[Xenomai-help] No core files for segmentation faults in realtime tasks

[Xenomai-help] No core files for segmentation faults in realtime tasks

[Christoph Permes]

Hi,

I have a realtime application running in user space that sometimes terminates with a segmentation fault. 
I get a message segfault at 4c4f5200 ip b7cb98d8 sp b7ef7354 error 4 in libc-2.7.so[b7c5e000+138000], 
but there is no core file writen, although ulimit is set to unlimited.

I wrote a small test program that produces a segmentation fault and I get a core file when the segfault
happens before the realtime task starts but not when it happens in the task.

Is it somehow possible to enable writing core files also for realtime tasks?

Thanks, 
Christoph

Re: [Xenomai-help] No core files for segmentation faults in realtime tasks

[Philippe Gerum]

On Mon, 2009-08-24 at 10:07 +0200, Christoph Permes wrote:
> Hi,
> 
> I have a realtime application running in user space that sometimes terminates with a segmentation fault. 
> I get a message segfault at 4c4f5200 ip b7cb98d8 sp b7ef7354 error 4 in libc-2.7.so[b7c5e000+138000], 
> but there is no core file writen, although ulimit is set to unlimited.
> 
> I wrote a small test program that produces a segmentation fault and I get a core file when the segfault
> happens before the realtime task starts but not when it happens in the task.
> 
> Is it somehow possible to enable writing core files also for realtime tasks?

This feature is not disabled for real-time tasks actually, so there must
be something else at work.

You may want to send your small test case, and give more details about
your hardware and software configuration, at the very least the CPU
architecture your code runs on, the Xenomai, I-pipe patch, and kernel
release numbers.

> 
> Thanks, 
> Christoph
> 
> 
> 
> _______________________________________________
> Xenomai-help mailing list
> Xenomai-help@domain.hid
> https://mail.gna.org/listinfo/xenomai-help
-- 
Philippe.

Re: [Xenomai-help] No core files for segmentation faults in realtime tasks

[Gilles Chanteperdrix]

Christoph Permes wrote:
> Hi,
> 
> I have a realtime application running in user space that sometimes terminates with a segmentation fault. 
> I get a message segfault at 4c4f5200 ip b7cb98d8 sp b7ef7354 error 4 in libc-2.7.so[b7c5e000+138000], 
> but there is no core file writen, although ulimit is set to unlimited.
> 
> I wrote a small test program that produces a segmentation fault and I get a core file when the segfault
> happens before the realtime task starts but not when it happens in the task.
> 
> Is it somehow possible to enable writing core files also for realtime tasks?

In addition to what Philippe asked, why do you not try to run your
application directly in gdb? Or with gdbserver? This would workardound
the lack of core file...

-- 
                                          Gilles

Re: [Xenomai-help] No core files for segmentation faults in realtime tasks

[Christoph Permes]

Am Montag, den 24.08.2009, 10:40 +0200 schrieb Philippe Gerum:
> On Mon, 2009-08-24 at 10:07 +0200, Christoph Permes wrote:
> > Hi,
> > 
> > I have a realtime application running in user space that sometimes terminates with a segmentation fault. 
> > I get a message segfault at 4c4f5200 ip b7cb98d8 sp b7ef7354 error 4 in libc-2.7.so[b7c5e000+138000], 
> > but there is no core file writen, although ulimit is set to unlimited.
> > 
> > I wrote a small test program that produces a segmentation fault and I get a core file when the segfault
> > happens before the realtime task starts but not when it happens in the task.
> > 
> > Is it somehow possible to enable writing core files also for realtime tasks?
> 
> This feature is not disabled for real-time tasks actually, so there must
> be something else at work.
> 
> You may want to send your small test case, and give more details about
> your hardware and software configuration, at the very least the CPU
> architecture your code runs on, the Xenomai, I-pipe patch, and kernel
> release numbers.
> 

I'm using kernel 2.6.29 now, it turned out that this kernel version
handles core dump files not the same way as 2.6.22 which I used
previously.
Now every core dump is written to the same file, while my old kernel
generated a separate file for every core dump. Therefore it seemed that
no core files are written.
I solved the problem by adjusting /proc/sys/kernel/core_pattern to get
the old behavior.

Christoph