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

[Davidlohr Bueso <dave@stgolabs.net>]

On Fri, 2015-01-30 at 17:51 -0800, Tim Chen wrote:
> On Fri, 2015-01-30 at 01:14 -0800, Davidlohr Bueso wrote:
> > When readers hold the semaphore, the ->owner is nil. As such,
> > and unlike mutexes, '!owner' does not necessarily imply that
> > the lock is free. This will cause writers to potentially spin
> > excessively as they've been mislead to thinking they have a
> > chance of acquiring the lock, instead of blocking.
> > 
> > This patch therefore enhances the counter check when the owner
> > is not set by the time we've broken out of the loop. Otherwise
> > we can return true as a new owner has the lock and thus we want
> > to continue spinning. While at it, we can make rwsem_spin_on_owner()
> > less ambiguos and return right away under need_resched conditions.
> > 
> > Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
> > ---
> >  kernel/locking/rwsem-xadd.c | 21 +++++++++++++++------
> >  1 file changed, 15 insertions(+), 6 deletions(-)
> > 
> > diff --git a/kernel/locking/rwsem-xadd.c b/kernel/locking/rwsem-xadd.c
> > index 07713e5..1c0d11e 100644
> > --- a/kernel/locking/rwsem-xadd.c
> > +++ b/kernel/locking/rwsem-xadd.c
> > @@ -337,21 +337,30 @@ static inline bool owner_running(struct rw_semaphore *sem,
> >  static noinline
> >  bool rwsem_spin_on_owner(struct rw_semaphore *sem, struct task_struct *owner)
> >  {
> > +	long count;
> > +
> >  	rcu_read_lock();
> >  	while (owner_running(sem, owner)) {
> > -		if (need_resched())
> > -			break;
> > +		/* abort spinning when need_resched */
> > +		if (need_resched()) {
> > +			rcu_read_unlock();
> > +			return false;
> > +		}
> >  
> >  		cpu_relax_lowlatency();
> >  	}
> >  	rcu_read_unlock();
> >  
> > +	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.

Thanks,
Davidlohr