-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);
-}
-
-/**
- * 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.
- */
-static void
-aub_write_large_trace_block(drm_intel_bo *bo, uint32_t type, uint32_t subtype,
- uint32_t offset, uint32_t size)
-{
- uint32_t block_size;
- uint32_t sub_offset;
-
- for (sub_offset = 0; sub_offset < size; sub_offset += block_size) {
- block_size = size - sub_offset;
-
- if (block_size > 8 * 4096)
- block_size = 8 * 4096;
-
- aub_write_trace_block(bo, type, subtype, offset + sub_offset,
- block_size);
- }
-}
-
-static void
-aub_write_bo(drm_intel_bo *bo)
-{
- drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *) bo;
- uint32_t offset = 0;
- unsigned i;
-
- aub_bo_get_address(bo);
-
- /* Write out each annotated section separately. */
- for (i = 0; i < bo_gem->aub_annotation_count; ++i) {
- drm_intel_aub_annotation *annotation =
- &bo_gem->aub_annotations[i];
- uint32_t ending_offset = annotation->ending_offset;
- if (ending_offset > bo->size)
- ending_offset = bo->size;
- if (ending_offset > offset) {
- aub_write_large_trace_block(bo, annotation->type,
- annotation->subtype,
- offset,
- ending_offset - offset);
- offset = ending_offset;
- }
- }
-
- /* Write out any remaining unannotated data */
- if (offset < bo->size) {
- aub_write_large_trace_block(bo, AUB_TRACE_TYPE_NOTYPE, 0,
- offset, bo->size - offset);
- }
-}
-
-/*
- * 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;
-}
-