[PATCH v2] drm/amdgpu: lock the eviction fence for wq signals it

[Christian König]

On 5/12/25 04:20, Prike Liang wrote:
> Lock and refer to the eviction fence before the eviction fence
> schedules work queue tries to signal it.
> 
> Suggested-by: Christian König <christian.koenig at amd.com>
> Signed-off-by: Prike Liang <Prike.Liang at amd.com>
> ---
>  drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c | 8 +++++++-
>  1 file changed, 7 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c
> index 1a7469543db5..d5f176fef357 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c
> @@ -108,12 +108,18 @@ amdgpu_eviction_fence_suspend_worker(struct work_struct *work)
>  	struct amdgpu_eviction_fence *ev_fence;
>  
>  	mutex_lock(&uq_mgr->userq_mutex);
> +	spin_lock(&evf_mgr->ev_fence_lock);
>  	ev_fence = evf_mgr->ev_fence;
> -	if (!ev_fence)
> +	spin_unlock(&evf_mgr->ev_fence_lock);

The dma_fence_get must be done while holding the lock, otherwise the change is rather pointless.

Regards,
Christian.

> +
> +	if (ev_fence)
> +		dma_fence_get(&ev_fence->base);
> +	else
>  		goto unlock;
>  
>  	amdgpu_userq_evict(uq_mgr, ev_fence);
>  
> +	dma_fence_put(&ev_fence->base);
>  unlock:
>  	mutex_unlock(&uq_mgr->userq_mutex);
>  }