[PATCH v2] rust: sync: implement Unpin for ARef

[PATCH v2] rust: sync: implement Unpin for ARef

[Alice Ryhl]

The default implementation of Unpin for ARef<T> is conditional on T
being Unpin due to its PhantomData<T> field. However, this is overly
strict as pointers to T are legal to move even if T itself cannot move.

Since commit 66f1ea83d9f8 ("rust: lock: Add a Pin<&mut T> accessor")
this causes build failures when combined with a Mutex that contains an
field ARef<T>, because almost any type that ARef is used with is !Unpin.

Reviewed-by: Daniel Almeida <daniel.almeida@collabora.com>
Signed-off-by: Alice Ryhl <aliceryhl@google.com>
---
Changes in v2:
- Add T: AlwaysRefCounted bound.
- Link to v1: https://lore.kernel.org/r/20251217-unpin-for-aref-v1-1-84711b747d02@google.com
---
 rust/kernel/sync/aref.rs | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/rust/kernel/sync/aref.rs b/rust/kernel/sync/aref.rs
index 0d24a0432015d09509a255df0a4f114171ae9fb0..a297477f23a3960928302709f6aeb322fd4d1bd0 100644
--- a/rust/kernel/sync/aref.rs
+++ b/rust/kernel/sync/aref.rs
@@ -83,6 +83,9 @@ unsafe impl<T: AlwaysRefCounted + Sync + Send> Send for ARef<T> {}
 // example, when the reference count reaches zero and `T` is dropped.
 unsafe impl<T: AlwaysRefCounted + Sync + Send> Sync for ARef<T> {}
 
+// Even if T is pinned, pointers to T can still move.
+impl<T: AlwaysRefCounted> Unpin for ARef<T> {}
+
 impl<T: AlwaysRefCounted> ARef<T> {
     /// Creates a new instance of [`ARef`].
     ///

---
base-commit: 8f0b4cce4481fb22653697cced8d0d04027cb1e8
change-id: 20251217-unpin-for-aref-3dc0fe3fd5b0

Best regards,
-- 
Alice Ryhl <aliceryhl@google.com>

Re: [PATCH v2] rust: sync: implement Unpin for ARef

[Benno Lossin]

On Thu Dec 18, 2025 at 9:25 AM CET, Alice Ryhl wrote:
> The default implementation of Unpin for ARef<T> is conditional on T
> being Unpin due to its PhantomData<T> field. However, this is overly
> strict as pointers to T are legal to move even if T itself cannot move.
>
> Since commit 66f1ea83d9f8 ("rust: lock: Add a Pin<&mut T> accessor")
> this causes build failures when combined with a Mutex that contains an
> field ARef<T>, because almost any type that ARef is used with is !Unpin.
>
> Reviewed-by: Daniel Almeida <daniel.almeida@collabora.com>
> Signed-off-by: Alice Ryhl <aliceryhl@google.com>

Reviewed-by: Benno Lossin <lossin@kernel.org>

Cheers,
Benno

> ---
> Changes in v2:
> - Add T: AlwaysRefCounted bound.
> - Link to v1: https://lore.kernel.org/r/20251217-unpin-for-aref-v1-1-84711b747d02@google.com
> ---
>  rust/kernel/sync/aref.rs | 3 +++
>  1 file changed, 3 insertions(+)

Re: [PATCH v2] rust: sync: implement Unpin for ARef

[Miguel Ojeda]

On Thu, Dec 18, 2025 at 9:25 AM Alice Ryhl <aliceryhl@google.com> wrote:
>
> +// Even if T is pinned, pointers to T can still move.

Boqun, if you pick this up, then please tweak this to:

    // Even if `T` is pinned, pointers to `T` can still move.

Otherwise, if you want me to pick it up, please let me know -- thanks!

Cheers,
Miguel

Re: [PATCH v2] rust: sync: implement Unpin for ARef

[Alexandre Courbot]

On Thu Dec 18, 2025 at 5:25 PM JST, Alice Ryhl wrote:
> The default implementation of Unpin for ARef<T> is conditional on T
> being Unpin due to its PhantomData<T> field. However, this is overly
> strict as pointers to T are legal to move even if T itself cannot move.
>
> Since commit 66f1ea83d9f8 ("rust: lock: Add a Pin<&mut T> accessor")
> this causes build failures when combined with a Mutex that contains an
> field ARef<T>, because almost any type that ARef is used with is !Unpin.
>
> Reviewed-by: Daniel Almeida <daniel.almeida@collabora.com>
> Signed-off-by: Alice Ryhl <aliceryhl@google.com>

Reviewed-by: Alexandre Courbot <acourbot@nvidia.com>

Re: [PATCH v2] rust: sync: implement Unpin for ARef

[Boqun Feng]

On Thu, Dec 18, 2025 at 11:26:19AM +0100, Miguel Ojeda wrote:
> On Thu, Dec 18, 2025 at 9:25 AM Alice Ryhl <aliceryhl@google.com> wrote:
> >
> > +// Even if T is pinned, pointers to T can still move.
> 
> Boqun, if you pick this up, then please tweak this to:
> 
>     // Even if `T` is pinned, pointers to `T` can still move.
> 

Done:

	https://git.kernel.org/pub/scm/linux/kernel/git/boqun/linux.git/ rust-sync

Thanks!

Regards,
Boqun

> Otherwise, if you want me to pick it up, please let me know -- thanks!
> 
> Cheers,
> Miguel