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> |
Andrey Ryabinin | b8c73fc | 2015-02-13 14:39:28 -0800 | [diff] [blame] | 19 | #include <linux/kasan.h> |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 20 | #include "internal.h" |
| 21 | |
Minchan Kim | 010fc29 | 2012-12-20 15:05:06 -0800 | [diff] [blame] | 22 | #ifdef CONFIG_COMPACTION |
| 23 | static inline void count_compact_event(enum vm_event_item item) |
| 24 | { |
| 25 | count_vm_event(item); |
| 26 | } |
| 27 | |
| 28 | static inline void count_compact_events(enum vm_event_item item, long delta) |
| 29 | { |
| 30 | count_vm_events(item, delta); |
| 31 | } |
| 32 | #else |
| 33 | #define count_compact_event(item) do { } while (0) |
| 34 | #define count_compact_events(item, delta) do { } while (0) |
| 35 | #endif |
| 36 | |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 37 | #if defined CONFIG_COMPACTION || defined CONFIG_CMA |
| 38 | |
Mel Gorman | b7aba69 | 2011-01-13 15:45:54 -0800 | [diff] [blame] | 39 | #define CREATE_TRACE_POINTS |
| 40 | #include <trace/events/compaction.h> |
| 41 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 42 | static unsigned long release_freepages(struct list_head *freelist) |
| 43 | { |
| 44 | struct page *page, *next; |
Vlastimil Babka | 6bace09 | 2014-12-10 15:43:31 -0800 | [diff] [blame] | 45 | unsigned long high_pfn = 0; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 46 | |
| 47 | list_for_each_entry_safe(page, next, freelist, lru) { |
Vlastimil Babka | 6bace09 | 2014-12-10 15:43:31 -0800 | [diff] [blame] | 48 | unsigned long pfn = page_to_pfn(page); |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 49 | list_del(&page->lru); |
| 50 | __free_page(page); |
Vlastimil Babka | 6bace09 | 2014-12-10 15:43:31 -0800 | [diff] [blame] | 51 | if (pfn > high_pfn) |
| 52 | high_pfn = pfn; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 53 | } |
| 54 | |
Vlastimil Babka | 6bace09 | 2014-12-10 15:43:31 -0800 | [diff] [blame] | 55 | return high_pfn; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 56 | } |
| 57 | |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 58 | static void map_pages(struct list_head *list) |
| 59 | { |
| 60 | struct page *page; |
| 61 | |
| 62 | list_for_each_entry(page, list, lru) { |
| 63 | arch_alloc_page(page, 0); |
| 64 | kernel_map_pages(page, 1, 1); |
Andrey Ryabinin | b8c73fc | 2015-02-13 14:39:28 -0800 | [diff] [blame] | 65 | kasan_alloc_pages(page, 0); |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 66 | } |
| 67 | } |
| 68 | |
Michal Nazarewicz | 47118af | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 69 | static inline bool migrate_async_suitable(int migratetype) |
| 70 | { |
| 71 | return is_migrate_cma(migratetype) || migratetype == MIGRATE_MOVABLE; |
| 72 | } |
| 73 | |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 74 | #ifdef CONFIG_COMPACTION |
Joonsoo Kim | 24e2716 | 2015-02-11 15:27:09 -0800 | [diff] [blame] | 75 | |
| 76 | /* Do not skip compaction more than 64 times */ |
| 77 | #define COMPACT_MAX_DEFER_SHIFT 6 |
| 78 | |
| 79 | /* |
| 80 | * Compaction is deferred when compaction fails to result in a page |
| 81 | * allocation success. 1 << compact_defer_limit compactions are skipped up |
| 82 | * to a limit of 1 << COMPACT_MAX_DEFER_SHIFT |
| 83 | */ |
| 84 | void defer_compaction(struct zone *zone, int order) |
| 85 | { |
| 86 | zone->compact_considered = 0; |
| 87 | zone->compact_defer_shift++; |
| 88 | |
| 89 | if (order < zone->compact_order_failed) |
| 90 | zone->compact_order_failed = order; |
| 91 | |
| 92 | if (zone->compact_defer_shift > COMPACT_MAX_DEFER_SHIFT) |
| 93 | zone->compact_defer_shift = COMPACT_MAX_DEFER_SHIFT; |
| 94 | |
| 95 | trace_mm_compaction_defer_compaction(zone, order); |
| 96 | } |
| 97 | |
| 98 | /* Returns true if compaction should be skipped this time */ |
| 99 | bool compaction_deferred(struct zone *zone, int order) |
| 100 | { |
| 101 | unsigned long defer_limit = 1UL << zone->compact_defer_shift; |
| 102 | |
| 103 | if (order < zone->compact_order_failed) |
| 104 | return false; |
| 105 | |
| 106 | /* Avoid possible overflow */ |
| 107 | if (++zone->compact_considered > defer_limit) |
| 108 | zone->compact_considered = defer_limit; |
| 109 | |
| 110 | if (zone->compact_considered >= defer_limit) |
| 111 | return false; |
| 112 | |
| 113 | trace_mm_compaction_deferred(zone, order); |
| 114 | |
| 115 | return true; |
| 116 | } |
| 117 | |
| 118 | /* |
| 119 | * Update defer tracking counters after successful compaction of given order, |
| 120 | * which means an allocation either succeeded (alloc_success == true) or is |
| 121 | * expected to succeed. |
| 122 | */ |
| 123 | void compaction_defer_reset(struct zone *zone, int order, |
| 124 | bool alloc_success) |
| 125 | { |
| 126 | if (alloc_success) { |
| 127 | zone->compact_considered = 0; |
| 128 | zone->compact_defer_shift = 0; |
| 129 | } |
| 130 | if (order >= zone->compact_order_failed) |
| 131 | zone->compact_order_failed = order + 1; |
| 132 | |
| 133 | trace_mm_compaction_defer_reset(zone, order); |
| 134 | } |
| 135 | |
| 136 | /* Returns true if restarting compaction after many failures */ |
| 137 | bool compaction_restarting(struct zone *zone, int order) |
| 138 | { |
| 139 | if (order < zone->compact_order_failed) |
| 140 | return false; |
| 141 | |
| 142 | return zone->compact_defer_shift == COMPACT_MAX_DEFER_SHIFT && |
| 143 | zone->compact_considered >= 1UL << zone->compact_defer_shift; |
| 144 | } |
| 145 | |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 146 | /* Returns true if the pageblock should be scanned for pages to isolate. */ |
| 147 | static inline bool isolation_suitable(struct compact_control *cc, |
| 148 | struct page *page) |
| 149 | { |
| 150 | if (cc->ignore_skip_hint) |
| 151 | return true; |
| 152 | |
| 153 | return !get_pageblock_skip(page); |
| 154 | } |
| 155 | |
Vlastimil Babka | 0233364 | 2015-09-08 15:02:42 -0700 | [diff] [blame] | 156 | static void reset_cached_positions(struct zone *zone) |
| 157 | { |
| 158 | zone->compact_cached_migrate_pfn[0] = zone->zone_start_pfn; |
| 159 | zone->compact_cached_migrate_pfn[1] = zone->zone_start_pfn; |
Joonsoo Kim | 623446e | 2016-03-15 14:57:45 -0700 | [diff] [blame] | 160 | zone->compact_cached_free_pfn = |
| 161 | round_down(zone_end_pfn(zone) - 1, pageblock_nr_pages); |
Vlastimil Babka | 0233364 | 2015-09-08 15:02:42 -0700 | [diff] [blame] | 162 | } |
| 163 | |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 164 | /* |
| 165 | * This function is called to clear all cached information on pageblocks that |
| 166 | * should be skipped for page isolation when the migrate and free page scanner |
| 167 | * meet. |
| 168 | */ |
Mel Gorman | 6299702 | 2012-10-08 16:32:47 -0700 | [diff] [blame] | 169 | static void __reset_isolation_suitable(struct zone *zone) |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 170 | { |
| 171 | unsigned long start_pfn = zone->zone_start_pfn; |
Cody P Schafer | 108bcc9 | 2013-02-22 16:35:23 -0800 | [diff] [blame] | 172 | unsigned long end_pfn = zone_end_pfn(zone); |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 173 | unsigned long pfn; |
| 174 | |
Mel Gorman | 6299702 | 2012-10-08 16:32:47 -0700 | [diff] [blame] | 175 | zone->compact_blockskip_flush = false; |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 176 | |
| 177 | /* Walk the zone and mark every pageblock as suitable for isolation */ |
| 178 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { |
| 179 | struct page *page; |
| 180 | |
| 181 | cond_resched(); |
| 182 | |
| 183 | if (!pfn_valid(pfn)) |
| 184 | continue; |
| 185 | |
| 186 | page = pfn_to_page(pfn); |
| 187 | if (zone != page_zone(page)) |
| 188 | continue; |
| 189 | |
| 190 | clear_pageblock_skip(page); |
| 191 | } |
Vlastimil Babka | 0233364 | 2015-09-08 15:02:42 -0700 | [diff] [blame] | 192 | |
| 193 | reset_cached_positions(zone); |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 194 | } |
| 195 | |
Mel Gorman | 6299702 | 2012-10-08 16:32:47 -0700 | [diff] [blame] | 196 | void reset_isolation_suitable(pg_data_t *pgdat) |
| 197 | { |
| 198 | int zoneid; |
| 199 | |
| 200 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { |
| 201 | struct zone *zone = &pgdat->node_zones[zoneid]; |
| 202 | if (!populated_zone(zone)) |
| 203 | continue; |
| 204 | |
| 205 | /* Only flush if a full compaction finished recently */ |
| 206 | if (zone->compact_blockskip_flush) |
| 207 | __reset_isolation_suitable(zone); |
| 208 | } |
| 209 | } |
| 210 | |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 211 | /* |
| 212 | * 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] | 213 | * future. The information is later cleared by __reset_isolation_suitable(). |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 214 | */ |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 215 | static void update_pageblock_skip(struct compact_control *cc, |
| 216 | struct page *page, unsigned long nr_isolated, |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 217 | bool migrate_scanner) |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 218 | { |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 219 | struct zone *zone = cc->zone; |
David Rientjes | 35979ef | 2014-06-04 16:08:27 -0700 | [diff] [blame] | 220 | unsigned long pfn; |
Joonsoo Kim | 6815bf3 | 2013-12-18 17:08:52 -0800 | [diff] [blame] | 221 | |
| 222 | if (cc->ignore_skip_hint) |
| 223 | return; |
| 224 | |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 225 | if (!page) |
| 226 | return; |
| 227 | |
David Rientjes | 35979ef | 2014-06-04 16:08:27 -0700 | [diff] [blame] | 228 | if (nr_isolated) |
| 229 | return; |
| 230 | |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 231 | set_pageblock_skip(page); |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 232 | |
David Rientjes | 35979ef | 2014-06-04 16:08:27 -0700 | [diff] [blame] | 233 | pfn = page_to_pfn(page); |
| 234 | |
| 235 | /* Update where async and sync compaction should restart */ |
| 236 | if (migrate_scanner) { |
David Rientjes | 35979ef | 2014-06-04 16:08:27 -0700 | [diff] [blame] | 237 | if (pfn > zone->compact_cached_migrate_pfn[0]) |
| 238 | zone->compact_cached_migrate_pfn[0] = pfn; |
David Rientjes | e0b9dae | 2014-06-04 16:08:28 -0700 | [diff] [blame] | 239 | if (cc->mode != MIGRATE_ASYNC && |
| 240 | pfn > zone->compact_cached_migrate_pfn[1]) |
David Rientjes | 35979ef | 2014-06-04 16:08:27 -0700 | [diff] [blame] | 241 | zone->compact_cached_migrate_pfn[1] = pfn; |
| 242 | } else { |
David Rientjes | 35979ef | 2014-06-04 16:08:27 -0700 | [diff] [blame] | 243 | if (pfn < zone->compact_cached_free_pfn) |
| 244 | zone->compact_cached_free_pfn = pfn; |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 245 | } |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 246 | } |
| 247 | #else |
| 248 | static inline bool isolation_suitable(struct compact_control *cc, |
| 249 | struct page *page) |
| 250 | { |
| 251 | return true; |
| 252 | } |
| 253 | |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 254 | static void update_pageblock_skip(struct compact_control *cc, |
| 255 | struct page *page, unsigned long nr_isolated, |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 256 | bool migrate_scanner) |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 257 | { |
| 258 | } |
| 259 | #endif /* CONFIG_COMPACTION */ |
| 260 | |
Vlastimil Babka | 8b44d27 | 2014-10-09 15:27:16 -0700 | [diff] [blame] | 261 | /* |
| 262 | * Compaction requires the taking of some coarse locks that are potentially |
| 263 | * very heavily contended. For async compaction, back out if the lock cannot |
| 264 | * be taken immediately. For sync compaction, spin on the lock if needed. |
| 265 | * |
| 266 | * Returns true if the lock is held |
| 267 | * Returns false if the lock is not held and compaction should abort |
| 268 | */ |
| 269 | static bool compact_trylock_irqsave(spinlock_t *lock, unsigned long *flags, |
| 270 | struct compact_control *cc) |
Mel Gorman | 2a1402a | 2012-10-08 16:32:33 -0700 | [diff] [blame] | 271 | { |
Vlastimil Babka | 8b44d27 | 2014-10-09 15:27:16 -0700 | [diff] [blame] | 272 | if (cc->mode == MIGRATE_ASYNC) { |
| 273 | if (!spin_trylock_irqsave(lock, *flags)) { |
| 274 | cc->contended = COMPACT_CONTENDED_LOCK; |
| 275 | return false; |
| 276 | } |
| 277 | } else { |
| 278 | spin_lock_irqsave(lock, *flags); |
| 279 | } |
Vlastimil Babka | 1f9efde | 2014-10-09 15:27:14 -0700 | [diff] [blame] | 280 | |
Vlastimil Babka | 8b44d27 | 2014-10-09 15:27:16 -0700 | [diff] [blame] | 281 | return true; |
Mel Gorman | 2a1402a | 2012-10-08 16:32:33 -0700 | [diff] [blame] | 282 | } |
| 283 | |
Michal Nazarewicz | 85aa125 | 2012-01-30 13:24:03 +0100 | [diff] [blame] | 284 | /* |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 285 | * Compaction requires the taking of some coarse locks that are potentially |
Vlastimil Babka | 8b44d27 | 2014-10-09 15:27:16 -0700 | [diff] [blame] | 286 | * very heavily contended. The lock should be periodically unlocked to avoid |
| 287 | * having disabled IRQs for a long time, even when there is nobody waiting on |
| 288 | * the lock. It might also be that allowing the IRQs will result in |
| 289 | * need_resched() becoming true. If scheduling is needed, async compaction |
| 290 | * aborts. Sync compaction schedules. |
| 291 | * Either compaction type will also abort if a fatal signal is pending. |
| 292 | * In either case if the lock was locked, it is dropped and not regained. |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 293 | * |
Vlastimil Babka | 8b44d27 | 2014-10-09 15:27:16 -0700 | [diff] [blame] | 294 | * Returns true if compaction should abort due to fatal signal pending, or |
| 295 | * async compaction due to need_resched() |
| 296 | * Returns false when compaction can continue (sync compaction might have |
| 297 | * scheduled) |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 298 | */ |
Vlastimil Babka | 8b44d27 | 2014-10-09 15:27:16 -0700 | [diff] [blame] | 299 | static bool compact_unlock_should_abort(spinlock_t *lock, |
| 300 | unsigned long flags, bool *locked, struct compact_control *cc) |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 301 | { |
Vlastimil Babka | 8b44d27 | 2014-10-09 15:27:16 -0700 | [diff] [blame] | 302 | if (*locked) { |
| 303 | spin_unlock_irqrestore(lock, flags); |
| 304 | *locked = false; |
| 305 | } |
Vlastimil Babka | 1f9efde | 2014-10-09 15:27:14 -0700 | [diff] [blame] | 306 | |
Vlastimil Babka | 8b44d27 | 2014-10-09 15:27:16 -0700 | [diff] [blame] | 307 | if (fatal_signal_pending(current)) { |
| 308 | cc->contended = COMPACT_CONTENDED_SCHED; |
| 309 | return true; |
| 310 | } |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 311 | |
Vlastimil Babka | 8b44d27 | 2014-10-09 15:27:16 -0700 | [diff] [blame] | 312 | if (need_resched()) { |
David Rientjes | e0b9dae | 2014-06-04 16:08:28 -0700 | [diff] [blame] | 313 | if (cc->mode == MIGRATE_ASYNC) { |
Vlastimil Babka | 8b44d27 | 2014-10-09 15:27:16 -0700 | [diff] [blame] | 314 | cc->contended = COMPACT_CONTENDED_SCHED; |
| 315 | return true; |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 316 | } |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 317 | cond_resched(); |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 318 | } |
| 319 | |
Vlastimil Babka | 8b44d27 | 2014-10-09 15:27:16 -0700 | [diff] [blame] | 320 | return false; |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 321 | } |
| 322 | |
Vlastimil Babka | be97657 | 2014-06-04 16:10:41 -0700 | [diff] [blame] | 323 | /* |
| 324 | * Aside from avoiding lock contention, compaction also periodically checks |
| 325 | * need_resched() and either schedules in sync compaction or aborts async |
Vlastimil Babka | 8b44d27 | 2014-10-09 15:27:16 -0700 | [diff] [blame] | 326 | * compaction. This is similar to what compact_unlock_should_abort() does, but |
Vlastimil Babka | be97657 | 2014-06-04 16:10:41 -0700 | [diff] [blame] | 327 | * is used where no lock is concerned. |
| 328 | * |
| 329 | * Returns false when no scheduling was needed, or sync compaction scheduled. |
| 330 | * Returns true when async compaction should abort. |
| 331 | */ |
| 332 | static inline bool compact_should_abort(struct compact_control *cc) |
| 333 | { |
| 334 | /* async compaction aborts if contended */ |
| 335 | if (need_resched()) { |
| 336 | if (cc->mode == MIGRATE_ASYNC) { |
Vlastimil Babka | 1f9efde | 2014-10-09 15:27:14 -0700 | [diff] [blame] | 337 | cc->contended = COMPACT_CONTENDED_SCHED; |
Vlastimil Babka | be97657 | 2014-06-04 16:10:41 -0700 | [diff] [blame] | 338 | return true; |
| 339 | } |
| 340 | |
| 341 | cond_resched(); |
| 342 | } |
| 343 | |
| 344 | return false; |
| 345 | } |
| 346 | |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 347 | /* |
Jerome Marchand | 9e4be47 | 2013-11-12 15:07:12 -0800 | [diff] [blame] | 348 | * Isolate free pages onto a private freelist. If @strict is true, will abort |
| 349 | * returning 0 on any invalid PFNs or non-free pages inside of the pageblock |
| 350 | * (even though it may still end up isolating some pages). |
Michal Nazarewicz | 85aa125 | 2012-01-30 13:24:03 +0100 | [diff] [blame] | 351 | */ |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 352 | static unsigned long isolate_freepages_block(struct compact_control *cc, |
Vlastimil Babka | e14c720 | 2014-10-09 15:27:20 -0700 | [diff] [blame] | 353 | unsigned long *start_pfn, |
Michal Nazarewicz | 85aa125 | 2012-01-30 13:24:03 +0100 | [diff] [blame] | 354 | unsigned long end_pfn, |
| 355 | struct list_head *freelist, |
| 356 | bool strict) |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 357 | { |
Mel Gorman | b7aba69 | 2011-01-13 15:45:54 -0800 | [diff] [blame] | 358 | int nr_scanned = 0, total_isolated = 0; |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 359 | struct page *cursor, *valid_page = NULL; |
Xiubo Li | b8b2d82 | 2014-10-09 15:28:21 -0700 | [diff] [blame] | 360 | unsigned long flags = 0; |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 361 | bool locked = false; |
Vlastimil Babka | e14c720 | 2014-10-09 15:27:20 -0700 | [diff] [blame] | 362 | unsigned long blockpfn = *start_pfn; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 363 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 364 | cursor = pfn_to_page(blockpfn); |
| 365 | |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 366 | /* Isolate free pages. */ |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 367 | for (; blockpfn < end_pfn; blockpfn++, cursor++) { |
| 368 | int isolated, i; |
| 369 | struct page *page = cursor; |
| 370 | |
Vlastimil Babka | 8b44d27 | 2014-10-09 15:27:16 -0700 | [diff] [blame] | 371 | /* |
| 372 | * Periodically drop the lock (if held) regardless of its |
| 373 | * contention, to give chance to IRQs. Abort if fatal signal |
| 374 | * pending or async compaction detects need_resched() |
| 375 | */ |
| 376 | if (!(blockpfn % SWAP_CLUSTER_MAX) |
| 377 | && compact_unlock_should_abort(&cc->zone->lock, flags, |
| 378 | &locked, cc)) |
| 379 | break; |
| 380 | |
Mel Gorman | b7aba69 | 2011-01-13 15:45:54 -0800 | [diff] [blame] | 381 | nr_scanned++; |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 382 | if (!pfn_valid_within(blockpfn)) |
Laura Abbott | 2af120b | 2014-03-10 15:49:44 -0700 | [diff] [blame] | 383 | goto isolate_fail; |
| 384 | |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 385 | if (!valid_page) |
| 386 | valid_page = page; |
Vlastimil Babka | 9fcd6d2 | 2015-09-08 15:02:49 -0700 | [diff] [blame] | 387 | |
| 388 | /* |
| 389 | * For compound pages such as THP and hugetlbfs, we can save |
| 390 | * potentially a lot of iterations if we skip them at once. |
| 391 | * The check is racy, but we can consider only valid values |
| 392 | * and the only danger is skipping too much. |
| 393 | */ |
| 394 | if (PageCompound(page)) { |
| 395 | unsigned int comp_order = compound_order(page); |
| 396 | |
| 397 | if (likely(comp_order < MAX_ORDER)) { |
| 398 | blockpfn += (1UL << comp_order) - 1; |
| 399 | cursor += (1UL << comp_order) - 1; |
| 400 | } |
| 401 | |
| 402 | goto isolate_fail; |
| 403 | } |
| 404 | |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 405 | if (!PageBuddy(page)) |
Laura Abbott | 2af120b | 2014-03-10 15:49:44 -0700 | [diff] [blame] | 406 | goto isolate_fail; |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 407 | |
| 408 | /* |
Vlastimil Babka | 69b7189 | 2014-10-09 15:27:18 -0700 | [diff] [blame] | 409 | * If we already hold the lock, we can skip some rechecking. |
| 410 | * Note that if we hold the lock now, checked_pageblock was |
| 411 | * already set in some previous iteration (or strict is true), |
| 412 | * so it is correct to skip the suitable migration target |
| 413 | * recheck as well. |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 414 | */ |
Vlastimil Babka | 69b7189 | 2014-10-09 15:27:18 -0700 | [diff] [blame] | 415 | if (!locked) { |
| 416 | /* |
| 417 | * The zone lock must be held to isolate freepages. |
| 418 | * Unfortunately this is a very coarse lock and can be |
| 419 | * heavily contended if there are parallel allocations |
| 420 | * or parallel compactions. For async compaction do not |
| 421 | * spin on the lock and we acquire the lock as late as |
| 422 | * possible. |
| 423 | */ |
Vlastimil Babka | 8b44d27 | 2014-10-09 15:27:16 -0700 | [diff] [blame] | 424 | locked = compact_trylock_irqsave(&cc->zone->lock, |
| 425 | &flags, cc); |
Vlastimil Babka | 69b7189 | 2014-10-09 15:27:18 -0700 | [diff] [blame] | 426 | if (!locked) |
| 427 | break; |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 428 | |
Vlastimil Babka | 69b7189 | 2014-10-09 15:27:18 -0700 | [diff] [blame] | 429 | /* Recheck this is a buddy page under lock */ |
| 430 | if (!PageBuddy(page)) |
| 431 | goto isolate_fail; |
| 432 | } |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 433 | |
| 434 | /* Found a free page, break it into order-0 pages */ |
| 435 | isolated = split_free_page(page); |
| 436 | total_isolated += isolated; |
| 437 | for (i = 0; i < isolated; i++) { |
| 438 | list_add(&page->lru, freelist); |
| 439 | page++; |
| 440 | } |
| 441 | |
| 442 | /* If a page was split, advance to the end of it */ |
| 443 | if (isolated) { |
Joonsoo Kim | 932ff6b | 2015-02-12 14:59:53 -0800 | [diff] [blame] | 444 | cc->nr_freepages += isolated; |
| 445 | if (!strict && |
| 446 | cc->nr_migratepages <= cc->nr_freepages) { |
| 447 | blockpfn += isolated; |
| 448 | break; |
| 449 | } |
| 450 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 451 | blockpfn += isolated - 1; |
| 452 | cursor += isolated - 1; |
Laura Abbott | 2af120b | 2014-03-10 15:49:44 -0700 | [diff] [blame] | 453 | continue; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 454 | } |
Laura Abbott | 2af120b | 2014-03-10 15:49:44 -0700 | [diff] [blame] | 455 | |
| 456 | isolate_fail: |
| 457 | if (strict) |
| 458 | break; |
| 459 | else |
| 460 | continue; |
| 461 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 462 | } |
| 463 | |
Vlastimil Babka | 9fcd6d2 | 2015-09-08 15:02:49 -0700 | [diff] [blame] | 464 | /* |
| 465 | * There is a tiny chance that we have read bogus compound_order(), |
| 466 | * so be careful to not go outside of the pageblock. |
| 467 | */ |
| 468 | if (unlikely(blockpfn > end_pfn)) |
| 469 | blockpfn = end_pfn; |
| 470 | |
Joonsoo Kim | e34d85f | 2015-02-11 15:27:04 -0800 | [diff] [blame] | 471 | trace_mm_compaction_isolate_freepages(*start_pfn, blockpfn, |
| 472 | nr_scanned, total_isolated); |
| 473 | |
Vlastimil Babka | e14c720 | 2014-10-09 15:27:20 -0700 | [diff] [blame] | 474 | /* Record how far we have got within the block */ |
| 475 | *start_pfn = blockpfn; |
| 476 | |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 477 | /* |
| 478 | * If strict isolation is requested by CMA then check that all the |
| 479 | * pages requested were isolated. If there were any failures, 0 is |
| 480 | * returned and CMA will fail. |
| 481 | */ |
Laura Abbott | 2af120b | 2014-03-10 15:49:44 -0700 | [diff] [blame] | 482 | if (strict && blockpfn < end_pfn) |
Mel Gorman | f40d1e4 | 2012-10-08 16:32:36 -0700 | [diff] [blame] | 483 | total_isolated = 0; |
| 484 | |
| 485 | if (locked) |
| 486 | spin_unlock_irqrestore(&cc->zone->lock, flags); |
| 487 | |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 488 | /* Update the pageblock-skip if the whole pageblock was scanned */ |
| 489 | if (blockpfn == end_pfn) |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 490 | update_pageblock_skip(cc, valid_page, total_isolated, false); |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 491 | |
Minchan Kim | 010fc29 | 2012-12-20 15:05:06 -0800 | [diff] [blame] | 492 | count_compact_events(COMPACTFREE_SCANNED, nr_scanned); |
Mel Gorman | 397487d | 2012-10-19 12:00:10 +0100 | [diff] [blame] | 493 | if (total_isolated) |
Minchan Kim | 010fc29 | 2012-12-20 15:05:06 -0800 | [diff] [blame] | 494 | count_compact_events(COMPACTISOLATED, total_isolated); |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 495 | return total_isolated; |
| 496 | } |
| 497 | |
Michal Nazarewicz | 85aa125 | 2012-01-30 13:24:03 +0100 | [diff] [blame] | 498 | /** |
| 499 | * isolate_freepages_range() - isolate free pages. |
| 500 | * @start_pfn: The first PFN to start isolating. |
| 501 | * @end_pfn: The one-past-last PFN. |
| 502 | * |
| 503 | * Non-free pages, invalid PFNs, or zone boundaries within the |
| 504 | * [start_pfn, end_pfn) range are considered errors, cause function to |
| 505 | * undo its actions and return zero. |
| 506 | * |
| 507 | * Otherwise, function returns one-past-the-last PFN of isolated page |
| 508 | * (which may be greater then end_pfn if end fell in a middle of |
| 509 | * a free page). |
| 510 | */ |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 511 | unsigned long |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 512 | isolate_freepages_range(struct compact_control *cc, |
| 513 | unsigned long start_pfn, unsigned long end_pfn) |
Michal Nazarewicz | 85aa125 | 2012-01-30 13:24:03 +0100 | [diff] [blame] | 514 | { |
Joonsoo Kim | e1409c3 | 2016-03-15 14:57:48 -0700 | [diff] [blame] | 515 | unsigned long isolated, pfn, block_start_pfn, block_end_pfn; |
Michal Nazarewicz | 85aa125 | 2012-01-30 13:24:03 +0100 | [diff] [blame] | 516 | LIST_HEAD(freelist); |
| 517 | |
Vlastimil Babka | 7d49d88 | 2014-10-09 15:27:11 -0700 | [diff] [blame] | 518 | pfn = start_pfn; |
Joonsoo Kim | e1409c3 | 2016-03-15 14:57:48 -0700 | [diff] [blame] | 519 | block_start_pfn = pfn & ~(pageblock_nr_pages - 1); |
| 520 | if (block_start_pfn < cc->zone->zone_start_pfn) |
| 521 | block_start_pfn = cc->zone->zone_start_pfn; |
Vlastimil Babka | 7d49d88 | 2014-10-09 15:27:11 -0700 | [diff] [blame] | 522 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); |
Michal Nazarewicz | 85aa125 | 2012-01-30 13:24:03 +0100 | [diff] [blame] | 523 | |
Vlastimil Babka | 7d49d88 | 2014-10-09 15:27:11 -0700 | [diff] [blame] | 524 | for (; pfn < end_pfn; pfn += isolated, |
Joonsoo Kim | e1409c3 | 2016-03-15 14:57:48 -0700 | [diff] [blame] | 525 | block_start_pfn = block_end_pfn, |
Vlastimil Babka | 7d49d88 | 2014-10-09 15:27:11 -0700 | [diff] [blame] | 526 | block_end_pfn += pageblock_nr_pages) { |
Vlastimil Babka | e14c720 | 2014-10-09 15:27:20 -0700 | [diff] [blame] | 527 | /* Protect pfn from changing by isolate_freepages_block */ |
| 528 | unsigned long isolate_start_pfn = pfn; |
Vlastimil Babka | 7d49d88 | 2014-10-09 15:27:11 -0700 | [diff] [blame] | 529 | |
Michal Nazarewicz | 85aa125 | 2012-01-30 13:24:03 +0100 | [diff] [blame] | 530 | block_end_pfn = min(block_end_pfn, end_pfn); |
| 531 | |
Joonsoo Kim | 5842001 | 2014-11-13 15:19:07 -0800 | [diff] [blame] | 532 | /* |
| 533 | * pfn could pass the block_end_pfn if isolated freepage |
| 534 | * is more than pageblock order. In this case, we adjust |
| 535 | * scanning range to right one. |
| 536 | */ |
| 537 | if (pfn >= block_end_pfn) { |
Joonsoo Kim | e1409c3 | 2016-03-15 14:57:48 -0700 | [diff] [blame] | 538 | block_start_pfn = pfn & ~(pageblock_nr_pages - 1); |
Joonsoo Kim | 5842001 | 2014-11-13 15:19:07 -0800 | [diff] [blame] | 539 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); |
| 540 | block_end_pfn = min(block_end_pfn, end_pfn); |
| 541 | } |
| 542 | |
Joonsoo Kim | e1409c3 | 2016-03-15 14:57:48 -0700 | [diff] [blame] | 543 | if (!pageblock_pfn_to_page(block_start_pfn, |
| 544 | block_end_pfn, cc->zone)) |
Vlastimil Babka | 7d49d88 | 2014-10-09 15:27:11 -0700 | [diff] [blame] | 545 | break; |
| 546 | |
Vlastimil Babka | e14c720 | 2014-10-09 15:27:20 -0700 | [diff] [blame] | 547 | isolated = isolate_freepages_block(cc, &isolate_start_pfn, |
| 548 | block_end_pfn, &freelist, true); |
Michal Nazarewicz | 85aa125 | 2012-01-30 13:24:03 +0100 | [diff] [blame] | 549 | |
| 550 | /* |
| 551 | * In strict mode, isolate_freepages_block() returns 0 if |
| 552 | * there are any holes in the block (ie. invalid PFNs or |
| 553 | * non-free pages). |
| 554 | */ |
| 555 | if (!isolated) |
| 556 | break; |
| 557 | |
| 558 | /* |
| 559 | * If we managed to isolate pages, it is always (1 << n) * |
| 560 | * pageblock_nr_pages for some non-negative n. (Max order |
| 561 | * page may span two pageblocks). |
| 562 | */ |
| 563 | } |
| 564 | |
| 565 | /* split_free_page does not map the pages */ |
| 566 | map_pages(&freelist); |
| 567 | |
| 568 | if (pfn < end_pfn) { |
| 569 | /* Loop terminated early, cleanup. */ |
| 570 | release_freepages(&freelist); |
| 571 | return 0; |
| 572 | } |
| 573 | |
| 574 | /* We don't use freelists for anything. */ |
| 575 | return pfn; |
| 576 | } |
| 577 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 578 | /* Update the number of anon and file isolated pages in the zone */ |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 579 | static void acct_isolated(struct zone *zone, struct compact_control *cc) |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 580 | { |
| 581 | struct page *page; |
Minchan Kim | b9e84ac | 2011-10-31 17:06:44 -0700 | [diff] [blame] | 582 | unsigned int count[2] = { 0, }; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 583 | |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 584 | if (list_empty(&cc->migratepages)) |
| 585 | return; |
| 586 | |
Minchan Kim | b9e84ac | 2011-10-31 17:06:44 -0700 | [diff] [blame] | 587 | list_for_each_entry(page, &cc->migratepages, lru) |
| 588 | count[!!page_is_file_cache(page)]++; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 589 | |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 590 | mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]); |
| 591 | mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]); |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 592 | } |
| 593 | |
| 594 | /* Similar to reclaim, but different enough that they don't share logic */ |
| 595 | static bool too_many_isolated(struct zone *zone) |
| 596 | { |
Minchan Kim | bc69304 | 2010-09-09 16:38:00 -0700 | [diff] [blame] | 597 | unsigned long active, inactive, isolated; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 598 | |
| 599 | inactive = zone_page_state(zone, NR_INACTIVE_FILE) + |
| 600 | zone_page_state(zone, NR_INACTIVE_ANON); |
Minchan Kim | bc69304 | 2010-09-09 16:38:00 -0700 | [diff] [blame] | 601 | active = zone_page_state(zone, NR_ACTIVE_FILE) + |
| 602 | zone_page_state(zone, NR_ACTIVE_ANON); |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 603 | isolated = zone_page_state(zone, NR_ISOLATED_FILE) + |
| 604 | zone_page_state(zone, NR_ISOLATED_ANON); |
| 605 | |
Minchan Kim | bc69304 | 2010-09-09 16:38:00 -0700 | [diff] [blame] | 606 | return isolated > (inactive + active) / 2; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 607 | } |
| 608 | |
Michal Nazarewicz | 2fe86e0 | 2012-01-30 13:16:26 +0100 | [diff] [blame] | 609 | /** |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 610 | * isolate_migratepages_block() - isolate all migrate-able pages within |
| 611 | * a single pageblock |
Michal Nazarewicz | 2fe86e0 | 2012-01-30 13:16:26 +0100 | [diff] [blame] | 612 | * @cc: Compaction control structure. |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 613 | * @low_pfn: The first PFN to isolate |
| 614 | * @end_pfn: The one-past-the-last PFN to isolate, within same pageblock |
| 615 | * @isolate_mode: Isolation mode to be used. |
Michal Nazarewicz | 2fe86e0 | 2012-01-30 13:16:26 +0100 | [diff] [blame] | 616 | * |
| 617 | * Isolate all pages that can be migrated from the range specified by |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 618 | * [low_pfn, end_pfn). The range is expected to be within same pageblock. |
| 619 | * Returns zero if there is a fatal signal pending, otherwise PFN of the |
| 620 | * first page that was not scanned (which may be both less, equal to or more |
| 621 | * than end_pfn). |
Michal Nazarewicz | 2fe86e0 | 2012-01-30 13:16:26 +0100 | [diff] [blame] | 622 | * |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 623 | * The pages are isolated on cc->migratepages list (not required to be empty), |
| 624 | * and cc->nr_migratepages is updated accordingly. The cc->migrate_pfn field |
| 625 | * is neither read nor updated. |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 626 | */ |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 627 | static unsigned long |
| 628 | isolate_migratepages_block(struct compact_control *cc, unsigned long low_pfn, |
| 629 | unsigned long end_pfn, isolate_mode_t isolate_mode) |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 630 | { |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 631 | struct zone *zone = cc->zone; |
Mel Gorman | b7aba69 | 2011-01-13 15:45:54 -0800 | [diff] [blame] | 632 | unsigned long nr_scanned = 0, nr_isolated = 0; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 633 | struct list_head *migratelist = &cc->migratepages; |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 634 | struct lruvec *lruvec; |
Xiubo Li | b8b2d82 | 2014-10-09 15:28:21 -0700 | [diff] [blame] | 635 | unsigned long flags = 0; |
Mel Gorman | 2a1402a | 2012-10-08 16:32:33 -0700 | [diff] [blame] | 636 | bool locked = false; |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 637 | struct page *page = NULL, *valid_page = NULL; |
Joonsoo Kim | e34d85f | 2015-02-11 15:27:04 -0800 | [diff] [blame] | 638 | unsigned long start_pfn = low_pfn; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 639 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 640 | /* |
| 641 | * Ensure that there are not too many pages isolated from the LRU |
| 642 | * list by either parallel reclaimers or compaction. If there are, |
| 643 | * delay for some time until fewer pages are isolated |
| 644 | */ |
| 645 | while (unlikely(too_many_isolated(zone))) { |
Mel Gorman | f9e35b3 | 2011-06-15 15:08:52 -0700 | [diff] [blame] | 646 | /* async migration should just abort */ |
David Rientjes | e0b9dae | 2014-06-04 16:08:28 -0700 | [diff] [blame] | 647 | if (cc->mode == MIGRATE_ASYNC) |
Michal Nazarewicz | 2fe86e0 | 2012-01-30 13:16:26 +0100 | [diff] [blame] | 648 | return 0; |
Mel Gorman | f9e35b3 | 2011-06-15 15:08:52 -0700 | [diff] [blame] | 649 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 650 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
| 651 | |
| 652 | if (fatal_signal_pending(current)) |
Michal Nazarewicz | 2fe86e0 | 2012-01-30 13:16:26 +0100 | [diff] [blame] | 653 | return 0; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 654 | } |
| 655 | |
Vlastimil Babka | be97657 | 2014-06-04 16:10:41 -0700 | [diff] [blame] | 656 | if (compact_should_abort(cc)) |
| 657 | return 0; |
David Rientjes | aeef4b8 | 2014-06-04 16:08:31 -0700 | [diff] [blame] | 658 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 659 | /* Time to isolate some pages for migration */ |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 660 | for (; low_pfn < end_pfn; low_pfn++) { |
Vlastimil Babka | 29c0dde | 2015-09-08 15:02:46 -0700 | [diff] [blame] | 661 | bool is_lru; |
| 662 | |
Vlastimil Babka | 8b44d27 | 2014-10-09 15:27:16 -0700 | [diff] [blame] | 663 | /* |
| 664 | * Periodically drop the lock (if held) regardless of its |
| 665 | * contention, to give chance to IRQs. Abort async compaction |
| 666 | * if contended. |
| 667 | */ |
| 668 | if (!(low_pfn % SWAP_CLUSTER_MAX) |
| 669 | && compact_unlock_should_abort(&zone->lru_lock, flags, |
| 670 | &locked, cc)) |
| 671 | break; |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 672 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 673 | if (!pfn_valid_within(low_pfn)) |
| 674 | continue; |
Mel Gorman | b7aba69 | 2011-01-13 15:45:54 -0800 | [diff] [blame] | 675 | nr_scanned++; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 676 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 677 | page = pfn_to_page(low_pfn); |
Mel Gorman | dc90860 | 2012-02-08 17:13:38 -0800 | [diff] [blame] | 678 | |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 679 | if (!valid_page) |
| 680 | valid_page = page; |
| 681 | |
Mel Gorman | 6c14466 | 2014-01-23 15:53:38 -0800 | [diff] [blame] | 682 | /* |
Vlastimil Babka | 99c0fd5 | 2014-10-09 15:27:23 -0700 | [diff] [blame] | 683 | * Skip if free. We read page order here without zone lock |
| 684 | * which is generally unsafe, but the race window is small and |
| 685 | * the worst thing that can happen is that we skip some |
| 686 | * potential isolation targets. |
Mel Gorman | 6c14466 | 2014-01-23 15:53:38 -0800 | [diff] [blame] | 687 | */ |
Vlastimil Babka | 99c0fd5 | 2014-10-09 15:27:23 -0700 | [diff] [blame] | 688 | if (PageBuddy(page)) { |
| 689 | unsigned long freepage_order = page_order_unsafe(page); |
| 690 | |
| 691 | /* |
| 692 | * Without lock, we cannot be sure that what we got is |
| 693 | * a valid page order. Consider only values in the |
| 694 | * valid order range to prevent low_pfn overflow. |
| 695 | */ |
| 696 | if (freepage_order > 0 && freepage_order < MAX_ORDER) |
| 697 | low_pfn += (1UL << freepage_order) - 1; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 698 | continue; |
Vlastimil Babka | 99c0fd5 | 2014-10-09 15:27:23 -0700 | [diff] [blame] | 699 | } |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 700 | |
Mel Gorman | 9927af74 | 2011-01-13 15:45:59 -0800 | [diff] [blame] | 701 | /* |
Rafael Aquini | bf6bddf | 2012-12-11 16:02:42 -0800 | [diff] [blame] | 702 | * Check may be lockless but that's ok as we recheck later. |
| 703 | * It's possible to migrate LRU pages and balloon pages |
| 704 | * Skip any other type of page |
| 705 | */ |
Vlastimil Babka | 29c0dde | 2015-09-08 15:02:46 -0700 | [diff] [blame] | 706 | is_lru = PageLRU(page); |
| 707 | if (!is_lru) { |
Rafael Aquini | bf6bddf | 2012-12-11 16:02:42 -0800 | [diff] [blame] | 708 | if (unlikely(balloon_page_movable(page))) { |
Konstantin Khlebnikov | d6d86c0 | 2014-10-09 15:29:27 -0700 | [diff] [blame] | 709 | if (balloon_page_isolate(page)) { |
Rafael Aquini | bf6bddf | 2012-12-11 16:02:42 -0800 | [diff] [blame] | 710 | /* Successfully isolated */ |
Joonsoo Kim | b6c7501 | 2014-04-07 15:37:07 -0700 | [diff] [blame] | 711 | goto isolate_success; |
Rafael Aquini | bf6bddf | 2012-12-11 16:02:42 -0800 | [diff] [blame] | 712 | } |
| 713 | } |
Rafael Aquini | bf6bddf | 2012-12-11 16:02:42 -0800 | [diff] [blame] | 714 | } |
Andrea Arcangeli | bc83501 | 2011-01-13 15:47:08 -0800 | [diff] [blame] | 715 | |
| 716 | /* |
Vlastimil Babka | 29c0dde | 2015-09-08 15:02:46 -0700 | [diff] [blame] | 717 | * Regardless of being on LRU, compound pages such as THP and |
| 718 | * hugetlbfs are not to be compacted. We can potentially save |
| 719 | * a lot of iterations if we skip them at once. The check is |
| 720 | * racy, but we can consider only valid values and the only |
| 721 | * danger is skipping too much. |
Andrea Arcangeli | bc83501 | 2011-01-13 15:47:08 -0800 | [diff] [blame] | 722 | */ |
Vlastimil Babka | 29c0dde | 2015-09-08 15:02:46 -0700 | [diff] [blame] | 723 | if (PageCompound(page)) { |
| 724 | unsigned int comp_order = compound_order(page); |
| 725 | |
| 726 | if (likely(comp_order < MAX_ORDER)) |
| 727 | low_pfn += (1UL << comp_order) - 1; |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 728 | |
Mel Gorman | 2a1402a | 2012-10-08 16:32:33 -0700 | [diff] [blame] | 729 | continue; |
| 730 | } |
| 731 | |
Vlastimil Babka | 29c0dde | 2015-09-08 15:02:46 -0700 | [diff] [blame] | 732 | if (!is_lru) |
| 733 | continue; |
| 734 | |
David Rientjes | 119d6d5 | 2014-04-03 14:48:00 -0700 | [diff] [blame] | 735 | /* |
| 736 | * Migration will fail if an anonymous page is pinned in memory, |
| 737 | * so avoid taking lru_lock and isolating it unnecessarily in an |
| 738 | * admittedly racy check. |
| 739 | */ |
| 740 | if (!page_mapping(page) && |
| 741 | page_count(page) > page_mapcount(page)) |
| 742 | continue; |
| 743 | |
Vlastimil Babka | 69b7189 | 2014-10-09 15:27:18 -0700 | [diff] [blame] | 744 | /* If we already hold the lock, we can skip some rechecking */ |
| 745 | if (!locked) { |
Vlastimil Babka | 8b44d27 | 2014-10-09 15:27:16 -0700 | [diff] [blame] | 746 | locked = compact_trylock_irqsave(&zone->lru_lock, |
| 747 | &flags, cc); |
Vlastimil Babka | 69b7189 | 2014-10-09 15:27:18 -0700 | [diff] [blame] | 748 | if (!locked) |
| 749 | break; |
Mel Gorman | 2a1402a | 2012-10-08 16:32:33 -0700 | [diff] [blame] | 750 | |
Vlastimil Babka | 29c0dde | 2015-09-08 15:02:46 -0700 | [diff] [blame] | 751 | /* Recheck PageLRU and PageCompound under lock */ |
Vlastimil Babka | 69b7189 | 2014-10-09 15:27:18 -0700 | [diff] [blame] | 752 | if (!PageLRU(page)) |
| 753 | continue; |
Vlastimil Babka | 29c0dde | 2015-09-08 15:02:46 -0700 | [diff] [blame] | 754 | |
| 755 | /* |
| 756 | * Page become compound since the non-locked check, |
| 757 | * and it's on LRU. It can only be a THP so the order |
| 758 | * is safe to read and it's 0 for tail pages. |
| 759 | */ |
| 760 | if (unlikely(PageCompound(page))) { |
| 761 | low_pfn += (1UL << compound_order(page)) - 1; |
Vlastimil Babka | 69b7189 | 2014-10-09 15:27:18 -0700 | [diff] [blame] | 762 | continue; |
| 763 | } |
Andrea Arcangeli | bc83501 | 2011-01-13 15:47:08 -0800 | [diff] [blame] | 764 | } |
| 765 | |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 766 | lruvec = mem_cgroup_page_lruvec(page, zone); |
| 767 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 768 | /* Try isolate the page */ |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 769 | if (__isolate_lru_page(page, isolate_mode) != 0) |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 770 | continue; |
| 771 | |
Vlastimil Babka | 29c0dde | 2015-09-08 15:02:46 -0700 | [diff] [blame] | 772 | VM_BUG_ON_PAGE(PageCompound(page), page); |
Andrea Arcangeli | bc83501 | 2011-01-13 15:47:08 -0800 | [diff] [blame] | 773 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 774 | /* Successfully isolated */ |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 775 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
Joonsoo Kim | b6c7501 | 2014-04-07 15:37:07 -0700 | [diff] [blame] | 776 | |
| 777 | isolate_success: |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 778 | list_add(&page->lru, migratelist); |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 779 | cc->nr_migratepages++; |
Mel Gorman | b7aba69 | 2011-01-13 15:45:54 -0800 | [diff] [blame] | 780 | nr_isolated++; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 781 | |
| 782 | /* Avoid isolating too much */ |
Hillf Danton | 31b8384 | 2012-01-10 15:07:59 -0800 | [diff] [blame] | 783 | if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) { |
| 784 | ++low_pfn; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 785 | break; |
Hillf Danton | 31b8384 | 2012-01-10 15:07:59 -0800 | [diff] [blame] | 786 | } |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 787 | } |
| 788 | |
Vlastimil Babka | 99c0fd5 | 2014-10-09 15:27:23 -0700 | [diff] [blame] | 789 | /* |
| 790 | * The PageBuddy() check could have potentially brought us outside |
| 791 | * the range to be scanned. |
| 792 | */ |
| 793 | if (unlikely(low_pfn > end_pfn)) |
| 794 | low_pfn = end_pfn; |
| 795 | |
Mel Gorman | c67fe37 | 2012-08-21 16:16:17 -0700 | [diff] [blame] | 796 | if (locked) |
| 797 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 798 | |
Vlastimil Babka | 50b5b09 | 2014-01-21 15:51:10 -0800 | [diff] [blame] | 799 | /* |
| 800 | * Update the pageblock-skip information and cached scanner pfn, |
| 801 | * if the whole pageblock was scanned without isolating any page. |
Vlastimil Babka | 50b5b09 | 2014-01-21 15:51:10 -0800 | [diff] [blame] | 802 | */ |
David Rientjes | 35979ef | 2014-06-04 16:08:27 -0700 | [diff] [blame] | 803 | if (low_pfn == end_pfn) |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 804 | update_pageblock_skip(cc, valid_page, nr_isolated, true); |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 805 | |
Joonsoo Kim | e34d85f | 2015-02-11 15:27:04 -0800 | [diff] [blame] | 806 | trace_mm_compaction_isolate_migratepages(start_pfn, low_pfn, |
| 807 | nr_scanned, nr_isolated); |
Mel Gorman | b7aba69 | 2011-01-13 15:45:54 -0800 | [diff] [blame] | 808 | |
Minchan Kim | 010fc29 | 2012-12-20 15:05:06 -0800 | [diff] [blame] | 809 | count_compact_events(COMPACTMIGRATE_SCANNED, nr_scanned); |
Mel Gorman | 397487d | 2012-10-19 12:00:10 +0100 | [diff] [blame] | 810 | if (nr_isolated) |
Minchan Kim | 010fc29 | 2012-12-20 15:05:06 -0800 | [diff] [blame] | 811 | count_compact_events(COMPACTISOLATED, nr_isolated); |
Mel Gorman | 397487d | 2012-10-19 12:00:10 +0100 | [diff] [blame] | 812 | |
Michal Nazarewicz | 2fe86e0 | 2012-01-30 13:16:26 +0100 | [diff] [blame] | 813 | return low_pfn; |
| 814 | } |
| 815 | |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 816 | /** |
| 817 | * isolate_migratepages_range() - isolate migrate-able pages in a PFN range |
| 818 | * @cc: Compaction control structure. |
| 819 | * @start_pfn: The first PFN to start isolating. |
| 820 | * @end_pfn: The one-past-last PFN. |
| 821 | * |
| 822 | * Returns zero if isolation fails fatally due to e.g. pending signal. |
| 823 | * Otherwise, function returns one-past-the-last PFN of isolated page |
| 824 | * (which may be greater than end_pfn if end fell in a middle of a THP page). |
| 825 | */ |
| 826 | unsigned long |
| 827 | isolate_migratepages_range(struct compact_control *cc, unsigned long start_pfn, |
| 828 | unsigned long end_pfn) |
| 829 | { |
Joonsoo Kim | e1409c3 | 2016-03-15 14:57:48 -0700 | [diff] [blame] | 830 | unsigned long pfn, block_start_pfn, block_end_pfn; |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 831 | |
| 832 | /* Scan block by block. First and last block may be incomplete */ |
| 833 | pfn = start_pfn; |
Joonsoo Kim | e1409c3 | 2016-03-15 14:57:48 -0700 | [diff] [blame] | 834 | block_start_pfn = pfn & ~(pageblock_nr_pages - 1); |
| 835 | if (block_start_pfn < cc->zone->zone_start_pfn) |
| 836 | block_start_pfn = cc->zone->zone_start_pfn; |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 837 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); |
| 838 | |
| 839 | for (; pfn < end_pfn; pfn = block_end_pfn, |
Joonsoo Kim | e1409c3 | 2016-03-15 14:57:48 -0700 | [diff] [blame] | 840 | block_start_pfn = block_end_pfn, |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 841 | block_end_pfn += pageblock_nr_pages) { |
| 842 | |
| 843 | block_end_pfn = min(block_end_pfn, end_pfn); |
| 844 | |
Joonsoo Kim | e1409c3 | 2016-03-15 14:57:48 -0700 | [diff] [blame] | 845 | if (!pageblock_pfn_to_page(block_start_pfn, |
| 846 | block_end_pfn, cc->zone)) |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 847 | continue; |
| 848 | |
| 849 | pfn = isolate_migratepages_block(cc, pfn, block_end_pfn, |
| 850 | ISOLATE_UNEVICTABLE); |
| 851 | |
| 852 | /* |
| 853 | * In case of fatal failure, release everything that might |
| 854 | * have been isolated in the previous iteration, and signal |
| 855 | * the failure back to caller. |
| 856 | */ |
| 857 | if (!pfn) { |
| 858 | putback_movable_pages(&cc->migratepages); |
| 859 | cc->nr_migratepages = 0; |
| 860 | break; |
| 861 | } |
Joonsoo Kim | 6ea41c0 | 2014-10-29 14:50:20 -0700 | [diff] [blame] | 862 | |
| 863 | if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) |
| 864 | break; |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 865 | } |
| 866 | acct_isolated(cc->zone, cc); |
| 867 | |
| 868 | return pfn; |
| 869 | } |
| 870 | |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 871 | #endif /* CONFIG_COMPACTION || CONFIG_CMA */ |
| 872 | #ifdef CONFIG_COMPACTION |
Andrew Morton | 018e9a4 | 2015-04-15 16:15:20 -0700 | [diff] [blame] | 873 | |
| 874 | /* Returns true if the page is within a block suitable for migration to */ |
| 875 | static bool suitable_migration_target(struct page *page) |
| 876 | { |
| 877 | /* If the page is a large free page, then disallow migration */ |
| 878 | if (PageBuddy(page)) { |
| 879 | /* |
| 880 | * We are checking page_order without zone->lock taken. But |
| 881 | * the only small danger is that we skip a potentially suitable |
| 882 | * pageblock, so it's not worth to check order for valid range. |
| 883 | */ |
| 884 | if (page_order_unsafe(page) >= pageblock_order) |
| 885 | return false; |
| 886 | } |
| 887 | |
| 888 | /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */ |
| 889 | if (migrate_async_suitable(get_pageblock_migratetype(page))) |
| 890 | return true; |
| 891 | |
| 892 | /* Otherwise skip the block */ |
| 893 | return false; |
| 894 | } |
| 895 | |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 896 | /* |
Vlastimil Babka | f2849aa | 2015-09-08 15:02:36 -0700 | [diff] [blame] | 897 | * Test whether the free scanner has reached the same or lower pageblock than |
| 898 | * the migration scanner, and compaction should thus terminate. |
| 899 | */ |
| 900 | static inline bool compact_scanners_met(struct compact_control *cc) |
| 901 | { |
| 902 | return (cc->free_pfn >> pageblock_order) |
| 903 | <= (cc->migrate_pfn >> pageblock_order); |
| 904 | } |
| 905 | |
| 906 | /* |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 907 | * Based on information in the current compact_control, find blocks |
| 908 | * suitable for isolating free pages from and then isolate them. |
| 909 | */ |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 910 | static void isolate_freepages(struct compact_control *cc) |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 911 | { |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 912 | struct zone *zone = cc->zone; |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 913 | struct page *page; |
Vlastimil Babka | c96b9e5 | 2014-06-04 16:07:26 -0700 | [diff] [blame] | 914 | unsigned long block_start_pfn; /* start of current pageblock */ |
Vlastimil Babka | e14c720 | 2014-10-09 15:27:20 -0700 | [diff] [blame] | 915 | unsigned long isolate_start_pfn; /* exact pfn we start at */ |
Vlastimil Babka | c96b9e5 | 2014-06-04 16:07:26 -0700 | [diff] [blame] | 916 | unsigned long block_end_pfn; /* end of current pageblock */ |
| 917 | unsigned long low_pfn; /* lowest pfn scanner is able to scan */ |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 918 | struct list_head *freelist = &cc->freepages; |
| 919 | |
| 920 | /* |
| 921 | * Initialise the free scanner. The starting point is where we last |
Vlastimil Babka | 49e068f | 2014-05-06 12:50:03 -0700 | [diff] [blame] | 922 | * successfully isolated from, zone-cached value, or the end of the |
Vlastimil Babka | e14c720 | 2014-10-09 15:27:20 -0700 | [diff] [blame] | 923 | * zone when isolating for the first time. For looping we also need |
| 924 | * this pfn aligned down to the pageblock boundary, because we do |
Vlastimil Babka | c96b9e5 | 2014-06-04 16:07:26 -0700 | [diff] [blame] | 925 | * block_start_pfn -= pageblock_nr_pages in the for loop. |
| 926 | * For ending point, take care when isolating in last pageblock of a |
| 927 | * a zone which ends in the middle of a pageblock. |
Vlastimil Babka | 49e068f | 2014-05-06 12:50:03 -0700 | [diff] [blame] | 928 | * The low boundary is the end of the pageblock the migration scanner |
| 929 | * is using. |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 930 | */ |
Vlastimil Babka | e14c720 | 2014-10-09 15:27:20 -0700 | [diff] [blame] | 931 | isolate_start_pfn = cc->free_pfn; |
Vlastimil Babka | c96b9e5 | 2014-06-04 16:07:26 -0700 | [diff] [blame] | 932 | block_start_pfn = cc->free_pfn & ~(pageblock_nr_pages-1); |
| 933 | block_end_pfn = min(block_start_pfn + pageblock_nr_pages, |
| 934 | zone_end_pfn(zone)); |
Vlastimil Babka | 7ed695e | 2014-01-21 15:51:09 -0800 | [diff] [blame] | 935 | low_pfn = ALIGN(cc->migrate_pfn + 1, pageblock_nr_pages); |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 936 | |
| 937 | /* |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 938 | * Isolate free pages until enough are available to migrate the |
| 939 | * pages on cc->migratepages. We stop searching if the migrate |
| 940 | * and free page scanners meet or enough free pages are isolated. |
| 941 | */ |
Vlastimil Babka | f5f61a3 | 2015-09-08 15:02:39 -0700 | [diff] [blame] | 942 | for (; block_start_pfn >= low_pfn; |
Vlastimil Babka | c96b9e5 | 2014-06-04 16:07:26 -0700 | [diff] [blame] | 943 | block_end_pfn = block_start_pfn, |
Vlastimil Babka | e14c720 | 2014-10-09 15:27:20 -0700 | [diff] [blame] | 944 | block_start_pfn -= pageblock_nr_pages, |
| 945 | isolate_start_pfn = block_start_pfn) { |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 946 | |
David Rientjes | f6ea3ad | 2013-09-30 13:45:03 -0700 | [diff] [blame] | 947 | /* |
| 948 | * This can iterate a massively long zone without finding any |
| 949 | * suitable migration targets, so periodically check if we need |
Vlastimil Babka | be97657 | 2014-06-04 16:10:41 -0700 | [diff] [blame] | 950 | * to schedule, or even abort async compaction. |
David Rientjes | f6ea3ad | 2013-09-30 13:45:03 -0700 | [diff] [blame] | 951 | */ |
Vlastimil Babka | be97657 | 2014-06-04 16:10:41 -0700 | [diff] [blame] | 952 | if (!(block_start_pfn % (SWAP_CLUSTER_MAX * pageblock_nr_pages)) |
| 953 | && compact_should_abort(cc)) |
| 954 | break; |
David Rientjes | f6ea3ad | 2013-09-30 13:45:03 -0700 | [diff] [blame] | 955 | |
Vlastimil Babka | 7d49d88 | 2014-10-09 15:27:11 -0700 | [diff] [blame] | 956 | page = pageblock_pfn_to_page(block_start_pfn, block_end_pfn, |
| 957 | zone); |
| 958 | if (!page) |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 959 | continue; |
| 960 | |
| 961 | /* Check the block is suitable for migration */ |
Linus Torvalds | 68e3e92 | 2012-06-03 20:05:57 -0700 | [diff] [blame] | 962 | if (!suitable_migration_target(page)) |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 963 | continue; |
Linus Torvalds | 68e3e92 | 2012-06-03 20:05:57 -0700 | [diff] [blame] | 964 | |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 965 | /* If isolation recently failed, do not retry */ |
| 966 | if (!isolation_suitable(cc, page)) |
| 967 | continue; |
| 968 | |
Vlastimil Babka | e14c720 | 2014-10-09 15:27:20 -0700 | [diff] [blame] | 969 | /* Found a block suitable for isolating free pages from. */ |
Joonsoo Kim | 932ff6b | 2015-02-12 14:59:53 -0800 | [diff] [blame] | 970 | isolate_freepages_block(cc, &isolate_start_pfn, |
Vlastimil Babka | c96b9e5 | 2014-06-04 16:07:26 -0700 | [diff] [blame] | 971 | block_end_pfn, freelist, false); |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 972 | |
| 973 | /* |
Vlastimil Babka | f5f61a3 | 2015-09-08 15:02:39 -0700 | [diff] [blame] | 974 | * If we isolated enough freepages, or aborted due to async |
| 975 | * compaction being contended, terminate the loop. |
Vlastimil Babka | e14c720 | 2014-10-09 15:27:20 -0700 | [diff] [blame] | 976 | * Remember where the free scanner should restart next time, |
| 977 | * which is where isolate_freepages_block() left off. |
| 978 | * But if it scanned the whole pageblock, isolate_start_pfn |
| 979 | * now points at block_end_pfn, which is the start of the next |
| 980 | * pageblock. |
| 981 | * In that case we will however want to restart at the start |
| 982 | * of the previous pageblock. |
| 983 | */ |
Vlastimil Babka | f5f61a3 | 2015-09-08 15:02:39 -0700 | [diff] [blame] | 984 | if ((cc->nr_freepages >= cc->nr_migratepages) |
| 985 | || cc->contended) { |
| 986 | if (isolate_start_pfn >= block_end_pfn) |
| 987 | isolate_start_pfn = |
| 988 | block_start_pfn - pageblock_nr_pages; |
Vlastimil Babka | be97657 | 2014-06-04 16:10:41 -0700 | [diff] [blame] | 989 | break; |
Vlastimil Babka | f5f61a3 | 2015-09-08 15:02:39 -0700 | [diff] [blame] | 990 | } else { |
| 991 | /* |
| 992 | * isolate_freepages_block() should not terminate |
| 993 | * prematurely unless contended, or isolated enough |
| 994 | */ |
| 995 | VM_BUG_ON(isolate_start_pfn < block_end_pfn); |
| 996 | } |
Michal Nazarewicz | 2fe86e0 | 2012-01-30 13:16:26 +0100 | [diff] [blame] | 997 | } |
| 998 | |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 999 | /* split_free_page does not map the pages */ |
| 1000 | map_pages(freelist); |
Michal Nazarewicz | 2fe86e0 | 2012-01-30 13:16:26 +0100 | [diff] [blame] | 1001 | |
Vlastimil Babka | 7ed695e | 2014-01-21 15:51:09 -0800 | [diff] [blame] | 1002 | /* |
Vlastimil Babka | f5f61a3 | 2015-09-08 15:02:39 -0700 | [diff] [blame] | 1003 | * Record where the free scanner will restart next time. Either we |
| 1004 | * broke from the loop and set isolate_start_pfn based on the last |
| 1005 | * call to isolate_freepages_block(), or we met the migration scanner |
| 1006 | * and the loop terminated due to isolate_start_pfn < low_pfn |
Vlastimil Babka | 7ed695e | 2014-01-21 15:51:09 -0800 | [diff] [blame] | 1007 | */ |
Vlastimil Babka | f5f61a3 | 2015-09-08 15:02:39 -0700 | [diff] [blame] | 1008 | cc->free_pfn = isolate_start_pfn; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1009 | } |
| 1010 | |
| 1011 | /* |
| 1012 | * This is a migrate-callback that "allocates" freepages by taking pages |
| 1013 | * from the isolated freelists in the block we are migrating to. |
| 1014 | */ |
| 1015 | static struct page *compaction_alloc(struct page *migratepage, |
| 1016 | unsigned long data, |
| 1017 | int **result) |
| 1018 | { |
| 1019 | struct compact_control *cc = (struct compact_control *)data; |
| 1020 | struct page *freepage; |
| 1021 | |
Vlastimil Babka | be97657 | 2014-06-04 16:10:41 -0700 | [diff] [blame] | 1022 | /* |
| 1023 | * Isolate free pages if necessary, and if we are not aborting due to |
| 1024 | * contention. |
| 1025 | */ |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1026 | if (list_empty(&cc->freepages)) { |
Vlastimil Babka | be97657 | 2014-06-04 16:10:41 -0700 | [diff] [blame] | 1027 | if (!cc->contended) |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 1028 | isolate_freepages(cc); |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1029 | |
| 1030 | if (list_empty(&cc->freepages)) |
| 1031 | return NULL; |
| 1032 | } |
| 1033 | |
| 1034 | freepage = list_entry(cc->freepages.next, struct page, lru); |
| 1035 | list_del(&freepage->lru); |
| 1036 | cc->nr_freepages--; |
| 1037 | |
| 1038 | return freepage; |
| 1039 | } |
| 1040 | |
| 1041 | /* |
David Rientjes | d53aea3 | 2014-06-04 16:08:26 -0700 | [diff] [blame] | 1042 | * This is a migrate-callback that "frees" freepages back to the isolated |
| 1043 | * freelist. All pages on the freelist are from the same zone, so there is no |
| 1044 | * special handling needed for NUMA. |
| 1045 | */ |
| 1046 | static void compaction_free(struct page *page, unsigned long data) |
| 1047 | { |
| 1048 | struct compact_control *cc = (struct compact_control *)data; |
| 1049 | |
| 1050 | list_add(&page->lru, &cc->freepages); |
| 1051 | cc->nr_freepages++; |
| 1052 | } |
| 1053 | |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 1054 | /* possible outcome of isolate_migratepages */ |
| 1055 | typedef enum { |
| 1056 | ISOLATE_ABORT, /* Abort compaction now */ |
| 1057 | ISOLATE_NONE, /* No pages isolated, continue scanning */ |
| 1058 | ISOLATE_SUCCESS, /* Pages isolated, migrate */ |
| 1059 | } isolate_migrate_t; |
| 1060 | |
| 1061 | /* |
Eric B Munson | 5bbe354 | 2015-04-15 16:13:20 -0700 | [diff] [blame] | 1062 | * Allow userspace to control policy on scanning the unevictable LRU for |
| 1063 | * compactable pages. |
| 1064 | */ |
| 1065 | int sysctl_compact_unevictable_allowed __read_mostly = 1; |
| 1066 | |
| 1067 | /* |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 1068 | * Isolate all pages that can be migrated from the first suitable block, |
| 1069 | * starting at the block pointed to by the migrate scanner pfn within |
| 1070 | * compact_control. |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 1071 | */ |
| 1072 | static isolate_migrate_t isolate_migratepages(struct zone *zone, |
| 1073 | struct compact_control *cc) |
| 1074 | { |
Joonsoo Kim | e1409c3 | 2016-03-15 14:57:48 -0700 | [diff] [blame] | 1075 | unsigned long block_start_pfn; |
| 1076 | unsigned long block_end_pfn; |
| 1077 | unsigned long low_pfn; |
Joonsoo Kim | 1a16718 | 2015-09-08 15:03:59 -0700 | [diff] [blame] | 1078 | unsigned long isolate_start_pfn; |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 1079 | struct page *page; |
| 1080 | const isolate_mode_t isolate_mode = |
Eric B Munson | 5bbe354 | 2015-04-15 16:13:20 -0700 | [diff] [blame] | 1081 | (sysctl_compact_unevictable_allowed ? ISOLATE_UNEVICTABLE : 0) | |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 1082 | (cc->mode == MIGRATE_ASYNC ? ISOLATE_ASYNC_MIGRATE : 0); |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 1083 | |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 1084 | /* |
| 1085 | * Start at where we last stopped, or beginning of the zone as |
| 1086 | * initialized by compact_zone() |
| 1087 | */ |
| 1088 | low_pfn = cc->migrate_pfn; |
Joonsoo Kim | e1409c3 | 2016-03-15 14:57:48 -0700 | [diff] [blame] | 1089 | block_start_pfn = cc->migrate_pfn & ~(pageblock_nr_pages - 1); |
| 1090 | if (block_start_pfn < zone->zone_start_pfn) |
| 1091 | block_start_pfn = zone->zone_start_pfn; |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 1092 | |
| 1093 | /* Only scan within a pageblock boundary */ |
Joonsoo Kim | e1409c3 | 2016-03-15 14:57:48 -0700 | [diff] [blame] | 1094 | block_end_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages); |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 1095 | |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 1096 | /* |
| 1097 | * Iterate over whole pageblocks until we find the first suitable. |
| 1098 | * Do not cross the free scanner. |
| 1099 | */ |
Joonsoo Kim | e1409c3 | 2016-03-15 14:57:48 -0700 | [diff] [blame] | 1100 | for (; block_end_pfn <= cc->free_pfn; |
| 1101 | low_pfn = block_end_pfn, |
| 1102 | block_start_pfn = block_end_pfn, |
| 1103 | block_end_pfn += pageblock_nr_pages) { |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 1104 | |
| 1105 | /* |
| 1106 | * This can potentially iterate a massively long zone with |
| 1107 | * many pageblocks unsuitable, so periodically check if we |
| 1108 | * need to schedule, or even abort async compaction. |
| 1109 | */ |
| 1110 | if (!(low_pfn % (SWAP_CLUSTER_MAX * pageblock_nr_pages)) |
| 1111 | && compact_should_abort(cc)) |
| 1112 | break; |
| 1113 | |
Joonsoo Kim | e1409c3 | 2016-03-15 14:57:48 -0700 | [diff] [blame] | 1114 | page = pageblock_pfn_to_page(block_start_pfn, block_end_pfn, |
| 1115 | zone); |
Vlastimil Babka | 7d49d88 | 2014-10-09 15:27:11 -0700 | [diff] [blame] | 1116 | if (!page) |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 1117 | continue; |
| 1118 | |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 1119 | /* If isolation recently failed, do not retry */ |
| 1120 | if (!isolation_suitable(cc, page)) |
| 1121 | continue; |
| 1122 | |
| 1123 | /* |
| 1124 | * For async compaction, also only scan in MOVABLE blocks. |
| 1125 | * Async compaction is optimistic to see if the minimum amount |
| 1126 | * of work satisfies the allocation. |
| 1127 | */ |
| 1128 | if (cc->mode == MIGRATE_ASYNC && |
| 1129 | !migrate_async_suitable(get_pageblock_migratetype(page))) |
| 1130 | continue; |
| 1131 | |
| 1132 | /* Perform the isolation */ |
Joonsoo Kim | 1a16718 | 2015-09-08 15:03:59 -0700 | [diff] [blame] | 1133 | isolate_start_pfn = low_pfn; |
Joonsoo Kim | e1409c3 | 2016-03-15 14:57:48 -0700 | [diff] [blame] | 1134 | low_pfn = isolate_migratepages_block(cc, low_pfn, |
| 1135 | block_end_pfn, isolate_mode); |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 1136 | |
Hugh Dickins | ff59909 | 2015-02-12 15:00:28 -0800 | [diff] [blame] | 1137 | if (!low_pfn || cc->contended) { |
| 1138 | acct_isolated(zone, cc); |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 1139 | return ISOLATE_ABORT; |
Hugh Dickins | ff59909 | 2015-02-12 15:00:28 -0800 | [diff] [blame] | 1140 | } |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 1141 | |
| 1142 | /* |
Joonsoo Kim | 1a16718 | 2015-09-08 15:03:59 -0700 | [diff] [blame] | 1143 | * Record where we could have freed pages by migration and not |
| 1144 | * yet flushed them to buddy allocator. |
| 1145 | * - this is the lowest page that could have been isolated and |
| 1146 | * then freed by migration. |
| 1147 | */ |
| 1148 | if (cc->nr_migratepages && !cc->last_migrated_pfn) |
| 1149 | cc->last_migrated_pfn = isolate_start_pfn; |
| 1150 | |
| 1151 | /* |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 1152 | * Either we isolated something and proceed with migration. Or |
| 1153 | * we failed and compact_zone should decide if we should |
| 1154 | * continue or not. |
| 1155 | */ |
| 1156 | break; |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 1157 | } |
| 1158 | |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 1159 | acct_isolated(zone, cc); |
Vlastimil Babka | f2849aa | 2015-09-08 15:02:36 -0700 | [diff] [blame] | 1160 | /* Record where migration scanner will be restarted. */ |
| 1161 | cc->migrate_pfn = low_pfn; |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 1162 | |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 1163 | return cc->nr_migratepages ? ISOLATE_SUCCESS : ISOLATE_NONE; |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 1164 | } |
| 1165 | |
Yaowei Bai | 21c527a | 2015-11-05 18:47:20 -0800 | [diff] [blame] | 1166 | /* |
| 1167 | * order == -1 is expected when compacting via |
| 1168 | * /proc/sys/vm/compact_memory |
| 1169 | */ |
| 1170 | static inline bool is_via_compact_memory(int order) |
| 1171 | { |
| 1172 | return order == -1; |
| 1173 | } |
| 1174 | |
Joonsoo Kim | 837d026 | 2015-02-11 15:27:06 -0800 | [diff] [blame] | 1175 | static int __compact_finished(struct zone *zone, struct compact_control *cc, |
David Rientjes | 6d7ce55 | 2014-10-09 15:27:27 -0700 | [diff] [blame] | 1176 | const int migratetype) |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1177 | { |
Mel Gorman | 8fb74b9 | 2013-01-11 14:32:16 -0800 | [diff] [blame] | 1178 | unsigned int order; |
Andrea Arcangeli | 5a03b05 | 2011-01-13 15:47:11 -0800 | [diff] [blame] | 1179 | unsigned long watermark; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1180 | |
Vlastimil Babka | be97657 | 2014-06-04 16:10:41 -0700 | [diff] [blame] | 1181 | if (cc->contended || fatal_signal_pending(current)) |
Vlastimil Babka | 2d1e104 | 2015-11-05 18:48:02 -0800 | [diff] [blame] | 1182 | return COMPACT_CONTENDED; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1183 | |
Mel Gorman | 753341a | 2012-10-08 16:32:40 -0700 | [diff] [blame] | 1184 | /* Compaction run completes if the migrate and free scanner meet */ |
Vlastimil Babka | f2849aa | 2015-09-08 15:02:36 -0700 | [diff] [blame] | 1185 | if (compact_scanners_met(cc)) { |
Vlastimil Babka | 55b7c4c | 2014-01-21 15:51:11 -0800 | [diff] [blame] | 1186 | /* Let the next compaction start anew. */ |
Vlastimil Babka | 0233364 | 2015-09-08 15:02:42 -0700 | [diff] [blame] | 1187 | reset_cached_positions(zone); |
Vlastimil Babka | 55b7c4c | 2014-01-21 15:51:11 -0800 | [diff] [blame] | 1188 | |
Mel Gorman | 6299702 | 2012-10-08 16:32:47 -0700 | [diff] [blame] | 1189 | /* |
| 1190 | * Mark that the PG_migrate_skip information should be cleared |
| 1191 | * by kswapd when it goes to sleep. kswapd does not set the |
| 1192 | * flag itself as the decision to be clear should be directly |
| 1193 | * based on an allocation request. |
| 1194 | */ |
| 1195 | if (!current_is_kswapd()) |
| 1196 | zone->compact_blockskip_flush = true; |
| 1197 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1198 | return COMPACT_COMPLETE; |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 1199 | } |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1200 | |
Yaowei Bai | 21c527a | 2015-11-05 18:47:20 -0800 | [diff] [blame] | 1201 | if (is_via_compact_memory(cc->order)) |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1202 | return COMPACT_CONTINUE; |
| 1203 | |
Michal Hocko | 3957c77 | 2011-06-15 15:08:25 -0700 | [diff] [blame] | 1204 | /* Compaction run is not finished if the watermark is not met */ |
| 1205 | watermark = low_wmark_pages(zone); |
Michal Hocko | 3957c77 | 2011-06-15 15:08:25 -0700 | [diff] [blame] | 1206 | |
Vlastimil Babka | ebff398 | 2014-12-10 15:43:22 -0800 | [diff] [blame] | 1207 | if (!zone_watermark_ok(zone, cc->order, watermark, cc->classzone_idx, |
| 1208 | cc->alloc_flags)) |
Michal Hocko | 3957c77 | 2011-06-15 15:08:25 -0700 | [diff] [blame] | 1209 | return COMPACT_CONTINUE; |
| 1210 | |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1211 | /* Direct compactor: Is a suitable page free? */ |
Mel Gorman | 8fb74b9 | 2013-01-11 14:32:16 -0800 | [diff] [blame] | 1212 | for (order = cc->order; order < MAX_ORDER; order++) { |
| 1213 | struct free_area *area = &zone->free_area[order]; |
Joonsoo Kim | 2149cda | 2015-04-14 15:45:21 -0700 | [diff] [blame] | 1214 | bool can_steal; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1215 | |
Mel Gorman | 8fb74b9 | 2013-01-11 14:32:16 -0800 | [diff] [blame] | 1216 | /* Job done if page is free of the right migratetype */ |
David Rientjes | 6d7ce55 | 2014-10-09 15:27:27 -0700 | [diff] [blame] | 1217 | if (!list_empty(&area->free_list[migratetype])) |
Mel Gorman | 8fb74b9 | 2013-01-11 14:32:16 -0800 | [diff] [blame] | 1218 | return COMPACT_PARTIAL; |
| 1219 | |
Joonsoo Kim | 2149cda | 2015-04-14 15:45:21 -0700 | [diff] [blame] | 1220 | #ifdef CONFIG_CMA |
| 1221 | /* MIGRATE_MOVABLE can fallback on MIGRATE_CMA */ |
| 1222 | if (migratetype == MIGRATE_MOVABLE && |
| 1223 | !list_empty(&area->free_list[MIGRATE_CMA])) |
| 1224 | return COMPACT_PARTIAL; |
| 1225 | #endif |
| 1226 | /* |
| 1227 | * Job done if allocation would steal freepages from |
| 1228 | * other migratetype buddy lists. |
| 1229 | */ |
| 1230 | if (find_suitable_fallback(area, order, migratetype, |
| 1231 | true, &can_steal) != -1) |
Mel Gorman | 8fb74b9 | 2013-01-11 14:32:16 -0800 | [diff] [blame] | 1232 | return COMPACT_PARTIAL; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1233 | } |
| 1234 | |
Joonsoo Kim | 837d026 | 2015-02-11 15:27:06 -0800 | [diff] [blame] | 1235 | return COMPACT_NO_SUITABLE_PAGE; |
| 1236 | } |
| 1237 | |
| 1238 | static int compact_finished(struct zone *zone, struct compact_control *cc, |
| 1239 | const int migratetype) |
| 1240 | { |
| 1241 | int ret; |
| 1242 | |
| 1243 | ret = __compact_finished(zone, cc, migratetype); |
| 1244 | trace_mm_compaction_finished(zone, cc->order, ret); |
| 1245 | if (ret == COMPACT_NO_SUITABLE_PAGE) |
| 1246 | ret = COMPACT_CONTINUE; |
| 1247 | |
| 1248 | return ret; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1249 | } |
| 1250 | |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1251 | /* |
| 1252 | * compaction_suitable: Is this suitable to run compaction on this zone now? |
| 1253 | * Returns |
| 1254 | * COMPACT_SKIPPED - If there are too few free pages for compaction |
| 1255 | * COMPACT_PARTIAL - If the allocation would succeed without compaction |
| 1256 | * COMPACT_CONTINUE - If compaction should run now |
| 1257 | */ |
Joonsoo Kim | 837d026 | 2015-02-11 15:27:06 -0800 | [diff] [blame] | 1258 | static unsigned long __compaction_suitable(struct zone *zone, int order, |
Vlastimil Babka | ebff398 | 2014-12-10 15:43:22 -0800 | [diff] [blame] | 1259 | int alloc_flags, int classzone_idx) |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1260 | { |
| 1261 | int fragindex; |
| 1262 | unsigned long watermark; |
| 1263 | |
Yaowei Bai | 21c527a | 2015-11-05 18:47:20 -0800 | [diff] [blame] | 1264 | if (is_via_compact_memory(order)) |
Michal Hocko | 3957c77 | 2011-06-15 15:08:25 -0700 | [diff] [blame] | 1265 | return COMPACT_CONTINUE; |
| 1266 | |
Vlastimil Babka | ebff398 | 2014-12-10 15:43:22 -0800 | [diff] [blame] | 1267 | watermark = low_wmark_pages(zone); |
| 1268 | /* |
| 1269 | * If watermarks for high-order allocation are already met, there |
| 1270 | * should be no need for compaction at all. |
| 1271 | */ |
| 1272 | if (zone_watermark_ok(zone, order, watermark, classzone_idx, |
| 1273 | alloc_flags)) |
| 1274 | return COMPACT_PARTIAL; |
| 1275 | |
Michal Hocko | 3957c77 | 2011-06-15 15:08:25 -0700 | [diff] [blame] | 1276 | /* |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1277 | * Watermarks for order-0 must be met for compaction. Note the 2UL. |
| 1278 | * This is because during migration, copies of pages need to be |
| 1279 | * allocated and for a short time, the footprint is higher |
| 1280 | */ |
Vlastimil Babka | ebff398 | 2014-12-10 15:43:22 -0800 | [diff] [blame] | 1281 | watermark += (2UL << order); |
| 1282 | if (!zone_watermark_ok(zone, 0, watermark, classzone_idx, alloc_flags)) |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1283 | return COMPACT_SKIPPED; |
| 1284 | |
| 1285 | /* |
| 1286 | * fragmentation index determines if allocation failures are due to |
| 1287 | * low memory or external fragmentation |
| 1288 | * |
Vlastimil Babka | ebff398 | 2014-12-10 15:43:22 -0800 | [diff] [blame] | 1289 | * index of -1000 would imply allocations might succeed depending on |
| 1290 | * watermarks, but we already failed the high-order watermark check |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1291 | * index towards 0 implies failure is due to lack of memory |
| 1292 | * index towards 1000 implies failure is due to fragmentation |
| 1293 | * |
| 1294 | * Only compact if a failure would be due to fragmentation. |
| 1295 | */ |
| 1296 | fragindex = fragmentation_index(zone, order); |
| 1297 | if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold) |
Joonsoo Kim | 837d026 | 2015-02-11 15:27:06 -0800 | [diff] [blame] | 1298 | return COMPACT_NOT_SUITABLE_ZONE; |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1299 | |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1300 | return COMPACT_CONTINUE; |
| 1301 | } |
| 1302 | |
Joonsoo Kim | 837d026 | 2015-02-11 15:27:06 -0800 | [diff] [blame] | 1303 | unsigned long compaction_suitable(struct zone *zone, int order, |
| 1304 | int alloc_flags, int classzone_idx) |
| 1305 | { |
| 1306 | unsigned long ret; |
| 1307 | |
| 1308 | ret = __compaction_suitable(zone, order, alloc_flags, classzone_idx); |
| 1309 | trace_mm_compaction_suitable(zone, order, ret); |
| 1310 | if (ret == COMPACT_NOT_SUITABLE_ZONE) |
| 1311 | ret = COMPACT_SKIPPED; |
| 1312 | |
| 1313 | return ret; |
| 1314 | } |
| 1315 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1316 | static int compact_zone(struct zone *zone, struct compact_control *cc) |
| 1317 | { |
| 1318 | int ret; |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 1319 | unsigned long start_pfn = zone->zone_start_pfn; |
Cody P Schafer | 108bcc9 | 2013-02-22 16:35:23 -0800 | [diff] [blame] | 1320 | unsigned long end_pfn = zone_end_pfn(zone); |
David Rientjes | 6d7ce55 | 2014-10-09 15:27:27 -0700 | [diff] [blame] | 1321 | const int migratetype = gfpflags_to_migratetype(cc->gfp_mask); |
David Rientjes | e0b9dae | 2014-06-04 16:08:28 -0700 | [diff] [blame] | 1322 | const bool sync = cc->mode != MIGRATE_ASYNC; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1323 | |
Vlastimil Babka | ebff398 | 2014-12-10 15:43:22 -0800 | [diff] [blame] | 1324 | ret = compaction_suitable(zone, cc->order, cc->alloc_flags, |
| 1325 | cc->classzone_idx); |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1326 | switch (ret) { |
| 1327 | case COMPACT_PARTIAL: |
| 1328 | case COMPACT_SKIPPED: |
| 1329 | /* Compaction is likely to fail */ |
| 1330 | return ret; |
| 1331 | case COMPACT_CONTINUE: |
| 1332 | /* Fall through to compaction */ |
| 1333 | ; |
| 1334 | } |
| 1335 | |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 1336 | /* |
Vlastimil Babka | d3132e4 | 2014-01-21 15:51:08 -0800 | [diff] [blame] | 1337 | * Clear pageblock skip if there were failures recently and compaction |
| 1338 | * is about to be retried after being deferred. kswapd does not do |
| 1339 | * this reset as it'll reset the cached information when going to sleep. |
| 1340 | */ |
| 1341 | if (compaction_restarting(zone, cc->order) && !current_is_kswapd()) |
| 1342 | __reset_isolation_suitable(zone); |
| 1343 | |
| 1344 | /* |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 1345 | * Setup to move all movable pages to the end of the zone. Used cached |
| 1346 | * information on where the scanners should start but check that it |
| 1347 | * is initialised by ensuring the values are within zone boundaries. |
| 1348 | */ |
David Rientjes | e0b9dae | 2014-06-04 16:08:28 -0700 | [diff] [blame] | 1349 | cc->migrate_pfn = zone->compact_cached_migrate_pfn[sync]; |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 1350 | cc->free_pfn = zone->compact_cached_free_pfn; |
Joonsoo Kim | 623446e | 2016-03-15 14:57:45 -0700 | [diff] [blame] | 1351 | if (cc->free_pfn < start_pfn || cc->free_pfn >= end_pfn) { |
| 1352 | cc->free_pfn = round_down(end_pfn - 1, pageblock_nr_pages); |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 1353 | zone->compact_cached_free_pfn = cc->free_pfn; |
| 1354 | } |
Joonsoo Kim | 623446e | 2016-03-15 14:57:45 -0700 | [diff] [blame] | 1355 | if (cc->migrate_pfn < start_pfn || cc->migrate_pfn >= end_pfn) { |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 1356 | cc->migrate_pfn = start_pfn; |
David Rientjes | 35979ef | 2014-06-04 16:08:27 -0700 | [diff] [blame] | 1357 | zone->compact_cached_migrate_pfn[0] = cc->migrate_pfn; |
| 1358 | zone->compact_cached_migrate_pfn[1] = cc->migrate_pfn; |
Mel Gorman | c89511a | 2012-10-08 16:32:45 -0700 | [diff] [blame] | 1359 | } |
Joonsoo Kim | 1a16718 | 2015-09-08 15:03:59 -0700 | [diff] [blame] | 1360 | cc->last_migrated_pfn = 0; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1361 | |
Joonsoo Kim | 16c4a09 | 2015-02-11 15:27:01 -0800 | [diff] [blame] | 1362 | trace_mm_compaction_begin(start_pfn, cc->migrate_pfn, |
| 1363 | cc->free_pfn, end_pfn, sync); |
Mel Gorman | 0eb927c | 2014-01-21 15:51:05 -0800 | [diff] [blame] | 1364 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1365 | migrate_prep_local(); |
| 1366 | |
David Rientjes | 6d7ce55 | 2014-10-09 15:27:27 -0700 | [diff] [blame] | 1367 | while ((ret = compact_finished(zone, cc, migratetype)) == |
| 1368 | COMPACT_CONTINUE) { |
Minchan Kim | 9d502c1 | 2011-03-22 16:30:39 -0700 | [diff] [blame] | 1369 | int err; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1370 | |
Mel Gorman | f9e35b3 | 2011-06-15 15:08:52 -0700 | [diff] [blame] | 1371 | switch (isolate_migratepages(zone, cc)) { |
| 1372 | case ISOLATE_ABORT: |
Vlastimil Babka | 2d1e104 | 2015-11-05 18:48:02 -0800 | [diff] [blame] | 1373 | ret = COMPACT_CONTENDED; |
Rafael Aquini | 5733c7d | 2012-12-11 16:02:47 -0800 | [diff] [blame] | 1374 | putback_movable_pages(&cc->migratepages); |
Shaohua Li | e64c523 | 2012-10-08 16:32:27 -0700 | [diff] [blame] | 1375 | cc->nr_migratepages = 0; |
Mel Gorman | f9e35b3 | 2011-06-15 15:08:52 -0700 | [diff] [blame] | 1376 | goto out; |
| 1377 | case ISOLATE_NONE: |
Vlastimil Babka | fdaf7f5 | 2014-12-10 15:43:34 -0800 | [diff] [blame] | 1378 | /* |
| 1379 | * We haven't isolated and migrated anything, but |
| 1380 | * there might still be unflushed migrations from |
| 1381 | * previous cc->order aligned block. |
| 1382 | */ |
| 1383 | goto check_drain; |
Mel Gorman | f9e35b3 | 2011-06-15 15:08:52 -0700 | [diff] [blame] | 1384 | case ISOLATE_SUCCESS: |
| 1385 | ; |
| 1386 | } |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1387 | |
David Rientjes | d53aea3 | 2014-06-04 16:08:26 -0700 | [diff] [blame] | 1388 | err = migrate_pages(&cc->migratepages, compaction_alloc, |
David Rientjes | e0b9dae | 2014-06-04 16:08:28 -0700 | [diff] [blame] | 1389 | compaction_free, (unsigned long)cc, cc->mode, |
Mel Gorman | 7b2a2d4 | 2012-10-19 14:07:31 +0100 | [diff] [blame] | 1390 | MR_COMPACTION); |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1391 | |
Vlastimil Babka | f8c9301 | 2014-06-04 16:08:32 -0700 | [diff] [blame] | 1392 | trace_mm_compaction_migratepages(cc->nr_migratepages, err, |
| 1393 | &cc->migratepages); |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1394 | |
Vlastimil Babka | f8c9301 | 2014-06-04 16:08:32 -0700 | [diff] [blame] | 1395 | /* All pages were either migrated or will be released */ |
| 1396 | cc->nr_migratepages = 0; |
Minchan Kim | 9d502c1 | 2011-03-22 16:30:39 -0700 | [diff] [blame] | 1397 | if (err) { |
Rafael Aquini | 5733c7d | 2012-12-11 16:02:47 -0800 | [diff] [blame] | 1398 | putback_movable_pages(&cc->migratepages); |
Vlastimil Babka | 7ed695e | 2014-01-21 15:51:09 -0800 | [diff] [blame] | 1399 | /* |
| 1400 | * migrate_pages() may return -ENOMEM when scanners meet |
| 1401 | * and we want compact_finished() to detect it |
| 1402 | */ |
Vlastimil Babka | f2849aa | 2015-09-08 15:02:36 -0700 | [diff] [blame] | 1403 | if (err == -ENOMEM && !compact_scanners_met(cc)) { |
Vlastimil Babka | 2d1e104 | 2015-11-05 18:48:02 -0800 | [diff] [blame] | 1404 | ret = COMPACT_CONTENDED; |
David Rientjes | 4bf2bba | 2012-07-11 14:02:13 -0700 | [diff] [blame] | 1405 | goto out; |
| 1406 | } |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1407 | } |
Vlastimil Babka | fdaf7f5 | 2014-12-10 15:43:34 -0800 | [diff] [blame] | 1408 | |
Vlastimil Babka | fdaf7f5 | 2014-12-10 15:43:34 -0800 | [diff] [blame] | 1409 | check_drain: |
| 1410 | /* |
| 1411 | * Has the migration scanner moved away from the previous |
| 1412 | * cc->order aligned block where we migrated from? If yes, |
| 1413 | * flush the pages that were freed, so that they can merge and |
| 1414 | * compact_finished() can detect immediately if allocation |
| 1415 | * would succeed. |
| 1416 | */ |
Joonsoo Kim | 1a16718 | 2015-09-08 15:03:59 -0700 | [diff] [blame] | 1417 | if (cc->order > 0 && cc->last_migrated_pfn) { |
Vlastimil Babka | fdaf7f5 | 2014-12-10 15:43:34 -0800 | [diff] [blame] | 1418 | int cpu; |
| 1419 | unsigned long current_block_start = |
| 1420 | cc->migrate_pfn & ~((1UL << cc->order) - 1); |
| 1421 | |
Joonsoo Kim | 1a16718 | 2015-09-08 15:03:59 -0700 | [diff] [blame] | 1422 | if (cc->last_migrated_pfn < current_block_start) { |
Vlastimil Babka | fdaf7f5 | 2014-12-10 15:43:34 -0800 | [diff] [blame] | 1423 | cpu = get_cpu(); |
| 1424 | lru_add_drain_cpu(cpu); |
| 1425 | drain_local_pages(zone); |
| 1426 | put_cpu(); |
| 1427 | /* No more flushing until we migrate again */ |
Joonsoo Kim | 1a16718 | 2015-09-08 15:03:59 -0700 | [diff] [blame] | 1428 | cc->last_migrated_pfn = 0; |
Vlastimil Babka | fdaf7f5 | 2014-12-10 15:43:34 -0800 | [diff] [blame] | 1429 | } |
| 1430 | } |
| 1431 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1432 | } |
| 1433 | |
Mel Gorman | f9e35b3 | 2011-06-15 15:08:52 -0700 | [diff] [blame] | 1434 | out: |
Vlastimil Babka | 6bace09 | 2014-12-10 15:43:31 -0800 | [diff] [blame] | 1435 | /* |
| 1436 | * Release free pages and update where the free scanner should restart, |
| 1437 | * so we don't leave any returned pages behind in the next attempt. |
| 1438 | */ |
| 1439 | if (cc->nr_freepages > 0) { |
| 1440 | unsigned long free_pfn = release_freepages(&cc->freepages); |
| 1441 | |
| 1442 | cc->nr_freepages = 0; |
| 1443 | VM_BUG_ON(free_pfn == 0); |
| 1444 | /* The cached pfn is always the first in a pageblock */ |
| 1445 | free_pfn &= ~(pageblock_nr_pages-1); |
| 1446 | /* |
| 1447 | * Only go back, not forward. The cached pfn might have been |
| 1448 | * already reset to zone end in compact_finished() |
| 1449 | */ |
| 1450 | if (free_pfn > zone->compact_cached_free_pfn) |
| 1451 | zone->compact_cached_free_pfn = free_pfn; |
| 1452 | } |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1453 | |
Joonsoo Kim | 16c4a09 | 2015-02-11 15:27:01 -0800 | [diff] [blame] | 1454 | trace_mm_compaction_end(start_pfn, cc->migrate_pfn, |
| 1455 | cc->free_pfn, end_pfn, sync, ret); |
Mel Gorman | 0eb927c | 2014-01-21 15:51:05 -0800 | [diff] [blame] | 1456 | |
Vlastimil Babka | 2d1e104 | 2015-11-05 18:48:02 -0800 | [diff] [blame] | 1457 | if (ret == COMPACT_CONTENDED) |
| 1458 | ret = COMPACT_PARTIAL; |
| 1459 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1460 | return ret; |
| 1461 | } |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1462 | |
David Rientjes | e0b9dae | 2014-06-04 16:08:28 -0700 | [diff] [blame] | 1463 | static unsigned long compact_zone_order(struct zone *zone, int order, |
Vlastimil Babka | ebff398 | 2014-12-10 15:43:22 -0800 | [diff] [blame] | 1464 | gfp_t gfp_mask, enum migrate_mode mode, int *contended, |
| 1465 | int alloc_flags, int classzone_idx) |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1466 | { |
Shaohua Li | e64c523 | 2012-10-08 16:32:27 -0700 | [diff] [blame] | 1467 | unsigned long ret; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1468 | struct compact_control cc = { |
| 1469 | .nr_freepages = 0, |
| 1470 | .nr_migratepages = 0, |
| 1471 | .order = order, |
David Rientjes | 6d7ce55 | 2014-10-09 15:27:27 -0700 | [diff] [blame] | 1472 | .gfp_mask = gfp_mask, |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1473 | .zone = zone, |
David Rientjes | e0b9dae | 2014-06-04 16:08:28 -0700 | [diff] [blame] | 1474 | .mode = mode, |
Vlastimil Babka | ebff398 | 2014-12-10 15:43:22 -0800 | [diff] [blame] | 1475 | .alloc_flags = alloc_flags, |
| 1476 | .classzone_idx = classzone_idx, |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1477 | }; |
| 1478 | INIT_LIST_HEAD(&cc.freepages); |
| 1479 | INIT_LIST_HEAD(&cc.migratepages); |
| 1480 | |
Shaohua Li | e64c523 | 2012-10-08 16:32:27 -0700 | [diff] [blame] | 1481 | ret = compact_zone(zone, &cc); |
| 1482 | |
| 1483 | VM_BUG_ON(!list_empty(&cc.freepages)); |
| 1484 | VM_BUG_ON(!list_empty(&cc.migratepages)); |
| 1485 | |
| 1486 | *contended = cc.contended; |
| 1487 | return ret; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1488 | } |
| 1489 | |
Mel Gorman | 5e77190 | 2010-05-24 14:32:31 -0700 | [diff] [blame] | 1490 | int sysctl_extfrag_threshold = 500; |
| 1491 | |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1492 | /** |
| 1493 | * try_to_compact_pages - Direct compact to satisfy a high-order allocation |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1494 | * @gfp_mask: The GFP mask of the current allocation |
Vlastimil Babka | 1a6d53a | 2015-02-11 15:25:44 -0800 | [diff] [blame] | 1495 | * @order: The order of the current allocation |
| 1496 | * @alloc_flags: The allocation flags of the current allocation |
| 1497 | * @ac: The context of current allocation |
David Rientjes | e0b9dae | 2014-06-04 16:08:28 -0700 | [diff] [blame] | 1498 | * @mode: The migration mode for async, sync light, or sync migration |
Vlastimil Babka | 1f9efde | 2014-10-09 15:27:14 -0700 | [diff] [blame] | 1499 | * @contended: Return value that determines if compaction was aborted due to |
| 1500 | * need_resched() or lock contention |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1501 | * |
| 1502 | * This is the main entry point for direct page compaction. |
| 1503 | */ |
Vlastimil Babka | 1a6d53a | 2015-02-11 15:25:44 -0800 | [diff] [blame] | 1504 | unsigned long try_to_compact_pages(gfp_t gfp_mask, unsigned int order, |
| 1505 | int alloc_flags, const struct alloc_context *ac, |
| 1506 | enum migrate_mode mode, int *contended) |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1507 | { |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1508 | int may_enter_fs = gfp_mask & __GFP_FS; |
| 1509 | int may_perform_io = gfp_mask & __GFP_IO; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1510 | struct zoneref *z; |
| 1511 | struct zone *zone; |
Vlastimil Babka | 53853e2 | 2014-10-09 15:27:02 -0700 | [diff] [blame] | 1512 | int rc = COMPACT_DEFERRED; |
Vlastimil Babka | 1f9efde | 2014-10-09 15:27:14 -0700 | [diff] [blame] | 1513 | int all_zones_contended = COMPACT_CONTENDED_LOCK; /* init for &= op */ |
| 1514 | |
| 1515 | *contended = COMPACT_CONTENDED_NONE; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1516 | |
Mel Gorman | 4ffb633 | 2012-10-08 16:29:09 -0700 | [diff] [blame] | 1517 | /* Check if the GFP flags allow compaction */ |
Andrea Arcangeli | c5a73c3 | 2011-01-13 15:47:11 -0800 | [diff] [blame] | 1518 | if (!order || !may_enter_fs || !may_perform_io) |
Vlastimil Babka | 53853e2 | 2014-10-09 15:27:02 -0700 | [diff] [blame] | 1519 | return COMPACT_SKIPPED; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1520 | |
Joonsoo Kim | 837d026 | 2015-02-11 15:27:06 -0800 | [diff] [blame] | 1521 | trace_mm_compaction_try_to_compact_pages(order, gfp_mask, mode); |
| 1522 | |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1523 | /* Compact each zone in the list */ |
Vlastimil Babka | 1a6d53a | 2015-02-11 15:25:44 -0800 | [diff] [blame] | 1524 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
| 1525 | ac->nodemask) { |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1526 | int status; |
Vlastimil Babka | 1f9efde | 2014-10-09 15:27:14 -0700 | [diff] [blame] | 1527 | int zone_contended; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1528 | |
Vlastimil Babka | 53853e2 | 2014-10-09 15:27:02 -0700 | [diff] [blame] | 1529 | if (compaction_deferred(zone, order)) |
| 1530 | continue; |
| 1531 | |
David Rientjes | e0b9dae | 2014-06-04 16:08:28 -0700 | [diff] [blame] | 1532 | status = compact_zone_order(zone, order, gfp_mask, mode, |
Vlastimil Babka | 1a6d53a | 2015-02-11 15:25:44 -0800 | [diff] [blame] | 1533 | &zone_contended, alloc_flags, |
| 1534 | ac->classzone_idx); |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1535 | rc = max(status, rc); |
Vlastimil Babka | 1f9efde | 2014-10-09 15:27:14 -0700 | [diff] [blame] | 1536 | /* |
| 1537 | * It takes at least one zone that wasn't lock contended |
| 1538 | * to clear all_zones_contended. |
| 1539 | */ |
| 1540 | all_zones_contended &= zone_contended; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1541 | |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1542 | /* If a normal allocation would succeed, stop compacting */ |
Vlastimil Babka | ebff398 | 2014-12-10 15:43:22 -0800 | [diff] [blame] | 1543 | if (zone_watermark_ok(zone, order, low_wmark_pages(zone), |
Vlastimil Babka | 1a6d53a | 2015-02-11 15:25:44 -0800 | [diff] [blame] | 1544 | ac->classzone_idx, alloc_flags)) { |
Vlastimil Babka | 53853e2 | 2014-10-09 15:27:02 -0700 | [diff] [blame] | 1545 | /* |
| 1546 | * We think the allocation will succeed in this zone, |
| 1547 | * but it is not certain, hence the false. The caller |
| 1548 | * will repeat this with true if allocation indeed |
| 1549 | * succeeds in this zone. |
| 1550 | */ |
| 1551 | compaction_defer_reset(zone, order, false); |
Vlastimil Babka | 1f9efde | 2014-10-09 15:27:14 -0700 | [diff] [blame] | 1552 | /* |
| 1553 | * It is possible that async compaction aborted due to |
| 1554 | * need_resched() and the watermarks were ok thanks to |
| 1555 | * somebody else freeing memory. The allocation can |
| 1556 | * however still fail so we better signal the |
| 1557 | * need_resched() contention anyway (this will not |
| 1558 | * prevent the allocation attempt). |
| 1559 | */ |
| 1560 | if (zone_contended == COMPACT_CONTENDED_SCHED) |
| 1561 | *contended = COMPACT_CONTENDED_SCHED; |
| 1562 | |
| 1563 | goto break_loop; |
| 1564 | } |
| 1565 | |
Vlastimil Babka | f866979 | 2014-12-10 15:43:28 -0800 | [diff] [blame] | 1566 | if (mode != MIGRATE_ASYNC && status == COMPACT_COMPLETE) { |
Vlastimil Babka | 53853e2 | 2014-10-09 15:27:02 -0700 | [diff] [blame] | 1567 | /* |
| 1568 | * We think that allocation won't succeed in this zone |
| 1569 | * so we defer compaction there. If it ends up |
| 1570 | * succeeding after all, it will be reset. |
| 1571 | */ |
| 1572 | defer_compaction(zone, order); |
| 1573 | } |
Vlastimil Babka | 1f9efde | 2014-10-09 15:27:14 -0700 | [diff] [blame] | 1574 | |
| 1575 | /* |
| 1576 | * We might have stopped compacting due to need_resched() in |
| 1577 | * async compaction, or due to a fatal signal detected. In that |
| 1578 | * case do not try further zones and signal need_resched() |
| 1579 | * contention. |
| 1580 | */ |
| 1581 | if ((zone_contended == COMPACT_CONTENDED_SCHED) |
| 1582 | || fatal_signal_pending(current)) { |
| 1583 | *contended = COMPACT_CONTENDED_SCHED; |
| 1584 | goto break_loop; |
| 1585 | } |
| 1586 | |
| 1587 | continue; |
| 1588 | break_loop: |
| 1589 | /* |
| 1590 | * We might not have tried all the zones, so be conservative |
| 1591 | * and assume they are not all lock contended. |
| 1592 | */ |
| 1593 | all_zones_contended = 0; |
| 1594 | break; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1595 | } |
| 1596 | |
Vlastimil Babka | 1f9efde | 2014-10-09 15:27:14 -0700 | [diff] [blame] | 1597 | /* |
| 1598 | * If at least one zone wasn't deferred or skipped, we report if all |
| 1599 | * zones that were tried were lock contended. |
| 1600 | */ |
| 1601 | if (rc > COMPACT_SKIPPED && all_zones_contended) |
| 1602 | *contended = COMPACT_CONTENDED_LOCK; |
| 1603 | |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1604 | return rc; |
| 1605 | } |
| 1606 | |
| 1607 | |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1608 | /* Compact all zones within a node */ |
Andrew Morton | 7103f16 | 2013-02-22 16:32:33 -0800 | [diff] [blame] | 1609 | static void __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc) |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1610 | { |
| 1611 | int zoneid; |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1612 | struct zone *zone; |
| 1613 | |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1614 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1615 | |
| 1616 | zone = &pgdat->node_zones[zoneid]; |
| 1617 | if (!populated_zone(zone)) |
| 1618 | continue; |
| 1619 | |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 1620 | cc->nr_freepages = 0; |
| 1621 | cc->nr_migratepages = 0; |
| 1622 | cc->zone = zone; |
| 1623 | INIT_LIST_HEAD(&cc->freepages); |
| 1624 | INIT_LIST_HEAD(&cc->migratepages); |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1625 | |
Gioh Kim | 195b0c6 | 2015-04-15 16:13:33 -0700 | [diff] [blame] | 1626 | /* |
| 1627 | * When called via /proc/sys/vm/compact_memory |
| 1628 | * this makes sure we compact the whole zone regardless of |
| 1629 | * cached scanner positions. |
| 1630 | */ |
Yaowei Bai | 21c527a | 2015-11-05 18:47:20 -0800 | [diff] [blame] | 1631 | if (is_via_compact_memory(cc->order)) |
Gioh Kim | 195b0c6 | 2015-04-15 16:13:33 -0700 | [diff] [blame] | 1632 | __reset_isolation_suitable(zone); |
| 1633 | |
Yaowei Bai | 21c527a | 2015-11-05 18:47:20 -0800 | [diff] [blame] | 1634 | if (is_via_compact_memory(cc->order) || |
| 1635 | !compaction_deferred(zone, cc->order)) |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 1636 | compact_zone(zone, cc); |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1637 | |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 1638 | VM_BUG_ON(!list_empty(&cc->freepages)); |
| 1639 | VM_BUG_ON(!list_empty(&cc->migratepages)); |
Joonsoo Kim | 7546934 | 2016-01-14 15:20:48 -0800 | [diff] [blame] | 1640 | |
| 1641 | if (is_via_compact_memory(cc->order)) |
| 1642 | continue; |
| 1643 | |
| 1644 | if (zone_watermark_ok(zone, cc->order, |
| 1645 | low_wmark_pages(zone), 0, 0)) |
| 1646 | compaction_defer_reset(zone, cc->order, false); |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1647 | } |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1648 | } |
| 1649 | |
Andrew Morton | 7103f16 | 2013-02-22 16:32:33 -0800 | [diff] [blame] | 1650 | void compact_pgdat(pg_data_t *pgdat, int order) |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 1651 | { |
| 1652 | struct compact_control cc = { |
| 1653 | .order = order, |
David Rientjes | e0b9dae | 2014-06-04 16:08:28 -0700 | [diff] [blame] | 1654 | .mode = MIGRATE_ASYNC, |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 1655 | }; |
| 1656 | |
Mel Gorman | 3a7200a | 2013-09-11 14:22:19 -0700 | [diff] [blame] | 1657 | if (!order) |
| 1658 | return; |
| 1659 | |
Andrew Morton | 7103f16 | 2013-02-22 16:32:33 -0800 | [diff] [blame] | 1660 | __compact_pgdat(pgdat, &cc); |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 1661 | } |
| 1662 | |
Andrew Morton | 7103f16 | 2013-02-22 16:32:33 -0800 | [diff] [blame] | 1663 | static void compact_node(int nid) |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 1664 | { |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 1665 | struct compact_control cc = { |
| 1666 | .order = -1, |
David Rientjes | e0b9dae | 2014-06-04 16:08:28 -0700 | [diff] [blame] | 1667 | .mode = MIGRATE_SYNC, |
David Rientjes | 91ca918 | 2014-04-03 14:47:23 -0700 | [diff] [blame] | 1668 | .ignore_skip_hint = true, |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 1669 | }; |
| 1670 | |
Andrew Morton | 7103f16 | 2013-02-22 16:32:33 -0800 | [diff] [blame] | 1671 | __compact_pgdat(NODE_DATA(nid), &cc); |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 1672 | } |
| 1673 | |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1674 | /* Compact all nodes in the system */ |
Jason Liu | 7964c06 | 2013-01-11 14:31:47 -0800 | [diff] [blame] | 1675 | static void compact_nodes(void) |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1676 | { |
| 1677 | int nid; |
| 1678 | |
Hugh Dickins | 8575ec2 | 2012-03-21 16:33:53 -0700 | [diff] [blame] | 1679 | /* Flush pending updates to the LRU lists */ |
| 1680 | lru_add_drain_all(); |
| 1681 | |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1682 | for_each_online_node(nid) |
| 1683 | compact_node(nid); |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1684 | } |
| 1685 | |
| 1686 | /* The written value is actually unused, all memory is compacted */ |
| 1687 | int sysctl_compact_memory; |
| 1688 | |
Yaowei Bai | fec4eb2 | 2016-01-14 15:20:09 -0800 | [diff] [blame] | 1689 | /* |
| 1690 | * This is the entry point for compacting all nodes via |
| 1691 | * /proc/sys/vm/compact_memory |
| 1692 | */ |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1693 | int sysctl_compaction_handler(struct ctl_table *table, int write, |
| 1694 | void __user *buffer, size_t *length, loff_t *ppos) |
| 1695 | { |
| 1696 | if (write) |
Jason Liu | 7964c06 | 2013-01-11 14:31:47 -0800 | [diff] [blame] | 1697 | compact_nodes(); |
Mel Gorman | 76ab0f5 | 2010-05-24 14:32:28 -0700 | [diff] [blame] | 1698 | |
| 1699 | return 0; |
| 1700 | } |
Mel Gorman | ed4a6d7 | 2010-05-24 14:32:29 -0700 | [diff] [blame] | 1701 | |
Mel Gorman | 5e77190 | 2010-05-24 14:32:31 -0700 | [diff] [blame] | 1702 | int sysctl_extfrag_handler(struct ctl_table *table, int write, |
| 1703 | void __user *buffer, size_t *length, loff_t *ppos) |
| 1704 | { |
| 1705 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
| 1706 | |
| 1707 | return 0; |
| 1708 | } |
| 1709 | |
Mel Gorman | ed4a6d7 | 2010-05-24 14:32:29 -0700 | [diff] [blame] | 1710 | #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA) |
Rashika Kheria | 74e77fb | 2014-04-03 14:48:01 -0700 | [diff] [blame] | 1711 | static ssize_t sysfs_compact_node(struct device *dev, |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 1712 | struct device_attribute *attr, |
Mel Gorman | ed4a6d7 | 2010-05-24 14:32:29 -0700 | [diff] [blame] | 1713 | const char *buf, size_t count) |
| 1714 | { |
Hugh Dickins | 8575ec2 | 2012-03-21 16:33:53 -0700 | [diff] [blame] | 1715 | int nid = dev->id; |
| 1716 | |
| 1717 | if (nid >= 0 && nid < nr_node_ids && node_online(nid)) { |
| 1718 | /* Flush pending updates to the LRU lists */ |
| 1719 | lru_add_drain_all(); |
| 1720 | |
| 1721 | compact_node(nid); |
| 1722 | } |
Mel Gorman | ed4a6d7 | 2010-05-24 14:32:29 -0700 | [diff] [blame] | 1723 | |
| 1724 | return count; |
| 1725 | } |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 1726 | static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node); |
Mel Gorman | ed4a6d7 | 2010-05-24 14:32:29 -0700 | [diff] [blame] | 1727 | |
| 1728 | int compaction_register_node(struct node *node) |
| 1729 | { |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 1730 | return device_create_file(&node->dev, &dev_attr_compact); |
Mel Gorman | ed4a6d7 | 2010-05-24 14:32:29 -0700 | [diff] [blame] | 1731 | } |
| 1732 | |
| 1733 | void compaction_unregister_node(struct node *node) |
| 1734 | { |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 1735 | return device_remove_file(&node->dev, &dev_attr_compact); |
Mel Gorman | ed4a6d7 | 2010-05-24 14:32:29 -0700 | [diff] [blame] | 1736 | } |
| 1737 | #endif /* CONFIG_SYSFS && CONFIG_NUMA */ |
Michal Nazarewicz | ff9543f | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 1738 | |
| 1739 | #endif /* CONFIG_COMPACTION */ |