Re: [PATCH v2 00/33] locking/atomic: convert all architectures to ARCH_ATOMIC

[Arnd Bergmann]

On Tue, Jun 29, 2021 at 12:21 AM Randy Dunlap <rdunlap@xxxxxxxxxxxxx> wrote:
> On 6/28/21 3:13 PM, Peter Zijlstra wrote:
> > On Mon, Jun 28, 2021 at 02:22:15PM -0700, Randy Dunlap wrote:
> >
> > 0day also reported something like that. We found that this was a
> > pre-existing error, and Mark's patches only changed the error, it never
> > successfully build.
> >
> > cmpxchg64 is an optional feature for 32bit architectures, one that ARMv7
> > does provide, and uses, for it's iommu bits. Building the ARM iommu
> > drivers on ARC seem somewhat daft but is a result of that COMPILE_TEST
> > config.
> >
>
> Yes, I realize it's just a COMPILE_TEST config. Not very interesting IMHO.

Agreed as well. However, there is an interesting question in who gets to decide
which compile-test configurations are relevant and which ones are not.

I'm definitely interested in seeing any compile-time regression on the
architectures
I most care about (arm, arm64, x86) and I can see the same thing making
sense for other architectures that have someone chasing randconfig regressions,
but less so on architectures that fail general randconfig builds already.

For the specific case of cmpxchg64(), I do think it would make sense to either
have a Kconfig symbol that controls the few users, or to provide a spinlock
based fallback for those that don't have a native implementation.

        Arnd