Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/mm/compaction.c |
| 3 | * |
| 4 | * Memory compaction for the reduction of external fragmentation. Note that |
| 5 | * this heavily depends upon page migration to do all the real heavy |
| 6 | * lifting |
| 7 | * |
| 8 | * Copyright IBM Corp. 2007-2010 Mel Gorman <mel@csn.ul.ie> |
| 9 | */ |
| 10 | #include <linux/swap.h> |
| 11 | #include <linux/migrate.h> |
| 12 | #include <linux/compaction.h> |
| 13 | #include <linux/mm_inline.h> |
| 14 | #include <linux/backing-dev.h> |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 15 | #include <linux/sysctl.h> |
Mel Gorman | ed4a6d7 | 2010-05-24 14:32:29 -0700 | [diff] [blame] | 16 | #include <linux/sysfs.h> |
Rafael Aquini | bf6bddf | 2012-12-11 16:02:42 -0800 | [diff] [blame] | 17 | #include <linux/balloon_compaction.h> |
Minchan Kim | 194159f | 2013-02-22 16:33:58 -0800 | [diff] [blame] | 18 | #include <linux/page-isolation.h> |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 19 | #include "internal.h" |
| 20 | |
Minchan Kim | 010fc29 | 2012-12-20 15:05:06 -0800 | [diff] [blame] | 21 | #ifdef CONFIG_COMPACTION |
| 22 | static inline void count_compact_event(enum vm_event_item item) |
| 23 | { |
| 24 | count_vm_event(item); |
| 25 | } |
| 26 | |
| 27 | static inline void count_compact_events(enum vm_event_item item, long delta) |
| 28 | { |
| 29 | count_vm_events(item, delta); |
| 30 | } |
| 31 | #else |
| 32 | #define count_compact_event(item) do { } while (0) |
| 33 | #define count_compact_events(item, delta) do { } while (0) |
| 34 | #endif |
| 35 | |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 36 | #if defined CONFIG_COMPACTION || defined CONFIG_CMA |
| 37 | |
Mel Gorman | b7aba69 | 2011-01-13 15:45:54 -0800 | [diff] [blame] | 38 | #define CREATE_TRACE_POINTS |
| 39 | #include <trace/events/compaction.h> |
| 40 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 41 | static unsigned long release_freepages(struct list_head *freelist) |
| 42 | { |
| 43 | struct page *page, *next; |
| 44 | unsigned long count = 0; |
| 45 | |
| 46 | list_for_each_entry_safe(page, next, freelist, lru) { |
| 47 | list_del(&page->lru); |
| 48 | __free_page(page); |
| 49 | count++; |
| 50 | } |
| 51 | |
| 52 | return count; |
| 53 | } |
| 54 | |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 55 | static void map_pages(struct list_head *list) |
| 56 | { |
| 57 | struct page *page; |
| 58 | |
| 59 | list_for_each_entry(page, list, lru) { |
| 60 | arch_alloc_page(page, 0); |
| 61 | kernel_map_pages(page, 1, 1); |
| 62 | } |
| 63 | } |
| 64 | |
Michal Nazarewicz | 47118af | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 65 | static inline bool migrate_async_suitable(int migratetype) |
| 66 | { |
| 67 | return is_migrate_cma(migratetype) || migratetype == MIGRATE_MOVABLE; |
| 68 | } |
| 69 | |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 70 | #ifdef CONFIG_COMPACTION |
| 71 | /* Returns true if the pageblock should be scanned for pages to isolate. */ |
| 72 | static inline bool isolation_suitable(struct compact_control *cc, |
| 73 | struct page *page) |
| 74 | { |
| 75 | if (cc->ignore_skip_hint) |
| 76 | return true; |
| 77 | |
| 78 | return !get_pageblock_skip(page); |
| 79 | } |
| 80 | |
| 81 | /* |
| 82 | * This function is called to clear all cached information on pageblocks that |
| 83 | * should be skipped for page isolation when the migrate and free page scanner |
| 84 | * meet. |
| 85 | */ |
Mel Gorman | 6299702 | 2012-10-08 16:32:47 -0700 | [diff] [blame] | 86 | static void __reset_isolation_suitable(struct zone *zone) |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 87 | { |
| 88 | unsigned long start_pfn = zone->zone_start_pfn; |
Cody P Schafer | 108bcc9 | 2013-02-22 16:35:23 -0800 | [diff] [blame] | 89 | unsigned long end_pfn = zone_end_pfn(zone); |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 90 | unsigned long pfn; |
| 91 | |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 92 | zone->compact_cached_migrate_pfn = start_pfn; |
| 93 | zone->compact_cached_free_pfn = end_pfn; |
Mel Gorman | 6299702 | 2012-10-08 16:32:47 -0700 | [diff] [blame] | 94 | zone->compact_blockskip_flush = false; |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 95 | |
| 96 | /* Walk the zone and mark every pageblock as suitable for isolation */ |
| 97 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { |
| 98 | struct page *page; |
| 99 | |
| 100 | cond_resched(); |
| 101 | |
| 102 | if (!pfn_valid(pfn)) |
| 103 | continue; |
| 104 | |
| 105 | page = pfn_to_page(pfn); |
| 106 | if (zone != page_zone(page)) |
| 107 | continue; |
| 108 | |
| 109 | clear_pageblock_skip(page); |
| 110 | } |
| 111 | } |
| 112 | |
Mel Gorman | 6299702 | 2012-10-08 16:32:47 -0700 | [diff] [blame] | 113 | void reset_isolation_suitable(pg_data_t *pgdat) |
| 114 | { |
| 115 | int zoneid; |
| 116 | |
| 117 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { |
| 118 | struct zone *zone = &pgdat->node_zones[zoneid]; |
| 119 | if (!populated_zone(zone)) |
| 120 | continue; |
| 121 | |
| 122 | /* Only flush if a full compaction finished recently */ |
| 123 | if (zone->compact_blockskip_flush) |
| 124 | __reset_isolation_suitable(zone); |
| 125 | } |
| 126 | } |
| 127 | |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 128 | /* |
| 129 | * If no pages were isolated then mark this pageblock to be skipped in the |
Mel Gorman | 6299702 | 2012-10-08 16:32:47 -0700 | [diff] [blame] | 130 | * future. The information is later cleared by __reset_isolation_suitable(). |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 131 | */ |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 132 | static void update_pageblock_skip(struct compact_control *cc, |
| 133 | struct page *page, unsigned long nr_isolated, |
| 134 | bool migrate_scanner) |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 135 | { |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 136 | struct zone *zone = cc->zone; |
Joonsoo Kim | 6815bf3 | 2013-12-18 17:08:52 -0800 | [diff] [blame] | 137 | |
| 138 | if (cc->ignore_skip_hint) |
| 139 | return; |
| 140 | |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 141 | if (!page) |
| 142 | return; |
| 143 | |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 144 | if (!nr_isolated) { |
| 145 | unsigned long pfn = page_to_pfn(page); |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 146 | set_pageblock_skip(page); |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 147 | |
| 148 | /* Update where compaction should restart */ |
| 149 | if (migrate_scanner) { |
| 150 | if (!cc->finished_update_migrate && |
| 151 | pfn > zone->compact_cached_migrate_pfn) |
| 152 | zone->compact_cached_migrate_pfn = pfn; |
| 153 | } else { |
| 154 | if (!cc->finished_update_free && |
| 155 | pfn < zone->compact_cached_free_pfn) |
| 156 | zone->compact_cached_free_pfn = pfn; |
| 157 | } |
| 158 | } |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 159 | } |
| 160 | #else |
| 161 | static inline bool isolation_suitable(struct compact_control *cc, |
| 162 | struct page *page) |
| 163 | { |
| 164 | return true; |
| 165 | } |
| 166 | |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 167 | static void update_pageblock_skip(struct compact_control *cc, |
| 168 | struct page *page, unsigned long nr_isolated, |
| 169 | bool migrate_scanner) |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 170 | { |
| 171 | } |
| 172 | #endif /* CONFIG_COMPACTION */ |
| 173 | |
Mel Gorman | 2a1402a | 2012-10-08 16:32:33 -0700 | [diff] [blame] | 174 | static inline bool should_release_lock(spinlock_t *lock) |
| 175 | { |
| 176 | return need_resched() || spin_is_contended(lock); |
| 177 | } |
| 178 | |
Michal Nazarewicz | 85aa125 | 2012-01-30 13:24:03 +0100 | [diff] [blame] | 179 | /* |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 180 | * Compaction requires the taking of some coarse locks that are potentially |
| 181 | * very heavily contended. Check if the process needs to be scheduled or |
| 182 | * if the lock is contended. For async compaction, back out in the event |
| 183 | * if contention is severe. For sync compaction, schedule. |
| 184 | * |
| 185 | * Returns true if the lock is held. |
| 186 | * Returns false if the lock is released and compaction should abort |
| 187 | */ |
| 188 | static bool compact_checklock_irqsave(spinlock_t *lock, unsigned long *flags, |
| 189 | bool locked, struct compact_control *cc) |
| 190 | { |
Mel Gorman | 2a1402a | 2012-10-08 16:32:33 -0700 | [diff] [blame] | 191 | if (should_release_lock(lock)) { |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 192 | if (locked) { |
| 193 | spin_unlock_irqrestore(lock, *flags); |
| 194 | locked = false; |
| 195 | } |
| 196 | |
| 197 | /* async aborts if taking too long or contended */ |
| 198 | if (!cc->sync) { |
Shaohua Li | e64c523 | 2012-10-08 16:32:27 -0700 | [diff] [blame] | 199 | cc->contended = true; |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 200 | return false; |
| 201 | } |
| 202 | |
| 203 | cond_resched(); |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 204 | } |
| 205 | |
| 206 | if (!locked) |
| 207 | spin_lock_irqsave(lock, *flags); |
| 208 | return true; |
| 209 | } |
| 210 | |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 211 | /* Returns true if the page is within a block suitable for migration to */ |
| 212 | static bool suitable_migration_target(struct page *page) |
| 213 | { |
Joonsoo Kim | 7d348b9 | 2014-04-07 15:37:03 -0700 | [diff] [blame] | 214 | /* If the page is a large free page, then disallow migration */ |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 215 | if (PageBuddy(page) && page_order(page) >= pageblock_order) |
Joonsoo Kim | 7d348b9 | 2014-04-07 15:37:03 -0700 | [diff] [blame] | 216 | return false; |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 217 | |
| 218 | /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */ |
Joonsoo Kim | 7d348b9 | 2014-04-07 15:37:03 -0700 | [diff] [blame] | 219 | if (migrate_async_suitable(get_pageblock_migratetype(page))) |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 220 | return true; |
| 221 | |
| 222 | /* Otherwise skip the block */ |
| 223 | return false; |
| 224 | } |
| 225 | |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 226 | /* |
Jerome Marchand | 9e4be47 | 2013-11-12 15:07:12 -0800 | [diff] [blame] | 227 | * Isolate free pages onto a private freelist. If @strict is true, will abort |
| 228 | * returning 0 on any invalid PFNs or non-free pages inside of the pageblock |
| 229 | * (even though it may still end up isolating some pages). |
Michal Nazarewicz | 85aa125 | 2012-01-30 13:24:03 +0100 | [diff] [blame] | 230 | */ |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 231 | static unsigned long isolate_freepages_block(struct compact_control *cc, |
| 232 | unsigned long blockpfn, |
Michal Nazarewicz | 85aa125 | 2012-01-30 13:24:03 +0100 | [diff] [blame] | 233 | unsigned long end_pfn, |
| 234 | struct list_head *freelist, |
| 235 | bool strict) |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 236 | { |
Mel Gorman | b7aba69 | 2011-01-13 15:45:54 -0800 | [diff] [blame] | 237 | int nr_scanned = 0, total_isolated = 0; |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 238 | struct page *cursor, *valid_page = NULL; |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 239 | unsigned long flags; |
| 240 | bool locked = false; |
Joonsoo Kim | 01ead53 | 2014-04-07 15:37:04 -0700 | [diff] [blame] | 241 | bool checked_pageblock = false; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 242 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 243 | cursor = pfn_to_page(blockpfn); |
| 244 | |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 245 | /* Isolate free pages. */ |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 246 | for (; blockpfn < end_pfn; blockpfn++, cursor++) { |
| 247 | int isolated, i; |
| 248 | struct page *page = cursor; |
| 249 | |
Mel Gorman | b7aba69 | 2011-01-13 15:45:54 -0800 | [diff] [blame] | 250 | nr_scanned++; |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 251 | if (!pfn_valid_within(blockpfn)) |
Laura Abbott | 2af120b | 2014-03-10 15:49:44 -0700 | [diff] [blame] | 252 | goto isolate_fail; |
| 253 | |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 254 | if (!valid_page) |
| 255 | valid_page = page; |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 256 | if (!PageBuddy(page)) |
Laura Abbott | 2af120b | 2014-03-10 15:49:44 -0700 | [diff] [blame] | 257 | goto isolate_fail; |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 258 | |
| 259 | /* |
| 260 | * The zone lock must be held to isolate freepages. |
| 261 | * Unfortunately this is a very coarse lock and can be |
| 262 | * heavily contended if there are parallel allocations |
| 263 | * or parallel compactions. For async compaction do not |
| 264 | * spin on the lock and we acquire the lock as late as |
| 265 | * possible. |
| 266 | */ |
| 267 | locked = compact_checklock_irqsave(&cc->zone->lock, &flags, |
| 268 | locked, cc); |
| 269 | if (!locked) |
| 270 | break; |
| 271 | |
| 272 | /* Recheck this is a suitable migration target under lock */ |
Joonsoo Kim | 01ead53 | 2014-04-07 15:37:04 -0700 | [diff] [blame] | 273 | if (!strict && !checked_pageblock) { |
| 274 | /* |
| 275 | * We need to check suitability of pageblock only once |
| 276 | * and this isolate_freepages_block() is called with |
| 277 | * pageblock range, so just check once is sufficient. |
| 278 | */ |
| 279 | checked_pageblock = true; |
| 280 | if (!suitable_migration_target(page)) |
| 281 | break; |
| 282 | } |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 283 | |
| 284 | /* Recheck this is a buddy page under lock */ |
| 285 | if (!PageBuddy(page)) |
Laura Abbott | 2af120b | 2014-03-10 15:49:44 -0700 | [diff] [blame] | 286 | goto isolate_fail; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 287 | |
| 288 | /* Found a free page, break it into order-0 pages */ |
| 289 | isolated = split_free_page(page); |
| 290 | total_isolated += isolated; |
| 291 | for (i = 0; i < isolated; i++) { |
| 292 | list_add(&page->lru, freelist); |
| 293 | page++; |
| 294 | } |
| 295 | |
| 296 | /* If a page was split, advance to the end of it */ |
| 297 | if (isolated) { |
| 298 | blockpfn += isolated - 1; |
| 299 | cursor += isolated - 1; |
Laura Abbott | 2af120b | 2014-03-10 15:49:44 -0700 | [diff] [blame] | 300 | continue; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 301 | } |
Laura Abbott | 2af120b | 2014-03-10 15:49:44 -0700 | [diff] [blame] | 302 | |
| 303 | isolate_fail: |
| 304 | if (strict) |
| 305 | break; |
| 306 | else |
| 307 | continue; |
| 308 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 309 | } |
| 310 | |
Mel Gorman | b7aba69 | 2011-01-13 15:45:54 -0800 | [diff] [blame] | 311 | trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated); |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 312 | |
| 313 | /* |
| 314 | * If strict isolation is requested by CMA then check that all the |
| 315 | * pages requested were isolated. If there were any failures, 0 is |
| 316 | * returned and CMA will fail. |
| 317 | */ |
Laura Abbott | 2af120b | 2014-03-10 15:49:44 -0700 | [diff] [blame] | 318 | if (strict && blockpfn < end_pfn) |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 319 | total_isolated = 0; |
| 320 | |
| 321 | if (locked) |
| 322 | spin_unlock_irqrestore(&cc->zone->lock, flags); |
| 323 | |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 324 | /* Update the pageblock-skip if the whole pageblock was scanned */ |
| 325 | if (blockpfn == end_pfn) |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 326 | update_pageblock_skip(cc, valid_page, total_isolated, false); |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 327 | |
Minchan Kim | 010fc29 | 2012-12-20 15:05:06 -0800 | [diff] [blame] | 328 | count_compact_events(COMPACTFREE_SCANNED, nr_scanned); |
Mel Gorman | 397487d | 2012-10-19 12:00:10 +0100 | [diff] [blame] | 329 | if (total_isolated) |
Minchan Kim | 010fc29 | 2012-12-20 15:05:06 -0800 | [diff] [blame] | 330 | count_compact_events(COMPACTISOLATED, total_isolated); |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 331 | return total_isolated; |
| 332 | } |
| 333 | |
Michal Nazarewicz | 85aa125 | 2012-01-30 13:24:03 +0100 | [diff] [blame] | 334 | /** |
| 335 | * isolate_freepages_range() - isolate free pages. |
| 336 | * @start_pfn: The first PFN to start isolating. |
| 337 | * @end_pfn: The one-past-last PFN. |
| 338 | * |
| 339 | * Non-free pages, invalid PFNs, or zone boundaries within the |
| 340 | * [start_pfn, end_pfn) range are considered errors, cause function to |
| 341 | * undo its actions and return zero. |
| 342 | * |
| 343 | * Otherwise, function returns one-past-the-last PFN of isolated page |
| 344 | * (which may be greater then end_pfn if end fell in a middle of |
| 345 | * a free page). |
| 346 | */ |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 347 | unsigned long |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 348 | isolate_freepages_range(struct compact_control *cc, |
| 349 | unsigned long start_pfn, unsigned long end_pfn) |
Michal Nazarewicz | 85aa125 | 2012-01-30 13:24:03 +0100 | [diff] [blame] | 350 | { |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 351 | unsigned long isolated, pfn, block_end_pfn; |
Michal Nazarewicz | 85aa125 | 2012-01-30 13:24:03 +0100 | [diff] [blame] | 352 | LIST_HEAD(freelist); |
| 353 | |
Michal Nazarewicz | 85aa125 | 2012-01-30 13:24:03 +0100 | [diff] [blame] | 354 | for (pfn = start_pfn; pfn < end_pfn; pfn += isolated) { |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 355 | if (!pfn_valid(pfn) || cc->zone != page_zone(pfn_to_page(pfn))) |
Michal Nazarewicz | 85aa125 | 2012-01-30 13:24:03 +0100 | [diff] [blame] | 356 | break; |
| 357 | |
| 358 | /* |
| 359 | * On subsequent iterations ALIGN() is actually not needed, |
| 360 | * but we keep it that we not to complicate the code. |
| 361 | */ |
| 362 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); |
| 363 | block_end_pfn = min(block_end_pfn, end_pfn); |
| 364 | |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 365 | isolated = isolate_freepages_block(cc, pfn, block_end_pfn, |
Michal Nazarewicz | 85aa125 | 2012-01-30 13:24:03 +0100 | [diff] [blame] | 366 | &freelist, true); |
Michal Nazarewicz | 85aa125 | 2012-01-30 13:24:03 +0100 | [diff] [blame] | 367 | |
| 368 | /* |
| 369 | * In strict mode, isolate_freepages_block() returns 0 if |
| 370 | * there are any holes in the block (ie. invalid PFNs or |
| 371 | * non-free pages). |
| 372 | */ |
| 373 | if (!isolated) |
| 374 | break; |
| 375 | |
| 376 | /* |
| 377 | * If we managed to isolate pages, it is always (1 << n) * |
| 378 | * pageblock_nr_pages for some non-negative n. (Max order |
| 379 | * page may span two pageblocks). |
| 380 | */ |
| 381 | } |
| 382 | |
| 383 | /* split_free_page does not map the pages */ |
| 384 | map_pages(&freelist); |
| 385 | |
| 386 | if (pfn < end_pfn) { |
| 387 | /* Loop terminated early, cleanup. */ |
| 388 | release_freepages(&freelist); |
| 389 | return 0; |
| 390 | } |
| 391 | |
| 392 | /* We don't use freelists for anything. */ |
| 393 | return pfn; |
| 394 | } |
| 395 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 396 | /* Update the number of anon and file isolated pages in the zone */ |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 397 | static void acct_isolated(struct zone *zone, bool locked, struct compact_control *cc) |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 398 | { |
| 399 | struct page *page; |
Minchan Kim | b9e84ac | 2011-10-31 17:06:44 -0700 | [diff] [blame] | 400 | unsigned int count[2] = { 0, }; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 401 | |
Minchan Kim | b9e84ac | 2011-10-31 17:06:44 -0700 | [diff] [blame] | 402 | list_for_each_entry(page, &cc->migratepages, lru) |
| 403 | count[!!page_is_file_cache(page)]++; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 404 | |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 405 | /* If locked we can use the interrupt unsafe versions */ |
| 406 | if (locked) { |
| 407 | __mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]); |
| 408 | __mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]); |
| 409 | } else { |
| 410 | mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]); |
| 411 | mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]); |
| 412 | } |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 413 | } |
| 414 | |
| 415 | /* Similar to reclaim, but different enough that they don't share logic */ |
| 416 | static bool too_many_isolated(struct zone *zone) |
| 417 | { |
Minchan Kim | bc69304 | 2010-09-09 16:38:00 -0700 | [diff] [blame] | 418 | unsigned long active, inactive, isolated; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 419 | |
| 420 | inactive = zone_page_state(zone, NR_INACTIVE_FILE) + |
| 421 | zone_page_state(zone, NR_INACTIVE_ANON); |
Minchan Kim | bc69304 | 2010-09-09 16:38:00 -0700 | [diff] [blame] | 422 | active = zone_page_state(zone, NR_ACTIVE_FILE) + |
| 423 | zone_page_state(zone, NR_ACTIVE_ANON); |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 424 | isolated = zone_page_state(zone, NR_ISOLATED_FILE) + |
| 425 | zone_page_state(zone, NR_ISOLATED_ANON); |
| 426 | |
Minchan Kim | bc69304 | 2010-09-09 16:38:00 -0700 | [diff] [blame] | 427 | return isolated > (inactive + active) / 2; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 428 | } |
| 429 | |
Michal Nazarewicz | 2fe86e0 | 2012-01-30 13:16:26 +0100 | [diff] [blame] | 430 | /** |
| 431 | * isolate_migratepages_range() - isolate all migrate-able pages in range. |
| 432 | * @zone: Zone pages are in. |
| 433 | * @cc: Compaction control structure. |
| 434 | * @low_pfn: The first PFN of the range. |
| 435 | * @end_pfn: The one-past-the-last PFN of the range. |
Minchan Kim | e46a287 | 2012-10-08 16:33:48 -0700 | [diff] [blame] | 436 | * @unevictable: true if it allows to isolate unevictable pages |
Michal Nazarewicz | 2fe86e0 | 2012-01-30 13:16:26 +0100 | [diff] [blame] | 437 | * |
| 438 | * Isolate all pages that can be migrated from the range specified by |
| 439 | * [low_pfn, end_pfn). Returns zero if there is a fatal signal |
| 440 | * pending), otherwise PFN of the first page that was not scanned |
| 441 | * (which may be both less, equal to or more then end_pfn). |
| 442 | * |
| 443 | * Assumes that cc->migratepages is empty and cc->nr_migratepages is |
| 444 | * zero. |
| 445 | * |
| 446 | * Apart from cc->migratepages and cc->nr_migratetypes this function |
| 447 | * does not modify any cc's fields, in particular it does not modify |
| 448 | * (or read for that matter) cc->migrate_pfn. |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 449 | */ |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 450 | unsigned long |
Michal Nazarewicz | 2fe86e0 | 2012-01-30 13:16:26 +0100 | [diff] [blame] | 451 | isolate_migratepages_range(struct zone *zone, struct compact_control *cc, |
Minchan Kim | e46a287 | 2012-10-08 16:33:48 -0700 | [diff] [blame] | 452 | unsigned long low_pfn, unsigned long end_pfn, bool unevictable) |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 453 | { |
Mel Gorman | 9927af74 | 2011-01-13 15:45:59 -0800 | [diff] [blame] | 454 | unsigned long last_pageblock_nr = 0, pageblock_nr; |
Mel Gorman | b7aba69 | 2011-01-13 15:45:54 -0800 | [diff] [blame] | 455 | unsigned long nr_scanned = 0, nr_isolated = 0; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 456 | struct list_head *migratelist = &cc->migratepages; |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 457 | struct lruvec *lruvec; |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 458 | unsigned long flags; |
Mel Gorman | 2a1402a | 2012-10-08 16:32:33 -0700 | [diff] [blame] | 459 | bool locked = false; |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 460 | struct page *page = NULL, *valid_page = NULL; |
Vlastimil Babka | 50b5b09 | 2014-01-21 15:51:10 -0800 | [diff] [blame] | 461 | bool skipped_async_unsuitable = false; |
David Rientjes | da1c67a | 2014-04-07 15:37:34 -0700 | [diff] [blame] | 462 | const isolate_mode_t mode = (!cc->sync ? ISOLATE_ASYNC_MIGRATE : 0) | |
| 463 | (unevictable ? ISOLATE_UNEVICTABLE : 0); |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 464 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 465 | /* |
| 466 | * Ensure that there are not too many pages isolated from the LRU |
| 467 | * list by either parallel reclaimers or compaction. If there are, |
| 468 | * delay for some time until fewer pages are isolated |
| 469 | */ |
| 470 | while (unlikely(too_many_isolated(zone))) { |
Mel Gorman | f9e35b3 | 2011-06-15 15:08:52 -0700 | [diff] [blame] | 471 | /* async migration should just abort */ |
Linus Torvalds | 68e3e92 | 2012-06-03 20:05:57 -0700 | [diff] [blame] | 472 | if (!cc->sync) |
Michal Nazarewicz | 2fe86e0 | 2012-01-30 13:16:26 +0100 | [diff] [blame] | 473 | return 0; |
Mel Gorman | f9e35b3 | 2011-06-15 15:08:52 -0700 | [diff] [blame] | 474 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 475 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
| 476 | |
| 477 | if (fatal_signal_pending(current)) |
Michal Nazarewicz | 2fe86e0 | 2012-01-30 13:16:26 +0100 | [diff] [blame] | 478 | return 0; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 479 | } |
| 480 | |
| 481 | /* Time to isolate some pages for migration */ |
Andrea Arcangeli | b2eef8c | 2011-03-22 16:33:10 -0700 | [diff] [blame] | 482 | cond_resched(); |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 483 | for (; low_pfn < end_pfn; low_pfn++) { |
Andrea Arcangeli | b2eef8c | 2011-03-22 16:33:10 -0700 | [diff] [blame] | 484 | /* give a chance to irqs before checking need_resched() */ |
Joonsoo Kim | be1aa03 | 2014-04-07 15:37:05 -0700 | [diff] [blame] | 485 | if (locked && !(low_pfn % SWAP_CLUSTER_MAX)) { |
Mel Gorman | 2a1402a | 2012-10-08 16:32:33 -0700 | [diff] [blame] | 486 | if (should_release_lock(&zone->lru_lock)) { |
| 487 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
| 488 | locked = false; |
| 489 | } |
Andrea Arcangeli | b2eef8c | 2011-03-22 16:33:10 -0700 | [diff] [blame] | 490 | } |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 491 | |
Mel Gorman | 0bf380b | 2012-02-03 15:37:18 -0800 | [diff] [blame] | 492 | /* |
| 493 | * migrate_pfn does not necessarily start aligned to a |
| 494 | * pageblock. Ensure that pfn_valid is called when moving |
| 495 | * into a new MAX_ORDER_NR_PAGES range in case of large |
| 496 | * memory holes within the zone |
| 497 | */ |
| 498 | if ((low_pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) { |
| 499 | if (!pfn_valid(low_pfn)) { |
| 500 | low_pfn += MAX_ORDER_NR_PAGES - 1; |
| 501 | continue; |
| 502 | } |
| 503 | } |
| 504 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 505 | if (!pfn_valid_within(low_pfn)) |
| 506 | continue; |
Mel Gorman | b7aba69 | 2011-01-13 15:45:54 -0800 | [diff] [blame] | 507 | nr_scanned++; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 508 | |
Mel Gorman | dc90860 | 2012-02-08 17:13:38 -0800 | [diff] [blame] | 509 | /* |
| 510 | * Get the page and ensure the page is within the same zone. |
| 511 | * See the comment in isolate_freepages about overlapping |
| 512 | * nodes. It is deliberate that the new zone lock is not taken |
| 513 | * as memory compaction should not move pages between nodes. |
| 514 | */ |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 515 | page = pfn_to_page(low_pfn); |
Mel Gorman | dc90860 | 2012-02-08 17:13:38 -0800 | [diff] [blame] | 516 | if (page_zone(page) != zone) |
| 517 | continue; |
| 518 | |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 519 | if (!valid_page) |
| 520 | valid_page = page; |
| 521 | |
| 522 | /* If isolation recently failed, do not retry */ |
| 523 | pageblock_nr = low_pfn >> pageblock_order; |
Joonsoo Kim | c122b20 | 2014-04-07 15:37:06 -0700 | [diff] [blame] | 524 | if (last_pageblock_nr != pageblock_nr) { |
| 525 | int mt; |
| 526 | |
| 527 | last_pageblock_nr = pageblock_nr; |
| 528 | if (!isolation_suitable(cc, page)) |
| 529 | goto next_pageblock; |
| 530 | |
| 531 | /* |
| 532 | * For async migration, also only scan in MOVABLE |
| 533 | * blocks. Async migration is optimistic to see if |
| 534 | * the minimum amount of work satisfies the allocation |
| 535 | */ |
| 536 | mt = get_pageblock_migratetype(page); |
| 537 | if (!cc->sync && !migrate_async_suitable(mt)) { |
| 538 | cc->finished_update_migrate = true; |
| 539 | skipped_async_unsuitable = true; |
| 540 | goto next_pageblock; |
| 541 | } |
| 542 | } |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 543 | |
Mel Gorman | 6c14466 | 2014-01-23 15:53:38 -0800 | [diff] [blame] | 544 | /* |
| 545 | * Skip if free. page_order cannot be used without zone->lock |
| 546 | * as nothing prevents parallel allocations or buddy merging. |
| 547 | */ |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 548 | if (PageBuddy(page)) |
| 549 | continue; |
| 550 | |
Mel Gorman | 9927af74 | 2011-01-13 15:45:59 -0800 | [diff] [blame] | 551 | /* |
Rafael Aquini | bf6bddf | 2012-12-11 16:02:42 -0800 | [diff] [blame] | 552 | * Check may be lockless but that's ok as we recheck later. |
| 553 | * It's possible to migrate LRU pages and balloon pages |
| 554 | * Skip any other type of page |
| 555 | */ |
| 556 | if (!PageLRU(page)) { |
| 557 | if (unlikely(balloon_page_movable(page))) { |
| 558 | if (locked && balloon_page_isolate(page)) { |
| 559 | /* Successfully isolated */ |
Joonsoo Kim | b6c7501 | 2014-04-07 15:37:07 -0700 | [diff] [blame] | 560 | goto isolate_success; |
Rafael Aquini | bf6bddf | 2012-12-11 16:02:42 -0800 | [diff] [blame] | 561 | } |
| 562 | } |
Andrea Arcangeli | bc83501 | 2011-01-13 15:47:08 -0800 | [diff] [blame] | 563 | continue; |
Rafael Aquini | bf6bddf | 2012-12-11 16:02:42 -0800 | [diff] [blame] | 564 | } |
Andrea Arcangeli | bc83501 | 2011-01-13 15:47:08 -0800 | [diff] [blame] | 565 | |
| 566 | /* |
Mel Gorman | 2a1402a | 2012-10-08 16:32:33 -0700 | [diff] [blame] | 567 | * PageLRU is set. lru_lock normally excludes isolation |
| 568 | * splitting and collapsing (collapsing has already happened |
| 569 | * if PageLRU is set) but the lock is not necessarily taken |
| 570 | * here and it is wasteful to take it just to check transhuge. |
| 571 | * Check TransHuge without lock and skip the whole pageblock if |
| 572 | * it's either a transhuge or hugetlbfs page, as calling |
| 573 | * compound_order() without preventing THP from splitting the |
| 574 | * page underneath us may return surprising results. |
Andrea Arcangeli | bc83501 | 2011-01-13 15:47:08 -0800 | [diff] [blame] | 575 | */ |
| 576 | if (PageTransHuge(page)) { |
Mel Gorman | 2a1402a | 2012-10-08 16:32:33 -0700 | [diff] [blame] | 577 | if (!locked) |
| 578 | goto next_pageblock; |
| 579 | low_pfn += (1 << compound_order(page)) - 1; |
| 580 | continue; |
| 581 | } |
| 582 | |
David Rientjes | 119d6d5 | 2014-04-03 14:48:00 -0700 | [diff] [blame] | 583 | /* |
| 584 | * Migration will fail if an anonymous page is pinned in memory, |
| 585 | * so avoid taking lru_lock and isolating it unnecessarily in an |
| 586 | * admittedly racy check. |
| 587 | */ |
| 588 | if (!page_mapping(page) && |
| 589 | page_count(page) > page_mapcount(page)) |
| 590 | continue; |
| 591 | |
Mel Gorman | 2a1402a | 2012-10-08 16:32:33 -0700 | [diff] [blame] | 592 | /* Check if it is ok to still hold the lock */ |
| 593 | locked = compact_checklock_irqsave(&zone->lru_lock, &flags, |
| 594 | locked, cc); |
| 595 | if (!locked || fatal_signal_pending(current)) |
| 596 | break; |
| 597 | |
| 598 | /* Recheck PageLRU and PageTransHuge under lock */ |
| 599 | if (!PageLRU(page)) |
| 600 | continue; |
| 601 | if (PageTransHuge(page)) { |
Andrea Arcangeli | bc83501 | 2011-01-13 15:47:08 -0800 | [diff] [blame] | 602 | low_pfn += (1 << compound_order(page)) - 1; |
| 603 | continue; |
| 604 | } |
| 605 | |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 606 | lruvec = mem_cgroup_page_lruvec(page, zone); |
| 607 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 608 | /* Try isolate the page */ |
Konstantin Khlebnikov | f3fd4a6 | 2012-05-29 15:06:54 -0700 | [diff] [blame] | 609 | if (__isolate_lru_page(page, mode) != 0) |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 610 | continue; |
| 611 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 612 | VM_BUG_ON_PAGE(PageTransCompound(page), page); |
Andrea Arcangeli | bc83501 | 2011-01-13 15:47:08 -0800 | [diff] [blame] | 613 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 614 | /* Successfully isolated */ |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 615 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
Joonsoo Kim | b6c7501 | 2014-04-07 15:37:07 -0700 | [diff] [blame] | 616 | |
| 617 | isolate_success: |
| 618 | cc->finished_update_migrate = true; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 619 | list_add(&page->lru, migratelist); |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 620 | cc->nr_migratepages++; |
Mel Gorman | b7aba69 | 2011-01-13 15:45:54 -0800 | [diff] [blame] | 621 | nr_isolated++; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 622 | |
| 623 | /* Avoid isolating too much */ |
Hillf Danton | 31b8384 | 2012-01-10 15:07:59 -0800 | [diff] [blame] | 624 | if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) { |
| 625 | ++low_pfn; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 626 | break; |
Hillf Danton | 31b8384 | 2012-01-10 15:07:59 -0800 | [diff] [blame] | 627 | } |
Mel Gorman | 2a1402a | 2012-10-08 16:32:33 -0700 | [diff] [blame] | 628 | |
| 629 | continue; |
| 630 | |
| 631 | next_pageblock: |
Mel Gorman | a9aacbc | 2013-02-22 16:32:25 -0800 | [diff] [blame] | 632 | low_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages) - 1; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 633 | } |
| 634 | |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 635 | acct_isolated(zone, locked, cc); |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 636 | |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 637 | if (locked) |
| 638 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 639 | |
Vlastimil Babka | 50b5b09 | 2014-01-21 15:51:10 -0800 | [diff] [blame] | 640 | /* |
| 641 | * Update the pageblock-skip information and cached scanner pfn, |
| 642 | * if the whole pageblock was scanned without isolating any page. |
| 643 | * This is not done when pageblock was skipped due to being unsuitable |
| 644 | * for async compaction, so that eventual sync compaction can try. |
| 645 | */ |
| 646 | if (low_pfn == end_pfn && !skipped_async_unsuitable) |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 647 | update_pageblock_skip(cc, valid_page, nr_isolated, true); |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 648 | |
Mel Gorman | b7aba69 | 2011-01-13 15:45:54 -0800 | [diff] [blame] | 649 | trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated); |
| 650 | |
Minchan Kim | 010fc29 | 2012-12-20 15:05:06 -0800 | [diff] [blame] | 651 | count_compact_events(COMPACTMIGRATE_SCANNED, nr_scanned); |
Mel Gorman | 397487d | 2012-10-19 12:00:10 +0100 | [diff] [blame] | 652 | if (nr_isolated) |
Minchan Kim | 010fc29 | 2012-12-20 15:05:06 -0800 | [diff] [blame] | 653 | count_compact_events(COMPACTISOLATED, nr_isolated); |
Mel Gorman | 397487d | 2012-10-19 12:00:10 +0100 | [diff] [blame] | 654 | |
Michal Nazarewicz | 2fe86e0 | 2012-01-30 13:16:26 +0100 | [diff] [blame] | 655 | return low_pfn; |
| 656 | } |
| 657 | |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 658 | #endif /* CONFIG_COMPACTION || CONFIG_CMA */ |
| 659 | #ifdef CONFIG_COMPACTION |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 660 | /* |
| 661 | * Based on information in the current compact_control, find blocks |
| 662 | * suitable for isolating free pages from and then isolate them. |
| 663 | */ |
| 664 | static void isolate_freepages(struct zone *zone, |
| 665 | struct compact_control *cc) |
| 666 | { |
| 667 | struct page *page; |
Vlastimil Babka | c96b9e5 | 2014-06-04 16:07:26 -0700 | [diff] [blame^] | 668 | unsigned long block_start_pfn; /* start of current pageblock */ |
| 669 | unsigned long block_end_pfn; /* end of current pageblock */ |
| 670 | unsigned long low_pfn; /* lowest pfn scanner is able to scan */ |
| 671 | unsigned long next_free_pfn; /* start pfn for scaning at next round */ |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 672 | int nr_freepages = cc->nr_freepages; |
| 673 | struct list_head *freelist = &cc->freepages; |
| 674 | |
| 675 | /* |
| 676 | * Initialise the free scanner. The starting point is where we last |
Vlastimil Babka | 49e068f | 2014-05-06 12:50:03 -0700 | [diff] [blame] | 677 | * successfully isolated from, zone-cached value, or the end of the |
| 678 | * zone when isolating for the first time. We need this aligned to |
Vlastimil Babka | c96b9e5 | 2014-06-04 16:07:26 -0700 | [diff] [blame^] | 679 | * the pageblock boundary, because we do |
| 680 | * block_start_pfn -= pageblock_nr_pages in the for loop. |
| 681 | * For ending point, take care when isolating in last pageblock of a |
| 682 | * a zone which ends in the middle of a pageblock. |
Vlastimil Babka | 49e068f | 2014-05-06 12:50:03 -0700 | [diff] [blame] | 683 | * The low boundary is the end of the pageblock the migration scanner |
| 684 | * is using. |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 685 | */ |
Vlastimil Babka | c96b9e5 | 2014-06-04 16:07:26 -0700 | [diff] [blame^] | 686 | block_start_pfn = cc->free_pfn & ~(pageblock_nr_pages-1); |
| 687 | block_end_pfn = min(block_start_pfn + pageblock_nr_pages, |
| 688 | zone_end_pfn(zone)); |
Vlastimil Babka | 7ed695e | 2014-01-21 15:51:09 -0800 | [diff] [blame] | 689 | low_pfn = ALIGN(cc->migrate_pfn + 1, pageblock_nr_pages); |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 690 | |
| 691 | /* |
Vlastimil Babka | c96b9e5 | 2014-06-04 16:07:26 -0700 | [diff] [blame^] | 692 | * If no pages are isolated, the block_start_pfn < low_pfn check |
| 693 | * will kick in. |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 694 | */ |
Vlastimil Babka | c96b9e5 | 2014-06-04 16:07:26 -0700 | [diff] [blame^] | 695 | next_free_pfn = 0; |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 696 | |
| 697 | /* |
| 698 | * Isolate free pages until enough are available to migrate the |
| 699 | * pages on cc->migratepages. We stop searching if the migrate |
| 700 | * and free page scanners meet or enough free pages are isolated. |
| 701 | */ |
Vlastimil Babka | c96b9e5 | 2014-06-04 16:07:26 -0700 | [diff] [blame^] | 702 | for (; block_start_pfn >= low_pfn && cc->nr_migratepages > nr_freepages; |
| 703 | block_end_pfn = block_start_pfn, |
| 704 | block_start_pfn -= pageblock_nr_pages) { |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 705 | unsigned long isolated; |
| 706 | |
David Rientjes | f6ea3ad | 2013-09-30 13:45:03 -0700 | [diff] [blame] | 707 | /* |
| 708 | * This can iterate a massively long zone without finding any |
| 709 | * suitable migration targets, so periodically check if we need |
| 710 | * to schedule. |
| 711 | */ |
| 712 | cond_resched(); |
| 713 | |
Vlastimil Babka | c96b9e5 | 2014-06-04 16:07:26 -0700 | [diff] [blame^] | 714 | if (!pfn_valid(block_start_pfn)) |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 715 | continue; |
| 716 | |
| 717 | /* |
| 718 | * Check for overlapping nodes/zones. It's possible on some |
| 719 | * configurations to have a setup like |
| 720 | * node0 node1 node0 |
| 721 | * i.e. it's possible that all pages within a zones range of |
| 722 | * pages do not belong to a single zone. |
| 723 | */ |
Vlastimil Babka | c96b9e5 | 2014-06-04 16:07:26 -0700 | [diff] [blame^] | 724 | page = pfn_to_page(block_start_pfn); |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 725 | if (page_zone(page) != zone) |
| 726 | continue; |
| 727 | |
| 728 | /* Check the block is suitable for migration */ |
Linus Torvalds | 68e3e92 | 2012-06-03 20:05:57 -0700 | [diff] [blame] | 729 | if (!suitable_migration_target(page)) |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 730 | continue; |
Linus Torvalds | 68e3e92 | 2012-06-03 20:05:57 -0700 | [diff] [blame] | 731 | |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 732 | /* If isolation recently failed, do not retry */ |
| 733 | if (!isolation_suitable(cc, page)) |
| 734 | continue; |
| 735 | |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 736 | /* Found a block suitable for isolating free pages from */ |
Vlastimil Babka | c96b9e5 | 2014-06-04 16:07:26 -0700 | [diff] [blame^] | 737 | isolated = isolate_freepages_block(cc, block_start_pfn, |
| 738 | block_end_pfn, freelist, false); |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 739 | nr_freepages += isolated; |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 740 | |
| 741 | /* |
| 742 | * Record the highest PFN we isolated pages from. When next |
| 743 | * looking for free pages, the search will restart here as |
| 744 | * page migration may have returned some pages to the allocator |
| 745 | */ |
Vlastimil Babka | c96b9e5 | 2014-06-04 16:07:26 -0700 | [diff] [blame^] | 746 | if (isolated && next_free_pfn == 0) { |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 747 | cc->finished_update_free = true; |
Vlastimil Babka | c96b9e5 | 2014-06-04 16:07:26 -0700 | [diff] [blame^] | 748 | next_free_pfn = block_start_pfn; |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 749 | } |
Michal Nazarewicz | 2fe86e0 | 2012-01-30 13:16:26 +0100 | [diff] [blame] | 750 | } |
| 751 | |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 752 | /* split_free_page does not map the pages */ |
| 753 | map_pages(freelist); |
Michal Nazarewicz | 2fe86e0 | 2012-01-30 13:16:26 +0100 | [diff] [blame] | 754 | |
Vlastimil Babka | 7ed695e | 2014-01-21 15:51:09 -0800 | [diff] [blame] | 755 | /* |
| 756 | * If we crossed the migrate scanner, we want to keep it that way |
| 757 | * so that compact_finished() may detect this |
| 758 | */ |
Vlastimil Babka | c96b9e5 | 2014-06-04 16:07:26 -0700 | [diff] [blame^] | 759 | if (block_start_pfn < low_pfn) |
| 760 | next_free_pfn = cc->migrate_pfn; |
| 761 | |
| 762 | cc->free_pfn = next_free_pfn; |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 763 | cc->nr_freepages = nr_freepages; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 764 | } |
| 765 | |
| 766 | /* |
| 767 | * This is a migrate-callback that "allocates" freepages by taking pages |
| 768 | * from the isolated freelists in the block we are migrating to. |
| 769 | */ |
| 770 | static struct page *compaction_alloc(struct page *migratepage, |
| 771 | unsigned long data, |
| 772 | int **result) |
| 773 | { |
| 774 | struct compact_control *cc = (struct compact_control *)data; |
| 775 | struct page *freepage; |
| 776 | |
| 777 | /* Isolate free pages if necessary */ |
| 778 | if (list_empty(&cc->freepages)) { |
| 779 | isolate_freepages(cc->zone, cc); |
| 780 | |
| 781 | if (list_empty(&cc->freepages)) |
| 782 | return NULL; |
| 783 | } |
| 784 | |
| 785 | freepage = list_entry(cc->freepages.next, struct page, lru); |
| 786 | list_del(&freepage->lru); |
| 787 | cc->nr_freepages--; |
| 788 | |
| 789 | return freepage; |
| 790 | } |
| 791 | |
| 792 | /* |
| 793 | * We cannot control nr_migratepages and nr_freepages fully when migration is |
| 794 | * running as migrate_pages() has no knowledge of compact_control. When |
| 795 | * migration is complete, we count the number of pages on the lists by hand. |
| 796 | */ |
| 797 | static void update_nr_listpages(struct compact_control *cc) |
| 798 | { |
| 799 | int nr_migratepages = 0; |
| 800 | int nr_freepages = 0; |
| 801 | struct page *page; |
| 802 | |
| 803 | list_for_each_entry(page, &cc->migratepages, lru) |
| 804 | nr_migratepages++; |
| 805 | list_for_each_entry(page, &cc->freepages, lru) |
| 806 | nr_freepages++; |
| 807 | |
| 808 | cc->nr_migratepages = nr_migratepages; |
| 809 | cc->nr_freepages = nr_freepages; |
| 810 | } |
| 811 | |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 812 | /* possible outcome of isolate_migratepages */ |
| 813 | typedef enum { |
| 814 | ISOLATE_ABORT, /* Abort compaction now */ |
| 815 | ISOLATE_NONE, /* No pages isolated, continue scanning */ |
| 816 | ISOLATE_SUCCESS, /* Pages isolated, migrate */ |
| 817 | } isolate_migrate_t; |
| 818 | |
| 819 | /* |
| 820 | * Isolate all pages that can be migrated from the block pointed to by |
| 821 | * the migrate scanner within compact_control. |
| 822 | */ |
| 823 | static isolate_migrate_t isolate_migratepages(struct zone *zone, |
| 824 | struct compact_control *cc) |
| 825 | { |
| 826 | unsigned long low_pfn, end_pfn; |
| 827 | |
| 828 | /* Do not scan outside zone boundaries */ |
| 829 | low_pfn = max(cc->migrate_pfn, zone->zone_start_pfn); |
| 830 | |
| 831 | /* Only scan within a pageblock boundary */ |
Mel Gorman | a9aacbc | 2013-02-22 16:32:25 -0800 | [diff] [blame] | 832 | end_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages); |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 833 | |
| 834 | /* Do not cross the free scanner or scan within a memory hole */ |
| 835 | if (end_pfn > cc->free_pfn || !pfn_valid(low_pfn)) { |
| 836 | cc->migrate_pfn = end_pfn; |
| 837 | return ISOLATE_NONE; |
| 838 | } |
| 839 | |
| 840 | /* Perform the isolation */ |
Minchan Kim | e46a287 | 2012-10-08 16:33:48 -0700 | [diff] [blame] | 841 | low_pfn = isolate_migratepages_range(zone, cc, low_pfn, end_pfn, false); |
Shaohua Li | e64c523 | 2012-10-08 16:32:27 -0700 | [diff] [blame] | 842 | if (!low_pfn || cc->contended) |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 843 | return ISOLATE_ABORT; |
| 844 | |
| 845 | cc->migrate_pfn = low_pfn; |
| 846 | |
| 847 | return ISOLATE_SUCCESS; |
| 848 | } |
| 849 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 850 | static int compact_finished(struct zone *zone, |
Andrea Arcangeli | 5a03b05 | 2011-01-13 15:47:11 -0800 | [diff] [blame] | 851 | struct compact_control *cc) |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 852 | { |
Mel Gorman | 8fb74b9 | 2013-01-11 14:32:16 -0800 | [diff] [blame] | 853 | unsigned int order; |
Andrea Arcangeli | 5a03b05 | 2011-01-13 15:47:11 -0800 | [diff] [blame] | 854 | unsigned long watermark; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 855 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 856 | if (fatal_signal_pending(current)) |
| 857 | return COMPACT_PARTIAL; |
| 858 | |
Mel Gorman | 753341a | 2012-10-08 16:32:40 -0700 | [diff] [blame] | 859 | /* Compaction run completes if the migrate and free scanner meet */ |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 860 | if (cc->free_pfn <= cc->migrate_pfn) { |
Vlastimil Babka | 55b7c4c | 2014-01-21 15:51:11 -0800 | [diff] [blame] | 861 | /* Let the next compaction start anew. */ |
| 862 | zone->compact_cached_migrate_pfn = zone->zone_start_pfn; |
| 863 | zone->compact_cached_free_pfn = zone_end_pfn(zone); |
| 864 | |
Mel Gorman | 6299702 | 2012-10-08 16:32:47 -0700 | [diff] [blame] | 865 | /* |
| 866 | * Mark that the PG_migrate_skip information should be cleared |
| 867 | * by kswapd when it goes to sleep. kswapd does not set the |
| 868 | * flag itself as the decision to be clear should be directly |
| 869 | * based on an allocation request. |
| 870 | */ |
| 871 | if (!current_is_kswapd()) |
| 872 | zone->compact_blockskip_flush = true; |
| 873 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 874 | return COMPACT_COMPLETE; |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 875 | } |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 876 | |
Johannes Weiner | 82478fb | 2011-01-20 14:44:21 -0800 | [diff] [blame] | 877 | /* |
| 878 | * order == -1 is expected when compacting via |
| 879 | * /proc/sys/vm/compact_memory |
| 880 | */ |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 881 | if (cc->order == -1) |
| 882 | return COMPACT_CONTINUE; |
| 883 | |
Michal Hocko | 3957c77 | 2011-06-15 15:08:25 -0700 | [diff] [blame] | 884 | /* Compaction run is not finished if the watermark is not met */ |
| 885 | watermark = low_wmark_pages(zone); |
| 886 | watermark += (1 << cc->order); |
| 887 | |
| 888 | if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0)) |
| 889 | return COMPACT_CONTINUE; |
| 890 | |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 891 | /* Direct compactor: Is a suitable page free? */ |
Mel Gorman | 8fb74b9 | 2013-01-11 14:32:16 -0800 | [diff] [blame] | 892 | for (order = cc->order; order < MAX_ORDER; order++) { |
| 893 | struct free_area *area = &zone->free_area[order]; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 894 | |
Mel Gorman | 8fb74b9 | 2013-01-11 14:32:16 -0800 | [diff] [blame] | 895 | /* Job done if page is free of the right migratetype */ |
| 896 | if (!list_empty(&area->free_list[cc->migratetype])) |
| 897 | return COMPACT_PARTIAL; |
| 898 | |
| 899 | /* Job done if allocation would set block type */ |
| 900 | if (cc->order >= pageblock_order && area->nr_free) |
| 901 | return COMPACT_PARTIAL; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 902 | } |
| 903 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 904 | return COMPACT_CONTINUE; |
| 905 | } |
| 906 | |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 907 | /* |
| 908 | * compaction_suitable: Is this suitable to run compaction on this zone now? |
| 909 | * Returns |
| 910 | * COMPACT_SKIPPED - If there are too few free pages for compaction |
| 911 | * COMPACT_PARTIAL - If the allocation would succeed without compaction |
| 912 | * COMPACT_CONTINUE - If compaction should run now |
| 913 | */ |
| 914 | unsigned long compaction_suitable(struct zone *zone, int order) |
| 915 | { |
| 916 | int fragindex; |
| 917 | unsigned long watermark; |
| 918 | |
| 919 | /* |
Michal Hocko | 3957c77 | 2011-06-15 15:08:25 -0700 | [diff] [blame] | 920 | * order == -1 is expected when compacting via |
| 921 | * /proc/sys/vm/compact_memory |
| 922 | */ |
| 923 | if (order == -1) |
| 924 | return COMPACT_CONTINUE; |
| 925 | |
| 926 | /* |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 927 | * Watermarks for order-0 must be met for compaction. Note the 2UL. |
| 928 | * This is because during migration, copies of pages need to be |
| 929 | * allocated and for a short time, the footprint is higher |
| 930 | */ |
| 931 | watermark = low_wmark_pages(zone) + (2UL << order); |
| 932 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) |
| 933 | return COMPACT_SKIPPED; |
| 934 | |
| 935 | /* |
| 936 | * fragmentation index determines if allocation failures are due to |
| 937 | * low memory or external fragmentation |
| 938 | * |
Shaohua Li | a582a73 | 2011-06-15 15:08:49 -0700 | [diff] [blame] | 939 | * index of -1000 implies allocations might succeed depending on |
| 940 | * watermarks |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 941 | * index towards 0 implies failure is due to lack of memory |
| 942 | * index towards 1000 implies failure is due to fragmentation |
| 943 | * |
| 944 | * Only compact if a failure would be due to fragmentation. |
| 945 | */ |
| 946 | fragindex = fragmentation_index(zone, order); |
| 947 | if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold) |
| 948 | return COMPACT_SKIPPED; |
| 949 | |
Shaohua Li | a582a73 | 2011-06-15 15:08:49 -0700 | [diff] [blame] | 950 | if (fragindex == -1000 && zone_watermark_ok(zone, order, watermark, |
| 951 | 0, 0)) |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 952 | return COMPACT_PARTIAL; |
| 953 | |
| 954 | return COMPACT_CONTINUE; |
| 955 | } |
| 956 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 957 | static int compact_zone(struct zone *zone, struct compact_control *cc) |
| 958 | { |
| 959 | int ret; |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 960 | unsigned long start_pfn = zone->zone_start_pfn; |
Cody P Schafer | 108bcc9 | 2013-02-22 16:35:23 -0800 | [diff] [blame] | 961 | unsigned long end_pfn = zone_end_pfn(zone); |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 962 | |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 963 | ret = compaction_suitable(zone, cc->order); |
| 964 | switch (ret) { |
| 965 | case COMPACT_PARTIAL: |
| 966 | case COMPACT_SKIPPED: |
| 967 | /* Compaction is likely to fail */ |
| 968 | return ret; |
| 969 | case COMPACT_CONTINUE: |
| 970 | /* Fall through to compaction */ |
| 971 | ; |
| 972 | } |
| 973 | |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 974 | /* |
Vlastimil Babka | d3132e4 | 2014-01-21 15:51:08 -0800 | [diff] [blame] | 975 | * Clear pageblock skip if there were failures recently and compaction |
| 976 | * is about to be retried after being deferred. kswapd does not do |
| 977 | * this reset as it'll reset the cached information when going to sleep. |
| 978 | */ |
| 979 | if (compaction_restarting(zone, cc->order) && !current_is_kswapd()) |
| 980 | __reset_isolation_suitable(zone); |
| 981 | |
| 982 | /* |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 983 | * Setup to move all movable pages to the end of the zone. Used cached |
| 984 | * information on where the scanners should start but check that it |
| 985 | * is initialised by ensuring the values are within zone boundaries. |
| 986 | */ |
| 987 | cc->migrate_pfn = zone->compact_cached_migrate_pfn; |
| 988 | cc->free_pfn = zone->compact_cached_free_pfn; |
| 989 | if (cc->free_pfn < start_pfn || cc->free_pfn > end_pfn) { |
| 990 | cc->free_pfn = end_pfn & ~(pageblock_nr_pages-1); |
| 991 | zone->compact_cached_free_pfn = cc->free_pfn; |
| 992 | } |
| 993 | if (cc->migrate_pfn < start_pfn || cc->migrate_pfn > end_pfn) { |
| 994 | cc->migrate_pfn = start_pfn; |
| 995 | zone->compact_cached_migrate_pfn = cc->migrate_pfn; |
| 996 | } |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 997 | |
Mel Gorman | 0eb927c | 2014-01-21 15:51:05 -0800 | [diff] [blame] | 998 | trace_mm_compaction_begin(start_pfn, cc->migrate_pfn, cc->free_pfn, end_pfn); |
| 999 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1000 | migrate_prep_local(); |
| 1001 | |
| 1002 | while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) { |
| 1003 | unsigned long nr_migrate, nr_remaining; |
Minchan Kim | 9d502c1 | 2011-03-22 16:30:39 -0700 | [diff] [blame] | 1004 | int err; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1005 | |
Mel Gorman | f9e35b3 | 2011-06-15 15:08:52 -0700 | [diff] [blame] | 1006 | switch (isolate_migratepages(zone, cc)) { |
| 1007 | case ISOLATE_ABORT: |
| 1008 | ret = COMPACT_PARTIAL; |
Rafael Aquini | 5733c7d | 2012-12-11 16:02:47 -0800 | [diff] [blame] | 1009 | putback_movable_pages(&cc->migratepages); |
Shaohua Li | e64c523 | 2012-10-08 16:32:27 -0700 | [diff] [blame] | 1010 | cc->nr_migratepages = 0; |
Mel Gorman | f9e35b3 | 2011-06-15 15:08:52 -0700 | [diff] [blame] | 1011 | goto out; |
| 1012 | case ISOLATE_NONE: |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1013 | continue; |
Mel Gorman | f9e35b3 | 2011-06-15 15:08:52 -0700 | [diff] [blame] | 1014 | case ISOLATE_SUCCESS: |
| 1015 | ; |
| 1016 | } |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1017 | |
| 1018 | nr_migrate = cc->nr_migratepages; |
Minchan Kim | 9d502c1 | 2011-03-22 16:30:39 -0700 | [diff] [blame] | 1019 | err = migrate_pages(&cc->migratepages, compaction_alloc, |
Hugh Dickins | 9c620e2 | 2013-02-22 16:35:14 -0800 | [diff] [blame] | 1020 | (unsigned long)cc, |
Mel Gorman | 7b2a2d4 | 2012-10-19 14:07:31 +0100 | [diff] [blame] | 1021 | cc->sync ? MIGRATE_SYNC_LIGHT : MIGRATE_ASYNC, |
| 1022 | MR_COMPACTION); |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1023 | update_nr_listpages(cc); |
| 1024 | nr_remaining = cc->nr_migratepages; |
| 1025 | |
Mel Gorman | b7aba69 | 2011-01-13 15:45:54 -0800 | [diff] [blame] | 1026 | trace_mm_compaction_migratepages(nr_migrate - nr_remaining, |
| 1027 | nr_remaining); |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1028 | |
Rafael Aquini | 5733c7d | 2012-12-11 16:02:47 -0800 | [diff] [blame] | 1029 | /* Release isolated pages not migrated */ |
Minchan Kim | 9d502c1 | 2011-03-22 16:30:39 -0700 | [diff] [blame] | 1030 | if (err) { |
Rafael Aquini | 5733c7d | 2012-12-11 16:02:47 -0800 | [diff] [blame] | 1031 | putback_movable_pages(&cc->migratepages); |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1032 | cc->nr_migratepages = 0; |
Vlastimil Babka | 7ed695e | 2014-01-21 15:51:09 -0800 | [diff] [blame] | 1033 | /* |
| 1034 | * migrate_pages() may return -ENOMEM when scanners meet |
| 1035 | * and we want compact_finished() to detect it |
| 1036 | */ |
| 1037 | if (err == -ENOMEM && cc->free_pfn > cc->migrate_pfn) { |
David Rientjes | 4bf2bba | 2012-07-11 14:02:13 -0700 | [diff] [blame] | 1038 | ret = COMPACT_PARTIAL; |
| 1039 | goto out; |
| 1040 | } |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1041 | } |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1042 | } |
| 1043 | |
Mel Gorman | f9e35b3 | 2011-06-15 15:08:52 -0700 | [diff] [blame] | 1044 | out: |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1045 | /* Release free pages and check accounting */ |
| 1046 | cc->nr_freepages -= release_freepages(&cc->freepages); |
| 1047 | VM_BUG_ON(cc->nr_freepages != 0); |
| 1048 | |
Mel Gorman | 0eb927c | 2014-01-21 15:51:05 -0800 | [diff] [blame] | 1049 | trace_mm_compaction_end(ret); |
| 1050 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1051 | return ret; |
| 1052 | } |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1053 | |
Kyungmin Park | d43a87e | 2011-10-31 17:09:08 -0700 | [diff] [blame] | 1054 | static unsigned long compact_zone_order(struct zone *zone, |
Andrea Arcangeli | 5a03b05 | 2011-01-13 15:47:11 -0800 | [diff] [blame] | 1055 | int order, gfp_t gfp_mask, |
Mel Gorman | 8fb74b9 | 2013-01-11 14:32:16 -0800 | [diff] [blame] | 1056 | bool sync, bool *contended) |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1057 | { |
Shaohua Li | e64c523 | 2012-10-08 16:32:27 -0700 | [diff] [blame] | 1058 | unsigned long ret; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1059 | struct compact_control cc = { |
| 1060 | .nr_freepages = 0, |
| 1061 | .nr_migratepages = 0, |
| 1062 | .order = order, |
| 1063 | .migratetype = allocflags_to_migratetype(gfp_mask), |
| 1064 | .zone = zone, |
Linus Torvalds | 68e3e92 | 2012-06-03 20:05:57 -0700 | [diff] [blame] | 1065 | .sync = sync, |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1066 | }; |
| 1067 | INIT_LIST_HEAD(&cc.freepages); |
| 1068 | INIT_LIST_HEAD(&cc.migratepages); |
| 1069 | |
Shaohua Li | e64c523 | 2012-10-08 16:32:27 -0700 | [diff] [blame] | 1070 | ret = compact_zone(zone, &cc); |
| 1071 | |
| 1072 | VM_BUG_ON(!list_empty(&cc.freepages)); |
| 1073 | VM_BUG_ON(!list_empty(&cc.migratepages)); |
| 1074 | |
| 1075 | *contended = cc.contended; |
| 1076 | return ret; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1077 | } |
| 1078 | |
Mel Gorman | 5e77190 | 2010-05-24 14:32:31 -0700 | [diff] [blame] | 1079 | int sysctl_extfrag_threshold = 500; |
| 1080 | |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1081 | /** |
| 1082 | * try_to_compact_pages - Direct compact to satisfy a high-order allocation |
| 1083 | * @zonelist: The zonelist used for the current allocation |
| 1084 | * @order: The order of the current allocation |
| 1085 | * @gfp_mask: The GFP mask of the current allocation |
| 1086 | * @nodemask: The allowed nodes to allocate from |
Mel Gorman | 77f1fe6 | 2011-01-13 15:45:57 -0800 | [diff] [blame] | 1087 | * @sync: Whether migration is synchronous or not |
Mel Gorman | 661c4cb | 2012-10-08 16:32:31 -0700 | [diff] [blame] | 1088 | * @contended: Return value that is true if compaction was aborted due to lock contention |
| 1089 | * @page: Optionally capture a free page of the requested order during compaction |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1090 | * |
| 1091 | * This is the main entry point for direct page compaction. |
| 1092 | */ |
| 1093 | unsigned long try_to_compact_pages(struct zonelist *zonelist, |
Mel Gorman | 77f1fe6 | 2011-01-13 15:45:57 -0800 | [diff] [blame] | 1094 | int order, gfp_t gfp_mask, nodemask_t *nodemask, |
Mel Gorman | 8fb74b9 | 2013-01-11 14:32:16 -0800 | [diff] [blame] | 1095 | bool sync, bool *contended) |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1096 | { |
| 1097 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); |
| 1098 | int may_enter_fs = gfp_mask & __GFP_FS; |
| 1099 | int may_perform_io = gfp_mask & __GFP_IO; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1100 | struct zoneref *z; |
| 1101 | struct zone *zone; |
| 1102 | int rc = COMPACT_SKIPPED; |
Bartlomiej Zolnierkiewicz | d95ea5d | 2012-10-08 16:32:05 -0700 | [diff] [blame] | 1103 | int alloc_flags = 0; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1104 | |
Mel Gorman | 4ffb633 | 2012-10-08 16:29:09 -0700 | [diff] [blame] | 1105 | /* Check if the GFP flags allow compaction */ |
Andrea Arcangeli | c5a73c3 | 2011-01-13 15:47:11 -0800 | [diff] [blame] | 1106 | if (!order || !may_enter_fs || !may_perform_io) |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1107 | return rc; |
| 1108 | |
Minchan Kim | 010fc29 | 2012-12-20 15:05:06 -0800 | [diff] [blame] | 1109 | count_compact_event(COMPACTSTALL); |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1110 | |
Bartlomiej Zolnierkiewicz | d95ea5d | 2012-10-08 16:32:05 -0700 | [diff] [blame] | 1111 | #ifdef CONFIG_CMA |
| 1112 | if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) |
| 1113 | alloc_flags |= ALLOC_CMA; |
| 1114 | #endif |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1115 | /* Compact each zone in the list */ |
| 1116 | for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx, |
| 1117 | nodemask) { |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1118 | int status; |
| 1119 | |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 1120 | status = compact_zone_order(zone, order, gfp_mask, sync, |
Mel Gorman | 8fb74b9 | 2013-01-11 14:32:16 -0800 | [diff] [blame] | 1121 | contended); |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1122 | rc = max(status, rc); |
| 1123 | |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1124 | /* If a normal allocation would succeed, stop compacting */ |
Bartlomiej Zolnierkiewicz | d95ea5d | 2012-10-08 16:32:05 -0700 | [diff] [blame] | 1125 | if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0, |
| 1126 | alloc_flags)) |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1127 | break; |
| 1128 | } |
| 1129 | |
| 1130 | return rc; |
| 1131 | } |
| 1132 | |
| 1133 | |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1134 | /* Compact all zones within a node */ |
Andrew Morton | 7103f16 | 2013-02-22 16:32:33 -0800 | [diff] [blame] | 1135 | static void __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc) |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1136 | { |
| 1137 | int zoneid; |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1138 | struct zone *zone; |
| 1139 | |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1140 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1141 | |
| 1142 | zone = &pgdat->node_zones[zoneid]; |
| 1143 | if (!populated_zone(zone)) |
| 1144 | continue; |
| 1145 | |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 1146 | cc->nr_freepages = 0; |
| 1147 | cc->nr_migratepages = 0; |
| 1148 | cc->zone = zone; |
| 1149 | INIT_LIST_HEAD(&cc->freepages); |
| 1150 | INIT_LIST_HEAD(&cc->migratepages); |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1151 | |
Dan Carpenter | aad6ec3 | 2012-03-21 16:33:54 -0700 | [diff] [blame] | 1152 | if (cc->order == -1 || !compaction_deferred(zone, cc->order)) |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 1153 | compact_zone(zone, cc); |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1154 | |
Rik van Riel | aff6224 | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 1155 | if (cc->order > 0) { |
Vlastimil Babka | de6c60a | 2014-01-21 15:51:07 -0800 | [diff] [blame] | 1156 | if (zone_watermark_ok(zone, cc->order, |
| 1157 | low_wmark_pages(zone), 0, 0)) |
| 1158 | compaction_defer_reset(zone, cc->order, false); |
Rik van Riel | aff6224 | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 1159 | } |
| 1160 | |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 1161 | VM_BUG_ON(!list_empty(&cc->freepages)); |
| 1162 | VM_BUG_ON(!list_empty(&cc->migratepages)); |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1163 | } |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1164 | } |
| 1165 | |
Andrew Morton | 7103f16 | 2013-02-22 16:32:33 -0800 | [diff] [blame] | 1166 | void compact_pgdat(pg_data_t *pgdat, int order) |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 1167 | { |
| 1168 | struct compact_control cc = { |
| 1169 | .order = order, |
Linus Torvalds | 68e3e92 | 2012-06-03 20:05:57 -0700 | [diff] [blame] | 1170 | .sync = false, |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 1171 | }; |
| 1172 | |
Mel Gorman | 3a7200a | 2013-09-11 14:22:19 -0700 | [diff] [blame] | 1173 | if (!order) |
| 1174 | return; |
| 1175 | |
Andrew Morton | 7103f16 | 2013-02-22 16:32:33 -0800 | [diff] [blame] | 1176 | __compact_pgdat(pgdat, &cc); |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 1177 | } |
| 1178 | |
Andrew Morton | 7103f16 | 2013-02-22 16:32:33 -0800 | [diff] [blame] | 1179 | static void compact_node(int nid) |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 1180 | { |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 1181 | struct compact_control cc = { |
| 1182 | .order = -1, |
Linus Torvalds | 68e3e92 | 2012-06-03 20:05:57 -0700 | [diff] [blame] | 1183 | .sync = true, |
David Rientjes | 91ca918 | 2014-04-03 14:47:23 -0700 | [diff] [blame] | 1184 | .ignore_skip_hint = true, |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 1185 | }; |
| 1186 | |
Andrew Morton | 7103f16 | 2013-02-22 16:32:33 -0800 | [diff] [blame] | 1187 | __compact_pgdat(NODE_DATA(nid), &cc); |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 1188 | } |
| 1189 | |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1190 | /* Compact all nodes in the system */ |
Jason Liu | 7964c06 | 2013-01-11 14:31:47 -0800 | [diff] [blame] | 1191 | static void compact_nodes(void) |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1192 | { |
| 1193 | int nid; |
| 1194 | |
Hugh Dickins | 8575ec2 | 2012-03-21 16:33:53 -0700 | [diff] [blame] | 1195 | /* Flush pending updates to the LRU lists */ |
| 1196 | lru_add_drain_all(); |
| 1197 | |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1198 | for_each_online_node(nid) |
| 1199 | compact_node(nid); |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1200 | } |
| 1201 | |
| 1202 | /* The written value is actually unused, all memory is compacted */ |
| 1203 | int sysctl_compact_memory; |
| 1204 | |
| 1205 | /* This is the entry point for compacting all nodes via /proc/sys/vm */ |
| 1206 | int sysctl_compaction_handler(struct ctl_table *table, int write, |
| 1207 | void __user *buffer, size_t *length, loff_t *ppos) |
| 1208 | { |
| 1209 | if (write) |
Jason Liu | 7964c06 | 2013-01-11 14:31:47 -0800 | [diff] [blame] | 1210 | compact_nodes(); |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1211 | |
| 1212 | return 0; |
| 1213 | } |
Mel Gorman | ed4a6d7 | 2010-05-24 14:32:29 -0700 | [diff] [blame] | 1214 | |
Mel Gorman | 5e77190 | 2010-05-24 14:32:31 -0700 | [diff] [blame] | 1215 | int sysctl_extfrag_handler(struct ctl_table *table, int write, |
| 1216 | void __user *buffer, size_t *length, loff_t *ppos) |
| 1217 | { |
| 1218 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
| 1219 | |
| 1220 | return 0; |
| 1221 | } |
| 1222 | |
Mel Gorman | ed4a6d7 | 2010-05-24 14:32:29 -0700 | [diff] [blame] | 1223 | #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA) |
Rashika Kheria | 74e77fb | 2014-04-03 14:48:01 -0700 | [diff] [blame] | 1224 | static ssize_t sysfs_compact_node(struct device *dev, |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 1225 | struct device_attribute *attr, |
Mel Gorman | ed4a6d7 | 2010-05-24 14:32:29 -0700 | [diff] [blame] | 1226 | const char *buf, size_t count) |
| 1227 | { |
Hugh Dickins | 8575ec2 | 2012-03-21 16:33:53 -0700 | [diff] [blame] | 1228 | int nid = dev->id; |
| 1229 | |
| 1230 | if (nid >= 0 && nid < nr_node_ids && node_online(nid)) { |
| 1231 | /* Flush pending updates to the LRU lists */ |
| 1232 | lru_add_drain_all(); |
| 1233 | |
| 1234 | compact_node(nid); |
| 1235 | } |
Mel Gorman | ed4a6d7 | 2010-05-24 14:32:29 -0700 | [diff] [blame] | 1236 | |
| 1237 | return count; |
| 1238 | } |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 1239 | static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node); |
Mel Gorman | ed4a6d7 | 2010-05-24 14:32:29 -0700 | [diff] [blame] | 1240 | |
| 1241 | int compaction_register_node(struct node *node) |
| 1242 | { |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 1243 | return device_create_file(&node->dev, &dev_attr_compact); |
Mel Gorman | ed4a6d7 | 2010-05-24 14:32:29 -0700 | [diff] [blame] | 1244 | } |
| 1245 | |
| 1246 | void compaction_unregister_node(struct node *node) |
| 1247 | { |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 1248 | return device_remove_file(&node->dev, &dev_attr_compact); |
Mel Gorman | ed4a6d7 | 2010-05-24 14:32:29 -0700 | [diff] [blame] | 1249 | } |
| 1250 | #endif /* CONFIG_SYSFS && CONFIG_NUMA */ |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 1251 | |
| 1252 | #endif /* CONFIG_COMPACTION */ |