Re: [PATCH v5 00/18] Rework READ_ONCE() to improve codegen

[Marco Elver]

On Tue, 12 May 2020 at 10:18, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> On Mon, May 11, 2020 at 09:41:32PM +0100, Will Deacon wrote:
> > Hi folks,
> >
> > (trimmed CC list since v4 since this is largely just a rebase)
> >
> > This is version five of the READ_ONCE() codegen improvement series that
> > I've previously posted here:
> >
> > RFC: https://lore.kernel.org/lkml/20200110165636.28035-1-will@xxxxxxxxxx
> > v2:  https://lore.kernel.org/lkml/20200123153341.19947-1-will@xxxxxxxxxx
> > v3:  https://lore.kernel.org/lkml/20200415165218.20251-1-will@xxxxxxxxxx
> > v4:  https://lore.kernel.org/lkml/20200421151537.19241-1-will@xxxxxxxxxx
> >
> > The main change since v4 is that this is now based on top of the KCSAN
> > changes queued in -tip (locking/kcsan) and therefore contains the patches
> > necessary to avoid breaking sparc32 as well as some cleanups to
> > consolidate {READ,WRITE}_ONCE() and data_race().
> >
> > Other changes include:
> >
> >   * Treat 'char' as distinct from 'signed char' and 'unsigned char' for
> >     __builtin_types_compatible_p()
> >
> >   * Add a compile-time assertion that the argument to READ_ONCE_NOCHECK()
> >     points at something the same size as 'unsigned long'
> >
> > I'm happy for all of this to go via -tip, or I can take it via arm64.
>
> Looks good to me; Thanks!
>
> Acked-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>

I just ran a bunch of KCSAN tests. While this series alone would have
passed the tests, there appears to be a problem with
__READ_ONCE/__WRITE_ONCE. I think they should already be using
'data_race()', as otherwise we will get lots of false positives in
future.

I noticed this when testing -tip/locking/kcsan, which breaks
unfortunately, because I see a bunch of spurious data races with
arch_atomic_{read,set} because "locking/atomics: Flip fallbacks and
instrumentation" changed them to use __READ_ONCE()/__WRITE_ONCE().
>From what I see, the intent was to not double-instrument,
unfortunately they are still double-instrumented because
__READ_ONCE/__WRITE_ONCE doesn't hide the access from KCSAN (nor KASAN
actually). I don't think we can use __no_sanitize_or_inline for the
arch_ functions, because we really want them to be __always_inline
(also to avoid calls to these functions in uaccess regions, which
objtool would notice).

I think the easiest way to resolve this is to wrap the accesses in
__*_ONCE with data_race().

Thanks,
-- Marco