x86/mce: Fix siginfo_t->si_addr value for non-recoverable memory faults

In commit dad1743 ("x86/mce: Only restart instruction after machine
check recovery if it is safe") we fixed mce_notify_process() to force a
signal to the current process if it was not restartable (RIPV bit not
set in MCG_STATUS). But doing it here means that the process doesn't
get told the virtual address of the fault via siginfo_t->si_addr. This
would prevent application level recovery from the fault.

Make a new MF_MUST_KILL flag bit for memory_failure() et al. to use so
that we will provide the right information with the signal.

Signed-off-by: Tony Luck <tony.luck@intel.com>
Acked-by: Borislav Petkov <borislav.petkov@amd.com>
Cc: stable@kernel.org    # 3.4+

arch/x86/kernel/cpu/mcheck/mce.c

@@ -1186,6 +1186,7 @@ void mce_notify_process(void)
 {
 	unsigned long pfn;
 	struct mce_info *mi = mce_find_info();
+	int flags = MF_ACTION_REQUIRED;
 
 	if (!mi)
 		mce_panic("Lost physical address for unconsumed uncorrectable error", NULL, NULL);
@@ -1200,8 +1201,9 @@ void mce_notify_process(void)
 	 * doomed. We still need to mark the page as poisoned and alert any
 	 * other users of the page.
 	 */
-	if (memory_failure(pfn, MCE_VECTOR, MF_ACTION_REQUIRED) < 0 ||
-			   mi->restartable == 0) {
+	if (!mi->restartable)
+		flags |= MF_MUST_KILL;
+	if (memory_failure(pfn, MCE_VECTOR, flags) < 0) {
 		pr_err("Memory error not recovered");
 		force_sig(SIGBUS, current);
 	}

include/linux/mm.h

@@ -1591,6 +1591,7 @@ void vmemmap_populate_print_last(void);
 enum mf_flags {
 	MF_COUNT_INCREASED = 1 << 0,
 	MF_ACTION_REQUIRED = 1 << 1,
+	MF_MUST_KILL = 1 << 2,
 };
 extern int memory_failure(unsigned long pfn, int trapno, int flags);
 extern void memory_failure_queue(unsigned long pfn, int trapno, int flags);

mm/memory-failure.c

@@ -345,14 +345,14 @@ static void add_to_kill(struct task_struct *tsk, struct page *p,
  * Also when FAIL is set do a force kill because something went
  * wrong earlier.
  */
-static void kill_procs(struct list_head *to_kill, int doit, int trapno,
+static void kill_procs(struct list_head *to_kill, int forcekill, int trapno,
 			  int fail, struct page *page, unsigned long pfn,
 			  int flags)
 {
 	struct to_kill *tk, *next;
 
 	list_for_each_entry_safe (tk, next, to_kill, nd) {
-		if (doit) {
+		if (forcekill) {
 			/*
 			 * In case something went wrong with munmapping
 			 * make sure the process doesn't catch the
@@ -858,7 +858,7 @@ static int hwpoison_user_mappings(struct page *p, unsigned long pfn,
 	struct address_space *mapping;
 	LIST_HEAD(tokill);
 	int ret;
-	int kill = 1;
+	int kill = 1, forcekill;
 	struct page *hpage = compound_head(p);
 	struct page *ppage;
 
@@ -888,7 +888,7 @@ static int hwpoison_user_mappings(struct page *p, unsigned long pfn,
 	 * be called inside page lock (it's recommended but not enforced).
 	 */
 	mapping = page_mapping(hpage);
-	if (!PageDirty(hpage) && mapping &&
+	if (!(flags & MF_MUST_KILL) && !PageDirty(hpage) && mapping &&
 	    mapping_cap_writeback_dirty(mapping)) {
 		if (page_mkclean(hpage)) {
 			SetPageDirty(hpage);
@@ -965,12 +965,14 @@ static int hwpoison_user_mappings(struct page *p, unsigned long pfn,
 	 * Now that the dirty bit has been propagated to the
 	 * struct page and all unmaps done we can decide if
 	 * killing is needed or not.  Only kill when the page
-	 * was dirty, otherwise the tokill list is merely
+	 * was dirty or the process is not restartable,
+	 * otherwise the tokill list is merely
 	 * freed.  When there was a problem unmapping earlier
 	 * use a more force-full uncatchable kill to prevent
 	 * any accesses to the poisoned memory.
 	 */
-	kill_procs(&tokill, !!PageDirty(ppage), trapno,
+	forcekill = PageDirty(ppage) || (flags & MF_MUST_KILL);
+	kill_procs(&tokill, forcekill, trapno,
 		      ret != SWAP_SUCCESS, p, pfn, flags);
 
 	return ret;