[MPlayer-advusers] Mplayer SEGV in module: ao2_init

[Stuart Foster]

On 08/01/11 18:21, Stuart Foster wrote:
> On 08/01/11 10:53, Clemens Ladisch wrote:
>> Reimar Döffinger wrote:
>>> On Thu, Jul 28, 2011 at 03:09:12PM +0100, Stuart Foster wrote:
>>>> With glibc 2.14 Mplayer is crashing in the ALSA set-up with a segv
>>>> (see attached) prior to the segv if run under valgrind a memory
>>>> error is reported in dl-close() (valgind report also in attachment).
>>>> I can repeat the problem both on my quad core phenom system and also
>>>> on a old IBM Thinkpad R51. Other applications linked against ALSA
>>>> report the valgrind problem but I am only experiencing crashes with
>>>> Mplayer. Please advise.
>>>
>>> Your ALSA setup continues to use a dlopen handle after it closed it
>>> again.
>>
>> The crash and the valgrind error are from two different places in
>> alsa-lib. In both cases, it looks as if the handles returned by dlopen
>> (or some internal dl* data structures) have been corrupted.
>>
>>> I don't see how this could be MPlayer's fault,
>>
>> If there were a bug in the dl* calls, it would be inside alsa-lib.
>> However, all these calls are straight dlopen-dlsym-dlcose sequences,
>> so I am at a loss to explain this with an ALSA bug.
>>
>> Stuart, is there a reason that you mention glibc 2.14?
>> Is this a regression from an earlier glibc?
>>
>>
>> Regards,
>> Clemens
>> _______________________________________________
>> MPlayer-advusers mailing list
>> MPlayer-advusers at mplayerhq.hu
>> https://lists.mplayerhq.hu/mailman/listinfo/mplayer-advusers
>
> The same binaries appeared fine at glibc-2.13 and earlier originally the
> only change was moving on to glibc-2.14. The glibc people have assured
> me this is an application issue, any way time will tell.
>
> Since the problem came up I have compiled all the components with
> different compiler options (glibc alsa-lib and mplayer) and on different
> hardware (all 32 bit LFS systems though) and for me the problem is
> consistent. I am currently working through all the various
> dlopen-dlsym-dlcose sequences to see if I can find the problematic piece
> of code.
> I currently think the symbol being processed at the time valgrind
> complains is: "snd_config_hook_load_for_all_cards".
> I know If I comment out the snd_dlclose on line 3327 of alsa-lib conf.c
> the problem goes away (obviously not the solution as I now have an
> in-balance in the dynamic symbol lookup processing).
>
> Thanks
>
> Stuart
> _______________________________________________
> MPlayer-advusers mailing list
> MPlayer-advusers at mplayerhq.hu
> https://lists.mplayerhq.hu/mailman/listinfo/mplayer-advusers
>

I have modified the dlopen call to include the RTLD_NODELETE flag in 
snd_config_hooks_call (line 3308 of alsa-lib conf.c).  On my system 
(with mplayer) this is the first call to dlopen in alsa-lib.

This has stopped valgrind complaining and applications are no longer 
crashing.

Now I have a scenario I understand I will try and reduce the problem to 
a piece of sample code for discussion with the glibc and alsa-lib people.

Thanks

Stuart