Re: [PATCH] garbage values in file /proc/net/sockstat

[Andi Kleen <ak@suse.de>]

On Wednesday 25 January 2006 20:59, Ravikiran G Thirumalai wrote:
> On Wed, Jan 25, 2006 at 02:31:15PM +0100, Andi Kleen wrote:
> > On Monday 23 January 2006 17:46, Eric Dumazet wrote:
> > 
> > I think the best course of action for this now for 2.6.16 is:
> > 
> > - mark percpu init data not __init
> > (this way it will still reference valid memory, although shared between
> > all impossible CPUs)
> > - keep the impossible CPUs per cpu data to point to the original reference  
> > version (== offset 0)
> > 
> 
> How about doing the above using a debug config option? So that when the
> config option is turned on, all per-cpu area references to not possible 
> cpus crash? and leave that option default on on -mm :)

In -mm* we could just apply Eric's patch and then someone should just
grep the tree for NR_CPUS and audit all users - that should
catch basically all occurrences. I can put it onto my todo list,
but I don't know when I'll get to it so it would be nice if someone
else could do this.

For 2.6.16 I think it's best to go forward with my hack.

> .  That way we can  
> quickly catch all references.  We can probably change the arch independent 
> setup_per_cpu_areas also to do allocations for cpu_possible cpus only while 
> we are at it?

Eric did that already.

-Andi