[LTP] [PATCH v2] sched: starvation: Autocallibrate the timeout

[Cyril Hrubis]

Hi!
> > > external/ltp/testcases/kernel/sched/cfs-scheduler/starvation.c:161: TINFO: do_test by pid 4523
> > > external/ltp/testcases/kernel/sched/cfs-scheduler/starvation.c:166: TINFO: main pid is 4523
> > > external/ltp/testcases/kernel/sched/cfs-scheduler/starvation.c:167: TINFO: child pid is 4524
> > > external/ltp/testcases/kernel/sched/cfs-scheduler/starvation.c:166: TINFO: main pid is 4524
> > > external/ltp/testcases/kernel/sched/cfs-scheduler/starvation.c:167: TINFO: child pid is 0
> > > external/ltp/testcases/kernel/sched/cfs-scheduler/starvation.c:176: TINFO: go loop, current pid is 4523
> > > external/ltp/testcases/kernel/sched/cfs-scheduler/starvation.c:145: TINFO: current ppid is 4523, current pid is 4524, go to child() start
> 
> > > main pid is 4523, child pid is 4524, and we only see child pid is working (checking by top)
> 
> > > 4524 root         20   0  14M 472K    0 R 85.7   0.0   0:14.93 starvation_v4 -t 1000000 -l 1000000 <-- cpu_load by top
> 
> > So this looks like we managed to reproduce the issue the test was trying
> > to catch. The child is hogging the CPU and the parent process cannot
> > proceed. And I suppose that this happens only on 32bit because 32bit is
> > less tested these days.
> 
> I guess we can merge this, right?
> Or we still not sure whether this is really kernel bug not related to the
> change?

It stil needs a runtime check at the end of the loop in a case that
timeout is multiplied. I will have to send another version, thanks for
reminding me.

-- 
Cyril Hrubis
chrubis@suse.cz