Merge remote branch 'origin/master' into modesetting-gem

Conflicts:

	libdrm/Makefile.am
	libdrm/dri_bufmgr.h
	linux-core/drm_irq.c
	linux-core/drm_sysfs.c
	linux-core/drm_ttm.c
	shared-core/i915_dma.c
	shared-core/i915_irq.c
	shared-core/nouveau_drv.h
	shared-core/radeon_cp.c

9 files changed, 625 insertions, 423 deletions

diff --git a/libdrm/Makefile.am b/libdrm/Makefile.am
index c6d66732..532ca138 100644
--- a/libdrm/Makefile.am
+++ b/libdrm/Makefile.am
@@ -26,10 +26,10 @@ libdrm_la_LDFLAGS = -version-number 2:3:0 -no-undefined
 
 AM_CFLAGS = -I$(top_srcdir)/shared-core
 libdrm_la_SOURCES = xf86drm.c xf86drmHash.c xf86drmRandom.c xf86drmSL.c \
-	xf86drmMode.c dri_bufmgr.c libdrm_lists.h
+	xf86drmMode.c
 libdrm_la_LIBADD = intel/libdrm_intel.la
 
 libdrmincludedir = ${includedir}
-libdrminclude_HEADERS = xf86drm.h xf86drmMode.h dri_bufmgr.h
+libdrminclude_HEADERS = xf86drm.h xf86drmMode.h
 
 EXTRA_DIST = ChangeLog TODO
diff --git a/libdrm/dri_bufmgr.h b/libdrm/dri_bufmgr.h
deleted file mode 100644
index 2005bdc6..00000000
--- a/libdrm/dri_bufmgr.h
+++ /dev/null
@@ -1,178 +0,0 @@
-/**************************************************************************
- * 
- * 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;
-
-struct _dri_bo {
-   /**
-    * Size in bytes of the buffer object.
-    *
-    * The size may be larger than the size originally requested for the
-    * allocation, such as being aligned to page size.
-    */
-   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;
-   /**
-    * MM-specific handle for accessing object
-    */
-   int handle;
-};
-
-/**
- * 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);
-
-   /** 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 execution on the
-    * buffer to complete, first.  The resulting mapping is available at
-    * buf->virtual.
-    */
-   int (*bo_map)(dri_bo *buf, int write_enable);
-
-   /** Reduces the refcount on the userspace mapping of the buffer object. */
-   int (*bo_unmap)(dri_bo *buf);
-
-   /**
-    * Write data into an object.
-    *
-    * This is an optional function, if missing,
-    * dri_bo will map/memcpy/unmap.
-    */
-   int (*bo_subdata) (dri_bo *buf, unsigned long offset,
-		      unsigned long size, const void *data);
-
-   /**
-    * Read data from an object
-    *
-    * This is an optional function, if missing,
-    * dri_bo will map/memcpy/unmap.
-    */
-   int (*bo_get_subdata) (dri_bo *bo, unsigned long offset,
-			  unsigned long size, void *data);
-
-   /**
-    * Waits for rendering to an object by the GPU to have completed.
-    *
-    * This is not required for any access to the BO by bo_map, bo_subdata, etc.
-    * It is merely a way for the driver to implement glFinish.
-    */
-   void (*bo_wait_rendering) (dri_bo *bo);
-
-   /**
-    * Tears down the buffer manager instance.
-    */
-   void (*destroy)(dri_bufmgr *bufmgr);
-
-   /**
-    * 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
-    * \return argument to be completed and passed to the execbuffers ioctl
-    *   (if any).
-    */
-   void *(*process_relocs)(dri_bo *batch_buf);
-
-   void (*post_submit)(dri_bo *batch_buf);
-
-   int (*check_aperture_space)(dri_bo **bo_array, int count);
-   int debug; /**< Enables verbose debugging printouts */
-};
-
-dri_bo *dri_bo_alloc(dri_bufmgr *bufmgr, const char *name, unsigned long size,
-		     unsigned int alignment);
-void dri_bo_reference(dri_bo *bo);
-void dri_bo_unreference(dri_bo *bo);
-int dri_bo_map(dri_bo *buf, int write_enable);
-int dri_bo_unmap(dri_bo *buf);
-
-int dri_bo_subdata(dri_bo *bo, unsigned long offset,
-		   unsigned long size, const void *data);
-int dri_bo_get_subdata(dri_bo *bo, unsigned long offset,
-		       unsigned long size, void *data);
-void dri_bo_wait_rendering(dri_bo *bo);
-
-void dri_bufmgr_set_debug(dri_bufmgr *bufmgr, int enable_debug);
-void dri_bufmgr_destroy(dri_bufmgr *bufmgr);
-
-void *dri_process_relocs(dri_bo *batch_buf);
-void dri_post_process_relocs(dri_bo *batch_buf);
-void dri_post_submit(dri_bo *batch_buf);
-int dri_bufmgr_check_aperture_space(dri_bo **bo_array, int count);
-
-#endif
diff --git a/libdrm/intel/Makefile.am b/libdrm/intel/Makefile.am
index 31a8512a..607c4765 100644
--- a/libdrm/intel/Makefile.am
+++ b/libdrm/intel/Makefile.am
@@ -25,11 +25,16 @@
 AM_CFLAGS = \
 	$(WARN_CFLAGS) \
 	-I$(top_srcdir)/libdrm \
+	-I$(top_srcdir)/libdrm/intel \
 	-I$(top_srcdir)/shared-core
 
-noinst_LTLIBRARIES = libdrm_intel.la
+libdrm_intel_la_LTLIBRARIES = libdrm_intel.la
+libdrm_intel_ladir = $(libdir)
+libdrm_intel_la_LDFLAGS = -version-number 1:0:0 -no-undefined
 
 libdrm_intel_la_SOURCES = \
+	intel_bufmgr.c \
+	intel_bufmgr_priv.h \
 	intel_bufmgr_fake.c \
 	intel_bufmgr_gem.c \
 	mm.c \
diff --git a/libdrm/dri_bufmgr.c b/libdrm/intel/intel_bufmgr.c
index a6eda3bd..fc7284b5 100644
--- a/libdrm/dri_bufmgr.c
+++ b/libdrm/intel/intel_bufmgr.c
@@ -25,10 +25,19 @@
  *
  */
 
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
 #include <string.h>
 #include <stdlib.h>
+#include <stdint.h>
 #include <assert.h>
-#include "dri_bufmgr.h"
+#include <errno.h>
+#include <drm.h>
+#include <i915_drm.h>
+#include "intel_bufmgr.h"
+#include "intel_bufmgr_priv.h"
 
 /** @file dri_bufmgr.c
  *
@@ -118,14 +127,12 @@ dri_bufmgr_destroy(dri_bufmgr *bufmgr)
    bufmgr->destroy(bufmgr);
 }
 
-void *dri_process_relocs(dri_bo *batch_buf)
-{
-   return batch_buf->bufmgr->process_relocs(batch_buf);
-}
-
-void dri_post_submit(dri_bo *batch_buf)
+int
+dri_bo_exec(dri_bo *bo, int used,
+	    drm_clip_rect_t *cliprects, int num_cliprects,
+	    int DR4)
 {
-   batch_buf->bufmgr->post_submit(batch_buf);
+   return bo->bufmgr->bo_exec(bo, used, cliprects, num_cliprects, DR4);
 }
 
 void
@@ -139,3 +146,49 @@ dri_bufmgr_check_aperture_space(dri_bo **bo_array, int count)
 {
 	return bo_array[0]->bufmgr->check_aperture_space(bo_array, count);
 }
+
+int
+dri_bo_flink(dri_bo *bo, uint32_t *name)
+{
+    if (bo->bufmgr->bo_flink)
+	return bo->bufmgr->bo_flink(bo, name);
+
+    return -ENODEV;
+}
+
+int
+dri_bo_emit_reloc(dri_bo *reloc_buf,
+		  uint32_t read_domains, uint32_t write_domain,
+		  uint32_t delta, uint32_t offset, dri_bo *target_buf)
+{
+    return reloc_buf->bufmgr->bo_emit_reloc(reloc_buf,
+					    read_domains, write_domain,
+					    delta, offset, target_buf);
+}
+
+int
+dri_bo_pin(dri_bo *bo, uint32_t alignment)
+{
+    if (bo->bufmgr->bo_pin)
+	return bo->bufmgr->bo_pin(bo, alignment);
+
+    return -ENODEV;
+}
+
+int
+dri_bo_unpin(dri_bo *bo)
+{
+    if (bo->bufmgr->bo_unpin)
+	return bo->bufmgr->bo_unpin(bo);
+
+    return -ENODEV;
+}
+
+int dri_bo_set_tiling(dri_bo *bo, uint32_t *tiling_mode)
+{
+    if (bo->bufmgr->bo_set_tiling)
+	return bo->bufmgr->bo_set_tiling(bo, tiling_mode);
+
+    *tiling_mode = I915_TILING_NONE;
+    return 0;
+}
diff --git a/libdrm/intel/intel_bufmgr.h b/libdrm/intel/intel_bufmgr.h
index 4d335210..c44d596b 100644
--- a/libdrm/intel/intel_bufmgr.h
+++ b/libdrm/intel/intel_bufmgr.h
@@ -31,65 +31,64 @@
  * Public definitions of Intel-specific bufmgr functions.
  */
 
-#ifndef INTEL_BUFMGR_GEM_H
-#define INTEL_BUFMGR_GEM_H
+#ifndef INTEL_BUFMGR_H
+#define INTEL_BUFMGR_H
 
-#include "dri_bufmgr.h"
+#include <stdint.h>
 
-/**
- * Intel-specific bufmgr bits that follow immediately after the
- * generic bufmgr structure.
- */
-struct intel_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 read_domains GEM read domains which the buffer will be read into
-     *	      by the command that this relocation is part of.
-     * \param write_domains GEM read domains which the buffer will be dirtied
-     *	      in by the command that this relocation is part of.
-     * \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,
-		      uint32_t read_domains, uint32_t write_domain,
-		      uint32_t delta, uint32_t offset, dri_bo *target);
+typedef struct _dri_bufmgr dri_bufmgr;
+typedef struct _dri_bo dri_bo;
+
+struct _dri_bo {
     /**
-     * Pin a buffer to the aperture and fix the offset until unpinned
+     * Size in bytes of the buffer object.
      *
-     * \param buf Buffer to pin
-     * \param alignment Required alignment for aperture, in bytes
+     * The size may be larger than the size originally requested for the
+     * allocation, such as being aligned to page size.
      */
-    int (*pin) (dri_bo *buf, uint32_t alignment);
+    unsigned long size;
     /**
-     * Unpin a buffer from the aperture, allowing it to be removed
-     *
-     * \param buf Buffer to unpin
+     * Card virtual address (offset from the beginning of the aperture) for the
+     * object.  Only valid while validated.
      */
-    int (*unpin) (dri_bo *buf);
+    unsigned long offset;
     /**
-     * Ask that the buffer be placed in tiling mode
-     *
-     * \param buf Buffer to set tiling mode for
-     * \param tiling_mode desired, and returned tiling mode
+     * Virtual address for accessing the buffer data.  Only valid while mapped.
      */
-    int (*set_tiling) (dri_bo *bo, uint32_t *tiling_mode);
-    /**
-     * Create a visible name for a buffer which can be used by other apps
-     *
-     * \param buf Buffer to create a name for
-     * \param name Returned name
-     */
-    int (*flink) (dri_bo *buf, uint32_t *name);
+    void *virtual;
+
+    /** Buffer manager context associated with this buffer object */
+    dri_bufmgr *bufmgr;
 };
 
+dri_bo *dri_bo_alloc(dri_bufmgr *bufmgr, const char *name, unsigned long size,
+		     unsigned int alignment);
+void dri_bo_reference(dri_bo *bo);
+void dri_bo_unreference(dri_bo *bo);
+int dri_bo_map(dri_bo *buf, int write_enable);
+int dri_bo_unmap(dri_bo *buf);
+
+int dri_bo_subdata(dri_bo *bo, unsigned long offset,
+		   unsigned long size, const void *data);
+int dri_bo_get_subdata(dri_bo *bo, unsigned long offset,
+		       unsigned long size, void *data);
+void dri_bo_wait_rendering(dri_bo *bo);
+
+void dri_bufmgr_set_debug(dri_bufmgr *bufmgr, int enable_debug);
+void dri_bufmgr_destroy(dri_bufmgr *bufmgr);
+int dri_bo_exec(dri_bo *bo, int used,
+		drm_clip_rect_t *cliprects, int num_cliprects,
+		int DR4);
+int dri_bufmgr_check_aperture_space(dri_bo **bo_array, int count);
+
+int dri_bo_emit_reloc(dri_bo *reloc_buf,
+		      uint32_t read_domains, uint32_t write_domain,
+		      uint32_t delta, uint32_t offset, dri_bo *target_buf);
+int dri_bo_pin(dri_bo *buf, uint32_t alignment);
+int dri_bo_unpin(dri_bo *buf);
+int dri_bo_set_tiling(dri_bo *buf, uint32_t *tiling_mode);
+int dri_bo_flink(dri_bo *buf, uint32_t *name);
+
 /* intel_bufmgr_gem.c */
 dri_bufmgr *intel_bufmgr_gem_init(int fd, int batch_size);
 dri_bo *intel_bo_gem_create_from_name(dri_bufmgr *bufmgr, const char *name,
@@ -97,34 +96,32 @@ dri_bo *intel_bo_gem_create_from_name(dri_bufmgr *bufmgr, const char *name,
 void intel_bufmgr_gem_enable_reuse(dri_bufmgr *bufmgr);
 
 /* intel_bufmgr_fake.c */
-dri_bufmgr *intel_bufmgr_fake_init(unsigned long low_offset, void *low_virtual,
+dri_bufmgr *intel_bufmgr_fake_init(int fd,
+				   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);
+				   volatile unsigned int *last_dispatch);
+void intel_bufmgr_fake_set_last_dispatch(dri_bufmgr *bufmgr,
+					 volatile unsigned int *last_dispatch);
+void intel_bufmgr_fake_set_exec_callback(dri_bufmgr *bufmgr,
+					 int (*exec)(dri_bo *bo,
+						     unsigned int used,
+						     void *priv),
+					 void *priv);
+void intel_bufmgr_fake_set_fence_callback(dri_bufmgr *bufmgr,
+					  unsigned int (*emit)(void *priv),
+					  void (*wait)(unsigned int fence,
+						       void *priv),
+					  void *priv);
 dri_bo *intel_bo_fake_alloc_static(dri_bufmgr *bufmgr, const char *name,
 				   unsigned long offset, unsigned long size,
 				   void *virtual);
-
-void intel_bufmgr_fake_contended_lock_take(dri_bufmgr *bufmgr);
 void intel_bo_fake_disable_backing_store(dri_bo *bo,
 					 void (*invalidate_cb)(dri_bo *bo,
 							       void *ptr),
 					 void *ptr);
-void intel_bufmgr_fake_evict_all(dri_bufmgr *bufmgr);
-
-int intel_bo_emit_reloc(dri_bo *reloc_buf,
-			uint32_t read_domains, uint32_t write_domain,
-			uint32_t delta, uint32_t offset, dri_bo *target_buf);
-
-int intel_bo_pin(dri_bo *buf, uint32_t alignment);
-
-int intel_bo_unpin(dri_bo *buf);
 
-int intel_bo_set_tiling(dri_bo *buf, uint32_t *tiling_mode);
-
-int intel_bo_flink(dri_bo *buf, uint32_t *name);
+void intel_bufmgr_fake_contended_lock_take(dri_bufmgr *bufmgr);
+void intel_bufmgr_fake_evict_all(dri_bufmgr *bufmgr);
 
-#endif /* INTEL_BUFMGR_GEM_H */
+#endif /* INTEL_BUFMGR_H */
 
diff --git a/libdrm/intel/intel_bufmgr_fake.c b/libdrm/intel/intel_bufmgr_fake.c
index 15da398f..c88e1cb0 100644
--- a/libdrm/intel/intel_bufmgr_fake.c
+++ b/libdrm/intel/intel_bufmgr_fake.c
@@ -34,11 +34,17 @@
  * the bugs in the old texture manager.
  */
 
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
 #include <stdlib.h>
 #include <string.h>
 #include <assert.h>
-#include "dri_bufmgr.h"
+#include <errno.h>
+#include <xf86drm.h>
 #include "intel_bufmgr.h"
+#include "intel_bufmgr_priv.h"
 #include "drm.h"
 #include "i915_drm.h"
 #include "mm.h"
@@ -105,7 +111,6 @@ struct block {
 
 typedef struct _bufmgr_fake {
    dri_bufmgr bufmgr;
-   struct intel_bufmgr intel_bufmgr;
 
    unsigned long low_offset;
    unsigned long size;
@@ -138,14 +143,32 @@ typedef struct _bufmgr_fake {
    /**
     * Driver callback to emit a fence, returning the cookie.
     *
+    * This allows the driver to hook in a replacement for the DRM usage in
+    * bufmgr_fake.
+    *
     * 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);
+   void (*fence_wait)(unsigned int fence, void *private);
+   void *fence_priv;
+
+   /**
+    * Driver callback to execute a buffer.
+    *
+    * This allows the driver to hook in a replacement for the DRM usage in
+    * bufmgr_fake.
+    */
+   int (*exec)(dri_bo *bo, unsigned int used, void *priv);
+   void *exec_priv;
+
    /** Driver-supplied argument to driver callbacks */
    void *driver_priv;
+   /* Pointer to kernel-updated sarea data for the last completed user irq */
+   volatile int *last_dispatch;
+
+   int fd;
 
    int debug;
 
@@ -211,24 +234,161 @@ static int FENCE_LTE( unsigned a, unsigned b )
    return 0;
 }
 
+void intel_bufmgr_fake_set_fence_callback(dri_bufmgr *bufmgr,
+					  unsigned int (*emit)(void *priv),
+					  void (*wait)(unsigned int fence,
+						       void *priv),
+					  void *priv)
+{
+   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)bufmgr;
+
+   bufmgr_fake->fence_emit = emit;
+   bufmgr_fake->fence_wait = wait;
+   bufmgr_fake->fence_priv = priv;
+}
+
 static unsigned int
 _fence_emit_internal(dri_bufmgr_fake *bufmgr_fake)
 {
-   bufmgr_fake->last_fence = bufmgr_fake->fence_emit(bufmgr_fake->driver_priv);
+   struct drm_i915_irq_emit ie;
+   int ret, seq = 1;
+
+   if (bufmgr_fake->fence_emit != NULL)
+      return bufmgr_fake->fence_emit(bufmgr_fake->fence_priv);
+
+   ie.irq_seq = &seq;
+   ret = drmCommandWriteRead(bufmgr_fake->fd, DRM_I915_IRQ_EMIT,
+			     &ie, sizeof(ie));
+   if (ret) {
+      drmMsg("%s: drm_i915_irq_emit: %d\n", __FUNCTION__, ret);
+      abort();
+   }
+
+   DBG("emit 0x%08x\n", seq);
+   bufmgr_fake->last_fence = seq;
    return bufmgr_fake->last_fence;
 }
 
 static void
-_fence_wait_internal(dri_bufmgr_fake *bufmgr_fake, unsigned int cookie)
+_fence_wait_internal(dri_bufmgr_fake *bufmgr_fake, int seq)
 {
+   struct drm_i915_irq_wait iw;
+   int hw_seq, busy_count = 0;
    int ret;
+   int kernel_lied;
+
+   if (bufmgr_fake->fence_wait != NULL) {
+      bufmgr_fake->fence_wait(seq, bufmgr_fake->fence_priv);
+      return;
+   }
+
+   DBG("wait 0x%08x\n", iw.irq_seq);
+
+   iw.irq_seq = seq;
+
+   /* The kernel IRQ_WAIT implementation is all sorts of broken.
+    * 1) It returns 1 to 0x7fffffff instead of using the full 32-bit unsigned
+    *    range.
+    * 2) It returns 0 if hw_seq >= seq, not seq - hw_seq < 0 on the 32-bit
+    *    signed range.
+    * 3) It waits if seq < hw_seq, not seq - hw_seq > 0 on the 32-bit
+    *    signed range.
+    * 4) It returns -EBUSY in 3 seconds even if the hardware is still
+    *    successfully chewing through buffers.
+    *
+    * Assume that in userland we treat sequence numbers as ints, which makes
+    * some of the comparisons convenient, since the sequence numbers are
+    * all postive signed integers.
+    *
+    * From this we get several cases we need to handle.  Here's a timeline.
+    * 0x2   0x7                                         0x7ffffff8   0x7ffffffd
+    *   |    |                                                   |    |
+    * -------------------------------------------------------------------
+    *
+    * A) Normal wait for hw to catch up
+    * hw_seq seq
+    *   |    |
+    * -------------------------------------------------------------------
+    * seq - hw_seq = 5.  If we call IRQ_WAIT, it will wait for hw to catch up.
+    *
+    * B) Normal wait for a sequence number that's already passed.
+    * seq    hw_seq
+    *   |    |
+    * -------------------------------------------------------------------
+    * seq - hw_seq = -5.  If we call IRQ_WAIT, it returns 0 quickly.
+    *
+    * C) Hardware has already wrapped around ahead of us
+    * hw_seq                                                         seq
+    *   |                                                             |
+    * -------------------------------------------------------------------
+    * seq - hw_seq = 0x80000000 - 5.  If we called IRQ_WAIT, it would wait
+    * for hw_seq >= seq, which may never occur.  Thus, we want to catch this
+    * in userland and return 0.
+    *
+    * D) We've wrapped around ahead of the hardware.
+    * seq                                                           hw_seq
+    *   |                                                             |
+    * -------------------------------------------------------------------
+    * seq - hw_seq = -(0x80000000 - 5).  If we called IRQ_WAIT, it would return
+    * 0 quickly because hw_seq >= seq, even though the hardware isn't caught up.
+    * Thus, we need to catch this early return in userland and bother the
+    * kernel until the hardware really does catch up.
+    *
+    * E) Hardware might wrap after we test in userland.
+    *                                                         hw_seq  seq
+    *                                                            |    |
+    * -------------------------------------------------------------------
+    * seq - hw_seq = 5.  If we call IRQ_WAIT, it will likely see seq >= hw_seq
+    * and wait.  However, suppose hw_seq wraps before we make it into the
+    * kernel.  The kernel sees hw_seq >= seq and waits for 3 seconds then
+    * returns -EBUSY.  This is case C).  We should catch this and then return
+    * successfully.
+    *
+    * F) Hardware might take a long time on a buffer.
+    * hw_seq seq
+    *   |    |
+    * -------------------------------------------------------------------
+    * seq - hw_seq = 5.  If we call IRQ_WAIT, if sequence 2 through 5 take too
+    * long, it will return -EBUSY.  Batchbuffers in the gltestperf demo were
+    * seen to take up to 7 seconds.  We should catch early -EBUSY return
+    * and keep trying.
+    */
+
+   do {
+      /* Keep a copy of last_dispatch so that if the wait -EBUSYs because the
+       * hardware didn't catch up in 3 seconds, we can see if it at least made
+       * progress and retry.
+       */
+      hw_seq = *bufmgr_fake->last_dispatch;
+
+      /* Catch case C */
+      if (seq - hw_seq > 0x40000000)
+	 return;
+
+      ret = drmCommandWrite(bufmgr_fake->fd, DRM_I915_IRQ_WAIT,
+			    &iw, sizeof(iw));
+      /* Catch case D */
+      kernel_lied = (ret == 0) && (seq - *bufmgr_fake->last_dispatch <
+				   -0x40000000);
+
+      /* Catch case E */
+      if (ret == -EBUSY && (seq - *bufmgr_fake->last_dispatch > 0x40000000))
+	 ret = 0;
+
+      /* Catch case F: Allow up to 15 seconds chewing on one buffer. */
+      if ((ret == -EBUSY) && (hw_seq != *bufmgr_fake->last_dispatch))
+	 busy_count = 0;
+      else
+	 busy_count++;
+   } while (kernel_lied || ret == -EAGAIN || ret == -EINTR ||
+	    (ret == -EBUSY && busy_count < 5));
 
-   ret = bufmgr_fake->fence_wait(bufmgr_fake->driver_priv, cookie);
    if (ret != 0) {
-      drmMsg("%s:%d: Error %d waiting for fence.\n", __FILE__, __LINE__);
+      drmMsg("%s:%d: Error waiting for fence: %s.\n", __FILE__, __LINE__,
+	     strerror(-ret));
       abort();
    }
-   clear_fenced(bufmgr_fake, cookie);
+   clear_fenced(bufmgr_fake, seq);
 }
 
 static int
@@ -540,7 +700,7 @@ dri_bufmgr_fake_wait_idle(dri_bufmgr_fake *bufmgr_fake)
 {
    unsigned int cookie;
 
-   cookie = bufmgr_fake->fence_emit(bufmgr_fake->driver_priv);
+   cookie = _fence_emit_internal(bufmgr_fake);
    _fence_wait_internal(bufmgr_fake, cookie);
 }
 
@@ -1052,38 +1212,6 @@ dri_fake_reloc_and_validate_buffer(dri_bo *bo)
    return dri_fake_bo_validate(bo);
 }
 
-static void *
-dri_fake_process_relocs(dri_bo *batch_buf)
-{
-   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 = 0;
-
-   dri_fake_calculate_domains(batch_buf);
-
-   batch_fake->read_domains = I915_GEM_DOMAIN_COMMAND;
-
-   /* 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);
-
-   return NULL;
-}
-
 static void
 dri_bo_fake_post_submit(dri_bo *bo)
 {
@@ -1110,12 +1238,74 @@ dri_bo_fake_post_submit(dri_bo *bo)
 }
 
 
-static void
-dri_fake_post_submit(dri_bo *batch_buf)
+void intel_bufmgr_fake_set_exec_callback(dri_bufmgr *bufmgr,
+					 int (*exec)(dri_bo *bo,
+						     unsigned int used,
+						     void *priv),
+					 void *priv)
+{
+   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)bufmgr;
+
+   bufmgr_fake->exec = exec;
+   bufmgr_fake->exec_priv = priv;
+}
+
+static int
+dri_fake_bo_exec(dri_bo *bo, int used,
+		 drm_clip_rect_t *cliprects, int num_cliprects,
+		 int DR4)
 {
-   dri_fake_fence_validated(batch_buf->bufmgr);
+   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)bo->bufmgr;
+   dri_bo_fake *batch_fake = (dri_bo_fake *)bo;
+   struct drm_i915_batchbuffer batch;
+   int ret;
+   int retry_count = 0;
+
+   bufmgr_fake->performed_rendering = 0;
+
+   dri_fake_calculate_domains(bo);
 
-   dri_bo_fake_post_submit(batch_buf);
+   batch_fake->read_domains = I915_GEM_DOMAIN_COMMAND;
+
+   /* we've ran out of RAM so blow the whole lot away and retry */
+ restart:
+   ret = dri_fake_reloc_and_validate_buffer(bo);
+   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);
+
+   if (bufmgr_fake->exec != NULL) {
+      int ret = bufmgr_fake->exec(bo, used, bufmgr_fake->exec_priv);
+      if (ret != 0)
+	 return ret;
+   } else {
+      batch.start = bo->offset;
+      batch.used = used;
+      batch.cliprects = cliprects;
+      batch.num_cliprects = num_cliprects;
+      batch.DR1 = 0;
+      batch.DR4 = DR4;
+
+      if (drmCommandWrite(bufmgr_fake->fd, DRM_I915_BATCHBUFFER, &batch,
+			  sizeof(batch))) {
+	 drmMsg("DRM_I915_BATCHBUFFER: %d\n", -errno);
+	 return -errno;
+      }
+   }
+
+   dri_fake_fence_validated(bo->bufmgr);
+
+   dri_bo_fake_post_submit(bo);
+
+   return 0;
 }
 
 /**
@@ -1187,13 +1377,19 @@ intel_bufmgr_fake_evict_all(dri_bufmgr *bufmgr)
       free_block(bufmgr_fake, block);
    }
 }
+void intel_bufmgr_fake_set_last_dispatch(dri_bufmgr *bufmgr,
+					 volatile unsigned int *last_dispatch)
+{
+   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)bufmgr;
+
+   bufmgr_fake->last_dispatch = (volatile int *)last_dispatch;
+}
 
 dri_bufmgr *
-intel_bufmgr_fake_init(unsigned long low_offset, void *low_virtual,
+intel_bufmgr_fake_init(int fd,
+		       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)
+		       volatile unsigned int *last_dispatch)
 {
    dri_bufmgr_fake *bufmgr_fake;
 
@@ -1216,16 +1412,14 @@ intel_bufmgr_fake_init(unsigned long low_offset, void *low_virtual,
    bufmgr_fake->bufmgr.bo_map = dri_fake_bo_map;
    bufmgr_fake->bufmgr.bo_unmap = dri_fake_bo_unmap;
    bufmgr_fake->bufmgr.bo_wait_rendering = dri_fake_bo_wait_rendering;
+   bufmgr_fake->bufmgr.bo_emit_reloc = dri_fake_emit_reloc;
    bufmgr_fake->bufmgr.destroy = dri_fake_destroy;
-   bufmgr_fake->bufmgr.process_relocs = dri_fake_process_relocs;
-   bufmgr_fake->bufmgr.post_submit = dri_fake_post_submit;
+   bufmgr_fake->bufmgr.bo_exec = dri_fake_bo_exec;
    bufmgr_fake->bufmgr.check_aperture_space = dri_fake_check_aperture_space;
    bufmgr_fake->bufmgr.debug = 0;
-   bufmgr_fake->intel_bufmgr.emit_reloc = dri_fake_emit_reloc;
 
-   bufmgr_fake->fence_emit = fence_emit;
-   bufmgr_fake->fence_wait = fence_wait;
-   bufmgr_fake->driver_priv = driver_priv;
+   bufmgr_fake->fd = fd;
+   bufmgr_fake->last_dispatch = (volatile int *)last_dispatch;
 
    return &bufmgr_fake->bufmgr;
 }
diff --git a/libdrm/intel/intel_bufmgr_gem.c b/libdrm/intel/intel_bufmgr_gem.c
index af20efb2..4ca49d0a 100644
--- a/libdrm/intel/intel_bufmgr_gem.c
+++ b/libdrm/intel/intel_bufmgr_gem.c
@@ -34,6 +34,10 @@
  *	    Dave Airlie <airlied@linux.ie>
  */
 
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
 #include <xf86drm.h>
 #include <stdio.h>
 #include <stdlib.h>
@@ -44,8 +48,8 @@
 #include <sys/mman.h>
 
 #include "errno.h"
-#include "dri_bufmgr.h"
 #include "intel_bufmgr.h"
+#include "intel_bufmgr_priv.h"
 #include "string.h"
 
 #include "i915_drm.h"
@@ -76,8 +80,6 @@ struct dri_gem_bo_bucket {
 typedef struct _dri_bufmgr_gem {
     dri_bufmgr bufmgr;
 
-    struct intel_bufmgr intel_bufmgr;
-
     int fd;
 
     int max_relocs;
@@ -89,8 +91,6 @@ typedef struct _dri_bufmgr_gem {
 
     /** Array of lists of cached gem objects of power-of-two sizes */
     struct dri_gem_bo_bucket cache_bucket[INTEL_GEM_BO_BUCKETS];
-
-    struct drm_i915_gem_execbuffer exec_arg;
 } dri_bufmgr_gem;
 
 struct _dri_bo_gem {
@@ -316,7 +316,7 @@ dri_gem_bo_alloc(dri_bufmgr *bufmgr, const char *name,
 
 	ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GEM_CREATE, &create);
 	bo_gem->gem_handle = create.handle;
-	bo_gem->bo.handle = bo_gem->gem_handle;
+	//bo_gem->bo.handle = bo_gem->gem_handle;
 	if (ret != 0) {
 	    free(bo_gem);
 	    return NULL;
@@ -358,7 +358,7 @@ intel_bo_gem_create_from_name(dri_bufmgr *bufmgr, const char *name,
     ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_GEM_OPEN, &open_arg);
     if (ret != 0) {
 	fprintf(stderr, "Couldn't reference %s handle 0x%08x: %s\n",
-	       name, handle, strerror(-ret));
+	       name, handle, strerror(errno));
 	free(bo_gem);
 	return NULL;
     }
@@ -370,6 +370,7 @@ intel_bo_gem_create_from_name(dri_bufmgr *bufmgr, const char *name,
     bo_gem->refcount = 1;
     bo_gem->validate_index = -1;
     bo_gem->gem_handle = open_arg.handle;
+    bo_gem->global_name = handle;
 
     DBG("bo_create_from_handle: %d (%s)\n", handle, bo_gem->name);
 
@@ -401,7 +402,7 @@ dri_gem_bo_free(dri_bo *bo)
     if (ret != 0) {
 	fprintf(stderr,
 		"DRM_IOCTL_GEM_CLOSE %d failed (%s): %s\n",
-		bo_gem->gem_handle, bo_gem->name, strerror(-ret));
+		bo_gem->gem_handle, bo_gem->name, strerror(errno));
     }
     free(bo);
 }
@@ -650,8 +651,8 @@ dri_bufmgr_gem_destroy(dri_bufmgr *bufmgr)
  * last known offset in target_bo.
  */
 static int
-dri_gem_emit_reloc(dri_bo *bo, uint32_t read_domains, uint32_t write_domain,
-		   uint32_t delta, uint32_t offset, dri_bo *target_bo)
+dri_gem_bo_emit_reloc(dri_bo *bo, uint32_t read_domains, uint32_t write_domain,
+		      uint32_t delta, uint32_t offset, dri_bo *target_bo)
 {
     dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr;
     dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
@@ -708,27 +709,6 @@ dri_gem_bo_process_reloc(dri_bo *bo)
     }
 }
 
-static void *
-dri_gem_process_reloc(dri_bo *batch_buf)
-{
-    dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *) batch_buf->bufmgr;
-
-    /* Update indices and set up the validate list. */
-    dri_gem_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);
-
-    bufmgr_gem->exec_arg.buffers_ptr = (uintptr_t)bufmgr_gem->exec_objects;
-    bufmgr_gem->exec_arg.buffer_count = bufmgr_gem->exec_count;
-    bufmgr_gem->exec_arg.batch_start_offset = 0;
-    bufmgr_gem->exec_arg.batch_len = 0;	/* written in intel_exec_ioctl */
-
-    return &bufmgr_gem->exec_arg;
-}
-
 static void
 intel_update_buffer_offsets (dri_bufmgr_gem *bufmgr_gem)
 {
@@ -748,11 +728,35 @@ intel_update_buffer_offsets (dri_bufmgr_gem *bufmgr_gem)
     }
 }
 
-static void
-dri_gem_post_submit(dri_bo *batch_buf)
+static int
+dri_gem_bo_exec(dri_bo *bo, int used,
+		drm_clip_rect_t *cliprects, int num_cliprects,
+		int DR4)
 {
-    dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)batch_buf->bufmgr;
-    int i;
+    dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr;
+    struct drm_i915_gem_execbuffer execbuf;
+    int ret, i;
+
+    /* Update indices and set up the validate list. */
+    dri_gem_bo_process_reloc(bo);
+
+    /* Add the batch buffer to the validation list.  There are no relocations
+     * pointing to it.
+     */
+    intel_add_validate_buffer(bo);
+
+    execbuf.buffers_ptr = (uintptr_t)bufmgr_gem->exec_objects;
+    execbuf.buffer_count = bufmgr_gem->exec_count;
+    execbuf.batch_start_offset = 0;
+    execbuf.batch_len = used;
+    execbuf.cliprects_ptr = (uintptr_t)cliprects;
+    execbuf.num_cliprects = num_cliprects;
+    execbuf.DR1 = 0;
+    execbuf.DR4 = DR4;
+
+    do {
+	ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GEM_EXECBUFFER, &execbuf);
+    } while (ret == -EAGAIN);
 
     intel_update_buffer_offsets (bufmgr_gem);
 
@@ -772,10 +776,12 @@ dri_gem_post_submit(dri_bo *batch_buf)
 	bufmgr_gem->exec_bos[i] = NULL;
     }
     bufmgr_gem->exec_count = 0;
+
+    return 0;
 }
 
 static int
-dri_gem_pin(dri_bo *bo, uint32_t alignment)
+dri_gem_bo_pin(dri_bo *bo, uint32_t alignment)
 {
     dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr;
     dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
@@ -794,7 +800,7 @@ dri_gem_pin(dri_bo *bo, uint32_t alignment)
 }
 
 static int
-dri_gem_unpin(dri_bo *bo)
+dri_gem_bo_unpin(dri_bo *bo)
 {
     dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr;
     dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
@@ -811,7 +817,7 @@ dri_gem_unpin(dri_bo *bo)
 }
 
 static int
-dri_gem_set_tiling(dri_bo *bo, uint32_t *tiling_mode)
+dri_gem_bo_set_tiling(dri_bo *bo, uint32_t *tiling_mode)
 {
     dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr;
     dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
@@ -832,7 +838,7 @@ dri_gem_set_tiling(dri_bo *bo, uint32_t *tiling_mode)
 }
 
 static int
-dri_gem_flink(dri_bo *bo, uint32_t *name)
+dri_gem_bo_flink(dri_bo *bo, uint32_t *name)
 {
     dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr;
     dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
@@ -910,16 +916,15 @@ intel_bufmgr_gem_init(int fd, int batch_size)
     bufmgr_gem->bufmgr.bo_subdata = dri_gem_bo_subdata;
     bufmgr_gem->bufmgr.bo_get_subdata = dri_gem_bo_get_subdata;
     bufmgr_gem->bufmgr.bo_wait_rendering = dri_gem_bo_wait_rendering;
+    bufmgr_gem->bufmgr.bo_emit_reloc = dri_gem_bo_emit_reloc;
+    bufmgr_gem->bufmgr.bo_pin = dri_gem_bo_pin;
+    bufmgr_gem->bufmgr.bo_unpin = dri_gem_bo_unpin;
+    bufmgr_gem->bufmgr.bo_set_tiling = dri_gem_bo_set_tiling;
+    bufmgr_gem->bufmgr.bo_flink = dri_gem_bo_flink;
+    bufmgr_gem->bufmgr.bo_exec = dri_gem_bo_exec;
     bufmgr_gem->bufmgr.destroy = dri_bufmgr_gem_destroy;
-    bufmgr_gem->bufmgr.process_relocs = dri_gem_process_reloc;
-    bufmgr_gem->bufmgr.post_submit = dri_gem_post_submit;
     bufmgr_gem->bufmgr.debug = 0;
     bufmgr_gem->bufmgr.check_aperture_space = dri_gem_check_aperture_space;
-    bufmgr_gem->intel_bufmgr.emit_reloc = dri_gem_emit_reloc;
-    bufmgr_gem->intel_bufmgr.pin = dri_gem_pin;
-    bufmgr_gem->intel_bufmgr.unpin = dri_gem_unpin;
-    bufmgr_gem->intel_bufmgr.set_tiling = dri_gem_set_tiling;
-    bufmgr_gem->intel_bufmgr.flink = dri_gem_flink;
     /* Initialize the linked lists for BO reuse cache. */
     for (i = 0; i < INTEL_GEM_BO_BUCKETS; i++)
 	bufmgr_gem->cache_bucket[i].tail = &bufmgr_gem->cache_bucket[i].head;
@@ -927,67 +932,3 @@ intel_bufmgr_gem_init(int fd, int batch_size)
     return &bufmgr_gem->bufmgr;
 }
 
-int
-intel_bo_emit_reloc(dri_bo *reloc_buf,
-		    uint32_t read_domains, uint32_t write_domain,
-		    uint32_t delta, uint32_t offset, dri_bo *target_buf)
-{
-    struct intel_bufmgr *intel_bufmgr;
-
-    intel_bufmgr = (struct intel_bufmgr *)(reloc_buf->bufmgr + 1);
-
-    return intel_bufmgr->emit_reloc(reloc_buf, read_domains, write_domain,
-				    delta, offset, target_buf);
-}
-
-int
-intel_bo_pin(dri_bo *bo, uint32_t alignment)
-{
-    struct intel_bufmgr *intel_bufmgr;
-
-    intel_bufmgr = (struct intel_bufmgr *)(bo->bufmgr + 1);
-
-    if (intel_bufmgr->pin)
-	return intel_bufmgr->pin(bo, alignment);
-
-    return 0;
-}
-
-int
-intel_bo_unpin(dri_bo *bo)
-{
-    struct intel_bufmgr *intel_bufmgr;
-
-    intel_bufmgr = (struct intel_bufmgr *)(bo->bufmgr + 1);
-
-    if (intel_bufmgr->unpin)
-	return intel_bufmgr->unpin(bo);
-
-    return 0;
-}
-
-int intel_bo_set_tiling(dri_bo *bo, uint32_t *tiling_mode)
-{
-    struct intel_bufmgr *intel_bufmgr;
-
-    intel_bufmgr = (struct intel_bufmgr *)(bo->bufmgr + 1);
-
-    if (intel_bufmgr->set_tiling)
-	return intel_bufmgr->set_tiling (bo, tiling_mode);
-
-    *tiling_mode = I915_TILING_NONE;
-    return 0;
-}
-
-int intel_bo_flink(dri_bo *bo, uint32_t *name)
-{
-    struct intel_bufmgr *intel_bufmgr;
-
-    intel_bufmgr = (struct intel_bufmgr *)(bo->bufmgr + 1);
-
-    if (intel_bufmgr->flink)
-	return intel_bufmgr->flink (bo, name);
-
-    return -ENODEV;
-}
-
diff --git a/libdrm/intel/intel_bufmgr_priv.h b/libdrm/intel/intel_bufmgr_priv.h
new file mode 100644
index 00000000..7f39bfc2
--- /dev/null
+++ b/libdrm/intel/intel_bufmgr_priv.h
@@ -0,0 +1,165 @@
+/*
+ * Copyright © 2008 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>
+ *
+ */
+
+/**
+ * @file intel_bufmgr_priv.h
+ *
+ * Private definitions of Intel-specific bufmgr functions and structures.
+ */
+
+#ifndef INTEL_BUFMGR_PRIV_H
+#define INTEL_BUFMGR_PRIV_H
+
+/**
+ * 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);
+
+   /** 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 execution on the
+    * buffer to complete, first.  The resulting mapping is available at
+    * buf->virtual.
+    */
+   int (*bo_map)(dri_bo *buf, int write_enable);
+
+   /** Reduces the refcount on the userspace mapping of the buffer object. */
+   int (*bo_unmap)(dri_bo *buf);
+
+   /**
+    * Write data into an object.
+    *
+    * This is an optional function, if missing,
+    * dri_bo will map/memcpy/unmap.
+    */
+   int (*bo_subdata) (dri_bo *buf, unsigned long offset,
+		      unsigned long size, const void *data);
+
+   /**
+    * Read data from an object
+    *
+    * This is an optional function, if missing,
+    * dri_bo will map/memcpy/unmap.
+    */
+   int (*bo_get_subdata) (dri_bo *bo, unsigned long offset,
+			  unsigned long size, void *data);
+
+   /**
+    * Waits for rendering to an object by the GPU to have completed.
+    *
+    * This is not required for any access to the BO by bo_map, bo_subdata, etc.
+    * It is merely a way for the driver to implement glFinish.
+    */
+   void (*bo_wait_rendering) (dri_bo *bo);
+
+   /**
+    * 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 read_domains GEM read domains which the buffer will be read into
+     *	      by the command that this relocation is part of.
+     * \param write_domains GEM read domains which the buffer will be dirtied
+     *	      in by the command that this relocation is part of.
+     * \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 (*bo_emit_reloc)(dri_bo *reloc_buf,
+			 uint32_t read_domains, uint32_t write_domain,
+			 uint32_t delta, uint32_t offset, dri_bo *target);
+
+    /** Executes the command buffer pointed to by bo. */
+    int (*bo_exec)(dri_bo *bo, int used,
+		   drm_clip_rect_t *cliprects, int num_cliprects,
+		   int DR4);
+
+    /**
+     * Pin a buffer to the aperture and fix the offset until unpinned
+     *
+     * \param buf Buffer to pin
+     * \param alignment Required alignment for aperture, in bytes
+     */
+    int (*bo_pin) (dri_bo *buf, uint32_t alignment);
+    /**
+     * Unpin a buffer from the aperture, allowing it to be removed
+     *
+     * \param buf Buffer to unpin
+     */
+    int (*bo_unpin) (dri_bo *buf);
+    /**
+     * Ask that the buffer be placed in tiling mode
+     *
+     * \param buf Buffer to set tiling mode for
+     * \param tiling_mode desired, and returned tiling mode
+     */
+    int (*bo_set_tiling) (dri_bo *bo, uint32_t *tiling_mode);
+    /**
+     * Create a visible name for a buffer which can be used by other apps
+     *
+     * \param buf Buffer to create a name for
+     * \param name Returned name
+     */
+    int (*bo_flink) (dri_bo *buf, uint32_t *name);
+
+    int (*check_aperture_space)(dri_bo **bo_array, int count);
+    int debug; /**< Enables verbose debugging printouts */
+};
+
+#endif /* INTEL_BUFMGR_PRIV_H */
+
diff --git a/libdrm/xf86drm.c b/libdrm/xf86drm.c
index 64765339..94f5dfad 100644
--- a/libdrm/xf86drm.c
+++ b/libdrm/xf86drm.c
@@ -178,7 +178,7 @@ static char *drmStrdup(const char *s)
  * Call ioctl, restarting if it is interupted
  */
 static int
-drmIoctl(int fd, int request, void *arg)
+drmIoctl(int fd, unsigned long request, void *arg)
 {
     int	ret;
 
@@ -300,6 +300,7 @@ static int drmOpenDevice(long dev, int minor, int type)
 	group = (serv_group >= 0) ? serv_group : DRM_DEV_GID;
     }
 
+#if !defined(UDEV)
     if (stat(DRM_DIR_NAME, &st)) {
 	if (!isroot)
 	    return DRM_ERR_NOT_ROOT;
@@ -320,6 +321,30 @@ static int drmOpenDevice(long dev, int minor, int type)
 	chown(buf, user, group);
 	chmod(buf, devmode);
     }
+#else
+    /* if we modprobed then wait for udev */
+    {
+	int udev_count = 0;
+wait_for_udev:
+        if (stat(DRM_DIR_NAME, &st)) {
+		usleep(20);
+		udev_count++;
+
+		if (udev_count == 50)
+			return -1;
+		goto wait_for_udev;
+	}
+
+    	if (stat(buf, &st)) {
+		usleep(20);
+		udev_count++;
+
+		if (udev_count == 50)
+			return -1;
+		goto wait_for_udev;
+    	}
+    }
+#endif
 
     fd = open(buf, O_RDWR, 0);
     drmMsg("drmOpenDevice: open result is %d, (%s)\n",