| 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> |
| 15 | #include <linux/sysctl.h> |
| 16 | #include <linux/sysfs.h> |
| 17 | #include <linux/balloon_compaction.h> |
| 18 | #include <linux/page-isolation.h> |
| 19 | #include <linux/kasan.h> |
| 20 | #include "internal.h" |
| 21 | |
| 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 | |
| 37 | #if defined CONFIG_COMPACTION || defined CONFIG_CMA |
| 38 | #ifdef CONFIG_TRACEPOINTS |
| 39 | static const char *const compaction_status_string[] = { |
| 40 | "deferred", |
| 41 | "skipped", |
| 42 | "continue", |
| 43 | "partial", |
| 44 | "complete", |
| 45 | "no_suitable_page", |
| 46 | "not_suitable_zone", |
| 47 | }; |
| 48 | #endif |
| 49 | |
| 50 | #define CREATE_TRACE_POINTS |
| 51 | #include <trace/events/compaction.h> |
| 52 | |
| 53 | static unsigned long release_freepages(struct list_head *freelist) |
| 54 | { |
| 55 | struct page *page, *next; |
| 56 | unsigned long high_pfn = 0; |
| 57 | |
| 58 | list_for_each_entry_safe(page, next, freelist, lru) { |
| 59 | unsigned long pfn = page_to_pfn(page); |
| 60 | list_del(&page->lru); |
| 61 | __free_page(page); |
| 62 | if (pfn > high_pfn) |
| 63 | high_pfn = pfn; |
| 64 | } |
| 65 | |
| 66 | return high_pfn; |
| 67 | } |
| 68 | |
| 69 | static void map_pages(struct list_head *list) |
| 70 | { |
| 71 | struct page *page; |
| 72 | |
| 73 | list_for_each_entry(page, list, lru) { |
| 74 | arch_alloc_page(page, 0); |
| 75 | kernel_map_pages(page, 1, 1); |
| 76 | kasan_alloc_pages(page, 0); |
| 77 | } |
| 78 | } |
| 79 | |
| 80 | static inline bool migrate_async_suitable(int migratetype) |
| 81 | { |
| 82 | return is_migrate_cma(migratetype) || migratetype == MIGRATE_MOVABLE; |
| 83 | } |
| 84 | |
| 85 | /* |
| 86 | * Check that the whole (or subset of) a pageblock given by the interval of |
| 87 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it |
| 88 | * with the migration of free compaction scanner. The scanners then need to |
| 89 | * use only pfn_valid_within() check for arches that allow holes within |
| 90 | * pageblocks. |
| 91 | * |
| 92 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. |
| 93 | * |
| 94 | * It's possible on some configurations to have a setup like node0 node1 node0 |
| 95 | * i.e. it's possible that all pages within a zones range of pages do not |
| 96 | * belong to a single zone. We assume that a border between node0 and node1 |
| 97 | * can occur within a single pageblock, but not a node0 node1 node0 |
| 98 | * interleaving within a single pageblock. It is therefore sufficient to check |
| 99 | * the first and last page of a pageblock and avoid checking each individual |
| 100 | * page in a pageblock. |
| 101 | */ |
| 102 | static struct page *pageblock_pfn_to_page(unsigned long start_pfn, |
| 103 | unsigned long end_pfn, struct zone *zone) |
| 104 | { |
| 105 | struct page *start_page; |
| 106 | struct page *end_page; |
| 107 | |
| 108 | /* end_pfn is one past the range we are checking */ |
| 109 | end_pfn--; |
| 110 | |
| 111 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) |
| 112 | return NULL; |
| 113 | |
| 114 | start_page = pfn_to_page(start_pfn); |
| 115 | |
| 116 | if (page_zone(start_page) != zone) |
| 117 | return NULL; |
| 118 | |
| 119 | end_page = pfn_to_page(end_pfn); |
| 120 | |
| 121 | /* This gives a shorter code than deriving page_zone(end_page) */ |
| 122 | if (page_zone_id(start_page) != page_zone_id(end_page)) |
| 123 | return NULL; |
| 124 | |
| 125 | return start_page; |
| 126 | } |
| 127 | |
| 128 | #ifdef CONFIG_COMPACTION |
| 129 | |
| 130 | /* Do not skip compaction more than 64 times */ |
| 131 | #define COMPACT_MAX_DEFER_SHIFT 6 |
| 132 | |
| 133 | /* |
| 134 | * Compaction is deferred when compaction fails to result in a page |
| 135 | * allocation success. 1 << compact_defer_limit compactions are skipped up |
| 136 | * to a limit of 1 << COMPACT_MAX_DEFER_SHIFT |
| 137 | */ |
| 138 | void defer_compaction(struct zone *zone, int order) |
| 139 | { |
| 140 | zone->compact_considered = 0; |
| 141 | zone->compact_defer_shift++; |
| 142 | |
| 143 | if (order < zone->compact_order_failed) |
| 144 | zone->compact_order_failed = order; |
| 145 | |
| 146 | if (zone->compact_defer_shift > COMPACT_MAX_DEFER_SHIFT) |
| 147 | zone->compact_defer_shift = COMPACT_MAX_DEFER_SHIFT; |
| 148 | |
| 149 | trace_mm_compaction_defer_compaction(zone, order); |
| 150 | } |
| 151 | |
| 152 | /* Returns true if compaction should be skipped this time */ |
| 153 | bool compaction_deferred(struct zone *zone, int order) |
| 154 | { |
| 155 | unsigned long defer_limit = 1UL << zone->compact_defer_shift; |
| 156 | |
| 157 | if (order < zone->compact_order_failed) |
| 158 | return false; |
| 159 | |
| 160 | /* Avoid possible overflow */ |
| 161 | if (++zone->compact_considered > defer_limit) |
| 162 | zone->compact_considered = defer_limit; |
| 163 | |
| 164 | if (zone->compact_considered >= defer_limit) |
| 165 | return false; |
| 166 | |
| 167 | trace_mm_compaction_deferred(zone, order); |
| 168 | |
| 169 | return true; |
| 170 | } |
| 171 | |
| 172 | /* |
| 173 | * Update defer tracking counters after successful compaction of given order, |
| 174 | * which means an allocation either succeeded (alloc_success == true) or is |
| 175 | * expected to succeed. |
| 176 | */ |
| 177 | void compaction_defer_reset(struct zone *zone, int order, |
| 178 | bool alloc_success) |
| 179 | { |
| 180 | if (alloc_success) { |
| 181 | zone->compact_considered = 0; |
| 182 | zone->compact_defer_shift = 0; |
| 183 | } |
| 184 | if (order >= zone->compact_order_failed) |
| 185 | zone->compact_order_failed = order + 1; |
| 186 | |
| 187 | trace_mm_compaction_defer_reset(zone, order); |
| 188 | } |
| 189 | |
| 190 | /* Returns true if restarting compaction after many failures */ |
| 191 | bool compaction_restarting(struct zone *zone, int order) |
| 192 | { |
| 193 | if (order < zone->compact_order_failed) |
| 194 | return false; |
| 195 | |
| 196 | return zone->compact_defer_shift == COMPACT_MAX_DEFER_SHIFT && |
| 197 | zone->compact_considered >= 1UL << zone->compact_defer_shift; |
| 198 | } |
| 199 | |
| 200 | /* Returns true if the pageblock should be scanned for pages to isolate. */ |
| 201 | static inline bool isolation_suitable(struct compact_control *cc, |
| 202 | struct page *page) |
| 203 | { |
| 204 | if (cc->ignore_skip_hint) |
| 205 | return true; |
| 206 | |
| 207 | return !get_pageblock_skip(page); |
| 208 | } |
| 209 | |
| 210 | static void reset_cached_positions(struct zone *zone) |
| 211 | { |
| 212 | zone->compact_cached_migrate_pfn[0] = zone->zone_start_pfn; |
| 213 | zone->compact_cached_migrate_pfn[1] = zone->zone_start_pfn; |
| 214 | zone->compact_cached_free_pfn = zone_end_pfn(zone); |
| 215 | } |
| 216 | |
| 217 | /* |
| 218 | * This function is called to clear all cached information on pageblocks that |
| 219 | * should be skipped for page isolation when the migrate and free page scanner |
| 220 | * meet. |
| 221 | */ |
| 222 | static void __reset_isolation_suitable(struct zone *zone) |
| 223 | { |
| 224 | unsigned long start_pfn = zone->zone_start_pfn; |
| 225 | unsigned long end_pfn = zone_end_pfn(zone); |
| 226 | unsigned long pfn; |
| 227 | |
| 228 | zone->compact_blockskip_flush = false; |
| 229 | |
| 230 | /* Walk the zone and mark every pageblock as suitable for isolation */ |
| 231 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { |
| 232 | struct page *page; |
| 233 | |
| 234 | cond_resched(); |
| 235 | |
| 236 | if (!pfn_valid(pfn)) |
| 237 | continue; |
| 238 | |
| 239 | page = pfn_to_page(pfn); |
| 240 | if (zone != page_zone(page)) |
| 241 | continue; |
| 242 | |
| 243 | clear_pageblock_skip(page); |
| 244 | } |
| 245 | |
| 246 | reset_cached_positions(zone); |
| 247 | } |
| 248 | |
| 249 | void reset_isolation_suitable(pg_data_t *pgdat) |
| 250 | { |
| 251 | int zoneid; |
| 252 | |
| 253 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { |
| 254 | struct zone *zone = &pgdat->node_zones[zoneid]; |
| 255 | if (!populated_zone(zone)) |
| 256 | continue; |
| 257 | |
| 258 | /* Only flush if a full compaction finished recently */ |
| 259 | if (zone->compact_blockskip_flush) |
| 260 | __reset_isolation_suitable(zone); |
| 261 | } |
| 262 | } |
| 263 | |
| 264 | /* |
| 265 | * If no pages were isolated then mark this pageblock to be skipped in the |
| 266 | * future. The information is later cleared by __reset_isolation_suitable(). |
| 267 | */ |
| 268 | static void update_pageblock_skip(struct compact_control *cc, |
| 269 | struct page *page, unsigned long nr_isolated, |
| 270 | bool migrate_scanner) |
| 271 | { |
| 272 | struct zone *zone = cc->zone; |
| 273 | unsigned long pfn; |
| 274 | |
| 275 | if (cc->ignore_skip_hint) |
| 276 | return; |
| 277 | |
| 278 | if (!page) |
| 279 | return; |
| 280 | |
| 281 | if (nr_isolated) |
| 282 | return; |
| 283 | |
| 284 | set_pageblock_skip(page); |
| 285 | |
| 286 | pfn = page_to_pfn(page); |
| 287 | |
| 288 | /* Update where async and sync compaction should restart */ |
| 289 | if (migrate_scanner) { |
| 290 | if (pfn > zone->compact_cached_migrate_pfn[0]) |
| 291 | zone->compact_cached_migrate_pfn[0] = pfn; |
| 292 | if (cc->mode != MIGRATE_ASYNC && |
| 293 | pfn > zone->compact_cached_migrate_pfn[1]) |
| 294 | zone->compact_cached_migrate_pfn[1] = pfn; |
| 295 | } else { |
| 296 | if (pfn < zone->compact_cached_free_pfn) |
| 297 | zone->compact_cached_free_pfn = pfn; |
| 298 | } |
| 299 | } |
| 300 | #else |
| 301 | static inline bool isolation_suitable(struct compact_control *cc, |
| 302 | struct page *page) |
| 303 | { |
| 304 | return true; |
| 305 | } |
| 306 | |
| 307 | static void update_pageblock_skip(struct compact_control *cc, |
| 308 | struct page *page, unsigned long nr_isolated, |
| 309 | bool migrate_scanner) |
| 310 | { |
| 311 | } |
| 312 | #endif /* CONFIG_COMPACTION */ |
| 313 | |
| 314 | /* |
| 315 | * Compaction requires the taking of some coarse locks that are potentially |
| 316 | * very heavily contended. For async compaction, back out if the lock cannot |
| 317 | * be taken immediately. For sync compaction, spin on the lock if needed. |
| 318 | * |
| 319 | * Returns true if the lock is held |
| 320 | * Returns false if the lock is not held and compaction should abort |
| 321 | */ |
| 322 | static bool compact_trylock_irqsave(spinlock_t *lock, unsigned long *flags, |
| 323 | struct compact_control *cc) |
| 324 | { |
| 325 | if (cc->mode == MIGRATE_ASYNC) { |
| 326 | if (!spin_trylock_irqsave(lock, *flags)) { |
| 327 | cc->contended = COMPACT_CONTENDED_LOCK; |
| 328 | return false; |
| 329 | } |
| 330 | } else { |
| 331 | spin_lock_irqsave(lock, *flags); |
| 332 | } |
| 333 | |
| 334 | return true; |
| 335 | } |
| 336 | |
| 337 | /* |
| 338 | * Compaction requires the taking of some coarse locks that are potentially |
| 339 | * very heavily contended. The lock should be periodically unlocked to avoid |
| 340 | * having disabled IRQs for a long time, even when there is nobody waiting on |
| 341 | * the lock. It might also be that allowing the IRQs will result in |
| 342 | * need_resched() becoming true. If scheduling is needed, async compaction |
| 343 | * aborts. Sync compaction schedules. |
| 344 | * Either compaction type will also abort if a fatal signal is pending. |
| 345 | * In either case if the lock was locked, it is dropped and not regained. |
| 346 | * |
| 347 | * Returns true if compaction should abort due to fatal signal pending, or |
| 348 | * async compaction due to need_resched() |
| 349 | * Returns false when compaction can continue (sync compaction might have |
| 350 | * scheduled) |
| 351 | */ |
| 352 | static bool compact_unlock_should_abort(spinlock_t *lock, |
| 353 | unsigned long flags, bool *locked, struct compact_control *cc) |
| 354 | { |
| 355 | if (*locked) { |
| 356 | spin_unlock_irqrestore(lock, flags); |
| 357 | *locked = false; |
| 358 | } |
| 359 | |
| 360 | if (fatal_signal_pending(current)) { |
| 361 | cc->contended = COMPACT_CONTENDED_SCHED; |
| 362 | return true; |
| 363 | } |
| 364 | |
| 365 | if (need_resched()) { |
| 366 | if (cc->mode == MIGRATE_ASYNC) { |
| 367 | cc->contended = COMPACT_CONTENDED_SCHED; |
| 368 | return true; |
| 369 | } |
| 370 | cond_resched(); |
| 371 | } |
| 372 | |
| 373 | return false; |
| 374 | } |
| 375 | |
| 376 | /* |
| 377 | * Aside from avoiding lock contention, compaction also periodically checks |
| 378 | * need_resched() and either schedules in sync compaction or aborts async |
| 379 | * compaction. This is similar to what compact_unlock_should_abort() does, but |
| 380 | * is used where no lock is concerned. |
| 381 | * |
| 382 | * Returns false when no scheduling was needed, or sync compaction scheduled. |
| 383 | * Returns true when async compaction should abort. |
| 384 | */ |
| 385 | static inline bool compact_should_abort(struct compact_control *cc) |
| 386 | { |
| 387 | /* async compaction aborts if contended */ |
| 388 | if (need_resched()) { |
| 389 | if (cc->mode == MIGRATE_ASYNC) { |
| 390 | cc->contended = COMPACT_CONTENDED_SCHED; |
| 391 | return true; |
| 392 | } |
| 393 | |
| 394 | cond_resched(); |
| 395 | } |
| 396 | |
| 397 | return false; |
| 398 | } |
| 399 | |
| 400 | /* |
| 401 | * Isolate free pages onto a private freelist. If @strict is true, will abort |
| 402 | * returning 0 on any invalid PFNs or non-free pages inside of the pageblock |
| 403 | * (even though it may still end up isolating some pages). |
| 404 | */ |
| 405 | static unsigned long isolate_freepages_block(struct compact_control *cc, |
| 406 | unsigned long *start_pfn, |
| 407 | unsigned long end_pfn, |
| 408 | struct list_head *freelist, |
| 409 | bool strict) |
| 410 | { |
| 411 | int nr_scanned = 0, total_isolated = 0; |
| 412 | struct page *cursor, *valid_page = NULL; |
| 413 | unsigned long flags = 0; |
| 414 | bool locked = false; |
| 415 | unsigned long blockpfn = *start_pfn; |
| 416 | |
| 417 | cursor = pfn_to_page(blockpfn); |
| 418 | |
| 419 | /* Isolate free pages. */ |
| 420 | for (; blockpfn < end_pfn; blockpfn++, cursor++) { |
| 421 | int isolated, i; |
| 422 | struct page *page = cursor; |
| 423 | |
| 424 | /* |
| 425 | * Periodically drop the lock (if held) regardless of its |
| 426 | * contention, to give chance to IRQs. Abort if fatal signal |
| 427 | * pending or async compaction detects need_resched() |
| 428 | */ |
| 429 | if (!(blockpfn % SWAP_CLUSTER_MAX) |
| 430 | && compact_unlock_should_abort(&cc->zone->lock, flags, |
| 431 | &locked, cc)) |
| 432 | break; |
| 433 | |
| 434 | nr_scanned++; |
| 435 | if (!pfn_valid_within(blockpfn)) |
| 436 | goto isolate_fail; |
| 437 | |
| 438 | if (!valid_page) |
| 439 | valid_page = page; |
| 440 | |
| 441 | /* |
| 442 | * For compound pages such as THP and hugetlbfs, we can save |
| 443 | * potentially a lot of iterations if we skip them at once. |
| 444 | * The check is racy, but we can consider only valid values |
| 445 | * and the only danger is skipping too much. |
| 446 | */ |
| 447 | if (PageCompound(page)) { |
| 448 | unsigned int comp_order = compound_order(page); |
| 449 | |
| 450 | if (likely(comp_order < MAX_ORDER)) { |
| 451 | blockpfn += (1UL << comp_order) - 1; |
| 452 | cursor += (1UL << comp_order) - 1; |
| 453 | } |
| 454 | |
| 455 | goto isolate_fail; |
| 456 | } |
| 457 | |
| 458 | if (!PageBuddy(page)) |
| 459 | goto isolate_fail; |
| 460 | |
| 461 | /* |
| 462 | * If we already hold the lock, we can skip some rechecking. |
| 463 | * Note that if we hold the lock now, checked_pageblock was |
| 464 | * already set in some previous iteration (or strict is true), |
| 465 | * so it is correct to skip the suitable migration target |
| 466 | * recheck as well. |
| 467 | */ |
| 468 | if (!locked) { |
| 469 | /* |
| 470 | * The zone lock must be held to isolate freepages. |
| 471 | * Unfortunately this is a very coarse lock and can be |
| 472 | * heavily contended if there are parallel allocations |
| 473 | * or parallel compactions. For async compaction do not |
| 474 | * spin on the lock and we acquire the lock as late as |
| 475 | * possible. |
| 476 | */ |
| 477 | locked = compact_trylock_irqsave(&cc->zone->lock, |
| 478 | &flags, cc); |
| 479 | if (!locked) |
| 480 | break; |
| 481 | |
| 482 | /* Recheck this is a buddy page under lock */ |
| 483 | if (!PageBuddy(page)) |
| 484 | goto isolate_fail; |
| 485 | } |
| 486 | |
| 487 | /* Found a free page, break it into order-0 pages */ |
| 488 | isolated = split_free_page(page); |
| 489 | total_isolated += isolated; |
| 490 | for (i = 0; i < isolated; i++) { |
| 491 | list_add(&page->lru, freelist); |
| 492 | page++; |
| 493 | } |
| 494 | |
| 495 | /* If a page was split, advance to the end of it */ |
| 496 | if (isolated) { |
| 497 | cc->nr_freepages += isolated; |
| 498 | if (!strict && |
| 499 | cc->nr_migratepages <= cc->nr_freepages) { |
| 500 | blockpfn += isolated; |
| 501 | break; |
| 502 | } |
| 503 | |
| 504 | blockpfn += isolated - 1; |
| 505 | cursor += isolated - 1; |
| 506 | continue; |
| 507 | } |
| 508 | |
| 509 | isolate_fail: |
| 510 | if (strict) |
| 511 | break; |
| 512 | else |
| 513 | continue; |
| 514 | |
| 515 | } |
| 516 | |
| 517 | /* |
| 518 | * There is a tiny chance that we have read bogus compound_order(), |
| 519 | * so be careful to not go outside of the pageblock. |
| 520 | */ |
| 521 | if (unlikely(blockpfn > end_pfn)) |
| 522 | blockpfn = end_pfn; |
| 523 | |
| 524 | trace_mm_compaction_isolate_freepages(*start_pfn, blockpfn, |
| 525 | nr_scanned, total_isolated); |
| 526 | |
| 527 | /* Record how far we have got within the block */ |
| 528 | *start_pfn = blockpfn; |
| 529 | |
| 530 | /* |
| 531 | * If strict isolation is requested by CMA then check that all the |
| 532 | * pages requested were isolated. If there were any failures, 0 is |
| 533 | * returned and CMA will fail. |
| 534 | */ |
| 535 | if (strict && blockpfn < end_pfn) |
| 536 | total_isolated = 0; |
| 537 | |
| 538 | if (locked) |
| 539 | spin_unlock_irqrestore(&cc->zone->lock, flags); |
| 540 | |
| 541 | /* Update the pageblock-skip if the whole pageblock was scanned */ |
| 542 | if (blockpfn == end_pfn) |
| 543 | update_pageblock_skip(cc, valid_page, total_isolated, false); |
| 544 | |
| 545 | count_compact_events(COMPACTFREE_SCANNED, nr_scanned); |
| 546 | if (total_isolated) |
| 547 | count_compact_events(COMPACTISOLATED, total_isolated); |
| 548 | return total_isolated; |
| 549 | } |
| 550 | |
| 551 | /** |
| 552 | * isolate_freepages_range() - isolate free pages. |
| 553 | * @start_pfn: The first PFN to start isolating. |
| 554 | * @end_pfn: The one-past-last PFN. |
| 555 | * |
| 556 | * Non-free pages, invalid PFNs, or zone boundaries within the |
| 557 | * [start_pfn, end_pfn) range are considered errors, cause function to |
| 558 | * undo its actions and return zero. |
| 559 | * |
| 560 | * Otherwise, function returns one-past-the-last PFN of isolated page |
| 561 | * (which may be greater then end_pfn if end fell in a middle of |
| 562 | * a free page). |
| 563 | */ |
| 564 | unsigned long |
| 565 | isolate_freepages_range(struct compact_control *cc, |
| 566 | unsigned long start_pfn, unsigned long end_pfn) |
| 567 | { |
| 568 | unsigned long isolated, pfn, block_end_pfn; |
| 569 | LIST_HEAD(freelist); |
| 570 | |
| 571 | pfn = start_pfn; |
| 572 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); |
| 573 | |
| 574 | for (; pfn < end_pfn; pfn += isolated, |
| 575 | block_end_pfn += pageblock_nr_pages) { |
| 576 | /* Protect pfn from changing by isolate_freepages_block */ |
| 577 | unsigned long isolate_start_pfn = pfn; |
| 578 | |
| 579 | block_end_pfn = min(block_end_pfn, end_pfn); |
| 580 | |
| 581 | /* |
| 582 | * pfn could pass the block_end_pfn if isolated freepage |
| 583 | * is more than pageblock order. In this case, we adjust |
| 584 | * scanning range to right one. |
| 585 | */ |
| 586 | if (pfn >= block_end_pfn) { |
| 587 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); |
| 588 | block_end_pfn = min(block_end_pfn, end_pfn); |
| 589 | } |
| 590 | |
| 591 | if (!pageblock_pfn_to_page(pfn, block_end_pfn, cc->zone)) |
| 592 | break; |
| 593 | |
| 594 | isolated = isolate_freepages_block(cc, &isolate_start_pfn, |
| 595 | block_end_pfn, &freelist, true); |
| 596 | |
| 597 | /* |
| 598 | * In strict mode, isolate_freepages_block() returns 0 if |
| 599 | * there are any holes in the block (ie. invalid PFNs or |
| 600 | * non-free pages). |
| 601 | */ |
| 602 | if (!isolated) |
| 603 | break; |
| 604 | |
| 605 | /* |
| 606 | * If we managed to isolate pages, it is always (1 << n) * |
| 607 | * pageblock_nr_pages for some non-negative n. (Max order |
| 608 | * page may span two pageblocks). |
| 609 | */ |
| 610 | } |
| 611 | |
| 612 | /* split_free_page does not map the pages */ |
| 613 | map_pages(&freelist); |
| 614 | |
| 615 | if (pfn < end_pfn) { |
| 616 | /* Loop terminated early, cleanup. */ |
| 617 | release_freepages(&freelist); |
| 618 | return 0; |
| 619 | } |
| 620 | |
| 621 | /* We don't use freelists for anything. */ |
| 622 | return pfn; |
| 623 | } |
| 624 | |
| 625 | /* Update the number of anon and file isolated pages in the zone */ |
| 626 | static void acct_isolated(struct zone *zone, struct compact_control *cc) |
| 627 | { |
| 628 | struct page *page; |
| 629 | unsigned int count[2] = { 0, }; |
| 630 | |
| 631 | if (list_empty(&cc->migratepages)) |
| 632 | return; |
| 633 | |
| 634 | list_for_each_entry(page, &cc->migratepages, lru) |
| 635 | count[!!page_is_file_cache(page)]++; |
| 636 | |
| 637 | mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]); |
| 638 | mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]); |
| 639 | } |
| 640 | |
| 641 | /* Similar to reclaim, but different enough that they don't share logic */ |
| 642 | static bool too_many_isolated(struct zone *zone) |
| 643 | { |
| 644 | unsigned long active, inactive, isolated; |
| 645 | |
| 646 | inactive = zone_page_state(zone, NR_INACTIVE_FILE) + |
| 647 | zone_page_state(zone, NR_INACTIVE_ANON); |
| 648 | active = zone_page_state(zone, NR_ACTIVE_FILE) + |
| 649 | zone_page_state(zone, NR_ACTIVE_ANON); |
| 650 | isolated = zone_page_state(zone, NR_ISOLATED_FILE) + |
| 651 | zone_page_state(zone, NR_ISOLATED_ANON); |
| 652 | |
| 653 | return isolated > (inactive + active) / 2; |
| 654 | } |
| 655 | |
| 656 | /** |
| 657 | * isolate_migratepages_block() - isolate all migrate-able pages within |
| 658 | * a single pageblock |
| 659 | * @cc: Compaction control structure. |
| 660 | * @low_pfn: The first PFN to isolate |
| 661 | * @end_pfn: The one-past-the-last PFN to isolate, within same pageblock |
| 662 | * @isolate_mode: Isolation mode to be used. |
| 663 | * |
| 664 | * Isolate all pages that can be migrated from the range specified by |
| 665 | * [low_pfn, end_pfn). The range is expected to be within same pageblock. |
| 666 | * Returns zero if there is a fatal signal pending, otherwise PFN of the |
| 667 | * first page that was not scanned (which may be both less, equal to or more |
| 668 | * than end_pfn). |
| 669 | * |
| 670 | * The pages are isolated on cc->migratepages list (not required to be empty), |
| 671 | * and cc->nr_migratepages is updated accordingly. The cc->migrate_pfn field |
| 672 | * is neither read nor updated. |
| 673 | */ |
| 674 | static unsigned long |
| 675 | isolate_migratepages_block(struct compact_control *cc, unsigned long low_pfn, |
| 676 | unsigned long end_pfn, isolate_mode_t isolate_mode) |
| 677 | { |
| 678 | struct zone *zone = cc->zone; |
| 679 | unsigned long nr_scanned = 0, nr_isolated = 0; |
| 680 | struct list_head *migratelist = &cc->migratepages; |
| 681 | struct lruvec *lruvec; |
| 682 | unsigned long flags = 0; |
| 683 | bool locked = false; |
| 684 | struct page *page = NULL, *valid_page = NULL; |
| 685 | unsigned long start_pfn = low_pfn; |
| 686 | |
| 687 | /* |
| 688 | * Ensure that there are not too many pages isolated from the LRU |
| 689 | * list by either parallel reclaimers or compaction. If there are, |
| 690 | * delay for some time until fewer pages are isolated |
| 691 | */ |
| 692 | while (unlikely(too_many_isolated(zone))) { |
| 693 | /* async migration should just abort */ |
| 694 | if (cc->mode == MIGRATE_ASYNC) |
| 695 | return 0; |
| 696 | |
| 697 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
| 698 | |
| 699 | if (fatal_signal_pending(current)) |
| 700 | return 0; |
| 701 | } |
| 702 | |
| 703 | if (compact_should_abort(cc)) |
| 704 | return 0; |
| 705 | |
| 706 | /* Time to isolate some pages for migration */ |
| 707 | for (; low_pfn < end_pfn; low_pfn++) { |
| 708 | bool is_lru; |
| 709 | |
| 710 | /* |
| 711 | * Periodically drop the lock (if held) regardless of its |
| 712 | * contention, to give chance to IRQs. Abort async compaction |
| 713 | * if contended. |
| 714 | */ |
| 715 | if (!(low_pfn % SWAP_CLUSTER_MAX) |
| 716 | && compact_unlock_should_abort(&zone->lru_lock, flags, |
| 717 | &locked, cc)) |
| 718 | break; |
| 719 | |
| 720 | if (!pfn_valid_within(low_pfn)) |
| 721 | continue; |
| 722 | nr_scanned++; |
| 723 | |
| 724 | page = pfn_to_page(low_pfn); |
| 725 | |
| 726 | if (!valid_page) |
| 727 | valid_page = page; |
| 728 | |
| 729 | /* |
| 730 | * Skip if free. We read page order here without zone lock |
| 731 | * which is generally unsafe, but the race window is small and |
| 732 | * the worst thing that can happen is that we skip some |
| 733 | * potential isolation targets. |
| 734 | */ |
| 735 | if (PageBuddy(page)) { |
| 736 | unsigned long freepage_order = page_order_unsafe(page); |
| 737 | |
| 738 | /* |
| 739 | * Without lock, we cannot be sure that what we got is |
| 740 | * a valid page order. Consider only values in the |
| 741 | * valid order range to prevent low_pfn overflow. |
| 742 | */ |
| 743 | if (freepage_order > 0 && freepage_order < MAX_ORDER) |
| 744 | low_pfn += (1UL << freepage_order) - 1; |
| 745 | continue; |
| 746 | } |
| 747 | |
| 748 | /* |
| 749 | * Check may be lockless but that's ok as we recheck later. |
| 750 | * It's possible to migrate LRU pages and balloon pages |
| 751 | * Skip any other type of page |
| 752 | */ |
| 753 | is_lru = PageLRU(page); |
| 754 | if (!is_lru) { |
| 755 | if (unlikely(balloon_page_movable(page))) { |
| 756 | if (balloon_page_isolate(page)) { |
| 757 | /* Successfully isolated */ |
| 758 | goto isolate_success; |
| 759 | } |
| 760 | } |
| 761 | } |
| 762 | |
| 763 | /* |
| 764 | * Regardless of being on LRU, compound pages such as THP and |
| 765 | * hugetlbfs are not to be compacted. We can potentially save |
| 766 | * a lot of iterations if we skip them at once. The check is |
| 767 | * racy, but we can consider only valid values and the only |
| 768 | * danger is skipping too much. |
| 769 | */ |
| 770 | if (PageCompound(page)) { |
| 771 | unsigned int comp_order = compound_order(page); |
| 772 | |
| 773 | if (likely(comp_order < MAX_ORDER)) |
| 774 | low_pfn += (1UL << comp_order) - 1; |
| 775 | |
| 776 | continue; |
| 777 | } |
| 778 | |
| 779 | if (!is_lru) |
| 780 | continue; |
| 781 | |
| 782 | /* |
| 783 | * Migration will fail if an anonymous page is pinned in memory, |
| 784 | * so avoid taking lru_lock and isolating it unnecessarily in an |
| 785 | * admittedly racy check. |
| 786 | */ |
| 787 | if (!page_mapping(page) && |
| 788 | page_count(page) > page_mapcount(page)) |
| 789 | continue; |
| 790 | |
| 791 | /* If we already hold the lock, we can skip some rechecking */ |
| 792 | if (!locked) { |
| 793 | locked = compact_trylock_irqsave(&zone->lru_lock, |
| 794 | &flags, cc); |
| 795 | if (!locked) |
| 796 | break; |
| 797 | |
| 798 | /* Recheck PageLRU and PageCompound under lock */ |
| 799 | if (!PageLRU(page)) |
| 800 | continue; |
| 801 | |
| 802 | /* |
| 803 | * Page become compound since the non-locked check, |
| 804 | * and it's on LRU. It can only be a THP so the order |
| 805 | * is safe to read and it's 0 for tail pages. |
| 806 | */ |
| 807 | if (unlikely(PageCompound(page))) { |
| 808 | low_pfn += (1UL << compound_order(page)) - 1; |
| 809 | continue; |
| 810 | } |
| 811 | } |
| 812 | |
| 813 | lruvec = mem_cgroup_page_lruvec(page, zone); |
| 814 | |
| 815 | /* Try isolate the page */ |
| 816 | if (__isolate_lru_page(page, isolate_mode) != 0) |
| 817 | continue; |
| 818 | |
| 819 | VM_BUG_ON_PAGE(PageCompound(page), page); |
| 820 | |
| 821 | /* Successfully isolated */ |
| 822 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
| 823 | |
| 824 | isolate_success: |
| 825 | list_add(&page->lru, migratelist); |
| 826 | cc->nr_migratepages++; |
| 827 | nr_isolated++; |
| 828 | |
| 829 | /* Avoid isolating too much */ |
| 830 | if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) { |
| 831 | ++low_pfn; |
| 832 | break; |
| 833 | } |
| 834 | } |
| 835 | |
| 836 | /* |
| 837 | * The PageBuddy() check could have potentially brought us outside |
| 838 | * the range to be scanned. |
| 839 | */ |
| 840 | if (unlikely(low_pfn > end_pfn)) |
| 841 | low_pfn = end_pfn; |
| 842 | |
| 843 | if (locked) |
| 844 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
| 845 | |
| 846 | /* |
| 847 | * Update the pageblock-skip information and cached scanner pfn, |
| 848 | * if the whole pageblock was scanned without isolating any page. |
| 849 | */ |
| 850 | if (low_pfn == end_pfn) |
| 851 | update_pageblock_skip(cc, valid_page, nr_isolated, true); |
| 852 | |
| 853 | trace_mm_compaction_isolate_migratepages(start_pfn, low_pfn, |
| 854 | nr_scanned, nr_isolated); |
| 855 | |
| 856 | count_compact_events(COMPACTMIGRATE_SCANNED, nr_scanned); |
| 857 | if (nr_isolated) |
| 858 | count_compact_events(COMPACTISOLATED, nr_isolated); |
| 859 | |
| 860 | return low_pfn; |
| 861 | } |
| 862 | |
| 863 | /** |
| 864 | * isolate_migratepages_range() - isolate migrate-able pages in a PFN range |
| 865 | * @cc: Compaction control structure. |
| 866 | * @start_pfn: The first PFN to start isolating. |
| 867 | * @end_pfn: The one-past-last PFN. |
| 868 | * |
| 869 | * Returns zero if isolation fails fatally due to e.g. pending signal. |
| 870 | * Otherwise, function returns one-past-the-last PFN of isolated page |
| 871 | * (which may be greater than end_pfn if end fell in a middle of a THP page). |
| 872 | */ |
| 873 | unsigned long |
| 874 | isolate_migratepages_range(struct compact_control *cc, unsigned long start_pfn, |
| 875 | unsigned long end_pfn) |
| 876 | { |
| 877 | unsigned long pfn, block_end_pfn; |
| 878 | |
| 879 | /* Scan block by block. First and last block may be incomplete */ |
| 880 | pfn = start_pfn; |
| 881 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); |
| 882 | |
| 883 | for (; pfn < end_pfn; pfn = block_end_pfn, |
| 884 | block_end_pfn += pageblock_nr_pages) { |
| 885 | |
| 886 | block_end_pfn = min(block_end_pfn, end_pfn); |
| 887 | |
| 888 | if (!pageblock_pfn_to_page(pfn, block_end_pfn, cc->zone)) |
| 889 | continue; |
| 890 | |
| 891 | pfn = isolate_migratepages_block(cc, pfn, block_end_pfn, |
| 892 | ISOLATE_UNEVICTABLE); |
| 893 | |
| 894 | /* |
| 895 | * In case of fatal failure, release everything that might |
| 896 | * have been isolated in the previous iteration, and signal |
| 897 | * the failure back to caller. |
| 898 | */ |
| 899 | if (!pfn) { |
| 900 | putback_movable_pages(&cc->migratepages); |
| 901 | cc->nr_migratepages = 0; |
| 902 | break; |
| 903 | } |
| 904 | |
| 905 | if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) |
| 906 | break; |
| 907 | } |
| 908 | acct_isolated(cc->zone, cc); |
| 909 | |
| 910 | return pfn; |
| 911 | } |
| 912 | |
| 913 | #endif /* CONFIG_COMPACTION || CONFIG_CMA */ |
| 914 | #ifdef CONFIG_COMPACTION |
| 915 | |
| 916 | /* Returns true if the page is within a block suitable for migration to */ |
| 917 | static bool suitable_migration_target(struct page *page) |
| 918 | { |
| 919 | /* If the page is a large free page, then disallow migration */ |
| 920 | if (PageBuddy(page)) { |
| 921 | /* |
| 922 | * We are checking page_order without zone->lock taken. But |
| 923 | * the only small danger is that we skip a potentially suitable |
| 924 | * pageblock, so it's not worth to check order for valid range. |
| 925 | */ |
| 926 | if (page_order_unsafe(page) >= pageblock_order) |
| 927 | return false; |
| 928 | } |
| 929 | |
| 930 | /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */ |
| 931 | if (migrate_async_suitable(get_pageblock_migratetype(page))) |
| 932 | return true; |
| 933 | |
| 934 | /* Otherwise skip the block */ |
| 935 | return false; |
| 936 | } |
| 937 | |
| 938 | /* |
| 939 | * Test whether the free scanner has reached the same or lower pageblock than |
| 940 | * the migration scanner, and compaction should thus terminate. |
| 941 | */ |
| 942 | static inline bool compact_scanners_met(struct compact_control *cc) |
| 943 | { |
| 944 | return (cc->free_pfn >> pageblock_order) |
| 945 | <= (cc->migrate_pfn >> pageblock_order); |
| 946 | } |
| 947 | |
| 948 | /* |
| 949 | * Based on information in the current compact_control, find blocks |
| 950 | * suitable for isolating free pages from and then isolate them. |
| 951 | */ |
| 952 | static void isolate_freepages(struct compact_control *cc) |
| 953 | { |
| 954 | struct zone *zone = cc->zone; |
| 955 | struct page *page; |
| 956 | unsigned long block_start_pfn; /* start of current pageblock */ |
| 957 | unsigned long isolate_start_pfn; /* exact pfn we start at */ |
| 958 | unsigned long block_end_pfn; /* end of current pageblock */ |
| 959 | unsigned long low_pfn; /* lowest pfn scanner is able to scan */ |
| 960 | struct list_head *freelist = &cc->freepages; |
| 961 | |
| 962 | /* |
| 963 | * Initialise the free scanner. The starting point is where we last |
| 964 | * successfully isolated from, zone-cached value, or the end of the |
| 965 | * zone when isolating for the first time. For looping we also need |
| 966 | * this pfn aligned down to the pageblock boundary, because we do |
| 967 | * block_start_pfn -= pageblock_nr_pages in the for loop. |
| 968 | * For ending point, take care when isolating in last pageblock of a |
| 969 | * a zone which ends in the middle of a pageblock. |
| 970 | * The low boundary is the end of the pageblock the migration scanner |
| 971 | * is using. |
| 972 | */ |
| 973 | isolate_start_pfn = cc->free_pfn; |
| 974 | block_start_pfn = cc->free_pfn & ~(pageblock_nr_pages-1); |
| 975 | block_end_pfn = min(block_start_pfn + pageblock_nr_pages, |
| 976 | zone_end_pfn(zone)); |
| 977 | low_pfn = ALIGN(cc->migrate_pfn + 1, pageblock_nr_pages); |
| 978 | |
| 979 | /* |
| 980 | * Isolate free pages until enough are available to migrate the |
| 981 | * pages on cc->migratepages. We stop searching if the migrate |
| 982 | * and free page scanners meet or enough free pages are isolated. |
| 983 | */ |
| 984 | for (; block_start_pfn >= low_pfn; |
| 985 | block_end_pfn = block_start_pfn, |
| 986 | block_start_pfn -= pageblock_nr_pages, |
| 987 | isolate_start_pfn = block_start_pfn) { |
| 988 | |
| 989 | /* |
| 990 | * This can iterate a massively long zone without finding any |
| 991 | * suitable migration targets, so periodically check if we need |
| 992 | * to schedule, or even abort async compaction. |
| 993 | */ |
| 994 | if (!(block_start_pfn % (SWAP_CLUSTER_MAX * pageblock_nr_pages)) |
| 995 | && compact_should_abort(cc)) |
| 996 | break; |
| 997 | |
| 998 | page = pageblock_pfn_to_page(block_start_pfn, block_end_pfn, |
| 999 | zone); |
| 1000 | if (!page) |
| 1001 | continue; |
| 1002 | |
| 1003 | /* Check the block is suitable for migration */ |
| 1004 | if (!suitable_migration_target(page)) |
| 1005 | continue; |
| 1006 | |
| 1007 | /* If isolation recently failed, do not retry */ |
| 1008 | if (!isolation_suitable(cc, page)) |
| 1009 | continue; |
| 1010 | |
| 1011 | /* Found a block suitable for isolating free pages from. */ |
| 1012 | isolate_freepages_block(cc, &isolate_start_pfn, |
| 1013 | block_end_pfn, freelist, false); |
| 1014 | |
| 1015 | /* |
| 1016 | * If we isolated enough freepages, or aborted due to async |
| 1017 | * compaction being contended, terminate the loop. |
| 1018 | * Remember where the free scanner should restart next time, |
| 1019 | * which is where isolate_freepages_block() left off. |
| 1020 | * But if it scanned the whole pageblock, isolate_start_pfn |
| 1021 | * now points at block_end_pfn, which is the start of the next |
| 1022 | * pageblock. |
| 1023 | * In that case we will however want to restart at the start |
| 1024 | * of the previous pageblock. |
| 1025 | */ |
| 1026 | if ((cc->nr_freepages >= cc->nr_migratepages) |
| 1027 | || cc->contended) { |
| 1028 | if (isolate_start_pfn >= block_end_pfn) |
| 1029 | isolate_start_pfn = |
| 1030 | block_start_pfn - pageblock_nr_pages; |
| 1031 | break; |
| 1032 | } else { |
| 1033 | /* |
| 1034 | * isolate_freepages_block() should not terminate |
| 1035 | * prematurely unless contended, or isolated enough |
| 1036 | */ |
| 1037 | VM_BUG_ON(isolate_start_pfn < block_end_pfn); |
| 1038 | } |
| 1039 | } |
| 1040 | |
| 1041 | /* split_free_page does not map the pages */ |
| 1042 | map_pages(freelist); |
| 1043 | |
| 1044 | /* |
| 1045 | * Record where the free scanner will restart next time. Either we |
| 1046 | * broke from the loop and set isolate_start_pfn based on the last |
| 1047 | * call to isolate_freepages_block(), or we met the migration scanner |
| 1048 | * and the loop terminated due to isolate_start_pfn < low_pfn |
| 1049 | */ |
| 1050 | cc->free_pfn = isolate_start_pfn; |
| 1051 | } |
| 1052 | |
| 1053 | /* |
| 1054 | * This is a migrate-callback that "allocates" freepages by taking pages |
| 1055 | * from the isolated freelists in the block we are migrating to. |
| 1056 | */ |
| 1057 | static struct page *compaction_alloc(struct page *migratepage, |
| 1058 | unsigned long data, |
| 1059 | int **result) |
| 1060 | { |
| 1061 | struct compact_control *cc = (struct compact_control *)data; |
| 1062 | struct page *freepage; |
| 1063 | |
| 1064 | /* |
| 1065 | * Isolate free pages if necessary, and if we are not aborting due to |
| 1066 | * contention. |
| 1067 | */ |
| 1068 | if (list_empty(&cc->freepages)) { |
| 1069 | if (!cc->contended) |
| 1070 | isolate_freepages(cc); |
| 1071 | |
| 1072 | if (list_empty(&cc->freepages)) |
| 1073 | return NULL; |
| 1074 | } |
| 1075 | |
| 1076 | freepage = list_entry(cc->freepages.next, struct page, lru); |
| 1077 | list_del(&freepage->lru); |
| 1078 | cc->nr_freepages--; |
| 1079 | |
| 1080 | return freepage; |
| 1081 | } |
| 1082 | |
| 1083 | /* |
| 1084 | * This is a migrate-callback that "frees" freepages back to the isolated |
| 1085 | * freelist. All pages on the freelist are from the same zone, so there is no |
| 1086 | * special handling needed for NUMA. |
| 1087 | */ |
| 1088 | static void compaction_free(struct page *page, unsigned long data) |
| 1089 | { |
| 1090 | struct compact_control *cc = (struct compact_control *)data; |
| 1091 | |
| 1092 | list_add(&page->lru, &cc->freepages); |
| 1093 | cc->nr_freepages++; |
| 1094 | } |
| 1095 | |
| 1096 | /* possible outcome of isolate_migratepages */ |
| 1097 | typedef enum { |
| 1098 | ISOLATE_ABORT, /* Abort compaction now */ |
| 1099 | ISOLATE_NONE, /* No pages isolated, continue scanning */ |
| 1100 | ISOLATE_SUCCESS, /* Pages isolated, migrate */ |
| 1101 | } isolate_migrate_t; |
| 1102 | |
| 1103 | /* |
| 1104 | * Allow userspace to control policy on scanning the unevictable LRU for |
| 1105 | * compactable pages. |
| 1106 | */ |
| 1107 | int sysctl_compact_unevictable_allowed __read_mostly = 1; |
| 1108 | |
| 1109 | /* |
| 1110 | * Isolate all pages that can be migrated from the first suitable block, |
| 1111 | * starting at the block pointed to by the migrate scanner pfn within |
| 1112 | * compact_control. |
| 1113 | */ |
| 1114 | static isolate_migrate_t isolate_migratepages(struct zone *zone, |
| 1115 | struct compact_control *cc) |
| 1116 | { |
| 1117 | unsigned long low_pfn, end_pfn; |
| 1118 | unsigned long isolate_start_pfn; |
| 1119 | struct page *page; |
| 1120 | const isolate_mode_t isolate_mode = |
| 1121 | (sysctl_compact_unevictable_allowed ? ISOLATE_UNEVICTABLE : 0) | |
| 1122 | (cc->mode == MIGRATE_ASYNC ? ISOLATE_ASYNC_MIGRATE : 0); |
| 1123 | |
| 1124 | /* |
| 1125 | * Start at where we last stopped, or beginning of the zone as |
| 1126 | * initialized by compact_zone() |
| 1127 | */ |
| 1128 | low_pfn = cc->migrate_pfn; |
| 1129 | |
| 1130 | /* Only scan within a pageblock boundary */ |
| 1131 | end_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages); |
| 1132 | |
| 1133 | /* |
| 1134 | * Iterate over whole pageblocks until we find the first suitable. |
| 1135 | * Do not cross the free scanner. |
| 1136 | */ |
| 1137 | for (; end_pfn <= cc->free_pfn; |
| 1138 | low_pfn = end_pfn, end_pfn += pageblock_nr_pages) { |
| 1139 | |
| 1140 | /* |
| 1141 | * This can potentially iterate a massively long zone with |
| 1142 | * many pageblocks unsuitable, so periodically check if we |
| 1143 | * need to schedule, or even abort async compaction. |
| 1144 | */ |
| 1145 | if (!(low_pfn % (SWAP_CLUSTER_MAX * pageblock_nr_pages)) |
| 1146 | && compact_should_abort(cc)) |
| 1147 | break; |
| 1148 | |
| 1149 | page = pageblock_pfn_to_page(low_pfn, end_pfn, zone); |
| 1150 | if (!page) |
| 1151 | continue; |
| 1152 | |
| 1153 | /* If isolation recently failed, do not retry */ |
| 1154 | if (!isolation_suitable(cc, page)) |
| 1155 | continue; |
| 1156 | |
| 1157 | /* |
| 1158 | * For async compaction, also only scan in MOVABLE blocks. |
| 1159 | * Async compaction is optimistic to see if the minimum amount |
| 1160 | * of work satisfies the allocation. |
| 1161 | */ |
| 1162 | if (cc->mode == MIGRATE_ASYNC && |
| 1163 | !migrate_async_suitable(get_pageblock_migratetype(page))) |
| 1164 | continue; |
| 1165 | |
| 1166 | /* Perform the isolation */ |
| 1167 | isolate_start_pfn = low_pfn; |
| 1168 | low_pfn = isolate_migratepages_block(cc, low_pfn, end_pfn, |
| 1169 | isolate_mode); |
| 1170 | |
| 1171 | if (!low_pfn || cc->contended) { |
| 1172 | acct_isolated(zone, cc); |
| 1173 | return ISOLATE_ABORT; |
| 1174 | } |
| 1175 | |
| 1176 | /* |
| 1177 | * Record where we could have freed pages by migration and not |
| 1178 | * yet flushed them to buddy allocator. |
| 1179 | * - this is the lowest page that could have been isolated and |
| 1180 | * then freed by migration. |
| 1181 | */ |
| 1182 | if (cc->nr_migratepages && !cc->last_migrated_pfn) |
| 1183 | cc->last_migrated_pfn = isolate_start_pfn; |
| 1184 | |
| 1185 | /* |
| 1186 | * Either we isolated something and proceed with migration. Or |
| 1187 | * we failed and compact_zone should decide if we should |
| 1188 | * continue or not. |
| 1189 | */ |
| 1190 | break; |
| 1191 | } |
| 1192 | |
| 1193 | acct_isolated(zone, cc); |
| 1194 | /* Record where migration scanner will be restarted. */ |
| 1195 | cc->migrate_pfn = low_pfn; |
| 1196 | |
| 1197 | return cc->nr_migratepages ? ISOLATE_SUCCESS : ISOLATE_NONE; |
| 1198 | } |
| 1199 | |
| 1200 | static int __compact_finished(struct zone *zone, struct compact_control *cc, |
| 1201 | const int migratetype) |
| 1202 | { |
| 1203 | unsigned int order; |
| 1204 | unsigned long watermark; |
| 1205 | |
| 1206 | if (cc->contended || fatal_signal_pending(current)) |
| 1207 | return COMPACT_PARTIAL; |
| 1208 | |
| 1209 | /* Compaction run completes if the migrate and free scanner meet */ |
| 1210 | if (compact_scanners_met(cc)) { |
| 1211 | /* Let the next compaction start anew. */ |
| 1212 | reset_cached_positions(zone); |
| 1213 | |
| 1214 | /* |
| 1215 | * Mark that the PG_migrate_skip information should be cleared |
| 1216 | * by kswapd when it goes to sleep. kswapd does not set the |
| 1217 | * flag itself as the decision to be clear should be directly |
| 1218 | * based on an allocation request. |
| 1219 | */ |
| 1220 | if (!current_is_kswapd()) |
| 1221 | zone->compact_blockskip_flush = true; |
| 1222 | |
| 1223 | return COMPACT_COMPLETE; |
| 1224 | } |
| 1225 | |
| 1226 | /* |
| 1227 | * order == -1 is expected when compacting via |
| 1228 | * /proc/sys/vm/compact_memory |
| 1229 | */ |
| 1230 | if (cc->order == -1) |
| 1231 | return COMPACT_CONTINUE; |
| 1232 | |
| 1233 | /* Compaction run is not finished if the watermark is not met */ |
| 1234 | watermark = low_wmark_pages(zone); |
| 1235 | |
| 1236 | if (!zone_watermark_ok(zone, cc->order, watermark, cc->classzone_idx, |
| 1237 | cc->alloc_flags)) |
| 1238 | return COMPACT_CONTINUE; |
| 1239 | |
| 1240 | /* Direct compactor: Is a suitable page free? */ |
| 1241 | for (order = cc->order; order < MAX_ORDER; order++) { |
| 1242 | struct free_area *area = &zone->free_area[order]; |
| 1243 | bool can_steal; |
| 1244 | |
| 1245 | /* Job done if page is free of the right migratetype */ |
| 1246 | if (!list_empty(&area->free_list[migratetype])) |
| 1247 | return COMPACT_PARTIAL; |
| 1248 | |
| 1249 | #ifdef CONFIG_CMA |
| 1250 | /* MIGRATE_MOVABLE can fallback on MIGRATE_CMA */ |
| 1251 | if (migratetype == MIGRATE_MOVABLE && |
| 1252 | !list_empty(&area->free_list[MIGRATE_CMA])) |
| 1253 | return COMPACT_PARTIAL; |
| 1254 | #endif |
| 1255 | /* |
| 1256 | * Job done if allocation would steal freepages from |
| 1257 | * other migratetype buddy lists. |
| 1258 | */ |
| 1259 | if (find_suitable_fallback(area, order, migratetype, |
| 1260 | true, &can_steal) != -1) |
| 1261 | return COMPACT_PARTIAL; |
| 1262 | } |
| 1263 | |
| 1264 | return COMPACT_NO_SUITABLE_PAGE; |
| 1265 | } |
| 1266 | |
| 1267 | static int compact_finished(struct zone *zone, struct compact_control *cc, |
| 1268 | const int migratetype) |
| 1269 | { |
| 1270 | int ret; |
| 1271 | |
| 1272 | ret = __compact_finished(zone, cc, migratetype); |
| 1273 | trace_mm_compaction_finished(zone, cc->order, ret); |
| 1274 | if (ret == COMPACT_NO_SUITABLE_PAGE) |
| 1275 | ret = COMPACT_CONTINUE; |
| 1276 | |
| 1277 | return ret; |
| 1278 | } |
| 1279 | |
| 1280 | /* |
| 1281 | * compaction_suitable: Is this suitable to run compaction on this zone now? |
| 1282 | * Returns |
| 1283 | * COMPACT_SKIPPED - If there are too few free pages for compaction |
| 1284 | * COMPACT_PARTIAL - If the allocation would succeed without compaction |
| 1285 | * COMPACT_CONTINUE - If compaction should run now |
| 1286 | */ |
| 1287 | static unsigned long __compaction_suitable(struct zone *zone, int order, |
| 1288 | int alloc_flags, int classzone_idx) |
| 1289 | { |
| 1290 | int fragindex; |
| 1291 | unsigned long watermark; |
| 1292 | |
| 1293 | /* |
| 1294 | * order == -1 is expected when compacting via |
| 1295 | * /proc/sys/vm/compact_memory |
| 1296 | */ |
| 1297 | if (order == -1) |
| 1298 | return COMPACT_CONTINUE; |
| 1299 | |
| 1300 | watermark = low_wmark_pages(zone); |
| 1301 | /* |
| 1302 | * If watermarks for high-order allocation are already met, there |
| 1303 | * should be no need for compaction at all. |
| 1304 | */ |
| 1305 | if (zone_watermark_ok(zone, order, watermark, classzone_idx, |
| 1306 | alloc_flags)) |
| 1307 | return COMPACT_PARTIAL; |
| 1308 | |
| 1309 | /* |
| 1310 | * Watermarks for order-0 must be met for compaction. Note the 2UL. |
| 1311 | * This is because during migration, copies of pages need to be |
| 1312 | * allocated and for a short time, the footprint is higher |
| 1313 | */ |
| 1314 | watermark += (2UL << order); |
| 1315 | if (!zone_watermark_ok(zone, 0, watermark, classzone_idx, alloc_flags)) |
| 1316 | return COMPACT_SKIPPED; |
| 1317 | |
| 1318 | /* |
| 1319 | * fragmentation index determines if allocation failures are due to |
| 1320 | * low memory or external fragmentation |
| 1321 | * |
| 1322 | * index of -1000 would imply allocations might succeed depending on |
| 1323 | * watermarks, but we already failed the high-order watermark check |
| 1324 | * index towards 0 implies failure is due to lack of memory |
| 1325 | * index towards 1000 implies failure is due to fragmentation |
| 1326 | * |
| 1327 | * Only compact if a failure would be due to fragmentation. |
| 1328 | */ |
| 1329 | fragindex = fragmentation_index(zone, order); |
| 1330 | if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold) |
| 1331 | return COMPACT_NOT_SUITABLE_ZONE; |
| 1332 | |
| 1333 | return COMPACT_CONTINUE; |
| 1334 | } |
| 1335 | |
| 1336 | unsigned long compaction_suitable(struct zone *zone, int order, |
| 1337 | int alloc_flags, int classzone_idx) |
| 1338 | { |
| 1339 | unsigned long ret; |
| 1340 | |
| 1341 | ret = __compaction_suitable(zone, order, alloc_flags, classzone_idx); |
| 1342 | trace_mm_compaction_suitable(zone, order, ret); |
| 1343 | if (ret == COMPACT_NOT_SUITABLE_ZONE) |
| 1344 | ret = COMPACT_SKIPPED; |
| 1345 | |
| 1346 | return ret; |
| 1347 | } |
| 1348 | |
| 1349 | static int compact_zone(struct zone *zone, struct compact_control *cc) |
| 1350 | { |
| 1351 | int ret; |
| 1352 | unsigned long start_pfn = zone->zone_start_pfn; |
| 1353 | unsigned long end_pfn = zone_end_pfn(zone); |
| 1354 | const int migratetype = gfpflags_to_migratetype(cc->gfp_mask); |
| 1355 | const bool sync = cc->mode != MIGRATE_ASYNC; |
| 1356 | |
| 1357 | ret = compaction_suitable(zone, cc->order, cc->alloc_flags, |
| 1358 | cc->classzone_idx); |
| 1359 | switch (ret) { |
| 1360 | case COMPACT_PARTIAL: |
| 1361 | case COMPACT_SKIPPED: |
| 1362 | /* Compaction is likely to fail */ |
| 1363 | return ret; |
| 1364 | case COMPACT_CONTINUE: |
| 1365 | /* Fall through to compaction */ |
| 1366 | ; |
| 1367 | } |
| 1368 | |
| 1369 | /* |
| 1370 | * Clear pageblock skip if there were failures recently and compaction |
| 1371 | * is about to be retried after being deferred. kswapd does not do |
| 1372 | * this reset as it'll reset the cached information when going to sleep. |
| 1373 | */ |
| 1374 | if (compaction_restarting(zone, cc->order) && !current_is_kswapd()) |
| 1375 | __reset_isolation_suitable(zone); |
| 1376 | |
| 1377 | /* |
| 1378 | * Setup to move all movable pages to the end of the zone. Used cached |
| 1379 | * information on where the scanners should start but check that it |
| 1380 | * is initialised by ensuring the values are within zone boundaries. |
| 1381 | */ |
| 1382 | cc->migrate_pfn = zone->compact_cached_migrate_pfn[sync]; |
| 1383 | cc->free_pfn = zone->compact_cached_free_pfn; |
| 1384 | if (cc->free_pfn < start_pfn || cc->free_pfn > end_pfn) { |
| 1385 | cc->free_pfn = end_pfn & ~(pageblock_nr_pages-1); |
| 1386 | zone->compact_cached_free_pfn = cc->free_pfn; |
| 1387 | } |
| 1388 | if (cc->migrate_pfn < start_pfn || cc->migrate_pfn > end_pfn) { |
| 1389 | cc->migrate_pfn = start_pfn; |
| 1390 | zone->compact_cached_migrate_pfn[0] = cc->migrate_pfn; |
| 1391 | zone->compact_cached_migrate_pfn[1] = cc->migrate_pfn; |
| 1392 | } |
| 1393 | cc->last_migrated_pfn = 0; |
| 1394 | |
| 1395 | trace_mm_compaction_begin(start_pfn, cc->migrate_pfn, |
| 1396 | cc->free_pfn, end_pfn, sync); |
| 1397 | |
| 1398 | migrate_prep_local(); |
| 1399 | |
| 1400 | while ((ret = compact_finished(zone, cc, migratetype)) == |
| 1401 | COMPACT_CONTINUE) { |
| 1402 | int err; |
| 1403 | |
| 1404 | switch (isolate_migratepages(zone, cc)) { |
| 1405 | case ISOLATE_ABORT: |
| 1406 | ret = COMPACT_PARTIAL; |
| 1407 | putback_movable_pages(&cc->migratepages); |
| 1408 | cc->nr_migratepages = 0; |
| 1409 | goto out; |
| 1410 | case ISOLATE_NONE: |
| 1411 | /* |
| 1412 | * We haven't isolated and migrated anything, but |
| 1413 | * there might still be unflushed migrations from |
| 1414 | * previous cc->order aligned block. |
| 1415 | */ |
| 1416 | goto check_drain; |
| 1417 | case ISOLATE_SUCCESS: |
| 1418 | ; |
| 1419 | } |
| 1420 | |
| 1421 | err = migrate_pages(&cc->migratepages, compaction_alloc, |
| 1422 | compaction_free, (unsigned long)cc, cc->mode, |
| 1423 | MR_COMPACTION); |
| 1424 | |
| 1425 | trace_mm_compaction_migratepages(cc->nr_migratepages, err, |
| 1426 | &cc->migratepages); |
| 1427 | |
| 1428 | /* All pages were either migrated or will be released */ |
| 1429 | cc->nr_migratepages = 0; |
| 1430 | if (err) { |
| 1431 | putback_movable_pages(&cc->migratepages); |
| 1432 | /* |
| 1433 | * migrate_pages() may return -ENOMEM when scanners meet |
| 1434 | * and we want compact_finished() to detect it |
| 1435 | */ |
| 1436 | if (err == -ENOMEM && !compact_scanners_met(cc)) { |
| 1437 | ret = COMPACT_PARTIAL; |
| 1438 | goto out; |
| 1439 | } |
| 1440 | } |
| 1441 | |
| 1442 | check_drain: |
| 1443 | /* |
| 1444 | * Has the migration scanner moved away from the previous |
| 1445 | * cc->order aligned block where we migrated from? If yes, |
| 1446 | * flush the pages that were freed, so that they can merge and |
| 1447 | * compact_finished() can detect immediately if allocation |
| 1448 | * would succeed. |
| 1449 | */ |
| 1450 | if (cc->order > 0 && cc->last_migrated_pfn) { |
| 1451 | int cpu; |
| 1452 | unsigned long current_block_start = |
| 1453 | cc->migrate_pfn & ~((1UL << cc->order) - 1); |
| 1454 | |
| 1455 | if (cc->last_migrated_pfn < current_block_start) { |
| 1456 | cpu = get_cpu(); |
| 1457 | lru_add_drain_cpu(cpu); |
| 1458 | drain_local_pages(zone); |
| 1459 | put_cpu(); |
| 1460 | /* No more flushing until we migrate again */ |
| 1461 | cc->last_migrated_pfn = 0; |
| 1462 | } |
| 1463 | } |
| 1464 | |
| 1465 | } |
| 1466 | |
| 1467 | out: |
| 1468 | /* |
| 1469 | * Release free pages and update where the free scanner should restart, |
| 1470 | * so we don't leave any returned pages behind in the next attempt. |
| 1471 | */ |
| 1472 | if (cc->nr_freepages > 0) { |
| 1473 | unsigned long free_pfn = release_freepages(&cc->freepages); |
| 1474 | |
| 1475 | cc->nr_freepages = 0; |
| 1476 | VM_BUG_ON(free_pfn == 0); |
| 1477 | /* The cached pfn is always the first in a pageblock */ |
| 1478 | free_pfn &= ~(pageblock_nr_pages-1); |
| 1479 | /* |
| 1480 | * Only go back, not forward. The cached pfn might have been |
| 1481 | * already reset to zone end in compact_finished() |
| 1482 | */ |
| 1483 | if (free_pfn > zone->compact_cached_free_pfn) |
| 1484 | zone->compact_cached_free_pfn = free_pfn; |
| 1485 | } |
| 1486 | |
| 1487 | trace_mm_compaction_end(start_pfn, cc->migrate_pfn, |
| 1488 | cc->free_pfn, end_pfn, sync, ret); |
| 1489 | |
| 1490 | return ret; |
| 1491 | } |
| 1492 | |
| 1493 | static unsigned long compact_zone_order(struct zone *zone, int order, |
| 1494 | gfp_t gfp_mask, enum migrate_mode mode, int *contended, |
| 1495 | int alloc_flags, int classzone_idx) |
| 1496 | { |
| 1497 | unsigned long ret; |
| 1498 | struct compact_control cc = { |
| 1499 | .nr_freepages = 0, |
| 1500 | .nr_migratepages = 0, |
| 1501 | .order = order, |
| 1502 | .gfp_mask = gfp_mask, |
| 1503 | .zone = zone, |
| 1504 | .mode = mode, |
| 1505 | .alloc_flags = alloc_flags, |
| 1506 | .classzone_idx = classzone_idx, |
| 1507 | }; |
| 1508 | INIT_LIST_HEAD(&cc.freepages); |
| 1509 | INIT_LIST_HEAD(&cc.migratepages); |
| 1510 | |
| 1511 | ret = compact_zone(zone, &cc); |
| 1512 | |
| 1513 | VM_BUG_ON(!list_empty(&cc.freepages)); |
| 1514 | VM_BUG_ON(!list_empty(&cc.migratepages)); |
| 1515 | |
| 1516 | *contended = cc.contended; |
| 1517 | return ret; |
| 1518 | } |
| 1519 | |
| 1520 | int sysctl_extfrag_threshold = 500; |
| 1521 | |
| 1522 | /** |
| 1523 | * try_to_compact_pages - Direct compact to satisfy a high-order allocation |
| 1524 | * @gfp_mask: The GFP mask of the current allocation |
| 1525 | * @order: The order of the current allocation |
| 1526 | * @alloc_flags: The allocation flags of the current allocation |
| 1527 | * @ac: The context of current allocation |
| 1528 | * @mode: The migration mode for async, sync light, or sync migration |
| 1529 | * @contended: Return value that determines if compaction was aborted due to |
| 1530 | * need_resched() or lock contention |
| 1531 | * |
| 1532 | * This is the main entry point for direct page compaction. |
| 1533 | */ |
| 1534 | unsigned long try_to_compact_pages(gfp_t gfp_mask, unsigned int order, |
| 1535 | int alloc_flags, const struct alloc_context *ac, |
| 1536 | enum migrate_mode mode, int *contended) |
| 1537 | { |
| 1538 | int may_enter_fs = gfp_mask & __GFP_FS; |
| 1539 | int may_perform_io = gfp_mask & __GFP_IO; |
| 1540 | struct zoneref *z; |
| 1541 | struct zone *zone; |
| 1542 | int rc = COMPACT_DEFERRED; |
| 1543 | int all_zones_contended = COMPACT_CONTENDED_LOCK; /* init for &= op */ |
| 1544 | |
| 1545 | *contended = COMPACT_CONTENDED_NONE; |
| 1546 | |
| 1547 | /* Check if the GFP flags allow compaction */ |
| 1548 | if (!order || !may_enter_fs || !may_perform_io) |
| 1549 | return COMPACT_SKIPPED; |
| 1550 | |
| 1551 | trace_mm_compaction_try_to_compact_pages(order, gfp_mask, mode); |
| 1552 | |
| 1553 | /* Compact each zone in the list */ |
| 1554 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
| 1555 | ac->nodemask) { |
| 1556 | int status; |
| 1557 | int zone_contended; |
| 1558 | |
| 1559 | if (compaction_deferred(zone, order)) |
| 1560 | continue; |
| 1561 | |
| 1562 | status = compact_zone_order(zone, order, gfp_mask, mode, |
| 1563 | &zone_contended, alloc_flags, |
| 1564 | ac->classzone_idx); |
| 1565 | rc = max(status, rc); |
| 1566 | /* |
| 1567 | * It takes at least one zone that wasn't lock contended |
| 1568 | * to clear all_zones_contended. |
| 1569 | */ |
| 1570 | all_zones_contended &= zone_contended; |
| 1571 | |
| 1572 | /* If a normal allocation would succeed, stop compacting */ |
| 1573 | if (zone_watermark_ok(zone, order, low_wmark_pages(zone), |
| 1574 | ac->classzone_idx, alloc_flags)) { |
| 1575 | /* |
| 1576 | * We think the allocation will succeed in this zone, |
| 1577 | * but it is not certain, hence the false. The caller |
| 1578 | * will repeat this with true if allocation indeed |
| 1579 | * succeeds in this zone. |
| 1580 | */ |
| 1581 | compaction_defer_reset(zone, order, false); |
| 1582 | /* |
| 1583 | * It is possible that async compaction aborted due to |
| 1584 | * need_resched() and the watermarks were ok thanks to |
| 1585 | * somebody else freeing memory. The allocation can |
| 1586 | * however still fail so we better signal the |
| 1587 | * need_resched() contention anyway (this will not |
| 1588 | * prevent the allocation attempt). |
| 1589 | */ |
| 1590 | if (zone_contended == COMPACT_CONTENDED_SCHED) |
| 1591 | *contended = COMPACT_CONTENDED_SCHED; |
| 1592 | |
| 1593 | goto break_loop; |
| 1594 | } |
| 1595 | |
| 1596 | if (mode != MIGRATE_ASYNC && status == COMPACT_COMPLETE) { |
| 1597 | /* |
| 1598 | * We think that allocation won't succeed in this zone |
| 1599 | * so we defer compaction there. If it ends up |
| 1600 | * succeeding after all, it will be reset. |
| 1601 | */ |
| 1602 | defer_compaction(zone, order); |
| 1603 | } |
| 1604 | |
| 1605 | /* |
| 1606 | * We might have stopped compacting due to need_resched() in |
| 1607 | * async compaction, or due to a fatal signal detected. In that |
| 1608 | * case do not try further zones and signal need_resched() |
| 1609 | * contention. |
| 1610 | */ |
| 1611 | if ((zone_contended == COMPACT_CONTENDED_SCHED) |
| 1612 | || fatal_signal_pending(current)) { |
| 1613 | *contended = COMPACT_CONTENDED_SCHED; |
| 1614 | goto break_loop; |
| 1615 | } |
| 1616 | |
| 1617 | continue; |
| 1618 | break_loop: |
| 1619 | /* |
| 1620 | * We might not have tried all the zones, so be conservative |
| 1621 | * and assume they are not all lock contended. |
| 1622 | */ |
| 1623 | all_zones_contended = 0; |
| 1624 | break; |
| 1625 | } |
| 1626 | |
| 1627 | /* |
| 1628 | * If at least one zone wasn't deferred or skipped, we report if all |
| 1629 | * zones that were tried were lock contended. |
| 1630 | */ |
| 1631 | if (rc > COMPACT_SKIPPED && all_zones_contended) |
| 1632 | *contended = COMPACT_CONTENDED_LOCK; |
| 1633 | |
| 1634 | return rc; |
| 1635 | } |
| 1636 | |
| 1637 | |
| 1638 | /* Compact all zones within a node */ |
| 1639 | static void __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc) |
| 1640 | { |
| 1641 | int zoneid; |
| 1642 | struct zone *zone; |
| 1643 | |
| 1644 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { |
| 1645 | |
| 1646 | zone = &pgdat->node_zones[zoneid]; |
| 1647 | if (!populated_zone(zone)) |
| 1648 | continue; |
| 1649 | |
| 1650 | cc->nr_freepages = 0; |
| 1651 | cc->nr_migratepages = 0; |
| 1652 | cc->zone = zone; |
| 1653 | INIT_LIST_HEAD(&cc->freepages); |
| 1654 | INIT_LIST_HEAD(&cc->migratepages); |
| 1655 | |
| 1656 | /* |
| 1657 | * When called via /proc/sys/vm/compact_memory |
| 1658 | * this makes sure we compact the whole zone regardless of |
| 1659 | * cached scanner positions. |
| 1660 | */ |
| 1661 | if (cc->order == -1) |
| 1662 | __reset_isolation_suitable(zone); |
| 1663 | |
| 1664 | if (cc->order == -1 || !compaction_deferred(zone, cc->order)) |
| 1665 | compact_zone(zone, cc); |
| 1666 | |
| 1667 | if (cc->order > 0) { |
| 1668 | if (zone_watermark_ok(zone, cc->order, |
| 1669 | low_wmark_pages(zone), 0, 0)) |
| 1670 | compaction_defer_reset(zone, cc->order, false); |
| 1671 | } |
| 1672 | |
| 1673 | VM_BUG_ON(!list_empty(&cc->freepages)); |
| 1674 | VM_BUG_ON(!list_empty(&cc->migratepages)); |
| 1675 | } |
| 1676 | } |
| 1677 | |
| 1678 | void compact_pgdat(pg_data_t *pgdat, int order) |
| 1679 | { |
| 1680 | struct compact_control cc = { |
| 1681 | .order = order, |
| 1682 | .mode = MIGRATE_ASYNC, |
| 1683 | }; |
| 1684 | |
| 1685 | if (!order) |
| 1686 | return; |
| 1687 | |
| 1688 | __compact_pgdat(pgdat, &cc); |
| 1689 | } |
| 1690 | |
| 1691 | static void compact_node(int nid) |
| 1692 | { |
| 1693 | struct compact_control cc = { |
| 1694 | .order = -1, |
| 1695 | .mode = MIGRATE_SYNC, |
| 1696 | .ignore_skip_hint = true, |
| 1697 | }; |
| 1698 | |
| 1699 | __compact_pgdat(NODE_DATA(nid), &cc); |
| 1700 | } |
| 1701 | |
| 1702 | /* Compact all nodes in the system */ |
| 1703 | static void compact_nodes(void) |
| 1704 | { |
| 1705 | int nid; |
| 1706 | |
| 1707 | /* Flush pending updates to the LRU lists */ |
| 1708 | lru_add_drain_all(); |
| 1709 | |
| 1710 | for_each_online_node(nid) |
| 1711 | compact_node(nid); |
| 1712 | } |
| 1713 | |
| 1714 | /* The written value is actually unused, all memory is compacted */ |
| 1715 | int sysctl_compact_memory; |
| 1716 | |
| 1717 | /* This is the entry point for compacting all nodes via /proc/sys/vm */ |
| 1718 | int sysctl_compaction_handler(struct ctl_table *table, int write, |
| 1719 | void __user *buffer, size_t *length, loff_t *ppos) |
| 1720 | { |
| 1721 | if (write) |
| 1722 | compact_nodes(); |
| 1723 | |
| 1724 | return 0; |
| 1725 | } |
| 1726 | |
| 1727 | int sysctl_extfrag_handler(struct ctl_table *table, int write, |
| 1728 | void __user *buffer, size_t *length, loff_t *ppos) |
| 1729 | { |
| 1730 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
| 1731 | |
| 1732 | return 0; |
| 1733 | } |
| 1734 | |
| 1735 | #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA) |
| 1736 | static ssize_t sysfs_compact_node(struct device *dev, |
| 1737 | struct device_attribute *attr, |
| 1738 | const char *buf, size_t count) |
| 1739 | { |
| 1740 | int nid = dev->id; |
| 1741 | |
| 1742 | if (nid >= 0 && nid < nr_node_ids && node_online(nid)) { |
| 1743 | /* Flush pending updates to the LRU lists */ |
| 1744 | lru_add_drain_all(); |
| 1745 | |
| 1746 | compact_node(nid); |
| 1747 | } |
| 1748 | |
| 1749 | return count; |
| 1750 | } |
| 1751 | static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node); |
| 1752 | |
| 1753 | int compaction_register_node(struct node *node) |
| 1754 | { |
| 1755 | return device_create_file(&node->dev, &dev_attr_compact); |
| 1756 | } |
| 1757 | |
| 1758 | void compaction_unregister_node(struct node *node) |
| 1759 | { |
| 1760 | return device_remove_file(&node->dev, &dev_attr_compact); |
| 1761 | } |
| 1762 | #endif /* CONFIG_SYSFS && CONFIG_NUMA */ |
| 1763 | |
| 1764 | #endif /* CONFIG_COMPACTION */ |