Re: [PATCH 06/14] mm: handle_pte_fault() use pte_offset_map_maywrite_nolock()

[David Hildenbrand <david@redhat.com>]

On 21.08.24 11:51, Qi Zheng wrote:
> 
> 
> On 2024/8/21 17:41, David Hildenbrand wrote:
>> On 21.08.24 11:24, Qi Zheng wrote:
>>>
>>>
>>> On 2024/8/21 17:17, LEROY Christophe wrote:
>>>>
>>>>
>>>> Le 21/08/2024 à 10:18, Qi Zheng a écrit :
>>>>> In handle_pte_fault(), we may modify the vmf->pte after acquiring the
>>>>> vmf->ptl, so convert it to using pte_offset_map_maywrite_nolock(). But
>>>>> since we already do the pte_same() check, so there is no need to get
>>>>> pmdval to do pmd_same() check, just pass NULL to pmdvalp parameter.
>>>>>
>>>>> Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
>>>>> ---
>>>>>      mm/memory.c | 9 +++++++--
>>>>>      1 file changed, 7 insertions(+), 2 deletions(-)
>>>>>
>>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>>> index 93c0c25433d02..d3378e98faf13 100644
>>>>> --- a/mm/memory.c
>>>>> +++ b/mm/memory.c
>>>>> @@ -5504,9 +5504,14 @@ static vm_fault_t handle_pte_fault(struct
>>>>> vm_fault *vmf)
>>>>>               * pmd by anon khugepaged, since that takes mmap_lock in
>>>>> write
>>>>>               * mode; but shmem or file collapse to THP could still
>>>>> morph
>>>>>               * it into a huge pmd: just retry later if so.
>>>>> +         *
>>>>> +         * Use the maywrite version to indicate that vmf->pte will be
>>>>> +         * modified, but since we will use pte_same() to detect the
>>>>> +         * change of the pte entry, there is no need to get pmdval.
>>>>>               */
>>>>> -        vmf->pte = pte_offset_map_nolock(vmf->vma->vm_mm, vmf->pmd,
>>>>> -                         vmf->address, &vmf->ptl);
>>>>> +        vmf->pte = pte_offset_map_maywrite_nolock(vmf->vma->vm_mm,
>>>>> +                              vmf->pmd, vmf->address,
>>>>> +                              NULL, &vmf->ptl);
>>
>> I think we discussed that passing NULL should be forbidden for that
>> function.
> 
> Yes, but for some maywrite case, there is no need to get pmdval to
> do pmd_same() check. So I passed NULL and added a comment to
> explain this.

I wonder if it's better to pass a dummy variable instead. One has to 
think harder why that is required compared to blindly passing "NULL" :)

-- 
Cheers,

David / dhildenb