glibc: Perform rseq(2) registration at C startup and thread creation (v8)
Subsystems (3)
A message that is not in the archive (sent off-list or not yet received): [email protected]
PATCH 1/5 glibc: Perform rseq(2) registration at C startup and thread creation (v8)
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
This patch is based on glibc-2.29. The rseq(2) system call was merged into Linux 4.18.
---
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.
--- NEWS | 15 ++++ csu/libc-start.c | 14 ++- misc/rseq-internal.h | 39 ++++++++ nptl/pthread_create.c | 9 ++ sysdeps/unix/sysv/linux/Makefile | 4 +- sysdeps/unix/sysv/linux/Versions | 4 + sysdeps/unix/sysv/linux/aarch64/bits/rseq.h | 32 +++++++ sysdeps/unix/sysv/linux/aarch64/libc.abilist | 2 + sysdeps/unix/sysv/linux/alpha/libc.abilist | 2 + sysdeps/unix/sysv/linux/arm/libc.abilist | 2 + sysdeps/unix/sysv/linux/bits/rseq.h | 30 +++++++ sysdeps/unix/sysv/linux/csky/libc.abilist | 2 + sysdeps/unix/sysv/linux/hppa/libc.abilist | 2 + sysdeps/unix/sysv/linux/i386/libc.abilist | 2 + sysdeps/unix/sysv/linux/ia64/libc.abilist | 2 + .../sysv/linux/m68k/coldfire/libc.abilist | 2 + .../unix/sysv/linux/m68k/m680x0/libc.abilist | 2 + .../unix/sysv/linux/microblaze/libc.abilist | 2 + .../sysv/linux/mips/mips32/fpu/libc.abilist | 2 + .../sysv/linux/mips/mips32/nofpu/libc.abilist | 2 + .../sysv/linux/mips/mips64/n32/libc.abilist | 2 + .../sysv/linux/mips/mips64/n64/libc.abilist | 2 + sysdeps/unix/sysv/linux/nios2/libc.abilist | 2 + .../linux/powerpc/powerpc32/fpu/libc.abilist | 2 + .../powerpc/powerpc32/nofpu/libc.abilist | 2 + .../linux/powerpc/powerpc64/be/libc.abilist | 2 + .../linux/powerpc/powerpc64/le/libc.abilist | 2 + .../unix/sysv/linux/riscv/rv64/libc.abilist | 2 + sysdeps/unix/sysv/linux/rseq-internal.h | 89 +++++++++++++++++++ sysdeps/unix/sysv/linux/rseq-sym.c | 64 +++++++++++++ sysdeps/unix/sysv/linux/s390/bits/rseq.h | 31 +++++++ .../unix/sysv/linux/s390/s390-32/libc.abilist | 2 + .../unix/sysv/linux/s390/s390-64/libc.abilist | 2 + sysdeps/unix/sysv/linux/sh/libc.abilist | 2 + .../sysv/linux/sparc/sparc32/libc.abilist | 2 + .../sysv/linux/sparc/sparc64/libc.abilist | 2 + sysdeps/unix/sysv/linux/sys/rseq.h | 51 +++++++++++ sysdeps/unix/sysv/linux/x86/bits/rseq.h | 31 +++++++ .../unix/sysv/linux/x86_64/64/libc.abilist | 2 + .../unix/sysv/linux/x86_64/x32/libc.abilist | 2 + 40 files changed, 462 insertions(+), 5 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/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 +15 -0
@@ -5,6 +5,21 @@ See the end for copying conditions. Please send GNU C library bug reports via <https://sourceware.org/bugzilla/> using `glibc' in the "product" field. +Version 2.30 + +Major new features: + +* 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. + See 'man 2 rseq' for the details of the ABI shared between libc and the + kernel. + + Version 2.29 Major new features:
csu/libc-start.c +11 -3
@@ -22,6 +22,7 @@#include <ldsodefs.h> #include <exit-thread.h> #include <libc-internal.h> +#include <rseq-internal.h> #include <elf/dl-tunables.h>
@@ -140,7 +141,12 @@ LIBC_START_MAIN (int (*main) (int, char **, char ** MAIN_AUXVEC_DECL), __libc_multiple_libcs = &_dl_starting_up && !_dl_starting_up; -#ifndef SHARED +rseq_init (); + +#ifdef SHARED +/* Register rseq ABI to the kernel. */ + (void )rseq_register_current_thread (); +#else _dl_relocate_static_pie ();char **ev = &argv[argc +1 ];
@@ -218,6 +224,9 @@ LIBC_START_MAIN (int (*main) (int, char **, char ** MAIN_AUXVEC_DECL), }# endif +/* Register rseq ABI to the kernel. */ + (void )rseq_register_current_thread (); +/* Initialize libpthread if linked in. */ if (__pthread_initialize_minimal !=NULL )__pthread_initialize_minimal ();
@@ -230,8 +239,7 @@ LIBC_START_MAIN (int (*main) (int, char **, char ** MAIN_AUXVEC_DECL),# else __pointer_chk_guard_local = pointer_chk_guard;# endif - -#endif /* !SHARED */ +#endif /* Register the destructor of the dynamic linker if there is any. */ if (__glibc_likely (rtld_fini !=NULL ))
misc/rseq-internal.h new file +39 -0
@@ -0,0 +1,39 @@ +/* Restartable Sequences internal API. Stub version. + +Copyright (C) 2019 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 +<http://www.gnu.org/licenses/>. */ + +#ifndef RSEQ_INTERNAL_H +#define RSEQ_INTERNAL_H + +static inline int +rseq_register_current_thread (void ) +{ +return -1 ; +} + +static inline int +rseq_unregister_current_thread (void ) +{ +return -1 ; +} + +static inline int +rseq_init (void ) +{ +} + +#endif /* rseq-internal.h */
nptl/pthread_create.c +9 -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>
@@ -378,6 +379,7 @@ __free_tcb (struct pthread *pd) START_THREAD_DEFN {struct pthread *pd =START_THREAD_SELF ; +bool has_rseq = false;#if HP_TIMING_AVAIL /* Remember the time when the thread was started. */
@@ -396,6 +398,9 @@ START_THREAD_DEFNif (__glibc_unlikely (atomic_exchange_acq (&pd->setxid_futex,0 ) ==-2 ))futex_wake (&pd->setxid_futex,1 ,FUTEX_PRIVATE ); +/* Register rseq TLS to the kernel. */ + has_rseq = !rseq_register_current_thread (); +#ifdef __NR_set_robust_list# ifndef __ASSUME_SET_ROBUST_LISTif (__set_robust_list_avail >=0 )
@@ -573,6 +578,10 @@ START_THREAD_DEFN }#endif +/* Unregister rseq TLS from kernel. */ +if (has_rseq &&rseq_unregister_current_thread ()) +abort (); +advise_stack_range (pd->stackblock, pd->stackblock_size, (uintptr_t ) pd, pd->guardsize);
sysdeps/unix/sysv/linux/Makefile +2 -2
@@ -1,5 +1,5 @@ ifeq ($(subdir),csu) -sysdep_routines += errno-loc +sysdep_routines += errno-loc rseq-sym endif ifeq ($(subdir),assert)
@@ -48,7 +48,7 @@ sysdep_headers += sys/mount.h sys/acct.h sys/sysctl.h \ bits/termios-c_iflag.h bits/termios-c_oflag.h \ bits/termios-baud.h bits/termios-c_cflag.h \ bits/termios-c_lflag.h bits/termios-tcflow.h \ - bits/termios-misc.h + bits/termios-misc.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 +4 -0
@@ -174,6 +174,10 @@ libc { GLIBC_2.29 { getcpu; } + GLIBC_2.30 { + __rseq_abi; + __rseq_handled; + } GLIBC_PRIVATE { # functions used in other libraries __syscall_rt_sigqueueinfo;
sysdeps/unix/sysv/linux/aarch64/bits/rseq.h new file +32 -0
@@ -0,0 +1,32 @@ +/* Restartable Sequences Linux aarch64 architecture header. + +Copyright (C) 2019 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 +<http://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. */ + +#define RSEQ_SIG 0xd428bc00/* BRK #0x45E0. */
sysdeps/unix/sysv/linux/aarch64/libc.abilist +2 -0
@@ -2141,3 +2141,5 @@ GLIBC_2.28 thrd_yield F GLIBC_2.29 getcpu F GLIBC_2.29 posix_spawn_file_actions_addchdir_np F GLIBC_2.29 posix_spawn_file_actions_addfchdir_np F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4
sysdeps/unix/sysv/linux/alpha/libc.abilist +2 -0
@@ -2204,6 +2204,8 @@ GLIBC_2.3.4 setipv4sourcefilter F GLIBC_2.3.4 setsourcefilter F GLIBC_2.3.4 xdr_quad_t F GLIBC_2.3.4 xdr_u_quad_t F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4 GLIBC_2.4 _IO_fprintf F GLIBC_2.4 _IO_printf F GLIBC_2.4 _IO_sprintf F
sysdeps/unix/sysv/linux/arm/libc.abilist +2 -0
@@ -126,6 +126,8 @@ GLIBC_2.28 thrd_yield F GLIBC_2.29 getcpu F GLIBC_2.29 posix_spawn_file_actions_addchdir_np F GLIBC_2.29 posix_spawn_file_actions_addfchdir_np F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4 GLIBC_2.4 _Exit F GLIBC_2.4 _IO_2_1_stderr_ D 0xa0 GLIBC_2.4 _IO_2_1_stdin_ D 0xa0
sysdeps/unix/sysv/linux/bits/rseq.h new file +30 -0
@@ -0,0 +1,30 @@ +/* Restartable Sequences architecture header. Stub version. + +Copyright (C) 2019 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 +<http://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 +2 -0
@@ -2085,3 +2085,5 @@ GLIBC_2.29 xdrstdio_create F GLIBC_2.29 xencrypt F GLIBC_2.29 xprt_register F GLIBC_2.29 xprt_unregister F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4
sysdeps/unix/sysv/linux/hppa/libc.abilist +2 -0
@@ -2037,6 +2037,8 @@ GLIBC_2.3.4 setipv4sourcefilter F GLIBC_2.3.4 setsourcefilter F GLIBC_2.3.4 xdr_quad_t F GLIBC_2.3.4 xdr_u_quad_t F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4 GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F
sysdeps/unix/sysv/linux/i386/libc.abilist +2 -0
@@ -2203,6 +2203,8 @@ GLIBC_2.3.4 setsourcefilter F GLIBC_2.3.4 vm86 F GLIBC_2.3.4 xdr_quad_t F GLIBC_2.3.4 xdr_u_quad_t F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4 GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F
sysdeps/unix/sysv/linux/ia64/libc.abilist +2 -0
@@ -2069,6 +2069,8 @@ GLIBC_2.3.4 setipv4sourcefilter F GLIBC_2.3.4 setsourcefilter F GLIBC_2.3.4 xdr_quad_t F GLIBC_2.3.4 xdr_u_quad_t F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4 GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F
sysdeps/unix/sysv/linux/m68k/coldfire/libc.abilist +2 -0
@@ -127,6 +127,8 @@ GLIBC_2.28 thrd_yield F GLIBC_2.29 getcpu F GLIBC_2.29 posix_spawn_file_actions_addchdir_np F GLIBC_2.29 posix_spawn_file_actions_addfchdir_np F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4 GLIBC_2.4 _Exit F GLIBC_2.4 _IO_2_1_stderr_ D 0x98 GLIBC_2.4 _IO_2_1_stdin_ D 0x98
sysdeps/unix/sysv/linux/m68k/m680x0/libc.abilist +2 -0
@@ -2146,6 +2146,8 @@ GLIBC_2.3.4 setipv4sourcefilter F GLIBC_2.3.4 setsourcefilter F GLIBC_2.3.4 xdr_quad_t F GLIBC_2.3.4 xdr_u_quad_t F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4 GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F
sysdeps/unix/sysv/linux/microblaze/libc.abilist +2 -0
@@ -2133,3 +2133,5 @@ GLIBC_2.28 thrd_yield F GLIBC_2.29 getcpu F GLIBC_2.29 posix_spawn_file_actions_addchdir_np F GLIBC_2.29 posix_spawn_file_actions_addfchdir_np F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4
sysdeps/unix/sysv/linux/mips/mips32/fpu/libc.abilist +2 -0
@@ -2120,6 +2120,8 @@ GLIBC_2.3.4 setipv4sourcefilter F GLIBC_2.3.4 setsourcefilter F GLIBC_2.3.4 xdr_quad_t F GLIBC_2.3.4 xdr_u_quad_t F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4 GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F
sysdeps/unix/sysv/linux/mips/mips32/nofpu/libc.abilist +2 -0
@@ -2118,6 +2118,8 @@ GLIBC_2.3.4 setipv4sourcefilter F GLIBC_2.3.4 setsourcefilter F GLIBC_2.3.4 xdr_quad_t F GLIBC_2.3.4 xdr_u_quad_t F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4 GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F
sysdeps/unix/sysv/linux/mips/mips64/n32/libc.abilist +2 -0
@@ -2126,6 +2126,8 @@ GLIBC_2.3.4 setipv4sourcefilter F GLIBC_2.3.4 setsourcefilter F GLIBC_2.3.4 xdr_quad_t F GLIBC_2.3.4 xdr_u_quad_t F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4 GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F
sysdeps/unix/sysv/linux/mips/mips64/n64/libc.abilist +2 -0
@@ -2120,6 +2120,8 @@ GLIBC_2.3.4 setipv4sourcefilter F GLIBC_2.3.4 setsourcefilter F GLIBC_2.3.4 xdr_quad_t F GLIBC_2.3.4 xdr_u_quad_t F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4 GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F
sysdeps/unix/sysv/linux/nios2/libc.abilist +2 -0
@@ -2174,3 +2174,5 @@ GLIBC_2.28 thrd_yield F GLIBC_2.29 getcpu F GLIBC_2.29 posix_spawn_file_actions_addchdir_np F GLIBC_2.29 posix_spawn_file_actions_addfchdir_np F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4
sysdeps/unix/sysv/linux/powerpc/powerpc32/fpu/libc.abilist +2 -0
@@ -2164,6 +2164,8 @@ GLIBC_2.3.4 siglongjmp F GLIBC_2.3.4 swapcontext F GLIBC_2.3.4 xdr_quad_t F GLIBC_2.3.4 xdr_u_quad_t F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4 GLIBC_2.4 _IO_fprintf F GLIBC_2.4 _IO_printf F GLIBC_2.4 _IO_sprintf F
sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/libc.abilist +2 -0
@@ -2197,6 +2197,8 @@ GLIBC_2.3.4 siglongjmp F GLIBC_2.3.4 swapcontext F GLIBC_2.3.4 xdr_quad_t F GLIBC_2.3.4 xdr_u_quad_t F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4 GLIBC_2.4 _IO_fprintf F GLIBC_2.4 _IO_printf F GLIBC_2.4 _IO_sprintf F
sysdeps/unix/sysv/linux/powerpc/powerpc64/be/libc.abilist +2 -0
@@ -2027,6 +2027,8 @@ GLIBC_2.3.4 siglongjmp F GLIBC_2.3.4 swapcontext F GLIBC_2.3.4 xdr_quad_t F GLIBC_2.3.4 xdr_u_quad_t F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4 GLIBC_2.4 _IO_fprintf F GLIBC_2.4 _IO_printf F GLIBC_2.4 _IO_sprintf F
sysdeps/unix/sysv/linux/powerpc/powerpc64/le/libc.abilist +2 -0
@@ -2231,3 +2231,5 @@ GLIBC_2.28 thrd_yield F GLIBC_2.29 getcpu F GLIBC_2.29 posix_spawn_file_actions_addchdir_np F GLIBC_2.29 posix_spawn_file_actions_addfchdir_np F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4
sysdeps/unix/sysv/linux/riscv/rv64/libc.abilist +2 -0
@@ -2103,3 +2103,5 @@ GLIBC_2.28 thrd_yield F GLIBC_2.29 getcpu F GLIBC_2.29 posix_spawn_file_actions_addchdir_np F GLIBC_2.29 posix_spawn_file_actions_addfchdir_np F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4
sysdeps/unix/sysv/linux/rseq-internal.h new file +89 -0
89 lines not on this page · raw message
sysdeps/unix/sysv/linux/rseq-sym.c new file +64 -0
64 lines not on this page · raw message
sysdeps/unix/sysv/linux/s390/bits/rseq.h new file +31 -0
@@ -0,0 +1,31 @@ +/* Restartable Sequences Linux s390 architecture header. + +Copyright (C) 2019 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 +<http://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 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 +2 -0
@@ -2159,6 +2159,8 @@ GLIBC_2.3.4 setipv4sourcefilter F GLIBC_2.3.4 setsourcefilter F GLIBC_2.3.4 xdr_quad_t F GLIBC_2.3.4 xdr_u_quad_t F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4 GLIBC_2.4 _IO_fprintf F GLIBC_2.4 _IO_printf F GLIBC_2.4 _IO_sprintf F
sysdeps/unix/sysv/linux/s390/s390-64/libc.abilist +2 -0
@@ -2063,6 +2063,8 @@ GLIBC_2.3.4 setipv4sourcefilter F GLIBC_2.3.4 setsourcefilter F GLIBC_2.3.4 xdr_quad_t F GLIBC_2.3.4 xdr_u_quad_t F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4 GLIBC_2.4 _IO_fprintf F GLIBC_2.4 _IO_printf F GLIBC_2.4 _IO_sprintf F
sysdeps/unix/sysv/linux/sh/libc.abilist +2 -0
@@ -2041,6 +2041,8 @@ GLIBC_2.3.4 setipv4sourcefilter F GLIBC_2.3.4 setsourcefilter F GLIBC_2.3.4 xdr_quad_t F GLIBC_2.3.4 xdr_u_quad_t F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4 GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F
sysdeps/unix/sysv/linux/sparc/sparc32/libc.abilist +2 -0
@@ -2153,6 +2153,8 @@ GLIBC_2.3.4 setipv4sourcefilter F GLIBC_2.3.4 setsourcefilter F GLIBC_2.3.4 xdr_quad_t F GLIBC_2.3.4 xdr_u_quad_t F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4 GLIBC_2.4 _IO_fprintf F GLIBC_2.4 _IO_printf F GLIBC_2.4 _IO_sprintf F
sysdeps/unix/sysv/linux/sparc/sparc64/libc.abilist +2 -0
@@ -2092,6 +2092,8 @@ GLIBC_2.3.4 setipv4sourcefilter F GLIBC_2.3.4 setsourcefilter F GLIBC_2.3.4 xdr_quad_t F GLIBC_2.3.4 xdr_u_quad_t F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4 GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F
sysdeps/unix/sysv/linux/sys/rseq.h new file +51 -0
51 lines not on this page · raw message
sysdeps/unix/sysv/linux/x86/bits/rseq.h new file +31 -0
@@ -0,0 +1,31 @@ +/* Restartable Sequences Linux x86 architecture header. + +Copyright (C) 2019 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 +<http://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 +2 -0
@@ -2050,6 +2050,8 @@ GLIBC_2.3.4 setipv4sourcefilter F GLIBC_2.3.4 setsourcefilter F GLIBC_2.3.4 xdr_quad_t F GLIBC_2.3.4 xdr_u_quad_t F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4 GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F
sysdeps/unix/sysv/linux/x86_64/x32/libc.abilist +2 -0
@@ -2149,3 +2149,5 @@ GLIBC_2.28 thrd_yield F GLIBC_2.29 getcpu F GLIBC_2.29 posix_spawn_file_actions_addchdir_np F GLIBC_2.29 posix_spawn_file_actions_addfchdir_np F +GLIBC_2.30 __rseq_abi T 0x20 +GLIBC_2.30 __rseq_handled D 0x4
Signature · 1 line
2.17.1
----- On Apr 16, 2019, at 1:32 PM, Mathieu Desnoyers [email protected] wrote:
[...]
Quoted 29 lines
sysdeps/unix/sysv/linux/aarch64/bits/rseq.h new file +22
-0@@ -0,0 +1,32 @@ +/* Restartable Sequences Linux aarch64 architecture header. + +Copyright (C) 2019 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 +<http://www.gnu.org/licenses/>. */ + +#ifndef _SYS_RSEQ_H +# error "Never use <bits/rseq.h> directly; include <sys/rseq.h> instead." +#endif +
sysdeps/unix/sysv/linux/aarch64/bits/rseq.h line 32
Quoted 10 lines
sysdeps/unix/sysv/linux/aarch64/bits/rseq.h new file +10
-0+/* 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. */ + +#define RSEQ_SIG 0xd428bc00/* BRK #0x45E0. */
After further investigation, we should probably do the following to handle compiling with -mbig-endian on aarch64, which generates binaries with mixed code vs data endianness (little endian code, big endian data):
#ifdef __ARM_BIG_ENDIAN
#define RSEQ_SIG 0x00bc28d4 /* BRK #0x45E0. */
#else
#define RSEQ_SIG 0xd428bc00 /* BRK #0x45E0. */
#endif
Else mismatch between code endianness for the generated signatures and data endianness for the RSEQ_SIG parameter passed to the rseq registration will trigger application segmentation faults when the kernel try to abort rseq critical sections.
For ARM32, the situation is a bit more complex. Only armv6+ generates mixed-endianness code vs data with -mbig-endian. Prior to armv6, the code and data endianness matches. Therefore, I plan to #ifdef the reversed endianness handling with:
#if __ARM_ARCH >= 6 && __ARM_BIG_ENDIAN
on arm32.
Thoughts ?
Thanks,
Mathieu
Signature · 3 lines
Mathieu Desnoyers EfficiOS Inc. http://www.efficios.com
Quoted 7 lines · Mathieu Desnoyers
Quoted 7 lines
sysdeps/unix/sysv/linux/aarch64/bits/rseq.h new file +7
-0+/* 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
Quoted 8 lines
sysdeps/unix/sysv/linux/aarch64/bits/rseq.h line 32
Quoted 3 lines
sysdeps/unix/sysv/linux/aarch64/bits/rseq.h new file +3
-0+execution efficiency in some cases. */ + +#define RSEQ_SIG 0xd428bc00/* BRK #0x45E0. */ After further investigation, we should probably do the following to handle compiling with -mbig-endian on aarch64, which generates binaries with mixed code vs data endianness (little endian code, big endian data):
First, the comment on RSEQ_SIG should specify whether it is to be interpreted in the code or the data endianness.
Quoted 8 lines
For ARM32, the situation is a bit more complex. Only armv6+ generates mixed-endianness code vs data with -mbig-endian. Prior to armv6, the code and data endianness matches. Therefore, I plan to #ifdef the reversed endianness handling with:
#if
__ARM_ARCH>= 6 &&__ARM_BIG_ENDIANon arm32.
That doesn't work well because BE code (.o files) can be built for v5te (for example) and used on a range of different architecture variants with both BE32 and BE8 - the choice between BE32 and BE8 is a link-time choice, not a compile-time choice. So if the value for Arm is a compile-time constant, it should also work for both BE32 and BE8.
In turn, that suggests to me that RSEQ_SIG should be defined to be a value that is always in the code endianness (and whatever corresponding kernel code handles RSEQ_SIG values should act accordingly on architectures where the two endiannesses can differ). If the kernel ABI is already fixed in a way that prevents such a definition of RSEQ_SIG semantics as using code endianness, a value should be chosen for Arm that works for both endiannesses.
(Also, installed glibc headers are supposed to work with older compilers, and support for __ARM_ARCH was only added in GCC 4.8. Before that you need to test lots of separate macros for different architecture variants to determine a version number.)
Signature · 2 lines
Joseph S. Myers [email protected]
Quoted 18 lines · ----- On Apr 17, 2019, at 12:17 PM, Joseph Myers [email protected] wrote:
Quoted 7 lines · Mathieu Desnoyers
Quoted 7 lines
sysdeps/unix/sysv/linux/aarch64/bits/rseq.h new file +7
-0+/* 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 Quoted 8 lines
sysdeps/unix/sysv/linux/aarch64/bits/rseq.h line 32
Quoted 3 lines
sysdeps/unix/sysv/linux/aarch64/bits/rseq.h new file +3
-0+execution efficiency in some cases. */ + +#define RSEQ_SIG 0xd428bc00/* BRK #0x45E0. */ After further investigation, we should probably do the following to handle compiling with -mbig-endian on aarch64, which generates binaries with mixed code vs data endianness (little endian code, big endian data):
Quoted 2 lines
First, the comment on
RSEQ_SIGshould specify whether it is to be interpreted in the code or the data endianness.
Right. The signature passed as argument to the rseq registration system call needs to be in data endianness (currently exposed kernel ABI).
Ideally for userspace, we want to define a signature in code endianness that happens to nicely match specific code patterns.
Quoted 8 lines
Quoted 8 lines
For ARM32, the situation is a bit more complex. Only armv6+ generates mixed-endianness code vs data with -mbig-endian. Prior to armv6, the code and data endianness matches. Therefore, I plan to #ifdef the reversed endianness handling with:
#if
__ARM_ARCH>= 6 &&__ARM_BIG_ENDIAN
Quoted 7 lines
Quoted 1 line
on arm32.
That doesn't work well because BE code (.o files) can be built for v5te (for example) and used on a range of different architecture variants with both BE32 and BE8 - the choice between BE32 and BE8 is a link-time choice, not a compile-time choice. So if the value for Arm is a compile-time constant, it should also work for both BE32 and BE8.
Good to know! Then we need to be even more careful.
Quoted 8 lines
In turn, that suggests to me that
RSEQ_SIGshould be defined to be a value that is always in the code endianness (and whatever corresponding kernel code handlesRSEQ_SIGvalues should act accordingly on architectures where the two endiannesses can differ). If the kernel ABI is already fixed in a way that prevents such a definition ofRSEQ_SIGsemantics as using code endianness, a value should be chosen for Arm that works for both endiannesses.
It might be tricky to pick up a trap instruction that is a palindrome endianness-wise.
Quoted 5 lines
(Also, installed glibc headers are supposed to work with older compilers, and support for
__ARM_ARCHwas only added in GCC 4.8. Before that you need to test lots of separate macros for different architecture variants to determine a version number.)
Good point!
Here is an alternative to the palindrome approach. I'm taking arm32 as an example:
* We define RSEQ_SIG_CODE in code endianness, meant to be used with .inst in rseq assembly:
#define RSEQ_SIG_CODE 0xe7f5def3
* We define RSEQ_SIG_DATA in data endianness:
#define RSEQ_SIG_DATA \
({ \
int sig; \
asm volatile ( "b 2f\n\t" \
".arm\n\t" \
"1: .inst 0xe7f5def3\n\t" \
"2:\n\t" \
"ldr %[sig], 1b\n\t" \
: [sig] "=r" (sig)); \
sig; \
})Technically, only glibc and early-adopter libraries wishing to register rseq need to use RSEQ_SIG_DATA. The RSEQ_SIG_CODE needs to be used from inline assembly to create the signatures before each abort handler.
Thoughts ?
Thanks,
Mathieu
Signature · 3 lines
Mathieu Desnoyers EfficiOS Inc. http://www.efficios.com
Quoted 90 lines · ----- On Apr 17, 2019, at 3:56 PM, Mathieu Desnoyers…
Quoted 18 lines · ----- On Apr 17, 2019, at 12:17 PM, Joseph Myers [email protected] wrote:
Quoted 7 lines · Mathieu Desnoyers
Quoted 7 lines
sysdeps/unix/sysv/linux/aarch64/bits/rseq.h new file +7
-0+/* 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 Quoted 8 lines
sysdeps/unix/sysv/linux/aarch64/bits/rseq.h line 32
Quoted 3 lines
sysdeps/unix/sysv/linux/aarch64/bits/rseq.h new file +3
-0+execution efficiency in some cases. */ + +#define RSEQ_SIG 0xd428bc00/* BRK #0x45E0. */ After further investigation, we should probably do the following to handle compiling with -mbig-endian on aarch64, which generates binaries with mixed code vs data endianness (little endian code, big endian data):
Quoted 2 lines
First, the comment on
RSEQ_SIGshould specify whether it is to be interpreted in the code or the data endianness.Right. The signature passed as argument to the rseq registration system call needs to be in data endianness (currently exposed kernel ABI).
Ideally for userspace, we want to define a signature in code endianness that happens to nicely match specific code patterns.
Quoted 8 lines
Quoted 8 lines
For ARM32, the situation is a bit more complex. Only armv6+ generates mixed-endianness code vs data with -mbig-endian. Prior to armv6, the code and data endianness matches. Therefore, I plan to #ifdef the reversed endianness handling with:
#if
__ARM_ARCH>= 6 &&__ARM_BIG_ENDIANQuoted 7 lines
Quoted 1 line
on arm32.
That doesn't work well because BE code (.o files) can be built for v5te (for example) and used on a range of different architecture variants with both BE32 and BE8 - the choice between BE32 and BE8 is a link-time choice, not a compile-time choice. So if the value for Arm is a compile-time constant, it should also work for both BE32 and BE8.
Good to know! Then we need to be even more careful.
Quoted 8 lines
In turn, that suggests to me that
RSEQ_SIGshould be defined to be a value that is always in the code endianness (and whatever corresponding kernel code handlesRSEQ_SIGvalues should act accordingly on architectures where the two endiannesses can differ). If the kernel ABI is already fixed in a way that prevents such a definition ofRSEQ_SIGsemantics as using code endianness, a value should be chosen for Arm that works for both endiannesses.It might be tricky to pick up a trap instruction that is a palindrome endianness-wise.
Quoted 5 lines
(Also, installed glibc headers are supposed to work with older compilers, and support for
__ARM_ARCHwas only added in GCC 4.8. Before that you need to test lots of separate macros for different architecture variants to determine a version number.)Good point!
Here is an alternative to the palindrome approach. I'm taking arm32 as an example:
* We define
RSEQ_SIG_CODEin code endianness, meant to be used with .inst in rseq assembly:#define
RSEQ_SIG_CODE0xe7f5def3* We define
RSEQ_SIG_DATAin data endianness:#define RSEQ_SIG_DATA \ ({ \ int sig; \ asm volatile ( "b 2f\n\t" \ ".arm\n\t" \ "1: .inst 0xe7f5def3\n\t" \ "2:\n\t" \ "ldr %[sig], 1b\n\t" \ : [sig] "=r" (sig)); \ sig; \ })
Quoted 4 lines
Technically, only glibc and early-adopter libraries wishing to register rseq need to use
RSEQ_SIG_DATA. TheRSEQ_SIG_CODEneeds to be used from inline assembly to create the signatures before each abort handler.
The approach above should work for arm32 be8 vs be32 linker weirdness.
For aarch64, I think we can simply do:
/*
* 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 __ARM_BIG_ENDIAN
#define RSEQ_SIG_DATA 0x00bc28d4 /* BRK #0x45E0. */
#else
#define RSEQ_SIG_DATA RSEQ_SIG_CODE
#endif
#define RSEQ_SIG RSEQ_SIG_DATA
Feedback is most welcome,
Thanks!
Mathieu
Signature · 3 lines
Mathieu Desnoyers EfficiOS Inc. http://www.efficios.com
Quoted 17 lines · Mathieu Desnoyers
The approach above should work for arm32 be8 vs be32 linker weirdness.
For aarch64, I think we can simply do:
/*
* aarch64 -mbig-endian generates mixed endianness code vs data:
* little-endian code and big-endian data. Ensure theRSEQ_SIGsignature
* matches code endianness.
*/
#defineRSEQ_SIG_CODE0xd428bc00/* BRK #0x45E0. */#ifdef
__ARM_BIG_ENDIAN
#defineRSEQ_SIG_DATA0x00bc28d4/* BRK #0x45E0. */
#else
#defineRSEQ_SIG_DATARSEQ_SIG_CODE
#endif
Quoted 3 lines
#define
RSEQ_SIGRSEQ_SIG_DATAFeedback is most welcome,
You'll also need __ASSEMBLER__ conditionals in the installed sys/rseq.h header so that it only defines constants and doesn't include any C declarations in that case, if RSEQ_SIG_CODE is meant to be usable in .S files rather than just inline asm in C files.
Signature · 2 lines
Joseph S. Myers [email protected]
Quoted 19 lines · ----- On Apr 18, 2019, at 10:48 AM, Joseph Myers [email protected] wrote:
Quoted 17 lines · Mathieu Desnoyers
The approach above should work for arm32 be8 vs be32 linker weirdness.
For aarch64, I think we can simply do:
/*
* aarch64 -mbig-endian generates mixed endianness code vs data:
* little-endian code and big-endian data. Ensure theRSEQ_SIGsignature
* matches code endianness.
*/
#defineRSEQ_SIG_CODE0xd428bc00/* BRK #0x45E0. */#ifdef
__ARM_BIG_ENDIAN
#defineRSEQ_SIG_DATA0x00bc28d4/* BRK #0x45E0. */
#else
#defineRSEQ_SIG_DATARSEQ_SIG_CODE
#endif
Quoted 8 lines
Quoted 3 lines
#define
RSEQ_SIGRSEQ_SIG_DATAFeedback is most welcome,
You'll also need
__ASSEMBLER__conditionals in the installedsys/rseq.hheader so that it only defines constants and doesn't include any C declarations in that case, ifRSEQ_SIG_CODEis meant to be usable in .S files rather than just inline asm in C files.
Good point!
Thanks,
Mathieu
Signature · 3 lines
Mathieu Desnoyers EfficiOS Inc. http://www.efficios.com
Quoted 29 lines · Mathieu Desnoyers
Quoted 28 lines · ----- On Apr 17, 2019, at 3:56 PM, Mathieu Desnoyers…
Quoted 18 lines · ----- On Apr 17, 2019, at 12:17 PM, Joseph Myers [email protected] wrote:
Quoted 7 lines · Mathieu Desnoyers
Quoted 7 lines
sysdeps/unix/sysv/linux/aarch64/bits/rseq.h new file +7
-0+/* 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 Quoted 8 lines
sysdeps/unix/sysv/linux/aarch64/bits/rseq.h line 32
Quoted 3 lines
sysdeps/unix/sysv/linux/aarch64/bits/rseq.h new file +3
-0+execution efficiency in some cases. */ + +#define RSEQ_SIG 0xd428bc00/* BRK #0x45E0. */ After further investigation, we should probably do the following to handle compiling with -mbig-endian on aarch64, which generates binaries with mixed code vs data endianness (little endian code, big endian data):
Quoted 2 lines
First, the comment on
RSEQ_SIGshould specify whether it is to be interpreted in the code or the data endianness.Right. The signature passed as argument to the rseq registration system call needs to be in data endianness (currently exposed kernel ABI).
Ideally for userspace, we want to define a signature in code endianness that happens to nicely match specific code patterns.
...
Quoted 15 lines
For aarch64, I think we can simply do:
/*
* aarch64 -mbig-endian generates mixed endianness code vs data:
* little-endian code and big-endian data. Ensure theRSEQ_SIGsignature
* matches code endianness.
*/
#defineRSEQ_SIG_CODE0xd428bc00/* BRK #0x45E0. */#ifdef
__ARM_BIG_ENDIAN
#defineRSEQ_SIG_DATA0x00bc28d4/* BRK #0x45E0. */
#else
#defineRSEQ_SIG_DATARSEQ_SIG_CODE
#endif
Quoted 3 lines
#define
RSEQ_SIGRSEQ_SIG_DATAFeedback is most welcome,
so the RSEQ_SIG value is supposed to be used with .word in asm instead of .inst?
i don't think we use __ARM_* in public headers currently, but hopefully aarch64_be compilers implement it.
otherwise this looks ok to me.
(i think a rare palindrome instruction would work too, e.g.0a5f5f0a and w10, w24, wzr, lsr #23 // shifted 02a5f5f2a orr w10, w25, wzr, lsr #23eb9f9feb negs x11, xzr, asr #39c83f3fc8 stxp wzr, x8, x15, [x30] // store to LR ignoring successd9ffffd9 stz2g x25, [x30, #-16]! // v8.5 tag+zero 2 granules around LR
etc. it does not need to be a guaranteed trap)
Quoted 47 lines · ----- On Apr 18, 2019, at 11:33 AM, Szabolcs Nagy [email protected] wrote:
Quoted 30 lines · Mathieu Desnoyers
----- On Apr 17, 2019, at 3:56 PM, Mathieu Desnoyers
Quoted 28 lines · [email protected] wrote:
Quoted 18 lines · ----- On Apr 17, 2019, at 12:17 PM, Joseph Myers [email protected] wrote:
Quoted 7 lines · Mathieu Desnoyers
Quoted 7 lines
sysdeps/unix/sysv/linux/aarch64/bits/rseq.h new file +7
-0+/* 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 Quoted 8 lines
sysdeps/unix/sysv/linux/aarch64/bits/rseq.h line 32
Quoted 3 lines
sysdeps/unix/sysv/linux/aarch64/bits/rseq.h new file +3
-0+execution efficiency in some cases. */ + +#define RSEQ_SIG 0xd428bc00/* BRK #0x45E0. */ After further investigation, we should probably do the following to handle compiling with -mbig-endian on aarch64, which generates binaries with mixed code vs data endianness (little endian code, big endian data):
Quoted 2 lines
First, the comment on
RSEQ_SIGshould specify whether it is to be interpreted in the code or the data endianness.Right. The signature passed as argument to the rseq registration system call needs to be in data endianness (currently exposed kernel ABI).
Ideally for userspace, we want to define a signature in code endianness that happens to nicely match specific code patterns.
...
Quoted 15 lines
For aarch64, I think we can simply do:
/*
* aarch64 -mbig-endian generates mixed endianness code vs data:
* little-endian code and big-endian data. Ensure theRSEQ_SIGsignature
* matches code endianness.
*/
#defineRSEQ_SIG_CODE0xd428bc00/* BRK #0x45E0. */#ifdef
__ARM_BIG_ENDIAN
#defineRSEQ_SIG_DATA0x00bc28d4/* BRK #0x45E0. */
#else
#defineRSEQ_SIG_DATARSEQ_SIG_CODE
#endif
Quoted 6 lines
Quoted 3 lines
#define
RSEQ_SIGRSEQ_SIG_DATAFeedback is most welcome,
so the
RSEQ_SIGvalue is supposed to be used with .word in asm instead of .inst?
We want a .inst so it translates into a valid trap instruction. It's better to trap in case program execution reaches this by mistake (makes debugging easier).
Quoted 3 lines
i don't think we use __ARM_* in public headers currently, but hopefully
aarch64_becompilers implement it.
Can I use #if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ then ?
Quoted 10 lines
otherwise this looks ok to me.
(i think a rare palindrome instruction would work too, e.g.
0a5f5f0aand w10, w24, wzr, lsr #23 // shifted 02a5f5f2aorr w10, w25, wzr, lsr #23eb9f9febnegs x11, xzr, asr #39c83f3fc8stxp wzr, x8, x15, [x30] // store to LR ignoring successd9ffffd9stz2g x25, [x30, #-16]! // v8.5 tag+zero 2 granules around LR
etc. it does not need to be a guaranteed trap)
Unfortunately it's not a trap :/
Thanks,
Mathieu
Signature · 3 lines
Mathieu Desnoyers EfficiOS Inc. http://www.efficios.com
Quoted 51 lines · Mathieu Desnoyers
Quoted 49 lines · ----- On Apr 18, 2019, at 11:33 AM, Szabolcs Nagy [email protected] wrote:
Quoted 30 lines · Mathieu Desnoyers
----- On Apr 17, 2019, at 3:56 PM, Mathieu Desnoyers
Quoted 28 lines · [email protected] wrote:
Quoted 18 lines · ----- On Apr 17, 2019, at 12:17 PM, Joseph Myers [email protected] wrote:
Quoted 7 lines · Mathieu Desnoyers
Quoted 7 lines
(quoted diff) +7
-0+/* 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 speculativeQuoted 8 lines
Review
Quoted 3 lines
(quoted diff) +3
-0+ execution efficiency in some cases. */ + +#define RSEQ_SIG 0xd428bc00 /* BRK #0x45E0. */After further investigation, we should probably do the following to handle compiling with -mbig-endian on aarch64, which generates binaries with mixed code vs data endianness (little endian code, big endian data):
Quoted 2 lines
First, the comment on
RSEQ_SIGshould specify whether it is to be interpreted in the code or the data endianness.Right. The signature passed as argument to the rseq registration system call needs to be in data endianness (currently exposed kernel ABI).
Ideally for userspace, we want to define a signature in code endianness that happens to nicely match specific code patterns.
...
Quoted 17 lines
For aarch64, I think we can simply do:
/*
* aarch64 -mbig-endian generates mixed endianness code vs data:
* little-endian code and big-endian data. Ensure theRSEQ_SIGsignature
* matches code endianness.
*/
#defineRSEQ_SIG_CODE0xd428bc00/* BRK #0x45E0. */#ifdef
__ARM_BIG_ENDIAN
#defineRSEQ_SIG_DATA0x00bc28d4/* BRK #0x45E0. */
#else
#defineRSEQ_SIG_DATARSEQ_SIG_CODE
#endif#define
RSEQ_SIGRSEQ_SIG_DATA
Quoted 8 lines
Quoted 4 lines
Quoted 1 line
Feedback is most welcome,
so the
RSEQ_SIGvalue is supposed to be used with .word in asm instead of .inst?We want a .inst so it translates into a valid trap instruction. It's better to trap in case program execution reaches this by mistake (makes debugging easier).
that does not make sense to me.
".inst" is an asm directive that requires a the value to be the same on BE and LE (normal insn encoding).
".word" is an asm directive that requires the value to use swapped encoding on BE (if it's used in the instruction stream it will create a data mapping symbol and disasm to .word value instead of the instruction mnemonics).
so which one is it?
Quoted 4 lines
Quoted 2 lines
i don't think we use __ARM_* in public headers currently, but hopefully
aarch64_becompilers implement it.Can I use #if
__BYTE_ORDER__==__ORDER_BIG_ENDIAN__then ?
hm, i'd either use #ifdef __AARCH64EB__ (since we already use it) or the portable #include endian.h and __BYTE_ORDER == __BIG_ENDIAN
Quoted 66 lines · ----- On Apr 18, 2019, at 12:07 PM, Szabolcs Nagy [email protected] wrote:
Quoted 51 lines · Mathieu Desnoyers
Quoted 49 lines · ----- On Apr 18, 2019, at 11:33 AM, Szabolcs Nagy [email protected] wrote:
Quoted 30 lines · Mathieu Desnoyers
----- On Apr 17, 2019, at 3:56 PM, Mathieu Desnoyers
Quoted 28 lines · [email protected] wrote:
Quoted 18 lines · ----- On Apr 17, 2019, at 12:17 PM, Joseph Myers [email protected] wrote:
Quoted 7 lines · Mathieu Desnoyers
Quoted 7 lines
(quoted diff) +7
-0+/* 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 speculativeQuoted 8 lines
Review
Quoted 3 lines
(quoted diff) +3
-0+ execution efficiency in some cases. */ + +#define RSEQ_SIG 0xd428bc00 /* BRK #0x45E0. */After further investigation, we should probably do the following to handle compiling with -mbig-endian on aarch64, which generates binaries with mixed code vs data endianness (little endian code, big endian data):
Quoted 2 lines
First, the comment on
RSEQ_SIGshould specify whether it is to be interpreted in the code or the data endianness.Right. The signature passed as argument to the rseq registration system call needs to be in data endianness (currently exposed kernel ABI).
Ideally for userspace, we want to define a signature in code endianness that happens to nicely match specific code patterns.
...
Quoted 17 lines
For aarch64, I think we can simply do:
/*
* aarch64 -mbig-endian generates mixed endianness code vs data:
* little-endian code and big-endian data. Ensure theRSEQ_SIGsignature
* matches code endianness.
*/
#defineRSEQ_SIG_CODE0xd428bc00/* BRK #0x45E0. */#ifdef
__ARM_BIG_ENDIAN
#defineRSEQ_SIG_DATA0x00bc28d4/* BRK #0x45E0. */
#else
#defineRSEQ_SIG_DATARSEQ_SIG_CODE
#endif#define
RSEQ_SIGRSEQ_SIG_DATAQuoted 8 lines
Quoted 4 lines
Quoted 1 line
Feedback is most welcome,
so the
RSEQ_SIGvalue is supposed to be used with .word in asm instead of .inst?We want a .inst so it translates into a valid trap instruction. It's better to trap in case program execution reaches this by mistake (makes debugging easier).
that does not make sense to me.
".inst" is an asm directive that requires a the value to be the same on BE and LE (normal insn encoding).
Quoted 6 lines
".word" is an asm directive that requires the value to use swapped encoding on BE (if it's used in the instruction stream it will create a data mapping symbol and disasm to .word value instead of the instruction mnemonics).
so which one is it?
We declare the signature with ".inst" in assembler.
However, we also need to pass that 32-bit signature value as argument to the rseq system call when registering rseq.
The signature comparison is performed by the kernel before moving the instruction pointer to the abort handler. It compares the signature received as parameter by sys_rseq (data) to the 4-byte signature preceding the abort IP.
On aarch64 big endian, AFAIU the signature in the code is in little endian, and the signature value passed as argument to the rseq system call is in big endian. One way to handle this is to swap the byte order of the signature "data" representation passed as argument to sys_rseq.
Quoted 8 lines
Quoted 5 lines
Quoted 3 lines
i don't think we use __ARM_* in public headers currently, but hopefully
aarch64_becompilers implement it.Can I use #if
__BYTE_ORDER__==__ORDER_BIG_ENDIAN__then ?hm, i'd either use #ifdef
__AARCH64EB__(since we already use it) or the portable #includeendian.hand__BYTE_ORDER==__BIG_ENDIAN
I'll use #ifdef __AARCH64EB__ given this header is specific to aarch64.
Thanks,
Mathieu
Signature · 3 lines
Mathieu Desnoyers EfficiOS Inc. http://www.efficios.com
Quoted 86 lines · Mathieu Desnoyers
Quoted 66 lines · ----- On Apr 18, 2019, at 12:07 PM, Szabolcs Nagy [email protected] wrote:
Quoted 51 lines · Mathieu Desnoyers
Quoted 49 lines · ----- On Apr 18, 2019, at 11:33 AM, Szabolcs Nagy [email protected] wrote:
Quoted 30 lines · Mathieu Desnoyers
----- On Apr 17, 2019, at 3:56 PM, Mathieu Desnoyers
Quoted 28 lines · [email protected] wrote:
Quoted 18 lines · ----- On Apr 17, 2019, at 12:17 PM, Joseph Myers [email protected] wrote:
Quoted 7 lines · Mathieu Desnoyers
Quoted 7 lines
(quoted diff) +7
-0+/* 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 speculativeQuoted 8 lines
Review
Quoted 3 lines
(quoted diff) +3
-0+ execution efficiency in some cases. */ + +#define RSEQ_SIG 0xd428bc00 /* BRK #0x45E0. */After further investigation, we should probably do the following to handle compiling with -mbig-endian on aarch64, which generates binaries with mixed code vs data endianness (little endian code, big endian data):
Quoted 2 lines
First, the comment on
RSEQ_SIGshould specify whether it is to be interpreted in the code or the data endianness.Right. The signature passed as argument to the rseq registration system call needs to be in data endianness (currently exposed kernel ABI).
Ideally for userspace, we want to define a signature in code endianness that happens to nicely match specific code patterns.
...
Quoted 17 lines
For aarch64, I think we can simply do:
/*
* aarch64 -mbig-endian generates mixed endianness code vs data:
* little-endian code and big-endian data. Ensure theRSEQ_SIGsignature
* matches code endianness.
*/
#defineRSEQ_SIG_CODE0xd428bc00/* BRK #0x45E0. */#ifdef
__ARM_BIG_ENDIAN
#defineRSEQ_SIG_DATA0x00bc28d4/* BRK #0x45E0. */
#else
#defineRSEQ_SIG_DATARSEQ_SIG_CODE
#endif#define
RSEQ_SIGRSEQ_SIG_DATAQuoted 8 lines
Quoted 4 lines
Quoted 1 line
Feedback is most welcome,
so the
RSEQ_SIGvalue is supposed to be used with .word in asm instead of .inst?We want a .inst so it translates into a valid trap instruction. It's better to trap in case program execution reaches this by mistake (makes debugging easier).
that does not make sense to me.
".inst" is an asm directive that requires a the value to be the same on BE and LE (normal insn encoding).
Quoted 6 lines
".word" is an asm directive that requires the value to use swapped encoding on BE (if it's used in the instruction stream it will create a data mapping symbol and disasm to .word value instead of the instruction mnemonics).
so which one is it?
We declare the signature with ".inst" in assembler.
However, we also need to pass that 32-bit signature value as argument to the rseq system call when registering rseq.
The signature comparison is performed by the kernel before moving the instruction pointer to the abort handler. It compares the signature received as parameter by
sys_rseq(data) to the 4-byte signature preceding the abort IP.
Quoted 5 lines
On aarch64 big endian, AFAIU the signature in the code is in little endian, and the signature value passed as argument to the rseq system call is in big endian. One way to handle this is to swap the byte order of the signature "data" representation passed as argument to
sys_rseq.
you have to add a documentation comment somewhere explaining if RSEQ_SIG is the value that's passed to the kernel and then aarch64 asm code has to use
.inst endianfixup(RSEQ_SIG) // or
.word RSEQ_SIG
or if RSEQ_SIG is used as
.inst RSEQ_SIG
in aarch64 asm and then endianfixup(RSEQ_SIG) should be passed to the syscall.
either way it can be a brk 0x45e0 on both LE and BE, but in the latter case you have to document this in arch independent way, since the syscall api must be portable (i assume "RSEQ_SIG" is part of the api).
Quoted 93 lines · ----- On Apr 18, 2019, at 1:37 PM, Szabolcs Nagy [email protected] wrote:
Quoted 86 lines · Mathieu Desnoyers
Quoted 66 lines · ----- On Apr 18, 2019, at 12:07 PM, Szabolcs Nagy [email protected] wrote:
Quoted 51 lines · Mathieu Desnoyers
Quoted 49 lines · ----- On Apr 18, 2019, at 11:33 AM, Szabolcs Nagy [email protected] wrote:
Quoted 30 lines · Mathieu Desnoyers
----- On Apr 17, 2019, at 3:56 PM, Mathieu Desnoyers
Quoted 28 lines · [email protected] wrote:
Quoted 18 lines · ----- On Apr 17, 2019, at 12:17 PM, Joseph Myers [email protected] wrote:
Quoted 7 lines · Mathieu Desnoyers
Quoted 7 lines
(quoted diff) +7
-0+/* 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 speculativeQuoted 8 lines
Review
Quoted 3 lines
(quoted diff) +3
-0+ execution efficiency in some cases. */ + +#define RSEQ_SIG 0xd428bc00 /* BRK #0x45E0. */After further investigation, we should probably do the following to handle compiling with -mbig-endian on aarch64, which generates binaries with mixed code vs data endianness (little endian code, big endian data):
Quoted 2 lines
First, the comment on
RSEQ_SIGshould specify whether it is to be interpreted in the code or the data endianness.Right. The signature passed as argument to the rseq registration system call needs to be in data endianness (currently exposed kernel ABI).
Ideally for userspace, we want to define a signature in code endianness that happens to nicely match specific code patterns.
...
Quoted 17 lines
For aarch64, I think we can simply do:
/*
* aarch64 -mbig-endian generates mixed endianness code vs data:
* little-endian code and big-endian data. Ensure theRSEQ_SIGsignature
* matches code endianness.
*/
#defineRSEQ_SIG_CODE0xd428bc00/* BRK #0x45E0. */#ifdef
__ARM_BIG_ENDIAN
#defineRSEQ_SIG_DATA0x00bc28d4/* BRK #0x45E0. */
#else
#defineRSEQ_SIG_DATARSEQ_SIG_CODE
#endif#define
RSEQ_SIGRSEQ_SIG_DATAQuoted 8 lines
Quoted 4 lines
Quoted 1 line
Feedback is most welcome,
so the
RSEQ_SIGvalue is supposed to be used with .word in asm instead of .inst?We want a .inst so it translates into a valid trap instruction. It's better to trap in case program execution reaches this by mistake (makes debugging easier).
that does not make sense to me.
".inst" is an asm directive that requires a the value to be the same on BE and LE (normal insn encoding).
Quoted 6 lines
".word" is an asm directive that requires the value to use swapped encoding on BE (if it's used in the instruction stream it will create a data mapping symbol and disasm to .word value instead of the instruction mnemonics).
so which one is it?
We declare the signature with ".inst" in assembler.
However, we also need to pass that 32-bit signature value as argument to the rseq system call when registering rseq.
The signature comparison is performed by the kernel before moving the instruction pointer to the abort handler. It compares the signature received as parameter by
sys_rseq(data) to the 4-byte signature preceding the abort IP.Quoted 5 lines
On aarch64 big endian, AFAIU the signature in the code is in little endian, and the signature value passed as argument to the rseq system call is in big endian. One way to handle this is to swap the byte order of the signature "data" representation passed as argument to
sys_rseq.
Quoted 6 lines
you have to add a documentation comment somewhere explaining if
RSEQ_SIGis the value that's passed to the kernel and then aarch64 asm code has to use.inst
endianfixup(RSEQ_SIG)// or
.wordRSEQ_SIG
Using ".word" won't allow objdump to show the instruction it maps to. It will consider it as data. So .inst is preferred here.
Quoted 7 lines
or if
RSEQ_SIGis used as.inst
RSEQ_SIGin aarch64 asm and then
endianfixup(RSEQ_SIG)should be passed to the syscall.
At this stage, we control the meaning of the definitions we publicly expose. They are part of glibc headers, not part of the kernel uapi.
On architectures where data and code endianness match, RSEQ_SIG can be used both as argument to sys_rseq and as value for .inst in assembler.
On architectures where data and code endianness differ, I am tempted to declare them separately:
* RSEQ_SIG_CODE: for use with .inst in assembly,
* RSEQ_SIG_DATA (mapping to RSEQ_SIG): to pass as parameter to sys_rseq.
So those specific architectures would use "RSEQ_SIG_CODE" with .inst in assembly, and we can still pass the RSEQ_SIG as parameter to sys_rseq in generic rseq registration code.
Quoted 4 lines
either way it can be a brk
0x45e0on both LE and BE, but in the latter case you have to document this in arch independent way, since the syscall api must be portable (i assume "RSEQ_SIG" is part of the api).
The RSEQ_SIG is defined by glibc bits/rseq.h which is included from sys/rseq.h. It's therefore not part of the Linux kernel uapi. So we can define whatever we need to at this point, but we won't be able to change it after it has been exposed for a given architecture.
All the kernel ABI expects is a data-endian value of the signature it needs to compare to when it loads the 4 bytes prior to the abort ip.
Thanks,
Mathieu
Signature · 3 lines
Mathieu Desnoyers EfficiOS Inc. http://www.efficios.com
Quoted 10 lines · Mathieu Desnoyers
Quoted 6 lines · ----- On Apr 18, 2019, at 1:37 PM, Szabolcs Nagy [email protected] wrote:
you have to add a documentation comment somewhere explaining if
RSEQ_SIGis the value that's passed to the kernel and then aarch64 asm code has to use.inst
endianfixup(RSEQ_SIG)// or
.wordRSEQ_SIGUsing ".word" won't allow objdump to show the instruction it maps to. It will consider it as data. So .inst is preferred here.
is there some specific reason you prefer .inst?
disassembling a canary value as data (that is never executed, but loaded and compared by the kernel as data) sounds more semantically correct to me than showing it as an instruction.
i guess having it as an instruction can avoid issues if some tools dislike .word in .text, but otherwise .word seems better.
Quoted 7 lines · Szabolcs Nagy
Quoted 5 lines · Mathieu Desnoyers
Quoted 4 lines · ----- On Apr 18, 2019, at 1:37 PM, Szabolcs Nagy [email protected] wrote:
you have to add a documentation comment somewhere explaining if
RSEQ_SIGis the value that's passed to the kernel and then aarch64 asm code has to use
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines
.inst
endianfixup(RSEQ_SIG)// or
.wordRSEQ_SIGUsing ".word" won't allow objdump to show the instruction it maps to. It will consider it as data. So .inst is preferred here.
is there some specific reason you prefer .inst?
I believe the reasoning here is that in the disassembly you want to see an instruction pattern for an architecture rather than a magic bit pattern that appears to be an "inline" literal pool entry. I would support the .inst variant so that the disassembler shows the instruction for what it is when debugging. If control reaches the marker instruction, something's gone wrong and thus from a user friendliness perspective I would prefer to see an instruction that clearly indicates that it's undefined .inst directive so that someone disassembling this in an assembly view in GDB sees the right thing (TM) instead of having to reach for the manual and disassembling this by hand.
Quoted 6 lines
disassembling a canary value as data (that is never executed, but loaded and compared by the kernel as data) sounds more semantically correct to me than showing it as an instruction.
Ramana
Quoted 3 lines
i guess having it as an instruction can avoid issues if some tools dislike .word in .text, but otherwise .word seems better.
Quoted 16 lines · ----- On Apr 23, 2019, at 7:59 AM, Ramana Radhakrishnan…
Quoted 7 lines · Szabolcs Nagy
Quoted 5 lines · Mathieu Desnoyers
Quoted 4 lines · ----- On Apr 18, 2019, at 1:37 PM, Szabolcs Nagy [email protected] wrote:
you have to add a documentation comment somewhere explaining if
RSEQ_SIGis the value that's passed to the kernel and then aarch64 asm code has to useQuoted 7 lines
Quoted 5 lines
Quoted 2 lines
.inst
endianfixup(RSEQ_SIG)// or
.wordRSEQ_SIGUsing ".word" won't allow objdump to show the instruction it maps to. It will consider it as data. So .inst is preferred here.
is there some specific reason you prefer .inst?
Quoted 11 lines
I believe the reasoning here is that in the disassembly you want to see an instruction pattern for an architecture rather than a magic bit pattern that appears to be an "inline" literal pool entry. I would support the .inst variant so that the disassembler shows the instruction for what it is when debugging. If control reaches the marker instruction, something's gone wrong and thus from a user friendliness perspective I would prefer to see an instruction that clearly indicates that it's undefined .inst directive so that someone disassembling this in an assembly view in GDB sees the right thing (TM) instead of having to reach for the manual and disassembling this by hand.
That's my line of thinking exactly. I might add that having data in a literal pool within the instruction stream might be unexpected in some compilation environments, e.g. when compiling with -mno-text-section-literals .
So even though the signature may likely end up being placed in a literal pool, it's preferable to ensure it is a valid uncommon trap instruction.
Thanks,
Mathieu
Quoted 16 lines
Quoted 7 lines
disassembling a canary value as data (that is never executed, but loaded and compared by the kernel as data) sounds more semantically correct to me than showing it as an instruction.
Ramana
Quoted 3 lines
i guess having it as an instruction can avoid issues if some tools dislike .word in .text, but otherwise .word seems better.
Signature · 3 lines
Mathieu Desnoyers EfficiOS Inc. http://www.efficios.com
PATCH 2/5 glibc: sched_getcpu(): use rseq cpu_id TLS on Linux (v2)
In reply to a message not in the archiveWhen 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.
--- sysdeps/unix/sysv/linux/sched_getcpu.c | 27 ++++++++++++++++++++++++-- 1 file changed, 25 insertions(+), 2 deletions(-)
sysdeps/unix/sysv/linux/sched_getcpu.c +25 -2
@@ -24,8 +24,8 @@#endif #include <sysdep-vdso.h> -int -sched_getcpu (void ) +static int +vsyscall_sched_getcpu (void ) {#ifdef __NR_getcpuunsigned int cpu;
@@ -37,3 +37,26 @@ sched_getcpu (void)return -1 ;#endif } + +#ifdef __NR_rseq +#include <sys/rseq.h> +#endif + +#if defined __NR_rseq && definedRSEQ_SIG +extern __attribute__ ((tls_model ("initial-exec" ))) +__threadvolatile struct rseq __rseq_abi; + +int +sched_getcpu (void ) +{ +int cpu_id = __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
Re: [PATCH 2/5] glibc: sched_getcpu(): use rseq cpu_id TLS on Linux (v2)
Quoted 4 lines · Mathieu Desnoyers
sysdeps/unix/sysv/linux/sched_getcpu.c +0
-0@@ -37,3 +37,26 @@ sched_getcpu (void)return -1 ;
sysdeps/unix/sysv/linux/sched_getcpu.c sched_getcpu (void) · line 47
Quoted 10 lines
sysdeps/unix/sysv/linux/sched_getcpu.c +8
-0#endif } + +#ifdef __NR_rseq +#include <sys/rseq.h> +#endif + +#if defined __NR_rseq && definedRSEQ_SIG +extern __attribute__ ((tls_model ("initial-exec" ))) +__threadvolatile struct rseq __rseq_abi;
i'd expect sys/rseq.h to provide this declaration.
Quoted 17 lines
sysdeps/unix/sysv/linux/sched_getcpu.c +15
-1+ +int +sched_getcpu (void ) +{ +int cpu_id = __rseq_abi.cpu_id; + +return cpu_id >=0 ? cpu_id :vsyscall_sched_getcpu (); +} +#else +int +sched_getcpu (void ) +{ +return vsyscall_sched_getcpu (); +} +#endif -- 2.17.1
Re: [PATCH 2/5] glibc: sched_getcpu(): use rseq cpu_id TLS on Linux (v2)
Quoted 17 lines · ----- On Apr 18, 2019, at 11:33 AM, Szabolcs Nagy [email protected] wrote:
Quoted 4 lines · Mathieu Desnoyers
sysdeps/unix/sysv/linux/sched_getcpu.c +0
-0@@ -37,3 +37,26 @@ sched_getcpu (void)return -1 ;sysdeps/unix/sysv/linux/sched_getcpu.c sched_getcpu (void) · line 47
Quoted 10 lines
sysdeps/unix/sysv/linux/sched_getcpu.c +8
-0#endif } + +#ifdef __NR_rseq +#include <sys/rseq.h> +#endif + +#if defined __NR_rseq && definedRSEQ_SIG +extern __attribute__ ((tls_model ("initial-exec" ))) +__threadvolatile struct rseq __rseq_abi;i'd expect
sys/rseq.hto provide this declaration.
And it actually does! Will remove this duplicate.
Thanks,
Mathieu
Quoted 17 lines
Quoted 17 lines
sysdeps/unix/sysv/linux/sched_getcpu.c +15
-1+ +int +sched_getcpu (void ) +{ +int cpu_id = __rseq_abi.cpu_id; + +return cpu_id >=0 ? cpu_id :vsyscall_sched_getcpu (); +} +#else +int +sched_getcpu (void ) +{ +return vsyscall_sched_getcpu (); +} +#endif -- 2.17.1
Signature · 3 lines
Mathieu Desnoyers EfficiOS Inc. http://www.efficios.com