On Thu, 19 May 2011, Peter Kruse wrote:

> you may remember, we have a server running 2.6.32.29
> and once in a while it just crashes, meaning that
> it just stops responding, and we have to reset it.
> The kernel itself and some processes are still running,
> and it is possible to initiate the sysRQ.
> The mentioned allocation failures are gone after the update
> to the kernel version, but the server crashed again
> after 40 days running without any problem.

Good to hear.
>
> At this time kswapd0/1 started to consume 99% CPU time
> and until the server was reset 8 hours later
> never used less than 50% CPU time.  Shortly after
> that (half an hour) the disk reads dropped down to zero.  And
> a program produced the attached Call Trace.  I also
> attach the process information of kswapd (provided by collectl).

We need the complete system log starting from bootup.

This looks like it has something to do with mutexes and pipes.