[lttng-dev] Segfault during process exit

[Jonathan Rajotte Julien]

Hi,

Don't forget to cc the mailing list.

On 2016-03-02 02:42 PM, Jeffrey Chen wrote:
>
> Even I share our executable, you will not be able to run on your 
> machine, as it depends on our environment.
>
This is expected, it's also why I asked you if you could provide a 
reproducer, a simpler version of your problem that reproduce the 
behavior even sporadically.

> The idea is that, if you spawn some threads, and these threads keep 
> writing traces through LTTng. Then, exit the main thread, as in return 
> main() function. LTTng's logic will do cleanup itself, but my other 
> threads are still writing to LTTng. This would cause segfault. It 
> would try to reference address 0x0.
>
This could be a good start for a simple reproducer.

Cheers
>
>
> #0 0x0000000000000000 in ?? ()
> #1  0x00007f7e31fdfd78 in __tracepoint_cb_processor___tracepoint_info (
>     taskNameArg=0x1cf3c30 "BackgroundTask", dataArg=0x1fd5200 "S")
>
> ...
>
> ------------------------------------------------------------------------
> *From:* Jonathan Rajotte Julien <Jonathan.rajotte-julien at efficios.com>
> *Sent:* Wednesday, March 2, 2016 11:21 AM
> *To:* Jeffrey Chen
> *Cc:* lttng-dev at lists.lttng.org
> *Subject:* Re: [lttng-dev] Segfault during process exit
> Hi,
>
> Could you provide a reproducer ?
>
> Could you give us some context ?
>
> Thanks
>
> On 2016-03-02 02:07 PM, Jeffrey Chen wrote:
>>
>> Hi:
>>
>>
>> Our system process will spawn several background threads to do tasks. 
>> During the process exit, there is no guarantee that these background 
>> threads are finished. I have noticed several times that the execution 
>> would segfault during exiting. The stacktrace is showing that the 
>> segfault happened at the call to write trace, tracepoint(). My guess 
>> is that LTTng cleaned up the tracing connection, and the background 
>> thread was still running. Has anyone noticed this issue? I thought 
>> about trying wait for all the background threads to complete, but 
>> this is very difficult to do in a complex system, as the background 
>> thread could spawn more threads during the process. Is there any way 
>> to solve this? Thanks.
>>
>>
>>
>> _______________________________________________
>> lttng-dev mailing list
>> lttng-dev at lists.lttng.org
>> http://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev
>
> -- 
> Jonathan R. Julien
> Efficios

-- 
Jonathan R. Julien
Efficios

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lttng.org/pipermail/lttng-dev/attachments/20160302/b1283489/attachment-0001.html>