diff options
-rw-r--r-- | include/linux/compaction.h | 4 | ||||
-rw-r--r-- | include/linux/mm.h | 1 | ||||
-rw-r--r-- | mm/compaction.c | 92 | ||||
-rw-r--r-- | mm/internal.h | 1 | ||||
-rw-r--r-- | mm/page_alloc.c | 35 |
5 files changed, 23 insertions, 110 deletions
diff --git a/include/linux/compaction.h b/include/linux/compaction.h index 6ecb6dc2f303..cc7bddeaf553 100644 --- a/include/linux/compaction.h +++ b/include/linux/compaction.h | |||
@@ -22,7 +22,7 @@ extern int sysctl_extfrag_handler(struct ctl_table *table, int write, | |||
22 | extern int fragmentation_index(struct zone *zone, unsigned int order); | 22 | extern int fragmentation_index(struct zone *zone, unsigned int order); |
23 | extern unsigned long try_to_compact_pages(struct zonelist *zonelist, | 23 | extern unsigned long try_to_compact_pages(struct zonelist *zonelist, |
24 | int order, gfp_t gfp_mask, nodemask_t *mask, | 24 | int order, gfp_t gfp_mask, nodemask_t *mask, |
25 | bool sync, bool *contended, struct page **page); | 25 | bool sync, bool *contended); |
26 | extern int compact_pgdat(pg_data_t *pgdat, int order); | 26 | extern int compact_pgdat(pg_data_t *pgdat, int order); |
27 | extern void reset_isolation_suitable(pg_data_t *pgdat); | 27 | extern void reset_isolation_suitable(pg_data_t *pgdat); |
28 | extern unsigned long compaction_suitable(struct zone *zone, int order); | 28 | extern unsigned long compaction_suitable(struct zone *zone, int order); |
@@ -75,7 +75,7 @@ static inline bool compaction_restarting(struct zone *zone, int order) | |||
75 | #else | 75 | #else |
76 | static inline unsigned long try_to_compact_pages(struct zonelist *zonelist, | 76 | static inline unsigned long try_to_compact_pages(struct zonelist *zonelist, |
77 | int order, gfp_t gfp_mask, nodemask_t *nodemask, | 77 | int order, gfp_t gfp_mask, nodemask_t *nodemask, |
78 | bool sync, bool *contended, struct page **page) | 78 | bool sync, bool *contended) |
79 | { | 79 | { |
80 | return COMPACT_CONTINUE; | 80 | return COMPACT_CONTINUE; |
81 | } | 81 | } |
diff --git a/include/linux/mm.h b/include/linux/mm.h index 63204078f72b..66e2f7c61e5c 100644 --- a/include/linux/mm.h +++ b/include/linux/mm.h | |||
@@ -455,7 +455,6 @@ void put_pages_list(struct list_head *pages); | |||
455 | 455 | ||
456 | void split_page(struct page *page, unsigned int order); | 456 | void split_page(struct page *page, unsigned int order); |
457 | int split_free_page(struct page *page); | 457 | int split_free_page(struct page *page); |
458 | int capture_free_page(struct page *page, int alloc_order, int migratetype); | ||
459 | 458 | ||
460 | /* | 459 | /* |
461 | * Compound pages have a destructor function. Provide a | 460 | * Compound pages have a destructor function. Provide a |
diff --git a/mm/compaction.c b/mm/compaction.c index f8f5c111b7d7..c62bd063d766 100644 --- a/mm/compaction.c +++ b/mm/compaction.c | |||
@@ -816,6 +816,7 @@ static isolate_migrate_t isolate_migratepages(struct zone *zone, | |||
816 | static int compact_finished(struct zone *zone, | 816 | static int compact_finished(struct zone *zone, |
817 | struct compact_control *cc) | 817 | struct compact_control *cc) |
818 | { | 818 | { |
819 | unsigned int order; | ||
819 | unsigned long watermark; | 820 | unsigned long watermark; |
820 | 821 | ||
821 | if (fatal_signal_pending(current)) | 822 | if (fatal_signal_pending(current)) |
@@ -850,22 +851,16 @@ static int compact_finished(struct zone *zone, | |||
850 | return COMPACT_CONTINUE; | 851 | return COMPACT_CONTINUE; |
851 | 852 | ||
852 | /* Direct compactor: Is a suitable page free? */ | 853 | /* Direct compactor: Is a suitable page free? */ |
853 | if (cc->page) { | 854 | for (order = cc->order; order < MAX_ORDER; order++) { |
854 | /* Was a suitable page captured? */ | 855 | struct free_area *area = &zone->free_area[order]; |
855 | if (*cc->page) | 856 | |
857 | /* Job done if page is free of the right migratetype */ | ||
858 | if (!list_empty(&area->free_list[cc->migratetype])) | ||
859 | return COMPACT_PARTIAL; | ||
860 | |||
861 | /* Job done if allocation would set block type */ | ||
862 | if (cc->order >= pageblock_order && area->nr_free) | ||
856 | return COMPACT_PARTIAL; | 863 | return COMPACT_PARTIAL; |
857 | } else { | ||
858 | unsigned int order; | ||
859 | for (order = cc->order; order < MAX_ORDER; order++) { | ||
860 | struct free_area *area = &zone->free_area[cc->order]; | ||
861 | /* Job done if page is free of the right migratetype */ | ||
862 | if (!list_empty(&area->free_list[cc->migratetype])) | ||
863 | return COMPACT_PARTIAL; | ||
864 | |||
865 | /* Job done if allocation would set block type */ | ||
866 | if (cc->order >= pageblock_order && area->nr_free) | ||
867 | return COMPACT_PARTIAL; | ||
868 | } | ||
869 | } | 864 | } |
870 | 865 | ||
871 | return COMPACT_CONTINUE; | 866 | return COMPACT_CONTINUE; |
@@ -921,60 +916,6 @@ unsigned long compaction_suitable(struct zone *zone, int order) | |||
921 | return COMPACT_CONTINUE; | 916 | return COMPACT_CONTINUE; |
922 | } | 917 | } |
923 | 918 | ||
924 | static void compact_capture_page(struct compact_control *cc) | ||
925 | { | ||
926 | unsigned long flags; | ||
927 | int mtype, mtype_low, mtype_high; | ||
928 | |||
929 | if (!cc->page || *cc->page) | ||
930 | return; | ||
931 | |||
932 | /* | ||
933 | * For MIGRATE_MOVABLE allocations we capture a suitable page ASAP | ||
934 | * regardless of the migratetype of the freelist is is captured from. | ||
935 | * This is fine because the order for a high-order MIGRATE_MOVABLE | ||
936 | * allocation is typically at least a pageblock size and overall | ||
937 | * fragmentation is not impaired. Other allocation types must | ||
938 | * capture pages from their own migratelist because otherwise they | ||
939 | * could pollute other pageblocks like MIGRATE_MOVABLE with | ||
940 | * difficult to move pages and making fragmentation worse overall. | ||
941 | */ | ||
942 | if (cc->migratetype == MIGRATE_MOVABLE) { | ||
943 | mtype_low = 0; | ||
944 | mtype_high = MIGRATE_PCPTYPES; | ||
945 | } else { | ||
946 | mtype_low = cc->migratetype; | ||
947 | mtype_high = cc->migratetype + 1; | ||
948 | } | ||
949 | |||
950 | /* Speculatively examine the free lists without zone lock */ | ||
951 | for (mtype = mtype_low; mtype < mtype_high; mtype++) { | ||
952 | int order; | ||
953 | for (order = cc->order; order < MAX_ORDER; order++) { | ||
954 | struct page *page; | ||
955 | struct free_area *area; | ||
956 | area = &(cc->zone->free_area[order]); | ||
957 | if (list_empty(&area->free_list[mtype])) | ||
958 | continue; | ||
959 | |||
960 | /* Take the lock and attempt capture of the page */ | ||
961 | if (!compact_trylock_irqsave(&cc->zone->lock, &flags, cc)) | ||
962 | return; | ||
963 | if (!list_empty(&area->free_list[mtype])) { | ||
964 | page = list_entry(area->free_list[mtype].next, | ||
965 | struct page, lru); | ||
966 | if (capture_free_page(page, cc->order, mtype)) { | ||
967 | spin_unlock_irqrestore(&cc->zone->lock, | ||
968 | flags); | ||
969 | *cc->page = page; | ||
970 | return; | ||
971 | } | ||
972 | } | ||
973 | spin_unlock_irqrestore(&cc->zone->lock, flags); | ||
974 | } | ||
975 | } | ||
976 | } | ||
977 | |||
978 | static int compact_zone(struct zone *zone, struct compact_control *cc) | 919 | static int compact_zone(struct zone *zone, struct compact_control *cc) |
979 | { | 920 | { |
980 | int ret; | 921 | int ret; |
@@ -1054,9 +995,6 @@ static int compact_zone(struct zone *zone, struct compact_control *cc) | |||
1054 | goto out; | 995 | goto out; |
1055 | } | 996 | } |
1056 | } | 997 | } |
1057 | |||
1058 | /* Capture a page now if it is a suitable size */ | ||
1059 | compact_capture_page(cc); | ||
1060 | } | 998 | } |
1061 | 999 | ||
1062 | out: | 1000 | out: |
@@ -1069,8 +1007,7 @@ out: | |||
1069 | 1007 | ||
1070 | static unsigned long compact_zone_order(struct zone *zone, | 1008 | static unsigned long compact_zone_order(struct zone *zone, |
1071 | int order, gfp_t gfp_mask, | 1009 | int order, gfp_t gfp_mask, |
1072 | bool sync, bool *contended, | 1010 | bool sync, bool *contended) |
1073 | struct page **page) | ||
1074 | { | 1011 | { |
1075 | unsigned long ret; | 1012 | unsigned long ret; |
1076 | struct compact_control cc = { | 1013 | struct compact_control cc = { |
@@ -1080,7 +1017,6 @@ static unsigned long compact_zone_order(struct zone *zone, | |||
1080 | .migratetype = allocflags_to_migratetype(gfp_mask), | 1017 | .migratetype = allocflags_to_migratetype(gfp_mask), |
1081 | .zone = zone, | 1018 | .zone = zone, |
1082 | .sync = sync, | 1019 | .sync = sync, |
1083 | .page = page, | ||
1084 | }; | 1020 | }; |
1085 | INIT_LIST_HEAD(&cc.freepages); | 1021 | INIT_LIST_HEAD(&cc.freepages); |
1086 | INIT_LIST_HEAD(&cc.migratepages); | 1022 | INIT_LIST_HEAD(&cc.migratepages); |
@@ -1110,7 +1046,7 @@ int sysctl_extfrag_threshold = 500; | |||
1110 | */ | 1046 | */ |
1111 | unsigned long try_to_compact_pages(struct zonelist *zonelist, | 1047 | unsigned long try_to_compact_pages(struct zonelist *zonelist, |
1112 | int order, gfp_t gfp_mask, nodemask_t *nodemask, | 1048 | int order, gfp_t gfp_mask, nodemask_t *nodemask, |
1113 | bool sync, bool *contended, struct page **page) | 1049 | bool sync, bool *contended) |
1114 | { | 1050 | { |
1115 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | 1051 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); |
1116 | int may_enter_fs = gfp_mask & __GFP_FS; | 1052 | int may_enter_fs = gfp_mask & __GFP_FS; |
@@ -1136,7 +1072,7 @@ unsigned long try_to_compact_pages(struct zonelist *zonelist, | |||
1136 | int status; | 1072 | int status; |
1137 | 1073 | ||
1138 | status = compact_zone_order(zone, order, gfp_mask, sync, | 1074 | status = compact_zone_order(zone, order, gfp_mask, sync, |
1139 | contended, page); | 1075 | contended); |
1140 | rc = max(status, rc); | 1076 | rc = max(status, rc); |
1141 | 1077 | ||
1142 | /* If a normal allocation would succeed, stop compacting */ | 1078 | /* If a normal allocation would succeed, stop compacting */ |
@@ -1192,7 +1128,6 @@ int compact_pgdat(pg_data_t *pgdat, int order) | |||
1192 | struct compact_control cc = { | 1128 | struct compact_control cc = { |
1193 | .order = order, | 1129 | .order = order, |
1194 | .sync = false, | 1130 | .sync = false, |
1195 | .page = NULL, | ||
1196 | }; | 1131 | }; |
1197 | 1132 | ||
1198 | return __compact_pgdat(pgdat, &cc); | 1133 | return __compact_pgdat(pgdat, &cc); |
@@ -1203,7 +1138,6 @@ static int compact_node(int nid) | |||
1203 | struct compact_control cc = { | 1138 | struct compact_control cc = { |
1204 | .order = -1, | 1139 | .order = -1, |
1205 | .sync = true, | 1140 | .sync = true, |
1206 | .page = NULL, | ||
1207 | }; | 1141 | }; |
1208 | 1142 | ||
1209 | return __compact_pgdat(NODE_DATA(nid), &cc); | 1143 | return __compact_pgdat(NODE_DATA(nid), &cc); |
diff --git a/mm/internal.h b/mm/internal.h index d597f94cc205..9ba21100ebf3 100644 --- a/mm/internal.h +++ b/mm/internal.h | |||
@@ -135,7 +135,6 @@ struct compact_control { | |||
135 | int migratetype; /* MOVABLE, RECLAIMABLE etc */ | 135 | int migratetype; /* MOVABLE, RECLAIMABLE etc */ |
136 | struct zone *zone; | 136 | struct zone *zone; |
137 | bool contended; /* True if a lock was contended */ | 137 | bool contended; /* True if a lock was contended */ |
138 | struct page **page; /* Page captured of requested size */ | ||
139 | }; | 138 | }; |
140 | 139 | ||
141 | unsigned long | 140 | unsigned long |
diff --git a/mm/page_alloc.c b/mm/page_alloc.c index c957805a7f0e..df2022ff0c8a 100644 --- a/mm/page_alloc.c +++ b/mm/page_alloc.c | |||
@@ -1384,14 +1384,8 @@ void split_page(struct page *page, unsigned int order) | |||
1384 | set_page_refcounted(page + i); | 1384 | set_page_refcounted(page + i); |
1385 | } | 1385 | } |
1386 | 1386 | ||
1387 | /* | 1387 | static int __isolate_free_page(struct page *page, unsigned int order) |
1388 | * Similar to the split_page family of functions except that the page | ||
1389 | * required at the given order and being isolated now to prevent races | ||
1390 | * with parallel allocators | ||
1391 | */ | ||
1392 | int capture_free_page(struct page *page, int alloc_order, int migratetype) | ||
1393 | { | 1388 | { |
1394 | unsigned int order; | ||
1395 | unsigned long watermark; | 1389 | unsigned long watermark; |
1396 | struct zone *zone; | 1390 | struct zone *zone; |
1397 | int mt; | 1391 | int mt; |
@@ -1399,7 +1393,6 @@ int capture_free_page(struct page *page, int alloc_order, int migratetype) | |||
1399 | BUG_ON(!PageBuddy(page)); | 1393 | BUG_ON(!PageBuddy(page)); |
1400 | 1394 | ||
1401 | zone = page_zone(page); | 1395 | zone = page_zone(page); |
1402 | order = page_order(page); | ||
1403 | mt = get_pageblock_migratetype(page); | 1396 | mt = get_pageblock_migratetype(page); |
1404 | 1397 | ||
1405 | if (mt != MIGRATE_ISOLATE) { | 1398 | if (mt != MIGRATE_ISOLATE) { |
@@ -1408,7 +1401,7 @@ int capture_free_page(struct page *page, int alloc_order, int migratetype) | |||
1408 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | 1401 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) |
1409 | return 0; | 1402 | return 0; |
1410 | 1403 | ||
1411 | __mod_zone_freepage_state(zone, -(1UL << alloc_order), mt); | 1404 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
1412 | } | 1405 | } |
1413 | 1406 | ||
1414 | /* Remove page from free list */ | 1407 | /* Remove page from free list */ |
@@ -1416,11 +1409,7 @@ int capture_free_page(struct page *page, int alloc_order, int migratetype) | |||
1416 | zone->free_area[order].nr_free--; | 1409 | zone->free_area[order].nr_free--; |
1417 | rmv_page_order(page); | 1410 | rmv_page_order(page); |
1418 | 1411 | ||
1419 | if (alloc_order != order) | 1412 | /* Set the pageblock if the isolated page is at least a pageblock */ |
1420 | expand(zone, page, alloc_order, order, | ||
1421 | &zone->free_area[order], migratetype); | ||
1422 | |||
1423 | /* Set the pageblock if the captured page is at least a pageblock */ | ||
1424 | if (order >= pageblock_order - 1) { | 1413 | if (order >= pageblock_order - 1) { |
1425 | struct page *endpage = page + (1 << order) - 1; | 1414 | struct page *endpage = page + (1 << order) - 1; |
1426 | for (; page < endpage; page += pageblock_nr_pages) { | 1415 | for (; page < endpage; page += pageblock_nr_pages) { |
@@ -1431,7 +1420,7 @@ int capture_free_page(struct page *page, int alloc_order, int migratetype) | |||
1431 | } | 1420 | } |
1432 | } | 1421 | } |
1433 | 1422 | ||
1434 | return 1UL << alloc_order; | 1423 | return 1UL << order; |
1435 | } | 1424 | } |
1436 | 1425 | ||
1437 | /* | 1426 | /* |
@@ -1449,10 +1438,9 @@ int split_free_page(struct page *page) | |||
1449 | unsigned int order; | 1438 | unsigned int order; |
1450 | int nr_pages; | 1439 | int nr_pages; |
1451 | 1440 | ||
1452 | BUG_ON(!PageBuddy(page)); | ||
1453 | order = page_order(page); | 1441 | order = page_order(page); |
1454 | 1442 | ||
1455 | nr_pages = capture_free_page(page, order, 0); | 1443 | nr_pages = __isolate_free_page(page, order); |
1456 | if (!nr_pages) | 1444 | if (!nr_pages) |
1457 | return 0; | 1445 | return 0; |
1458 | 1446 | ||
@@ -2136,8 +2124,6 @@ __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |||
2136 | bool *contended_compaction, bool *deferred_compaction, | 2124 | bool *contended_compaction, bool *deferred_compaction, |
2137 | unsigned long *did_some_progress) | 2125 | unsigned long *did_some_progress) |
2138 | { | 2126 | { |
2139 | struct page *page = NULL; | ||
2140 | |||
2141 | if (!order) | 2127 | if (!order) |
2142 | return NULL; | 2128 | return NULL; |
2143 | 2129 | ||
@@ -2149,16 +2135,12 @@ __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |||
2149 | current->flags |= PF_MEMALLOC; | 2135 | current->flags |= PF_MEMALLOC; |
2150 | *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask, | 2136 | *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask, |
2151 | nodemask, sync_migration, | 2137 | nodemask, sync_migration, |
2152 | contended_compaction, &page); | 2138 | contended_compaction); |
2153 | current->flags &= ~PF_MEMALLOC; | 2139 | current->flags &= ~PF_MEMALLOC; |
2154 | 2140 | ||
2155 | /* If compaction captured a page, prep and use it */ | ||
2156 | if (page) { | ||
2157 | prep_new_page(page, order, gfp_mask); | ||
2158 | goto got_page; | ||
2159 | } | ||
2160 | |||
2161 | if (*did_some_progress != COMPACT_SKIPPED) { | 2141 | if (*did_some_progress != COMPACT_SKIPPED) { |
2142 | struct page *page; | ||
2143 | |||
2162 | /* Page migration frees to the PCP lists but we want merging */ | 2144 | /* Page migration frees to the PCP lists but we want merging */ |
2163 | drain_pages(get_cpu()); | 2145 | drain_pages(get_cpu()); |
2164 | put_cpu(); | 2146 | put_cpu(); |
@@ -2168,7 +2150,6 @@ __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |||
2168 | alloc_flags & ~ALLOC_NO_WATERMARKS, | 2150 | alloc_flags & ~ALLOC_NO_WATERMARKS, |
2169 | preferred_zone, migratetype); | 2151 | preferred_zone, migratetype); |
2170 | if (page) { | 2152 | if (page) { |
2171 | got_page: | ||
2172 | preferred_zone->compact_blockskip_flush = false; | 2153 | preferred_zone->compact_blockskip_flush = false; |
2173 | preferred_zone->compact_considered = 0; | 2154 | preferred_zone->compact_considered = 0; |
2174 | preferred_zone->compact_defer_shift = 0; | 2155 | preferred_zone->compact_defer_shift = 0; |