linux/mm/compaction.c
<<
>>
Prefs
   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 "internal.h"
  20
  21#ifdef CONFIG_COMPACTION
  22static inline void count_compact_event(enum vm_event_item item)
  23{
  24        count_vm_event(item);
  25}
  26
  27static inline void count_compact_events(enum vm_event_item item, long delta)
  28{
  29        count_vm_events(item, delta);
  30}
  31#else
  32#define count_compact_event(item) do { } while (0)
  33#define count_compact_events(item, delta) do { } while (0)
  34#endif
  35
  36#if defined CONFIG_COMPACTION || defined CONFIG_CMA
  37
  38#define CREATE_TRACE_POINTS
  39#include <trace/events/compaction.h>
  40
  41static unsigned long release_freepages(struct list_head *freelist)
  42{
  43        struct page *page, *next;
  44        unsigned long count = 0;
  45
  46        list_for_each_entry_safe(page, next, freelist, lru) {
  47                list_del(&page->lru);
  48                __free_page(page);
  49                count++;
  50        }
  51
  52        return count;
  53}
  54
  55static void map_pages(struct list_head *list)
  56{
  57        struct page *page;
  58
  59        list_for_each_entry(page, list, lru) {
  60                arch_alloc_page(page, 0);
  61                kernel_map_pages(page, 1, 1);
  62        }
  63}
  64
  65static inline bool migrate_async_suitable(int migratetype)
  66{
  67        return is_migrate_cma(migratetype) || migratetype == MIGRATE_MOVABLE;
  68}
  69
  70#ifdef CONFIG_COMPACTION
  71/* Returns true if the pageblock should be scanned for pages to isolate. */
  72static inline bool isolation_suitable(struct compact_control *cc,
  73                                        struct page *page)
  74{
  75        if (cc->ignore_skip_hint)
  76                return true;
  77
  78        return !get_pageblock_skip(page);
  79}
  80
  81/*
  82 * This function is called to clear all cached information on pageblocks that
  83 * should be skipped for page isolation when the migrate and free page scanner
  84 * meet.
  85 */
  86static void __reset_isolation_suitable(struct zone *zone)
  87{
  88        unsigned long start_pfn = zone->zone_start_pfn;
  89        unsigned long end_pfn = zone_end_pfn(zone);
  90        unsigned long pfn;
  91
  92        zone->compact_cached_migrate_pfn = start_pfn;
  93        zone->compact_cached_free_pfn = end_pfn;
  94        zone->compact_blockskip_flush = false;
  95
  96        /* Walk the zone and mark every pageblock as suitable for isolation */
  97        for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
  98                struct page *page;
  99
 100                cond_resched();
 101
 102                if (!pfn_valid(pfn))
 103                        continue;
 104
 105                page = pfn_to_page(pfn);
 106                if (zone != page_zone(page))
 107                        continue;
 108
 109                clear_pageblock_skip(page);
 110        }
 111}
 112
 113void reset_isolation_suitable(pg_data_t *pgdat)
 114{
 115        int zoneid;
 116
 117        for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) {
 118                struct zone *zone = &pgdat->node_zones[zoneid];
 119                if (!populated_zone(zone))
 120                        continue;
 121
 122                /* Only flush if a full compaction finished recently */
 123                if (zone->compact_blockskip_flush)
 124                        __reset_isolation_suitable(zone);
 125        }
 126}
 127
 128/*
 129 * If no pages were isolated then mark this pageblock to be skipped in the
 130 * future. The information is later cleared by __reset_isolation_suitable().
 131 */
 132static void update_pageblock_skip(struct compact_control *cc,
 133                        struct page *page, unsigned long nr_isolated,
 134                        bool migrate_scanner)
 135{
 136        struct zone *zone = cc->zone;
 137        if (!page)
 138                return;
 139
 140        if (!nr_isolated) {
 141                unsigned long pfn = page_to_pfn(page);
 142                set_pageblock_skip(page);
 143
 144                /* Update where compaction should restart */
 145                if (migrate_scanner) {
 146                        if (!cc->finished_update_migrate &&
 147                            pfn > zone->compact_cached_migrate_pfn)
 148                                zone->compact_cached_migrate_pfn = pfn;
 149                } else {
 150                        if (!cc->finished_update_free &&
 151                            pfn < zone->compact_cached_free_pfn)
 152                                zone->compact_cached_free_pfn = pfn;
 153                }
 154        }
 155}
 156#else
 157static inline bool isolation_suitable(struct compact_control *cc,
 158                                        struct page *page)
 159{
 160        return true;
 161}
 162
 163static void update_pageblock_skip(struct compact_control *cc,
 164                        struct page *page, unsigned long nr_isolated,
 165                        bool migrate_scanner)
 166{
 167}
 168#endif /* CONFIG_COMPACTION */
 169
 170static inline bool should_release_lock(spinlock_t *lock)
 171{
 172        return need_resched() || spin_is_contended(lock);
 173}
 174
 175/*
 176 * Compaction requires the taking of some coarse locks that are potentially
 177 * very heavily contended. Check if the process needs to be scheduled or
 178 * if the lock is contended. For async compaction, back out in the event
 179 * if contention is severe. For sync compaction, schedule.
 180 *
 181 * Returns true if the lock is held.
 182 * Returns false if the lock is released and compaction should abort
 183 */
 184static bool compact_checklock_irqsave(spinlock_t *lock, unsigned long *flags,
 185                                      bool locked, struct compact_control *cc)
 186{
 187        if (should_release_lock(lock)) {
 188                if (locked) {
 189                        spin_unlock_irqrestore(lock, *flags);
 190                        locked = false;
 191                }
 192
 193                /* async aborts if taking too long or contended */
 194                if (!cc->sync) {
 195                        cc->contended = true;
 196                        return false;
 197                }
 198
 199                cond_resched();
 200        }
 201
 202        if (!locked)
 203                spin_lock_irqsave(lock, *flags);
 204        return true;
 205}
 206
 207static inline bool compact_trylock_irqsave(spinlock_t *lock,
 208                        unsigned long *flags, struct compact_control *cc)
 209{
 210        return compact_checklock_irqsave(lock, flags, false, cc);
 211}
 212
 213/* Returns true if the page is within a block suitable for migration to */
 214static bool suitable_migration_target(struct page *page)
 215{
 216        int migratetype = get_pageblock_migratetype(page);
 217
 218        /* Don't interfere with memory hot-remove or the min_free_kbytes blocks */
 219        if (migratetype == MIGRATE_RESERVE)
 220                return false;
 221
 222        if (is_migrate_isolate(migratetype))
 223                return false;
 224
 225        /* If the page is a large free page, then allow migration */
 226        if (PageBuddy(page) && page_order(page) >= pageblock_order)
 227                return true;
 228
 229        /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
 230        if (migrate_async_suitable(migratetype))
 231                return true;
 232
 233        /* Otherwise skip the block */
 234        return false;
 235}
 236
 237/*
 238 * Isolate free pages onto a private freelist. Caller must hold zone->lock.
 239 * If @strict is true, will abort returning 0 on any invalid PFNs or non-free
 240 * pages inside of the pageblock (even though it may still end up isolating
 241 * some pages).
 242 */
 243static unsigned long isolate_freepages_block(struct compact_control *cc,
 244                                unsigned long blockpfn,
 245                                unsigned long end_pfn,
 246                                struct list_head *freelist,
 247                                bool strict)
 248{
 249        int nr_scanned = 0, total_isolated = 0;
 250        struct page *cursor, *valid_page = NULL;
 251        unsigned long nr_strict_required = end_pfn - blockpfn;
 252        unsigned long flags;
 253        bool locked = false;
 254
 255        cursor = pfn_to_page(blockpfn);
 256
 257        /* Isolate free pages. */
 258        for (; blockpfn < end_pfn; blockpfn++, cursor++) {
 259                int isolated, i;
 260                struct page *page = cursor;
 261
 262                nr_scanned++;
 263                if (!pfn_valid_within(blockpfn))
 264                        continue;
 265                if (!valid_page)
 266                        valid_page = page;
 267                if (!PageBuddy(page))
 268                        continue;
 269
 270                /*
 271                 * The zone lock must be held to isolate freepages.
 272                 * Unfortunately this is a very coarse lock and can be
 273                 * heavily contended if there are parallel allocations
 274                 * or parallel compactions. For async compaction do not
 275                 * spin on the lock and we acquire the lock as late as
 276                 * possible.
 277                 */
 278                locked = compact_checklock_irqsave(&cc->zone->lock, &flags,
 279                                                                locked, cc);
 280                if (!locked)
 281                        break;
 282
 283                /* Recheck this is a suitable migration target under lock */
 284                if (!strict && !suitable_migration_target(page))
 285                        break;
 286
 287                /* Recheck this is a buddy page under lock */
 288                if (!PageBuddy(page))
 289                        continue;
 290
 291                /* Found a free page, break it into order-0 pages */
 292                isolated = split_free_page(page);
 293                if (!isolated && strict)
 294                        break;
 295                total_isolated += isolated;
 296                for (i = 0; i < isolated; i++) {
 297                        list_add(&page->lru, freelist);
 298                        page++;
 299                }
 300
 301                /* If a page was split, advance to the end of it */
 302                if (isolated) {
 303                        blockpfn += isolated - 1;
 304                        cursor += isolated - 1;
 305                }
 306        }
 307
 308        trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated);
 309
 310        /*
 311         * If strict isolation is requested by CMA then check that all the
 312         * pages requested were isolated. If there were any failures, 0 is
 313         * returned and CMA will fail.
 314         */
 315        if (strict && nr_strict_required > total_isolated)
 316                total_isolated = 0;
 317
 318        if (locked)
 319                spin_unlock_irqrestore(&cc->zone->lock, flags);
 320
 321        /* Update the pageblock-skip if the whole pageblock was scanned */
 322        if (blockpfn == end_pfn)
 323                update_pageblock_skip(cc, valid_page, total_isolated, false);
 324
 325        count_compact_events(COMPACTFREE_SCANNED, nr_scanned);
 326        if (total_isolated)
 327                count_compact_events(COMPACTISOLATED, total_isolated);
 328        return total_isolated;
 329}
 330
 331/**
 332 * isolate_freepages_range() - isolate free pages.
 333 * @start_pfn: The first PFN to start isolating.
 334 * @end_pfn:   The one-past-last PFN.
 335 *
 336 * Non-free pages, invalid PFNs, or zone boundaries within the
 337 * [start_pfn, end_pfn) range are considered errors, cause function to
 338 * undo its actions and return zero.
 339 *
 340 * Otherwise, function returns one-past-the-last PFN of isolated page
 341 * (which may be greater then end_pfn if end fell in a middle of
 342 * a free page).
 343 */
 344unsigned long
 345isolate_freepages_range(struct compact_control *cc,
 346                        unsigned long start_pfn, unsigned long end_pfn)
 347{
 348        unsigned long isolated, pfn, block_end_pfn;
 349        LIST_HEAD(freelist);
 350
 351        for (pfn = start_pfn; pfn < end_pfn; pfn += isolated) {
 352                if (!pfn_valid(pfn) || cc->zone != page_zone(pfn_to_page(pfn)))
 353                        break;
 354
 355                /*
 356                 * On subsequent iterations ALIGN() is actually not needed,
 357                 * but we keep it that we not to complicate the code.
 358                 */
 359                block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
 360                block_end_pfn = min(block_end_pfn, end_pfn);
 361
 362                isolated = isolate_freepages_block(cc, pfn, block_end_pfn,
 363                                                   &freelist, true);
 364
 365                /*
 366                 * In strict mode, isolate_freepages_block() returns 0 if
 367                 * there are any holes in the block (ie. invalid PFNs or
 368                 * non-free pages).
 369                 */
 370                if (!isolated)
 371                        break;
 372
 373                /*
 374                 * If we managed to isolate pages, it is always (1 << n) *
 375                 * pageblock_nr_pages for some non-negative n.  (Max order
 376                 * page may span two pageblocks).
 377                 */
 378        }
 379
 380        /* split_free_page does not map the pages */
 381        map_pages(&freelist);
 382
 383        if (pfn < end_pfn) {
 384                /* Loop terminated early, cleanup. */
 385                release_freepages(&freelist);
 386                return 0;
 387        }
 388
 389        /* We don't use freelists for anything. */
 390        return pfn;
 391}
 392
 393/* Update the number of anon and file isolated pages in the zone */
 394static void acct_isolated(struct zone *zone, bool locked, struct compact_control *cc)
 395{
 396        struct page *page;
 397        unsigned int count[2] = { 0, };
 398
 399        list_for_each_entry(page, &cc->migratepages, lru)
 400                count[!!page_is_file_cache(page)]++;
 401
 402        /* If locked we can use the interrupt unsafe versions */
 403        if (locked) {
 404                __mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]);
 405                __mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]);
 406        } else {
 407                mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]);
 408                mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]);
 409        }
 410}
 411
 412/* Similar to reclaim, but different enough that they don't share logic */
 413static bool too_many_isolated(struct zone *zone)
 414{
 415        unsigned long active, inactive, isolated;
 416
 417        inactive = zone_page_state(zone, NR_INACTIVE_FILE) +
 418                                        zone_page_state(zone, NR_INACTIVE_ANON);
 419        active = zone_page_state(zone, NR_ACTIVE_FILE) +
 420                                        zone_page_state(zone, NR_ACTIVE_ANON);
 421        isolated = zone_page_state(zone, NR_ISOLATED_FILE) +
 422                                        zone_page_state(zone, NR_ISOLATED_ANON);
 423
 424        return isolated > (inactive + active) / 2;
 425}
 426
 427/**
 428 * isolate_migratepages_range() - isolate all migrate-able pages in range.
 429 * @zone:       Zone pages are in.
 430 * @cc:         Compaction control structure.
 431 * @low_pfn:    The first PFN of the range.
 432 * @end_pfn:    The one-past-the-last PFN of the range.
 433 * @unevictable: true if it allows to isolate unevictable pages
 434 *
 435 * Isolate all pages that can be migrated from the range specified by
 436 * [low_pfn, end_pfn).  Returns zero if there is a fatal signal
 437 * pending), otherwise PFN of the first page that was not scanned
 438 * (which may be both less, equal to or more then end_pfn).
 439 *
 440 * Assumes that cc->migratepages is empty and cc->nr_migratepages is
 441 * zero.
 442 *
 443 * Apart from cc->migratepages and cc->nr_migratetypes this function
 444 * does not modify any cc's fields, in particular it does not modify
 445 * (or read for that matter) cc->migrate_pfn.
 446 */
 447unsigned long
 448isolate_migratepages_range(struct zone *zone, struct compact_control *cc,
 449                unsigned long low_pfn, unsigned long end_pfn, bool unevictable)
 450{
 451        unsigned long last_pageblock_nr = 0, pageblock_nr;
 452        unsigned long nr_scanned = 0, nr_isolated = 0;
 453        struct list_head *migratelist = &cc->migratepages;
 454        isolate_mode_t mode = 0;
 455        struct lruvec *lruvec;
 456        unsigned long flags;
 457        bool locked = false;
 458        struct page *page = NULL, *valid_page = NULL;
 459
 460        /*
 461         * Ensure that there are not too many pages isolated from the LRU
 462         * list by either parallel reclaimers or compaction. If there are,
 463         * delay for some time until fewer pages are isolated
 464         */
 465        while (unlikely(too_many_isolated(zone))) {
 466                /* async migration should just abort */
 467                if (!cc->sync)
 468                        return 0;
 469
 470                congestion_wait(BLK_RW_ASYNC, HZ/10);
 471
 472                if (fatal_signal_pending(current))
 473                        return 0;
 474        }
 475
 476        /* Time to isolate some pages for migration */
 477        cond_resched();
 478        for (; low_pfn < end_pfn; low_pfn++) {
 479                /* give a chance to irqs before checking need_resched() */
 480                if (locked && !((low_pfn+1) % SWAP_CLUSTER_MAX)) {
 481                        if (should_release_lock(&zone->lru_lock)) {
 482                                spin_unlock_irqrestore(&zone->lru_lock, flags);
 483                                locked = false;
 484                        }
 485                }
 486
 487                /*
 488                 * migrate_pfn does not necessarily start aligned to a
 489                 * pageblock. Ensure that pfn_valid is called when moving
 490                 * into a new MAX_ORDER_NR_PAGES range in case of large
 491                 * memory holes within the zone
 492                 */
 493                if ((low_pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) {
 494                        if (!pfn_valid(low_pfn)) {
 495                                low_pfn += MAX_ORDER_NR_PAGES - 1;
 496                                continue;
 497                        }
 498                }
 499
 500                if (!pfn_valid_within(low_pfn))
 501                        continue;
 502                nr_scanned++;
 503
 504                /*
 505                 * Get the page and ensure the page is within the same zone.
 506                 * See the comment in isolate_freepages about overlapping
 507                 * nodes. It is deliberate that the new zone lock is not taken
 508                 * as memory compaction should not move pages between nodes.
 509                 */
 510                page = pfn_to_page(low_pfn);
 511                if (page_zone(page) != zone)
 512                        continue;
 513
 514                if (!valid_page)
 515                        valid_page = page;
 516
 517                /* If isolation recently failed, do not retry */
 518                pageblock_nr = low_pfn >> pageblock_order;
 519                if (!isolation_suitable(cc, page))
 520                        goto next_pageblock;
 521
 522                /* Skip if free */
 523                if (PageBuddy(page))
 524                        continue;
 525
 526                /*
 527                 * For async migration, also only scan in MOVABLE blocks. Async
 528                 * migration is optimistic to see if the minimum amount of work
 529                 * satisfies the allocation
 530                 */
 531                if (!cc->sync && last_pageblock_nr != pageblock_nr &&
 532                    !migrate_async_suitable(get_pageblock_migratetype(page))) {
 533                        cc->finished_update_migrate = true;
 534                        goto next_pageblock;
 535                }
 536
 537                /*
 538                 * Check may be lockless but that's ok as we recheck later.
 539                 * It's possible to migrate LRU pages and balloon pages
 540                 * Skip any other type of page
 541                 */
 542                if (!PageLRU(page)) {
 543                        if (unlikely(balloon_page_movable(page))) {
 544                                if (locked && balloon_page_isolate(page)) {
 545                                        /* Successfully isolated */
 546                                        cc->finished_update_migrate = true;
 547                                        list_add(&page->lru, migratelist);
 548                                        cc->nr_migratepages++;
 549                                        nr_isolated++;
 550                                        goto check_compact_cluster;
 551                                }
 552                        }
 553                        continue;
 554                }
 555
 556                /*
 557                 * PageLRU is set. lru_lock normally excludes isolation
 558                 * splitting and collapsing (collapsing has already happened
 559                 * if PageLRU is set) but the lock is not necessarily taken
 560                 * here and it is wasteful to take it just to check transhuge.
 561                 * Check TransHuge without lock and skip the whole pageblock if
 562                 * it's either a transhuge or hugetlbfs page, as calling
 563                 * compound_order() without preventing THP from splitting the
 564                 * page underneath us may return surprising results.
 565                 */
 566                if (PageTransHuge(page)) {
 567                        if (!locked)
 568                                goto next_pageblock;
 569                        low_pfn += (1 << compound_order(page)) - 1;
 570                        continue;
 571                }
 572
 573                /* Check if it is ok to still hold the lock */
 574                locked = compact_checklock_irqsave(&zone->lru_lock, &flags,
 575                                                                locked, cc);
 576                if (!locked || fatal_signal_pending(current))
 577                        break;
 578
 579                /* Recheck PageLRU and PageTransHuge under lock */
 580                if (!PageLRU(page))
 581                        continue;
 582                if (PageTransHuge(page)) {
 583                        low_pfn += (1 << compound_order(page)) - 1;
 584                        continue;
 585                }
 586
 587                if (!cc->sync)
 588                        mode |= ISOLATE_ASYNC_MIGRATE;
 589
 590                if (unevictable)
 591                        mode |= ISOLATE_UNEVICTABLE;
 592
 593                lruvec = mem_cgroup_page_lruvec(page, zone);
 594
 595                /* Try isolate the page */
 596                if (__isolate_lru_page(page, mode) != 0)
 597                        continue;
 598
 599                VM_BUG_ON(PageTransCompound(page));
 600
 601                /* Successfully isolated */
 602                cc->finished_update_migrate = true;
 603                del_page_from_lru_list(page, lruvec, page_lru(page));
 604                list_add(&page->lru, migratelist);
 605                cc->nr_migratepages++;
 606                nr_isolated++;
 607
 608check_compact_cluster:
 609                /* Avoid isolating too much */
 610                if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) {
 611                        ++low_pfn;
 612                        break;
 613                }
 614
 615                continue;
 616
 617next_pageblock:
 618                low_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages) - 1;
 619                last_pageblock_nr = pageblock_nr;
 620        }
 621
 622        acct_isolated(zone, locked, cc);
 623
 624        if (locked)
 625                spin_unlock_irqrestore(&zone->lru_lock, flags);
 626
 627        /* Update the pageblock-skip if the whole pageblock was scanned */
 628        if (low_pfn == end_pfn)
 629                update_pageblock_skip(cc, valid_page, nr_isolated, true);
 630
 631        trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated);
 632
 633        count_compact_events(COMPACTMIGRATE_SCANNED, nr_scanned);
 634        if (nr_isolated)
 635                count_compact_events(COMPACTISOLATED, nr_isolated);
 636
 637        return low_pfn;
 638}
 639
 640#endif /* CONFIG_COMPACTION || CONFIG_CMA */
 641#ifdef CONFIG_COMPACTION
 642/*
 643 * Based on information in the current compact_control, find blocks
 644 * suitable for isolating free pages from and then isolate them.
 645 */
 646static void isolate_freepages(struct zone *zone,
 647                                struct compact_control *cc)
 648{
 649        struct page *page;
 650        unsigned long high_pfn, low_pfn, pfn, z_end_pfn, end_pfn;
 651        int nr_freepages = cc->nr_freepages;
 652        struct list_head *freelist = &cc->freepages;
 653
 654        /*
 655         * Initialise the free scanner. The starting point is where we last
 656         * scanned from (or the end of the zone if starting). The low point
 657         * is the end of the pageblock the migration scanner is using.
 658         */
 659        pfn = cc->free_pfn;
 660        low_pfn = cc->migrate_pfn + pageblock_nr_pages;
 661
 662        /*
 663         * Take care that if the migration scanner is at the end of the zone
 664         * that the free scanner does not accidentally move to the next zone
 665         * in the next isolation cycle.
 666         */
 667        high_pfn = min(low_pfn, pfn);
 668
 669        z_end_pfn = zone_end_pfn(zone);
 670
 671        /*
 672         * Isolate free pages until enough are available to migrate the
 673         * pages on cc->migratepages. We stop searching if the migrate
 674         * and free page scanners meet or enough free pages are isolated.
 675         */
 676        for (; pfn > low_pfn && cc->nr_migratepages > nr_freepages;
 677                                        pfn -= pageblock_nr_pages) {
 678                unsigned long isolated;
 679
 680                if (!pfn_valid(pfn))
 681                        continue;
 682
 683                /*
 684                 * Check for overlapping nodes/zones. It's possible on some
 685                 * configurations to have a setup like
 686                 * node0 node1 node0
 687                 * i.e. it's possible that all pages within a zones range of
 688                 * pages do not belong to a single zone.
 689                 */
 690                page = pfn_to_page(pfn);
 691                if (page_zone(page) != zone)
 692                        continue;
 693
 694                /* Check the block is suitable for migration */
 695                if (!suitable_migration_target(page))
 696                        continue;
 697
 698                /* If isolation recently failed, do not retry */
 699                if (!isolation_suitable(cc, page))
 700                        continue;
 701
 702                /* Found a block suitable for isolating free pages from */
 703                isolated = 0;
 704
 705                /*
 706                 * As pfn may not start aligned, pfn+pageblock_nr_page
 707                 * may cross a MAX_ORDER_NR_PAGES boundary and miss
 708                 * a pfn_valid check. Ensure isolate_freepages_block()
 709                 * only scans within a pageblock
 710                 */
 711                end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
 712                end_pfn = min(end_pfn, z_end_pfn);
 713                isolated = isolate_freepages_block(cc, pfn, end_pfn,
 714                                                   freelist, false);
 715                nr_freepages += isolated;
 716
 717                /*
 718                 * Record the highest PFN we isolated pages from. When next
 719                 * looking for free pages, the search will restart here as
 720                 * page migration may have returned some pages to the allocator
 721                 */
 722                if (isolated) {
 723                        cc->finished_update_free = true;
 724                        high_pfn = max(high_pfn, pfn);
 725                }
 726        }
 727
 728        /* split_free_page does not map the pages */
 729        map_pages(freelist);
 730
 731        cc->free_pfn = high_pfn;
 732        cc->nr_freepages = nr_freepages;
 733}
 734
 735/*
 736 * This is a migrate-callback that "allocates" freepages by taking pages
 737 * from the isolated freelists in the block we are migrating to.
 738 */
 739static struct page *compaction_alloc(struct page *migratepage,
 740                                        unsigned long data,
 741                                        int **result)
 742{
 743        struct compact_control *cc = (struct compact_control *)data;
 744        struct page *freepage;
 745
 746        /* Isolate free pages if necessary */
 747        if (list_empty(&cc->freepages)) {
 748                isolate_freepages(cc->zone, cc);
 749
 750                if (list_empty(&cc->freepages))
 751                        return NULL;
 752        }
 753
 754        freepage = list_entry(cc->freepages.next, struct page, lru);
 755        list_del(&freepage->lru);
 756        cc->nr_freepages--;
 757
 758        return freepage;
 759}
 760
 761/*
 762 * We cannot control nr_migratepages and nr_freepages fully when migration is
 763 * running as migrate_pages() has no knowledge of compact_control. When
 764 * migration is complete, we count the number of pages on the lists by hand.
 765 */
 766static void update_nr_listpages(struct compact_control *cc)
 767{
 768        int nr_migratepages = 0;
 769        int nr_freepages = 0;
 770        struct page *page;
 771
 772        list_for_each_entry(page, &cc->migratepages, lru)
 773                nr_migratepages++;
 774        list_for_each_entry(page, &cc->freepages, lru)
 775                nr_freepages++;
 776
 777        cc->nr_migratepages = nr_migratepages;
 778        cc->nr_freepages = nr_freepages;
 779}
 780
 781/* possible outcome of isolate_migratepages */
 782typedef enum {
 783        ISOLATE_ABORT,          /* Abort compaction now */
 784        ISOLATE_NONE,           /* No pages isolated, continue scanning */
 785        ISOLATE_SUCCESS,        /* Pages isolated, migrate */
 786} isolate_migrate_t;
 787
 788/*
 789 * Isolate all pages that can be migrated from the block pointed to by
 790 * the migrate scanner within compact_control.
 791 */
 792static isolate_migrate_t isolate_migratepages(struct zone *zone,
 793                                        struct compact_control *cc)
 794{
 795        unsigned long low_pfn, end_pfn;
 796
 797        /* Do not scan outside zone boundaries */
 798        low_pfn = max(cc->migrate_pfn, zone->zone_start_pfn);
 799
 800        /* Only scan within a pageblock boundary */
 801        end_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages);
 802
 803        /* Do not cross the free scanner or scan within a memory hole */
 804        if (end_pfn > cc->free_pfn || !pfn_valid(low_pfn)) {
 805                cc->migrate_pfn = end_pfn;
 806                return ISOLATE_NONE;
 807        }
 808
 809        /* Perform the isolation */
 810        low_pfn = isolate_migratepages_range(zone, cc, low_pfn, end_pfn, false);
 811        if (!low_pfn || cc->contended)
 812                return ISOLATE_ABORT;
 813
 814        cc->migrate_pfn = low_pfn;
 815
 816        return ISOLATE_SUCCESS;
 817}
 818
 819static int compact_finished(struct zone *zone,
 820                            struct compact_control *cc)
 821{
 822        unsigned int order;
 823        unsigned long watermark;
 824
 825        if (fatal_signal_pending(current))
 826                return COMPACT_PARTIAL;
 827
 828        /* Compaction run completes if the migrate and free scanner meet */
 829        if (cc->free_pfn <= cc->migrate_pfn) {
 830                /*
 831                 * Mark that the PG_migrate_skip information should be cleared
 832                 * by kswapd when it goes to sleep. kswapd does not set the
 833                 * flag itself as the decision to be clear should be directly
 834                 * based on an allocation request.
 835                 */
 836                if (!current_is_kswapd())
 837                        zone->compact_blockskip_flush = true;
 838
 839                return COMPACT_COMPLETE;
 840        }
 841
 842        /*
 843         * order == -1 is expected when compacting via
 844         * /proc/sys/vm/compact_memory
 845         */
 846        if (cc->order == -1)
 847                return COMPACT_CONTINUE;
 848
 849        /* Compaction run is not finished if the watermark is not met */
 850        watermark = low_wmark_pages(zone);
 851        watermark += (1 << cc->order);
 852
 853        if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0))
 854                return COMPACT_CONTINUE;
 855
 856        /* Direct compactor: Is a suitable page free? */
 857        for (order = cc->order; order < MAX_ORDER; order++) {
 858                struct free_area *area = &zone->free_area[order];
 859
 860                /* Job done if page is free of the right migratetype */
 861                if (!list_empty(&area->free_list[cc->migratetype]))
 862                        return COMPACT_PARTIAL;
 863
 864                /* Job done if allocation would set block type */
 865                if (cc->order >= pageblock_order && area->nr_free)
 866                        return COMPACT_PARTIAL;
 867        }
 868
 869        return COMPACT_CONTINUE;
 870}
 871
 872/*
 873 * compaction_suitable: Is this suitable to run compaction on this zone now?
 874 * Returns
 875 *   COMPACT_SKIPPED  - If there are too few free pages for compaction
 876 *   COMPACT_PARTIAL  - If the allocation would succeed without compaction
 877 *   COMPACT_CONTINUE - If compaction should run now
 878 */
 879unsigned long compaction_suitable(struct zone *zone, int order)
 880{
 881        int fragindex;
 882        unsigned long watermark;
 883
 884        /*
 885         * order == -1 is expected when compacting via
 886         * /proc/sys/vm/compact_memory
 887         */
 888        if (order == -1)
 889                return COMPACT_CONTINUE;
 890
 891        /*
 892         * Watermarks for order-0 must be met for compaction. Note the 2UL.
 893         * This is because during migration, copies of pages need to be
 894         * allocated and for a short time, the footprint is higher
 895         */
 896        watermark = low_wmark_pages(zone) + (2UL << order);
 897        if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
 898                return COMPACT_SKIPPED;
 899
 900        /*
 901         * fragmentation index determines if allocation failures are due to
 902         * low memory or external fragmentation
 903         *
 904         * index of -1000 implies allocations might succeed depending on
 905         * watermarks
 906         * index towards 0 implies failure is due to lack of memory
 907         * index towards 1000 implies failure is due to fragmentation
 908         *
 909         * Only compact if a failure would be due to fragmentation.
 910         */
 911        fragindex = fragmentation_index(zone, order);
 912        if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold)
 913                return COMPACT_SKIPPED;
 914
 915        if (fragindex == -1000 && zone_watermark_ok(zone, order, watermark,
 916            0, 0))
 917                return COMPACT_PARTIAL;
 918
 919        return COMPACT_CONTINUE;
 920}
 921
 922static int compact_zone(struct zone *zone, struct compact_control *cc)
 923{
 924        int ret;
 925        unsigned long start_pfn = zone->zone_start_pfn;
 926        unsigned long end_pfn = zone_end_pfn(zone);
 927
 928        ret = compaction_suitable(zone, cc->order);
 929        switch (ret) {
 930        case COMPACT_PARTIAL:
 931        case COMPACT_SKIPPED:
 932                /* Compaction is likely to fail */
 933                return ret;
 934        case COMPACT_CONTINUE:
 935                /* Fall through to compaction */
 936                ;
 937        }
 938
 939        /*
 940         * Setup to move all movable pages to the end of the zone. Used cached
 941         * information on where the scanners should start but check that it
 942         * is initialised by ensuring the values are within zone boundaries.
 943         */
 944        cc->migrate_pfn = zone->compact_cached_migrate_pfn;
 945        cc->free_pfn = zone->compact_cached_free_pfn;
 946        if (cc->free_pfn < start_pfn || cc->free_pfn > end_pfn) {
 947                cc->free_pfn = end_pfn & ~(pageblock_nr_pages-1);
 948                zone->compact_cached_free_pfn = cc->free_pfn;
 949        }
 950        if (cc->migrate_pfn < start_pfn || cc->migrate_pfn > end_pfn) {
 951                cc->migrate_pfn = start_pfn;
 952                zone->compact_cached_migrate_pfn = cc->migrate_pfn;
 953        }
 954
 955        /*
 956         * Clear pageblock skip if there were failures recently and compaction
 957         * is about to be retried after being deferred. kswapd does not do
 958         * this reset as it'll reset the cached information when going to sleep.
 959         */
 960        if (compaction_restarting(zone, cc->order) && !current_is_kswapd())
 961                __reset_isolation_suitable(zone);
 962
 963        migrate_prep_local();
 964
 965        while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) {
 966                unsigned long nr_migrate, nr_remaining;
 967                int err;
 968
 969                switch (isolate_migratepages(zone, cc)) {
 970                case ISOLATE_ABORT:
 971                        ret = COMPACT_PARTIAL;
 972                        putback_movable_pages(&cc->migratepages);
 973                        cc->nr_migratepages = 0;
 974                        goto out;
 975                case ISOLATE_NONE:
 976                        continue;
 977                case ISOLATE_SUCCESS:
 978                        ;
 979                }
 980
 981                nr_migrate = cc->nr_migratepages;
 982                err = migrate_pages(&cc->migratepages, compaction_alloc,
 983                                (unsigned long)cc,
 984                                cc->sync ? MIGRATE_SYNC_LIGHT : MIGRATE_ASYNC,
 985                                MR_COMPACTION);
 986                update_nr_listpages(cc);
 987                nr_remaining = cc->nr_migratepages;
 988
 989                trace_mm_compaction_migratepages(nr_migrate - nr_remaining,
 990                                                nr_remaining);
 991
 992                /* Release isolated pages not migrated */
 993                if (err) {
 994                        putback_movable_pages(&cc->migratepages);
 995                        cc->nr_migratepages = 0;
 996                        if (err == -ENOMEM) {
 997                                ret = COMPACT_PARTIAL;
 998                                goto out;
 999                        }
1000                }
1001        }
1002
1003out:
1004        /* Release free pages and check accounting */
1005        cc->nr_freepages -= release_freepages(&cc->freepages);
1006        VM_BUG_ON(cc->nr_freepages != 0);
1007
1008        return ret;
1009}
1010
1011static unsigned long compact_zone_order(struct zone *zone,
1012                                 int order, gfp_t gfp_mask,
1013                                 bool sync, bool *contended)
1014{
1015        unsigned long ret;
1016        struct compact_control cc = {
1017                .nr_freepages = 0,
1018                .nr_migratepages = 0,
1019                .order = order,
1020                .migratetype = allocflags_to_migratetype(gfp_mask),
1021                .zone = zone,
1022                .sync = sync,
1023        };
1024        INIT_LIST_HEAD(&cc.freepages);
1025        INIT_LIST_HEAD(&cc.migratepages);
1026
1027        ret = compact_zone(zone, &cc);
1028
1029        VM_BUG_ON(!list_empty(&cc.freepages));
1030        VM_BUG_ON(!list_empty(&cc.migratepages));
1031
1032        *contended = cc.contended;
1033        return ret;
1034}
1035
1036int sysctl_extfrag_threshold = 500;
1037
1038/**
1039 * try_to_compact_pages - Direct compact to satisfy a high-order allocation
1040 * @zonelist: The zonelist used for the current allocation
1041 * @order: The order of the current allocation
1042 * @gfp_mask: The GFP mask of the current allocation
1043 * @nodemask: The allowed nodes to allocate from
1044 * @sync: Whether migration is synchronous or not
1045 * @contended: Return value that is true if compaction was aborted due to lock contention
1046 * @page: Optionally capture a free page of the requested order during compaction
1047 *
1048 * This is the main entry point for direct page compaction.
1049 */
1050unsigned long try_to_compact_pages(struct zonelist *zonelist,
1051                        int order, gfp_t gfp_mask, nodemask_t *nodemask,
1052                        bool sync, bool *contended)
1053{
1054        enum zone_type high_zoneidx = gfp_zone(gfp_mask);
1055        int may_enter_fs = gfp_mask & __GFP_FS;
1056        int may_perform_io = gfp_mask & __GFP_IO;
1057        struct zoneref *z;
1058        struct zone *zone;
1059        int rc = COMPACT_SKIPPED;
1060        int alloc_flags = 0;
1061
1062        /* Check if the GFP flags allow compaction */
1063        if (!order || !may_enter_fs || !may_perform_io)
1064                return rc;
1065
1066        count_compact_event(COMPACTSTALL);
1067
1068#ifdef CONFIG_CMA
1069        if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
1070                alloc_flags |= ALLOC_CMA;
1071#endif
1072        /* Compact each zone in the list */
1073        for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx,
1074                                                                nodemask) {
1075                int status;
1076
1077                status = compact_zone_order(zone, order, gfp_mask, sync,
1078                                                contended);
1079                rc = max(status, rc);
1080
1081                /* If a normal allocation would succeed, stop compacting */
1082                if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0,
1083                                      alloc_flags))
1084                        break;
1085        }
1086
1087        return rc;
1088}
1089
1090
1091/* Compact all zones within a node */
1092static void __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc)
1093{
1094        int zoneid;
1095        struct zone *zone;
1096
1097        for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) {
1098
1099                zone = &pgdat->node_zones[zoneid];
1100                if (!populated_zone(zone))
1101                        continue;
1102
1103                cc->nr_freepages = 0;
1104                cc->nr_migratepages = 0;
1105                cc->zone = zone;
1106                INIT_LIST_HEAD(&cc->freepages);
1107                INIT_LIST_HEAD(&cc->migratepages);
1108
1109                if (cc->order == -1 || !compaction_deferred(zone, cc->order))
1110                        compact_zone(zone, cc);
1111
1112                if (cc->order > 0) {
1113                        int ok = zone_watermark_ok(zone, cc->order,
1114                                                low_wmark_pages(zone), 0, 0);
1115                        if (ok && cc->order >= zone->compact_order_failed)
1116                                zone->compact_order_failed = cc->order + 1;
1117                        /* Currently async compaction is never deferred. */
1118                        else if (!ok && cc->sync)
1119                                defer_compaction(zone, cc->order);
1120                }
1121
1122                VM_BUG_ON(!list_empty(&cc->freepages));
1123                VM_BUG_ON(!list_empty(&cc->migratepages));
1124        }
1125}
1126
1127void compact_pgdat(pg_data_t *pgdat, int order)
1128{
1129        struct compact_control cc = {
1130                .order = order,
1131                .sync = false,
1132        };
1133
1134        __compact_pgdat(pgdat, &cc);
1135}
1136
1137static void compact_node(int nid)
1138{
1139        struct compact_control cc = {
1140                .order = -1,
1141                .sync = true,
1142        };
1143
1144        __compact_pgdat(NODE_DATA(nid), &cc);
1145}
1146
1147/* Compact all nodes in the system */
1148static void compact_nodes(void)
1149{
1150        int nid;
1151
1152        /* Flush pending updates to the LRU lists */
1153        lru_add_drain_all();
1154
1155        for_each_online_node(nid)
1156                compact_node(nid);
1157}
1158
1159/* The written value is actually unused, all memory is compacted */
1160int sysctl_compact_memory;
1161
1162/* This is the entry point for compacting all nodes via /proc/sys/vm */
1163int sysctl_compaction_handler(struct ctl_table *table, int write,
1164                        void __user *buffer, size_t *length, loff_t *ppos)
1165{
1166        if (write)
1167                compact_nodes();
1168
1169        return 0;
1170}
1171
1172int sysctl_extfrag_handler(struct ctl_table *table, int write,
1173                        void __user *buffer, size_t *length, loff_t *ppos)
1174{
1175        proc_dointvec_minmax(table, write, buffer, length, ppos);
1176
1177        return 0;
1178}
1179
1180#if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA)
1181ssize_t sysfs_compact_node(struct device *dev,
1182                        struct device_attribute *attr,
1183                        const char *buf, size_t count)
1184{
1185        int nid = dev->id;
1186
1187        if (nid >= 0 && nid < nr_node_ids && node_online(nid)) {
1188                /* Flush pending updates to the LRU lists */
1189                lru_add_drain_all();
1190
1191                compact_node(nid);
1192        }
1193
1194        return count;
1195}
1196static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node);
1197
1198int compaction_register_node(struct node *node)
1199{
1200        return device_create_file(&node->dev, &dev_attr_compact);
1201}
1202
1203void compaction_unregister_node(struct node *node)
1204{
1205        return device_remove_file(&node->dev, &dev_attr_compact);
1206}
1207#endif /* CONFIG_SYSFS && CONFIG_NUMA */
1208
1209#endif /* CONFIG_COMPACTION */
1210
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.