Re: 2.5.38-mm1

[Andrew Morton]

John Levon wrote:
> 
> On Sun, Sep 22, 2002 at 05:54:31PM -0700, Andrew Morton wrote:
> 
> > It found a bug.  Someone is calling kmem_cache_create() in an
> > atomic region.
> 
> And kmem_cache_alloc() has jumped to the top of the profile (checked
> with readprofile) in 2.3.38-linus.
> 

Linus disabled the cpu-local caches if slab debugging is
enabled.  This is because they were not being poisoned,
and so SMP machines were not getting the full debug benefit
of slab poisoning.

If you disable kernel debugging (either in config, or locally
in slab) then it should be fine.

Slab performance has always been sucky with debug enabled,
so no real loss there.