glibc: Perform rseq(2) registration at C startup and thread creation (v18)
Subsystems (3)
RFC PATCH 1/3 glibc: Perform rseq(2) registration at C startup and thread creation (v18)
Register rseq(2) TLS for each thread (including main), and unregister for each thread (excluding main). "rseq" stands for Restartable Sequences.
See the rseq(2) man page proposed here:
https://lkml.org/lkml/2018/9/19/647
those are based on glibc master branch commit a9bfa4353cd39ae2eae3c111844a32f9d3abbc19. The rseq(2) system call was merged into Linux 4.18.
This patch depends on "elf: Add initial flag argument to __libc_early_init" from Florian Weimer.
---
Changes since v1:
- Move __rseq_refcount to an extra field at the end of __rseq_abi to eliminate one symbol.
All libraries/programs which try to register rseq (glibc, early-adopter applications, early-adopter libraries) should use the rseq refcount. It becomes part of the ABI within a user-space process, but it's not part of the ABI shared with the kernel per se.
- Restructure how this code is organized so glibc keeps building on non-Linux targets.
- Use non-weak symbol for __rseq_abi.
- Move rseq registration/unregistration implementation into its own nptl/rseq.c compile unit.
- Move __rseq_abi symbol under GLIBC_2.29.
Changes since v2:
- Move __rseq_refcount to its own symbol, which is less ugly than trying to play tricks with the rseq uapi.
- Move __rseq_abi from nptl to csu (C start up), so it can be used across glibc, including memory allocator and sched_getcpu(). The __rseq_refcount symbol is kept in nptl, because there is no reason to use it elsewhere in glibc.
Changes since v3:
- Set __rseq_refcount TLS to 1 on register/set to 0 on unregister because glibc is the first/last user.
- Unconditionally register/unregister rseq at thread start/exit, because glibc is the first/last user.
- Add missing abilist items.
- Rebase on glibc master commit a502c5294.
- Add NEWS entry.
Changes since v4:
- Do not use "weak" symbols for __rseq_abi and __rseq_refcount. Based on "System V Application Binary Interface", weak only affects the link editor, not the dynamic linker.
- Install a new sys/rseq.h system header on Linux, which contains the RSEQ_SIG definition, __rseq_abi declaration and __rseq_refcount declaration. Move those definition/declarations from rseq-internal.h to the installed sys/rseq.h header.
- Considering that rseq is only available on Linux, move csu/rseq.c to sysdeps/unix/sysv/linux/rseq-sym.c.
- Move __rseq_refcount from nptl/rseq.c to sysdeps/unix/sysv/linux/rseq-sym.c, so it is only defined on Linux.
- Move both ABI definitions for __rseq_abi and __rseq_refcount to sysdeps/unix/sysv/linux/Versions, so they only appear on Linux.
- Document __rseq_abi and __rseq_refcount volatile.
- Document the RSEQ_SIG signature define.
- Move registration functions from rseq.c to rseq-internal.h static inline functions. Introduce empty stubs in misc/rseq-internal.h, which can be overridden by architecture code in sysdeps/unix/sysv/linux/rseq-internal.h.
- Rename __rseq_register_current_thread and __rseq_unregister_current_thread to rseq_register_current_thread and rseq_unregister_current_thread, now that those are only visible as internal static inline functions.
- Invoke rseq_register_current_thread() from libc-start.c LIBC_START_MAIN rather than nptl init, so applications not linked against libpthread.so have rseq registered for their main() thread. Note that it is invoked separately for SHARED and !SHARED builds.
Changes since v5:
- Replace __rseq_refcount by __rseq_lib_abi, which contains two
uint32_t: register_state and refcount. The "register_state" field allows inhibiting rseq registration from signal handlers nested on top of glibc registration and occuring after rseq unregistration by glibc.
- Introduce enum rseq_register_state, which contains the states allowed for the struct rseq_lib_abi register_state field.
Changes since v6:
- Introduce bits/rseq.h to define RSEQ_SIG for each architecture. The generic bits/rseq.h does not define RSEQ_SIG, meaning that each architecture implementing rseq needs to implement bits/rseq.h.
- Rename enum item RSEQ_REGISTER_NESTED to RSEQ_REGISTER_ONGOING.
- Port to glibc-2.29.
Changes since v7:
- Remove __rseq_lib_abi symbol, including refcount and register_state fields.
- Remove reference counting and nested signals handling from registration/unregistration functions.
- Introduce new __rseq_handled exported symbol, which is set to 1 by glibc on C startup when it handles restartable sequences. This allows glibc to coexist with early adopter libraries and applications wishing to register restartable sequences when it is not handled by glibc.
- Introduce rseq_init (), which sets __rseq_handled to 1 from C startup.
- Update NEWS entry.
- Update comments at the beginning of new files.
- Registration depends on both __NR_rseq and RSEQ_SIG.
- Remove ARM, powerpc, MIPS RSEQ_SIG until we agree with maintainers on the signature choice.
- Update x86, s390 RSEQ_SIG based on discussion with arch maintainers.
- Remove rseq-internal.h from headers list of misc/Makefile, so it it not installed by make install.
Changes since v8:
- Introduce RSEQ_SIG_CODE and RSEQ_SIG_DATA on aarch64 to handle compiling with -mbig-endian.
Changes since v9:
- Update Changelog.
- Remove unneeded new file comment header newlines.
Changes since v10:
- Remove volatile from __rseq_abi declaration.
- Document that __rseq_handled is about library managing rseq registration, independently of whether rseq is available or not.
- Move __rseq_handled symbol to ld.so, initialize this symbol within the dynamic linker initialization for both shared (rtld.c) and static (dl-support.c) builds.
- Only register the rseq TLS on initialization once in multiple-libc scenarios. Use rtld_active () for this purpose.
- In the static libc case, register the rseq TLS after LD_PRELOAD constructors are run, so it matches the order of this initialization vs LD_PRELOAD contructors execution for the shared libc.
- Agreed on signature choice with powerpc and MIPS maintainers, re-adding those signatures,
- The main architecture still left out signature-wise is ARM32.
Changes since v11:
- Rebase on glibc 2.30.
- Re-introduce ARM RSEQ_SIG following feedback from Will Deacon.
Changes since v12:
- Remove __rseq_handled,
- Rely on OS implicit rseq unregistration on thread teardown,
- Register main thread in __libc_early_init ().
- Add Restartable Sequences entry to threads manual.
Changes since v13:
- Update following be/le abilist split for arm, microblaze, and sh.
- Update manual to add the __rseq_abi variable and RSEQ_SIG macro to generate manual index entries, and add missing "Restartable Sequences" menu entry to the threads chapter.
Changes since v14:
- Update copyright range to include 2020.
- Introduce __ASSUME_RSEQ defined for --enable-kernel=4.18.0 and higher.
- Use ifdef __ASSUME_RSEQ rather than ifdef __NR_rseq to discover rseq availability. This is necessary now that the system call numbers are integrated within glibc.
Changes since v15:
- Remove __ASSUME_RSEQ from kernel features.
- rseq internal: remove assume rseq
- remove assume rseq and struct rseq def from sysdeps/unix/sysv/linux/rseq-sym.c
- sys/rseq.h: detect rseq header, implement fallback
- sysdeps/unix/sysv/linux/sys/rseq.h include cdefs.h, add _Static_assert to validate struct rseq and struct rseq_cs alignment.
- sys/rseq.h: document that posix_memalign should be used rather than malloc if allocating struct rseq or struct rseq_cs on the heap. This is required to guarantee 32-byte alignement.
Changes since v16:
- Move rseq NEWS entry under 2.32.
- Move new __rseq_abi symbol to GLIBC_2.32.
Changes since v17:
- Change copyright year to 2020.
- Refer to GNU C Library manual rather than rseq manpage in NEWS.
- Use "initial" parameter from __libc_early_init ().
- Manual: rseq is Linux rather than GNU std.
- Remove rseq_unregister_current_thread () (unused).
- rseq_register_current_thread () returns void.
- Coding style fixes.
- sys/rseq.h: use "32" for alignment.
- Change http:// for https:// in comments.
- Add const struct rseq_cs * field to rseq_cs union.
--- NEWS | 10 + elf/libc_early_init.c | 4 + manual/threads.texi | 30 ++- misc/rseq-internal.h | 26 +++ nptl/pthread_create.c | 13 ++ sysdeps/unix/sysv/linux/Makefile | 5 +- sysdeps/unix/sysv/linux/Versions | 1 + sysdeps/unix/sysv/linux/aarch64/bits/rseq.h | 43 ++++ sysdeps/unix/sysv/linux/aarch64/libc.abilist | 1 + sysdeps/unix/sysv/linux/alpha/libc.abilist | 1 + sysdeps/unix/sysv/linux/arm/be/libc.abilist | 1 + sysdeps/unix/sysv/linux/arm/bits/rseq.h | 83 +++++++ sysdeps/unix/sysv/linux/arm/le/libc.abilist | 1 + sysdeps/unix/sysv/linux/bits/rseq.h | 29 +++ sysdeps/unix/sysv/linux/csky/libc.abilist | 1 + sysdeps/unix/sysv/linux/hppa/libc.abilist | 1 + sysdeps/unix/sysv/linux/i386/libc.abilist | 1 + sysdeps/unix/sysv/linux/ia64/libc.abilist | 1 + .../sysv/linux/m68k/coldfire/libc.abilist | 1 + .../unix/sysv/linux/m68k/m680x0/libc.abilist | 1 + .../sysv/linux/microblaze/be/libc.abilist | 1 + .../sysv/linux/microblaze/le/libc.abilist | 1 + sysdeps/unix/sysv/linux/mips/bits/rseq.h | 62 ++++++ .../sysv/linux/mips/mips32/fpu/libc.abilist | 1 + .../sysv/linux/mips/mips32/nofpu/libc.abilist | 1 + .../sysv/linux/mips/mips64/n32/libc.abilist | 1 + .../sysv/linux/mips/mips64/n64/libc.abilist | 1 + sysdeps/unix/sysv/linux/nios2/libc.abilist | 1 + sysdeps/unix/sysv/linux/powerpc/bits/rseq.h | 37 ++++ .../linux/powerpc/powerpc32/fpu/libc.abilist | 1 + .../powerpc/powerpc32/nofpu/libc.abilist | 1 + .../linux/powerpc/powerpc64/be/libc.abilist | 1 + .../linux/powerpc/powerpc64/le/libc.abilist | 1 + .../unix/sysv/linux/riscv/rv64/libc.abilist | 1 + sysdeps/unix/sysv/linux/rseq-internal.h | 47 ++++ sysdeps/unix/sysv/linux/rseq-sym.c | 26 +++ sysdeps/unix/sysv/linux/s390/bits/rseq.h | 37 ++++ .../unix/sysv/linux/s390/s390-32/libc.abilist | 1 + .../unix/sysv/linux/s390/s390-64/libc.abilist | 1 + sysdeps/unix/sysv/linux/sh/be/libc.abilist | 1 + sysdeps/unix/sysv/linux/sh/le/libc.abilist | 1 + .../sysv/linux/sparc/sparc32/libc.abilist | 1 + .../sysv/linux/sparc/sparc64/libc.abilist | 1 + sysdeps/unix/sysv/linux/sys/rseq.h | 207 ++++++++++++++++++ sysdeps/unix/sysv/linux/x86/bits/rseq.h | 30 +++ .../unix/sysv/linux/x86_64/64/libc.abilist | 1 + .../unix/sysv/linux/x86_64/x32/libc.abilist | 1 + 47 files changed, 716 insertions(+), 4 deletions(-) create mode 100644 misc/rseq-internal.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/rseq-sym.c 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/x86/bits/rseq.h
NEWS +10 -0
@@ -18,6 +18,16 @@ Major new features: * The GNU C Library now loads audit modules listed in the DT_AUDIT and DT_DEPAUDIT dynamic section entries of the main executable. +* Support for automatically registering threads with the Linux rseq(2) + 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(2) support by using the documented ABI. + The GNU C Library manual has details on integration of Restartable + Sequences. + Deprecated and removed features, and other changes affecting compatibility: * The deprecated <sys/sysctl.h> header and the sysctl function have been
elf/libc_early_init.c +4 -0
@@ -18,10 +18,14 @@#include <ctype.h> #include <libc-early-init.h> +#include <rseq-internal.h> void __libc_early_init (_Bool initial) {/* Initialize ctype data. */ __ctype_init (); +/* Register rseq ABI to the kernel for the main program's libc. */ +if (initial) +rseq_register_current_thread (); }
manual/threads.texi +28 -2
@@ -9,8 +9,10 @@ This chapter describes functions used for managing threads. POSIX threads. @menu -* ISO C Threads:: Threads based on the ISO C specification. -* POSIX Threads:: Threads based on the POSIX specification. +* ISO C Threads:: Threads based on the ISO C specification. +* POSIX Threads:: Threads based on the POSIX specification. +* Restartable Sequences:: Linux-specific Restartable Sequences + integration. @end menu
@@ -881,3 +883,27 @@ Behaves like @code{pthread_timedjoin_np} except that the absolute time in @c pthread_spin_unlock @c pthread_testcancel @c pthread_yield + +@node Restartable Sequences +@section Restartable Sequences +@cindex rseq + +This section describes @theglibc{} Restartable Sequences integration. + +@deftypevar {struct rseq} __rseq_abi +@standards{Linux, sys/rseq.h} +@Theglibc{} implements a @code{__rseq_abi} TLS symbol to interact with the +Restartable Sequences system call (Linux-specific). The layout of this +structure is defined by the Linux kernel @file{linux/rseq.h} UAPI. +Registration of each thread's @code{__rseq_abi} is performed by +@theglibc{} at libc initialization and pthread creation. +@end deftypevar + +@deftypevr Macro int RSEQ_SIG +@standards{Linux, sys/rseq.h} +Each supported architecture provide 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
misc/rseq-internal.h new file +26 -0
@@ -0,0 +1,26 @@ +/* Restartable Sequences internal API. Stub version. +Copyright (C) 2020 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 + +static inline void +rseq_register_current_thread (void ) +{ +} + +#endif /* rseq-internal.h */
nptl/pthread_create.c +13 -0
@@ -33,6 +33,7 @@#include <default-sched.h> #include <futex-internal.h> #include <tls-setup.h> +#include <rseq-internal.h> #include "libioP.h" #include <shlib-compat.h>
@@ -384,6 +385,9 @@ START_THREAD_DEFN/* Initialize pointers to locale data. */ __ctype_init (); +/* Register rseq TLS to the kernel. */ +rseq_register_current_thread (); +#ifndef __ASSUME_SET_ROBUST_LIST if (__set_robust_list_avail >= 0)#endif
@@ -578,6 +582,15 @@ START_THREAD_DEFNprocess 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). */ __exit_thread ();
sysdeps/unix/sysv/linux/Makefile +3 -2
@@ -41,7 +41,7 @@ update-syscall-lists: arch-syscall.h endif ifeq ($(subdir),csu) -sysdep_routines += errno-loc +sysdep_routines += errno-loc rseq-sym endif ifeq ($(subdir),assert)
@@ -90,7 +90,8 @@ sysdep_headers += sys/mount.h sys/acct.h \ bits/termios-baud.h bits/termios-c_cflag.h \ bits/termios-c_lflag.h bits/termios-tcflow.h \ bits/termios-misc.h \ - bits/ipc-perm.h + bits/ipc-perm.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 \
sysdeps/unix/sysv/linux/Versions +1 -0
@@ -178,6 +178,7 @@ libc { getdents64; gettid; tgkill; } GLIBC_2.32 { + __rseq_abi; } GLIBC_PRIVATE { # functions used in other libraries
sysdeps/unix/sysv/linux/aarch64/bits/rseq.h new file +43 -0
43 lines not on this page · raw message
sysdeps/unix/sysv/linux/aarch64/libc.abilist +1 -0
@@ -2146,4 +2146,5 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F
sysdeps/unix/sysv/linux/alpha/libc.abilist +1 -0
@@ -2226,6 +2226,7 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F GLIBC_2.4 _IO_fprintf F GLIBC_2.4 _IO_printf F
sysdeps/unix/sysv/linux/arm/be/libc.abilist +1 -0
@@ -133,6 +133,7 @@ GLIBC_2.30 twalk_r F GLIBC_2.31 msgctl F GLIBC_2.31 semctl F GLIBC_2.31 shmctl F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F GLIBC_2.4 _Exit F GLIBC_2.4 _IO_2_1_stderr_ D 0xa0
sysdeps/unix/sysv/linux/arm/bits/rseq.h new file +83 -0
83 lines not on this page · raw message
sysdeps/unix/sysv/linux/arm/le/libc.abilist +1 -0
@@ -130,6 +130,7 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F GLIBC_2.4 _Exit F GLIBC_2.4 _IO_2_1_stderr_ D 0xa0
sysdeps/unix/sysv/linux/bits/rseq.h new file +29 -0
@@ -0,0 +1,29 @@ +/* Restartable Sequences architecture header. Stub version. +Copyright (C) 2020 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. */
sysdeps/unix/sysv/linux/csky/libc.abilist +1 -0
@@ -2090,4 +2090,5 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F
sysdeps/unix/sysv/linux/hppa/libc.abilist +1 -0
@@ -2047,6 +2047,7 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F
sysdeps/unix/sysv/linux/i386/libc.abilist +1 -0
@@ -2213,6 +2213,7 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F
sysdeps/unix/sysv/linux/ia64/libc.abilist +1 -0
@@ -2079,6 +2079,7 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F
sysdeps/unix/sysv/linux/m68k/coldfire/libc.abilist +1 -0
@@ -134,6 +134,7 @@ GLIBC_2.30 twalk_r F GLIBC_2.31 msgctl F GLIBC_2.31 semctl F GLIBC_2.31 shmctl F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F GLIBC_2.4 _Exit F GLIBC_2.4 _IO_2_1_stderr_ D 0x98
sysdeps/unix/sysv/linux/m68k/m680x0/libc.abilist +1 -0
@@ -2159,6 +2159,7 @@ GLIBC_2.30 twalk_r F GLIBC_2.31 msgctl F GLIBC_2.31 semctl F GLIBC_2.31 shmctl F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F
sysdeps/unix/sysv/linux/microblaze/be/libc.abilist +1 -0
@@ -2141,4 +2141,5 @@ GLIBC_2.30 twalk_r F GLIBC_2.31 msgctl F GLIBC_2.31 semctl F GLIBC_2.31 shmctl F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F
sysdeps/unix/sysv/linux/microblaze/le/libc.abilist +1 -0
@@ -2138,4 +2138,5 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F
sysdeps/unix/sysv/linux/mips/bits/rseq.h new file +62 -0
62 lines not on this page · raw message
sysdeps/unix/sysv/linux/mips/mips32/fpu/libc.abilist +1 -0
@@ -2130,6 +2130,7 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F
sysdeps/unix/sysv/linux/mips/mips32/nofpu/libc.abilist +1 -0
@@ -2128,6 +2128,7 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F
sysdeps/unix/sysv/linux/mips/mips64/n32/libc.abilist +1 -0
@@ -2136,6 +2136,7 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F
sysdeps/unix/sysv/linux/mips/mips64/n64/libc.abilist +1 -0
@@ -2130,6 +2130,7 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F
sysdeps/unix/sysv/linux/nios2/libc.abilist +1 -0
@@ -2179,4 +2179,5 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F
sysdeps/unix/sysv/linux/powerpc/bits/rseq.h new file +37 -0
@@ -0,0 +1,37 @@ +/* Restartable Sequences Linux powerpc architecture header. +Copyright (C) 2020 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
sysdeps/unix/sysv/linux/powerpc/powerpc32/fpu/libc.abilist +1 -0
@@ -2186,6 +2186,7 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F GLIBC_2.4 _IO_fprintf F GLIBC_2.4 _IO_printf F
sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/libc.abilist +1 -0
@@ -2219,6 +2219,7 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F GLIBC_2.4 _IO_fprintf F GLIBC_2.4 _IO_printf F
sysdeps/unix/sysv/linux/powerpc/powerpc64/be/libc.abilist +1 -0
@@ -2049,6 +2049,7 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F GLIBC_2.4 _IO_fprintf F GLIBC_2.4 _IO_printf F
sysdeps/unix/sysv/linux/powerpc/powerpc64/le/libc.abilist +1 -0
@@ -2248,4 +2248,5 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F
sysdeps/unix/sysv/linux/riscv/rv64/libc.abilist +1 -0
@@ -2108,4 +2108,5 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F
sysdeps/unix/sysv/linux/rseq-internal.h new file +47 -0
47 lines not on this page · raw message
sysdeps/unix/sysv/linux/rseq-sym.c new file +26 -0
@@ -0,0 +1,26 @@ +/* Restartable Sequences exported symbols. Linux Implementation. +Copyright (C) 2020 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 <sys/syscall.h> +#include <stdint.h> +#include <kernel-features.h> +#include <sys/rseq.h> + +__threadstruct rseq __rseq_abi = + { + .cpu_id =RSEQ_CPU_ID_UNINITIALIZED , + };
sysdeps/unix/sysv/linux/s390/bits/rseq.h new file +37 -0
@@ -0,0 +1,37 @@ +/* Restartable Sequences Linux s390 architecture header. +Copyright (C) 2020 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
sysdeps/unix/sysv/linux/s390/s390-32/libc.abilist +1 -0
@@ -2184,6 +2184,7 @@ GLIBC_2.30 twalk_r F GLIBC_2.31 msgctl F GLIBC_2.31 semctl F GLIBC_2.31 shmctl F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F GLIBC_2.4 _IO_fprintf F GLIBC_2.4 _IO_printf F
sysdeps/unix/sysv/linux/s390/s390-64/libc.abilist +1 -0
@@ -2085,6 +2085,7 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F GLIBC_2.4 _IO_fprintf F GLIBC_2.4 _IO_printf F
sysdeps/unix/sysv/linux/sh/be/libc.abilist +1 -0
@@ -2054,6 +2054,7 @@ GLIBC_2.30 twalk_r F GLIBC_2.31 msgctl F GLIBC_2.31 semctl F GLIBC_2.31 shmctl F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F
sysdeps/unix/sysv/linux/sh/le/libc.abilist +1 -0
@@ -2051,6 +2051,7 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F
sysdeps/unix/sysv/linux/sparc/sparc32/libc.abilist +1 -0
@@ -2175,6 +2175,7 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F GLIBC_2.4 _IO_fprintf F GLIBC_2.4 _IO_printf F
sysdeps/unix/sysv/linux/sparc/sparc64/libc.abilist +1 -0
@@ -2102,6 +2102,7 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F
sysdeps/unix/sysv/linux/sys/rseq.h new file +207 -0
207 lines not on this page · raw message
sysdeps/unix/sysv/linux/x86/bits/rseq.h new file +30 -0
@@ -0,0 +1,30 @@ +/* Restartable Sequences Linux x86 architecture header. +Copyright (C) 2020 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
sysdeps/unix/sysv/linux/x86_64/64/libc.abilist +1 -0
@@ -2060,6 +2060,7 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F
sysdeps/unix/sysv/linux/x86_64/x32/libc.abilist +1 -0
@@ -2159,4 +2159,5 @@ GLIBC_2.30 getdents64 F GLIBC_2.30 gettid F GLIBC_2.30 tgkill F GLIBC_2.30 twalk_r F +GLIBC_2.32 __rseq_abi T 0x20 GLIBC_2.32 pthread_sigmask F
Signature · 1 line
2.17.1
RFC PATCH 2/3 glibc: sched_getcpu(): use rseq cpu_id TLS on Linux (v7)
When available, use the cpu_id field from __rseq_abi on Linux to implement sched_getcpu(). Fall-back on the vgetcpu vDSO if unavailable.
Benchmarks:
x86-64: Intel E5-2630 [email protected], 16-core, hyperthreading
glibc sched_getcpu(): 13.7 ns (baseline) glibc sched_getcpu() using rseq: 2.5 ns (speedup: 5.5x)
inline load cpuid from __rseq_abi TLS: 0.8 ns (speedup: 17.1x)
---
Changes since v1:
- rseq is only used if both __NR_rseq and RSEQ_SIG are defined.
Changes since v2:
- remove duplicated __rseq_abi extern declaration.
Changes since v3:
- update ChangeLog.
Changes since v4:
- Use atomic_load_relaxed to load the __rseq_abi.cpu_id field, a consequence of the fact that __rseq_abi is not volatile anymore.
- Include atomic.h which provides atomic_load_relaxed.
Changes since v5:
- Use __ASSUME_RSEQ to detect rseq availability.
Changes since v6:
- Remove use of __ASSUME_RSEQ.
--- sysdeps/unix/sysv/linux/sched_getcpu.c | 27 ++++++++++++++++++++++++-- 1 file changed, 25 insertions(+), 2 deletions(-)
sysdeps/unix/sysv/linux/sched_getcpu.c +25 -2
@@ -18,10 +18,15 @@#include <errno.h> #include <sched.h> #include <sysdep.h> +#include <atomic.h> #include <sysdep-vdso.h> -int -sched_getcpu (void ) +#ifdef HAVE_GETCPU_VSYSCALL +# define HAVE_VSYSCALL +#endif + +static int +vsyscall_sched_getcpu (void ) {unsigned int cpu;int r =-1 ;
@@ -32,3 +37,21 @@ sched_getcpu (void)#endif return r ==-1 ? r : cpu; } + +#include <sys/rseq.h> + +#ifdef RSEQ_SIG +int +sched_getcpu (void ) +{ +int cpu_id =atomic_load_relaxed (&__rseq_abi.cpu_id); + +return cpu_id >=0 ? cpu_id :vsyscall_sched_getcpu (); +} +#else +int +sched_getcpu (void ) +{ +return vsyscall_sched_getcpu (); +} +#endif
Signature · 1 line
2.17.1
* Mathieu Desnoyers:
Quoted 7 lines
NEWS +0
-0@@ -18,6 +18,16 @@ Major new features: * The GNU C Library now loads audit modules listed in the DT_AUDIT and DT_DEPAUDIT dynamic section entries of the main executable.
NEWS Major new features: · line 29
Quoted 10 lines
NEWS +9
-0+* Support for automatically registering threads with the Linux rseq(2) + 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(2) support by using the documented ABI. + The GNU C Library manual has details on integration of Restartable + Sequences.
GNU style doesn't use (2) here, I think.
manual/threads.texi Behaves like @code{pthread_timedjoin_np} except that the absolute time in · line 889
Quoted 13 lines
manual/threads.texi +4
-0@@ -881,3 +883,27 @@ Behaves like @code{pthread_timedjoin_np} except that the absolute time in @c pthread_spin_unlock @c pthread_testcancel @c pthread_yield + +@node Restartable Sequences +@section Restartable Sequences +@cindex rseq
Suggest: @cindex Restartable Sequences
manual/threads.texi Behaves like @code{pthread_timedjoin_np} except that the absolute time in · line 891
Quoted 2 lines
manual/threads.texi +2
-0+ +This section describes @theglibc{} Restartable Sequences integration.
Suggest: This section describes Restartable Sequences integration for @theglibc{}. (Avoids an excessively long noun phrase.)
Maybe mention which uses of the rseq syscall are permitted behind the back of glibc? And that code should not leave dangling rseq cs pointers behind (the dlopen interaction)?
manual/threads.texi Behaves like @code{pthread_timedjoin_np} except that the absolute time in · line 897
Quoted 5 lines
manual/threads.texi +5
-0+@deftypevar {struct rseq} __rseq_abi +@standards{Linux, sys/rseq.h} +@Theglibc{} implements a @code{__rseq_abi} TLS symbol to interact with the +Restartable Sequences system call (Linux-specific). The layout of this +structure is defined by the Linux kernel @file{linux/rseq.h} UAPI.
The linux/rseq.h reference seems redundant, given that sys/rseq.h covers it as well.
manual/threads.texi Behaves like @code{pthread_timedjoin_np} except that the absolute time in · line 899
Quoted 2 lines
manual/threads.texi +2
-0+Registration of each thread's @code{__rseq_abi} is performed by +@theglibc{} at libc initialization and pthread creation.
Suggest: library initialization and thread creation
manual/threads.texi Behaves like @code{pthread_timedjoin_np} except that the absolute time in · line 904
Quoted 5 lines
manual/threads.texi +5
-0+@end deftypevar + +@deftypevr Macro int RSEQ_SIG +@standards{Linux, sys/rseq.h} +Each supported architecture provide a @code{RSEQ_SIG} macro in
Typo: provides
manual/threads.texi Behaves like @code{pthread_timedjoin_np} except that the absolute time in · line 908
Quoted 4 lines
manual/threads.texi +4
-0+@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.
Suggest: segmentation fault (no capitalization)
misc/rseq-internal.h
Quoted 5 lines
misc/rseq-internal.h new file +0
-0
Maybe this should go in to sysdeps/generic instead of misc? (See the recent discussion about elf_machine_sym_no_match.)
Quoted 34 lines
sysdeps/unix/sysv/linux/rseq-internal.h new file +28
-0@@ -0,0 +1,47 @@ +/* Restartable Sequences internal API. Linux implementation. +Copyright (C) 2020 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 <sys/rseq.h> + +#ifdef RSEQ_SIG +static inline void +rseq_register_current_thread (void )
sysdeps/unix/sysv/linux/rseq-internal.h line 38
Quoted 10 lines
sysdeps/unix/sysv/linux/rseq-internal.h new file +10
-0+{ +int ret; + +if (__rseq_abi.cpu_id ==RSEQ_CPU_ID_REGISTRATION_FAILED ) +return ; + ret =INTERNAL_SYSCALL_CALL (rseq, &__rseq_abi,sizeof (struct rseq ), +0 ,RSEQ_SIG ); +if (INTERNAL_SYSCALL_ERROR_P (ret) && +INTERNAL_SYSCALL_ERRNO (ret) !=EBUSY ) + __rseq_abi.cpu_id =RSEQ_CPU_ID_REGISTRATION_FAILED ;
Sorry, I forgot: Please add a comment that the EBUSY error is ignored because registration may have already happened in a legacy library.
sysdeps/unix/sysv/linux/sys/rseq.h line 44
Quoted 13 lines
sysdeps/unix/sysv/linux/sys/rseq.h new file +7
-0+#ifdef __GLIBC_HAVE_KERNEL_RSEQ +/* We use the structures declarations from the kernel headers. */ +# include <linux/rseq.h> +#else +/* We use a copy of the include/uapi/linux/rseq.h kernel header. */ + +#include <asm/byteorder.h>
Missing “# include“ indentation.
Quoted 1 line
+#ifdef
__LP64__
Likewise (more indentation needed below, include double-space indentation).
sysdeps/unix/sysv/linux/sys/rseq.h line 205
Quoted 7 lines
sysdeps/unix/sysv/linux/sys/rseq.h new file +7
-0+/* Allocations of struct rseq and struct rseq_cs on the heap need to +be aligned on 32 bytes. Therefore, use of malloc is discouraged +because it does not guarantee alignment. posix_memalign should be +used instead. */ + +extern __thread struct rseq __rseq_abi +__attribute__ ((tls_model ("initial-exec")));
Please indent the __attribute__ with two spaces.
Actual code looks good now. Thanks. I don't think there are any remaining issues except maybe more documentation.
Florian
Quoted 10 lines · ----- On Apr 30, 2020, at 8:20 AM, Florian Weimer [email protected] wrote:
* Mathieu Desnoyers:
Quoted 8 lines
NEWS +0
-0@@ -18,6 +18,16 @@ Major new features: * The GNU C Library now loads audit modules listed in the DT_AUDIT and DT_DEPAUDIT dynamic section entries of the main executable.
Quoted 11 lines
NEWS Major new features: · line 29
Quoted 9 lines
NEWS +9
-0+* Support for automatically registering threads with the Linux rseq(2) + 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(2) support by using the documented ABI. + The GNU C Library manual has details on integration of Restartable + Sequences.GNU style doesn't use (2) here, I think.
OK
Quoted 6 lines
Quoted 5 lines
manual/threads.texi +0
-0
Quoted 11 lines
Quoted 9 lines
manual/threads.texi +0
-0@@ -881,3 +883,27 @@ Behaves like @code{pthread_timedjoin_np} except that theabsolute time in
@cpthread_spin_unlock
@cpthread_testcancel
@cpthread_yield
+
+@node Restartable Sequences
+@section Restartable Sequences
+@cindex rseqSuggest: @cindex Restartable Sequences
OK
Quoted 6 lines
manual/threads.texi
Quoted 3 lines
manual/threads.texi +2
-0+ +This section describes @theglibc{} Restartable Sequences integration.Suggest: This section describes Restartable Sequences integration for @theglibc{}. (Avoids an excessively long noun phrase.)
OK
Quoted 4 lines
Maybe mention which uses of the rseq syscall are permitted behind the back of glibc? And that code should not leave dangling rseq cs pointers behind (the dlopen interaction)?
Here is the entire updated section, let me know if I missed anything:
@deftypevar {struct rseq} __rseq_abi
@standards{Linux, sys/rseq.h}
@Theglibc{} implements a @code{__rseq_abi} TLS symbol to interact with the Restartable Sequences system call (Linux-specific). The layout of this structure is defined by the @file{sys/rseq.h} header. Registration of each thread's @code{__rseq_abi} is performed by @theglibc{} at libc library initialization and thread creation.
The main executable and shared libraries may either have an undefined @code{__rseq_abi} TLS symbol, or define their own, with the same declaration as the one present in @file{sys/rseq.h}. The dynamic linker will ensure that only one of those available symbols will be used at runtime across the process.
If the main executable or shared libraries observe an uninitialized @code{__rseq_abi.cpu_id} field (value @code{RSEQ_CPU_ID_UNINITIALIZED}), they may perform rseq registration to the kernel: this means either glibc was prevented from doing the registration, or an older glibc version, which does not include rseq support, is in use. When the main executable or a library thus takes ownership of the registration, the memory used to hold the @code{__rseq_abi} TLS variable must stay allocated, and is not re-used, until the very end of the thread lifetime or until an explicit rseq unregistration for that thread is performed. It is not recommended to dlclose() libraries owning the @code{__rseq_abi} TLS variable.
Users of the @code{__rseq_abi} TLS symbol can store the address of a @code{struct rseq_cs} to the @code{__rseq_abi.rseq_cs.uptr.ptr} TLS variable, thus informing the kernel that it 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_abi.rseq_cs.uptr.ptr} TLS variable to @code{NULL} in each thread's TLS 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_abi.rseq_cs.uptr.ptr} TLS variable to @code{NULL} before returning from library functions which use Restartable Sequences.
Quoted 9 lines
manual/threads.texi Behaves like @code{pthread_timedjoin_np} except that the absolute time in · line 897
Quoted 6 lines
+@deftypevar {struct rseq}
__rseq_abi
+@standards{Linux,sys/rseq.h}
+@Theglibc{} implements a @code{__rseq_abi} TLS symbol to interact with themanual/threads.texi +2
-0+Restartable Sequences system call (Linux-specific). The layout of this +structure is defined by the Linux kernel @file{linux/rseq.h} UAPI.The
linux/rseq.hreference seems redundant, given thatsys/rseq.hcovers it as well.
OK
Quoted 5 lines
Quoted 3 lines
+Registration of each thread's @code{__rseq_abi} is performed by +@theglibc{} at libc initialization and pthread creation.
Suggest: library initialization and thread creation
OK
Quoted 8 lines
manual/threads.texi Behaves like @code{pthread_timedjoin_np} except that the absolute time in · line 904
Quoted 6 lines
+@end deftypevar
manual/threads.texi +4
-0+ +@deftypevr Macro int RSEQ_SIG +@standards{Linux, sys/rseq.h} +Each supported architecture provide a @code{RSEQ_SIG} macro inTypo: provides
OK
Quoted 7 lines
manual/threads.texi Behaves like @code{pthread_timedjoin_np} except that the absolute time in · line 908
Quoted 5 lines
+@file{sys/rseq.h} which contains a signature. That signature is expected to be
manual/threads.texi +3
-0+present in the code before each Restartable Sequences abort handler. Failure +to provide the expected signature may terminate the process with a Segmentation +fault.Suggest: segmentation fault (no capitalization)
OK
Quoted 9 lines
misc/rseq-internal.h
Quoted 6 lines
misc/rseq-internal.h new file +0
-0Maybe this should go in to sysdeps/generic instead of misc? (See the recent discussion about
elf_machine_sym_no_match.)
OK
Quoted 41 lines
Quoted 41 lines
sysdeps/unix/sysv/linux/rseq-internal.h new file +33
-0@@ -0,0 +1,47 @@ +/* Restartable Sequences internal API. Linux implementation. +Copyright (C) 2020 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 <sys/rseq.h> + +#ifdef RSEQ_SIG +static inline void +rseq_register_current_thread (void) +{ + int ret; + + if (__rseq_abi.cpu_id == RSEQ_CPU_ID_REGISTRATION_FAILED) + return;
Quoted 8 lines
sysdeps/unix/sysv/linux/rseq-internal.h line 38
Quoted 5 lines
sysdeps/unix/sysv/linux/rseq-internal.h new file +5
-0+ ret =INTERNAL_SYSCALL_CALL (rseq, &__rseq_abi,sizeof (struct rseq ), +0 ,RSEQ_SIG ); +if (INTERNAL_SYSCALL_ERROR_P (ret) && +INTERNAL_SYSCALL_ERRNO (ret) !=EBUSY ) + __rseq_abi.cpu_id =RSEQ_CPU_ID_REGISTRATION_FAILED ;Sorry, I forgot: Please add a comment that the EBUSY error is ignored because registration may have already happened in a legacy library.
Considering that we now disable signals across thread creation, and that glibc's initialization happens before other libraries' constructors (as far as I remember even before LD_PRELOADed library constructors), in which scenario can we expect to have EBUSY here ?
Not setting __rseq_abi.cpu_id to RSEQ_CPU_ID_REGISTRATION_FAILED in case of EBUSY is more a way to handle "unforeseen" scenarios where somehow the registration would already be done. But I cannot find an "expected" scenario which would lead to this now.
So if EBUSY really is unexpected, how should we treat that ? I don't think setting REGISTRATION_FAILED would be appropriate, because then it would break assumption of the prior successful registration that have already been done by this thread.
Quoted 8 lines
Quoted 7 lines
sysdeps/unix/sysv/linux/sys/rseq.h new file +0
-0
Quoted 9 lines
sysdeps/unix/sysv/linux/sys/rseq.h line 44
Quoted 7 lines
sysdeps/unix/sysv/linux/sys/rseq.h new file +7
-0+#ifdef __GLIBC_HAVE_KERNEL_RSEQ +/* We use the structures declarations from the kernel headers. */ +# include <linux/rseq.h> +#else +/* We use a copy of the include/uapi/linux/rseq.h kernel header. */ + +#include <asm/byteorder.h> Missing “# include“ indentation.
OK
Quoted 5 lines
Quoted 2 lines
+#ifdef
__LP64__Likewise (more indentation needed below, include double-space indentation).
OK
Quoted 10 lines
sysdeps/unix/sysv/linux/sys/rseq.h line 205
Quoted 8 lines
sysdeps/unix/sysv/linux/sys/rseq.h new file +7
-0+/* Allocations of struct rseq and struct rseq_cs on the heap need to +be aligned on 32 bytes. Therefore, use of malloc is discouraged +because it does not guarantee alignment. posix_memalign should be +used instead. */ + +extern __thread struct rseq __rseq_abi +__attribute__ ((tls_model ("initial-exec")));Please indent the
__attribute__with two spaces.
OK
Quoted 3 lines
Actual code looks good now. Thanks. I don't think there are any remaining issues except maybe more documentation.
I raised a few questions in my reply to yours, so I'll wait for your input on those topics before the next iteration.
Thanks,
Mathieu
Signature · 3 lines
Mathieu Desnoyers EfficiOS Inc. http://www.efficios.com
* Mathieu Desnoyers:
Quoted 7 lines
@deftypevar {struct rseq}
__rseq_abi
@standards{Linux,sys/rseq.h}
@Theglibc{} implements a @code{__rseq_abi} TLS symbol to interact with the Restartable Sequences system call (Linux-specific). The layout of this structure is defined by the @file{sys/rseq.h} header. Registration of each thread's @code{__rseq_abi} is performed by @theglibc{} at libc library initialization and thread creation.
s/libc library/library/
Quoted 6 lines
The main executable and shared libraries may either have an undefined @code{__rseq_abi} TLS symbol, or define their own, with the same declaration as the one present in @file{sys/rseq.h}. The dynamic linker will ensure that only one of those available symbols will be used at runtime across the process.
Quoted 10 lines
If the main executable or shared libraries observe an uninitialized @code{__rseq_abi.cpu_id} field (value @code{RSEQ_CPU_ID_UNINITIALIZED}), they may perform rseq registration to the kernel: this means either glibc was prevented from doing the registration, or an older glibc version, which does not include rseq support, is in use. When the main executable or a library thus takes ownership of the registration, the memory used to hold the @code{__rseq_abi} TLS variable must stay allocated, and is not re-used, until the very end of the thread lifetime or until an explicit rseq unregistration for that thread is performed. It is not recommended to
dlclose()libraries owning the @code{__rseq_abi} TLS variable.
s/dlclose()/@code{dlclose}/ (no parentheses)
Rest looks okay.
Quoted 8 lines
Quoted 8 lines
Quoted 7 lines
sysdeps/unix/sysv/linux/rseq-internal.h new file +7
-0+if (__rseq_abi.cpu_id ==RSEQ_CPU_ID_REGISTRATION_FAILED ) +return ; + ret =INTERNAL_SYSCALL_CALL (rseq, &__rseq_abi,sizeof (struct rseq ), +0 ,RSEQ_SIG ); +if (INTERNAL_SYSCALL_ERROR_P (ret) && +INTERNAL_SYSCALL_ERRNO (ret) !=EBUSY ) + __rseq_abi.cpu_id =RSEQ_CPU_ID_REGISTRATION_FAILED ;
Quoted 7 lines
Quoted 2 lines
Sorry, I forgot: Please add a comment that the EBUSY error is ignored because registration may have already happened in a legacy library.
Considering that we now disable signals across thread creation, and that glibc's initialization happens before other libraries' constructors (as far as I remember even before
LD_PRELOADedlibrary constructors), in which scenario can we expect to have EBUSY here ?
That's a good point.
Quoted 9 lines
Not setting __rseq_abi.cpu_id to
RSEQ_CPU_ID_REGISTRATION_FAILEDin case of EBUSY is more a way to handle "unforeseen" scenarios where somehow the registration would already be done. But I cannot find an "expected" scenario which would lead to this now.So if EBUSY really is unexpected, how should we treat that ? I don't think setting
REGISTRATION_FAILEDwould be appropriate, because then it would break assumption of the prior successful registration that have already been done by this thread.
You could call __libc_fatal with an error message. ENOSYS is definitely an expected error code here, and EPERM (and perhaps EACCES) can happen with seccomp filters.
Thanks,
Florian
Quoted 2 lines · ----- On Apr 30, 2020, at 12:36 PM, Florian Weimer [email protected] wrote:
* Mathieu Desnoyers:
[...]
Quoted 24 lines
Quoted 9 lines
Quoted 9 lines
Quoted 8 lines
sysdeps/unix/sysv/linux/rseq-internal.h new file +7
-0+if (__rseq_abi.cpu_id ==RSEQ_CPU_ID_REGISTRATION_FAILED ) +return ; + ret =INTERNAL_SYSCALL_CALL (rseq, &__rseq_abi,sizeof (struct rseq ), +0 ,RSEQ_SIG ); +if (INTERNAL_SYSCALL_ERROR_P (ret) && +INTERNAL_SYSCALL_ERRNO (ret) !=EBUSY ) + __rseq_abi.cpu_id =RSEQ_CPU_ID_REGISTRATION_FAILED ;Quoted 7 lines
Quoted 2 lines
Sorry, I forgot: Please add a comment that the EBUSY error is ignored because registration may have already happened in a legacy library.
Considering that we now disable signals across thread creation, and that glibc's initialization happens before other libraries' constructors (as far as I remember even before
LD_PRELOADedlibrary constructors), in which scenario can we expect to have EBUSY here ?That's a good point.
Quoted 5 lines
Not setting __rseq_abi.cpu_id to
RSEQ_CPU_ID_REGISTRATION_FAILEDin case of EBUSY is more a way to handle "unforeseen" scenarios where somehow the registration would already be done. But I cannot find an "expected" scenario which would lead to this now.
Quoted 8 lines
Quoted 4 lines
So if EBUSY really is unexpected, how should we treat that ? I don't think setting
REGISTRATION_FAILEDwould be appropriate, because then it would break assumption of the prior successful registration that have already been done by this thread.You could call
__libc_fatalwith an error message. ENOSYS is definitely an expected error code here, and EPERM (and perhaps EACCES) can happen with seccomp filters.
If we go this way, I'd also recommend to treat any situation where __rseq_abi.cpu_id is already initialized as a fatal error. Does the code below seem OK to you ?
static inline voidrseq_register_current_thread (void)
{
int ret;
if (__rseq_abi.cpu_id != RSEQ_CPU_ID_UNINITIALIZED)
__libc_fatal ("rseq already initialized for this thread\n");
ret = INTERNAL_SYSCALL_CALL (rseq, &__rseq_abi, sizeof (struct rseq),
0, RSEQ_SIG);
if (INTERNAL_SYSCALL_ERROR_P (ret))
{
if (INTERNAL_SYSCALL_ERRNO (ret) == EBUSY)
__libc_fatal ("rseq already registered for this thread\n");
__rseq_abi.cpu_id = RSEQ_CPU_ID_REGISTRATION_FAILED;
}
}Thanks,
Mathieu
Signature · 3 lines
Mathieu Desnoyers EfficiOS Inc. http://www.efficios.com
* Mathieu Desnoyers:
Quoted 11 lines
If we go this way, I'd also recommend to treat any situation where __rseq_abi.cpu_id is already initialized as a fatal error. Does the code below seem OK to you ?
static inline void
rseq_register_current_thread(void)
{
int ret;if (__rseq_abi.cpu_id !=
RSEQ_CPU_ID_UNINITIALIZED)__libc_fatal ("rseq already initialized for this thread\n");
Agreed; this should work because this code runs after relocation processing.
Quoted 9 lines
ret =
INTERNAL_SYSCALL_CALL(rseq,&__rseq_abi, sizeof (struct rseq),0, RSEQ_SIG); if (INTERNAL_SYSCALL_ERROR_P (ret)) { if (INTERNAL_SYSCALL_ERRNO (ret) == EBUSY) __libc_fatal ("rseq already registered for this thread\n"); __rseq_abi.cpu_id = RSEQ_CPU_ID_REGISTRATION_FAILED; } }
__libc_fatal does not attribute the error to glibc, so I suggest to start the error messages with “glibc fatal error: ”, so that people know where to look.
Thanks,
Florian
----- On Apr 30, 2020, at 1:07 PM, Florian Weimer [email protected] wrote:
[...]
Quoted 3 lines
__libc_fataldoes not attribute the error to glibc, so I suggest to start the error messages with “glibc fatal error: ”, so that people know where to look.
OK. Is there a strict requirement on limiting to 80 columns for code including an error message string in glibc ? IOW:
if (__rseq_abi.cpu_id != RSEQ_CPU_ID_UNINITIALIZED)
__libc_fatal ("glibc fatal error: rseq already initialized for this thread\n");or
if (__rseq_abi.cpu_id != RSEQ_CPU_ID_UNINITIALIZED)
__libc_fatal ("glibc fatal error: "
"rseq already initialized for this thread\n");
?
Thanks,
Mathieu
Signature · 3 lines
Mathieu Desnoyers EfficiOS Inc. http://www.efficios.com