Re: [PATCH v9 2/8] mm/huge_memory: add two new (not yet used) functions for folio_split()

[Zi Yan <ziy@nvidia.com>]

On 27 Feb 2025, at 10:14, Matthew Wilcox wrote:

> On Thu, Feb 27, 2025 at 05:55:43AM +0000, Matthew Wilcox wrote:
>> On Wed, Feb 26, 2025 at 04:00:25PM -0500, Zi Yan wrote:
>>> +static int __split_unmapped_folio(struct folio *folio, int new_order,
>>> +		struct page *split_at, struct page *lock_at,
>>> +		struct list_head *list, pgoff_t end,
>>> +		struct xa_state *xas, struct address_space *mapping,
>>> +		bool uniform_split)
>>> +{
>> [...]
>>> +		/* complete memcg works before add pages to LRU */
>>> +		split_page_memcg(&folio->page, old_order, split_order);
>>> +		split_page_owner(&folio->page, old_order, split_order);
>>> +		pgalloc_tag_split(folio, old_order, split_order);
>>
>> At least split_page_memcg() needs to become aware of 'uniform_split'.
>>
>>         if (folio_memcg_kmem(folio))
>>                 obj_cgroup_get_many(__folio_objcg(folio), old_nr / new_nr - 1);
>>
>> If we're doing uniform_split, that calculation should be
>> 	old_order - new_order - 1
>
> umm, old_order - new_order.  Anyway, here's a patch I've done on top of
> your work, but it probably needs to be massaged slightly and placed
> before your work?

Wait. uniform_split is the existing splitting one order-9 to 512 order-0
approach, so split_page_memcg() still works. For !uniform_split,
split_page_memcg() is called multiple times,
each time old_order = new_order + 1, so what split_page_memcg() does
is:
1. two order-8 folios get their memcg, and ref count is increased by 1;
2. one of the order-8s is split into two order-7, each of which gets
their memcg, and ref count is increased by 1;
…

8. one of the order-1s is split into two order-0, each of which gets
their memcg, and ref count is increased by 1.

At the end, the refcount is increased by old_order - new_order like
you described above. Let me know if it makes sense to you.


>
> From 190e13ed77e562eb59fa1fa4bfefdefe5d0416ed Mon Sep 17 00:00:00 2001
> From: "Matthew Wilcox (Oracle)" <willy@infradead.org>
> Date: Mon, 28 Oct 2024 16:23:30 -0400
> Subject: [PATCH] mm: Separate folio_split_memcg() from split_page_memcg()
>
> Folios always use memcg_data to refer to the mem_cgroup while pages
> allocated with GFP_ACCOUNT have a pointer to the obj_cgroup.  Since the
> caller already knows what it has, split the function into two and then
> we don't need to check.
>
> Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
> ---
>  include/linux/memcontrol.h |  7 +++++++
>  mm/huge_memory.c           |  6 ++++--
>  mm/memcontrol.c            | 18 +++++++++++++++---
>  3 files changed, 26 insertions(+), 5 deletions(-)
>
> diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
> index 57664e2a8fb7..155c3f81f4df 100644
> --- a/include/linux/memcontrol.h
> +++ b/include/linux/memcontrol.h
> @@ -1039,6 +1039,8 @@ static inline void memcg_memory_event_mm(struct mm_struct *mm,
>  }
>
>  void split_page_memcg(struct page *head, int old_order, int new_order);
> +void folio_split_memcg(struct folio *folio, unsigned old_order,
> +		unsigned new_order, bool uniform_split);
>
>  static inline u64 cgroup_id_from_mm(struct mm_struct *mm)
>  {
> @@ -1463,6 +1465,11 @@ static inline void split_page_memcg(struct page *head, int old_order, int new_or
>  {
>  }
>
> +static inline void folio_split_memcg(struct folio *folio, unsigned old_order,
> +		unsigned new_order, bool uniform)
> +{
> +}
> +
>  static inline u64 cgroup_id_from_mm(struct mm_struct *mm)
>  {
>  	return 0;
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index 1e45064046a0..75fa9c9d9ec9 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -3401,6 +3401,9 @@ static void __split_folio_to_order(struct folio *folio, int new_order)
>  			folio_set_young(new_folio);
>  		if (folio_test_idle(folio))
>  			folio_set_idle(new_folio);
> +#ifdef CONFIG_MEMCG
> +		new_folio->memcg_data = folio->memcg_data;
> +#endif
>
>  		folio_xchg_last_cpupid(new_folio, folio_last_cpupid(folio));
>  	}
> @@ -3529,8 +3532,7 @@ static int __split_unmapped_folio(struct folio *folio, int new_order,
>  			}
>  		}
>
> -		/* complete memcg works before add pages to LRU */
> -		split_page_memcg(&folio->page, old_order, split_order);
> +		folio_split_memcg(folio, old_order, split_order, uniform_split);
>  		split_page_owner(&folio->page, old_order, split_order);
>  		pgalloc_tag_split(folio, old_order, split_order);
>
> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> index 16f3bdbd37d8..c2d41e1337cb 100644
> --- a/mm/memcontrol.c
> +++ b/mm/memcontrol.c
> @@ -3064,10 +3064,22 @@ void split_page_memcg(struct page *head, int old_order, int new_order)
>  	for (i = new_nr; i < old_nr; i += new_nr)
>  		folio_page(folio, i)->memcg_data = folio->memcg_data;
>
> -	if (folio_memcg_kmem(folio))
> -		obj_cgroup_get_many(__folio_objcg(folio), old_nr / new_nr - 1);
> +	obj_cgroup_get_many(__folio_objcg(folio), old_nr / new_nr - 1);
> +}
> +
> +void folio_split_memcg(struct folio *folio, unsigned old_order,
> +		unsigned new_order, bool uniform_split)
> +{
> +	unsigned new_refs;
> +
> +	if (mem_cgroup_disabled() || !folio_memcg_charged(folio))
> +		return;
> +
> +	if (uniform_split)
> +		new_refs = (1 << (old_order - new_order)) - 1;
>  	else
> -		css_get_many(&folio_memcg(folio)->css, old_nr / new_nr - 1);
> +		new_refs = old_order - new_order;
> +	css_get_many(&__folio_memcg(folio)->css, new_refs);
>  }
>
>  unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
> -- 
> 2.47.2


Best Regards,
Yan, Zi