intel/decoders: handle decoding MI_BBS from ring

An MI_BATCH_BUFFER_START in the ring buffer acts as a second level
batchbuffer (aka jump back to ring buffer when running into a
MI_BATCH_BUFFER_END).

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>

src/intel/common/gen_batch_decoder.c

@@ -805,7 +805,7 @@ struct custom_decoder {
 void
 gen_print_batch(struct gen_batch_decode_ctx *ctx,
                 const uint32_t *batch, uint32_t batch_size,
-                uint64_t batch_addr)
+                uint64_t batch_addr, bool from_ring)
 {
    const uint32_t *p, *end = batch + batch_size / sizeof(uint32_t);
    int length;
@@ -887,7 +887,7 @@ gen_print_batch(struct gen_batch_decode_ctx *ctx,
                     next_batch_addr);
          } else {
             gen_print_batch(ctx, next_batch.map, next_batch.size,
-                            next_batch.addr);
+                            next_batch.addr, false);
          }
          if (second_level) {
             /* MI_BATCH_BUFFER_START with "2nd Level Batch Buffer" set acts
@@ -896,7 +896,7 @@ gen_print_batch(struct gen_batch_decode_ctx *ctx,
              * MI_BATCH_BUFFER_END.
              */
             continue;
-         } else {
+         } else if (!from_ring) {
             /* MI_BATCH_BUFFER_START with "2nd Level Batch Buffer" unset acts
              * like a goto.  Nothing after it will ever get processed.  In
              * order to prevent the recursion from growing, we just reset the

src/intel/common/gen_decoder.h

@@ -254,7 +254,7 @@ void gen_batch_decode_ctx_finish(struct gen_batch_decode_ctx *ctx);
 
 void gen_print_batch(struct gen_batch_decode_ctx *ctx,
                      const uint32_t *batch, uint32_t batch_size,
-                     uint64_t batch_addr);
+                     uint64_t batch_addr, bool from_ring);
 
 #ifdef __cplusplus
 }

src/intel/tools/aubinator.c

@@ -166,7 +166,7 @@ handle_execlist_write(void *user_data, enum drm_i915_gem_engine_class engine, ui
    batch_ctx.engine = engine;
    gen_print_batch(&batch_ctx, commands,
                    MIN2(ring_buffer_tail - ring_buffer_head, ring_buffer_length),
-                   ring_bo.addr + ring_buffer_head);
+                   ring_bo.addr + ring_buffer_head, true);
    aub_mem_clear_bo_maps(&mem);
 }
 
@@ -184,7 +184,7 @@ handle_ring_write(void *user_data, enum drm_i915_gem_engine_class engine,
    batch_ctx.get_bo = get_legacy_bo;
 
    batch_ctx.engine = engine;
-   gen_print_batch(&batch_ctx, data, data_len, 0);
+   gen_print_batch(&batch_ctx, data, data_len, 0, false);
 
    aub_mem_clear_bo_maps(&mem);
 }

src/intel/tools/aubinator_error_decode.c

@@ -612,7 +612,7 @@ read_data_file(FILE *file)
          batch_ctx.engine = class;
          gen_print_batch(&batch_ctx, sections[s].data,
                          sections[s].dword_count * 4,
-                         sections[s].gtt_offset);
+                         sections[s].gtt_offset, false);
       }
    }
 

src/intel/tools/aubinator_viewer.cpp

@@ -702,7 +702,7 @@ display_batch_ring_write(void *user_data, enum drm_i915_gem_engine_class engine,
 
    window->uses_ppgtt = false;
 
-   aub_viewer_render_batch(&window->decode_ctx, data, data_len, 0);
+   aub_viewer_render_batch(&window->decode_ctx, data, data_len, 0, false);
 }
 
 static void
@@ -737,7 +737,7 @@ display_batch_execlist_write(void *user_data,
    window->decode_ctx.engine = engine;
    aub_viewer_render_batch(&window->decode_ctx, commands,
                            MIN2(ring_buffer_tail - ring_buffer_head, ring_buffer_length),
-                           ring_buffer_start + ring_buffer_head);
+                           ring_buffer_start + ring_buffer_head, true);
 }
 
 static void

src/intel/tools/aubinator_viewer.h

@@ -95,6 +95,6 @@ void aub_viewer_decode_ctx_init(struct aub_viewer_decode_ctx *ctx,
 
 void aub_viewer_render_batch(struct aub_viewer_decode_ctx *ctx,
                              const void *batch, uint32_t batch_size,
-                             uint64_t batch_addr);
+                             uint64_t batch_addr, bool from_ring);
 
 #endif /* AUBINATOR_VIEWER_H */

src/intel/tools/aubinator_viewer_decoder.cpp

@@ -892,7 +892,7 @@ struct custom_decoder info_decoders[] = {
 void
 aub_viewer_render_batch(struct aub_viewer_decode_ctx *ctx,
                         const void *_batch, uint32_t batch_size,
-                        uint64_t batch_addr)
+                        uint64_t batch_addr, bool from_ring)
 {
    struct gen_group *inst;
    const uint32_t *p, *batch = (const uint32_t *) _batch, *end = batch + batch_size / sizeof(uint32_t);
@@ -970,7 +970,7 @@ aub_viewer_render_batch(struct aub_viewer_decode_ctx *ctx,
                                next_batch_addr);
          } else {
             aub_viewer_render_batch(ctx, next_batch.map, next_batch.size,
-                                    next_batch.addr);
+                                    next_batch.addr, false);
          }
          if (second_level) {
             /* MI_BATCH_BUFFER_START with "2nd Level Batch Buffer" set acts
@@ -979,7 +979,7 @@ aub_viewer_render_batch(struct aub_viewer_decode_ctx *ctx,
              * MI_BATCH_BUFFER_END.
              */
             continue;
-         } else {
+         } else if (!from_ring) {
             /* MI_BATCH_BUFFER_START with "2nd Level Batch Buffer" unset acts
              * like a goto.  Nothing after it will ever get processed.  In
              * order to prevent the recursion from growing, we just reset the

src/intel/vulkan/anv_batch_chain.c

@@ -1735,7 +1735,7 @@ anv_cmd_buffer_execbuf(struct anv_device *device,
 
          device->cmd_buffer_being_decoded = cmd_buffer;
          gen_print_batch(&device->decoder_ctx, (*bo)->bo.map,
-                         (*bo)->bo.size, (*bo)->bo.offset);
+                         (*bo)->bo.size, (*bo)->bo.offset, false);
          device->cmd_buffer_being_decoded = NULL;
       }
 

src/intel/vulkan/anv_device.c

@@ -1795,11 +1795,13 @@ get_bo_from_pool(struct gen_batch_decode_bo *ret,
 
 /* Finding a buffer for batch decoding */
 static struct gen_batch_decode_bo
-decode_get_bo(void *v_batch, uint64_t address)
+decode_get_bo(void *v_batch, bool ppgtt, uint64_t address)
 {
    struct anv_device *device = v_batch;
    struct gen_batch_decode_bo ret_bo = {};
 
+   assert(ppgtt);
+
    if (get_bo_from_pool(&ret_bo, &device->dynamic_state_pool.block_pool, address))
       return ret_bo;
    if (get_bo_from_pool(&ret_bo, &device->instruction_state_pool.block_pool, address))

src/intel/vulkan/anv_queue.c

@@ -101,7 +101,7 @@ anv_device_submit_simple_batch(struct anv_device *device,
    execbuf.rsvd2 = 0;
 
    if (unlikely(INTEL_DEBUG & DEBUG_BATCH))
-      gen_print_batch(&device->decoder_ctx, bo.map, bo.size, bo.offset);
+      gen_print_batch(&device->decoder_ctx, bo.map, bo.size, bo.offset, false);
 
    result = anv_device_execbuf(device, &execbuf, exec_bos);
    if (result != VK_SUCCESS)

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

@@ -828,7 +828,7 @@ submit_batch(struct brw_context *brw, int in_fence_fd, int *out_fence_fd)
    if (unlikely(INTEL_DEBUG & DEBUG_BATCH)) {
       gen_print_batch(&batch->decoder, batch->batch.map,
                       4 * USED_BATCH(*batch),
-                      batch->batch.bo->gtt_offset);
+                      batch->batch.bo->gtt_offset, false);
    }
 
    if (brw->ctx.Const.ResetStrategy == GL_LOSE_CONTEXT_ON_RESET_ARB)