Re: [PATCH] Revert "x86_64: Suppress false positive valgrind error"

["H.J. Lu" <hjl.tools@gmail.com>]

On Sat, Apr 13, 2024 at 8:27 AM H.J. Lu <hjl.tools@gmail.com> wrote:
>
> On Sat, Apr 13, 2024 at 8:20 AM Florian Weimer <fweimer@redhat.com> wrote:
> >
> > * H. J. Lu:
> >
> > > On Sat, Apr 13, 2024 at 8:03 AM Florian Weimer <fweimer@redhat.com> wrote:
> > >>
> > >> * H. J. Lu:
> > >>
> > >> > On Sat, Apr 13, 2024 at 7:49 AM Florian Weimer <fweimer@redhat.com> wrote:
> > >> >>
> > >> >> This reverts commit a1735e0aa858f0c8b15e5ee9975bff4279423680.
> > >> >>
> > >> >> The test failure is a real valgrind bug that needs to be fixed before
> > >> >> valgrind is usable with a glibc that has been built with
> > >> >> CC="gcc -march=x86-64-v3".  The valgrind fix addresses the observed
> > >> >> test failure as well.
> > >> >>
> > >> >> Upstream bug: <https://bugs.kde.org/show_bug.cgi?id=485487>
> > >> >>
> > >> >> Tested on x86_64-linux-gnu with and without CC="gcc -march=x86-64-v3".
> > >> >> The elf/tst-valgrind-smoke tests again fails with an unpatched valgrind,
> > >> >> and the proposed upstream patch fixes it.  (Apparently, valgrind
> > >> >> matches on soname, and not the file name, which is ld.so in the test.)
> > >> >
> > >> > Please clarify that your patch fixes the glibc valgrind test by not using
> > >> > strcmp-avx2.S in the valgrind test.
> > >>
> > >> Sorry, clarify where?  In the commit message here?  On the upstream bug?
> > >
> > > In the glibc commit message.
> > >
> > >> My valgrind patch replaces the strcmp implementation in ld.so with its
> > >> own.
> > >>
> > >
> > > Please mention it in the glibc commit message.
> >
> > Okay, I'll send a v2.
> >
> > Could someone proficient in x86-64 assembly have a look at my strcmp
> > implementation?
>
> Can you adapt sysdeps/i386/i686/strcmp.S for valgrind?
>

You can use sysdeps/x86_64/strcmp.S from glibc 2.10:

ENTRY (BP_SYM (strcmp))
L(oop): movb (%rdi), %al
cmpb (%rsi), %al
jne L(neq)
incq %rdi
incq %rsi
testb %al, %al
jnz L(oop)

xorl %eax, %eax
ret

L(neq): movl $1, %eax
movl $-1, %ecx
cmovbl %ecx, %eax
ret
END (BP_SYM (strcmp))


-- 
H.J.