Re: [PATCH -next] x86/mm/pat: fix a data race in cpa_inc_4k_install

[Marco Elver]

On Tue, 21 Jan 2020 at 10:16, Borislav Petkov <bp@xxxxxxxxx> wrote:
>
> On Tue, Jan 21, 2020 at 09:19:18AM +0100, Marco Elver wrote:
> > As I said in my email that you also copied to the message, this is
> > just a stats counter. For the general case, I think we reached
> > consensus that such accesses should intentionally remain data races:
> >   https://lore.kernel.org/linux-fsdevel/CAHk-=wg5CkOEF8DTez1Qu0XTEFw_oHhxN98bDnFqbY7HL5AB2g@xxxxxxxxxxxxxx/T/#u
>
> Yap, I agree with Linus on the legibility aspect.
>
> > Either you can use the data_race() macro, making this
> > 'data_race(cpa_4k_install++)' -- this effectively documents the
> > intentional data race --
> >
> > or just blacklist the entire file by putting
> >   KCSAN_SANITIZE_set_memory.o := n
> > into the Makefile.
> >
> > [ Note that there are 2 more ways to blacklist:
> >   - __no_kcsan function attribute, for blacklisting entire functions.
> >   - KCSAN_SANITIZE :=n in the Makefile, blacklisting all compilation
> > units in the Makefile. ]
>
> Do we have all those official methods how to make KCSAN happy,
> documented somewhere?

Yes, it's in Documentation/dev-tools/kcsan.rst. I sent a patch last
month, which is in the -rcu tree:
  http://lkml.kernel.org/r/20191212000709.166889-1-elver@xxxxxxxxxx

> > I leave it to you what makes more sense. I don't know if there are
> > other data races lurking here, since cpa_4k_install is not the only
> > stats counter.
>
> In this particular case and if it were me, I'd prefer the __no_kcsan
> function attribute because it is kept outside of the function body. But
> I can't find __no_kcsan in current tip:
>
> $ git grep __no_kcsan
> .h:204:static __no_kcsan_or_inline bool constant_test_bit(long nr, const volatile unsigned long *addr)
> include/linux/compiler.h:215:# define __no_kcsan_or_inline __no_sanitize_thread notrace __maybe_unused
> include/linux/compiler.h:216:# define __no_sanitize_or_inline __no_kcsan_or_inline
> include/linux/compiler.h:218:# define __no_kcsan_or_inline __always_inline
> include/linux/compiler.h:225:static __no_kcsan_or_inline
> include/linux/compiler.h:238:static __no_kcsan_or_inline
>
> just this "glued together" thing __no_kcsan_or_inline.

As far as I can tell it's not yet in -tip, but only in -rcu (and
-next). I believe it should be in -tip soon.

Thanks,
-- Marco