And only move them in on validation. This allows for better control
when multiple processes are fighting over those resources.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Tested-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
 #endif
 
        bo->tbo.bdev = &adev->mman.bdev;
-       amdgpu_bo_placement_from_domain(bo, bp->domain);
+       if (bp->domain & (AMDGPU_GEM_DOMAIN_GWS | AMDGPU_GEM_DOMAIN_OA |
+                         AMDGPU_GEM_DOMAIN_GDS))
+               amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_CPU);
+       else
+               amdgpu_bo_placement_from_domain(bo, bp->domain);
        if (bp->type == ttm_bo_type_kernel)
                bo->tbo.priority = 1;