[PATCH v4 0/6] Optimize posix_spawn signal setup with clone3

Carlos O'Donell carlos@redhat.com
Sun Jan 22 23:21:53 GMT 2023


On 1/19/23 09:39, Adhemerval Zanella Netto wrote:
> Just to confirm that it is for 2.38, since you replied as 2.28 on the patches :}.

Yes, 2.38, not 2.28 :-)

Thank you, and sorry.
 
> On 18/01/23 19:24, Carlos O'Donell wrote:
>> On 1/18/23 12:53, Adhemerval Zanella Netto wrote:
>>> Carlos, do we still want this for 2.37?
>>
>> I think we should just push this to 2.38.
>>
>> Start by committing everything we have for x86_64 and aarch64.
>>
>> Then repost the other arches for review.
>>  
>>> On 12/01/23 10:58, Adhemerval Zanella wrote:
>>>> Currently posix_spawn has to issue at least NSIG sigaction syscalls
>>>> to obtain current signal disposition and more if they there are not
>>>> either SIG_IGN or SIG_DFL.  The new clone3 CLONE_CLEAR_SIGHAND flags
>>>> introduced with Linux 5.5 resets all signal handlers of the child not
>>>> set to SIG_IGN to SIG_DFL, thus allowing posix_spawn to skip this
>>>> preparation phase.
>>>>
>>>> The exception is the signals defined by posix_spawnattr_setsigdefault
>>>> when POSIX_SPAWN_SETSIGDEF is set (since they can be SIG_IGN).  In
>>>> this case posix_spawn helper process needs to issue a sigction to
>>>> set the signal disposition to SIG_DFL.
>>>>
>>>> The patchset also adds clone3 implementation for aarch64.  The clone3
>>>> for x86_64 is change to not align the stack, since this is not
>>>> async-signal safe (although current calls does mask/unmask signals).
>>>>
>>>> * Changes from v3:
>>>>   - Changes commit message and comments wording.
>>>>   - Removed powerpc, s390x, riscv, arm, and mips clone3 optimizations.
>>>>
>>>> * Changes from v2:
>>>>   - Improve clone3 internal documentation.
>>>>
>>>> * Changes from v1:
>>>>   - Extend clone3 documentation.
>>>>   - Remove x86_64 stack alignment on clone3 child.
>>>>   - Fixed powerpc64 missing initial stack frame.
>>>>   - Fixed missing save/restore r7 for arm.
>>>>
>>>> Adhemerval Zanella (6):
>>>>   linux: Do not reset signal handler in posix_spawn if it is already
>>>>     SIG_DFL
>>>>   linux: Extend internal clone3 documentation
>>>>   Linux: Do not align the stack for __clone3
>>>>   linux: Add clone3 CLONE_CLEAR_SIGHAND optimization to posix_spawn
>>>>   aarch64: Add the clone3 wrapper
>>>>   Linux: optimize clone3 internal usage
>>>>
>>>>  include/clone_internal.h                      |  37 +++-
>>>>  posix/Makefile                                |   3 +-
>>>>  posix/tst-spawn7.c                            | 179 ++++++++++++++++++
>>>>  sysdeps/unix/sysv/linux/Makefile              |   3 +-
>>>>  sysdeps/unix/sysv/linux/aarch64/clone3.S      |  85 +++++++++
>>>>  sysdeps/unix/sysv/linux/aarch64/sysdep.h      |   2 +
>>>>  sysdeps/unix/sysv/linux/clone-internal.c      |  59 ++++--
>>>>  sysdeps/unix/sysv/linux/clone3.h              |  15 +-
>>>>  sysdeps/unix/sysv/linux/kernel-features.h     |   9 +
>>>>  sysdeps/unix/sysv/linux/spawni.c              |  33 +++-
>>>>  .../sysv/linux/tst-misalign-clone-internal.c  |  74 --------
>>>>  sysdeps/unix/sysv/linux/x86_64/clone3.S       |   3 -
>>>>  12 files changed, 384 insertions(+), 118 deletions(-)
>>>>  create mode 100644 posix/tst-spawn7.c
>>>>  create mode 100644 sysdeps/unix/sysv/linux/aarch64/clone3.S
>>>>  delete mode 100644 sysdeps/unix/sysv/linux/tst-misalign-clone-internal.c
>>>>
>>>
>>
> 

-- 
Cheers,
Carlos.



More information about the Libc-alpha mailing list