(dri_bufmgr_fake *bufmgr_fake, unsigned fence)
+{
+   /* Slight problem with wrap-around:
+    */
+   return fence == 0 || FENCE_LTE(fence, bufmgr_fake->last_fence);
+}
+
+/**
+ * Allocate a memory manager block for the buffer.
+ */
+static GLboolean
+alloc_block(dri_bo *bo)
+{
+   dri_bo_fake *bo_fake = (dri_bo_fake *)bo;
+   dri_bufmgr_fake *bufmgr_fake= (dri_bufmgr_fake *)bo->bufmgr;
+   struct block *block = (struct block *)calloc(sizeof *block, 1);
+   unsigned int align_log2 = _mesa_ffs(bo_fake->alignment) - 1;
+   GLuint sz;
+
+   if (!block)
+      return GL_FALSE;
+
+   sz = (bo->size + bo_fake->alignment - 1) & ~(bo_fake->alignment - 1);
+
+   block->mem = mmAllocMem(bufmgr_fake->heap, sz, align_log2, 0);
+   if (!block->mem) {
+      free(block);
+      return GL_FALSE;
+   }
+
+   make_empty_list(block);
+
+   /* Insert at head or at tail???   
+    */
+   insert_at_tail(&bufmgr_fake->lru, block);
+
+   block->virtual = bufmgr_fake->virtual +
+      block->mem->ofs - bufmgr_fake->low_offset;
+   block->bo = bo;
+
+   bo_fake->block = block;
+
+   return GL_TRUE;
+}
+
+/* Release the card storage associated with buf:
+ */
+static void free_block(dri_bufmgr_fake *bufmgr_fake, struct block *block)
+{
+   dri_bo_fake *bo_fake;
+   DBG("free block %p %08x %d %d\n", block, block->mem->ofs, block->on_hardware, block->fenced);
+
+   if (!block)
+      return;
+
+   bo_fake = (dri_bo_fake *)block->bo;
+   if (!(bo_fake->flags & BM_NO_BACKING_STORE) && (bo_fake->card_dirty == 1)) {
+     memcpy(bo_fake->backing_store, block->virtual, block->bo->size);
+     bo_fake->card_dirty = 1;
+     bo_fake->dirty = 1;
+   }
+
+   if (block->on_hardware) {
+      block->bo = NULL;
+   }
+   else if (block->fenced) {
+      block->bo = NULL;
+   }
+   else {
+      DBG("    - free immediately\n");
+      remove_from_list(block);
+
+      mmFreeMem(block->mem);
+      free(block);
+   }
+}
+
+static void
+alloc_backing_store(dri_bo *bo)
+{
+   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)bo->bufmgr;
+   dri_bo_fake *bo_fake = (dri_bo_fake *)bo;
+   assert(!bo_fake->backing_store);
+   assert(!(bo_fake->flags & (BM_PINNED|BM_NO_BACKING_STORE)));
+
+   bo_fake->backing_store = ALIGN_MALLOC(bo->size, 64);
+
+   DBG("alloc_backing - buf %d %p %d\n", bo_fake->id, bo_fake->backing_store, bo->size);
+   assert(bo_fake->backing_store);
+}
+
+static void
+free_backing_store(dri_bo *bo)
+{
+   dri_bo_fake *bo_fake = (dri_bo_fake *)bo;
+
+   if (bo_fake->backing_store) {
+      assert(!(bo_fake->flags & (BM_PINNED|BM_NO_BACKING_STORE)));
+      ALIGN_FREE(bo_fake->backing_store);
+      bo_fake->backing_store = NULL;
+   }
+}
+
+static void
+set_dirty(dri_bo *bo)
+{
+   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)bo->bufmgr;
+   dri_bo_fake *bo_fake = (dri_bo_fake *)bo;
+
+   if (bo_fake->flags & BM_NO_BACKING_STORE && bo_fake->invalidate_cb != NULL)
+      bo_fake->invalidate_cb(bo, bo_fake->invalidate_ptr);
+
+   assert(!(bo_fake->flags & BM_PINNED));
+
+   DBG("set_dirty - buf %d\n", bo_fake->id);
+   bo_fake->dirty = 1;
+}
+
+static GLboolean
+evict_lru(dri_bufmgr_fake *bufmgr_fake, GLuint max_fence)
+{
+   struct block *block, *tmp;
+
+   DBG("%s\n", __FUNCTION__);
+
+   foreach_s(block, tmp, &bufmgr_fake->lru) {
+      dri_bo_fake *bo_fake = (dri_bo_fake *)block->bo;
+
+      if (bo_fake != NULL && (bo_fake->flags & BM_NO_FENCE_SUBDATA))
+	 continue;
+
+      if (block->fence && max_fence && !FENCE_LTE(block->fence, max_fence))
+	 return 0;
+
+      set_dirty(&bo_fake->bo);
+      bo_fake->block = NULL;
+
+      free_block(bufmgr_fake, block);
+      return GL_TRUE;
+   }
+
+   return GL_FALSE;
+}
+
+#define foreach_s_rev(ptr, t, list)   \
+        for(ptr=(list)->prev,t=(ptr)->prev; list != ptr; ptr=t, t=(t)->prev)
+
+static GLboolean
+evict_mru(dri_bufmgr_fake *bufmgr_fake)
+{
+   struct block *block, *tmp;
+
+   DBG("%s\n", __FUNCTION__);
+
+   foreach_s_rev(block, tmp, &bufmgr_fake->lru) {
+      dri_bo_fake *bo_fake = (dri_bo_fake *)block->bo;
+
+      if (bo_fake && (bo_fake->flags & BM_NO_FENCE_SUBDATA))
+	 continue;
+
+      set_dirty(&bo_fake->bo);
+      bo_fake->block = NULL;
+
+      free_block(bufmgr_fake, block);
+      return GL_TRUE;
+   }
+
+   return GL_FALSE;
+}
+
+/**
+ * Removes all objects from the fenced list older than the given fence.
+ */
+static int clear_fenced(dri_bufmgr_fake *bufmgr_fake,
+			unsigned int fence_cookie)
+{
+   struct block *block, *tmp;
+   int ret = 0;
+
+   foreach_s(block, tmp, &bufmgr_fake->fenced) {
+      assert(block->fenced);
+
+      if (_fence_test(bufmgr_fake, block->fence)) {
+
+	 block->fenced = 0;
+
+	 if (!block->bo) {
+	    DBG("delayed free: offset %x sz %x\n",
+		block->mem->ofs, block->mem->size);
+	    remove_from_list(block);
+	    mmFreeMem(block->mem);
+	    free(block);
+	 }
+	 else {
+	    DBG("return to lru: offset %x sz %x\n",
+		block->mem->ofs, block->mem->size);
+	    move_to_tail(&bufmgr_fake->lru, block);
+	 }
+
+	 ret = 1;
+      }
+      else {
+	 /* Blocks are ordered by fence, so if one fails, all from
+	  * here will fail also:
+	  */
+	DBG("fence not passed: offset %x sz %x %d %d \n",
+	    block->mem->ofs, block->mem->size, block->fence, bufmgr_fake->last_fence);
+	 break;
+      }
+   }
+
+   DBG("%s: %d\n", __FUNCTION__, ret);
+   return ret;
+}
+
+static void fence_blocks(dri_bufmgr_fake *bufmgr_fake, unsigned fence)
+{
+   struct block *block, *tmp;
+
+   foreach_s (block, tmp, &bufmgr_fake->on_hardware) {
+      DBG("Fence block %p (sz 0x%x ofs %x buf %p) with fence %d\n", block,
+	  block->mem->size, block->mem->ofs, block->bo, fence);
+      block->fence = fence;
+
+      block->on_hardware = 0;
+      block->fenced = 1;
+
+      /* Move to tail of pending list here
+       */
+      move_to_tail(&bufmgr_fake->fenced, block);
+   }
+
+   assert(is_empty_list(&bufmgr_fake->on_hardware));
+}
+
+static GLboolean evict_and_alloc_block(dri_bo *bo)
+{
+   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)bo->bufmgr;
+   dri_bo_fake *bo_fake = (dri_bo_fake *)bo;
+
+   assert(bo_fake->block == NULL);
+
+   /* Search for already free memory:
+    */
+   if (alloc_block(bo))
+      return GL_TRUE;
+
+   /* If we're not thrashing, allow lru eviction to dig deeper into
+    * recently used textures.  We'll probably be thrashing soon:
+    */
+   if (!bufmgr_fake->thrashing) {
+      while (evict_lru(bufmgr_fake, 0))
+	 if (alloc_block(bo))
+	    return GL_TRUE;
+   }
+
+   /* Keep thrashing counter alive?
+    */
+   if (bufmgr_fake->thrashing)
+      bufmgr_fake->thrashing = 20;
+
+   /* Wait on any already pending fences - here we are waiting for any
+    * freed memory that has been submitted to hardware and fenced to
+    * become available:
+    */
+   while (!is_empty_list(&bufmgr_fake->fenced)) {
+      GLuint fence = bufmgr_fake->fenced.next->fence;
+      _fence_wait_internal(bufmgr_fake, fence);
+
+      if (alloc_block(bo))
+	 return GL_TRUE;
+   }
+
+   if (!is_empty_list(&bufmgr_fake->on_hardware)) {
+      while (!is_empty_list(&bufmgr_fake->fenced)) {
+	 GLuint fence = bufmgr_fake->fenced.next->fence;
+	 _fence_wait_internal(bufmgr_fake, fence);
+      }
+
+      if (!bufmgr_fake->thrashing) {
+	 DBG("thrashing\n");
+      }
+      bufmgr_fake->thrashing = 20;
+
+      if (alloc_block(bo))
+	 return GL_TRUE;
+   }
+
+   while (evict_mru(bufmgr_fake))
+      if (alloc_block(bo))
+	 return GL_TRUE;
+
+   DBG("%s 0x%x bytes failed\n", __FUNCTION__, bo->size);
+
+   return GL_FALSE;
+}
+
+/***********************************************************************
+ * Public functions
+ */
+
+/**
+ * Wait for hardware idle by emitting a fence and waiting for it.
+ */
+static void
+dri_bufmgr_fake_wait_idle(dri_bufmgr_fake *bufmgr_fake)
+{
+   unsigned int cookie;
+
+   cookie = bufmgr_fake->fence_emit(bufmgr_fake->driver_priv);
+   _fence_wait_internal(bufmgr_fake, cookie);
+}
+
+/**
+ * Wait for execution pending on a buffer
+ */
+static void
+dri_bufmgr_fake_bo_wait_idle(dri_bo *bo)
+{
+   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)bo->bufmgr;
+   dri_bo_fake *bo_fake = (dri_bo_fake *)bo;
+
+   if (bo_fake->block == NULL || !bo_fake->block->fenced)
+      return;
+
+   _fence_wait_internal(bufmgr_fake, bo_fake->block->fence);
+}
+
+/* Specifically ignore texture memory sharing.
+ *  -- just evict everything
+ *  -- and wait for idle
+ */
+void
+dri_bufmgr_fake_contended_lock_take(dri_bufmgr *bufmgr)
+{
+   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)bufmgr;
+   struct block *block, *tmp;
+
+   bufmgr_fake->need_fence = 1;
+   bufmgr_fake->fail = 0;
+
+   /* Wait for hardware idle.  We don't know where acceleration has been
+    * happening, so we'll need to wait anyway before letting anything get
+    * put on the card again.
+    */
+   dri_bufmgr_fake_wait_idle(bufmgr_fake);
+
+   /* Check that we hadn't released the lock without having fenced the last
+    * set of buffers.
+    */
+   assert(is_empty_list(&bufmgr_fake->fenced));
+   assert(is_empty_list(&bufmgr_fake->on_hardware));
+
+   foreach_s(block, tmp, &bufmgr_fake->lru) {
+      assert(_fence_test(bufmgr_fake, block->fence));
+      set_dirty(block->bo);
+   }
+}
+
+static dri_bo *
+dri_fake_bo_alloc(dri_bufmgr *bufmgr, const char *name,
+		  unsigned long size, unsigned int alignment,
+		  uint64_t location_mask)
+{
+   dri_bufmgr_fake *bufmgr_fake;
+   dri_bo_fake *bo_fake;
+
+   bufmgr_fake = (dri_bufmgr_fake *)bufmgr;
+
+   assert(size != 0);
+
+   bo_fake = calloc(1, sizeof(*bo_fake));
+   if (!bo_fake)
+      return NULL;
+
+   bo_fake->bo.size = size;
+   bo_fake->bo.offset = -1;
+   bo_fake->bo.virtual = NULL;
+   bo_fake->bo.bufmgr = bufmgr;
+   bo_fake->refcount = 1;
+
+   /* Alignment must be a power of two */
+   assert((alignment & (alignment - 1)) == 0);
+   if (alignment == 0)
+      alignment = 1;
+   bo_fake->alignment = alignment;
+   bo_fake->id = ++bufmgr_fake->buf_nr;
+   bo_fake->name = name;
+   bo_fake->flags = 0;
+   bo_fake->is_static = GL_FALSE;
+
+   DBG("drm_bo_alloc: (buf %d: %s, %d kb)\n", bo_fake->id, bo_fake->name,
+       bo_fake->bo.size / 1024);
+
+   return &bo_fake->bo;
+}
+
+static dri_bo *
+dri_fake_bo_alloc_static(dri_bufmgr *bufmgr, const char *name,
+			 unsigned long offset, unsigned long size,
+			 void *virtual, uint64_t location_mask)
+{
+   dri_bufmgr_fake *bufmgr_fake;
+   dri_bo_fake *bo_fake;
+
+   bufmgr_fake = (dri_bufmgr_fake *)bufmgr;
+
+   assert(size != 0);
+
+   bo_fake = calloc(1, sizeof(*bo_fake));
+   if (!bo_fake)
+      return NULL;
+
+   bo_fake->bo.size = size;
+   bo_fake->bo.offset = offset;
+   bo_fake->bo.virtual = virtual;
+   bo_fake->bo.bufmgr = bufmgr;
+   bo_fake->refcount = 1;
+   bo_fake->id = ++bufmgr_fake->buf_nr;
+   bo_fake->name = name;
+   bo_fake->flags = BM_PINNED | DRM_BO_FLAG_NO_MOVE;
+   bo_fake->is_static = GL_TRUE;
+
+   DBG("drm_bo_alloc_static: (buf %d: %s, %d kb)\n", bo_fake->id, bo_fake->name,
+       bo_fake->bo.size / 1024);
+
+   return &bo_fake->bo;
+}
+
+static void
+dri_fake_bo_reference(dri_bo *bo)
+{
+   dri_bo_fake *bo_fake = (dri_bo_fake *)bo;
+
+   bo_fake->refcount++;
+}
+
+static void
+dri_fake_bo_unreference(dri_bo *bo)
+{
+   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)bo->bufmgr;
+   dri_bo_fake *bo_fake = (dri_bo_fake *)bo;
+   int i;
+
+   if (!bo)
+      return;
+
+   if (--bo_fake->refcount == 0) {
+      assert(bo_fake->map_count == 0);
+      /* No remaining references, so free it */
+      if (bo_fake->block)
+	 free_block(bufmgr_fake, bo_fake->block);
+      free_backing_store(bo);
+
+      for (i = 0; i < bo_fake->nr_relocs; i++)
+	 dri_bo_unreference(bo_fake->relocs[i].target_buf);
+
+      DBG("drm_bo_unreference: free buf %d %s\n", bo_fake->id, bo_fake->name);
+
+      free(bo_fake->relocs);
+      free(bo);
+
+      return;
+   }
+}
+
+/**
+ * Set the buffer as not requiring backing store, and instead get the callback
+ * invoked whenever it would be set dirty.
+ */
+void dri_bo_fake_disable_backing_store(dri_bo *bo,
+				       void (*invalidate_cb)(dri_bo *bo,
+							     void *ptr),
+				       void *ptr)
+{
+   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)bo->bufmgr;
+   dri_bo_fake *bo_fake = (dri_bo_fake *)bo;
+
+   if (bo_fake->backing_store)
+      free_backing_store(bo);
+
+   bo_fake->flags |= BM_NO_BACKING_STORE;
+
+   DBG("disable_backing_store set buf %d dirty\n", bo_fake->id);
+   bo_fake->dirty = 1;
+   bo_fake->invalidate_cb = invalidate_cb;
+   bo_fake->invalidate_ptr = ptr;
+
+   /* Note that it is invalid right from the start.  Also note
+    * invalidate_cb is called with the bufmgr locked, so cannot
+    * itself make bufmgr calls.
+    */
+   if (invalidate_cb != NULL)
+      invalidate_cb(bo, ptr);
+}
+
+/**
+ * Map a buffer into bo->virtual, allocating either card memory space (If
+ * BM_NO_BACKING_STORE or BM_PINNED) or backing store, as necessary.
+ */
+static int
+dri_fake_bo_map(dri_bo *bo, GLboolean write_enable)
+{
+   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)bo->bufmgr;
+   dri_bo_fake *bo_fake = (dri_bo_fake *)bo;
+
+   /* Static buffers are always mapped. */
+   if (bo_fake->is_static)
+      return 0;
+
+   /* Allow recursive mapping.  Mesa may recursively map buffers with
+    * nested display loops, and it is used internally in bufmgr_fake
+    * for relocation.
+    */
+   if (bo_fake->map_count++ != 0)
+      return 0;
+
+   {
+      DBG("drm_bo_map: (buf %d: %s, %d kb)\n", bo_fake->id, bo_fake->name,
+	  bo_fake->bo.size / 1024);
+
+      if (bo->virtual != NULL) {
+	 _mesa_printf("%s: already mapped\n", __FUNCTION__);
+	 abort();
+      }
+      else if (bo_fake->flags & (BM_NO_BACKING_STORE|BM_PINNED)) {
+
+	 if (!bo_fake->block && !evict_and_alloc_block(bo)) {
+	    DBG("%s: alloc failed\n", __FUNCTION__);
+	    bufmgr_fake->fail = 1;
+	    return 1;
+	 }
+	 else {
+	    assert(bo_fake->block);
+	    bo_fake->dirty = 0;
+
+	    if (!(bo_fake->flags & BM_NO_FENCE_SUBDATA) &&
+		bo_fake->block->fenced) {
+	       dri_bufmgr_fake_bo_wait_idle(bo);
+	    }
+
+	    bo->virtual = bo_fake->block->virtual;
+	 }
+      }
+      else {
+	 if (write_enable)
+	    set_dirty(bo);
+
+	 if (bo_fake->backing_store == 0)
+	    alloc_backing_store(bo);
+
+	 bo->virtual = bo_fake->backing_store;
+      }
+   }
+
+   return 0;
+}
+
+static int
+dri_fake_bo_unmap(dri_bo *bo)
+{
+   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)bo->bufmgr;
+   dri_bo_fake *bo_fake = (dri_bo_fake *)bo;
+
+   /* Static buffers are always mapped. */
+   if (bo_fake->is_static)
+      return 0;
+
+   assert(bo_fake->map_count != 0);
+   if (--bo_fake->map_count != 0)
+      return 0;
+
+   DBG("drm_bo_unmap: (buf %d: %s, %d kb)\n", bo_fake->id, bo_fake->name,
+       bo_fake->bo.size / 1024);
+
+   bo->virtual = NULL;
+
+   return 0;
+}
+
+static void
+dri_fake_kick_all(dri_bufmgr_fake *bufmgr_fake)
+{
+   struct block *block, *tmp;
+
+   bufmgr_fake->performed_rendering = GL_FALSE;
+   /* okay for ever BO that is on the HW kick it off.
+      seriously not afraid of the POLICE right now */
+   foreach_s(block, tmp, &bufmgr_fake->on_hardware) {
+      dri_bo_fake *bo_fake = (dri_bo_fake *)block->bo;
+
+      block->on_hardware = 0;
+      free_block(bufmgr_fake, block);
+      bo_fake->block = NULL;
+      bo_fake->validated = GL_FALSE;
+      if (!(bo_fake->flags & BM_NO_BACKING_STORE))
+         bo_fake->dirty = 1;
+   }
+}
+
+static int
+dri_fake_bo_validate(dri_bo *bo, uint64_t flags)
+{
+   dri_bufmgr_fake *bufmgr_fake;
+   dri_bo_fake *bo_fake = (dri_bo_fake *)bo;
+
+   /* XXX: Sanity-check whether we've already validated this one under
+    * different flags.  See drmAddValidateItem().
+    */
+   bufmgr_fake = (dri_bufmgr_fake *)bo->bufmgr;
+
+   DBG("drm_bo_validate: (buf %d: %s, %d kb)\n", bo_fake->id, bo_fake->name,
+       bo_fake->bo.size / 1024);
+
+   /* Sanity check: Buffers should be unmapped before being validated.
+    * This is not so much of a problem for bufmgr_fake, but TTM refuses,
+    * and the problem is harder to debug there.
+    */
+   assert(bo_fake->map_count == 0);
+
+   if (bo_fake->is_static) {
+      /* Add it to the needs-fence list */
+      bufmgr_fake->need_fence = 1;
+      return 0;
+   }
+
+   /* reset size accounted */
+   bo_fake->size_accounted = 0;
+
+   /* Allocate the card memory */
+   if (!bo_fake->block && !evict_and_alloc_block(bo)) {
+      bufmgr_fake->fail = 1;
+      DBG("Failed to validate buf %d:%s\n", bo_fake->id, bo_fake->name);
+      return -1;
+   }
+
+   assert(bo_fake->block);
+   assert(bo_fake->block->bo == &bo_fake->bo);
+
+   bo->offset = bo_fake->block->mem->ofs;
+
+   /* Upload the buffer contents if necessary */
+   if (bo_fake->dirty) {
+      DBG("Upload dirty buf %d:%s, sz %d offset 0x%x\n", bo_fake->id,
+	  bo_fake->name, bo->size, bo_fake->block->mem->ofs);
+
+      assert(!(bo_fake->flags &
+	       (BM_NO_BACKING_STORE|BM_PINNED)));
+
+      /* Actually, should be able to just wait for a fence on the memory,
+       * which we would be tracking when we free it.  Waiting for idle is
+       * a sufficiently large hammer for now.
+       */
+      dri_bufmgr_fake_wait_idle(bufmgr_fake);
+
+      /* we may never have mapped this BO so it might not have any backing
+       * store if this happens it should be rare, but 0 the card memory
+       * in any case */
+      if (bo_fake->backing_store)
+         memcpy(bo_fake->block->virtual, bo_fake->backing_store, bo->size);
+      else
+         memset(bo_fake->block->virtual, 0, bo->size);
+
+      bo_fake->dirty = 0;
+   }
+
+   bo_fake->block->fenced = 0;
+   bo_fake->block->on_hardware = 1;
+   move_to_tail(&bufmgr_fake->on_hardware, bo_fake->block);
+
+   bo_fake->validated = GL_TRUE;
+   bufmgr_fake->need_fence = 1;
+
+   return 0;
+}
+
+static dri_fence *
+dri_fake_fence_validated(dri_bufmgr *bufmgr, const char *name,
+			 GLboolean flushed)
+{
+   dri_fence_fake *fence_fake;
+   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)bufmgr;
+   unsigned int cookie;
+
+   fence_fake = malloc(sizeof(*fence_fake));
+   if (!fence_fake)
+      return NULL;
+
+   fence_fake->refcount = 1;
+   fence_fake->name = name;
+   fence_fake->flushed = flushed;
+   fence_fake->fence.bufmgr = bufmgr;
+
+   cookie = _fence_emit_internal(bufmgr_fake);
+   fence_fake->fence_cookie = cookie;
+   fence_blocks(bufmgr_fake, cookie);
+
+   DBG("drm_fence_validated: 0x%08x cookie\n", fence_fake->fence_cookie);
+
+   return &fence_fake->fence;
+}
+
+static void
+dri_fake_fence_reference(dri_fence *fence)
+{
+   dri_fence_fake *fence_fake = (dri_fence_fake *)fence;
+
+   ++fence_fake->refcount;
+}
+
+static void
+dri_fake_fence_unreference(dri_fence *fence)
+{
+   dri_fence_fake *fence_fake = (dri_fence_fake *)fence;
+
+   if (!fence)
+      return;
+
+   if (--fence_fake->refcount == 0) {
+      free(fence);
+      return;
+   }
+}
+
+static void
+dri_fake_fence_wait(dri_fence *fence)
+{
+   dri_fence_fake *fence_fake = (dri_fence_fake *)fence;
+   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)fence->bufmgr;
+
+   DBG("drm_fence_wait: 0x%08x cookie\n", fence_fake->fence_cookie);
+
+   _fence_wait_internal(bufmgr_fake, fence_fake->fence_cookie);
+}
+
+static void
+dri_fake_destroy(dri_bufmgr *bufmgr)
+{
+   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)bufmgr;
+
+   mmDestroy(bufmgr_fake->heap);
+   free(bufmgr);
+}
+
+static int
+dri_fake_emit_reloc(dri_bo *reloc_buf, uint64_t flags, GLuint delta,
+		    GLuint offset, dri_bo *target_buf)
+{
+   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)reloc_buf->bufmgr;
+   struct fake_buffer_reloc *r;
+   dri_bo_fake *reloc_fake = (dri_bo_fake *)reloc_buf;
+   dri_bo_fake *target_fake = (dri_bo_fake *)target_buf;
+   int i;
+
+   assert(reloc_buf);
+   assert(target_buf);
+
+   assert(target_fake->is_static || target_fake->size_accounted);
+
+   if (reloc_fake->relocs == NULL) {
+      reloc_fake->relocs = malloc(sizeof(struct fake_buffer_reloc) *
+				  MAX_RELOCS);
+   }
+
+   r = &reloc_fake->relocs[reloc_fake->nr_relocs++];
+
+   assert(reloc_fake->nr_relocs <= MAX_RELOCS);
+
+   dri_bo_reference(target_buf);
+
+   r->target_buf = target_buf;
+   r->offset = offset;
+   r->last_target_offset = target_buf->offset;
+   r->delta = delta;
+   r->validate_flags = flags;
+
+   if (bufmgr_fake->debug) {
+      /* Check that a conflicting relocation hasn't already been emitted. */
+      for (i = 0; i < reloc_fake->nr_relocs - 1; i++) {
+	 struct fake_buffer_reloc *r2 = &reloc_fake->relocs[i];
+
+	 assert(r->offset != r2->offset);
+      }
+   }
+
+   return 0;
+}
+
+/**
+ * Incorporates the validation flags associated with each relocation into
+ * the combined validation flags for the buffer on this batchbuffer submission.
+ */
+static void
+dri_fake_calculate_validate_flags(dri_bo *bo)
+{
+   dri_bo_fake *bo_fake = (dri_bo_fake *)bo;
+   int i;
+
+   for (i = 0; i < bo_fake->nr_relocs; i++) {
+      struct fake_buffer_reloc *r = &bo_fake->relocs[i];
+      dri_bo_fake *target_fake = (dri_bo_fake *)r->target_buf;
+
+      /* Do the same for the tree of buffers we depend on */
+      dri_fake_calculate_validate_flags(r->target_buf);
+
+      if (target_fake->validate_flags == 0) {
+	 target_fake->validate_flags = r->validate_flags;
+      } else {
+	 /* Mask the memory location to the intersection of all the memory
+	  * locations the buffer is being validated to.
+	  */
+	 target_fake->validate_flags =
+	    (target_fake->validate_flags & ~DRM_BO_MASK_MEM) |
+	    (r->validate_flags & target_fake->validate_flags &
+	     DRM_BO_MASK_MEM);
+	 /* All the other flags just accumulate. */
+	 target_fake->validate_flags |= r->validate_flags & ~DRM_BO_MASK_MEM;
+      }
+   }
+}
+
+
+static int
+dri_fake_reloc_and_validate_buffer(dri_bo *bo)
+{
+   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)bo->bufmgr;
+   dri_bo_fake *bo_fake = (dri_bo_fake *)bo;
+   int i, ret;
+
+   assert(bo_fake->map_count == 0);
+
+   for (i = 0; i < bo_fake->nr_relocs; i++) {
+      struct fake_buffer_reloc *r = &bo_fake->relocs[i];
+      dri_bo_fake *target_fake = (dri_bo_fake *)r->target_buf;
+      uint32_t reloc_data;
+
+      /* Validate the target buffer if that hasn't been done. */
+      if (!target_fake->validated) {
+         ret = dri_fake_reloc_and_validate_buffer(r->target_buf);
+         if (ret != 0) {
+            if (bo->virtual != NULL)
+                dri_bo_unmap(bo);
+            return ret;
+         }
+      }
+
+      /* Calculate the value of the relocation entry. */
+      if (r->target_buf->offset != r->last_target_offset) {
+	 reloc_data = r->target_buf->offset + r->delta;
+
+	 if (bo->virtual == NULL)
+	    dri_bo_map(bo, GL_TRUE);
+
+	 *(uint32_t *)(bo->virtual + r->offset) = reloc_data;
+
+	 r->last_target_offset = r->target_buf->offset;
+      }
+   }
+
+   if (bo->virtual != NULL)
+      dri_bo_unmap(bo);
+
+   if (bo_fake->validate_flags & DRM_BO_FLAG_WRITE) {
+      if (!(bo_fake->flags & (BM_NO_BACKING_STORE|BM_PINNED))) {
+         if (bo_fake->backing_store == 0)
+            alloc_backing_store(bo);
+
+         bo_fake->card_dirty = 1;
+      }
+      bufmgr_fake->performed_rendering = GL_TRUE;
+   }
+
+   return dri_fake_bo_validate(bo, bo_fake->validate_flags);
+}
+
+static void *
+dri_fake_process_relocs(dri_bo *batch_buf, GLuint *count_p)
+{
+   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)batch_buf->bufmgr;
+   dri_bo_fake *batch_fake = (dri_bo_fake *)batch_buf;
+   int ret;
+   int retry_count = 0;
+
+   bufmgr_fake->performed_rendering = GL_FALSE;
+
+   dri_fake_calculate_validate_flags(batch_buf);
+
+   batch_fake->validate_flags = DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ;
+
+   /* we've ran out of RAM so blow the whole lot away and retry */
+ restart:
+   ret = dri_fake_reloc_and_validate_buffer(batch_buf);
+   if (bufmgr_fake->fail == 1) {
+      if (retry_count == 0) {
+         retry_count++;
+         dri_fake_kick_all(bufmgr_fake);
+         bufmgr_fake->fail = 0;
+         goto restart;
+      } else /* dump out the memory here */
+         mmDumpMemInfo(bufmgr_fake->heap);
+   }
+
+   assert(ret == 0);
+
+   *count_p = 0; /* junk */
+
+   bufmgr_fake->current_total_size = 0;
+   return NULL;
+}
+
+static void
+dri_bo_fake_post_submit(dri_bo *bo)
+{
+   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)bo->bufmgr;
+   dri_bo_fake *bo_fake = (dri_bo_fake *)bo;
+   int i;
+
+   for (i = 0; i < bo_fake->nr_relocs; i++) {
+      struct fake_buffer_reloc *r = &bo_fake->relocs[i];
+      dri_bo_fake *target_fake = (dri_bo_fake *)r->target_buf;
+
+      if (target_fake->validated)
+	 dri_bo_fake_post_submit(r->target_buf);
+
+      DBG("%s@0x%08x + 0x%08x -> %s@0x%08x + 0x%08x\n",
+	  bo_fake->name, (uint32_t)bo->offset, r->offset,
+	  target_fake->name, (uint32_t)r->target_buf->offset, r->delta);
+   }
+
+   assert(bo_fake->map_count == 0);
+   bo_fake->validated = GL_FALSE;
+   bo_fake->validate_flags = 0;
+}
+
+
+static void
+dri_fake_post_submit(dri_bo *batch_buf, dri_fence **last_fence)
+{
+   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)batch_buf->bufmgr;
+   dri_fence *fo;
+
+   fo = dri_fake_fence_validated(batch_buf->bufmgr, "Batch fence", GL_TRUE);
+
+   if (bufmgr_fake->performed_rendering) {
+      dri_fence_unreference(*last_fence);
+      *last_fence = fo;
+   } else {
+      dri_fence_unreference(fo);
+   }
+
+   dri_bo_fake_post_submit(batch_buf);
+}
+
+static int
+dri_fake_check_aperture_space(dri_bo *bo)
+{
+   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)bo->bufmgr;
+   dri_bo_fake *bo_fake = (dri_bo_fake *)bo;
+   GLuint sz;
+
+   sz = (bo->size + bo_fake->alignment - 1) & ~(bo_fake->alignment - 1);
+
+   if (bo_fake->size_accounted || bo_fake->is_static)
+      return 0;
+
+   if (bufmgr_fake->current_total_size + sz > bufmgr_fake->size) {
+     DBG("check_space: %s bo %d %d overflowed bufmgr size %d\n", bo_fake->name, bo_fake->id, sz, bufmgr_fake->size);
+      return -1;
+   }
+
+   bufmgr_fake->current_total_size += sz;
+   bo_fake->size_accounted = 1;
+   DBG("drm_check_space: buf %d, %s %d %d\n", bo_fake->id, bo_fake->name, bo->size, bufmgr_fake->current_total_size);
+   return 0;
+}
+
+dri_bufmgr *
+dri_bufmgr_fake_init(unsigned long low_offset, void *low_virtual,
+		     unsigned long size,
+		     unsigned int (*fence_emit)(void *private),
+		     int (*fence_wait)(void *private, unsigned int cookie),
+		     void *driver_priv)
+{
+   dri_bufmgr_fake *bufmgr_fake;
+
+   bufmgr_fake = calloc(1, sizeof(*bufmgr_fake));
+
+   /* Initialize allocator */
+   make_empty_list(&bufmgr_fake->fenced);
+   make_empty_list(&bufmgr_fake->on_hardware);
+   make_empty_list(&bufmgr_fake->lru);
+
+   bufmgr_fake->low_offset = low_offset;
+   bufmgr_fake->virtual = low_virtual;
+   bufmgr_fake->size = size;
+   bufmgr_fake->heap = mmInit(low_offset, size);
+
+   /* Hook in methods */
+   bufmgr_fake->bufmgr.bo_alloc = dri_fake_bo_alloc;
+   bufmgr_fake->bufmgr.bo_alloc_static = dri_fake_bo_alloc_static;
+   bufmgr_fake->bufmgr.bo_reference = dri_fake_bo_reference;
+   bufmgr_fake->bufmgr.bo_unreference = dri_fake_bo_unreference;
+   bufmgr_fake->bufmgr.bo_map = dri_fake_bo_map;
+   bufmgr_fake->bufmgr.bo_unmap = dri_fake_bo_unmap;
+   bufmgr_fake->bufmgr.fence_wait = dri_fake_fence_wait;
+   bufmgr_fake->bufmgr.fence_reference = dri_fake_fence_reference;
+   bufmgr_fake->bufmgr.fence_unreference = dri_fake_fence_unreference;
+   bufmgr_fake->bufmgr.destroy = dri_fake_destroy;
+   bufmgr_fake->bufmgr.emit_reloc = dri_fake_emit_reloc;
+   bufmgr_fake->bufmgr.process_relocs = dri_fake_process_relocs;
+   bufmgr_fake->bufmgr.post_submit = dri_fake_post_submit;
+   bufmgr_fake->bufmgr.check_aperture_space = dri_fake_check_aperture_space;
+   bufmgr_fake->bufmgr.debug = GL_FALSE;
+
+   bufmgr_fake->fence_emit = fence_emit;
+   bufmgr_fake->fence_wait = fence_wait;
+   bufmgr_fake->driver_priv = driver_priv;
+
+   return &bufmgr_fake->bufmgr;
+}
+
diff --git a/src/mesa/drivers/dri/common/spantmp2.h b/src/mesa/drivers/dri/common/spantmp2.h
index a1e56eb148..53f5f846a0 100644
--- a/src/mesa/drivers/dri/common/spantmp2.h
+++ b/src/mesa/drivers/dri/common/spantmp2.h
@@ -48,34 +48,36 @@
 #define HW_WRITE_CLIPLOOP()	HW_CLIPLOOP()
 #endif
 
+
 #if (SPANTMP_PIXEL_FMT == GL_RGB)  && (SPANTMP_PIXEL_TYPE == GL_UNSIGNED_SHORT_5_6_5)
 
 /**
  ** GL_RGB, GL_UNSIGNED_SHORT_5_6_5
  **/
 
-#ifndef GET_VALUE
 #ifndef GET_PTR
 #define GET_PTR(_x, _y) (buf + (_x) * 2 + (_y) * pitch)
 #endif
 
-#define GET_VALUE(_x, _y) *(volatile GLushort *)(GET_PTR(_x, _y))
-#define PUT_VALUE(_x, _y, _v) *(volatile GLushort *)(GET_PTR(_x, _y)) = (_v)
-#endif /* GET_VALUE */
-
 #define INIT_MONO_PIXEL(p, color) \
   p = PACK_COLOR_565( color[0], color[1], color[2] )
 
 #define WRITE_RGBA( _x, _y, r, g, b, a )				\
-   PUT_VALUE(_x, _y, ((((int)r & 0xf8) << 8) |				\
-		      (((int)g & 0xfc) << 3) |				\
-		      (((int)b & 0xf8) >> 3)))				\
+    do {                                                                \
+       GLshort * _p = (GLshort *) GET_PTR(_x, _y);                      \
+       _p[0] = ((((int)r & 0xf8) << 8) | (((int)g & 0xfc) << 3) |	\
+		   (((int)b & 0xf8) >> 3));                             \
+   } while(0)
 
-#define WRITE_PIXEL( _x, _y, p ) PUT_VALUE(_x, _y, p)
+#define WRITE_PIXEL( _x, _y, p )					\
+   do {                                                                 \
+      GLushort * _p = (GLushort *) GET_PTR(_x, _y);                     \
+      _p[0] = p;                                                        \
+   } while(0)
 
 #define READ_RGBA( rgba, _x, _y )					\
    do {									\
-      GLushort p = GET_VALUE(_x, _y);					\
+      GLushort p = *(volatile GLshort *) GET_PTR(_x, _y);               \
       rgba[0] = ((p >> 8) & 0xf8) * 255 / 0xf8;				\
       rgba[1] = ((p >> 3) & 0xfc) * 255 / 0xfc;				\
       rgba[2] = ((p << 3) & 0xf8) * 255 / 0xf8;				\
@@ -88,30 +90,29 @@
  ** GL_BGRA, GL_UNSIGNED_INT_8_8_8_8_REV
  **/
 
-#ifndef GET_VALUE
 #ifndef GET_PTR
 #define GET_PTR(_x, _y) (     buf + (_x) * 4 + (_y) * pitch)
 #endif
 
-#define GET_VALUE(_x, _y) *(volatile GLuint *)(GET_PTR(_x, _y))
-#define PUT_VALUE(_x, _y, _v) *(volatile GLuint *)(GET_PTR(_x, _y)) = (_v)
-#endif /* GET_VALUE */
-
 # define INIT_MONO_PIXEL(p, color)                       \
      p = PACK_COLOR_8888(color[3], color[0], color[1], color[2]) 
 
 # define WRITE_RGBA(_x, _y, r, g, b, a)                                 \
-   PUT_VALUE(_x, _y, ((r << 16) |					\
-		      (g << 8) |					\
-		      (b << 0) |					\
-		      (a << 24)))
+    do {                                                                \
+       GLuint * _p = (GLuint *) GET_PTR(_x, _y);                        \
+       _p[0] = ((r << 16) | (g << 8) | (b << 0) | (a << 24));           \
+    } while(0)
 
-#define WRITE_PIXEL(_x, _y, p) PUT_VALUE(_x, _y, p)
+#define WRITE_PIXEL(_x, _y, p)                                          \
+    do {                                                                \
+       GLuint * _p = (GLuint *) GET_PTR(_x, _y);                        \
+       _p[0] = p;                                                       \
+    } while(0)
 
 # if defined( USE_X86_ASM )
 #  define READ_RGBA(rgba, _x, _y)                                       \
     do {                                                                \
-       GLuint p = GET_VALUE(_x, _y);					\
+        GLuint p = *(volatile GLuint *) GET_PTR(_x, _y);                \
        __asm__ __volatile__( "bswap	%0; rorl $8, %0"                \
 				: "=r" (p) : "0" (p) );                 \
        ((GLuint *)rgba)[0] = p;                                         \
@@ -122,14 +123,14 @@
      */
 #  define READ_RGBA( rgba, _x, _y )				        \
      do {								\
-        GLuint p = GET_VALUE(_x, _y);					\
+        GLuint p = *(volatile GLuint *) GET_PTR(_x, _y);                \
         GLuint t = p;                                                   \
         *((uint32_t *) rgba) = (t >> 24) | (p << 8);                    \
      } while (0)
 # else
 #  define READ_RGBA( rgba, _x, _y )				        \
      do {								\
-        GLuint p = GET_VALUE(_x, _y);					\
+        GLuint p = *(volatile GLuint *) GET_PTR(_x, _y);                \
 	rgba[0] = (p >> 16) & 0xff;					\
 	rgba[1] = (p >>  8) & 0xff;					\
 	rgba[2] = (p >>  0) & 0xff;					\
@@ -388,8 +389,7 @@ static void TAG(ReadRGBASpan)( GLcontext *ctx,
 }
 
 
-#if defined(GET_PTR) && \
-   defined(USE_MMX_ASM) && \
+#if defined(USE_MMX_ASM) && \
    (((SPANTMP_PIXEL_FMT == GL_BGRA) && \
 	(SPANTMP_PIXEL_TYPE == GL_UNSIGNED_INT_8_8_8_8_REV)) || \
     ((SPANTMP_PIXEL_FMT == GL_RGB) && \
@@ -440,8 +440,7 @@ static void TAG2(ReadRGBASpan,_MMX)( GLcontext *ctx,
 #endif
 
 
-#if defined(GET_PTR) &&	\
-   defined(USE_SSE_ASM) && \
+#if defined(USE_SSE_ASM) && \
    (SPANTMP_PIXEL_FMT == GL_BGRA) && \
      (SPANTMP_PIXEL_TYPE == GL_UNSIGNED_INT_8_8_8_8_REV)
 static void TAG2(ReadRGBASpan,_SSE2)( GLcontext *ctx,
@@ -475,8 +474,7 @@ static void TAG2(ReadRGBASpan,_SSE2)( GLcontext *ctx,
 }
 #endif
 
-#if defined(GET_PTR) &&	\
-   defined(USE_SSE_ASM) && \
+#if defined(USE_SSE_ASM) && \
    (SPANTMP_PIXEL_FMT == GL_BGRA) && \
      (SPANTMP_PIXEL_TYPE == GL_UNSIGNED_INT_8_8_8_8_REV)
 static void TAG2(ReadRGBASpan,_SSE)( GLcontext *ctx,
@@ -569,7 +567,6 @@ static void TAG(InitPointers)(struct gl_renderbuffer *rb)
    rb->PutMonoValues = TAG(WriteMonoRGBAPixels);
    rb->GetValues = TAG(ReadRGBAPixels);
 
-#if defined(GET_PTR)
 #if defined(USE_SSE_ASM) && \
    (SPANTMP_PIXEL_FMT == GL_BGRA) && \
      (SPANTMP_PIXEL_TYPE == GL_UNSIGNED_INT_8_8_8_8_REV)
@@ -599,7 +596,6 @@ static void TAG(InitPointers)(struct gl_renderbuffer *rb)
    }
    else
 #endif
-#endif /* GET_PTR */
    {
       if (DBG) fprintf( stderr, "Using %s version of GetRow\n", "C" );
       rb->GetRow = TAG(ReadRGBASpan);
@@ -614,8 +610,6 @@ static void TAG(InitPointers)(struct gl_renderbuffer *rb)
 #undef READ_RGBA
 #undef TAG
 #undef TAG2
-#undef GET_VALUE
-#undef PUT_VALUE
 #undef GET_PTR
 #undef SPANTMP_PIXEL_FMT
 #undef SPANTMP_PIXEL_TYPE
diff --git a/src/mesa/drivers/dri/i915/Makefile b/src/mesa/drivers/dri/i915/Makefile
index fd77980ebf..bdd3b85260 100644
--- a/src/mesa/drivers/dri/i915/Makefile
+++ b/src/mesa/drivers/dri/i915/Makefile
@@ -30,8 +30,8 @@ DRIVER_SOURCES = \
 	intel_pixel.c \
 	intel_pixel_bitmap.c \
 	intel_pixel_copy.c \
-	intel_pixel_draw.c \
 	intel_pixel_read.c \
+	intel_pixel_draw.c \
 	intel_buffers.c \
 	intel_blit.c \
 	i915_tex.c \
@@ -52,10 +52,12 @@ DRIVER_SOURCES = \
 	intel_state.c \
 	intel_tris.c \
 	intel_fbo.c \
-	intel_depthstencil.c
+	intel_depthstencil.c \
+	intel_bufmgr_ttm.c
 
 C_SOURCES = \
 	$(COMMON_SOURCES) \
+	$(COMMON_BM_SOURCES) \
 	$(DRIVER_SOURCES)
 
 ASM_SOURCES = 
@@ -67,7 +69,6 @@ DRIVER_DEFINES = -I../intel -I../intel/server -DI915 \
 include ../Makefile.template
 
 intel_decode.o: ../intel/intel_decode.c
-
 intel_tex_layout.o: ../intel/intel_tex_layout.c
 
 symlinks:
diff --git a/src/mesa/drivers/dri/i915/i830_context.c b/src/mesa/drivers/dri/i915/i830_context.c
index 116b52511f..16c8a8d24f 100644
--- a/src/mesa/drivers/dri/i915/i830_context.c
+++ b/src/mesa/drivers/dri/i915/i830_context.c
@@ -81,9 +81,6 @@ i830CreateContext(const __GLcontextModes * mesaVis,
    _tnl_destroy_pipeline(ctx);
    _tnl_install_pipeline(ctx, intel_pipeline);
 
-   if (intel->no_rast)
-      FALLBACK(intel, INTEL_FALLBACK_USER, 1);
-
    intel->ctx.Const.MaxTextureUnits = I830_TEX_UNITS;
    intel->ctx.Const.MaxTextureImageUnits = I830_TEX_UNITS;
    intel->ctx.Const.MaxTextureCoordUnits = I830_TEX_UNITS;
diff --git a/src/mesa/drivers/dri/i915/i830_reg.h b/src/mesa/drivers/dri/i915/i830_reg.h
index d210c2d08e..41280bca7c 100644
--- a/src/mesa/drivers/dri/i915/i830_reg.h
+++ b/src/mesa/drivers/dri/i915/i830_reg.h
@@ -494,6 +494,10 @@
 #define VFT1_TEX0_FMT(x)	(x)
 #define VFT1_TEX0_MASK          3
 #define VFT1_TEX1_SHIFT         2
+#define TEXCOORDFMT_2D		0
+#define TEXCOORDFMT_3D		1
+#define TEXCOORDFMT_4D		2
+#define TEXCOORDFMT_1D		3
 
 /*New stuff picked up along the way */
 
@@ -631,4 +635,8 @@
 #define ENABLE_TEX_STREAM_MAP_IDX	(1<<3)
 #define TEX_STREAM_MAP_IDX(x)		(x)
 
+
+#define MI_FLUSH           ((0<<29)|(4<<23))
+#define FLUSH_MAP_CACHE    (1<<0)
+
 #endif
diff --git a/src/mesa/drivers/dri/i915/i830_vtbl.c b/src/mesa/drivers/dri/i915/i830_vtbl.c
index 40a50ff772..c5a85fe035 100644
--- a/src/mesa/drivers/dri/i915/i830_vtbl.c
+++ b/src/mesa/drivers/dri/i915/i830_vtbl.c
@@ -31,7 +31,6 @@
 #include "i830_reg.h"
 #include "intel_batchbuffer.h"
 #include "intel_regions.h"
-#include "intel_tris.h"
 #include "tnl/t_context.h"
 #include "tnl/t_vertex.h"
 
@@ -420,12 +419,10 @@ i830_emit_state(struct intel_context *intel)
 {
    struct i830_context *i830 = i830_context(&intel->ctx);
    struct i830_hw_state *state = i830->current;
-   int i, count;
+   int i, ret, count;
    GLuint dirty;
    GET_CURRENT_CONTEXT(ctx);
    BATCH_LOCALS;
-   dri_bo *aper_array[3 + I830_TEX_UNITS];
-   int aper_count;
 
    /* We don't hold the lock at this point, so want to make sure that
     * there won't be a buffer wrap between the state emits and the primitive
@@ -438,28 +435,26 @@ i830_emit_state(struct intel_context *intel)
     * Set the space as LOOP_CLIPRECTS now, since that's what our primitives
     * will be emitted under.
     */
-   intel_batchbuffer_require_space(intel->batch,
-				   get_state_size(state) + INTEL_PRIM_EMIT_SIZE,
+   intel_batchbuffer_require_space(intel->batch, get_state_size(state) + 8,
 				   LOOP_CLIPRECTS);
    count = 0;
  again:
-   aper_count = 0;
    dirty = get_dirty(state);
 
-   aper_array[aper_count++] = intel->batch->buf;
+   ret = 0;
    if (dirty & I830_UPLOAD_BUFFERS) {
-      aper_array[aper_count++] = state->draw_region->buffer;
-      aper_array[aper_count++] = state->depth_region->buffer;
+     ret |= dri_bufmgr_check_aperture_space(state->draw_region->buffer);
+     ret |= dri_bufmgr_check_aperture_space(state->depth_region->buffer);
    }
-
+   
    for (i = 0; i < I830_TEX_UNITS; i++)
      if (dirty & I830_UPLOAD_TEX(i)) {
 	if (state->tex_buffer[i]) {
-	   aper_array[aper_count++] = state->tex_buffer[i];
+	  ret |= dri_bufmgr_check_aperture_space(state->tex_buffer[i]);
 	}
      }
 
-   if (dri_bufmgr_check_aperture_space(aper_array, aper_count)) {
+   if (ret) {
        if (count == 0) {
 	   count++;
 	   intel_batchbuffer_flush(intel->batch);
@@ -495,14 +490,14 @@ i830_emit_state(struct intel_context *intel)
       OUT_BATCH(state->Buffer[I830_DESTREG_CBUFADDR0]);
       OUT_BATCH(state->Buffer[I830_DESTREG_CBUFADDR1]);
       OUT_RELOC(state->draw_region->buffer,
-		I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
+                DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
                 state->draw_region->draw_offset);
 
       if (state->depth_region) {
          OUT_BATCH(state->Buffer[I830_DESTREG_DBUFADDR0]);
          OUT_BATCH(state->Buffer[I830_DESTREG_DBUFADDR1]);
          OUT_RELOC(state->depth_region->buffer,
-		   I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
+                   DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
                    state->depth_region->draw_offset);
       }
 
@@ -529,7 +524,7 @@ i830_emit_state(struct intel_context *intel)
 
          if (state->tex_buffer[i]) {
             OUT_RELOC(state->tex_buffer[i],
-		      I915_GEM_DOMAIN_SAMPLER, 0,
+                      DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
                       state->tex_offset[i] | TM0S0_USE_FENCE);
          }
 	 else if (state == &i830->meta) {
@@ -722,5 +717,4 @@ i830InitVtbl(struct i830_context *i830)
    i830->intel.vtbl.render_prevalidate = i830_render_prevalidate;
    i830->intel.vtbl.assert_not_dirty = i830_assert_not_dirty;
    i830->intel.vtbl.note_unlock = i830_note_unlock; 
-   i830->intel.vtbl.finish_batch = intel_finish_vb;
 }
diff --git a/src/mesa/drivers/dri/i915/i915_context.c b/src/mesa/drivers/dri/i915/i915_context.c
index 940b02513d..bd9f1d5f5d 100644
--- a/src/mesa/drivers/dri/i915/i915_context.c
+++ b/src/mesa/drivers/dri/i915/i915_context.c
@@ -138,9 +138,6 @@ i915CreateContext(const __GLcontextModes * mesaVis,
    _tnl_destroy_pipeline(ctx);
    _tnl_install_pipeline(ctx, intel_pipeline);
 
-   if (intel->no_rast)
-      FALLBACK(intel, INTEL_FALLBACK_USER, 1);
-
    ctx->Const.MaxTextureUnits = I915_TEX_UNITS;
    ctx->Const.MaxTextureImageUnits = I915_TEX_UNITS;
    ctx->Const.MaxTextureCoordUnits = I915_TEX_UNITS;
diff --git a/src/mesa/drivers/dri/i915/i915_reg.h b/src/mesa/drivers/dri/i915/i915_reg.h
index 8891e11c6f..b5585e70e7 100644
--- a/src/mesa/drivers/dri/i915/i915_reg.h
+++ b/src/mesa/drivers/dri/i915/i915_reg.h
@@ -325,6 +325,118 @@
 #define SCISSOR_RECT_0_YMAX(x)		((x)<<16)
 #define SCISSOR_RECT_0_XMAX(x)		(x)
 
+/* p189 */
+#define _3DSTATE_LOAD_STATE_IMMEDIATE_1   ((0x3<<29)|(0x1d<<24)|(0x04<<16))
+#define I1_LOAD_S(n)                      (1<<(4+n))
+
+#define S0_VB_OFFSET_MASK              0xffffffc
+#define S0_AUTO_CACHE_INV_DISABLE      (1<<0)
+
+#define S1_VERTEX_WIDTH_SHIFT          24
+#define S1_VERTEX_WIDTH_MASK           (0x3f<<24)
+#define S1_VERTEX_PITCH_SHIFT          16
+#define S1_VERTEX_PITCH_MASK           (0x3f<<16)
+
+#define TEXCOORDFMT_2D                 0x0
+#define TEXCOORDFMT_3D                 0x1
+#define TEXCOORDFMT_4D                 0x2
+#define TEXCOORDFMT_1D                 0x3
+#define TEXCOORDFMT_2D_16              0x4
+#define TEXCOORDFMT_4D_16              0x5
+#define TEXCOORDFMT_NOT_PRESENT        0xf
+#define S2_TEXCOORD_FMT0_MASK            0xf
+#define S2_TEXCOORD_FMT1_SHIFT           4
+#define S2_TEXCOORD_FMT(unit, type)    ((type)<<(unit*4))
+#define S2_TEXCOORD_NONE               (~0)
+
+/* S3 not interesting */
+
+#define S4_POINT_WIDTH_SHIFT           23
+#define S4_POINT_WIDTH_MASK            (0x1ff<<23)
+#define S4_LINE_WIDTH_SHIFT            19
+#define S4_LINE_WIDTH_ONE              (0x2<<19)
+#define S4_LINE_WIDTH_MASK             (0xf<<19)
+#define S4_FLATSHADE_ALPHA             (1<<18)
+#define S4_FLATSHADE_FOG               (1<<17)
+#define S4_FLATSHADE_SPECULAR          (1<<16)
+#define S4_FLATSHADE_COLOR             (1<<15)
+#define S4_CULLMODE_BOTH	       (0<<13)
+#define S4_CULLMODE_NONE	       (1<<13)
+#define S4_CULLMODE_CW		       (2<<13)
+#define S4_CULLMODE_CCW		       (3<<13)
+#define S4_CULLMODE_MASK	       (3<<13)
+#define S4_VFMT_POINT_WIDTH            (1<<12)
+#define S4_VFMT_SPEC_FOG               (1<<11)
+#define S4_VFMT_COLOR                  (1<<10)
+#define S4_VFMT_DEPTH_OFFSET           (1<<9)
+#define S4_VFMT_XYZ     	       (1<<6)
+#define S4_VFMT_XYZW     	       (2<<6)
+#define S4_VFMT_XY     		       (3<<6)
+#define S4_VFMT_XYW     	       (4<<6)
+#define S4_VFMT_XYZW_MASK              (7<<6)
+#define S4_FORCE_DEFAULT_DIFFUSE       (1<<5)
+#define S4_FORCE_DEFAULT_SPECULAR      (1<<4)
+#define S4_LOCAL_DEPTH_OFFSET_ENABLE   (1<<3)
+#define S4_VFMT_FOG_PARAM              (1<<2)
+#define S4_SPRITE_POINT_ENABLE         (1<<1)
+#define S4_LINE_ANTIALIAS_ENABLE       (1<<0)
+
+#define S4_VFMT_MASK (S4_VFMT_POINT_WIDTH   | 	\
+		      S4_VFMT_SPEC_FOG      |	\
+		      S4_VFMT_COLOR         |	\
+		      S4_VFMT_DEPTH_OFFSET  |	\
+		      S4_VFMT_XYZW_MASK     |	\
+		      S4_VFMT_FOG_PARAM)
+
+
+#define S5_WRITEDISABLE_ALPHA          (1<<31)
+#define S5_WRITEDISABLE_RED            (1<<30)
+#define S5_WRITEDISABLE_GREEN          (1<<29)
+#define S5_WRITEDISABLE_BLUE           (1<<28)
+#define S5_WRITEDISABLE_MASK           (0xf<<28)
+#define S5_FORCE_DEFAULT_POINT_SIZE    (1<<27)
+#define S5_LAST_PIXEL_ENABLE           (1<<26)
+#define S5_GLOBAL_DEPTH_OFFSET_ENABLE  (1<<25)
+#define S5_FOG_ENABLE                  (1<<24)
+#define S5_STENCIL_REF_SHIFT           16
+#define S5_STENCIL_REF_MASK            (0xff<<16)
+#define S5_STENCIL_TEST_FUNC_SHIFT     13
+#define S5_STENCIL_TEST_FUNC_MASK      (0x7<<13)
+#define S5_STENCIL_FAIL_SHIFT          10
+#define S5_STENCIL_FAIL_MASK           (0x7<<10)
+#define S5_STENCIL_PASS_Z_FAIL_SHIFT   7
+#define S5_STENCIL_PASS_Z_FAIL_MASK    (0x7<<7)
+#define S5_STENCIL_PASS_Z_PASS_SHIFT   4
+#define S5_STENCIL_PASS_Z_PASS_MASK    (0x7<<4)
+#define S5_STENCIL_WRITE_ENABLE        (1<<3)
+#define S5_STENCIL_TEST_ENABLE         (1<<2)
+#define S5_COLOR_DITHER_ENABLE         (1<<1)
+#define S5_LOGICOP_ENABLE              (1<<0)
+
+
+#define S6_ALPHA_TEST_ENABLE           (1<<31)
+#define S6_ALPHA_TEST_FUNC_SHIFT       28
+#define S6_ALPHA_TEST_FUNC_MASK        (0x7<<28)
+#define S6_ALPHA_REF_SHIFT             20
+#define S6_ALPHA_REF_MASK              (0xff<<20)
+#define S6_DEPTH_TEST_ENABLE           (1<<19)
+#define S6_DEPTH_TEST_FUNC_SHIFT       16
+#define S6_DEPTH_TEST_FUNC_MASK        (0x7<<16)
+#define S6_CBUF_BLEND_ENABLE           (1<<15)
+#define S6_CBUF_BLEND_FUNC_SHIFT       12
+#define S6_CBUF_BLEND_FUNC_MASK        (0x7<<12)
+#define S6_CBUF_SRC_BLEND_FACT_SHIFT   8
+#define S6_CBUF_SRC_BLEND_FACT_MASK    (0xf<<8)
+#define S6_CBUF_DST_BLEND_FACT_SHIFT   4
+#define S6_CBUF_DST_BLEND_FACT_MASK    (0xf<<4)
+#define S6_DEPTH_WRITE_ENABLE          (1<<3)
+#define S6_COLOR_WRITE_ENABLE          (1<<2)
+#define S6_TRISTRIP_PV_SHIFT           0
+#define S6_TRISTRIP_PV_MASK            (0x3<<0)
+
+#define S7_DEPTH_OFFSET_CONST_MASK     ~0
+
+
 /* Helper macros for blend factors
  */
 #define DST_BLND_FACT(f) ((f)<<S6_CBUF_DST_BLEND_FACT_SHIFT)
@@ -743,4 +855,10 @@
 #define _3DSTATE_DEFAULT_DIFFUSE    ((0x3<<29)|(0x1d<<24)|(0x99<<16))
 #define _3DSTATE_DEFAULT_SPECULAR   ((0x3<<29)|(0x1d<<24)|(0x9a<<16))
 
+
+#define MI_FLUSH                   ((0<<29)|(4<<23))
+#define FLUSH_MAP_CACHE            (1<<0)
+#define INHIBIT_FLUSH_RENDER_CACHE (1<<2)
+
+
 #endif
diff --git a/src/mesa/drivers/dri/i915/i915_vtbl.c b/src/mesa/drivers/dri/i915/i915_vtbl.c
index 19f2206285..135bfaa265 100644
--- a/src/mesa/drivers/dri/i915/i915_vtbl.c
+++ b/src/mesa/drivers/dri/i915/i915_vtbl.c
@@ -39,7 +39,6 @@
 #include "intel_batchbuffer.h"
 #include "intel_tex.h"
 #include "intel_regions.h"
-#include "intel_tris.h"
 
 #include "i915_reg.h"
 #include "i915_context.h"
@@ -297,9 +296,9 @@ i915_emit_state(struct intel_context *intel)
 {
    struct i915_context *i915 = i915_context(&intel->ctx);
    struct i915_hw_state *state = i915->current;
-   int i, count, aper_count;
+   int i;
+   int ret, count;
    GLuint dirty;
-   dri_bo *aper_array[3 + I915_TEX_UNITS];
    GET_CURRENT_CONTEXT(ctx);
    BATCH_LOCALS;
 
@@ -314,32 +313,28 @@ i915_emit_state(struct intel_context *intel)
     * Set the space as LOOP_CLIPRECTS now, since that's what our primitives
     * will be emitted under.
     */
-   intel_batchbuffer_require_space(intel->batch,
-				   get_state_size(state) + INTEL_PRIM_EMIT_SIZE,
+   intel_batchbuffer_require_space(intel->batch, get_state_size(state) + 8,
 				   LOOP_CLIPRECTS);
    count = 0;
  again:
-   aper_count = 0;
    dirty = get_dirty(state);
 
-   aper_array[aper_count++] = intel->batch->buf;
+   ret = 0;
    if (dirty & I915_UPLOAD_BUFFERS) {
-      aper_array[aper_count++] = state->draw_region->buffer;
-      if (state->depth_region)
-	 aper_array[aper_count++] = state->depth_region->buffer;
+     ret |= dri_bufmgr_check_aperture_space(state->draw_region->buffer);
+     if (state->depth_region)
+        ret |= dri_bufmgr_check_aperture_space(state->depth_region->buffer);
    }
 
    if (dirty & I915_UPLOAD_TEX_ALL) {
-      for (i = 0; i < I915_TEX_UNITS; i++) {
-	 if (dirty & I915_UPLOAD_TEX(i)) {
-	    if (state->tex_buffer[i]) {
-	       aper_array[aper_count++] = state->tex_buffer[i];
-	    }
-	 }
-      }
+     for (i = 0; i < I915_TEX_UNITS; i++)
+       if (dirty & I915_UPLOAD_TEX(i)) {
+	   if (state->tex_buffer[i]) {
+	       ret |= dri_bufmgr_check_aperture_space(state->tex_buffer[i]);
+	   }
+       }
    }
-
-   if (dri_bufmgr_check_aperture_space(aper_array, aper_count)) {
+   if (ret) {
        if (count == 0) {
 	   count++;
 	   intel_batchbuffer_flush(intel->batch);
@@ -382,14 +377,14 @@ i915_emit_state(struct intel_context *intel)
       OUT_BATCH(state->Buffer[I915_DESTREG_CBUFADDR0]);
       OUT_BATCH(state->Buffer[I915_DESTREG_CBUFADDR1]);
       OUT_RELOC(state->draw_region->buffer,
-		I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
+                DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
                 state->draw_region->draw_offset);
 
       if (state->depth_region) {
          OUT_BATCH(state->Buffer[I915_DESTREG_DBUFADDR0]);
          OUT_BATCH(state->Buffer[I915_DESTREG_DBUFADDR1]);
          OUT_RELOC(state->depth_region->buffer,
-		   I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
+                   DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
                    state->depth_region->draw_offset);
       }
 
@@ -432,7 +427,7 @@ i915_emit_state(struct intel_context *intel)
 
             if (state->tex_buffer[i]) {
                OUT_RELOC(state->tex_buffer[i],
-			 I915_GEM_DOMAIN_SAMPLER, 0,
+                         DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
                          state->tex_offset[i]);
             }
             else if (state == &i915->meta) {
@@ -634,5 +629,4 @@ i915InitVtbl(struct i915_context *i915)
    i915->intel.vtbl.flush_cmd = i915_flush_cmd;
    i915->intel.vtbl.assert_not_dirty = i915_assert_not_dirty;
    i915->intel.vtbl.note_unlock = i915_note_unlock; 
-   i915->intel.vtbl.finish_batch = intel_finish_vb;
 }
diff --git a/src/mesa/drivers/dri/i915/intel_bufmgr_ttm.c b/src/mesa/drivers/dri/i915/intel_bufmgr_ttm.c
new file mode 120000
index 0000000000..e9df5c6279
--- /dev/null
+++ b/src/mesa/drivers/dri/i915/intel_bufmgr_ttm.c
@@ -0,0 +1 @@
+../intel/intel_bufmgr_ttm.c
\ No newline at end of file
diff --git a/src/mesa/drivers/dri/i915/intel_pixel_read.c b/src/mesa/drivers/dri/i915/intel_pixel_read.c
index d009590a4b..2e31656e98 100644
--- a/src/mesa/drivers/dri/i915/intel_pixel_read.c
+++ b/src/mesa/drivers/dri/i915/intel_pixel_read.c
@@ -173,6 +173,7 @@ do_blit_readpixels(GLcontext * ctx,
    struct intel_buffer_object *dst = intel_buffer_object(pack->BufferObj);
    GLuint dst_offset;
    GLuint rowLength;
+   dri_fence *fence = NULL;
 
    if (INTEL_DEBUG & DEBUG_PIXEL)
       _mesa_printf("%s\n", __FUNCTION__);
@@ -263,7 +264,7 @@ do_blit_readpixels(GLcontext * ctx,
 
          intelEmitCopyBlit(intel,
                            src->cpp,
-                           src->pitch, src->buffer, 0, src->tiling,
+                           src->pitch, src->buffer, 0, src->tiled,
                            rowLength, dst_buffer, dst_offset, GL_FALSE,
                            rect.x1,
                            rect.y1,
@@ -272,9 +273,19 @@ do_blit_readpixels(GLcontext * ctx,
                            rect.x2 - rect.x1, rect.y2 - rect.y1,
 			   GL_COPY);
       }
+
+      intel_batchbuffer_flush(intel->batch);
+      fence = intel->batch->last_fence;
+      dri_fence_reference(fence);
+
    }
    UNLOCK_HARDWARE(intel);
 
+   if (fence) {
+      dri_fence_wait(fence);
+      dri_fence_unreference(fence);
+   }
+
    if (INTEL_DEBUG & DEBUG_PIXEL)
       _mesa_printf("%s - DONE\n", __FUNCTION__);
 
diff --git a/src/mesa/drivers/dri/i915/intel_render.c b/src/mesa/drivers/dri/i915/intel_render.c
index 838d450378..5e6500cfa1 100644
--- a/src/mesa/drivers/dri/i915/intel_render.c
+++ b/src/mesa/drivers/dri/i915/intel_render.c
@@ -67,7 +67,7 @@
 
 #define HAVE_ELTS        0
 
-static uint32_t hw_prim[GL_POLYGON + 1] = {
+static GLuint hw_prim[GL_POLYGON + 1] = {
    0,
    PRIM3D_LINELIST,
    PRIM3D_LINESTRIP,
@@ -114,7 +114,7 @@ intelDmaPrimitive(struct intel_context *intel, GLenum prim)
       fprintf(stderr, "%s %s\n", __FUNCTION__, _mesa_lookup_enum_by_nr(prim));
    INTEL_FIREVERTICES(intel);
    intel->vtbl.reduced_primitive_state(intel, reduced_prim[prim]);
-   intel_set_prim(intel, hw_prim[prim]);
+   intelStartInlinePrimitive(intel, hw_prim[prim], LOOP_CLIPRECTS);
 }
 
 
@@ -126,11 +126,12 @@ do {						\
 
 #define FLUSH() INTEL_FIREVERTICES(intel)
 
-#define GET_SUBSEQUENT_VB_MAX_VERTS() (INTEL_VB_SIZE / (intel->vertex_size * 4))
-#define GET_CURRENT_VB_MAX_VERTS() \
-   ((INTEL_VB_SIZE - intel->prim.current_offset) / (intel->vertex_size * 4))
+#define GET_SUBSEQUENT_VB_MAX_VERTS() \
+  ((intel->batch->size - 1500) / (intel->vertex_size*4))
+#define GET_CURRENT_VB_MAX_VERTS() GET_SUBSEQUENT_VB_MAX_VERTS()
 
-#define ALLOC_VERTS(nr) intel_get_prim_space(intel, nr)
+#define ALLOC_VERTS( nr ) \
+   intelExtendInlinePrimitive( intel, (nr) * intel->vertex_size )
 
 #define EMIT_VERTS( ctx, j, nr, buf ) \
   _tnl_emit_vertices_to_buffer(ctx, j, (j)+(nr), buf )
diff --git a/src/mesa/drivers/dri/i915/intel_tris.c b/src/mesa/drivers/dri/i915/intel_tris.c
index 5f4b852212..bbb4e0f3cd 100644
--- a/src/mesa/drivers/dri/i915/intel_tris.c
+++ b/src/mesa/drivers/dri/i915/intel_tris.c
@@ -25,12 +25,6 @@
  * 
  **************************************************************************/
 
-/** @file intel_tris.c
- *
- * This file contains functions for managing the vertex buffer and emitting
- * primitives into it.
- */
-
 #include "glheader.h"
 #include "context.h"
 #include "macros.h"
@@ -53,185 +47,111 @@
 #include "intel_reg.h"
 #include "intel_span.h"
 #include "intel_tex.h"
-#include "intel_chipset.h"
-#include "i830_context.h"
-#include "i830_reg.h"
 
 static void intelRenderPrimitive(GLcontext * ctx, GLenum prim);
 static void intelRasterPrimitive(GLcontext * ctx, GLenum rprim,
                                  GLuint hwprim);
 
-/** Sets the primitive type for a primitive sequence, flushing as needed. */
-void intel_set_prim(struct intel_context *intel, uint32_t prim)
+/*
+ */
+static void
+intel_flush_inline_primitive(struct intel_context *intel)
 {
-   if (prim != intel->prim.primitive) {
-      INTEL_FIREVERTICES(intel);
-      intel->prim.primitive = prim;
-   }
-}
+   GLuint used = intel->batch->ptr - intel->prim.start_ptr;
 
-/** Returns mapped VB space for the given number of vertices */
-uint32_t *intel_get_prim_space(struct intel_context *intel, unsigned int count)
-{
-   uint32_t *addr;
-
-   /* Check for space in the existing VB */
-   if (intel->prim.vb_bo == NULL ||
-       (intel->prim.current_offset +
-	count * intel->vertex_size * 4) > INTEL_VB_SIZE ||
-       (intel->prim.count + count) >= (1 << 16)) {
-      /* Flush existing prim if any */
-      INTEL_FIREVERTICES(intel);
+   assert(intel->prim.primitive != ~0);
 
-      intel_finish_vb(intel);
+/*    _mesa_printf("/\n"); */
 
-      /* Start a new VB */
-      if (intel->prim.vb == NULL)
-	 intel->prim.vb = malloc(INTEL_VB_SIZE);
-      intel->prim.vb_bo = dri_bo_alloc(intel->bufmgr, "vb",
-				       INTEL_VB_SIZE, 4);
-      intel->prim.start_offset = 0;
-      intel->prim.current_offset = 0;
-   }
+   if (used < 8)
+      goto do_discard;
+
+   *(int *) intel->prim.start_ptr = (_3DPRIMITIVE |
+                                     intel->prim.primitive | (used / 4 - 2));
 
-   intel->prim.flush = intel_flush_prim;
+   goto finished;
 
-   addr = (uint32_t *)(intel->prim.vb + intel->prim.current_offset);
-   intel->prim.current_offset += intel->vertex_size * 4 * count;
-   intel->prim.count += count;
+ do_discard:
+   intel->batch->ptr -= used;
 
-   return addr;
+ finished:
+   intel->prim.primitive = ~0;
+   intel->prim.start_ptr = 0;
+   intel->prim.flush = 0;
 }
 
-/** Dispatches the accumulated primitive to the batchbuffer. */
-void intel_flush_prim(struct intel_context *intel)
+
+/* Emit a primitive referencing vertices in a vertex buffer.
+ */
+void
+intelStartInlinePrimitive(struct intel_context *intel,
+                          GLuint prim, GLuint batch_flags)
 {
    BATCH_LOCALS;
-   dri_bo *aper_array[2];
-   dri_bo *vb_bo;
-
-   /* Must be called after an intel_start_prim. */
-   assert(intel->prim.primitive != ~0);
-
-   if (intel->prim.count == 0)
-      return;
-
-   /* Keep a reference on the BO as it may get finished as we start the
-    * batch emit.
-    */
-   vb_bo = intel->prim.vb_bo;
-   dri_bo_reference(vb_bo);
 
    intel_wait_flips(intel);
 
    intel->vtbl.emit_state(intel);
 
-   aper_array[0] = intel->batch->buf;
-   aper_array[1] = vb_bo;
-   if (dri_bufmgr_check_aperture_space(aper_array, 2)) {
-      intel_batchbuffer_flush(intel->batch);
-      intel->vtbl.emit_state(intel);
-   }
-
-   /* Ensure that we don't start a new batch for the following emit, which
-    * depends on the state just emitted. emit_state should be making sure we
-    * have the space for this.
-    */
    intel->no_batch_wrap = GL_TRUE;
 
-   /* Check that we actually emitted the state into this batch, using the
-    * UPLOAD_CTX bit as the signal.
+/*    _mesa_printf("%s *", __progname); */
+
+   /* Emit a slot which will be filled with the inline primitive
+    * command later.
     */
+   BEGIN_BATCH(2, batch_flags);
+   OUT_BATCH(0);
+
    assert((intel->batch->dirty_state & (1<<1)) == 0);
 
-#if 0
-   printf("emitting %d..%d=%d vertices size %d\n", intel->prim.start_offset,
-	  intel->prim.current_offset, intel->prim.count,
-	  intel->vertex_size * 4);
-#endif
+   intel->prim.start_ptr = intel->batch->ptr;
+   intel->prim.primitive = prim;
+   intel->prim.flush = intel_flush_inline_primitive;
 
-   if (IS_9XX(intel->intelScreen->deviceID)) {
-      BEGIN_BATCH(5, LOOP_CLIPRECTS);
-      OUT_BATCH(_3DSTATE_LOAD_STATE_IMMEDIATE_1 |
-		I1_LOAD_S(0) | I1_LOAD_S(1) | 1);
-      assert((intel->prim.start_offset & !S0_VB_OFFSET_MASK) == 0);
-      OUT_RELOC(vb_bo, I915_GEM_DOMAIN_VERTEX, 0,
-		intel->prim.start_offset);
-      OUT_BATCH((intel->vertex_size << S1_VERTEX_WIDTH_SHIFT) |
-		(intel->vertex_size << S1_VERTEX_PITCH_SHIFT));
-
-      OUT_BATCH(_3DPRIMITIVE |
-		PRIM_INDIRECT |
-		PRIM_INDIRECT_SEQUENTIAL |
-		intel->prim.primitive |
-		intel->prim.count);
-      OUT_BATCH(0); /* Beginning vertex index */
-      ADVANCE_BATCH();
-   } else {
-      struct i830_context *i830 = i830_context(&intel->ctx);
-
-      BEGIN_BATCH(5, LOOP_CLIPRECTS);
-      OUT_BATCH(_3DSTATE_LOAD_STATE_IMMEDIATE_1 |
-		I1_LOAD_S(0) | I1_LOAD_S(2) | 1);
-      /* S0 */
-      assert((intel->prim.start_offset & !S0_VB_OFFSET_MASK_830) == 0);
-      OUT_RELOC(vb_bo, I915_GEM_DOMAIN_VERTEX, 0,
-		intel->prim.start_offset |
-		(intel->vertex_size << S0_VB_PITCH_SHIFT_830) |
-		S0_VB_ENABLE_830);
-      /* S1
-       * This is somewhat unfortunate -- VB width is tied up with
-       * vertex format data that we've already uploaded through
-       * _3DSTATE_VFT[01]_CMD.  We may want to replace emits of VFT state with
-       * STATE_IMMEDIATE_1 like this to avoid duplication.
-       */
-      OUT_BATCH((i830->state.Ctx[I830_CTXREG_VF] & VFT0_TEX_COUNT_MASK) >>
-		VFT0_TEX_COUNT_SHIFT << S2_TEX_COUNT_SHIFT_830 |
-		(i830->state.Ctx[I830_CTXREG_VF2] << 16) |
-		intel->vertex_size << S2_VERTEX_0_WIDTH_SHIFT_830);
-
-      OUT_BATCH(_3DPRIMITIVE |
-		PRIM_INDIRECT |
-		PRIM_INDIRECT_SEQUENTIAL |
-		intel->prim.primitive |
-		intel->prim.count);
-      OUT_BATCH(0); /* Beginning vertex index */
-      ADVANCE_BATCH();
-   }
+   OUT_BATCH(0);
+   ADVANCE_BATCH();
 
    intel->no_batch_wrap = GL_FALSE;
 
-   intel->prim.flush = NULL;
-   intel->prim.start_offset = intel->prim.current_offset;
-   if (!IS_9XX(intel->intelScreen->deviceID))
-      intel->prim.start_offset = ALIGN(intel->prim.start_offset, 128);
-   intel->prim.count = 0;
+/*    _mesa_printf(">"); */
+}
+
+
+void
+intelWrapInlinePrimitive(struct intel_context *intel)
+{
+   GLuint prim = intel->prim.primitive;
+   enum cliprect_mode cliprect_mode = intel->batch->cliprect_mode;
 
-   dri_bo_unreference(vb_bo);
+   intel_flush_inline_primitive(intel);
+   intel_batchbuffer_flush(intel->batch);
+   intelStartInlinePrimitive(intel, prim, cliprect_mode);  /* ??? */
 }
 
-/**
- * Uploads the locally-accumulated VB into the buffer object.
- *
- * This avoids us thrashing the cachelines in and out as the buffer gets
- * filled, dispatched, then reused as the hardware completes rendering from it,
- * and also lets us clflush less if we dispatch with a partially-filled VB.
- *
- * This is called normally from get_space when we're finishing a BO, but also
- * at batch flush time so that we don't try accessing the contents of a
- * just-dispatched buffer.
- */
-void intel_finish_vb(struct intel_context *intel)
+GLuint *
+intelExtendInlinePrimitive(struct intel_context *intel, GLuint dwords)
 {
-   if (intel->prim.vb_bo == NULL)
-      return;
+   GLuint sz = dwords * sizeof(GLuint);
+   GLuint *ptr;
+
+   assert(intel->prim.flush == intel_flush_inline_primitive);
+
+   if (intel_batchbuffer_space(intel->batch) < sz)
+      intelWrapInlinePrimitive(intel);
+
+/*    _mesa_printf("."); */
+
+   intel->vtbl.assert_not_dirty(intel);
 
-   dri_bo_subdata(intel->prim.vb_bo, 0, intel->prim.start_offset,
-		  intel->prim.vb);
-   dri_bo_unreference(intel->prim.vb_bo);
-   intel->prim.vb_bo = NULL;
+   ptr = (GLuint *) intel->batch->ptr;
+   intel->batch->ptr += sz;
+
+   return ptr;
 }
 
+
+
 /***********************************************************************
  *                    Emit primitives as inline vertices               *
  ***********************************************************************/
@@ -262,7 +182,7 @@ intel_draw_quad(struct intel_context *intel,
                 intelVertexPtr v1, intelVertexPtr v2, intelVertexPtr v3)
 {
    GLuint vertsize = intel->vertex_size;
-   GLuint *vb = intel_get_prim_space(intel, 6);
+   GLuint *vb = intelExtendInlinePrimitive(intel, 6 * vertsize);
    int j;
 
    COPY_DWORDS(j, vb, vertsize, v0);
@@ -290,7 +210,7 @@ intel_draw_triangle(struct intel_context *intel,
                     intelVertexPtr v0, intelVertexPtr v1, intelVertexPtr v2)
 {
    GLuint vertsize = intel->vertex_size;
-   GLuint *vb = intel_get_prim_space(intel, 3);
+   GLuint *vb = intelExtendInlinePrimitive(intel, 3 * vertsize);
    int j;
 
    COPY_DWORDS(j, vb, vertsize, v0);
@@ -304,7 +224,7 @@ intel_draw_line(struct intel_context *intel,
                 intelVertexPtr v0, intelVertexPtr v1)
 {
    GLuint vertsize = intel->vertex_size;
-   GLuint *vb = intel_get_prim_space(intel, 2);
+   GLuint *vb = intelExtendInlinePrimitive(intel, 2 * vertsize);
    int j;
 
    COPY_DWORDS(j, vb, vertsize, v0);
@@ -316,7 +236,7 @@ static void
 intel_draw_point(struct intel_context *intel, intelVertexPtr v0)
 {
    GLuint vertsize = intel->vertex_size;
-   GLuint *vb = intel_get_prim_space(intel, 1);
+   GLuint *vb = intelExtendInlinePrimitive(intel, vertsize);
    int j;
 
    /* Adjust for sub pixel position -- still required for conform. */
@@ -825,7 +745,7 @@ intelFastRenderClippedPoly(GLcontext * ctx, const GLuint * elts, GLuint n)
 {
    struct intel_context *intel = intel_context(ctx);
    const GLuint vertsize = intel->vertex_size;
-   GLuint *vb = intel_get_prim_space(intel, (n - 2) * 3);
+   GLuint *vb = intelExtendInlinePrimitive(intel, (n - 2) * 3 * vertsize);
    GLubyte *vertptr = (GLubyte *) intel->verts;
    const GLuint *start = (const GLuint *) V(elts[0]);
    int i, j;
@@ -1030,7 +950,7 @@ intelRasterPrimitive(GLcontext * ctx, GLenum rprim, GLuint hwprim)
    if (hwprim != intel->prim.primitive) {
       INTEL_FIREVERTICES(intel);
 
-      intel_set_prim(intel, hwprim);
+      intelStartInlinePrimitive(intel, hwprim, LOOP_CLIPRECTS);
    }
 }
 
@@ -1163,18 +1083,15 @@ intel_meta_draw_poly(struct intel_context *intel,
    union fi *vb;
    GLint i;
    GLboolean was_locked = intel->locked;
-   unsigned int saved_vertex_size = intel->vertex_size;
 
    if (!was_locked)
        LOCK_HARDWARE(intel);
 
-   intel->vertex_size = 6;
-
    /* All 3d primitives should be emitted with LOOP_CLIPRECTS,
     * otherwise the drawing origin (DR4) might not be set correctly.
     */
-   intel_set_prim(intel, PRIM3D_TRIFAN);
-   vb = (union fi *) intel_get_prim_space(intel, n);
+   intelStartInlinePrimitive(intel, PRIM3D_TRIFAN, LOOP_CLIPRECTS);
+   vb = (union fi *) intelExtendInlinePrimitive(intel, n * 6);
 
    for (i = 0; i < n; i++) {
       vb[0].f = xy[i][0];
@@ -1188,8 +1105,6 @@ intel_meta_draw_poly(struct intel_context *intel,
 
    INTEL_FIREVERTICES(intel);
 
-   intel->vertex_size = saved_vertex_size;
-
    if (!was_locked)
        UNLOCK_HARDWARE(intel);
 }
diff --git a/src/mesa/drivers/dri/i915/intel_tris.h b/src/mesa/drivers/dri/i915/intel_tris.h
index 0e08986221..021e5c6450 100644
--- a/src/mesa/drivers/dri/i915/intel_tris.h
+++ b/src/mesa/drivers/dri/i915/intel_tris.h
@@ -30,9 +30,7 @@
 
 #include "mtypes.h"
 
-#define INTEL_VB_SIZE		(32 * 1024)
-/** 3 dwords of state_immediate and 2 of 3dprim, in intel_flush_prim */
-#define INTEL_PRIM_EMIT_SIZE	(5 * 4)
+
 
 #define _INTEL_NEW_RENDERSTATE (_DD_NEW_LINE_STIPPLE |		\
 			       _DD_NEW_TRI_UNFILLED |		\
@@ -46,9 +44,11 @@ extern void intelInitTriFuncs(GLcontext * ctx);
 
 extern void intelChooseRenderState(GLcontext * ctx);
 
-void intel_set_prim(struct intel_context *intel, uint32_t prim);
-GLuint *intel_get_prim_space(struct intel_context *intel, unsigned int count);
-void intel_flush_prim(struct intel_context *intel);
-void intel_finish_vb(struct intel_context *intel);
+extern void intelStartInlinePrimitive(struct intel_context *intel,
+                                      GLuint prim, GLuint flags);
+extern void intelWrapInlinePrimitive(struct intel_context *intel);
+
+GLuint *intelExtendInlinePrimitive(struct intel_context *intel,
+                                   GLuint dwords);
 
 #endif
diff --git a/src/mesa/drivers/dri/i965/Makefile b/src/mesa/drivers/dri/i965/Makefile
index 8fa205e79f..c561b744d1 100644
--- a/src/mesa/drivers/dri/i965/Makefile
+++ b/src/mesa/drivers/dri/i965/Makefile
@@ -9,6 +9,7 @@ DRIVER_SOURCES = \
 	intel_blit.c \
 	intel_buffer_objects.c \
 	intel_buffers.c \
+	intel_bufmgr_ttm.c \
 	intel_context.c \
 	intel_decode.c \
 	intel_depthstencil.c \
@@ -84,6 +85,7 @@ DRIVER_SOURCES = \
 
 C_SOURCES = \
 	$(COMMON_SOURCES) \
+	$(COMMON_BM_SOURCES) \
 	$(MINIGLX_SOURCES) \
 	$(DRIVER_SOURCES)
 
diff --git a/src/mesa/drivers/dri/i965/brw_cc.c b/src/mesa/drivers/dri/i965/brw_cc.c
index d662cf7521..9d8984f05c 100644
--- a/src/mesa/drivers/dri/i965/brw_cc.c
+++ b/src/mesa/drivers/dri/i965/brw_cc.c
@@ -37,7 +37,7 @@
 #include "macros.h"
 #include "enums.h"
 
-static void prepare_cc_vp( struct brw_context *brw )
+static int upload_cc_vp( struct brw_context *brw )
 {
    struct brw_cc_viewport ccv;
 
@@ -48,6 +48,7 @@ static void prepare_cc_vp( struct brw_context *brw )
 
    dri_bo_unreference(brw->cc.vp_bo);
    brw->cc.vp_bo = brw_cache_data( &brw->cache, BRW_CC_VP, &ccv, NULL, 0 );
+   return dri_bufmgr_check_aperture_space(brw->cc.vp_bo);
 }
 
 const struct brw_tracked_state brw_cc_vp = {
@@ -56,7 +57,7 @@ const struct brw_tracked_state brw_cc_vp = {
       .brw = BRW_NEW_CONTEXT,
       .cache = 0
    },
-   .prepare = prepare_cc_vp
+   .prepare = upload_cc_vp
 };
 
 struct brw_cc_unit_key {
@@ -255,17 +256,16 @@ cc_unit_create_from_key(struct brw_context *brw, struct brw_cc_unit_key *key)
 			 NULL, NULL);
 
    /* Emit CC viewport relocation */
-   intel_bo_emit_reloc(bo,
-		       I915_GEM_DOMAIN_INSTRUCTION,
-		       0,
-		       0,
-		       offsetof(struct brw_cc_unit_state, cc4),
-		       brw->cc.vp_bo);
+   dri_emit_reloc(bo,
+		  DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
+		  0,
+		  offsetof(struct brw_cc_unit_state, cc4),
+		  brw->cc.vp_bo);
 
    return bo;
 }
 
-static void prepare_cc_unit( struct brw_context *brw )
+static int prepare_cc_unit( struct brw_context *brw )
 {
    struct brw_cc_unit_key key;
 
@@ -279,6 +279,7 @@ static void prepare_cc_unit( struct brw_context *brw )
 
    if (brw->cc.state_bo == NULL)
       brw->cc.state_bo = cc_unit_create_from_key(brw, &key);
+   return dri_bufmgr_check_aperture_space(brw->cc.state_bo);
 }
 
 const struct brw_tracked_state brw_cc_unit = {
diff --git a/src/mesa/drivers/dri/i965/brw_clip.c b/src/mesa/drivers/dri/i965/brw_clip.c
index 22981fd2d9..540108e5f4 100644
--- a/src/mesa/drivers/dri/i965/brw_clip.c
+++ b/src/mesa/drivers/dri/i965/brw_clip.c
@@ -131,7 +131,7 @@ static void compile_clip_prog( struct brw_context *brw,
 
 /* Calculate interpolants for triangle and line rasterization.
  */
-static void upload_clip_prog(struct brw_context *brw)
+static int upload_clip_prog( struct brw_context *brw )
 {
    GLcontext *ctx = &brw->intel.ctx;
    struct brw_clip_prog_key key;
@@ -242,6 +242,8 @@ static void upload_clip_prog(struct brw_context *brw)
 					&brw->clip.prog_data);
    if (brw->clip.prog_bo == NULL)
       compile_clip_prog( brw, &key );
+
+   return dri_bufmgr_check_aperture_space(brw->clip.prog_bo);
 }
 
 
diff --git a/src/mesa/drivers/dri/i965/brw_clip_state.c b/src/mesa/drivers/dri/i965/brw_clip_state.c
index ae904c6253..2d0b24c5ca 100644
--- a/src/mesa/drivers/dri/i965/brw_clip_state.c
+++ b/src/mesa/drivers/dri/i965/brw_clip_state.c
@@ -119,19 +119,19 @@ clip_unit_create_from_key(struct brw_context *brw,
 
    /* Emit clip program relocation */
    assert(brw->clip.prog_bo);
-   intel_bo_emit_reloc(bo,
-		       I915_GEM_DOMAIN_INSTRUCTION,
-		       0,
-		       clip.thread0.grf_reg_count << 1,
-		       offsetof(struct brw_clip_unit_state, thread0),
-		       brw->clip.prog_bo);
+   dri_emit_reloc(bo,
+		  DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
+		  clip.thread0.grf_reg_count << 1,
+		  offsetof(struct brw_clip_unit_state, thread0),
+		  brw->clip.prog_bo);
 
    return bo;
 }
 
-static void upload_clip_unit( struct brw_context *brw )
+static int upload_clip_unit( struct brw_context *brw )
 {
    struct brw_clip_unit_key key;
+   int ret = 0;
 
    clip_unit_populate_key(brw, &key);
 
@@ -143,6 +143,9 @@ static void upload_clip_unit( struct brw_context *brw )
    if (brw->clip.state_bo == NULL) {
       brw->clip.state_bo = clip_unit_create_from_key(brw, &key);
    }
+
+   ret = dri_bufmgr_check_aperture_space(brw->clip.state_bo);
+   return ret;
 }
 
 const struct brw_tracked_state brw_clip_unit = {
diff --git a/src/mesa/drivers/dri/i965/brw_context.c b/src/mesa/drivers/dri/i965/brw_context.c
index b4e2619929..1c7ad5c9c8 100644
--- a/src/mesa/drivers/dri/i965/brw_context.c
+++ b/src/mesa/drivers/dri/i965/brw_context.c
@@ -134,6 +134,7 @@ GLboolean brwCreateContext( const __GLcontextModes *mesaVis,
    ctx->Const.Max3DTextureLevels = 9;
    ctx->Const.MaxCubeTextureLevels = 12;
    ctx->Const.MaxTextureRectSize = (1<<11);
+   ctx->Const.MaxTextureUnits = BRW_MAX_TEX_UNIT;
    
 /*    ctx->Const.MaxNativeVertexProgramTemps = 32; */
 
diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h
index 330d5714da..32e05542e0 100644
--- a/src/mesa/drivers/dri/i965/brw_context.h
+++ b/src/mesa/drivers/dri/i965/brw_context.h
@@ -135,8 +135,6 @@ struct brw_context;
 #define BRW_NEW_METAOPS                 0x1000
 #define BRW_NEW_FENCE                   0x2000
 #define BRW_NEW_LOCK                    0x4000
-#define BRW_NEW_INDICES			0x8000
-#define BRW_NEW_VERTICES		0x10000
 /**
  * Used for any batch entry with a relocated pointer that will be used
  * by any 3D rendering.
@@ -334,7 +332,7 @@ struct brw_state_pointers {
  */
 struct brw_tracked_state {
    struct brw_state_flags dirty;
-   void (*prepare)( struct brw_context *brw );
+   int (*prepare)( struct brw_context *brw );
    void (*emit)( struct brw_context *brw );
 };
 
@@ -452,22 +450,9 @@ struct brw_context
        * for changes to this state:
        */
       struct brw_vertex_info info;
-      unsigned int min_index, max_index;
    } vb;
 
    struct {
-      /**
-       * Index buffer for this draw_prims call.
-       *
-       * Updates are signaled by BRW_NEW_INDICES.
-       */
-      const struct _mesa_index_buffer *ib;
-
-      dri_bo *bo;
-      unsigned int offset;
-   } ib;
-
-   struct {
       /* Will be allocated on demand if needed.   
        */
       struct brw_state_pointers attribs;
@@ -656,7 +641,7 @@ GLboolean brwCreateContext( const __GLcontextModes *mesaVis,
 /*======================================================================
  * brw_state.c
  */
-void brw_validate_state( struct brw_context *brw );
+int brw_validate_state( struct brw_context *brw );
 void brw_init_state( struct brw_context *brw );
 void brw_destroy_state( struct brw_context *brw );
 
diff --git a/src/mesa/drivers/dri/i965/brw_curbe.c b/src/mesa/drivers/dri/i965/brw_curbe.c
index 0a3600193b..5ff4e2964e 100644
--- a/src/mesa/drivers/dri/i965/brw_curbe.c
+++ b/src/mesa/drivers/dri/i965/brw_curbe.c
@@ -46,7 +46,7 @@
 
 /* Partition the CURBE between the various users of constant values:
  */
-static void calculate_curbe_offsets( struct brw_context *brw )
+static int calculate_curbe_offsets( struct brw_context *brw )
 {
    /* CACHE_NEW_WM_PROG */
    GLuint nr_fp_regs = (brw->wm.prog_data->nr_params + 15) / 16;
@@ -117,6 +117,7 @@ static void calculate_curbe_offsets( struct brw_context *brw )
 
       brw->state.dirty.brw |= BRW_NEW_CURBE_OFFSETS;
    }
+   return 0;
 }
 
 
@@ -155,7 +156,19 @@ void brw_upload_constant_buffer_state(struct brw_context *brw)
 
    assert(brw->urb.nr_cs_entries);
    BRW_CACHED_BATCH_STRUCT(brw, &cbs);
-}
+}      
+
+#if 0
+const struct brw_tracked_state brw_constant_buffer_state = {
+   .dirty = {
+      .mesa = 0,
+      .brw = BRW_NEW_URB_FENCE,
+      .cache = 0
+   },
+   .update = brw_upload_constant_buffer_state
+};
+#endif
+
 
 static GLfloat fixed_plane[6][4] = {
    { 0,    0,   -1, 1 },
@@ -170,7 +183,7 @@ static GLfloat fixed_plane[6][4] = {
  * cache mechanism, but maybe would benefit from a comparison against
  * the current uploaded set of constants.
  */
-static void prepare_constant_buffer(struct brw_context *brw)
+static int prepare_constant_buffer(struct brw_context *brw)
 {
    GLcontext *ctx = &brw->intel.ctx;
    struct brw_vertex_program *vp = (struct brw_vertex_program *)brw->vertex_program;
@@ -194,8 +207,8 @@ static void prepare_constant_buffer(struct brw_context *brw)
 	 brw->curbe.last_buf = NULL;
 	 brw->curbe.last_bufsz  = 0;
       }
-
-      return;
+       
+      return 0;
    }
 
    buf = (GLfloat *)malloc(bufsz);
@@ -293,7 +306,10 @@ static void prepare_constant_buffer(struct brw_context *brw)
 	  * They're generally around 64b.
 	  */
 	 brw->curbe.curbe_bo = dri_bo_alloc(brw->intel.bufmgr, "CURBE",
-					    4096, 1 << 6);
+					    4096, 1 << 6,
+					    DRM_BO_FLAG_MEM_LOCAL |
+					    DRM_BO_FLAG_CACHED |
+					    DRM_BO_FLAG_CACHED_MAPPED);
 	 brw->curbe.curbe_next_offset = 0;
       }
 
@@ -320,6 +336,9 @@ static void prepare_constant_buffer(struct brw_context *brw)
     * flushes as necessary when doublebuffering of CURBEs isn't
     * possible.
     */
+
+   /* check aperture space for this bo */
+   return dri_bufmgr_check_aperture_space(brw->curbe.curbe_bo);
 }
 
 
@@ -327,13 +346,6 @@ static void emit_constant_buffer(struct brw_context *brw)
 {
    struct intel_context *intel = &brw->intel;
    GLuint sz = brw->curbe.total_size;
-   dri_bo *aper_array[] = {
-      brw->intel.batch->buf,
-      brw->curbe.curbe_bo,
-   };
-
-   if (dri_bufmgr_check_aperture_space(aper_array, ARRAY_SIZE(aper_array)))
-      intel_batchbuffer_flush(intel->batch);
 
    BEGIN_BATCH(2, IGNORE_CLIPRECTS);
    if (sz == 0) {
@@ -341,8 +353,7 @@ static void emit_constant_buffer(struct brw_context *brw)
       OUT_BATCH(0);
    } else {
       OUT_BATCH((CMD_CONST_BUFFER << 16) | (1 << 8) | (2 - 2));
-      OUT_RELOC(brw->curbe.curbe_bo,
-		I915_GEM_DOMAIN_INSTRUCTION, 0,
+      OUT_RELOC(brw->curbe.curbe_bo, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
 		(sz - 1) + brw->curbe.curbe_offset);
    }
    ADVANCE_BATCH();
diff --git a/src/mesa/drivers/dri/i965/brw_defines.h b/src/mesa/drivers/dri/i965/brw_defines.h
index 0593e8d5f5..92c058ade8 100644
--- a/src/mesa/drivers/dri/i965/brw_defines.h
+++ b/src/mesa/drivers/dri/i965/brw_defines.h
@@ -33,6 +33,69 @@
 #ifndef BRW_DEFINES_H
 #define BRW_DEFINES_H
 
+/*
+ */
+#define MI_NOOP                              0x00
+#define MI_USER_INTERRUPT                    0x02
+#define MI_WAIT_FOR_EVENT                    0x03
+#define MI_FLUSH                             0x04
+#define MI_REPORT_HEAD                       0x07
+#define MI_ARB_ON_OFF                        0x08
+#define MI_BATCH_BUFFER_END                  0x0A
+#define MI_OVERLAY_FLIP                      0x11
+#define MI_LOAD_SCAN_LINES_INCL              0x12
+#define MI_LOAD_SCAN_LINES_EXCL              0x13
+#define MI_DISPLAY_BUFFER_INFO               0x14
+#define MI_SET_CONTEXT                       0x18
+#define MI_STORE_DATA_IMM                    0x20
+#define MI_STORE_DATA_INDEX                  0x21
+#define MI_LOAD_REGISTER_IMM                 0x22
+#define MI_STORE_REGISTER_MEM                0x24
+#define MI_BATCH_BUFFER_START                0x31
+
+#define MI_SYNCHRONOUS_FLIP                  0x0 
+#define MI_ASYNCHRONOUS_FLIP                 0x1
+
+#define MI_BUFFER_SECURE                     0x0 
+#define MI_BUFFER_NONSECURE                  0x1
+
+#define MI_ARBITRATE_AT_CHAIN_POINTS         0x0 
+#define MI_ARBITRATE_BETWEEN_INSTS           0x1
+#define MI_NO_ARBITRATION                    0x3 
+
+#define MI_CONDITION_CODE_WAIT_DISABLED      0x0
+#define MI_CONDITION_CODE_WAIT_0             0x1
+#define MI_CONDITION_CODE_WAIT_1             0x2
+#define MI_CONDITION_CODE_WAIT_2             0x3
+#define MI_CONDITION_CODE_WAIT_3             0x4
+#define MI_CONDITION_CODE_WAIT_4             0x5
+
+#define MI_DISPLAY_PIPE_A                    0x0
+#define MI_DISPLAY_PIPE_B                    0x1
+
+#define MI_DISPLAY_PLANE_A                   0x0 
+#define MI_DISPLAY_PLANE_B                   0x1
+#define MI_DISPLAY_PLANE_C                   0x2
+
+#define MI_STANDARD_FLIP                                 0x0
+#define MI_ENQUEUE_FLIP_PERFORM_BASE_FRAME_NUMBER_LOAD   0x1
+#define MI_ENQUEUE_FLIP_TARGET_FRAME_NUMBER_RELATIVE     0x2
+#define MI_ENQUEUE_FLIP_ABSOLUTE_TARGET_FRAME_NUMBER     0x3
+
+#define MI_PHYSICAL_ADDRESS                  0x0
+#define MI_VIRTUAL_ADDRESS                   0x1
+
+#define MI_BUFFER_MEMORY_MAIN                0x0 
+#define MI_BUFFER_MEMORY_GTT                 0x2
+#define MI_BUFFER_MEMORY_PER_PROCESS_GTT     0x3 
+
+#define MI_FLIP_CONTINUE                     0x0
+#define MI_FLIP_ON                           0x1
+#define MI_FLIP_OFF                          0x2
+
+#define MI_UNTRUSTED_REGISTER_SPACE          0x0
+#define MI_TRUSTED_REGISTER_SPACE            0x1
+
 /* 3D state:
  */
 #define _3DOP_3DSTATE_PIPELINED       0x0
@@ -56,6 +119,7 @@
 #define _3DSTATE_LINE_STIPPLE                 0x08
 #define _3DSTATE_GLOBAL_DEPTH_OFFSET_CLAMP    0x09
 #define _3DCONTROL    0x00
+#define _3DPRIMITIVE  0x00
 
 #define PIPE_CONTROL_NOWRITE          0x00
 #define PIPE_CONTROL_WRITEIMMEDIATE   0x01
diff --git a/src/mesa/drivers/dri/i965/brw_draw.c b/src/mesa/drivers/dri/i965/brw_draw.c
index d43b52c2c7..f90c5f7b08 100644
--- a/src/mesa/drivers/dri/i965/brw_draw.c
+++ b/src/mesa/drivers/dri/i965/brw_draw.c
@@ -83,8 +83,9 @@ static const GLenum reduced_prim[GL_POLYGON+1] = {
  * programs be immune to the active primitive (ie. cope with all
  * possibilities).  That may not be realistic however.
  */
-static GLuint brw_set_prim(struct brw_context *brw, GLenum prim)
+static GLuint brw_set_prim(struct brw_context *brw, GLenum prim, GLboolean *need_flush)
 {
+   int ret;
    if (INTEL_DEBUG & DEBUG_PRIMS)
       _mesa_printf("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim));
    
@@ -105,7 +106,9 @@ static GLuint brw_set_prim(struct brw_context *brw, GLenum prim)
 	 brw->state.dirty.brw |= BRW_NEW_REDUCED_PRIMITIVE;
       }
 
-      brw_validate_state(brw);
+      ret = brw_validate_state(brw);
+      if (ret)
+         *need_flush = GL_TRUE;
    }
 
    return hw_prim[prim];
@@ -128,6 +131,7 @@ static void brw_emit_prim( struct brw_context *brw,
 
 {
    struct brw_3d_primitive prim_packet;
+   GLboolean need_flush = GL_FALSE;
 
    if (INTEL_DEBUG & DEBUG_PRIMS)
       _mesa_printf("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim->mode), 
@@ -136,7 +140,7 @@ static void brw_emit_prim( struct brw_context *brw,
    prim_packet.header.opcode = CMD_3D_PRIM;
    prim_packet.header.length = sizeof(prim_packet)/4 - 2;
    prim_packet.header.pad = 0;
-   prim_packet.header.topology = brw_set_prim(brw, prim->mode);
+   prim_packet.header.topology = brw_set_prim(brw, prim->mode, &need_flush);
    prim_packet.header.indexed = prim->indexed;
 
    prim_packet.verts_per_instance = trim(prim->mode, prim->count);
@@ -145,13 +149,12 @@ static void brw_emit_prim( struct brw_context *brw,
    prim_packet.start_instance_location = 0;
    prim_packet.base_vert_location = 0;
 
-   /* Can't wrap here, since we rely on the validated state. */
-   brw->no_batch_wrap = GL_TRUE;
    if (prim_packet.verts_per_instance) {
       intel_batchbuffer_data( brw->intel.batch, &prim_packet,
 			      sizeof(prim_packet), LOOP_CLIPRECTS);
    }
-   brw->no_batch_wrap = GL_FALSE;
+
+   assert(need_flush == GL_FALSE);
 }
 
 static void brw_merge_inputs( struct brw_context *brw,
@@ -255,6 +258,10 @@ static GLboolean brw_try_draw_prims( GLcontext *ctx,
    struct brw_context *brw = brw_context(ctx);
    GLboolean retval = GL_FALSE;
    GLuint i;
+   GLuint ib_offset;
+   dri_bo *ib_bo;
+   GLboolean force_flush = GL_FALSE;
+   int ret;
 
    if (ctx->NewState)
       _mesa_update_state( ctx );
@@ -264,13 +271,7 @@ static GLboolean brw_try_draw_prims( GLcontext *ctx,
    /* Bind all inputs, derive varying and size information:
     */
    brw_merge_inputs( brw, arrays );
-
-   brw->ib.ib = ib;
-   brw->state.dirty.brw |= BRW_NEW_INDICES;
-
-   brw->vb.min_index = min_index;
-   brw->vb.max_index = max_index;
-   brw->state.dirty.brw |= BRW_NEW_VERTICES;
+      
    /* Have to validate state quite late.  Will rebuild tnl_program,
     * which depends on varying information.  
     * 
@@ -293,18 +294,29 @@ static GLboolean brw_try_draw_prims( GLcontext *ctx,
        * an upper bound of how much we might emit in a single
        * brw_try_draw_prims().
        */
+   flush:
+      if (force_flush)
+         brw->no_batch_wrap = GL_FALSE;
+
       if (intel->batch->ptr - intel->batch->map > intel->batch->size * 3 / 4
 	/* brw_emit_prim may change the cliprect_mode to LOOP_CLIPRECTS */
-	  || intel->batch->cliprect_mode != LOOP_CLIPRECTS)
+	  || intel->batch->cliprect_mode != LOOP_CLIPRECTS || (force_flush == GL_TRUE))
 	      intel_batchbuffer_flush(intel->batch);
 
+      force_flush = GL_FALSE;
+      brw->no_batch_wrap = GL_TRUE;
+
       /* Set the first primitive early, ahead of validate_state:
        */
-      brw_set_prim(brw, prim[0].mode);
+      brw_set_prim(brw, prim[0].mode, &force_flush);
 
       /* XXX:  Need to separate validate and upload of state.  
        */
-      brw_validate_state( brw );
+      ret = brw_validate_state( brw );
+      if (ret) {
+         force_flush = GL_TRUE;
+         goto flush;
+      }
 
       /* Various fallback checks:
        */
@@ -314,6 +326,31 @@ static GLboolean brw_try_draw_prims( GLcontext *ctx,
       if (check_fallbacks( brw, prim, nr_prims ))
 	 goto out;
 
+      /* need to account for index buffer and vertex buffer */
+      if (ib) {
+         ret = brw_prepare_indices( brw, ib , &ib_bo, &ib_offset);
+         if (ret) {
+            force_flush = GL_TRUE;
+            goto flush;
+         }
+      }
+
+      ret = brw_prepare_vertices( brw, min_index, max_index);
+      if (ret < 0)
+         goto out;
+
+      if (ret > 0) {
+         force_flush = GL_TRUE;
+         goto flush;
+      }
+	  
+      /* Upload index, vertex data: 
+       */
+      if (ib)
+	brw_emit_indices( brw, ib, ib_bo, ib_offset);
+
+      brw_emit_vertices( brw, min_index, max_index);
+
       for (i = 0; i < nr_prims; i++) {
 	 brw_emit_prim(brw, &prim[i]);
       }
@@ -322,6 +359,9 @@ static GLboolean brw_try_draw_prims( GLcontext *ctx,
    }
 
  out:
+
+   brw->no_batch_wrap = GL_FALSE;
+
    UNLOCK_HARDWARE(intel);
 
    if (!retval)
diff --git a/src/mesa/drivers/dri/i965/brw_draw.h b/src/mesa/drivers/dri/i965/brw_draw.h
index 2a3e0c1c5b..b3547400d4 100644
--- a/src/mesa/drivers/dri/i965/brw_draw.h
+++ b/src/mesa/drivers/dri/i965/brw_draw.h
@@ -51,4 +51,27 @@ void brw_draw_destroy( struct brw_context *brw );
 void brw_init_current_values(GLcontext *ctx,
 			     struct gl_client_array *arrays);
 
+
+/* brw_draw_upload.c
+ */
+int brw_prepare_indices( struct brw_context *brw,
+			 const struct _mesa_index_buffer *index_buffer,
+			 dri_bo **bo_return,
+			 GLuint *offset_return);
+
+void brw_emit_indices( struct brw_context *brw,
+		       const struct _mesa_index_buffer *index_buffer,
+		       dri_bo *bo,
+		       GLuint offset);
+
+int brw_prepare_vertices( struct brw_context *brw,
+			       GLuint min_index,
+			       GLuint max_index );
+
+void brw_emit_vertices( struct brw_context *brw,
+			       GLuint min_index,
+			       GLuint max_index );
+
+
+
 #endif
diff --git a/src/mesa/drivers/dri/i965/brw_draw_upload.c b/src/mesa/drivers/dri/i965/brw_draw_upload.c
index 0122d18f82..7946ffd9d9 100644
--- a/src/mesa/drivers/dri/i965/brw_draw_upload.c
+++ b/src/mesa/drivers/dri/i965/brw_draw_upload.c
@@ -228,7 +228,10 @@ static void wrap_buffers( struct brw_context *brw,
    if (brw->vb.upload.bo != NULL)
       dri_bo_unreference(brw->vb.upload.bo);
    brw->vb.upload.bo = dri_bo_alloc(brw->intel.bufmgr, "temporary VBO",
-				    size, 1);
+				    size, 1,
+				    DRM_BO_FLAG_MEM_LOCAL |
+				    DRM_BO_FLAG_CACHED |
+				    DRM_BO_FLAG_CACHED_MAPPED);
 
    /* Set the internal VBO\ to no-backing-store.  We only use them as a
     * temporary within a brw_try_draw_prims while the lock is held.
@@ -301,7 +304,9 @@ copy_array_to_vbo_array( struct brw_context *brw,
    }
 }
 
-static void brw_prepare_vertices(struct brw_context *brw)
+int brw_prepare_vertices( struct brw_context *brw,
+			       GLuint min_index,
+			       GLuint max_index )
 {
    GLcontext *ctx = &brw->intel.ctx;
    struct intel_context *intel = intel_context(ctx);
@@ -309,8 +314,7 @@ static void brw_prepare_vertices(struct brw_context *brw)
    GLuint i;
    const unsigned char *ptr = NULL;
    GLuint interleave = 0;
-   unsigned int min_index = brw->vb.min_index;
-   unsigned int max_index = brw->vb.max_index;
+   int ret = 0;
 
    struct brw_vertex_element *enabled[VERT_ATTRIB_MAX];
    GLuint nr_enabled = 0;
@@ -338,10 +342,8 @@ static void brw_prepare_vertices(struct brw_context *brw)
     * cases with > 17 vertex attributes enabled, so it probably
     * isn't an issue at this point.
     */
-   if (nr_enabled >= BRW_VEP_MAX) {
-      intel->Fallback = 1;
-      return;
-   }
+   if (nr_enabled >= BRW_VEP_MAX)
+       return -1;
 
    for (i = 0; i < nr_enabled; i++) {
       struct brw_vertex_element *input = enabled[i];
@@ -359,6 +361,8 @@ static void brw_prepare_vertices(struct brw_context *brw)
 	 dri_bo_reference(input->bo);
 	 input->offset = (unsigned long)input->glarray->Ptr;
 	 input->stride = input->glarray->StrideB;
+
+	 ret |= dri_bufmgr_check_aperture_space(input->bo);
       } else {
 	 /* Queue the buffer object up to be uploaded in the next pass,
 	  * when we've decided if we're doing interleaved or not.
@@ -367,7 +371,7 @@ static void brw_prepare_vertices(struct brw_context *brw)
 	    /* Position array not properly enabled:
 	     */
 	    if (input->glarray->StrideB == 0)
-	      return;
+	      return -1;
 
 	    interleave = input->glarray->StrideB;
 	    ptr = input->glarray->Ptr;
@@ -399,6 +403,7 @@ static void brw_prepare_vertices(struct brw_context *brw)
        */
       copy_array_to_vbo_array(brw, upload[0], interleave);
 
+      ret |= dri_bufmgr_check_aperture_space(upload[0]->bo);
       for (i = 1; i < nr_uploads; i++) {
 	 /* Then, just point upload[i] at upload[0]'s buffer. */
 	 upload[i]->stride = interleave;
@@ -412,11 +417,23 @@ static void brw_prepare_vertices(struct brw_context *brw)
       /* Upload non-interleaved arrays */
       for (i = 0; i < nr_uploads; i++) {
           copy_array_to_vbo_array(brw, upload[i], upload[i]->element_size);
+          if (upload[i]->bo) {
+              ret |= dri_bufmgr_check_aperture_space(upload[i]->bo);
+          }
       }
    }
+
+
+   if (ret)
+     return 1;
+
+
+   return 0;
 }
 
-static void brw_emit_vertices(struct brw_context *brw)
+void brw_emit_vertices( struct brw_context *brw,
+                        GLuint min_index,
+                        GLuint max_index )
 {
    GLcontext *ctx = &brw->intel.ctx;
    struct intel_context *intel = intel_context(ctx);
@@ -452,9 +469,9 @@ static void brw_emit_vertices(struct brw_context *brw)
 		BRW_VB0_ACCESS_VERTEXDATA |
 		(input->stride << BRW_VB0_PITCH_SHIFT));
       OUT_RELOC(input->bo,
-		I915_GEM_DOMAIN_VERTEX, 0,
+		DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
 		input->offset);
-      OUT_BATCH(brw->vb.max_index);
+      OUT_BATCH(max_index);
       OUT_BATCH(0); /* Instance data step rate */
 
       /* Unreference the buffer so it can get freed, now that we won't
@@ -498,31 +515,18 @@ static void brw_emit_vertices(struct brw_context *brw)
    ADVANCE_BATCH();
 }
 
-const struct brw_tracked_state brw_vertices = {
-   .dirty = {
-      .mesa = 0,
-      .brw = BRW_NEW_BATCH | BRW_NEW_VERTICES,
-      .cache = 0,
-   },
-   .prepare = brw_prepare_vertices,
-   .emit = brw_emit_vertices,
-};
-
-static void brw_prepare_indices(struct brw_context *brw)
+int brw_prepare_indices( struct brw_context *brw,
+			 const struct _mesa_index_buffer *index_buffer,
+			 dri_bo **bo_return,
+			 GLuint *offset_return)
 {
    GLcontext *ctx = &brw->intel.ctx;
    struct intel_context *intel = &brw->intel;
-   const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
-   GLuint ib_size;
+   GLuint ib_size = get_size(index_buffer->type) * index_buffer->count;
    dri_bo *bo;
-   struct gl_buffer_object *bufferobj;
-   GLuint offset;
-
-   if (index_buffer == NULL)
-      return;
-
-   ib_size = get_size(index_buffer->type) * index_buffer->count;
-   bufferobj = index_buffer->obj;;
+   struct gl_buffer_object *bufferobj = index_buffer->obj;
+   GLuint offset = (GLuint)index_buffer->ptr;
+   int ret;
 
    /* Turn into a proper VBO:
     */
@@ -536,8 +540,6 @@ static void brw_prepare_indices(struct brw_context *brw)
        */
       dri_bo_subdata(bo, offset, ib_size, index_buffer->ptr);
    } else {
-      offset = (GLuint)index_buffer->ptr;
-
       /* If the index buffer isn't aligned to its element size, we have to
        * rebase it into a temporary.
        */
@@ -560,22 +562,19 @@ static void brw_prepare_indices(struct brw_context *brw)
        }
    }
 
-   dri_bo_unreference(brw->ib.bo);
-   brw->ib.bo = bo;
-   brw->ib.offset = offset;
+   *bo_return = bo;
+   *offset_return = offset;
+   ret = dri_bufmgr_check_aperture_space(bo);
+   return ret;
 }
 
-static void brw_emit_indices(struct brw_context *brw)
+void brw_emit_indices(struct brw_context *brw,
+                      const struct _mesa_index_buffer *index_buffer,
+                      dri_bo *bo,
+                      GLuint offset)
 {
    struct intel_context *intel = &brw->intel;
-   const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
-   GLuint ib_size;
-
-   if (index_buffer == NULL)
-      return;
-
-   ib_size = get_size(index_buffer->type) * index_buffer->count;
-
+   GLuint ib_size = get_size(index_buffer->type) * index_buffer->count;
    /* Emit the indexbuffer packet:
     */
    {
@@ -591,23 +590,13 @@ static void brw_emit_indices(struct brw_context *brw)
 
       BEGIN_BATCH(4, IGNORE_CLIPRECTS);
       OUT_BATCH( ib.header.dword );
-      OUT_RELOC(brw->ib.bo,
-		I915_GEM_DOMAIN_VERTEX, 0,
-		brw->ib.offset);
-      OUT_RELOC(brw->ib.bo,
-		I915_GEM_DOMAIN_VERTEX, 0,
-		brw->ib.offset + ib_size);
+      OUT_RELOC( bo, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ, offset);
+      OUT_RELOC( bo, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
+		 offset + ib_size);
       OUT_BATCH( 0 );
       ADVANCE_BATCH();
+
+      dri_bo_unreference(bo);
    }
 }
 
-const struct brw_tracked_state brw_indices = {
-   .dirty = {
-      .mesa = 0,
-      .brw = BRW_NEW_BATCH | BRW_NEW_INDICES,
-      .cache = 0,
-   },
-   .prepare = brw_prepare_indices,
-   .emit = brw_emit_indices,
-};
diff --git a/src/mesa/drivers/dri/i965/brw_fallback.c b/src/mesa/drivers/dri/i965/brw_fallback.c
index 693f68f32a..8a8fb50cb9 100644
--- a/src/mesa/drivers/dri/i965/brw_fallback.c
+++ b/src/mesa/drivers/dri/i965/brw_fallback.c
@@ -95,9 +95,10 @@ static GLboolean do_check_fallback(struct brw_context *brw)
    return GL_FALSE;
 }
 
-static void check_fallback(struct brw_context *brw)
+static int check_fallback(struct brw_context *brw)
 {
    brw->intel.Fallback = do_check_fallback(brw);
+   return 0;
 }
 
 const struct brw_tracked_state brw_check_fallback = {
diff --git a/src/mesa/drivers/dri/i965/brw_gs.c b/src/mesa/drivers/dri/i965/brw_gs.c
index 2daef0093b..9419315c7a 100644
--- a/src/mesa/drivers/dri/i965/brw_gs.c
+++ b/src/mesa/drivers/dri/i965/brw_gs.c
@@ -162,9 +162,10 @@ static void populate_key( struct brw_context *brw,
 
 /* Calculate interpolants for triangle and line rasterization.
  */
-static void prepare_gs_prog(struct brw_context *brw)
+static int prepare_gs_prog( struct brw_context *brw )
 {
    struct brw_gs_prog_key key;
+   int ret = 0;
    /* Populate the key:
     */
    populate_key(brw, &key);
@@ -182,7 +183,11 @@ static void prepare_gs_prog(struct brw_context *brw)
 					 &brw->gs.prog_data);
       if (brw->gs.prog_bo == NULL)
 	 compile_gs_prog( brw, &key );
+
+      ret |= dri_bufmgr_check_aperture_space(brw->gs.prog_bo);
    }
+
+   return ret;
 }
 
 
diff --git a/src/mesa/drivers/dri/i965/brw_gs_state.c b/src/mesa/drivers/dri/i965/brw_gs_state.c
index ff2e3ab059..f1f9e018f1 100644
--- a/src/mesa/drivers/dri/i965/brw_gs_state.c
+++ b/src/mesa/drivers/dri/i965/brw_gs_state.c
@@ -106,17 +106,17 @@ gs_unit_create_from_key(struct brw_context *brw, struct brw_gs_unit_key *key)
 
    if (key->prog_active) {
       /* Emit GS program relocation */
-      intel_bo_emit_reloc(bo,
-			  I915_GEM_DOMAIN_INSTRUCTION, 0,
-			  gs.thread0.grf_reg_count << 1,
-			  offsetof(struct brw_gs_unit_state, thread0),
-			  brw->gs.prog_bo);
+      dri_emit_reloc(bo,
+		     DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
+		     gs.thread0.grf_reg_count << 1,
+		     offsetof(struct brw_gs_unit_state, thread0),
+		     brw->gs.prog_bo);
    }
 
    return bo;
 }
 
-static void prepare_gs_unit(struct brw_context *brw)
+static int prepare_gs_unit( struct brw_context *brw )
 {
    struct brw_gs_unit_key key;
 
@@ -130,6 +130,7 @@ static void prepare_gs_unit(struct brw_context *brw)
    if (brw->gs.state_bo == NULL) {
       brw->gs.state_bo = gs_unit_create_from_key(brw, &key);
    }
+   return dri_bufmgr_check_aperture_space(brw->gs.state_bo);
 }
 
 const struct brw_tracked_state brw_gs_unit = {
diff --git a/src/mesa/drivers/dri/i965/brw_misc_state.c b/src/mesa/drivers/dri/i965/brw_misc_state.c
index 487c638ce2..62df2590f3 100644
--- a/src/mesa/drivers/dri/i965/brw_misc_state.c
+++ b/src/mesa/drivers/dri/i965/brw_misc_state.c
@@ -81,13 +81,6 @@ const struct brw_tracked_state brw_blend_constant_color = {
 static void upload_binding_table_pointers(struct brw_context *brw)
 {
    struct intel_context *intel = &brw->intel;
-   dri_bo *aper_array[] = {
-      intel->batch->buf,
-      brw->wm.bind_bo,
-   };
-
-   if (dri_bufmgr_check_aperture_space(aper_array, ARRAY_SIZE(aper_array)))
-      intel_batchbuffer_flush(intel->batch);
 
    BEGIN_BATCH(6, IGNORE_CLIPRECTS);
    OUT_BATCH(CMD_BINDING_TABLE_PTRS << 16 | (6 - 2));
@@ -95,9 +88,7 @@ static void upload_binding_table_pointers(struct brw_context *brw)
    OUT_BATCH(0); /* gs */
    OUT_BATCH(0); /* clip */
    OUT_BATCH(0); /* sf */
-   OUT_RELOC(brw->wm.bind_bo,
-	     I915_GEM_DOMAIN_SAMPLER, 0,
-	     0);
+   OUT_RELOC(brw->wm.bind_bo, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ, 0);
    ADVANCE_BATCH();
 }
 
@@ -123,43 +114,49 @@ static void upload_pipelined_state_pointers(struct brw_context *brw )
 
    BEGIN_BATCH(7, IGNORE_CLIPRECTS);
    OUT_BATCH(CMD_PIPELINED_STATE_POINTERS << 16 | (7 - 2));
-   OUT_RELOC(brw->vs.state_bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 0);
+   OUT_RELOC(brw->vs.state_bo, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ, 0);
    if (brw->gs.prog_active)
-      OUT_RELOC(brw->gs.state_bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 1);
+      OUT_RELOC(brw->gs.state_bo, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ, 1);
    else
       OUT_BATCH(0);
    if (!brw->metaops.active)
-      OUT_RELOC(brw->clip.state_bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 1);
+      OUT_RELOC(brw->clip.state_bo, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ, 1);
    else
       OUT_BATCH(0);
-   OUT_RELOC(brw->sf.state_bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 0);
-   OUT_RELOC(brw->wm.state_bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 0);
-   OUT_RELOC(brw->cc.state_bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 0);
+   OUT_RELOC(brw->sf.state_bo, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ, 0);
+   OUT_RELOC(brw->wm.state_bo, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ, 0);
+   OUT_RELOC(brw->cc.state_bo, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ, 0);
    ADVANCE_BATCH();
 
    brw->state.dirty.brw |= BRW_NEW_PSP;
 }
 
+#if 0
+/* Combined into brw_psp_urb_cbs */
+const struct brw_tracked_state brw_pipelined_state_pointers = {
+   .dirty = {
+      .mesa = 0,
+      .brw = BRW_NEW_METAOPS | BRW_NEW_BATCH,
+      .cache = (CACHE_NEW_VS_UNIT | 
+		CACHE_NEW_GS_UNIT | 
+		CACHE_NEW_GS_PROG | 
+		CACHE_NEW_CLIP_UNIT | 
+		CACHE_NEW_SF_UNIT | 
+		CACHE_NEW_WM_UNIT | 
+		CACHE_NEW_CC_UNIT)
+   },
+   .emit = upload_pipelined_state_pointers
+};
+#endif
+
 static void upload_psp_urb_cbs(struct brw_context *brw )
 {
-   struct intel_context *intel = &brw->intel;
-   dri_bo *aper_array[] = {
-      intel->batch->buf,
-      brw->vs.state_bo,
-      brw->gs.state_bo,
-      brw->clip.state_bo,
-      brw->wm.state_bo,
-      brw->cc.state_bo,
-   };
-
-   if (dri_bufmgr_check_aperture_space(aper_array, ARRAY_SIZE(aper_array)))
-      intel_batchbuffer_flush(intel->batch);
-
    upload_pipelined_state_pointers(brw);
    brw_upload_urb_fence(brw);
    brw_upload_constant_buffer_state(brw);
 }
 
+
 const struct brw_tracked_state brw_psp_urb_cbs = {
    .dirty = {
       .mesa = 0,
@@ -175,6 +172,22 @@ const struct brw_tracked_state brw_psp_urb_cbs = {
    .emit = upload_psp_urb_cbs,
 };
 
+/**
+ * Upload the depthbuffer offset and format.
+ *
+ * We have to do this per state validation as we need to emit the relocation
+ * in the batch buffer.
+ */
+
+static int prepare_depthbuffer(struct brw_context *brw)
+{
+   struct intel_region *region = brw->state.depth_region;
+
+   if (!region || !region->buffer)
+      return 0;
+   return dri_bufmgr_check_aperture_space(region->buffer);
+}
+
 static void emit_depthbuffer(struct brw_context *brw)
 {
    struct intel_context *intel = &brw->intel;
@@ -196,10 +209,6 @@ static void emit_depthbuffer(struct brw_context *brw)
       ADVANCE_BATCH();
    } else {
       unsigned int format;
-      dri_bo *aper_array[] = {
-	 intel->batch->buf,
-	 region->buffer
-      };
 
       switch (region->cpp) {
       case 2:
@@ -216,19 +225,15 @@ static void emit_depthbuffer(struct brw_context *brw)
 	 return;
       }
 
-      if (dri_bufmgr_check_aperture_space(aper_array, ARRAY_SIZE(aper_array)))
-	 intel_batchbuffer_flush(intel->batch);
-
       BEGIN_BATCH(len, IGNORE_CLIPRECTS);
       OUT_BATCH(CMD_DEPTH_BUFFER << 16 | (len - 2));
       OUT_BATCH(((region->pitch * region->cpp) - 1) |
 		(format << 18) |
 		(BRW_TILEWALK_YMAJOR << 26) |
-		((region->tiling != I915_TILING_NONE) << 27) |
+		(region->tiled << 27) |
 		(BRW_SURFACE_2D << 29));
       OUT_RELOC(region->buffer,
-		I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
-		0);
+		DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE, 0);
       OUT_BATCH((BRW_SURFACE_MIPMAPLAYOUT_BELOW << 1) |
 		((region->pitch - 1) << 6) |
 		((region->height - 1) << 19));
@@ -247,6 +252,7 @@ const struct brw_tracked_state brw_depthbuffer = {
       .brw = BRW_NEW_DEPTH_BUFFER | BRW_NEW_BATCH,
       .cache = 0,
    },
+   .prepare = prepare_depthbuffer,
    .emit = emit_depthbuffer,
 };
 
@@ -374,6 +380,40 @@ const struct brw_tracked_state brw_line_stipple = {
 };
 
 
+
+/***********************************************************************
+ * Misc constant state packets
+ */
+
+static void upload_pipe_control(struct brw_context *brw)
+{
+   struct brw_pipe_control pc;
+
+   return;
+
+   memset(&pc, 0, sizeof(pc));
+
+   pc.header.opcode = CMD_PIPE_CONTROL;
+   pc.header.length = sizeof(pc)/4 - 2;
+   pc.header.post_sync_operation = PIPE_CONTROL_NOWRITE;
+
+   pc.header.instruction_state_cache_flush_enable = 1;
+
+   pc.bits1.dest_addr_type = PIPE_CONTROL_GTTWRITE_GLOBAL;
+
+   BRW_BATCH_STRUCT(brw, &pc);
+}
+
+const struct brw_tracked_state brw_pipe_control = {
+   .dirty = {
+      .mesa = 0,
+      .brw = BRW_NEW_BATCH,
+      .cache = 0
+   },
+   .emit = upload_pipe_control
+};
+
+
 /***********************************************************************
  * Misc invarient state packets
  */
diff --git a/src/mesa/drivers/dri/i965/brw_sf.c b/src/mesa/drivers/dri/i965/brw_sf.c
index 5bb9e11310..0b61748321 100644
--- a/src/mesa/drivers/dri/i965/brw_sf.c
+++ b/src/mesa/drivers/dri/i965/brw_sf.c
@@ -125,7 +125,7 @@ static void compile_sf_prog( struct brw_context *brw,
 
 /* Calculate interpolants for triangle and line rasterization.
  */
-static void upload_sf_prog(struct brw_context *brw)
+static int upload_sf_prog( struct brw_context *brw )
 {
    struct brw_sf_prog_key key;
 
@@ -174,6 +174,7 @@ static void upload_sf_prog(struct brw_context *brw)
 				      &brw->sf.prog_data);
    if (brw->sf.prog_bo == NULL)
       compile_sf_prog( brw, &key );
+   return dri_bufmgr_check_aperture_space(brw->sf.prog_bo);
 }
 
 
diff --git a/src/mesa/drivers/dri/i965/brw_sf_state.c b/src/mesa/drivers/dri/i965/brw_sf_state.c
index 2478872b82..24388b79a5 100644
--- a/src/mesa/drivers/dri/i965/brw_sf_state.c
+++ b/src/mesa/drivers/dri/i965/brw_sf_state.c
@@ -37,7 +37,7 @@
 #include "macros.h"
 #include "intel_fbo.h"
 
-static void upload_sf_vp(struct brw_context *brw)
+static int upload_sf_vp(struct brw_context *brw)
 {
    GLcontext *ctx = &brw->intel.ctx;
    const GLfloat depth_scale = 1.0F / ctx->DrawBuffer->_DepthMaxF;
@@ -98,6 +98,8 @@ static void upload_sf_vp(struct brw_context *brw)
 
    dri_bo_unreference(brw->sf.vp_bo);
    brw->sf.vp_bo = brw_cache_data( &brw->cache, BRW_SF_VP, &sfv, NULL, 0 );
+
+   return dri_bufmgr_check_aperture_space(brw->sf.vp_bo);
 }
 
 const struct brw_tracked_state brw_sf_vp = {
@@ -251,26 +253,27 @@ sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key,
 			 NULL, NULL);
 
    /* Emit SF program relocation */
-   intel_bo_emit_reloc(bo,
-		       I915_GEM_DOMAIN_INSTRUCTION, 0,
-		       sf.thread0.grf_reg_count << 1,
-		       offsetof(struct brw_sf_unit_state, thread0),
-		       brw->sf.prog_bo);
+   dri_emit_reloc(bo,
+		  DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
+		  sf.thread0.grf_reg_count << 1,
+		  offsetof(struct brw_sf_unit_state, thread0),
+		  brw->sf.prog_bo);
 
    /* Emit SF viewport relocation */
-   intel_bo_emit_reloc(bo,
-		       I915_GEM_DOMAIN_INSTRUCTION, 0,
-		       sf.sf5.front_winding | (sf.sf5.viewport_transform << 1),
-		       offsetof(struct brw_sf_unit_state, sf5),
-		       brw->sf.vp_bo);
+   dri_emit_reloc(bo,
+		  DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
+		  sf.sf5.front_winding | (sf.sf5.viewport_transform << 1),
+		  offsetof(struct brw_sf_unit_state, sf5),
+		  brw->sf.vp_bo);
 
    return bo;
 }
 
-static void upload_sf_unit( struct brw_context *brw )
+static int upload_sf_unit( struct brw_context *brw )
 {
    struct brw_sf_unit_key key;
    dri_bo *reloc_bufs[2];
+   int ret = 0;
 
    sf_unit_populate_key(brw, &key);
 
@@ -285,6 +288,15 @@ static void upload_sf_unit( struct brw_context *brw )
    if (brw->sf.state_bo == NULL) {
       brw->sf.state_bo = sf_unit_create_from_key(brw, &key, reloc_bufs);
    }
+
+   if (reloc_bufs[0])
+     ret |= dri_bufmgr_check_aperture_space(reloc_bufs[0]);
+
+   if (reloc_bufs[1])
+     ret |= dri_bufmgr_check_aperture_space(reloc_bufs[1]);
+
+   ret |= dri_bufmgr_check_aperture_space(brw->sf.state_bo);
+   return ret;
 }
 
 const struct brw_tracked_state brw_sf_unit = {
diff --git a/src/mesa/drivers/dri/i965/brw_state.h b/src/mesa/drivers/dri/i965/brw_state.h
index 3ea6151ae9..d1fca051ec 100644
--- a/src/mesa/drivers/dri/i965/brw_state.h
+++ b/src/mesa/drivers/dri/i965/brw_state.h
@@ -80,9 +80,6 @@ const struct brw_tracked_state brw_pipe_control;
 const struct brw_tracked_state brw_clear_surface_cache;
 const struct brw_tracked_state brw_clear_batch_cache;
 
-const struct brw_tracked_state brw_indices;
-const struct brw_tracked_state brw_vertices;
-
 /***********************************************************************
  * brw_state_cache.c
  */
diff --git a/src/mesa/drivers/dri/i965/brw_state_cache.c b/src/mesa/drivers/dri/i965/brw_state_cache.c
index fc0c3bd9ff..d617650fad 100644
--- a/src/mesa/drivers/dri/i965/brw_state_cache.c
+++ b/src/mesa/drivers/dri/i965/brw_state_cache.c
@@ -214,7 +214,10 @@ brw_upload_cache( struct brw_cache *cache,
 
    /* Create the buffer object to contain the data */
    bo = dri_bo_alloc(cache->brw->intel.bufmgr,
-		     cache->name[cache_id], data_size, 1 << 6);
+		     cache->name[cache_id], data_size, 1 << 6,
+		     DRM_BO_FLAG_MEM_LOCAL |
+		     DRM_BO_FLAG_CACHED |
+		     DRM_BO_FLAG_CACHED_MAPPED);
 
 
    /* Set up the memory containing the key, aux_data, and reloc_bufs */
diff --git a/src/mesa/drivers/dri/i965/brw_state_upload.c b/src/mesa/drivers/dri/i965/brw_state_upload.c
index d1d319d92e..3b2ccd48c3 100644
--- a/src/mesa/drivers/dri/i965/brw_state_upload.c
+++ b/src/mesa/drivers/dri/i965/brw_state_upload.c
@@ -80,6 +80,7 @@ const struct brw_tracked_state *atoms[] =
     */
    &brw_invarient_state,
    &brw_state_base_address,
+   &brw_pipe_control,
 
    &brw_binding_table_pointers,
    &brw_blend_constant_color,
@@ -101,8 +102,6 @@ const struct brw_tracked_state *atoms[] =
    &brw_psp_urb_cbs,
 #endif
 
-   &brw_indices,
-   &brw_vertices,
 
    NULL,			/* brw_constant_buffer */
 };
@@ -174,12 +173,10 @@ static void xor_states( struct brw_state_flags *result,
 /***********************************************************************
  * Emit all state:
  */
-void brw_validate_state( struct brw_context *brw )
+int brw_validate_state( struct brw_context *brw )
 {
-   struct intel_context *intel = &brw->intel;
    struct brw_state_flags *state = &brw->state.dirty;
-   GLuint i, count, pass = 0;
-   dri_bo *last_batch_bo = NULL;
+   GLuint i, ret, count;
 
    state->mesa |= brw->intel.NewGLState;
    brw->intel.NewGLState = 0;
@@ -205,7 +202,7 @@ void brw_validate_state( struct brw_context *brw )
    if (state->mesa == 0 &&
        state->cache == 0 &&
        state->brw == 0)
-      return;
+      return 0;
 
    if (brw->state.dirty.brw & BRW_NEW_CONTEXT)
       brw_clear_batch_cache_flush(brw);
@@ -223,23 +220,15 @@ void brw_validate_state( struct brw_context *brw )
 
       if (check_state(state, &atom->dirty)) {
          if (atom->prepare) {
-            atom->prepare(brw);
+            ret = atom->prepare(brw);
+            if (ret)
+               return ret;
         }
       }
    }
 
    if (brw->intel.Fallback)
-      return;
-
-   /* We're about to try to set up a coherent state in the batchbuffer for
-    * the emission of primitives.  If we exceed the aperture size in any of the
-    * emit() calls, we need to go back to square 1 and try setting up again.
-    */
-got_flushed:
-   dri_bo_unreference(last_batch_bo);
-   last_batch_bo = intel->batch->buf;
-   dri_bo_reference(last_batch_bo);
-   assert(pass++ <= 2);
+      return 0;
 
    if (INTEL_DEBUG) {
       /* Debug version which enforces various sanity checks on the
@@ -262,11 +251,8 @@ got_flushed:
 	    break;
 
 	 if (check_state(state, &atom->dirty)) {
-	    if (atom->emit) {
+	    if (atom->emit)
 	       atom->emit( brw );
-	       if (intel->batch->buf != last_batch_bo)
-		  goto got_flushed;
-	    }
 	 }
 
 	 accumulate_state(&examined, &atom->dirty);
@@ -288,17 +274,13 @@ got_flushed:
 	    break;
 
 	 if (check_state(state, &atom->dirty)) {
-	    if (atom->emit) {
+	    if (atom->emit)
 	       atom->emit( brw );
-	       if (intel->batch->buf != last_batch_bo)
-		  goto got_flushed;
-	    }
 	 }
       }
    }
 
-   dri_bo_unreference(last_batch_bo);
-
    if (!brw->intel.Fallback)
       memset(state, 0, sizeof(*state));
+   return 0;
 }
diff --git a/src/mesa/drivers/dri/i965/brw_urb.c b/src/mesa/drivers/dri/i965/brw_urb.c
index 1116ade0a4..c423dbe7d7 100644
--- a/src/mesa/drivers/dri/i965/brw_urb.c
+++ b/src/mesa/drivers/dri/i965/brw_urb.c
@@ -74,7 +74,7 @@ static GLboolean check_urb_layout( struct brw_context *brw )
 /* Most minimal update, forces re-emit of URB fence packet after GS
  * unit turned on/off.
  */
-static void recalculate_urb_fence( struct brw_context *brw )
+static int recalculate_urb_fence( struct brw_context *brw )
 {
    GLuint csize = brw->curbe.total_size;
    GLuint vsize = brw->vs.prog_data->urb_entry_size;
@@ -142,6 +142,7 @@ static void recalculate_urb_fence( struct brw_context *brw )
       
       brw->state.dirty.brw |= BRW_NEW_URB_FENCE;
    }
+   return 0;
 }
 
 
@@ -186,3 +187,15 @@ void brw_upload_urb_fence(struct brw_context *brw)
 
    BRW_BATCH_STRUCT(brw, &uf);
 }
+
+
+#if 0
+const struct brw_tracked_state brw_urb_fence = {
+   .dirty = {
+      .mesa = 0,
+      .brw = BRW_NEW_URB_FENCE | BRW_NEW_PSP,
+      .cache = 0
+   },
+   .update = brw_upload_urb_fence
+};
+#endif
diff --git a/src/mesa/drivers/dri/i965/brw_vs.c b/src/mesa/drivers/dri/i965/brw_vs.c
index 1db7ceebcf..f89b0e14a1 100644
--- a/src/mesa/drivers/dri/i965/brw_vs.c
+++ b/src/mesa/drivers/dri/i965/brw_vs.c
@@ -83,7 +83,7 @@ static void do_vs_prog( struct brw_context *brw,
 }
 
 
-static void brw_upload_vs_prog(struct brw_context *brw)
+static int brw_upload_vs_prog( struct brw_context *brw )
 {
    struct brw_vs_prog_key key;
    struct brw_vertex_program *vp = 
@@ -115,6 +115,7 @@ static void brw_upload_vs_prog(struct brw_context *brw)
 				      &brw->vs.prog_data);
    if (brw->vs.prog_bo == NULL)
       do_vs_prog(brw, vp, &key);
+   return dri_bufmgr_check_aperture_space(brw->vs.prog_bo);
 }
 
 
diff --git a/src/mesa/drivers/dri/i965/brw_vs_constval.c b/src/mesa/drivers/dri/i965/brw_vs_constval.c
index 734a926e96..a0106b8975 100644
--- a/src/mesa/drivers/dri/i965/brw_vs_constval.c
+++ b/src/mesa/drivers/dri/i965/brw_vs_constval.c
@@ -166,7 +166,7 @@ static GLuint get_input_size(struct brw_context *brw,
 /* Calculate sizes of vertex program outputs.  Size is the largest
  * component index which might vary from [0,0,0,1]
  */
-static void calc_wm_input_sizes( struct brw_context *brw )
+static int calc_wm_input_sizes( struct brw_context *brw )
 {
    /* BRW_NEW_VERTEX_PROGRAM */
    struct brw_vertex_program *vp = 
@@ -210,6 +210,7 @@ static void calc_wm_input_sizes( struct brw_context *brw )
       memcpy(brw->wm.input_size_masks, t.size_masks, sizeof(t.size_masks));
       brw->state.dirty.brw |= BRW_NEW_WM_INPUT_DIMENSIONS;
    }
+   return 0;
 }
 
 const struct brw_tracked_state brw_wm_input_sizes = {
diff --git a/src/mesa/drivers/dri/i965/brw_vs_state.c b/src/mesa/drivers/dri/i965/brw_vs_state.c
index 909b942610..2a64f3df33 100644
--- a/src/mesa/drivers/dri/i965/brw_vs_state.c
+++ b/src/mesa/drivers/dri/i965/brw_vs_state.c
@@ -115,16 +115,16 @@ vs_unit_create_from_key(struct brw_context *brw, struct brw_vs_unit_key *key)
 			 NULL, NULL);
 
    /* Emit VS program relocation */
-   intel_bo_emit_reloc(bo,
-		       I915_GEM_DOMAIN_INSTRUCTION, 0,
-		       vs.thread0.grf_reg_count << 1,
-		       offsetof(struct brw_vs_unit_state, thread0),
-		       brw->vs.prog_bo);
+   dri_emit_reloc(bo,
+		  DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
+		  vs.thread0.grf_reg_count << 1,
+		  offsetof(struct brw_vs_unit_state, thread0),
+		  brw->vs.prog_bo);
 
    return bo;
 }
 
-static void prepare_vs_unit(struct brw_context *brw)
+static int prepare_vs_unit( struct brw_context *brw )
 {
    struct brw_vs_unit_key key;
 
@@ -138,6 +138,7 @@ static void prepare_vs_unit(struct brw_context *brw)
    if (brw->vs.state_bo == NULL) {
       brw->vs.state_bo = vs_unit_create_from_key(brw, &key);
    }
+   return dri_bufmgr_check_aperture_space(brw->vs.state_bo);
 }
 
 const struct brw_tracked_state brw_vs_unit = {
diff --git a/src/mesa/drivers/dri/i965/brw_vs_tnl.c b/src/mesa/drivers/dri/i965/brw_vs_tnl.c
index 2caa020e61..e409620bbf 100644
--- a/src/mesa/drivers/dri/i965/brw_vs_tnl.c
+++ b/src/mesa/drivers/dri/i965/brw_vs_tnl.c
@@ -1581,7 +1581,7 @@ static GLuint hash_key( struct state_key *key )
    return hash;
 }
 
-static void prepare_tnl_program( struct brw_context *brw )
+static int prepare_tnl_program( struct brw_context *brw )
 {
    GLcontext *ctx = &brw->intel.ctx;
    struct state_key key;
@@ -1590,7 +1590,7 @@ static void prepare_tnl_program( struct brw_context *brw )
 
    /* _NEW_PROGRAM */
    if (brw->attribs.VertexProgram->_Current) 
-      return;
+      return 0;
       
    /* Grab all the relevent state and put it in a single structure:
     */
@@ -1623,7 +1623,7 @@ static void prepare_tnl_program( struct brw_context *brw )
 
    if (old != brw->tnl_program)
       brw->state.dirty.brw |= BRW_NEW_TNL_PROGRAM;
-   return;
+   return 0;
 }
 
 /* Note: See brw_draw.c - the vertex program must not rely on
@@ -1649,7 +1649,7 @@ const struct brw_tracked_state brw_tnl_vertprog = {
 
 
 
-static void prepare_active_vertprog( struct brw_context *brw )
+static int prepare_active_vertprog( struct brw_context *brw )
 {
    const struct gl_vertex_program *prev = brw->vertex_program;
 
@@ -1664,6 +1664,8 @@ static void prepare_active_vertprog( struct brw_context *brw )
 
    if (brw->vertex_program != prev) 
       brw->state.dirty.brw |= BRW_NEW_VERTEX_PROGRAM;
+
+   return 0;
 }
 
 
diff --git a/src/mesa/drivers/dri/i965/brw_wm.c b/src/mesa/drivers/dri/i965/brw_wm.c
index 361312c2ca..a470a25636 100644
--- a/src/mesa/drivers/dri/i965/brw_wm.c
+++ b/src/mesa/drivers/dri/i965/brw_wm.c
@@ -322,7 +322,7 @@ static void brw_wm_populate_key( struct brw_context *brw,
 }
 
 
-static void brw_prepare_wm_prog(struct brw_context *brw)
+static int brw_prepare_wm_prog( struct brw_context *brw )
 {
    struct brw_wm_prog_key key;
    struct brw_fragment_program *fp = (struct brw_fragment_program *)
@@ -339,6 +339,8 @@ static void brw_prepare_wm_prog(struct brw_context *brw)
 				      &brw->wm.prog_data);
    if (brw->wm.prog_bo == NULL)
       do_wm_prog(brw, fp, &key);
+
+   return dri_bufmgr_check_aperture_space(brw->wm.prog_bo);
 }
 
 
diff --git a/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c b/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c
index 08d01823de..d40332e9ae 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c
@@ -255,10 +255,11 @@ brw_wm_sampler_populate_key(struct brw_context *brw,
  * complicates various things.  However, this is still too confusing -
  * FIXME: simplify all the different new texture state flags.
  */
-static void upload_wm_samplers( struct brw_context *brw )
+static int upload_wm_samplers( struct brw_context *brw )
 {
    struct wm_sampler_key key;
    int i;
+   int ret = 0;
 
    brw_wm_sampler_populate_key(brw, &key);
 
@@ -270,7 +271,7 @@ static void upload_wm_samplers( struct brw_context *brw )
    dri_bo_unreference(brw->wm.sampler_bo);
    brw->wm.sampler_bo = NULL;
    if (brw->wm.sampler_count == 0)
-      return;
+      return 0;
 
    brw->wm.sampler_bo = brw_search_cache(&brw->cache, BRW_SAMPLER,
 					 &key, sizeof(key),
@@ -303,14 +304,19 @@ static void upload_wm_samplers( struct brw_context *brw )
 	 if (!brw->attribs.Texture->Unit[i]._ReallyEnabled)
 	    continue;
 
-	 intel_bo_emit_reloc(brw->wm.sampler_bo,
-			     I915_GEM_DOMAIN_INSTRUCTION, 0,
-			     0,
-			     i * sizeof(struct brw_sampler_state) +
-			     offsetof(struct brw_sampler_state, ss2),
-			     brw->wm.sdc_bo[i]);
+	 ret |= dri_bufmgr_check_aperture_space(brw->wm.sdc_bo[i]);
+	 dri_emit_reloc(brw->wm.sampler_bo,
+			DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
+			0,
+			i * sizeof(struct brw_sampler_state) +
+			offsetof(struct brw_sampler_state, ss2),
+			brw->wm.sdc_bo[i]);
       }
    }
+
+   ret |= dri_bufmgr_check_aperture_space(brw->wm.sampler_bo);
+   return ret;
+
 }
 
 const struct brw_tracked_state brw_wm_samplers = {
diff --git a/src/mesa/drivers/dri/i965/brw_wm_state.c b/src/mesa/drivers/dri/i965/brw_wm_state.c
index f97d0dc285..f4da0f279e 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_state.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_state.c
@@ -199,39 +199,40 @@ wm_unit_create_from_key(struct brw_context *brw, struct brw_wm_unit_key *key,
 			 NULL, NULL);
 
    /* Emit WM program relocation */
-   intel_bo_emit_reloc(bo,
-		       I915_GEM_DOMAIN_INSTRUCTION, 0,
-		       wm.thread0.grf_reg_count << 1,
-		       offsetof(struct brw_wm_unit_state, thread0),
-		       brw->wm.prog_bo);
+   dri_emit_reloc(bo,
+		  DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
+		  wm.thread0.grf_reg_count << 1,
+		  offsetof(struct brw_wm_unit_state, thread0),
+		  brw->wm.prog_bo);
 
    /* Emit scratch space relocation */
    if (key->total_scratch != 0) {
-      intel_bo_emit_reloc(bo,
-			  0, 0,
-			  wm.thread2.per_thread_scratch_space,
-			  offsetof(struct brw_wm_unit_state, thread2),
-			  brw->wm.scratch_buffer);
+      dri_emit_reloc(bo,
+		     DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE,
+		     wm.thread2.per_thread_scratch_space,
+		     offsetof(struct brw_wm_unit_state, thread2),
+		     brw->wm.scratch_buffer);
    }
 
    /* Emit sampler state relocation */
    if (key->sampler_count != 0) {
-      intel_bo_emit_reloc(bo,
-			  I915_GEM_DOMAIN_INSTRUCTION, 0,
-			  wm.wm4.stats_enable | (wm.wm4.sampler_count << 2),
-			  offsetof(struct brw_wm_unit_state, wm4),
-			  brw->wm.sampler_bo);
+      dri_emit_reloc(bo,
+		     DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
+		     wm.wm4.stats_enable | (wm.wm4.sampler_count << 2),
+		     offsetof(struct brw_wm_unit_state, wm4),
+		     brw->wm.sampler_bo);
    }
 
    return bo;
 }
 
 
-static void upload_wm_unit( struct brw_context *brw )
+static int upload_wm_unit( struct brw_context *brw )
 {
    struct intel_context *intel = &brw->intel;
    struct brw_wm_unit_key key;
    dri_bo *reloc_bufs[3];
+   int ret = 0, i;
    wm_unit_populate_key(brw, &key);
 
    /* Allocate the necessary scratch space if we haven't already.  Don't
@@ -250,7 +251,7 @@ static void upload_wm_unit( struct brw_context *brw )
 	 brw->wm.scratch_buffer = dri_bo_alloc(intel->bufmgr,
 					       "wm scratch",
 					       total,
-					       4096);
+					       4096, DRM_BO_FLAG_MEM_TT);
       }
    }
 
@@ -266,6 +267,12 @@ static void upload_wm_unit( struct brw_context *brw )
    if (brw->wm.state_bo == NULL) {
       brw->wm.state_bo = wm_unit_create_from_key(brw, &key, reloc_bufs);
    }
+
+   for (i = 0; i < 3; i++)
+     if (reloc_bufs[i])
+       ret |= dri_bufmgr_check_aperture_space(reloc_bufs[i]);
+   ret |= dri_bufmgr_check_aperture_space(brw->wm.state_bo);
+   return ret;
 }
 
 const struct brw_tracked_state brw_wm_unit = {
diff --git a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
index af1401b34b..2ba3eb4de8 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
@@ -154,29 +154,10 @@ struct brw_wm_surface_key {
    GLint first_level, last_level;
    GLint width, height, depth;
    GLint pitch, cpp;
-   uint32_t tiling;
+   GLboolean tiled;
    GLuint offset;
 };
 
-static void
-brw_set_surface_tiling(struct brw_surface_state *surf, uint32_t tiling)
-{
-   switch (tiling) {
-   case I915_TILING_NONE:
-      surf->ss3.tiled_surface = 0;
-      surf->ss3.tile_walk = 0;
-      break;
-   case I915_TILING_X:
-      surf->ss3.tiled_surface = 1;
-      surf->ss3.tile_walk = BRW_TILEWALK_XMAJOR;
-      break;
-   case I915_TILING_Y:
-      surf->ss3.tiled_surface = 1;
-      surf->ss3.tile_walk = BRW_TILEWALK_YMAJOR;
-      break;
-   }
-}
-
 static dri_bo *
 brw_create_texture_surface( struct brw_context *brw,
 			    struct brw_wm_surface_key *key )
@@ -211,7 +192,9 @@ brw_create_texture_surface( struct brw_context *brw,
    surf.ss2.mip_count = key->last_level - key->first_level;
    surf.ss2.width = key->width - 1;
    surf.ss2.height = key->height - 1;
-   brw_set_surface_tiling(&surf, key->tiling);
+
+   surf.ss3.tile_walk = BRW_TILEWALK_XMAJOR;
+   surf.ss3.tiled_surface = key->tiled;
    surf.ss3.pitch = (key->pitch * key->cpp) - 1;
    surf.ss3.depth = key->depth - 1;
 
@@ -231,19 +214,18 @@ brw_create_texture_surface( struct brw_context *brw,
 			 &key->bo, key->bo ? 1 : 0,
 			 &surf, sizeof(surf),
 			 NULL, NULL);
-
    if (key->bo) {
       /* Emit relocation to surface contents */
-      intel_bo_emit_reloc(bo,
-			  I915_GEM_DOMAIN_SAMPLER, 0,
-			  0,
-			  offsetof(struct brw_surface_state, ss1),
-			  key->bo);
+      dri_emit_reloc(bo,
+		     DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
+		     0,
+		     offsetof(struct brw_surface_state, ss1),
+		     key->bo);
    }
    return bo;
 }
 
-static void
+static int
 brw_update_texture_surface( GLcontext *ctx, GLuint unit )
 {
    struct brw_context *brw = brw_context(ctx);
@@ -251,6 +233,7 @@ brw_update_texture_surface( GLcontext *ctx, GLuint unit )
    struct intel_texture_object *intelObj = intel_texture_object(tObj);
    struct gl_texture_image *firstImage = tObj->Image[0][intelObj->firstLevel];
    struct brw_wm_surface_key key;
+   int ret = 0;
 
    memset(&key, 0, sizeof(key));
 
@@ -265,6 +248,7 @@ brw_update_texture_surface( GLcontext *ctx, GLuint unit )
       key.depth = firstImage->Depth;
       key.bo = intelObj->mt->region->buffer;
       key.offset = 0;
+      ret |= dri_bufmgr_check_aperture_space(key.bo);
    }
 
    key.target = tObj->Target;
@@ -274,8 +258,7 @@ brw_update_texture_surface( GLcontext *ctx, GLuint unit )
    key.width = firstImage->Width;
    key.height = firstImage->Height;
    key.cpp = intelObj->mt->cpp;
-   key.depth = firstImage->Depth;
-   key.tiling = intelObj->mt->region->tiling;
+   key.tiled = intelObj->mt->region->tiled;
 
    dri_bo_unreference(brw->wm.surf_bo[unit + MAX_DRAW_BUFFERS]);
    brw->wm.surf_bo[unit + MAX_DRAW_BUFFERS] = brw_search_cache(&brw->cache, BRW_SS_SURFACE,
@@ -285,6 +268,9 @@ brw_update_texture_surface( GLcontext *ctx, GLuint unit )
    if (brw->wm.surf_bo[unit + MAX_DRAW_BUFFERS] == NULL) {
       brw->wm.surf_bo[unit + MAX_DRAW_BUFFERS] = brw_create_texture_surface(brw, &key);
    }
+
+   ret |= dri_bufmgr_check_aperture_space(brw->wm.surf_bo[unit + MAX_DRAW_BUFFERS]);
+   return ret;
 }
 
 /**
@@ -292,18 +278,18 @@ brw_update_texture_surface( GLcontext *ctx, GLuint unit )
  * While it is only used for the front/back buffer currently, it should be
  * usable for further buffers when doing ARB_draw_buffer support.
  */
-static void
+static int
 brw_update_region_surface(struct brw_context *brw, struct intel_region *region,
 			  unsigned int unit, GLboolean cached)
 {
    dri_bo *region_bo = NULL;
+   int ret = 0;
    struct {
       unsigned int surface_type;
       unsigned int surface_format;
       unsigned int width, height, cpp;
       GLubyte color_mask[4];
-      GLboolean color_blend;
-      uint32_t tiling;
+      GLboolean tiled, color_blend;
    } key;
 
    memset(&key, 0, sizeof(key));
@@ -316,14 +302,16 @@ brw_update_region_surface(struct brw_context *brw, struct intel_region *region,
 	 key.surface_format = BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
       else
 	 key.surface_format = BRW_SURFACEFORMAT_B5G6R5_UNORM;
-      key.tiling = region->tiling;
+      key.tiled = region->tiled;
       key.width = region->pitch; /* XXX: not really! */
       key.height = region->height;
       key.cpp = region->cpp;
+
+      ret |= dri_bufmgr_check_aperture_space(region->buffer);
    } else {
       key.surface_type = BRW_SURFACE_NULL;
       key.surface_format = BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
-      key.tiling = 0;
+      key.tiled = 0;
       key.width = 1;
       key.height = 1;
       key.cpp = 4;
@@ -353,7 +341,8 @@ brw_update_region_surface(struct brw_context *brw, struct intel_region *region,
 
       surf.ss2.width = key.width - 1;
       surf.ss2.height = key.height - 1;
-      brw_set_surface_tiling(&surf, key.tiling);
+      surf.ss3.tile_walk = BRW_TILEWALK_XMAJOR;
+      surf.ss3.tiled_surface = key.tiled;
       surf.ss3.pitch = (key.width * key.cpp) - 1;
 
       /* _NEW_COLOR */
@@ -370,19 +359,19 @@ brw_update_region_surface(struct brw_context *brw, struct intel_region *region,
 					       &surf, sizeof(surf),
 					       NULL, NULL);
       if (region_bo != NULL) {
-	 /* We might sample from it, and we might render to it, so flag
-	  * them both.  We might be able to figure out from other state
-	  * a more restrictive relocation to emit.
-	  */
-	 intel_bo_emit_reloc(brw->wm.surf_bo[unit],
-			     I915_GEM_DOMAIN_RENDER |
-			     I915_GEM_DOMAIN_SAMPLER,
-			     I915_GEM_DOMAIN_RENDER,
-			     0,
-			     offsetof(struct brw_surface_state, ss1),
-			     region_bo);
+	 dri_emit_reloc(brw->wm.surf_bo[unit],
+			DRM_BO_FLAG_MEM_TT |
+			DRM_BO_FLAG_READ |
+			DRM_BO_FLAG_WRITE,
+			0,
+			offsetof(struct brw_surface_state, ss1),
+			region_bo);
       }
    }
+
+   ret |= dri_bufmgr_check_aperture_space(brw->wm.surf_bo[unit]);
+
+   return ret;
 }
 
 
@@ -420,11 +409,13 @@ brw_wm_get_binding_table(struct brw_context *brw)
       /* Emit binding table relocations to surface state */
       for (i = 0; i < BRW_WM_MAX_SURF; i++) {
 	 if (brw->wm.surf_bo[i] != NULL) {
-	    intel_bo_emit_reloc(bind_bo,
-				I915_GEM_DOMAIN_INSTRUCTION, 0,
-				0,
-				i * sizeof(GLuint),
-				brw->wm.surf_bo[i]);
+	    dri_emit_reloc(bind_bo,
+			   DRM_BO_FLAG_MEM_TT |
+			   DRM_BO_FLAG_READ |
+			   DRM_BO_FLAG_WRITE,
+			   0,
+			   i * sizeof(GLuint),
+			   brw->wm.surf_bo[i]);
 	 }
       }
 
@@ -434,19 +425,23 @@ brw_wm_get_binding_table(struct brw_context *brw)
    return bind_bo;
 }
 
-static void prepare_wm_surfaces(struct brw_context *brw )
+static int prepare_wm_surfaces(struct brw_context *brw )
 {
    GLcontext *ctx = &brw->intel.ctx;
    struct intel_context *intel = &brw->intel;
-   GLuint i;
+   GLuint i, ret;
 
    if (brw->state.nr_draw_regions  > 1) {
       for (i = 0; i < brw->state.nr_draw_regions; i++) {
-         brw_update_region_surface(brw, brw->state.draw_regions[i], i,
-				   GL_FALSE);
+         ret = brw_update_region_surface(brw, brw->state.draw_regions[i], i,
+                                         GL_FALSE);
+         if (ret)
+            return ret;
       }
    }else {
-      brw_update_region_surface(brw, brw->state.draw_regions[0], 0, GL_TRUE);
+      ret = brw_update_region_surface(brw, brw->state.draw_regions[0], 0, GL_TRUE);
+      if (ret)
+         return ret;
    }
 
    brw->wm.nr_surfaces = MAX_DRAW_BUFFERS;
@@ -462,8 +457,11 @@ static void prepare_wm_surfaces(struct brw_context *brw )
             dri_bo_reference(brw->wm.surf_bo[i+MAX_DRAW_BUFFERS]);
             brw->wm.nr_surfaces = i + MAX_DRAW_BUFFERS + 1;
          } else {
-            brw_update_texture_surface(ctx, i);
+            ret = brw_update_texture_surface(ctx, i);
             brw->wm.nr_surfaces = i + MAX_DRAW_BUFFERS + 1;
+
+            if (ret)
+               return ret;
          }
       } else {
          dri_bo_unreference(brw->wm.surf_bo[i+MAX_DRAW_BUFFERS]);
@@ -474,6 +472,8 @@ static void prepare_wm_surfaces(struct brw_context *brw )
 
    dri_bo_unreference(brw->wm.bind_bo);
    brw->wm.bind_bo = brw_wm_get_binding_table(brw);
+
+   return dri_bufmgr_check_aperture_space(brw->wm.bind_bo);
 }
 
 
diff --git a/src/mesa/drivers/dri/i965/intel_bufmgr_ttm.c b/src/mesa/drivers/dri/i965/intel_bufmgr_ttm.c
new file mode 120000
index 0000000000..e9df5c6279
--- /dev/null
+++ b/src/mesa/drivers/dri/i965/intel_bufmgr_ttm.c
@@ -0,0 +1 @@
+../intel/intel_bufmgr_ttm.c
\ No newline at end of file
diff --git a/src/mesa/drivers/dri/intel/intel_batchbuffer.c b/src/mesa/drivers/dri/intel/intel_batchbuffer.c
index 5afaad070c..a594fb6cc4 100644
--- a/src/mesa/drivers/dri/intel/intel_batchbuffer.c
+++ b/src/mesa/drivers/dri/intel/intel_batchbuffer.c
@@ -29,7 +29,6 @@
 #include "intel_ioctl.h"
 #include "intel_decode.h"
 #include "intel_reg.h"
-#include "intel_bufmgr.h"
 
 /* Relocations in kernel space:
  *    - pass dma buffer seperately
@@ -79,21 +78,19 @@ intel_batchbuffer_reset(struct intel_batchbuffer *batch)
       batch->buf = NULL;
    }
 
-   if (!batch->buffer && intel->ttm == GL_TRUE)
-      batch->buffer = malloc (intel->maxBatchSize);
-
    batch->buf = dri_bo_alloc(intel->bufmgr, "batchbuffer",
-			     intel->maxBatchSize, 4096);
-   if (batch->buffer)
-      batch->map = batch->buffer;
-   else {
-      dri_bo_map(batch->buf, GL_TRUE);
-      batch->map = batch->buf->virtual;
-   }
+			     intel->maxBatchSize, 4096,
+			     DRM_BO_FLAG_MEM_LOCAL | DRM_BO_FLAG_CACHED | DRM_BO_FLAG_CACHED_MAPPED);
+   dri_bo_map(batch->buf, GL_TRUE);
+   batch->map = batch->buf->virtual;
    batch->size = intel->maxBatchSize;
    batch->ptr = batch->map;
    batch->dirty_state = ~0;
    batch->cliprect_mode = IGNORE_CLIPRECTS;
+
+   /* account batchbuffer in aperture */
+   dri_bufmgr_check_aperture_space(batch->buf);
+
 }
 
 struct intel_batchbuffer *
@@ -102,6 +99,7 @@ intel_batchbuffer_alloc(struct intel_context *intel)
    struct intel_batchbuffer *batch = calloc(sizeof(*batch), 1);
 
    batch->intel = intel;
+   batch->last_fence = NULL;
    intel_batchbuffer_reset(batch);
 
    return batch;
@@ -110,13 +108,14 @@ intel_batchbuffer_alloc(struct intel_context *intel)
 void
 intel_batchbuffer_free(struct intel_batchbuffer *batch)
 {
-   if (batch->buffer)
-      free (batch->buffer);
-   else {
-      if (batch->map) {
-	 dri_bo_unmap(batch->buf);
-	 batch->map = NULL;
-      }
+   if (batch->last_fence) {
+      dri_fence_wait(batch->last_fence);
+      dri_fence_unreference(batch->last_fence);
+      batch->last_fence = NULL;
+   }
+   if (batch->map) {
+      dri_bo_unmap(batch->buf);
+      batch->map = NULL;
    }
    dri_bo_unreference(batch->buf);
    batch->buf = NULL;
@@ -132,12 +131,11 @@ do_flush_locked(struct intel_batchbuffer *batch,
 		GLuint used, GLboolean allow_unlock)
 {
    struct intel_context *intel = batch->intel;
-   int ret = 0;
+   void *start;
+   GLuint count;
 
-   if (batch->buffer)
-      dri_bo_subdata (batch->buf, 0, used, batch->buffer);
-   else
-      dri_bo_unmap(batch->buf);
+   dri_bo_unmap(batch->buf);
+   start = dri_process_relocs(batch->buf, &count);
 
    batch->map = NULL;
    batch->ptr = NULL;
@@ -150,25 +148,21 @@ do_flush_locked(struct intel_batchbuffer *batch,
    if (!(intel->numClipRects == 0 &&
 	 batch->cliprect_mode == LOOP_CLIPRECTS)) {
       if (intel->ttm == GL_TRUE) {
-	 struct drm_i915_gem_execbuffer *execbuf;
-
-	 execbuf = dri_process_relocs(batch->buf);
-	 ret = intel_exec_ioctl(batch->intel,
-				used,
-				batch->cliprect_mode != LOOP_CLIPRECTS,
-				allow_unlock,
-				execbuf);
+	 intel_exec_ioctl(batch->intel,
+			  used,
+			  batch->cliprect_mode != LOOP_CLIPRECTS,
+			  allow_unlock,
+			  start, count, &batch->last_fence);
       } else {
-	 dri_process_relocs(batch->buf);
-	 ret = intel_batch_ioctl(batch->intel,
-				 batch->buf->offset,
-				 used,
-				 batch->cliprect_mode != LOOP_CLIPRECTS,
-				 allow_unlock);
+	 intel_batch_ioctl(batch->intel,
+			   batch->buf->offset,
+			   used,
+			   batch->cliprect_mode != LOOP_CLIPRECTS,
+			   allow_unlock);
       }
    }
-
-   dri_post_submit(batch->buf);
+      
+   dri_post_submit(batch->buf, &batch->last_fence);
 
    if (intel->numClipRects == 0 &&
        batch->cliprect_mode == LOOP_CLIPRECTS) {
@@ -193,10 +187,6 @@ do_flush_locked(struct intel_batchbuffer *batch,
 	 intel->vtbl.debug_batch(intel);
    }
 
-   if (ret != 0) {
-      UNLOCK_HARDWARE(intel);
-      exit(1);
-   }
    intel->vtbl.new_batch(intel);
 }
 
@@ -214,27 +204,21 @@ _intel_batchbuffer_flush(struct intel_batchbuffer *batch, const char *file,
    if (INTEL_DEBUG & DEBUG_BATCH)
       fprintf(stderr, "%s:%d: Batchbuffer flush with %db used\n", file, line,
 	      used);
-
-   /* Emit a flush if the bufmgr doesn't do it for us. */
-   if (!intel->ttm) {
-      *(GLuint *) (batch->ptr) = intel->vtbl.flush_cmd();
-      batch->ptr += 4;
-      used = batch->ptr - batch->map;
+   /* Add the MI_BATCH_BUFFER_END.  Always add an MI_FLUSH - this is a
+    * performance drain that we would like to avoid.
+    */
+   if (used & 4) {
+      ((int *) batch->ptr)[0] = intel->vtbl.flush_cmd();
+      ((int *) batch->ptr)[1] = 0;
+      ((int *) batch->ptr)[2] = MI_BATCH_BUFFER_END;
+      used += 12;
    }
-
-   /* Round batchbuffer usage to 2 DWORDs. */
-
-   if ((used & 4) == 0) {
-      *(GLuint *) (batch->ptr) = 0; /* noop */
-      batch->ptr += 4;
-      used = batch->ptr - batch->map;
+   else {
+      ((int *) batch->ptr)[0] = intel->vtbl.flush_cmd();
+      ((int *) batch->ptr)[1] = MI_BATCH_BUFFER_END;
+      used += 8;
    }
 
-   /* Mark the end of the buffer. */
-   *(GLuint *) (batch->ptr) = MI_BATCH_BUFFER_END; /* noop */
-   batch->ptr += 4;
-   used = batch->ptr - batch->map;
-
    /* Workaround for recursive batchbuffer flushing: If the window is
     * moved, we can get into a case where we try to flush during a
     * flush.  What happens is that when we try to grab the lock for
@@ -246,9 +230,6 @@ _intel_batchbuffer_flush(struct intel_batchbuffer *batch, const char *file,
     * avoid that in the first place. */
    batch->ptr = batch->map;
 
-   if (intel->vtbl.finish_batch)
-      intel->vtbl.finish_batch(intel);
-
    /* TODO: Just pass the relocation list and dma buffer up to the
     * kernel.
     */
@@ -261,13 +242,9 @@ _intel_batchbuffer_flush(struct intel_batchbuffer *batch, const char *file,
       UNLOCK_HARDWARE(intel);
 
    if (INTEL_DEBUG & DEBUG_SYNC) {
-      int irq;
-
       fprintf(stderr, "waiting for idle\n");
-      LOCK_HARDWARE(intel);
-      irq = intelEmitIrqLocked(intel);
-      UNLOCK_HARDWARE(intel);
-      intelWaitIrq(intel, irq);
+      if (batch->last_fence != NULL)
+	 dri_fence_wait(batch->last_fence);
    }
 
    /* Reset the buffer:
@@ -275,22 +252,25 @@ _intel_batchbuffer_flush(struct intel_batchbuffer *batch, const char *file,
    intel_batchbuffer_reset(batch);
 }
 
+void
+intel_batchbuffer_finish(struct intel_batchbuffer *batch)
+{
+   intel_batchbuffer_flush(batch);
+   if (batch->last_fence != NULL)
+      dri_fence_wait(batch->last_fence);
+}
+
 
 /*  This is the only way buffers get added to the validate list.
  */
 GLboolean
 intel_batchbuffer_emit_reloc(struct intel_batchbuffer *batch,
                              dri_bo *buffer,
-                             uint32_t read_domains, uint32_t write_domain,
-			     uint32_t delta)
+                             GLuint flags, GLuint delta)
 {
    int ret;
 
-   if (batch->ptr - batch->map > batch->buf->size)
-    _mesa_printf ("bad relocation ptr %p map %p offset %d size %d\n",
-		  batch->ptr, batch->map, batch->ptr - batch->map, batch->buf->size);
-   ret = intel_bo_emit_reloc(batch->buf, read_domains, write_domain,
-			     delta, batch->ptr - batch->map, buffer);
+   ret = dri_emit_reloc(batch->buf, flags, delta, batch->ptr - batch->map, buffer);
 
    /*
     * Using the old buffer offset, write in what the right data would be, in case
diff --git a/src/mesa/drivers/dri/intel/intel_batchbuffer.h b/src/mesa/drivers/dri/intel/intel_batchbuffer.h
index 52d6ecc223..0da602010e 100644
--- a/src/mesa/drivers/dri/intel/intel_batchbuffer.h
+++ b/src/mesa/drivers/dri/intel/intel_batchbuffer.h
@@ -4,7 +4,6 @@
 #include "mtypes.h"
 
 #include "dri_bufmgr.h"
-#include "intel_reg.h"
 
 struct intel_context;
 
@@ -41,8 +40,7 @@ struct intel_batchbuffer
    struct intel_context *intel;
 
    dri_bo *buf;
-
-   GLubyte *buffer;
+   dri_fence *last_fence;
 
    GLubyte *map;
    GLubyte *ptr;
@@ -60,6 +58,8 @@ struct intel_batchbuffer *intel_batchbuffer_alloc(struct intel_context
 void intel_batchbuffer_free(struct intel_batchbuffer *batch);
 
 
+void intel_batchbuffer_finish(struct intel_batchbuffer *batch);
+
 void _intel_batchbuffer_flush(struct intel_batchbuffer *batch,
 			      const char *file, int line);
 
@@ -82,16 +82,14 @@ void intel_batchbuffer_release_space(struct intel_batchbuffer *batch,
 
 GLboolean intel_batchbuffer_emit_reloc(struct intel_batchbuffer *batch,
                                        dri_bo *buffer,
-				       uint32_t read_domains,
-				       uint32_t write_domain,
-				       uint32_t offset);
+                                       GLuint flags, GLuint offset);
 
 /* Inline functions - might actually be better off with these
  * non-inlined.  Certainly better off switching all command packets to
  * be passed as structs rather than dwords, but that's a little bit of
  * work...
  */
-static INLINE GLint
+static INLINE GLuint
 intel_batchbuffer_space(struct intel_batchbuffer *batch)
 {
    return (batch->size - BATCH_RESERVED) - (batch->ptr - batch->map);
@@ -138,20 +136,12 @@ intel_batchbuffer_require_space(struct intel_batchbuffer *batch,
 
 #define OUT_BATCH(d)  intel_batchbuffer_emit_dword(intel->batch, d)
 
-#define OUT_RELOC(buf, read_domains, write_domain, delta) do {		\
+#define OUT_RELOC(buf, cliprect_mode, delta) do { 			\
    assert((delta) >= 0);						\
-   intel_batchbuffer_emit_reloc(intel->batch, buf,			\
-				read_domains, write_domain, delta);	\
+   intel_batchbuffer_emit_reloc(intel->batch, buf, cliprect_mode, delta); \
 } while (0)
 
 #define ADVANCE_BATCH() do { } while(0)
 
 
-static INLINE void
-intel_batchbuffer_emit_mi_flush(struct intel_batchbuffer *batch)
-{
-   intel_batchbuffer_require_space(batch, 4, IGNORE_CLIPRECTS);
-   intel_batchbuffer_emit_dword(batch, MI_FLUSH);
-}
-
 #endif
diff --git a/src/mesa/drivers/dri/intel/intel_blit.c b/src/mesa/drivers/dri/intel/intel_blit.c
index 7129a4ba91..25ac609f13 100644
--- a/src/mesa/drivers/dri/intel/intel_blit.c
+++ b/src/mesa/drivers/dri/intel/intel_blit.c
@@ -54,6 +54,7 @@ intelCopyBuffer(const __DRIdrawablePrivate * dPriv,
 
    struct intel_context *intel;
    const intelScreenPrivate *intelScreen;
+   int ret;
 
    DBG("%s\n", __FUNCTION__);
 
@@ -65,6 +66,14 @@ intelCopyBuffer(const __DRIdrawablePrivate * dPriv,
 
    intelScreen = intel->intelScreen;
 
+   if (intel->last_swap_fence) {
+      dri_fence_wait(intel->last_swap_fence);
+      dri_fence_unreference(intel->last_swap_fence);
+      intel->last_swap_fence = NULL;
+   }
+   intel->last_swap_fence = intel->first_swap_fence;
+   intel->first_swap_fence = NULL;
+
    /* The LOCK_HARDWARE is required for the cliprects.  Buffer offsets
     * should work regardless.
     */
@@ -80,7 +89,6 @@ intelCopyBuffer(const __DRIdrawablePrivate * dPriv,
       unsigned short src_x, src_y;
       int BR13, CMD;
       int i;
-      dri_bo *aper_array[3];
 
       src = intel_get_rb_region(&intel_fb->Base, BUFFER_BACK_LEFT);
       dst = intel_get_rb_region(&intel_fb->Base, BUFFER_FRONT_LEFT);
@@ -106,28 +114,26 @@ intelCopyBuffer(const __DRIdrawablePrivate * dPriv,
       }
 
 #ifndef I915
-      if (src->tiling != I915_TILING_NONE) {
+      if (src->tiled) {
 	 CMD |= XY_SRC_TILED;
 	 src_pitch /= 4;
       }
-      if (dst->tiling != I915_TILING_NONE) {
+      if (dst->tiled) {
 	 CMD |= XY_DST_TILED;
 	 dst_pitch /= 4;
       }
 #endif
       /* do space/cliprects check before going any further */
-      intel_batchbuffer_require_space(intel->batch, 8 * 4,
-				      REFERENCES_CLIPRECTS);
+      intel_batchbuffer_require_space(intel->batch, 8 * 4, REFERENCES_CLIPRECTS);
    again:
-      aper_array[0] = intel->batch->buf;
-      aper_array[1] = dst->buffer;
-      aper_array[2] = src->buffer;
-
-      if (dri_bufmgr_check_aperture_space(aper_array, 3) != 0) {
+      ret = dri_bufmgr_check_aperture_space(dst->buffer);
+      ret |= dri_bufmgr_check_aperture_space(src->buffer);
+      
+      if (ret) {
 	intel_batchbuffer_flush(intel->batch);
 	goto again;
       }
-
+      
       for (i = 0; i < nbox; i++, pbox++) {
 	 drm_clip_rect_t box = *pbox;
 
@@ -151,22 +157,19 @@ intelCopyBuffer(const __DRIdrawablePrivate * dPriv,
 	 OUT_BATCH((box.y1 << 16) | box.x1);
 	 OUT_BATCH((box.y2 << 16) | box.x2);
 
-	 OUT_RELOC(dst->buffer,
-		   I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
-		   0);
+	 OUT_RELOC(dst->buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE, 0);
 	 OUT_BATCH((src_y << 16) | src_x);
 	 OUT_BATCH(src_pitch);
-	 OUT_RELOC(src->buffer,
-		   I915_GEM_DOMAIN_RENDER, 0,
-		   0);
+	 OUT_RELOC(src->buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ, 0);
 	 ADVANCE_BATCH();
       }
 
-      /* Flush the rendering and the batch so that the results all land on the
-       * screen in a timely fashion.
-       */
-      intel_batchbuffer_emit_mi_flush(intel->batch);
+      if (intel->first_swap_fence)
+	 dri_fence_unreference(intel->first_swap_fence);
       intel_batchbuffer_flush(intel->batch);
+      intel->first_swap_fence = intel->batch->last_fence;
+      if (intel->first_swap_fence)
+	 dri_fence_reference(intel->first_swap_fence);
    }
 
    UNLOCK_HARDWARE(intel);
@@ -181,7 +184,7 @@ intelEmitFillBlit(struct intel_context *intel,
 		  GLshort dst_pitch,
 		  dri_bo *dst_buffer,
 		  GLuint dst_offset,
-		  uint32_t dst_tiling,
+		  GLboolean dst_tiled,
 		  GLshort x, GLshort y,
 		  GLshort w, GLshort h,
 		  GLuint color)
@@ -206,7 +209,7 @@ intelEmitFillBlit(struct intel_context *intel,
       return;
    }
 #ifndef I915
-   if (dst_tiling != I915_TILING_NONE) {
+   if (dst_tiled) {
       CMD |= XY_DST_TILED;
       dst_pitch /= 4;
    }
@@ -223,9 +226,7 @@ intelEmitFillBlit(struct intel_context *intel,
    OUT_BATCH(BR13 | dst_pitch);
    OUT_BATCH((y << 16) | x);
    OUT_BATCH(((y + h) << 16) | (x + w));
-   OUT_RELOC(dst_buffer,
-	     I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
-	     dst_offset);
+   OUT_RELOC(dst_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE, dst_offset);
    OUT_BATCH(color);
    ADVANCE_BATCH();
 }
@@ -262,11 +263,11 @@ intelEmitCopyBlit(struct intel_context *intel,
 		  GLshort src_pitch,
 		  dri_bo *src_buffer,
 		  GLuint src_offset,
-		  uint32_t src_tiling,
+		  GLboolean src_tiled,
 		  GLshort dst_pitch,
 		  dri_bo *dst_buffer,
 		  GLuint dst_offset,
-		  uint32_t dst_tiling,
+		  GLboolean dst_tiled,
 		  GLshort src_x, GLshort src_y,
 		  GLshort dst_x, GLshort dst_y,
 		  GLshort w, GLshort h,
@@ -275,19 +276,17 @@ intelEmitCopyBlit(struct intel_context *intel,
    GLuint CMD, BR13;
    int dst_y2 = dst_y + h;
    int dst_x2 = dst_x + w;
-   dri_bo *aper_array[3];
+   int ret;
    BATCH_LOCALS;
 
    /* do space/cliprects check before going any further */
    intel_batchbuffer_require_space(intel->batch, 8 * 4, NO_LOOP_CLIPRECTS);
  again:
-   aper_array[0] = intel->batch->buf;
-   aper_array[1] = dst_buffer;
-   aper_array[2] = src_buffer;
-
-   if (dri_bufmgr_check_aperture_space(aper_array, 3) != 0) {
-      intel_batchbuffer_flush(intel->batch);
-      goto again;
+   ret = dri_bufmgr_check_aperture_space(dst_buffer);
+   ret |= dri_bufmgr_check_aperture_space(src_buffer);
+   if (ret) {
+     intel_batchbuffer_flush(intel->batch);
+     goto again;
    }
 
    DBG("%s src:buf(%p)/%d+%d %d,%d dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n",
@@ -316,11 +315,11 @@ intelEmitCopyBlit(struct intel_context *intel,
    }
 
 #ifndef I915
-   if (dst_tiling != I915_TILING_NONE) {
+   if (dst_tiled) {
       CMD |= XY_DST_TILED;
       dst_pitch /= 4;
    }
-   if (src_tiling != I915_TILING_NONE) {
+   if (src_tiled) {
       CMD |= XY_SRC_TILED;
       src_pitch /= 4;
    }
@@ -346,13 +345,11 @@ intelEmitCopyBlit(struct intel_context *intel,
       OUT_BATCH(BR13 | dst_pitch);
       OUT_BATCH((dst_y << 16) | dst_x);
       OUT_BATCH((dst_y2 << 16) | dst_x2);
-      OUT_RELOC(dst_buffer,
-		I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
+      OUT_RELOC(dst_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
 		dst_offset);
       OUT_BATCH((src_y << 16) | src_x);
       OUT_BATCH(src_pitch);
-      OUT_RELOC(src_buffer,
-		I915_GEM_DOMAIN_RENDER, 0,
+      OUT_RELOC(src_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
 		src_offset);
       ADVANCE_BATCH();
    }
@@ -365,17 +362,14 @@ intelEmitCopyBlit(struct intel_context *intel,
       OUT_BATCH(BR13 | ((uint16_t)dst_pitch));
       OUT_BATCH((0 << 16) | dst_x);
       OUT_BATCH((h << 16) | dst_x2);
-      OUT_RELOC(dst_buffer,
-		I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
+      OUT_RELOC(dst_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
 		dst_offset + dst_y * dst_pitch);
       OUT_BATCH((0 << 16) | src_x);
       OUT_BATCH(src_pitch);
-      OUT_RELOC(src_buffer,
-		I915_GEM_DOMAIN_RENDER, 0,
+      OUT_RELOC(src_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
 		src_offset + src_y * src_pitch);
       ADVANCE_BATCH();
    }
-   intel_batchbuffer_emit_mi_flush(intel->batch);
 }
 
 
@@ -519,7 +513,7 @@ intelClearWithBlit(GLcontext *ctx, GLbitfield mask)
                }
 
 #ifndef I915
-	       if (irb_region->tiling != I915_TILING_NONE) {
+	       if (irb_region->tiled) {
 		  CMD |= XY_DST_TILED;
 		  pitch /= 4;
 	       }
@@ -547,8 +541,7 @@ intelClearWithBlit(GLcontext *ctx, GLbitfield mask)
                OUT_BATCH(BR13);
                OUT_BATCH((b.y1 << 16) | b.x1);
                OUT_BATCH((b.y2 << 16) | b.x2);
-               OUT_RELOC(write_buffer,
-			 I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
+               OUT_RELOC(write_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
                          irb_region->draw_offset);
                OUT_BATCH(clearVal);
                ADVANCE_BATCH();
@@ -556,7 +549,7 @@ intelClearWithBlit(GLcontext *ctx, GLbitfield mask)
             }
          }
       }
-      intel_batchbuffer_emit_mi_flush(intel->batch);
+      intel_batchbuffer_flush(intel->batch);
    }
 
    UNLOCK_HARDWARE(intel);
@@ -570,7 +563,7 @@ intelEmitImmediateColorExpandBlit(struct intel_context *intel,
 				  GLshort dst_pitch,
 				  dri_bo *dst_buffer,
 				  GLuint dst_offset,
-				  uint32_t dst_tiling,
+				  GLboolean dst_tiled,
 				  GLshort x, GLshort y,
 				  GLshort w, GLshort h,
 				  GLenum logic_op)
@@ -594,13 +587,13 @@ intelEmitImmediateColorExpandBlit(struct intel_context *intel,
 				    (8 * 4) +
 				    (3 * 4) +
 				    dwords,
-				    REFERENCES_CLIPRECTS );
+				    NO_LOOP_CLIPRECTS );
 
    opcode = XY_SETUP_BLT_CMD;
    if (cpp == 4)
       opcode |= XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB;
 #ifndef I915
-   if (dst_tiling != I915_TILING_NONE) {
+   if (dst_tiled) {
       opcode |= XY_DST_TILED;
       dst_pitch /= 4;
    }
@@ -613,17 +606,15 @@ intelEmitImmediateColorExpandBlit(struct intel_context *intel,
       br13 |= BR13_8888;
 
    blit_cmd = XY_TEXT_IMMEDIATE_BLIT_CMD | XY_TEXT_BYTE_PACKED; /* packing? */
-   if (dst_tiling != I915_TILING_NONE)
+   if (dst_tiled)
       blit_cmd |= XY_DST_TILED;
 
-   BEGIN_BATCH(8 + 3, REFERENCES_CLIPRECTS);
+   BEGIN_BATCH(8 + 3, NO_LOOP_CLIPRECTS);
    OUT_BATCH(opcode);
    OUT_BATCH(br13);
    OUT_BATCH((0 << 16) | 0); /* clip x1, y1 */
    OUT_BATCH((100 << 16) | 100); /* clip x2, y2 */
-   OUT_RELOC(dst_buffer,
-	     I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
-	     dst_offset);
+   OUT_RELOC(dst_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE, dst_offset);
    OUT_BATCH(0); /* bg */
    OUT_BATCH(fg_color); /* fg */
    OUT_BATCH(0); /* pattern base addr */
@@ -636,7 +627,5 @@ intelEmitImmediateColorExpandBlit(struct intel_context *intel,
    intel_batchbuffer_data( intel->batch,
 			   src_bits,
 			   dwords * 4,
-			   REFERENCES_CLIPRECTS );
-
-   intel_batchbuffer_emit_mi_flush(intel->batch);
+			   NO_LOOP_CLIPRECTS );
 }
diff --git a/src/mesa/drivers/dri/intel/intel_blit.h b/src/mesa/drivers/dri/intel/intel_blit.h
index 0881cc4fdc..fc0620caba 100644
--- a/src/mesa/drivers/dri/intel/intel_blit.h
+++ b/src/mesa/drivers/dri/intel/intel_blit.h
@@ -42,11 +42,11 @@ extern void intelEmitCopyBlit(struct intel_context *intel,
                               GLshort src_pitch,
                               dri_bo *src_buffer,
                               GLuint src_offset,
-			      uint32_t src_tiling,
+			      GLboolean src_tiled,
                               GLshort dst_pitch,
                               dri_bo *dst_buffer,
                               GLuint dst_offset,
-			      uint32_t dst_tiling,
+			      GLboolean dst_tiled,
                               GLshort srcx, GLshort srcy,
                               GLshort dstx, GLshort dsty,
                               GLshort w, GLshort h,
@@ -57,7 +57,7 @@ extern void intelEmitFillBlit(struct intel_context *intel,
                               GLshort dst_pitch,
                               dri_bo *dst_buffer,
                               GLuint dst_offset,
-			      uint32_t dst_tiling,
+			      GLboolean dst_tiled,
                               GLshort x, GLshort y,
                               GLshort w, GLshort h, GLuint color);
 
@@ -69,7 +69,7 @@ intelEmitImmediateColorExpandBlit(struct intel_context *intel,
 				  GLshort dst_pitch,
 				  dri_bo *dst_buffer,
 				  GLuint dst_offset,
-				  uint32_t dst_tiling,
+				  GLboolean dst_tiled,
 				  GLshort x, GLshort y,
 				  GLshort w, GLshort h,
 				  GLenum logic_op);
diff --git a/src/mesa/drivers/dri/intel/intel_buffer_objects.c b/src/mesa/drivers/dri/intel/intel_buffer_objects.c
index 1923a21516..951b8cbfb7 100644
--- a/src/mesa/drivers/dri/intel/intel_buffer_objects.c
+++ b/src/mesa/drivers/dri/intel/intel_buffer_objects.c
@@ -32,7 +32,6 @@
 
 #include "intel_context.h"
 #include "intel_buffer_objects.h"
-#include "intel_batchbuffer.h"
 #include "intel_regions.h"
 #include "dri_bufmgr.h"
 
@@ -46,7 +45,8 @@ intel_bufferobj_alloc_buffer(struct intel_context *intel,
 			     struct intel_buffer_object *intel_obj)
 {
    intel_obj->buffer = dri_bo_alloc(intel->bufmgr, "bufferobj",
-				    intel_obj->Base.Size, 64);
+				    intel_obj->Base.Size, 64,
+				    DRM_BO_FLAG_MEM_LOCAL | DRM_BO_FLAG_CACHED | DRM_BO_FLAG_CACHED_MAPPED);
 }
 
 /**
diff --git a/src/mesa/drivers/dri/intel/intel_buffers.c b/src/mesa/drivers/dri/intel/intel_buffers.c
index 77352e6a53..75542a9775 100644
--- a/src/mesa/drivers/dri/intel/intel_buffers.c
+++ b/src/mesa/drivers/dri/intel/intel_buffers.c
@@ -819,8 +819,6 @@ intelSwapBuffers(__DRIdrawablePrivate * dPriv)
 
 	 intel_fb->swap_ust = ust;
       }
-      drmCommandNone(intel->driFd, DRM_I915_GEM_THROTTLE);
-
    }
    else {
       /* XXX this shouldn't be an error but we can't handle it for now */
diff --git a/src/mesa/drivers/dri/intel/intel_bufmgr_ttm.c b/src/mesa/drivers/dri/intel/intel_bufmgr_ttm.c
new file mode 100644
index 0000000000..194814e8fb
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_bufmgr_ttm.c
@@ -0,0 +1,1122 @@
+/**************************************************************************
+ *
+ * Copyright © 2007 Red Hat Inc.
+ * Copyright © 2007 Intel Corporation
+ * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS 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.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ *
+ **************************************************************************/
+/*
+ * Authors: Thomas Hellström <thomas-at-tungstengraphics-dot-com>
+ *          Keith Whitwell <keithw-at-tungstengraphics-dot-com>
+ *	    Eric Anholt <eric@anholt.net>
+ *	    Dave Airlie <airlied@linux.ie>
+ */
+
+#include <xf86drm.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+#include <assert.h>
+
+#include "errno.h"
+#include "mtypes.h"
+#include "dri_bufmgr.h"
+#include "string.h"
+#include "imports.h"
+
+#include "i915_drm.h"
+
+#include "intel_bufmgr_ttm.h"
+#ifdef TTM_API
+
+#define DBG(...) do {					\
+   if (bufmgr_ttm->bufmgr.debug)			\
+      fprintf(stderr, __VA_ARGS__);			\
+} while (0)
+
+/*
+ * These bits are always specified in each validation
+ * request. Other bits are not supported at this point
+ * as it would require a bit of investigation to figure
+ * out what mask value should be used.
+ */
+#define INTEL_BO_MASK  (DRM_BO_MASK_MEM | \
+			DRM_BO_FLAG_READ | \
+			DRM_BO_FLAG_WRITE | \
+			DRM_BO_FLAG_EXE)
+
+struct intel_validate_entry {
+    dri_bo *bo;
+    struct drm_i915_op_arg bo_arg;
+};
+
+struct dri_ttm_bo_bucket_entry {
+   drmBO drm_bo;
+   struct dri_ttm_bo_bucket_entry *next;
+};
+
+struct dri_ttm_bo_bucket {
+   struct dri_ttm_bo_bucket_entry *head;
+   struct dri_ttm_bo_bucket_entry **tail;
+   /**
+    * Limit on the number of entries in this bucket.
+    *
+    * 0 means that this caching at this bucket size is disabled.
+    * -1 means that there is no limit to caching at this size.
+    */
+   int max_entries;
+   int num_entries;
+};
+
+/* Arbitrarily chosen, 16 means that the maximum size we'll cache for reuse
+ * is 1 << 16 pages, or 256MB.
+ */
+#define INTEL_TTM_BO_BUCKETS	16
+typedef struct _dri_bufmgr_ttm {
+    dri_bufmgr bufmgr;
+
+    int fd;
+    unsigned int fence_type;
+    unsigned int fence_type_flush;
+
+    uint32_t max_relocs;
+
+    struct intel_validate_entry *validate_array;
+    int validate_array_size;
+    int validate_count;
+
+    /** Array of lists of cached drmBOs of power-of-two sizes */
+    struct dri_ttm_bo_bucket cache_bucket[INTEL_TTM_BO_BUCKETS];
+} dri_bufmgr_ttm;
+
+/**
+ * Private information associated with a relocation that isn't already stored
+ * in the relocation buffer to be passed to the kernel.
+ */
+struct dri_ttm_reloc {
+    dri_bo *target_buf;
+    uint64_t validate_flags;
+    /** Offset of target_buf after last execution of this relocation entry. */
+    unsigned int last_target_offset;
+};
+
+typedef struct _dri_bo_ttm {
+    dri_bo bo;
+
+    int refcount;
+    unsigned int map_count;
+    drmBO drm_bo;
+    const char *name;
+
+    uint64_t last_flags;
+
+    /**
+     * Index of the buffer within the validation list while preparing a
+     * batchbuffer execution.
+     */
+    int validate_index;
+
+    /** DRM buffer object containing relocation list */
+    uint32_t *reloc_buf_data;
+    struct dri_ttm_reloc *relocs;
+
+    /**
+     * Indicates that the buffer may be shared with other processes, so we
+     * can't hold maps beyond when the user does.
+     */
+    GLboolean shared;
+
+    GLboolean delayed_unmap;
+    /* Virtual address from the dri_bo_map whose unmap was delayed. */
+    void *saved_virtual;
+} dri_bo_ttm;
+
+typedef struct _dri_fence_ttm
+{
+    dri_fence fence;
+
+    int refcount;
+    const char *name;
+    drmFence drm_fence;
+} dri_fence_ttm;
+
+static int
+logbase2(int n)
+{
+   GLint i = 1;
+   GLint log2 = 0;
+
+   while (n > i) {
+      i *= 2;
+      log2++;
+   }
+
+   return log2;
+}
+
+static struct dri_ttm_bo_bucket *
+dri_ttm_bo_bucket_for_size(dri_bufmgr_ttm *bufmgr_ttm, unsigned long size)
+{
+    int i;
+
+    /* We only do buckets in power of two increments */
+    if ((size & (size - 1)) != 0)
+	return NULL;
+
+    /* We should only see sizes rounded to pages. */
+    assert((size % 4096) == 0);
+
+    /* We always allocate in units of pages */
+    i = ffs(size / 4096) - 1;
+    if (i >= INTEL_TTM_BO_BUCKETS)
+	return NULL;
+
+    return &bufmgr_ttm->cache_bucket[i];
+}
+
+
+static void dri_ttm_dump_validation_list(dri_bufmgr_ttm *bufmgr_ttm)
+{
+    int i, j;
+
+    for (i = 0; i < bufmgr_ttm->validate_count; i++) {
+	dri_bo *bo = bufmgr_ttm->validate_array[i].bo;
+	dri_bo_ttm *bo_ttm = (dri_bo_ttm *)bo;
+
+	if (bo_ttm->reloc_buf_data != NULL) {
+	    for (j = 0; j < (bo_ttm->reloc_buf_data[0] & 0xffff); j++) {
+		uint32_t *reloc_entry = bo_ttm->reloc_buf_data +
+		    I915_RELOC_HEADER +
+		    j * I915_RELOC0_STRIDE;
+		dri_bo *target_bo = bo_ttm->relocs[j].target_buf;
+		dri_bo_ttm *target_ttm = (dri_bo_ttm *)target_bo;
+
+		DBG("%2d: %s@0x%08x -> %s@0x%08lx + 0x%08x\n",
+		    i,
+		    bo_ttm->name, reloc_entry[0],
+		    target_ttm->name, target_bo->offset,
+		    reloc_entry[1]);
+	    }
+	} else {
+	    DBG("%2d: %s\n", i, bo_ttm->name);
+	}
+    }
+}
+
+/**
+ * Adds the given buffer to the list of buffers to be validated (moved into the
+ * appropriate memory type) with the next batch submission.
+ *
+ * If a buffer is validated multiple times in a batch submission, it ends up
+ * with the intersection of the memory type flags and the union of the
+ * access flags.
+ */
+static void
+intel_add_validate_buffer(dri_bo *buf,
+			  uint64_t flags)
+{
+    dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)buf->bufmgr;
+    dri_bo_ttm *ttm_buf = (dri_bo_ttm *)buf;
+
+    /* If we delayed doing an unmap to mitigate map/unmap syscall thrashing,
+     * do that now.
+     */
+    if (ttm_buf->delayed_unmap) {
+	drmBOUnmap(bufmgr_ttm->fd, &ttm_buf->drm_bo);
+	ttm_buf->delayed_unmap = GL_FALSE;
+    }
+
+    if (ttm_buf->validate_index == -1) {
+	struct intel_validate_entry *entry;
+	struct drm_i915_op_arg *arg;
+	struct drm_bo_op_req *req;
+	int index;
+
+	/* Extend the array of validation entries as necessary. */
+	if (bufmgr_ttm->validate_count == bufmgr_ttm->validate_array_size) {
+	    int i, new_size = bufmgr_ttm->validate_array_size * 2;
+
+	    if (new_size == 0)
+		new_size = 5;
+
+	    bufmgr_ttm->validate_array =
+	       realloc(bufmgr_ttm->validate_array,
+		       sizeof(struct intel_validate_entry) * new_size);
+	    bufmgr_ttm->validate_array_size = new_size;
+
+	    /* Update pointers for realloced mem. */
+	    for (i = 0; i < bufmgr_ttm->validate_count - 1; i++) {
+	       bufmgr_ttm->validate_array[i].bo_arg.next = (unsigned long)
+		  &bufmgr_ttm->validate_array[i + 1].bo_arg;
+	    }
+	}
+
+	/* Pick out the new array entry for ourselves */
+	index = bufmgr_ttm->validate_count;
+	ttm_buf->validate_index = index;
+	entry = &bufmgr_ttm->validate_array[index];
+	bufmgr_ttm->validate_count++;
+
+	/* Fill in array entry */
+	entry->bo = buf;
+	dri_bo_reference(buf);
+
+	/* Fill in kernel arg */
+	arg = &entry->bo_arg;
+	req = &arg->d.req;
+
+	memset(arg, 0, sizeof(*arg));
+	req->bo_req.handle = ttm_buf->drm_bo.handle;
+	req->op = drm_bo_validate;
+	req->bo_req.flags = flags;
+	req->bo_req.hint = 0;
+#ifdef DRM_BO_HINT_PRESUMED_OFFSET
+	/* PRESUMED_OFFSET indicates that all relocations pointing at this
+	 * buffer have the correct offset.  If any of our relocations don't,
+	 * this flag will be cleared off the buffer later in the relocation
+	 * processing.
+	 */
+	req->bo_req.hint |= DRM_BO_HINT_PRESUMED_OFFSET;
+	req->bo_req.presumed_offset = buf->offset;
+#endif
+	req->bo_req.mask = INTEL_BO_MASK;
+	req->bo_req.fence_class = 0; /* Backwards compat. */
+
+	if (ttm_buf->reloc_buf_data != NULL)
+ 	    arg->reloc_ptr = (unsigned long)(void *)ttm_buf->reloc_buf_data;
+	else
+	    arg->reloc_ptr = 0;
+
+	/* Hook up the linked list of args for the kernel */
+	arg->next = 0;
+	if (index != 0) {
+	    bufmgr_ttm->validate_array[index - 1].bo_arg.next =
+		(unsigned long)arg;
+	}
+    } else {
+	struct intel_validate_entry *entry =
+	    &bufmgr_ttm->validate_array[ttm_buf->validate_index];
+	struct drm_i915_op_arg *arg = &entry->bo_arg;
+	struct drm_bo_op_req *req = &arg->d.req;
+	uint64_t memFlags = req->bo_req.flags & flags & DRM_BO_MASK_MEM;
+	uint64_t modeFlags = (req->bo_req.flags | flags) & ~DRM_BO_MASK_MEM;
+
+	/* Buffer was already in the validate list.  Extend its flags as
+	 * necessary.
+	 */
+
+	if (memFlags == 0) {
+	    fprintf(stderr,
+		    "%s: No shared memory types between "
+		    "0x%16llx and 0x%16llx\n",
+		    __FUNCTION__, req->bo_req.flags, flags);
+	    abort();
+	}
+	if (flags & ~INTEL_BO_MASK) {
+	    fprintf(stderr,
+		    "%s: Flags bits 0x%16llx are not supposed to be used in a relocation\n",
+		    __FUNCTION__, flags & ~INTEL_BO_MASK);
+	    abort();
+	}
+	req->bo_req.flags = memFlags | modeFlags;
+    }
+}
+
+
+#define RELOC_BUF_SIZE(x) ((I915_RELOC_HEADER + x * I915_RELOC0_STRIDE) * \
+	sizeof(uint32_t))
+
+static int
+intel_setup_reloc_list(dri_bo *bo)
+{
+    dri_bo_ttm *bo_ttm = (dri_bo_ttm *)bo;
+    dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)bo->bufmgr;
+
+    bo_ttm->relocs = calloc(bufmgr_ttm->max_relocs,
+			    sizeof(struct dri_ttm_reloc));
+    bo_ttm->reloc_buf_data = calloc(1, RELOC_BUF_SIZE(bufmgr_ttm->max_relocs));
+
+    /* Initialize the relocation list with the header:
+     * DWORD 0: relocation count
+     * DWORD 1: relocation type  
+     * DWORD 2+3: handle to next relocation list (currently none) 64-bits
+     */
+    bo_ttm->reloc_buf_data[0] = 0;
+    bo_ttm->reloc_buf_data[1] = I915_RELOC_TYPE_0;
+    bo_ttm->reloc_buf_data[2] = 0;
+    bo_ttm->reloc_buf_data[3] = 0;
+
+    return 0;
+}
+
+#if 0
+int
+driFenceSignaled(DriFenceObject * fence, unsigned type)
+{
+    int signaled;
+    int ret;
+
+    if (fence == NULL)
+	return GL_TRUE;
+
+    ret = drmFenceSignaled(bufmgr_ttm->fd, &fence->fence, type, &signaled);
+    BM_CKFATAL(ret);
+    return signaled;
+}
+#endif
+
+static dri_bo *
+dri_ttm_alloc(dri_bufmgr *bufmgr, const char *name,
+	      unsigned long size, unsigned int alignment,
+	      uint64_t location_mask)
+{
+    dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)bufmgr;
+    dri_bo_ttm *ttm_buf;
+    unsigned int pageSize = getpagesize();
+    int ret;
+    uint64_t flags;
+    unsigned int hint;
+    unsigned long alloc_size;
+    struct dri_ttm_bo_bucket *bucket;
+    GLboolean alloc_from_cache = GL_FALSE;
+
+    ttm_buf = calloc(1, sizeof(*ttm_buf));
+    if (!ttm_buf)
+	return NULL;
+
+    /* The mask argument doesn't do anything for us that we want other than
+     * determine which pool (TTM or local) the buffer is allocated into, so
+     * just pass all of the allocation class flags.
+     */
+    flags = location_mask | DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE |
+	DRM_BO_FLAG_EXE;
+    /* No hints we want to use. */
+    hint = 0;
+
+    /* Round the allocated size up to a power of two number of pages. */
+    alloc_size = 1 << logbase2(size);
+    if (alloc_size < pageSize)
+	alloc_size = pageSize;
+    bucket = dri_ttm_bo_bucket_for_size(bufmgr_ttm, alloc_size);
+
+    /* If we don't have caching at this size, don't actually round the
+     * allocation up.
+     */
+    if (bucket == NULL || bucket->max_entries == 0)
+	alloc_size = size;
+
+    /* Get a buffer out of the cache if available */
+    if (bucket != NULL && bucket->num_entries > 0) {
+	struct dri_ttm_bo_bucket_entry *entry = bucket->head;
+	int busy;
+
+	/* Check if the buffer is still in flight.  If not, reuse it. */
+	ret = drmBOBusy(bufmgr_ttm->fd, &entry->drm_bo, &busy);
+	alloc_from_cache = (ret == 0 && busy == 0);
+
+	if (alloc_from_cache) {
+	    bucket->head = entry->next;
+	    if (entry->next == NULL)
+		bucket->tail = &bucket->head;
+	    bucket->num_entries--;
+
+	    ttm_buf->drm_bo = entry->drm_bo;
+	    free(entry);
+	}
+    }
+
+    if (!alloc_from_cache) {
+	ret = drmBOCreate(bufmgr_ttm->fd, alloc_size, alignment / pageSize,
+			  NULL, flags, hint, &ttm_buf->drm_bo);
+	if (ret != 0) {
+	    free(ttm_buf);
+	    return NULL;
+	}
+    }
+
+    ttm_buf->bo.size = size;
+    ttm_buf->bo.offset = ttm_buf->drm_bo.offset;
+    ttm_buf->bo.virtual = NULL;
+    ttm_buf->bo.bufmgr = bufmgr;
+    ttm_buf->name = name;
+    ttm_buf->refcount = 1;
+    ttm_buf->reloc_buf_data = NULL;
+    ttm_buf->relocs = NULL;
+    ttm_buf->last_flags = ttm_buf->drm_bo.flags;
+    ttm_buf->shared = GL_FALSE;
+    ttm_buf->delayed_unmap = GL_FALSE;
+    ttm_buf->validate_index = -1;
+
+    DBG("bo_create: %p (%s) %ldb\n", &ttm_buf->bo, ttm_buf->name, size);
+
+    return &ttm_buf->bo;
+}
+
+/* Our TTM backend doesn't allow creation of static buffers, as that requires
+ * privelege for the non-fake case, and the lock in the fake case where we were
+ * working around the X Server not creating buffers and passing handles to us.
+ */
+static dri_bo *
+dri_ttm_alloc_static(dri_bufmgr *bufmgr, const char *name,
+		     unsigned long offset, unsigned long size, void *virtual,
+		     uint64_t location_mask)
+{
+    return NULL;
+}
+
+/**
+ * Returns a dri_bo wrapping the given buffer object handle.
+ *
+ * This can be used when one application needs to pass a buffer object
+ * to another.
+ */
+dri_bo *
+intel_ttm_bo_create_from_handle(dri_bufmgr *bufmgr, const char *name,
+			      unsigned int handle)
+{
+    dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)bufmgr;
+    dri_bo_ttm *ttm_buf;
+    int ret;
+
+    ttm_buf = calloc(1, sizeof(*ttm_buf));
+    if (!ttm_buf)
+	return NULL;
+
+    ret = drmBOReference(bufmgr_ttm->fd, handle, &ttm_buf->drm_bo);
+    if (ret != 0) {
+       fprintf(stderr, "Couldn't reference %s handle 0x%08x: %s\n",
+	       name, handle, strerror(-ret));
+	free(ttm_buf);
+	return NULL;
+    }
+    ttm_buf->bo.size = ttm_buf->drm_bo.size;
+    ttm_buf->bo.offset = ttm_buf->drm_bo.offset;
+    ttm_buf->bo.virtual = NULL;
+    ttm_buf->bo.bufmgr = bufmgr;
+    ttm_buf->name = name;
+    ttm_buf->refcount = 1;
+    ttm_buf->reloc_buf_data = NULL;
+    ttm_buf->relocs = NULL;
+    ttm_buf->last_flags = ttm_buf->drm_bo.flags;
+    ttm_buf->shared = GL_TRUE;
+    ttm_buf->delayed_unmap = GL_FALSE;
+    ttm_buf->validate_index = -1;
+
+    DBG("bo_create_from_handle: %p %08x (%s)\n",
+	&ttm_buf->bo, handle, ttm_buf->name);
+
+    return &ttm_buf->bo;
+}
+
+static void
+dri_ttm_bo_reference(dri_bo *buf)
+{
+    dri_bo_ttm *ttm_buf = (dri_bo_ttm *)buf;
+
+    ttm_buf->refcount++;
+}
+
+static void
+dri_ttm_bo_unreference(dri_bo *buf)
+{
+    dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)buf->bufmgr;
+    dri_bo_ttm *ttm_buf = (dri_bo_ttm *)buf;
+
+    if (!buf)
+	return;
+
+    if (--ttm_buf->refcount == 0) {
+	struct dri_ttm_bo_bucket *bucket;
+	int ret;
+
+	assert(ttm_buf->map_count == 0);
+
+	if (ttm_buf->reloc_buf_data) {
+	    int i;
+
+	    /* Unreference all the target buffers */
+	    for (i = 0; i < (ttm_buf->reloc_buf_data[0] & 0xffff); i++)
+		 dri_bo_unreference(ttm_buf->relocs[i].target_buf);
+	    free(ttm_buf->relocs);
+
+	    /* Free the kernel BO containing relocation entries */
+	    free(ttm_buf->reloc_buf_data);
+	    ttm_buf->reloc_buf_data = NULL;
+	}
+
+	if (ttm_buf->delayed_unmap) {
+	    int ret = drmBOUnmap(bufmgr_ttm->fd, &ttm_buf->drm_bo);
+
+	    if (ret != 0) {
+		fprintf(stderr, "%s:%d: Error unmapping buffer %s: %s.\n",
+			__FILE__, __LINE__, ttm_buf->name, strerror(-ret));
+	   }
+	}
+
+	bucket = dri_ttm_bo_bucket_for_size(bufmgr_ttm, ttm_buf->drm_bo.size);
+	/* Put the buffer into our internal cache for reuse if we can. */
+	if (!ttm_buf->shared &&
+	    bucket != NULL &&
+	    (bucket->max_entries == -1 ||
+	     (bucket->max_entries > 0 &&
+	      bucket->num_entries < bucket->max_entries)))
+	{
+	    struct dri_ttm_bo_bucket_entry *entry;
+
+	    entry = calloc(1, sizeof(*entry));
+	    entry->drm_bo = ttm_buf->drm_bo;
+
+	    entry->next = NULL;
+	    *bucket->tail = entry;
+	    bucket->tail = &entry->next;
+	    bucket->num_entries++;
+	} else {
+	    /* Decrement the kernel refcount for the buffer. */
+	    ret = drmBOUnreference(bufmgr_ttm->fd, &ttm_buf->drm_bo);
+	    if (ret != 0) {
+	       fprintf(stderr, "drmBOUnreference failed (%s): %s\n",
+		       ttm_buf->name, strerror(-ret));
+	    }
+	}
+
+	DBG("bo_unreference final: %p (%s)\n", &ttm_buf->bo, ttm_buf->name);
+
+	free(buf);
+	return;
+    }
+}
+
+static int
+dri_ttm_bo_map(dri_bo *buf, GLboolean write_enable)
+{
+    dri_bufmgr_ttm *bufmgr_ttm;
+    dri_bo_ttm *ttm_buf = (dri_bo_ttm *)buf;
+    uint64_t flags;
+    int ret;
+
+    bufmgr_ttm = (dri_bufmgr_ttm *)buf->bufmgr;
+
+    flags = DRM_BO_FLAG_READ;
+    if (write_enable)
+	flags |= DRM_BO_FLAG_WRITE;
+
+    /* Allow recursive mapping. Mesa may recursively map buffers with
+     * nested display loops.
+     */
+    if (ttm_buf->map_count++ != 0)
+	return 0;
+
+    assert(buf->virtual == NULL);
+
+    DBG("bo_map: %p (%s)\n", &ttm_buf->bo, ttm_buf->name);
+
+    /* XXX: What about if we're upgrading from READ to WRITE? */
+    if (ttm_buf->delayed_unmap) {
+	buf->virtual = ttm_buf->saved_virtual;
+	return 0;
+    }
+
+    ret = drmBOMap(bufmgr_ttm->fd, &ttm_buf->drm_bo, flags, 0, &buf->virtual);
+    if (ret != 0) {
+        fprintf(stderr, "%s:%d: Error mapping buffer %s: %s .\n",
+		__FILE__, __LINE__, ttm_buf->name, strerror(-ret));
+    }
+
+    return ret;
+}
+
+static int
+dri_ttm_bo_unmap(dri_bo *buf)
+{
+    dri_bufmgr_ttm *bufmgr_ttm;
+    dri_bo_ttm *ttm_buf = (dri_bo_ttm *)buf;
+    int ret;
+
+    if (buf == NULL)
+	return 0;
+
+    assert(ttm_buf->map_count != 0);
+    if (--ttm_buf->map_count != 0)
+	return 0;
+
+    bufmgr_ttm = (dri_bufmgr_ttm *)buf->bufmgr;
+
+    assert(buf->virtual != NULL);
+
+    DBG("bo_unmap: %p (%s)\n", &ttm_buf->bo, ttm_buf->name);
+
+    if (!ttm_buf->shared) {
+	ttm_buf->saved_virtual = buf->virtual;
+	ttm_buf->delayed_unmap = GL_TRUE;
+	buf->virtual = NULL;
+
+	return 0;
+    }
+
+    buf->virtual = NULL;
+
+    ret = drmBOUnmap(bufmgr_ttm->fd, &ttm_buf->drm_bo);
+    if (ret != 0) {
+        fprintf(stderr, "%s:%d: Error unmapping buffer %s: %s.\n",
+		__FILE__, __LINE__, ttm_buf->name, strerror(-ret));
+    }
+
+    return ret;
+}
+
+/**
+ * Returns a dri_bo wrapping the given buffer object handle.
+ *
+ * This can be used when one application needs to pass a buffer object
+ * to another.
+ */
+dri_fence *
+intel_ttm_fence_create_from_arg(dri_bufmgr *bufmgr, const char *name,
+				drm_fence_arg_t *arg)
+{
+    dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)bufmgr;
+    dri_fence_ttm *ttm_fence;
+
+    ttm_fence = malloc(sizeof(*ttm_fence));
+    if (!ttm_fence)
+	return NULL;
+
+    ttm_fence->drm_fence.handle = arg->handle;
+    ttm_fence->drm_fence.fence_class = arg->fence_class;
+    ttm_fence->drm_fence.type = arg->type;
+    ttm_fence->drm_fence.flags = arg->flags;
+    ttm_fence->drm_fence.signaled = 0;
+    ttm_fence->drm_fence.sequence = arg->sequence;
+
+    ttm_fence->fence.bufmgr = bufmgr;
+    ttm_fence->name = name;
+    ttm_fence->refcount = 1;
+
+    DBG("fence_create_from_handle: %p (%s)\n",
+	&ttm_fence->fence, ttm_fence->name);
+
+    return &ttm_fence->fence;
+}
+
+
+static void
+dri_ttm_fence_reference(dri_fence *fence)
+{
+    dri_fence_ttm *fence_ttm = (dri_fence_ttm *)fence;
+    dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)fence->bufmgr;
+
+    ++fence_ttm->refcount;
+    DBG("fence_reference: %p (%s)\n", &fence_ttm->fence, fence_ttm->name);
+}
+
+static void
+dri_ttm_fence_unreference(dri_fence *fence)
+{
+    dri_fence_ttm *fence_ttm = (dri_fence_ttm *)fence;
+    dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)fence->bufmgr;
+
+    if (!fence)
+	return;
+
+    DBG("fence_unreference: %p (%s)\n", &fence_ttm->fence, fence_ttm->name);
+
+    if (--fence_ttm->refcount == 0) {
+	int ret;
+
+	ret = drmFenceUnreference(bufmgr_ttm->fd, &fence_ttm->drm_fence);
+	if (ret != 0) {
+	    fprintf(stderr, "drmFenceUnreference failed (%s): %s\n",
+		    fence_ttm->name, strerror(-ret));
+	}
+
+	free(fence);
+	return;
+    }
+}
+
+static void
+dri_ttm_fence_wait(dri_fence *fence)
+{
+    dri_fence_ttm *fence_ttm = (dri_fence_ttm *)fence;
+    dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)fence->bufmgr;
+    int ret;
+
+    ret = drmFenceWait(bufmgr_ttm->fd, DRM_FENCE_FLAG_WAIT_LAZY, &fence_ttm->drm_fence, 0);
+    if (ret != 0) {
+        fprintf(stderr, "%s:%d: Error waiting for fence %s: %s.\n",
+		__FILE__, __LINE__, fence_ttm->name, strerror(-ret));
+	abort();
+    }
+
+    DBG("fence_wait: %p (%s)\n", &fence_ttm->fence, fence_ttm->name);
+}
+
+static void
+dri_bufmgr_ttm_destroy(dri_bufmgr *bufmgr)
+{
+    dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)bufmgr;
+    int i;
+
+    free(bufmgr_ttm->validate_array);
+
+    /* Free any cached buffer objects we were going to reuse */
+    for (i = 0; i < INTEL_TTM_BO_BUCKETS; i++) {
+	struct dri_ttm_bo_bucket *bucket = &bufmgr_ttm->cache_bucket[i];
+	struct dri_ttm_bo_bucket_entry *entry;
+
+	while ((entry = bucket->head) != NULL) {
+	    int ret;
+
+	    bucket->head = entry->next;
+	    if (entry->next == NULL)
+		bucket->tail = &bucket->head;
+	    bucket->num_entries--;
+
+	    /* Decrement the kernel refcount for the buffer. */
+	    ret = drmBOUnreference(bufmgr_ttm->fd, &entry->drm_bo);
+	    if (ret != 0) {
+	       fprintf(stderr, "drmBOUnreference failed: %s\n",
+		       strerror(-ret));
+	    }
+
+	    free(entry);
+	}
+    }
+
+    free(bufmgr);
+}
+
+/**
+ * Adds the target buffer to the validation list and adds the relocation
+ * to the reloc_buffer's relocation list.
+ *
+ * The relocation entry at the given offset must already contain the
+ * precomputed relocation value, because the kernel will optimize out
+ * the relocation entry write when the buffer hasn't moved from the
+ * last known offset in target_buf.
+ */
+static int
+dri_ttm_emit_reloc(dri_bo *reloc_buf, uint64_t flags, GLuint delta,
+		   GLuint offset, dri_bo *target_buf)
+{
+    dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)reloc_buf->bufmgr;
+    dri_bo_ttm *reloc_buf_ttm = (dri_bo_ttm *)reloc_buf;
+    dri_bo_ttm *target_buf_ttm = (dri_bo_ttm *)target_buf;
+    int num_relocs;
+    uint32_t *this_reloc;
+
+    /* Create a new relocation list if needed */
+    if (reloc_buf_ttm->reloc_buf_data == NULL)
+	intel_setup_reloc_list(reloc_buf);
+
+    num_relocs = reloc_buf_ttm->reloc_buf_data[0];
+
+    /* Check overflow */
+    assert(num_relocs < bufmgr_ttm->max_relocs);
+
+    this_reloc = reloc_buf_ttm->reloc_buf_data + I915_RELOC_HEADER +
+	num_relocs * I915_RELOC0_STRIDE;
+
+    this_reloc[0] = offset;
+    this_reloc[1] = delta;
+    this_reloc[2] = target_buf_ttm->drm_bo.handle; /* To be filled in at exec time */
+    this_reloc[3] = 0;
+
+    reloc_buf_ttm->relocs[num_relocs].validate_flags = flags;
+    reloc_buf_ttm->relocs[num_relocs].target_buf = target_buf;
+    dri_bo_reference(target_buf);
+
+    reloc_buf_ttm->reloc_buf_data[0]++; /* Increment relocation count */
+    /* Check wraparound */
+    assert(reloc_buf_ttm->reloc_buf_data[0] != 0);
+    return 0;
+}
+
+/**
+ * Walk the tree of relocations rooted at BO and accumulate the list of
+ * validations to be performed and update the relocation buffers with
+ * index values into the validation list.
+ */
+static void
+dri_ttm_bo_process_reloc(dri_bo *bo)
+{
+    dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)bo->bufmgr;
+    dri_bo_ttm *bo_ttm = (dri_bo_ttm *)bo;
+    unsigned int nr_relocs;
+    int i;
+
+    if (bo_ttm->reloc_buf_data == NULL)
+	return;
+
+    nr_relocs = bo_ttm->reloc_buf_data[0] & 0xffff;
+
+    for (i = 0; i < nr_relocs; i++) {
+	struct dri_ttm_reloc *r = &bo_ttm->relocs[i];
+
+	/* Continue walking the tree depth-first. */
+	dri_ttm_bo_process_reloc(r->target_buf);
+
+	/* Add the target to the validate list */
+	intel_add_validate_buffer(r->target_buf, r->validate_flags);
+
+	/* Clear the PRESUMED_OFFSET flag from the validate list entry of the
+	 * target if this buffer has a stale relocated pointer at it.
+	 */
+	if (r->last_target_offset != r->target_buf->offset) {
+	   dri_bo_ttm *target_buf_ttm = (dri_bo_ttm *)r->target_buf;
+	   struct intel_validate_entry *entry =
+	      &bufmgr_ttm->validate_array[target_buf_ttm->validate_index];
+
+	   entry->bo_arg.d.req.bo_req.hint &= ~DRM_BO_HINT_PRESUMED_OFFSET;
+	}
+    }
+}
+
+static void *
+dri_ttm_process_reloc(dri_bo *batch_buf, GLuint *count)
+{
+    dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)batch_buf->bufmgr;
+
+    /* Update indices and set up the validate list. */
+    dri_ttm_bo_process_reloc(batch_buf);
+
+    /* Add the batch buffer to the validation list.  There are no relocations
+     * pointing to it.
+     */
+    intel_add_validate_buffer(batch_buf,
+			      DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_EXE);
+
+    *count = bufmgr_ttm->validate_count;
+    return &bufmgr_ttm->validate_array[0].bo_arg;
+}
+
+static const char *
+intel_get_flags_mem_type_string(uint64_t flags)
+{
+    switch (flags & DRM_BO_MASK_MEM) {
+    case DRM_BO_FLAG_MEM_LOCAL: return "local";
+    case DRM_BO_FLAG_MEM_TT: return "ttm";
+    case DRM_BO_FLAG_MEM_VRAM: return "vram";
+    case DRM_BO_FLAG_MEM_PRIV0: return "priv0";
+    case DRM_BO_FLAG_MEM_PRIV1: return "priv1";
+    case DRM_BO_FLAG_MEM_PRIV2: return "priv2";
+    case DRM_BO_FLAG_MEM_PRIV3: return "priv3";
+    case DRM_BO_FLAG_MEM_PRIV4: return "priv4";
+    default: return NULL;
+    }
+}
+
+static const char *
+intel_get_flags_caching_string(uint64_t flags)
+{
+    switch (flags & (DRM_BO_FLAG_CACHED | DRM_BO_FLAG_CACHED_MAPPED)) {
+    case 0: return "UU";
+    case DRM_BO_FLAG_CACHED: return "CU";
+    case DRM_BO_FLAG_CACHED_MAPPED: return "UC";
+    case DRM_BO_FLAG_CACHED | DRM_BO_FLAG_CACHED_MAPPED: return "CC";
+    default: return NULL;
+    }
+}
+
+static void
+intel_update_buffer_offsets (dri_bufmgr_ttm *bufmgr_ttm)
+{
+    int i;
+
+    for (i = 0; i < bufmgr_ttm->validate_count; i++) {
+	dri_bo *bo = bufmgr_ttm->validate_array[i].bo;
+	dri_bo_ttm *bo_ttm = (dri_bo_ttm *)bo;
+	struct drm_i915_op_arg *arg = &bufmgr_ttm->validate_array[i].bo_arg;
+	struct drm_bo_arg_rep *rep = &arg->d.rep;
+
+	/* Update the flags */
+	if (rep->bo_info.flags != bo_ttm->last_flags) {
+	    DBG("BO %s migrated: %s/%s -> %s/%s\n",
+		bo_ttm->name,
+		intel_get_flags_mem_type_string(bo_ttm->last_flags),
+		intel_get_flags_caching_string(bo_ttm->last_flags),
+		intel_get_flags_mem_type_string(rep->bo_info.flags),
+		intel_get_flags_caching_string(rep->bo_info.flags));
+
+	    bo_ttm->last_flags = rep->bo_info.flags;
+	}
+	/* Update the buffer offset */
+	if (rep->bo_info.offset != bo->offset) {
+	    DBG("BO %s migrated: 0x%08lx -> 0x%08lx\n",
+		bo_ttm->name, bo->offset, (unsigned long)rep->bo_info.offset);
+	    bo->offset = rep->bo_info.offset;
+	}
+    }
+}
+
+/**
+ * Update the last target offset field of relocation entries for PRESUMED_OFFSET
+ * computation.
+ */
+static void
+dri_ttm_bo_post_submit(dri_bo *bo)
+{
+    dri_bo_ttm *bo_ttm = (dri_bo_ttm *)bo;
+    unsigned int nr_relocs;
+    int i;
+
+    if (bo_ttm->reloc_buf_data == NULL)
+	return;
+
+    nr_relocs = bo_ttm->reloc_buf_data[0] & 0xffff;
+
+    for (i = 0; i < nr_relocs; i++) {
+	struct dri_ttm_reloc *r = &bo_ttm->relocs[i];
+
+	/* Continue walking the tree depth-first. */
+	dri_ttm_bo_post_submit(r->target_buf);
+
+	r->last_target_offset = r->target_buf->offset;
+    }
+}
+
+static void
+dri_ttm_post_submit(dri_bo *batch_buf, dri_fence **last_fence)
+{
+    dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)batch_buf->bufmgr;
+    int i;
+
+    intel_update_buffer_offsets (bufmgr_ttm);
+
+    dri_ttm_bo_post_submit(batch_buf);
+
+    if (bufmgr_ttm->bufmgr.debug)
+	dri_ttm_dump_validation_list(bufmgr_ttm);
+
+    for (i = 0; i < bufmgr_ttm->validate_count; i++) {
+	dri_bo *bo = bufmgr_ttm->validate_array[i].bo;
+	dri_bo_ttm *bo_ttm = (dri_bo_ttm *)bo;
+
+	/* Disconnect the buffer from the validate list */
+	bo_ttm->validate_index = -1;
+	dri_bo_unreference(bo);
+	bufmgr_ttm->validate_array[i].bo = NULL;
+    }
+    bufmgr_ttm->validate_count = 0;
+}
+
+/**
+ * Enables unlimited caching of buffer objects for reuse.
+ *
+ * This is potentially very memory expensive, as the cache at each bucket
+ * size is only bounded by how many buffers of that size we've managed to have
+ * in flight at once.
+ */
+void
+intel_ttm_enable_bo_reuse(dri_bufmgr *bufmgr)
+{
+    dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)bufmgr;
+    int i;
+
+    for (i = 0; i < INTEL_TTM_BO_BUCKETS; i++) {
+	bufmgr_ttm->cache_bucket[i].max_entries = -1;
+    }
+}
+
+/*
+ *
+ */
+static int
+dri_ttm_check_aperture_space(dri_bo *bo)
+{
+    return 0;
+}
+
+/**
+ * Initializes the TTM buffer manager, which uses the kernel to allocate, map,
+ * and manage map buffer objections.
+ *
+ * \param fd File descriptor of the opened DRM device.
+ * \param fence_type Driver-specific fence type used for fences with no flush.
+ * \param fence_type_flush Driver-specific fence type used for fences with a
+ *	  flush.
+ */
+dri_bufmgr *
+intel_bufmgr_ttm_init(int fd, unsigned int fence_type,
+		      unsigned int fence_type_flush, int batch_size)
+{
+    dri_bufmgr_ttm *bufmgr_ttm;
+    int i;
+
+    bufmgr_ttm = calloc(1, sizeof(*bufmgr_ttm));
+    bufmgr_ttm->fd = fd;
+    bufmgr_ttm->fence_type = fence_type;
+    bufmgr_ttm->fence_type_flush = fence_type_flush;
+
+    /* Let's go with one relocation per every 2 dwords (but round down a bit
+     * since a power of two will mean an extra page allocation for the reloc
+     * buffer).
+     *
+     * Every 4 was too few for the blender benchmark.
+     */
+    bufmgr_ttm->max_relocs = batch_size / sizeof(uint32_t) / 2 - 2;
+
+    bufmgr_ttm->bufmgr.bo_alloc = dri_ttm_alloc;
+    bufmgr_ttm->bufmgr.bo_alloc_static = dri_ttm_alloc_static;
+    bufmgr_ttm->bufmgr.bo_reference = dri_ttm_bo_reference;
+    bufmgr_ttm->bufmgr.bo_unreference = dri_ttm_bo_unreference;
+    bufmgr_ttm->bufmgr.bo_map = dri_ttm_bo_map;
+    bufmgr_ttm->bufmgr.bo_unmap = dri_ttm_bo_unmap;
+    bufmgr_ttm->bufmgr.fence_reference = dri_ttm_fence_reference;
+    bufmgr_ttm->bufmgr.fence_unreference = dri_ttm_fence_unreference;
+    bufmgr_ttm->bufmgr.fence_wait = dri_ttm_fence_wait;
+    bufmgr_ttm->bufmgr.destroy = dri_bufmgr_ttm_destroy;
+    bufmgr_ttm->bufmgr.emit_reloc = dri_ttm_emit_reloc;
+    bufmgr_ttm->bufmgr.process_relocs = dri_ttm_process_reloc;
+    bufmgr_ttm->bufmgr.post_submit = dri_ttm_post_submit;
+    bufmgr_ttm->bufmgr.debug = GL_FALSE;
+    bufmgr_ttm->bufmgr.check_aperture_space = dri_ttm_check_aperture_space;
+    /* Initialize the linked lists for BO reuse cache. */
+    for (i = 0; i < INTEL_TTM_BO_BUCKETS; i++)
+	bufmgr_ttm->cache_bucket[i].tail = &bufmgr_ttm->cache_bucket[i].head;
+
+    return &bufmgr_ttm->bufmgr;
+}
+#else
+dri_bufmgr *
+intel_bufmgr_ttm_init(int fd, unsigned int fence_type,
+		      unsigned int fence_type_flush, int batch_size)
+{
+    return NULL;
+}
+
+dri_bo *
+intel_ttm_bo_create_from_handle(dri_bufmgr *bufmgr, const char *name,
+			      unsigned int handle)
+{
+    return NULL;
+}
+
+void
+intel_ttm_enable_bo_reuse(dri_bufmgr *bufmgr)
+{
+}
+#endif
diff --git a/src/mesa/drivers/dri/intel/intel_bufmgr_ttm.h b/src/mesa/drivers/dri/intel/intel_bufmgr_ttm.h
new file mode 100644
index 0000000000..f5bd64c90f
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_bufmgr_ttm.h
@@ -0,0 +1,28 @@
+
+#ifndef INTEL_BUFMGR_TTM_H
+#define INTEL_BUFMGR_TTM_H
+
+#include "dri_bufmgr.h"
+
+extern dri_bo *intel_ttm_bo_create_from_handle(dri_bufmgr *bufmgr, const char *name,
+					       unsigned int handle);
+
+#ifdef TTM_API
+dri_fence *intel_ttm_fence_create_from_arg(dri_bufmgr *bufmgr, const char *name,
+					   drm_fence_arg_t *arg);
+#endif
+
+
+dri_bufmgr *intel_bufmgr_ttm_init(int fd, unsigned int fence_type,
+				  unsigned int fence_type_flush, int batch_size);
+
+void
+intel_ttm_enable_bo_reuse(dri_bufmgr *bufmgr);
+
+#ifndef TTM_API
+#define DRM_I915_FENCE_CLASS_ACCEL 0
+#define DRM_I915_FENCE_TYPE_RW 2
+#define DRM_I915_FENCE_FLAG_FLUSHED 0x01000000
+#endif
+
+#endif
diff --git a/src/mesa/drivers/dri/intel/intel_context.c b/src/mesa/drivers/dri/intel/intel_context.c
index 50c1964d87..5fa9d95e11 100644
--- a/src/mesa/drivers/dri/intel/intel_context.c
+++ b/src/mesa/drivers/dri/intel/intel_context.c
@@ -59,7 +59,7 @@
 #include "intel_buffer_objects.h"
 #include "intel_fbo.h"
 #include "intel_decode.h"
-#include "intel_bufmgr.h"
+#include "intel_bufmgr_ttm.h"
 
 #include "drirenderbuffer.h"
 #include "vblank.h"
@@ -96,13 +96,11 @@ int INTEL_DEBUG = (0);
 
 #include "extension_helper.h"
 
-#define DRIVER_DATE                     "20080716"
-#define DRIVER_DATE_GEM                 "GEM " DRIVER_DATE
+#define DRIVER_DATE                     "20061102"
 
 static const GLubyte *
 intelGetString(GLcontext * ctx, GLenum name)
 {
-   const struct intel_context *const intel = intel_context(ctx);
    const char *chipset;
    static char buffer[128];
 
@@ -112,7 +110,7 @@ intelGetString(GLcontext * ctx, GLenum name)
       break;
 
    case GL_RENDERER:
-      switch (intel->intelScreen->deviceID) {
+      switch (intel_context(ctx)->intelScreen->deviceID) {
       case PCI_CHIP_845_G:
          chipset = "Intel(R) 845G";
          break;
@@ -185,9 +183,7 @@ intelGetString(GLcontext * ctx, GLenum name)
          break;
       }
 
-      (void) driGetRendererString(buffer, chipset, 
-				  (intel->ttm) ? DRIVER_DATE_GEM : DRIVER_DATE,
-				  0);
+      (void) driGetRendererString(buffer, chipset, DRIVER_DATE, 0);
       return (GLubyte *) buffer;
 
    default:
@@ -371,34 +367,22 @@ intelFlush(GLcontext * ctx)
    if (!IS_965(intel->intelScreen->deviceID))
       INTEL_FIREVERTICES(intel);
 
-   /* Emit a flush so that any frontbuffer rendering that might have occurred
-    * lands onscreen in a timely manner, even if the X Server doesn't trigger
-    * a flush for us.
-    */
-   intel_batchbuffer_emit_mi_flush(intel->batch);
-
    if (intel->batch->map != intel->batch->ptr)
       intel_batchbuffer_flush(intel->batch);
+
+   /* XXX: Need to do an MI_FLUSH here.
+    */
 }
 
 void
 intelFinish(GLcontext * ctx)
 {
-   struct gl_framebuffer *fb = ctx->DrawBuffer;
-   int i;
-
+   struct intel_context *intel = intel_context(ctx);
    intelFlush(ctx);
-
-   for (i = 0; i < fb->_NumColorDrawBuffers; i++) {
-       struct intel_renderbuffer *irb;
-
-       irb = intel_renderbuffer(fb->_ColorDrawBuffers[i]);
-
-       if (irb->region)
-	  dri_bo_wait_rendering(irb->region->buffer);
-   }
-   if (fb->_DepthBuffer) {
-      /* XXX: Wait on buffer idle */
+   if (intel->batch->last_fence) {
+      dri_fence_wait(intel->batch->last_fence);
+      dri_fence_unreference(intel->batch->last_fence);
+      intel->batch->last_fence = NULL;
    }
 }
 
@@ -464,32 +448,28 @@ static GLboolean
 intel_init_bufmgr(struct intel_context *intel)
 {
    intelScreenPrivate *intelScreen = intel->intelScreen;
-   GLboolean gem_disable = getenv("INTEL_NO_GEM") != NULL;
-   int gem_kernel = 0;
-   GLboolean gem_supported;
-   struct drm_i915_getparam gp;
-
-   gp.param = I915_PARAM_HAS_GEM;
-   gp.value = &gem_kernel;
+   GLboolean ttm_disable = getenv("INTEL_NO_TTM") != NULL;
+   GLboolean ttm_supported;
 
-   (void) drmCommandWriteRead(intel->driFd, DRM_I915_GETPARAM, &gp, sizeof(gp));
-
-   /* If we've got a new enough DDX that's initializing GEM and giving us
+   /* If we've got a new enough DDX that's initializing TTM and giving us
     * object handles for the shared buffers, use that.
     */
    intel->ttm = GL_FALSE;
    if (intel->intelScreen->driScrnPriv->dri2.enabled)
-       gem_supported = GL_TRUE;
+       ttm_supported = GL_TRUE;
    else if (intel->intelScreen->driScrnPriv->ddx_version.minor >= 9 &&
-	    gem_kernel &&
+	    intel->intelScreen->drmMinor >= 11 &&
 	    intel->intelScreen->front.bo_handle != -1)
-       gem_supported = GL_TRUE;
+       ttm_supported = GL_TRUE;
    else
-       gem_supported = GL_FALSE;
+       ttm_supported = GL_FALSE;
 
-   if (!gem_disable && gem_supported) {
+   if (!ttm_disable && ttm_supported) {
       int bo_reuse_mode;
-      intel->bufmgr = intel_bufmgr_gem_init(intel->driFd,
+      intel->bufmgr = intel_bufmgr_ttm_init(intel->driFd,
+					    DRM_FENCE_TYPE_EXE,
+					    DRM_FENCE_TYPE_EXE |
+					    DRM_I915_FENCE_TYPE_RW,
 					    BATCH_SZ);
       if (intel->bufmgr != NULL)
 	 intel->ttm = GL_TRUE;
@@ -499,16 +479,16 @@ intel_init_bufmgr(struct intel_context *intel)
       case DRI_CONF_BO_REUSE_DISABLED:
 	 break;
       case DRI_CONF_BO_REUSE_ALL:
-	 intel_bufmgr_gem_enable_reuse(intel->bufmgr);
+	 intel_ttm_enable_bo_reuse(intel->bufmgr);
 	 break;
       }
    }
    /* Otherwise, use the classic buffer manager. */
    if (intel->bufmgr == NULL) {
-      if (gem_disable) {
-	 fprintf(stderr, "GEM disabled.  Using classic.\n");
+      if (ttm_disable) {
+	 fprintf(stderr, "TTM buffer manager disabled.  Using classic.\n");
       } else {
-	 fprintf(stderr, "Failed to initialize GEM.  "
+	 fprintf(stderr, "Failed to initialize TTM buffer manager.  "
 		 "Falling back to classic.\n");
       }
 
@@ -518,17 +498,14 @@ intel_init_bufmgr(struct intel_context *intel)
 	 return GL_FALSE;
       }
 
-      intel->bufmgr = intel_bufmgr_fake_init(intelScreen->tex.offset,
-					     intelScreen->tex.map,
-					     intelScreen->tex.size,
-					     intel_fence_emit,
-					     intel_fence_wait,
-					     intel);
+      intel->bufmgr = dri_bufmgr_fake_init(intelScreen->tex.offset,
+					   intelScreen->tex.map,
+					   intelScreen->tex.size,
+					   intel_fence_emit,
+					   intel_fence_wait,
+					   intel);
    }
 
-   /* XXX bufmgr should be per-screen, not per-context */
-   intelScreen->ttm = intel->ttm;
-
    return GL_TRUE;
 }
 
@@ -696,6 +673,8 @@ intelInitContext(struct intel_context *intel,
       intel_recreate_static_regions(intel);
 
    intel->batch = intel_batchbuffer_alloc(intel);
+   intel->last_swap_fence = NULL;
+   intel->first_swap_fence = NULL;
 
    intel_bufferobj_init(intel);
    intel_fbo_init(intel);
@@ -713,6 +692,7 @@ intelInitContext(struct intel_context *intel,
    /* Force all software fallbacks */
    if (driQueryOptionb(&intel->optionCache, "no_rast")) {
       fprintf(stderr, "disabling 3D rasterization\n");
+      FALLBACK(intel, INTEL_FALLBACK_USER, 1);
       intel->no_rast = 1;
    }
 
@@ -747,7 +727,17 @@ intelDestroyContext(__DRIcontextPrivate * driContextPriv)
       intel->Fallback = 0;      /* don't call _swrast_Flush later */
 
       intel_batchbuffer_free(intel->batch);
-      free(intel->prim.vb);
+
+      if (intel->last_swap_fence) {
+	 dri_fence_wait(intel->last_swap_fence);
+	 dri_fence_unreference(intel->last_swap_fence);
+	 intel->last_swap_fence = NULL;
+      }
+      if (intel->first_swap_fence) {
+	 dri_fence_wait(intel->first_swap_fence);
+	 dri_fence_unreference(intel->first_swap_fence);
+	 intel->first_swap_fence = NULL;
+      }
 
       if (release_texture_heaps) {
          /* This share group is about to go away, free our private
@@ -899,7 +889,7 @@ intelContendedLock(struct intel_context *intel, GLuint flags)
     */
    if (!intel->ttm && sarea->texAge != intel->hHWContext) {
       sarea->texAge = intel->hHWContext;
-      intel_bufmgr_fake_contended_lock_take(intel->bufmgr);
+      dri_bufmgr_fake_contended_lock_take(intel->bufmgr);
       if (INTEL_DEBUG & DEBUG_BATCH)
 	 intel_decode_context_reset();
       if (INTEL_DEBUG & DEBUG_BUFMGR)
@@ -1027,7 +1017,6 @@ void UNLOCK_HARDWARE( struct intel_context *intel )
     * Nothing should be left in batch outside of LOCK/UNLOCK which references
     * cliprects.
     */
-   if (intel->batch->cliprect_mode == REFERENCES_CLIPRECTS)
-      intel_batchbuffer_flush(intel->batch);
+   assert(intel->batch->cliprect_mode != REFERENCES_CLIPRECTS);
 }
 
diff --git a/src/mesa/drivers/dri/intel/intel_context.h b/src/mesa/drivers/dri/intel/intel_context.h
index f9a373cf74..df79ab8897 100644
--- a/src/mesa/drivers/dri/intel/intel_context.h
+++ b/src/mesa/drivers/dri/intel/intel_context.h
@@ -35,7 +35,6 @@
 #include "mm.h"
 #include "texmem.h"
 #include "dri_bufmgr.h"
-#include "intel_bufmgr.h"
 
 #include "intel_screen.h"
 #include "intel_tex_obj.h"
@@ -86,7 +85,6 @@ struct intel_context
    {
       void (*destroy) (struct intel_context * intel);
       void (*emit_state) (struct intel_context * intel);
-      void (*finish_batch) (struct intel_context * intel);
       void (*new_batch) (struct intel_context * intel);
       void (*emit_invarient_state) (struct intel_context * intel);
       void (*note_fence) (struct intel_context *intel, GLuint fence);
@@ -176,6 +174,9 @@ struct intel_context
     */
    GLboolean ttm;
 
+   dri_fence *last_swap_fence;
+   dri_fence *first_swap_fence;
+
    struct intel_batchbuffer *batch;
    GLboolean no_batch_wrap;
    unsigned batch_id;
@@ -183,13 +184,9 @@ struct intel_context
    struct
    {
       GLuint id;
-      uint32_t primitive;	/**< Current hardware primitive type */
+      GLuint primitive;
+      GLubyte *start_ptr;
       void (*flush) (struct intel_context *);
-      dri_bo *vb_bo;
-      uint8_t *vb;
-      unsigned int start_offset; /**< Byte offset of primitive sequence */
-      unsigned int current_offset; /**< Byte offset of next vertex */
-      unsigned int count;	/**< Number of vertices in current primitive */
    } prim;
 
    GLuint stats_wm;
@@ -294,7 +291,6 @@ extern char *__progname;
 #define SUBPIXEL_X 0.125
 #define SUBPIXEL_Y 0.125
 
-#define ARRAY_SIZE(x) (sizeof(x) / sizeof(x[0]))
 #define ALIGN(value, alignment)  ((value + alignment - 1) & ~(alignment - 1))
 
 #define INTEL_FIREVERTICES(intel)		\
diff --git a/src/mesa/drivers/dri/intel/intel_decode.c b/src/mesa/drivers/dri/intel/intel_decode.c
index 9c105013c0..a1240639f4 100644
--- a/src/mesa/drivers/dri/intel/intel_decode.c
+++ b/src/mesa/drivers/dri/intel/intel_decode.c
@@ -183,10 +183,9 @@ decode_2d(uint32_t *data, int count, uint32_t hw_offset, int *failures)
     switch ((data[0] & 0x1fc00000) >> 22) {
     case 0x50:
 	instr_out(data, hw_offset, 0,
-		  "XY_COLOR_BLT (rgb %sabled, alpha %sabled, dst tile %d)\n",
+		  "XY_COLOR_BLT (rgb %sabled, alpha %sabled)\n",
 		  (data[0] & (1 << 20)) ? "en" : "dis",
-		  (data[0] & (1 << 21)) ? "en" : "dis",
-		  (data[0] >> 11) & 1);
+		  (data[0] & (1 << 21)) ? "en" : "dis");
 
 	len = (data[0] & 0x000000ff) + 2;
 	if (len != 6)
@@ -211,8 +210,7 @@ decode_2d(uint32_t *data, int count, uint32_t hw_offset, int *failures)
 
 	instr_out(data, hw_offset, 1, "format %s, pitch %d, "
 		  "clipping %sabled\n", format,
-		  (short)(data[1] & 0xffff),
-		  data[1] & (1 << 30) ? "en" : "dis");
+		  data[1] & 0xffff, data[1] & (1 << 30) ? "en" : "dis");
 	instr_out(data, hw_offset, 2, "(%d,%d)\n",
 		  data[2] & 0xffff, data[2] >> 16);
 	instr_out(data, hw_offset, 3, "(%d,%d)\n",
@@ -222,12 +220,9 @@ decode_2d(uint32_t *data, int count, uint32_t hw_offset, int *failures)
 	return len;
     case 0x53:
 	instr_out(data, hw_offset, 0,
-		  "XY_SRC_COPY_BLT (rgb %sabled, alpha %sabled, "
-		  "src tile %d, dst tile %d)\n",
+		  "XY_SRC_COPY_BLT (rgb %sabled, alpha %sabled)\n",
 		  (data[0] & (1 << 20)) ? "en" : "dis",
-		  (data[0] & (1 << 21)) ? "en" : "dis",
-		  (data[0] >> 15) & 1,
-		  (data[0] >> 11) & 1);
+		  (data[0] & (1 << 21)) ? "en" : "dis");
 
 	len = (data[0] & 0x000000ff) + 2;
 	if (len != 8)
@@ -252,17 +247,16 @@ decode_2d(uint32_t *data, int count, uint32_t hw_offset, int *failures)
 
 	instr_out(data, hw_offset, 1, "format %s, dst pitch %d, "
 		  "clipping %sabled\n", format,
-		  (short)(data[1] & 0xffff),
-		  data[1] & (1 << 30) ? "en" : "dis");
+		  data[1] & 0xffff, data[1] & (1 << 30) ? "en" : "dis");
 	instr_out(data, hw_offset, 2, "dst (%d,%d)\n",
 		  data[2] & 0xffff, data[2] >> 16);
 	instr_out(data, hw_offset, 3, "dst (%d,%d)\n",
-		  data[3] & 0xffff, data[3] >> 16);
+		  data[2] & 0xffff, data[2] >> 16);
 	instr_out(data, hw_offset, 4, "dst offset 0x%08x\n", data[4]);
 	instr_out(data, hw_offset, 5, "src (%d,%d)\n",
 		  data[5] & 0xffff, data[5] >> 16);
 	instr_out(data, hw_offset, 6, "src pitch %d\n",
-		  (short)(data[6] & 0xffff));
+		  data[6] & 0xffff);
 	instr_out(data, hw_offset, 7, "src offset 0x%08x\n", data[7]);
 	return len;
     }
diff --git a/src/mesa/drivers/dri/intel/intel_depthstencil.c b/src/mesa/drivers/dri/intel/intel_depthstencil.c
index 70ba68e9e3..90baecd8c2 100644
--- a/src/mesa/drivers/dri/intel/intel_depthstencil.c
+++ b/src/mesa/drivers/dri/intel/intel_depthstencil.c
@@ -39,7 +39,7 @@
 #include "intel_fbo.h"
 #include "intel_depthstencil.h"
 #include "intel_regions.h"
-#include "intel_span.h"
+
 
 /**
  * The GL_EXT_framebuffer_object allows the user to create their own
@@ -86,33 +86,68 @@
  *
  */
 
+
+
+static void
+map_regions(GLcontext * ctx,
+            struct intel_renderbuffer *depthRb,
+            struct intel_renderbuffer *stencilRb)
+{
+   struct intel_context *intel = intel_context(ctx);
+   if (depthRb && depthRb->region) {
+      intel_region_map(intel, depthRb->region);
+      depthRb->pfMap = depthRb->region->map;
+      depthRb->pfPitch = depthRb->region->pitch;
+   }
+   if (stencilRb && stencilRb->region) {
+      intel_region_map(intel, stencilRb->region);
+      stencilRb->pfMap = stencilRb->region->map;
+      stencilRb->pfPitch = stencilRb->region->pitch;
+   }
+}
+
+static void
+unmap_regions(GLcontext * ctx,
+              struct intel_renderbuffer *depthRb,
+              struct intel_renderbuffer *stencilRb)
+{
+   struct intel_context *intel = intel_context(ctx);
+   if (depthRb && depthRb->region) {
+      intel_region_unmap(intel, depthRb->region);
+      depthRb->pfMap = NULL;
+      depthRb->pfPitch = 0;
+   }
+   if (stencilRb && stencilRb->region) {
+      intel_region_unmap(intel, stencilRb->region);
+      stencilRb->pfMap = NULL;
+      stencilRb->pfPitch = 0;
+   }
+}
+
+
+
 /**
  * Undo the pairing/interleaving between depth and stencil buffers.
  * irb should be a depth/stencil or stencil renderbuffer.
  */
 void
-intel_unpair_depth_stencil(GLcontext *ctx, struct intel_renderbuffer *irb)
+intel_unpair_depth_stencil(GLcontext * ctx, struct intel_renderbuffer *irb)
 {
-   struct intel_context *intel = intel_context(ctx);
-   struct gl_renderbuffer *rb = &irb->Base;
-
    if (irb->PairedStencil) {
       /* irb is a depth/stencil buffer */
       struct gl_renderbuffer *stencilRb;
       struct intel_renderbuffer *stencilIrb;
 
-      ASSERT(rb->_ActualFormat == GL_DEPTH24_STENCIL8_EXT);
+      ASSERT(irb->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT);
 
       stencilRb = _mesa_lookup_renderbuffer(ctx, irb->PairedStencil);
       stencilIrb = intel_renderbuffer(stencilRb);
       if (stencilIrb) {
          /* need to extract stencil values from the depth buffer */
-	 ASSERT(stencilIrb->PairedDepth == rb->Name);
-	 intel_renderbuffer_map(intel, rb);
-	 intel_renderbuffer_map(intel, stencilRb);
-	 _mesa_extract_stencil(ctx, rb, stencilRb);
-	 intel_renderbuffer_unmap(intel, stencilRb);
-	 intel_renderbuffer_unmap(intel, rb);
+         ASSERT(stencilIrb->PairedDepth == irb->Base.Name);
+         map_regions(ctx, irb, stencilIrb);
+         _mesa_extract_stencil(ctx, &irb->Base, &stencilIrb->Base);
+         unmap_regions(ctx, irb, stencilIrb);
          stencilIrb->PairedDepth = 0;
       }
       irb->PairedStencil = 0;
@@ -122,19 +157,17 @@ intel_unpair_depth_stencil(GLcontext *ctx, struct intel_renderbuffer *irb)
       struct gl_renderbuffer *depthRb;
       struct intel_renderbuffer *depthIrb;
 
-      ASSERT(rb->_ActualFormat == GL_STENCIL_INDEX8_EXT ||
-             rb->_ActualFormat == GL_DEPTH24_STENCIL8_EXT);
+      ASSERT(irb->Base._ActualFormat == GL_STENCIL_INDEX8_EXT ||
+             irb->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT);
 
       depthRb = _mesa_lookup_renderbuffer(ctx, irb->PairedDepth);
       depthIrb = intel_renderbuffer(depthRb);
       if (depthIrb) {
          /* need to extract stencil values from the depth buffer */
-	 ASSERT(depthIrb->PairedStencil == rb->Name);
-	 intel_renderbuffer_map(intel, rb);
-	 intel_renderbuffer_map(intel, depthRb);
-	 _mesa_extract_stencil(ctx, depthRb, rb);
-	 intel_renderbuffer_unmap(intel, depthRb);
-	 intel_renderbuffer_unmap(intel, rb);
+         ASSERT(depthIrb->PairedStencil == irb->Base.Name);
+         map_regions(ctx, depthIrb, irb);
+         _mesa_extract_stencil(ctx, &depthIrb->Base, &irb->Base);
+         unmap_regions(ctx, depthIrb, irb);
          depthIrb->PairedStencil = 0;
       }
       irb->PairedDepth = 0;
@@ -161,7 +194,6 @@ void
 intel_validate_paired_depth_stencil(GLcontext * ctx,
                                     struct gl_framebuffer *fb)
 {
-   struct intel_context *intel = intel_context(ctx);
    struct intel_renderbuffer *depthRb, *stencilRb;
 
    depthRb = intel_get_renderbuffer(fb, BUFFER_DEPTH);
@@ -198,11 +230,9 @@ intel_validate_paired_depth_stencil(GLcontext * ctx,
                    stencilRb->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT);
 
             /* establish new pairing: interleave stencil into depth buffer */
-	    intel_renderbuffer_map(intel, &depthRb->Base);
-	    intel_renderbuffer_map(intel, &stencilRb->Base);
+            map_regions(ctx, depthRb, stencilRb);
             _mesa_insert_stencil(ctx, &depthRb->Base, &stencilRb->Base);
-	    intel_renderbuffer_unmap(intel, &stencilRb->Base);
-	    intel_renderbuffer_unmap(intel, &depthRb->Base);
+            unmap_regions(ctx, depthRb, stencilRb);
             depthRb->PairedStencil = stencilRb->Base.Name;
             stencilRb->PairedDepth = depthRb->Base.Name;
          }
diff --git a/src/mesa/drivers/dri/intel/intel_fbo.c b/src/mesa/drivers/dri/intel/intel_fbo.c
index 5bd2ebfdcf..b3f6610546 100644
--- a/src/mesa/drivers/dri/intel/intel_fbo.c
+++ b/src/mesa/drivers/dri/intel/intel_fbo.c
@@ -153,9 +153,6 @@ intel_delete_renderbuffer(struct gl_renderbuffer *rb)
       intel_unpair_depth_stencil(ctx, irb);
    }
 
-   if (irb->span_cache != NULL)
-      _mesa_free(irb->span_cache);
-
    if (intel && irb->region) {
       intel_region_release(&irb->region);
    }
@@ -212,14 +209,6 @@ intel_alloc_renderbuffer_storage(GLcontext * ctx, struct gl_renderbuffer *rb,
    case GL_RGB10:
    case GL_RGB12:
    case GL_RGB16:
-      rb->_ActualFormat = GL_RGB8;
-      rb->DataType = GL_UNSIGNED_BYTE;
-      rb->RedBits = 8;
-      rb->GreenBits = 8;
-      rb->BlueBits = 8;
-      rb->AlphaBits = 0;
-      cpp = 4;
-      break;
    case GL_RGBA:
    case GL_RGBA2:
    case GL_RGBA4:
@@ -305,6 +294,9 @@ intel_alloc_renderbuffer_storage(GLcontext * ctx, struct gl_renderbuffer *rb,
       rb->Width = width;
       rb->Height = height;
 
+      /* This sets the Get/PutRow/Value functions */
+      intel_set_span_functions(&irb->Base);
+
       return GL_TRUE;
    }
 }
@@ -374,6 +366,7 @@ intel_renderbuffer_set_region(struct intel_renderbuffer *rb,
    intel_region_reference(&rb->region, region);
    intel_region_release(&old);
 
+   rb->pfMap = region->map;
    rb->pfPitch = region->pitch;
 }
 
@@ -453,6 +446,8 @@ intel_create_renderbuffer(GLenum intFormat)
    irb->Base.Delete = intel_delete_renderbuffer;
    irb->Base.AllocStorage = intel_alloc_window_storage;
    irb->Base.GetPointer = intel_get_pointer;
+   /* This sets the Get/PutRow/Value functions */
+   intel_set_span_functions(&irb->Base);
 
    return irb;
 }
@@ -524,7 +519,7 @@ intel_framebuffer_renderbuffer(GLcontext * ctx,
 
 static GLboolean
 intel_update_wrapper(GLcontext *ctx, struct intel_renderbuffer *irb, 
-		     struct gl_texture_image *texImage)
+                          struct gl_texture_image *texImage)
 {
    if (texImage->TexFormat == &_mesa_texformat_argb8888) {
       irb->Base._ActualFormat = GL_RGBA8;
@@ -563,6 +558,7 @@ intel_update_wrapper(GLcontext *ctx, struct intel_renderbuffer *irb,
 
    irb->Base.Delete = intel_delete_renderbuffer;
    irb->Base.AllocStorage = intel_nop_alloc_storage;
+   intel_set_span_functions(&irb->Base);
 
    irb->RenderToTexture = GL_TRUE;
 
diff --git a/src/mesa/drivers/dri/intel/intel_fbo.h b/src/mesa/drivers/dri/intel/intel_fbo.h
index 9d15582d78..c90c84b48c 100644
--- a/src/mesa/drivers/dri/intel/intel_fbo.h
+++ b/src/mesa/drivers/dri/intel/intel_fbo.h
@@ -28,9 +28,9 @@
 #ifndef INTEL_FBO_H
 #define INTEL_FBO_H
 
-#include "intel_screen.h"
 
 struct intel_context;
+struct intel_region;
 
 /**
  * Intel framebuffer, derived from gl_framebuffer.
@@ -70,6 +70,7 @@ struct intel_renderbuffer
 {
    struct gl_renderbuffer Base;
    struct intel_region *region;
+   void *pfMap;                 /* possibly paged flipped map pointer */
    GLuint pfPitch;              /* possibly paged flipped pitch */
    GLboolean RenderToTexture;   /* RTT? */
 
@@ -79,9 +80,6 @@ struct intel_renderbuffer
    GLuint pf_pending;  /**< sequence number of pending flip */
 
    GLuint vbl_pending;   /**< vblank sequence number of pending flip */
-
-   uint8_t *span_cache;
-   unsigned long span_cache_offset;
 };
 
 extern struct intel_renderbuffer *intel_renderbuffer(struct gl_renderbuffer
diff --git a/src/mesa/drivers/dri/intel/intel_ioctl.c b/src/mesa/drivers/dri/intel/intel_ioctl.c
index 58c81766cd..f4566ba89c 100644
--- a/src/mesa/drivers/dri/intel/intel_ioctl.c
+++ b/src/mesa/drivers/dri/intel/intel_ioctl.c
@@ -30,8 +30,6 @@
 #include <unistd.h>
 #include <errno.h>
 #include <sched.h>
-#include <sys/types.h>
-#include <sys/ioctl.h>
 
 #include "mtypes.h"
 #include "context.h"
@@ -45,7 +43,7 @@
 #include "drm.h"
 #include "i915_drm.h"
 
-#include "intel_bufmgr.h"
+#include "intel_bufmgr_ttm.h"
 
 #define FILE_DEBUG_FLAG DEBUG_IOCTL
 
@@ -106,7 +104,7 @@ intelWaitIrq(struct intel_context *intel, int seq)
 }
 
 
-int
+void
 intel_batch_ioctl(struct intel_context *intel,
                   GLuint start_offset,
                   GLuint used,
@@ -115,7 +113,7 @@ intel_batch_ioctl(struct intel_context *intel,
    struct drm_i915_batchbuffer batch;
 
    if (intel->no_hw)
-      return 0;
+      return;
 
    assert(intel->locked);
    assert(used);
@@ -144,42 +142,82 @@ intel_batch_ioctl(struct intel_context *intel,
    if (drmCommandWrite(intel->driFd, DRM_I915_BATCHBUFFER, &batch,
                        sizeof(batch))) {
       fprintf(stderr, "DRM_I915_BATCHBUFFER: %d\n", -errno);
-      return -errno;
+      UNLOCK_HARDWARE(intel);
+      exit(1);
    }
-
-   return 0;
 }
 
-int
+#ifdef TTM_API
+void
 intel_exec_ioctl(struct intel_context *intel,
 		 GLuint used,
 		 GLboolean ignore_cliprects, GLboolean allow_unlock,
-		 struct drm_i915_gem_execbuffer *execbuf)
+		 void *start, GLuint count, dri_fence **fence)
 {
+   struct drm_i915_execbuffer execbuf;
+   dri_fence *fo;
    int ret;
 
    assert(intel->locked);
    assert(used);
 
    if (intel->no_hw)
-      return 0;
+      return;
+
+   if (*fence) {
+     dri_fence_unreference(*fence);
+   }
 
-   execbuf->batch_start_offset = 0;
-   execbuf->batch_len = used;
-   execbuf->cliprects_ptr = (uintptr_t)intel->pClipRects;
-   execbuf->num_cliprects = ignore_cliprects ? 0 : intel->numClipRects;
-   execbuf->DR1 = 0;
-   execbuf->DR4 = ((((GLuint) intel->drawX) & 0xffff) |
-		   (((GLuint) intel->drawY) << 16));
+   memset(&execbuf, 0, sizeof(execbuf));
+
+   execbuf.num_buffers = count;
+   execbuf.batch.used = used;
+   execbuf.batch.cliprects = intel->pClipRects;
+   execbuf.batch.num_cliprects = ignore_cliprects ? 0 : intel->numClipRects;
+   execbuf.batch.DR1 = 0;
+   execbuf.batch.DR4 = ((((GLuint) intel->drawX) & 0xffff) |
+			(((GLuint) intel->drawY) << 16));
+
+   execbuf.ops_list = (unsigned long)start; // TODO
+   execbuf.fence_arg.flags = DRM_FENCE_FLAG_SHAREABLE | DRM_I915_FENCE_FLAG_FLUSHED;
 
    do {
-      ret = ioctl(intel->driFd, DRM_IOCTL_I915_GEM_EXECBUFFER, execbuf);
+      ret = drmCommandWriteRead(intel->driFd, DRM_I915_EXECBUFFER, &execbuf,
+				sizeof(execbuf));
    } while (ret == -EAGAIN);
 
    if (ret != 0) {
-      fprintf(stderr, "DRM_I915_GEM_EXECBUFFER: %d\n", -errno);
-      return -errno;
+      fprintf(stderr, "DRM_I915_EXECBUFFER: %d\n", -errno);
+      UNLOCK_HARDWARE(intel);
+      exit(1);
    }
 
-   return 0;
+   if (execbuf.fence_arg.error != 0) {
+
+      /*
+       * Fence creation has failed, but the GPU has been
+       * idled by the kernel. Safe to continue.
+       */ 
+
+      *fence = NULL;
+      return;
+   }
+
+   fo = intel_ttm_fence_create_from_arg(intel->bufmgr, "fence buffers",
+					&execbuf.fence_arg);
+   if (!fo) {
+      fprintf(stderr, "failed to fence handle: %08x\n", execbuf.fence_arg.handle);
+      UNLOCK_HARDWARE(intel);
+      exit(1);
+   }
+   *fence = fo;
+}
+#else
+void
+intel_exec_ioctl(struct intel_context *intel,
+		 GLuint used,
+		 GLboolean ignore_cliprects, GLboolean allow_unlock,
+		 void *start, GLuint count, dri_fence **fence)
+{
 }
+#endif
diff --git a/src/mesa/drivers/dri/intel/intel_ioctl.h b/src/mesa/drivers/dri/intel/intel_ioctl.h
index 526e38358c..8674aef723 100644
--- a/src/mesa/drivers/dri/intel/intel_ioctl.h
+++ b/src/mesa/drivers/dri/intel/intel_ioctl.h
@@ -33,14 +33,14 @@
 void intelWaitIrq( struct intel_context *intel, int seq );
 int intelEmitIrqLocked( struct intel_context *intel );
 
-int intel_batch_ioctl(struct intel_context *intel,
-		      GLuint start_offset,
+void intel_batch_ioctl( struct intel_context *intel, 
+			GLuint start_offset,
+			GLuint used,
+			GLboolean ignore_cliprects,
+			GLboolean allow_unlock );
+void intel_exec_ioctl(struct intel_context *intel,
 		      GLuint used,
-		      GLboolean ignore_cliprects,
-		      GLboolean allow_unlock);
-int intel_exec_ioctl(struct intel_context *intel,
-		     GLuint used,
-		     GLboolean ignore_cliprects, GLboolean allow_unlock,
-		     struct drm_i915_gem_execbuffer *execbuf);
+		      GLboolean ignore_cliprects, GLboolean allow_unlock,
+		      void *start, GLuint count, dri_fence **fence);
 
 #endif
diff --git a/src/mesa/drivers/dri/intel/intel_pixel_bitmap.c b/src/mesa/drivers/dri/intel/intel_pixel_bitmap.c
index 7e0d20e681..81238acfe4 100644
--- a/src/mesa/drivers/dri/intel/intel_pixel_bitmap.c
+++ b/src/mesa/drivers/dri/intel/intel_pixel_bitmap.c
@@ -43,7 +43,7 @@
 #include "intel_buffer_objects.h"
 #include "intel_buffers.h"
 #include "intel_pixel.h"
-#include "intel_reg.h"
+
 
 
 #define FILE_DEBUG_FLAG DEBUG_PIXEL
@@ -293,7 +293,7 @@ do_blit_bitmap( GLcontext *ctx,
 						  dst->pitch,
 						  dst->buffer,
 						  0,
-						  dst->tiling,
+						  dst->tiled,
 						  rect.x1 + px,
 						  rect.y2 - (py + h),
 						  w, h,
@@ -301,8 +301,9 @@ do_blit_bitmap( GLcontext *ctx,
 	    } 
 	 } 
       }
+   out:
+      intel_batchbuffer_flush(intel->batch);
    }
-out:
    UNLOCK_HARDWARE(intel);
 
 
diff --git a/src/mesa/drivers/dri/intel/intel_pixel_copy.c b/src/mesa/drivers/dri/intel/intel_pixel_copy.c
index 3093ccf7c6..45f72bac52 100644
--- a/src/mesa/drivers/dri/intel/intel_pixel_copy.c
+++ b/src/mesa/drivers/dri/intel/intel_pixel_copy.c
@@ -229,7 +229,7 @@ do_texture_copypixels(GLcontext * ctx,
 
     out:
       intel->vtbl.leave_meta_state(intel);
-      intel_batchbuffer_emit_mi_flush(intel->batch);
+      intel_batchbuffer_flush(intel->batch);
    }
    UNLOCK_HARDWARE(intel);
 
@@ -337,16 +337,18 @@ do_blit_copypixels(GLcontext * ctx,
             continue;
 
          intelEmitCopyBlit(intel, dst->cpp,
-			   src->pitch, src->buffer, 0, src->tiling,
-			   dst->pitch, dst->buffer, 0, dst->tiling,
+			   src->pitch, src->buffer, 0, src->tiled,
+			   dst->pitch, dst->buffer, 0, dst->tiled,
 			   clip_x + delta_x, clip_y + delta_y, /* srcx, srcy */
 			   clip_x, clip_y, /* dstx, dsty */
 			   clip_w, clip_h,
 			   ctx->Color.ColorLogicOpEnabled ?
 			   ctx->Color.LogicOp : GL_COPY);
       }
+
+    out:
+      intel_batchbuffer_flush(intel->batch);
    }
-out:
    UNLOCK_HARDWARE(intel);
 
    DBG("%s: success\n", __FUNCTION__);
diff --git a/src/mesa/drivers/dri/intel/intel_pixel_draw.c b/src/mesa/drivers/dri/intel/intel_pixel_draw.c
index 5675084da5..34813d2aa0 100644
--- a/src/mesa/drivers/dri/intel/intel_pixel_draw.c
+++ b/src/mesa/drivers/dri/intel/intel_pixel_draw.c
@@ -81,8 +81,7 @@ do_texture_drawpixels(GLcontext * ctx,
    else {
       /* PBO only for now:
        */
-      if (INTEL_DEBUG & DEBUG_PIXEL)
-	 _mesa_printf("%s - not PBO\n", __FUNCTION__);
+/*       _mesa_printf("%s - not PBO\n", __FUNCTION__); */
       return GL_FALSE;
    }
 
@@ -181,7 +180,7 @@ do_texture_drawpixels(GLcontext * ctx,
 				 srcx, srcx + width, srcy + height, srcy);
     out:
       intel->vtbl.leave_meta_state(intel);
-      intel_batchbuffer_emit_mi_flush(intel->batch);
+      intel_batchbuffer_flush(intel->batch);
    }
    UNLOCK_HARDWARE(intel);
    return GL_TRUE;
@@ -219,6 +218,7 @@ do_blit_drawpixels(GLcontext * ctx,
    struct intel_buffer_object *src = intel_buffer_object(unpack->BufferObj);
    GLuint src_offset;
    GLuint rowLength;
+   dri_fence *fence = NULL;
 
    if (INTEL_DEBUG & DEBUG_PIXEL)
       _mesa_printf("%s\n", __FUNCTION__);
@@ -314,7 +314,7 @@ do_blit_drawpixels(GLcontext * ctx,
          intelEmitCopyBlit(intel,
                            dest->cpp,
                            rowLength, src_buffer, src_offset, GL_FALSE,
-                           dest->pitch, dest->buffer, 0, dest->tiling,
+                           dest->pitch, dest->buffer, 0, dest->tiled,
                            rect.x1 - dest_rect.x1,
                            rect.y2 - dest_rect.y2,
                            rect.x1,
@@ -322,9 +322,17 @@ do_blit_drawpixels(GLcontext * ctx,
 			   ctx->Color.ColorLogicOpEnabled ?
 			   ctx->Color.LogicOp : GL_COPY);
       }
+      intel_batchbuffer_flush(intel->batch);
+      fence = intel->batch->last_fence;
+      dri_fence_reference(fence);
    }
    UNLOCK_HARDWARE(intel);
 
+   if (fence) {
+      dri_fence_wait(fence);
+      dri_fence_unreference(fence);
+   }
+
    if (INTEL_DEBUG & DEBUG_PIXEL)
       _mesa_printf("%s - DONE\n", __FUNCTION__);
 
diff --git a/src/mesa/drivers/dri/intel/intel_reg.h b/src/mesa/drivers/dri/intel/intel_reg.h
index 96af7e1a03..37629c07e2 100644
--- a/src/mesa/drivers/dri/intel/intel_reg.h
+++ b/src/mesa/drivers/dri/intel/intel_reg.h
@@ -31,140 +31,11 @@
 
 #define MI_BATCH_BUFFER_END		(CMD_MI | 0xA << 23)
 
-#define MI_FLUSH			(CMD_MI | (4 << 23))
-#define FLUSH_MAP_CACHE				(1 << 0)
-#define INHIBIT_FLUSH_RENDER_CACHE		(1 << 2)
-
 /* Stalls command execution waiting for the given events to have occurred. */
 #define MI_WAIT_FOR_EVENT               (CMD_MI | (0x3 << 23))
 #define MI_WAIT_FOR_PLANE_B_FLIP        (1<<6)
 #define MI_WAIT_FOR_PLANE_A_FLIP        (1<<2)
 
-/* p189 */
-#define _3DSTATE_LOAD_STATE_IMMEDIATE_1   (CMD_3D | (0x1d<<24) | (0x04<<16))
-#define I1_LOAD_S(n)                      (1<<(4+n))
-
-/** @{
- * 915 definitions
- */
-#define S0_VB_OFFSET_MASK		0xffffffc
-#define S0_AUTO_CACHE_INV_DISABLE	(1<<0)
-/** @} */
-
-/** @{
- * 830 definitions
- */
-#define S0_VB_OFFSET_MASK_830		0xffffff8
-#define S0_VB_PITCH_SHIFT_830		1
-#define S0_VB_ENABLE_830		0
-/** @} */
-
-#define S1_VERTEX_WIDTH_SHIFT          24
-#define S1_VERTEX_WIDTH_MASK           (0x3f<<24)
-#define S1_VERTEX_PITCH_SHIFT          16
-#define S1_VERTEX_PITCH_MASK           (0x3f<<16)
-
-#define TEXCOORDFMT_2D                 0x0
-#define TEXCOORDFMT_3D                 0x1
-#define TEXCOORDFMT_4D                 0x2
-#define TEXCOORDFMT_1D                 0x3
-#define TEXCOORDFMT_2D_16              0x4
-#define TEXCOORDFMT_4D_16              0x5
-#define TEXCOORDFMT_NOT_PRESENT        0xf
-#define S2_TEXCOORD_FMT0_MASK            0xf
-#define S2_TEXCOORD_FMT1_SHIFT           4
-#define S2_TEXCOORD_FMT(unit, type)    ((type)<<(unit*4))
-#define S2_TEXCOORD_NONE               (~0)
-#define S2_TEX_COUNT_SHIFT_830		12
-#define S2_VERTEX_0_WIDTH_SHIFT_830	0
-#define S2_VERTEX_1_WIDTH_SHIFT_830	6
-/* S3 not interesting */
-
-#define S4_POINT_WIDTH_SHIFT           23
-#define S4_POINT_WIDTH_MASK            (0x1ff<<23)
-#define S4_LINE_WIDTH_SHIFT            19
-#define S4_LINE_WIDTH_ONE              (0x2<<19)
-#define S4_LINE_WIDTH_MASK             (0xf<<19)
-#define S4_FLATSHADE_ALPHA             (1<<18)
-#define S4_FLATSHADE_FOG               (1<<17)
-#define S4_FLATSHADE_SPECULAR          (1<<16)
-#define S4_FLATSHADE_COLOR             (1<<15)
-#define S4_CULLMODE_BOTH	       (0<<13)
-#define S4_CULLMODE_NONE	       (1<<13)
-#define S4_CULLMODE_CW		       (2<<13)
-#define S4_CULLMODE_CCW		       (3<<13)
-#define S4_CULLMODE_MASK	       (3<<13)
-#define S4_VFMT_POINT_WIDTH            (1<<12)
-#define S4_VFMT_SPEC_FOG               (1<<11)
-#define S4_VFMT_COLOR                  (1<<10)
-#define S4_VFMT_DEPTH_OFFSET           (1<<9)
-#define S4_VFMT_XYZ     	       (1<<6)
-#define S4_VFMT_XYZW     	       (2<<6)
-#define S4_VFMT_XY     		       (3<<6)
-#define S4_VFMT_XYW     	       (4<<6)
-#define S4_VFMT_XYZW_MASK              (7<<6)
-#define S4_FORCE_DEFAULT_DIFFUSE       (1<<5)
-#define S4_FORCE_DEFAULT_SPECULAR      (1<<4)
-#define S4_LOCAL_DEPTH_OFFSET_ENABLE   (1<<3)
-#define S4_VFMT_FOG_PARAM              (1<<2)
-#define S4_SPRITE_POINT_ENABLE         (1<<1)
-#define S4_LINE_ANTIALIAS_ENABLE       (1<<0)
-
-#define S4_VFMT_MASK (S4_VFMT_POINT_WIDTH   | 	\
-		      S4_VFMT_SPEC_FOG      |	\
-		      S4_VFMT_COLOR         |	\
-		      S4_VFMT_DEPTH_OFFSET  |	\
-		      S4_VFMT_XYZW_MASK     |	\
-		      S4_VFMT_FOG_PARAM)
-
-
-#define S5_WRITEDISABLE_ALPHA          (1<<31)
-#define S5_WRITEDISABLE_RED            (1<<30)
-#define S5_WRITEDISABLE_GREEN          (1<<29)
-#define S5_WRITEDISABLE_BLUE           (1<<28)
-#define S5_WRITEDISABLE_MASK           (0xf<<28)
-#define S5_FORCE_DEFAULT_POINT_SIZE    (1<<27)
-#define S5_LAST_PIXEL_ENABLE           (1<<26)
-#define S5_GLOBAL_DEPTH_OFFSET_ENABLE  (1<<25)
-#define S5_FOG_ENABLE                  (1<<24)
-#define S5_STENCIL_REF_SHIFT           16
-#define S5_STENCIL_REF_MASK            (0xff<<16)
-#define S5_STENCIL_TEST_FUNC_SHIFT     13
-#define S5_STENCIL_TEST_FUNC_MASK      (0x7<<13)
-#define S5_STENCIL_FAIL_SHIFT          10
-#define S5_STENCIL_FAIL_MASK           (0x7<<10)
-#define S5_STENCIL_PASS_Z_FAIL_SHIFT   7
-#define S5_STENCIL_PASS_Z_FAIL_MASK    (0x7<<7)
-#define S5_STENCIL_PASS_Z_PASS_SHIFT   4
-#define S5_STENCIL_PASS_Z_PASS_MASK    (0x7<<4)
-#define S5_STENCIL_WRITE_ENABLE        (1<<3)
-#define S5_STENCIL_TEST_ENABLE         (1<<2)
-#define S5_COLOR_DITHER_ENABLE         (1<<1)
-#define S5_LOGICOP_ENABLE              (1<<0)
-
-
-#define S6_ALPHA_TEST_ENABLE           (1<<31)
-#define S6_ALPHA_TEST_FUNC_SHIFT       28
-#define S6_ALPHA_TEST_FUNC_MASK        (0x7<<28)
-#define S6_ALPHA_REF_SHIFT             20
-#define S6_ALPHA_REF_MASK              (0xff<<20)
-#define S6_DEPTH_TEST_ENABLE           (1<<19)
-#define S6_DEPTH_TEST_FUNC_SHIFT       16
-#define S6_DEPTH_TEST_FUNC_MASK        (0x7<<16)
-#define S6_CBUF_BLEND_ENABLE           (1<<15)
-#define S6_CBUF_BLEND_FUNC_SHIFT       12
-#define S6_CBUF_BLEND_FUNC_MASK        (0x7<<12)
-#define S6_CBUF_SRC_BLEND_FACT_SHIFT   8
-#define S6_CBUF_SRC_BLEND_FACT_MASK    (0xf<<8)
-#define S6_CBUF_DST_BLEND_FACT_SHIFT   4
-#define S6_CBUF_DST_BLEND_FACT_MASK    (0xf<<4)
-#define S6_DEPTH_WRITE_ENABLE          (1<<3)
-#define S6_COLOR_WRITE_ENABLE          (1<<2)
-#define S6_TRISTRIP_PV_SHIFT           0
-#define S6_TRISTRIP_PV_MASK            (0x3<<0)
-
-#define S7_DEPTH_OFFSET_CONST_MASK     ~0
-
 /* Primitive dispatch on 830-945 */
 #define _3DPRIMITIVE			(CMD_3D | (0x1f << 24))
 #define PRIM_INDIRECT            (1<<23)
diff --git a/src/mesa/drivers/dri/intel/intel_regions.c b/src/mesa/drivers/dri/intel/intel_regions.c
index ddfdce3835..35ab46afea 100644
--- a/src/mesa/drivers/dri/intel/intel_regions.c
+++ b/src/mesa/drivers/dri/intel/intel_regions.c
@@ -39,17 +39,13 @@
  * last moment.
  */
 
-#include <sys/ioctl.h>
-#include <errno.h>
-
 #include "intel_context.h"
 #include "intel_regions.h"
 #include "intel_blit.h"
 #include "intel_buffer_objects.h"
 #include "dri_bufmgr.h"
-#include "intel_bufmgr.h"
+#include "intel_bufmgr_ttm.h"
 #include "intel_batchbuffer.h"
-#include "intel_chipset.h"
 
 #define FILE_DEBUG_FLAG DEBUG_REGION
 
@@ -80,34 +76,10 @@ intel_region_unmap(struct intel_context *intel, struct intel_region *region)
    }
 }
 
-static int
-intel_set_region_tiling_gem(struct intel_context *intel,
-			    struct intel_region *region,
-			    uint32_t bo_handle)
-{
-   struct drm_i915_gem_get_tiling get_tiling;
-   int ret;
-
-   memset(&get_tiling, 0, sizeof(get_tiling));
-
-   get_tiling.handle = bo_handle;
-   ret = ioctl(intel->driFd, DRM_IOCTL_I915_GEM_GET_TILING, &get_tiling);
-   if (ret != 0) {
-      fprintf(stderr, "Failed to get tiling state for region: %s\n",
-	      strerror(errno));
-      return ret;
-   }
-
-   region->tiling = get_tiling.tiling_mode;
-   region->bit_6_swizzle = get_tiling.swizzle_mode;
-
-   return 0;
-}
-
 static struct intel_region *
 intel_region_alloc_internal(struct intel_context *intel,
 			    GLuint cpp, GLuint pitch, GLuint height,
-			    dri_bo *buffer)
+			    GLuint tiled, dri_bo *buffer)
 {
    struct intel_region *region;
 
@@ -121,12 +93,9 @@ intel_region_alloc_internal(struct intel_context *intel,
    region->pitch = pitch;
    region->height = height;     /* needed? */
    region->refcount = 1;
+   region->tiled = tiled;
    region->buffer = buffer;
 
-   /* Default to no tiling */
-   region->tiling = I915_TILING_NONE;
-   region->bit_6_swizzle = I915_BIT_6_SWIZZLE_NONE;
-
    return region;
 }
 
@@ -137,28 +106,25 @@ intel_region_alloc(struct intel_context *intel,
    dri_bo *buffer;
 
    buffer = dri_bo_alloc(intel->bufmgr, "region",
-			 pitch * cpp * height, 64);
+			 pitch * cpp * height, 64,
+			 DRM_BO_FLAG_MEM_LOCAL |
+			 DRM_BO_FLAG_CACHED |
+			 DRM_BO_FLAG_CACHED_MAPPED);
 
-   return intel_region_alloc_internal(intel, cpp, pitch, height, buffer);
+   return intel_region_alloc_internal(intel, cpp, pitch, height, 0, buffer);
 }
 
 struct intel_region *
 intel_region_alloc_for_handle(struct intel_context *intel,
 			      GLuint cpp, GLuint pitch, GLuint height,
-			      GLuint handle)
+			      GLuint tiled, GLuint handle)
 {
-   struct intel_region *region;
    dri_bo *buffer;
 
-   buffer = intel_bo_gem_create_from_name(intel->bufmgr, "dri2 region", handle);
+   buffer = intel_ttm_bo_create_from_handle(intel->bufmgr, "region", handle);
 
-   region = intel_region_alloc_internal(intel, cpp, pitch, height, buffer);
-   if (region == NULL)
-      return region;
-
-   intel_set_region_tiling_gem(intel, region, handle);
-
-   return region;
+   return intel_region_alloc_internal(intel,
+				      cpp, pitch, height, tiled, buffer);
 }
 
 void
@@ -172,34 +138,26 @@ intel_region_reference(struct intel_region **dst, struct intel_region *src)
 }
 
 void
-intel_region_release(struct intel_region **region_handle)
+intel_region_release(struct intel_region **region)
 {
-   struct intel_region *region = *region_handle;
-
-   if (region == NULL)
+   if (!*region)
       return;
 
-   DBG("%s %d\n", __FUNCTION__, region->refcount - 1);
-
-   ASSERT(region->refcount > 0);
-   region->refcount--;
+   DBG("%s %d\n", __FUNCTION__, (*region)->refcount - 1);
 
-   if (region->refcount == 0) {
-      assert(region->map_refcount == 0);
+   ASSERT((*region)->refcount > 0);
+   (*region)->refcount--;
 
-      if (region->pbo)
-	 region->pbo->region = NULL;
-      region->pbo = NULL;
-      dri_bo_unreference(region->buffer);
-
-      if (region->classic_map != NULL) {
-	 drmUnmap(region->classic_map,
-			region->pitch * region->cpp * region->height);
-      }
+   if ((*region)->refcount == 0) {
+      assert((*region)->map_refcount == 0);
 
-      free(region);
+      if ((*region)->pbo)
+	 (*region)->pbo->region = NULL;
+      (*region)->pbo = NULL;
+      dri_bo_unreference((*region)->buffer);
+      free(*region);
    }
-   *region_handle = NULL;
+   *region = NULL;
 }
 
 /*
@@ -314,8 +272,8 @@ intel_region_copy(struct intel_context *intel,
 
    intelEmitCopyBlit(intel,
                      dst->cpp,
-                     src->pitch, src->buffer, src_offset, src->tiling,
-                     dst->pitch, dst->buffer, dst_offset, dst->tiling,
+                     src->pitch, src->buffer, src_offset, src->tiled,
+                     dst->pitch, dst->buffer, dst_offset, dst->tiled,
                      srcx, srcy, dstx, dsty, width, height,
 		     GL_COPY);
 }
@@ -345,7 +303,7 @@ intel_region_fill(struct intel_context *intel,
 
    intelEmitFillBlit(intel,
                      dst->cpp,
-                     dst->pitch, dst->buffer, dst_offset, dst->tiling,
+                     dst->pitch, dst->buffer, dst_offset, dst->tiled,
                      dstx, dsty, width, height, color);
 }
 
@@ -397,7 +355,10 @@ intel_region_release_pbo(struct intel_context *intel,
 
    region->buffer = dri_bo_alloc(intel->bufmgr, "region",
 				 region->pitch * region->cpp * region->height,
-				 64);
+				 64,
+				 DRM_BO_FLAG_MEM_LOCAL |
+				 DRM_BO_FLAG_CACHED |
+				 DRM_BO_FLAG_CACHED_MAPPED);
 }
 
 /* Break the COW tie to the pbo.  Both the pbo and the region end up
@@ -421,19 +382,23 @@ intel_region_cow(struct intel_context *intel, struct intel_region *region)
    /* Now blit from the texture buffer to the new buffer: 
     */
 
+   intel_batchbuffer_flush(intel->batch);
+
    was_locked = intel->locked;
-   if (!was_locked)
+   if (intel->locked)
       LOCK_HARDWARE(intel);
 
    intelEmitCopyBlit(intel,
 		     region->cpp,
-		     region->pitch, region->buffer, 0, region->tiling,
-		     region->pitch, pbo->buffer, 0, region->tiling,
+		     region->pitch, region->buffer, 0, region->tiled,
+		     region->pitch, pbo->buffer, 0, region->tiled,
 		     0, 0, 0, 0,
 		     region->pitch, region->height,
 		     GL_COPY);
 
-   if (!was_locked)
+   intel_batchbuffer_flush(intel->batch);
+
+   if (was_locked)
       UNLOCK_HARDWARE(intel);
 }
 
@@ -458,7 +423,6 @@ intel_recreate_static(struct intel_context *intel,
 		      intelRegion *region_desc)
 {
    intelScreenPrivate *intelScreen = intel->intelScreen;
-   int ret;
 
    if (region == NULL) {
       region = calloc(sizeof(*region), 1);
@@ -471,45 +435,21 @@ intel_recreate_static(struct intel_context *intel,
       region->cpp = intel->ctx.Visual.rgbBits / 8;
    region->pitch = intelScreen->pitch;
    region->height = intelScreen->height;     /* needed? */
+   region->tiled = region_desc->tiled;
 
    if (intel->ttm) {
       assert(region_desc->bo_handle != -1);
-      region->buffer = intel_bo_gem_create_from_name(intel->bufmgr,
-						     name,
-						     region_desc->bo_handle);
-
-      intel_set_region_tiling_gem(intel, region, region_desc->bo_handle);
+      region->buffer = intel_ttm_bo_create_from_handle(intel->bufmgr,
+						       name,
+						       region_desc->bo_handle);
    } else {
-      ret = drmMap(intel->driFd, region_desc->handle,
-		   region->pitch * region->cpp * region->height,
-		   &region->classic_map);
-      if (ret != 0) {
-	 fprintf(stderr, "Failed to drmMap %s buffer\n", name);
-	 free(region);
-	 return NULL;
-      }
-
-      region->buffer = intel_bo_fake_alloc_static(intel->bufmgr,
-						  name,
-						  region_desc->offset,
-						  region->pitch * region->cpp *
-						  region->height,
-						  region->classic_map);
-
-      /* The sarea just gives us a boolean for whether it's tiled or not,
-       * instead of which tiling mode it is.  Guess.
-       */
-      if (region_desc->tiled) {
-	 if (IS_965(intel->intelScreen->deviceID) &&
-	     region_desc == &intelScreen->depth)
-	    region->tiling = I915_TILING_Y;
-	 else
-	    region->tiling = I915_TILING_X;
-      } else {
-	 region->tiling = I915_TILING_NONE;
-      }
-
-      region->bit_6_swizzle = I915_BIT_6_SWIZZLE_NONE;
+      region->buffer = dri_bo_alloc_static(intel->bufmgr,
+					   name,
+					   region_desc->offset,
+					   intelScreen->pitch *
+					   intelScreen->height,
+					   region_desc->map,
+					   DRM_BO_FLAG_MEM_TT);
    }
 
    assert(region->buffer != NULL);
diff --git a/src/mesa/drivers/dri/intel/intel_regions.h b/src/mesa/drivers/dri/intel/intel_regions.h
index e5f19fbb45..229f79aeba 100644
--- a/src/mesa/drivers/dri/intel/intel_regions.h
+++ b/src/mesa/drivers/dri/intel/intel_regions.h
@@ -28,12 +28,6 @@
 #ifndef INTEL_REGIONS_H
 #define INTEL_REGIONS_H
 
-/** @file intel_regions.h
- *
- * Structure definitions and prototypes for intel_region handling, which is
- * the basic structure for rectangular collections of pixels stored in a dri_bo.
- */
-
 #include "mtypes.h"
 #include "dri_bufmgr.h"
 
@@ -59,9 +53,8 @@ struct intel_region
    GLuint map_refcount;  /**< Reference count for mapping */
 
    GLuint draw_offset; /**< Offset of drawing address within the region */
-   uint32_t tiling; /**< Which tiling mode the region is in */
-   uint32_t bit_6_swizzle; /**< GEM flag for address swizzling requirement */
-   drmAddress classic_map; /**< drmMap of the region when not in GEM mode */
+   GLboolean tiled; /**< True if the region is X or Y-tiled.  Used on 965. */
+
    struct intel_buffer_object *pbo;     /* zero-copy uploads */
 };
 
@@ -76,7 +69,7 @@ struct intel_region *intel_region_alloc(struct intel_context *intel,
 struct intel_region *
 intel_region_alloc_for_handle(struct intel_context *intel,
 			      GLuint cpp, GLuint pitch, GLuint height,
-			      unsigned int handle);
+			      GLuint tiled, unsigned int handle);
 
 void intel_region_reference(struct intel_region **dst,
                             struct intel_region *src);
diff --git a/src/mesa/drivers/dri/intel/intel_screen.c b/src/mesa/drivers/dri/intel/intel_screen.c
index 9ed89906d5..5dded4b167 100644
--- a/src/mesa/drivers/dri/intel/intel_screen.c
+++ b/src/mesa/drivers/dri/intel/intel_screen.c
@@ -49,7 +49,7 @@
 #include "i830_dri.h"
 #include "intel_regions.h"
 #include "intel_batchbuffer.h"
-#include "intel_bufmgr.h"
+#include "intel_bufmgr_ttm.h"
 
 PUBLIC const char __driConfigOptions[] =
    DRI_CONF_BEGIN
@@ -59,7 +59,7 @@ PUBLIC const char __driConfigOptions[] =
       /* Options correspond to DRI_CONF_BO_REUSE_DISABLED,
        * DRI_CONF_BO_REUSE_ALL
        */
-      DRI_CONF_OPT_BEGIN_V(bo_reuse, enum, 1, "0:1")
+      DRI_CONF_OPT_BEGIN_V(bo_reuse, enum, 0, "0:1")
 	 DRI_CONF_DESC_BEGIN(en, "Buffer object reuse")
 	    DRI_CONF_ENUM(0, "Disable buffer object reuse")
 	    DRI_CONF_ENUM(1, "Enable reuse of all sizes of buffer objects")
@@ -90,6 +90,51 @@ intelMapScreenRegions(__DRIscreenPrivate * sPriv)
 {
    intelScreenPrivate *intelScreen = (intelScreenPrivate *) sPriv->private;
 
+   if (intelScreen->front.handle) {
+      if (drmMap(sPriv->fd,
+                 intelScreen->front.handle,
+                 intelScreen->front.size,
+                 (drmAddress *) & intelScreen->front.map) != 0) {
+         _mesa_problem(NULL, "drmMap(frontbuffer) failed!");
+         return GL_FALSE;
+      }
+   }
+   else {
+      _mesa_warning(NULL, "no front buffer handle in intelMapScreenRegions!");
+   }
+
+   if (0)
+      _mesa_printf("Back 0x%08x ", intelScreen->back.handle);
+   if (drmMap(sPriv->fd,
+              intelScreen->back.handle,
+              intelScreen->back.size,
+              (drmAddress *) & intelScreen->back.map) != 0) {
+      intelUnmapScreenRegions(intelScreen);
+      return GL_FALSE;
+   }
+
+   if (intelScreen->third.handle) {
+      if (0)
+	 _mesa_printf("Third 0x%08x ", intelScreen->third.handle);
+      if (drmMap(sPriv->fd,
+		 intelScreen->third.handle,
+		 intelScreen->third.size,
+		 (drmAddress *) & intelScreen->third.map) != 0) {
+	 intelUnmapScreenRegions(intelScreen);
+	 return GL_FALSE;
+      }
+   }
+
+   if (0)
+      _mesa_printf("Depth 0x%08x ", intelScreen->depth.handle);
+   if (drmMap(sPriv->fd,
+              intelScreen->depth.handle,
+              intelScreen->depth.size,
+              (drmAddress *) & intelScreen->depth.map) != 0) {
+      intelUnmapScreenRegions(intelScreen);
+      return GL_FALSE;
+   }
+
    if (0)
       _mesa_printf("TEX 0x%08x ", intelScreen->tex.handle);
    if (intelScreen->tex.size != 0) {
@@ -102,15 +147,50 @@ intelMapScreenRegions(__DRIscreenPrivate * sPriv)
       }
    }
 
+   if (0)
+      printf("Mappings:  front: %p  back: %p  third: %p  depth: %p  tex: %p\n",
+             intelScreen->front.map,
+             intelScreen->back.map, intelScreen->third.map,
+             intelScreen->depth.map, intelScreen->tex.map);
    return GL_TRUE;
 }
 
 void
 intelUnmapScreenRegions(intelScreenPrivate * intelScreen)
 {
+#define REALLY_UNMAP 1
+   if (intelScreen->front.map) {
+#if REALLY_UNMAP
+      if (drmUnmap(intelScreen->front.map, intelScreen->front.size) != 0)
+         printf("drmUnmap front failed!\n");
+#endif
+      intelScreen->front.map = NULL;
+   }
+   if (intelScreen->back.map) {
+#if REALLY_UNMAP
+      if (drmUnmap(intelScreen->back.map, intelScreen->back.size) != 0)
+         printf("drmUnmap back failed!\n");
+#endif
+      intelScreen->back.map = NULL;
+   }
+   if (intelScreen->third.map) {
+#if REALLY_UNMAP
+      if (drmUnmap(intelScreen->third.map, intelScreen->third.size) != 0)
+         printf("drmUnmap third failed!\n");
+#endif
+      intelScreen->third.map = NULL;
+   }
+   if (intelScreen->depth.map) {
+#if REALLY_UNMAP
+      drmUnmap(intelScreen->depth.map, intelScreen->depth.size);
+      intelScreen->depth.map = NULL;
+#endif
+   }
    if (intelScreen->tex.map) {
+#if REALLY_UNMAP
       drmUnmap(intelScreen->tex.map, intelScreen->tex.size);
       intelScreen->tex.map = NULL;
+#endif
    }
 }
 
@@ -141,16 +221,16 @@ intelPrintSAREA(const struct drm_i915_sarea * sarea)
            sarea->height);
    fprintf(stderr, "SAREA: pitch: %d\n", sarea->pitch);
    fprintf(stderr,
-           "SAREA: front offset: 0x%08x  size: 0x%x  handle: 0x%x tiled: %d\n",
+           "SAREA: front offset: 0x%08x  size: 0x%x  handle: 0x%x\n",
            sarea->front_offset, sarea->front_size,
-           (unsigned) sarea->front_handle, sarea->front_tiled);
+           (unsigned) sarea->front_handle);
    fprintf(stderr,
-           "SAREA: back  offset: 0x%08x  size: 0x%x  handle: 0x%x tiled: %d\n",
+           "SAREA: back  offset: 0x%08x  size: 0x%x  handle: 0x%x\n",
            sarea->back_offset, sarea->back_size,
-           (unsigned) sarea->back_handle, sarea->back_tiled);
-   fprintf(stderr, "SAREA: depth offset: 0x%08x  size: 0x%x  handle: 0x%x tiled: %d\n",
+           (unsigned) sarea->back_handle);
+   fprintf(stderr, "SAREA: depth offset: 0x%08x  size: 0x%x  handle: 0x%x\n",
            sarea->depth_offset, sarea->depth_size,
-           (unsigned) sarea->depth_handle, sarea->depth_tiled);
+           (unsigned) sarea->depth_handle);
    fprintf(stderr, "SAREA: tex   offset: 0x%08x  size: 0x%x  handle: 0x%x\n",
            sarea->tex_offset, sarea->tex_size, (unsigned) sarea->tex_handle);
 }
@@ -254,6 +334,8 @@ intelHandleDrawableConfig(__DRIdrawablePrivate *dPriv,
     * attached. */
 }
 
+#define BUFFER_FLAG_TILED 0x0100
+
 /**
  * DRI2 entrypoint
  */
@@ -266,6 +348,7 @@ intelHandleBufferAttach(__DRIdrawablePrivate *dPriv,
    struct intel_renderbuffer *rb;
    struct intel_region *region;
    struct intel_context *intel = pcp->driverPrivate;
+   GLuint tiled;
 
    switch (ba->buffer.attachment) {
    case DRI_DRAWABLE_BUFFER_FRONT_LEFT:
@@ -299,9 +382,10 @@ intelHandleBufferAttach(__DRIdrawablePrivate *dPriv,
       return;
 #endif
 
+   tiled = (ba->buffer.flags & BUFFER_FLAG_TILED) > 0;
    region = intel_region_alloc_for_handle(intel, ba->buffer.cpp,
 					  ba->buffer.pitch / ba->buffer.cpp,
-					  dPriv->h,
+					  dPriv->h, tiled,
 					  ba->buffer.handle);
 
    intel_renderbuffer_set_region(rb, region);
@@ -446,13 +530,14 @@ intelCreateBuffer(__DRIscreenPrivate * driScrnPriv,
       _mesa_initialize_framebuffer(&intel_fb->Base, mesaVis);
 
       /* setup the hardware-based renderbuffers */
-      intel_fb->color_rb[0] = intel_create_renderbuffer(rgbFormat);
-      _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_FRONT_LEFT,
-			     &intel_fb->color_rb[0]->Base);
+      {
+         intel_fb->color_rb[0] = intel_create_renderbuffer(rgbFormat);
+         _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_FRONT_LEFT,
+				&intel_fb->color_rb[0]->Base);
+      }
 
       if (mesaVis->doubleBufferMode) {
-	 intel_fb->color_rb[1] = intel_create_renderbuffer(rgbFormat);
-
+         intel_fb->color_rb[1] = intel_create_renderbuffer(rgbFormat);
          _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_BACK_LEFT,
 				&intel_fb->color_rb[1]->Base);
 
@@ -484,7 +569,7 @@ intelCreateBuffer(__DRIscreenPrivate * driScrnPriv,
       else if (mesaVis->depthBits == 16) {
          /* just 16-bit depth buffer, no hw stencil */
          struct intel_renderbuffer *depthRb
-	    = intel_create_renderbuffer(GL_DEPTH_COMPONENT16);
+            = intel_create_renderbuffer(GL_DEPTH_COMPONENT16);
          _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_DEPTH, &depthRb->Base);
       }
 
diff --git a/src/mesa/drivers/dri/intel/intel_screen.h b/src/mesa/drivers/dri/intel/intel_screen.h
index 9a73b13951..e62b2d7c89 100644
--- a/src/mesa/drivers/dri/intel/intel_screen.h
+++ b/src/mesa/drivers/dri/intel/intel_screen.h
@@ -74,8 +74,6 @@ typedef struct
    int irq_active;
    int allow_batchbuffer;
 
-   int ttm;
-
    /**
    * Configuration cache with default values for all contexts
    */
diff --git a/src/mesa/drivers/dri/intel/intel_span.c b/src/mesa/drivers/dri/intel/intel_span.c
index edede3a74b..742b1b8735 100644
--- a/src/mesa/drivers/dri/intel/intel_span.c
+++ b/src/mesa/drivers/dri/intel/intel_span.c
@@ -39,224 +39,6 @@
 
 #include "swrast/swrast.h"
 
-static void
-intel_set_span_functions(struct intel_context *intel,
-			 struct gl_renderbuffer *rb);
-
-#define SPAN_CACHE_SIZE		4096
-
-static void
-get_span_cache(struct intel_renderbuffer *irb, uint32_t offset)
-{
-   if (irb->span_cache == NULL) {
-      irb->span_cache = _mesa_malloc(SPAN_CACHE_SIZE);
-      irb->span_cache_offset = -1;
-   }
-
-   if ((offset & ~(SPAN_CACHE_SIZE - 1)) != irb->span_cache_offset) {
-      irb->span_cache_offset = offset & ~(SPAN_CACHE_SIZE - 1);
-      dri_bo_get_subdata(irb->region->buffer, irb->span_cache_offset,
-			 SPAN_CACHE_SIZE, irb->span_cache);
-   }
-}
-
-static void
-clear_span_cache(struct intel_renderbuffer *irb)
-{
-   irb->span_cache_offset = -1;
-}
-
-static uint32_t
-pread_32(struct intel_renderbuffer *irb, uint32_t offset)
-{
-   get_span_cache(irb, offset);
-
-   return *(uint32_t *)(irb->span_cache + (offset & (SPAN_CACHE_SIZE - 1)));
-}
-
-static uint32_t
-pread_xrgb8888(struct intel_renderbuffer *irb, uint32_t offset)
-{
-   get_span_cache(irb, offset);
-
-   return *(uint32_t *)(irb->span_cache + (offset & (SPAN_CACHE_SIZE - 1))) |
-      0xff000000;
-}
-
-static uint16_t
-pread_16(struct intel_renderbuffer *irb, uint32_t offset)
-{
-   get_span_cache(irb, offset);
-
-   return *(uint16_t *)(irb->span_cache + (offset & (SPAN_CACHE_SIZE - 1)));
-}
-
-static uint8_t
-pread_8(struct intel_renderbuffer *irb, uint32_t offset)
-{
-   get_span_cache(irb, offset);
-
-   return *(uint8_t *)(irb->span_cache + (offset & (SPAN_CACHE_SIZE - 1)));
-}
-
-static void
-pwrite_32(struct intel_renderbuffer *irb, uint32_t offset, uint32_t val)
-{
-   clear_span_cache(irb);
-
-   dri_bo_subdata(irb->region->buffer, offset, 4, &val);
-}
-
-static void
-pwrite_xrgb8888(struct intel_renderbuffer *irb, uint32_t offset, uint32_t val)
-{
-   clear_span_cache(irb);
-
-   dri_bo_subdata(irb->region->buffer, offset, 3, &val);
-}
-
-static void
-pwrite_16(struct intel_renderbuffer *irb, uint32_t offset, uint16_t val)
-{
-   clear_span_cache(irb);
-
-   dri_bo_subdata(irb->region->buffer, offset, 2, &val);
-}
-
-static void
-pwrite_8(struct intel_renderbuffer *irb, uint32_t offset, uint8_t val)
-{
-   clear_span_cache(irb);
-
-   dri_bo_subdata(irb->region->buffer, offset, 1, &val);
-}
-
-static uint32_t no_tile_swizzle(struct intel_renderbuffer *irb,
-				struct intel_context *intel,
-				int x, int y)
-{
-	x += intel->drawX;
-	y += intel->drawY;
-
-	return (y * irb->region->pitch + x) * irb->region->cpp;
-}
-
-/*
- * Deal with tiled surfaces
- */
-
-static uint32_t x_tile_swizzle(struct intel_renderbuffer *irb,
-			       struct intel_context *intel,
-			       int x, int y)
-{
-	int	tile_stride;
-	int	xbyte;
-	int	x_tile_off, y_tile_off;
-	int	x_tile_number, y_tile_number;
-	int	tile_off, tile_base;
-	
-	tile_stride = (irb->pfPitch * irb->region->cpp) << 3;
-	
-	x += intel->drawX;
-	y += intel->drawY;
-
-	xbyte = x * irb->region->cpp;
-
-	x_tile_off = xbyte & 0x1ff;
-	y_tile_off = y & 7;
-
-	x_tile_number = xbyte >> 9;
-	y_tile_number = y >> 3;
-
-	tile_off = (y_tile_off << 9) + x_tile_off;
-
-	switch (irb->region->bit_6_swizzle) {
-	case I915_BIT_6_SWIZZLE_NONE:
-	   break;
-	case I915_BIT_6_SWIZZLE_9:
-	   tile_off ^= ((tile_off >> 3) & 64);
-	   break;
-	case I915_BIT_6_SWIZZLE_9_10:
-	   tile_off ^= ((tile_off >> 3) & 64) ^ ((tile_off >> 4) & 64);
-	   break;
-	case I915_BIT_6_SWIZZLE_9_11:
-	   tile_off ^= ((tile_off >> 3) & 64) ^ ((tile_off >> 5) & 64);
-	   break;
-	case I915_BIT_6_SWIZZLE_9_10_11:
-	   tile_off ^= ((tile_off >> 3) & 64) ^ ((tile_off >> 4) & 64) ^
-	      ((tile_off >> 5) & 64);
-	   break;
-	default:
-	   fprintf(stderr, "Unknown tile swizzling mode %d\n",
-		   irb->region->bit_6_swizzle);
-	   exit(1);
-	}
-
-	tile_base = (x_tile_number << 12) + y_tile_number * tile_stride;
-
-#if 0
-	printf("(%d,%d) -> %d + %d = %d (pitch = %d, tstride = %d)\n",
-	       x, y, tile_off, tile_base,
-	       tile_off + tile_base,
-	       irb->pfPitch, tile_stride);
-#endif
-
-	return tile_base + tile_off;
-}
-
-static uint32_t y_tile_swizzle(struct intel_renderbuffer *irb,
-			       struct intel_context *intel,
-			       int x, int y)
-{
-	int	tile_stride;
-	int	xbyte;
-	int	x_tile_off, y_tile_off;
-	int	x_tile_number, y_tile_number;
-	int	tile_off, tile_base;
-	
-	tile_stride = (irb->pfPitch * irb->region->cpp) << 5;
-	
-	x += intel->drawX;
-	y += intel->drawY;
-
-	xbyte = x * irb->region->cpp;
-
-	x_tile_off = xbyte & 0x7f;
-	y_tile_off = y & 0x1f;
-
-	x_tile_number = xbyte >> 7;
-	y_tile_number = y >> 5;
-
-	tile_off = ((x_tile_off & ~0xf) << 5) + (y_tile_off << 4) +
-	   (x_tile_off & 0xf);
-
-	switch (irb->region->bit_6_swizzle) {
-	case I915_BIT_6_SWIZZLE_NONE:
-	   break;
-	case I915_BIT_6_SWIZZLE_9:
-	   tile_off ^= ((tile_off >> 3) & 64);
-	   break;
-	case I915_BIT_6_SWIZZLE_9_10:
-	   tile_off ^= ((tile_off >> 3) & 64) ^ ((tile_off >> 4) & 64);
-	   break;
-	case I915_BIT_6_SWIZZLE_9_11:
-	   tile_off ^= ((tile_off >> 3) & 64) ^ ((tile_off >> 5) & 64);
-	   break;
-	case I915_BIT_6_SWIZZLE_9_10_11:
-	   tile_off ^= ((tile_off >> 3) & 64) ^ ((tile_off >> 4) & 64) ^
-	      ((tile_off >> 5) & 64);
-	   break;
-	default:
-	   fprintf(stderr, "Unknown tile swizzling mode %d\n",
-		   irb->region->bit_6_swizzle);
-	   exit(1);
-	}
-
-	tile_base = (x_tile_number << 12) + y_tile_number * tile_stride;
-
-	return tile_base + tile_off;
-}
-
 /*
   break intelWriteRGBASpan_ARGB8888
 */
@@ -269,7 +51,10 @@ static uint32_t y_tile_swizzle(struct intel_renderbuffer *irb,
    struct intel_renderbuffer *irb = intel_renderbuffer(rb);		\
    const GLint yScale = irb->RenderToTexture ? 1 : -1;			\
    const GLint yBias = irb->RenderToTexture ? 0 : irb->Base.Height - 1;	\
+   GLubyte *buf = (GLubyte *) irb->pfMap				\
+      + (intel->drawY * irb->pfPitch + intel->drawX) * irb->region->cpp;\
    GLuint p;								\
+   assert(irb->pfMap);\
    (void) p;
 
 /* XXX FBO: this is identical to the macro in spantmp2.h except we get
@@ -284,14 +69,12 @@ static uint32_t y_tile_swizzle(struct intel_renderbuffer *irb,
 	 int miny = intel->pClipRects[_nc].y1 - intel->drawY;		\
 	 int maxx = intel->pClipRects[_nc].x2 - intel->drawX;		\
 	 int maxy = intel->pClipRects[_nc].y2 - intel->drawY;
-	
-#if 0
-      }}
-#endif
+
+
+
 
 #define Y_FLIP(_y) ((_y) * yScale + yBias)
 
-/* XXX with GEM, these need to tell the kernel */
 #define HW_LOCK()
 
 #define HW_UNLOCK()
@@ -303,8 +86,7 @@ static uint32_t y_tile_swizzle(struct intel_renderbuffer *irb,
 
 #define TAG(x)    intel##x##_RGB565
 #define TAG2(x,y) intel##x##_RGB565##y
-#define GET_VALUE(X, Y) pread_16(irb, no_tile_swizzle(irb, intel, X, Y))
-#define PUT_VALUE(X, Y, V) pwrite_16(irb, no_tile_swizzle(irb, intel, X, Y), V)
+#define GET_PTR(X,Y) (buf + ((Y) * irb->pfPitch + (X)) * 2)
 #include "spantmp2.h"
 
 /* 32 bit, ARGB8888 color spanline and pixel functions
@@ -314,89 +96,17 @@ static uint32_t y_tile_swizzle(struct intel_renderbuffer *irb,
 
 #define TAG(x)    intel##x##_ARGB8888
 #define TAG2(x,y) intel##x##_ARGB8888##y
-#define GET_VALUE(X, Y) pread_32(irb, no_tile_swizzle(irb, intel, X, Y))
-#define PUT_VALUE(X, Y, V) pwrite_32(irb, no_tile_swizzle(irb, intel, X, Y), V)
-#include "spantmp2.h"
-
-/* 32 bit, xRGB8888 color spanline and pixel functions
- */
-#define SPANTMP_PIXEL_FMT GL_BGRA
-#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV
-
-#define TAG(x)    intel##x##_xRGB8888
-#define TAG2(x,y) intel##x##_xRGB8888##y
-#define GET_VALUE(X, Y) pread_xrgb8888(irb, no_tile_swizzle(irb, intel, X, Y))
-#define PUT_VALUE(X, Y, V) pwrite_xrgb8888(irb, no_tile_swizzle(irb, intel, X, Y), V)
-#include "spantmp2.h"
-
-/* 16 bit RGB565 color tile spanline and pixel functions
- */
-
-#define SPANTMP_PIXEL_FMT GL_RGB
-#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_SHORT_5_6_5
-
-#define TAG(x)    intel_XTile_##x##_RGB565
-#define TAG2(x,y) intel_XTile_##x##_RGB565##y
-#define GET_VALUE(X, Y) pread_16(irb, x_tile_swizzle(irb, intel, X, Y))
-#define PUT_VALUE(X, Y, V) pwrite_16(irb, x_tile_swizzle(irb, intel, X, Y), V)
-#include "spantmp2.h"
-
-#define SPANTMP_PIXEL_FMT GL_RGB
-#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_SHORT_5_6_5
-
-#define TAG(x)    intel_YTile_##x##_RGB565
-#define TAG2(x,y) intel_YTile_##x##_RGB565##y
-#define GET_VALUE(X, Y) pread_16(irb, y_tile_swizzle(irb, intel, X, Y))
-#define PUT_VALUE(X, Y, V) pwrite_16(irb, y_tile_swizzle(irb, intel, X, Y), V)
-#include "spantmp2.h"
-
-/* 32 bit ARGB888 color tile spanline and pixel functions
- */
-
-#define SPANTMP_PIXEL_FMT GL_BGRA
-#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV
-
-#define TAG(x)    intel_XTile_##x##_ARGB8888
-#define TAG2(x,y) intel_XTile_##x##_ARGB8888##y
-#define GET_VALUE(X, Y) pread_32(irb, x_tile_swizzle(irb, intel, X, Y))
-#define PUT_VALUE(X, Y, V) pwrite_32(irb, x_tile_swizzle(irb, intel, X, Y), V)
-#include "spantmp2.h"
-
-#define SPANTMP_PIXEL_FMT GL_BGRA
-#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV
-
-#define TAG(x)    intel_YTile_##x##_ARGB8888
-#define TAG2(x,y) intel_YTile_##x##_ARGB8888##y
-#define GET_VALUE(X, Y) pread_32(irb, y_tile_swizzle(irb, intel, X, Y))
-#define PUT_VALUE(X, Y, V) pwrite_32(irb, y_tile_swizzle(irb, intel, X, Y), V)
-#include "spantmp2.h"
-
-/* 32 bit xRGB888 color tile spanline and pixel functions
- */
-
-#define SPANTMP_PIXEL_FMT GL_BGRA
-#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV
-
-#define TAG(x)    intel_XTile_##x##_xRGB8888
-#define TAG2(x,y) intel_XTile_##x##_xRGB8888##y
-#define GET_VALUE(X, Y) pread_xrgb8888(irb, x_tile_swizzle(irb, intel, X, Y))
-#define PUT_VALUE(X, Y, V) pwrite_xrgb8888(irb, x_tile_swizzle(irb, intel, X, Y), V)
-#include "spantmp2.h"
-
-#define SPANTMP_PIXEL_FMT GL_BGRA
-#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV
-
-#define TAG(x)    intel_YTile_##x##_xRGB8888
-#define TAG2(x,y) intel_YTile_##x##_xRGB8888##y
-#define GET_VALUE(X, Y) pread_xrgb8888(irb, y_tile_swizzle(irb, intel, X, Y))
-#define PUT_VALUE(X, Y, V) pwrite_xrgb8888(irb, y_tile_swizzle(irb, intel, X, Y), V)
+#define GET_PTR(X,Y) (buf + ((Y) * irb->pfPitch + (X)) * 4)
 #include "spantmp2.h"
 
 #define LOCAL_DEPTH_VARS						\
    struct intel_context *intel = intel_context(ctx);			\
    struct intel_renderbuffer *irb = intel_renderbuffer(rb);		\
+   const GLuint pitch = irb->pfPitch/***XXX region->pitch*/; /* in pixels */ \
    const GLint yScale = irb->RenderToTexture ? 1 : -1;			\
-   const GLint yBias = irb->RenderToTexture ? 0 : irb->Base.Height - 1;
+   const GLint yBias = irb->RenderToTexture ? 0 : irb->Base.Height - 1;	\
+   char *buf = (char *) irb->pfMap/*XXX use region->map*/ +             \
+      (intel->drawY * pitch + intel->drawX) * irb->region->cpp;
 
 
 #define LOCAL_STENCIL_VARS LOCAL_DEPTH_VARS
@@ -405,34 +115,15 @@ static uint32_t y_tile_swizzle(struct intel_renderbuffer *irb,
  ** 16-bit depthbuffer functions.
  **/
 #define VALUE_TYPE GLushort
-#define WRITE_DEPTH(_x, _y, d) \
-   pwrite_16(irb, no_tile_swizzle(irb, intel, _x, _y), d)
-#define READ_DEPTH(d, _x, _y) \
-   d = pread_16(irb, no_tile_swizzle(irb, intel, _x, _y))
-#define TAG(x) intel##x##_z16
-#include "depthtmp.h"
 
+#define WRITE_DEPTH( _x, _y, d ) \
+   ((GLushort *)buf)[(_x) + (_y) * pitch] = d;
 
-/**
- ** 16-bit x tile depthbuffer functions.
- **/
-#define VALUE_TYPE GLushort
-#define WRITE_DEPTH(_x, _y, d) \
-   pwrite_16(irb, x_tile_swizzle(irb, intel, _x, _y), d)
-#define READ_DEPTH(d, _x, _y) \
-   d = pread_16(irb, x_tile_swizzle(irb, intel, _x, _y))
-#define TAG(x) intel_XTile_##x##_z16
-#include "depthtmp.h"
+#define READ_DEPTH( d, _x, _y )	\
+   d = ((GLushort *)buf)[(_x) + (_y) * pitch];
 
-/**
- ** 16-bit y tile depthbuffer functions.
- **/
-#define VALUE_TYPE GLushort
-#define WRITE_DEPTH(_x, _y, d) \
-   pwrite_16(irb, y_tile_swizzle(irb, intel, _x, _y), d)
-#define READ_DEPTH(d, _x, _y) \
-   d = pread_16(irb, y_tile_swizzle(irb, intel, _x, _y))
-#define TAG(x) intel_YTile_##x##_z16
+
+#define TAG(x) intel##x##_z16
 #include "depthtmp.h"
 
 
@@ -445,129 +136,38 @@ static uint32_t y_tile_swizzle(struct intel_renderbuffer *irb,
 #define VALUE_TYPE GLuint
 
 /* Change ZZZS -> SZZZ */
-#define WRITE_DEPTH(_x, _y, d)					\
-   pwrite_32(irb, no_tile_swizzle(irb, intel, _x, _y),		\
-	     ((d) >> 8) | ((d) << 24))
-
-/* Change SZZZ -> ZZZS */
-#define READ_DEPTH( d, _x, _y ) {				\
-   GLuint tmp = pread_32(irb, no_tile_swizzle(irb, intel, _x, _y));	\
-   d = (tmp << 8) | (tmp >> 24);				\
+#define WRITE_DEPTH( _x, _y, d ) {				\
+   GLuint tmp = ((d) >> 8) | ((d) << 24);			\
+   ((GLuint *)buf)[(_x) + (_y) * pitch] = tmp;			\
 }
 
-#define TAG(x) intel##x##_z24_s8
-#include "depthtmp.h"
-
-
-/**
- ** 24/8-bit x-tile interleaved depth/stencil functions
- ** Note: we're actually reading back combined depth+stencil values.
- ** The wrappers in main/depthstencil.c are used to extract the depth
- ** and stencil values.
- **/
-#define VALUE_TYPE GLuint
-
-/* Change ZZZS -> SZZZ */
-#define WRITE_DEPTH(_x, _y, d)					\
-   pwrite_32(irb, x_tile_swizzle(irb, intel, _x, _y),		\
-	     ((d) >> 8) | ((d) << 24))				\
-
 /* Change SZZZ -> ZZZS */
 #define READ_DEPTH( d, _x, _y ) {				\
-   GLuint tmp = pread_32(irb, x_tile_swizzle(irb, intel, _x, _y));	\
+   GLuint tmp = ((GLuint *)buf)[(_x) + (_y) * pitch];		\
    d = (tmp << 8) | (tmp >> 24);				\
 }
 
-#define TAG(x) intel_XTile_##x##_z24_s8
-#include "depthtmp.h"
-
-/**
- ** 24/8-bit y-tile interleaved depth/stencil functions
- ** Note: we're actually reading back combined depth+stencil values.
- ** The wrappers in main/depthstencil.c are used to extract the depth
- ** and stencil values.
- **/
-#define VALUE_TYPE GLuint
-
-/* Change ZZZS -> SZZZ */
-#define WRITE_DEPTH(_x, _y, d)					\
-   pwrite_32(irb, y_tile_swizzle(irb, intel, _x, _y),		\
-	     ((d) >> 8) | ((d) << 24))
-
-/* Change SZZZ -> ZZZS */
-#define READ_DEPTH( d, _x, _y ) {				\
-   GLuint tmp = pread_32(irb, y_tile_swizzle(irb, intel, _x, _y));	\
-   d = (tmp << 8) | (tmp >> 24);				\
-}
-
-#define TAG(x) intel_YTile_##x##_z24_s8
+#define TAG(x) intel##x##_z24_s8
 #include "depthtmp.h"
 
 
 /**
  ** 8-bit stencil function (XXX FBO: This is obsolete)
  **/
-#define WRITE_STENCIL(_x, _y, d)				\
-   pwrite_8(irb, no_tile_swizzle(irb, intel, _x, _y) + 3, d)
+#define WRITE_STENCIL( _x, _y, d ) {				\
+   GLuint tmp = ((GLuint *)buf)[(_x) + (_y) * pitch];		\
+   tmp &= 0xffffff;						\
+   tmp |= ((d) << 24);						\
+   ((GLuint *) buf)[(_x) + (_y) * pitch] = tmp;			\
+}
 
-#define READ_STENCIL(d, _x, _y)					\
-   d = pread_8(irb, no_tile_swizzle(irb, intel, _x, _y) + 3);
+#define READ_STENCIL( d, _x, _y )				\
+   d = ((GLuint *)buf)[(_x) + (_y) * pitch] >> 24;
 
 #define TAG(x) intel##x##_z24_s8
 #include "stenciltmp.h"
 
-/**
- ** 8-bit x-tile stencil function (XXX FBO: This is obsolete)
- **/
-#define WRITE_STENCIL(_x, _y, d)				\
-   pwrite_8(irb, x_tile_swizzle(irb, intel, _x, _y) + 3, d)
 
-#define READ_STENCIL(d, _x, _y)					\
-   d = pread_8(irb, x_tile_swizzle(irb, intel, _x, _y) + 3);
-
-#define TAG(x) intel_XTile_##x##_z24_s8
-#include "stenciltmp.h"
-
-/**
- ** 8-bit y-tile stencil function (XXX FBO: This is obsolete)
- **/
-#define WRITE_STENCIL(_x, _y, d)				\
-   pwrite_8(irb, y_tile_swizzle(irb, intel, _x, _y) + 3, d)
-
-#define READ_STENCIL(d, _x, _y)					\
-   d = pread_8(irb, y_tile_swizzle(irb, intel, _x, _y) + 3)
-
-#define TAG(x) intel_YTile_##x##_z24_s8
-#include "stenciltmp.h"
-
-void
-intel_renderbuffer_map(struct intel_context *intel, struct gl_renderbuffer *rb)
-{
-   struct intel_renderbuffer *irb = intel_renderbuffer(rb);
-
-   if (irb == NULL || irb->region == NULL)
-      return;
-
-   irb->pfPitch = irb->region->pitch;
-
-   intel_set_span_functions(intel, rb);
-}
-
-void
-intel_renderbuffer_unmap(struct intel_context *intel,
-			 struct gl_renderbuffer *rb)
-{
-   struct intel_renderbuffer *irb = intel_renderbuffer(rb);
-
-   if (irb == NULL || irb->region == NULL)
-      return;
-
-   clear_span_cache(irb);
-   irb->pfPitch = 0;
-
-   rb->GetRow = NULL;
-   rb->PutRow = NULL;
-}
 
 /**
  * Map or unmap all the renderbuffers which we may need during
@@ -586,13 +186,23 @@ intel_map_unmap_buffers(struct intel_context *intel, GLboolean map)
 {
    GLcontext *ctx = &intel->ctx;
    GLuint i, j;
+   struct intel_renderbuffer *irb;
 
    /* color draw buffers */
    for (j = 0; j < ctx->DrawBuffer->_NumColorDrawBuffers; j++) {
-      if (map)
-	 intel_renderbuffer_map(intel, ctx->DrawBuffer->_ColorDrawBuffers[j]);
-      else
-	 intel_renderbuffer_unmap(intel, ctx->DrawBuffer->_ColorDrawBuffers[j]);
+      struct gl_renderbuffer *rb = ctx->DrawBuffer->_ColorDrawBuffers[j];
+      irb = intel_renderbuffer(rb);
+      if (irb) {
+         /* this is a user-created intel_renderbuffer */
+         if (irb->region) {
+            if (map)
+               intel_region_map(intel, irb->region);
+            else
+               intel_region_unmap(intel, irb->region);
+            irb->pfMap = irb->region->map;
+            irb->pfPitch = irb->region->pitch;
+         }
+      }
    }
 
    /* check for render to textures */
@@ -615,28 +225,77 @@ intel_map_unmap_buffers(struct intel_context *intel, GLboolean map)
    }
 
    /* color read buffers */
-   if (map)
-      intel_renderbuffer_map(intel, ctx->ReadBuffer->_ColorReadBuffer);
-   else
-      intel_renderbuffer_unmap(intel, ctx->ReadBuffer->_ColorReadBuffer);
+   irb = intel_renderbuffer(ctx->ReadBuffer->_ColorReadBuffer);
+   if (irb && irb->region) {
+      if (map)
+         intel_region_map(intel, irb->region);
+      else
+         intel_region_unmap(intel, irb->region);
+      irb->pfMap = irb->region->map;
+      irb->pfPitch = irb->region->pitch;
+   }
+
+   /* Account for front/back color page flipping.
+    * The span routines use the pfMap and pfPitch fields which will
+    * swap the front/back region map/pitch if we're page flipped.
+    * Do this after mapping, above, so the map field is valid.
+    */
+#if 0
+   if (map && ctx->DrawBuffer->Name == 0) {
+      struct intel_renderbuffer *irbFront
+         = intel_get_renderbuffer(ctx->DrawBuffer, BUFFER_FRONT_LEFT);
+      struct intel_renderbuffer *irbBack
+         = intel_get_renderbuffer(ctx->DrawBuffer, BUFFER_BACK_LEFT);
+      if (irbBack) {
+         /* double buffered */
+         if (intel->sarea->pf_current_page == 0) {
+            irbFront->pfMap = irbFront->region->map;
+            irbFront->pfPitch = irbFront->region->pitch;
+            irbBack->pfMap = irbBack->region->map;
+            irbBack->pfPitch = irbBack->region->pitch;
+         }
+         else {
+            irbFront->pfMap = irbBack->region->map;
+            irbFront->pfPitch = irbBack->region->pitch;
+            irbBack->pfMap = irbFront->region->map;
+            irbBack->pfPitch = irbFront->region->pitch;
+         }
+      }
+   }
+#endif
 
    /* depth buffer (Note wrapper!) */
    if (ctx->DrawBuffer->_DepthBuffer) {
-      if (map)
-	 intel_renderbuffer_map(intel, ctx->DrawBuffer->_DepthBuffer->Wrapped);
-      else
-	 intel_renderbuffer_unmap(intel,
-				  ctx->DrawBuffer->_DepthBuffer->Wrapped);
+      irb = intel_renderbuffer(ctx->DrawBuffer->_DepthBuffer->Wrapped);
+      if (irb && irb->region) {
+         if (map) {
+            intel_region_map(intel, irb->region);
+            irb->pfMap = irb->region->map;
+            irb->pfPitch = irb->region->pitch;
+         }
+         else {
+            intel_region_unmap(intel, irb->region);
+            irb->pfMap = irb->region->map;
+            irb->pfPitch = irb->region->pitch;
+         }
+      }
    }
 
    /* stencil buffer (Note wrapper!) */
    if (ctx->DrawBuffer->_StencilBuffer) {
-      if (map)
-	 intel_renderbuffer_map(intel,
-				ctx->DrawBuffer->_StencilBuffer->Wrapped);
-      else
-	 intel_renderbuffer_unmap(intel,
-				  ctx->DrawBuffer->_StencilBuffer->Wrapped);
+      irb = intel_renderbuffer(ctx->DrawBuffer->_StencilBuffer->Wrapped);
+      if (irb && irb->region) {
+         if (map) {
+            intel_region_map(intel, irb->region);
+            irb->pfMap = irb->region->map;
+            irb->pfPitch = irb->region->pitch;
+         }
+         else {
+            intel_region_unmap(intel, irb->region);
+            irb->pfMap = irb->region->map;
+            irb->pfPitch = irb->region->pitch;
+         }
+      }
    }
 }
 
@@ -654,9 +313,18 @@ intelSpanRenderStart(GLcontext * ctx)
    struct intel_context *intel = intel_context(ctx);
    GLuint i;
 
-   intelFlush(&intel->ctx);
+   intelFinish(&intel->ctx);
    LOCK_HARDWARE(intel);
 
+#if 0
+   /* Just map the framebuffer and all textures.  Bufmgr code will
+    * take care of waiting on the necessary fences:
+    */
+   intel_region_map(intel, intel->front_region);
+   intel_region_map(intel, intel->back_region);
+   intel_region_map(intel, intel->depth_region);
+#endif
+
    for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
       if (ctx->Texture.Unit[i]._ReallyEnabled) {
          struct gl_texture_object *texObj = ctx->Texture.Unit[i]._Current;
@@ -679,6 +347,14 @@ intelSpanRenderFinish(GLcontext * ctx)
 
    _swrast_flush(ctx);
 
+   /* Now unmap the framebuffer:
+    */
+#if 0
+   intel_region_unmap(intel, intel->front_region);
+   intel_region_unmap(intel, intel->back_region);
+   intel_region_unmap(intel, intel->depth_region);
+#endif
+
    for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
       if (ctx->Texture.Unit[i]._ReallyEnabled) {
          struct gl_texture_object *texObj = ctx->Texture.Unit[i]._Current;
@@ -705,108 +381,26 @@ intelInitSpanFuncs(GLcontext * ctx)
  * Plug in appropriate span read/write functions for the given renderbuffer.
  * These are used for the software fallbacks.
  */
-static void
-intel_set_span_functions(struct intel_context *intel,
-			 struct gl_renderbuffer *rb)
+void
+intel_set_span_functions(struct gl_renderbuffer *rb)
 {
-   struct intel_renderbuffer *irb = (struct intel_renderbuffer *) rb;
-   uint32_t tiling;
-
-   /* If in GEM mode, we need to do the tile address swizzling ourselves,
-    * instead of the fence registers handling it.
-    */
-   if (intel->ttm)
-      tiling = irb->region->tiling;
-   else
-      tiling = I915_TILING_NONE;
-
    if (rb->_ActualFormat == GL_RGB5) {
       /* 565 RGB */
-      switch (tiling) {
-      case I915_TILING_NONE:
-      default:
-	 intelInitPointers_RGB565(rb);
-	 break;
-      case I915_TILING_X:
-	 intel_XTile_InitPointers_RGB565(rb);
-	 break;
-      case I915_TILING_Y:
-	 intel_YTile_InitPointers_RGB565(rb);
-	 break;
-      }
-   }
-   else if (rb->_ActualFormat == GL_RGB8) {
-      /* 8888 RGBx */
-      switch (tiling) {
-      case I915_TILING_NONE:
-      default:
-	 intelInitPointers_xRGB8888(rb);
-	 break;
-      case I915_TILING_X:
-	 intel_XTile_InitPointers_xRGB8888(rb);
-	 break;
-      case I915_TILING_Y:
-	 intel_YTile_InitPointers_xRGB8888(rb);
-	 break;
-      }
+      intelInitPointers_RGB565(rb);
    }
    else if (rb->_ActualFormat == GL_RGBA8) {
       /* 8888 RGBA */
-      switch (tiling) {
-      case I915_TILING_NONE:
-      default:
-	 intelInitPointers_ARGB8888(rb);
-	 break;
-      case I915_TILING_X:
-	 intel_XTile_InitPointers_ARGB8888(rb);
-	 break;
-      case I915_TILING_Y:
-	 intel_YTile_InitPointers_ARGB8888(rb);
-	 break;
-      }
+      intelInitPointers_ARGB8888(rb);
    }
    else if (rb->_ActualFormat == GL_DEPTH_COMPONENT16) {
-      switch (tiling) {
-      case I915_TILING_NONE:
-      default:
-	 intelInitDepthPointers_z16(rb);
-	 break;
-      case I915_TILING_X:
-	 intel_XTile_InitDepthPointers_z16(rb);
-	 break;
-      case I915_TILING_Y:
-	 intel_YTile_InitDepthPointers_z16(rb);
-	 break;
-      }
+      intelInitDepthPointers_z16(rb);
    }
    else if (rb->_ActualFormat == GL_DEPTH_COMPONENT24 ||        /* XXX FBO remove */
             rb->_ActualFormat == GL_DEPTH24_STENCIL8_EXT) {
-      switch (tiling) {
-      case I915_TILING_NONE:
-      default:
-	 intelInitDepthPointers_z24_s8(rb);
-	 break;
-      case I915_TILING_X:
-	 intel_XTile_InitDepthPointers_z24_s8(rb);
-	 break;
-      case I915_TILING_Y:
-	 intel_YTile_InitDepthPointers_z24_s8(rb);
-	 break;
-      }
+      intelInitDepthPointers_z24_s8(rb);
    }
-   else if (rb->_ActualFormat == GL_STENCIL_INDEX8_EXT) {
-      switch (tiling) {
-      case I915_TILING_NONE:
-      default:
-	 intelInitStencilPointers_z24_s8(rb);
-	 break;
-      case I915_TILING_X:
-	 intel_XTile_InitStencilPointers_z24_s8(rb);
-	 break;
-      case I915_TILING_Y:
-	 intel_YTile_InitStencilPointers_z24_s8(rb);
-	 break;
-      }
+   else if (rb->_ActualFormat == GL_STENCIL_INDEX8_EXT) {       /* XXX FBO remove */
+      intelInitStencilPointers_z24_s8(rb);
    }
    else {
       _mesa_problem(NULL,
diff --git a/src/mesa/drivers/dri/intel/intel_span.h b/src/mesa/drivers/dri/intel/intel_span.h
index acbeb4abe1..5201f6d6c6 100644
--- a/src/mesa/drivers/dri/intel/intel_span.h
+++ b/src/mesa/drivers/dri/intel/intel_span.h
@@ -32,9 +32,7 @@ extern void intelInitSpanFuncs(GLcontext * ctx);
 
 extern void intelSpanRenderFinish(GLcontext * ctx);
 extern void intelSpanRenderStart(GLcontext * ctx);
-void intel_renderbuffer_map(struct intel_context *intel,
-			    struct gl_renderbuffer *rb);
-void intel_renderbuffer_unmap(struct intel_context *intel,
-			      struct gl_renderbuffer *rb);
+
+extern void intel_set_span_functions(struct gl_renderbuffer *rb);
 
 #endif
diff --git a/src/mesa/drivers/dri/intel/intel_tex_copy.c b/src/mesa/drivers/dri/intel/intel_tex_copy.c
index d0ab464a1c..1add7c6188 100644
--- a/src/mesa/drivers/dri/intel/intel_tex_copy.c
+++ b/src/mesa/drivers/dri/intel/intel_tex_copy.c
@@ -144,13 +144,15 @@ do_copy_texsubimage(struct intel_context *intel,
                            -src->pitch,
                            src->buffer,
                            src->height * src->pitch * src->cpp,
-			   src->tiling,
+			   GL_FALSE,
                            intelImage->mt->pitch,
                            intelImage->mt->region->buffer,
                            image_offset,
-			   intelImage->mt->region->tiling,
+			   intelImage->mt->region->tiled,
                            x, y + height, dstx, dsty, width, height,
 			   GL_COPY); /* ? */
+
+         intel_batchbuffer_flush(intel->batch);
       }
    }
 
diff --git a/src/mesa/drivers/dri/intel/intel_tex_image.c b/src/mesa/drivers/dri/intel/intel_tex_image.c
index c11687a2cf..f261034c18 100644
--- a/src/mesa/drivers/dri/intel/intel_tex_image.c
+++ b/src/mesa/drivers/dri/intel/intel_tex_image.c
@@ -238,6 +238,8 @@ try_pbo_upload(struct intel_context *intel,
                         dst_stride, dst_buffer, dst_offset, GL_FALSE,
                         0, 0, 0, 0, width, height,
 			GL_COPY);
+
+      intel_batchbuffer_flush(intel->batch);
    }
    UNLOCK_HARDWARE(intel);
 
@@ -398,26 +400,11 @@ intelTexImage(GLcontext * ctx,
 
       intel_miptree_reference(&intelImage->mt, intelObj->mt);
       assert(intelImage->mt);
-   } else if (intelImage->base.Border == 0) {
-      int comp_byte = 0;
-
-      if (intelImage->base.IsCompressed) {
-	 comp_byte =
-	    intel_compressed_num_bytes(intelImage->base.TexFormat->MesaFormat);
-      }
-
-      /* Didn't fit in the object miptree, but it's suitable for inclusion in
-       * a miptree, so create one just for our level and store it in the image.
-       * It'll get moved into the object miptree at validate time.
-       */
-      intelImage->mt = intel_miptree_create(intel, target, internalFormat,
-					    level, level,
-					    width, height, depth,
-					    intelImage->base.TexFormat->TexelBytes,
-					    comp_byte);
-
    }
 
+   if (!intelImage->mt)
+      DBG("XXX: Image did not fit into tree - storing in local memory!\n");
+
    /* PBO fastpaths:
     */
    if (dims <= 2 &&
diff --git a/src/mesa/drivers/dri/intel/intel_tex_validate.c b/src/mesa/drivers/dri/intel/intel_tex_validate.c
index 5763f4ae1f..1b3aa89c05 100644
--- a/src/mesa/drivers/dri/intel/intel_tex_validate.c
+++ b/src/mesa/drivers/dri/intel/intel_tex_validate.c
@@ -125,10 +125,13 @@ intel_finalize_mipmap_tree(struct intel_context *intel, GLuint unit)
    struct intel_texture_object *intelObj = intel_texture_object(tObj);
    int comp_byte = 0;
    int cpp;
+
    GLuint face, i;
    GLuint nr_faces = 0;
    struct intel_texture_image *firstImage;
 
+   GLboolean need_flush = GL_FALSE;
+
    /* We know/require this is true by now: 
     */
    assert(intelObj->base._Complete);
@@ -224,10 +227,21 @@ intel_finalize_mipmap_tree(struct intel_context *intel, GLuint unit)
           */
          if (intelObj->mt != intelImage->mt) {
             copy_image_data_to_tree(intel, intelObj, intelImage);
+	    need_flush = GL_TRUE;
          }
       }
    }
 
+#ifdef I915
+   /* XXX: what is this flush about?
+    * On 965, it causes a batch flush in the middle of the state relocation
+    * emits, which means that the eventual rendering doesn't have all of the
+    * required relocations in place.
+    */
+   if (need_flush)
+      intel_batchbuffer_flush(intel->batch);
+#endif
+
    return GL_TRUE;
 }