intel/ds: lock submissions to u_trace_context

This allows for submissions to be outside the global anv_device mutex.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24744>

src/gallium/drivers/iris/iris_fence.c

@@ -270,8 +270,7 @@ iris_fence_flush(struct pipe_context *ctx,
       iris_measure_frame_end(ice);
    }
 
-   u_trace_context_process(&ice->ds.trace_context,
+   intel_ds_device_process(&ice->ds, flags & PIPE_FLUSH_END_OF_FRAME);
-                           flags & PIPE_FLUSH_END_OF_FRAME);
 
    if (!out_fence)
       return;

src/gallium/drivers/iris/iris_utrace.c

@@ -168,7 +168,7 @@ void iris_utrace_flush(struct iris_batch *batch, uint64_t submission_id)
 {
    struct intel_ds_flush_data *flush_data = malloc(sizeof(*flush_data));
    intel_ds_flush_data_init(flush_data, &batch->ds, submission_id);
-   u_trace_flush(&batch->trace, flush_data, false);
+   intel_ds_queue_flush_data(&batch->ds, &batch->trace, flush_data, false);
 }
 
 void iris_utrace_init(struct iris_context *ice)

src/intel/ds/intel_driver_ds.cc

@@ -555,12 +555,14 @@ intel_ds_device_init(struct intel_ds_device *device,
    device->iid = get_iid();
    device->api = api;
    list_inithead(&device->queues);
+   simple_mtx_init(&device->trace_context_mutex, mtx_plain);
 }
 
 void
 intel_ds_device_fini(struct intel_ds_device *device)
 {
    u_trace_context_fini(&device->trace_context);
+   simple_mtx_destroy(&device->trace_context_mutex);
 }
 
 struct intel_ds_queue *
@@ -606,6 +608,24 @@ void intel_ds_flush_data_fini(struct intel_ds_flush_data *data)
    u_trace_fini(&data->trace);
 }
 
+void intel_ds_queue_flush_data(struct intel_ds_queue *queue,
+                               struct u_trace *ut,
+                               struct intel_ds_flush_data *data,
+                               bool free_data)
+{
+   simple_mtx_lock(&queue->device->trace_context_mutex);
+   u_trace_flush(ut, data, free_data);
+   simple_mtx_unlock(&queue->device->trace_context_mutex);
+}
+
+void intel_ds_device_process(struct intel_ds_device *device,
+                             bool eof)
+{
+   simple_mtx_lock(&device->trace_context_mutex);
+   u_trace_context_process(&device->trace_context, eof);
+   simple_mtx_unlock(&device->trace_context_mutex);
+}
+
 #ifdef __cplusplus
 }
 #endif

src/intel/ds/intel_driver_ds.h

@@ -112,6 +112,9 @@ struct intel_ds_device {
     */
    uint64_t event_id;
 
+   /* Protects submissions of u_trace data to trace_context */
+   simple_mtx_t trace_context_mutex;
+
    struct u_trace_context trace_context;
 
    /* List of intel_ds_queue */
@@ -183,6 +186,13 @@ void intel_ds_flush_data_init(struct intel_ds_flush_data *data,
 
 void intel_ds_flush_data_fini(struct intel_ds_flush_data *data);
 
+void intel_ds_queue_flush_data(struct intel_ds_queue *queue,
+                               struct u_trace *ut,
+                               struct intel_ds_flush_data *data,
+                               bool free_data);
+
+void intel_ds_device_process(struct intel_ds_device *device, bool eof);
+
 #ifdef HAVE_PERFETTO
 
 uint64_t intel_ds_begin_submit(struct intel_ds_queue *queue);

src/intel/vulkan/anv_batch_chain.c

@@ -1481,10 +1481,10 @@ anv_queue_submit(struct vk_queue *vk_queue,
    /* Take submission ID under lock */
    intel_ds_end_submit(&queue->ds, start_ts);
 
-   u_trace_context_process(&device->ds.trace_context, true);
-
    pthread_mutex_unlock(&device->mutex);
 
+   intel_ds_device_process(&device->ds, true);
+
    return result;
 }
 

src/intel/vulkan/anv_utrace.c

@@ -232,7 +232,8 @@ anv_device_utrace_flush_cmd_buffers(struct anv_queue *queue,
                                                      &submit->batch);
          for (uint32_t i = 0; i < cmd_buffer_count; i++) {
             if (cmd_buffers[i]->usage_flags & VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT) {
-               u_trace_flush(&cmd_buffers[i]->trace, submit, false);
+               intel_ds_queue_flush_data(&queue->ds, &cmd_buffers[i]->trace,
+                                         &submit->ds, false);
             } else {
                u_trace_clone_append(u_trace_begin_iterator(&cmd_buffers[i]->trace),
                                     u_trace_end_iterator(&cmd_buffers[i]->trace),
@@ -258,7 +259,8 @@ anv_device_utrace_flush_cmd_buffers(struct anv_queue *queue,
 
          for (uint32_t i = 0; i < cmd_buffer_count; i++) {
             if (cmd_buffers[i]->usage_flags & VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT) {
-               u_trace_flush(&cmd_buffers[i]->trace, submit, false);
+               intel_ds_queue_flush_data(&queue->ds, &cmd_buffers[i]->trace,
+                                         &submit->ds, false);
             } else {
                u_trace_clone_append(u_trace_begin_iterator(&cmd_buffers[i]->trace),
                                     u_trace_end_iterator(&cmd_buffers[i]->trace),
@@ -271,7 +273,7 @@ anv_device_utrace_flush_cmd_buffers(struct anv_queue *queue,
          anv_genX(device->info, emit_simple_shader_end)(&submit->simple_state);
       }
 
-      u_trace_flush(&submit->ds.trace, submit, true);
+      intel_ds_queue_flush_data(&queue->ds, &submit->ds.trace, &submit->ds, true);
 
       if (submit->batch.status != VK_SUCCESS) {
          result = submit->batch.status;
@@ -280,7 +282,8 @@ anv_device_utrace_flush_cmd_buffers(struct anv_queue *queue,
    } else {
       for (uint32_t i = 0; i < cmd_buffer_count; i++) {
          assert(cmd_buffers[i]->usage_flags & VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT);
-         u_trace_flush(&cmd_buffers[i]->trace, submit, i == (cmd_buffer_count - 1));
+         intel_ds_queue_flush_data(&queue->ds, &cmd_buffers[i]->trace,
+                                   &submit->ds, i == (cmd_buffer_count - 1));
       }
    }
 
@@ -445,7 +448,7 @@ anv_device_utrace_init(struct anv_device *device)
 void
 anv_device_utrace_finish(struct anv_device *device)
 {
-   u_trace_context_process(&device->ds.trace_context, true);
+   intel_ds_device_process(&device->ds, true);
    intel_ds_device_fini(&device->ds);
    anv_bo_pool_finish(&device->utrace_bo_pool);
 }
@@ -574,7 +577,7 @@ anv_queue_trace(struct anv_queue *queue, const char *label, bool frame, bool beg
       goto error_reloc_list;
    }
 
-   u_trace_flush(&submit->ds.trace, submit, true);
+   intel_ds_queue_flush_data(&queue->ds, &submit->ds.trace, &submit->ds, true);
 
    pthread_mutex_lock(&device->mutex);
    device->kmd_backend->queue_exec_trace(queue, submit);
@@ -617,7 +620,7 @@ anv_QueueEndDebugUtilsLabelEXT(VkQueue _queue)
       anv_queue_trace(queue, label->pLabelName,
                       false /* frame */, false /* begin */);
 
-      u_trace_context_process(&queue->device->ds.trace_context, true);
+      intel_ds_device_process(&queue->device->ds, true);
    }
 
    vk_common_QueueEndDebugUtilsLabelEXT(_queue);

src/intel/vulkan_hasvk/anv_utrace.c

@@ -148,7 +148,8 @@ anv_device_utrace_flush_cmd_buffers(struct anv_queue *queue,
                                                    &flush->batch);
       for (uint32_t i = 0; i < cmd_buffer_count; i++) {
          if (cmd_buffers[i]->usage_flags & VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT) {
-            u_trace_flush(&cmd_buffers[i]->trace, flush, false);
+           intel_ds_queue_flush_data(&queue->ds, &cmd_buffers[i]->trace,
+                                     &flush->ds, false);
          } else {
             u_trace_clone_append(u_trace_begin_iterator(&cmd_buffers[i]->trace),
                                  u_trace_end_iterator(&cmd_buffers[i]->trace),
@@ -159,7 +160,7 @@ anv_device_utrace_flush_cmd_buffers(struct anv_queue *queue,
       }
       anv_genX(device->info, emit_so_memcpy_fini)(&flush->memcpy_state);
 
-      u_trace_flush(&flush->ds.trace, flush, true);
+      intel_ds_queue_flush_data(&queue->ds, &flush->ds.trace, &flush->ds, true);
 
       if (flush->batch.status != VK_SUCCESS) {
          result = flush->batch.status;
@@ -168,7 +169,8 @@ anv_device_utrace_flush_cmd_buffers(struct anv_queue *queue,
    } else {
       for (uint32_t i = 0; i < cmd_buffer_count; i++) {
          assert(cmd_buffers[i]->usage_flags & VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT);
-         u_trace_flush(&cmd_buffers[i]->trace, flush, i == (cmd_buffer_count - 1));
+         intel_ds_queue_flush_data(&queue->ds, &cmd_buffers[i]->trace,
+                                   &flush->ds, i == (cmd_buffer_count - 1));
       }
    }
 
@@ -293,7 +295,7 @@ anv_device_utrace_init(struct anv_device *device)
 void
 anv_device_utrace_finish(struct anv_device *device)
 {
-   u_trace_context_process(&device->ds.trace_context, true);
+   intel_ds_device_process(&device->ds, true);
    intel_ds_device_fini(&device->ds);
    anv_bo_pool_finish(&device->utrace_bo_pool);
 }

src/intel/vulkan_hasvk/anv_wsi.c

@@ -112,7 +112,7 @@ VkResult anv_QueuePresentKHR(
                                      _queue, 0,
                                      pPresentInfo);
 
-   u_trace_context_process(&device->ds.trace_context, true);
+   intel_ds_device_process(&device->ds, true);
 
    return result;
 }