mm: clear pmd_numa before invalidating

commit 67f87463d3a3362424efcbe8b40e4772fd34fc61 upstream.

On x86, PMD entries are similar to _PAGE_PROTNONE protection and are
handled as NUMA hinting faults.  The following two page table protection
bits are what defines them

	_PAGE_NUMA:set	_PAGE_PRESENT:clear

A PMD is considered present if any of the _PAGE_PRESENT, _PAGE_PROTNONE,
_PAGE_PSE or _PAGE_NUMA bits are set.  If pmdp_invalidate encounters a
pmd_numa, it clears the present bit leaving _PAGE_NUMA which will be
considered not present by the CPU but present by pmd_present.  The
existing caller of pmdp_invalidate should handle it but it's an
inconsistent state for a PMD.  This patch keeps the state consistent
when calling pmdp_invalidate.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Alex Thorlton <athorlton@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

diff --git a/mm/pgtable-generic.c b/mm/pgtable-generic.c
index 0c8323fe6c8f610b4068a2232eb4459c23f765f6..eb900bbaa71325b804dde5490612c3000d88d62a 100644 (file)
--- a/mm/pgtable-generic.c
+++ b/mm/pgtable-generic.c
@@ -166,6 +166,9 @@ pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm)
 void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
                     pmd_t *pmdp)
 {
+       pmd_t entry = *pmdp;
+       if (pmd_numa(entry))
+               entry = pmd_mknonnuma(entry);
        set_pmd_at(vma->vm_mm, address, pmdp, pmd_mknotpresent(*pmdp));
        flush_tlb_range(vma, address, address + HPAGE_PMD_SIZE);
 }