Re: [PATCH v6 4/4] mm: thp: reparent the split queue during memcg offline

[Muchun Song <muchun.song@linux.dev>]

> On Jan 14, 2026, at 14:25, Qi Zheng <qi.zheng@linux.dev> wrote:
> 
> 
> 
> On 1/14/26 1:36 PM, Harry Yoo wrote:
>> On Tue, Jan 13, 2026 at 02:00:43PM -0800, Chris Mason wrote:
>>> On Mon, 10 Nov 2025 16:17:58 +0800 Qi Zheng <qi.zheng@linux.dev> wrote:
>>> 
>>>> From: Qi Zheng <zhengqi.arch@bytedance.com>
>>>> 
>>>> Similar to list_lru, the split queue is relatively independent and does
>>>> not need to be reparented along with objcg and LRU folios (holding
>>>> objcg lock and lru lock). So let's apply the similar mechanism as list_lru
>>>> to reparent the split queue separately when memcg is offine.
>>>> 
>>> 
>>> Hi everyone,
>>> 
>>> I'm running all of the MM commits since v6.18 through the AI patch review
>>> tools, and this one was flagged.  I'm much less confident in this report,
>>> AI tends to get these races wrong:
>>> 
>>> [AI review output]
>>> 
>>>> diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
>>>> index b0c6a4635c67b..cc6db20d7dca0 100644
>>>> --- a/include/linux/memcontrol.h
>>>> +++ b/include/linux/memcontrol.h
>>>> @@ -1775,6 +1775,12 @@ static inline void count_objcg_events(struct obj_cgroup *objcg,
>>>>  bool mem_cgroup_node_allowed(struct mem_cgroup *memcg, int nid);
>>>> 
>>>>  void mem_cgroup_show_protected_memory(struct mem_cgroup *memcg);
>>>> +
>>>> +static inline bool memcg_is_dying(struct mem_cgroup *memcg)
>>>> +{
>>>> + return memcg ? css_is_dying(&memcg->css) : false;
>>>> +}
>>> 
>>> [ ... ]
>>> 
>>>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
>>>> index 13684e5376e82..d17d3810a882c 100644
>>> 
>>> [ ... ]
>>> 
>>>> @@ -1129,8 +1140,14 @@ split_queue_lock_irqsave(int nid, struct mem_cgroup *memcg, unsigned long *flags
>>>>  {
>>>>   struct deferred_split *queue;
>>>> 
>>>> +retry:
>>>>   queue = memcg_split_queue(nid, memcg);
>>>>   spin_lock_irqsave(&queue->split_queue_lock, *flags);
>>>> + if (unlikely(memcg_is_dying(memcg))) {
>>>> + spin_unlock_irqrestore(&queue->split_queue_lock, *flags);
>>>> + memcg = parent_mem_cgroup(memcg);
>>>> + goto retry;
>>>> + }
>>>> 
>>>>   return queue;
>>>>  }
>>> 
>>> There appears to be a race window that can cause split_queue_len counter
>>> mismatch. Between the time CSS_DYING is set (in kill_css()) and when
>>> reparent_deferred_split_queue() actually runs (asynchronously in
>>> css_offline), folios can be unqueued via __folio_unqueue_deferred_split().
>>> 
>>> In this scenario:
>>>   __folio_unqueue_deferred_split()
>>>     -> folio_split_queue_lock_irqsave()
>>>        -> split_queue_lock_irqsave() sees memcg_is_dying() == true
>>>        -> returns parent's queue (locked)
>>>     -> ds_queue->split_queue_len-- decrements parent's counter
>>>     -> list_del_init() removes folio from child's list
>>> 
>>> The folio is correctly removed from the child's list (list_del_init is
>>> self-referential), but the wrong queue's counter is decremented.
>> Good point. Sounds pretty possible to me?
>> I don't think there's anything that prevents it from unqueued
>> before it's reparented.
>>> If the parent queue was empty (split_queue_len == 0), this would cause an
>>> unsigned long underflow to ULONG_MAX.
>> Although the accounting mismatch will only persist until
>> reparent_deferred_split_queue() reparents the deferred split queue.
> 
> This period is very short, and the only thing affected should be
> deferred_split_count(), but it does not cause a system error.
> 
> So I think maybe we can leave it unrepaired.

I didn’t look closely at the specific issue, but based on Qi’s point,
we can actually take a cue from list_lru_count_one and provide a similar
workaround to keep deferred_split_count from returning an enormous
value—see commit 41d17431df4aa.

> 
>> Ideally this should be fixed by checking if the folio has been
>> reparented after acquiring the split queue lock, but since we don't reparent
>> LRU pages yet ... do we need a band-aid before then?
>> Do we want to have is_dying property in the split queue as it was in v2? [1]
>> [1] https://lore.kernel.org/linux-mm/55370bda7b2df617033ac12116c1712144bb7591.1758618527.git.zhengqi.arch@bytedance.com
>>> Could this be addressed by checking that the folio is actually in the
>>> queue being unlocked, perhaps by verifying the list head matches before
>>> decrementing split_queue_len?
>>> [ ... ]
>>> 
>>>> @@ -3920,6 +3920,7 @@ static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
>>>>   zswap_memcg_offline_cleanup(memcg);
>>>> 
>>>>   memcg_offline_kmem(memcg);
>>>> + reparent_deferred_split_queue(memcg);
>>> 
>>> The race window exists because CSS_DYING is set in kill_css() before
>>> css_offline() callbacks run asynchronously. The comment in split_queue_lock
>>> mentions that THPs will be "hidden from the shrinker side" during this
>>> period, but does not address the counter accounting issue when folios
>>> are unqueued.
>