2.6.7-vanilla-SMP kernel: pagebuf_get: failed to lookup pages

2.6.7-vanilla-SMP kernel: pagebuf_get: failed to lookup pages

[L A Walsh]

Jul 20 09:07:34 ishtar kernel: pagebuf_get: failed to lookup pages
Jul 20 09:07:34 ishtar last message repeated 25 times
Jul 20 09:26:38 ishtar kernel: pagebuf_get: failed to lookup pages
Jul 20 09:27:09 ishtar last message repeated 354 times
Jul 20 09:27:52 ishtar last message repeated 274 times
Jul 20 09:45:46 ishtar kernel: pagebuf_get: failed to lookup pages
Jul 20 09:45:46 ishtar last message repeated 2 times
Jul 20 10:00:10 ishtar kernel: pagebuf_get: failed to lookup pages
Jul 21 02:30:00 ishtar su: (to backup) root on none
Jul 21 02:30:01 ishtar kernel: pagebuf_get: failed to lookup pages
Jul 21 02:30:04 ishtar last message repeated 16 times
Jul 21 02:30:30 ishtar su: (to backup) root on none
Jul 21 02:31:55 ishtar su: (to backup) root on none
Jul 21 02:31:55 ishtar last message repeated 3 times
Jul 21 03:15:09 ishtar kernel: pagebuf_get: failed to lookup pages
Jul 21 03:15:09 ishtar last message repeated 4 times
Jul 21 04:07:34 ishtar kernel: pagebuf_get: failed to lookup pages
Jul 21 04:07:34 ishtar last message repeated 9 times
Jul 21 04:26:44 ishtar kernel: pagebuf_get: failed to lookup pages
Jul 21 04:27:45 ishtar last message repeated 1516 times
Jul 21 04:27:54 ishtar last message repeated 36 times
Jul 21 04:45:51 ishtar kernel: pagebuf_get: failed to lookup pages
Jul 21 04:45:51 ishtar last message repeated 7 times

----

    Any idea what this message means?  I especially notice a high 
frequency during
high disk i/o.  File systems are all xfs if that is pertinent.  Backups 
run in early
AM backing up SCSI disks to a large IDE.  However, the messages around 
9:27 on the
20th wouldn't have been backup related but possibly processing a backlog of
email after some system maintenance -- and that would have all been on 
SCSI disks.

Re: 2.6.7-vanilla-SMP kernel: pagebuf_get: failed to lookup pages

[Chris Wedgwood]

On Wed, Jul 21, 2004 at 05:04:09PM -0700, L A Walsh wrote:

> Any idea what this message means?

it means "try the CVS tree" (i think hch fixed this and it's in CVS
but not mainline)


  --cw

Re: 2.6.7-vanilla-SMP kernel: pagebuf_get: failed to lookup pages

[L A Walsh]

Will this be included/fixed in 2.6.8?

How serious is the problem?  The system doesn't seem to panic or 
indicate backup
failures.

Setting up a CVS tree to get a patch for a "stable-series" kernel seems 
a bit
unstable.  I'm not sure what I'd pull in besides the fix or even if I'd pull
down a coherent/stable CVS image if I downloaded in the middle of when some
other patch was being checked in.  Maybe I'm sounding like a wimp, but 
the idea
of pulling in freshly checked in CVS code for use on a 'stable' machine is
bordering on my discomfort zone. :-)

-l

Chris Wedgwood wrote:

>On Wed, Jul 21, 2004 at 05:04:09PM -0700, L A Walsh wrote:
>
>  
>
>>Any idea what this message means?
>>    
>>
>
>it means "try the CVS tree" (i think hch fixed this and it's in CVS
>but not mainline)
>
>
>  --cw
>  
>

Re: 2.6.7-vanilla-SMP kernel: pagebuf_get: failed to lookup pages

[Chris Wedgwood]

On Wed, Jul 21, 2004 at 05:21:25PM -0700, L A Walsh wrote:

> Will this be included/fixed in 2.6.8?

i assume that's the intention but i don't know when 2.6.8 is and how
much time the sgi people have before then.  my guess is yes though

> How serious is the problem?  The system doesn't seem to panic or
> indicate backup failures.

not sure, hch can you comment here maybe?

> Setting up a CVS tree to get a patch for a "stable-series" kernel
> seems a bit unstable.

CVS is "stable linux releases + XFS fixes" --- it's really not that
bad (whilst i personally don't use it, my tree is derieved from it and
i don't have problems)

> I'm not sure what I'd pull in besides the fix or even if I'd pull
> down a coherent/stable CVS image if I downloaded in the middle of
> when some other patch was being checked in.

  cd path/to/workarea
  cvs -qz9 -d :pserver:cvs@oss.sgi.com:/cvs co linux-2.6-xfs
  cd linux-2.6-xfs
  cp path/to/old/.config .config
  make oldconfig
  make ...

> Maybe I'm sounding like a wimp, but the idea of pulling in freshly
> checked in CVS code for use on a 'stable' machine is bordering on my
> discomfort zone. :-)

FWIW, the CVS tree isn't freshly checked in, it's a reflection of the
internal ptools tree where in theory you shouldn't get adhoc checkins
like lots of places (but yeah, it does sometimes break but not usually
badly).

Anyhow, if you don't like it you can (1) ingnore the problems (2) use
official binary releases from vendors (3) use ext3, etc.

whatever works for you...


  --cw

Re: 2.6.7-vanilla-SMP kernel: pagebuf_get: failed to lookup pages

[Nathan Scott]

On Wed, Jul 21, 2004 at 05:33:57PM -0700, Chris Wedgwood wrote:
> On Wed, Jul 21, 2004 at 05:21:25PM -0700, L A Walsh wrote:
> 
> > Will this be included/fixed in 2.6.8?
> 
> i assume that's the intention but i don't know when 2.6.8 is and how
> much time the sgi people have before then.  my guess is yes though

The fix has been included in the 2.6.8-pre/rc kernels for some
time now, so yes it'll be in 2.6.8.

> > How serious is the problem?  The system doesn't seem to panic or
> > indicate backup failures.
> 
> not sure, hch can you comment here maybe?

This leaked locked pages on metadata readahead failure (which
could occur when free memory becomes low), which is serious.

cheers.

-- 
Nathan