Re: [PATCH v2] rwsem: Support optimistic spinning

[Peter Zijlstra <peterz@infradead.org>]

On Wed, Apr 30, 2014 at 09:32:17AM -0700, Davidlohr Bueso wrote:
> On Wed, 2014-04-30 at 11:21 +0200, Peter Zijlstra wrote:
> > On Mon, Apr 28, 2014 at 03:09:01PM -0700, Davidlohr Bueso wrote:
> > > +#ifdef CONFIG_SMP
> > 
> > > +#else
> > > +static bool rwsem_optimistic_spin(struct rw_semaphore *sem)
> > > +{
> > > +	return false;
> > > +}
> > > +#endif
> > 
> > On the mutex side we guard this with MUTEX_SPIN_ON_OWNER, do we want to
> > use that here too?
> 
> I thought of it, but hated adding mutex naming to rwsem code -- we
> already do it for cpu_relax() thanks to s390.

Yah, we need to find a better name for that thing.

> MUTEX_SPIN_ON_OWNER depends on SMP && !DEBUG_MUTEXES.
> Right now rwsem optimistic spinning depends on SMP && RWSEM_XCHGADD_ALGORITHM.
> 
> It might sense to add RWSEM_SPIN_ON_OWNER to encapsulate what we already
> have. I don't think we want DEBUG_MUTEX dependencies in rwsems. Would
> you accept such thing?

Ah, I remember why we have that MUTEX_SPIN_ON_OWNER, its in the comment
in mutex.c right under using it. It the games we play with ->owner isn't
compatible with what DEBUG_MUTEXES wants it for.

So no, you're fine with the rwsem code as is.