drm/amdgpu: re-validate per VM BOs if required v2
authorChristian König <christian.koenig@amd.com>
Mon, 19 Mar 2018 10:49:14 +0000 (11:49 +0100)
committerAlex Deucher <alexander.deucher@amd.com>
Tue, 15 May 2018 18:43:20 +0000 (13:43 -0500)
If a per VM BO ends up in a allowed domain it never moves back into the
prefered domain.

v2: move the extra handling into amdgpu_vm_bo_update when we exit the
    state machine. Make memory type handling generic.

Signed-off-by: Christian König <christian.koenig@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_vm.c

index da55a78..f0fbc33 100644 (file)
@@ -1556,7 +1556,20 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev,
        }
 
        spin_lock(&vm->status_lock);
-       list_del_init(&bo_va->base.vm_status);
+       if (bo && bo->tbo.resv == vm->root.base.bo->tbo.resv) {
+               unsigned mem_type = bo->tbo.mem.mem_type;
+
+               /* If the BO is not in its preferred location add it back to
+                * the evicted list so that it gets validated again on the
+                * next command submission.
+                */
+               if (!(bo->preferred_domains & amdgpu_mem_type_to_domain(mem_type)))
+                       list_add_tail(&bo_va->base.vm_status, &vm->evicted);
+               else
+                       list_del_init(&bo_va->base.vm_status);
+       } else {
+               list_del_init(&bo_va->base.vm_status);
+       }
        spin_unlock(&vm->status_lock);
 
        list_splice_init(&bo_va->invalids, &bo_va->valids);