mm/swap.c

   1 /*
   2  *  linux/mm/swap.c
   3  *
   4  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
   5  */
   6
   7 /*
   8  * This file contains the default values for the operation of the
   9  * Linux VM subsystem. Fine-tuning documentation can be found in
  10  * Documentation/sysctl/vm.txt.
  11  * Started 18.12.91
  12  * Swap aging added 23.2.95, Stephen Tweedie.
  13  * Buffermem limits added 12.3.98, Rik van Riel.
  14  */
  15
  16 #include <linux/mm.h>
  17 #include <linux/sched.h>
  18 #include <linux/kernel_stat.h>
  19 #include <linux/swap.h>
  20 #include <linux/mman.h>
  21 #include <linux/pagemap.h>
  22 #include <linux/pagevec.h>
  23 #include <linux/init.h>
  24 #include <linux/module.h>
  25 #include <linux/mm_inline.h>
  26 #include <linux/buffer_head.h>  /* for try_to_release_page() */
  27 #include <linux/percpu_counter.h>
  28 #include <linux/percpu.h>
  29 #include <linux/cpu.h>
  30 #include <linux/notifier.h>
  31 #include <linux/backing-dev.h>
  32 #include <linux/memcontrol.h>
  33
  34 /* How many pages do we try to swap or page in/out together? */
  35 int page_cluster;
  36
  37 static DEFINE_PER_CPU(struct pagevec[NR_LRU_LISTS], lru_add_pvecs);
  38 static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
  39
  40 /*
  41  * This path almost never happens for VM activity - pages are normally
  42  * freed via pagevecs.  But it gets used by networking.
  43  */
  44 static void __page_cache_release(struct page *page)
  45 {
  46         if (PageLRU(page)) {
  47                 unsigned long flags;
  48                 struct zone *zone = page_zone(page);
  49
  50                 spin_lock_irqsave(&zone->lru_lock, flags);
  51                 VM_BUG_ON(!PageLRU(page));
  52                 __ClearPageLRU(page);
  53                 del_page_from_lru(zone, page);
  54                 spin_unlock_irqrestore(&zone->lru_lock, flags);
  55         }
  56         free_hot_page(page);
  57 }
  58
  59 static void put_compound_page(struct page *page)
  60 {
  61         page = compound_head(page);
  62         if (put_page_testzero(page)) {
  63                 compound_page_dtor *dtor;
  64
  65                 dtor = get_compound_page_dtor(page);
  66                 (*dtor)(page);
  67         }
  68 }
  69
  70 void put_page(struct page *page)
  71 {
  72         if (unlikely(PageCompound(page)))
  73                 put_compound_page(page);
  74         else if (put_page_testzero(page))
  75                 __page_cache_release(page);
  76 }
  77 EXPORT_SYMBOL(put_page);
  78
  79 /**
  80  * put_pages_list() - release a list of pages
  81  * @pages: list of pages threaded on page->lru
  82  *
  83  * Release a list of pages which are strung together on page.lru.  Currently
  84  * used by read_cache_pages() and related error recovery code.
  85  */
  86 void put_pages_list(struct list_head *pages)
  87 {
  88         while (!list_empty(pages)) {
  89                 struct page *victim;
  90
  91                 victim = list_entry(pages->prev, struct page, lru);
  92                 list_del(&victim->lru);
  93                 page_cache_release(victim);
  94         }
  95 }
  96 EXPORT_SYMBOL(put_pages_list);
  97
  98 /*
  99  * pagevec_move_tail() must be called with IRQ disabled.
 100  * Otherwise this may cause nasty races.
 101  */
 102 static void pagevec_move_tail(struct pagevec *pvec)
 103 {
 104         int i;
 105         int pgmoved = 0;
 106         struct zone *zone = NULL;
 107
 108         for (i = 0; i < pagevec_count(pvec); i++) {
 109                 struct page *page = pvec->pages[i];
 110                 struct zone *pagezone = page_zone(page);
 111
 112                 if (pagezone != zone) {
 113                         if (zone)
 114                                 spin_unlock(&zone->lru_lock);
 115                         zone = pagezone;
 116                         spin_lock(&zone->lru_lock);
 117                 }
 118                 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
 119                         int lru = page_is_file_cache(page);
 120                         list_move_tail(&page->lru, &zone->lru[lru].list);
 121                         pgmoved++;
 122                 }
 123         }
 124         if (zone)
 125                 spin_unlock(&zone->lru_lock);
 126         __count_vm_events(PGROTATED, pgmoved);
 127         release_pages(pvec->pages, pvec->nr, pvec->cold);
 128         pagevec_reinit(pvec);
 129 }
 130
 131 /*
 132  * Writeback is about to end against a page which has been marked for immediate
 133  * reclaim.  If it still appears to be reclaimable, move it to the tail of the
 134  * inactive list.
 135  */
 136 void  rotate_reclaimable_page(struct page *page)
 137 {
 138         if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
 139             !PageUnevictable(page) && PageLRU(page)) {
 140                 struct pagevec *pvec;
 141                 unsigned long flags;
 142
 143                 page_cache_get(page);
 144                 local_irq_save(flags);
 145                 pvec = &__get_cpu_var(lru_rotate_pvecs);
 146                 if (!pagevec_add(pvec, page))
 147                         pagevec_move_tail(pvec);
 148                 local_irq_restore(flags);
 149         }
 150 }
 151
 152 /*
 153  * FIXME: speed this up?
 154  */
 155 void activate_page(struct page *page)
 156 {
 157         struct zone *zone = page_zone(page);
 158
 159         spin_lock_irq(&zone->lru_lock);
 160         if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
 161                 int file = page_is_file_cache(page);
 162                 int lru = LRU_BASE + file;
 163                 del_page_from_lru_list(zone, page, lru);
 164
 165                 SetPageActive(page);
 166                 lru += LRU_ACTIVE;
 167                 add_page_to_lru_list(zone, page, lru);
 168                 __count_vm_event(PGACTIVATE);
 169                 mem_cgroup_move_lists(page, lru);
 170
 171                 zone->recent_rotated[!!file]++;
 172                 zone->recent_scanned[!!file]++;
 173         }
 174         spin_unlock_irq(&zone->lru_lock);
 175 }
 176
 177 /*
 178  * Mark a page as having seen activity.
 179  *
 180  * inactive,unreferenced        ->      inactive,referenced
 181  * inactive,referenced          ->      active,unreferenced
 182  * active,unreferenced          ->      active,referenced
 183  */
 184 void mark_page_accessed(struct page *page)
 185 {
 186         if (!PageActive(page) && !PageUnevictable(page) &&
 187                         PageReferenced(page) && PageLRU(page)) {
 188                 activate_page(page);
 189                 ClearPageReferenced(page);
 190         } else if (!PageReferenced(page)) {
 191                 SetPageReferenced(page);
 192         }
 193 }
 194
 195 EXPORT_SYMBOL(mark_page_accessed);
 196
 197 void __lru_cache_add(struct page *page, enum lru_list lru)
 198 {
 199         struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru];
 200
 201         page_cache_get(page);
 202         if (!pagevec_add(pvec, page))
 203                 ____pagevec_lru_add(pvec, lru);
 204         put_cpu_var(lru_add_pvecs);
 205 }
 206
 207 /**
 208  * lru_cache_add_lru - add a page to a page list
 209  * @page: the page to be added to the LRU.
 210  * @lru: the LRU list to which the page is added.
 211  */
 212 void lru_cache_add_lru(struct page *page, enum lru_list lru)
 213 {
 214         if (PageActive(page)) {
 215                 VM_BUG_ON(PageUnevictable(page));
 216                 ClearPageActive(page);
 217         } else if (PageUnevictable(page)) {
 218                 VM_BUG_ON(PageActive(page));
 219                 ClearPageUnevictable(page);
 220         }
 221
 222         VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page));
 223         __lru_cache_add(page, lru);
 224 }
 225
 226 /**
 227  * add_page_to_unevictable_list - add a page to the unevictable list
 228  * @page:  the page to be added to the unevictable list
 229  *
 230  * Add page directly to its zone's unevictable list.  To avoid races with
 231  * tasks that might be making the page evictable, through eg. munlock,
 232  * munmap or exit, while it's not on the lru, we want to add the page
 233  * while it's locked or otherwise "invisible" to other tasks.  This is
 234  * difficult to do when using the pagevec cache, so bypass that.
 235  */
 236 void add_page_to_unevictable_list(struct page *page)
 237 {
 238         struct zone *zone = page_zone(page);
 239
 240         spin_lock_irq(&zone->lru_lock);
 241         SetPageUnevictable(page);
 242         SetPageLRU(page);
 243         add_page_to_lru_list(zone, page, LRU_UNEVICTABLE);
 244         spin_unlock_irq(&zone->lru_lock);
 245 }
 246
 247 /*
 248  * Drain pages out of the cpu's pagevecs.
 249  * Either "cpu" is the current CPU, and preemption has already been
 250  * disabled; or "cpu" is being hot-unplugged, and is already dead.
 251  */
 252 static void drain_cpu_pagevecs(int cpu)
 253 {
 254         struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu);
 255         struct pagevec *pvec;
 256         int lru;
 257
 258         for_each_lru(lru) {
 259                 pvec = &pvecs[lru - LRU_BASE];
 260                 if (pagevec_count(pvec))
 261                         ____pagevec_lru_add(pvec, lru);
 262         }
 263
 264         pvec = &per_cpu(lru_rotate_pvecs, cpu);
 265         if (pagevec_count(pvec)) {
 266                 unsigned long flags;
 267
 268                 /* No harm done if a racing interrupt already did this */
 269                 local_irq_save(flags);
 270                 pagevec_move_tail(pvec);
 271                 local_irq_restore(flags);
 272         }
 273 }
 274
 275 void lru_add_drain(void)
 276 {
 277         drain_cpu_pagevecs(get_cpu());
 278         put_cpu();
 279 }
 280
 281 #if defined(CONFIG_NUMA) || defined(CONFIG_UNEVICTABLE_LRU)
 282 static void lru_add_drain_per_cpu(struct work_struct *dummy)
 283 {
 284         lru_add_drain();
 285 }
 286
 287 /*
 288  * Returns 0 for success
 289  */
 290 int lru_add_drain_all(void)
 291 {
 292         return schedule_on_each_cpu(lru_add_drain_per_cpu);
 293 }
 294
 295 #else
 296
 297 /*
 298  * Returns 0 for success
 299  */
 300 int lru_add_drain_all(void)
 301 {
 302         lru_add_drain();
 303         return 0;
 304 }
 305 #endif
 306
 307 /*
 308  * Batched page_cache_release().  Decrement the reference count on all the
 309  * passed pages.  If it fell to zero then remove the page from the LRU and
 310  * free it.
 311  *
 312  * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
 313  * for the remainder of the operation.
 314  *
 315  * The locking in this function is against shrink_inactive_list(): we recheck
 316  * the page count inside the lock to see whether shrink_inactive_list()
 317  * grabbed the page via the LRU.  If it did, give up: shrink_inactive_list()
 318  * will free it.
 319  */
 320 void release_pages(struct page **pages, int nr, int cold)
 321 {
 322         int i;
 323         struct pagevec pages_to_free;
 324         struct zone *zone = NULL;
 325         unsigned long uninitialized_var(flags);
 326
 327         pagevec_init(&pages_to_free, cold);
 328         for (i = 0; i < nr; i++) {
 329                 struct page *page = pages[i];
 330
 331                 if (unlikely(PageCompound(page))) {
 332                         if (zone) {
 333                                 spin_unlock_irqrestore(&zone->lru_lock, flags);
 334                                 zone = NULL;
 335                         }
 336                         put_compound_page(page);
 337                         continue;
 338                 }
 339
 340                 if (!put_page_testzero(page))
 341                         continue;
 342
 343                 if (PageLRU(page)) {
 344                         struct zone *pagezone = page_zone(page);
 345
 346                         if (pagezone != zone) {
 347                                 if (zone)
 348                                         spin_unlock_irqrestore(&zone->lru_lock,
 349                                                                         flags);
 350                                 zone = pagezone;
 351                                 spin_lock_irqsave(&zone->lru_lock, flags);
 352                         }
 353                         VM_BUG_ON(!PageLRU(page));
 354                         __ClearPageLRU(page);
 355                         del_page_from_lru(zone, page);
 356                 }
 357
 358                 if (!pagevec_add(&pages_to_free, page)) {
 359                         if (zone) {
 360                                 spin_unlock_irqrestore(&zone->lru_lock, flags);
 361                                 zone = NULL;
 362                         }
 363                         __pagevec_free(&pages_to_free);
 364                         pagevec_reinit(&pages_to_free);
 365                 }
 366         }
 367         if (zone)
 368                 spin_unlock_irqrestore(&zone->lru_lock, flags);
 369
 370         pagevec_free(&pages_to_free);
 371 }
 372
 373 /*
 374  * The pages which we're about to release may be in the deferred lru-addition
 375  * queues.  That would prevent them from really being freed right now.  That's
 376  * OK from a correctness point of view but is inefficient - those pages may be
 377  * cache-warm and we want to give them back to the page allocator ASAP.
 378  *
 379  * So __pagevec_release() will drain those queues here.  __pagevec_lru_add()
 380  * and __pagevec_lru_add_active() call release_pages() directly to avoid
 381  * mutual recursion.
 382  */
 383 void __pagevec_release(struct pagevec *pvec)
 384 {
 385         lru_add_drain();
 386         release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
 387         pagevec_reinit(pvec);
 388 }
 389
 390 EXPORT_SYMBOL(__pagevec_release);
 391
 392 /*
 393  * pagevec_release() for pages which are known to not be on the LRU
 394  *
 395  * This function reinitialises the caller's pagevec.
 396  */
 397 void __pagevec_release_nonlru(struct pagevec *pvec)
 398 {
 399         int i;
 400         struct pagevec pages_to_free;
 401
 402         pagevec_init(&pages_to_free, pvec->cold);
 403         for (i = 0; i < pagevec_count(pvec); i++) {
 404                 struct page *page = pvec->pages[i];
 405
 406                 VM_BUG_ON(PageLRU(page));
 407                 if (put_page_testzero(page))
 408                         pagevec_add(&pages_to_free, page);
 409         }
 410         pagevec_free(&pages_to_free);
 411         pagevec_reinit(pvec);
 412 }
 413
 414 /*
 415  * Add the passed pages to the LRU, then drop the caller's refcount
 416  * on them.  Reinitialises the caller's pagevec.
 417  */
 418 void ____pagevec_lru_add(struct pagevec *pvec, enum lru_list lru)
 419 {
 420         int i;
 421         struct zone *zone = NULL;
 422         VM_BUG_ON(is_unevictable_lru(lru));
 423
 424         for (i = 0; i < pagevec_count(pvec); i++) {
 425                 struct page *page = pvec->pages[i];
 426                 struct zone *pagezone = page_zone(page);
 427
 428                 if (pagezone != zone) {
 429                         if (zone)
 430                                 spin_unlock_irq(&zone->lru_lock);
 431                         zone = pagezone;
 432                         spin_lock_irq(&zone->lru_lock);
 433                 }
 434                 VM_BUG_ON(PageActive(page));
 435                 VM_BUG_ON(PageUnevictable(page));
 436                 VM_BUG_ON(PageLRU(page));
 437                 SetPageLRU(page);
 438                 if (is_active_lru(lru))
 439                         SetPageActive(page);
 440                 add_page_to_lru_list(zone, page, lru);
 441         }
 442         if (zone)
 443                 spin_unlock_irq(&zone->lru_lock);
 444         release_pages(pvec->pages, pvec->nr, pvec->cold);
 445         pagevec_reinit(pvec);
 446 }
 447
 448 EXPORT_SYMBOL(____pagevec_lru_add);
 449
 450 /*
 451  * Try to drop buffers from the pages in a pagevec
 452  */
 453 void pagevec_strip(struct pagevec *pvec)
 454 {
 455         int i;
 456
 457         for (i = 0; i < pagevec_count(pvec); i++) {
 458                 struct page *page = pvec->pages[i];
 459
 460                 if (PagePrivate(page) && trylock_page(page)) {
 461                         if (PagePrivate(page))
 462                                 try_to_release_page(page, 0);
 463                         unlock_page(page);
 464                 }
 465         }
 466 }
 467
 468 /**
 469  * pagevec_swap_free - try to free swap space from the pages in a pagevec
 470  * @pvec: pagevec with swapcache pages to free the swap space of
 471  *
 472  * The caller needs to hold an extra reference to each page and
 473  * not hold the page lock on the pages.  This function uses a
 474  * trylock on the page lock so it may not always free the swap
 475  * space associated with a page.
 476  */
 477 void pagevec_swap_free(struct pagevec *pvec)
 478 {
 479         int i;
 480
 481         for (i = 0; i < pagevec_count(pvec); i++) {
 482                 struct page *page = pvec->pages[i];
 483
 484                 if (PageSwapCache(page) && trylock_page(page)) {
 485                         if (PageSwapCache(page))
 486                                 remove_exclusive_swap_page_ref(page);
 487                         unlock_page(page);
 488                 }
 489         }
 490 }
 491
 492 /**
 493  * pagevec_lookup - gang pagecache lookup
 494  * @pvec:       Where the resulting pages are placed
 495  * @mapping:    The address_space to search
 496  * @start:      The starting page index
 497  * @nr_pages:   The maximum number of pages
 498  *
 499  * pagevec_lookup() will search for and return a group of up to @nr_pages pages
 500  * in the mapping.  The pages are placed in @pvec.  pagevec_lookup() takes a
 501  * reference against the pages in @pvec.
 502  *
 503  * The search returns a group of mapping-contiguous pages with ascending
 504  * indexes.  There may be holes in the indices due to not-present pages.
 505  *
 506  * pagevec_lookup() returns the number of pages which were found.
 507  */
 508 unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
 509                 pgoff_t start, unsigned nr_pages)
 510 {
 511         pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
 512         return pagevec_count(pvec);
 513 }
 514
 515 EXPORT_SYMBOL(pagevec_lookup);
 516
 517 unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
 518                 pgoff_t *index, int tag, unsigned nr_pages)
 519 {
 520         pvec->nr = find_get_pages_tag(mapping, index, tag,
 521                                         nr_pages, pvec->pages);
 522         return pagevec_count(pvec);
 523 }
 524
 525 EXPORT_SYMBOL(pagevec_lookup_tag);
 526
 527 #ifdef CONFIG_SMP
 528 /*
 529  * We tolerate a little inaccuracy to avoid ping-ponging the counter between
 530  * CPUs
 531  */
 532 #define ACCT_THRESHOLD  max(16, NR_CPUS * 2)
 533
 534 static DEFINE_PER_CPU(long, committed_space);
 535
 536 void vm_acct_memory(long pages)
 537 {
 538         long *local;
 539
 540         preempt_disable();
 541         local = &__get_cpu_var(committed_space);
 542         *local += pages;
 543         if (*local > ACCT_THRESHOLD || *local < -ACCT_THRESHOLD) {
 544                 atomic_long_add(*local, &vm_committed_space);
 545                 *local = 0;
 546         }
 547         preempt_enable();
 548 }
 549
 550 #ifdef CONFIG_HOTPLUG_CPU
 551
 552 /* Drop the CPU's cached committed space back into the central pool. */
 553 static int cpu_swap_callback(struct notifier_block *nfb,
 554                              unsigned long action,
 555                              void *hcpu)
 556 {
 557         long *committed;
 558
 559         committed = &per_cpu(committed_space, (long)hcpu);
 560         if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
 561                 atomic_long_add(*committed, &vm_committed_space);
 562                 *committed = 0;
 563                 drain_cpu_pagevecs((long)hcpu);
 564         }
 565         return NOTIFY_OK;
 566 }
 567 #endif /* CONFIG_HOTPLUG_CPU */
 568 #endif /* CONFIG_SMP */
 569
 570 /*
 571  * Perform any setup for the swap system
 572  */
 573 void __init swap_setup(void)
 574 {
 575         unsigned long megs = num_physpages >> (20 - PAGE_SHIFT);
 576
 577 #ifdef CONFIG_SWAP
 578         bdi_init(swapper_space.backing_dev_info);
 579 #endif
 580
 581         /* Use a smaller cluster for small-memory machines */
 582         if (megs < 16)
 583                 page_cluster = 2;
 584         else
 585                 page_cluster = 3;
 586         /*
 587          * Right now other parts of the system means that we
 588          * _really_ don't want to cluster much more
 589          */
 590 #ifdef CONFIG_HOTPLUG_CPU
 591         hotcpu_notifier(cpu_swap_callback, 0);
 592 #endif
 593 }