On Wed, 2017-04-05 at 15:10 +0800, Huang, Ying wrote:
> To solve the issue, the per-CPU buffer is sorted according to the
> swap
> device before freeing the swap entries.  Test shows that the time
> spent by swapcache_free_entries() could be reduced after the patch.

That makes a lot of sense.

> @@ -1075,6 +1083,8 @@ void swapcache_free_entries(swp_entry_t
> *entries, int n)
>  
>  	prev = NULL;
>  	p = NULL;
> +	if (nr_swapfiles > 1)
> +		sort(entries, n, sizeof(entries[0]), swp_entry_cmp,
> NULL);

But it really wants a comment in the code, so people
reading the code a few years from now can see why
we are sorting things we are about to free.

Maybe something like:
        /* Sort swap entries by swap device, so each lock is only taken
once. */

>  	for (i = 0; i < n; ++i) {
>  		p = swap_info_get_cont(entries[i], prev);
>  		if (p)
-- 
All rights reversed