drm/amdgpu: add a fence after the VM flush

This way we can track when the flush is done.
Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

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

@@ -880,6 +880,7 @@ struct amdgpu_vm_id {
 	struct list_head	list;
 	struct fence		*first;
 	struct amdgpu_sync	active;
+	struct fence		*last_flush;
 	atomic_long_t		owner;
 
 	uint64_t		pd_gpu_addr;
@@ -926,11 +927,11 @@ void amdgpu_vm_move_pt_bos_in_lru(struct amdgpu_device *adev,
 int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
 		      struct amdgpu_sync *sync, struct fence *fence,
 		      unsigned *vm_id, uint64_t *vm_pd_addr);
-void amdgpu_vm_flush(struct amdgpu_ring *ring,
-		     unsigned vm_id, uint64_t pd_addr,
-		     uint32_t gds_base, uint32_t gds_size,
-		     uint32_t gws_base, uint32_t gws_size,
-		     uint32_t oa_base, uint32_t oa_size);
+int amdgpu_vm_flush(struct amdgpu_ring *ring,
+		    unsigned vm_id, uint64_t pd_addr,
+		    uint32_t gds_base, uint32_t gds_size,
+		    uint32_t gws_base, uint32_t gws_size,
+		    uint32_t oa_base, uint32_t oa_size);
 void amdgpu_vm_reset_id(struct amdgpu_device *adev, unsigned vm_id);
 uint64_t amdgpu_vm_map_gart(const dma_addr_t *pages_addr, uint64_t addr);
 int amdgpu_vm_update_page_directory(struct amdgpu_device *adev,

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

@@ -155,10 +155,14 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
 
 	if (vm) {
 		/* do context switch */
-		amdgpu_vm_flush(ring, ib->vm_id, ib->vm_pd_addr,
-				ib->gds_base, ib->gds_size,
-				ib->gws_base, ib->gws_size,
-				ib->oa_base, ib->oa_size);
+		r = amdgpu_vm_flush(ring, ib->vm_id, ib->vm_pd_addr,
+				    ib->gds_base, ib->gds_size,
+				    ib->gws_base, ib->gws_size,
+				    ib->oa_base, ib->oa_size);
+		if (r) {
+			amdgpu_ring_undo(ring);
+			return r;
+		}
 
 		if (ring->funcs->emit_hdp_flush)
 			amdgpu_ring_emit_hdp_flush(ring);

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

@@ -236,6 +236,9 @@ int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
 	fence_put(id->first);
 	id->first = fence_get(fence);
 
+	fence_put(id->last_flush);
+	id->last_flush = NULL;
+
 	fence_put(id->flushed_updates);
 	id->flushed_updates = fence_get(updates);
 
@@ -263,11 +266,11 @@ int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
  *
  * Emit a VM flush when it is necessary.
  */
-void amdgpu_vm_flush(struct amdgpu_ring *ring,
-		     unsigned vm_id, uint64_t pd_addr,
-		     uint32_t gds_base, uint32_t gds_size,
-		     uint32_t gws_base, uint32_t gws_size,
-		     uint32_t oa_base, uint32_t oa_size)
+int amdgpu_vm_flush(struct amdgpu_ring *ring,
+		    unsigned vm_id, uint64_t pd_addr,
+		    uint32_t gds_base, uint32_t gds_size,
+		    uint32_t gws_base, uint32_t gws_size,
+		    uint32_t oa_base, uint32_t oa_size)
 {
 	struct amdgpu_device *adev = ring->adev;
 	struct amdgpu_vm_id *id = &adev->vm_manager.ids[vm_id];
@@ -278,14 +281,25 @@ void amdgpu_vm_flush(struct amdgpu_ring *ring,
 		id->gws_size != gws_size ||
 		id->oa_base != oa_base ||
 		id->oa_size != oa_size);
+	int r;
 
 	if (ring->funcs->emit_pipeline_sync && (
 	    pd_addr != AMDGPU_VM_NO_FLUSH || gds_switch_needed))
 		amdgpu_ring_emit_pipeline_sync(ring);
 
 	if (pd_addr != AMDGPU_VM_NO_FLUSH) {
+		struct fence *fence;
+
 		trace_amdgpu_vm_flush(pd_addr, ring->idx, vm_id);
 		amdgpu_ring_emit_vm_flush(ring, vm_id, pd_addr);
+		r = amdgpu_fence_emit(ring, &fence);
+		if (r)
+			return r;
+
+		mutex_lock(&adev->vm_manager.lock);
+		fence_put(id->last_flush);
+		id->last_flush = fence;
+		mutex_unlock(&adev->vm_manager.lock);
 	}
 
 	if (gds_switch_needed) {
@@ -300,6 +314,8 @@ void amdgpu_vm_flush(struct amdgpu_ring *ring,
 					    gws_base, gws_size,
 					    oa_base, oa_size);
 	}
+
+	return 0;
 }
 
 /**