Re: [PATCH v2 2/5] mm: damon must atomically clear young on ptes and pmds

[Ryan Roberts <ryan.roberts@arm.com>]

On 19/05/2023 20:54, SeongJae Park wrote:
> On Fri, 19 May 2023 10:02:48 +0100 Ryan Roberts <ryan.roberts@arm.com> wrote:
> 
>> On 19/05/2023 00:19, Yu Zhao wrote:
>>> On Thu, May 18, 2023 at 5:07 AM Ryan Roberts <ryan.roberts@arm.com> wrote:
>>>>
>>>> It is racy to non-atomically read a pte, then clear the young bit, then
>>>> write it back as this could discard dirty information. Further, it is
>>>> bad practice to directly set a pte entry within a table. Instead
>>>> clearing young must go through the arch-provided helper,
>>>> ptep_test_and_clear_young() to ensure it is modified atomically and to
>>>> give the arch code visibility and allow it to check (and potentially
>>>> modify) the operation.
>>>>
>>>> Fixes: 46c3a0accdc4 ("mm/damon/vaddr: separate commonly usable functions")
>>>
>>> This should be a separate patch, since it's independent from what the
>>> series tries to do.
>>>
>>> And that patch should cc stable, since it fixes user data corruption.
>>
>> Fair point. The first 3 patches are fixes for issues I found during the
>> conversion. So if you're ok with it, I'll split into 2 series; the first 3
>> patches in the first, and the conversion to ptep_deref() (or ptep_get(), as per
>> your comments in the other thread) in the second. I guess the whole first series
>> should go to stable.
>>
>> Let me know if you disagree.
>>
>>>
>>>> Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
>>>> Reviewed-by: Zi Yan <ziy@nvidia.com>
>>>> ---
>>>>  mm/damon/ops-common.c | 16 ++++++----------
>>>>  mm/damon/ops-common.h |  4 ++--
>>>>  mm/damon/paddr.c      |  4 ++--
>>>>  mm/damon/vaddr.c      |  4 ++--
>>>>  4 files changed, 12 insertions(+), 16 deletions(-)
>>>>
>>>> diff --git a/mm/damon/ops-common.c b/mm/damon/ops-common.c
>>>> index cc63cf953636..acc264b97903 100644
>>>> --- a/mm/damon/ops-common.c
>>>> +++ b/mm/damon/ops-common.c
>>>> @@ -37,7 +37,7 @@ struct folio *damon_get_folio(unsigned long pfn)
>>>>         return folio;
>>>>  }
>>>>
>>>> -void damon_ptep_mkold(pte_t *pte, struct mm_struct *mm, unsigned long addr)
>>>> +void damon_ptep_mkold(pte_t *pte, struct vm_area_struct *vma, unsigned long addr)
>>>>  {
>>>>         bool referenced = false;
>>>>         struct folio *folio = damon_get_folio(pte_pfn(*pte));
>>>> @@ -45,13 +45,11 @@ void damon_ptep_mkold(pte_t *pte, struct mm_struct *mm, unsigned long addr)
>>>>         if (!folio)
>>>>                 return;
>>>>
>>>> -       if (pte_young(*pte)) {
>>>> +       if (ptep_test_and_clear_young(vma, addr, pte))
>>>>                 referenced = true;
>>>> -               *pte = pte_mkold(*pte);
>>>> -       }
>>>>
>>>>  #ifdef CONFIG_MMU_NOTIFIER
>>>> -       if (mmu_notifier_clear_young(mm, addr, addr + PAGE_SIZE))
>>>> +       if (mmu_notifier_clear_young(vma->vm_mm, addr, addr + PAGE_SIZE))
>>>>                 referenced = true;
>>>>  #endif /* CONFIG_MMU_NOTIFIER */
>>>
>>> Use ptep_clear_young_notify(). Similar below.
>>
>> This looks sensible but I'd like clarification from SeongJae: Are you happy for
>> me to do this refactoring as part of the patch?
> 
> Yes, I would be happy for that :)  Nevertheless, because that's a refactoring
> rather than a fix, I'd be happier if you could do that with a separate patch.

Great - I'll do it like this for v3.

> 
> 
> Thanks,
> SJ
> 
>>
>> Thanks,
>> Ryan
>>