Re: 2.6.32.5 regression: page allocation failure. order:1,

[Mark Lord <kernel@teksavvy.com>]

David Rientjes wrote:
> On Wed, 27 Jan 2010, Mark Lord wrote:
>> It's rock solid again with 2.6.31.12 on it now.
>>
> 
> Is there something specific about the workload that makes it easily 
> reproducible?  Are you saying that 2.6.31.12 is "rock solid" because it 
> has survived a certain workload that caused these page allocation failures 
> with 2.6.32.5, or simply because it has a longer uptime and hasn't shown 
> a problem?  It would be very helpful to describe the load so that we can 
> attempt to reproduce it locally without a sacrifice to your server.
..

Good questions, and I do "feel for you" here.  :)

That machine has a light workload.  Web server, email server,
dhcp for the small local network, name server, etc.

It sits at one end of a 6mb/1.5mb DSL connection, hardly busy at all.

But the logfile posted shows many of those "allocation failures"
within the first (only)36 hours of running 2.6.32.5:

  mirrordir (doing a backup to a USB drive)
  apcupsd  (UPS monitoring; hardly an intensive activity)
  apache2  (the web server receives rather light use, and no fancy php or anything)
  apache2  (again!)
  apache2  (and again)
  apache2  (another)
  apache2  (yet again)
  ...
  apache2  (and again again)
  vim      (hey, can't a guy edit his driver sources ??)
  cc1      (or compile them?)

So, not much happening, really.
It's a slow machine, a 600Mhz C7 ("VIA Samuel 2"), with only 512MB of RAM.

We've got 2.6.32.5 running on several other machines here with nary a glitch,
but all of those have 2-4GB of RAM.

Cheers