hugetlb: skip to end of PT page mapping when pte not present

Patch series "hugetlb: speed up linear address scanning", v2.

At unmap, fork and remap time hugetlb address ranges are linearly scanned.
We can optimize these scans if the ranges are sparsely populated.

Also, enable page table "Lazy copy" for hugetlb at fork.

NOTE: Architectures not defining CONFIG_ARCH_WANT_GENERAL_HUGETLB need to
add an arch specific version hugetlb_mask_last_page() to take advantage of
sparse address scanning improvements.  Baolin Wang added the routine for
arm64.  Other architectures which could be optimized are: ia64, mips,
parisc, powerpc, s390, sh and sparc.


This patch (of 4):

HugeTLB address ranges are linearly scanned during fork, unmap and remap
operations.  If a non-present entry is encountered, the code currently
continues to the next huge page aligned address.  However, a non-present
entry implies that the page table page for that entry is not present. 
Therefore, the linear scan can skip to the end of range mapped by the page
table page.  This can speed operations on large sparsely populated hugetlb
mappings.

Create a new routine hugetlb_mask_last_page() that will return an address
mask.  When the mask is ORed with an address, the result will be the
address of the last huge page mapped by the associated page table page. 
Use this mask to update addresses in routines which linearly scan hugetlb
address ranges when a non-present pte is encountered.

hugetlb_mask_last_page is related to the implementation of huge_pte_offset
as hugetlb_mask_last_page is called when huge_pte_offset returns NULL. 
This patch only provides a complete hugetlb_mask_last_page implementation
when CONFIG_ARCH_WANT_GENERAL_HUGETLB is defined.  Architectures which
provide their own versions of huge_pte_offset can also provide their own
version of hugetlb_mask_last_page.

Link: https://lkml.kernel.org/r/20220621235620.291305-1-mike.kravetz@oracle.com
Link: https://lkml.kernel.org/r/20220621235620.291305-2-mike.kravetz@oracle.com
Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Tested-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Acked-by: Muchun Song <songmuchun@bytedance.com>
Reported-by: kernel test robot <lkp@intel.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Naoya Horiguchi <naoya.horiguchi@linux.dev>
Cc: James Houghton <jthoughton@google.com>
Cc: Mina Almasry <almasrymina@google.com>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Christian Borntraeger <borntraeger@linux.ibm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Rolf Eike Beer <eike-kernel@sf-tec.de>
Cc: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

1 files changed, 51 insertions, 5 deletions

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index ffdf3fc4a83f..95fd1c36c17f 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -4727,6 +4727,7 @@ int copy_hugetlb_page_range(struct mm_struct *dst, struct mm_struct *src,
 	unsigned long npages = pages_per_huge_page(h);
 	struct address_space *mapping = src_vma->vm_file->f_mapping;
 	struct mmu_notifier_range range;
+	unsigned long last_addr_mask;
 	int ret = 0;
 
 	if (cow) {
@@ -4746,11 +4747,14 @@ int copy_hugetlb_page_range(struct mm_struct *dst, struct mm_struct *src,
 		i_mmap_lock_read(mapping);
 	}
 
+	last_addr_mask = hugetlb_mask_last_page(h);
 	for (addr = src_vma->vm_start; addr < src_vma->vm_end; addr += sz) {
 		spinlock_t *src_ptl, *dst_ptl;
 		src_pte = huge_pte_offset(src, addr, sz);
-		if (!src_pte)
+		if (!src_pte) {
+			addr |= last_addr_mask;
 			continue;
+		}
 		dst_pte = huge_pte_alloc(dst, dst_vma, addr, sz);
 		if (!dst_pte) {
 			ret = -ENOMEM;
@@ -4767,8 +4771,10 @@ int copy_hugetlb_page_range(struct mm_struct *dst, struct mm_struct *src,
 		 * after taking the lock below.
 		 */
 		dst_entry = huge_ptep_get(dst_pte);
-		if ((dst_pte == src_pte) || !huge_pte_none(dst_entry))
+		if ((dst_pte == src_pte) || !huge_pte_none(dst_entry)) {
+			addr |= last_addr_mask;
 			continue;
+		}
 
 		dst_ptl = huge_pte_lock(h, dst, dst_pte);
 		src_ptl = huge_pte_lockptr(h, src, src_pte);
@@ -4928,6 +4934,7 @@ int move_hugetlb_page_tables(struct vm_area_struct *vma,
 	unsigned long sz = huge_page_size(h);
 	struct mm_struct *mm = vma->vm_mm;
 	unsigned long old_end = old_addr + len;
+	unsigned long last_addr_mask;
 	unsigned long old_addr_copy;
 	pte_t *src_pte, *dst_pte;
 	struct mmu_notifier_range range;
@@ -4943,12 +4950,16 @@ int move_hugetlb_page_tables(struct vm_area_struct *vma,
 	flush_cache_range(vma, range.start, range.end);
 
 	mmu_notifier_invalidate_range_start(&range);
+	last_addr_mask = hugetlb_mask_last_page(h);
 	/* Prevent race with file truncation */
 	i_mmap_lock_write(mapping);
 	for (; old_addr < old_end; old_addr += sz, new_addr += sz) {
 		src_pte = huge_pte_offset(mm, old_addr, sz);
-		if (!src_pte)
+		if (!src_pte) {
+			old_addr |= last_addr_mask;
+			new_addr |= last_addr_mask;
 			continue;
+		}
 		if (huge_pte_none(huge_ptep_get(src_pte)))
 			continue;
 
@@ -4993,6 +5004,7 @@ static void __unmap_hugepage_range(struct mmu_gather *tlb, struct vm_area_struct
 	struct hstate *h = hstate_vma(vma);
 	unsigned long sz = huge_page_size(h);
 	struct mmu_notifier_range range;
+	unsigned long last_addr_mask;
 	bool force_flush = false;
 
 	WARN_ON(!is_vm_hugetlb_page(vma));
@@ -5013,11 +5025,14 @@ static void __unmap_hugepage_range(struct mmu_gather *tlb, struct vm_area_struct
 				end);
 	adjust_range_if_pmd_sharing_possible(vma, &range.start, &range.end);
 	mmu_notifier_invalidate_range_start(&range);
+	last_addr_mask = hugetlb_mask_last_page(h);
 	address = start;
 	for (; address < end; address += sz) {
 		ptep = huge_pte_offset(mm, address, sz);
-		if (!ptep)
+		if (!ptep) {
+			address |= last_addr_mask;
 			continue;
+		}
 
 		ptl = huge_pte_lock(h, mm, ptep);
 		if (huge_pmd_unshare(mm, vma, &address, ptep)) {
@@ -6285,6 +6300,7 @@ unsigned long hugetlb_change_protection(struct vm_area_struct *vma,
 	unsigned long pages = 0, psize = huge_page_size(h);
 	bool shared_pmd = false;
 	struct mmu_notifier_range range;
+	unsigned long last_addr_mask;
 	bool uffd_wp = cp_flags & MM_CP_UFFD_WP;
 	bool uffd_wp_resolve = cp_flags & MM_CP_UFFD_WP_RESOLVE;
 
@@ -6301,12 +6317,15 @@ unsigned long hugetlb_change_protection(struct vm_area_struct *vma,
 	flush_cache_range(vma, range.start, range.end);
 
 	mmu_notifier_invalidate_range_start(&range);
+	last_addr_mask = hugetlb_mask_last_page(h);
 	i_mmap_lock_write(vma->vm_file->f_mapping);
 	for (; address < end; address += psize) {
 		spinlock_t *ptl;
 		ptep = huge_pte_offset(mm, address, psize);
-		if (!ptep)
+		if (!ptep) {
+			address |= last_addr_mask;
 			continue;
+		}
 		ptl = huge_pte_lock(h, mm, ptep);
 		if (huge_pmd_unshare(mm, vma, &address, ptep)) {
 			/*
@@ -6856,6 +6875,33 @@ pte_t *huge_pte_offset(struct mm_struct *mm,
 	return (pte_t *)pmd;
 }
 
+/*
+ * Return a mask that can be used to update an address to the last huge
+ * page in a page table page mapping size.  Used to skip non-present
+ * page table entries when linearly scanning address ranges.  Architectures
+ * with unique huge page to page table relationships can define their own
+ * version of this routine.
+ */
+unsigned long hugetlb_mask_last_page(struct hstate *h)
+{
+	unsigned long hp_size = huge_page_size(h);
+
+	if (hp_size == PUD_SIZE)
+		return P4D_SIZE - PUD_SIZE;
+	else if (hp_size == PMD_SIZE)
+		return PUD_SIZE - PMD_SIZE;
+	else
+		return 0UL;
+}
+
+#else
+
+/* See description above.  Architectures can provide their own version. */
+__weak unsigned long hugetlb_mask_last_page(struct hstate *h)
+{
+	return 0UL;
+}
+
 #endif /* CONFIG_ARCH_WANT_GENERAL_HUGETLB */
 
 /*