Re: [PATCH 1/5] mm: khugepaged: expand the is_refcount_suitable() to support file folios

[Baolin Wang <baolin.wang@linux.alibaba.com>]

On 2024/8/19 16:36, David Hildenbrand wrote:
> On 19.08.24 10:14, Baolin Wang wrote:
>> Expand the is_refcount_suitable() to support reference checks for file 
>> folios,
>> as preparation for supporting shmem mTHP collapse.
>>
>> Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
>> ---
>>   mm/khugepaged.c | 11 ++++++++---
>>   1 file changed, 8 insertions(+), 3 deletions(-)
>>
>> diff --git a/mm/khugepaged.c b/mm/khugepaged.c
>> index cdd1d8655a76..f11b4f172e61 100644
>> --- a/mm/khugepaged.c
>> +++ b/mm/khugepaged.c
>> @@ -549,8 +549,14 @@ static bool is_refcount_suitable(struct folio 
>> *folio)
>>       int expected_refcount;
>>       expected_refcount = folio_mapcount(folio);
>> -    if (folio_test_swapcache(folio))
>> +    if (folio_test_anon(folio)) {
>> +        expected_refcount += folio_test_swapcache(folio) ?
>> +                    folio_nr_pages(folio) : 0;
>> +    } else {
>>           expected_refcount += folio_nr_pages(folio);
>> +        if (folio_test_private(folio))
>> +            expected_refcount++;
>> +    }
> 
> Alternatively, a bit neater
> 
> if (!folio_test_anon(folio) || folio_test_swapcache(folio))
>      expected_refcount += folio_nr_pages(folio);
> if (folio_test_private(folio))
>      expected_refcount++;
> 
> The latter check should be fine even for anon folios (although always 
> false)

Looks better. Will do in v2.

>>       return folio_ref_count(folio) == expected_refcount;
>>   }
>> @@ -2285,8 +2291,7 @@ static int hpage_collapse_scan_file(struct 
>> mm_struct *mm, unsigned long addr,
>>               break;
>>           }
>> -        if (folio_ref_count(folio) !=
>> -            1 + folio_mapcount(folio) + folio_test_private(folio)) {
> 
> The "1" is due to the pagecache, right? IIUC, we don't hold a raised 
> folio refcount as we do the xas_for_each().

Right.