Re: [PATCH v2 4/4] mm: thp: reparent the split queue during memcg offline

[Harry Yoo <harry.yoo@oracle.com>]

On Tue, Sep 23, 2025 at 05:16:25PM +0800, Qi Zheng wrote:
> In the future, we will reparent LRU folios during memcg offline to
> eliminate dying memory cgroups, which requires reparenting the split queue
> to its parent.
> 
> 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 same mechanism as list_lru
> to reparent the split queue separately when memcg is offine.
>
> Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
> ---
>  include/linux/huge_mm.h |  2 ++
>  include/linux/mmzone.h  |  1 +
>  mm/huge_memory.c        | 39 +++++++++++++++++++++++++++++++++++++++
>  mm/memcontrol.c         |  1 +
>  mm/mm_init.c            |  1 +
>  5 files changed, 44 insertions(+)
> 
> diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h
> index f327d62fc9852..a0d4b751974d2 100644
> --- a/include/linux/huge_mm.h
> +++ b/include/linux/huge_mm.h
> @@ -417,6 +417,7 @@ static inline int split_huge_page(struct page *page)
>  	return split_huge_page_to_list_to_order(page, NULL, ret);
>  }
>  void deferred_split_folio(struct folio *folio, bool partially_mapped);
> +void reparent_deferred_split_queue(struct mem_cgroup *memcg);
>  
>  void __split_huge_pmd(struct vm_area_struct *vma, pmd_t *pmd,
>  		unsigned long address, bool freeze);
> @@ -611,6 +612,7 @@ static inline int try_folio_split(struct folio *folio, struct page *page,
>  }
>  
>  static inline void deferred_split_folio(struct folio *folio, bool partially_mapped) {}
> +static inline void reparent_deferred_split_queue(struct mem_cgroup *memcg) {}
>  #define split_huge_pmd(__vma, __pmd, __address)	\
>  	do { } while (0)
>  
> diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
> index 7fb7331c57250..f3eb81fee056a 100644
> --- a/include/linux/mmzone.h
> +++ b/include/linux/mmzone.h
> @@ -1346,6 +1346,7 @@ struct deferred_split {
>  	spinlock_t split_queue_lock;
>  	struct list_head split_queue;
>  	unsigned long split_queue_len;
> +	bool is_dying;
>  };
>  #endif

The scheme in Muchun's version was:

retry:
queue = folio_split_queue(folio);
spin_lock(&queue->split_queue_lock);
if (folio_memcg(folio) != folio_split_queue_memcg(folio, queue)) {
    /* split queue was reparented, retry */
    spin_unlock(&queue->split_queue_lock);
    goto retry;
}
/* now we have a stable mapping between the folio and the split queue */
spin_unlock(&queue->split_queue_lock);

Oh, I see. We can't use this scheme yet because we don't reparent LRU
folios. (I was wondering why we're adding is_dying property)

> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index 48b51e6230a67..de7806f759cba 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -1094,9 +1094,15 @@ static struct deferred_split *folio_split_queue_lock(struct folio *folio)
>  	struct deferred_split *queue;


For now it's safe to not call rcu_read_lock() here because memcgs won't
disappear under us as long as there are folios to split (we don't reparent
LRU folios), right?

>  	memcg = folio_memcg(folio);
> +retry:
>  	queue = memcg ? &memcg->deferred_split_queue :
>  			&NODE_DATA(folio_nid(folio))->deferred_split_queue;
>  	spin_lock(&queue->split_queue_lock);
> +	if (unlikely(queue->is_dying == true)) {
> +		spin_unlock(&queue->split_queue_lock);
> +		memcg = parent_mem_cgroup(memcg);
> +		goto retry;
> +	}
>  	return queue;
>  }

-- 
Cheers,
Harry / Hyeonggon