thp: set compound tail page _count to zero
authorYouquan Song <youquan.song@intel.com>
Thu, 8 Dec 2011 22:34:18 +0000 (14:34 -0800)
committerGreg Kroah-Hartman <gregkh@suse.de>
Wed, 21 Dec 2011 20:58:22 +0000 (12:58 -0800)
commit7f37071e9df2344a7d7d066235ec72a0d1a4b013
tree159499fa82f3aa12ee6539c7f5f7d5d507d693f8
parent75aaef5468ceb118aa55825cf6a52d60d09dfde0
thp: set compound tail page _count to zero

commit 58a84aa92723d1ac3e1cc4e3b0ff49291663f7e1 upstream.

Commit 70b50f94f1644 ("mm: thp: tail page refcounting fix") keeps all
page_tail->_count zero at all times.  But the current kernel does not
set page_tail->_count to zero if a 1GB page is utilized.  So when an
IOMMU 1GB page is used by KVM, it wil result in a kernel oops because a
tail page's _count does not equal zero.

  kernel BUG at include/linux/mm.h:386!
  invalid opcode: 0000 [#1] SMP
  Call Trace:
    gup_pud_range+0xb8/0x19d
    get_user_pages_fast+0xcb/0x192
    ? trace_hardirqs_off+0xd/0xf
    hva_to_pfn+0x119/0x2f2
    gfn_to_pfn_memslot+0x2c/0x2e
    kvm_iommu_map_pages+0xfd/0x1c1
    kvm_iommu_map_memslots+0x7c/0xbd
    kvm_iommu_map_guest+0xaa/0xbf
    kvm_vm_ioctl_assigned_device+0x2ef/0xa47
    kvm_vm_ioctl+0x36c/0x3a2
    do_vfs_ioctl+0x49e/0x4e4
    sys_ioctl+0x5a/0x7c
    system_call_fastpath+0x16/0x1b
  RIP  gup_huge_pud+0xf2/0x159

Signed-off-by: Youquan Song <youquan.song@intel.com>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.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@suse.de>
mm/hugetlb.c
mm/page_alloc.c