aboutsummaryrefslogtreecommitdiffstats
path: root/mm/memory-failure.c
diff options
context:
space:
mode:
Diffstat (limited to 'mm/memory-failure.c')
-rw-r--r--mm/memory-failure.c39
1 files changed, 22 insertions, 17 deletions
diff --git a/mm/memory-failure.c b/mm/memory-failure.c
index 620b0b461593..1ec68c80788e 100644
--- a/mm/memory-failure.c
+++ b/mm/memory-failure.c
@@ -45,6 +45,7 @@
45#include <linux/page-isolation.h> 45#include <linux/page-isolation.h>
46#include <linux/suspend.h> 46#include <linux/suspend.h>
47#include <linux/slab.h> 47#include <linux/slab.h>
48#include <linux/hugetlb.h>
48#include "internal.h" 49#include "internal.h"
49 50
50int sysctl_memory_failure_early_kill __read_mostly = 0; 51int sysctl_memory_failure_early_kill __read_mostly = 0;
@@ -837,6 +838,7 @@ static int hwpoison_user_mappings(struct page *p, unsigned long pfn,
837 int ret; 838 int ret;
838 int i; 839 int i;
839 int kill = 1; 840 int kill = 1;
841 struct page *hpage = compound_head(p);
840 842
841 if (PageReserved(p) || PageSlab(p)) 843 if (PageReserved(p) || PageSlab(p))
842 return SWAP_SUCCESS; 844 return SWAP_SUCCESS;
@@ -845,10 +847,10 @@ static int hwpoison_user_mappings(struct page *p, unsigned long pfn,
845 * This check implies we don't kill processes if their pages 847 * This check implies we don't kill processes if their pages
846 * are in the swap cache early. Those are always late kills. 848 * are in the swap cache early. Those are always late kills.
847 */ 849 */
848 if (!page_mapped(p)) 850 if (!page_mapped(hpage))
849 return SWAP_SUCCESS; 851 return SWAP_SUCCESS;
850 852
851 if (PageCompound(p) || PageKsm(p)) 853 if (PageKsm(p))
852 return SWAP_FAIL; 854 return SWAP_FAIL;
853 855
854 if (PageSwapCache(p)) { 856 if (PageSwapCache(p)) {
@@ -863,10 +865,11 @@ static int hwpoison_user_mappings(struct page *p, unsigned long pfn,
863 * XXX: the dirty test could be racy: set_page_dirty() may not always 865 * XXX: the dirty test could be racy: set_page_dirty() may not always
864 * be called inside page lock (it's recommended but not enforced). 866 * be called inside page lock (it's recommended but not enforced).
865 */ 867 */
866 mapping = page_mapping(p); 868 mapping = page_mapping(hpage);
867 if (!PageDirty(p) && mapping && mapping_cap_writeback_dirty(mapping)) { 869 if (!PageDirty(hpage) && mapping &&
868 if (page_mkclean(p)) { 870 mapping_cap_writeback_dirty(mapping)) {
869 SetPageDirty(p); 871 if (page_mkclean(hpage)) {
872 SetPageDirty(hpage);
870 } else { 873 } else {
871 kill = 0; 874 kill = 0;
872 ttu |= TTU_IGNORE_HWPOISON; 875 ttu |= TTU_IGNORE_HWPOISON;
@@ -885,14 +888,14 @@ static int hwpoison_user_mappings(struct page *p, unsigned long pfn,
885 * there's nothing that can be done. 888 * there's nothing that can be done.
886 */ 889 */
887 if (kill) 890 if (kill)
888 collect_procs(p, &tokill); 891 collect_procs(hpage, &tokill);
889 892
890 /* 893 /*
891 * try_to_unmap can fail temporarily due to races. 894 * try_to_unmap can fail temporarily due to races.
892 * Try a few times (RED-PEN better strategy?) 895 * Try a few times (RED-PEN better strategy?)
893 */ 896 */
894 for (i = 0; i < N_UNMAP_TRIES; i++) { 897 for (i = 0; i < N_UNMAP_TRIES; i++) {
895 ret = try_to_unmap(p, ttu); 898 ret = try_to_unmap(hpage, ttu);
896 if (ret == SWAP_SUCCESS) 899 if (ret == SWAP_SUCCESS)
897 break; 900 break;
898 pr_debug("MCE %#lx: try_to_unmap retry needed %d\n", pfn, ret); 901 pr_debug("MCE %#lx: try_to_unmap retry needed %d\n", pfn, ret);
@@ -900,7 +903,7 @@ static int hwpoison_user_mappings(struct page *p, unsigned long pfn,
900 903
901 if (ret != SWAP_SUCCESS) 904 if (ret != SWAP_SUCCESS)
902 printk(KERN_ERR "MCE %#lx: failed to unmap page (mapcount=%d)\n", 905 printk(KERN_ERR "MCE %#lx: failed to unmap page (mapcount=%d)\n",
903 pfn, page_mapcount(p)); 906 pfn, page_mapcount(hpage));
904 907
905 /* 908 /*
906 * Now that the dirty bit has been propagated to the 909 * Now that the dirty bit has been propagated to the
@@ -911,7 +914,7 @@ static int hwpoison_user_mappings(struct page *p, unsigned long pfn,
911 * use a more force-full uncatchable kill to prevent 914 * use a more force-full uncatchable kill to prevent
912 * any accesses to the poisoned memory. 915 * any accesses to the poisoned memory.
913 */ 916 */
914 kill_procs_ao(&tokill, !!PageDirty(p), trapno, 917 kill_procs_ao(&tokill, !!PageDirty(hpage), trapno,
915 ret != SWAP_SUCCESS, pfn); 918 ret != SWAP_SUCCESS, pfn);
916 919
917 return ret; 920 return ret;
@@ -921,6 +924,7 @@ int __memory_failure(unsigned long pfn, int trapno, int flags)
921{ 924{
922 struct page_state *ps; 925 struct page_state *ps;
923 struct page *p; 926 struct page *p;
927 struct page *hpage;
924 int res; 928 int res;
925 929
926 if (!sysctl_memory_failure_recovery) 930 if (!sysctl_memory_failure_recovery)
@@ -934,6 +938,7 @@ int __memory_failure(unsigned long pfn, int trapno, int flags)
934 } 938 }
935 939
936 p = pfn_to_page(pfn); 940 p = pfn_to_page(pfn);
941 hpage = compound_head(p);
937 if (TestSetPageHWPoison(p)) { 942 if (TestSetPageHWPoison(p)) {
938 printk(KERN_ERR "MCE %#lx: already hardware poisoned\n", pfn); 943 printk(KERN_ERR "MCE %#lx: already hardware poisoned\n", pfn);
939 return 0; 944 return 0;
@@ -953,7 +958,7 @@ int __memory_failure(unsigned long pfn, int trapno, int flags)
953 * that may make page_freeze_refs()/page_unfreeze_refs() mismatch. 958 * that may make page_freeze_refs()/page_unfreeze_refs() mismatch.
954 */ 959 */
955 if (!(flags & MF_COUNT_INCREASED) && 960 if (!(flags & MF_COUNT_INCREASED) &&
956 !get_page_unless_zero(compound_head(p))) { 961 !get_page_unless_zero(hpage)) {
957 if (is_free_buddy_page(p)) { 962 if (is_free_buddy_page(p)) {
958 action_result(pfn, "free buddy", DELAYED); 963 action_result(pfn, "free buddy", DELAYED);
959 return 0; 964 return 0;
@@ -971,9 +976,9 @@ int __memory_failure(unsigned long pfn, int trapno, int flags)
971 * The check (unnecessarily) ignores LRU pages being isolated and 976 * The check (unnecessarily) ignores LRU pages being isolated and
972 * walked by the page reclaim code, however that's not a big loss. 977 * walked by the page reclaim code, however that's not a big loss.
973 */ 978 */
974 if (!PageLRU(p)) 979 if (!PageLRU(p) && !PageHuge(p))
975 shake_page(p, 0); 980 shake_page(p, 0);
976 if (!PageLRU(p)) { 981 if (!PageLRU(p) && !PageHuge(p)) {
977 /* 982 /*
978 * shake_page could have turned it free. 983 * shake_page could have turned it free.
979 */ 984 */
@@ -991,7 +996,7 @@ int __memory_failure(unsigned long pfn, int trapno, int flags)
991 * It's very difficult to mess with pages currently under IO 996 * It's very difficult to mess with pages currently under IO
992 * and in many cases impossible, so we just avoid it here. 997 * and in many cases impossible, so we just avoid it here.
993 */ 998 */
994 lock_page_nosync(p); 999 lock_page_nosync(hpage);
995 1000
996 /* 1001 /*
997 * unpoison always clear PG_hwpoison inside page lock 1002 * unpoison always clear PG_hwpoison inside page lock
@@ -1004,8 +1009,8 @@ int __memory_failure(unsigned long pfn, int trapno, int flags)
1004 if (hwpoison_filter(p)) { 1009 if (hwpoison_filter(p)) {
1005 if (TestClearPageHWPoison(p)) 1010 if (TestClearPageHWPoison(p))
1006 atomic_long_dec(&mce_bad_pages); 1011 atomic_long_dec(&mce_bad_pages);
1007 unlock_page(p); 1012 unlock_page(hpage);
1008 put_page(p); 1013 put_page(hpage);
1009 return 0; 1014 return 0;
1010 } 1015 }
1011 1016
@@ -1038,7 +1043,7 @@ int __memory_failure(unsigned long pfn, int trapno, int flags)
1038 } 1043 }
1039 } 1044 }
1040out: 1045out:
1041 unlock_page(p); 1046 unlock_page(hpage);
1042 return res; 1047 return res;
1043} 1048}
1044EXPORT_SYMBOL_GPL(__memory_failure); 1049EXPORT_SYMBOL_GPL(__memory_failure);