Message-Id: <DBBPI9ZJVO64.3A83G118BMVLI@kernel.org>
Date: Mon, 14 Jul 2025 12:30:12 +0200
From: "Benno Lossin" <lossin@...nel.org>
To: "Boqun Feng" <boqun.feng@...il.com>, <linux-kernel@...r.kernel.org>,
 <rust-for-linux@...r.kernel.org>, <lkmm@...ts.linux.dev>,
 <linux-arch@...r.kernel.org>
Cc: "Miguel Ojeda" <ojeda@...nel.org>, "Alex Gaynor"
 <alex.gaynor@...il.com>, "Gary Guo" <gary@...yguo.net>,
 Björn Roy Baron <bjorn3_gh@...tonmail.com>, "Andreas
 Hindborg" <a.hindborg@...nel.org>, "Alice Ryhl" <aliceryhl@...gle.com>,
 "Trevor Gross" <tmgross@...ch.edu>, "Danilo Krummrich" <dakr@...nel.org>,
 "Will Deacon" <will@...nel.org>, "Peter Zijlstra" <peterz@...radead.org>,
 "Mark Rutland" <mark.rutland@....com>, "Wedson Almeida Filho"
 <wedsonaf@...il.com>, "Viresh Kumar" <viresh.kumar@...aro.org>, "Lyude
 Paul" <lyude@...hat.com>, "Ingo Molnar" <mingo@...nel.org>, "Mitchell Levy"
 <levymitchell0@...il.com>, "Paul E. McKenney" <paulmck@...nel.org>, "Greg
 Kroah-Hartman" <gregkh@...uxfoundation.org>, "Linus Torvalds"
 <torvalds@...ux-foundation.org>, "Thomas Gleixner" <tglx@...utronix.de>,
 "Alan Stern" <stern@...land.harvard.edu>
Subject: Re: [PATCH v7 4/9] rust: sync: atomic: Add generic atomics

On Mon Jul 14, 2025 at 7:36 AM CEST, Boqun Feng wrote:
> To provide using LKMM atomics for Rust code, a generic `Atomic<T>` is
> added, currently `T` needs to be Send + Copy because these are the
> straightforward usages and all basic types support this.
>
> Implement `AllowAtomic` for `i32` and `i64`, and so far only basic
> operations load() and store() are introduced.
>
> Reviewed-by: Alice Ryhl <aliceryhl@...gle.com>
> Signed-off-by: Boqun Feng <boqun.feng@...il.com>
> ---
>  rust/kernel/sync/atomic.rs         |  14 ++
>  rust/kernel/sync/atomic/generic.rs | 285 +++++++++++++++++++++++++++++
>  2 files changed, 299 insertions(+)
>  create mode 100644 rust/kernel/sync/atomic/generic.rs
>
> diff --git a/rust/kernel/sync/atomic.rs b/rust/kernel/sync/atomic.rs
> index e80ac049f36b..c5193c1c90fe 100644
> --- a/rust/kernel/sync/atomic.rs
> +++ b/rust/kernel/sync/atomic.rs
> @@ -16,7 +16,21 @@
>  //!
>  //! [`LKMM`]: srctree/tools/memory-model/
>  
> +pub mod generic;

Hmm, maybe just re-export the stuff? I don't think there's an advantage
to having the generic module be public.

>  pub mod ops;
>  pub mod ordering;
>  
> +pub use generic::Atomic;
>  pub use ordering::{Acquire, Full, Relaxed, Release};
> +
> +// SAFETY: `i32` has the same size and alignment with itself, and is round-trip transmutable to
> +// itself.
> +unsafe impl generic::AllowAtomic for i32 {
> +    type Repr = i32;
> +}
> +
> +// SAFETY: `i64` has the same size and alignment with itself, and is round-trip transmutable to
> +// itself.
> +unsafe impl generic::AllowAtomic for i64 {
> +    type Repr = i64;
> +}
> diff --git a/rust/kernel/sync/atomic/generic.rs b/rust/kernel/sync/atomic/generic.rs
> new file mode 100644
> index 000000000000..b3e07328d857
> --- /dev/null
> +++ b/rust/kernel/sync/atomic/generic.rs
> @@ -0,0 +1,285 @@
> +// SPDX-License-Identifier: GPL-2.0
> +
> +//! Generic atomic primitives.
> +
> +use super::ops::{AtomicHasBasicOps, AtomicImpl};
> +use super::{ordering, ordering::OrderingType};
> +use crate::build_error;
> +use core::cell::UnsafeCell;
> +
> +/// A memory location which can be safely modified from multiple execution contexts.
> +///
> +/// This has the same size, alignment and bit validity as the underlying type `T`.

Let's also mention that this disables any niche optimizations (due to
the unsafe cell).

> +///
> +/// The atomic operations are implemented in a way that is fully compatible with the [Linux Kernel
> +/// Memory (Consistency) Model][LKMM], hence they should be modeled as the corresponding
> +/// [`LKMM`][LKMM] atomic primitives. With the help of [`Atomic::from_ptr()`] and
> +/// [`Atomic::as_ptr()`], this provides a way to interact with [C-side atomic operations]
> +/// (including those without the `atomic` prefix, e.g. `READ_ONCE()`, `WRITE_ONCE()`,
> +/// `smp_load_acquire()` and `smp_store_release()`).
> +///
> +/// [LKMM]: srctree/tools/memory-model/
> +/// [C-side atomic operations]: srctree/Documentation/atomic_t.txt

Did you check that these links looks good in rustdoc?

> +#[repr(transparent)]
> +pub struct Atomic<T: AllowAtomic>(UnsafeCell<T>);
> +
> +// SAFETY: `Atomic<T>` is safe to share among execution contexts because all accesses are atomic.
> +unsafe impl<T: AllowAtomic> Sync for Atomic<T> {}
> +
> +/// Types that support basic atomic operations.
> +///
> +/// # Round-trip transmutability
> +///
> +/// `T` is round-trip transmutable to `U` if and only if both of these properties hold:
> +///
> +/// - Any valid bit pattern for `T` is also a valid bit pattern for `U`.
> +/// - Transmuting (e.g. using [`transmute()`]) a value of type `T` to `U` and then to `T` again
> +///   yields a value that is in all aspects equivalent to the original value.
> +///
> +/// # Safety
> +///
> +/// - [`Self`] must have the same size and alignment as [`Self::Repr`].
> +/// - [`Self`] must be [round-trip transmutable] to  [`Self::Repr`].
> +///
> +/// Note that this is more relaxed than requiring the bi-directional transmutability (i.e.
> +/// [`transmute()`] is always sound between `U` to `T`) because of the support for atomic variables

s/ to / and /

> +/// over unit-only enums, see [Examples].
> +///
> +/// # Limitations
> +///
> +/// Because C primitives are used to implement the atomic operations, and a C function requires a
> +/// valid object of a type to operate on (i.e. no `MaybeUninit<_>`), hence at the Rust <-> C
> +/// surface, only types with no uninitialized bits can be passed. As a result, types like `(u8,

s/no uninitialized/initialized/

> +/// u16)` (a tuple with a `MaybeUninit` hole in it) are currently not supported. Note that

s/a tuple with a `MaybeUninit` hole in it/padding bytes are uninitialized/

> +/// technically these types can be supported if some APIs are removed for them and the inner
> +/// implementation is tweaked, but the justification of support such a type is not strong enough at
> +/// the moment. This should be resolved if there is an implementation for `MaybeUninit<i32>` as
> +/// `AtomicImpl`.
> +///
> +/// # Examples
> +///
> +/// A unit-only enum that implements [`AllowAtomic`]:
> +///
> +/// ```
> +/// use kernel::sync::atomic::{generic::AllowAtomic, Atomic, Relaxed};
> +///
> +/// #[derive(Clone, Copy, PartialEq, Eq)]
> +/// #[repr(i32)]
> +/// enum State {
> +///     Uninit = 0,
> +///     Working = 1,
> +///     Done = 2,
> +/// };
> +///
> +/// // SAFETY: `State` and `i32` has the same size and alignment, and it's round-trip
> +/// // transmutable to `i32`.
> +/// unsafe impl AllowAtomic for State {
> +///     type Repr = i32;
> +/// }
> +///
> +/// let s = Atomic::new(State::Uninit);
> +///
> +/// assert_eq!(State::Uninit, s.load(Relaxed));
> +/// ```
> +/// [`transmute()`]: core::mem::transmute
> +/// [round-trip transmutable]: AllowAtomic#round-trip-transmutability
> +/// [Examples]: AllowAtomic#examples
> +pub unsafe trait AllowAtomic: Sized + Send + Copy {
> +    /// The backing atomic implementation type.
> +    type Repr: AtomicImpl;
> +}
> +
> +#[inline(always)]
> +const fn into_repr<T: AllowAtomic>(v: T) -> T::Repr {
> +    // SAFETY: Per the safety requirement of `AllowAtomic`, the transmute operation is sound.

Please explain the concrete parts of the safety requirements that you
are using here (ie round-trip-transmutability).

> +    unsafe { core::mem::transmute_copy(&v) }
> +}
> +
> +/// # Safety
> +///
> +/// `r` must be a valid bit pattern of `T`.
> +#[inline(always)]
> +const unsafe fn from_repr<T: AllowAtomic>(r: T::Repr) -> T {
> +    // SAFETY: Per the safety requirement of the function, the transmute operation is sound.
> +    unsafe { core::mem::transmute_copy(&r) }
> +}
> +
> +impl<T: AllowAtomic> Atomic<T> {
> +    /// Creates a new atomic `T`.
> +    pub const fn new(v: T) -> Self {
> +        Self(UnsafeCell::new(v))
> +    }
> +
> +    /// Creates a reference to an atomic `T` from a pointer of `T`.
> +    ///
> +    /// # Safety
> +    ///
> +    /// - `ptr` is aligned to `align_of::<T>()`.
> +    /// - `ptr` is valid for reads and writes for `'a`.
> +    /// - For the duration of `'a`, other accesses to `*ptr` must not cause data races (defined
> +    ///   by [`LKMM`]) against atomic operations on the returned reference. Note that if all other
> +    ///   accesses are atomic, then this safety requirement is trivially fulfilled.
> +    ///
> +    /// [`LKMM`]: srctree/tools/memory-model
> +    ///
> +    /// # Examples
> +    ///
> +    /// Using [`Atomic::from_ptr()`] combined with [`Atomic::load()`] or [`Atomic::store()`] can
> +    /// achieve the same functionality as `READ_ONCE()`/`smp_load_acquire()` or
> +    /// `WRITE_ONCE()`/`smp_store_release()` in C side:
> +    ///
> +    /// ```
> +    /// # use kernel::types::Opaque;
> +    /// use kernel::sync::atomic::{Atomic, Relaxed, Release};
> +    ///
> +    /// // Assume there is a C struct `foo`.
> +    /// mod cbindings {
> +    ///     #[repr(C)]
> +    ///     pub(crate) struct foo {
> +    ///         pub(crate) a: i32,
> +    ///         pub(crate) b: i32
> +    ///     }
> +    /// }
> +    ///
> +    /// let tmp = Opaque::new(cbindings::foo { a: 1, b: 2 });
> +    ///
> +    /// // struct foo *foo_ptr = ..;
> +    /// let foo_ptr = tmp.get();
> +    ///
> +    /// // SAFETY: `foo_ptr` is valid, and `.a` is in bounds.
> +    /// let foo_a_ptr = unsafe { &raw mut (*foo_ptr).a };
> +    ///
> +    /// // a = READ_ONCE(foo_ptr->a);
> +    /// //
> +    /// // SAFETY: `foo_a_ptr` is valid for read, and all other accesses on it is atomic, so no
> +    /// // data race.
> +    /// let a = unsafe { Atomic::from_ptr(foo_a_ptr) }.load(Relaxed);
> +    /// # assert_eq!(a, 1);
> +    ///
> +    /// // smp_store_release(&foo_ptr->a, 2);
> +    /// //
> +    /// // SAFETY: `foo_a_ptr` is valid for writes, and all other accesses on it is atomic, so
> +    /// // no data race.
> +    /// unsafe { Atomic::from_ptr(foo_a_ptr) }.store(2, Release);
> +    /// ```
> +    ///
> +    /// However, this should be only used when communicating with C side or manipulating a C
> +    /// struct.

This sentence should be above the `Safety` section.

> +    pub unsafe fn from_ptr<'a>(ptr: *mut T) -> &'a Self
> +    where
> +        T: Sync,
> +    {
> +        // CAST: `T` is transparent to `Atomic<T>`.
> +        // SAFETY: Per function safety requirement, `ptr` is a valid pointer and the object will
> +        // live long enough. It's safe to return a `&Atomic<T>` because function safety requirement
> +        // guarantees other accesses won't cause data races.
> +        unsafe { &*ptr.cast::<Self>() }
> +    }
> +
> +    /// Returns a pointer to the underlying atomic `T`.
> +    ///
> +    /// Note that use of the return pointer must not cause data races defined by [`LKMM`].
> +    ///
> +    /// # Guarantees
> +    ///
> +    /// The returned pointer is properly aligned (i.e. aligned to [`align_of::<T>()`])

You really only need this guarantee? Not validity etc?

> +    ///
> +    /// [`LKMM`]: srctree/tools/memory-model
> +    /// [`align_of::<T>()`]: core::mem::align_of
> +    pub const fn as_ptr(&self) -> *mut T {
> +        // GUARANTEE: `self.0` has the same alignment of `T`.
> +        self.0.get()
> +    }
> +
> +    /// Returns a mutable reference to the underlying atomic `T`.
> +    ///
> +    /// This is safe because the mutable reference of the atomic `T` guarantees the exclusive

s/the exclusive/exclusive/

---
Cheers,
Benno

> +    /// access.
> +    pub fn get_mut(&mut self) -> &mut T {
> +        // SAFETY: `self.as_ptr()` is a valid pointer to `T`. `&mut self` guarantees the exclusive
> +        // access, so it's safe to reborrow mutably.
> +        unsafe { &mut *self.as_ptr() }
> +    }
> +}

Powered by blists - more mailing lists