releases/3.5.1/x86-mce-fix-siginfo_t-si_addr-value-for-non-recoverable-memory-faults.patch

   1 From 6751ed65dc6642af64f7b8a440a75563c8aab7ae Mon Sep 17 00:00:00 2001
   2 From: Tony Luck <tony.luck@intel.com>
   3 Date: Wed, 11 Jul 2012 10:20:47 -0700
   4 Subject: x86/mce: Fix siginfo_t->si_addr value for non-recoverable memory faults
   5
   6 From: Tony Luck <tony.luck@intel.com>
   7
   8 commit 6751ed65dc6642af64f7b8a440a75563c8aab7ae upstream.
   9
  10 In commit dad1743e5993f1 ("x86/mce: Only restart instruction after machine
  11 check recovery if it is safe") we fixed mce_notify_process() to force a
  12 signal to the current process if it was not restartable (RIPV bit not
  13 set in MCG_STATUS). But doing it here means that the process doesn't
  14 get told the virtual address of the fault via siginfo_t->si_addr. This
  15 would prevent application level recovery from the fault.
  16
  17 Make a new MF_MUST_KILL flag bit for memory_failure() et al. to use so
  18 that we will provide the right information with the signal.
  19
  20 Signed-off-by: Tony Luck <tony.luck@intel.com>
  21 Acked-by: Borislav Petkov <borislav.petkov@amd.com>
  22 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  23
  24 ---
  25  arch/x86/kernel/cpu/mcheck/mce.c |    6 ++++--
  26  include/linux/mm.h               |    1 +
  27  mm/memory-failure.c              |   14 ++++++++------
  28  3 files changed, 13 insertions(+), 8 deletions(-)
  29
  30 --- a/arch/x86/kernel/cpu/mcheck/mce.c
  31 +++ b/arch/x86/kernel/cpu/mcheck/mce.c
  32 @@ -1186,6 +1186,7 @@ void mce_notify_process(void)
  33  {
  34         unsigned long pfn;
  35         struct mce_info *mi = mce_find_info();
  36 +       int flags = MF_ACTION_REQUIRED;
  37
  38         if (!mi)
  39                 mce_panic("Lost physical address for unconsumed uncorrectable error", NULL, NULL);
  40 @@ -1200,8 +1201,9 @@ void mce_notify_process(void)
  41          * doomed. We still need to mark the page as poisoned and alert any
  42          * other users of the page.
  43          */
  44 -       if (memory_failure(pfn, MCE_VECTOR, MF_ACTION_REQUIRED) < 0 ||
  45 -                          mi->restartable == 0) {
  46 +       if (!mi->restartable)
  47 +               flags |= MF_MUST_KILL;
  48 +       if (memory_failure(pfn, MCE_VECTOR, flags) < 0) {
  49                 pr_err("Memory error not recovered");
  50                 force_sig(SIGBUS, current);
  51         }
  52 --- a/include/linux/mm.h
  53 +++ b/include/linux/mm.h
  54 @@ -1591,6 +1591,7 @@ void vmemmap_populate_print_last(void);
  55  enum mf_flags {
  56         MF_COUNT_INCREASED = 1 << 0,
  57         MF_ACTION_REQUIRED = 1 << 1,
  58 +       MF_MUST_KILL = 1 << 2,
  59  };
  60  extern int memory_failure(unsigned long pfn, int trapno, int flags);
  61  extern void memory_failure_queue(unsigned long pfn, int trapno, int flags);
  62 --- a/mm/memory-failure.c
  63 +++ b/mm/memory-failure.c
  64 @@ -345,14 +345,14 @@ static void add_to_kill(struct task_stru
  65   * Also when FAIL is set do a force kill because something went
  66   * wrong earlier.
  67   */
  68 -static void kill_procs(struct list_head *to_kill, int doit, int trapno,
  69 +static void kill_procs(struct list_head *to_kill, int forcekill, int trapno,
  70                           int fail, struct page *page, unsigned long pfn,
  71                           int flags)
  72  {
  73         struct to_kill *tk, *next;
  74
  75         list_for_each_entry_safe (tk, next, to_kill, nd) {
  76 -               if (doit) {
  77 +               if (forcekill) {
  78                         /*
  79                          * In case something went wrong with munmapping
  80                          * make sure the process doesn't catch the
  81 @@ -858,7 +858,7 @@ static int hwpoison_user_mappings(struct
  82         struct address_space *mapping;
  83         LIST_HEAD(tokill);
  84         int ret;
  85 -       int kill = 1;
  86 +       int kill = 1, forcekill;
  87         struct page *hpage = compound_head(p);
  88         struct page *ppage;
  89
  90 @@ -888,7 +888,7 @@ static int hwpoison_user_mappings(struct
  91          * be called inside page lock (it's recommended but not enforced).
  92          */
  93         mapping = page_mapping(hpage);
  94 -       if (!PageDirty(hpage) && mapping &&
  95 +       if (!(flags & MF_MUST_KILL) && !PageDirty(hpage) && mapping &&
  96             mapping_cap_writeback_dirty(mapping)) {
  97                 if (page_mkclean(hpage)) {
  98                         SetPageDirty(hpage);
  99 @@ -965,12 +965,14 @@ static int hwpoison_user_mappings(struct
 100          * Now that the dirty bit has been propagated to the
 101          * struct page and all unmaps done we can decide if
 102          * killing is needed or not.  Only kill when the page
 103 -        * was dirty, otherwise the tokill list is merely
 104 +        * was dirty or the process is not restartable,
 105 +        * otherwise the tokill list is merely
 106          * freed.  When there was a problem unmapping earlier
 107          * use a more force-full uncatchable kill to prevent
 108          * any accesses to the poisoned memory.
 109          */
 110 -       kill_procs(&tokill, !!PageDirty(ppage), trapno,
 111 +       forcekill = PageDirty(ppage) || (flags & MF_MUST_KILL);
 112 +       kill_procs(&tokill, forcekill, trapno,
 113                       ret != SWAP_SUCCESS, p, pfn, flags);
 114
 115         return ret;