* [PATCH] drm/amdgpu: amdgpu_fence: remove ops->release()
@ 2026-10-27 13:37 Philipp Stanner
2026-10-27 13:40 ` Christian König
0 siblings, 1 reply; 2+ messages in thread
From: Philipp Stanner @ 2026-10-27 13:37 UTC (permalink / raw)
To: Alex Deucher, Christian König, David Airlie, Simona Vetter,
Sumit Semwal, Jesse Zhang, Timur Kristóf, Kexin Sun,
Philipp Stanner, Srinivasan Shanmugam, Sakari Ailus
Cc: amd-gfx, dri-devel, linux-kernel, linux-media
amdgpu's fence implements the deprecated fence_ops->release() callback
which prevents the driver from ever unloading robustly.
amdgpu's implementation merely uses a call_rcu() that performs a kfree()
on the fence - this functionality is already centrally implemented by
the dma_fence backend.
The dma_fence backend invokes a kfree_rcu() on the dma_fence, which is
fine since this is always the first member of a struct amdgpu_fence.
Thus, amdgpu's implementation is redundant.
Remove fence_ops->release().
Signed-off-by: Philipp Stanner <phasta@kernel.org>
---
Just saw that this should be unnecessary? Don't have a setup for amdgpu
testing right now, so please take a close look, just compile-tested.
P.
---
drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 29 -----------------------
1 file changed, 29 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
index 3043ad041bb4..3f8b2571d74a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
@@ -892,39 +892,10 @@ static bool amdgpu_fence_enable_signaling(struct dma_fence *f)
return true;
}
-/**
- * amdgpu_fence_free - free up the fence memory
- *
- * @rcu: RCU callback head
- *
- * Free up the fence memory after the RCU grace period.
- */
-static void amdgpu_fence_free(struct rcu_head *rcu)
-{
- struct dma_fence *f = container_of(rcu, struct dma_fence, rcu);
-
- /* free fence_slab if it's separated fence*/
- kfree(to_amdgpu_fence(f));
-}
-
-/**
- * amdgpu_fence_release - callback that fence can be freed
- *
- * @f: fence
- *
- * This function is called when the reference count becomes zero.
- * It just RCU schedules freeing up the fence.
- */
-static void amdgpu_fence_release(struct dma_fence *f)
-{
- call_rcu(&f->rcu, amdgpu_fence_free);
-}
-
static const struct dma_fence_ops amdgpu_fence_ops = {
.get_driver_name = amdgpu_fence_get_driver_name,
.get_timeline_name = amdgpu_fence_get_timeline_name,
.enable_signaling = amdgpu_fence_enable_signaling,
- .release = amdgpu_fence_release,
};
/*
base-commit: 731ba09867520c047c7de53f92e384808e0b638e
--
2.56.0
^ permalink raw reply [flat|nested] 2+ messages in thread* Re: [PATCH] drm/amdgpu: amdgpu_fence: remove ops->release()
2026-10-27 13:37 [PATCH] drm/amdgpu: amdgpu_fence: remove ops->release() Philipp Stanner
@ 2026-10-27 13:40 ` Christian König
0 siblings, 0 replies; 2+ messages in thread
From: Christian König @ 2026-10-27 13:40 UTC (permalink / raw)
To: Philipp Stanner, Alex Deucher, David Airlie, Simona Vetter,
Sumit Semwal, Jesse Zhang, Timur Kristóf, Kexin Sun,
Srinivasan Shanmugam, Sakari Ailus
Cc: amd-gfx, dri-devel, linux-kernel, linux-media
On 10/27/26 14:37, Philipp Stanner wrote:
> amdgpu's fence implements the deprecated fence_ops->release() callback
> which prevents the driver from ever unloading robustly.
>
> amdgpu's implementation merely uses a call_rcu() that performs a kfree()
> on the fence - this functionality is already centrally implemented by
> the dma_fence backend.
>
> The dma_fence backend invokes a kfree_rcu() on the dma_fence, which is
> fine since this is always the first member of a struct amdgpu_fence.
> Thus, amdgpu's implementation is redundant.
>
> Remove fence_ops->release().
>
> Signed-off-by: Philipp Stanner <phasta@kernel.org>
I have exactly the same patch in my pipeline as well.
Reviewed-by: Christian König <christian.koenig@amd.com>
> ---
> Just saw that this should be unnecessary? Don't have a setup for amdgpu
> testing right now, so please take a close look, just compile-tested.
>
> P.
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 29 -----------------------
> 1 file changed, 29 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
> index 3043ad041bb4..3f8b2571d74a 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
> @@ -892,39 +892,10 @@ static bool amdgpu_fence_enable_signaling(struct dma_fence *f)
> return true;
> }
>
> -/**
> - * amdgpu_fence_free - free up the fence memory
> - *
> - * @rcu: RCU callback head
> - *
> - * Free up the fence memory after the RCU grace period.
> - */
> -static void amdgpu_fence_free(struct rcu_head *rcu)
> -{
> - struct dma_fence *f = container_of(rcu, struct dma_fence, rcu);
> -
> - /* free fence_slab if it's separated fence*/
> - kfree(to_amdgpu_fence(f));
> -}
> -
> -/**
> - * amdgpu_fence_release - callback that fence can be freed
> - *
> - * @f: fence
> - *
> - * This function is called when the reference count becomes zero.
> - * It just RCU schedules freeing up the fence.
> - */
> -static void amdgpu_fence_release(struct dma_fence *f)
> -{
> - call_rcu(&f->rcu, amdgpu_fence_free);
> -}
> -
> static const struct dma_fence_ops amdgpu_fence_ops = {
> .get_driver_name = amdgpu_fence_get_driver_name,
> .get_timeline_name = amdgpu_fence_get_timeline_name,
> .enable_signaling = amdgpu_fence_enable_signaling,
> - .release = amdgpu_fence_release,
> };
>
> /*
>
> base-commit: 731ba09867520c047c7de53f92e384808e0b638e
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2026-10-27 13:41 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-10-27 13:37 [PATCH] drm/amdgpu: amdgpu_fence: remove ops->release() Philipp Stanner
2026-10-27 13:40 ` Christian König
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®