Re: [PATCH 5/8] locking, mutex: Cancelable MCS lock for adaptive spinning

[Peter Zijlstra <peterz@infradead.org>]

On Tue, Feb 25, 2014 at 11:56:19AM -0800, Jason Low wrote:
> On Mon, 2014-02-10 at 20:58 +0100, Peter Zijlstra wrote:
> 
> > +unqueue:
> > +	/*
> > +	 * Step - A  -- stabilize @prev
> > +	 *
> > +	 * Undo our @prev->next assignment; this will make @prev's
> > +	 * unlock()/unqueue() wait for a next pointer since @lock points to us
> > +	 * (or later).
> > +	 */
> > +
> > +	for (;;) {
> > +		if (prev->next == node &&
> > +		    cmpxchg(&prev->next, node, NULL) == node)
> > +			break;
> > +
> > +		/*
> > +		 * We can only fail the cmpxchg() racing against an unlock(),
> > +		 * in which case we should observe @node->locked becomming
> > +		 * true.
> > +		 */
> > +		if (smp_load_acquire(&node->locked))
> > +			return true;

I've stuck on in right about here. So that we don't unduly delay the
cmpxchg() after the load of prev.

> > +
> > +		/*
> > +		 * Or we race against a concurrent unqueue()'s step-B, in which
> > +		 * case its step-C will write us a new @node->prev pointer.
> > +		 */
> > +		prev = ACCESS_ONCE(node->prev);
> 
> Should we also add an arch_mutex_cpu_relax() to this loop?
> 
>