On 23 Mar 2026, at 15:53, David Hildenbrand (Arm) wrote:

> On 3/23/26 20:47, Zi Yan wrote:
>> On 23 Mar 2026, at 15:41, David Hildenbrand (Arm) wrote:
>>
>>> On 3/23/26 20:06, Zi Yan wrote:
>>>> READ_ONLY_THP_FOR_FS is no longer present, remove corresponding code.
>>>>
>>>> Signed-off-by: Zi Yan <[email protected]>
>>>> ---
>>>>  mm/khugepaged.c | 159 +++++++++++-------------------------------------
>>>>  1 file changed, 34 insertions(+), 125 deletions(-)
>>>>
>>>> diff --git a/mm/khugepaged.c b/mm/khugepaged.c
>>>> index b2ac28ddd480..39f0b8959535 100644
>>>> --- a/mm/khugepaged.c
>>>> +++ b/mm/khugepaged.c
>>>> @@ -1899,7 +1899,7 @@ static enum scan_result collapse_file(struct 
>>>> mm_struct *mm, unsigned long addr,
>>>>    int nr_none = 0;
>>>>    bool is_shmem = shmem_file(file);
>>>>
>>>> -  VM_BUG_ON(!IS_ENABLED(CONFIG_READ_ONLY_THP_FOR_FS) && !is_shmem);
>>>> +  VM_WARN_ON_ONCE(!is_shmem);
>>>
>>> Oh, but if that means that khugepaged cannot collapse large folios in
>>> FSes anymore, then this is the wrong approach I suppose?
>>>
>>> I would have assumed that we would now collapse for any files that
>>> support large folios (in PMD size), not stopping to collapse entirely.
>>
>> My understanding is that collapse_file() is only used for
>> READ_ONLY_THP_FOR_FS. If FSes with large folio support also use it,
>> I can replace IS_ENABLE with mapping_large_folio_support().
>
> Otherwise we'd be losing support for THP collapse in files? We'd have to
> cross fingers that readahead gives us some.
>
> So we have to be a bit careful here. We want khugepaged to collapse THPs
> in filesystems that support large folios even without READ_ONLY_THP_FOR_FS.

Based on[1], collapse_file() only works on read-only fd. So it is not
as general as all files. If we do what you suggested, the function
will be collapse read-only folios from FSes with large folio support.

BTW, has anyone used/relied on collapse_file() on any FS with large folio
support?

[1] https://elixir.bootlin.com/linux/v6.19.9/source/mm/khugepaged.c#L1956


Best Regards,
Yan, Zi

Reply via email to