drm/amdgpu: Remove job->s_entity to avoid keeping reference to stale pointer.

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/amd/amdgpu/amdgpu_cs.c

@@ -1203,7 +1203,7 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 	amdgpu_ring_priority_get(job->ring, job->base.s_priority);
 
 	trace_amdgpu_cs_ioctl(job);
-	amd_sched_entity_push_job(&job->base);
+	amd_sched_entity_push_job(&job->base, entity);
 
 	ttm_eu_fence_buffer_objects(&p->ticket, &p->validated, p->fence);
 	amdgpu_mn_unlock(p->mn);

drivers/gpu/drm/amd/amdgpu/amdgpu_job.c

@@ -142,12 +142,13 @@ int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring,
 	*f = dma_fence_get(&job->base.s_fence->finished);
 	amdgpu_job_free_resources(job);
 	amdgpu_ring_priority_get(job->ring, job->base.s_priority);
-	amd_sched_entity_push_job(&job->base);
+	amd_sched_entity_push_job(&job->base, entity);
 
 	return 0;
 }
 
-static struct dma_fence *amdgpu_job_dependency(struct amd_sched_job *sched_job)
+static struct dma_fence *amdgpu_job_dependency(struct amd_sched_job *sched_job,
+					       struct amd_sched_entity *s_entity)
 {
 	struct amdgpu_job *job = to_amdgpu_job(sched_job);
 	struct amdgpu_vm *vm = job->vm;
@@ -155,7 +156,7 @@ static struct dma_fence *amdgpu_job_dependency(struct amd_sched_job *sched_job)
 	struct dma_fence *fence = amdgpu_sync_get_fence(&job->dep_sync);
 	int r;
 
-	if (amd_sched_dependency_optimized(fence, sched_job->s_entity)) {
+	if (amd_sched_dependency_optimized(fence, s_entity)) {
 		r = amdgpu_sync_fence(job->adev, &job->sched_sync, fence);
 		if (r)
 			DRM_ERROR("Error adding fence to sync (%d)\n", r);

drivers/gpu/drm/amd/scheduler/gpu_sched_trace.h

@@ -13,8 +13,8 @@
 #define TRACE_INCLUDE_FILE gpu_sched_trace
 
 TRACE_EVENT(amd_sched_job,
-	    TP_PROTO(struct amd_sched_job *sched_job),
-	    TP_ARGS(sched_job),
+	    TP_PROTO(struct amd_sched_job *sched_job, struct amd_sched_entity *entity),
+	    TP_ARGS(sched_job, entity),
 	    TP_STRUCT__entry(
 			     __field(struct amd_sched_entity *, entity)
 			     __field(struct dma_fence *, fence)
@@ -25,12 +25,11 @@ TRACE_EVENT(amd_sched_job,
 			     ),
 
 	    TP_fast_assign(
-			   __entry->entity = sched_job->s_entity;
+			   __entry->entity = entity;
 			   __entry->id = sched_job->id;
 			   __entry->fence = &sched_job->s_fence->finished;
 			   __entry->name = sched_job->sched->name;
-			   __entry->job_count = spsc_queue_count(
-				   &sched_job->s_entity->job_queue);
+			   __entry->job_count = spsc_queue_count(&entity->job_queue);
 			   __entry->hw_job_count = atomic_read(
 				   &sched_job->sched->hw_rq_count);
 			   ),

drivers/gpu/drm/amd/scheduler/gpu_scheduler.c

@@ -341,11 +341,10 @@ amd_sched_entity_pop_job(struct amd_sched_entity *entity)
 	if (!sched_job)
 		return NULL;
 
-	while ((entity->dependency = sched->ops->dependency(sched_job)))
+	while ((entity->dependency = sched->ops->dependency(sched_job, entity)))
 		if (amd_sched_entity_add_dependency_cb(entity))
 			return NULL;
 
-	sched_job->s_entity = NULL;
 	spsc_queue_pop(&entity->job_queue);
 	return sched_job;
 }
@@ -357,13 +356,13 @@ amd_sched_entity_pop_job(struct amd_sched_entity *entity)
  *
  * Returns 0 for success, negative error code otherwise.
  */
-void amd_sched_entity_push_job(struct amd_sched_job *sched_job)
+void amd_sched_entity_push_job(struct amd_sched_job *sched_job,
+			       struct amd_sched_entity *entity)
 {
 	struct amd_gpu_scheduler *sched = sched_job->sched;
-	struct amd_sched_entity *entity = sched_job->s_entity;
 	bool first = false;
 
-	trace_amd_sched_job(sched_job);
+	trace_amd_sched_job(sched_job, entity);
 
 	spin_lock(&entity->queue_lock);
 	first = spsc_queue_push(&entity->job_queue, &sched_job->queue_node);
@@ -442,11 +441,12 @@ static void amd_sched_job_timedout(struct work_struct *work)
 	job->sched->ops->timedout_job(job);
 }
 
-static void amd_sched_set_guilty(struct amd_sched_job *s_job)
+static void amd_sched_set_guilty(struct amd_sched_job *s_job,
+				 struct amd_sched_entity *s_entity)
 {
 	if (atomic_inc_return(&s_job->karma) > s_job->sched->hang_limit)
-		if (s_job->s_entity->guilty)
-			atomic_set(s_job->s_entity->guilty, 1);
+		if (s_entity->guilty)
+			atomic_set(s_entity->guilty, 1);
 }
 
 void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched, struct amd_sched_job *bad)
@@ -477,7 +477,7 @@ void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched, struct amd_sched_jo
 			list_for_each_entry_safe(entity, tmp, &rq->entities, list) {
 				if (bad->s_fence->scheduled.context == entity->fence_context) {
 					found = true;
-					amd_sched_set_guilty(bad);
+					amd_sched_set_guilty(bad, entity);
 					break;
 				}
 			}
@@ -541,7 +541,6 @@ int amd_sched_job_init(struct amd_sched_job *job,
 		       void *owner)
 {
 	job->sched = sched;
-	job->s_entity = entity;
 	job->s_priority = entity->rq - sched->sched_rq;
 	job->s_fence = amd_sched_fence_create(entity, owner);
 	if (!job->s_fence)

drivers/gpu/drm/amd/scheduler/gpu_scheduler.h

@@ -91,7 +91,6 @@ struct amd_sched_fence {
 struct amd_sched_job {
 	struct spsc_node queue_node;
 	struct amd_gpu_scheduler        *sched;
-	struct amd_sched_entity         *s_entity;
 	struct amd_sched_fence          *s_fence;
 	struct dma_fence_cb		finish_cb;
 	struct work_struct		finish_work;
@@ -125,7 +124,8 @@ static inline bool amd_sched_invalidate_job(struct amd_sched_job *s_job, int thr
  * these functions should be implemented in driver side
 */
 struct amd_sched_backend_ops {
-	struct dma_fence *(*dependency)(struct amd_sched_job *sched_job);
+	struct dma_fence *(*dependency)(struct amd_sched_job *sched_job,
+					struct amd_sched_entity *s_entity);
 	struct dma_fence *(*run_job)(struct amd_sched_job *sched_job);
 	void (*timedout_job)(struct amd_sched_job *sched_job);
 	void (*free_job)(struct amd_sched_job *sched_job);
@@ -161,7 +161,8 @@ int amd_sched_entity_init(struct amd_gpu_scheduler *sched,
 			  uint32_t jobs, atomic_t* guilty);
 void amd_sched_entity_fini(struct amd_gpu_scheduler *sched,
 			   struct amd_sched_entity *entity);
-void amd_sched_entity_push_job(struct amd_sched_job *sched_job);
+void amd_sched_entity_push_job(struct amd_sched_job *sched_job,
+			       struct amd_sched_entity *entity);
 void amd_sched_entity_set_rq(struct amd_sched_entity *entity,
 			     struct amd_sched_rq *rq);