Re: [PATCH RFC 1/2] mm: collect the number of anon large folios

From: Barry Song
Date: Sun Aug 11 2024 - 05:22:43 EST


On Sun, Aug 11, 2024 at 8:51 PM David Hildenbrand <david@xxxxxxxxxx> wrote:
>
> > the correct place should be:
> >
> > @@ -1329,6 +1326,10 @@ static int migrate_folio_move(free_folio_t
> > put_new_folio, unsigned long private,
> > if (anon_vma)
> > put_anon_vma(anon_vma);
> > folio_unlock(src);
> > +
> > + if (folio_test_anon(src))
> > + mod_mthp_stat(folio_order(src), MTHP_STAT_NR_ANON, 1);
> > +
> > migrate_folio_done(src, reason);
> >
> > return rc;
> >
> > Without this modification in migration code, my tests fail, anon_num can
> > become negative.
>
> I was wondering if we should do it in __folio_migrate_mapping().
>
> There, we set newfolio->mapping.

Correct! To handle cases where the destination might not be migrated to but
could be put or freed, migrate_folio_undo_dst() will be called to release the
destination earlier. It's better to simply place the count where the
anon mapping
is set and cleared.

--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -423,6 +423,8 @@ static int __folio_migrate_mapping(struct
address_space *mapping,
/* No turning back from here */
newfolio->index = folio->index;
newfolio->mapping = folio->mapping;
+ if (folio_test_anon(folio) && folio_test_large(folio))
+ mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, 1);
if (folio_test_swapbacked(folio))
__folio_set_swapbacked(newfolio);

@@ -446,6 +448,9 @@ static int __folio_migrate_mapping(struct
address_space *mapping,
* no turning back from here.
*/
newfolio->index = folio->index;
+
+ if (folio_test_anon(folio) && folio_test_large(folio))
+ mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, 1);
newfolio->mapping = folio->mapping;
folio_ref_add(newfolio, nr); /* add cache reference */
if (folio_test_swapbacked(folio)) {


>
> --
> Cheers,
>
> David / dhildenb
>