Re: [PATCH v3] mm/rmap: do not add fully unmapped large folio to deferred split list

[David Hildenbrand <david@redhat.com>]

On 25.04.24 09:27, Lance Yang wrote:
> On Thu, Apr 25, 2024 at 3:21 PM David Hildenbrand <david@redhat.com> wrote:
>>
>> On 25.04.24 05:45, Lance Yang wrote:
>>> Hey Zi,
>>>
>>> On Thu, Apr 25, 2024 at 6:46 AM Zi Yan <zi.yan@sent.com> wrote:
>>>>
>>>> From: Zi Yan <ziy@nvidia.com>
>>>>
>>>> In __folio_remove_rmap(), a large folio is added to deferred split list
>>>> if any page in a folio loses its final mapping. It is possible that
>>>> the folio is unmapped fully, but it is unnecessary to add the folio
>>>
>>> Agreed. If a folio is fully unmapped, then that's unnecessary to add
>>> to the deferred split list.
>>>
>>>> to deferred split list at all. Fix it by checking folio->_nr_pages_mapped
>>>> before adding a folio to deferred split list. If the folio is already
>>>> on the deferred split list, it will be skipped. This issue applies to
>>>> both PTE-mapped THP and mTHP.
>>>>
>>>> Commit 98046944a159 ("mm: huge_memory: add the missing
>>>> folio_test_pmd_mappable() for THP split statistics") tried to exclude
>>>> mTHP deferred split stats from THP_DEFERRED_SPLIT_PAGE, but it does not
>>>> fix the above issue. A fully unmapped PTE-mapped order-9 THP was still
>>>> added to deferred split list and counted as THP_DEFERRED_SPLIT_PAGE,
>>>> since nr is 512 (non zero), level is RMAP_LEVEL_PTE, and inside
>>>> deferred_split_folio() the order-9 folio is folio_test_pmd_mappable().
>>>> However, this miscount was present even earlier due to implementation,
>>>> since PTEs are unmapped individually and first PTE unmapping adds the THP
>>>> into the deferred split list.
>>>>
>>>> With commit b06dc281aa99 ("mm/rmap: introduce
>>>> folio_remove_rmap_[pte|ptes|pmd]()"), kernel is able to unmap PTE-mapped
>>>> folios in one shot without causing the miscount, hence this patch.
>>>>
>>>> Signed-off-by: Zi Yan <ziy@nvidia.com>
>>>> Reviewed-by: Yang Shi <shy828301@gmail.com>
>>>> ---
>>>>    mm/rmap.c | 7 ++++---
>>>>    1 file changed, 4 insertions(+), 3 deletions(-)
>>>>
>>>> diff --git a/mm/rmap.c b/mm/rmap.c
>>>> index a7913a454028..2809348add7b 100644
>>>> --- a/mm/rmap.c
>>>> +++ b/mm/rmap.c
>>>> @@ -1553,9 +1553,10 @@ static __always_inline void __folio_remove_rmap(struct folio *folio,
>>>>                    * page of the folio is unmapped and at least one page
>>>>                    * is still mapped.
>>>>                    */
>>>> -               if (folio_test_large(folio) && folio_test_anon(folio))
>>>> -                       if (level == RMAP_LEVEL_PTE || nr < nr_pmdmapped)
>>>> -                               deferred_split_folio(folio);
>>>> +               if (folio_test_large(folio) && folio_test_anon(folio) &&
>>>> +                   ((level == RMAP_LEVEL_PTE && atomic_read(mapped)) ||
>>>> +                    (level == RMAP_LEVEL_PMD && nr < nr_pmdmapped)))
>>>
>>> Perhaps we only need to check the mapcount?
>>>
>>> IIUC, if a large folio that was PMD/PTE mapped is fully unmapped here,
>>> then folio_mapcount() will return 0.
>>
>> See discussion on v1. folio_large_mapcount() would achieve the same
>> without another folio_test_large() check, but in the context of this
>> patch it doesn't really matter.
> 
> Got it. Thanks for pointing that out!
> I'll take a closer look at the discussion in v1.

Forgot to add: as long as the large mapcount patches are not upstream, 
folio_large_mapcount() would be expensive. So this patch can be added 
independent of the other stuff.

-- 
Cheers,

David / dhildenb