drm/sched: Set error to s_fence if HW job submission failed.
authorAndrey Grodzovsky <andrey.grodzovsky@amd.com>
Thu, 24 Oct 2019 19:39:06 +0000 (15:39 -0400)
committerAlex Deucher <alexander.deucher@amd.com>
Fri, 25 Oct 2019 20:50:11 +0000 (16:50 -0400)
Problem:
When run_job fails and HW fence returned is NULL we still signal
the s_fence to avoid hangs but the user has no way of knowing if
the actual HW job was ran and finished.

Fix:
Allow .run_job implementations to return ERR_PTR in the fence pointer
returned and then set this error for s_fence->finished fence so whoever
wait on this fence can inspect the signaled fence for an error.

Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drivers/gpu/drm/scheduler/sched_main.c

index 9a0ee74d82dc2b6c39859781c44e4e4be615cd1e..f39b97ed4ade4db6730fb5a31029c16701fd1d7d 100644 (file)
@@ -479,6 +479,7 @@ void drm_sched_resubmit_jobs(struct drm_gpu_scheduler *sched)
        struct drm_sched_job *s_job, *tmp;
        uint64_t guilty_context;
        bool found_guilty = false;
+       struct dma_fence *fence;
 
        list_for_each_entry_safe(s_job, tmp, &sched->ring_mirror_list, node) {
                struct drm_sched_fence *s_fence = s_job->s_fence;
@@ -492,7 +493,16 @@ void drm_sched_resubmit_jobs(struct drm_gpu_scheduler *sched)
                        dma_fence_set_error(&s_fence->finished, -ECANCELED);
 
                dma_fence_put(s_job->s_fence->parent);
-               s_job->s_fence->parent = sched->ops->run_job(s_job);
+               fence = sched->ops->run_job(s_job);
+
+               if (IS_ERR_OR_NULL(fence)) {
+                       s_job->s_fence->parent = NULL;
+                       dma_fence_set_error(&s_fence->finished, PTR_ERR(fence));
+               } else {
+                       s_job->s_fence->parent = fence;
+               }
+
+
        }
 }
 EXPORT_SYMBOL(drm_sched_resubmit_jobs);
@@ -720,7 +730,7 @@ static int drm_sched_main(void *param)
                fence = sched->ops->run_job(sched_job);
                drm_sched_fence_scheduled(s_fence);
 
-               if (fence) {
+               if (!IS_ERR_OR_NULL(fence)) {
                        s_fence->parent = dma_fence_get(fence);
                        r = dma_fence_add_callback(fence, &sched_job->cb,
                                                   drm_sched_process_job);
@@ -730,8 +740,11 @@ static int drm_sched_main(void *param)
                                DRM_ERROR("fence add callback failed (%d)\n",
                                          r);
                        dma_fence_put(fence);
-               } else
+               } else {
+
+                       dma_fence_set_error(&s_fence->finished, PTR_ERR(fence));
                        drm_sched_process_job(NULL, &sched_job->cb);
+               }
 
                wake_up(&sched->job_scheduled);
        }