drm/amdgpu: Get rid of dep_sync as a seperate object.

Instead mark fence as explicit in it's amdgpu_sync_entry.

v2:
Fix use after free bug and add new parameter description.
Signed-off-by: Andrey Grodzovsky <[e-mail]>
Reviewed-by: Christian König <[e-mail]>
Signed-off-by: Alex Deucher <[e-mail]>

drivers/gpu/drm/amd/amdgpu/amdgpu.h

···
 	struct amdgpu_vm	*vm;
 	struct amdgpu_ring	*ring;
 	struct amdgpu_sync	sync;
-	struct amdgpu_sync	dep_sync;
 	struct amdgpu_sync	sched_sync;
 	struct amdgpu_ib	*ibs;
 	struct dma_fence	*fence; /* the hw fence */

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

···
 		return r;
 
 	r = amdgpu_sync_fence(adev, &p->job->sync,
-			      fpriv->prt_va->last_pt_update);
+			      fpriv->prt_va->last_pt_update, false);
 	if (r)
 		return r;
 
···
 			return r;
 
 		f = bo_va->last_pt_update;
-		r = amdgpu_sync_fence(adev, &p->job->sync, f);
+		r = amdgpu_sync_fence(adev, &p->job->sync, f, false);
 		if (r)
 			return r;
 	}
···
 				return r;
 
 			f = bo_va->last_pt_update;
-			r = amdgpu_sync_fence(adev, &p->job->sync, f);
+			r = amdgpu_sync_fence(adev, &p->job->sync, f, false);
 			if (r)
 				return r;
 		}
···
 	if (r)
 		return r;
 
-	r = amdgpu_sync_fence(adev, &p->job->sync, vm->last_update);
+	r = amdgpu_sync_fence(adev, &p->job->sync, vm->last_update, false);
 	if (r)
 		return r;
 
···
 			amdgpu_ctx_put(ctx);
 			return r;
 		} else if (fence) {
-			r = amdgpu_sync_fence(p->adev, &p->job->dep_sync,
-					      fence);
+			r = amdgpu_sync_fence(p->adev, &p->job->sync, fence,
+					true);
 			dma_fence_put(fence);
 			amdgpu_ctx_put(ctx);
 			if (r)
···
 	if (r)
 		return r;
 
-	r = amdgpu_sync_fence(p->adev, &p->job->dep_sync, fence);
+	r = amdgpu_sync_fence(p->adev, &p->job->sync, fence, true);
 	dma_fence_put(fence);
 
 	return r;

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

···
 	}
 
 	if (ring->funcs->emit_pipeline_sync && job &&
-	    ((tmp = amdgpu_sync_get_fence(&job->sched_sync)) ||
+	    ((tmp = amdgpu_sync_get_fence(&job->sched_sync, NULL)) ||
 	     amdgpu_vm_need_pipeline_sync(ring, job))) {
 		need_pipe_sync = true;
 		dma_fence_put(tmp);

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

···
 	(*job)->num_ibs = num_ibs;
 
 	amdgpu_sync_create(&(*job)->sync);
-	amdgpu_sync_create(&(*job)->dep_sync);
 	amdgpu_sync_create(&(*job)->sched_sync);
 	(*job)->vram_lost_counter = atomic_read(&adev->vram_lost_counter);
 
···
 	amdgpu_ring_priority_put(job->ring, s_job->s_priority);
 	dma_fence_put(job->fence);
 	amdgpu_sync_free(&job->sync);
-	amdgpu_sync_free(&job->dep_sync);
 	amdgpu_sync_free(&job->sched_sync);
 	kfree(job);
 }
···
 
 	dma_fence_put(job->fence);
 	amdgpu_sync_free(&job->sync);
-	amdgpu_sync_free(&job->dep_sync);
 	amdgpu_sync_free(&job->sched_sync);
 	kfree(job);
 }
···
 {
 	struct amdgpu_job *job = to_amdgpu_job(sched_job);
 	struct amdgpu_vm *vm = job->vm;
-
-	struct dma_fence *fence = amdgpu_sync_get_fence(&job->dep_sync);
+	bool explicit = false;
 	int r;
+	struct dma_fence *fence = amdgpu_sync_get_fence(&job->sync, &explicit);
 
-	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);
+	if (fence && explicit) {
+		if (amd_sched_dependency_optimized(fence, s_entity)) {
+			r = amdgpu_sync_fence(job->adev, &job->sched_sync, fence, false);
+			if (r)
+				DRM_ERROR("Error adding fence to sync (%d)\n", r);
+		}
 	}
-	if (!fence)
-		fence = amdgpu_sync_get_fence(&job->sync);
+
 	while (fence == NULL && vm && !job->vm_id) {
 		struct amdgpu_ring *ring = job->ring;
 
···
 		if (r)
 			DRM_ERROR("Error getting VM ID (%d)\n", r);
 
-		fence = amdgpu_sync_get_fence(&job->sync);
+		fence = amdgpu_sync_get_fence(&job->sync, NULL);
 	}
 
 	return fence;

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

···
 struct amdgpu_sync_entry {
 	struct hlist_node	node;
 	struct dma_fence	*fence;
+	bool	explicit;
 };
 
 static struct kmem_cache *amdgpu_sync_slab;
···
  *
  */
 int amdgpu_sync_fence(struct amdgpu_device *adev, struct amdgpu_sync *sync,
-		      struct dma_fence *f)
+		      struct dma_fence *f, bool explicit)
 {
 	struct amdgpu_sync_entry *e;
 
···
 	e = kmem_cache_alloc(amdgpu_sync_slab, GFP_KERNEL);
 	if (!e)
 		return -ENOMEM;
+
+	e->explicit = explicit;
 
 	hash_add(sync->fences, &e->node, f->context);
 	e->fence = dma_fence_get(f);
···
 
 	/* always sync to the exclusive fence */
 	f = reservation_object_get_excl(resv);
-	r = amdgpu_sync_fence(adev, sync, f);
+	r = amdgpu_sync_fence(adev, sync, f, false);
 
 	if (explicit_sync)
 		return r;
···
 				continue;
 		}
 
-		r = amdgpu_sync_fence(adev, sync, f);
+		r = amdgpu_sync_fence(adev, sync, f, false);
 		if (r)
 			break;
 	}
···
  * amdgpu_sync_get_fence - get the next fence from the sync object
  *
  * @sync: sync object to use
+ * @explicit: true if the next fence is explicit
  *
  * Get and removes the next fence from the sync object not signaled yet.
  */
-struct dma_fence *amdgpu_sync_get_fence(struct amdgpu_sync *sync)
+struct dma_fence *amdgpu_sync_get_fence(struct amdgpu_sync *sync, bool *explicit)
 {
 	struct amdgpu_sync_entry *e;
 	struct hlist_node *tmp;
 	struct dma_fence *f;
 	int i;
-
 	hash_for_each_safe(sync->fences, i, tmp, e, node) {
 
 		f = e->fence;
+		if (explicit)
+			*explicit = e->explicit;
 
 		hash_del(&e->node);
 		kmem_cache_free(amdgpu_sync_slab, e);

drivers/gpu/drm/amd/amdgpu/amdgpu_sync.h

···
 
 void amdgpu_sync_create(struct amdgpu_sync *sync);
 int amdgpu_sync_fence(struct amdgpu_device *adev, struct amdgpu_sync *sync,
-		      struct dma_fence *f);
+		      struct dma_fence *f, bool explicit);
 int amdgpu_sync_resv(struct amdgpu_device *adev,
 		     struct amdgpu_sync *sync,
 		     struct reservation_object *resv,
···
 		     bool explicit_sync);
 struct dma_fence *amdgpu_sync_peek_fence(struct amdgpu_sync *sync,
 				     struct amdgpu_ring *ring);
-struct dma_fence *amdgpu_sync_get_fence(struct amdgpu_sync *sync);
+struct dma_fence *amdgpu_sync_get_fence(struct amdgpu_sync *sync, bool *explicit);
 int amdgpu_sync_wait(struct amdgpu_sync *sync, bool intr);
 void amdgpu_sync_free(struct amdgpu_sync *sync);
 int amdgpu_sync_init(void);

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

···
 		id->pd_gpu_addr = 0;
 		tmp = amdgpu_sync_peek_fence(&id->active, ring);
 		if (tmp) {
-			r = amdgpu_sync_fence(adev, sync, tmp);
+			r = amdgpu_sync_fence(adev, sync, tmp, false);
 			return r;
 		}
 	}
···
 	/* Good we can use this VMID. Remember this submission as
 	* user of the VMID.
 	*/
-	r = amdgpu_sync_fence(ring->adev, &id->active, fence);
+	r = amdgpu_sync_fence(ring->adev, &id->active, fence, false);
 	if (r)
 		goto out;
 
···
 		}
 
 
-		r = amdgpu_sync_fence(ring->adev, sync, &array->base);
+		r = amdgpu_sync_fence(ring->adev, sync, &array->base, false);
 		dma_fence_put(&array->base);
 		if (r)
 			goto error;
···
 		/* Good we can use this VMID. Remember this submission as
 		 * user of the VMID.
 		 */
-		r = amdgpu_sync_fence(ring->adev, &id->active, fence);
+		r = amdgpu_sync_fence(ring->adev, &id->active, fence, false);
 		if (r)
 			goto error;
 
···
 	id = idle;
 
 	/* Remember this submission as user of the VMID */
-	r = amdgpu_sync_fence(ring->adev, &id->active, fence);
+	r = amdgpu_sync_fence(ring->adev, &id->active, fence, false);
 	if (r)
 		goto error;
 
···
 		addr = 0;
 	}
 
-	r = amdgpu_sync_fence(adev, &job->sync, exclusive);
+	r = amdgpu_sync_fence(adev, &job->sync, exclusive, false);
 	if (r)
 		goto error_free;