Bug#679811: dh_shlibdeps crashes on s390x

[Niko Tyni]

merge 604128 679811
thanks

On Sun, Jul 01, 2012 at 07:27:23PM +0200, Soeren Sonnenburg wrote:

> When building shogun dh_shlibdeps crashes with 
> 
> dh_shlibdeps -pshogun-ruby-modular  -l
> :/usr/lib/atlas:/build/buildd-shogun_1.1.0-5-s390x-FQcWd3/shogun-1.1.0/src/shogun:/build/buildd-shogun_1.1.0-5-s390x-FQcWd3/shogun-1.1.0/debian/libshogun11/usr/lib  
> *** glibc detected *** /usr/bin/perl: free(): invalid pointer:
> 0x000000008013a578 ***
> ======= Backtrace: =========
> /lib/s390x-linux-gnu/libc.so.6(+0x86b90)[0x20000344b90]
> /lib/s390x-linux-gnu/libc.so.6(cfree+0x76)[0x2000034a69e]
> /usr/lib/libperl.so.5.14(Perl_vivify_ref+0x124)[0x2000011188c]
> /usr/lib/libperl.so.5.14(Perl_pp_helem+0x43a)[0x200001123b6]
> /usr/lib/libperl.so.5.14(Perl_runops_standard+0x20)[0x20000107094]
> /usr/lib/libperl.so.5.14(perl_run+0x3e8)[0x2000009ba98]
> /usr/bin/perl(main+0x1c0)[0x8000126c]
> /lib/s390x-linux-gnu/libc.so.6(__libc_start_main+0x120)[0x200002dea50]
> /usr/bin/perl[0x800012ba]

[...]

> dpkg-shlibdeps: error: objdump died from signal 6
> dh_shlibdeps: dpkg-shlibdeps -Tdebian/shogun-ruby-modular.substvars
> debian/shogun-ruby-modular/usr/lib/site_ruby/1.9.1/s390x-linux/modshogun.so
> returned exit code 255

> https://buildd.debian.org/status/fetch.php?pkg=shogun&arch=s390x&ver=1.1.0-5&stamp=1341102794

Looks like a duplicate of #604128 which ended up as unreproducible.

It would be good to know if a retry makes this happens again on the buildd.

Even better would be to have a reproducible test case. Building the
package manually on s390 and testing dh_shlibdeps and dpkg-shlibdeps
invocations would indeed go a long way towards that. Valgrind might help
once we've got something that always crashes.

Hm, AFAICT signal 6 is ABORT on s390* too, and the main dpkg-shlibdeps
perl process is clearly still alive when that signal gets delivered to
objdump as it's printing "error: objdump died from signal 6". I suppose
its memory could be corrupt already at that point though. In any case,
I'm afraid the backtrace doesn't help much.
-- 
Niko Tyni   ntyni at debian.org