On 03/07/2012 06:51 PM, Yuanhan Liu wrote:
On Wed, Mar 07, 2012 at 11:21:07AM -0800, Eric Anholt wrote:
From: Kenneth Graunke<kenn...@whitecape.org>

This will allow the driver to capture all of its execution state to a
file for later debugging.  intel_gpu_dump is limited in that it only
captures batchbuffers, and Mesa's captures, while more complete, still
capture only a portion of the state involved in execution.

It also enables us to load traces in our internal simulator.

Signed-off-by: Eric Anholt<e...@anholt.net>
Signed-off-by: Yuanhan Liu<yuanhan....@linux.intel.com>
Signed-off-by: Kenneth Graunke<kenn...@whitecape.org>
---
  intel/Makefile.am        |    1 +
  intel/intel_aub.h        |  123 ++++++++++++++++++
  intel/intel_bufmgr.h     |   14 ++
  intel/intel_bufmgr_gem.c |  315 ++++++++++++++++++++++++++++++++++++++++++++++
  4 files changed, 453 insertions(+), 0 deletions(-)
  create mode 100644 intel/intel_aub.h

diff --git a/intel/Makefile.am b/intel/Makefile.am
index 06362b6..dc01a96 100644
--- a/intel/Makefile.am
+++ b/intel/Makefile.am
@@ -53,6 +53,7 @@ intel_bufmgr_gem_o_CFLAGS = $(AM_CFLAGS) -c99

  libdrm_intelincludedir = ${includedir}/libdrm
  libdrm_intelinclude_HEADERS = intel_bufmgr.h \
+                             intel_aub.h \
                              intel_debug.h

  # This may be interesting even outside of "make check", due to the -dump 
option.
diff --git a/intel/intel_aub.h b/intel/intel_aub.h
new file mode 100644
index 0000000..a36fd53
--- /dev/null
+++ b/intel/intel_aub.h
@@ -0,0 +1,123 @@
+/*
+ * Copyright © 2010 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<e...@anholt.net>
+ *
+ */
+
+/** @file intel_aub.h
+ *
+ * The AUB file is a file format used by Intel's internal simulation
+ * and other validation tools.  It can be used at various levels by a
+ * driver to input state to the simulated hardware or a replaying
+ * debugger.
+ *
+ * We choose to dump AUB files using the trace block format for ease
+ * of implementation -- dump out the blocks of memory as plain blobs
+ * and insert ring commands to execute the batchbuffer blob.
+ */
+
+#ifndef _INTEL_AUB_H
+#define _INTEL_AUB_H
+
+#define AUB_MI_NOOP                    (0)
+#define AUB_MI_BATCH_BUFFER_START      (0x31<<  23)
+#define AUB_PIPE_CONTROL               (0x7a000002)
+
+/* DW0: instruction type. */
+
+#define CMD_AUB                        (7<<  29)
+
+#define CMD_AUB_HEADER         (CMD_AUB | (1<<  23) | (0x05<<  16))
+/* DW1 */
+# define AUB_HEADER_MAJOR_SHIFT                24
+# define AUB_HEADER_MINOR_SHIFT                16
+
+#define CMD_AUB_TRACE_HEADER_BLOCK (CMD_AUB | (1<<  23) | (0x41<<  16))
+#define CMD_AUB_DUMP_BMP           (CMD_AUB | (1<<  23) | (0x9e<<  16))
+
+/* DW1 */
+#define AUB_TRACE_OPERATION_MASK       0x000000ff
+#define AUB_TRACE_OP_COMMENT           0x00000000
+#define AUB_TRACE_OP_DATA_WRITE                0x00000001
+#define AUB_TRACE_OP_COMMAND_WRITE     0x00000002
+#define AUB_TRACE_OP_MMIO_WRITE                0x00000003
+// operation = TRACE_DATA_WRITE, Type
+#define AUB_TRACE_TYPE_MASK            0x0000ff00
+#define AUB_TRACE_TYPE_NOTYPE          (0<<  8)
+#define AUB_TRACE_TYPE_BATCH           (1<<  8)
+#define AUB_TRACE_TYPE_VERTEX_BUFFER   (5<<  8)
+#define AUB_TRACE_TYPE_2D_MAP          (6<<  8)
+#define AUB_TRACE_TYPE_CUBE_MAP                (7<<  8)
+#define AUB_TRACE_TYPE_VOLUME_MAP      (9<<  8)
+#define AUB_TRACE_TYPE_1D_MAP          (10<<  8)
+#define AUB_TRACE_TYPE_CONSTANT_BUFFER (11<<  8)
+#define AUB_TRACE_TYPE_CONSTANT_URB    (12<<  8)
+#define AUB_TRACE_TYPE_INDEX_BUFFER    (13<<  8)
+#define AUB_TRACE_TYPE_GENERAL         (14<<  8)
+#define AUB_TRACE_TYPE_SURFACE         (15<<  8)
+
+
+// operation = TRACE_COMMAND_WRITE, Type =
+#define AUB_TRACE_TYPE_RING_HWB                (1<<  8)
+#define AUB_TRACE_TYPE_RING_PRB0       (2<<  8)
+#define AUB_TRACE_TYPE_RING_PRB1       (3<<  8)
+#define AUB_TRACE_TYPE_RING_PRB2       (4<<  8)
+
+// Address space
+#define AUB_TRACE_ADDRESS_SPACE_MASK   0x00ff0000
+#define AUB_TRACE_MEMTYPE_GTT          (0<<  16)
+#define AUB_TRACE_MEMTYPE_LOCAL                (1<<  16)
+#define AUB_TRACE_MEMTYPE_NONLOCAL     (2<<  16)
+#define AUB_TRACE_MEMTYPE_PCI          (3<<  16)
+#define AUB_TRACE_MEMTYPE_GTT_ENTRY     (4<<  16)
+
+/* DW2 */
+// operation = TRACE_DATA_WRITE, Type = TRACE_DATA_WRITE_GENERAL_STATE
+#define AUB_TRACE_GENERAL_STATE_MASK   0x000000ff
+
+#define AUB_TRACE_VS_STATE             0x00000001
+#define AUB_TRACE_GS_STATE             0x00000002
+#define AUB_TRACE_CL_STATE             0x00000003
+#define AUB_TRACE_SF_STATE             0x00000004
+#define AUB_TRACE_WM_STATE             0x00000005
+#define AUB_TRACE_CC_STATE             0x00000006
+#define AUB_TRACE_CL_VP                        0x00000007
+#define AUB_TRACE_SF_VP                        0x00000008
+#define AUB_TRACE_CC_VP                        0x00000009
+#define AUB_TRACE_SAMPLER_STATE                0x0000000a
+#define AUB_TRACE_KERNEL               0x0000000b
+#define AUB_TRACE_SCRATCH              0x0000000c
+#define AUB_TRACE_SDC                  0x0000000d
+#define AUB_TRACE_BLEND_STATE          0x00000016
+#define AUB_TRACE_DEPTH_STENCIL_STATE  0x00000017
+
+// operation = TRACE_DATA_WRITE, Type = TRACE_DATA_WRITE_SURFACE_STATE
+#define AUB_TRACE_SURFACE_STATE_MASK   0x00000ff00
+#define AUB_TRACE_BINDING_TABLE                0x000000100
+#define AUB_TRACE_SURFACE_STATE                0x000000200
+
+/* DW3: address */
+/* DW4: len */
+
+#endif /* _INTEL_AUB_H */
diff --git a/intel/intel_bufmgr.h b/intel/intel_bufmgr.h
index 8036031..fa6f2b8 100644
--- a/intel/intel_bufmgr.h
+++ b/intel/intel_bufmgr.h
@@ -36,6 +36,7 @@

  #include<stdio.h>
  #include<stdint.h>
+#include<stdio.h>

  struct drm_clip_rect;

@@ -84,6 +85,13 @@ struct _drm_intel_bo {
        int handle;
  };

+enum aub_dump_bmp_format {
+       AUB_DUMP_BMP_FORMAT_8BIT = 1,
+       AUB_DUMP_BMP_FORMAT_ARGB_4444 = 4,
+       AUB_DUMP_BMP_FORMAT_ARGB_0888 = 6,
+       AUB_DUMP_BMP_FORMAT_ARGB_8888 = 7,
+};
+
  #define BO_ALLOC_FOR_RENDER (1<<0)

  drm_intel_bo *drm_intel_bo_alloc(drm_intel_bufmgr *bufmgr, const char *name,
@@ -154,6 +162,12 @@ int drm_intel_gem_bo_get_reloc_count(drm_intel_bo *bo);
  void drm_intel_gem_bo_clear_relocs(drm_intel_bo *bo, int start);
  void drm_intel_gem_bo_start_gtt_access(drm_intel_bo *bo, int write_enable);

+void drm_intel_bufmgr_gem_set_aub_dump(drm_intel_bufmgr *bufmgr, int enable);
+void drm_intel_gem_bo_aub_dump_bmp(drm_intel_bo *bo,
+                                  int x1, int y1, int width, int height,
+                                  enum aub_dump_bmp_format format,
+                                  int pitch, int offset);
+
  int drm_intel_get_pipe_from_crtc_id(drm_intel_bufmgr *bufmgr, int crtc_id);

  int drm_intel_get_aperture_sizes(int fd, size_t *mappable, size_t *total);
diff --git a/intel/intel_bufmgr_gem.c b/intel/intel_bufmgr_gem.c
index ba38e50..45585f7 100644
--- a/intel/intel_bufmgr_gem.c
+++ b/intel/intel_bufmgr_gem.c
@@ -58,6 +58,7 @@
  #include "intel_bufmgr.h"
  #include "intel_bufmgr_priv.h"
  #include "intel_chipset.h"
+#include "intel_aub.h"
  #include "string.h"

  #include "i915_drm.h"
@@ -121,6 +122,9 @@ typedef struct _drm_intel_bufmgr_gem {
        unsigned int bo_reuse : 1;
        unsigned int no_exec : 1;
        bool fenced_relocs;
+
+       FILE *aub_file;
+       uint32_t aub_offset;
  } drm_intel_bufmgr_gem;

  #define DRM_INTEL_RELOC_FENCE (1<<0)
@@ -215,6 +219,8 @@ struct _drm_intel_bo_gem {

        /** Flags that we may need to do the SW_FINSIH ioctl on unmap. */
        bool mapped_cpu_write;
+
+       uint32_t aub_offset;
  };

  static unsigned int
@@ -1715,6 +1721,247 @@ drm_intel_update_buffer_offsets2 (drm_intel_bufmgr_gem 
*bufmgr_gem)
        }
  }

+static void
+aub_out(drm_intel_bufmgr_gem *bufmgr_gem, uint32_t data)
+{
+       fwrite(&data, 1, 4, bufmgr_gem->aub_file);
+}
+
+static void
+aub_out_data(drm_intel_bufmgr_gem *bufmgr_gem, void *data, size_t size)
+{
+       fwrite(data, 1, size, bufmgr_gem->aub_file);
+}
+
+static void
+aub_write_bo_data(drm_intel_bo *bo, uint32_t offset, uint32_t size)
+{
+       drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *) bo->bufmgr;
+       drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *) bo;
+       uint32_t *data;
+       unsigned int i;
+
+       data = malloc(bo->size);
+       drm_intel_bo_get_subdata(bo, offset, size, data);
+
+       /* Easy mode: write out bo with no relocations */
+       if (!bo_gem->reloc_count) {
+               aub_out_data(bufmgr_gem, data, size);
+               free(data);
+               return;
+       }
+
+       /* Otherwise, handle the relocations while writing. */
+       for (i = 0; i<  size / 4; i++) {
+               int r;
+               for (r = 0; r<  bo_gem->reloc_count; r++) {
+                       struct drm_i915_gem_relocation_entry *reloc;
+                       drm_intel_reloc_target *info;
+
+                       reloc =&bo_gem->relocs[r];
+                       info =&bo_gem->reloc_target_info[r];
+
+                       if (reloc->offset == offset + i * 4) {
+                               drm_intel_bo_gem *target_gem;
+                               uint32_t val;
+
+                               target_gem = (drm_intel_bo_gem *)info->bo;
+
+                               val = reloc->delta;
+                               val += target_gem->aub_offset;
+
+                               aub_out(bufmgr_gem, val);
+                               data[i] = val;
+                               break;
+                       }
+               }
+               if (r == bo_gem->reloc_count) {
+                       /* no relocation, just the data */
+                       aub_out(bufmgr_gem, data[i]);
+               }
+       }
+
+       free(data);
+}
+
+static void
+aub_bo_get_address(drm_intel_bo *bo)
+{
+       drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *) bo->bufmgr;
+       drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *) bo;
+
+       /* Give the object a graphics address in the AUB file.  We
+        * don't just use the GEM object address because we do AUB
+        * dumping before execution -- we want to successfully log
+        * when the hardware might hang, and we might even want to aub
+        * capture for a driver trying to execute on a different
+        * generation of hardware by disabling the actual kernel exec
+        * call.
+        */
+       bo_gem->aub_offset = bufmgr_gem->aub_offset;
+       bufmgr_gem->aub_offset += bo->size;
+       /* XXX: Handle aperture overflow. */
+       assert(bufmgr_gem->aub_offset<  256 * 1024 * 1024);
+}
+
+static void
+aub_write_trace_block(drm_intel_bo *bo, uint32_t type, uint32_t subtype,
+                     uint32_t offset, uint32_t size)
+{
+       drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *) bo->bufmgr;
+       drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *) bo;
+
+       aub_out(bufmgr_gem,
+               CMD_AUB_TRACE_HEADER_BLOCK |
+               (5 - 2));
+       aub_out(bufmgr_gem,
+               AUB_TRACE_MEMTYPE_GTT | type | AUB_TRACE_OP_DATA_WRITE);
+       aub_out(bufmgr_gem, subtype);
+       aub_out(bufmgr_gem, bo_gem->aub_offset + offset);
+       aub_out(bufmgr_gem, size);
+       aub_write_bo_data(bo, offset, size);
+}
+
+static void
+aub_write_bo(drm_intel_bo *bo)
+{
+       uint32_t block_size;
+       uint32_t offset;
+
+       aub_bo_get_address(bo);
+
+       /* Break up large objects into multiple writes.  Otherwise a
+        * 128kb VBO would overflow the 16 bits of size field in the
+        * packet header and everything goes badly after that.
+        */
+       for (offset = 0; offset<  bo->size; offset += block_size) {
+               block_size = bo->size - offset;
+
+               if (block_size>  8 * 4096)
+                       block_size = 8 * 4096;
+
+               aub_write_trace_block(bo, AUB_TRACE_TYPE_NOTYPE, 0,
+                                     offset, block_size);
+       }
+}
+
+/*
+ * Make a ringbuffer on fly and dump it
+ */
+static void
+aub_build_dump_ringbuffer(drm_intel_bufmgr_gem *bufmgr_gem,
+                         uint32_t batch_buffer, int ring_flag)
+{
+       uint32_t ringbuffer[4096];
+       int ring = AUB_TRACE_TYPE_RING_PRB0; /* The default ring */
+       int ring_count = 0;
+
+       if (ring_flag == I915_EXEC_BSD)
+               ring = AUB_TRACE_TYPE_RING_PRB1;
+
+       /* Make a ring buffer to execute our batchbuffer. */
+       memset(ringbuffer, 0, sizeof(ringbuffer));
+       ringbuffer[ring_count++] = AUB_MI_BATCH_BUFFER_START;
+       ringbuffer[ring_count++] = batch_buffer;
+
+       /* Write out the ring.  This appears to trigger execution of
+        * the ring in the simulator.
+        */
+       aub_out(bufmgr_gem,
+               CMD_AUB_TRACE_HEADER_BLOCK |
+               (5 - 2));
+       aub_out(bufmgr_gem,
+               AUB_TRACE_MEMTYPE_GTT | ring | AUB_TRACE_OP_COMMAND_WRITE);
+       aub_out(bufmgr_gem, 0); /* general/surface subtype */
+       aub_out(bufmgr_gem, bufmgr_gem->aub_offset);
+       aub_out(bufmgr_gem, ring_count * 4);
+
+       /* FIXME: Need some flush operations here? */
+       aub_out_data(bufmgr_gem, ringbuffer, ring_count * 4);
+
+       /* Update offset pointer */
+       bufmgr_gem->aub_offset += 4096;
+}
+
+void
+drm_intel_gem_bo_aub_dump_bmp(drm_intel_bo *bo,
+                             int x1, int y1, int width, int height,
+                             enum aub_dump_bmp_format format,
+                             int pitch, int offset)
+{
+       drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *) bo->bufmgr;
+       drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
+       uint32_t cpp;
+
+       switch (format) {
+       case AUB_DUMP_BMP_FORMAT_8BIT:
+               cpp = 1;
+               break;
+       case AUB_DUMP_BMP_FORMAT_ARGB_4444:
+               cpp = 2;
+               break;
+       case AUB_DUMP_BMP_FORMAT_ARGB_0888:
+       case AUB_DUMP_BMP_FORMAT_ARGB_8888:
+               cpp = 4;
+               break;
+       default:
+               printf("Unknown AUB dump format %d\n", format);
+               return;
+       }
+
+       if (!bufmgr_gem->aub_file)
+               return;
+
+       aub_out(bufmgr_gem, CMD_AUB_DUMP_BMP | 4);
+       aub_out(bufmgr_gem, (y1<<  16) | x1);
+       aub_out(bufmgr_gem,
+               (format<<  24) |
+               (cpp<<  19) |
+               pitch / 4);
+       aub_out(bufmgr_gem, (height<<  16) | width);
+       aub_out(bufmgr_gem, bo_gem->aub_offset + offset);
+       aub_out(bufmgr_gem,
+               ((bo_gem->tiling_mode != I915_TILING_NONE) ? (1<<  2) : 0) |
+               ((bo_gem->tiling_mode == I915_TILING_Y) ? (1<<  3) : 0));
+}
+
+static void
+aub_exec(drm_intel_bo *bo, int ring_flag, int used)
+{
+       drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *) bo->bufmgr;
+       drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *) bo;
+       int i;
+
+       if (!bufmgr_gem->aub_file)
+               return;
+
+       /* Write out all but the batchbuffer to AUB memory */
+       for (i = 0; i<  bufmgr_gem->exec_count - 1; i++) {
+               if (bufmgr_gem->exec_bos[i] != bo)
+                       aub_write_bo(bufmgr_gem->exec_bos[i]);
+       }
+
+       aub_bo_get_address(bo);
+
+       /* Dump the batchbuffer. */
+       aub_write_trace_block(bo, AUB_TRACE_TYPE_BATCH, 0,
+                             0, used);
+       aub_write_trace_block(bo, AUB_TRACE_TYPE_NOTYPE, 0,
+                             used, bo->size - used);
+
+       /* Dump ring buffer */
+       aub_build_dump_ringbuffer(bufmgr_gem, bo_gem->aub_offset, ring_flag);
+
+       fflush(bufmgr_gem->aub_file);
+
+       /*
+        * One frame has been dumped. So reset the aub_offset for the next 
frame.
+        *
+        * FIXME: Can we do this?
+        */
+       bufmgr_gem->aub_offset = 0x10000;
+}
+
  static int
  drm_intel_gem_bo_exec(drm_intel_bo *bo, int used,
                      drm_clip_rect_t * cliprects, int num_cliprects, int DR4)
@@ -1830,6 +2077,8 @@ drm_intel_gem_bo_mrb_exec2(drm_intel_bo *bo, int used,
        execbuf.rsvd1 = 0;
        execbuf.rsvd2 = 0;

+       aub_exec(bo, flags, used);
+
        if (bufmgr_gem->no_exec)
                goto skip_execution;

@@ -2360,6 +2609,72 @@ drm_intel_bufmgr_gem_get_devid(drm_intel_bufmgr *bufmgr)
  }

  /**
+ * Sets up AUB dumping.
+ *
+ * This is a trace file format that can be used with the simulator.
+ * Packets are emitted in a format somewhat like GPU command packets.
+ * You can set up a GTT and upload your objects into the referenced
+ * space, then send off batchbuffers and get BMPs out the other end.
+ */
+void
+drm_intel_bufmgr_gem_set_aub_dump(drm_intel_bufmgr *bufmgr, int enable)
+{
+       drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bufmgr;
+       int entry = 0x200003;
+       int i;
+       int gtt_size = 0x10000;
+
+       if (!enable) {
+               if (bufmgr_gem->aub_file) {
+                       fclose(bufmgr_gem->aub_file);
+                       bufmgr_gem->aub_file = NULL;
+               }
+       }
+
+       if (geteuid() != getuid())
+               return;
+
+       bufmgr_gem->aub_file = fopen("intel.aub", "w+");

I guess it's would be better that we can name the aub dump file
according to the program we are tracing, like if we run:
  $ INTEL_DEBUG=aub glxgears

I guess it would be good if we get a glxgears.aub but not intel.aub.
Otherwise, it would override the former one we dumped. It's somehow a
little un-convenient.

That would be a nice feature, though I'm okay with intel.aub for now.

+       if (!bufmgr_gem->aub_file)
+               return;
+
+       /* Start allocating objects from just after the GTT. */
+       bufmgr_gem->aub_offset = gtt_size;
+
+       /* Start with a (required) version packet. */
+       aub_out(bufmgr_gem, CMD_AUB_HEADER | (13 - 2));
+       aub_out(bufmgr_gem,
+               (4<<  AUB_HEADER_MAJOR_SHIFT) |
+               (0<<  AUB_HEADER_MINOR_SHIFT));
+       for (i = 0; i<  8; i++) {
+               aub_out(bufmgr_gem, 0); /* app name */
+       }
+       aub_out(bufmgr_gem, 0); /* timestamp */
+       aub_out(bufmgr_gem, 0); /* timestamp */
+       aub_out(bufmgr_gem, 0); /* comment len */
+
+       /* Set up the GTT. The max we can handle is 256M */
+       aub_out(bufmgr_gem, CMD_AUB_TRACE_HEADER_BLOCK | (5 - 2));
+       aub_out(bufmgr_gem, AUB_TRACE_MEMTYPE_NONLOCAL | 0 | 
AUB_TRACE_OP_DATA_WRITE);
+       aub_out(bufmgr_gem, 0); /* subtype */
+       aub_out(bufmgr_gem, 0); /* offset */
+       aub_out(bufmgr_gem, gtt_size); /* size */
+       for (i = 0x000; i<  gtt_size; i += 4, entry += 0x1000) {
+               aub_out(bufmgr_gem, entry);
+       }
+
+       /* MI_FLUSH enable */
+       if (bufmgr_gem->gen>= 6) {
+               aub_out(bufmgr_gem, CMD_AUB_TRACE_HEADER_BLOCK | (5 - 2));
+               aub_out(bufmgr_gem, AUB_TRACE_OP_MMIO_WRITE);
+               aub_out(bufmgr_gem, 0);
+               aub_out(bufmgr_gem, 0x209c);  /* reg addr */
+               aub_out(bufmgr_gem, 0x04);    /* Length in byte */
+               aub_out(bufmgr_gem, ((1<<  12)<<  16) | (1<<  12));
+       }

Zhenyu and me came to an agreement that it's the driver side to do the
MI_FLUSH enable stuff. Since using MI_FLUSH is deprecated, and if you still
use it, it is your job to set up the MI_FLUSH enable bit.

Oh...yeah, we should definitely drop that.

+}
+
+/**
   * Initializes the GEM buffer manager, which uses the kernel to allocate, map,
   * and manage map buffer objections.
   *
--
1.7.9.1

Otherwise,
Reviewed-by: Kenneth Graunke <kenn...@whitecape.org>

Please set yourself as the author, though - I just squashed this and fixed up a few minor things.
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

Reply via email to