Re: [RFC PATCH 2/3] mm/huge_memory: add kernel-doc for folio_split_supported()

["David Hildenbrand (Red Hat)" <david@kernel.org>]

On 11/20/25 15:48, Zi Yan wrote:
> On 20 Nov 2025, at 4:27, David Hildenbrand (Red Hat) wrote:
> 
>> On 11/20/25 04:59, Zi Yan wrote:
>>> It clarifies that folio_split_supported() does not check folio->mapping and
>>> can dereference it.
>>>
>>> Signed-off-by: Zi Yan <ziy@nvidia.com>
>>> ---
>>>    mm/huge_memory.c | 17 +++++++++++++++++
>>>    1 file changed, 17 insertions(+)
>>>
>>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
>>> index efea42d68157..15e555f1b85d 100644
>>> --- a/mm/huge_memory.c
>>> +++ b/mm/huge_memory.c
>>> @@ -3688,6 +3688,23 @@ static int __split_unmapped_folio(struct folio *folio, int new_order,
>>>    	return 0;
>>>    }
>>>   +/**
>>> + * folio_split_supported() - check if a folio can be split to a given order
>>> + * @folio: folio to be split
>>> + * @new_order: the smallest order of the after split folios (since buddy
>>> + *             allocator like split generates folios with orders from @folio's
>>> + *             order - 1 to new_order).
>>> + * @split_type: uniform or non-uniform split
>>> + * @warns: whether gives warnings or not for the checks in the function
>>> + *
>>> + * folio_split_supported() checks if @folio can be split to @new_order using
>>> + * @split_type method.
>>> + *
>>> + * Context: Caller must make sure folio->mapping is not NULL, since the
>>> + *          function does not check it and can dereference folio->mapping
>>
>> Only for anon folios. Also, I would drop the detail about dereference.
> 
> OK.
> 
>>
>> I guess we really need the folio lock to prevent concurrent truncation.
>>
>> Maybe something like:
>>
>> "The folio must be locked. For non-anon folios, the caller must make sure that folio->mapping is not NULL (e.g., not truncated)."
> 
> Sure. Do you think it is worth adding VM_WARN_ONCE_ON(!folio_test_locked);
> and VM_WARN_ONCE_ON(!folio->mapping); ?

Makes sense. Or we allow !folio->mapping, return false and do something 
like the following. Still wondering how we could handle that case better.

if (!folio_split_supported(folio)) {
	if (folio_split_temporarily_unsupported(folio))
		return -EBUSY;
	return -EINVAL;
}

hmmmm

-- 
Cheers

David