[PATCH v6 09/11] mm/mremap: Fix race between mremap and pageout

Aneesh Kumar K.V aneesh.kumar at linux.ibm.com
Mon May 24 19:01:12 AEST 2021


CPU 1				CPU 2					CPU 3

mremap(old_addr, new_addr)      page_shrinker/try_to_unmap_one

				addr = old_addr
				lock(pte_ptl)
lock(pmd_ptl)
pmd = *old_pmd
pmd_clear(old_pmd)
flush_tlb_range(old_addr)

*new_pmd = pmd
									*new_addr = 10; and fills
									TLB with new addr
									and old pfn

unlock(pmd_ptl)
				ptep_get_and_clear()
				flush_tlb_range(old_addr)

				old pfn is free.
									Stale TLB entry

Avoid the above race with MOVE_PMD by holding pte ptl in mremap and waiting for
parallel pagetable walk to finish operating on pte before updating new_pmd

With MOVE_PUD only enable MOVE_PUD only if USE_SPLIT_PTE_PTLOCKS is disabled.
In this case both pte ptl and pud ptl points to mm->page_table_lock.

Fixes: c49dd3401802 ("mm: speedup mremap on 1GB or larger regions")
Fixes: 2c91bd4a4e2e ("mm: speed up mremap by 20x on large regions")
Link: https://lore.kernel.org/linux-mm/CAHk-=wgXVR04eBNtxQfevontWnP6FDm+oj5vauQXP3S-huwbPw@mail.gmail.com
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar at linux.ibm.com>
---
 mm/mremap.c | 24 +++++++++++++++++++++++-
 1 file changed, 23 insertions(+), 1 deletion(-)

diff --git a/mm/mremap.c b/mm/mremap.c
index 8967a3707332..e70b8e3b9568 100644
--- a/mm/mremap.c
+++ b/mm/mremap.c
@@ -224,7 +224,7 @@ static inline void flush_pte_tlb_pwc_range(struct vm_area_struct *vma,
 static bool move_normal_pmd(struct vm_area_struct *vma, unsigned long old_addr,
 		  unsigned long new_addr, pmd_t *old_pmd, pmd_t *new_pmd)
 {
-	spinlock_t *old_ptl, *new_ptl;
+	spinlock_t *pte_ptl, *old_ptl, *new_ptl;
 	struct mm_struct *mm = vma->vm_mm;
 	pmd_t pmd;
 
@@ -254,6 +254,7 @@ static bool move_normal_pmd(struct vm_area_struct *vma, unsigned long old_addr,
 	if (WARN_ON_ONCE(!pmd_none(*new_pmd)))
 		return false;
 
+
 	/*
 	 * We don't have to worry about the ordering of src and dst
 	 * ptlocks because exclusive mmap_lock prevents deadlock.
@@ -263,6 +264,14 @@ static bool move_normal_pmd(struct vm_area_struct *vma, unsigned long old_addr,
 	if (new_ptl != old_ptl)
 		spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
 
+	if (pmd_none(*old_pmd))
+		goto unlock_out;
+
+	/*
+	 * Take the ptl here so that we wait for parallel page table walk
+	 * and operations (eg: pageout)using old addr to finish.
+	 */
+	pte_ptl = pte_lockptr(mm, old_pmd);
 	/* Clear the pmd */
 	pmd = *old_pmd;
 	pmd_clear(old_pmd);
@@ -270,9 +279,14 @@ static bool move_normal_pmd(struct vm_area_struct *vma, unsigned long old_addr,
 	 * flush the TLB before we move the page table entries.
 	 */
 	flush_pte_tlb_pwc_range(vma, old_addr, old_addr + PMD_SIZE);
+
+	spin_lock(pte_ptl);
+
 	VM_BUG_ON(!pmd_none(*new_pmd));
 	pmd_populate(mm, new_pmd, pmd_pgtable(pmd));
+	spin_unlock(pte_ptl);
 
+unlock_out:
 	if (new_ptl != old_ptl)
 		spin_unlock(new_ptl);
 	spin_unlock(old_ptl);
@@ -296,6 +310,14 @@ static bool move_normal_pud(struct vm_area_struct *vma, unsigned long old_addr,
 	struct mm_struct *mm = vma->vm_mm;
 	pud_t pud;
 
+	/*
+	 * Disable MOVE_PUD until we get the pageout done with all
+	 * higher level page table locks held. With SPLIT_PTE_PTLOCKS
+	 * we use mm->page_table_lock for both pte ptl and pud ptl
+	 */
+	if (USE_SPLIT_PTE_PTLOCKS)
+		return false;
+
 	/*
 	 * The destination pud shouldn't be established, free_pgtables()
 	 * should have released it.
-- 
2.31.1



More information about the Linuxppc-dev mailing list