[Bug libc/21182] New: segfaults with glibc-2.25 on i686

["blog at pozimski dot eu" <sourceware-bugzilla at sourceware dot org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=21182

            Bug ID: 21182
           Summary: segfaults with glibc-2.25 on i686
           Product: glibc
           Version: 2.25
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: libc
          Assignee: unassigned at sourceware dot org
          Reporter: blog at pozimski dot eu
                CC: drepper.fsp at gmail dot com
  Target Milestone: ---

After Void Linux (http://www.voidlinux.eu/) updated their glibc package to
2.25, several users reported segmentation faults on the i686 platform. I
experienced them myself on a physical computer using an Intel Atom processor
(cpuinfo: https://owncloud.helmut-pozimski.de/index.php/s/3cOwICY2qNemETN ).
Specifically I saw segfaults while using ps, grep (on bigger files) or some
python relying software like salt-minion, in some instances gdb or kodi.

The segfaults seem to trace back to __memchr_sse2, this is a backtrace produced
via gdb while running grep on a 32M file:

https://owncloud.helmut-pozimski.de/index.php/s/03jB43CoEQtEXay

I tried to gather some more pieces of information together with one of the Void
developers, but was not able to produce a more detailed backtrace. The last
change in this function happened in git commit
https://sourceware.org/git/?p=glibc.git;a=commit;h=23d27709a423aec32821e9a5198a10267107bae2
, I tried to revert that one and recompile glibc. This seems to resolve the
issue at least one my machine.

It might also be relevant that no one managed to reproduce the issue inside a
VM yet, it seemingly is only reproducable on physical hardware and I don't know
if it affects all i686 era systems or only a certain range of Intel processors.

To reproduce the issue, run grep with any pattern on a larger file (in my tests
32K were not enough, several 100K produced a segfault and my /var/log/messages
which is 32M also did). Also just running "ps aux" on my system produced a
segfault as well as just starting either kodi or salt-minion. With gdb I do not
know under which circumstances it occurs, it wasn't really reliably
reproducable, just happened sometimes.

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