[LITMUS^RT] running a task as its execution time

[Björn Brandenburg]

On Apr 30, 2012, at 4:11 PM, Giovani Gracioli wrote:
> I am getting successive kernel panics at the same litmus/budget.c line 65 using GSN-EDF scheduler. The application code (test.c) and the task sets are attached. You can run ./test 125 125_tasks_litmus.txt for 125 tasks or 150 150_tasks_litmus.txt for 150 tasks.
> 
> The processor is an intel i7-2600 with 4 cores and 2 threads per core.

I am able to reproduce the problem. After some debugging, I think this is what happens:

1) A task with precise enforcement has zero budget
2) It gets unlinked before the timer is handled. The budget/job completion is not processed because of this check in GSN-EDF's schedule():

471:        if (!np && (out_of_time || sleep) && !blocks && !preempt)                                                                                                                                                                             
472:                job_completion(entry->scheduled, !sleep);

3) The task gets relinked elsewhere.
4) When it is scheduled, the BUG_ON() triggers.

However, just removing the !preempt check will cause trouble since timers. This is not straightforward.
I'll continue to investigate tomorrow. In the mean while, suggestions anyone? Glenn? Chris? Jonathan?

Thanks,
Björn