Re: PATCH] debug: add notifier chain debugging

[Christoph Hellwig <hch@infradead.org>]

On Fri, Aug 22, 2008 at 07:00:54AM -0700, Arjan van de Ven wrote:
> On Tue, 19 Aug 2008 21:09:32 -0700
> Andrew Morton <akpm@linux-foundation.org> wrote:
> 
> > 
> > If we remove the first two checks then surely we can afford to add the
> > remaining check unconditionally and lose the new config option and its
> > ifdef.
> > 
> 
> Andi told me that this notifier is a hotpath for kprobes (systemtap)
> use models; at which point making this a config option makes sort of
> sense to get the last bit of performance out of that. Personally I would
> always want this option on for anything I run, and I suspect
> (enterprise) distros will just always turn it on as well since it's a
> good sanity check at low cost.
> 
> I'd be happy to lose the config (I think we have way too many unneeded
> ones of those already); I'll make a patch.

What about checking yourself? At least the braindead use of notiiers in
the pagefault path for kprobes (even unconditional on x86_64) is gone
now and replaced with proper kprobes hooks. The die notifier that's
also used by kprobes shouldn't be in a performance critical path.