Re: [PATCH v2 1/2] mm/huge_memory.c: skip zero-page remapping when underused THP shrinker is disabled

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] mm: pointless underused THP splits in deferred_split_scan()
--

commit ebf8a156b1409b53d86c1c5c41f9ce1ca1b64265
Author: Nico Pache (Red Hat) <nico.pache@linux.dev>

mm/huge_memory.c: skip zero-page remapping when underused THP shrinker is disabled

This commit gates shared zeropage remapping in __folio_freeze_split_anon() on
split_underused_thp so subpages are not scanned for zero content during THP
splits when the underused shrinker is disabled.

> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index bd44d58e86845..2d56d1bff0eab 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -4104,7 +4104,8 @@ static int __folio_freeze_split_anon(struct folio *folio,
>  out_no_split:
>  	local_irq_enable();
>  	if (anon_vma) {
> -		if (!ret && !folio_is_device_private(folio))
> +		if (!ret && !folio_is_device_private(folio) &&
> +		    split_underused_thp)
>  			ttu_flags = TTU_USE_SHARED_ZEROPAGE;
>  		remap_anon_folio(folio, 1 << old_order, ttu_flags);
>  	}

[Severity: Medium]
What happens in deferred_split_scan() if shrink_underused is disabled at
runtime after anonymous folios have already been queued onto
deferred_split_lru?

Since split_underused_thp defaults to true at boot, anonymous folios
allocated via map_anon_folio_pmd_nopf() or collapsed via collapse_huge_page()
are queued via deferred_split_folio(folio, false) without setting the
partially mapped flag. If a user later sets shrink_underused to 0 (setting
split_underused_thp to false), those previously queued folios remain on
deferred_split_lru.

When memory reclaim subsequently runs deferred_split_scan(), the
!folio_test_partially_mapped(folio) path does not check split_underused_thp:

mm/huge_memory.c:deferred_split_scan() {
    ...
		if (!folio_test_partially_mapped(folio)) {
			if (folio_test_mlocked(folio))
				goto next;
			underused = thp_underused(folio);
			if (!underused)
				goto next;
		}
		if (!folio_trylock(folio))
			goto requeue;
		if (!split_folio(folio)) {
			did_split = true;
			if (underused)
				count_vm_event(THP_UNDERUSED_SPLIT_PAGE);
			split++;
		}
    ...
}

Because __folio_freeze_split_anon() now omits the shared zeropage flag when
split_underused_thp is false, remap_anon_folio() -> remove_migration_ptes()
skips try_to_map_unused_to_zeropage() and remaps all subpages (including
zero-filled ones) back to their split order-0 folios.

Doesn't this cause deferred_split_scan() to still scan subpages via
thp_underused(), shatter the huge page into order-0 pages, and increment the
underused split counters while freeing zero pages? Should
deferred_split_scan() also check split_underused_thp (for example, skipping
non-partially-mapped folios via goto next when !split_underused_thp)?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261025-b4-ksm_fixv2-v2-0-8b61ad0491f1@linux.dev?part=1