Re: [PATCH v3] mm: mglru: fix stale batch updates after memcg reparenting

[Harry Yoo <harry@kernel.org>]

[-- Attachment #1.1: Type: text/plain, Size: 5175 bytes --]



On 6/26/26 11:27 AM, Qi Zheng wrote:
> Hi Johannes,
> 
> On 6/26/26 2:41 AM, Johannes Weiner wrote:
>> On Thu, Jun 25, 2026 at 11:15:54PM +0800, Qi Zheng wrote:
>>> From: Qi Zheng <zhengqi.arch@bytedance.com>
>>>
>>> The mglru page table walker batches per-generation size deltas in
>>> walk->nr_pages while walking page tables without holding the lruvec
>>> lock.
>>> The reset_batch_size() later folds those deltas into walk->lruvec under
>>> the lruvec lock.
>>>
>>> The page table walker can run concurrently with the memcg reparenting
>>> path
>>> as follows:
>>>
>>> CPU0                           CPU1
>>> ====                           ====
>>>
>>> walk_mm
>>> --> walk_page_range
>>>      --> update_batch_size
>>>          --> walk->nr_pages += delta
>>>
>>>                                mem_cgroup_css_offline
>>>                                --> memcg_reparent_objcgs
>>>                                    --> lock lruvec
>>>                                        lru_gen_reparent_memcg
>>>                                        --> reparent child folios to
>>> parent
>>>                                        unlock lruvec
>>>
>>>      lock lruvec
>>>      reset_batch_size
>>>      --> child lrugen->nr_pages += delta
>>>
>>> This will trigger the following warning in lru_gen_exit_memcg():
>>>
>>>     VM_WARN_ON_ONCE(memchr_inv(lruvec->lrugen.nr_pages, 0,
>>>                    sizeof(lruvec->lrugen.nr_pages)));
>>>
>>> And the user-visible impact of underestimated nr_pages in MGLRU was
>>> premature OOMs because MGLRU does not try to reclaim memory when
>>> nr_pages
>>> reaches zero, but there are still more pages.
>>>
>>> To fix it, make reset_batch_size() check CSS_DYING under RCU before
>>> flushing the pending batch. A non-dying memcg keeps the original lruvec
>>> stable against RCU-delayed offlining; a dying memcg redirects the deltas
>>> to the first non-dying ancestor.
>>>
>>> Reported-by: Peiyang He <peiyang_he@smail.nju.edu.cn>
>>> Closes: https://lore.kernel.org/all/5A9E929D82717101+12fcf643-
>>> efb8-4b9a-a53a-1e28cc894f0b@smail.nju.edu.cn
>>> Fixes: f304652609ea ("mm: vmscan: prepare for reparenting MGLRU folios")
>>> Cc: <stable@vger.kernel.org>
>>> Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
>>> ---
>>> Changes in v3:
>>>   - re-implement lock_batch_lruvec() by checking CSS_DYING under the
>>> RCU lock
>>>     (suggested by Harry)
>>>   - update the commit message (suggested by Harry)
>>>   - temporarily drop the previous Reviewed-by tags
>>>     (since the sync method has changed)
>>>   - rebase onto the next-20260624
>>>
>>> Changes in v2:
>>>   - update the commit message (pointed by Barry)
>>>   - collect Reviewed-by
>>>
>>>   mm/vmscan.c | 45 ++++++++++++++++++++++++++++++++++++++-------
>>>   1 file changed, 38 insertions(+), 7 deletions(-)
>>>
>>> diff --git a/mm/vmscan.c b/mm/vmscan.c
>>> index 35c3bb15ae96..1ec8c23c72b9 100644
>>> --- a/mm/vmscan.c
>>> +++ b/mm/vmscan.c
>>> @@ -3262,10 +3262,44 @@ static void update_batch_size(struct
>>> lru_gen_mm_walk *walk, struct folio *folio,
>>>       walk->nr_pages[new_gen][type][zone] += delta;
>>>   }
>>>   +#ifdef CONFIG_MEMCG
>>> +static struct lruvec *lock_batch_lruvec(struct lruvec *lruvec)
>>> +{
>>> +    struct pglist_data *pgdat = lruvec_pgdat(lruvec);
>>> +    struct mem_cgroup *memcg = lruvec_memcg(lruvec);
>>> +
>>> +    rcu_read_lock();
>>
>> Where is this unlocked?
> 
> The lruvec_unlock_irq() in reset_batch_size() will handle the unlocking.
> 
>>
>>> +    /*
>>> +     * The memcg can be NULL when the memory controller is disabled.
>>> +     * Otherwise, the caller keeps the memcg owning @lruvec alive.
>>> +     */
>>> +    if (!memcg || !css_is_dying(&memcg->css))
>>> +        goto lock;
>>> +
>>> +    do {
>>> +        memcg = parent_mem_cgroup(memcg);
>>> +    } while (memcg && css_is_dying(&memcg->css));
>>> +    lruvec = mem_cgroup_lruvec(memcg, pgdat);
>>
>>     while (unlikely(memcg && css_is_dying(&memcg->css))) {
>>         memcg = parent_mem_cgroup(memcg);
>>         lruvec = mem_cgroup_lruvec(memcg, pgdat);
> 
> There is no need to acquire the lruvec before finding the first
> non-dying memcg.

struct pglist_data *pgdat = lruvec_pgdat(lruvec);
struct mem_cgroup *memcg = lruvec_memcg(lruvec);

rcu_read_lock()

while (unlikely(memcg_is_dying(memcg)))
        memcg = parent_mem_cgroup(memcg);

lruvec = mem_cgroup_lruvec(memcg, pgdat);
spin_lock_irq(&lruvec->lru_lock);

return lruvec;

should work?

if the memory controller is disabled, it's equivalent to:

rcu_read_lock();
spin_lock_irq(&lruvec->lru_lock);
return lruvec;

-- 
Cheers,
Harry / Hyeonggon


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 228 bytes --]