Re: [PATCH 00/33] locking/atomic: convert all architectures to ARCH_ATOMIC
From: Mark Rutland
Date: Tue May 11 2021 - 05:27:16 EST
On Mon, May 10, 2021 at 04:56:03PM +0200, Arnd Bergmann wrote:
> On Mon, May 10, 2021 at 11:39 AM Mark Rutland <mark.rutland@xxxxxxx> wrote:
> > This series (based on v5.13-rc1) converts all architectures to
> > ARCH_ATOMIC. This will allow the use of instrumented atomics on all
> > architectures (e.g. for KASAN and similar), and simplifies the core
> > atomic code (which should allow for easier rework of the fallbacks and
> > other bits in future).
> > The series is split into three parts:
> > 1) Some preparatory work is done to prepare architectures and common
> > code for the conversion. In this phase h8300 and microblaze are
> > converted to use the asm-generic atomics exclusively, and the
> > asm-generic implementations are made to function with or without
> > ARCH_ATOMIC.
> > 2) Architectures are converted one-by-one to use the ARCH_ATOMIC
> > interface. I've converted each architecture with its own patch (even
> > where the conversion is trivial) to make review and bisection easier.
> > 3) The code handling !ARCH_ATOMIC is removed.
> > Note: I've generated the patches with:
> > git format-patch -C -M -D
> > ... so the preimage of include/linux/atomic-fallback.h is not included
> > in the diff when it is deleted.
> > The series can also be found in my atomics/arch-atomic branch on
> > kernel.org:
> > https://git.kernel.org/pub/scm/linux/kernel/git/mark/linux.git/log/?h=atomics/arch-atomic
> > git://git.kernel.org/pub/scm/linux/kernel/git/mark/linux.git atomics/arch-atomic
> Nice work!
> I had recently tried experimented with cleanups around cmpxchg(),
> adding fixed-length
> versions of that which can be used to implement the atomics with less bloated
> preprocessor output. My patches will clearly conflict with your
> series, but it seems
> that we should do yours first in any case.
Ah; sorry for stepping on your toes with this. Simplifying the
xchg/cmpxchg glue sounds like a nice cleanup too!
> I'm happy to put this into the asm-generic tree once you have
> collected enough Acks.
Great! Going via asm-generic sounds good to me. The kbuild test robot
has found a few issues, so I'll probably have a v2 with those fixed and
some acks accumulated.
Peter, does that make sense to you, or did you have reasons to want this
to go via tip?