Re: [PATCH v7 3/5] locking/qspinlock: Introduce CNA into the slow path of qspinlock

[Peter Zijlstra]

On Wed, Jan 22, 2020 at 06:04:56PM +0100, Peter Zijlstra wrote:
> +/*
> + * cna_splice_head -- splice the entire secondary queue onto the head of the
> + * primary queue.
> + */
> +static cna_splice_head(struct qspinlock *lock, u32 val,
> +		       struct mcs_spinlock *node, struct mcs_spinlock *next)
> +{
> +	struct mcs_spinlock *head_2nd, *tail_2nd;
> +
> +	tail_2nd = decode_tail(node->locked);
> +	head_2nd = tail_2nd->next;
> +
> +	if (lock) {

That should be: if (!next) {

> +		u32 new = ((struct cna_node *)tail_2nd)->encoded_tail | _Q_LOCKED_VAL;
> +		if (!atomic_try_cmpxchg_relaxed(&lock->val, &val, new))
> +			return NULL;
> +
> +		/*
> +		 * The moment we've linked the primary tail we can race with
> +		 * the WRITE_ONCE(prev->next, node) store from new waiters.
> +		 * That store must win.
> +		 */

And that still is a shit comment; I'll go try again.

> +		cmpxchg_relaxed(&tail_2nd->next, head_2nd, next);
> +	} else {
> +		tail_2nd->next = next;
> +	}
> +
> +	return head_2nd;
> +}