Re: [Xenomai-help] real time task disapears... memory problem ?

[Jan Kiszka <jan.kiszka@domain.hid>]

[-- Attachment #1: Type: text/plain, Size: 2020 bytes --]

desvages@domain.hid wrote:
>> From my kernel console after running your program for a few seconds:
>>
>> Xenomai: watchdog triggered -- killing runaway thread 'MyAlarmServer'
>>
>> Previously, "rt_alarm_wait" was printed on the console, thus gsl_qp()
>> likely entered some infinite loop. You have the watchdog enabled as
>> well, so you should see the same effect, right?
>>
>> Jan
> 
> In my console, I only see the "rt_alarm_wait". Is there something to
> activate to enable xenomai errors messages ?

Not local console, _kernel console_. Check your systems log files (eg.
/var/log/syslog or .../messages), or run "dmesg" to see the last kernel
messages.

> 
> As this gsl_qp() function takes about 6/7 secondes to solve my quadratic
> problem, the watchdog could effectively be the thing that goes wrong...

Well, don't blame others until you are sure (== you measured) that it is
not your own code. :)

> 
> To be sure to understand how works the watchdog in xenomai:
> - Is it a limit on the execution time (so time effectively used by the task)
> - or is it a limit on the response time (the time between the beginning
> and the end of the task, even if it is preempted) ?

See documentation. Linux must be able to run at least once in _4_ seconds.

> 
> Because, if it is the first case, maybe that putting the gsl_qp() function
> in primary mode increases the execution time measured by xenomai, and this
> activate the watchdog ? whereas in secondary mode, time is not counted by
> xenomai watchdog ?

Yes, secondary mode would be accounted to Linux, thus the watchdog would
be happy (and your Linux subsystem as well because it would be allowed
to breath).

> 
> I will try to increase the watchdog time to see if this is only a matter
> of a task taking to much time.

Wrong approach. First check how long your loop actually takes. If you
have to tweak the watchdog (and there is no bug in the trivial watchdog
code), it's your code that is broken.

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 250 bytes --]