Bug#597489: kswapd ownes the whole cpu resources, unable to reboot

[Jonathan Nieder <jrnieder@gmail.com>]

retitle 597489 kswapd using 100% CPU, unable to reboot (soft lockup in pagevec_lookup)
found 597489 linux-2.6/2.6.32-21
fixed 597489 linux-2.6/2.6.38-2
tags 597489 = upstream fixed-upstream
quit

Hi,

Giuseppe Lavagetto wrote:

> I backported 2.6.38 from sid to squeeze, and the machine is running fine
> since then. I tried to reproduce the problem with my test case, with no
> luck. So, it seems that changes that have been performed to the MM
> subsystem (specifically, the patch I referred to in a previous post) do
> solve this issue.

Thanks for checking.  Marking accordingly.

Out of curiosity, how did you find v2.6.38-rc1~216 (mm: page
allocator: adjust the per-cpu counter threshold when memory is low,
2011-01-13)?  Is there a corresponding RH bugzilla entry with details?

I thought that patch fixed a regression introduced by v2.6.32.23~22
(mm: page allocator: calculate a better estimate of NR_FREE_PAGES when
memory is low and kswapd is awake, 2010-09-09), which would correspond
to version 2.6.32-24 of the linux-2.6 package, but the original report
was against 2.6.32-21.  Moreover, 2.6.32-24 was not uploaded until
after a couple of weeks after the original report.  Hmm.

Thomas, if you get a chance to try a kernel from sid or experimental,
that would be interesting.

Regards,
Jonathan