Re: [PATCH v5 04/17] mm/memcg: add folio-based lruvec live helper

From: Kairui Song

Date: Sun Nov 22 2026 - 12:03:56 EST


On Sun, Nov 22, 2026 at 03:53:44PM +0800, Muchun Song wrote:
>
>
> On 2026/11/19 05:03, Kairui Song via B4 Relay wrote:
> > From: Kairui Song <kasong@xxxxxxxxxxx>
> >
> > Add a helper that resolves a live lruvec for a folio under RCU
> > without taking the lruvec lock. It takes a folio directly so the
> > lruvec lookup happens inside the RCU read-side critical section,
> > which a lruvec-based interface cannot guarantee. The first user is
> > the lockless folio_inc_lru_refs() path later in the series.
> >
> > Factor the ancestor walk out of lruvec_live_lock_irq() into
> > __lruvec_get_live(), so that both entry points share it, and spell
> > out why the walk is enough while only the RCU read lock is held:
> > reparenting of a memcg that starts dying waits for a grace period,
> > so it cannot race with the caller until the lock is released.
> >
> > Move memcg_is_dying() up in memcontrol.h, next to the other memcg state
> > predicates, and use it for the walk rather than repeating the open coded
> > css_is_dying() test a few definitions away from where it is explained.
> >
> > No functional change for existing callers.
> >
> > Tested-by: Zhao Zhenzhuo <zhaozhengzhuo@xxxxxxxxxxxxx>
> > Signed-off-by: Kairui Song <kasong@xxxxxxxxxxx>
>
> Reviewed-by: Muchun Song (ByteDance) <muchun.song@xxxxxxxxx>
>
> Some nits bellow:
>
> > ---
> > include/linux/memcontrol.h | 71 ++++++++++++++++++++++++++++++++++++----------
> > 1 file changed, 56 insertions(+), 15 deletions(-)
> >
> > diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
> > index 980358bc6f49..441a5d57e723 100644
> > --- a/include/linux/memcontrol.h
> > +++ b/include/linux/memcontrol.h
> > @@ -571,6 +571,11 @@ static inline bool mem_cgroup_disabled(void)
> > return !cgroup_subsys_enabled(memory_cgrp_subsys);
> > }
> > +static inline bool memcg_is_dying(struct mem_cgroup *memcg)
>
> Let's make it const for @memcg along the way.

Good idea, but css_is_dying isn't constified, maybe we can do it later I think.

> > @@ -1510,28 +1520,41 @@ static inline void lruvec_lock_irq(struct lruvec *lruvec)
> > spin_lock_irq(&lruvec->lru_lock);
> > }
> > -static inline struct lruvec *lruvec_live_lock_irq(struct lruvec *lruvec)
> > +/*
> > + * Reparenting moves a dying memcg's LRU state to its parent at any time,
> > + * so anything added to that LRU afterwards would be lost. This helper
> > + * walks up to the nearest live ancestor while the caller must hold the
> > + * RCU read lock: reparenting of a memcg that starts dying waits for a
> > + * grace period, so it cannot race with the caller until the lock is
> > + * released.
> > + */
> > +static inline struct lruvec *__lruvec_get_live(struct lruvec *lruvec)
>
> Cound @lruvec be const as well?

The argument is fine, but tricky for the return value: when nothing is dying it
hands back @lruvec unchanged, so a const argument means a const return (which
isn't right) or a cast dropping it again.

> > @@ -1561,6 +1584,34 @@ static inline bool folio_matches_lruvec(const struct folio *folio,
> > lruvec_memcg(lruvec) == folio_memcg(folio);
> > }
> > +/**
> > + * folio_lruvec_live_get - get a live lruvec for a folio under RCU
> > + * @folio: the folio
> > + *
> > + * Same as folio_lruvec(), except that a folio whose memcg is dying is
> > + * accounted to the nearest live ancestor, which is where reparenting
> > + * would put it anyway.
> > + *
> > + * Return: the live lruvec, with rcu_read_lock() held. Pair it with
> > + * folio_lruvec_live_put().
> > + */
> > +static inline struct lruvec *folio_lruvec_live_get(struct folio *folio)
>
> const folio?

This is doable, will do!

Thanks for the review!