Re: [PATCH RFC ticketlock] v2 Auto-queued ticketlock

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

On Tue, Jun 11, 2013 at 01:36:27PM -0400, Steven Rostedt wrote:
> On Tue, 2013-06-11 at 10:02 -0700, Paul E. McKenney wrote:
> 
> > +#ifdef CONFIG_TICKET_LOCK_QUEUED
> > +
> > +#define __TKT_SPIN_INC 2
> > +bool tkt_spin_pass(arch_spinlock_t *ap, struct __raw_tickets inc);
> > +
> > +#else /* #ifdef CONFIG_TICKET_LOCK_QUEUED */
> > +
> > +#define __TKT_SPIN_INC 1
> > +static inline bool tkt_spin_pass(arch_spinlock_t *ap, struct __raw_tickets inc)
> > +{
> > +	return false;
> > +}
> > +
> > +#endif /* #else #ifdef CONFIG_TICKET_LOCK_QUEUED */
> > +
> >  /*
> >   * Ticket locks are conceptually two parts, one indicating the current head of
> >   * the queue, and the other indicating the current tail. The lock is acquired
> > @@ -49,17 +64,15 @@
> >   */
> >  static __always_inline void __ticket_spin_lock(arch_spinlock_t *lock)
> >  {
> > -	register struct __raw_tickets inc = { .tail = 1 };
> > +	register struct __raw_tickets inc = { .tail = __TKT_SPIN_INC };
> >  
> >  	inc = xadd(&lock->tickets, inc);
> > -
> >  	for (;;) {
> > -		if (inc.head == inc.tail)
> > +		if (inc.head == inc.tail || tkt_spin_pass(lock, inc))
> >  			break;
> > -		cpu_relax();
> 
> Overheating the CPU are we ;-)
> 
> Keeping the cpu_relax() doesn't hurt, even when TICKET_LOCK_QUEUE is
> enabled. As the only latency to worry about is when tkt_spin_pass()
> returns true, where it breaks out of the loop anyway.
> 
> But if you really don't want the double call to cpu_relax(), we can
> probably remove the cpu_relax from tkt_spin_pass() and keep this one, or
> in the above tkt_spin_pass() where TICK_LOCK_QUEUED is not set, we can
> do:
> 
> static inline bool tkt_spin_pass(arch_spinlock_t *ap, struct
> __raw_tickets inc)
> {
> 	cpu_relax();
> 	return false;
> }
> 
> Honesty, I would say remove it from tkt_spin_pass() when returning
> false.

Sold!  I moved the cpu_relax() from tkt_spin_pass()'s false return to
the spin loop in __ticket_spin_lock().  Misguided attempt on my part to
minimize __ticket_spin_lock()'s size.

							Thanx, Paul

> -- Steve
> 
> 
> >  		inc.head = ACCESS_ONCE(lock->tickets.head);
> >  	}
> > -	barrier();		/* make sure nothing creeps before the lock is taken */
> > +	barrier(); /* Make sure nothing creeps in before the lock is taken. */
> >  }
> >  
> 
>