hugetlb: call mmu notifiers on hugepage cow
Doug Doan [Tue, 10 Aug 2010 00:18:30 +0000 (17:18 -0700)]
When a copy-on-write occurs, we take one of two paths in handle_mm_fault:
through handle_pte_fault for normal pages, or through hugetlb_fault for
huge pages.

In the normal page case, we eventually get to do_wp_page and call mmu
notifiers via ptep_clear_flush_notify.  There is no callout to the mmmu
notifiers in the huge page case.  This patch fixes that.

Signed-off-by: Doug Doan <dougd@cray.com>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/hugetlb.c

index 54d42b0..b61d2db 100644 (file)
@@ -2349,11 +2349,17 @@ retry_avoidcopy:
        ptep = huge_pte_offset(mm, address & huge_page_mask(h));
        if (likely(pte_same(huge_ptep_get(ptep), pte))) {
                /* Break COW */
+               mmu_notifier_invalidate_range_start(mm,
+                       address & huge_page_mask(h),
+                       (address & huge_page_mask(h)) + huge_page_size(h));
                huge_ptep_clear_flush(vma, address, ptep);
                set_huge_pte_at(mm, address, ptep,
                                make_huge_pte(vma, new_page, 1));
                /* Make the old page be freed below */
                new_page = old_page;
+               mmu_notifier_invalidate_range_end(mm,
+                       address & huge_page_mask(h),
+                       (address & huge_page_mask(h)) + huge_page_size(h));
        }
        page_cache_release(new_page);
        page_cache_release(old_page);