Re: lookup on memory shortage

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Fri, Oct 03, 2008 at 01:29:11PM +0000, Andrew Doran wrote:
> On Fri, Oct 03, 2008 at 01:11:05PM +0200, Manuel Bouyer wrote:
> 
> > |CPU states:  0.0% user,  0.0% nice,  100% system,  0.0% interrupt,  0.0% 
> > idle
> 
> Presumably the pagedaemon is spinning.
> 
> > |              21 2       204           ca2729e0             xenbus
> 
> Will disk I/O stop if this thread cannot get CPU time?

No, it should not. It's there only to handle devices appearing/dissaparing.
There's no specific thread for disk I/O in Xen domUs

> 
> > ddb bt/a isn't of much use:
> > |db> bt/a ca271040
> 
> ca271040 is running on the CPU so it won't be of much use, right. 'bt' with
> no arguments would have shown it:

But it did the same: get a trap crossing the interrupt frame.

> a stupid UI quirk in ddb. Can you add some
> counters to the pagedaemon main loop to see what it is doing?
> 
> > The map for cc1 looks large but I'm not sure it explains where the
> > memory did go. I wonder if there's some memory leak in the kernel that
> > only show up with certain usage patterns (the box can build packages for
> > days with almost no swap in use so it's not a slow memory leak).
> 
> Wired usage is low, so it does not look like a simple kernel memory leak.
> There are plenty of pages that can be reclaimed but the pagedaemon is
> failing to do so.

OK, I'll try to get more details from pagedaemon next time.

-- 
Manuel Bouyer, LIP6, Universite Paris VI.           
Manuel.Bouyer%lip6.fr@localhost
     NetBSD: 26 ans d'experience feront toujours la difference
--