Re: [PATCH] kasan: remove hwasan-kernel-mem-intrinsic-prefix=1 for clang-14

[Marco Elver]

On Fri, 14 Apr 2023 at 18:26, Nathan Chancellor <nathan@xxxxxxxxxx> wrote:
>
> On Fri, Apr 14, 2023 at 10:29:27AM +0200, Arnd Bergmann wrote:
> > From: Arnd Bergmann <arnd@xxxxxxxx>
> >
> > Unknown -mllvm options don't cause an error to be returned by clang, so
> > the cc-option helper adds the unknown hwasan-kernel-mem-intrinsic-prefix=1
> > flag to CFLAGS with compilers that are new enough for hwasan but too
>
> Hmmm, how did a change like commit 0e1aa5b62160 ("kcsan: Restrict
> supported compilers") work if cc-option does not work with unknown
> '-mllvm' flags (or did it)? That definitely seems like a problem, as I
> see a few different places where '-mllvm' options are used with
> cc-option. I guess I will leave that up to the sanitizer folks to
> comment on that further, one small comment below.

Urgh, this one turns out to be rather ridiculous. It's only a problem
with hwasan...

If you try it for yourself, e.g. with something "normal" like:

> clang -Werror -mllvm -asan-does-not-exist -c -x c /dev/null -o /dev/null

It errors as expected. But with:

> clang -Werror -mllvm -hwasan-does-not-exist -c -x c /dev/null -o /dev/null

It ends up printing _help_ text, because anything "-h..." (if it
doesn't recognize it as a long-form argument), will make it produce
the help text.

> > old for this option. This causes a rather unreadable build failure:
> >
> > fixdep: error opening file: scripts/mod/.empty.o.d: No such file or directory
> > make[4]: *** [/home/arnd/arm-soc/scripts/Makefile.build:252: scripts/mod/empty.o] Error 2
> > fixdep: error opening file: scripts/mod/.devicetable-offsets.s.d: No such file or directory
> > make[4]: *** [/home/arnd/arm-soc/scripts/Makefile.build:114: scripts/mod/devicetable-offsets.s] Error 2
> >
> > Add a version check to only allow this option with clang-15, gcc-13
> > or later versions.
> >
> > Fixes: 51287dcb00cc ("kasan: emit different calls for instrumentable memintrinsics")
> > Signed-off-by: Arnd Bergmann <arnd@xxxxxxxx>
> > ---
> > There is probably a better way to do this than to add version checks,
> > but I could not figure it out.
> > ---
> >  scripts/Makefile.kasan | 5 +++++
> >  1 file changed, 5 insertions(+)
> >
> > diff --git a/scripts/Makefile.kasan b/scripts/Makefile.kasan
> > index c186110ffa20..2cea0592e343 100644
> > --- a/scripts/Makefile.kasan
> > +++ b/scripts/Makefile.kasan
> > @@ -69,7 +69,12 @@ CFLAGS_KASAN := -fsanitize=kernel-hwaddress \
> >               $(instrumentation_flags)
> >
> >  # Instrument memcpy/memset/memmove calls by using instrumented __hwasan_mem*().
> > +ifeq ($(call clang-min-version, 150000),y)
> >  CFLAGS_KASAN += $(call cc-param,hwasan-kernel-mem-intrinsic-prefix=1)
> > +endif
> > +ifeq ($(call gcc-min-version, 130000),y)
> > +CFLAGS_KASAN += $(call cc-param,hwasan-kernel-mem-intrinsic-prefix=1)
> > +endif
>
> I do not think you need to duplicate this block, I think
>
>   ifeq ($(call clang-min-version, 150000)$(call gcc-min-version, 130000),y)
>   CFLAGS_KASAN += $(call cc-param,hwasan-kernel-mem-intrinsic-prefix=1)
>   endif

We just need the clang version check. If the compiler is gcc, it'll do
the "right thing" (i.e. not print help text). So at a minimum, we need
if "clang version >= 15 or gcc". Checking if gcc is 13 or later
doesn't hurt though, so I don't mind either way.

So on a whole this patch is the right thing to do because fixing old
clang versions to not interpret unrecognized options that start with
"-h.." as help isn't something we can realistically do.

Thanks,
-- Marco