Re: [PATCH 04/12] mm: add support for async page locking

[Matthew Wilcox]

On Tue, May 26, 2020 at 01:51:15PM -0600, Jens Axboe wrote:
> +static int __wait_on_page_locked_async(struct page *page,
> +				       struct wait_page_queue *wait, bool set)
> +{
> +	struct wait_queue_head *q = page_waitqueue(page);
> +	int ret = 0;
> +
> +	wait->page = page;
> +	wait->bit_nr = PG_locked;
> +
> +	spin_lock_irq(&q->lock);
> +	if (set)
> +		ret = !trylock_page(page);
> +	else
> +		ret = PageLocked(page);
> +	if (ret) {
> +		__add_wait_queue_entry_tail(q, &wait->wait);
> +		SetPageWaiters(page);
> +		if (set)
> +			ret = !trylock_page(page);
> +		else
> +			ret = PageLocked(page);

Between the callers and this function, we actually look at PG_lock three
times; once in the caller, then after taking the spinlock, then after
adding ourselves to the waitqueue.  I understand the first and third, but
is it really worth doing the second test?  It feels unlikely to succeed
and only saves us setting PageWaiters.