[Xenomai-help] remove_proc_entry problem semi-tracked down

[Xenomai-help] remove_proc_entry problem semi-tracked down

[Steven Seeger]

The kernel panic in remove_proc_entry happens when a userspace fusion app
calls a xenomai function that deletes something from the register in atexit.
For example, I have a set of c++ classes that wrap mutexes, queues, etc. If
I have a global instance, it sometimes causes a kernel panic in the
destructor where the rt_mutex_delete, rt_queue_delete, etc, is called. This
seems to always happen with rt_queue_delete, but not always with
rt_mutex_delete. (A couple of starts/stops seem to create the problem.)

This is when exiting with a call to exit(0), not with a control-c.

Steven

Re: [Xenomai-help] remove_proc_entry problem semi-tracked down

[Philippe Gerum]

Steven Seeger wrote:
> The kernel panic in remove_proc_entry happens when a userspace fusion app
> calls a xenomai function that deletes something from the register in atexit.
> For example, I have a set of c++ classes that wrap mutexes, queues, etc. If
> I have a global instance, it sometimes causes a kernel panic in the
> destructor where the rt_mutex_delete, rt_queue_delete, etc, is called. This
> seems to always happen with rt_queue_delete, but not always with
> rt_mutex_delete. (A couple of starts/stops seem to create the problem.)
> 
> This is when exiting with a call to exit(0), not with a control-c.
> 

Please (re)send the panic stack dump you get.

-- 

Philippe.

Re: [Xenomai-help] remove_proc_entry problem semi-tracked down

[Steven Seeger]

EIP is at remove_proc_entry+0x20/0x120

__registry_proc_callback
__registry_proc_callback
__registry_proc_callback
worker_thread
default_wake_function
worker_thread
kthread
kthread
kernel_thread_helper


On 10/12/05 7:45 AM, "Philippe Gerum" <rpm@xenomai.org> wrote:

> Steven Seeger wrote:
>> The kernel panic in remove_proc_entry happens when a userspace fusion app
>> calls a xenomai function that deletes something from the register in atexit.
>> For example, I have a set of c++ classes that wrap mutexes, queues, etc. If
>> I have a global instance, it sometimes causes a kernel panic in the
>> destructor where the rt_mutex_delete, rt_queue_delete, etc, is called. This
>> seems to always happen with rt_queue_delete, but not always with
>> rt_mutex_delete. (A couple of starts/stops seem to create the problem.)
>> 
>> This is when exiting with a call to exit(0), not with a control-c.
>> 
> 
> Please (re)send the panic stack dump you get.

Re: [Xenomai-help] remove_proc_entry problem semi-tracked down

[Steven Seeger]

I only seem to get this problem if I start and stop my app a few times
without unloading the xenomai modules and reloading them, btw.

Steven

Re: [Xenomai-help] remove_proc_entry problem semi-tracked down

[Philippe Gerum]

Steven Seeger wrote:
> I only seem to get this problem if I start and stop my app a few times
> without unloading the xenomai modules and reloading them, btw.
> 

Ok, this might be a useful hint. Please try crafting a simple test case creating 
a hundred mutexes then destroying them in a row, restarting this app a few times 
without unloading the xeno_native module. It would help to know if it breaks too.

-- 

Philippe.

Re: [Xenomai-help] remove_proc_entry problem semi-tracked down

[Steven Seeger]

It doesn't break when all my app does is this. This issue requires more
investigation. Still, nothing I do in userspace should make it kernel panic.
:)

Steven

On 10/12/05 7:54 AM, "Philippe Gerum" <rpm@xenomai.org> wrote:

> Steven Seeger wrote:
>> I only seem to get this problem if I start and stop my app a few times
>> without unloading the xenomai modules and reloading them, btw.
>> 
> 
> Ok, this might be a useful hint. Please try crafting a simple test case
> creating 
> a hundred mutexes then destroying them in a row, restarting this app a few
> times 
> without unloading the xeno_native module. It would help to know if it breaks
> too.

Re: [Xenomai-help] remove_proc_entry problem semi-tracked down

[Philippe Gerum]

Steven Seeger wrote:
> It doesn't break when all my app does is this. This issue requires more
> investigation. Still, nothing I do in userspace should make it kernel panic.

Sure, this exactely why we are trying to find this bug.

> :)
> 
> Steven
> 
> On 10/12/05 7:54 AM, "Philippe Gerum" <rpm@xenomai.org> wrote:
> 
> 
>>Steven Seeger wrote:
>>
>>>I only seem to get this problem if I start and stop my app a few times
>>>without unloading the xenomai modules and reloading them, btw.
>>>
>>
>>Ok, this might be a useful hint. Please try crafting a simple test case
>>creating 
>>a hundred mutexes then destroying them in a row, restarting this app a few
>>times 
>>without unloading the xeno_native module. It would help to know if it breaks
>>too.
> 
> 
> 


-- 

Philippe.

Re: [Xenomai-help] remove_proc_entry problem semi-tracked down

[Philippe Gerum]

Steven Seeger wrote:
> It doesn't break when all my app does is this. This issue requires more
> investigation. Still, nothing I do in userspace should make it kernel panic.
> :)

Dmitry found a serious flaw in the recent I-pipe patches that I've just fixed. 
You might want to try this one out:
http://download.gna.org/adeos/patches/v2.6/adeos/i386/adeos-ipipe-2.6.13-i386-1.0-07.patch

> 
> Steven
> 
> On 10/12/05 7:54 AM, "Philippe Gerum" <rpm@xenomai.org> wrote:
> 
> 
>>Steven Seeger wrote:
>>
>>>I only seem to get this problem if I start and stop my app a few times
>>>without unloading the xenomai modules and reloading them, btw.
>>>
>>
>>Ok, this might be a useful hint. Please try crafting a simple test case
>>creating 
>>a hundred mutexes then destroying them in a row, restarting this app a few
>>times 
>>without unloading the xeno_native module. It would help to know if it breaks
>>too.
> 
> 
> 


-- 

Philippe.