Re: [PATCH v2 3/5] locking/qspinlock: Introduce CNA into the slow path of qspinlock

[Peter Zijlstra]

On Fri, Mar 29, 2019 at 11:20:04AM -0400, Alex Kogan wrote:
> +static inline void pass_mcs_lock(struct mcs_spinlock *node,
> +				 struct mcs_spinlock *next)
> +{
> +	struct mcs_spinlock *succ = NULL;
> +
> +	succ = find_successor(node);
> +
> +	if (succ) {
> +		arch_mcs_spin_unlock_contended(&succ->locked, node->locked);
> +	} else if (node->locked > 1) {
> +		/*
> +		 * If the secondary queue is not empty, pass the lock
> +		 * to the first node in that queue.
> +		 */
> +		succ = MCS_NODE(node->locked);
> +		succ->tail->next = next;
> +		arch_mcs_spin_unlock_contended(&succ->locked, 1);
> +	} else {
> +		/*
> +		 * Otherwise, pass the lock to the immediate successor
> +		 * in the main queue.
> +		 */
> +		arch_mcs_spin_unlock_contended(&next->locked, 1);
> +	}
> +}

Note that something like:

static inline void
pass_mcs_lock(struct mcs_spinlock *node, struct mcs_spinlock *next)
{
	struct mcs_spinlock *succ = NULL;
	uintptr_t *var = &next->locked;
	uintptr_t val = 1;

	succ = find_successor(node);
	if (succ) {
		var = &succ->locked;
		val = node->locked;
	} else if (node->locked > 1) {
		succ = MCS_NODE(node->locked);
		succ->tail->next = next; /* WRITE_ONCE() !?!? */
		var = &node->locked;
	}

	arch_mcs_spin_unlock_contended(var, val);
}

is shorter and generates much better code if
arch_mcs_spin_unlock_contended() is asm volatile ().