Re: [RFC PATCH 15/26] hugetlb: make unmapping compatible with high-granularity mappings

["manish.mishra" <manish.mishra@nutanix.com>]

On 24/06/22 11:06 pm, James Houghton wrote:
> This enlightens __unmap_hugepage_range to deal with high-granularity
> mappings. This doesn't change its API; it still must be called with
> hugepage alignment, but it will correctly unmap hugepages that have been
> mapped at high granularity.
>
> Analogous to the mapcount rules introduced by hugetlb_no_page, we only
> drop mapcount in this case if we are unmapping an entire hugepage in one
> operation. This is the case when a VMA is destroyed.
>
> Eventually, functionality here can be expanded to allow users to call
> MADV_DONTNEED on PAGE_SIZE-aligned sections of a hugepage, but that is
> not done here.

Sorry i may have misunderstood something here, but allowing something like

MADV_DONTNEED on PAGE_SIZE in hugetlbfs can cause fragmentation

in hugetlbfs pool which kind of looks opposite of prupose of hugetlbfs?

>
> Signed-off-by: James Houghton <jthoughton@google.com>
> ---
>   include/asm-generic/tlb.h |  6 +--
>   mm/hugetlb.c              | 85 ++++++++++++++++++++++++++-------------
>   2 files changed, 59 insertions(+), 32 deletions(-)
>
> diff --git a/include/asm-generic/tlb.h b/include/asm-generic/tlb.h
> index ff3e82553a76..8daa3ae460d9 100644
> --- a/include/asm-generic/tlb.h
> +++ b/include/asm-generic/tlb.h
> @@ -562,9 +562,9 @@ static inline void tlb_flush_p4d_range(struct mmu_gather *tlb,
>   		__tlb_remove_tlb_entry(tlb, ptep, address);	\
>   	} while (0)
>   
> -#define tlb_remove_huge_tlb_entry(h, tlb, ptep, address)	\
> +#define tlb_remove_huge_tlb_entry(tlb, hpte, address)	\
>   	do {							\
> -		unsigned long _sz = huge_page_size(h);		\
> +		unsigned long _sz = hugetlb_pte_size(&hpte);	\
>   		if (_sz >= P4D_SIZE)				\
>   			tlb_flush_p4d_range(tlb, address, _sz);	\
>   		else if (_sz >= PUD_SIZE)			\
> @@ -573,7 +573,7 @@ static inline void tlb_flush_p4d_range(struct mmu_gather *tlb,
>   			tlb_flush_pmd_range(tlb, address, _sz);	\
>   		else						\
>   			tlb_flush_pte_range(tlb, address, _sz);	\
> -		__tlb_remove_tlb_entry(tlb, ptep, address);	\
> +		__tlb_remove_tlb_entry(tlb, hpte.ptep, address);\
>   	} while (0)
>   
>   /**
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index da30621656b8..51fc1d3f122f 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -5120,24 +5120,20 @@ static void __unmap_hugepage_range(struct mmu_gather *tlb, struct vm_area_struct
>   {
>   	struct mm_struct *mm = vma->vm_mm;
>   	unsigned long address;
> -	pte_t *ptep;
> +	struct hugetlb_pte hpte;
>   	pte_t pte;
>   	spinlock_t *ptl;
> -	struct page *page;
> +	struct page *hpage, *subpage;
>   	struct hstate *h = hstate_vma(vma);
>   	unsigned long sz = huge_page_size(h);
>   	struct mmu_notifier_range range;
>   	bool force_flush = false;
> +	bool hgm_enabled = hugetlb_hgm_enabled(vma);
>   
>   	WARN_ON(!is_vm_hugetlb_page(vma));
>   	BUG_ON(start & ~huge_page_mask(h));
>   	BUG_ON(end & ~huge_page_mask(h));
>   
> -	/*
> -	 * This is a hugetlb vma, all the pte entries should point
> -	 * to huge page.
> -	 */
> -	tlb_change_page_size(tlb, sz);
>   	tlb_start_vma(tlb, vma);
>   
>   	/*
> @@ -5148,25 +5144,43 @@ static void __unmap_hugepage_range(struct mmu_gather *tlb, struct vm_area_struct
>   	adjust_range_if_pmd_sharing_possible(vma, &range.start, &range.end);
>   	mmu_notifier_invalidate_range_start(&range);
>   	address = start;
> -	for (; address < end; address += sz) {
> -		ptep = huge_pte_offset(mm, address, sz);
> -		if (!ptep)
> +
> +	while (address < end) {
> +		pte_t *ptep = huge_pte_offset(mm, address, sz);
> +
> +		if (!ptep) {
> +			address += sz;
>   			continue;
> +		}
> +		hugetlb_pte_populate(&hpte, ptep, huge_page_shift(h));
> +		if (hgm_enabled) {
> +			int ret = huge_pte_alloc_high_granularity(
> +					&hpte, mm, vma, address, PAGE_SHIFT,
> +					HUGETLB_SPLIT_NEVER,
> +					/*write_locked=*/true);

I see huge_pte_alloc_high_granularity with HUGETLB_SPLIT_NEVER just

do huge_tlb_walk. So is HUGETLB_SPLIT_NEVER even required, i mean

for those cases you can directly do huge_tlb_walk? I mean name

huge_pte_alloc_high_granularity confuses for those cases.

> +			/*
> +			 * We will never split anything, so this should always
> +			 * succeed.
> +			 */
> +			BUG_ON(ret);
> +		}
>   
> -		ptl = huge_pte_lock(h, mm, ptep);
> -		if (huge_pmd_unshare(mm, vma, &address, ptep)) {
> +		ptl = hugetlb_pte_lock(mm, &hpte);
> +		if (!hgm_enabled && huge_pmd_unshare(
> +					mm, vma, &address, hpte.ptep)) {
>   			spin_unlock(ptl);
>   			tlb_flush_pmd_range(tlb, address & PUD_MASK, PUD_SIZE);
>   			force_flush = true;
> -			continue;
> +			goto next_hpte;
>   		}
>   
> -		pte = huge_ptep_get(ptep);
> -		if (huge_pte_none(pte)) {
> +		if (hugetlb_pte_none(&hpte)) {
>   			spin_unlock(ptl);
> -			continue;
> +			goto next_hpte;
>   		}
>   
> +		pte = hugetlb_ptep_get(&hpte);
> +
>   		/*
>   		 * Migrating hugepage or HWPoisoned hugepage is already
>   		 * unmapped and its refcount is dropped, so just clear pte here.
> @@ -5180,24 +5194,27 @@ static void __unmap_hugepage_range(struct mmu_gather *tlb, struct vm_area_struct
>   			 */
>   			if (pte_swp_uffd_wp_any(pte) &&
>   			    !(zap_flags & ZAP_FLAG_DROP_MARKER))
> -				set_huge_pte_at(mm, address, ptep,
> +				set_huge_pte_at(mm, address, hpte.ptep,
>   						make_pte_marker(PTE_MARKER_UFFD_WP));
>   			else
> -				huge_pte_clear(mm, address, ptep, sz);
> +				huge_pte_clear(mm, address, hpte.ptep,
> +						hugetlb_pte_size(&hpte));
>   			spin_unlock(ptl);
> -			continue;
> +			goto next_hpte;
>   		}
>   
> -		page = pte_page(pte);
> +		subpage = pte_page(pte);
> +		BUG_ON(!subpage);
> +		hpage = compound_head(subpage);
>   		/*
>   		 * If a reference page is supplied, it is because a specific
>   		 * page is being unmapped, not a range. Ensure the page we
>   		 * are about to unmap is the actual page of interest.
>   		 */
>   		if (ref_page) {
> -			if (page != ref_page) {
> +			if (hpage != ref_page) {
>   				spin_unlock(ptl);
> -				continue;
> +				goto next_hpte;
>   			}
>   			/*
>   			 * Mark the VMA as having unmapped its page so that
> @@ -5207,25 +5224,35 @@ static void __unmap_hugepage_range(struct mmu_gather *tlb, struct vm_area_struct
>   			set_vma_resv_flags(vma, HPAGE_RESV_UNMAPPED);
>   		}
>   
> -		pte = huge_ptep_get_and_clear(mm, address, ptep);
> -		tlb_remove_huge_tlb_entry(h, tlb, ptep, address);
> +		pte = huge_ptep_get_and_clear(mm, address, hpte.ptep);
> +		tlb_change_page_size(tlb, hugetlb_pte_size(&hpte));
> +		tlb_remove_huge_tlb_entry(tlb, hpte, address);
>   		if (huge_pte_dirty(pte))
> -			set_page_dirty(page);
> +			set_page_dirty(hpage);
>   		/* Leave a uffd-wp pte marker if needed */
>   		if (huge_pte_uffd_wp(pte) &&
>   		    !(zap_flags & ZAP_FLAG_DROP_MARKER))
> -			set_huge_pte_at(mm, address, ptep,
> +			set_huge_pte_at(mm, address, hpte.ptep,
>   					make_pte_marker(PTE_MARKER_UFFD_WP));
> -		hugetlb_count_sub(pages_per_huge_page(h), mm);
> -		page_remove_rmap(page, vma, true);
> +
> +		hugetlb_count_sub(hugetlb_pte_size(&hpte)/PAGE_SIZE, mm);
> +
> +		/*
> +		 * If we are unmapping the entire page, remove it from the
> +		 * rmap.
> +		 */
> +		if (IS_ALIGNED(address, sz) && address + sz <= end)
> +			page_remove_rmap(hpage, vma, true);
>   
>   		spin_unlock(ptl);
> -		tlb_remove_page_size(tlb, page, huge_page_size(h));
> +		tlb_remove_page_size(tlb, subpage, hugetlb_pte_size(&hpte));
>   		/*
>   		 * Bail out after unmapping reference page if supplied
>   		 */
>   		if (ref_page)
>   			break;
> +next_hpte:
> +		address += hugetlb_pte_size(&hpte);
>   	}
>   	mmu_notifier_invalidate_range_end(&range);
>   	tlb_end_vma(tlb, vma);