khugepaged: collapse_pte_mapped_thp() flush the right range
authorHugh Dickins <hughd@google.com>
Fri, 7 Aug 2020 06:26:15 +0000 (23:26 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 7 Aug 2020 18:33:29 +0000 (11:33 -0700)
pmdp_collapse_flush() should be given the start address at which the huge
page is mapped, haddr: it was given addr, which at that point has been
used as a local variable, incremented to the end address of the extent.

Found by source inspection while chasing a hugepage locking bug, which I
then could not explain by this.  At first I thought this was very bad;
then saw that all of the page translations that were not flushed would
actually still point to the right pages afterwards, so harmless; then
realized that I know nothing of how different architectures and models
cache intermediate paging structures, so maybe it matters after all -
particularly since the page table concerned is immediately freed.

Much easier to fix than to think about.

Fixes: 27e1f8273113 ("khugepaged: enable collapse pmd for pte-mapped THP")
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Song Liu <songliubraving@fb.com>
Cc: <stable@vger.kernel.org> [5.4+]
Link: http://lkml.kernel.org/r/alpine.LSU.2.11.2008021204390.27773@eggly.anvils
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/khugepaged.c

index 700f5160f3e4dcaaed5199923e1aedaa57e9ad08..28aae7f45c63fa156e6cd6302955543045e10f0e 100644 (file)
@@ -1502,7 +1502,7 @@ void collapse_pte_mapped_thp(struct mm_struct *mm, unsigned long addr)
 
        /* step 4: collapse pmd */
        ptl = pmd_lock(vma->vm_mm, pmd);
-       _pmd = pmdp_collapse_flush(vma, addr, pmd);
+       _pmd = pmdp_collapse_flush(vma, haddr, pmd);
        spin_unlock(ptl);
        mm_dec_nr_ptes(mm);
        pte_free(mm, pmd_pgtable(_pmd));