* [PATCH v5] x86: Add a testcase for BZ #29863
@ 2022-12-14 21:30 H.J. Lu
2022-12-19 14:38 ` Noah Goldstein
0 siblings, 1 reply; 5+ messages in thread
From: H.J. Lu @ 2022-12-14 21:30 UTC (permalink / raw)
To: libc-alpha; +Cc: Noah Goldstein
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);
+ 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);
+ }
+
+ return 0;
+}
+
+#include <support/test-driver.c>
--
2.38.1
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH v5] x86: Add a testcase for BZ #29863
2022-12-14 21:30 [PATCH v5] x86: Add a testcase for BZ #29863 H.J. Lu
@ 2022-12-19 14:38 ` Noah Goldstein
2022-12-19 16:55 ` H.J. Lu
0 siblings, 1 reply; 5+ messages in thread
From: Noah Goldstein @ 2022-12-19 14:38 UTC (permalink / raw)
To: H.J. Lu; +Cc: libc-alpha
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).
> + 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.
> + }
> +
> + 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.
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH v5] x86: Add a testcase for BZ #29863
2022-12-19 14:38 ` Noah Goldstein
@ 2022-12-19 16:55 ` H.J. Lu
2022-12-19 17:50 ` Noah Goldstein
0 siblings, 1 reply; 5+ messages in thread
From: H.J. Lu @ 2022-12-19 16:55 UTC (permalink / raw)
To: Noah Goldstein; +Cc: libc-alpha
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.
>
> > + 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.
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH v5] x86: Add a testcase for BZ #29863
2022-12-19 16:55 ` H.J. Lu
@ 2022-12-19 17:50 ` Noah Goldstein
2022-12-19 18:04 ` H.J. Lu
0 siblings, 1 reply; 5+ messages in thread
From: Noah Goldstein @ 2022-12-19 17:50 UTC (permalink / raw)
To: H.J. Lu; +Cc: libc-alpha
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.
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH v5] x86: Add a testcase for BZ #29863
2022-12-19 17:50 ` Noah Goldstein
@ 2022-12-19 18:04 ` H.J. Lu
0 siblings, 0 replies; 5+ messages in thread
From: H.J. Lu @ 2022-12-19 18:04 UTC (permalink / raw)
To: Noah Goldstein; +Cc: libc-alpha
On Mon, Dec 19, 2022 at 9:50 AM Noah Goldstein <goldstein.w.n@gmail.com> wrote:
>
> 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.
We need to verify that it triggers the segfault without the fix.
>
> > >
> > > > + 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.
--
H.J.
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2022-12-19 18:05 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-12-14 21:30 [PATCH v5] x86: Add a testcase for BZ #29863 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
2022-12-19 18:04 ` H.J. Lu
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).