mm/compaction.c

   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 "internal.h"
  18
  19 #define CREATE_TRACE_POINTS
  20 #include <trace/events/compaction.h>
  21
  22 /*
  23  * compact_control is used to track pages being migrated and the free pages
  24  * they are being migrated to during memory compaction. The free_pfn starts
  25  * at the end of a zone and migrate_pfn begins at the start. Movable pages
  26  * are moved to the end of a zone during a compaction run and the run
  27  * completes when free_pfn <= migrate_pfn
  28  */
  29 struct compact_control {
  30         struct list_head freepages;     /* List of free pages to migrate to */
  31         struct list_head migratepages;  /* List of pages being migrated */
  32         unsigned long nr_freepages;     /* Number of isolated free pages */
  33         unsigned long nr_migratepages;  /* Number of pages to migrate */
  34         unsigned long free_pfn;         /* isolate_freepages search base */
  35         unsigned long migrate_pfn;      /* isolate_migratepages search base */
  36         bool sync;                      /* Synchronous migration */
  37
  38         unsigned int order;             /* order a direct compactor needs */
  39         int migratetype;                /* MOVABLE, RECLAIMABLE etc */
  40         struct zone *zone;
  41 };
  42
  43 static unsigned long release_freepages(struct list_head *freelist)
  44 {
  45         struct page *page, *next;
  46         unsigned long count = 0;
  47
  48         list_for_each_entry_safe(page, next, freelist, lru) {
  49                 list_del(&page->lru);
  50                 __free_page(page);
  51                 count++;
  52         }
  53
  54         return count;
  55 }
  56
  57 /* Isolate free pages onto a private freelist. Must hold zone->lock */
  58 static unsigned long isolate_freepages_block(struct zone *zone,
  59                                 unsigned long blockpfn,
  60                                 struct list_head *freelist)
  61 {
  62         unsigned long zone_end_pfn, end_pfn;
  63         int nr_scanned = 0, total_isolated = 0;
  64         struct page *cursor;
  65
  66         /* Get the last PFN we should scan for free pages at */
  67         zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
  68         end_pfn = min(blockpfn + pageblock_nr_pages, zone_end_pfn);
  69
  70         /* Find the first usable PFN in the block to initialse page cursor */
  71         for (; blockpfn < end_pfn; blockpfn++) {
  72                 if (pfn_valid_within(blockpfn))
  73                         break;
  74         }
  75         cursor = pfn_to_page(blockpfn);
  76
  77         /* Isolate free pages. This assumes the block is valid */
  78         for (; blockpfn < end_pfn; blockpfn++, cursor++) {
  79                 int isolated, i;
  80                 struct page *page = cursor;
  81
  82                 if (!pfn_valid_within(blockpfn))
  83                         continue;
  84                 nr_scanned++;
  85
  86                 if (!PageBuddy(page))
  87                         continue;
  88
  89                 /* Found a free page, break it into order-0 pages */
  90                 isolated = split_free_page(page);
  91                 total_isolated += isolated;
  92                 for (i = 0; i < isolated; i++) {
  93                         list_add(&page->lru, freelist);
  94                         page++;
  95                 }
  96
  97                 /* If a page was split, advance to the end of it */
  98                 if (isolated) {
  99                         blockpfn += isolated - 1;
 100                         cursor += isolated - 1;
 101                 }
 102         }
 103
 104         trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated);
 105         return total_isolated;
 106 }
 107
 108 /* Returns true if the page is within a block suitable for migration to */
 109 static bool suitable_migration_target(struct page *page)
 110 {
 111
 112         int migratetype = get_pageblock_migratetype(page);
 113
 114         /* Don't interfere with memory hot-remove or the min_free_kbytes blocks */
 115         if (migratetype == MIGRATE_ISOLATE || migratetype == MIGRATE_RESERVE)
 116                 return false;
 117
 118         /* If the page is a large free page, then allow migration */
 119         if (PageBuddy(page) && page_order(page) >= pageblock_order)
 120                 return true;
 121
 122         /* If the block is MIGRATE_MOVABLE, allow migration */
 123         if (migratetype == MIGRATE_MOVABLE)
 124                 return true;
 125
 126         /* Otherwise skip the block */
 127         return false;
 128 }
 129
 130 /*
 131  * Based on information in the current compact_control, find blocks
 132  * suitable for isolating free pages from and then isolate them.
 133  */
 134 static void isolate_freepages(struct zone *zone,
 135                                 struct compact_control *cc)
 136 {
 137         struct page *page;
 138         unsigned long high_pfn, low_pfn, pfn;
 139         unsigned long flags;
 140         int nr_freepages = cc->nr_freepages;
 141         struct list_head *freelist = &cc->freepages;
 142
 143         /*
 144          * Initialise the free scanner. The starting point is where we last
 145          * scanned from (or the end of the zone if starting). The low point
 146          * is the end of the pageblock the migration scanner is using.
 147          */
 148         pfn = cc->free_pfn;
 149         low_pfn = cc->migrate_pfn + pageblock_nr_pages;
 150
 151         /*
 152          * Take care that if the migration scanner is at the end of the zone
 153          * that the free scanner does not accidentally move to the next zone
 154          * in the next isolation cycle.
 155          */
 156         high_pfn = min(low_pfn, pfn);
 157
 158         /*
 159          * Isolate free pages until enough are available to migrate the
 160          * pages on cc->migratepages. We stop searching if the migrate
 161          * and free page scanners meet or enough free pages are isolated.
 162          */
 163         for (; pfn > low_pfn && cc->nr_migratepages > nr_freepages;
 164                                         pfn -= pageblock_nr_pages) {
 165                 unsigned long isolated;
 166
 167                 if (!pfn_valid(pfn))
 168                         continue;
 169
 170                 /*
 171                  * Check for overlapping nodes/zones. It's possible on some
 172                  * configurations to have a setup like
 173                  * node0 node1 node0
 174                  * i.e. it's possible that all pages within a zones range of
 175                  * pages do not belong to a single zone.
 176                  */
 177                 page = pfn_to_page(pfn);
 178                 if (page_zone(page) != zone)
 179                         continue;
 180
 181                 /* Check the block is suitable for migration */
 182                 if (!suitable_migration_target(page))
 183                         continue;
 184
 185                 /*
 186                  * Found a block suitable for isolating free pages from. Now
 187                  * we disabled interrupts, double check things are ok and
 188                  * isolate the pages. This is to minimise the time IRQs
 189                  * are disabled
 190                  */
 191                 isolated = 0;
 192                 spin_lock_irqsave(&zone->lock, flags);
 193                 if (suitable_migration_target(page)) {
 194                         isolated = isolate_freepages_block(zone, pfn, freelist);
 195                         nr_freepages += isolated;
 196                 }
 197                 spin_unlock_irqrestore(&zone->lock, flags);
 198
 199                 /*
 200                  * Record the highest PFN we isolated pages from. When next
 201                  * looking for free pages, the search will restart here as
 202                  * page migration may have returned some pages to the allocator
 203                  */
 204                 if (isolated)
 205                         high_pfn = max(high_pfn, pfn);
 206         }
 207
 208         /* split_free_page does not map the pages */
 209         list_for_each_entry(page, freelist, lru) {
 210                 arch_alloc_page(page, 0);
 211                 kernel_map_pages(page, 1, 1);
 212         }
 213
 214         cc->free_pfn = high_pfn;
 215         cc->nr_freepages = nr_freepages;
 216 }
 217
 218 /* Update the number of anon and file isolated pages in the zone */
 219 static void acct_isolated(struct zone *zone, struct compact_control *cc)
 220 {
 221         struct page *page;
 222         unsigned int count[2] = { 0, };
 223
 224         list_for_each_entry(page, &cc->migratepages, lru)
 225                 count[!!page_is_file_cache(page)]++;
 226
 227         __mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]);
 228         __mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]);
 229 }
 230
 231 /* Similar to reclaim, but different enough that they don't share logic */
 232 static bool too_many_isolated(struct zone *zone)
 233 {
 234         unsigned long active, inactive, isolated;
 235
 236         inactive = zone_page_state(zone, NR_INACTIVE_FILE) +
 237                                         zone_page_state(zone, NR_INACTIVE_ANON);
 238         active = zone_page_state(zone, NR_ACTIVE_FILE) +
 239                                         zone_page_state(zone, NR_ACTIVE_ANON);
 240         isolated = zone_page_state(zone, NR_ISOLATED_FILE) +
 241                                         zone_page_state(zone, NR_ISOLATED_ANON);
 242
 243         return isolated > (inactive + active) / 2;
 244 }
 245
 246 /* possible outcome of isolate_migratepages */
 247 typedef enum {
 248         ISOLATE_ABORT,          /* Abort compaction now */
 249         ISOLATE_NONE,           /* No pages isolated, continue scanning */
 250         ISOLATE_SUCCESS,        /* Pages isolated, migrate */
 251 } isolate_migrate_t;
 252
 253 /*
 254  * Isolate all pages that can be migrated from the block pointed to by
 255  * the migrate scanner within compact_control.
 256  */
 257 static isolate_migrate_t isolate_migratepages(struct zone *zone,
 258                                         struct compact_control *cc)
 259 {
 260         unsigned long low_pfn, end_pfn;
 261         unsigned long last_pageblock_nr = 0, pageblock_nr;
 262         unsigned long nr_scanned = 0, nr_isolated = 0;
 263         struct list_head *migratelist = &cc->migratepages;
 264         isolate_mode_t mode = ISOLATE_ACTIVE|ISOLATE_INACTIVE;
 265
 266         /* Do not scan outside zone boundaries */
 267         low_pfn = max(cc->migrate_pfn, zone->zone_start_pfn);
 268
 269         /* Only scan within a pageblock boundary */
 270         end_pfn = ALIGN(low_pfn + pageblock_nr_pages, pageblock_nr_pages);
 271
 272         /* Do not cross the free scanner or scan within a memory hole */
 273         if (end_pfn > cc->free_pfn || !pfn_valid(low_pfn)) {
 274                 cc->migrate_pfn = end_pfn;
 275                 return ISOLATE_NONE;
 276         }
 277
 278         /*
 279          * Ensure that there are not too many pages isolated from the LRU
 280          * list by either parallel reclaimers or compaction. If there are,
 281          * delay for some time until fewer pages are isolated
 282          */
 283         while (unlikely(too_many_isolated(zone))) {
 284                 /* async migration should just abort */
 285                 if (!cc->sync)
 286                         return ISOLATE_ABORT;
 287
 288                 congestion_wait(BLK_RW_ASYNC, HZ/10);
 289
 290                 if (fatal_signal_pending(current))
 291                         return ISOLATE_ABORT;
 292         }
 293
 294         /* Time to isolate some pages for migration */
 295         cond_resched();
 296         spin_lock_irq(&zone->lru_lock);
 297         for (; low_pfn < end_pfn; low_pfn++) {
 298                 struct page *page;
 299                 bool locked = true;
 300
 301                 /* give a chance to irqs before checking need_resched() */
 302                 if (!((low_pfn+1) % SWAP_CLUSTER_MAX)) {
 303                         spin_unlock_irq(&zone->lru_lock);
 304                         locked = false;
 305                 }
 306                 if (need_resched() || spin_is_contended(&zone->lru_lock)) {
 307                         if (locked)
 308                                 spin_unlock_irq(&zone->lru_lock);
 309                         cond_resched();
 310                         spin_lock_irq(&zone->lru_lock);
 311                         if (fatal_signal_pending(current))
 312                                 break;
 313                 } else if (!locked)
 314                         spin_lock_irq(&zone->lru_lock);
 315
 316                 if (!pfn_valid_within(low_pfn))
 317                         continue;
 318                 nr_scanned++;
 319
 320                 /* Get the page and skip if free */
 321                 page = pfn_to_page(low_pfn);
 322                 if (PageBuddy(page))
 323                         continue;
 324
 325                 /*
 326                  * For async migration, also only scan in MOVABLE blocks. Async
 327                  * migration is optimistic to see if the minimum amount of work
 328                  * satisfies the allocation
 329                  */
 330                 pageblock_nr = low_pfn >> pageblock_order;
 331                 if (!cc->sync && last_pageblock_nr != pageblock_nr &&
 332                                 get_pageblock_migratetype(page) != MIGRATE_MOVABLE) {
 333                         low_pfn += pageblock_nr_pages;
 334                         low_pfn = ALIGN(low_pfn, pageblock_nr_pages) - 1;
 335                         last_pageblock_nr = pageblock_nr;
 336                         continue;
 337                 }
 338
 339                 if (!PageLRU(page))
 340                         continue;
 341
 342                 /*
 343                  * PageLRU is set, and lru_lock excludes isolation,
 344                  * splitting and collapsing (collapsing has already
 345                  * happened if PageLRU is set).
 346                  */
 347                 if (PageTransHuge(page)) {
 348                         low_pfn += (1 << compound_order(page)) - 1;
 349                         continue;
 350                 }
 351
 352                 /* Try isolate the page */
 353                 if (__isolate_lru_page(page, mode, 0) != 0)
 354                         continue;
 355
 356                 VM_BUG_ON(PageTransCompound(page));
 357
 358                 /* Successfully isolated */
 359                 del_page_from_lru_list(zone, page, page_lru(page));
 360                 list_add(&page->lru, migratelist);
 361                 cc->nr_migratepages++;
 362                 nr_isolated++;
 363
 364                 /* Avoid isolating too much */
 365                 if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) {
 366                         ++low_pfn;
 367                         break;
 368                 }
 369         }
 370
 371         acct_isolated(zone, cc);
 372
 373         spin_unlock_irq(&zone->lru_lock);
 374         cc->migrate_pfn = low_pfn;
 375
 376         trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated);
 377
 378         return ISOLATE_SUCCESS;
 379 }
 380
 381 /*
 382  * This is a migrate-callback that "allocates" freepages by taking pages
 383  * from the isolated freelists in the block we are migrating to.
 384  */
 385 static struct page *compaction_alloc(struct page *migratepage,
 386                                         unsigned long data,
 387                                         int **result)
 388 {
 389         struct compact_control *cc = (struct compact_control *)data;
 390         struct page *freepage;
 391
 392         /* Isolate free pages if necessary */
 393         if (list_empty(&cc->freepages)) {
 394                 isolate_freepages(cc->zone, cc);
 395
 396                 if (list_empty(&cc->freepages))
 397                         return NULL;
 398         }
 399
 400         freepage = list_entry(cc->freepages.next, struct page, lru);
 401         list_del(&freepage->lru);
 402         cc->nr_freepages--;
 403
 404         return freepage;
 405 }
 406
 407 /*
 408  * We cannot control nr_migratepages and nr_freepages fully when migration is
 409  * running as migrate_pages() has no knowledge of compact_control. When
 410  * migration is complete, we count the number of pages on the lists by hand.
 411  */
 412 static void update_nr_listpages(struct compact_control *cc)
 413 {
 414         int nr_migratepages = 0;
 415         int nr_freepages = 0;
 416         struct page *page;
 417
 418         list_for_each_entry(page, &cc->migratepages, lru)
 419                 nr_migratepages++;
 420         list_for_each_entry(page, &cc->freepages, lru)
 421                 nr_freepages++;
 422
 423         cc->nr_migratepages = nr_migratepages;
 424         cc->nr_freepages = nr_freepages;
 425 }
 426
 427 static int compact_finished(struct zone *zone,
 428                             struct compact_control *cc)
 429 {
 430         unsigned int order;
 431         unsigned long watermark;
 432
 433         if (fatal_signal_pending(current))
 434                 return COMPACT_PARTIAL;
 435
 436         /* Compaction run completes if the migrate and free scanner meet */
 437         if (cc->free_pfn <= cc->migrate_pfn)
 438                 return COMPACT_COMPLETE;
 439
 440         /*
 441          * order == -1 is expected when compacting via
 442          * /proc/sys/vm/compact_memory
 443          */
 444         if (cc->order == -1)
 445                 return COMPACT_CONTINUE;
 446
 447         /* Compaction run is not finished if the watermark is not met */
 448         watermark = low_wmark_pages(zone);
 449         watermark += (1 << cc->order);
 450
 451         if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0))
 452                 return COMPACT_CONTINUE;
 453
 454         /* Direct compactor: Is a suitable page free? */
 455         for (order = cc->order; order < MAX_ORDER; order++) {
 456                 /* Job done if page is free of the right migratetype */
 457                 if (!list_empty(&zone->free_area[order].free_list[cc->migratetype]))
 458                         return COMPACT_PARTIAL;
 459
 460                 /* Job done if allocation would set block type */
 461                 if (order >= pageblock_order && zone->free_area[order].nr_free)
 462                         return COMPACT_PARTIAL;
 463         }
 464
 465         return COMPACT_CONTINUE;
 466 }
 467
 468 /*
 469  * compaction_suitable: Is this suitable to run compaction on this zone now?
 470  * Returns
 471  *   COMPACT_SKIPPED  - If there are too few free pages for compaction
 472  *   COMPACT_PARTIAL  - If the allocation would succeed without compaction
 473  *   COMPACT_CONTINUE - If compaction should run now
 474  */
 475 unsigned long compaction_suitable(struct zone *zone, int order)
 476 {
 477         int fragindex;
 478         unsigned long watermark;
 479
 480         /*
 481          * order == -1 is expected when compacting via
 482          * /proc/sys/vm/compact_memory
 483          */
 484         if (order == -1)
 485                 return COMPACT_CONTINUE;
 486
 487         /*
 488          * Watermarks for order-0 must be met for compaction. Note the 2UL.
 489          * This is because during migration, copies of pages need to be
 490          * allocated and for a short time, the footprint is higher
 491          */
 492         watermark = low_wmark_pages(zone) + (2UL << order);
 493         if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
 494                 return COMPACT_SKIPPED;
 495
 496         /*
 497          * fragmentation index determines if allocation failures are due to
 498          * low memory or external fragmentation
 499          *
 500          * index of -1000 implies allocations might succeed depending on
 501          * watermarks
 502          * index towards 0 implies failure is due to lack of memory
 503          * index towards 1000 implies failure is due to fragmentation
 504          *
 505          * Only compact if a failure would be due to fragmentation.
 506          */
 507         fragindex = fragmentation_index(zone, order);
 508         if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold)
 509                 return COMPACT_SKIPPED;
 510
 511         if (fragindex == -1000 && zone_watermark_ok(zone, order, watermark,
 512             0, 0))
 513                 return COMPACT_PARTIAL;
 514
 515         return COMPACT_CONTINUE;
 516 }
 517
 518 static int compact_zone(struct zone *zone, struct compact_control *cc)
 519 {
 520         int ret;
 521
 522         ret = compaction_suitable(zone, cc->order);
 523         switch (ret) {
 524         case COMPACT_PARTIAL:
 525         case COMPACT_SKIPPED:
 526                 /* Compaction is likely to fail */
 527                 return ret;
 528         case COMPACT_CONTINUE:
 529                 /* Fall through to compaction */
 530                 ;
 531         }
 532
 533         /* Setup to move all movable pages to the end of the zone */
 534         cc->migrate_pfn = zone->zone_start_pfn;
 535         cc->free_pfn = cc->migrate_pfn + zone->spanned_pages;
 536         cc->free_pfn &= ~(pageblock_nr_pages-1);
 537
 538         migrate_prep_local();
 539
 540         while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) {
 541                 unsigned long nr_migrate, nr_remaining;
 542                 int err;
 543
 544                 switch (isolate_migratepages(zone, cc)) {
 545                 case ISOLATE_ABORT:
 546                         ret = COMPACT_PARTIAL;
 547                         goto out;
 548                 case ISOLATE_NONE:
 549                         continue;
 550                 case ISOLATE_SUCCESS:
 551                         ;
 552                 }
 553
 554                 nr_migrate = cc->nr_migratepages;
 555                 err = migrate_pages(&cc->migratepages, compaction_alloc,
 556                                 (unsigned long)cc, false,
 557                                 cc->sync);
 558                 update_nr_listpages(cc);
 559                 nr_remaining = cc->nr_migratepages;
 560
 561                 count_vm_event(COMPACTBLOCKS);
 562                 count_vm_events(COMPACTPAGES, nr_migrate - nr_remaining);
 563                 if (nr_remaining)
 564                         count_vm_events(COMPACTPAGEFAILED, nr_remaining);
 565                 trace_mm_compaction_migratepages(nr_migrate - nr_remaining,
 566                                                 nr_remaining);
 567
 568                 /* Release LRU pages not migrated */
 569                 if (err) {
 570                         putback_lru_pages(&cc->migratepages);
 571                         cc->nr_migratepages = 0;
 572                 }
 573
 574         }
 575
 576 out:
 577         /* Release free pages and check accounting */
 578         cc->nr_freepages -= release_freepages(&cc->freepages);
 579         VM_BUG_ON(cc->nr_freepages != 0);
 580
 581         return ret;
 582 }
 583
 584 static unsigned long compact_zone_order(struct zone *zone,
 585                                  int order, gfp_t gfp_mask,
 586                                  bool sync)
 587 {
 588         struct compact_control cc = {
 589                 .nr_freepages = 0,
 590                 .nr_migratepages = 0,
 591                 .order = order,
 592                 .migratetype = allocflags_to_migratetype(gfp_mask),
 593                 .zone = zone,
 594                 .sync = sync,
 595         };
 596         INIT_LIST_HEAD(&cc.freepages);
 597         INIT_LIST_HEAD(&cc.migratepages);
 598
 599         return compact_zone(zone, &cc);
 600 }
 601
 602 int sysctl_extfrag_threshold = 500;
 603
 604 /**
 605  * try_to_compact_pages - Direct compact to satisfy a high-order allocation
 606  * @zonelist: The zonelist used for the current allocation
 607  * @order: The order of the current allocation
 608  * @gfp_mask: The GFP mask of the current allocation
 609  * @nodemask: The allowed nodes to allocate from
 610  * @sync: Whether migration is synchronous or not
 611  *
 612  * This is the main entry point for direct page compaction.
 613  */
 614 unsigned long try_to_compact_pages(struct zonelist *zonelist,
 615                         int order, gfp_t gfp_mask, nodemask_t *nodemask,
 616                         bool sync)
 617 {
 618         enum zone_type high_zoneidx = gfp_zone(gfp_mask);
 619         int may_enter_fs = gfp_mask & __GFP_FS;
 620         int may_perform_io = gfp_mask & __GFP_IO;
 621         struct zoneref *z;
 622         struct zone *zone;
 623         int rc = COMPACT_SKIPPED;
 624
 625         /*
 626          * Check whether it is worth even starting compaction. The order check is
 627          * made because an assumption is made that the page allocator can satisfy
 628          * the "cheaper" orders without taking special steps
 629          */
 630         if (!order || !may_enter_fs || !may_perform_io)
 631                 return rc;
 632
 633         count_vm_event(COMPACTSTALL);
 634
 635         /* Compact each zone in the list */
 636         for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx,
 637                                                                 nodemask) {
 638                 int status;
 639
 640                 status = compact_zone_order(zone, order, gfp_mask, sync);
 641                 rc = max(status, rc);
 642
 643                 /* If a normal allocation would succeed, stop compacting */
 644                 if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0, 0))
 645                         break;
 646         }
 647
 648         return rc;
 649 }
 650
 651
 652 /* Compact all zones within a node */
 653 static int compact_node(int nid)
 654 {
 655         int zoneid;
 656         pg_data_t *pgdat;
 657         struct zone *zone;
 658
 659         if (nid < 0 || nid >= nr_node_ids || !node_online(nid))
 660                 return -EINVAL;
 661         pgdat = NODE_DATA(nid);
 662
 663         /* Flush pending updates to the LRU lists */
 664         lru_add_drain_all();
 665
 666         for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) {
 667                 struct compact_control cc = {
 668                         .nr_freepages = 0,
 669                         .nr_migratepages = 0,
 670                         .order = -1,
 671                 };
 672
 673                 zone = &pgdat->node_zones[zoneid];
 674                 if (!populated_zone(zone))
 675                         continue;
 676
 677                 cc.zone = zone;
 678                 INIT_LIST_HEAD(&cc.freepages);
 679                 INIT_LIST_HEAD(&cc.migratepages);
 680
 681                 compact_zone(zone, &cc);
 682
 683                 VM_BUG_ON(!list_empty(&cc.freepages));
 684                 VM_BUG_ON(!list_empty(&cc.migratepages));
 685         }
 686
 687         return 0;
 688 }
 689
 690 /* Compact all nodes in the system */
 691 static int compact_nodes(void)
 692 {
 693         int nid;
 694
 695         for_each_online_node(nid)
 696                 compact_node(nid);
 697
 698         return COMPACT_COMPLETE;
 699 }
 700
 701 /* The written value is actually unused, all memory is compacted */
 702 int sysctl_compact_memory;
 703
 704 /* This is the entry point for compacting all nodes via /proc/sys/vm */
 705 int sysctl_compaction_handler(struct ctl_table *table, int write,
 706                         void __user *buffer, size_t *length, loff_t *ppos)
 707 {
 708         if (write)
 709                 return compact_nodes();
 710
 711         return 0;
 712 }
 713
 714 int sysctl_extfrag_handler(struct ctl_table *table, int write,
 715                         void __user *buffer, size_t *length, loff_t *ppos)
 716 {
 717         proc_dointvec_minmax(table, write, buffer, length, ppos);
 718
 719         return 0;
 720 }
 721
 722 #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA)
 723 ssize_t sysfs_compact_node(struct device *dev,
 724                         struct device_attribute *attr,
 725                         const char *buf, size_t count)
 726 {
 727         compact_node(dev->id);
 728
 729         return count;
 730 }
 731 static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node);
 732
 733 int compaction_register_node(struct node *node)
 734 {
 735         return device_create_file(&node->dev, &dev_attr_compact);
 736 }
 737
 738 void compaction_unregister_node(struct node *node)
 739 {
 740         return device_remove_file(&node->dev, &dev_attr_compact);
 741 }
 742 #endif /* CONFIG_SYSFS && CONFIG_NUMA */