On 11/21/22 17:03, Mel Gorman wrote:
> On Mon, Nov 21, 2022 at 12:01:23PM +0000, Mel Gorman wrote:
>> On Fri, Nov 18, 2022 at 03:30:57PM +0100, Vlastimil Babka wrote:
>> > On 11/18/22 11:17, Mel Gorman wrote:
>> > AFAICS if this block was just "locked_zone = NULL;" then the existing code
>> > would do the right thing.
>> > Or maybe to have simpler code, just do batch_count++ here and
>> > make the relocking check do
>> > if (zone != locked_zone || batch_count == SWAP_CLUSTER_MAX)
>> >
>>
>> While I think you're right, I think it's a bit subtle, the batch reset would
>> need to move, rechecked within the "Different zone, different pcp lock."
>> block and it would be easy to forget exactly why it's structured like
>> that in the future. Rather than being a fix, it could be a standalone
>> patch so it would be obvious in git blame but I don't feel particularly
>> strongly about it.
>>
>
> Ok, less subtle than I initially thought but still deserving of a separate
> patch instead of being a fix. This?
Yeah, thanks!
> --8<--
> mm/page_alloc: Simplify locking during free_unref_page_list
>
> While freeing a large list, the zone lock will be released and reacquired
> to avoid long hold times since commit c24ad77d962c ("mm/page_alloc.c: avoid
> excessive IRQ disabled times in free_unref_page_list()"). As suggested
> by Vlastimil Babka, the lockrelease/reacquire logic can be simplified by
> reusing the logic that acquires a different lock when changing zones.
>
> Signed-off-by: Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx>
Reviewed-by: Vlastimil Babka <vbabka@xxxxxxx>
> ---
> mm/page_alloc.c | 25 +++++++++----------------
> 1 file changed, 9 insertions(+), 16 deletions(-)
>
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 445066617204..08e32daf0918 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -3518,13 +3518,19 @@ void free_unref_page_list(struct list_head *list)
> list_del(&page->lru);
> migratetype = get_pcppage_migratetype(page);
>
> - /* Different zone, different pcp lock. */
> - if (zone != locked_zone) {
> + /*
> + * Either different zone requiring a different pcp lock or
> + * excessive lock hold times when freeing a large list of
> + * pages.
> + */
> + if (zone != locked_zone || batch_count == SWAP_CLUSTER_MAX) {
> if (pcp) {
> pcp_spin_unlock(pcp);
> pcp_trylock_finish(UP_flags);
> }
>
> + batch_count = 0;
> +
> /*
> * trylock is necessary as pages may be getting freed
> * from IRQ or SoftIRQ context after an IO completion.
> @@ -3539,7 +3545,6 @@ void free_unref_page_list(struct list_head *list)
> continue;
> }
> locked_zone = zone;
> - batch_count = 0;
> }
>
> /*
> @@ -3551,19 +3556,7 @@ void free_unref_page_list(struct list_head *list)
>
> trace_mm_page_free_batched(page);
> free_unref_page_commit(zone, pcp, page, migratetype, 0);
> -
> - /*
> - * Guard against excessive lock hold times when freeing
> - * a large list of pages. Lock will be reacquired if
> - * necessary on the next iteration.
> - */
> - if (++batch_count == SWAP_CLUSTER_MAX) {
> - pcp_spin_unlock(pcp);
> - pcp_trylock_finish(UP_flags);
> - batch_count = 0;
> - pcp = NULL;
> - locked_zone = NULL;
> - }
> + batch_count++;
> }
>
> if (pcp) {
