]> www.infradead.org Git - users/hch/xfs.git/commitdiff
mm/list_lru: fix false warning of negative counter
authorKairui Song <kasong@tencent.com>
Mon, 23 Dec 2024 15:09:07 +0000 (23:09 +0800)
committerAndrew Morton <akpm@linux-foundation.org>
Tue, 31 Dec 2024 01:59:10 +0000 (17:59 -0800)
commit 2788cf0c401c ("memcg: reparent list_lrus and free kmemcg_id on css
offline") removed sanity checks for the nr_items counter's value because
it implemented list_lru re-parenting in a way that will redirect
children's list_lru to the parent before re-parenting the items in
list_lru.  This will make item counter uncharging happen in the parent
while the item is still being held by the child.  As a result, the
parent's counter value may become negative.  This is acceptable because
re-parenting will sum up the children's counter values, and the parent's
counter will be fixed.

Later commit fb56fdf8b9a2 ("mm/list_lru: split the lock to per-cgroup
scope") reworked the re-parenting process, and removed the redirect.  So
it added the sanity check back, assuming that as long as items are still
in the children's list_lru, parent's counter will not be uncharged.

But that assumption is incorrect.  The xas_store in
memcg_reparent_list_lrus will set children's list_lru to NULL before
re-parenting the items, it redirects list_lru helpers to use parent's
list_lru just like before.  But still, it's not a problem as re-parenting
will fix the counter.

Therefore, remove this sanity check, but add a new check to ensure that
the counter won't go negative in a different way: the child's list_lru
being re-parented should never have a negative counter, since re-parenting
should occur in order and fixes counters.

Link: https://lkml.kernel.org/r/20241223150907.1591-1-ryncsn@gmail.com
Fixes: fb56fdf8b9a2 ("mm/list_lru: split the lock to per-cgroup scope")
Signed-off-by: Kairui Song <kasong@tencent.com>
Closes: https://lore.kernel.org/lkml/Z2Bz9t92Be9l1xqj@lappy/
Cc: Chengming Zhou <zhouchengming@bytedance.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Roman Gushchin <roman.gushchin@linux.dev>
Cc: Sasha Levin <sashal@kernel.org>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Yu Zhao <yuzhao@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
mm/list_lru.c

index f93ada6a207b13bd87775539a0b8d757b1070348..7d69434c70e0accf515ad8ee4ac6383167d2235b 100644 (file)
@@ -77,7 +77,6 @@ again:
                        spin_lock(&l->lock);
                nr_items = READ_ONCE(l->nr_items);
                if (likely(nr_items != LONG_MIN)) {
-                       WARN_ON(nr_items < 0);
                        rcu_read_unlock();
                        return l;
                }
@@ -450,6 +449,7 @@ static void memcg_reparent_list_lru_one(struct list_lru *lru, int nid,
 
        list_splice_init(&src->list, &dst->list);
        if (src->nr_items) {
+               WARN_ON(src->nr_items < 0);
                dst->nr_items += src->nr_items;
                set_shrinker_bit(dst_memcg, nid, lru_shrinker_id(lru));
        }