[PATCH] x86: Add a testcase for BZ #29863

[H.J. Lu]

On Wed, Dec 14, 2022 at 9:45 AM Noah Goldstein <goldstein.w.n@gmail.com> wrote:
>
> On Wed, Dec 14, 2022 at 9:29 AM H.J. Lu <hjl.tools@gmail.com> wrote:
> >
> > When a thread is updating the memory area of memcmp and another thread
> > is doing memcmp, the memcmp return value is undefined.  Add a testcase
> > to verify that memcmp won't segfault in this case.
> > ---
> >  sysdeps/x86/Makefile               | 11 +++++
> >  sysdeps/x86/tst-memcmp-race-sse2.c |  1 +
> >  sysdeps/x86/tst-memcmp-race.c      | 75 ++++++++++++++++++++++++++++++
> >  3 files changed, 87 insertions(+)
> >  create mode 100644 sysdeps/x86/tst-memcmp-race-sse2.c
> >  create mode 100644 sysdeps/x86/tst-memcmp-race.c
> >
> > diff --git a/sysdeps/x86/Makefile b/sysdeps/x86/Makefile
> > index 56fd5fc805..3e07e18c7d 100644
> > --- a/sysdeps/x86/Makefile
> > +++ b/sysdeps/x86/Makefile
> > @@ -257,3 +257,14 @@ tests += \
> >  tests-static += \
> >    tst-sysconf-cache-linesize-static
> >  endif
> > +
> > +ifeq ($(subdir),nptl)
> > +tests += \
> > +  tst-memcmp-race-sse2 \
> > +  tst-memcmp-race \
> > +# tests
> > +
> > +CFLAGS-tst-memcmp-race-sse2.c += -O0
> > +CFLAGS-tst-memcmp-race.c += -O0
> > +tst-memcmp-race-sse2-ENV = GLIBC_TUNABLES=glibc.cpu.hwcaps=-AVX2
> > +endif
> > diff --git a/sysdeps/x86/tst-memcmp-race-sse2.c b/sysdeps/x86/tst-memcmp-race-sse2.c
> > new file mode 100644
> > index 0000000000..2986e418f3
> > --- /dev/null
> > +++ b/sysdeps/x86/tst-memcmp-race-sse2.c
> > @@ -0,0 +1 @@
> > +#include "tst-memcmp-race.c"
> > diff --git a/sysdeps/x86/tst-memcmp-race.c b/sysdeps/x86/tst-memcmp-race.c
> > new file mode 100644
> > index 0000000000..f8f8e47f6e
> > --- /dev/null
> > +++ b/sysdeps/x86/tst-memcmp-race.c
> > @@ -0,0 +1,75 @@
> > +/* Test case for memcmp with race condition.
> > +   Copyright (C) 2022 Free Software Foundation, Inc.
> > +   This file is part of the GNU C Library.
> > +
> > +   The GNU C Library is free software; you can redistribute it and/or
> > +   modify it under the terms of the GNU Lesser General Public
> > +   License as published by the Free Software Foundation; either
> > +   version 2.1 of the License, or (at your option) any later version.
> > +
> > +   The GNU C Library is distributed in the hope that it will be useful,
> > +   but WITHOUT ANY WARRANTY; without even the implied warranty of
> > +   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> > +   Lesser General Public License for more details.
> > +
> > +   You should have received a copy of the GNU Lesser General Public
> > +   License along with the GNU C Library; if not, see
> > +   <https://www.gnu.org/licenses/>.  */
> > +
> > +#include <stdio.h>
> > +#include <stdint.h>
> > +#include <string.h>
> > +#include <support/xthread.h>
> > +
> > +#define        NUM_THREADS     2
> > +#define        LEN             17
>
> Imo if we test this `LEN` should be a variable and we should test
> all 2^N + 1 for N in [0, 12]. I wouldn't be suprised if there are
> other cases.

Fixd in v2.

> > +#define        STR             "abcdefghijklmnopq"
> > +#define LOOP1          100
> > +#define LOOP2          1000000
> > +
> > +char buf1[LEN], buf2[LEN];
>
> I had to make this volatile to reproduce the failure
> with GCC 12, does this properly segfault w.o the
> fix patch?

Yes, with -O0.

> > +
> > +static void *
> > +childThread (void *tArgs)
> > +{
> > +  int i;
> > +
> > +  if (0 == (size_t)tArgs % 2)
> > +    {
> > +      for (i = 0; i < LOOP1; i++)
> > +       {
> > +         int result;
> > +
> > +         result = memcmp (buf1, buf2, LEN);
>
> If we are testing this I think it should test each impl.

Fixed in v2.

> > +         printf ("i = %d : result = %d\n", i, result);
>
> IMO remove the print and make `result` volatile.

There is a race condition.  -O0 can reproduce it with GCC 12.

> > +       }
> > +    }
> > +  else
> > +    {
> > +      for (i = 0; i < LOOP2; i++)
> > +       buf1[1] = (0 == (i % 2)) ? 'b' : 'c';
> > +    }
> > +
> > +  return NULL;
> > +}
> > +
> > +static int
> > +do_test (void)
> > +{
> > +  int i;
> > +  pthread_t threads[NUM_THREADS];
> > +
> > +  memcpy (buf1, STR, LEN);
> > +  memcpy (buf2, STR, LEN);
> > +
> > +  for (i = 0; i < NUM_THREADS; ++i)
> > +    threads[i] = xpthread_create (NULL, childThread,
> > +                                 (void *)(uintptr_t) i);
> > +
> > +  for (i = 0; i < NUM_THREADS; ++i)
> > +    xpthread_join (threads[i]);
> > +
> > +  return 0;
> > +}
> > +
> > +#include <support/test-driver.c>
> > --
> > 2.38.1
> >
>
> Not sure we want this, given its not behavior we have decided
> to actively support its failures could be misleading.

This testcase only checks segfault, not memcmp return values.

-- 
H.J.