Re: irqbalance on movidis crashes the machine (was: movidis x16 hard lockup using 2.6.33)

[Andreas Barth <aba@not.so.argh.org>]

* Andreas Barth (aba@not.so.argh.org) [100415 20:43]:
> * Peter 'p2' De Schrijver (p2@debian.org) [100403 17:43]:
> > http://zobel.ftbfs.de/.x/lucatelli-nmi-watchdog-output.txt 
> > Dump of one of those hangs. Most cores seem to be stuck in wait 
> > (0xffffffff81100b80), except for core 1 which is in octeon_irq_ciu0_ack 
> > (octeon_irq_ciu0_ack).
> 
> On further investigation we found out that this happens when
> irqbalance is started. The version of irqbalance being run is 0.55.
> 
> We removed this program from the affected machine, but of course this
> still should be fixed (and we still get a few reboots on another
> machine without irqbalance).

Clarification:

Running irqbalance itself doesn't crash the machine, but increases the
probability of crashes dramatically. Usually the next few (< 10)
commands crash the machine.

The crashs however look similar to the ones we have without irqbalance
- just way less often then with irqbalance. This seems like irqbalance
exposes the crash way better than we do without.


Andi