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 #include <linux/gfp.h>
  34
  35 #include "internal.h"
  36
  37 /* How many pages do we try to swap or page in/out together? */
  38 int page_cluster;
  39
  40 static DEFINE_PER_CPU(struct pagevec[NR_LRU_LISTS], lru_add_pvecs);
  41 static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
  42
  43 /*
  44  * This path almost never happens for VM activity - pages are normally
  45  * freed via pagevecs.  But it gets used by networking.
  46  */
  47 static void __page_cache_release(struct page *page)
  48 {
  49         if (PageLRU(page)) {
  50                 unsigned long flags;
  51                 struct zone *zone = page_zone(page);
  52
  53                 spin_lock_irqsave(&zone->lru_lock, flags);
  54                 VM_BUG_ON(!PageLRU(page));
  55                 __ClearPageLRU(page);
  56                 del_page_from_lru(zone, page);
  57                 spin_unlock_irqrestore(&zone->lru_lock, flags);
  58         }
  59         free_hot_cold_page(page, 0);
  60 }
  61
  62 static void put_compound_page(struct page *page)
  63 {
  64         page = compound_head(page);
  65         if (put_page_testzero(page)) {
  66                 compound_page_dtor *dtor;
  67
  68                 dtor = get_compound_page_dtor(page);
  69                 (*dtor)(page);
  70         }
  71 }
  72
  73 void put_page(struct page *page)
  74 {
  75         if (unlikely(PageCompound(page)))
  76                 put_compound_page(page);
  77         else if (put_page_testzero(page))
  78                 __page_cache_release(page);
  79 }
  80 EXPORT_SYMBOL(put_page);
  81
  82 /**
  83  * put_pages_list() - release a list of pages
  84  * @pages: list of pages threaded on page->lru
  85  *
  86  * Release a list of pages which are strung together on page.lru.  Currently
  87  * used by read_cache_pages() and related error recovery code.
  88  */
  89 void put_pages_list(struct list_head *pages)
  90 {
  91         while (!list_empty(pages)) {
  92                 struct page *victim;
  93
  94                 victim = list_entry(pages->prev, struct page, lru);
  95                 list_del(&victim->lru);
  96                 page_cache_release(victim);
  97         }
  98 }
  99 EXPORT_SYMBOL(put_pages_list);
 100
 101 /*
 102  * pagevec_move_tail() must be called with IRQ disabled.
 103  * Otherwise this may cause nasty races.
 104  */
 105 static void pagevec_move_tail(struct pagevec *pvec)
 106 {
 107         int i;
 108         int pgmoved = 0;
 109         struct zone *zone = NULL;
 110
 111         for (i = 0; i < pagevec_count(pvec); i++) {
 112                 struct page *page = pvec->pages[i];
 113                 struct zone *pagezone = page_zone(page);
 114
 115                 if (pagezone != zone) {
 116                         if (zone)
 117                                 spin_unlock(&zone->lru_lock);
 118                         zone = pagezone;
 119                         spin_lock(&zone->lru_lock);
 120                 }
 121                 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
 122                         int lru = page_lru_base_type(page);
 123                         list_move_tail(&page->lru, &zone->lru[lru].list);
 124                         pgmoved++;
 125                 }
 126         }
 127         if (zone)
 128                 spin_unlock(&zone->lru_lock);
 129         __count_vm_events(PGROTATED, pgmoved);
 130         release_pages(pvec->pages, pvec->nr, pvec->cold);
 131         pagevec_reinit(pvec);
 132 }
 133
 134 /*
 135  * Writeback is about to end against a page which has been marked for immediate
 136  * reclaim.  If it still appears to be reclaimable, move it to the tail of the
 137  * inactive list.
 138  */
 139 void  rotate_reclaimable_page(struct page *page)
 140 {
 141         if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
 142             !PageUnevictable(page) && PageLRU(page)) {
 143                 struct pagevec *pvec;
 144                 unsigned long flags;
 145
 146                 page_cache_get(page);
 147                 local_irq_save(flags);
 148                 pvec = &__get_cpu_var(lru_rotate_pvecs);
 149                 if (!pagevec_add(pvec, page))
 150                         pagevec_move_tail(pvec);
 151                 local_irq_restore(flags);
 152         }
 153 }
 154
 155 static void update_page_reclaim_stat(struct zone *zone, struct page *page,
 156                                      int file, int rotated)
 157 {
 158         struct zone_reclaim_stat *reclaim_stat = &zone->reclaim_stat;
 159         struct zone_reclaim_stat *memcg_reclaim_stat;
 160
 161         memcg_reclaim_stat = mem_cgroup_get_reclaim_stat_from_page(page);
 162
 163         reclaim_stat->recent_scanned[file]++;
 164         if (rotated)
 165                 reclaim_stat->recent_rotated[file]++;
 166
 167         if (!memcg_reclaim_stat)
 168                 return;
 169
 170         memcg_reclaim_stat->recent_scanned[file]++;
 171         if (rotated)
 172                 memcg_reclaim_stat->recent_rotated[file]++;
 173 }
 174
 175 /*
 176  * FIXME: speed this up?
 177  */
 178 void activate_page(struct page *page)
 179 {
 180         struct zone *zone = page_zone(page);
 181
 182         spin_lock_irq(&zone->lru_lock);
 183         if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
 184                 int file = page_is_file_cache(page);
 185                 int lru = page_lru_base_type(page);
 186                 del_page_from_lru_list(zone, page, lru);
 187
 188                 SetPageActive(page);
 189                 lru += LRU_ACTIVE;
 190                 add_page_to_lru_list(zone, page, lru);
 191                 __count_vm_event(PGACTIVATE);
 192
 193                 update_page_reclaim_stat(zone, page, file, 1);
 194         }
 195         spin_unlock_irq(&zone->lru_lock);
 196 }
 197
 198 /*
 199  * Mark a page as having seen activity.
 200  *
 201  * inactive,unreferenced        ->      inactive,referenced
 202  * inactive,referenced          ->      active,unreferenced
 203  * active,unreferenced          ->      active,referenced
 204  */
 205 void mark_page_accessed(struct page *page)
 206 {
 207         if (!PageActive(page) && !PageUnevictable(page) &&
 208                         PageReferenced(page) && PageLRU(page)) {
 209                 activate_page(page);
 210                 ClearPageReferenced(page);
 211         } else if (!PageReferenced(page)) {
 212                 SetPageReferenced(page);
 213         }
 214 }
 215
 216 EXPORT_SYMBOL(mark_page_accessed);
 217
 218 void __lru_cache_add(struct page *page, enum lru_list lru)
 219 {
 220         struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru];
 221
 222         page_cache_get(page);
 223         if (!pagevec_add(pvec, page))
 224                 ____pagevec_lru_add(pvec, lru);
 225         put_cpu_var(lru_add_pvecs);
 226 }
 227
 228 /**
 229  * lru_cache_add_lru - add a page to a page list
 230  * @page: the page to be added to the LRU.
 231  * @lru: the LRU list to which the page is added.
 232  */
 233 void lru_cache_add_lru(struct page *page, enum lru_list lru)
 234 {
 235         if (PageActive(page)) {
 236                 VM_BUG_ON(PageUnevictable(page));
 237                 ClearPageActive(page);
 238         } else if (PageUnevictable(page)) {
 239                 VM_BUG_ON(PageActive(page));
 240                 ClearPageUnevictable(page);
 241         }
 242
 243         VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page));
 244         __lru_cache_add(page, lru);
 245 }
 246
 247 /**
 248  * add_page_to_unevictable_list - add a page to the unevictable list
 249  * @page:  the page to be added to the unevictable list
 250  *
 251  * Add page directly to its zone's unevictable list.  To avoid races with
 252  * tasks that might be making the page evictable, through eg. munlock,
 253  * munmap or exit, while it's not on the lru, we want to add the page
 254  * while it's locked or otherwise "invisible" to other tasks.  This is
 255  * difficult to do when using the pagevec cache, so bypass that.
 256  */
 257 void add_page_to_unevictable_list(struct page *page)
 258 {
 259         struct zone *zone = page_zone(page);
 260
 261         spin_lock_irq(&zone->lru_lock);
 262         SetPageUnevictable(page);
 263         SetPageLRU(page);
 264         add_page_to_lru_list(zone, page, LRU_UNEVICTABLE);
 265         spin_unlock_irq(&zone->lru_lock);
 266 }
 267
 268 /*
 269  * Drain pages out of the cpu's pagevecs.
 270  * Either "cpu" is the current CPU, and preemption has already been
 271  * disabled; or "cpu" is being hot-unplugged, and is already dead.
 272  */
 273 static void drain_cpu_pagevecs(int cpu)
 274 {
 275         struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu);
 276         struct pagevec *pvec;
 277         int lru;
 278
 279         for_each_lru(lru) {
 280                 pvec = &pvecs[lru - LRU_BASE];
 281                 if (pagevec_count(pvec))
 282                         ____pagevec_lru_add(pvec, lru);
 283         }
 284
 285         pvec = &per_cpu(lru_rotate_pvecs, cpu);
 286         if (pagevec_count(pvec)) {
 287                 unsigned long flags;
 288
 289                 /* No harm done if a racing interrupt already did this */
 290                 local_irq_save(flags);
 291                 pagevec_move_tail(pvec);
 292                 local_irq_restore(flags);
 293         }
 294 }
 295
 296 void lru_add_drain(void)
 297 {
 298         drain_cpu_pagevecs(get_cpu());
 299         put_cpu();
 300 }
 301
 302 static void lru_add_drain_per_cpu(struct work_struct *dummy)
 303 {
 304         lru_add_drain();
 305 }
 306
 307 /*
 308  * Returns 0 for success
 309  */
 310 int lru_add_drain_all(void)
 311 {
 312         return schedule_on_each_cpu(lru_add_drain_per_cpu);
 313 }
 314
 315 /*
 316  * Batched page_cache_release().  Decrement the reference count on all the
 317  * passed pages.  If it fell to zero then remove the page from the LRU and
 318  * free it.
 319  *
 320  * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
 321  * for the remainder of the operation.
 322  *
 323  * The locking in this function is against shrink_inactive_list(): we recheck
 324  * the page count inside the lock to see whether shrink_inactive_list()
 325  * grabbed the page via the LRU.  If it did, give up: shrink_inactive_list()
 326  * will free it.
 327  */
 328 void release_pages(struct page **pages, int nr, int cold)
 329 {
 330         int i;
 331         struct pagevec pages_to_free;
 332         struct zone *zone = NULL;
 333         unsigned long uninitialized_var(flags);
 334
 335         pagevec_init(&pages_to_free, cold);
 336         for (i = 0; i < nr; i++) {
 337                 struct page *page = pages[i];
 338
 339                 if (unlikely(PageCompound(page))) {
 340                         if (zone) {
 341                                 spin_unlock_irqrestore(&zone->lru_lock, flags);
 342                                 zone = NULL;
 343                         }
 344                         put_compound_page(page);
 345                         continue;
 346                 }
 347
 348                 if (!put_page_testzero(page))
 349                         continue;
 350
 351                 if (PageLRU(page)) {
 352                         struct zone *pagezone = page_zone(page);
 353
 354                         if (pagezone != zone) {
 355                                 if (zone)
 356                                         spin_unlock_irqrestore(&zone->lru_lock,
 357                                                                         flags);
 358                                 zone = pagezone;
 359                                 spin_lock_irqsave(&zone->lru_lock, flags);
 360                         }
 361                         VM_BUG_ON(!PageLRU(page));
 362                         __ClearPageLRU(page);
 363                         del_page_from_lru(zone, page);
 364                 }
 365
 366                 if (!pagevec_add(&pages_to_free, page)) {
 367                         if (zone) {
 368                                 spin_unlock_irqrestore(&zone->lru_lock, flags);
 369                                 zone = NULL;
 370                         }
 371                         __pagevec_free(&pages_to_free);
 372                         pagevec_reinit(&pages_to_free);
 373                 }
 374         }
 375         if (zone)
 376                 spin_unlock_irqrestore(&zone->lru_lock, flags);
 377
 378         pagevec_free(&pages_to_free);
 379 }
 380
 381 /*
 382  * The pages which we're about to release may be in the deferred lru-addition
 383  * queues.  That would prevent them from really being freed right now.  That's
 384  * OK from a correctness point of view but is inefficient - those pages may be
 385  * cache-warm and we want to give them back to the page allocator ASAP.
 386  *
 387  * So __pagevec_release() will drain those queues here.  __pagevec_lru_add()
 388  * and __pagevec_lru_add_active() call release_pages() directly to avoid
 389  * mutual recursion.
 390  */
 391 void __pagevec_release(struct pagevec *pvec)
 392 {
 393         lru_add_drain();
 394         release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
 395         pagevec_reinit(pvec);
 396 }
 397
 398 EXPORT_SYMBOL(__pagevec_release);
 399
 400 /*
 401  * Add the passed pages to the LRU, then drop the caller's refcount
 402  * on them.  Reinitialises the caller's pagevec.
 403  */
 404 void ____pagevec_lru_add(struct pagevec *pvec, enum lru_list lru)
 405 {
 406         int i;
 407         struct zone *zone = NULL;
 408
 409         VM_BUG_ON(is_unevictable_lru(lru));
 410
 411         for (i = 0; i < pagevec_count(pvec); i++) {
 412                 struct page *page = pvec->pages[i];
 413                 struct zone *pagezone = page_zone(page);
 414                 int file;
 415                 int active;
 416
 417                 if (pagezone != zone) {
 418                         if (zone)
 419                                 spin_unlock_irq(&zone->lru_lock);
 420                         zone = pagezone;
 421                         spin_lock_irq(&zone->lru_lock);
 422                 }
 423                 VM_BUG_ON(PageActive(page));
 424                 VM_BUG_ON(PageUnevictable(page));
 425                 VM_BUG_ON(PageLRU(page));
 426                 SetPageLRU(page);
 427                 active = is_active_lru(lru);
 428                 file = is_file_lru(lru);
 429                 if (active)
 430                         SetPageActive(page);
 431                 update_page_reclaim_stat(zone, page, file, active);
 432                 add_page_to_lru_list(zone, page, lru);
 433         }
 434         if (zone)
 435                 spin_unlock_irq(&zone->lru_lock);
 436         release_pages(pvec->pages, pvec->nr, pvec->cold);
 437         pagevec_reinit(pvec);
 438 }
 439
 440 EXPORT_SYMBOL(____pagevec_lru_add);
 441
 442 /*
 443  * Try to drop buffers from the pages in a pagevec
 444  */
 445 void pagevec_strip(struct pagevec *pvec)
 446 {
 447         int i;
 448
 449         for (i = 0; i < pagevec_count(pvec); i++) {
 450                 struct page *page = pvec->pages[i];
 451
 452                 if (page_has_private(page) && trylock_page(page)) {
 453                         if (page_has_private(page))
 454                                 try_to_release_page(page, 0);
 455                         unlock_page(page);
 456                 }
 457         }
 458 }
 459
 460 /**
 461  * pagevec_lookup - gang pagecache lookup
 462  * @pvec:       Where the resulting pages are placed
 463  * @mapping:    The address_space to search
 464  * @start:      The starting page index
 465  * @nr_pages:   The maximum number of pages
 466  *
 467  * pagevec_lookup() will search for and return a group of up to @nr_pages pages
 468  * in the mapping.  The pages are placed in @pvec.  pagevec_lookup() takes a
 469  * reference against the pages in @pvec.
 470  *
 471  * The search returns a group of mapping-contiguous pages with ascending
 472  * indexes.  There may be holes in the indices due to not-present pages.
 473  *
 474  * pagevec_lookup() returns the number of pages which were found.
 475  */
 476 unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
 477                 pgoff_t start, unsigned nr_pages)
 478 {
 479         pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
 480         return pagevec_count(pvec);
 481 }
 482
 483 EXPORT_SYMBOL(pagevec_lookup);
 484
 485 unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
 486                 pgoff_t *index, int tag, unsigned nr_pages)
 487 {
 488         pvec->nr = find_get_pages_tag(mapping, index, tag,
 489                                         nr_pages, pvec->pages);
 490         return pagevec_count(pvec);
 491 }
 492
 493 EXPORT_SYMBOL(pagevec_lookup_tag);
 494
 495 /*
 496  * Perform any setup for the swap system
 497  */
 498 void __init swap_setup(void)
 499 {
 500         unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
 501
 502 #ifdef CONFIG_SWAP
 503         bdi_init(swapper_space.backing_dev_info);
 504 #endif
 505
 506         /* Use a smaller cluster for small-memory machines */
 507         if (megs < 16)
 508                 page_cluster = 2;
 509         else
 510                 page_cluster = 3;
 511         /*
 512          * Right now other parts of the system means that we
 513          * _really_ don't want to cluster much more
 514          */
 515 }