Re: [PATCH RFC ticketlock] Auto-queued ticketlock

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

On Mon, Jun 10, 2013 at 02:35:06PM -0700, Eric Dumazet wrote:
> On Sun, 2013-06-09 at 12:36 -0700, Paul E. McKenney wrote:
> > Breaking up locks is better than implementing high-contention locks, but
> > if we must have high-contention locks, why not make them automatically
> > switch between light-weight ticket locks at low contention and queued
> > locks at high contention?
> > 
> > This commit therefore allows ticket locks to automatically switch between
> > pure ticketlock and queued-lock operation as needed.  If too many CPUs
> > are spinning on a given ticket lock, a queue structure will be allocated
> > and the lock will switch to queued-lock operation.  When the lock becomes
> > free, it will switch back into ticketlock operation.  The low-order bit
> > of the head counter is used to indicate that the lock is in queued mode,
> > which forces an unconditional mismatch between the head and tail counters.
> > This approach means that the common-case code path under conditions of
> > low contention is very nearly that of a plain ticket lock.
> > 
> > A fixed number of queueing structures is statically allocated in an
> > array.  The ticket-lock address is used to hash into an initial element,
> > but if that element is already in use, it moves to the next element.  If
> > the entire array is already in use, continue to spin in ticket mode.
> > 
> > This has been only lightly tested in the kernel, though a userspace
> > implementation has survived substantial testing.
> > 
> > Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
> 
> This looks a great idea ;)

Glad you like it!  Hopefully workloads like it as well.  ;-)

> > +
> > +static __always_inline void __ticket_spin_unlock(arch_spinlock_t *lock)
> > +{
> > +	__ticket_t head = 2;
> > +
> > +	head = xadd(&lock->tickets.head, 2);
> 
> 	head = xadd(&lock->tickets.head, head);

Yikes!  Good catch, fixed.

> > +	if (head & 0x1)
> > +		tkt_q_do_wake(lock);
> > +}
> > +#endif /* #else #ifndef CONFIG_TICKET_LOCK_QUEUED */
> 
> > + */
> > +void tkt_q_do_wake(arch_spinlock_t *asp)
> > +{
> > +	struct tkt_q_head *tqhp;
> > +	struct tkt_q *tqp;
> > +
> > +	/* If the queue is still being set up, wait for it. */
> > +	while ((tqhp = tkt_q_find_head(asp)) == NULL)
> > +		cpu_relax();
> > +
> > +	for (;;) {
> > +
> > +		/* Find the first queue element. */
> > +		tqp = ACCESS_ONCE(tqhp->spin);
> > +		if (tqp != NULL)
> > +			break;  /* Element exists, hand off lock. */
> > +		if (tkt_q_try_unqueue(asp, tqhp))
> > +			return; /* No element, successfully removed queue. */
> > +		cpu_relax();
> > +	}
> > +	if (ACCESS_ONCE(tqhp->head_tkt) != -1)
> > +		ACCESS_ONCE(tqhp->head_tkt) = -1;
> > +	smp_mb(); /* Order pointer fetch and assignment against handoff. */
> > +	ACCESS_ONCE(tqp->cpu) = -1;
> > +}
> 
> EXPORT_SYMBOL(tkt_q_do_wake) ?

Good point, just in case we want to use spinlocks in modules.  ;-)
Same for tkt_spin_pass(), I guess.

> Hmm, unfortunately I lack time this week to fully read the patch !

I suspect that there is very little danger of this patch going in this
week, so you should have some additional time.  ;-)

							Thanx, Paul