[PATCH] drm/radeon: delete radeon_fence_process in is_signaled, no deadlock

[PATCH] drm/radeon: delete radeon_fence_process in is_signaled, no deadlock

[Robert McClinton]

Delete the attempt to progress the queue when checking if fence is
signaled. This avoids deadlock.

dma-fence_ops::signaled can be called with the fence lock in unknown
state. For radeon, the fence lock is also the wait queue lock. This can
cause a self deadlock when signaled() tries to make forward progress on
the wait queue. But advancing the queue is unneeded because incorrectly
returning false from signaled() is perfectly acceptable.

Link: https://github.com/brave/brave-browser/issues/49182

Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4641

Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
Signed-off-by: Robert McClinton <rbmccav@gmail.com>
---
 drivers/gpu/drm/radeon/radeon_fence.c | 7 -------
 1 file changed, 7 deletions(-)

diff --git a/drivers/gpu/drm/radeon/radeon_fence.c b/drivers/gpu/drm/radeon/radeon_fence.c
index 5b5b54e876d4..167d6f122b8e 100644
--- a/drivers/gpu/drm/radeon/radeon_fence.c
+++ b/drivers/gpu/drm/radeon/radeon_fence.c
@@ -360,13 +360,6 @@ static bool radeon_fence_is_signaled(struct dma_fence *f)
 	if (atomic64_read(&rdev->fence_drv[ring].last_seq) >= seq)
 		return true;
 
-	if (down_read_trylock(&rdev->exclusive_lock)) {
-		radeon_fence_process(rdev, ring);
-		up_read(&rdev->exclusive_lock);
-
-		if (atomic64_read(&rdev->fence_drv[ring].last_seq) >= seq)
-			return true;
-	}
 	return false;
 }
 
-- 
2.51.2

Re: [PATCH] drm/radeon: delete radeon_fence_process in is_signaled, no deadlock

[Alex Deucher]

Applied.  Thanks!

Alex

On Mon, Nov 17, 2025 at 3:42 AM Robert McClinton <rbmccav@gmail.com> wrote:
>
> Delete the attempt to progress the queue when checking if fence is
> signaled. This avoids deadlock.
>
> dma-fence_ops::signaled can be called with the fence lock in unknown
> state. For radeon, the fence lock is also the wait queue lock. This can
> cause a self deadlock when signaled() tries to make forward progress on
> the wait queue. But advancing the queue is unneeded because incorrectly
> returning false from signaled() is perfectly acceptable.
>
> Link: https://github.com/brave/brave-browser/issues/49182
>
> Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4641
>
> Cc: Alex Deucher <alexander.deucher@amd.com>
> Cc: stable@vger.kernel.org
> Signed-off-by: Robert McClinton <rbmccav@gmail.com>
> ---
>  drivers/gpu/drm/radeon/radeon_fence.c | 7 -------
>  1 file changed, 7 deletions(-)
>
> diff --git a/drivers/gpu/drm/radeon/radeon_fence.c b/drivers/gpu/drm/radeon/radeon_fence.c
> index 5b5b54e876d4..167d6f122b8e 100644
> --- a/drivers/gpu/drm/radeon/radeon_fence.c
> +++ b/drivers/gpu/drm/radeon/radeon_fence.c
> @@ -360,13 +360,6 @@ static bool radeon_fence_is_signaled(struct dma_fence *f)
>         if (atomic64_read(&rdev->fence_drv[ring].last_seq) >= seq)
>                 return true;
>
> -       if (down_read_trylock(&rdev->exclusive_lock)) {
> -               radeon_fence_process(rdev, ring);
> -               up_read(&rdev->exclusive_lock);
> -
> -               if (atomic64_read(&rdev->fence_drv[ring].last_seq) >= seq)
> -                       return true;
> -       }
>         return false;
>  }
>
> --
> 2.51.2
>