On 11/13/26 00:37, Erhan Bilgili wrote:
> If a fallback wait is interrupted, USERQ_WAIT leaves the unwrap loop
> without releasing the reference taken by dma_fence_unwrap_first().
> That reference is normally dropped by dma_fence_chain_walk().
>
> Drop iter.chain before returning the error.
>
> Fixes: 15e30a6e4792 ("drm/amdgpu: Add wait IOCTL timeline syncobj support")
> Signed-off-by: Erhan Bilgili <[email protected]>
Good catch, but we should probably fix dma_fence_unwrap() instead and add a
cleanup.
I will take care of that.
Regards,
Christian.
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c
> index dd976a50e053..e7b63c227421 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c
> @@ -796,6 +796,7 @@ amdgpu_userq_wait_return_fence_info(struct drm_device
> *dev, struct drm_file *fil
> r = amdgpu_userq_wait_add_fence(wait_info, fences,
> &num_fences, f);
> if (r) {
> + dma_fence_put(iter.chain);
> dma_fence_put(fence);
> goto free_fences;
> }