Re: [RFC PATCH 0/3] support large folio for mlock

[Yu Zhao]

On Fri, Jul 7, 2023 at 10:52 AM Yin Fengwei <fengwei.yin@xxxxxxxxx> wrote:
>
> Yu mentioned at [1] about the mlock() can't be applied to large folio.
>
> I leant the related code and here is my understanding:
> - For RLIMIT_MEMLOCK related, there is no problem. Becuase the
>   RLIMIT_MEMLOCK statistics is not related underneath page. That means
>   underneath page mlock or munlock doesn't impact the RLIMIT_MEMLOCK
>   statistics collection which is always correct.
>
> - For keeping the page in RAM, there is no problem either. At least,
>   during try_to_unmap_one(), once detect the VMA has VM_LOCKED bit
>   set in vm_flags, the folio will be kept whatever the folio is
>   mlocked or not.
>
> So the function of mlock for large folio works. But it's not optimized
> because the page reclaim needs scan these large folio and may split
> them.
>
> This series identified the large folio for mlock to two types:
>   - The large folio is in VM_LOCKED VMA range
>   - The large folio cross VM_LOCKED VMA boundary
>
> For the first type, we mlock large folio so page relcaim will skip it.
> For the second type, we don't mlock large folio. It's allowed to be
> picked by page reclaim and be split. So the pages not in VM_LOCKED VMA
> range are allowed to be reclaimed/released.

This is a sound design, which is also what I have in mind. I see the
rationales are being spelled out in this thread, and hopefully
everyone can be convinced.

> patch1 introduce API to check whether large folio is in VMA range.
> patch2 make page reclaim/mlock_vma_folio/munlock_vma_folio support
> large folio mlock/munlock.
> patch3 make mlock/munlock syscall support large folio.

Could you tidy up the last patch a little bit? E.g., Saying "mlock the
4K folio" is obviously not the best idea.

And if it's possible, make the loop just look like before, i.e.,

  if (!can_mlock_entire_folio())
    continue;
  if (vma->vm_flags & VM_LOCKED)
    mlock_folio_range();
  else
    munlock_folio_range();

Thanks.