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         /* Account for isolated anon and file pages */
  39         unsigned long nr_anon;
  40         unsigned long nr_file;
  41
  42         unsigned int order;             /* order a direct compactor needs */
  43         int migratetype;                /* MOVABLE, RECLAIMABLE etc */
  44         struct zone *zone;
  45 };
  46
  47 static unsigned long release_freepages(struct list_head *freelist)
  48 {
  49         struct page *page, *next;
  50         unsigned long count = 0;
  51
  52         list_for_each_entry_safe(page, next, freelist, lru) {
  53                 list_del(&page->lru);
  54                 __free_page(page);
  55                 count++;
  56         }
  57
  58         return count;
  59 }
  60
  61 /* Isolate free pages onto a private freelist. Must hold zone->lock */
  62 static unsigned long isolate_freepages_block(struct zone *zone,
  63                                 unsigned long blockpfn,
  64                                 struct list_head *freelist)
  65 {
  66         unsigned long zone_end_pfn, end_pfn;
  67         int nr_scanned = 0, total_isolated = 0;
  68         struct page *cursor;
  69
  70         /* Get the last PFN we should scan for free pages at */
  71         zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
  72         end_pfn = min(blockpfn + pageblock_nr_pages, zone_end_pfn);
  73
  74         /* Find the first usable PFN in the block to initialse page cursor */
  75         for (; blockpfn < end_pfn; blockpfn++) {
  76                 if (pfn_valid_within(blockpfn))
  77                         break;
  78         }
  79         cursor = pfn_to_page(blockpfn);
  80
  81         /* Isolate free pages. This assumes the block is valid */
  82         for (; blockpfn < end_pfn; blockpfn++, cursor++) {
  83                 int isolated, i;
  84                 struct page *page = cursor;
  85
  86                 if (!pfn_valid_within(blockpfn))
  87                         continue;
  88                 nr_scanned++;
  89
  90                 if (!PageBuddy(page))
  91                         continue;
  92
  93                 /* Found a free page, break it into order-0 pages */
  94                 isolated = split_free_page(page);
  95                 total_isolated += isolated;
  96                 for (i = 0; i < isolated; i++) {
  97                         list_add(&page->lru, freelist);
  98                         page++;
  99                 }
 100
 101                 /* If a page was split, advance to the end of it */
 102                 if (isolated) {
 103                         blockpfn += isolated - 1;
 104                         cursor += isolated - 1;
 105                 }
 106         }
 107
 108         trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated);
 109         return total_isolated;
 110 }
 111
 112 /* Returns true if the page is within a block suitable for migration to */
 113 static bool suitable_migration_target(struct page *page)
 114 {
 115
 116         int migratetype = get_pageblock_migratetype(page);
 117
 118         /* Don't interfere with memory hot-remove or the min_free_kbytes blocks */
 119         if (migratetype == MIGRATE_ISOLATE || migratetype == MIGRATE_RESERVE)
 120                 return false;
 121
 122         /* If the page is a large free page, then allow migration */
 123         if (PageBuddy(page) && page_order(page) >= pageblock_order)
 124                 return true;
 125
 126         /* If the block is MIGRATE_MOVABLE, allow migration */
 127         if (migratetype == MIGRATE_MOVABLE)
 128                 return true;
 129
 130         /* Otherwise skip the block */
 131         return false;
 132 }
 133
 134 /*
 135  * Based on information in the current compact_control, find blocks
 136  * suitable for isolating free pages from and then isolate them.
 137  */
 138 static void isolate_freepages(struct zone *zone,
 139                                 struct compact_control *cc)
 140 {
 141         struct page *page;
 142         unsigned long high_pfn, low_pfn, pfn;
 143         unsigned long flags;
 144         int nr_freepages = cc->nr_freepages;
 145         struct list_head *freelist = &cc->freepages;
 146
 147         /*
 148          * Initialise the free scanner. The starting point is where we last
 149          * scanned from (or the end of the zone if starting). The low point
 150          * is the end of the pageblock the migration scanner is using.
 151          */
 152         pfn = cc->free_pfn;
 153         low_pfn = cc->migrate_pfn + pageblock_nr_pages;
 154
 155         /*
 156          * Take care that if the migration scanner is at the end of the zone
 157          * that the free scanner does not accidentally move to the next zone
 158          * in the next isolation cycle.
 159          */
 160         high_pfn = min(low_pfn, pfn);
 161
 162         /*
 163          * Isolate free pages until enough are available to migrate the
 164          * pages on cc->migratepages. We stop searching if the migrate
 165          * and free page scanners meet or enough free pages are isolated.
 166          */
 167         for (; pfn > low_pfn && cc->nr_migratepages > nr_freepages;
 168                                         pfn -= pageblock_nr_pages) {
 169                 unsigned long isolated;
 170
 171                 if (!pfn_valid(pfn))
 172                         continue;
 173
 174                 /*
 175                  * Check for overlapping nodes/zones. It's possible on some
 176                  * configurations to have a setup like
 177                  * node0 node1 node0
 178                  * i.e. it's possible that all pages within a zones range of
 179                  * pages do not belong to a single zone.
 180                  */
 181                 page = pfn_to_page(pfn);
 182                 if (page_zone(page) != zone)
 183                         continue;
 184
 185                 /* Check the block is suitable for migration */
 186                 if (!suitable_migration_target(page))
 187                         continue;
 188
 189                 /*
 190                  * Found a block suitable for isolating free pages from. Now
 191                  * we disabled interrupts, double check things are ok and
 192                  * isolate the pages. This is to minimise the time IRQs
 193                  * are disabled
 194                  */
 195                 isolated = 0;
 196                 spin_lock_irqsave(&zone->lock, flags);
 197                 if (suitable_migration_target(page)) {
 198                         isolated = isolate_freepages_block(zone, pfn, freelist);
 199                         nr_freepages += isolated;
 200                 }
 201                 spin_unlock_irqrestore(&zone->lock, flags);
 202
 203                 /*
 204                  * Record the highest PFN we isolated pages from. When next
 205                  * looking for free pages, the search will restart here as
 206                  * page migration may have returned some pages to the allocator
 207                  */
 208                 if (isolated)
 209                         high_pfn = max(high_pfn, pfn);
 210         }
 211
 212         /* split_free_page does not map the pages */
 213         list_for_each_entry(page, freelist, lru) {
 214                 arch_alloc_page(page, 0);
 215                 kernel_map_pages(page, 1, 1);
 216         }
 217
 218         cc->free_pfn = high_pfn;
 219         cc->nr_freepages = nr_freepages;
 220 }
 221
 222 /* Update the number of anon and file isolated pages in the zone */
 223 static void acct_isolated(struct zone *zone, struct compact_control *cc)
 224 {
 225         struct page *page;
 226         unsigned int count[NR_LRU_LISTS] = { 0, };
 227
 228         list_for_each_entry(page, &cc->migratepages, lru) {
 229                 int lru = page_lru_base_type(page);
 230                 count[lru]++;
 231         }
 232
 233         cc->nr_anon = count[LRU_ACTIVE_ANON] + count[LRU_INACTIVE_ANON];
 234         cc->nr_file = count[LRU_ACTIVE_FILE] + count[LRU_INACTIVE_FILE];
 235         __mod_zone_page_state(zone, NR_ISOLATED_ANON, cc->nr_anon);
 236         __mod_zone_page_state(zone, NR_ISOLATED_FILE, cc->nr_file);
 237 }
 238
 239 /* Similar to reclaim, but different enough that they don't share logic */
 240 static bool too_many_isolated(struct zone *zone)
 241 {
 242         unsigned long active, inactive, isolated;
 243
 244         inactive = zone_page_state(zone, NR_INACTIVE_FILE) +
 245                                         zone_page_state(zone, NR_INACTIVE_ANON);
 246         active = zone_page_state(zone, NR_ACTIVE_FILE) +
 247                                         zone_page_state(zone, NR_ACTIVE_ANON);
 248         isolated = zone_page_state(zone, NR_ISOLATED_FILE) +
 249                                         zone_page_state(zone, NR_ISOLATED_ANON);
 250
 251         return isolated > (inactive + active) / 2;
 252 }
 253
 254 /* possible outcome of isolate_migratepages */
 255 typedef enum {
 256         ISOLATE_ABORT,          /* Abort compaction now */
 257         ISOLATE_NONE,           /* No pages isolated, continue scanning */
 258         ISOLATE_SUCCESS,        /* Pages isolated, migrate */
 259 } isolate_migrate_t;
 260
 261 /*
 262  * Isolate all pages that can be migrated from the block pointed to by
 263  * the migrate scanner within compact_control.
 264  */
 265 static isolate_migrate_t isolate_migratepages(struct zone *zone,
 266                                         struct compact_control *cc)
 267 {
 268         unsigned long low_pfn, end_pfn;
 269         unsigned long last_pageblock_nr = 0, pageblock_nr;
 270         unsigned long nr_scanned = 0, nr_isolated = 0;
 271         struct list_head *migratelist = &cc->migratepages;
 272
 273         /* Do not scan outside zone boundaries */
 274         low_pfn = max(cc->migrate_pfn, zone->zone_start_pfn);
 275
 276         /* Only scan within a pageblock boundary */
 277         end_pfn = ALIGN(low_pfn + pageblock_nr_pages, pageblock_nr_pages);
 278
 279         /* Do not cross the free scanner or scan within a memory hole */
 280         if (end_pfn > cc->free_pfn || !pfn_valid(low_pfn)) {
 281                 cc->migrate_pfn = end_pfn;
 282                 return ISOLATE_NONE;
 283         }
 284
 285         /*
 286          * Ensure that there are not too many pages isolated from the LRU
 287          * list by either parallel reclaimers or compaction. If there are,
 288          * delay for some time until fewer pages are isolated
 289          */
 290         while (unlikely(too_many_isolated(zone))) {
 291                 /* async migration should just abort */
 292                 if (!cc->sync)
 293                         return ISOLATE_ABORT;
 294
 295                 congestion_wait(BLK_RW_ASYNC, HZ/10);
 296
 297                 if (fatal_signal_pending(current))
 298                         return ISOLATE_ABORT;
 299         }
 300
 301         /* Time to isolate some pages for migration */
 302         cond_resched();
 303         spin_lock_irq(&zone->lru_lock);
 304         for (; low_pfn < end_pfn; low_pfn++) {
 305                 struct page *page;
 306                 bool locked = true;
 307
 308                 /* give a chance to irqs before checking need_resched() */
 309                 if (!((low_pfn+1) % SWAP_CLUSTER_MAX)) {
 310                         spin_unlock_irq(&zone->lru_lock);
 311                         locked = false;
 312                 }
 313                 if (need_resched() || spin_is_contended(&zone->lru_lock)) {
 314                         if (locked)
 315                                 spin_unlock_irq(&zone->lru_lock);
 316                         cond_resched();
 317                         spin_lock_irq(&zone->lru_lock);
 318                         if (fatal_signal_pending(current))
 319                                 break;
 320                 } else if (!locked)
 321                         spin_lock_irq(&zone->lru_lock);
 322
 323                 if (!pfn_valid_within(low_pfn))
 324                         continue;
 325                 nr_scanned++;
 326
 327                 /* Get the page and skip if free */
 328                 page = pfn_to_page(low_pfn);
 329                 if (PageBuddy(page))
 330                         continue;
 331
 332                 /*
 333                  * For async migration, also only scan in MOVABLE blocks. Async
 334                  * migration is optimistic to see if the minimum amount of work
 335                  * satisfies the allocation
 336                  */
 337                 pageblock_nr = low_pfn >> pageblock_order;
 338                 if (!cc->sync && last_pageblock_nr != pageblock_nr &&
 339                                 get_pageblock_migratetype(page) != MIGRATE_MOVABLE) {
 340                         low_pfn += pageblock_nr_pages;
 341                         low_pfn = ALIGN(low_pfn, pageblock_nr_pages) - 1;
 342                         last_pageblock_nr = pageblock_nr;
 343                         continue;
 344                 }
 345
 346                 if (!PageLRU(page))
 347                         continue;
 348
 349                 /*
 350                  * PageLRU is set, and lru_lock excludes isolation,
 351                  * splitting and collapsing (collapsing has already
 352                  * happened if PageLRU is set).
 353                  */
 354                 if (PageTransHuge(page)) {
 355                         low_pfn += (1 << compound_order(page)) - 1;
 356                         continue;
 357                 }
 358
 359                 /* Try isolate the page */
 360                 if (__isolate_lru_page(page, ISOLATE_BOTH, 0) != 0)
 361                         continue;
 362
 363                 VM_BUG_ON(PageTransCompound(page));
 364
 365                 /* Successfully isolated */
 366                 del_page_from_lru_list(zone, page, page_lru(page));
 367                 list_add(&page->lru, migratelist);
 368                 cc->nr_migratepages++;
 369                 nr_isolated++;
 370
 371                 /* Avoid isolating too much */
 372                 if (cc->nr_migratepages == COMPACT_CLUSTER_MAX)
 373                         break;
 374         }
 375
 376         acct_isolated(zone, cc);
 377
 378         spin_unlock_irq(&zone->lru_lock);
 379         cc->migrate_pfn = low_pfn;
 380
 381         trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated);
 382
 383         return ISOLATE_SUCCESS;
 384 }
 385
 386 /*
 387  * This is a migrate-callback that "allocates" freepages by taking pages
 388  * from the isolated freelists in the block we are migrating to.
 389  */
 390 static struct page *compaction_alloc(struct page *migratepage,
 391                                         unsigned long data,
 392                                         int **result)
 393 {
 394         struct compact_control *cc = (struct compact_control *)data;
 395         struct page *freepage;
 396
 397         /* Isolate free pages if necessary */
 398         if (list_empty(&cc->freepages)) {
 399                 isolate_freepages(cc->zone, cc);
 400
 401                 if (list_empty(&cc->freepages))
 402                         return NULL;
 403         }
 404
 405         freepage = list_entry(cc->freepages.next, struct page, lru);
 406         list_del(&freepage->lru);
 407         cc->nr_freepages--;
 408
 409         return freepage;
 410 }
 411
 412 /*
 413  * We cannot control nr_migratepages and nr_freepages fully when migration is
 414  * running as migrate_pages() has no knowledge of compact_control. When
 415  * migration is complete, we count the number of pages on the lists by hand.
 416  */
 417 static void update_nr_listpages(struct compact_control *cc)
 418 {
 419         int nr_migratepages = 0;
 420         int nr_freepages = 0;
 421         struct page *page;
 422
 423         list_for_each_entry(page, &cc->migratepages, lru)
 424                 nr_migratepages++;
 425         list_for_each_entry(page, &cc->freepages, lru)
 426                 nr_freepages++;
 427
 428         cc->nr_migratepages = nr_migratepages;
 429         cc->nr_freepages = nr_freepages;
 430 }
 431
 432 static int compact_finished(struct zone *zone,
 433                             struct compact_control *cc)
 434 {
 435         unsigned int order;
 436         unsigned long watermark;
 437
 438         if (fatal_signal_pending(current))
 439                 return COMPACT_PARTIAL;
 440
 441         /* Compaction run completes if the migrate and free scanner meet */
 442         if (cc->free_pfn <= cc->migrate_pfn)
 443                 return COMPACT_COMPLETE;
 444
 445         /*
 446          * order == -1 is expected when compacting via
 447          * /proc/sys/vm/compact_memory
 448          */
 449         if (cc->order == -1)
 450                 return COMPACT_CONTINUE;
 451
 452         /* Compaction run is not finished if the watermark is not met */
 453         watermark = low_wmark_pages(zone);
 454         watermark += (1 << cc->order);
 455
 456         if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0))
 457                 return COMPACT_CONTINUE;
 458
 459         /* Direct compactor: Is a suitable page free? */
 460         for (order = cc->order; order < MAX_ORDER; order++) {
 461                 /* Job done if page is free of the right migratetype */
 462                 if (!list_empty(&zone->free_area[order].free_list[cc->migratetype]))
 463                         return COMPACT_PARTIAL;
 464
 465                 /* Job done if allocation would set block type */
 466                 if (order >= pageblock_order && zone->free_area[order].nr_free)
 467                         return COMPACT_PARTIAL;
 468         }
 469
 470         return COMPACT_CONTINUE;
 471 }
 472
 473 /*
 474  * compaction_suitable: Is this suitable to run compaction on this zone now?
 475  * Returns
 476  *   COMPACT_SKIPPED  - If there are too few free pages for compaction
 477  *   COMPACT_PARTIAL  - If the allocation would succeed without compaction
 478  *   COMPACT_CONTINUE - If compaction should run now
 479  */
 480 unsigned long compaction_suitable(struct zone *zone, int order)
 481 {
 482         int fragindex;
 483         unsigned long watermark;
 484
 485         /*
 486          * order == -1 is expected when compacting via
 487          * /proc/sys/vm/compact_memory
 488          */
 489         if (order == -1)
 490                 return COMPACT_CONTINUE;
 491
 492         /*
 493          * Watermarks for order-0 must be met for compaction. Note the 2UL.
 494          * This is because during migration, copies of pages need to be
 495          * allocated and for a short time, the footprint is higher
 496          */
 497         watermark = low_wmark_pages(zone) + (2UL << order);
 498         if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
 499                 return COMPACT_SKIPPED;
 500
 501         /*
 502          * fragmentation index determines if allocation failures are due to
 503          * low memory or external fragmentation
 504          *
 505          * index of -1000 implies allocations might succeed depending on
 506          * watermarks
 507          * index towards 0 implies failure is due to lack of memory
 508          * index towards 1000 implies failure is due to fragmentation
 509          *
 510          * Only compact if a failure would be due to fragmentation.
 511          */
 512         fragindex = fragmentation_index(zone, order);
 513         if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold)
 514                 return COMPACT_SKIPPED;
 515
 516         if (fragindex == -1000 && zone_watermark_ok(zone, order, watermark,
 517             0, 0))
 518                 return COMPACT_PARTIAL;
 519
 520         return COMPACT_CONTINUE;
 521 }
 522
 523 static int compact_zone(struct zone *zone, struct compact_control *cc)
 524 {
 525         int ret;
 526
 527         ret = compaction_suitable(zone, cc->order);
 528         switch (ret) {
 529         case COMPACT_PARTIAL:
 530         case COMPACT_SKIPPED:
 531                 /* Compaction is likely to fail */
 532                 return ret;
 533         case COMPACT_CONTINUE:
 534                 /* Fall through to compaction */
 535                 ;
 536         }
 537
 538         /* Setup to move all movable pages to the end of the zone */
 539         cc->migrate_pfn = zone->zone_start_pfn;
 540         cc->free_pfn = cc->migrate_pfn + zone->spanned_pages;
 541         cc->free_pfn &= ~(pageblock_nr_pages-1);
 542
 543         migrate_prep_local();
 544
 545         while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) {
 546                 unsigned long nr_migrate, nr_remaining;
 547                 int err;
 548
 549                 switch (isolate_migratepages(zone, cc)) {
 550                 case ISOLATE_ABORT:
 551                         ret = COMPACT_PARTIAL;
 552                         goto out;
 553                 case ISOLATE_NONE:
 554                         continue;
 555                 case ISOLATE_SUCCESS:
 556                         ;
 557                 }
 558
 559                 nr_migrate = cc->nr_migratepages;
 560                 err = migrate_pages(&cc->migratepages, compaction_alloc,
 561                                 (unsigned long)cc, false,
 562                                 cc->sync);
 563                 update_nr_listpages(cc);
 564                 nr_remaining = cc->nr_migratepages;
 565
 566                 count_vm_event(COMPACTBLOCKS);
 567                 count_vm_events(COMPACTPAGES, nr_migrate - nr_remaining);
 568                 if (nr_remaining)
 569                         count_vm_events(COMPACTPAGEFAILED, nr_remaining);
 570                 trace_mm_compaction_migratepages(nr_migrate - nr_remaining,
 571                                                 nr_remaining);
 572
 573                 /* Release LRU pages not migrated */
 574                 if (err) {
 575                         putback_lru_pages(&cc->migratepages);
 576                         cc->nr_migratepages = 0;
 577                 }
 578
 579         }
 580
 581 out:
 582         /* Release free pages and check accounting */
 583         cc->nr_freepages -= release_freepages(&cc->freepages);
 584         VM_BUG_ON(cc->nr_freepages != 0);
 585
 586         return ret;
 587 }
 588
 589 unsigned long compact_zone_order(struct zone *zone,
 590                                  int order, gfp_t gfp_mask,
 591                                  bool sync)
 592 {
 593         struct compact_control cc = {
 594                 .nr_freepages = 0,
 595                 .nr_migratepages = 0,
 596                 .order = order,
 597                 .migratetype = allocflags_to_migratetype(gfp_mask),
 598                 .zone = zone,
 599                 .sync = sync,
 600         };
 601         INIT_LIST_HEAD(&cc.freepages);
 602         INIT_LIST_HEAD(&cc.migratepages);
 603
 604         return compact_zone(zone, &cc);
 605 }
 606
 607 int sysctl_extfrag_threshold = 500;
 608
 609 /**
 610  * try_to_compact_pages - Direct compact to satisfy a high-order allocation
 611  * @zonelist: The zonelist used for the current allocation
 612  * @order: The order of the current allocation
 613  * @gfp_mask: The GFP mask of the current allocation
 614  * @nodemask: The allowed nodes to allocate from
 615  * @sync: Whether migration is synchronous or not
 616  *
 617  * This is the main entry point for direct page compaction.
 618  */
 619 unsigned long try_to_compact_pages(struct zonelist *zonelist,
 620                         int order, gfp_t gfp_mask, nodemask_t *nodemask,
 621                         bool sync)
 622 {
 623         enum zone_type high_zoneidx = gfp_zone(gfp_mask);
 624         int may_enter_fs = gfp_mask & __GFP_FS;
 625         int may_perform_io = gfp_mask & __GFP_IO;
 626         struct zoneref *z;
 627         struct zone *zone;
 628         int rc = COMPACT_SKIPPED;
 629
 630         /*
 631          * Check whether it is worth even starting compaction. The order check is
 632          * made because an assumption is made that the page allocator can satisfy
 633          * the "cheaper" orders without taking special steps
 634          */
 635         if (!order || !may_enter_fs || !may_perform_io)
 636                 return rc;
 637
 638         count_vm_event(COMPACTSTALL);
 639
 640         /* Compact each zone in the list */
 641         for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx,
 642                                                                 nodemask) {
 643                 int status;
 644
 645                 status = compact_zone_order(zone, order, gfp_mask, sync);
 646                 rc = max(status, rc);
 647
 648                 /* If a normal allocation would succeed, stop compacting */
 649                 if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0, 0))
 650                         break;
 651         }
 652
 653         return rc;
 654 }
 655
 656
 657 /* Compact all zones within a node */
 658 static int compact_node(int nid)
 659 {
 660         int zoneid;
 661         pg_data_t *pgdat;
 662         struct zone *zone;
 663
 664         if (nid < 0 || nid >= nr_node_ids || !node_online(nid))
 665                 return -EINVAL;
 666         pgdat = NODE_DATA(nid);
 667
 668         /* Flush pending updates to the LRU lists */
 669         lru_add_drain_all();
 670
 671         for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) {
 672                 struct compact_control cc = {
 673                         .nr_freepages = 0,
 674                         .nr_migratepages = 0,
 675                         .order = -1,
 676                 };
 677
 678                 zone = &pgdat->node_zones[zoneid];
 679                 if (!populated_zone(zone))
 680                         continue;
 681
 682                 cc.zone = zone;
 683                 INIT_LIST_HEAD(&cc.freepages);
 684                 INIT_LIST_HEAD(&cc.migratepages);
 685
 686                 compact_zone(zone, &cc);
 687
 688                 VM_BUG_ON(!list_empty(&cc.freepages));
 689                 VM_BUG_ON(!list_empty(&cc.migratepages));
 690         }
 691
 692         return 0;
 693 }
 694
 695 /* Compact all nodes in the system */
 696 static int compact_nodes(void)
 697 {
 698         int nid;
 699
 700         for_each_online_node(nid)
 701                 compact_node(nid);
 702
 703         return COMPACT_COMPLETE;
 704 }
 705
 706 /* The written value is actually unused, all memory is compacted */
 707 int sysctl_compact_memory;
 708
 709 /* This is the entry point for compacting all nodes via /proc/sys/vm */
 710 int sysctl_compaction_handler(struct ctl_table *table, int write,
 711                         void __user *buffer, size_t *length, loff_t *ppos)
 712 {
 713         if (write)
 714                 return compact_nodes();
 715
 716         return 0;
 717 }
 718
 719 int sysctl_extfrag_handler(struct ctl_table *table, int write,
 720                         void __user *buffer, size_t *length, loff_t *ppos)
 721 {
 722         proc_dointvec_minmax(table, write, buffer, length, ppos);
 723
 724         return 0;
 725 }
 726
 727 #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA)
 728 ssize_t sysfs_compact_node(struct sys_device *dev,
 729                         struct sysdev_attribute *attr,
 730                         const char *buf, size_t count)
 731 {
 732         compact_node(dev->id);
 733
 734         return count;
 735 }
 736 static SYSDEV_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node);
 737
 738 int compaction_register_node(struct node *node)
 739 {
 740         return sysdev_create_file(&node->sysdev, &attr_compact);
 741 }
 742
 743 void compaction_unregister_node(struct node *node)
 744 {
 745         return sysdev_remove_file(&node->sysdev, &attr_compact);
 746 }
 747 #endif /* CONFIG_SYSFS && CONFIG_NUMA */