Re: [PATCH v5 tip/core/locking 5/7] Documentation/memory-barriers.txt: Downgrade UNLOCK+LOCK

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

On Tue, Dec 10, 2013 at 09:43:20AM -0800, Josh Triplett wrote:
> On Tue, Dec 10, 2013 at 06:25:28PM +0100, Peter Zijlstra wrote:
> > On Tue, Dec 10, 2013 at 09:12:47AM -0800, Paul E. McKenney wrote:
> > > > The way I read the above it says that you need
> > > > smp_mb__after_unlock_lock() when the UNLOCK and LOCK are on the same
> > > > variable. That doesn't make sense, I thought that was the one case we
> > > > all agreed on it would indeed be a full barrier without extra trickery.
> > > 
> > > On x86, sure, but smp_mb__after_unlock_lock() is nothingness on x86
> > > anyway.  Other architectures might benefit from requiring that the
> > > smp_mb__after_unlock_lock() be used in this case.
> > 
> > Confused, UNLOCK X, LOCK X, must always be fully serializing. That's the
> > entire purpose of the thing.
> > 
> > The only place you can go play games (and clearly we are going there) is
> > when the UNLOCK and LOCK are on different variables.
> 
> That would certainly be a good assumption to preserve, and it would
> eliminate most of the need for smp_mb__after_unlock_lock().

Perhaps RCU is an outlier, but most of the places where I added
smp_mb__after_unlock_lock() had an unlock of one rcu_node structure's
->lock followed by a lock of another rcu_node structure's ->lock.

							Thanx, Paul