Re: [PATCH 4/5] locking/rwsem: Avoid deceiving lock spinners

[Tim Chen]

On Tue, 2015-02-03 at 09:54 -0800, Jason Low wrote:
> On Tue, 2015-02-03 at 09:16 -0800, Tim Chen wrote:
> > > > >  
> > > > > +	if (READ_ONCE(sem->owner))
> > > > > +		return true; /* new owner, continue spinning */
> > > > > +
> > > > 
> > > > Do you have some comparison data of whether it is more advantageous
> > > > to continue spinning when owner changes?  After the above change, 
> > > > rwsem will behave more like a spin lock for write lock and 
> > > > will keep spinning when the lock changes ownership.
> > > 
> > > But recall we still abort when need_resched, so the spinning isn't
> > > infinite. Never has been.
> > > 
> > > >  Now during heavy
> > > > lock contention, if we don't continue spinning and sleep, we may use the
> > > > clock cycles for actually running other threads. 
> > > 
> > > Under heavy contention, time spinning will force us to ultimately block
> > > anyway.
> > 
> > The question is under heavy contention, if we are going to block anyway,
> > won't it be more advantageous not to continue spinning so we can use
> > the cycles for useful task?
> 
> Hi Tim,
> 
> Now that we have the OSQ logic, under heavy contention, there will still
> only be 1 thread that is spinning on owner at a time. 

That's true.  We cannot have the lock grabbed by a new write 
contender as any new writer contender of the lock will be 
queued by the OSQ logic. Only the
thread doing the optimistic spin is attempting write lock.  
In other word, switching of write owner of the rwsem to a new
owner cannot happen.  Either write owner stay as the original one, or
we don't have a write owner.  So using test of write owner
switching as an indicator of congestion is incorrect.

If my reasoning above is sound, then the check 

+       if (READ_ONCE(sem->owner))
+               return true; /* new owner, continue spinning */
+

is unnecessary and can be removed, as we cannot have a 
new write owner of the rwsem, other than the thread
doing optimistic spinning.

Tim


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/