[PATCH v2 06/14] mm: handle_pte_fault() use pte_offset_map_rw_nolock()

[Qi Zheng]

On 2024/8/26 23:36, David Hildenbrand wrote:
> On 22.08.24 09:13, Qi Zheng wrote:
>> In handle_pte_fault(), we may modify the vmf->pte after acquiring the
>> vmf->ptl, so convert it to using pte_offset_map_rw_nolock(). But since we
>> will do the pte_same() check, so there is no need to get pmdval to do
>> pmd_same() check, just pass a dummy variable to it.
>>
>> Signed-off-by: Qi Zheng <zhengqi.arch at bytedance.com>
>> ---
>>   mm/memory.c | 12 ++++++++++--
>>   1 file changed, 10 insertions(+), 2 deletions(-)
>>
>> diff --git a/mm/memory.c b/mm/memory.c
>> index 93c0c25433d02..7b6071a0e21e2 100644
>> --- a/mm/memory.c
>> +++ b/mm/memory.c
>> @@ -5499,14 +5499,22 @@ static vm_fault_t handle_pte_fault(struct 
>> vm_fault *vmf)
>>           vmf->pte = NULL;
>>           vmf->flags &= ~FAULT_FLAG_ORIG_PTE_VALID;
>>       } else {
>> +        pmd_t dummy_pmdval;
>> +
>>           /*
>>            * A regular pmd is established and it can't morph into a huge
>>            * 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, so
>> +         * just pass a dummy variable to it.
>>            */
>> -        vmf->pte = pte_offset_map_nolock(vmf->vma->vm_mm, vmf->pmd,
>> -                         vmf->address, &vmf->ptl);
>> +        vmf->pte = pte_offset_map_rw_nolock(vmf->vma->vm_mm, vmf->pmd,
>> +                            vmf->address, &dummy_pmdval,
>> +                            &vmf->ptl);
>>           if (unlikely(!vmf->pte))
>>               return 0;
>>           vmf->orig_pte = ptep_get_lockless(vmf->pte);
> 
> No I understand why we don't need the PMD val in these cases ... the PTE 
> would also be pte_none() at the point the page table is freed, so we 
> would detect the change as well.

Yes.

> 
> I do enjoy documenting why we use a dummy value, though. Likely without 
> that, new users will just pass NULL and call it a day.

OK, how about the following:

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_none() entry, there is no need to recheck
the pmdval. Here we chooes to pass a dummy variable instead
of NULL, which helps new user think about why this place is
special.

> 
> Acked-by: David Hildenbrand <david at redhat.com>

Thanks!

>