Re: [RFC PATCH 5/5] Revert "mm/migrate: __unmap_and_move() push good newpage to LRU"

[Yosry Ahmed]

On Sun, Jun 18, 2023 at 7:00 PM Huang, Ying <ying.huang@xxxxxxxxx> wrote:
>
> Hi, Yosry,
>
> Yosry Ahmed <yosryahmed@xxxxxxxxxx> writes:
>
> > This reverts commit c3096e6782b733158bf34f6bbb4567808d4e0740.
> >
> > That commit made sure we immediately add the new page to the LRU before
> > remove_migration_ptes() is called in migrate_move_folio() (used to be
> > __unmap_and_move() back then), such that the rmap walk will rebuild the
> > correct mlock_count for the page again. This was needed because the
> > mlock_count was lost when the page is isolated. This is no longer the
> > case since mlock_count no longer overlays page->lru.
> >
> > Revert the commit (the code was foliated afterward the commit, so the
> > revert is updated as such).
> >
> > Signed-off-by: Yosry Ahmed <yosryahmed@xxxxxxxxxx>
> > ---
> >  mm/migrate.c | 24 +++++++++---------------
> >  1 file changed, 9 insertions(+), 15 deletions(-)
> >
> > diff --git a/mm/migrate.c b/mm/migrate.c
> > index 01cac26a3127..68f693731865 100644
> > --- a/mm/migrate.c
> > +++ b/mm/migrate.c
> > @@ -1279,19 +1279,6 @@ static int migrate_folio_move(free_page_t put_new_page, unsigned long private,
> >       if (unlikely(!is_lru))
> >               goto out_unlock_both;
>
> The patch itself looks good to me!  Thanks!
>
> Reviewed-by: "Huang, Ying" <ying.huang@xxxxxxxxx>

Thanks for taking a look!

>
> And, it seems that we can remove the above 2 lines and "out_unlock_both"
> label now.  That can make the code simpler a little.  Right?

I am not familiar with this code. If we remove the above condition
then pages that have is_lru == 0 (i.e __PageMovable(src) is true) and
page_was_mapped == 1 will call remove_migration_ptes(). This wouldn't
happen without removing the above 2 lines. If this combination is
impossible (is_lru == 0 && page_was_mapped == 1), then yeah we can
remove the above condition.

It looks like __SetPageMovable() is only called by zsmalloc, z3fold,
and balloon_page_insert(). The former 2 will never have those pages
mapped into userspace. I am not familiar with balloon_page_insert(),
but my gut feeling is that those are pages used by the driver and are
also not mapped into userspace.

So I guess we can just remove the condition, but a confirmation for
the above would be reassuring :)

>
> Best Regards,
> Huang, Ying
>
> > -     /*
> > -      * When successful, push dst to LRU immediately: so that if it
> > -      * turns out to be an mlocked page, remove_migration_ptes() will
> > -      * automatically build up the correct dst->mlock_count for it.
> > -      *
> > -      * We would like to do something similar for the old page, when
> > -      * unsuccessful, and other cases when a page has been temporarily
> > -      * isolated from the unevictable LRU: but this case is the easiest.
> > -      */
> > -     folio_add_lru(dst);
> > -     if (page_was_mapped)
> > -             lru_add_drain();
> > -
> >       if (page_was_mapped)
> >               remove_migration_ptes(src, dst, false);
> >
> > @@ -1301,9 +1288,16 @@ static int migrate_folio_move(free_page_t put_new_page, unsigned long private,
> >       /*
> >        * If migration is successful, decrease refcount of dst,
> >        * which will not free the page because new page owner increased
> > -      * refcounter.
> > +      * refcounter. As well, if it is LRU folio, add the folio to LRU
> > +      * list in here. Use the old state of the isolated source folio to
> > +      * determine if we migrated a LRU folio. dst was already unlocked
> > +      * and possibly modified by its owner - don't rely on the folio
> > +      * state.
> >        */
> > -     folio_put(dst);
> > +     if (unlikely(!is_lru))
> > +             folio_put(dst);
> > +     else
> > +             folio_putback_lru(dst);
> >
> >       /*
> >        * A folio that has been migrated has all references removed
>