On 9/5/23 10:20, Boris Brezillon wrote:
> On Sun, 3 Sep 2023 20:07:22 +0300
> Dmitry Osipenko <[email protected]> wrote:
>
>> Prepare virtio_gpu_object_create() to addition of memory shrinker support
>> by replacing open-coded drm_gem_shmem_free() with drm_gem_object_put() that
>> decrements GEM refcount to 0, which becomes important for drm-shmem because
>> it will start to use GEM's refcount during the shmem's BO freeing time in
>> order to prevent spurious lockdep warning about resv lock ordering vs
>> fs_reclaim code paths.
>
> I think I'm okay with the change (assuming virtio_gpu_free_object()
> can deal with partially initialized objects), not with the explanation
> :-). I don't really see why we need to take the resv lock in
> drm_gem_shmem_free(). As said in my v15 review, I think we should
> replace the drm_gem_shmem_put_pages() call we have in
> drm_gem_shmem_free() by a call to a new drm_gem_shmem_free_pages()
> helper that does exactly what drm_gem_shmem_put_pages() does without
> the refcounting/locking, because all that should remain at the time
> drm_gem_shmem_free() is called is the implicit pages ref owned by
> shmem->sgt, and there's no risk of other threads accessing the GEM
> object at that point.
Apparently I forgot to drop these drm_gem_object_put() patches by
accident in v16, the drm_gem_shmem_free() doesn't touch resv lock
anymore. Will re-check for v17.
--
Best regards,
Dmitry