radv: Avoid deadlock on bo_list.

With the kernel timeline sysncobj changes, the kernel submits do
not necessarily happen in global vkQueueSubmit order. Which should
be fine, we added the appropriate waits for that. (See
DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT in the winsys)

However, all kernel submissions take a lock on the bo_list mutex,
and since we do the wait in the winsys, we wait while having the
bo_list mutex held. This means that as soon as a wait and a signal
submission are out of order we have a deadlock on the bo_list mutex
and the wait.

Solution is to use a shared reader lock during the kernel submission,
as we only need read access for the submission.

Fixes: 6bc5ce7a91 "radv: Add timeline syncobj for timeline semaphores."
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3446
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6478>

src/amd/vulkan/radv_device.c

@@ -2372,7 +2372,7 @@ radv_queue_finish(struct radv_queue *queue)
 static void
 radv_bo_list_init(struct radv_bo_list *bo_list)
 {
-	pthread_mutex_init(&bo_list->mutex, NULL);
+	pthread_rwlock_init(&bo_list->rwlock, NULL);
 	bo_list->list.count = bo_list->capacity = 0;
 	bo_list->list.bos = NULL;
 }
@@ -2381,7 +2381,7 @@ static void
 radv_bo_list_finish(struct radv_bo_list *bo_list)
 {
 	free(bo_list->list.bos);
-	pthread_mutex_destroy(&bo_list->mutex);
+	pthread_rwlock_destroy(&bo_list->rwlock);
 }
 
 VkResult radv_bo_list_add(struct radv_device *device,
@@ -2395,13 +2395,13 @@ VkResult radv_bo_list_add(struct radv_device *device,
 	if (unlikely(!device->use_global_bo_list))
 		return VK_SUCCESS;
 
-	pthread_mutex_lock(&bo_list->mutex);
+	pthread_rwlock_wrlock(&bo_list->rwlock);
 	if (bo_list->list.count == bo_list->capacity) {
 		unsigned capacity = MAX2(4, bo_list->capacity * 2);
 		void *data = realloc(bo_list->list.bos, capacity * sizeof(struct radeon_winsys_bo*));
 
 		if (!data) {
-			pthread_mutex_unlock(&bo_list->mutex);
+			pthread_rwlock_unlock(&bo_list->rwlock);
 			return VK_ERROR_OUT_OF_HOST_MEMORY;
 		}
 
@@ -2410,7 +2410,7 @@ VkResult radv_bo_list_add(struct radv_device *device,
 	}
 
 	bo_list->list.bos[bo_list->list.count++] = bo;
-	pthread_mutex_unlock(&bo_list->mutex);
+	pthread_rwlock_unlock(&bo_list->rwlock);
 	return VK_SUCCESS;
 }
 
@@ -2425,7 +2425,7 @@ void radv_bo_list_remove(struct radv_device *device,
 	if (unlikely(!device->use_global_bo_list))
 		return;
 
-	pthread_mutex_lock(&bo_list->mutex);
+	pthread_rwlock_wrlock(&bo_list->rwlock);
 	/* Loop the list backwards so we find the most recently added
 	 * memory first. */
 	for(unsigned i = bo_list->list.count; i-- > 0;) {
@@ -2435,7 +2435,7 @@ void radv_bo_list_remove(struct radv_device *device,
 			break;
 		}
 	}
-	pthread_mutex_unlock(&bo_list->mutex);
+	pthread_rwlock_unlock(&bo_list->rwlock);
 }
 
 static void
@@ -4549,7 +4549,7 @@ radv_queue_submit_deferred(struct radv_deferred_queue_submission *submission,
 			sem_info.cs_emit_signal = j + advance == submission->cmd_buffer_count;
 
 			if (unlikely(queue->device->use_global_bo_list)) {
-				pthread_mutex_lock(&queue->device->bo_list.mutex);
+				pthread_rwlock_rdlock(&queue->device->bo_list.rwlock);
 				bo_list = &queue->device->bo_list.list;
 			}
 
@@ -4559,7 +4559,7 @@ radv_queue_submit_deferred(struct radv_deferred_queue_submission *submission,
 							      can_patch, base_fence);
 
 			if (unlikely(queue->device->use_global_bo_list))
-				pthread_mutex_unlock(&queue->device->bo_list.mutex);
+				pthread_rwlock_unlock(&queue->device->bo_list.rwlock);
 
 			if (result != VK_SUCCESS)
 				goto fail;

src/amd/vulkan/radv_private.h

@@ -741,7 +741,7 @@ struct radv_queue {
 struct radv_bo_list {
 	struct radv_winsys_bo_list list;
 	unsigned capacity;
-	pthread_mutex_t mutex;
+	pthread_rwlock_t rwlock;
 };
 
 VkResult radv_bo_list_add(struct radv_device *device,