Re: [RFC][PATCH 2/6] x86, nmi: create new NMI handler routines

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Wed, Aug 24, 2011 at 08:19:54PM +0200, Peter Zijlstra wrote:
> On Wed, 2011-08-24 at 14:16 -0400, Don Zickus wrote:
> > On Wed, Aug 24, 2011 at 07:51:00PM +0200, Peter Zijlstra wrote:
> > > On Wed, 2011-08-24 at 13:44 -0400, Don Zickus wrote:
> > > > > > +   rcu_read_lock();
> > > > > > +   a = rcu_dereference_raw(*ap);
> > > > > 
> > > > > The reason for rcu_dereference_raw() is to prevent lockdep from choking
> > > > > due to being called from an NMI handler, correct?  If so, please add a
> > > > > comment to this effect on this and similar uses.
> > > > 
> > > > That sounds right.  But honestly, I just copied what notifier_call_chain
> > > > had.  Regardless, I will make sure to document that in my next version.
> > > > Thanks! 
> > > 
> > > Not quite right, nmi_enter() does lockdep_disable() and makes
> > > lock_is_held() return always true.
> > > 
> > > I think this (and the other sites) could do with rcu_dereference_check(,
> > > lockdep_is_held(&desc->lock)); not that it wouldn't be anything but
> > > documentation since the actual test isn't working from NMI context but I
> > > do think its worth it for that alone.
> > 
> > So you want me to remove the _raw part of the dereference?  I can test
> > that with lockdep enabled to verify things don't go splat.
> 
> Ah, right, its never used from the desc->lock context and we always hold
> rcu_read_lock(), so a simple rcu_dereference() should indeed suffice. 

Even better!  ;-)

							Thanx, Paul