Re: [PATCH] [LSM] Rework LSM hooks

[Chris Wright <chrisw@osdl.org>]

* James Morris (jmorris@redhat.com) wrote:
> On Tue, 10 Aug 2004, Alan Cox wrote:
> 
> > On Maw, 2004-08-10 at 15:16, James Morris wrote:
> > > On Tue, 10 Aug 2004, Kurt Garloff wrote:
> > > 
> > > > * Even with selinux=0 and capability loaded, the kernel takes a 
> > > >   few percents in networking benchmarks (measured by HP on ia64); 
> > > >   this is caused by the slowliness of indirect jumps on ia64.
> > > 
> > > Is this just an ia64 issue?  If so, then perhaps we should look at only
> > > penalising ia64?  Otherwise, loading an LSM module is going to cause
> > > expensive false unlikely() on _every_ LSM hook.
> > 
> > I see this on x86-32 to an extent. Its quite visible with gigabit as
> > you'd expect. ia64 ought to be less affected providing the compiler is
> > doing the right things with the unconditional jumps.
> 
> I did some benchmarking (full results below), and I'm not seeing anything
> significant on a P4 Xeon.

Is this new (i.e. you just did this)?  It's basically the same result we
had from a few years ago.

thanks,
-chris
-- 
Linux Security Modules     http://lsm.immunix.org     http://lsm.bkbits.net