drm/amdgpu: avoid restore process run into dead loop.

In restore process worker, pinned BO cause update PTE fail, then
the function re-schedule the restore_work. This will generate dead loop.

Signed-off-by: gaba <gaba@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

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

@@ -2881,6 +2881,9 @@ int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence **ef)
 			if (!attachment->is_mapped)
 				continue;
 
+			if (attachment->bo_va->base.bo->tbo.pin_count)
+				continue;
+
 			kfd_mem_dmaunmap_attachment(mem, attachment);
 			ret = update_gpuvm_pte(mem, attachment, &sync_obj);
 			if (ret) {