1 files changed, 40 insertions, 5 deletions

diff --git a/mm/khugepaged.c b/mm/khugepaged.c
index 3c2326568193..f1f98305433e 100644
--- a/mm/khugepaged.c
+++ b/mm/khugepaged.c
@@ -1060,6 +1060,7 @@ static void collapse_huge_page(struct mm_struct *mm,
 	_pmd = pmdp_collapse_flush(vma, address, pmd);
 	spin_unlock(pmd_ptl);
 	mmu_notifier_invalidate_range_end(&range);
+	tlb_remove_table_sync_one();
 
 	spin_lock(pte_ptl);
 	isolated = __collapse_huge_page_isolate(vma, address, pte);
@@ -1312,6 +1313,7 @@ void collapse_pte_mapped_thp(struct mm_struct *mm, unsigned long addr)
 	spinlock_t *ptl;
 	int count = 0;
 	int i;
+	struct mmu_notifier_range range;
 
 	if (!vma || !vma->vm_file ||
 	    vma->vm_start > haddr || vma->vm_end < haddr + HPAGE_PMD_SIZE)
@@ -1338,6 +1340,19 @@ void collapse_pte_mapped_thp(struct mm_struct *mm, unsigned long addr)
 	if (!pmd)
 		goto drop_hpage;
 
+	/*
+	 * We need to lock the mapping so that from here on, only GUP-fast and
+	 * hardware page walks can access the parts of the page tables that
+	 * we're operating on.
+	 */
+	i_mmap_lock_write(vma->vm_file->f_mapping);
+
+	/*
+	 * This spinlock should be unnecessary: Nobody else should be accessing
+	 * the page tables under spinlock protection here, only
+	 * lockless_pages_from_mm() and the hardware page walker can access page
+	 * tables while all the high-level locks are held in write mode.
+	 */
 	start_pte = pte_offset_map_lock(mm, pmd, haddr, &ptl);
 
 	/* step 1: check all mapped PTEs are to the right huge page */
@@ -1384,12 +1399,23 @@ void collapse_pte_mapped_thp(struct mm_struct *mm, unsigned long addr)
 	}
 
 	/* step 4: collapse pmd */
-	ptl = pmd_lock(vma->vm_mm, pmd);
+	/* we make no change to anon, but protect concurrent anon page lookup */
+	if (vma->anon_vma)
+		anon_vma_lock_write(vma->anon_vma);
+
+	mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, NULL, mm, haddr,
+				haddr + HPAGE_PMD_SIZE);
+	mmu_notifier_invalidate_range_start(&range);
 	_pmd = pmdp_collapse_flush(vma, haddr, pmd);
-	spin_unlock(ptl);
 	mm_dec_nr_ptes(mm);
+	tlb_remove_table_sync_one();
+	mmu_notifier_invalidate_range_end(&range);
 	pte_free(mm, pmd_pgtable(_pmd));
 
+	if (vma->anon_vma)
+		anon_vma_unlock_write(vma->anon_vma);
+	i_mmap_unlock_write(vma->vm_file->f_mapping);
+
 drop_hpage:
 	unlock_page(hpage);
 	put_page(hpage);
@@ -1397,6 +1423,7 @@ drop_hpage:
 
 abort:
 	pte_unmap_unlock(start_pte, ptl);
+	i_mmap_unlock_write(vma->vm_file->f_mapping);
 	goto drop_hpage;
 }
 
@@ -1446,7 +1473,8 @@ static void retract_page_tables(struct address_space *mapping, pgoff_t pgoff)
 		 * An alternative would be drop the check, but check that page
 		 * table is clear before calling pmdp_collapse_flush() under
 		 * ptl. It has higher chance to recover THP for the VMA, but
-		 * has higher cost too.
+		 * has higher cost too. It would also probably require locking
+		 * the anon_vma.
 		 */
 		if (vma->anon_vma)
 			continue;
@@ -1468,12 +1496,19 @@ static void retract_page_tables(struct address_space *mapping, pgoff_t pgoff)
 		 */
 		if (down_write_trylock(&mm->mmap_sem)) {
 			if (!khugepaged_test_exit(mm)) {
-				spinlock_t *ptl = pmd_lock(mm, pmd);
+				struct mmu_notifier_range range;
+
+				mmu_notifier_range_init(&range,
+							MMU_NOTIFY_CLEAR, 0,
+							NULL, mm, addr,
+							addr + HPAGE_PMD_SIZE);
+				mmu_notifier_invalidate_range_start(&range);
 				/* assume page table is clear */
 				_pmd = pmdp_collapse_flush(vma, addr, pmd);
-				spin_unlock(ptl);
 				mm_dec_nr_ptes(mm);
+				tlb_remove_table_sync_one();
 				pte_free(mm, pmd_pgtable(_pmd));
+				mmu_notifier_invalidate_range_end(&range);
 			}
 			up_write(&mm->mmap_sem);
 		} else {