Re: [v2 PATCH] mm: thp: fix false negative of shmem vma's THP eligibility

[Yang Shi <yang.shi@linux.alibaba.com>]

On 4/24/19 9:17 AM, Vlastimil Babka wrote:
> On 4/24/19 5:47 PM, Yang Shi wrote:
>>
>> On 4/24/19 6:10 AM, Vlastimil Babka wrote:
>>> On 4/23/19 6:43 PM, Yang Shi wrote:
>>>> The commit 7635d9cbe832 ("mm, thp, proc: report THP eligibility for each
>>>> vma") introduced THPeligible bit for processes' smaps. But, when checking
>>>> the eligibility for shmem vma, __transparent_hugepage_enabled() is
>>>> called to override the result from shmem_huge_enabled().  It may result
>>>> in the anonymous vma's THP flag override shmem's.  For example, running a
>>>> simple test which create THP for shmem, but with anonymous THP disabled,
>>>> when reading the process's smaps, it may show:
>>>>
>>>> 7fc92ec00000-7fc92f000000 rw-s 00000000 00:14 27764 /dev/shm/test
>>>> Size:               4096 kB
>>>> ...
>>>> [snip]
>>>> ...
>>>> ShmemPmdMapped:     4096 kB
>>> But how does this happen in the first place?
>>> In __handle_mm_fault() we do:
>>>
>>>           if (pmd_none(*vmf.pmd) && __transparent_hugepage_enabled(vma)) {
>>>                   ret = create_huge_pmd(&vmf);
>>>                   if (!(ret & VM_FAULT_FALLBACK))
>>>                           return ret;
>>>
>>> And __transparent_hugepage_enabled() checks the global THP settings.
>>> If THP is not enabled / is only for madvise and the vma is not madvised,
>>> then this should fail, and also khugepaged shouldn't either run at all,
>>> or don't do its job for such non-madvised vma.
>> If __transparent_hugepage_enabled() returns false, the code will not
>> reach create_huge_pmd() at all. If it returns true, create_huge_pmd()
>> actually will return VM_FAULT_FALLBACK for shmem since shmem doesn't
>> have huge_fault (or pmd_fault in earlier versions) method.
>>
>> Then it will get into handle_pte_fault(), finally shmem_fault() is
>> called, which allocates THP by checking some global flag (i.e.
>> VM_NOHUGEPAGE and MMF_DISABLE_THP) and  shmem THP knobs.
> Aha, thanks! What a mess...

Yes, it does look convoluted. I'm wondering we may consider refactor the 
shmem THP fault handling.

>
>> 4.8 (the first version has shmem THP merged) behaves exactly in the same
>> way. So, I suspect this may be intended behavior.
> Still looks like an oversight to me. And it's inconsistent... it might
> fault huge shmem pages when THPs are globally disabled, but khugepaged
> is still not running. I think it should just check the global THP flags
> as well...

It does looks inconsistent, particularly for the khugepaged part.