drm/nouveau/mmu: flush tlbs before deleting page tables

Even though we've zeroed the PDE, the GPU may have cached the PD, so we
need to flush when deleting them.

Noticed while working on replacement MMU code, but a backport might be a
good idea, so let's fix it in the current code too.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Cc: stable@vger.kernel.org

diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c
index d06ad2c372bf30efb6b8ecc5978776def5721222..455da298227f65c2b4c2cfc6a2cedebe12661877 100644 (file)
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c
@@ -241,6 +241,8 @@ nvkm_vm_unmap_pgt(struct nvkm_vm *vm, int big, u32 fpde, u32 lpde)
                        mmu->func->map_pgt(vpgd->obj, pde, vpgt->mem);
                }
 
+               mmu->func->flush(vm);
+
                nvkm_memory_del(&pgt);
        }
 }