1 files changed, 264 insertions, 126 deletions

diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index cca80d96e509..95d1acb0f3d2 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -27,11 +27,12 @@
 #include "internal.h"
 
 /*
- * By default transparent hugepage support is enabled for all mappings
- * and khugepaged scans all mappings. Defrag is only invoked by
- * khugepaged hugepage allocations and by page faults inside
- * MADV_HUGEPAGE regions to avoid the risk of slowing down short lived
- * allocations.
+ * By default transparent hugepage support is disabled in order that avoid
+ * to risk increase the memory footprint of applications without a guaranteed
+ * benefit. When transparent hugepage support is enabled, is for all mappings,
+ * and khugepaged scans all mappings.
+ * Defrag is invoked by khugepaged hugepage allocations and by page faults
+ * for all hugepage allocations.
  */
 unsigned long transparent_hugepage_flags __read_mostly =
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE_ALWAYS
@@ -709,6 +710,7 @@ static int __do_huge_pmd_anonymous_page(struct mm_struct *mm,
 					struct page *page)
 {
 	pgtable_t pgtable;
+	spinlock_t *ptl;
 
 	VM_BUG_ON(!PageCompound(page));
 	pgtable = pte_alloc_one(mm, haddr);
@@ -723,9 +725,9 @@ static int __do_huge_pmd_anonymous_page(struct mm_struct *mm,
 	 */
 	__SetPageUptodate(page);
 
-	spin_lock(&mm->page_table_lock);
+	ptl = pmd_lock(mm, pmd);
 	if (unlikely(!pmd_none(*pmd))) {
-		spin_unlock(&mm->page_table_lock);
+		spin_unlock(ptl);
 		mem_cgroup_uncharge_page(page);
 		put_page(page);
 		pte_free(mm, pgtable);
@@ -737,8 +739,8 @@ static int __do_huge_pmd_anonymous_page(struct mm_struct *mm,
 		pgtable_trans_huge_deposit(mm, pmd, pgtable);
 		set_pmd_at(mm, haddr, pmd, entry);
 		add_mm_counter(mm, MM_ANONPAGES, HPAGE_PMD_NR);
-		mm->nr_ptes++;
-		spin_unlock(&mm->page_table_lock);
+		atomic_long_inc(&mm->nr_ptes);
+		spin_unlock(ptl);
 	}
 
 	return 0;
@@ -758,14 +760,7 @@ static inline struct page *alloc_hugepage_vma(int defrag,
 			       HPAGE_PMD_ORDER, vma, haddr, nd);
 }
 
-#ifndef CONFIG_NUMA
-static inline struct page *alloc_hugepage(int defrag)
-{
-	return alloc_pages(alloc_hugepage_gfpmask(defrag, 0),
-			   HPAGE_PMD_ORDER);
-}
-#endif
-
+/* Caller must hold page table lock. */
 static bool set_huge_zero_page(pgtable_t pgtable, struct mm_struct *mm,
 		struct vm_area_struct *vma, unsigned long haddr, pmd_t *pmd,
 		struct page *zero_page)
@@ -778,7 +773,7 @@ static bool set_huge_zero_page(pgtable_t pgtable, struct mm_struct *mm,
 	entry = pmd_mkhuge(entry);
 	pgtable_trans_huge_deposit(mm, pmd, pgtable);
 	set_pmd_at(mm, haddr, pmd, entry);
-	mm->nr_ptes++;
+	atomic_long_inc(&mm->nr_ptes);
 	return true;
 }
 
@@ -797,6 +792,7 @@ int do_huge_pmd_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
 		return VM_FAULT_OOM;
 	if (!(flags & FAULT_FLAG_WRITE) &&
 			transparent_hugepage_use_zero_page()) {
+		spinlock_t *ptl;
 		pgtable_t pgtable;
 		struct page *zero_page;
 		bool set;
@@ -809,10 +805,10 @@ int do_huge_pmd_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
 			count_vm_event(THP_FAULT_FALLBACK);
 			return VM_FAULT_FALLBACK;
 		}
-		spin_lock(&mm->page_table_lock);
+		ptl = pmd_lock(mm, pmd);
 		set = set_huge_zero_page(pgtable, mm, vma, haddr, pmd,
 				zero_page);
-		spin_unlock(&mm->page_table_lock);
+		spin_unlock(ptl);
 		if (!set) {
 			pte_free(mm, pgtable);
 			put_huge_zero_page();
@@ -845,6 +841,7 @@ int copy_huge_pmd(struct mm_struct *dst_mm, struct mm_struct *src_mm,
 		  pmd_t *dst_pmd, pmd_t *src_pmd, unsigned long addr,
 		  struct vm_area_struct *vma)
 {
+	spinlock_t *dst_ptl, *src_ptl;
 	struct page *src_page;
 	pmd_t pmd;
 	pgtable_t pgtable;
@@ -855,8 +852,9 @@ int copy_huge_pmd(struct mm_struct *dst_mm, struct mm_struct *src_mm,
 	if (unlikely(!pgtable))
 		goto out;
 
-	spin_lock(&dst_mm->page_table_lock);
-	spin_lock_nested(&src_mm->page_table_lock, SINGLE_DEPTH_NESTING);
+	dst_ptl = pmd_lock(dst_mm, dst_pmd);
+	src_ptl = pmd_lockptr(src_mm, src_pmd);
+	spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
 
 	ret = -EAGAIN;
 	pmd = *src_pmd;
@@ -865,7 +863,7 @@ int copy_huge_pmd(struct mm_struct *dst_mm, struct mm_struct *src_mm,
 		goto out_unlock;
 	}
 	/*
-	 * mm->page_table_lock is enough to be sure that huge zero pmd is not
+	 * When page table lock is held, the huge zero pmd should not be
 	 * under splitting since we don't split the page itself, only pmd to
 	 * a page table.
 	 */
@@ -884,10 +882,11 @@ int copy_huge_pmd(struct mm_struct *dst_mm, struct mm_struct *src_mm,
 		ret = 0;
 		goto out_unlock;
 	}
+
 	if (unlikely(pmd_trans_splitting(pmd))) {
 		/* split huge page running from under us */
-		spin_unlock(&src_mm->page_table_lock);
-		spin_unlock(&dst_mm->page_table_lock);
+		spin_unlock(src_ptl);
+		spin_unlock(dst_ptl);
 		pte_free(dst_mm, pgtable);
 
 		wait_split_huge_page(vma->anon_vma, src_pmd); /* src_vma */
@@ -903,12 +902,12 @@ int copy_huge_pmd(struct mm_struct *dst_mm, struct mm_struct *src_mm,
 	pmd = pmd_mkold(pmd_wrprotect(pmd));
 	pgtable_trans_huge_deposit(dst_mm, dst_pmd, pgtable);
 	set_pmd_at(dst_mm, addr, dst_pmd, pmd);
-	dst_mm->nr_ptes++;
+	atomic_long_inc(&dst_mm->nr_ptes);
 
 	ret = 0;
 out_unlock:
-	spin_unlock(&src_mm->page_table_lock);
-	spin_unlock(&dst_mm->page_table_lock);
+	spin_unlock(src_ptl);
+	spin_unlock(dst_ptl);
 out:
 	return ret;
 }
@@ -919,10 +918,11 @@ void huge_pmd_set_accessed(struct mm_struct *mm,
 			   pmd_t *pmd, pmd_t orig_pmd,
 			   int dirty)
 {
+	spinlock_t *ptl;
 	pmd_t entry;
 	unsigned long haddr;
 
-	spin_lock(&mm->page_table_lock);
+	ptl = pmd_lock(mm, pmd);
 	if (unlikely(!pmd_same(*pmd, orig_pmd)))
 		goto unlock;
 
@@ -932,13 +932,14 @@ void huge_pmd_set_accessed(struct mm_struct *mm,
 		update_mmu_cache_pmd(vma, address, pmd);
 
 unlock:
-	spin_unlock(&mm->page_table_lock);
+	spin_unlock(ptl);
 }
 
 static int do_huge_pmd_wp_zero_page_fallback(struct mm_struct *mm,
 		struct vm_area_struct *vma, unsigned long address,
 		pmd_t *pmd, pmd_t orig_pmd, unsigned long haddr)
 {
+	spinlock_t *ptl;
 	pgtable_t pgtable;
 	pmd_t _pmd;
 	struct page *page;
@@ -965,7 +966,7 @@ static int do_huge_pmd_wp_zero_page_fallback(struct mm_struct *mm,
 	mmun_end   = haddr + HPAGE_PMD_SIZE;
 	mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
 
-	spin_lock(&mm->page_table_lock);
+	ptl = pmd_lock(mm, pmd);
 	if (unlikely(!pmd_same(*pmd, orig_pmd)))
 		goto out_free_page;
 
@@ -992,7 +993,7 @@ static int do_huge_pmd_wp_zero_page_fallback(struct mm_struct *mm,
 	}
 	smp_wmb(); /* make pte visible before pmd */
 	pmd_populate(mm, pmd, pgtable);
-	spin_unlock(&mm->page_table_lock);
+	spin_unlock(ptl);
 	put_huge_zero_page();
 	inc_mm_counter(mm, MM_ANONPAGES);
 
@@ -1002,7 +1003,7 @@ static int do_huge_pmd_wp_zero_page_fallback(struct mm_struct *mm,
 out:
 	return ret;
 out_free_page:
-	spin_unlock(&mm->page_table_lock);
+	spin_unlock(ptl);
 	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
 	mem_cgroup_uncharge_page(page);
 	put_page(page);
@@ -1016,6 +1017,7 @@ static int do_huge_pmd_wp_page_fallback(struct mm_struct *mm,
 					struct page *page,
 					unsigned long haddr)
 {
+	spinlock_t *ptl;
 	pgtable_t pgtable;
 	pmd_t _pmd;
 	int ret = 0, i;
@@ -1062,7 +1064,7 @@ static int do_huge_pmd_wp_page_fallback(struct mm_struct *mm,
 	mmun_end   = haddr + HPAGE_PMD_SIZE;
 	mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
 
-	spin_lock(&mm->page_table_lock);
+	ptl = pmd_lock(mm, pmd);
 	if (unlikely(!pmd_same(*pmd, orig_pmd)))
 		goto out_free_pages;
 	VM_BUG_ON(!PageHead(page));
@@ -1088,7 +1090,7 @@ static int do_huge_pmd_wp_page_fallback(struct mm_struct *mm,
 	smp_wmb(); /* make pte visible before pmd */
 	pmd_populate(mm, pmd, pgtable);
 	page_remove_rmap(page);
-	spin_unlock(&mm->page_table_lock);
+	spin_unlock(ptl);
 
 	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
 
@@ -1099,7 +1101,7 @@ out:
 	return ret;
 
 out_free_pages:
-	spin_unlock(&mm->page_table_lock);
+	spin_unlock(ptl);
 	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
 	mem_cgroup_uncharge_start();
 	for (i = 0; i < HPAGE_PMD_NR; i++) {
@@ -1114,17 +1116,19 @@ out_free_pages:
 int do_huge_pmd_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
 			unsigned long address, pmd_t *pmd, pmd_t orig_pmd)
 {
+	spinlock_t *ptl;
 	int ret = 0;
 	struct page *page = NULL, *new_page;
 	unsigned long haddr;
 	unsigned long mmun_start;	/* For mmu_notifiers */
 	unsigned long mmun_end;		/* For mmu_notifiers */
 
+	ptl = pmd_lockptr(mm, pmd);
 	VM_BUG_ON(!vma->anon_vma);
 	haddr = address & HPAGE_PMD_MASK;
 	if (is_huge_zero_pmd(orig_pmd))
 		goto alloc;
-	spin_lock(&mm->page_table_lock);
+	spin_lock(ptl);
 	if (unlikely(!pmd_same(*pmd, orig_pmd)))
 		goto out_unlock;
 
@@ -1140,7 +1144,7 @@ int do_huge_pmd_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
 		goto out_unlock;
 	}
 	get_page(page);
-	spin_unlock(&mm->page_table_lock);
+	spin_unlock(ptl);
 alloc:
 	if (transparent_hugepage_enabled(vma) &&
 	    !transparent_hugepage_debug_cow())
@@ -1150,7 +1154,7 @@ alloc:
 		new_page = NULL;
 
 	if (unlikely(!new_page)) {
-		if (is_huge_zero_pmd(orig_pmd)) {
+		if (!page) {
 			ret = do_huge_pmd_wp_zero_page_fallback(mm, vma,
 					address, pmd, orig_pmd, haddr);
 		} else {
@@ -1177,7 +1181,7 @@ alloc:
 
 	count_vm_event(THP_FAULT_ALLOC);
 
-	if (is_huge_zero_pmd(orig_pmd))
+	if (!page)
 		clear_huge_page(new_page, haddr, HPAGE_PMD_NR);
 	else
 		copy_user_huge_page(new_page, page, haddr, vma, HPAGE_PMD_NR);
@@ -1187,11 +1191,11 @@ alloc:
 	mmun_end   = haddr + HPAGE_PMD_SIZE;
 	mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
 
-	spin_lock(&mm->page_table_lock);
+	spin_lock(ptl);
 	if (page)
 		put_page(page);
 	if (unlikely(!pmd_same(*pmd, orig_pmd))) {
-		spin_unlock(&mm->page_table_lock);
+		spin_unlock(ptl);
 		mem_cgroup_uncharge_page(new_page);
 		put_page(new_page);
 		goto out_mn;
@@ -1203,7 +1207,7 @@ alloc:
 		page_add_new_anon_rmap(new_page, vma, haddr);
 		set_pmd_at(mm, haddr, pmd, entry);
 		update_mmu_cache_pmd(vma, address, pmd);
-		if (is_huge_zero_pmd(orig_pmd)) {
+		if (!page) {
 			add_mm_counter(mm, MM_ANONPAGES, HPAGE_PMD_NR);
 			put_huge_zero_page();
 		} else {
@@ -1213,13 +1217,13 @@ alloc:
 		}
 		ret |= VM_FAULT_WRITE;
 	}
-	spin_unlock(&mm->page_table_lock);
+	spin_unlock(ptl);
 out_mn:
 	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
 out:
 	return ret;
 out_unlock:
-	spin_unlock(&mm->page_table_lock);
+	spin_unlock(ptl);
 	return ret;
 }
 
@@ -1231,7 +1235,7 @@ struct page *follow_trans_huge_pmd(struct vm_area_struct *vma,
 	struct mm_struct *mm = vma->vm_mm;
 	struct page *page = NULL;
 
-	assert_spin_locked(&mm->page_table_lock);
+	assert_spin_locked(pmd_lockptr(mm, pmd));
 
 	if (flags & FOLL_WRITE && !pmd_write(*pmd))
 		goto out;
@@ -1240,6 +1244,10 @@ struct page *follow_trans_huge_pmd(struct vm_area_struct *vma,
 	if ((flags & FOLL_DUMP) && is_huge_zero_pmd(*pmd))
 		return ERR_PTR(-EFAULT);
 
+	/* Full NUMA hinting faults to serialise migration in fault paths */
+	if ((flags & FOLL_NUMA) && pmd_numa(*pmd))
+		goto out;
+
 	page = pmd_page(*pmd);
 	VM_BUG_ON(!PageHead(page));
 	if (flags & FOLL_TOUCH) {
@@ -1278,23 +1286,48 @@ out:
 int do_huge_pmd_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,
 				unsigned long addr, pmd_t pmd, pmd_t *pmdp)
 {
+	spinlock_t *ptl;
 	struct anon_vma *anon_vma = NULL;
 	struct page *page;
 	unsigned long haddr = addr & HPAGE_PMD_MASK;
 	int page_nid = -1, this_nid = numa_node_id();
-	int target_nid;
+	int target_nid, last_cpupid = -1;
 	bool page_locked;
 	bool migrated = false;
+	int flags = 0;
 
-	spin_lock(&mm->page_table_lock);
+	ptl = pmd_lock(mm, pmdp);
 	if (unlikely(!pmd_same(pmd, *pmdp)))
 		goto out_unlock;
 
+	/*
+	 * If there are potential migrations, wait for completion and retry
+	 * without disrupting NUMA hinting information. Do not relock and
+	 * check_same as the page may no longer be mapped.
+	 */
+	if (unlikely(pmd_trans_migrating(*pmdp))) {
+		spin_unlock(ptl);
+		wait_migrate_huge_page(vma->anon_vma, pmdp);
+		goto out;
+	}
+
 	page = pmd_page(pmd);
+	BUG_ON(is_huge_zero_page(page));
 	page_nid = page_to_nid(page);
+	last_cpupid = page_cpupid_last(page);
 	count_vm_numa_event(NUMA_HINT_FAULTS);
-	if (page_nid == this_nid)
+	if (page_nid == this_nid) {
 		count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL);
+		flags |= TNF_FAULT_LOCAL;
+	}
+
+	/*
+	 * Avoid grouping on DSO/COW pages in specific and RO pages
+	 * in general, RO pages shouldn't hurt as much anyway since
+	 * they can be in shared cache state.
+	 */
+	if (!pmd_write(pmd))
+		flags |= TNF_NO_GROUP;
 
 	/*
 	 * Acquire the page lock to serialise THP migrations but avoid dropping
@@ -1306,27 +1339,26 @@ int do_huge_pmd_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,
 		/* If the page was locked, there are no parallel migrations */
 		if (page_locked)
 			goto clear_pmdnuma;
+	}
 
-		/*
-		 * Otherwise wait for potential migrations and retry. We do
-		 * relock and check_same as the page may no longer be mapped.
-		 * As the fault is being retried, do not account for it.
-		 */
-		spin_unlock(&mm->page_table_lock);
+	/* Migration could have started since the pmd_trans_migrating check */
+	if (!page_locked) {
+		spin_unlock(ptl);
 		wait_on_page_locked(page);
 		page_nid = -1;
 		goto out;
 	}
 
-	/* Page is misplaced, serialise migrations and parallel THP splits */
+	/*
+	 * Page is misplaced. Page lock serialises migrations. Acquire anon_vma
+	 * to serialises splits
+	 */
 	get_page(page);
-	spin_unlock(&mm->page_table_lock);
-	if (!page_locked)
-		lock_page(page);
+	spin_unlock(ptl);
 	anon_vma = page_lock_anon_vma_read(page);
 
-	/* Confirm the PTE did not while locked */
-	spin_lock(&mm->page_table_lock);
+	/* Confirm the PMD did not change while page_table_lock was released */
+	spin_lock(ptl);
 	if (unlikely(!pmd_same(pmd, *pmdp))) {
 		unlock_page(page);
 		put_page(page);
@@ -1334,15 +1366,24 @@ int do_huge_pmd_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,
 		goto out_unlock;
 	}
 
+	/* Bail if we fail to protect against THP splits for any reason */
+	if (unlikely(!anon_vma)) {
+		put_page(page);
+		page_nid = -1;
+		goto clear_pmdnuma;
+	}
+
 	/*
 	 * Migrate the THP to the requested node, returns with page unlocked
 	 * and pmd_numa cleared.
 	 */
-	spin_unlock(&mm->page_table_lock);
+	spin_unlock(ptl);
 	migrated = migrate_misplaced_transhuge_page(mm, vma,
 				pmdp, pmd, addr, page, target_nid);
-	if (migrated)
+	if (migrated) {
+		flags |= TNF_MIGRATED;
 		page_nid = target_nid;
+	}
 
 	goto out;
 clear_pmdnuma:
@@ -1353,14 +1394,14 @@ clear_pmdnuma:
 	update_mmu_cache_pmd(vma, addr, pmdp);
 	unlock_page(page);
 out_unlock:
-	spin_unlock(&mm->page_table_lock);
+	spin_unlock(ptl);
 
 out:
 	if (anon_vma)
 		page_unlock_anon_vma_read(anon_vma);
 
 	if (page_nid != -1)
-		task_numa_fault(page_nid, HPAGE_PMD_NR, migrated);
+		task_numa_fault(last_cpupid, page_nid, HPAGE_PMD_NR, flags);
 
 	return 0;
 }
@@ -1368,9 +1409,10 @@ out:
 int zap_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
 		 pmd_t *pmd, unsigned long addr)
 {
+	spinlock_t *ptl;
 	int ret = 0;
 
-	if (__pmd_trans_huge_lock(pmd, vma) == 1) {
+	if (__pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
 		struct page *page;
 		pgtable_t pgtable;
 		pmd_t orig_pmd;
@@ -1384,8 +1426,8 @@ int zap_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
 		tlb_remove_pmd_tlb_entry(tlb, pmd, addr);
 		pgtable = pgtable_trans_huge_withdraw(tlb->mm, pmd);
 		if (is_huge_zero_pmd(orig_pmd)) {
-			tlb->mm->nr_ptes--;
-			spin_unlock(&tlb->mm->page_table_lock);
+			atomic_long_dec(&tlb->mm->nr_ptes);
+			spin_unlock(ptl);
 			put_huge_zero_page();
 		} else {
 			page = pmd_page(orig_pmd);
@@ -1393,8 +1435,8 @@ int zap_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
 			VM_BUG_ON(page_mapcount(page) < 0);
 			add_mm_counter(tlb->mm, MM_ANONPAGES, -HPAGE_PMD_NR);
 			VM_BUG_ON(!PageHead(page));
-			tlb->mm->nr_ptes--;
-			spin_unlock(&tlb->mm->page_table_lock);
+			atomic_long_dec(&tlb->mm->nr_ptes);
+			spin_unlock(ptl);
 			tlb_remove_page(tlb, page);
 		}
 		pte_free(tlb->mm, pgtable);
@@ -1407,14 +1449,15 @@ int mincore_huge_pmd(struct vm_area_struct *vma, pmd_t *pmd,
 		unsigned long addr, unsigned long end,
 		unsigned char *vec)
 {
+	spinlock_t *ptl;
 	int ret = 0;
 
-	if (__pmd_trans_huge_lock(pmd, vma) == 1) {
+	if (__pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
 		/*
 		 * All logical pages in the range are present
 		 * if backed by a huge page.
 		 */
-		spin_unlock(&vma->vm_mm->page_table_lock);
+		spin_unlock(ptl);
 		memset(vec, 1, (end - addr) >> PAGE_SHIFT);
 		ret = 1;
 	}
@@ -1427,6 +1470,7 @@ int move_huge_pmd(struct vm_area_struct *vma, struct vm_area_struct *new_vma,
 		  unsigned long new_addr, unsigned long old_end,
 		  pmd_t *old_pmd, pmd_t *new_pmd)
 {
+	spinlock_t *old_ptl, *new_ptl;
 	int ret = 0;
 	pmd_t pmd;
 
@@ -1447,41 +1491,81 @@ int move_huge_pmd(struct vm_area_struct *vma, struct vm_area_struct *new_vma,
 		goto out;
 	}
 
-	ret = __pmd_trans_huge_lock(old_pmd, vma);
+	/*
+	 * We don't have to worry about the ordering of src and dst
+	 * ptlocks because exclusive mmap_sem prevents deadlock.
+	 */
+	ret = __pmd_trans_huge_lock(old_pmd, vma, &old_ptl);
 	if (ret == 1) {
+		new_ptl = pmd_lockptr(mm, new_pmd);
+		if (new_ptl != old_ptl)
+			spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
 		pmd = pmdp_get_and_clear(mm, old_addr, old_pmd);
 		VM_BUG_ON(!pmd_none(*new_pmd));
 		set_pmd_at(mm, new_addr, new_pmd, pmd_mksoft_dirty(pmd));
-		spin_unlock(&mm->page_table_lock);
+		if (new_ptl != old_ptl) {
+			pgtable_t pgtable;
+
+			/*
+			 * Move preallocated PTE page table if new_pmd is on
+			 * different PMD page table.
+			 */
+			pgtable = pgtable_trans_huge_withdraw(mm, old_pmd);
+			pgtable_trans_huge_deposit(mm, new_pmd, pgtable);
+
+			spin_unlock(new_ptl);
+		}
+		spin_unlock(old_ptl);
 	}
 out:
 	return ret;
 }
 
+/*
+ * Returns
+ *  - 0 if PMD could not be locked
+ *  - 1 if PMD was locked but protections unchange and TLB flush unnecessary
+ *  - HPAGE_PMD_NR is protections changed and TLB flush necessary
+ */
 int change_huge_pmd(struct vm_area_struct *vma, pmd_t *pmd,
 		unsigned long addr, pgprot_t newprot, int prot_numa)
 {
 	struct mm_struct *mm = vma->vm_mm;
+	spinlock_t *ptl;
 	int ret = 0;
 
-	if (__pmd_trans_huge_lock(pmd, vma) == 1) {
+	if (__pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
 		pmd_t entry;
-		entry = pmdp_get_and_clear(mm, addr, pmd);
+		ret = 1;
 		if (!prot_numa) {
+			entry = pmdp_get_and_clear(mm, addr, pmd);
+			if (pmd_numa(entry))
+				entry = pmd_mknonnuma(entry);
 			entry = pmd_modify(entry, newprot);
+			ret = HPAGE_PMD_NR;
 			BUG_ON(pmd_write(entry));
 		} else {
 			struct page *page = pmd_page(*pmd);
 
-			/* only check non-shared pages */
-			if (page_mapcount(page) == 1 &&
+			/*
+			 * Do not trap faults against the zero page. The
+			 * read-only data is likely to be read-cached on the
+			 * local CPU cache and it is less useful to know about
+			 * local vs remote hits on the zero page.
+			 */
+			if (!is_huge_zero_page(page) &&
 			    !pmd_numa(*pmd)) {
+				entry = *pmd;
 				entry = pmd_mknuma(entry);
+				ret = HPAGE_PMD_NR;
 			}
 		}
-		set_pmd_at(mm, addr, pmd, entry);
-		spin_unlock(&vma->vm_mm->page_table_lock);
-		ret = 1;
+
+		/* Set PMD if cleared earlier */
+		if (ret == HPAGE_PMD_NR)
+			set_pmd_at(mm, addr, pmd, entry);
+
+		spin_unlock(ptl);
 	}
 
 	return ret;
@@ -1494,12 +1578,13 @@ int change_huge_pmd(struct vm_area_struct *vma, pmd_t *pmd,
  * Note that if it returns 1, this routine returns without unlocking page
  * table locks. So callers must unlock them.
  */
-int __pmd_trans_huge_lock(pmd_t *pmd, struct vm_area_struct *vma)
+int __pmd_trans_huge_lock(pmd_t *pmd, struct vm_area_struct *vma,
+		spinlock_t **ptl)
 {
-	spin_lock(&vma->vm_mm->page_table_lock);
+	*ptl = pmd_lock(vma->vm_mm, pmd);
 	if (likely(pmd_trans_huge(*pmd))) {
 		if (unlikely(pmd_trans_splitting(*pmd))) {
-			spin_unlock(&vma->vm_mm->page_table_lock);
+			spin_unlock(*ptl);
 			wait_split_huge_page(vma->anon_vma, pmd);
 			return -1;
 		} else {
@@ -1508,27 +1593,37 @@ int __pmd_trans_huge_lock(pmd_t *pmd, struct vm_area_struct *vma)
 			return 1;
 		}
 	}
-	spin_unlock(&vma->vm_mm->page_table_lock);
+	spin_unlock(*ptl);
 	return 0;
 }
 
+/*
+ * This function returns whether a given @page is mapped onto the @address
+ * in the virtual space of @mm.
+ *
+ * When it's true, this function returns *pmd with holding the page table lock
+ * and passing it back to the caller via @ptl.
+ * If it's false, returns NULL without holding the page table lock.
+ */
 pmd_t *page_check_address_pmd(struct page *page,
 			      struct mm_struct *mm,
 			      unsigned long address,
-			      enum page_check_address_pmd_flag flag)
+			      enum page_check_address_pmd_flag flag,
+			      spinlock_t **ptl)
 {
-	pmd_t *pmd, *ret = NULL;
+	pmd_t *pmd;
 
 	if (address & ~HPAGE_PMD_MASK)
-		goto out;
+		return NULL;
 
 	pmd = mm_find_pmd(mm, address);
 	if (!pmd)
-		goto out;
+		return NULL;
+	*ptl = pmd_lock(mm, pmd);
 	if (pmd_none(*pmd))
-		goto out;
+		goto unlock;
 	if (pmd_page(*pmd) != page)
-		goto out;
+		goto unlock;
 	/*
 	 * split_vma() may create temporary aliased mappings. There is
 	 * no risk as long as all huge pmd are found and have their
@@ -1538,14 +1633,15 @@ pmd_t *page_check_address_pmd(struct page *page,
 	 */
 	if (flag == PAGE_CHECK_ADDRESS_PMD_NOTSPLITTING_FLAG &&
 	    pmd_trans_splitting(*pmd))
-		goto out;
+		goto unlock;
 	if (pmd_trans_huge(*pmd)) {
 		VM_BUG_ON(flag == PAGE_CHECK_ADDRESS_PMD_SPLITTING_FLAG &&
 			  !pmd_trans_splitting(*pmd));
-		ret = pmd;
+		return pmd;
 	}
-out:
-	return ret;
+unlock:
+	spin_unlock(*ptl);
+	return NULL;
 }
 
 static int __split_huge_page_splitting(struct page *page,
@@ -1553,6 +1649,7 @@ static int __split_huge_page_splitting(struct page *page,
 				       unsigned long address)
 {
 	struct mm_struct *mm = vma->vm_mm;
+	spinlock_t *ptl;
 	pmd_t *pmd;
 	int ret = 0;
 	/* For mmu_notifiers */
@@ -1560,9 +1657,8 @@ static int __split_huge_page_splitting(struct page *page,
 	const unsigned long mmun_end   = address + HPAGE_PMD_SIZE;
 
 	mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
-	spin_lock(&mm->page_table_lock);
 	pmd = page_check_address_pmd(page, mm, address,
-				     PAGE_CHECK_ADDRESS_PMD_NOTSPLITTING_FLAG);
+			PAGE_CHECK_ADDRESS_PMD_NOTSPLITTING_FLAG, &ptl);
 	if (pmd) {
 		/*
 		 * We can't temporarily set the pmd to null in order
@@ -1573,8 +1669,8 @@ static int __split_huge_page_splitting(struct page *page,
 		 */
 		pmdp_splitting_flush(vma, address, pmd);
 		ret = 1;
+		spin_unlock(ptl);
 	}
-	spin_unlock(&mm->page_table_lock);
 	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
 
 	return ret;
@@ -1662,7 +1758,7 @@ static void __split_huge_page_refcount(struct page *page,
 		page_tail->mapping = page->mapping;
 
 		page_tail->index = page->index + i;
-		page_nid_xchg_last(page_tail, page_nid_last(page));
+		page_cpupid_xchg_last(page_tail, page_cpupid_last(page));
 
 		BUG_ON(!PageAnon(page_tail));
 		BUG_ON(!PageUptodate(page_tail));
@@ -1705,14 +1801,14 @@ static int __split_huge_page_map(struct page *page,
 				 unsigned long address)
 {
 	struct mm_struct *mm = vma->vm_mm;
+	spinlock_t *ptl;
 	pmd_t *pmd, _pmd;
 	int ret = 0, i;
 	pgtable_t pgtable;
 	unsigned long haddr;
 
-	spin_lock(&mm->page_table_lock);
 	pmd = page_check_address_pmd(page, mm, address,
-				     PAGE_CHECK_ADDRESS_PMD_SPLITTING_FLAG);
+			PAGE_CHECK_ADDRESS_PMD_SPLITTING_FLAG, &ptl);
 	if (pmd) {
 		pgtable = pgtable_trans_huge_withdraw(mm, pmd);
 		pmd_populate(mm, &_pmd, pgtable);
@@ -1767,8 +1863,8 @@ static int __split_huge_page_map(struct page *page,
 		pmdp_invalidate(vma, address, pmd);
 		pmd_populate(mm, pmd, pgtable);
 		ret = 1;
+		spin_unlock(ptl);
 	}
-	spin_unlock(&mm->page_table_lock);
 
 	return ret;
 }
@@ -2165,7 +2261,34 @@ static void khugepaged_alloc_sleep(void)
 			msecs_to_jiffies(khugepaged_alloc_sleep_millisecs));
 }
 
+static int khugepaged_node_load[MAX_NUMNODES];
+
 #ifdef CONFIG_NUMA
+static int khugepaged_find_target_node(void)
+{
+	static int last_khugepaged_target_node = NUMA_NO_NODE;
+	int nid, target_node = 0, max_value = 0;
+
+	/* find first node with max normal pages hit */
+	for (nid = 0; nid < MAX_NUMNODES; nid++)
+		if (khugepaged_node_load[nid] > max_value) {
+			max_value = khugepaged_node_load[nid];
+			target_node = nid;
+		}
+
+	/* do some balance if several nodes have the same hit record */
+	if (target_node <= last_khugepaged_target_node)
+		for (nid = last_khugepaged_target_node + 1; nid < MAX_NUMNODES;
+				nid++)
+			if (max_value == khugepaged_node_load[nid]) {
+				target_node = nid;
+				break;
+			}
+
+	last_khugepaged_target_node = target_node;
+	return target_node;
+}
+
 static bool khugepaged_prealloc_page(struct page **hpage, bool *wait)
 {
 	if (IS_ERR(*hpage)) {
@@ -2199,9 +2322,8 @@ static struct page
 	 * mmap_sem in read mode is good idea also to allow greater
 	 * scalability.
 	 */
-	*hpage  = alloc_hugepage_vma(khugepaged_defrag(), vma, address,
-				      node, __GFP_OTHER_NODE);
-
+	*hpage = alloc_pages_exact_node(node, alloc_hugepage_gfpmask(
+		khugepaged_defrag(), __GFP_OTHER_NODE), HPAGE_PMD_ORDER);
 	/*
 	 * After allocating the hugepage, release the mmap_sem read lock in
 	 * preparation for taking it in write mode.
@@ -2217,6 +2339,17 @@ static struct page
 	return *hpage;
 }
 #else
+static int khugepaged_find_target_node(void)
+{
+	return 0;
+}
+
+static inline struct page *alloc_hugepage(int defrag)
+{
+	return alloc_pages(alloc_hugepage_gfpmask(defrag, 0),
+			   HPAGE_PMD_ORDER);
+}
+
 static struct page *khugepaged_alloc_hugepage(bool *wait)
 {
 	struct page *hpage;
@@ -2283,7 +2416,7 @@ static void collapse_huge_page(struct mm_struct *mm,
 	pte_t *pte;
 	pgtable_t pgtable;
 	struct page *new_page;
-	spinlock_t *ptl;
+	spinlock_t *pmd_ptl, *pte_ptl;
 	int isolated;
 	unsigned long hstart, hend;
 	unsigned long mmun_start;	/* For mmu_notifiers */
@@ -2326,12 +2459,12 @@ static void collapse_huge_page(struct mm_struct *mm,
 	anon_vma_lock_write(vma->anon_vma);
 
 	pte = pte_offset_map(pmd, address);
-	ptl = pte_lockptr(mm, pmd);
+	pte_ptl = pte_lockptr(mm, pmd);
 
 	mmun_start = address;
 	mmun_end   = address + HPAGE_PMD_SIZE;
 	mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
-	spin_lock(&mm->page_table_lock); /* probably unnecessary */
+	pmd_ptl = pmd_lock(mm, pmd); /* probably unnecessary */
 	/*
 	 * After this gup_fast can't run anymore. This also removes
 	 * any huge TLB entry from the CPU so we won't allow
@@ -2339,16 +2472,16 @@ static void collapse_huge_page(struct mm_struct *mm,
 	 * to avoid the risk of CPU bugs in that area.
 	 */
 	_pmd = pmdp_clear_flush(vma, address, pmd);
-	spin_unlock(&mm->page_table_lock);
+	spin_unlock(pmd_ptl);
 	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
 
-	spin_lock(ptl);
+	spin_lock(pte_ptl);
 	isolated = __collapse_huge_page_isolate(vma, address, pte);
-	spin_unlock(ptl);
+	spin_unlock(pte_ptl);
 
 	if (unlikely(!isolated)) {
 		pte_unmap(pte);
-		spin_lock(&mm->page_table_lock);
+		spin_lock(pmd_ptl);
 		BUG_ON(!pmd_none(*pmd));
 		/*
 		 * We can only use set_pmd_at when establishing
@@ -2356,7 +2489,7 @@ static void collapse_huge_page(struct mm_struct *mm,
 		 * points to regular pagetables. Use pmd_populate for that
 		 */
 		pmd_populate(mm, pmd, pmd_pgtable(_pmd));
-		spin_unlock(&mm->page_table_lock);
+		spin_unlock(pmd_ptl);
 		anon_vma_unlock_write(vma->anon_vma);
 		goto out;
 	}
@@ -2367,7 +2500,7 @@ static void collapse_huge_page(struct mm_struct *mm,
 	 */
 	anon_vma_unlock_write(vma->anon_vma);
 
-	__collapse_huge_page_copy(pte, new_page, vma, address, ptl);
+	__collapse_huge_page_copy(pte, new_page, vma, address, pte_ptl);
 	pte_unmap(pte);
 	__SetPageUptodate(new_page);
 	pgtable = pmd_pgtable(_pmd);
@@ -2382,13 +2515,13 @@ static void collapse_huge_page(struct mm_struct *mm,
 	 */
 	smp_wmb();
 
-	spin_lock(&mm->page_table_lock);
+	spin_lock(pmd_ptl);
 	BUG_ON(!pmd_none(*pmd));
 	page_add_new_anon_rmap(new_page, vma, address);
 	pgtable_trans_huge_deposit(mm, pmd, pgtable);
 	set_pmd_at(mm, address, pmd, _pmd);
 	update_mmu_cache_pmd(vma, address, pmd);
-	spin_unlock(&mm->page_table_lock);
+	spin_unlock(pmd_ptl);
 
 	*hpage = NULL;
 
@@ -2423,6 +2556,7 @@ static int khugepaged_scan_pmd(struct mm_struct *mm,
 	if (pmd_trans_huge(*pmd))
 		goto out;
 
+	memset(khugepaged_node_load, 0, sizeof(khugepaged_node_load));
 	pte = pte_offset_map_lock(mm, pmd, address, &ptl);
 	for (_address = address, _pte = pte; _pte < pte+HPAGE_PMD_NR;
 	     _pte++, _address += PAGE_SIZE) {
@@ -2439,12 +2573,13 @@ static int khugepaged_scan_pmd(struct mm_struct *mm,
 		if (unlikely(!page))
 			goto out_unmap;
 		/*
-		 * Chose the node of the first page. This could
-		 * be more sophisticated and look at more pages,
-		 * but isn't for now.
+		 * Record which node the original page is from and save this
+		 * information to khugepaged_node_load[].
+		 * Khupaged will allocate hugepage from the node has the max
+		 * hit record.
 		 */
-		if (node == NUMA_NO_NODE)
-			node = page_to_nid(page);
+		node = page_to_nid(page);
+		khugepaged_node_load[node]++;
 		VM_BUG_ON(PageCompound(page));
 		if (!PageLRU(page) || PageLocked(page) || !PageAnon(page))
 			goto out_unmap;
@@ -2459,9 +2594,11 @@ static int khugepaged_scan_pmd(struct mm_struct *mm,
 		ret = 1;
 out_unmap:
 	pte_unmap_unlock(pte, ptl);
-	if (ret)
+	if (ret) {
+		node = khugepaged_find_target_node();
 		/* collapse_huge_page will return with the mmap_sem released */
 		collapse_huge_page(mm, address, hpage, vma, node);
+	}
 out:
 	return ret;
 }
@@ -2713,6 +2850,7 @@ static void __split_huge_zero_page_pmd(struct vm_area_struct *vma,
 void __split_huge_page_pmd(struct vm_area_struct *vma, unsigned long address,
 		pmd_t *pmd)
 {
+	spinlock_t *ptl;
 	struct page *page;
 	struct mm_struct *mm = vma->vm_mm;
 	unsigned long haddr = address & HPAGE_PMD_MASK;
@@ -2725,22 +2863,22 @@ void __split_huge_page_pmd(struct vm_area_struct *vma, unsigned long address,
 	mmun_end   = haddr + HPAGE_PMD_SIZE;
 again:
 	mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
-	spin_lock(&mm->page_table_lock);
+	ptl = pmd_lock(mm, pmd);
 	if (unlikely(!pmd_trans_huge(*pmd))) {
-		spin_unlock(&mm->page_table_lock);
+		spin_unlock(ptl);
 		mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
 		return;
 	}
 	if (is_huge_zero_pmd(*pmd)) {
 		__split_huge_zero_page_pmd(vma, haddr, pmd);
-		spin_unlock(&mm->page_table_lock);
+		spin_unlock(ptl);
 		mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
 		return;
 	}
 	page = pmd_page(*pmd);
 	VM_BUG_ON(!page_count(page));
 	get_page(page);
-	spin_unlock(&mm->page_table_lock);
+	spin_unlock(ptl);
 	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
 
 	split_huge_page(page);