drm/amdgpu: fix seq in ctx_add_fence

if enabling scheduler, then the queued seq is assigned
when pushing job before emitting job.

Signed-off-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Christian K?nig <christian.koenig@amd.com>

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

@@ -419,7 +419,6 @@ struct amdgpu_user_fence {
 	struct amdgpu_bo 	*bo;
 	/* write-back address offset to bo start */
 	uint32_t                offset;
-	uint64_t                sequence;
 };
 
 int amdgpu_fence_driver_init(struct amdgpu_device *adev);
@@ -1031,7 +1030,7 @@ struct amdgpu_ctx *amdgpu_ctx_get(struct amdgpu_fpriv *fpriv, uint32_t id);
 int amdgpu_ctx_put(struct amdgpu_ctx *ctx);
 
 uint64_t amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx, struct amdgpu_ring *ring,
-			      struct fence *fence);
+			      struct fence *fence, uint64_t queued_seq);
 struct fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx,
 				   struct amdgpu_ring *ring, uint64_t seq);
 

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

@@ -739,7 +739,6 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
 			ib->oa_size = amdgpu_bo_size(oa);
 		}
 	}
-
 	/* wrap the last IB with user fence */
 	if (parser->uf.bo) {
 		struct amdgpu_ib *ib = &parser->ibs[parser->num_ibs - 1];
@@ -908,7 +907,7 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 	if (amdgpu_enable_scheduler && parser->num_ibs) {
 		struct amdgpu_ring * ring =
 			amdgpu_cs_parser_get_ring(adev, parser);
-		parser->uf.sequence = atomic64_inc_return(
+		parser->ibs[parser->num_ibs - 1].sequence = atomic64_inc_return(
 			&parser->ctx->rings[ring->idx].c_entity.last_queued_v_seq);
 		if (ring->is_pte_ring || (parser->bo_list && parser->bo_list->has_userptr)) {
 			r = amdgpu_cs_parser_prepare_job(parser);
@@ -922,7 +921,7 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 		amd_sched_push_job(ring->scheduler,
 				   &parser->ctx->rings[ring->idx].c_entity,
 				   parser);
-		cs->out.handle = parser->uf.sequence;
+		cs->out.handle = parser->ibs[parser->num_ibs - 1].sequence;
 		up_read(&adev->exclusive_lock);
 		return 0;
 	}

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

@@ -258,7 +258,7 @@ int amdgpu_ctx_put(struct amdgpu_ctx *ctx)
 }
 
 uint64_t amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx, struct amdgpu_ring *ring,
-			      struct fence *fence)
+			      struct fence *fence, uint64_t queued_seq)
 {
 	struct amdgpu_ctx_ring *cring = & ctx->rings[ring->idx];
 	uint64_t seq = 0;
@@ -266,7 +266,7 @@ uint64_t amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx, struct amdgpu_ring *ring,
 	struct fence *other = NULL;
 
 	if (amdgpu_enable_scheduler)
-		seq = atomic64_read(&cring->c_entity.last_queued_v_seq);
+		seq = queued_seq;
 	else
 		seq = cring->sequence;
 	idx = seq % AMDGPU_CTX_MAX_CS_PENDING;

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

@@ -143,6 +143,7 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs,
 	struct amdgpu_ring *ring;
 	struct amdgpu_ctx *ctx, *old_ctx;
 	struct amdgpu_vm *vm;
+	uint64_t sequence;
 	unsigned i;
 	int r = 0;
 
@@ -215,9 +216,12 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs,
 		return r;
 	}
 
+	sequence = amdgpu_enable_scheduler ? ib->sequence : 0;
+
 	if (ib->ctx)
 		ib->sequence = amdgpu_ctx_add_fence(ib->ctx, ring,
-						    &ib->fence->base);
+						    &ib->fence->base,
+						    sequence);
 
 	/* wrap the last IB with fence */
 	if (ib->user) {

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

@@ -62,7 +62,7 @@ static void amdgpu_sched_run_job(struct amd_gpu_scheduler *sched,
 			goto err;
 	}
 	atomic64_set(&c_entity->last_emitted_v_seq,
-		     sched_job->uf.sequence);
+		     sched_job->ibs[sched_job->num_ibs - 1].sequence);
 	wake_up_all(&c_entity->wait_emit);
 
 	mutex_unlock(&sched_job->job_lock);
@@ -93,7 +93,7 @@ static void amdgpu_sched_process_job(struct amd_gpu_scheduler *sched, void *job)
 	if (sched_job->ctx) {
 		c_entity = &sched_job->ctx->rings[ring->idx].c_entity;
 		atomic64_set(&c_entity->last_signaled_v_seq,
-			     sched_job->uf.sequence);
+			     sched_job->ibs[sched_job->num_ibs - 1].sequence);
 	}
 
 	/* wake up users waiting for time stamp */

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

@@ -380,7 +380,7 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev,
 		sched_job->run_job = amdgpu_vm_run_job;
 		sched_job->free_job = amdgpu_vm_free_job;
 		v_seq = atomic64_inc_return(&adev->kernel_ctx->rings[ring->idx].c_entity.last_queued_v_seq);
-		sched_job->uf.sequence = v_seq;
+		ib->sequence = v_seq;
 		amd_sched_push_job(ring->scheduler,
 				   &adev->kernel_ctx->rings[ring->idx].c_entity,
 				   sched_job);
@@ -531,7 +531,7 @@ int amdgpu_vm_update_page_directory(struct amdgpu_device *adev,
 			sched_job->run_job = amdgpu_vm_run_job;
 			sched_job->free_job = amdgpu_vm_free_job;
 			v_seq = atomic64_inc_return(&adev->kernel_ctx->rings[ring->idx].c_entity.last_queued_v_seq);
-			sched_job->uf.sequence = v_seq;
+			ib->sequence = v_seq;
 			amd_sched_push_job(ring->scheduler,
 					   &adev->kernel_ctx->rings[ring->idx].c_entity,
 					   sched_job);
@@ -884,7 +884,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
 		sched_job->run_job = amdgpu_vm_bo_update_mapping_run_job;
 		sched_job->free_job = amdgpu_vm_free_job;
 		v_seq = atomic64_inc_return(&adev->kernel_ctx->rings[ring->idx].c_entity.last_queued_v_seq);
-		sched_job->uf.sequence = v_seq;
+		ib->sequence = v_seq;
 		amd_sched_push_job(ring->scheduler,
 				   &adev->kernel_ctx->rings[ring->idx].c_entity,
 				   sched_job);