From: Noah Goldstein <goldstein.w.n@gmail.com>
To: "H.J. Lu" <hjl.tools@gmail.com>
Cc: libc-alpha@sourceware.org
Subject: Re: [PATCH v5] x86: Add a testcase for BZ #29863
Date: Mon, 19 Dec 2022 09:50:06 -0800 [thread overview]
Message-ID: <CAFUsyf+eKtLjRK+H3kV6dC+XRKKBRd-8rnvx4ReAE+dP7vrpdw@mail.gmail.com> (raw)
In-Reply-To: <CAMe9rOrEWP11eBJ6z4Rj--oBdpM+v1J_n3+gXM19fsqOWNJxPQ@mail.gmail.com>
On Mon, Dec 19, 2022 at 8:56 AM H.J. Lu <hjl.tools@gmail.com> wrote:
>
> On Mon, Dec 19, 2022 at 6:39 AM Noah Goldstein <goldstein.w.n@gmail.com> wrote:
> >
> > On Wed, Dec 14, 2022 at 1:30 PM 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 | 8 +++
> > > sysdeps/x86/tst-memcmp-race.c | 106 ++++++++++++++++++++++++++++++++++
> > > 2 files changed, 114 insertions(+)
> > > create mode 100644 sysdeps/x86/tst-memcmp-race.c
> > >
> > > diff --git a/sysdeps/x86/Makefile b/sysdeps/x86/Makefile
> > > index 56fd5fc805..c3f05a06ab 100644
> > > --- a/sysdeps/x86/Makefile
> > > +++ b/sysdeps/x86/Makefile
> > > @@ -257,3 +257,11 @@ tests += \
> > > tests-static += \
> > > tst-sysconf-cache-linesize-static
> > > endif
> > > +
> > > +ifeq ($(subdir),nptl)
> > > +tests += \
> > > + tst-memcmp-race \
> > > +# tests
> > > +
> > > +CFLAGS-tst-memcmp-race.c += -O0
> > > +endif
> > > diff --git a/sysdeps/x86/tst-memcmp-race.c b/sysdeps/x86/tst-memcmp-race.c
> > > new file mode 100644
> > > index 0000000000..eefa9e43c4
> > > --- /dev/null
> > > +++ b/sysdeps/x86/tst-memcmp-race.c
> > > @@ -0,0 +1,106 @@
> > > +/* 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/>. */
> > > +
> > > +/* Verify that there is no segfault when one thread is updating the
> > > + memory block of memcmp and the other thread is doing memcmp. */
> > > +
> > > +#define TEST_MAIN
> > > +#define TEST_NAME "memcmp"
> > > +
> > > +#include <stdio.h>
> > > +#include <stdint.h>
> > > +#include <string.h>
> > > +#include <string/test-string.h>
> > > +#include <support/xthread.h>
> > > +
> > > +#define NUM_THREADS 2
> > > +#define LOOP1 10000
> > > +#define LOOP2 1000000
> > > +
> > > +typedef int (*proto_t) (const void *, const void *, size_t);
> > > +
> > > +IMPL (memcmp, 1)
> > > +
> > > +struct arg
> > > +{
> > > + proto_t func;
> > > + size_t len;
> > > + int i;
> > > +};
> > > +
> > > +static void *
> > > +childThread (void *tArgs)
> > > +{
> > > + struct arg *a = (struct arg *) tArgs;
> > > + int i;
> > > +
> > > + if (0 == a->i % 2)
> > > + {
> > > + for (i = 0; i < LOOP1; i++)
> > > + {
> > > + int result = a->func (buf1, buf2, a->len);
> >
> > So the test can catch more subtle errors can we test
> > with memcmp args at begining and end of a page.
> >
> > i.e buf1/buf2 and (buf1 + page_size - a->len)/(buf2 + page_size - a->len).
>
> I'd like to keep it as simple as possible to cover this
> particular race condition. Make it more complex may
> not provide additional coverage for race conditions.
>
Think the idea is since we want to make this test generic,
it should provide actual coverage, not just test for the exact
bug we had in memcmp-sse2.
For example one of the toy impls brough up:
for (i = 0; i + 4; i < n; i += 4) {
u32 a32 = *(u32 *)(a + i);
u32 b32 = *(u32 *)(b + i);
if(a32 != b32) {
for(; a[i] - b[i] == 0; ++i);
return a[i] - b[i];
}
}
might overread just a few bytes and we probably
want to at least be aware of that.
> >
> > > + printf ("i = %d : result = %d\n", i, result);
> > > + }
> > > + }
> > > + else
> > > + {
> > > + size_t offset = a->len > 16 ? a->len - 16 : 1;
> > > + for (i = 0; i < LOOP2; i++)
> > > + buf1[offset] = i & 1;
> > > + }
> > > +
> > > + return NULL;
> > > +}
> > > +
> > > +static void
> > > +do_one_test (proto_t func, size_t len)
> > > +{
> > > + int i;
> > > + pthread_t threads[NUM_THREADS];
> > > + struct arg a[NUM_THREADS];
> > > +
> > > + for (i = 0; i < NUM_THREADS; ++i)
> > > + {
> > > + a[i].func = func;
> > > + a[i].len = len;
> > > + a[i].i = i;
> > > + threads[i] = xpthread_create (NULL, childThread, (void *)&a[i]);
> > > + }
> > > +
> > > + for (i = 0; i < NUM_THREADS; ++i)
> > > + xpthread_join (threads[i]);
> > > +}
> > > +
> > > +int
> > > +test_main (void)
> > > +{
> > > + test_init ();
> > > +
> > > + memset (buf1, 1, page_size);
> > > + memset (buf2, 1, page_size);
> > > +
> > > + for (size_t i = 0; i <= 11; i++)
> > > + {
> > > + FOR_EACH_IMPL (impl, 0)
> > > + do_one_test ((proto_t) impl->fn, (1 << i) + 1);
> > can you also test 2^i - 1.
>
> Will add it.
>
> > > + }
> > > +
> > > + return 0;
> > > +}
> > > +
> > > +#include <support/test-driver.c>
> > > --
> > > 2.38.1
> > >
> >
> > Can you move this to the string directory (so we run this on all
> > arch) and add corresponding tests for memcmpeq/wmemcmp.
>
> I can move it to nptl since it is a threaded test. I will take a look
> at memcmpeq and wmemcmp.
>
> --
> H.J.
next prev parent reply other threads:[~2022-12-19 17:50 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-12-14 21:30 H.J. Lu
2022-12-19 14:38 ` Noah Goldstein
2022-12-19 16:55 ` H.J. Lu
2022-12-19 17:50 ` Noah Goldstein [this message]
2022-12-19 18:04 ` H.J. Lu
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=CAFUsyf+eKtLjRK+H3kV6dC+XRKKBRd-8rnvx4ReAE+dP7vrpdw@mail.gmail.com \
--to=goldstein.w.n@gmail.com \
--cc=hjl.tools@gmail.com \
--cc=libc-alpha@sourceware.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).