[RFC v2 02/13] rust: sync: Add basic atomic operation mapping framework

[Boqun Feng]

On Thu, Dec 12, 2024 at 11:51:23AM +0100, Alice Ryhl wrote:
> On Fri, Nov 1, 2024 at 7:03 AM Boqun Feng <boqun.feng at gmail.com> wrote:
> >
> > Preparation for generic atomic implementation. To unify the
> > ipmlementation of a generic method over `i32` and `i64`, the C side
> > atomic methods need to be grouped so that in a generic method, they can
> > be referred as <type>::<method>, otherwise their parameters and return
> > value are different between `i32` and `i64`, which would require using
> > `transmute()` to unify the type into a `T`.
> >
> > Introduce `AtomicIpml` to represent a basic type in Rust that has the
> > direct mapping to an atomic implementation from C. This trait is sealed,
> > and currently only `i32` and `i64` ipml this.
> 
> There seems to be quite a few instances of "impl" spelled as "ipml" here.
> 

Will fix!

> > Further, different methods are put into different `*Ops` trait groups,
> > and this is for the future when smaller types like `i8`/`i16` are
> > supported but only with a limited set of API (e.g. only set(), load(),
> > xchg() and cmpxchg(), no add() or sub() etc).
> >
> > While the atomic mod is introduced, documentation is also added for
> > memory models and data races.
> >
> > Also bump my role to the maintainer of ATOMIC INFRASTRUCTURE to reflect
> > my responsiblity on the Rust atomic mod.
> >
> > Signed-off-by: Boqun Feng <boqun.feng at gmail.com>
> > ---
> >  MAINTAINERS                    |   4 +-
> >  rust/kernel/sync.rs            |   1 +
> >  rust/kernel/sync/atomic.rs     |  19 ++++
> >  rust/kernel/sync/atomic/ops.rs | 199 +++++++++++++++++++++++++++++++++
> >  4 files changed, 222 insertions(+), 1 deletion(-)
> >  create mode 100644 rust/kernel/sync/atomic.rs
> >  create mode 100644 rust/kernel/sync/atomic/ops.rs
> >
> > diff --git a/MAINTAINERS b/MAINTAINERS
> > index b77f4495dcf4..e09471027a63 100644
> > --- a/MAINTAINERS
> > +++ b/MAINTAINERS
> > @@ -3635,7 +3635,7 @@ F:        drivers/input/touchscreen/atmel_mxt_ts.c
> >  ATOMIC INFRASTRUCTURE
> >  M:     Will Deacon <will at kernel.org>
> >  M:     Peter Zijlstra <peterz at infradead.org>
> > -R:     Boqun Feng <boqun.feng at gmail.com>
> > +M:     Boqun Feng <boqun.feng at gmail.com>
> >  R:     Mark Rutland <mark.rutland at arm.com>
> >  L:     linux-kernel at vger.kernel.org
> >  S:     Maintained
> > @@ -3644,6 +3644,8 @@ F:        arch/*/include/asm/atomic*.h
> >  F:     include/*/atomic*.h
> >  F:     include/linux/refcount.h
> >  F:     scripts/atomic/
> > +F:     rust/kernel/sync/atomic.rs
> > +F:     rust/kernel/sync/atomic/
> 
> This is why mod.rs files are superior :)
> 

;-) Not going to do anything right now, but let me think about this.

> > @@ -0,0 +1,19 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +
> > +//! Atomic primitives.
> > +//!
> > +//! These primitives have the same semantics as their C counterparts: and the precise definitions of
> > +//! semantics can be found at [`LKMM`]. Note that Linux Kernel Memory (Consistency) Model is the
> > +//! only model for Rust code in kernel, and Rust's own atomics should be avoided.
> > +//!
> > +//! # Data races
> > +//!
> > +//! [`LKMM`] atomics have different rules regarding data races:
> > +//!
> > +//! - A normal read doesn't data-race with an atomic read.
> 
> This was fixed:
> https://github.com/rust-lang/rust/pull/128778
> 

Yeah, I was aware of that effort, and good to know it's finally merged.
Thanks!

This will be in 1.83, right? If so, we will still need the above until
we bump up the minimal rustc version to 1.83 or beyond. I will handle
this properly with the minimal rustc 1.83 (i.e. if this goes in first,
will send a follow up patch). I will also mention in the above that this
has been changed in 1.83.

This also reminds that I should add that LKMM allows mixed-size atomic
accesses (as non data race), I will add that in the version.

> > +mod private {
> > +    /// Sealed trait marker to disable customized impls on atomic implementation traits.
> > +    pub trait Sealed {}
> > +}
> 
> Just make the trait unsafe?
> 

And make the safety requirement of `AtomicImpl` something like:

    The type must have the implementation for atomic operations.

? Hmm.. I don't think that's a good safety requirement TBH. Actually the
reason that we need to restrict `AtomicImpl` types is more of an
iplementation issue (the implementation need to be done if we want to
support i8 or i16) rather than safety issue. So a sealed trait is proper
here. Does this make sense? Or am I missing something?

Regards,
Boqun

> Alice