* [PATCH 1/5] nptl: Add <thread_pointer.h> for defining __thread_pointer
2021-12-06 13:45 [PATCH 0/5] Extensible rseq support for glibc Florian Weimer
@ 2021-12-06 13:46 ` Florian Weimer
2021-12-06 16:44 ` Mathieu Desnoyers
2021-12-06 13:46 ` [PATCH 2/5] nptl: Add rseq registration Florian Weimer
` (4 subsequent siblings)
5 siblings, 1 reply; 25+ messages in thread
From: Florian Weimer @ 2021-12-06 13:46 UTC (permalink / raw)
To: libc-alpha; +Cc: Mathieu Desnoyers
<tls.h> already contains a definition that is quite similar,
but it is not readily accessible.
---
sysdeps/csky/nptl/thread_pointer.h | 25 ++++++++++++++++++
sysdeps/hppa/nptl/thread_pointer.h | 25 ++++++++++++++++++
sysdeps/ia64/nptl/thread_pointer.h | 19 ++++++++++++++
sysdeps/m68k/nptl/thread_pointer.h | 20 ++++++++++++++
sysdeps/microblaze/nptl/thread_pointer.h | 19 ++++++++++++++
sysdeps/nptl/thread_pointer.h | 19 ++++++++++++++
sysdeps/powerpc/nptl/thread_pointer.h | 23 +++++++++++++++++
sysdeps/sparc/nptl/thread_pointer.h | 19 ++++++++++++++
sysdeps/x86/nptl/thread_pointer.h | 33 ++++++++++++++++++++++++
9 files changed, 202 insertions(+)
create mode 100644 sysdeps/csky/nptl/thread_pointer.h
create mode 100644 sysdeps/hppa/nptl/thread_pointer.h
create mode 100644 sysdeps/ia64/nptl/thread_pointer.h
create mode 100644 sysdeps/m68k/nptl/thread_pointer.h
create mode 100644 sysdeps/microblaze/nptl/thread_pointer.h
create mode 100644 sysdeps/nptl/thread_pointer.h
create mode 100644 sysdeps/powerpc/nptl/thread_pointer.h
create mode 100644 sysdeps/sparc/nptl/thread_pointer.h
create mode 100644 sysdeps/x86/nptl/thread_pointer.h
diff --git a/sysdeps/csky/nptl/thread_pointer.h b/sysdeps/csky/nptl/thread_pointer.h
new file mode 100644
index 0000000000..c8be0b1e84
--- /dev/null
+++ b/sysdeps/csky/nptl/thread_pointer.h
@@ -0,0 +1,25 @@
+/* __thread_pointer definition. csky version.
+ Copyright (C) 2021 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/>. */
+
+static inline void *
+__thread_pointer (void)
+{
+ void *__result;
+ __asm__ volatile ("mov %0, r31" : "=r" (__result));
+ return __result;
+}
diff --git a/sysdeps/hppa/nptl/thread_pointer.h b/sysdeps/hppa/nptl/thread_pointer.h
new file mode 100644
index 0000000000..118b7f9385
--- /dev/null
+++ b/sysdeps/hppa/nptl/thread_pointer.h
@@ -0,0 +1,25 @@
+/* __thread_pointer definition. hppa version.
+ Copyright (C) 2021 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/>. */
+
+static inline void *
+__thread_pointer (void)
+{
+ void *__result;
+ __asm__ ("mfctl %%cr27, %0" : "=r" (__result));
+ return __result;
+}
diff --git a/sysdeps/ia64/nptl/thread_pointer.h b/sysdeps/ia64/nptl/thread_pointer.h
new file mode 100644
index 0000000000..412b9f0460
--- /dev/null
+++ b/sysdeps/ia64/nptl/thread_pointer.h
@@ -0,0 +1,19 @@
+/* __thread_pointer definition. ia64 version.
+ Copyright (C) 2021 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/>. */
+
+#define __thread_pointer() ({ register void *__reg asm ("r13"); __reg; })
diff --git a/sysdeps/m68k/nptl/thread_pointer.h b/sysdeps/m68k/nptl/thread_pointer.h
new file mode 100644
index 0000000000..932567fddb
--- /dev/null
+++ b/sysdeps/m68k/nptl/thread_pointer.h
@@ -0,0 +1,20 @@
+/* __thread_pointer definition. m68k version.
+ Copyright (C) 2021 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/>. */
+
+extern void * __m68k_read_tp (void);
+#define __thread_pointer() __m68k_read_tp ()
diff --git a/sysdeps/microblaze/nptl/thread_pointer.h b/sysdeps/microblaze/nptl/thread_pointer.h
new file mode 100644
index 0000000000..020c73113a
--- /dev/null
+++ b/sysdeps/microblaze/nptl/thread_pointer.h
@@ -0,0 +1,19 @@
+/* __thread_pointer definition. microblaze version.
+ Copyright (C) 2021 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/>. */
+
+#define __thread_pointer() ({ register void *__reg asm ("r21"); __reg; })
diff --git a/sysdeps/nptl/thread_pointer.h b/sysdeps/nptl/thread_pointer.h
new file mode 100644
index 0000000000..6368b1083e
--- /dev/null
+++ b/sysdeps/nptl/thread_pointer.h
@@ -0,0 +1,19 @@
+/* __thread_pointer definition. Generic version.
+ Copyright (C) 2021 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/>. */
+
+#define __thread_pointer() __builtin_thread_pointer ()
diff --git a/sysdeps/powerpc/nptl/thread_pointer.h b/sysdeps/powerpc/nptl/thread_pointer.h
new file mode 100644
index 0000000000..69943a2f24
--- /dev/null
+++ b/sysdeps/powerpc/nptl/thread_pointer.h
@@ -0,0 +1,23 @@
+/* __thread_pointer definition. powerpc version.
+ Copyright (C) 2021 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/>. */
+
+#ifdef __powerpc64__
+# define __thread_pointer() ({ register void *__reg asm ("r13"); __reg; })
+#else
+# define __thread_pointer() ({ register void *__reg asm ("r2"); __reg; })
+#endif
diff --git a/sysdeps/sparc/nptl/thread_pointer.h b/sysdeps/sparc/nptl/thread_pointer.h
new file mode 100644
index 0000000000..6a176cba28
--- /dev/null
+++ b/sysdeps/sparc/nptl/thread_pointer.h
@@ -0,0 +1,19 @@
+/* __thread_pointer definition. sparc version.
+ Copyright (C) 2021 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/>. */
+
+#define __thread_pointer() ({ register void *__reg asm ("%g7"); __reg; })
diff --git a/sysdeps/x86/nptl/thread_pointer.h b/sysdeps/x86/nptl/thread_pointer.h
new file mode 100644
index 0000000000..d636839ecf
--- /dev/null
+++ b/sysdeps/x86/nptl/thread_pointer.h
@@ -0,0 +1,33 @@
+/* __thread_pointer definition. x86 version.
+ Copyright (C) 2021 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/>. */
+
+#if __GNUC_PREREQ (11, 1)
+# define __thread_pointer() __builtin_thread_pointer ()
+#else
+static inline void *
+__thread_pointer (void)
+{
+ void *__result;
+# ifdef __x86_64__
+ __asm__ volatile ("mov %%fs:0, %0" : "=r" (__result));
+# else
+ __asm__ volatile ("mov %%gs:0, %0" : "=r" (__result));
+#endif
+ return __result;
+}
+#endif
--
2.33.1
^ permalink raw reply [flat|nested] 25+ messages in thread* Re: [PATCH 1/5] nptl: Add <thread_pointer.h> for defining __thread_pointer
2021-12-06 13:46 ` [PATCH 1/5] nptl: Add <thread_pointer.h> for defining __thread_pointer Florian Weimer
@ 2021-12-06 16:44 ` Mathieu Desnoyers
2021-12-06 17:01 ` Florian Weimer
0 siblings, 1 reply; 25+ messages in thread
From: Mathieu Desnoyers @ 2021-12-06 16:44 UTC (permalink / raw)
To: Florian Weimer; +Cc: libc-alpha
----- On Dec 6, 2021, at 8:46 AM, Florian Weimer fweimer@redhat.com wrote:
> <tls.h> already contains a definition that is quite similar,
> but it is not readily accessible.
[...]
> +++ b/sysdeps/csky/nptl/thread_pointer.h
[...]
> +static inline void *
> +__thread_pointer (void)
> +{
> + void *__result;
> + __asm__ volatile ("mov %0, r31" : "=r" (__result));
Do we want volatile here ?
> + return __result;
> +}
[...]
> +++ b/sysdeps/hppa/nptl/thread_pointer.h
[...]
> +
> +static inline void *
> +__thread_pointer (void)
> +{
> + void *__result;
> + __asm__ ("mfctl %%cr27, %0" : "=r" (__result));
> + return __result;
> +}
[...]
> +++ b/sysdeps/ia64/nptl/thread_pointer.h
[...]
> +
> +#define __thread_pointer() ({ register void *__reg asm ("r13"); __reg; })
[...]
Why do we find a mix of macros and static inline across architectures ?
I would expect that we stick to one or the other.
Static inline have some advantages in terms of type-awareness. Arguably here
everything is a (void *), so there is little benefit from it.
Defining macros end up clobbering the preprocessor namespace (output of "gcc -dM -E"),
which is one reason why static inlines are sometimes preferred. Or do we intend users
to use this for feature detection ? e.g. #ifdef __thread_pointer ?
OTOH, static inline functions will not be inlined when compiling with "-fno-inline"
(default at -O0). An attribute "always_inline" is needed for this.
> +++ b/sysdeps/m68k/nptl/thread_pointer.h
[...]
> +extern void * __m68k_read_tp (void);
> +#define __thread_pointer() __m68k_read_tp ()
[...]
> +++ b/sysdeps/microblaze/nptl/thread_pointer.h
[...]
> +#define __thread_pointer() ({ register void *__reg asm ("r21"); __reg; })
Not that applications are supposed to use variable names with a leading
"__" prefix (AFAIU this is reserved to libc), but one downside of these
statement expressions is that they would shadow any "__reg" variables.
One example where things could theoretically be a problem is found in
sysdeps/x86/sys/platform/x86.h within glibc, which defines a __reg
variable.
[...]
> +++ b/sysdeps/nptl/thread_pointer.h
[...]
> +#define __thread_pointer() __builtin_thread_pointer ()
[...]
> +++ b/sysdeps/powerpc/nptl/thread_pointer.h
[...]
> +#ifdef __powerpc64__
> +# define __thread_pointer() ({ register void *__reg asm ("r13"); __reg; })
> +#else
> +# define __thread_pointer() ({ register void *__reg asm ("r2"); __reg; })
> +#endif
[ ...]
> +++ b/sysdeps/sparc/nptl/thread_pointer.h
[...]
> +#define __thread_pointer() ({ register void *__reg asm ("%g7"); __reg; })
[...]
> +++ b/sysdeps/x86/nptl/thread_pointer.h
[...]
> +#if __GNUC_PREREQ (11, 1)
I suspect this requires a #include "features.h" ?
> +# define __thread_pointer() __builtin_thread_pointer ()
> +#else
> +static inline void *
> +__thread_pointer (void)
> +{
> + void *__result;
> +# ifdef __x86_64__
> + __asm__ volatile ("mov %%fs:0, %0" : "=r" (__result));
Just out of curiosity, is x32 supported by glibc, and is this the
expected behavior ?
> +# else
> + __asm__ volatile ("mov %%gs:0, %0" : "=r" (__result));
Do we really want a "volatile" asm here ? E.g. see:
sysdeps/x86_64/nptl/tls.h:
/* Return the thread descriptor for the current thread.
The contained asm must *not* be marked volatile since otherwise
assignments like
pthread_descr self = thread_self();
do not get optimized away. */
# if __GNUC_PREREQ (6, 0)
# define THREAD_SELF \
(*(struct pthread *__seg_fs *) offsetof (struct pthread, header.self))
# else
# define THREAD_SELF \
({ struct pthread *__self; \
asm ("mov %%fs:%c1,%0" : "=r" (__self) \
: "i" (offsetof (struct pthread, header.self))); \
__self;})
# endif
Is this expected to be a complete list of all architectures ? I notably identify
that arm, aarch64, and s390 are missing. Or is it that all the other architectures
map to "sysdeps/nptl/thread_pointers.h" ? Maybe it would be good to document this
in the commit message.
Thanks,
Mathieu
> +#endif
> + return __result;
> +}
> +#endif
--
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com
^ permalink raw reply [flat|nested] 25+ messages in thread* Re: [PATCH 1/5] nptl: Add <thread_pointer.h> for defining __thread_pointer
2021-12-06 16:44 ` Mathieu Desnoyers
@ 2021-12-06 17:01 ` Florian Weimer
2021-12-06 19:55 ` Florian Weimer
0 siblings, 1 reply; 25+ messages in thread
From: Florian Weimer @ 2021-12-06 17:01 UTC (permalink / raw)
To: Mathieu Desnoyers; +Cc: libc-alpha
* Mathieu Desnoyers:
> Why do we find a mix of macros and static inline across architectures ?
It's following the existing snippets in <tls.h> for those architectures
which cannot use __builtin_thread_pointer.
> Defining macros end up clobbering the preprocessor namespace (output
> of "gcc -dM -E"), which is one reason why static inlines are sometimes
> preferred. Or do we intend users to use this for feature detection ?
> e.g. #ifdef __thread_pointer ?
It's currently not an installed header, so this isn't a concern.
>
> OTOH, static inline functions will not be inlined when compiling with "-fno-inline"
> (default at -O0). An attribute "always_inline" is needed for this.
>
>
>> +++ b/sysdeps/m68k/nptl/thread_pointer.h
> [...]
>> +extern void * __m68k_read_tp (void);
>> +#define __thread_pointer() __m68k_read_tp ()
> [...]
>
>> +++ b/sysdeps/microblaze/nptl/thread_pointer.h
> [...]
>> +#define __thread_pointer() ({ register void *__reg asm ("r21"); __reg; })
>
> Not that applications are supposed to use variable names with a leading
> "__" prefix (AFAIU this is reserved to libc), but one downside of these
> statement expressions is that they would shadow any "__reg" variables.
It doesn't matter in this context because there are no macro arguments.
The only thing that would break this would be a __reg macro.
>> +++ b/sysdeps/x86/nptl/thread_pointer.h
> [...]
>> +#if __GNUC_PREREQ (11, 1)
>
> I suspect this requires a #include "features.h" ?
It's an internal header. We'd also need include guards if we made this
into an installed header.
>> +# define __thread_pointer() __builtin_thread_pointer ()
>> +#else
>> +static inline void *
>> +__thread_pointer (void)
>> +{
>> + void *__result;
>> +# ifdef __x86_64__
>> + __asm__ volatile ("mov %%fs:0, %0" : "=r" (__result));
>
> Just out of curiosity, is x32 supported by glibc, and is this the
> expected behavior ?
x32 is supported and uses %fs.
>> +# else
>> + __asm__ volatile ("mov %%gs:0, %0" : "=r" (__result));
>
> Do we really want a "volatile" asm here ? E.g. see:
With sufficiently separate compilation, we do not need it. It's
currently not needed.
THREAD_SELF should not use it because it is often used in ways that
isn't needed after all, e.g. if THREAD_SELF is passed to THREAD_GETMEM
on an architecture that does not need a materialized thread pointer to
change TCB fields.
> Is this expected to be a complete list of all architectures ? I
> notably identify that arm, aarch64, and s390 are missing. Or is it
> that all the other architectures map to
> "sysdeps/nptl/thread_pointers.h" ? Maybe it would be good to document
> this in the commit message.
Yes, those can use __builtin_thread_pointer. The patch as is builds on
all targets with GCC 11.
Thanks,
Florian
^ permalink raw reply [flat|nested] 25+ messages in thread* Re: [PATCH 1/5] nptl: Add <thread_pointer.h> for defining __thread_pointer
2021-12-06 17:01 ` Florian Weimer
@ 2021-12-06 19:55 ` Florian Weimer
0 siblings, 0 replies; 25+ messages in thread
From: Florian Weimer @ 2021-12-06 19:55 UTC (permalink / raw)
To: Mathieu Desnoyers; +Cc: libc-alpha
* Florian Weimer:
> * Mathieu Desnoyers:
>
>> Why do we find a mix of macros and static inline across architectures ?
>
> It's following the existing snippets in <tls.h> for those architectures
> which cannot use __builtin_thread_pointer.
I'm reposting this separately. I was able to restrict the set of
architectures that need this.
Thanks,
Florian
^ permalink raw reply [flat|nested] 25+ messages in thread
* [PATCH 2/5] nptl: Add rseq registration
2021-12-06 13:45 [PATCH 0/5] Extensible rseq support for glibc Florian Weimer
2021-12-06 13:46 ` [PATCH 1/5] nptl: Add <thread_pointer.h> for defining __thread_pointer Florian Weimer
@ 2021-12-06 13:46 ` Florian Weimer
2021-12-06 16:53 ` Mathieu Desnoyers
2021-12-06 16:59 ` Mathieu Desnoyers
2021-12-06 13:46 ` [PATCH 3/5] Linux: Use rseq to accelerate sched_getcpu Florian Weimer
` (3 subsequent siblings)
5 siblings, 2 replies; 25+ messages in thread
From: Florian Weimer @ 2021-12-06 13:46 UTC (permalink / raw)
To: libc-alpha; +Cc: Mathieu Desnoyers
The rseq area is placed directly into struct pthread. rseq
registration failure is not treated as an error, so it is possible
that threads run with inconsistent registration status.
<sys/rseq.h> is not yet installed as a public header.
Co-Authored-By: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
---
csu/libc-tls.c | 6 +
nptl/descr.h | 4 +
nptl/pthread_create.c | 13 +
sysdeps/nptl/dl-tls_init_tp.c | 8 +-
sysdeps/unix/sysv/linux/Makefile | 9 +-
sysdeps/unix/sysv/linux/aarch64/bits/rseq.h | 43 ++++
sysdeps/unix/sysv/linux/arm/bits/rseq.h | 83 +++++++
sysdeps/unix/sysv/linux/bits/rseq.h | 29 +++
sysdeps/unix/sysv/linux/mips/bits/rseq.h | 62 +++++
sysdeps/unix/sysv/linux/powerpc/bits/rseq.h | 37 +++
sysdeps/unix/sysv/linux/rseq-internal.h | 45 ++++
sysdeps/unix/sysv/linux/s390/bits/rseq.h | 37 +++
sysdeps/unix/sysv/linux/sys/rseq.h | 174 +++++++++++++
sysdeps/unix/sysv/linux/tst-rseq-nptl.c | 260 ++++++++++++++++++++
sysdeps/unix/sysv/linux/tst-rseq.c | 64 +++++
sysdeps/unix/sysv/linux/tst-rseq.h | 57 +++++
sysdeps/unix/sysv/linux/x86/bits/rseq.h | 30 +++
17 files changed, 958 insertions(+), 3 deletions(-)
create mode 100644 sysdeps/unix/sysv/linux/aarch64/bits/rseq.h
create mode 100644 sysdeps/unix/sysv/linux/arm/bits/rseq.h
create mode 100644 sysdeps/unix/sysv/linux/bits/rseq.h
create mode 100644 sysdeps/unix/sysv/linux/mips/bits/rseq.h
create mode 100644 sysdeps/unix/sysv/linux/powerpc/bits/rseq.h
create mode 100644 sysdeps/unix/sysv/linux/rseq-internal.h
create mode 100644 sysdeps/unix/sysv/linux/s390/bits/rseq.h
create mode 100644 sysdeps/unix/sysv/linux/sys/rseq.h
create mode 100644 sysdeps/unix/sysv/linux/tst-rseq-nptl.c
create mode 100644 sysdeps/unix/sysv/linux/tst-rseq.c
create mode 100644 sysdeps/unix/sysv/linux/tst-rseq.h
create mode 100644 sysdeps/unix/sysv/linux/x86/bits/rseq.h
diff --git a/csu/libc-tls.c b/csu/libc-tls.c
index 5515204863..0d8146c0fa 100644
--- a/csu/libc-tls.c
+++ b/csu/libc-tls.c
@@ -112,6 +112,12 @@ __libc_setup_tls (void)
size_t tcb_offset;
const ElfW(Phdr) *phdr;
+ /* libc.so with rseq has TLS with 32-byte alignment. Static dlopen
+ requires at least 32-byte alignment as well, otherwise loading
+ libc.so will always fail. */
+ if (max_align < 32)
+ max_align = 32;
+
struct link_map *main_map = GL(dl_ns)[LM_ID_BASE]._ns_loaded;
__tls_pre_init_tp ();
diff --git a/nptl/descr.h b/nptl/descr.h
index af2a6ab87a..92db305913 100644
--- a/nptl/descr.h
+++ b/nptl/descr.h
@@ -34,6 +34,7 @@
#include <bits/types/res_state.h>
#include <kernel-features.h>
#include <tls-internal-struct.h>
+#include <sys/rseq.h>
#ifndef TCB_ALIGNMENT
# define TCB_ALIGNMENT 32
@@ -406,6 +407,9 @@ struct pthread
/* Used on strsignal. */
struct tls_internal_t tls_state;
+ /* rseq area registered with the kernel. */
+ struct rseq rseq_area;
+
/* This member must be last. */
char end_padding[];
diff --git a/nptl/pthread_create.c b/nptl/pthread_create.c
index bad9eeb52f..ea0d79341e 100644
--- a/nptl/pthread_create.c
+++ b/nptl/pthread_create.c
@@ -32,6 +32,7 @@
#include <default-sched.h>
#include <futex-internal.h>
#include <tls-setup.h>
+#include <rseq-internal.h>
#include "libioP.h"
#include <sys/single_threaded.h>
#include <version.h>
@@ -366,6 +367,9 @@ start_thread (void *arg)
/* Initialize pointers to locale data. */
__ctype_init ();
+ /* Register rseq TLS to the kernel. */
+ rseq_register_current_thread (pd);
+
#ifndef __ASSUME_SET_ROBUST_LIST
if (__nptl_set_robust_list_avail)
#endif
@@ -571,6 +575,15 @@ out:
process is really dead since 'clone' got passed the CLONE_CHILD_CLEARTID
flag. The 'tid' field in the TCB will be set to zero.
+ rseq TLS is still registered at this point. Rely on implicit
+ unregistration performed by the kernel on thread teardown. This is not a
+ problem because the rseq TLS lives on the stack, and the stack outlives
+ the thread. If TCB allocation is ever changed, additional steps may be
+ required, such as performing explicit rseq unregistration before
+ reclaiming the rseq TLS area memory. It is NOT sufficient to block
+ signals because the kernel may write to the rseq area even without
+ signals.
+
The exit code is zero since in case all threads exit by calling
'pthread_exit' the exit status must be 0 (zero). */
while (1)
diff --git a/sysdeps/nptl/dl-tls_init_tp.c b/sysdeps/nptl/dl-tls_init_tp.c
index ca494dd3a5..fedb876fdb 100644
--- a/sysdeps/nptl/dl-tls_init_tp.c
+++ b/sysdeps/nptl/dl-tls_init_tp.c
@@ -21,6 +21,7 @@
#include <list.h>
#include <pthreadP.h>
#include <tls.h>
+#include <rseq-internal.h>
#ifndef __ASSUME_SET_ROBUST_LIST
bool __nptl_set_robust_list_avail;
@@ -57,11 +58,12 @@ __tls_pre_init_tp (void)
void
__tls_init_tp (void)
{
+ struct pthread *pd = THREAD_SELF;
+
/* Set up thread stack list management. */
- list_add (&THREAD_SELF->list, &GL (dl_stack_user));
+ list_add (&pd->list, &GL (dl_stack_user));
/* Early initialization of the TCB. */
- struct pthread *pd = THREAD_SELF;
pd->tid = INTERNAL_SYSCALL_CALL (set_tid_address, &pd->tid);
THREAD_SETMEM (pd, specific[0], &pd->specific_1stblock[0]);
THREAD_SETMEM (pd, user_stack, true);
@@ -90,6 +92,8 @@ __tls_init_tp (void)
}
}
+ rseq_register_current_thread (pd);
+
/* Set initial thread's stack block from 0 up to __libc_stack_end.
It will be bigger than it actually is, but for unwind.c/pt-longjmp.c
purposes this is good enough. */
diff --git a/sysdeps/unix/sysv/linux/Makefile b/sysdeps/unix/sysv/linux/Makefile
index 29c6c78f98..eb0f5fc021 100644
--- a/sysdeps/unix/sysv/linux/Makefile
+++ b/sysdeps/unix/sysv/linux/Makefile
@@ -131,7 +131,10 @@ ifeq ($(have-GLIBC_2.27)$(build-shared),yesyes)
tests += tst-ofdlocks-compat
endif
-tests-internal += tst-sigcontext-get_pc
+tests-internal += \
+ tst-rseq \
+ tst-sigcontext-get_pc \
+ # tests-internal
tests-time64 += \
tst-adjtimex-time64 \
@@ -357,4 +360,8 @@ endif
ifeq ($(subdir),nptl)
tests += tst-align-clone tst-getpid1
+
+# tst-rseq-nptl is an internal test because it requires a definition of
+# __NR_rseq from the internal system call list.
+tests-internal += tst-rseq-nptl
endif
diff --git a/sysdeps/unix/sysv/linux/aarch64/bits/rseq.h b/sysdeps/unix/sysv/linux/aarch64/bits/rseq.h
new file mode 100644
index 0000000000..9ba92725c7
--- /dev/null
+++ b/sysdeps/unix/sysv/linux/aarch64/bits/rseq.h
@@ -0,0 +1,43 @@
+/* Restartable Sequences Linux aarch64 architecture header.
+ Copyright (C) 2021 Free Software Foundation, Inc.
+
+ 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/>. */
+
+#ifndef _SYS_RSEQ_H
+# error "Never use <bits/rseq.h> directly; include <sys/rseq.h> instead."
+#endif
+
+/* RSEQ_SIG is a signature required before each abort handler code.
+
+ It is a 32-bit value that maps to actual architecture code compiled
+ into applications and libraries. It needs to be defined for each
+ architecture. When choosing this value, it needs to be taken into
+ account that generating invalid instructions may have ill effects on
+ tools like objdump, and may also have impact on the CPU speculative
+ execution efficiency in some cases.
+
+ aarch64 -mbig-endian generates mixed endianness code vs data:
+ little-endian code and big-endian data. Ensure the RSEQ_SIG signature
+ matches code endianness. */
+
+#define RSEQ_SIG_CODE 0xd428bc00 /* BRK #0x45E0. */
+
+#ifdef __AARCH64EB__
+# define RSEQ_SIG_DATA 0x00bc28d4 /* BRK #0x45E0. */
+#else
+# define RSEQ_SIG_DATA RSEQ_SIG_CODE
+#endif
+
+#define RSEQ_SIG RSEQ_SIG_DATA
diff --git a/sysdeps/unix/sysv/linux/arm/bits/rseq.h b/sysdeps/unix/sysv/linux/arm/bits/rseq.h
new file mode 100644
index 0000000000..0542b26f6a
--- /dev/null
+++ b/sysdeps/unix/sysv/linux/arm/bits/rseq.h
@@ -0,0 +1,83 @@
+/* Restartable Sequences Linux arm architecture header.
+ Copyright (C) 2021 Free Software Foundation, Inc.
+
+ 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/>. */
+
+#ifndef _SYS_RSEQ_H
+# error "Never use <bits/rseq.h> directly; include <sys/rseq.h> instead."
+#endif
+
+/*
+ RSEQ_SIG is a signature required before each abort handler code.
+
+ It is a 32-bit value that maps to actual architecture code compiled
+ into applications and libraries. It needs to be defined for each
+ architecture. When choosing this value, it needs to be taken into
+ account that generating invalid instructions may have ill effects on
+ tools like objdump, and may also have impact on the CPU speculative
+ execution efficiency in some cases.
+
+ - ARM little endian
+
+ RSEQ_SIG uses the udf A32 instruction with an uncommon immediate operand
+ value 0x5de3. This traps if user-space reaches this instruction by mistake,
+ and the uncommon operand ensures the kernel does not move the instruction
+ pointer to attacker-controlled code on rseq abort.
+
+ The instruction pattern in the A32 instruction set is:
+
+ e7f5def3 udf #24035 ; 0x5de3
+
+ This translates to the following instruction pattern in the T16 instruction
+ set:
+
+ little endian:
+ def3 udf #243 ; 0xf3
+ e7f5 b.n <7f5>
+
+ - ARMv6+ big endian (BE8):
+
+ ARMv6+ -mbig-endian generates mixed endianness code vs data: little-endian
+ code and big-endian data. The data value of the signature needs to have its
+ byte order reversed to generate the trap instruction:
+
+ Data: 0xf3def5e7
+
+ Translates to this A32 instruction pattern:
+
+ e7f5def3 udf #24035 ; 0x5de3
+
+ Translates to this T16 instruction pattern:
+
+ def3 udf #243 ; 0xf3
+ e7f5 b.n <7f5>
+
+ - Prior to ARMv6 big endian (BE32):
+
+ Prior to ARMv6, -mbig-endian generates big-endian code and data
+ (which match), so the endianness of the data representation of the
+ signature should not be reversed. However, the choice between BE32
+ and BE8 is done by the linker, so we cannot know whether code and
+ data endianness will be mixed before the linker is invoked. So rather
+ than try to play tricks with the linker, the rseq signature is simply
+ data (not a trap instruction) prior to ARMv6 on big endian. This is
+ why the signature is expressed as data (.word) rather than as
+ instruction (.inst) in assembler. */
+
+#ifdef __ARMEB__
+# define RSEQ_SIG 0xf3def5e7 /* udf #24035 ; 0x5de3 (ARMv6+) */
+#else
+# define RSEQ_SIG 0xe7f5def3 /* udf #24035 ; 0x5de3 */
+#endif
diff --git a/sysdeps/unix/sysv/linux/bits/rseq.h b/sysdeps/unix/sysv/linux/bits/rseq.h
new file mode 100644
index 0000000000..46cf5d1c74
--- /dev/null
+++ b/sysdeps/unix/sysv/linux/bits/rseq.h
@@ -0,0 +1,29 @@
+/* Restartable Sequences architecture header. Stub version.
+ Copyright (C) 2021 Free Software Foundation, Inc.
+
+ 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/>. */
+
+#ifndef _SYS_RSEQ_H
+# error "Never use <bits/rseq.h> directly; include <sys/rseq.h> instead."
+#endif
+
+/* RSEQ_SIG is a signature required before each abort handler code.
+
+ It is a 32-bit value that maps to actual architecture code compiled
+ into applications and libraries. It needs to be defined for each
+ architecture. When choosing this value, it needs to be taken into
+ account that generating invalid instructions may have ill effects on
+ tools like objdump, and may also have impact on the CPU speculative
+ execution efficiency in some cases. */
diff --git a/sysdeps/unix/sysv/linux/mips/bits/rseq.h b/sysdeps/unix/sysv/linux/mips/bits/rseq.h
new file mode 100644
index 0000000000..a9defee568
--- /dev/null
+++ b/sysdeps/unix/sysv/linux/mips/bits/rseq.h
@@ -0,0 +1,62 @@
+/* Restartable Sequences Linux mips architecture header.
+ Copyright (C) 2021 Free Software Foundation, Inc.
+
+ 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/>. */
+
+#ifndef _SYS_RSEQ_H
+# error "Never use <bits/rseq.h> directly; include <sys/rseq.h> instead."
+#endif
+
+/* RSEQ_SIG is a signature required before each abort handler code.
+
+ It is a 32-bit value that maps to actual architecture code compiled
+ into applications and libraries. It needs to be defined for each
+ architecture. When choosing this value, it needs to be taken into
+ account that generating invalid instructions may have ill effects on
+ tools like objdump, and may also have impact on the CPU speculative
+ execution efficiency in some cases.
+
+ RSEQ_SIG uses the break instruction. The instruction pattern is:
+
+ On MIPS:
+ 0350000d break 0x350
+
+ On nanoMIPS:
+ 00100350 break 0x350
+
+ On microMIPS:
+ 0000d407 break 0x350
+
+ For nanoMIPS32 and microMIPS, the instruction stream is encoded as
+ 16-bit halfwords, so the signature halfwords need to be swapped
+ accordingly for little-endian. */
+
+#if defined (__nanomips__)
+# ifdef __MIPSEL__
+# define RSEQ_SIG 0x03500010
+# else
+# define RSEQ_SIG 0x00100350
+# endif
+#elif defined (__mips_micromips)
+# ifdef __MIPSEL__
+# define RSEQ_SIG 0xd4070000
+# else
+# define RSEQ_SIG 0x0000d407
+# endif
+#elif defined (__mips__)
+# define RSEQ_SIG 0x0350000d
+#else
+/* Unknown MIPS architecture. */
+#endif
diff --git a/sysdeps/unix/sysv/linux/powerpc/bits/rseq.h b/sysdeps/unix/sysv/linux/powerpc/bits/rseq.h
new file mode 100644
index 0000000000..05b3cf7b8f
--- /dev/null
+++ b/sysdeps/unix/sysv/linux/powerpc/bits/rseq.h
@@ -0,0 +1,37 @@
+/* Restartable Sequences Linux powerpc architecture header.
+ Copyright (C) 2021 Free Software Foundation, Inc.
+
+ 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/>. */
+
+#ifndef _SYS_RSEQ_H
+# error "Never use <bits/rseq.h> directly; include <sys/rseq.h> instead."
+#endif
+
+/* RSEQ_SIG is a signature required before each abort handler code.
+
+ It is a 32-bit value that maps to actual architecture code compiled
+ into applications and libraries. It needs to be defined for each
+ architecture. When choosing this value, it needs to be taken into
+ account that generating invalid instructions may have ill effects on
+ tools like objdump, and may also have impact on the CPU speculative
+ execution efficiency in some cases.
+
+ RSEQ_SIG uses the following trap instruction:
+
+ powerpc-be: 0f e5 00 0b twui r5,11
+ powerpc64-le: 0b 00 e5 0f twui r5,11
+ powerpc64-be: 0f e5 00 0b twui r5,11 */
+
+#define RSEQ_SIG 0x0fe5000b
diff --git a/sysdeps/unix/sysv/linux/rseq-internal.h b/sysdeps/unix/sysv/linux/rseq-internal.h
new file mode 100644
index 0000000000..909f547825
--- /dev/null
+++ b/sysdeps/unix/sysv/linux/rseq-internal.h
@@ -0,0 +1,45 @@
+/* Restartable Sequences internal API. Linux implementation.
+ Copyright (C) 2021 Free Software Foundation, Inc.
+
+ 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/>. */
+
+#ifndef RSEQ_INTERNAL_H
+#define RSEQ_INTERNAL_H
+
+#include <sysdep.h>
+#include <errno.h>
+#include <kernel-features.h>
+#include <stdio.h>
+#include <sys/rseq.h>
+
+#ifdef RSEQ_SIG
+static inline void
+rseq_register_current_thread (struct pthread *self)
+{
+ int ret = INTERNAL_SYSCALL_CALL (rseq,
+ &self->rseq_area, sizeof (self->rseq_area),
+ 0, RSEQ_SIG);
+ if (INTERNAL_SYSCALL_ERROR_P (ret))
+ THREAD_SETMEM (self, rseq_area.cpu_id, RSEQ_CPU_ID_REGISTRATION_FAILED);
+}
+#else /* RSEQ_SIG */
+static inline void
+rseq_register_current_thread (struct pthread *self)
+{
+ THREAD_SETMEM (self, rseq_area.cpu_id, RSEQ_CPU_ID_REGISTRATION_FAILED);
+}
+#endif /* RSEQ_SIG */
+
+#endif /* rseq-internal.h */
diff --git a/sysdeps/unix/sysv/linux/s390/bits/rseq.h b/sysdeps/unix/sysv/linux/s390/bits/rseq.h
new file mode 100644
index 0000000000..3030e38f40
--- /dev/null
+++ b/sysdeps/unix/sysv/linux/s390/bits/rseq.h
@@ -0,0 +1,37 @@
+/* Restartable Sequences Linux s390 architecture header.
+ Copyright (C) 2021 Free Software Foundation, Inc.
+
+ 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/>. */
+
+#ifndef _SYS_RSEQ_H
+# error "Never use <bits/rseq.h> directly; include <sys/rseq.h> instead."
+#endif
+
+/* RSEQ_SIG is a signature required before each abort handler code.
+
+ It is a 32-bit value that maps to actual architecture code compiled
+ into applications and libraries. It needs to be defined for each
+ architecture. When choosing this value, it needs to be taken into
+ account that generating invalid instructions may have ill effects on
+ tools like objdump, and may also have impact on the CPU speculative
+ execution efficiency in some cases.
+
+ RSEQ_SIG uses the trap4 instruction. As Linux does not make use of the
+ access-register mode nor the linkage stack this instruction will always
+ cause a special-operation exception (the trap-enabled bit in the DUCT
+ is and will stay 0). The instruction pattern is
+ b2 ff 0f ff trap4 4095(%r0) */
+
+#define RSEQ_SIG 0xB2FF0FFF
diff --git a/sysdeps/unix/sysv/linux/sys/rseq.h b/sysdeps/unix/sysv/linux/sys/rseq.h
new file mode 100644
index 0000000000..c8edff50d4
--- /dev/null
+++ b/sysdeps/unix/sysv/linux/sys/rseq.h
@@ -0,0 +1,174 @@
+/* Restartable Sequences exported symbols. Linux header.
+ Copyright (C) 2021 Free Software Foundation, Inc.
+
+ 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/>. */
+
+#ifndef _SYS_RSEQ_H
+#define _SYS_RSEQ_H 1
+
+/* Architecture-specific rseq signature. */
+#include <bits/rseq.h>
+
+#include <stdint.h>
+#include <sys/cdefs.h>
+#include <bits/endian.h>
+
+#ifdef __has_include
+# if __has_include ("linux/rseq.h")
+# define __GLIBC_HAVE_KERNEL_RSEQ
+# endif
+#else
+# include <linux/version.h>
+# if LINUX_VERSION_CODE >= KERNEL_VERSION (4, 18, 0)
+# define __GLIBC_HAVE_KERNEL_RSEQ
+# endif
+#endif
+
+#ifdef __GLIBC_HAVE_KERNEL_RSEQ
+/* We use the structures declarations from the kernel headers. */
+# include <linux/rseq.h>
+#else /* __GLIBC_HAVE_KERNEL_RSEQ */
+/* We use a copy of the include/uapi/linux/rseq.h kernel header. */
+
+enum rseq_cpu_id_state
+ {
+ RSEQ_CPU_ID_UNINITIALIZED = -1,
+ RSEQ_CPU_ID_REGISTRATION_FAILED = -2,
+ };
+
+enum rseq_flags
+ {
+ RSEQ_FLAG_UNREGISTER = (1 << 0),
+ };
+
+enum rseq_cs_flags_bit
+ {
+ RSEQ_CS_FLAG_NO_RESTART_ON_PREEMPT_BIT = 0,
+ RSEQ_CS_FLAG_NO_RESTART_ON_SIGNAL_BIT = 1,
+ RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE_BIT = 2,
+ };
+
+enum rseq_cs_flags
+ {
+ RSEQ_CS_FLAG_NO_RESTART_ON_PREEMPT =
+ (1U << RSEQ_CS_FLAG_NO_RESTART_ON_PREEMPT_BIT),
+ RSEQ_CS_FLAG_NO_RESTART_ON_SIGNAL =
+ (1U << RSEQ_CS_FLAG_NO_RESTART_ON_SIGNAL_BIT),
+ RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE =
+ (1U << RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE_BIT),
+ };
+
+/* struct rseq_cs is aligned on 32 bytes to ensure it is always
+ contained within a single cache-line. It is usually declared as
+ link-time constant data. */
+struct rseq_cs
+ {
+ /* Version of this structure. */
+ uint32_t version;
+ /* enum rseq_cs_flags. */
+ uint32_t flags;
+ uint64_t start_ip;
+ /* Offset from start_ip. */
+ uint64_t post_commit_offset;
+ uint64_t abort_ip;
+ } __attribute__ ((__aligned__ (32)));
+
+/* struct rseq is aligned on 32 bytes to ensure it is always
+ contained within a single cache-line.
+
+ A single struct rseq per thread is allowed. */
+struct rseq
+ {
+ /* Restartable sequences cpu_id_start field. Updated by the
+ kernel. Read by user-space with single-copy atomicity
+ semantics. This field should only be read by the thread which
+ registered this data structure. Aligned on 32-bit. Always
+ contains a value in the range of possible CPUs, although the
+ value may not be the actual current CPU (e.g. if rseq is not
+ initialized). This CPU number value should always be compared
+ against the value of the cpu_id field before performing a rseq
+ commit or returning a value read from a data structure indexed
+ using the cpu_id_start value. */
+ uint32_t cpu_id_start;
+ /* Restartable sequences cpu_id field. Updated by the kernel.
+ Read by user-space with single-copy atomicity semantics. This
+ field should only be read by the thread which registered this
+ data structure. Aligned on 32-bit. Values
+ RSEQ_CPU_ID_UNINITIALIZED and RSEQ_CPU_ID_REGISTRATION_FAILED
+ have a special semantic: the former means "rseq uninitialized",
+ and latter means "rseq initialization failed". This value is
+ meant to be read within rseq critical sections and compared
+ with the cpu_id_start value previously read, before performing
+ the commit instruction, or read and compared with the
+ cpu_id_start value before returning a value loaded from a data
+ structure indexed using the cpu_id_start value. */
+ uint32_t cpu_id;
+ /* Restartable sequences rseq_cs field.
+
+ Contains NULL when no critical section is active for the current
+ thread, or holds a pointer to the currently active struct rseq_cs.
+
+ Updated by user-space, which sets the address of the currently
+ active rseq_cs at the beginning of assembly instruction sequence
+ block, and set to NULL by the kernel when it restarts an assembly
+ instruction sequence block, as well as when the kernel detects that
+ it is preempting or delivering a signal outside of the range
+ targeted by the rseq_cs. Also needs to be set to NULL by user-space
+ before reclaiming memory that contains the targeted struct rseq_cs.
+
+ Read and set by the kernel. Set by user-space with single-copy
+ atomicity semantics. This field should only be updated by the
+ thread which registered this data structure. Aligned on 64-bit. */
+ union
+ {
+ uint64_t ptr64;
+# ifdef __LP64__
+ uint64_t ptr;
+# else /* __LP64__ */
+ struct
+ {
+#if __BYTE_ORDER == __BIG_ENDIAN
+ uint32_t padding; /* Initialized to zero. */
+ uint32_t ptr32;
+# else /* LITTLE */
+ uint32_t ptr32;
+ uint32_t padding; /* Initialized to zero. */
+# endif /* ENDIAN */
+ } ptr;
+# endif /* __LP64__ */
+ } rseq_cs;
+
+ /* Restartable sequences flags field.
+
+ This field should only be updated by the thread which
+ registered this data structure. Read by the kernel.
+ Mainly used for single-stepping through rseq critical sections
+ with debuggers.
+
+ - RSEQ_CS_FLAG_NO_RESTART_ON_PREEMPT
+ Inhibit instruction sequence block restart on preemption
+ for this thread.
+ - RSEQ_CS_FLAG_NO_RESTART_ON_SIGNAL
+ Inhibit instruction sequence block restart on signal
+ delivery for this thread.
+ - RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE
+ Inhibit instruction sequence block restart on migration for
+ this thread. */
+ uint32_t flags;
+ } __attribute__ ((__aligned__ (32)));
+
+#endif /* __GLIBC_HAVE_KERNEL_RSEQ */
+
+#endif /* sys/rseq.h */
diff --git a/sysdeps/unix/sysv/linux/tst-rseq-nptl.c b/sysdeps/unix/sysv/linux/tst-rseq-nptl.c
new file mode 100644
index 0000000000..d31d94445c
--- /dev/null
+++ b/sysdeps/unix/sysv/linux/tst-rseq-nptl.c
@@ -0,0 +1,260 @@
+/* Restartable Sequences NPTL test.
+ Copyright (C) 2021 Free Software Foundation, Inc.
+
+ 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/>. */
+
+/* These tests validate that rseq is registered from various execution
+ contexts (main thread, destructor, other threads, other threads created
+ from destructor, forked process (without exec), pthread_atfork handlers,
+ pthread setspecific destructors, signal handlers, atexit handlers).
+
+ See the Linux kernel selftests for extensive rseq stress-tests. */
+
+#include <stdio.h>
+#include <support/check.h>
+#include <support/xthread.h>
+#include <sys/rseq.h>
+#include <unistd.h>
+
+#ifdef RSEQ_SIG
+# include <array_length.h>
+# include <errno.h>
+# include <error.h>
+# include <pthread.h>
+# include <signal.h>
+# include <stdlib.h>
+# include <string.h>
+# include <support/namespace.h>
+# include <support/xsignal.h>
+# include <syscall.h>
+# include <sys/types.h>
+# include <sys/wait.h>
+# include "tst-rseq.h"
+
+static pthread_key_t rseq_test_key;
+
+static void
+atfork_prepare (void)
+{
+ if (!rseq_thread_registered ())
+ {
+ printf ("error: rseq not registered in pthread atfork prepare\n");
+ support_record_failure ();
+ }
+}
+
+static void
+atfork_parent (void)
+{
+ if (!rseq_thread_registered ())
+ {
+ printf ("error: rseq not registered in pthread atfork parent\n");
+ support_record_failure ();
+ }
+}
+
+static void
+atfork_child (void)
+{
+ if (!rseq_thread_registered ())
+ {
+ printf ("error: rseq not registered in pthread atfork child\n");
+ support_record_failure ();
+ }
+}
+
+static void
+rseq_key_destructor (void *arg)
+{
+ /* Cannot use deferred failure reporting after main returns. */
+ if (!rseq_thread_registered ())
+ FAIL_EXIT1 ("rseq not registered in pthread key destructor");
+}
+
+static void
+atexit_handler (void)
+{
+ /* Cannot use deferred failure reporting after main returns. */
+ if (!rseq_thread_registered ())
+ FAIL_EXIT1 ("rseq not registered in atexit handler");
+}
+
+/* Used to avoid -Werror=stringop-overread warning with
+ pthread_setspecific and GCC 11. */
+static char one = 1;
+
+static void
+do_rseq_main_test (void)
+{
+ TEST_COMPARE (atexit (atexit_handler), 0);
+ rseq_test_key = xpthread_key_create (rseq_key_destructor);
+ TEST_COMPARE (pthread_atfork (atfork_prepare, atfork_parent, atfork_child), 0);
+ xraise (SIGUSR1);
+ TEST_COMPARE (pthread_setspecific (rseq_test_key, &one), 0);
+ TEST_VERIFY_EXIT (rseq_thread_registered ());
+}
+
+static void
+cancel_routine (void *arg)
+{
+ if (!rseq_thread_registered ())
+ {
+ printf ("error: rseq not registered in cancel routine\n");
+ support_record_failure ();
+ }
+}
+
+static pthread_barrier_t cancel_thread_barrier;
+static pthread_cond_t cancel_thread_cond = PTHREAD_COND_INITIALIZER;
+static pthread_mutex_t cancel_thread_mutex = PTHREAD_MUTEX_INITIALIZER;
+
+static void
+test_cancel_thread (void)
+{
+ pthread_cleanup_push (cancel_routine, NULL);
+ (void) xpthread_barrier_wait (&cancel_thread_barrier);
+ /* Wait forever until cancellation. */
+ xpthread_cond_wait (&cancel_thread_cond, &cancel_thread_mutex);
+ pthread_cleanup_pop (0);
+}
+
+static void *
+thread_function (void * arg)
+{
+ int i = (int) (intptr_t) arg;
+
+ xraise (SIGUSR1);
+ if (i == 0)
+ test_cancel_thread ();
+ TEST_COMPARE (pthread_setspecific (rseq_test_key, &one), 0);
+ return rseq_thread_registered () ? NULL : (void *) 1l;
+}
+
+static void
+sighandler (int sig)
+{
+ if (!rseq_thread_registered ())
+ {
+ printf ("error: rseq not registered in signal handler\n");
+ support_record_failure ();
+ }
+}
+
+static void
+setup_signals (void)
+{
+ struct sigaction sa;
+
+ sigemptyset (&sa.sa_mask);
+ sigaddset (&sa.sa_mask, SIGUSR1);
+ sa.sa_flags = 0;
+ sa.sa_handler = sighandler;
+ xsigaction (SIGUSR1, &sa, NULL);
+}
+
+static int
+do_rseq_threads_test (int nr_threads)
+{
+ pthread_t th[nr_threads];
+ int i;
+ int result = 0;
+
+ xpthread_barrier_init (&cancel_thread_barrier, NULL, 2);
+
+ for (i = 0; i < nr_threads; ++i)
+ th[i] = xpthread_create (NULL, thread_function,
+ (void *) (intptr_t) i);
+
+ (void) xpthread_barrier_wait (&cancel_thread_barrier);
+
+ xpthread_cancel (th[0]);
+
+ for (i = 0; i < nr_threads; ++i)
+ {
+ void *v;
+
+ v = xpthread_join (th[i]);
+ if (i != 0 && v != NULL)
+ {
+ printf ("error: join %d successful, but child failed\n", i);
+ result = 1;
+ }
+ else if (i == 0 && v == NULL)
+ {
+ printf ("error: join %d successful, child did not fail as expected\n", i);
+ result = 1;
+ }
+ }
+
+ xpthread_barrier_destroy (&cancel_thread_barrier);
+
+ return result;
+}
+
+static void
+subprocess_callback (void *closure)
+{
+ do_rseq_main_test ();
+}
+
+static void
+do_rseq_fork_test (void)
+{
+ support_isolate_in_subprocess (subprocess_callback, NULL);
+}
+
+static int
+do_rseq_test (void)
+{
+ int t[] = { 1, 2, 6, 5, 4, 3, 50 };
+ int i, result = 0;
+
+ if (!rseq_available ())
+ FAIL_UNSUPPORTED ("kernel does not support rseq, skipping test");
+ setup_signals ();
+ xraise (SIGUSR1);
+ do_rseq_main_test ();
+ for (i = 0; i < array_length (t); i++)
+ if (do_rseq_threads_test (t[i]))
+ result = 1;
+ do_rseq_fork_test ();
+ return result;
+}
+
+static void __attribute__ ((destructor))
+do_rseq_destructor_test (void)
+{
+ /* Cannot use deferred failure reporting after main returns. */
+ if (do_rseq_test ())
+ FAIL_EXIT1 ("rseq not registered within destructor");
+ xpthread_key_delete (rseq_test_key);
+}
+
+#else /* RSEQ_SIG */
+static int
+do_rseq_test (void)
+{
+ FAIL_UNSUPPORTED ("glibc does not define RSEQ_SIG, skipping test");
+ return 0;
+}
+#endif /* RSEQ_SIG */
+
+static int
+do_test (void)
+{
+ return do_rseq_test ();
+}
+
+#include <support/test-driver.c>
diff --git a/sysdeps/unix/sysv/linux/tst-rseq.c b/sysdeps/unix/sysv/linux/tst-rseq.c
new file mode 100644
index 0000000000..926376b6a5
--- /dev/null
+++ b/sysdeps/unix/sysv/linux/tst-rseq.c
@@ -0,0 +1,64 @@
+/* Restartable Sequences single-threaded tests.
+ Copyright (C) 2021 Free Software Foundation, Inc.
+
+ 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/>. */
+
+/* These tests validate that rseq is registered from main in an executable
+ not linked against libpthread. */
+
+#include <support/check.h>
+#include <stdio.h>
+#include <sys/rseq.h>
+#include <unistd.h>
+
+#ifdef RSEQ_SIG
+# include <errno.h>
+# include <error.h>
+# include <stdlib.h>
+# include <string.h>
+# include <syscall.h>
+# include "tst-rseq.h"
+
+static void
+do_rseq_main_test (void)
+{
+ TEST_VERIFY_EXIT (rseq_thread_registered ());
+}
+
+static void
+do_rseq_test (void)
+{
+ if (!rseq_available ())
+ {
+ FAIL_UNSUPPORTED ("kernel does not support rseq, skipping test");
+ }
+ do_rseq_main_test ();
+}
+#else /* RSEQ_SIG */
+static void
+do_rseq_test (void)
+{
+ FAIL_UNSUPPORTED ("glibc does not define RSEQ_SIG, skipping test");
+}
+#endif /* RSEQ_SIG */
+
+static int
+do_test (void)
+{
+ do_rseq_test ();
+ return 0;
+}
+
+#include <support/test-driver.c>
diff --git a/sysdeps/unix/sysv/linux/tst-rseq.h b/sysdeps/unix/sysv/linux/tst-rseq.h
new file mode 100644
index 0000000000..12d4639874
--- /dev/null
+++ b/sysdeps/unix/sysv/linux/tst-rseq.h
@@ -0,0 +1,57 @@
+/* Restartable Sequences tests header.
+ Copyright (C) 2021 Free Software Foundation, Inc.
+
+ 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 <errno.h>
+#include <error.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <support/check.h>
+#include <syscall.h>
+#include <sys/rseq.h>
+#include <tls.h>
+
+static inline bool
+rseq_thread_registered (void)
+{
+ return THREAD_GETMEM (THREAD_SELF, rseq_area.cpu_id) >= 0;
+}
+
+static inline int
+sys_rseq (struct rseq *rseq_abi, uint32_t rseq_len, int flags, uint32_t sig)
+{
+ return syscall (__NR_rseq, rseq_abi, rseq_len, flags, sig);
+}
+
+static inline bool
+rseq_available (void)
+{
+ int rc;
+
+ rc = sys_rseq (NULL, 0, 0, 0);
+ if (rc != -1)
+ FAIL_EXIT1 ("Unexpected rseq return value %d", rc);
+ switch (errno)
+ {
+ case ENOSYS:
+ return false;
+ case EINVAL:
+ /* rseq is implemented, but detected an invalid rseq_len parameter. */
+ return true;
+ default:
+ FAIL_EXIT1 ("Unexpected rseq error %s", strerror (errno));
+ }
+}
diff --git a/sysdeps/unix/sysv/linux/x86/bits/rseq.h b/sysdeps/unix/sysv/linux/x86/bits/rseq.h
new file mode 100644
index 0000000000..9fc909e7c8
--- /dev/null
+++ b/sysdeps/unix/sysv/linux/x86/bits/rseq.h
@@ -0,0 +1,30 @@
+/* Restartable Sequences Linux x86 architecture header.
+ Copyright (C) 2021 Free Software Foundation, Inc.
+
+ 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/>. */
+
+#ifndef _SYS_RSEQ_H
+# error "Never use <bits/rseq.h> directly; include <sys/rseq.h> instead."
+#endif
+
+/* RSEQ_SIG is a signature required before each abort handler code.
+
+ RSEQ_SIG is used with the following reserved undefined instructions, which
+ trap in user-space:
+
+ x86-32: 0f b9 3d 53 30 05 53 ud1 0x53053053,%edi
+ x86-64: 0f b9 3d 53 30 05 53 ud1 0x53053053(%rip),%edi */
+
+#define RSEQ_SIG 0x53053053
--
2.33.1
^ permalink raw reply [flat|nested] 25+ messages in thread* Re: [PATCH 2/5] nptl: Add rseq registration
2021-12-06 13:46 ` [PATCH 2/5] nptl: Add rseq registration Florian Weimer
@ 2021-12-06 16:53 ` Mathieu Desnoyers
2021-12-06 17:10 ` Florian Weimer
2021-12-06 16:59 ` Mathieu Desnoyers
1 sibling, 1 reply; 25+ messages in thread
From: Mathieu Desnoyers @ 2021-12-06 16:53 UTC (permalink / raw)
To: Florian Weimer; +Cc: libc-alpha
----- On Dec 6, 2021, at 8:46 AM, Florian Weimer fweimer@redhat.com wrote:
> The rseq area is placed directly into struct pthread. rseq
> registration failure is not treated as an error, so it is possible
> that threads run with inconsistent registration status.
>
> <sys/rseq.h> is not yet installed as a public header.
>
> Co-Authored-By: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
> ---
[...]
> --- a/csu/libc-tls.c
> +++ b/csu/libc-tls.c
> @@ -112,6 +112,12 @@ __libc_setup_tls (void)
> size_t tcb_offset;
> const ElfW(Phdr) *phdr;
>
> + /* libc.so with rseq has TLS with 32-byte alignment. Static dlopen
> + requires at least 32-byte alignment as well, otherwise loading
> + libc.so will always fail. */
> + if (max_align < 32)
> + max_align = 32;
Now that the rseq area is move into struct pthread, do we still require a
32-byte alignment on the TLS ?
[...]
> --- a/nptl/pthread_create.c
> +++ b/nptl/pthread_create.c
[...]
>
> + rseq TLS is still registered at this point. Rely on implicit
> + unregistration performed by the kernel on thread teardown. This is not a
> + problem because the rseq TLS lives on the stack, and the stack outlives
I think we need to update this comment, because it does not make sense anymore to
discuss the placement of the rseq TLS now that it moves to struct pthread.
Is struct pthread placed on the thread stack ?
> + the thread. If TCB allocation is ever changed, additional steps may be
> + required, such as performing explicit rseq unregistration before
> + reclaiming the rseq TLS area memory. It is NOT sufficient to block
> + signals because the kernel may write to the rseq area even without
> + signals.
> +
Thanks,
Mathieu
--
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com
^ permalink raw reply [flat|nested] 25+ messages in thread
* Re: [PATCH 2/5] nptl: Add rseq registration
2021-12-06 16:53 ` Mathieu Desnoyers
@ 2021-12-06 17:10 ` Florian Weimer
0 siblings, 0 replies; 25+ messages in thread
From: Florian Weimer @ 2021-12-06 17:10 UTC (permalink / raw)
To: Mathieu Desnoyers; +Cc: libc-alpha
* Mathieu Desnoyers:
>> --- a/csu/libc-tls.c
>> +++ b/csu/libc-tls.c
>> @@ -112,6 +112,12 @@ __libc_setup_tls (void)
>> size_t tcb_offset;
>> const ElfW(Phdr) *phdr;
>>
>> + /* libc.so with rseq has TLS with 32-byte alignment. Static dlopen
>> + requires at least 32-byte alignment as well, otherwise loading
>> + libc.so will always fail. */
>> + if (max_align < 32)
>> + max_align = 32;
>
> Now that the rseq area is move into struct pthread, do we still require a
> 32-byte alignment on the TLS ?
Yes, we can remove that. In fact, I increased TCB_ALIGNMENT
specifically for that.
>> --- a/nptl/pthread_create.c
>> +++ b/nptl/pthread_create.c
>
> [...]
>>
>> + rseq TLS is still registered at this point. Rely on implicit
>> + unregistration performed by the kernel on thread teardown. This is not a
>> + problem because the rseq TLS lives on the stack, and the stack outlives
>
> I think we need to update this comment, because it does not make sense
> anymore to discuss the placement of the rseq TLS now that it moves to
> struct pthread.
>
> Is struct pthread placed on the thread stack ?
Yes, it is, along with the TLS data, which I thought the current wording
was still fine.
Thanks,
Florian
^ permalink raw reply [flat|nested] 25+ messages in thread
* Re: [PATCH 2/5] nptl: Add rseq registration
2021-12-06 13:46 ` [PATCH 2/5] nptl: Add rseq registration Florian Weimer
2021-12-06 16:53 ` Mathieu Desnoyers
@ 2021-12-06 16:59 ` Mathieu Desnoyers
2021-12-06 17:14 ` Florian Weimer
1 sibling, 1 reply; 25+ messages in thread
From: Mathieu Desnoyers @ 2021-12-06 16:59 UTC (permalink / raw)
To: Florian Weimer; +Cc: libc-alpha
----- On Dec 6, 2021, at 8:46 AM, Florian Weimer fweimer@redhat.com wrote:
[...]
> @@ -406,6 +407,9 @@ struct pthread
> /* Used on strsignal. */
> struct tls_internal_t tls_state;
>
> + /* rseq area registered with the kernel. */
> + struct rseq rseq_area;
The rseq UAPI requires that the fields within the rseq_area
are read-written with single-copy atomicity semantics.
So either we define a "volatile struct rseq" here, or we'll need
to wrap all accesses with the proper volatile casts, or use the
relaxed_mo atomic accesses.
Thanks,
Mathieu
> +
> /* This member must be last. */
> char end_padding[];
>
--
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com
^ permalink raw reply [flat|nested] 25+ messages in thread
* Re: [PATCH 2/5] nptl: Add rseq registration
2021-12-06 16:59 ` Mathieu Desnoyers
@ 2021-12-06 17:14 ` Florian Weimer
2021-12-06 18:52 ` Mathieu Desnoyers
0 siblings, 1 reply; 25+ messages in thread
From: Florian Weimer @ 2021-12-06 17:14 UTC (permalink / raw)
To: Mathieu Desnoyers; +Cc: libc-alpha
* Mathieu Desnoyers:
> ----- On Dec 6, 2021, at 8:46 AM, Florian Weimer fweimer@redhat.com wrote:
> [...]
>> @@ -406,6 +407,9 @@ struct pthread
>> /* Used on strsignal. */
>> struct tls_internal_t tls_state;
>>
>> + /* rseq area registered with the kernel. */
>> + struct rseq rseq_area;
>
> The rseq UAPI requires that the fields within the rseq_area
> are read-written with single-copy atomicity semantics.
>
> So either we define a "volatile struct rseq" here, or we'll need
> to wrap all accesses with the proper volatile casts, or use the
> relaxed_mo atomic accesses.
Under the C memory model, neither volatile nor relaxed MO result in
single-copy atomicity semantics. So I'm not sure what to make of this.
Surely switching to inline assembly on all targets is over the top.
I think we can rely on a plain read doing the right thing for us.
Thanks,
Florian
^ permalink raw reply [flat|nested] 25+ messages in thread
* Re: [PATCH 2/5] nptl: Add rseq registration
2021-12-06 17:14 ` Florian Weimer
@ 2021-12-06 18:52 ` Mathieu Desnoyers
2021-12-06 19:03 ` Florian Weimer
0 siblings, 1 reply; 25+ messages in thread
From: Mathieu Desnoyers @ 2021-12-06 18:52 UTC (permalink / raw)
To: Florian Weimer, paulmck, Boqun Feng, Peter Zijlstra
Cc: libc-alpha, linux-kernel
[ Adding other kernel rseq maintainers in CC. ]
----- On Dec 6, 2021, at 12:14 PM, Florian Weimer fweimer@redhat.com wrote:
> * Mathieu Desnoyers:
>
>> ----- On Dec 6, 2021, at 8:46 AM, Florian Weimer fweimer@redhat.com wrote:
>> [...]
>>> @@ -406,6 +407,9 @@ struct pthread
>>> /* Used on strsignal. */
>>> struct tls_internal_t tls_state;
>>>
>>> + /* rseq area registered with the kernel. */
>>> + struct rseq rseq_area;
>>
>> The rseq UAPI requires that the fields within the rseq_area
>> are read-written with single-copy atomicity semantics.
>>
>> So either we define a "volatile struct rseq" here, or we'll need
>> to wrap all accesses with the proper volatile casts, or use the
>> relaxed_mo atomic accesses.
>
> Under the C memory model, neither volatile nor relaxed MO result in
> single-copy atomicity semantics. So I'm not sure what to make of this.
> Surely switching to inline assembly on all targets is over the top.
>
> I think we can rely on a plain read doing the right thing for us.
AFAIU, the plain read does not prevent the compiler from re-loading the
value in case of high register pressure.
Accesses to rseq fields such as cpu_id need to be done as if those were
concurrently modified by a signal handler nesting on top of the user-space
code, with the particular twist that blocking signals has no effect on
concurrent updates.
I do not think we need to do the load in assembly. I was under the impression
that both volatile load and relaxed MO result in single-copy atomicity
semantics for an aligned pointer. Perhaps Paul, Peter, Boqun have something
to add here ?
Thanks,
Mathieu
--
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com
^ permalink raw reply [flat|nested] 25+ messages in thread
* Re: [PATCH 2/5] nptl: Add rseq registration
2021-12-06 18:52 ` Mathieu Desnoyers
@ 2021-12-06 19:03 ` Florian Weimer
2021-12-06 20:11 ` Paul E. McKenney
0 siblings, 1 reply; 25+ messages in thread
From: Florian Weimer @ 2021-12-06 19:03 UTC (permalink / raw)
To: Mathieu Desnoyers
Cc: paulmck, Boqun Feng, Peter Zijlstra, libc-alpha, linux-kernel
* Mathieu Desnoyers:
> [ Adding other kernel rseq maintainers in CC. ]
>
> ----- On Dec 6, 2021, at 12:14 PM, Florian Weimer fweimer@redhat.com wrote:
>
>> * Mathieu Desnoyers:
>>
>>> ----- On Dec 6, 2021, at 8:46 AM, Florian Weimer fweimer@redhat.com wrote:
>>> [...]
>>>> @@ -406,6 +407,9 @@ struct pthread
>>>> /* Used on strsignal. */
>>>> struct tls_internal_t tls_state;
>>>>
>>>> + /* rseq area registered with the kernel. */
>>>> + struct rseq rseq_area;
>>>
>>> The rseq UAPI requires that the fields within the rseq_area
>>> are read-written with single-copy atomicity semantics.
>>>
>>> So either we define a "volatile struct rseq" here, or we'll need
>>> to wrap all accesses with the proper volatile casts, or use the
>>> relaxed_mo atomic accesses.
>>
>> Under the C memory model, neither volatile nor relaxed MO result in
>> single-copy atomicity semantics. So I'm not sure what to make of this.
>> Surely switching to inline assembly on all targets is over the top.
>>
>> I think we can rely on a plain read doing the right thing for us.
>
> AFAIU, the plain read does not prevent the compiler from re-loading the
> value in case of high register pressure.
>
> Accesses to rseq fields such as cpu_id need to be done as if those were
> concurrently modified by a signal handler nesting on top of the user-space
> code, with the particular twist that blocking signals has no effect on
> concurrent updates.
>
> I do not think we need to do the load in assembly. I was under the impression
> that both volatile load and relaxed MO result in single-copy atomicity
> semantics for an aligned pointer. Perhaps Paul, Peter, Boqun have something
> to add here ?
The C memory model is broken and does not prevent out-of-thin-air
values. As far as I know, this breaks single-copy atomicity. In
practice, compilers will not exercise the latitude offered by the memory
model. volatile does not ensure absence of data races.
Using atomics or volatile would require us to materialize the thread
pointer, given the current internal interfaces we have, and I don't want
to do this because this is supposed to be performance-critical code.
The compiler barrier inherent to the function call will have to be
enough. I can add a comment to this effect:
/* This load has single-copy atomicity semantics (as required for
rseq) because the function call implies a compiler barrier. */
Thanks,
Florian
^ permalink raw reply [flat|nested] 25+ messages in thread* Re: [PATCH 2/5] nptl: Add rseq registration
2021-12-06 19:03 ` Florian Weimer
@ 2021-12-06 20:11 ` Paul E. McKenney
2021-12-06 20:26 ` Florian Weimer
0 siblings, 1 reply; 25+ messages in thread
From: Paul E. McKenney @ 2021-12-06 20:11 UTC (permalink / raw)
To: Florian Weimer
Cc: Mathieu Desnoyers, Boqun Feng, Peter Zijlstra, libc-alpha, linux-kernel
On Mon, Dec 06, 2021 at 08:03:26PM +0100, Florian Weimer wrote:
> * Mathieu Desnoyers:
>
> > [ Adding other kernel rseq maintainers in CC. ]
> >
> > ----- On Dec 6, 2021, at 12:14 PM, Florian Weimer fweimer@redhat.com wrote:
> >
> >> * Mathieu Desnoyers:
> >>
> >>> ----- On Dec 6, 2021, at 8:46 AM, Florian Weimer fweimer@redhat.com wrote:
> >>> [...]
> >>>> @@ -406,6 +407,9 @@ struct pthread
> >>>> /* Used on strsignal. */
> >>>> struct tls_internal_t tls_state;
> >>>>
> >>>> + /* rseq area registered with the kernel. */
> >>>> + struct rseq rseq_area;
> >>>
> >>> The rseq UAPI requires that the fields within the rseq_area
> >>> are read-written with single-copy atomicity semantics.
> >>>
> >>> So either we define a "volatile struct rseq" here, or we'll need
> >>> to wrap all accesses with the proper volatile casts, or use the
> >>> relaxed_mo atomic accesses.
> >>
> >> Under the C memory model, neither volatile nor relaxed MO result in
> >> single-copy atomicity semantics. So I'm not sure what to make of this.
> >> Surely switching to inline assembly on all targets is over the top.
> >>
> >> I think we can rely on a plain read doing the right thing for us.
> >
> > AFAIU, the plain read does not prevent the compiler from re-loading the
> > value in case of high register pressure.
> >
> > Accesses to rseq fields such as cpu_id need to be done as if those were
> > concurrently modified by a signal handler nesting on top of the user-space
> > code, with the particular twist that blocking signals has no effect on
> > concurrent updates.
> >
> > I do not think we need to do the load in assembly. I was under the impression
> > that both volatile load and relaxed MO result in single-copy atomicity
> > semantics for an aligned pointer. Perhaps Paul, Peter, Boqun have something
> > to add here ?
>
> The C memory model is broken and does not prevent out-of-thin-air
> values. As far as I know, this breaks single-copy atomicity. In
> practice, compilers will not exercise the latitude offered by the memory
> model. volatile does not ensure absence of data races.
Within the confines of the standard, agreed, use of the volatile keyword
does not explicitly prevent data races.
However, volatile accesses are (informally) defined to suffice for
device-driver memory accesses that communicate with devices, whether via
MMIO or DMA-style shared memory. The device-driver firmware is often
written in C or C++. So doesn't this informal device-driver guarantee
need to also do what is needed for userspace code that is communicating
with kernel code? If not, why not?
> Using atomics or volatile would require us to materialize the thread
> pointer, given the current internal interfaces we have, and I don't want
> to do this because this is supposed to be performance-critical code.
> The compiler barrier inherent to the function call will have to be
> enough. I can add a comment to this effect:
>
> /* This load has single-copy atomicity semantics (as required for
> rseq) because the function call implies a compiler barrier. */
Agreed on the need to be very careful to avoid degrading performance on
fast paths!
Thanx, Paul
^ permalink raw reply [flat|nested] 25+ messages in thread
* Re: [PATCH 2/5] nptl: Add rseq registration
2021-12-06 20:11 ` Paul E. McKenney
@ 2021-12-06 20:26 ` Florian Weimer
2021-12-06 21:08 ` Paul E. McKenney
0 siblings, 1 reply; 25+ messages in thread
From: Florian Weimer @ 2021-12-06 20:26 UTC (permalink / raw)
To: Paul E. McKenney via Libc-alpha
Cc: paulmck, Peter Zijlstra, Boqun Feng, Mathieu Desnoyers, linux-kernel
* Paul E. McKenney via Libc-alpha:
>> The C memory model is broken and does not prevent out-of-thin-air
>> values. As far as I know, this breaks single-copy atomicity. In
>> practice, compilers will not exercise the latitude offered by the memory
>> model. volatile does not ensure absence of data races.
>
> Within the confines of the standard, agreed, use of the volatile keyword
> does not explicitly prevent data races.
>
> However, volatile accesses are (informally) defined to suffice for
> device-driver memory accesses that communicate with devices, whether via
> MMIO or DMA-style shared memory. The device-driver firmware is often
> written in C or C++. So doesn't this informal device-driver guarantee
> need to also do what is needed for userspace code that is communicating
> with kernel code? If not, why not?
The informal guarantee is probably good enough here, too. However, the
actual accesses are behind macros, and those macros use either
non-volatile plain reads or inline assembler (which use
single-instruction naturally aligned reads).
THanks,
Florian
^ permalink raw reply [flat|nested] 25+ messages in thread
* Re: [PATCH 2/5] nptl: Add rseq registration
2021-12-06 20:26 ` Florian Weimer
@ 2021-12-06 21:08 ` Paul E. McKenney
0 siblings, 0 replies; 25+ messages in thread
From: Paul E. McKenney @ 2021-12-06 21:08 UTC (permalink / raw)
To: Florian Weimer
Cc: Paul E. McKenney via Libc-alpha, Peter Zijlstra, Boqun Feng,
Mathieu Desnoyers, linux-kernel
On Mon, Dec 06, 2021 at 09:26:51PM +0100, Florian Weimer wrote:
> * Paul E. McKenney via Libc-alpha:
>
> >> The C memory model is broken and does not prevent out-of-thin-air
> >> values. As far as I know, this breaks single-copy atomicity. In
> >> practice, compilers will not exercise the latitude offered by the memory
> >> model. volatile does not ensure absence of data races.
> >
> > Within the confines of the standard, agreed, use of the volatile keyword
> > does not explicitly prevent data races.
> >
> > However, volatile accesses are (informally) defined to suffice for
> > device-driver memory accesses that communicate with devices, whether via
> > MMIO or DMA-style shared memory. The device-driver firmware is often
> > written in C or C++. So doesn't this informal device-driver guarantee
> > need to also do what is needed for userspace code that is communicating
> > with kernel code? If not, why not?
>
> The informal guarantee is probably good enough here, too. However, the
> actual accesses are behind macros, and those macros use either
> non-volatile plain reads or inline assembler (which use
> single-instruction naturally aligned reads).
Agreed, a non-volatile plain read is quite dangerous in this context.
Thanx, Paul
^ permalink raw reply [flat|nested] 25+ messages in thread
* [PATCH 3/5] Linux: Use rseq to accelerate sched_getcpu
2021-12-06 13:45 [PATCH 0/5] Extensible rseq support for glibc Florian Weimer
2021-12-06 13:46 ` [PATCH 1/5] nptl: Add <thread_pointer.h> for defining __thread_pointer Florian Weimer
2021-12-06 13:46 ` [PATCH 2/5] nptl: Add rseq registration Florian Weimer
@ 2021-12-06 13:46 ` Florian Weimer
2021-12-06 16:50 ` Szabolcs Nagy
2021-12-06 13:46 ` [PATCH 4/5] nptl: Add glibc.pthread.rseq tunable to control rseq registration Florian Weimer
` (2 subsequent siblings)
5 siblings, 1 reply; 25+ messages in thread
From: Florian Weimer @ 2021-12-06 13:46 UTC (permalink / raw)
To: libc-alpha; +Cc: Mathieu Desnoyers
Co-Authored-By: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
---
sysdeps/unix/sysv/linux/sched_getcpu.c | 19 +++++++++++++++++--
1 file changed, 17 insertions(+), 2 deletions(-)
diff --git a/sysdeps/unix/sysv/linux/sched_getcpu.c b/sysdeps/unix/sysv/linux/sched_getcpu.c
index c41e986f2c..91250f9d0c 100644
--- a/sysdeps/unix/sysv/linux/sched_getcpu.c
+++ b/sysdeps/unix/sysv/linux/sched_getcpu.c
@@ -20,8 +20,8 @@
#include <sysdep.h>
#include <sysdep-vdso.h>
-int
-sched_getcpu (void)
+static int
+vsyscall_sched_getcpu (void)
{
unsigned int cpu;
int r = -1;
@@ -32,3 +32,18 @@ sched_getcpu (void)
#endif
return r == -1 ? r : cpu;
}
+
+#ifdef RSEQ_SIG
+int
+sched_getcpu (void)
+{
+ int cpu_id = THREAD_GETMEM (THREAD_SELF, rseq_area.cpu_id);
+ return __glibc_likely (cpu_id >= 0) ? cpu_id : vsyscall_sched_getcpu ();
+}
+#else /* RSEQ_SIG */
+int
+sched_getcpu (void)
+{
+ return vsyscall_sched_getcpu ();
+}
+#endif /* RSEQ_SIG */
--
2.33.1
^ permalink raw reply [flat|nested] 25+ messages in thread* Re: [PATCH 3/5] Linux: Use rseq to accelerate sched_getcpu
2021-12-06 13:46 ` [PATCH 3/5] Linux: Use rseq to accelerate sched_getcpu Florian Weimer
@ 2021-12-06 16:50 ` Szabolcs Nagy
2021-12-06 17:06 ` Florian Weimer
0 siblings, 1 reply; 25+ messages in thread
From: Szabolcs Nagy @ 2021-12-06 16:50 UTC (permalink / raw)
To: Florian Weimer; +Cc: libc-alpha, Mathieu Desnoyers
The 12/06/2021 14:46, Florian Weimer via Libc-alpha wrote:
> Co-Authored-By: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
> ---
> sysdeps/unix/sysv/linux/sched_getcpu.c | 19 +++++++++++++++++--
> 1 file changed, 17 insertions(+), 2 deletions(-)
>
> diff --git a/sysdeps/unix/sysv/linux/sched_getcpu.c b/sysdeps/unix/sysv/linux/sched_getcpu.c
> index c41e986f2c..91250f9d0c 100644
> --- a/sysdeps/unix/sysv/linux/sched_getcpu.c
> +++ b/sysdeps/unix/sysv/linux/sched_getcpu.c
> @@ -20,8 +20,8 @@
> #include <sysdep.h>
> #include <sysdep-vdso.h>
>
> -int
> -sched_getcpu (void)
> +static int
> +vsyscall_sched_getcpu (void)
> {
> unsigned int cpu;
> int r = -1;
> @@ -32,3 +32,18 @@ sched_getcpu (void)
> #endif
> return r == -1 ? r : cpu;
> }
> +
> +#ifdef RSEQ_SIG
> +int
> +sched_getcpu (void)
> +{
> + int cpu_id = THREAD_GETMEM (THREAD_SELF, rseq_area.cpu_id);
> + return __glibc_likely (cpu_id >= 0) ? cpu_id : vsyscall_sched_getcpu ();
> +}
i think there is a formal memory model issue here since
tp->rseq_area.cpu_id can be modified by the kernel asynchronously
but we don't use volatile or relaxed_mo atomic access.
(assuming THREAD_GETMEM is defined as a normal access)
i think this is fine here (unlikely to cause miscompilation),
but may be worth a note?
> +#else /* RSEQ_SIG */
> +int
> +sched_getcpu (void)
> +{
> + return vsyscall_sched_getcpu ();
> +}
> +#endif /* RSEQ_SIG */
> --
> 2.33.1
^ permalink raw reply [flat|nested] 25+ messages in thread* Re: [PATCH 3/5] Linux: Use rseq to accelerate sched_getcpu
2021-12-06 16:50 ` Szabolcs Nagy
@ 2021-12-06 17:06 ` Florian Weimer
2021-12-06 17:45 ` Szabolcs Nagy
0 siblings, 1 reply; 25+ messages in thread
From: Florian Weimer @ 2021-12-06 17:06 UTC (permalink / raw)
To: Szabolcs Nagy; +Cc: libc-alpha, Mathieu Desnoyers
* Szabolcs Nagy:
> The 12/06/2021 14:46, Florian Weimer via Libc-alpha wrote:
>> Co-Authored-By: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
>> ---
>> sysdeps/unix/sysv/linux/sched_getcpu.c | 19 +++++++++++++++++--
>> 1 file changed, 17 insertions(+), 2 deletions(-)
>>
>> diff --git a/sysdeps/unix/sysv/linux/sched_getcpu.c b/sysdeps/unix/sysv/linux/sched_getcpu.c
>> index c41e986f2c..91250f9d0c 100644
>> --- a/sysdeps/unix/sysv/linux/sched_getcpu.c
>> +++ b/sysdeps/unix/sysv/linux/sched_getcpu.c
>> @@ -20,8 +20,8 @@
>> #include <sysdep.h>
>> #include <sysdep-vdso.h>
>>
>> -int
>> -sched_getcpu (void)
>> +static int
>> +vsyscall_sched_getcpu (void)
>> {
>> unsigned int cpu;
>> int r = -1;
>> @@ -32,3 +32,18 @@ sched_getcpu (void)
>> #endif
>> return r == -1 ? r : cpu;
>> }
>> +
>> +#ifdef RSEQ_SIG
>> +int
>> +sched_getcpu (void)
>> +{
>> + int cpu_id = THREAD_GETMEM (THREAD_SELF, rseq_area.cpu_id);
>> + return __glibc_likely (cpu_id >= 0) ? cpu_id : vsyscall_sched_getcpu ();
>> +}
>
> i think there is a formal memory model issue here since
> tp->rseq_area.cpu_id can be modified by the kernel asynchronously
> but we don't use volatile or relaxed_mo atomic access.
> (assuming THREAD_GETMEM is defined as a normal access)
The kernel only modifies cpu_id if the user code is not running on the
thread, so I don't think this doesn't matter.
> i think this is fine here (unlikely to cause miscompilation),
> but may be worth a note?
The kernel doesn't use the C memory model, so using that terminology in
the kernel context would only be more confusing.
Thanks,
Florian
^ permalink raw reply [flat|nested] 25+ messages in thread* Re: [PATCH 3/5] Linux: Use rseq to accelerate sched_getcpu
2021-12-06 17:06 ` Florian Weimer
@ 2021-12-06 17:45 ` Szabolcs Nagy
2021-12-07 15:48 ` Florian Weimer
0 siblings, 1 reply; 25+ messages in thread
From: Szabolcs Nagy @ 2021-12-06 17:45 UTC (permalink / raw)
To: Florian Weimer; +Cc: libc-alpha, Mathieu Desnoyers
The 12/06/2021 18:06, Florian Weimer wrote:
> * Szabolcs Nagy:
>
> > The 12/06/2021 14:46, Florian Weimer via Libc-alpha wrote:
> >> Co-Authored-By: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
> >> ---
> >> sysdeps/unix/sysv/linux/sched_getcpu.c | 19 +++++++++++++++++--
> >> 1 file changed, 17 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/sysdeps/unix/sysv/linux/sched_getcpu.c b/sysdeps/unix/sysv/linux/sched_getcpu.c
> >> index c41e986f2c..91250f9d0c 100644
> >> --- a/sysdeps/unix/sysv/linux/sched_getcpu.c
> >> +++ b/sysdeps/unix/sysv/linux/sched_getcpu.c
> >> @@ -20,8 +20,8 @@
> >> #include <sysdep.h>
> >> #include <sysdep-vdso.h>
> >>
> >> -int
> >> -sched_getcpu (void)
> >> +static int
> >> +vsyscall_sched_getcpu (void)
> >> {
> >> unsigned int cpu;
> >> int r = -1;
> >> @@ -32,3 +32,18 @@ sched_getcpu (void)
> >> #endif
> >> return r == -1 ? r : cpu;
> >> }
> >> +
> >> +#ifdef RSEQ_SIG
> >> +int
> >> +sched_getcpu (void)
> >> +{
> >> + int cpu_id = THREAD_GETMEM (THREAD_SELF, rseq_area.cpu_id);
> >> + return __glibc_likely (cpu_id >= 0) ? cpu_id : vsyscall_sched_getcpu ();
> >> +}
> >
> > i think there is a formal memory model issue here since
> > tp->rseq_area.cpu_id can be modified by the kernel asynchronously
> > but we don't use volatile or relaxed_mo atomic access.
> > (assuming THREAD_GETMEM is defined as a normal access)
>
> The kernel only modifies cpu_id if the user code is not running on the
> thread, so I don't think this doesn't matter.
it does not look very different from async signal handler
modifying a global. there we would use volatile eg. to
avoid splitting a read access into multiple reads.
> > i think this is fine here (unlikely to cause miscompilation),
> > but may be worth a note?
>
> The kernel doesn't use the C memory model, so using that terminology in
> the kernel context would only be more confusing.
ok, i'm happy with the code, i just noticed that the
previous patchset used relaxed mo.
^ permalink raw reply [flat|nested] 25+ messages in thread* Re: [PATCH 3/5] Linux: Use rseq to accelerate sched_getcpu
2021-12-06 17:45 ` Szabolcs Nagy
@ 2021-12-07 15:48 ` Florian Weimer
0 siblings, 0 replies; 25+ messages in thread
From: Florian Weimer @ 2021-12-07 15:48 UTC (permalink / raw)
To: Szabolcs Nagy; +Cc: libc-alpha, Mathieu Desnoyers
* Szabolcs Nagy:
>> > i think there is a formal memory model issue here since
>> > tp->rseq_area.cpu_id can be modified by the kernel asynchronously
>> > but we don't use volatile or relaxed_mo atomic access.
>> > (assuming THREAD_GETMEM is defined as a normal access)
>>
>> The kernel only modifies cpu_id if the user code is not running on the
>> thread, so I don't think this doesn't matter.
>
> it does not look very different from async signal handler
> modifying a global. there we would use volatile eg. to
> avoid splitting a read access into multiple reads.
I added the necessary scaffolding to use a voltile read here in the v2
series.
Thanks,
Florian
^ permalink raw reply [flat|nested] 25+ messages in thread
* [PATCH 4/5] nptl: Add glibc.pthread.rseq tunable to control rseq registration
2021-12-06 13:45 [PATCH 0/5] Extensible rseq support for glibc Florian Weimer
` (2 preceding siblings ...)
2021-12-06 13:46 ` [PATCH 3/5] Linux: Use rseq to accelerate sched_getcpu Florian Weimer
@ 2021-12-06 13:46 ` Florian Weimer
2021-12-06 13:53 ` [PATCH 5/5] nptl: Add public rseq symbols and <sys/rseq.h> Florian Weimer
2021-12-06 16:13 ` [PATCH 0/5] Extensible rseq support for glibc Mathieu Desnoyers
5 siblings, 0 replies; 25+ messages in thread
From: Florian Weimer @ 2021-12-06 13:46 UTC (permalink / raw)
To: libc-alpha; +Cc: Mathieu Desnoyers
This tunable allows applications to register the rseq area instead
of glibc.
---
manual/tunables.texi | 10 +++
nptl/pthread_create.c | 10 ++-
sysdeps/nptl/dl-tls_init_tp.c | 11 ++-
sysdeps/nptl/dl-tunables.list | 6 ++
sysdeps/nptl/internaltypes.h | 1 +
sysdeps/unix/sysv/linux/Makefile | 8 ++
sysdeps/unix/sysv/linux/rseq-internal.h | 19 +++--
sysdeps/unix/sysv/linux/tst-rseq-disable.c | 89 ++++++++++++++++++++++
8 files changed, 145 insertions(+), 9 deletions(-)
create mode 100644 sysdeps/unix/sysv/linux/tst-rseq-disable.c
diff --git a/manual/tunables.texi b/manual/tunables.texi
index 10f4d75993..5d50b90f64 100644
--- a/manual/tunables.texi
+++ b/manual/tunables.texi
@@ -424,6 +424,16 @@ The value is measured in bytes. The default is @samp{41943040}
(fourty mibibytes).
@end deftp
+@deftp Tunable glibc.pthread.rseq
+The @code{glibc.pthread.rseq} tunable can be set to @samp{0}, to disable
+restartable sequences support in @theglibc{}. This enables applications
+to perform direct restartable sequence registration with the kernel.
+The default is @samp{1}, which means that @theglibc{} performs
+registration on behalf of the application.
+
+Restartable sequences are a Linux-specific extension.
+@end deftp
+
@node Hardware Capability Tunables
@section Hardware Capability Tunables
@cindex hardware capability tunables
diff --git a/nptl/pthread_create.c b/nptl/pthread_create.c
index ea0d79341e..036ca15ea0 100644
--- a/nptl/pthread_create.c
+++ b/nptl/pthread_create.c
@@ -368,7 +368,10 @@ start_thread (void *arg)
__ctype_init ();
/* Register rseq TLS to the kernel. */
- rseq_register_current_thread (pd);
+ {
+ bool do_rseq = THREAD_GETMEM (pd, flags) & ATTR_FLAG_DO_RSEQ;
+ rseq_register_current_thread (pd, do_rseq);
+ }
#ifndef __ASSUME_SET_ROBUST_LIST
if (__nptl_set_robust_list_avail)
@@ -677,6 +680,11 @@ __pthread_create_2_1 (pthread_t *newthread, const pthread_attr_t *attr,
pd->flags = ((iattr->flags & ~(ATTR_FLAG_SCHED_SET | ATTR_FLAG_POLICY_SET))
| (self->flags & (ATTR_FLAG_SCHED_SET | ATTR_FLAG_POLICY_SET)));
+ /* Inherit rseq registration state. Without seccomp filters, rseq
+ registration will either always fail or always succeed. */
+ if ((int) THREAD_GETMEM (self, rseq_area.cpu_id) >= 0)
+ pd->flags |= ATTR_FLAG_DO_RSEQ;
+
/* Initialize the field for the ID of the thread which is waiting
for us. This is a self-reference in case the thread is created
detached. */
diff --git a/sysdeps/nptl/dl-tls_init_tp.c b/sysdeps/nptl/dl-tls_init_tp.c
index fedb876fdb..b39dfbff2c 100644
--- a/sysdeps/nptl/dl-tls_init_tp.c
+++ b/sysdeps/nptl/dl-tls_init_tp.c
@@ -23,6 +23,9 @@
#include <tls.h>
#include <rseq-internal.h>
+#define TUNABLE_NAMESPACE pthread
+#include <dl-tunables.h>
+
#ifndef __ASSUME_SET_ROBUST_LIST
bool __nptl_set_robust_list_avail;
rtld_hidden_data_def (__nptl_set_robust_list_avail)
@@ -92,7 +95,13 @@ __tls_init_tp (void)
}
}
- rseq_register_current_thread (pd);
+ {
+ bool do_rseq = true;
+#if HAVE_TUNABLES
+ do_rseq = TUNABLE_GET (rseq, int, NULL);
+#endif
+ rseq_register_current_thread (pd, do_rseq);
+ }
/* Set initial thread's stack block from 0 up to __libc_stack_end.
It will be bigger than it actually is, but for unwind.c/pt-longjmp.c
diff --git a/sysdeps/nptl/dl-tunables.list b/sysdeps/nptl/dl-tunables.list
index ac5d053298..d24f4be0d0 100644
--- a/sysdeps/nptl/dl-tunables.list
+++ b/sysdeps/nptl/dl-tunables.list
@@ -27,5 +27,11 @@ glibc {
type: SIZE_T
default: 41943040
}
+ rseq {
+ type: INT_32
+ minval: 0
+ maxval: 1
+ default: 1
+ }
}
}
diff --git a/sysdeps/nptl/internaltypes.h b/sysdeps/nptl/internaltypes.h
index 6032a6b785..dec8c5b5ff 100644
--- a/sysdeps/nptl/internaltypes.h
+++ b/sysdeps/nptl/internaltypes.h
@@ -48,6 +48,7 @@ struct pthread_attr
#define ATTR_FLAG_OLDATTR 0x0010
#define ATTR_FLAG_SCHED_SET 0x0020
#define ATTR_FLAG_POLICY_SET 0x0040
+#define ATTR_FLAG_DO_RSEQ 0x0080
/* Used to allocate a pthread_attr_t object which is also accessed
internally. */
diff --git a/sysdeps/unix/sysv/linux/Makefile b/sysdeps/unix/sysv/linux/Makefile
index eb0f5fc021..62a796f214 100644
--- a/sysdeps/unix/sysv/linux/Makefile
+++ b/sysdeps/unix/sysv/linux/Makefile
@@ -136,6 +136,12 @@ tests-internal += \
tst-sigcontext-get_pc \
# tests-internal
+ifneq (no,$(have-tunables))
+tests-internal += \
+ tst-rseq-disable \
+ # tests-internal $(have-tunables)
+endif
+
tests-time64 += \
tst-adjtimex-time64 \
tst-clock_adjtime-time64 \
@@ -227,6 +233,8 @@ $(objpfx)tst-mman-consts.out: ../sysdeps/unix/sysv/linux/tst-mman-consts.py
< /dev/null > $@ 2>&1; $(evaluate-test)
$(objpfx)tst-mman-consts.out: $(sysdeps-linux-python-deps)
+tst-rseq-disable-ENV = GLIBC_TUNABLES=glibc.pthread.rseq=0
+
endif # $(subdir) == misc
ifeq ($(subdir),time)
diff --git a/sysdeps/unix/sysv/linux/rseq-internal.h b/sysdeps/unix/sysv/linux/rseq-internal.h
index 909f547825..15bc7ffd6e 100644
--- a/sysdeps/unix/sysv/linux/rseq-internal.h
+++ b/sysdeps/unix/sysv/linux/rseq-internal.h
@@ -21,22 +21,27 @@
#include <sysdep.h>
#include <errno.h>
#include <kernel-features.h>
+#include <stdbool.h>
#include <stdio.h>
#include <sys/rseq.h>
#ifdef RSEQ_SIG
static inline void
-rseq_register_current_thread (struct pthread *self)
+rseq_register_current_thread (struct pthread *self, bool do_rseq)
{
- int ret = INTERNAL_SYSCALL_CALL (rseq,
- &self->rseq_area, sizeof (self->rseq_area),
- 0, RSEQ_SIG);
- if (INTERNAL_SYSCALL_ERROR_P (ret))
- THREAD_SETMEM (self, rseq_area.cpu_id, RSEQ_CPU_ID_REGISTRATION_FAILED);
+ if (do_rseq)
+ {
+ int ret = INTERNAL_SYSCALL_CALL (rseq, &self->rseq_area,
+ sizeof (self->rseq_area),
+ 0, RSEQ_SIG);
+ if (!INTERNAL_SYSCALL_ERROR_P (ret))
+ return;
+ }
+ THREAD_SETMEM (self, rseq_area.cpu_id, RSEQ_CPU_ID_REGISTRATION_FAILED);
}
#else /* RSEQ_SIG */
static inline void
-rseq_register_current_thread (struct pthread *self)
+rseq_register_current_thread (struct pthread *self, bool do_rseq)
{
THREAD_SETMEM (self, rseq_area.cpu_id, RSEQ_CPU_ID_REGISTRATION_FAILED);
}
diff --git a/sysdeps/unix/sysv/linux/tst-rseq-disable.c b/sysdeps/unix/sysv/linux/tst-rseq-disable.c
new file mode 100644
index 0000000000..000e351872
--- /dev/null
+++ b/sysdeps/unix/sysv/linux/tst-rseq-disable.c
@@ -0,0 +1,89 @@
+/* Test disabling of rseq registration via tunable.
+ Copyright (C) 2021 Free Software Foundation, Inc.
+
+ 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 <errno.h>
+#include <stdio.h>
+#include <support/check.h>
+#include <support/namespace.h>
+#include <support/xthread.h>
+#include <sysdep.h>
+#include <unistd.h>
+
+#ifdef RSEQ_SIG
+
+/* Check that rseq can be registered and has not been taken by glibc. */
+static void
+check_rseq_disabled (void)
+{
+ struct pthread *pd = THREAD_SELF;
+ TEST_COMPARE ((int) pd->rseq_area.cpu_id, RSEQ_CPU_ID_REGISTRATION_FAILED);
+
+ int ret = syscall (__NR_rseq, &pd->rseq_area, sizeof (pd->rseq_area),
+ 0, RSEQ_SIG);
+ if (ret == 0)
+ {
+ ret = syscall (__NR_rseq, &pd->rseq_area, sizeof (pd->rseq_area),
+ RSEQ_FLAG_UNREGISTER, RSEQ_SIG);
+ TEST_COMPARE (ret, 0);
+ pd->rseq_area.cpu_id = RSEQ_CPU_ID_REGISTRATION_FAILED;
+ }
+ else
+ {
+ TEST_VERIFY (errno != -EINVAL);
+ TEST_VERIFY (errno != -EBUSY);
+ }
+}
+
+static void *
+thread_func (void *ignored)
+{
+ check_rseq_disabled ();
+ return NULL;
+}
+
+static void
+proc_func (void *ignored)
+{
+ check_rseq_disabled ();
+}
+
+static int
+do_test (void)
+{
+ puts ("info: checking main thread");
+ check_rseq_disabled ();
+
+ puts ("info: checking main thread (2)");
+ check_rseq_disabled ();
+
+ puts ("info: checking new thread");
+ xpthread_join (xpthread_create (NULL, thread_func, NULL));
+
+ puts ("info: checking subprocess");
+ support_isolate_in_subprocess (proc_func, NULL);
+
+ return 0;
+}
+#else /* !RSEQ_SIG */
+static int
+do_test (void)
+{
+ FAIL_UNSUPPORTED ("glibc does not define RSEQ_SIG, skipping test");
+}
+#endif
+
+#include <support/test-driver.c>
--
2.33.1
^ permalink raw reply [flat|nested] 25+ messages in thread* [PATCH 5/5] nptl: Add public rseq symbols and <sys/rseq.h>
2021-12-06 13:45 [PATCH 0/5] Extensible rseq support for glibc Florian Weimer
` (3 preceding siblings ...)
2021-12-06 13:46 ` [PATCH 4/5] nptl: Add glibc.pthread.rseq tunable to control rseq registration Florian Weimer
@ 2021-12-06 13:53 ` Florian Weimer
2021-12-07 11:01 ` Szabolcs Nagy
2021-12-06 16:13 ` [PATCH 0/5] Extensible rseq support for glibc Mathieu Desnoyers
5 siblings, 1 reply; 25+ messages in thread
From: Florian Weimer @ 2021-12-06 13:53 UTC (permalink / raw)
To: libc-alpha
The relationship between the thread pointer and the rseq area
is made explicit. The constant offset can be used by JIT compilers
to optimize rseq access (e.g., for really fast sched_getcpu).
Extensibility is provided through __rseq_size and __rseq_flags.
(In the future, the kernel could request a different rseq size
via the auxiliary vector.)
Co-Authored-By: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Tested on aarch64-linux-gnu, i686-linux-gnu, powerpc64le-linux-gnu,
s390x-linux-gnu, x86_64-linux-gnu. Built with build-many-glibcs.py
(except for a late fix to actually install the public headers).
---
NEWS | 11 +++
manual/threads.texi | 81 ++++++++++++++++++++++
sysdeps/nptl/dl-tls_init_tp.c | 18 ++++-
sysdeps/unix/sysv/linux/Makefile | 3 +-
sysdeps/unix/sysv/linux/Versions | 5 ++
sysdeps/unix/sysv/linux/aarch64/ld.abilist | 3 +
sysdeps/unix/sysv/linux/alpha/ld.abilist | 3 +
sysdeps/unix/sysv/linux/arc/ld.abilist | 3 +
sysdeps/unix/sysv/linux/arm/be/ld.abilist | 3 +
sysdeps/unix/sysv/linux/arm/le/ld.abilist | 3 +
sysdeps/unix/sysv/linux/csky/ld.abilist | 3 +
sysdeps/unix/sysv/linux/hppa/ld.abilist | 3 +
sysdeps/unix/sysv/linux/i386/ld.abilist | 3 +
sysdeps/unix/sysv/linux/ia64/ld.abilist | 3 +
sysdeps/unix/sysv/linux/m68k/coldfire/ld.abilist | 3 +
sysdeps/unix/sysv/linux/m68k/m680x0/ld.abilist | 3 +
sysdeps/unix/sysv/linux/microblaze/ld.abilist | 3 +
sysdeps/unix/sysv/linux/mips/mips32/ld.abilist | 3 +
sysdeps/unix/sysv/linux/mips/mips64/n32/ld.abilist | 3 +
sysdeps/unix/sysv/linux/mips/mips64/n64/ld.abilist | 3 +
sysdeps/unix/sysv/linux/nios2/ld.abilist | 3 +
.../unix/sysv/linux/powerpc/powerpc32/ld.abilist | 3 +
.../sysv/linux/powerpc/powerpc64/be/ld.abilist | 3 +
.../sysv/linux/powerpc/powerpc64/le/ld.abilist | 3 +
sysdeps/unix/sysv/linux/riscv/rv32/ld.abilist | 3 +
sysdeps/unix/sysv/linux/riscv/rv64/ld.abilist | 3 +
sysdeps/unix/sysv/linux/rseq-internal.h | 8 ++-
sysdeps/unix/sysv/linux/s390/s390-32/ld.abilist | 3 +
sysdeps/unix/sysv/linux/s390/s390-64/ld.abilist | 3 +
sysdeps/unix/sysv/linux/sh/be/ld.abilist | 3 +
sysdeps/unix/sysv/linux/sh/le/ld.abilist | 3 +
sysdeps/unix/sysv/linux/sparc/sparc32/ld.abilist | 3 +
sysdeps/unix/sysv/linux/sparc/sparc64/ld.abilist | 3 +
sysdeps/unix/sysv/linux/sys/rseq.h | 10 +++
sysdeps/unix/sysv/linux/tst-rseq-disable.c | 6 ++
sysdeps/unix/sysv/linux/tst-rseq.c | 8 +++
sysdeps/unix/sysv/linux/x86_64/64/ld.abilist | 3 +
sysdeps/unix/sysv/linux/x86_64/x32/ld.abilist | 3 +
38 files changed, 232 insertions(+), 5 deletions(-)
diff --git a/NEWS b/NEWS
index 1398cf2e87..8744a92532 100644
--- a/NEWS
+++ b/NEWS
@@ -68,6 +68,17 @@ Major new features:
to be used by compilers for optimizing usage of 'memcmp' when its
return value is only used for its boolean status.
+* Support for automatically registering threads with the Linux rseq
+ system call has been added. This system call is implemented starting
+ from Linux 4.18. The Restartable Sequences ABI accelerates user-space
+ operations on per-cpu data. It allows user-space to perform updates
+ on per-cpu data without requiring heavy-weight atomic operations.
+ Automatically registering threads allows all libraries, including
+ libc, to make immediate use of the rseq support by using the
+ documented ABI, via the __rseq_flags, __rseq_offset, and __rseq_size
+ variables. The GNU C Library manual has details on integration of
+ Restartable Sequences.
+
Deprecated and removed features, and other changes affecting compatibility:
* The r_version update in the debugger interface makes the glibc binary
diff --git a/manual/threads.texi b/manual/threads.texi
index 06b6b277a1..7b6170cbd0 100644
--- a/manual/threads.texi
+++ b/manual/threads.texi
@@ -629,6 +629,8 @@ the standard.
* Waiting with Explicit Clocks:: Functions for waiting with an
explicit clock specification.
* Single-Threaded:: Detecting single-threaded execution.
+* Restartable Sequences:: Linux-specific restartable sequences
+ integration.
@end menu
@node Default Thread Attributes
@@ -958,6 +960,85 @@ application-created thread because future versions of @theglibc{} may
create background threads after the first thread has been created, and
the application has no way of knowning that these threads are present.
+@node Restartable Sequences
+@subsubsection Restartable Sequences
+
+This section describes restartable sequences integration for
+@theglibc{}. This functionality is only available on Linux.
+
+@deftp {Data Type} {struct rseq}
+@standards{Linux, sys/rseq.h}
+The type of the restartable sequences area. Future versions
+of Linux may add additional fields to the end of this structure.
+
+
+Users need to obtain the address of the restartable sequences area using
+the thread pointer and the @code{__rseq_offset} variable, described
+below.
+
+One use of the restartable sequences area is to read the current CPU
+number from its @code{cpu_id} field, as an inline version of
+@code{sched_getcpu}. @Theglibc{} sets the @code{cpu_id} field to
+@code{RSEQ_CPU_ID_REGISTRATION_FAILED} if registration failed or was
+explicitly disabled.
+
+Furthermore, users can store the address of a @code{struct rseq_cs}
+object into the @code{rseq_cs} field of @code{struct rseq}, thus
+informing the kernel that the thread enters a restartable sequence
+critical section. This pointer and the code areas it itself points to
+must not be left pointing to memory areas which are freed or re-used.
+Several approaches can guarantee this. If the application or library
+can guarantee that the memory used to hold the @code{struct rseq_cs} and
+the code areas it refers to are never freed or re-used, no special
+action must be taken. Else, before that memory is re-used of freed, the
+application is responsible for setting the @code{rseq_cs} field to
+@code{NULL} in each thread's restartable sequence area to guarantee that
+it does not leak dangling references. Because the application does not
+typically have knowledge of libraries' use of restartable sequences, it
+is recommended that libraries using restartable sequences which may end
+up freeing or re-using their memory set the @code{rseq_cs} field to
+@code{NULL} before returning from library functions which use
+restartable sequences.
+
+The manual for the @code{rseq} system call can be found
+at @uref{https://git.kernel.org/pub/scm/libs/librseq/librseq.git/tree/doc/man/rseq.2}.
+@end deftp
+
+@deftypevar {int} __rseq_offset
+@standards{Linux, sys/rseq.h}
+This variable contains the offset between the thread pointer (as defined
+by @code{__builtin_thread_pointer} or the thread pointer register for
+the architecture) and the restartable sequences area. This value is the
+same for all threads in the process. If the restartable sequences area
+is located at a lower address than the location to whic the thread
+pointer points, the value is negative.
+@end deftypevar
+
+@deftypevar {int} __rseq_size
+@standards{Linux, sys/rseq.h}
+This variable is either zero (if restartable sequence registration
+failed or has been disabled) or the size of the restartable sequence
+registration. This can be less can be different from the size of
+@code{struct rseq} if the kernel has extended the size of the
+registration. If registration is successful, @code{__rseq_size} is at
+least 32 (the initial size of @code{struct rseq}.
+@end deftypevar
+
+@deftypevar {unsigned int} __rseq_flags
+@standards{Linux, sys/rseq.h}
+The flags used during restartable sequence registration with the kernel.
+Currently zero.
+@end deftypevar
+
+@deftypevr Macro int RSEQ_SIG
+@standards{Linux, sys/rseq.h}
+Each supported architecture provides a @code{RSEQ_SIG} macro in
+@file{sys/rseq.h} which contains a signature. That signature is
+expected to be present in the code before each restartable sequences
+abort handler. Failure to provide the expected signature may terminate
+the process with a segmentation fault.
+@end deftypevr
+
@c FIXME these are undocumented:
@c pthread_atfork
@c pthread_attr_destroy
diff --git a/sysdeps/nptl/dl-tls_init_tp.c b/sysdeps/nptl/dl-tls_init_tp.c
index b39dfbff2c..82aa7083c1 100644
--- a/sysdeps/nptl/dl-tls_init_tp.c
+++ b/sysdeps/nptl/dl-tls_init_tp.c
@@ -22,6 +22,7 @@
#include <pthreadP.h>
#include <tls.h>
#include <rseq-internal.h>
+#include <thread_pointer.h>
#define TUNABLE_NAMESPACE pthread
#include <dl-tunables.h>
@@ -43,6 +44,10 @@ rtld_mutex_dummy (pthread_mutex_t *lock)
}
#endif
+const unsigned int __rseq_flags;
+const unsigned int __rseq_size attribute_relro;
+const int __rseq_offset attribute_relro;
+
void
__tls_pre_init_tp (void)
{
@@ -100,7 +105,18 @@ __tls_init_tp (void)
#if HAVE_TUNABLES
do_rseq = TUNABLE_GET (rseq, int, NULL);
#endif
- rseq_register_current_thread (pd, do_rseq);
+ if (rseq_register_current_thread (pd, do_rseq))
+ {
+ /* We need a writable view of the variables. They are in
+ .data.relro and are not yet write-protected. */
+ extern unsigned int size __asm__ ("__rseq_size");
+ size = sizeof (pd->rseq_area);
+ }
+
+ /* This should be a compile-time constant, but the current
+ infrastructure makes it difficult to determine its value. */
+ extern int offset __asm__ ("__rseq_offset");
+ offset = (char *) &pd->rseq_area - (char *) __thread_pointer ();
}
/* Set initial thread's stack block from 0 up to __libc_stack_end.
diff --git a/sysdeps/unix/sysv/linux/Makefile b/sysdeps/unix/sysv/linux/Makefile
index 62a796f214..61acc1987d 100644
--- a/sysdeps/unix/sysv/linux/Makefile
+++ b/sysdeps/unix/sysv/linux/Makefile
@@ -112,7 +112,8 @@ sysdep_headers += sys/mount.h sys/acct.h \
bits/types/struct_semid64_ds_helper.h \
bits/types/struct_shmid64_ds.h \
bits/types/struct_shmid64_ds_helper.h \
- bits/pthread_stack_min.h bits/pthread_stack_min-dynamic.h
+ bits/pthread_stack_min.h bits/pthread_stack_min-dynamic.h \
+ sys/rseq.h bits/rseq.h
tests += tst-clone tst-clone2 tst-clone3 tst-fanotify tst-personality \
tst-quota tst-sync_file_range tst-sysconf-iov_max tst-ttyname \
diff --git a/sysdeps/unix/sysv/linux/Versions b/sysdeps/unix/sysv/linux/Versions
index 26452f3f17..3f8809a158 100644
--- a/sysdeps/unix/sysv/linux/Versions
+++ b/sysdeps/unix/sysv/linux/Versions
@@ -316,6 +316,11 @@ librt {
}
ld {
+ GLIBC_2.35 {
+ __rseq_flags;
+ __rseq_offset;
+ __rseq_size;
+ }
GLIBC_PRIVATE {
__nptl_change_stack_perm;
}
diff --git a/sysdeps/unix/sysv/linux/aarch64/ld.abilist b/sysdeps/unix/sysv/linux/aarch64/ld.abilist
index 80b2fe6725..717a35f242 100644
--- a/sysdeps/unix/sysv/linux/aarch64/ld.abilist
+++ b/sysdeps/unix/sysv/linux/aarch64/ld.abilist
@@ -3,3 +3,6 @@ GLIBC_2.17 __stack_chk_guard D 0x8
GLIBC_2.17 __tls_get_addr F
GLIBC_2.17 _dl_mcount F
GLIBC_2.17 _r_debug D 0x28
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
diff --git a/sysdeps/unix/sysv/linux/alpha/ld.abilist b/sysdeps/unix/sysv/linux/alpha/ld.abilist
index 98a03f611f..76911bd7f8 100644
--- a/sysdeps/unix/sysv/linux/alpha/ld.abilist
+++ b/sysdeps/unix/sysv/linux/alpha/ld.abilist
@@ -2,4 +2,7 @@ GLIBC_2.0 _r_debug D 0x28
GLIBC_2.1 __libc_stack_end D 0x8
GLIBC_2.1 _dl_mcount F
GLIBC_2.3 __tls_get_addr F
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
GLIBC_2.4 __stack_chk_guard D 0x8
diff --git a/sysdeps/unix/sysv/linux/arc/ld.abilist b/sysdeps/unix/sysv/linux/arc/ld.abilist
index 048f17c848..71c67f9803 100644
--- a/sysdeps/unix/sysv/linux/arc/ld.abilist
+++ b/sysdeps/unix/sysv/linux/arc/ld.abilist
@@ -3,3 +3,6 @@ GLIBC_2.32 __stack_chk_guard D 0x4
GLIBC_2.32 __tls_get_addr F
GLIBC_2.32 _dl_mcount F
GLIBC_2.32 _r_debug D 0x14
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
diff --git a/sysdeps/unix/sysv/linux/arm/be/ld.abilist b/sysdeps/unix/sysv/linux/arm/be/ld.abilist
index cc8825c3bc..3859433b21 100644
--- a/sysdeps/unix/sysv/linux/arm/be/ld.abilist
+++ b/sysdeps/unix/sysv/linux/arm/be/ld.abilist
@@ -1,3 +1,6 @@
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
GLIBC_2.4 __libc_stack_end D 0x4
GLIBC_2.4 __stack_chk_guard D 0x4
GLIBC_2.4 __tls_get_addr F
diff --git a/sysdeps/unix/sysv/linux/arm/le/ld.abilist b/sysdeps/unix/sysv/linux/arm/le/ld.abilist
index cc8825c3bc..3859433b21 100644
--- a/sysdeps/unix/sysv/linux/arm/le/ld.abilist
+++ b/sysdeps/unix/sysv/linux/arm/le/ld.abilist
@@ -1,3 +1,6 @@
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
GLIBC_2.4 __libc_stack_end D 0x4
GLIBC_2.4 __stack_chk_guard D 0x4
GLIBC_2.4 __tls_get_addr F
diff --git a/sysdeps/unix/sysv/linux/csky/ld.abilist b/sysdeps/unix/sysv/linux/csky/ld.abilist
index 564ac09737..6bfc582b73 100644
--- a/sysdeps/unix/sysv/linux/csky/ld.abilist
+++ b/sysdeps/unix/sysv/linux/csky/ld.abilist
@@ -3,3 +3,6 @@ GLIBC_2.29 __stack_chk_guard D 0x4
GLIBC_2.29 __tls_get_addr F
GLIBC_2.29 _dl_mcount F
GLIBC_2.29 _r_debug D 0x14
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
diff --git a/sysdeps/unix/sysv/linux/hppa/ld.abilist b/sysdeps/unix/sysv/linux/hppa/ld.abilist
index d155a59843..efccd6a023 100644
--- a/sysdeps/unix/sysv/linux/hppa/ld.abilist
+++ b/sysdeps/unix/sysv/linux/hppa/ld.abilist
@@ -2,4 +2,7 @@ GLIBC_2.2 __libc_stack_end D 0x4
GLIBC_2.2 _dl_mcount F
GLIBC_2.2 _r_debug D 0x14
GLIBC_2.3 __tls_get_addr F
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
GLIBC_2.4 __stack_chk_guard D 0x4
diff --git a/sysdeps/unix/sysv/linux/i386/ld.abilist b/sysdeps/unix/sysv/linux/i386/ld.abilist
index 0478e22071..1eb94ae75f 100644
--- a/sysdeps/unix/sysv/linux/i386/ld.abilist
+++ b/sysdeps/unix/sysv/linux/i386/ld.abilist
@@ -3,3 +3,6 @@ GLIBC_2.1 __libc_stack_end D 0x4
GLIBC_2.1 _dl_mcount F
GLIBC_2.3 ___tls_get_addr F
GLIBC_2.3 __tls_get_addr F
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
diff --git a/sysdeps/unix/sysv/linux/ia64/ld.abilist b/sysdeps/unix/sysv/linux/ia64/ld.abilist
index 33f91199bf..2cc68bcf7b 100644
--- a/sysdeps/unix/sysv/linux/ia64/ld.abilist
+++ b/sysdeps/unix/sysv/linux/ia64/ld.abilist
@@ -2,3 +2,6 @@ GLIBC_2.2 __libc_stack_end D 0x8
GLIBC_2.2 _dl_mcount F
GLIBC_2.2 _r_debug D 0x28
GLIBC_2.3 __tls_get_addr F
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
diff --git a/sysdeps/unix/sysv/linux/m68k/coldfire/ld.abilist b/sysdeps/unix/sysv/linux/m68k/coldfire/ld.abilist
index cc8825c3bc..3859433b21 100644
--- a/sysdeps/unix/sysv/linux/m68k/coldfire/ld.abilist
+++ b/sysdeps/unix/sysv/linux/m68k/coldfire/ld.abilist
@@ -1,3 +1,6 @@
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
GLIBC_2.4 __libc_stack_end D 0x4
GLIBC_2.4 __stack_chk_guard D 0x4
GLIBC_2.4 __tls_get_addr F
diff --git a/sysdeps/unix/sysv/linux/m68k/m680x0/ld.abilist b/sysdeps/unix/sysv/linux/m68k/m680x0/ld.abilist
index 3ba474c27f..e62b2742af 100644
--- a/sysdeps/unix/sysv/linux/m68k/m680x0/ld.abilist
+++ b/sysdeps/unix/sysv/linux/m68k/m680x0/ld.abilist
@@ -2,4 +2,7 @@ GLIBC_2.0 _r_debug D 0x14
GLIBC_2.1 __libc_stack_end D 0x4
GLIBC_2.1 _dl_mcount F
GLIBC_2.3 __tls_get_addr F
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
GLIBC_2.4 __stack_chk_guard D 0x4
diff --git a/sysdeps/unix/sysv/linux/microblaze/ld.abilist b/sysdeps/unix/sysv/linux/microblaze/ld.abilist
index a4933c3541..5d63d74e8f 100644
--- a/sysdeps/unix/sysv/linux/microblaze/ld.abilist
+++ b/sysdeps/unix/sysv/linux/microblaze/ld.abilist
@@ -3,3 +3,6 @@ GLIBC_2.18 __stack_chk_guard D 0x4
GLIBC_2.18 __tls_get_addr F
GLIBC_2.18 _dl_mcount F
GLIBC_2.18 _r_debug D 0x14
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
diff --git a/sysdeps/unix/sysv/linux/mips/mips32/ld.abilist b/sysdeps/unix/sysv/linux/mips/mips32/ld.abilist
index be09641a48..53ca22de2f 100644
--- a/sysdeps/unix/sysv/linux/mips/mips32/ld.abilist
+++ b/sysdeps/unix/sysv/linux/mips/mips32/ld.abilist
@@ -2,4 +2,7 @@ GLIBC_2.0 _r_debug D 0x14
GLIBC_2.2 __libc_stack_end D 0x4
GLIBC_2.2 _dl_mcount F
GLIBC_2.3 __tls_get_addr F
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
GLIBC_2.4 __stack_chk_guard D 0x4
diff --git a/sysdeps/unix/sysv/linux/mips/mips64/n32/ld.abilist b/sysdeps/unix/sysv/linux/mips/mips64/n32/ld.abilist
index be09641a48..53ca22de2f 100644
--- a/sysdeps/unix/sysv/linux/mips/mips64/n32/ld.abilist
+++ b/sysdeps/unix/sysv/linux/mips/mips64/n32/ld.abilist
@@ -2,4 +2,7 @@ GLIBC_2.0 _r_debug D 0x14
GLIBC_2.2 __libc_stack_end D 0x4
GLIBC_2.2 _dl_mcount F
GLIBC_2.3 __tls_get_addr F
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
GLIBC_2.4 __stack_chk_guard D 0x4
diff --git a/sysdeps/unix/sysv/linux/mips/mips64/n64/ld.abilist b/sysdeps/unix/sysv/linux/mips/mips64/n64/ld.abilist
index 1ea36e13f2..d1cdd68333 100644
--- a/sysdeps/unix/sysv/linux/mips/mips64/n64/ld.abilist
+++ b/sysdeps/unix/sysv/linux/mips/mips64/n64/ld.abilist
@@ -2,4 +2,7 @@ GLIBC_2.0 _r_debug D 0x28
GLIBC_2.2 __libc_stack_end D 0x8
GLIBC_2.2 _dl_mcount F
GLIBC_2.3 __tls_get_addr F
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
GLIBC_2.4 __stack_chk_guard D 0x8
diff --git a/sysdeps/unix/sysv/linux/nios2/ld.abilist b/sysdeps/unix/sysv/linux/nios2/ld.abilist
index 52178802dd..bcbba1823e 100644
--- a/sysdeps/unix/sysv/linux/nios2/ld.abilist
+++ b/sysdeps/unix/sysv/linux/nios2/ld.abilist
@@ -3,3 +3,6 @@ GLIBC_2.21 __stack_chk_guard D 0x4
GLIBC_2.21 __tls_get_addr F
GLIBC_2.21 _dl_mcount F
GLIBC_2.21 _r_debug D 0x14
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
diff --git a/sysdeps/unix/sysv/linux/powerpc/powerpc32/ld.abilist b/sysdeps/unix/sysv/linux/powerpc/powerpc32/ld.abilist
index 4bbfba7a61..0d033cb8bd 100644
--- a/sysdeps/unix/sysv/linux/powerpc/powerpc32/ld.abilist
+++ b/sysdeps/unix/sysv/linux/powerpc/powerpc32/ld.abilist
@@ -4,3 +4,6 @@ GLIBC_2.1 _dl_mcount F
GLIBC_2.22 __tls_get_addr_opt F
GLIBC_2.23 __parse_hwcap_and_convert_at_platform F
GLIBC_2.3 __tls_get_addr F
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
diff --git a/sysdeps/unix/sysv/linux/powerpc/powerpc64/be/ld.abilist b/sysdeps/unix/sysv/linux/powerpc/powerpc64/be/ld.abilist
index 283fb4510b..9c627b1ddf 100644
--- a/sysdeps/unix/sysv/linux/powerpc/powerpc64/be/ld.abilist
+++ b/sysdeps/unix/sysv/linux/powerpc/powerpc64/be/ld.abilist
@@ -4,3 +4,6 @@ GLIBC_2.3 __libc_stack_end D 0x8
GLIBC_2.3 __tls_get_addr F
GLIBC_2.3 _dl_mcount F
GLIBC_2.3 _r_debug D 0x28
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
diff --git a/sysdeps/unix/sysv/linux/powerpc/powerpc64/le/ld.abilist b/sysdeps/unix/sysv/linux/powerpc/powerpc64/le/ld.abilist
index b1f313c7cd..3a748c2817 100644
--- a/sysdeps/unix/sysv/linux/powerpc/powerpc64/le/ld.abilist
+++ b/sysdeps/unix/sysv/linux/powerpc/powerpc64/le/ld.abilist
@@ -4,3 +4,6 @@ GLIBC_2.17 _dl_mcount F
GLIBC_2.17 _r_debug D 0x28
GLIBC_2.22 __tls_get_addr_opt F
GLIBC_2.23 __parse_hwcap_and_convert_at_platform F
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
diff --git a/sysdeps/unix/sysv/linux/riscv/rv32/ld.abilist b/sysdeps/unix/sysv/linux/riscv/rv32/ld.abilist
index 94ca64c43d..4c67ea18d6 100644
--- a/sysdeps/unix/sysv/linux/riscv/rv32/ld.abilist
+++ b/sysdeps/unix/sysv/linux/riscv/rv32/ld.abilist
@@ -3,3 +3,6 @@ GLIBC_2.33 __stack_chk_guard D 0x4
GLIBC_2.33 __tls_get_addr F
GLIBC_2.33 _dl_mcount F
GLIBC_2.33 _r_debug D 0x14
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
diff --git a/sysdeps/unix/sysv/linux/riscv/rv64/ld.abilist b/sysdeps/unix/sysv/linux/riscv/rv64/ld.abilist
index 845f356c3c..09596f09e2 100644
--- a/sysdeps/unix/sysv/linux/riscv/rv64/ld.abilist
+++ b/sysdeps/unix/sysv/linux/riscv/rv64/ld.abilist
@@ -3,3 +3,6 @@ GLIBC_2.27 __stack_chk_guard D 0x8
GLIBC_2.27 __tls_get_addr F
GLIBC_2.27 _dl_mcount F
GLIBC_2.27 _r_debug D 0x28
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
diff --git a/sysdeps/unix/sysv/linux/rseq-internal.h b/sysdeps/unix/sysv/linux/rseq-internal.h
index 15bc7ffd6e..9e8f99fd51 100644
--- a/sysdeps/unix/sysv/linux/rseq-internal.h
+++ b/sysdeps/unix/sysv/linux/rseq-internal.h
@@ -26,7 +26,7 @@
#include <sys/rseq.h>
#ifdef RSEQ_SIG
-static inline void
+static inline bool
rseq_register_current_thread (struct pthread *self, bool do_rseq)
{
if (do_rseq)
@@ -35,15 +35,17 @@ rseq_register_current_thread (struct pthread *self, bool do_rseq)
sizeof (self->rseq_area),
0, RSEQ_SIG);
if (!INTERNAL_SYSCALL_ERROR_P (ret))
- return;
+ return true;
}
THREAD_SETMEM (self, rseq_area.cpu_id, RSEQ_CPU_ID_REGISTRATION_FAILED);
+ return false;
}
#else /* RSEQ_SIG */
-static inline void
+static inline bool
rseq_register_current_thread (struct pthread *self, bool do_rseq)
{
THREAD_SETMEM (self, rseq_area.cpu_id, RSEQ_CPU_ID_REGISTRATION_FAILED);
+ return false;
}
#endif /* RSEQ_SIG */
diff --git a/sysdeps/unix/sysv/linux/s390/s390-32/ld.abilist b/sysdeps/unix/sysv/linux/s390/s390-32/ld.abilist
index b56f005beb..2c47004bae 100644
--- a/sysdeps/unix/sysv/linux/s390/s390-32/ld.abilist
+++ b/sysdeps/unix/sysv/linux/s390/s390-32/ld.abilist
@@ -2,3 +2,6 @@ GLIBC_2.0 _r_debug D 0x14
GLIBC_2.1 __libc_stack_end D 0x4
GLIBC_2.1 _dl_mcount F
GLIBC_2.3 __tls_get_offset F
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
diff --git a/sysdeps/unix/sysv/linux/s390/s390-64/ld.abilist b/sysdeps/unix/sysv/linux/s390/s390-64/ld.abilist
index 6f788a086d..385a73a257 100644
--- a/sysdeps/unix/sysv/linux/s390/s390-64/ld.abilist
+++ b/sysdeps/unix/sysv/linux/s390/s390-64/ld.abilist
@@ -2,3 +2,6 @@ GLIBC_2.2 __libc_stack_end D 0x8
GLIBC_2.2 _dl_mcount F
GLIBC_2.2 _r_debug D 0x28
GLIBC_2.3 __tls_get_offset F
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
diff --git a/sysdeps/unix/sysv/linux/sh/be/ld.abilist b/sysdeps/unix/sysv/linux/sh/be/ld.abilist
index d155a59843..efccd6a023 100644
--- a/sysdeps/unix/sysv/linux/sh/be/ld.abilist
+++ b/sysdeps/unix/sysv/linux/sh/be/ld.abilist
@@ -2,4 +2,7 @@ GLIBC_2.2 __libc_stack_end D 0x4
GLIBC_2.2 _dl_mcount F
GLIBC_2.2 _r_debug D 0x14
GLIBC_2.3 __tls_get_addr F
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
GLIBC_2.4 __stack_chk_guard D 0x4
diff --git a/sysdeps/unix/sysv/linux/sh/le/ld.abilist b/sysdeps/unix/sysv/linux/sh/le/ld.abilist
index d155a59843..efccd6a023 100644
--- a/sysdeps/unix/sysv/linux/sh/le/ld.abilist
+++ b/sysdeps/unix/sysv/linux/sh/le/ld.abilist
@@ -2,4 +2,7 @@ GLIBC_2.2 __libc_stack_end D 0x4
GLIBC_2.2 _dl_mcount F
GLIBC_2.2 _r_debug D 0x14
GLIBC_2.3 __tls_get_addr F
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
GLIBC_2.4 __stack_chk_guard D 0x4
diff --git a/sysdeps/unix/sysv/linux/sparc/sparc32/ld.abilist b/sysdeps/unix/sysv/linux/sparc/sparc32/ld.abilist
index 0c6610e3c2..8fb5ff3ef3 100644
--- a/sysdeps/unix/sysv/linux/sparc/sparc32/ld.abilist
+++ b/sysdeps/unix/sysv/linux/sparc/sparc32/ld.abilist
@@ -2,3 +2,6 @@ GLIBC_2.0 _r_debug D 0x14
GLIBC_2.1 __libc_stack_end D 0x4
GLIBC_2.1 _dl_mcount F
GLIBC_2.3 __tls_get_addr F
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
diff --git a/sysdeps/unix/sysv/linux/sparc/sparc64/ld.abilist b/sysdeps/unix/sysv/linux/sparc/sparc64/ld.abilist
index 33f91199bf..2cc68bcf7b 100644
--- a/sysdeps/unix/sysv/linux/sparc/sparc64/ld.abilist
+++ b/sysdeps/unix/sysv/linux/sparc/sparc64/ld.abilist
@@ -2,3 +2,6 @@ GLIBC_2.2 __libc_stack_end D 0x8
GLIBC_2.2 _dl_mcount F
GLIBC_2.2 _r_debug D 0x28
GLIBC_2.3 __tls_get_addr F
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
diff --git a/sysdeps/unix/sysv/linux/sys/rseq.h b/sysdeps/unix/sysv/linux/sys/rseq.h
index c8edff50d4..1215b5d086 100644
--- a/sysdeps/unix/sysv/linux/sys/rseq.h
+++ b/sysdeps/unix/sysv/linux/sys/rseq.h
@@ -171,4 +171,14 @@ struct rseq
#endif /* __GLIBC_HAVE_KERNEL_RSEQ */
+/* Offset from the thread pointer to the rseq area. */
+extern const int __rseq_offset;
+
+/* Size of the registered rseq area. 0 if the registration was
+ unsuccessful. */
+extern const unsigned int __rseq_size;
+
+/* Flags used during rseq registration. */
+extern const unsigned int __rseq_flags;
+
#endif /* sys/rseq.h */
diff --git a/sysdeps/unix/sysv/linux/tst-rseq-disable.c b/sysdeps/unix/sysv/linux/tst-rseq-disable.c
index 000e351872..6d73f77e96 100644
--- a/sysdeps/unix/sysv/linux/tst-rseq-disable.c
+++ b/sysdeps/unix/sysv/linux/tst-rseq-disable.c
@@ -21,6 +21,7 @@
#include <support/namespace.h>
#include <support/xthread.h>
#include <sysdep.h>
+#include <thread_pointer.h>
#include <unistd.h>
#ifdef RSEQ_SIG
@@ -30,6 +31,11 @@ static void
check_rseq_disabled (void)
{
struct pthread *pd = THREAD_SELF;
+
+ TEST_COMPARE (__rseq_flags, 0);
+ TEST_VERIFY ((char *) __thread_pointer () + __rseq_offset
+ == (char *) &pd->rseq_area);
+ TEST_COMPARE (__rseq_size, 0);
TEST_COMPARE ((int) pd->rseq_area.cpu_id, RSEQ_CPU_ID_REGISTRATION_FAILED);
int ret = syscall (__NR_rseq, &pd->rseq_area, sizeof (pd->rseq_area),
diff --git a/sysdeps/unix/sysv/linux/tst-rseq.c b/sysdeps/unix/sysv/linux/tst-rseq.c
index 926376b6a5..572c11166f 100644
--- a/sysdeps/unix/sysv/linux/tst-rseq.c
+++ b/sysdeps/unix/sysv/linux/tst-rseq.c
@@ -29,12 +29,20 @@
# include <stdlib.h>
# include <string.h>
# include <syscall.h>
+# include <thread_pointer.h>
+# include <tls.h>
# include "tst-rseq.h"
static void
do_rseq_main_test (void)
{
+ struct pthread *pd = THREAD_SELF;
+
TEST_VERIFY_EXIT (rseq_thread_registered ());
+ TEST_COMPARE (__rseq_flags, 0);
+ TEST_VERIFY ((char *) __thread_pointer () + __rseq_offset
+ == (char *) &pd->rseq_area);
+ TEST_COMPARE (__rseq_size, sizeof (pd->rseq_area));
}
static void
diff --git a/sysdeps/unix/sysv/linux/x86_64/64/ld.abilist b/sysdeps/unix/sysv/linux/x86_64/64/ld.abilist
index d3cdf7611e..49a8f31c93 100644
--- a/sysdeps/unix/sysv/linux/x86_64/64/ld.abilist
+++ b/sysdeps/unix/sysv/linux/x86_64/64/ld.abilist
@@ -2,3 +2,6 @@ GLIBC_2.2.5 __libc_stack_end D 0x8
GLIBC_2.2.5 _dl_mcount F
GLIBC_2.2.5 _r_debug D 0x28
GLIBC_2.3 __tls_get_addr F
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
diff --git a/sysdeps/unix/sysv/linux/x86_64/x32/ld.abilist b/sysdeps/unix/sysv/linux/x86_64/x32/ld.abilist
index c70bccf782..ce68cc6304 100644
--- a/sysdeps/unix/sysv/linux/x86_64/x32/ld.abilist
+++ b/sysdeps/unix/sysv/linux/x86_64/x32/ld.abilist
@@ -2,3 +2,6 @@ GLIBC_2.16 __libc_stack_end D 0x4
GLIBC_2.16 __tls_get_addr F
GLIBC_2.16 _dl_mcount F
GLIBC_2.16 _r_debug D 0x14
+GLIBC_2.35 __rseq_flags D 0x4
+GLIBC_2.35 __rseq_offset D 0x4
+GLIBC_2.35 __rseq_size D 0x4
^ permalink raw reply [flat|nested] 25+ messages in thread* Re: [PATCH 5/5] nptl: Add public rseq symbols and <sys/rseq.h>
2021-12-06 13:53 ` [PATCH 5/5] nptl: Add public rseq symbols and <sys/rseq.h> Florian Weimer
@ 2021-12-07 11:01 ` Szabolcs Nagy
2021-12-07 11:28 ` Florian Weimer
0 siblings, 1 reply; 25+ messages in thread
From: Szabolcs Nagy @ 2021-12-07 11:01 UTC (permalink / raw)
To: Florian Weimer; +Cc: libc-alpha
The 12/06/2021 14:53, Florian Weimer via Libc-alpha wrote:
> The relationship between the thread pointer and the rseq area
> is made explicit. The constant offset can be used by JIT compilers
> to optimize rseq access (e.g., for really fast sched_getcpu).
>
> Extensibility is provided through __rseq_size and __rseq_flags.
> (In the future, the kernel could request a different rseq size
> via the auxiliary vector.)
>
> Co-Authored-By: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
>
> Tested on aarch64-linux-gnu, i686-linux-gnu, powerpc64le-linux-gnu,
> s390x-linux-gnu, x86_64-linux-gnu. Built with build-many-glibcs.py
> (except for a late fix to actually install the public headers).
minor comments below.
>
> ---
> NEWS | 11 +++
> manual/threads.texi | 81 ++++++++++++++++++++++
> sysdeps/nptl/dl-tls_init_tp.c | 18 ++++-
> sysdeps/unix/sysv/linux/Makefile | 3 +-
> sysdeps/unix/sysv/linux/Versions | 5 ++
> sysdeps/unix/sysv/linux/aarch64/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/alpha/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/arc/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/arm/be/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/arm/le/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/csky/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/hppa/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/i386/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/ia64/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/m68k/coldfire/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/m68k/m680x0/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/microblaze/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/mips/mips32/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/mips/mips64/n32/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/mips/mips64/n64/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/nios2/ld.abilist | 3 +
> .../unix/sysv/linux/powerpc/powerpc32/ld.abilist | 3 +
> .../sysv/linux/powerpc/powerpc64/be/ld.abilist | 3 +
> .../sysv/linux/powerpc/powerpc64/le/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/riscv/rv32/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/riscv/rv64/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/rseq-internal.h | 8 ++-
> sysdeps/unix/sysv/linux/s390/s390-32/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/s390/s390-64/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/sh/be/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/sh/le/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/sparc/sparc32/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/sparc/sparc64/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/sys/rseq.h | 10 +++
> sysdeps/unix/sysv/linux/tst-rseq-disable.c | 6 ++
> sysdeps/unix/sysv/linux/tst-rseq.c | 8 +++
> sysdeps/unix/sysv/linux/x86_64/64/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/x86_64/x32/ld.abilist | 3 +
> 38 files changed, 232 insertions(+), 5 deletions(-)
>
> diff --git a/NEWS b/NEWS
> index 1398cf2e87..8744a92532 100644
> --- a/NEWS
> +++ b/NEWS
> @@ -68,6 +68,17 @@ Major new features:
> to be used by compilers for optimizing usage of 'memcmp' when its
> return value is only used for its boolean status.
>
> +* Support for automatically registering threads with the Linux rseq
> + system call has been added. This system call is implemented starting
> + from Linux 4.18. The Restartable Sequences ABI accelerates user-space
> + operations on per-cpu data. It allows user-space to perform updates
> + on per-cpu data without requiring heavy-weight atomic operations.
> + Automatically registering threads allows all libraries, including
> + libc, to make immediate use of the rseq support by using the
> + documented ABI, via the __rseq_flags, __rseq_offset, and __rseq_size
> + variables. The GNU C Library manual has details on integration of
> + Restartable Sequences.
> +
> Deprecated and removed features, and other changes affecting compatibility:
>
> * The r_version update in the debugger interface makes the glibc binary
> diff --git a/manual/threads.texi b/manual/threads.texi
> index 06b6b277a1..7b6170cbd0 100644
> --- a/manual/threads.texi
> +++ b/manual/threads.texi
> @@ -629,6 +629,8 @@ the standard.
> * Waiting with Explicit Clocks:: Functions for waiting with an
> explicit clock specification.
> * Single-Threaded:: Detecting single-threaded execution.
> +* Restartable Sequences:: Linux-specific restartable sequences
> + integration.
> @end menu
>
> @node Default Thread Attributes
> @@ -958,6 +960,85 @@ application-created thread because future versions of @theglibc{} may
> create background threads after the first thread has been created, and
> the application has no way of knowning that these threads are present.
>
> +@node Restartable Sequences
> +@subsubsection Restartable Sequences
> +
> +This section describes restartable sequences integration for
> +@theglibc{}. This functionality is only available on Linux.
> +
> +@deftp {Data Type} {struct rseq}
> +@standards{Linux, sys/rseq.h}
> +The type of the restartable sequences area. Future versions
> +of Linux may add additional fields to the end of this structure.
> +
> +
> +Users need to obtain the address of the restartable sequences area using
> +the thread pointer and the @code{__rseq_offset} variable, described
> +below.
> +
> +One use of the restartable sequences area is to read the current CPU
> +number from its @code{cpu_id} field, as an inline version of
> +@code{sched_getcpu}. @Theglibc{} sets the @code{cpu_id} field to
> +@code{RSEQ_CPU_ID_REGISTRATION_FAILED} if registration failed or was
> +explicitly disabled.
is it better to use RSEQ_CPU_ID_UNINITIALIZED for the disabled case?
so code that uses rseq itself can distinguish enabled but failed rseq
and disabled rseq (in the latter case it would try to register, in the
former it would assume rseq is not supported).
> +
> +Furthermore, users can store the address of a @code{struct rseq_cs}
> +object into the @code{rseq_cs} field of @code{struct rseq}, thus
> +informing the kernel that the thread enters a restartable sequence
> +critical section. This pointer and the code areas it itself points to
> +must not be left pointing to memory areas which are freed or re-used.
> +Several approaches can guarantee this. If the application or library
> +can guarantee that the memory used to hold the @code{struct rseq_cs} and
> +the code areas it refers to are never freed or re-used, no special
> +action must be taken. Else, before that memory is re-used of freed, the
> +application is responsible for setting the @code{rseq_cs} field to
> +@code{NULL} in each thread's restartable sequence area to guarantee that
> +it does not leak dangling references. Because the application does not
> +typically have knowledge of libraries' use of restartable sequences, it
> +is recommended that libraries using restartable sequences which may end
> +up freeing or re-using their memory set the @code{rseq_cs} field to
> +@code{NULL} before returning from library functions which use
> +restartable sequences.
> +
> +The manual for the @code{rseq} system call can be found
> +at @uref{https://git.kernel.org/pub/scm/libs/librseq/librseq.git/tree/doc/man/rseq.2}.
> +@end deftp
> +
> +@deftypevar {int} __rseq_offset
> +@standards{Linux, sys/rseq.h}
> +This variable contains the offset between the thread pointer (as defined
> +by @code{__builtin_thread_pointer} or the thread pointer register for
> +the architecture) and the restartable sequences area. This value is the
> +same for all threads in the process. If the restartable sequences area
> +is located at a lower address than the location to whic the thread
s/whic/which/
> +pointer points, the value is negative.
> +@end deftypevar
> +
> +@deftypevar {int} __rseq_size
declaration below uses unsigned int.
> +@standards{Linux, sys/rseq.h}
> +This variable is either zero (if restartable sequence registration
> +failed or has been disabled) or the size of the restartable sequence
> +registration. This can be less can be different from the size of
s/can be less//
> +@code{struct rseq} if the kernel has extended the size of the
> +registration. If registration is successful, @code{__rseq_size} is at
> +least 32 (the initial size of @code{struct rseq}.
missing )
> +@end deftypevar
> +
> +@deftypevar {unsigned int} __rseq_flags
> +@standards{Linux, sys/rseq.h}
> +The flags used during restartable sequence registration with the kernel.
the syscall argument is documented to be int and our header
declares it as enum rseq_flags, not sure if this matters.
> +Currently zero.
> +@end deftypevar
> +
> +@deftypevr Macro int RSEQ_SIG
> +@standards{Linux, sys/rseq.h}
> +Each supported architecture provides a @code{RSEQ_SIG} macro in
> +@file{sys/rseq.h} which contains a signature. That signature is
> +expected to be present in the code before each restartable sequences
> +abort handler. Failure to provide the expected signature may terminate
> +the process with a segmentation fault.
> +@end deftypevr
> +
> @c FIXME these are undocumented:
> @c pthread_atfork
> @c pthread_attr_destroy
> diff --git a/sysdeps/nptl/dl-tls_init_tp.c b/sysdeps/nptl/dl-tls_init_tp.c
> index b39dfbff2c..82aa7083c1 100644
> --- a/sysdeps/nptl/dl-tls_init_tp.c
> +++ b/sysdeps/nptl/dl-tls_init_tp.c
> @@ -22,6 +22,7 @@
> #include <pthreadP.h>
> #include <tls.h>
> #include <rseq-internal.h>
> +#include <thread_pointer.h>
>
> #define TUNABLE_NAMESPACE pthread
> #include <dl-tunables.h>
> @@ -43,6 +44,10 @@ rtld_mutex_dummy (pthread_mutex_t *lock)
> }
> #endif
>
> +const unsigned int __rseq_flags;
> +const unsigned int __rseq_size attribute_relro;
> +const int __rseq_offset attribute_relro;
> +
> void
> __tls_pre_init_tp (void)
> {
> @@ -100,7 +105,18 @@ __tls_init_tp (void)
> #if HAVE_TUNABLES
> do_rseq = TUNABLE_GET (rseq, int, NULL);
> #endif
> - rseq_register_current_thread (pd, do_rseq);
> + if (rseq_register_current_thread (pd, do_rseq))
> + {
> + /* We need a writable view of the variables. They are in
> + .data.relro and are not yet write-protected. */
> + extern unsigned int size __asm__ ("__rseq_size");
> + size = sizeof (pd->rseq_area);
> + }
> +
> + /* This should be a compile-time constant, but the current
> + infrastructure makes it difficult to determine its value. */
> + extern int offset __asm__ ("__rseq_offset");
> + offset = (char *) &pd->rseq_area - (char *) __thread_pointer ();
> }
>
> /* Set initial thread's stack block from 0 up to __libc_stack_end.
> diff --git a/sysdeps/unix/sysv/linux/Makefile b/sysdeps/unix/sysv/linux/Makefile
> index 62a796f214..61acc1987d 100644
> --- a/sysdeps/unix/sysv/linux/Makefile
> +++ b/sysdeps/unix/sysv/linux/Makefile
> @@ -112,7 +112,8 @@ sysdep_headers += sys/mount.h sys/acct.h \
> bits/types/struct_semid64_ds_helper.h \
> bits/types/struct_shmid64_ds.h \
> bits/types/struct_shmid64_ds_helper.h \
> - bits/pthread_stack_min.h bits/pthread_stack_min-dynamic.h
> + bits/pthread_stack_min.h bits/pthread_stack_min-dynamic.h \
> + sys/rseq.h bits/rseq.h
>
> tests += tst-clone tst-clone2 tst-clone3 tst-fanotify tst-personality \
> tst-quota tst-sync_file_range tst-sysconf-iov_max tst-ttyname \
> diff --git a/sysdeps/unix/sysv/linux/Versions b/sysdeps/unix/sysv/linux/Versions
> index 26452f3f17..3f8809a158 100644
> --- a/sysdeps/unix/sysv/linux/Versions
> +++ b/sysdeps/unix/sysv/linux/Versions
> @@ -316,6 +316,11 @@ librt {
> }
>
> ld {
> + GLIBC_2.35 {
> + __rseq_flags;
> + __rseq_offset;
> + __rseq_size;
> + }
> GLIBC_PRIVATE {
> __nptl_change_stack_perm;
> }
> diff --git a/sysdeps/unix/sysv/linux/aarch64/ld.abilist b/sysdeps/unix/sysv/linux/aarch64/ld.abilist
> index 80b2fe6725..717a35f242 100644
> --- a/sysdeps/unix/sysv/linux/aarch64/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/aarch64/ld.abilist
> @@ -3,3 +3,6 @@ GLIBC_2.17 __stack_chk_guard D 0x8
> GLIBC_2.17 __tls_get_addr F
> GLIBC_2.17 _dl_mcount F
> GLIBC_2.17 _r_debug D 0x28
> +GLIBC_2.35 __rseq_flags D 0x4
> +GLIBC_2.35 __rseq_offset D 0x4
> +GLIBC_2.35 __rseq_size D 0x4
> diff --git a/sysdeps/unix/sysv/linux/alpha/ld.abilist b/sysdeps/unix/sysv/linux/alpha/ld.abilist
> index 98a03f611f..76911bd7f8 100644
> --- a/sysdeps/unix/sysv/linux/alpha/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/alpha/ld.abilist
> @@ -2,4 +2,7 @@ GLIBC_2.0 _r_debug D 0x28
> GLIBC_2.1 __libc_stack_end D 0x8
> GLIBC_2.1 _dl_mcount F
> GLIBC_2.3 __tls_get_addr F
> +GLIBC_2.35 __rseq_flags D 0x4
> +GLIBC_2.35 __rseq_offset D 0x4
> +GLIBC_2.35 __rseq_size D 0x4
> GLIBC_2.4 __stack_chk_guard D 0x8
> diff --git a/sysdeps/unix/sysv/linux/arc/ld.abilist b/sysdeps/unix/sysv/linux/arc/ld.abilist
> index 048f17c848..71c67f9803 100644
> --- a/sysdeps/unix/sysv/linux/arc/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/arc/ld.abilist
> @@ -3,3 +3,6 @@ GLIBC_2.32 __stack_chk_guard D 0x4
> GLIBC_2.32 __tls_get_addr F
> GLIBC_2.32 _dl_mcount F
> GLIBC_2.32 _r_debug D 0x14
> +GLIBC_2.35 __rseq_flags D 0x4
> +GLIBC_2.35 __rseq_offset D 0x4
> +GLIBC_2.35 __rseq_size D 0x4
> diff --git a/sysdeps/unix/sysv/linux/arm/be/ld.abilist b/sysdeps/unix/sysv/linux/arm/be/ld.abilist
> index cc8825c3bc..3859433b21 100644
> --- a/sysdeps/unix/sysv/linux/arm/be/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/arm/be/ld.abilist
> @@ -1,3 +1,6 @@
> +GLIBC_2.35 __rseq_flags D 0x4
> +GLIBC_2.35 __rseq_offset D 0x4
> +GLIBC_2.35 __rseq_size D 0x4
> GLIBC_2.4 __libc_stack_end D 0x4
> GLIBC_2.4 __stack_chk_guard D 0x4
> GLIBC_2.4 __tls_get_addr F
> diff --git a/sysdeps/unix/sysv/linux/arm/le/ld.abilist b/sysdeps/unix/sysv/linux/arm/le/ld.abilist
> index cc8825c3bc..3859433b21 100644
> --- a/sysdeps/unix/sysv/linux/arm/le/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/arm/le/ld.abilist
> @@ -1,3 +1,6 @@
> +GLIBC_2.35 __rseq_flags D 0x4
> +GLIBC_2.35 __rseq_offset D 0x4
> +GLIBC_2.35 __rseq_size D 0x4
> GLIBC_2.4 __libc_stack_end D 0x4
> GLIBC_2.4 __stack_chk_guard D 0x4
> GLIBC_2.4 __tls_get_addr F
> diff --git a/sysdeps/unix/sysv/linux/csky/ld.abilist b/sysdeps/unix/sysv/linux/csky/ld.abilist
> index 564ac09737..6bfc582b73 100644
> --- a/sysdeps/unix/sysv/linux/csky/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/csky/ld.abilist
> @@ -3,3 +3,6 @@ GLIBC_2.29 __stack_chk_guard D 0x4
> GLIBC_2.29 __tls_get_addr F
> GLIBC_2.29 _dl_mcount F
> GLIBC_2.29 _r_debug D 0x14
> +GLIBC_2.35 __rseq_flags D 0x4
> +GLIBC_2.35 __rseq_offset D 0x4
> +GLIBC_2.35 __rseq_size D 0x4
> diff --git a/sysdeps/unix/sysv/linux/hppa/ld.abilist b/sysdeps/unix/sysv/linux/hppa/ld.abilist
> index d155a59843..efccd6a023 100644
> --- a/sysdeps/unix/sysv/linux/hppa/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/hppa/ld.abilist
> @@ -2,4 +2,7 @@ GLIBC_2.2 __libc_stack_end D 0x4
> GLIBC_2.2 _dl_mcount F
> GLIBC_2.2 _r_debug D 0x14
> GLIBC_2.3 __tls_get_addr F
> +GLIBC_2.35 __rseq_flags D 0x4
> +GLIBC_2.35 __rseq_offset D 0x4
> +GLIBC_2.35 __rseq_size D 0x4
> GLIBC_2.4 __stack_chk_guard D 0x4
> diff --git a/sysdeps/unix/sysv/linux/i386/ld.abilist b/sysdeps/unix/sysv/linux/i386/ld.abilist
> index 0478e22071..1eb94ae75f 100644
> --- a/sysdeps/unix/sysv/linux/i386/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/i386/ld.abilist
> @@ -3,3 +3,6 @@ GLIBC_2.1 __libc_stack_end D 0x4
> GLIBC_2.1 _dl_mcount F
> GLIBC_2.3 ___tls_get_addr F
> GLIBC_2.3 __tls_get_addr F
> +GLIBC_2.35 __rseq_flags D 0x4
> +GLIBC_2.35 __rseq_offset D 0x4
> +GLIBC_2.35 __rseq_size D 0x4
> diff --git a/sysdeps/unix/sysv/linux/ia64/ld.abilist b/sysdeps/unix/sysv/linux/ia64/ld.abilist
> index 33f91199bf..2cc68bcf7b 100644
> --- a/sysdeps/unix/sysv/linux/ia64/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/ia64/ld.abilist
> @@ -2,3 +2,6 @@ GLIBC_2.2 __libc_stack_end D 0x8
> GLIBC_2.2 _dl_mcount F
> GLIBC_2.2 _r_debug D 0x28
> GLIBC_2.3 __tls_get_addr F
> +GLIBC_2.35 __rseq_flags D 0x4
> +GLIBC_2.35 __rseq_offset D 0x4
> +GLIBC_2.35 __rseq_size D 0x4
> diff --git a/sysdeps/unix/sysv/linux/m68k/coldfire/ld.abilist b/sysdeps/unix/sysv/linux/m68k/coldfire/ld.abilist
> index cc8825c3bc..3859433b21 100644
> --- a/sysdeps/unix/sysv/linux/m68k/coldfire/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/m68k/coldfire/ld.abilist
> @@ -1,3 +1,6 @@
> +GLIBC_2.35 __rseq_flags D 0x4
> +GLIBC_2.35 __rseq_offset D 0x4
> +GLIBC_2.35 __rseq_size D 0x4
> GLIBC_2.4 __libc_stack_end D 0x4
> GLIBC_2.4 __stack_chk_guard D 0x4
> GLIBC_2.4 __tls_get_addr F
> diff --git a/sysdeps/unix/sysv/linux/m68k/m680x0/ld.abilist b/sysdeps/unix/sysv/linux/m68k/m680x0/ld.abilist
> index 3ba474c27f..e62b2742af 100644
> --- a/sysdeps/unix/sysv/linux/m68k/m680x0/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/m68k/m680x0/ld.abilist
> @@ -2,4 +2,7 @@ GLIBC_2.0 _r_debug D 0x14
> GLIBC_2.1 __libc_stack_end D 0x4
> GLIBC_2.1 _dl_mcount F
> GLIBC_2.3 __tls_get_addr F
> +GLIBC_2.35 __rseq_flags D 0x4
> +GLIBC_2.35 __rseq_offset D 0x4
> +GLIBC_2.35 __rseq_size D 0x4
> GLIBC_2.4 __stack_chk_guard D 0x4
> diff --git a/sysdeps/unix/sysv/linux/microblaze/ld.abilist b/sysdeps/unix/sysv/linux/microblaze/ld.abilist
> index a4933c3541..5d63d74e8f 100644
> --- a/sysdeps/unix/sysv/linux/microblaze/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/microblaze/ld.abilist
> @@ -3,3 +3,6 @@ GLIBC_2.18 __stack_chk_guard D 0x4
> GLIBC_2.18 __tls_get_addr F
> GLIBC_2.18 _dl_mcount F
> GLIBC_2.18 _r_debug D 0x14
> +GLIBC_2.35 __rseq_flags D 0x4
> +GLIBC_2.35 __rseq_offset D 0x4
> +GLIBC_2.35 __rseq_size D 0x4
> diff --git a/sysdeps/unix/sysv/linux/mips/mips32/ld.abilist b/sysdeps/unix/sysv/linux/mips/mips32/ld.abilist
> index be09641a48..53ca22de2f 100644
> --- a/sysdeps/unix/sysv/linux/mips/mips32/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/mips/mips32/ld.abilist
> @@ -2,4 +2,7 @@ GLIBC_2.0 _r_debug D 0x14
> GLIBC_2.2 __libc_stack_end D 0x4
> GLIBC_2.2 _dl_mcount F
> GLIBC_2.3 __tls_get_addr F
> +GLIBC_2.35 __rseq_flags D 0x4
> +GLIBC_2.35 __rseq_offset D 0x4
> +GLIBC_2.35 __rseq_size D 0x4
> GLIBC_2.4 __stack_chk_guard D 0x4
> diff --git a/sysdeps/unix/sysv/linux/mips/mips64/n32/ld.abilist b/sysdeps/unix/sysv/linux/mips/mips64/n32/ld.abilist
> index be09641a48..53ca22de2f 100644
> --- a/sysdeps/unix/sysv/linux/mips/mips64/n32/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/mips/mips64/n32/ld.abilist
> @@ -2,4 +2,7 @@ GLIBC_2.0 _r_debug D 0x14
> GLIBC_2.2 __libc_stack_end D 0x4
> GLIBC_2.2 _dl_mcount F
> GLIBC_2.3 __tls_get_addr F
> +GLIBC_2.35 __rseq_flags D 0x4
> +GLIBC_2.35 __rseq_offset D 0x4
> +GLIBC_2.35 __rseq_size D 0x4
> GLIBC_2.4 __stack_chk_guard D 0x4
> diff --git a/sysdeps/unix/sysv/linux/mips/mips64/n64/ld.abilist b/sysdeps/unix/sysv/linux/mips/mips64/n64/ld.abilist
> index 1ea36e13f2..d1cdd68333 100644
> --- a/sysdeps/unix/sysv/linux/mips/mips64/n64/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/mips/mips64/n64/ld.abilist
> @@ -2,4 +2,7 @@ GLIBC_2.0 _r_debug D 0x28
> GLIBC_2.2 __libc_stack_end D 0x8
> GLIBC_2.2 _dl_mcount F
> GLIBC_2.3 __tls_get_addr F
> +GLIBC_2.35 __rseq_flags D 0x4
> +GLIBC_2.35 __rseq_offset D 0x4
> +GLIBC_2.35 __rseq_size D 0x4
> GLIBC_2.4 __stack_chk_guard D 0x8
> diff --git a/sysdeps/unix/sysv/linux/nios2/ld.abilist b/sysdeps/unix/sysv/linux/nios2/ld.abilist
> index 52178802dd..bcbba1823e 100644
> --- a/sysdeps/unix/sysv/linux/nios2/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/nios2/ld.abilist
> @@ -3,3 +3,6 @@ GLIBC_2.21 __stack_chk_guard D 0x4
> GLIBC_2.21 __tls_get_addr F
> GLIBC_2.21 _dl_mcount F
> GLIBC_2.21 _r_debug D 0x14
> +GLIBC_2.35 __rseq_flags D 0x4
> +GLIBC_2.35 __rseq_offset D 0x4
> +GLIBC_2.35 __rseq_size D 0x4
> diff --git a/sysdeps/unix/sysv/linux/powerpc/powerpc32/ld.abilist b/sysdeps/unix/sysv/linux/powerpc/powerpc32/ld.abilist
> index 4bbfba7a61..0d033cb8bd 100644
> --- a/sysdeps/unix/sysv/linux/powerpc/powerpc32/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/powerpc/powerpc32/ld.abilist
> @@ -4,3 +4,6 @@ GLIBC_2.1 _dl_mcount F
> GLIBC_2.22 __tls_get_addr_opt F
> GLIBC_2.23 __parse_hwcap_and_convert_at_platform F
> GLIBC_2.3 __tls_get_addr F
> +GLIBC_2.35 __rseq_flags D 0x4
> +GLIBC_2.35 __rseq_offset D 0x4
> +GLIBC_2.35 __rseq_size D 0x4
> diff --git a/sysdeps/unix/sysv/linux/powerpc/powerpc64/be/ld.abilist b/sysdeps/unix/sysv/linux/powerpc/powerpc64/be/ld.abilist
> index 283fb4510b..9c627b1ddf 100644
> --- a/sysdeps/unix/sysv/linux/powerpc/powerpc64/be/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/powerpc/powerpc64/be/ld.abilist
> @@ -4,3 +4,6 @@ GLIBC_2.3 __libc_stack_end D 0x8
> GLIBC_2.3 __tls_get_addr F
> GLIBC_2.3 _dl_mcount F
> GLIBC_2.3 _r_debug D 0x28
> +GLIBC_2.35 __rseq_flags D 0x4
> +GLIBC_2.35 __rseq_offset D 0x4
> +GLIBC_2.35 __rseq_size D 0x4
> diff --git a/sysdeps/unix/sysv/linux/powerpc/powerpc64/le/ld.abilist b/sysdeps/unix/sysv/linux/powerpc/powerpc64/le/ld.abilist
> index b1f313c7cd..3a748c2817 100644
> --- a/sysdeps/unix/sysv/linux/powerpc/powerpc64/le/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/powerpc/powerpc64/le/ld.abilist
> @@ -4,3 +4,6 @@ GLIBC_2.17 _dl_mcount F
> GLIBC_2.17 _r_debug D 0x28
> GLIBC_2.22 __tls_get_addr_opt F
> GLIBC_2.23 __parse_hwcap_and_convert_at_platform F
> +GLIBC_2.35 __rseq_flags D 0x4
> +GLIBC_2.35 __rseq_offset D 0x4
> +GLIBC_2.35 __rseq_size D 0x4
> diff --git a/sysdeps/unix/sysv/linux/riscv/rv32/ld.abilist b/sysdeps/unix/sysv/linux/riscv/rv32/ld.abilist
> index 94ca64c43d..4c67ea18d6 100644
> --- a/sysdeps/unix/sysv/linux/riscv/rv32/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/riscv/rv32/ld.abilist
> @@ -3,3 +3,6 @@ GLIBC_2.33 __stack_chk_guard D 0x4
> GLIBC_2.33 __tls_get_addr F
> GLIBC_2.33 _dl_mcount F
> GLIBC_2.33 _r_debug D 0x14
> +GLIBC_2.35 __rseq_flags D 0x4
> +GLIBC_2.35 __rseq_offset D 0x4
> +GLIBC_2.35 __rseq_size D 0x4
> diff --git a/sysdeps/unix/sysv/linux/riscv/rv64/ld.abilist b/sysdeps/unix/sysv/linux/riscv/rv64/ld.abilist
> index 845f356c3c..09596f09e2 100644
> --- a/sysdeps/unix/sysv/linux/riscv/rv64/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/riscv/rv64/ld.abilist
> @@ -3,3 +3,6 @@ GLIBC_2.27 __stack_chk_guard D 0x8
> GLIBC_2.27 __tls_get_addr F
> GLIBC_2.27 _dl_mcount F
> GLIBC_2.27 _r_debug D 0x28
> +GLIBC_2.35 __rseq_flags D 0x4
> +GLIBC_2.35 __rseq_offset D 0x4
> +GLIBC_2.35 __rseq_size D 0x4
> diff --git a/sysdeps/unix/sysv/linux/rseq-internal.h b/sysdeps/unix/sysv/linux/rseq-internal.h
> index 15bc7ffd6e..9e8f99fd51 100644
> --- a/sysdeps/unix/sysv/linux/rseq-internal.h
> +++ b/sysdeps/unix/sysv/linux/rseq-internal.h
> @@ -26,7 +26,7 @@
> #include <sys/rseq.h>
>
> #ifdef RSEQ_SIG
> -static inline void
> +static inline bool
> rseq_register_current_thread (struct pthread *self, bool do_rseq)
> {
> if (do_rseq)
> @@ -35,15 +35,17 @@ rseq_register_current_thread (struct pthread *self, bool do_rseq)
> sizeof (self->rseq_area),
> 0, RSEQ_SIG);
> if (!INTERNAL_SYSCALL_ERROR_P (ret))
> - return;
> + return true;
> }
> THREAD_SETMEM (self, rseq_area.cpu_id, RSEQ_CPU_ID_REGISTRATION_FAILED);
> + return false;
> }
> #else /* RSEQ_SIG */
> -static inline void
> +static inline bool
> rseq_register_current_thread (struct pthread *self, bool do_rseq)
> {
> THREAD_SETMEM (self, rseq_area.cpu_id, RSEQ_CPU_ID_REGISTRATION_FAILED);
> + return false;
> }
> #endif /* RSEQ_SIG */
this and the rest looks good.
^ permalink raw reply [flat|nested] 25+ messages in thread* Re: [PATCH 5/5] nptl: Add public rseq symbols and <sys/rseq.h>
2021-12-07 11:01 ` Szabolcs Nagy
@ 2021-12-07 11:28 ` Florian Weimer
0 siblings, 0 replies; 25+ messages in thread
From: Florian Weimer @ 2021-12-07 11:28 UTC (permalink / raw)
To: Szabolcs Nagy via Libc-alpha; +Cc: Szabolcs Nagy
* Szabolcs Nagy via Libc-alpha:
>> +One use of the restartable sequences area is to read the current CPU
>> +number from its @code{cpu_id} field, as an inline version of
>> +@code{sched_getcpu}. @Theglibc{} sets the @code{cpu_id} field to
>> +@code{RSEQ_CPU_ID_REGISTRATION_FAILED} if registration failed or was
>> +explicitly disabled.
>
> is it better to use RSEQ_CPU_ID_UNINITIALIZED for the disabled case?
>
> so code that uses rseq itself can distinguish enabled but failed rseq
> and disabled rseq (in the latter case it would try to register, in the
> former it would assume rseq is not supported).
I think RSEQ_CPU_ID_UNINITIALIZED was previously intended for a scenario
that is just not observable at all in the current glibc implementation,
basically an initialization ordering issue.
My expectation is that an application that knows about glibc's rseq
implementation will not want to use rseq if it has been explicitly
disabled. That tunable is only for getting old rseq users to work
again. If the application goes behind glibc's back and starts rseq
using again, it defeats the tunable (because there could be another rseq
user in the process image).
>> +@deftypevar {int} __rseq_offset
>> +@standards{Linux, sys/rseq.h}
>> +This variable contains the offset between the thread pointer (as defined
>> +by @code{__builtin_thread_pointer} or the thread pointer register for
>> +the architecture) and the restartable sequences area. This value is the
>> +same for all threads in the process. If the restartable sequences area
>> +is located at a lower address than the location to whic the thread
>
> s/whic/which/
>
>> +pointer points, the value is negative.
>> +@end deftypevar
>> +
>> +@deftypevar {int} __rseq_size
>
> declaration below uses unsigned int.
>
>> +@standards{Linux, sys/rseq.h}
>> +This variable is either zero (if restartable sequence registration
>> +failed or has been disabled) or the size of the restartable sequence
>> +registration. This can be less can be different from the size of
>
> s/can be less//
>
>> +@code{struct rseq} if the kernel has extended the size of the
>> +registration. If registration is successful, @code{__rseq_size} is at
>> +least 32 (the initial size of @code{struct rseq}.
>
> missing )
All fixed, thanks.
>> +@end deftypevar
>> +
>> +@deftypevar {unsigned int} __rseq_flags
>> +@standards{Linux, sys/rseq.h}
>> +The flags used during restartable sequence registration with the kernel.
>
> the syscall argument is documented to be int and our header
> declares it as enum rseq_flags, not sure if this matters.
It shouldn't matter, unsigned int seems more conservative. I'm always a
bit wary about enums and bitmasks.
Thanks,
Florian
^ permalink raw reply [flat|nested] 25+ messages in thread
* Re: [PATCH 0/5] Extensible rseq support for glibc
2021-12-06 13:45 [PATCH 0/5] Extensible rseq support for glibc Florian Weimer
` (4 preceding siblings ...)
2021-12-06 13:53 ` [PATCH 5/5] nptl: Add public rseq symbols and <sys/rseq.h> Florian Weimer
@ 2021-12-06 16:13 ` Mathieu Desnoyers
5 siblings, 0 replies; 25+ messages in thread
From: Mathieu Desnoyers @ 2021-12-06 16:13 UTC (permalink / raw)
To: Florian Weimer; +Cc: libc-alpha
----- On Dec 6, 2021, at 8:45 AM, Florian Weimer fweimer@redhat.com wrote:
> This series again adds rseq support, this time in such a way that the
> kernel can eventually request a larger rseq area size. Registration can
> be disabled using a new tunable.
>
> The series is based mostly on Mathieu's earlier work, especially the
> definition of RSEQ_SIG and the tests.
Hi Florian,
Thanks for doing this work. I will review it over the next few days.
Mathieu
>
> Thanks,
> Florian
>
> Florian Weimer (5):
> nptl: Add <thread_pointer.h> for defining __thread_pointer
> nptl: Add rseq registration
> Linux: Use rseq to accelerate sched_getcpu
> nptl: Add glibc.pthread.rseq tunable to control rseq registration
> nptl: Add public rseq symbols
>
> NEWS | 11 +
> csu/libc-tls.c | 6 +
> manual/threads.texi | 81 ++++++
> manual/tunables.texi | 10 +
> nptl/descr.h | 4 +
> nptl/pthread_create.c | 21 ++
> sysdeps/csky/nptl/thread_pointer.h | 25 ++
> sysdeps/hppa/nptl/thread_pointer.h | 25 ++
> sysdeps/ia64/nptl/thread_pointer.h | 19 ++
> sysdeps/m68k/nptl/thread_pointer.h | 20 ++
> sysdeps/microblaze/nptl/thread_pointer.h | 19 ++
> sysdeps/nptl/dl-tls_init_tp.c | 33 ++-
> sysdeps/nptl/dl-tunables.list | 6 +
> sysdeps/nptl/internaltypes.h | 1 +
> sysdeps/nptl/thread_pointer.h | 19 ++
> sysdeps/powerpc/nptl/thread_pointer.h | 23 ++
> sysdeps/sparc/nptl/thread_pointer.h | 19 ++
> sysdeps/unix/sysv/linux/Makefile | 17 +-
> sysdeps/unix/sysv/linux/Versions | 5 +
> sysdeps/unix/sysv/linux/aarch64/bits/rseq.h | 43 +++
> sysdeps/unix/sysv/linux/aarch64/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/alpha/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/arc/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/arm/be/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/arm/bits/rseq.h | 83 ++++++
> sysdeps/unix/sysv/linux/arm/le/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/bits/rseq.h | 29 ++
> sysdeps/unix/sysv/linux/csky/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/hppa/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/i386/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/ia64/ld.abilist | 3 +
> .../unix/sysv/linux/m68k/coldfire/ld.abilist | 3 +
> .../unix/sysv/linux/m68k/m680x0/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/microblaze/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/mips/bits/rseq.h | 62 +++++
> .../unix/sysv/linux/mips/mips32/ld.abilist | 3 +
> .../sysv/linux/mips/mips64/n32/ld.abilist | 3 +
> .../sysv/linux/mips/mips64/n64/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/nios2/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/powerpc/bits/rseq.h | 37 +++
> .../sysv/linux/powerpc/powerpc32/ld.abilist | 3 +
> .../linux/powerpc/powerpc64/be/ld.abilist | 3 +
> .../linux/powerpc/powerpc64/le/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/riscv/rv32/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/riscv/rv64/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/rseq-internal.h | 52 ++++
> sysdeps/unix/sysv/linux/s390/bits/rseq.h | 37 +++
> .../unix/sysv/linux/s390/s390-32/ld.abilist | 3 +
> .../unix/sysv/linux/s390/s390-64/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/sched_getcpu.c | 19 +-
> sysdeps/unix/sysv/linux/sh/be/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/sh/le/ld.abilist | 3 +
> .../unix/sysv/linux/sparc/sparc32/ld.abilist | 3 +
> .../unix/sysv/linux/sparc/sparc64/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/sys/rseq.h | 184 +++++++++++++
> sysdeps/unix/sysv/linux/tst-rseq-disable.c | 95 +++++++
> sysdeps/unix/sysv/linux/tst-rseq-nptl.c | 260 ++++++++++++++++++
> sysdeps/unix/sysv/linux/tst-rseq.c | 72 +++++
> sysdeps/unix/sysv/linux/tst-rseq.h | 57 ++++
> sysdeps/unix/sysv/linux/x86/bits/rseq.h | 30 ++
> sysdeps/unix/sysv/linux/x86_64/64/ld.abilist | 3 +
> sysdeps/unix/sysv/linux/x86_64/x32/ld.abilist | 3 +
> sysdeps/x86/nptl/thread_pointer.h | 33 +++
> 63 files changed, 1539 insertions(+), 5 deletions(-)
> create mode 100644 sysdeps/csky/nptl/thread_pointer.h
> create mode 100644 sysdeps/hppa/nptl/thread_pointer.h
> create mode 100644 sysdeps/ia64/nptl/thread_pointer.h
> create mode 100644 sysdeps/m68k/nptl/thread_pointer.h
> create mode 100644 sysdeps/microblaze/nptl/thread_pointer.h
> create mode 100644 sysdeps/nptl/thread_pointer.h
> create mode 100644 sysdeps/powerpc/nptl/thread_pointer.h
> create mode 100644 sysdeps/sparc/nptl/thread_pointer.h
> create mode 100644 sysdeps/unix/sysv/linux/aarch64/bits/rseq.h
> create mode 100644 sysdeps/unix/sysv/linux/arm/bits/rseq.h
> create mode 100644 sysdeps/unix/sysv/linux/bits/rseq.h
> create mode 100644 sysdeps/unix/sysv/linux/mips/bits/rseq.h
> create mode 100644 sysdeps/unix/sysv/linux/powerpc/bits/rseq.h
> create mode 100644 sysdeps/unix/sysv/linux/rseq-internal.h
> create mode 100644 sysdeps/unix/sysv/linux/s390/bits/rseq.h
> create mode 100644 sysdeps/unix/sysv/linux/sys/rseq.h
> create mode 100644 sysdeps/unix/sysv/linux/tst-rseq-disable.c
> create mode 100644 sysdeps/unix/sysv/linux/tst-rseq-nptl.c
> create mode 100644 sysdeps/unix/sysv/linux/tst-rseq.c
> create mode 100644 sysdeps/unix/sysv/linux/tst-rseq.h
> create mode 100644 sysdeps/unix/sysv/linux/x86/bits/rseq.h
> create mode 100644 sysdeps/x86/nptl/thread_pointer.h
>
>
> base-commit: 68007900beef12000ed90f38c251eaf32fbc0490
> --
> 2.33.1
--
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com
^ permalink raw reply [flat|nested] 25+ messages in thread