Re: [PATCH v7 3/4] mm/rmap: integrate PMD-mapped folio splitting into pagewalk loop

[David Hildenbrand <david@redhat.com>]

On 13.06.24 10:45, David Hildenbrand wrote:
> On 13.06.24 10:34, David Hildenbrand wrote:
>> On 10.06.24 14:06, Lance Yang wrote:
>>> In preparation for supporting try_to_unmap_one() to unmap PMD-mapped
>>> folios, start the pagewalk first, then call split_huge_pmd_address() to
>>> split the folio.
>>>
>>> Suggested-by: David Hildenbrand <david@redhat.com>
>>> Suggested-by: Baolin Wang <baolin.wang@linux.alibaba.com>
>>> Signed-off-by: Lance Yang <ioworker0@gmail.com>
>>> ---
>>>     include/linux/huge_mm.h |  6 ++++++
>>>     mm/huge_memory.c        | 42 +++++++++++++++++++++--------------------
>>>     mm/rmap.c               | 21 +++++++++++++++------
>>>     3 files changed, 43 insertions(+), 26 deletions(-)
>>>
>>> diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h
>>> index 088d66a54643..4670c6ee118b 100644
>>> --- a/include/linux/huge_mm.h
>>> +++ b/include/linux/huge_mm.h
>>> @@ -415,6 +415,9 @@ static inline bool thp_migration_supported(void)
>>>     	return IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION);
>>>     }
>>>     
>>> +void split_huge_pmd_locked(struct vm_area_struct *vma, unsigned long address,
>>> +			   pmd_t *pmd, bool freeze, struct folio *folio);
>>> +
>>>     #else /* CONFIG_TRANSPARENT_HUGEPAGE */
>>>     
>>>     static inline bool folio_test_pmd_mappable(struct folio *folio)
>>> @@ -477,6 +480,9 @@ static inline void __split_huge_pmd(struct vm_area_struct *vma, pmd_t *pmd,
>>>     		unsigned long address, bool freeze, struct folio *folio) {}
>>>     static inline void split_huge_pmd_address(struct vm_area_struct *vma,
>>>     		unsigned long address, bool freeze, struct folio *folio) {}
>>> +static inline void split_huge_pmd_locked(struct vm_area_struct *vma,
>>> +					 unsigned long address, pmd_t *pmd,
>>> +					 bool freeze, struct folio *folio) {}
>>>     
>>>     #define split_huge_pud(__vma, __pmd, __address)	\
>>>     	do { } while (0)
>>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
>>> index e6d26c2eb670..d2697cc8f9d4 100644
>>> --- a/mm/huge_memory.c
>>> +++ b/mm/huge_memory.c
>>> @@ -2581,6 +2581,27 @@ static void __split_huge_pmd_locked(struct vm_area_struct *vma, pmd_t *pmd,
>>>     	pmd_populate(mm, pmd, pgtable);
>>>     }
>>>     
>>> +void split_huge_pmd_locked(struct vm_area_struct *vma, unsigned long address,
>>> +			   pmd_t *pmd, bool freeze, struct folio *folio)
>>> +{
>>> +	VM_WARN_ON_ONCE(folio && !folio_test_pmd_mappable(folio));
>>> +	VM_WARN_ON_ONCE(!IS_ALIGNED(address, HPAGE_PMD_SIZE));
>>> +	VM_WARN_ON_ONCE(folio && !folio_test_locked(folio));
>>> +	VM_BUG_ON(freeze && !folio);
>>
>> Curious: could we actually end up here without a folio right now? That
>> would mean, that try_to_unmap_one() would be called with folio==NULL.
>>
>>> +
>>> +	/*
>>> +	 * When the caller requests to set up a migration entry, we
>>> +	 * require a folio to check the PMD against. Otherwise, there
>>> +	 * is a risk of replacing the wrong folio.
>>> +	 */
>>> +	if (pmd_trans_huge(*pmd) || pmd_devmap(*pmd) ||
>>> +	    is_pmd_migration_entry(*pmd)) {
>>> +		if (folio && folio != pmd_folio(*pmd))
>>> +			return;
>>> +		__split_huge_pmd_locked(vma, pmd, address, freeze);
>>> +	}
>>> +}
>>> +
>>>     void __split_huge_pmd(struct vm_area_struct *vma, pmd_t *pmd,
>>>     		unsigned long address, bool freeze, struct folio *folio)
>>>     {
>>> @@ -2592,26 +2613,7 @@ void __split_huge_pmd(struct vm_area_struct *vma, pmd_t *pmd,
>>>     				(address & HPAGE_PMD_MASK) + HPAGE_PMD_SIZE);
>>>     	mmu_notifier_invalidate_range_start(&range);
>>>     	ptl = pmd_lock(vma->vm_mm, pmd);
>>> -
>>> -	/*
>>> -	 * If caller asks to setup a migration entry, we need a folio to check
>>> -	 * pmd against. Otherwise we can end up replacing wrong folio.
>>> -	 */
>>> -	VM_BUG_ON(freeze && !folio);
>>> -	VM_WARN_ON_ONCE(folio && !folio_test_locked(folio));
>>> -
>>> -	if (pmd_trans_huge(*pmd) || pmd_devmap(*pmd) ||
>>> -	    is_pmd_migration_entry(*pmd)) {
>>> -		/*
>>> -		 * It's safe to call pmd_page when folio is set because it's
>>> -		 * guaranteed that pmd is present.
>>> -		 */
>>> -		if (folio && folio != pmd_folio(*pmd))
>>> -			goto out;
>>> -		__split_huge_pmd_locked(vma, pmd, range.start, freeze);
>>> -	}
>>> -
>>> -out:
>>> +	split_huge_pmd_locked(vma, range.start, pmd, freeze, folio);
>>>     	spin_unlock(ptl);
>>>     	mmu_notifier_invalidate_range_end(&range);
>>>     }
>>> diff --git a/mm/rmap.c b/mm/rmap.c
>>> index ddffa30c79fb..b77f88695588 100644
>>> --- a/mm/rmap.c
>>> +++ b/mm/rmap.c
>>> @@ -1640,9 +1640,6 @@ static bool try_to_unmap_one(struct folio *folio, struct vm_area_struct *vma,
>>>     	if (flags & TTU_SYNC)
>>>     		pvmw.flags = PVMW_SYNC;
>>>     
>>> -	if (flags & TTU_SPLIT_HUGE_PMD)
>>> -		split_huge_pmd_address(vma, address, false, folio);
>>> -
>>>     	/*
>>>     	 * For THP, we have to assume the worse case ie pmd for invalidation.
>>>     	 * For hugetlb, it could be much worse if we need to do pud
>>> @@ -1668,9 +1665,6 @@ static bool try_to_unmap_one(struct folio *folio, struct vm_area_struct *vma,
>>>     	mmu_notifier_invalidate_range_start(&range);
>>>     
>>>     	while (page_vma_mapped_walk(&pvmw)) {
>>> -		/* Unexpected PMD-mapped THP? */
>>> -		VM_BUG_ON_FOLIO(!pvmw.pte, folio);
>>> -
>>>     		/*
>>>     		 * If the folio is in an mlock()d vma, we must not swap it out.
>>>     		 */
>>> @@ -1682,6 +1676,21 @@ static bool try_to_unmap_one(struct folio *folio, struct vm_area_struct *vma,
>>>     			goto walk_done_err;
>>>     		}
>>>     
>>> +		if (!pvmw.pte && (flags & TTU_SPLIT_HUGE_PMD)) {
>>> +			/*
>>> +			 * We temporarily have to drop the PTL and start once
>>> +			 * again from that now-PTE-mapped page table.
>>> +			 */
>>> +			split_huge_pmd_locked(vma, pvmw.address, pvmw.pmd,
>>> +					      false, folio);
>>> +			flags &= ~TTU_SPLIT_HUGE_PMD;
>>> +			page_vma_mapped_walk_restart(&pvmw);
>>
>> If, for some reason, split_huge_pmd_locked() would fail, we would keep
>> looping and never hit the VM_BUG_ON_FOLIO() below. Maybe we'd want to
>> let split_huge_pmd_locked() return whether splitting succeeded, and
>> handle that case differently?
> 
> I assume it could fail if we race with concurrent split? Or isn't that
> possible?
> 

(I assume we hold the PTL, so such races should not be possible)

-- 
Cheers,

David / dhildenb