[Bug libc/26994] New: Performance regression with raise()

[Bug libc/26994] New: Performance regression with raise()

[su-lifan at linux dot alibaba.com]

https://sourceware.org/bugzilla/show_bug.cgi?id=26994

            Bug ID: 26994
           Summary: Performance regression with raise()
           Product: glibc
           Version: 2.24
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: libc
          Assignee: unassigned at sourceware dot org
          Reporter: su-lifan at linux dot alibaba.com
                CC: drepper.fsp at gmail dot com
  Target Milestone: ---

With the following commit introduced in glibc-2.24:
```
commit 2ac88eecc57ff00e0b5ff803ebcc3465d2d640dd
Author: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Date:   Fri Apr 22 09:25:20 2016 -0300

    Refactor Linux raise implementation (BZ#15368)

    This patch changes both the nptl and libc Linux raise implementation
    to avoid the issues described in BZ#15368.  The strategy used is
    summarized in bug report first comment:

     1. Block all signals (including internal NPTL ones);
     2. Get pid and tid directly from syscall (not relying on cached
        values);
     3. Call tgkill;
     4. Restore old signal mask.

    Tested on x86_64 and i686.
```

This causes performance regression (around 20%) of will-it-scale/signal1_thread
with 1 worker. I examined the perf and confirmed the extra rt_sigprocmask
syscalls performed by raise().

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/26994] Performance regression with raise()

[adhemerval.zanella at linaro dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=26994

Adhemerval Zanella <adhemerval.zanella at linaro dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|---                         |NOTABUG
                 CC|                            |adhemerval.zanella at linaro dot o
                   |                            |rg

--- Comment #1 from Adhemerval Zanella <adhemerval.zanella at linaro dot org> ---
POSIX specifies raise to be async-signal-safe, and NPTL's raise() is
implemented as a call to pthread_kill(pthread_self(), sig) and this sequence is
is not async-signal-safe. If a signal handler forks during raise, the the
thread in the parent could receive the signal twice, once from itself and once
from the child.

The signal blocking is required to make obtaining the tid/pid atomically.

-- 
You are receiving this mail because:
You are on the CC list for the bug.