On 2018å¹´03æ??05æ?¥ 12:09, Monk Liu wrote:
> bo_do_create will unreserve in in the end if @resv
> is NULL, which cause the following bo_create_shadow
> run without lock held on @resv
>
> Change-Id: Iaad24b8aea60522f25188874ab9d5c5f13f1a941
> Signed-off-by: Monk Liu <Monk.Liu at amd.com>
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 9 ++++++---
> 1 file changed, 6 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> index 216799cc..d8a818d 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> @@ -507,9 +507,12 @@ int amdgpu_bo_create(struct amdgpu_device *adev,
> return r;
>
> if ((flags & AMDGPU_GEM_CREATE_SHADOW) && amdgpu_need_backup(adev)) {
> - if (!resv)
> - WARN_ON(reservation_object_lock((*bo_ptr)->tbo.resv,
> - NULL));
> + if (!resv) {
> + WARN_ON(reservation_object_lock((*bo_ptr)->tbo.resv, NULL));
isn't this locking resv?
> + r = amdgpu_bo_reserve(*bo_ptr, kernel?true:false);
I don't think this is need.
Regards,
David Zhou
> + if (unlikely(r))
> + return r;
> + }
>
> r = amdgpu_bo_create_shadow(adev, size, byte_align, (*bo_ptr));
>
