Re: swapper: page allocation failure.

[Nick Piggin <nickpiggin@yahoo.com.au>]

Chris Wedgwood wrote:
> On Sun, May 21, 2006 at 11:31:12AM +0200, Haar J?nos wrote:
> 
> 
>>[root@st-0001 /]# uname -a
>>Linux st-0001 2.6.17-rc3-git1 #2 SMP Sun May 21 01:12:22 CEST 2006 i686 i686 i386 GNU/Linux
> 
> 
> did earlier kernels work OK?
> 
> 
>>This is a simple disk node.
>>It serves the md0 array, and uses mem for buffering-caching.
> 
> 
> odd, i looks like you've leaked alot of lowmem but i can't think why
> 
> i've got major (induced) brain-fog right now so i'll have to think
> about it tomorrow sorry

The buffers are buffercache rather than the usual pagecache; due to
nbd I guess. Buffercache cannot be satisfied by highmem.

This would be a relatively uncommon setup, which explains why it
isn't working 100%. I don't know of any reason why reclaim speed
should be worse for buffercache, however one notable thing will be
that zone_normal's lowmem reserve that is untouchable by pagecache
will be eaten by buffercache...

Anyway, increasing /proc/sys/vm/min_free_kbytes should help. Janos,
perhaps you could try doubling it and see how you go?

-- 
SUSE Labs, Novell Inc.
Send instant messages to your online friends http://au.messenger.yahoo.com