Re: iotop: khugepaged at 99.99% (2.6.38.X)

[Andrea Arcangeli <aarcange@redhat.com>]

On Fri, May 06, 2011 at 10:49:45AM +0200, Thomas Sattler wrote:
> > You can already run "grep threshold /proc/zoneinfo" on the system
> > where you reproduced the hang the last time (the one running 2.6.38.4)
> > the one with 1.5G of ram. They all should be well below 512 (so in
> > theory not causing troubles because of the per-cpu stats, and with so
> > few cpus it shouldn't have been such a longstanding problem anyway).
> 
> 'grep threshold /proc/zoneinfo' returns nothing on this machine after
> a reboot and before a crash. Did I tell it's a single core machine?

Hmm single core machine doesn't seem to fit my theory of per-cpu stats
too well... I think it's not impossible but it should only be possible
on smp and maybe it's not reproducible and probably not what you're
dealing with.

The new make-it-worse-patch should help to reproduce though.

If you can "cat /proc/zoneinfo" _during_ the hang (and then run
sysrq+t again to see the status, no need of sysrq+l), it'll help to
try to track this down. I'll recheck your last sysrq+t once again to
see if I get other ideas.

A 'cat /proc/zoneinfo' after the hang resolves itself and the system
is mostly idle, will also be interesting to verify the nr_isolated_*
are all zero.

Thanks,
Andrea