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 opereation 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/module.h>
  28 #include <linux/percpu_counter.h>
  29 #include <linux/percpu.h>
  30 #include <linux/cpu.h>
  31 #include <linux/notifier.h>
  32 #include <linux/init.h>
  33
  34 /* How many pages do we try to swap or page in/out together? */
  35 int page_cluster;
  36
  37 static void put_compound_page(struct page *page)
  38 {
  39         page = (struct page *)page_private(page);
  40         if (put_page_testzero(page)) {
  41                 void (*dtor)(struct page *page);
  42
  43                 dtor = (void (*)(struct page *))page[1].lru.next;
  44                 (*dtor)(page);
  45         }
  46 }
  47
  48 void put_page(struct page *page)
  49 {
  50         if (unlikely(PageCompound(page)))
  51                 put_compound_page(page);
  52         else if (put_page_testzero(page))
  53                 __page_cache_release(page);
  54 }
  55 EXPORT_SYMBOL(put_page);
  56
  57 /**
  58  * put_pages_list(): release a list of pages
  59  *
  60  * Release a list of pages which are strung together on page.lru.  Currently
  61  * used by read_cache_pages() and related error recovery code.
  62  *
  63  * @pages: list of pages threaded on page->lru
  64  */
  65 void put_pages_list(struct list_head *pages)
  66 {
  67         while (!list_empty(pages)) {
  68                 struct page *victim;
  69
  70                 victim = list_entry(pages->prev, struct page, lru);
  71                 list_del(&victim->lru);
  72                 page_cache_release(victim);
  73         }
  74 }
  75 EXPORT_SYMBOL(put_pages_list);
  76
  77 /*
  78  * Writeback is about to end against a page which has been marked for immediate
  79  * reclaim.  If it still appears to be reclaimable, move it to the tail of the
  80  * inactive list.  The page still has PageWriteback set, which will pin it.
  81  *
  82  * We don't expect many pages to come through here, so don't bother batching
  83  * things up.
  84  *
  85  * To avoid placing the page at the tail of the LRU while PG_writeback is still
  86  * set, this function will clear PG_writeback before performing the page
  87  * motion.  Do that inside the lru lock because once PG_writeback is cleared
  88  * we may not touch the page.
  89  *
  90  * Returns zero if it cleared PG_writeback.
  91  */
  92 int rotate_reclaimable_page(struct page *page)
  93 {
  94         struct zone *zone;
  95         unsigned long flags;
  96
  97         if (PageLocked(page))
  98                 return 1;
  99         if (PageDirty(page))
 100                 return 1;
 101         if (PageActive(page))
 102                 return 1;
 103         if (!PageLRU(page))
 104                 return 1;
 105
 106         zone = page_zone(page);
 107         spin_lock_irqsave(&zone->lru_lock, flags);
 108         if (PageLRU(page) && !PageActive(page)) {
 109                 list_move_tail(&page->lru, &zone->inactive_list);
 110                 __count_vm_event(PGROTATED);
 111         }
 112         if (!test_clear_page_writeback(page))
 113                 BUG();
 114         spin_unlock_irqrestore(&zone->lru_lock, flags);
 115         return 0;
 116 }
 117
 118 /*
 119  * FIXME: speed this up?
 120  */
 121 void fastcall activate_page(struct page *page)
 122 {
 123         struct zone *zone = page_zone(page);
 124
 125         spin_lock_irq(&zone->lru_lock);
 126         if (PageLRU(page) && !PageActive(page)) {
 127                 del_page_from_inactive_list(zone, page);
 128                 SetPageActive(page);
 129                 add_page_to_active_list(zone, page);
 130                 __count_vm_event(PGACTIVATE);
 131         }
 132         spin_unlock_irq(&zone->lru_lock);
 133 }
 134
 135 /*
 136  * Mark a page as having seen activity.
 137  *
 138  * inactive,unreferenced        ->      inactive,referenced
 139  * inactive,referenced          ->      active,unreferenced
 140  * active,unreferenced          ->      active,referenced
 141  */
 142 void fastcall mark_page_accessed(struct page *page)
 143 {
 144         if (!PageActive(page) && PageReferenced(page) && PageLRU(page)) {
 145                 activate_page(page);
 146                 ClearPageReferenced(page);
 147         } else if (!PageReferenced(page)) {
 148                 SetPageReferenced(page);
 149         }
 150 }
 151
 152 EXPORT_SYMBOL(mark_page_accessed);
 153
 154 /**
 155  * lru_cache_add: add a page to the page lists
 156  * @page: the page to add
 157  */
 158 static DEFINE_PER_CPU(struct pagevec, lru_add_pvecs) = { 0, };
 159 static DEFINE_PER_CPU(struct pagevec, lru_add_active_pvecs) = { 0, };
 160
 161 void fastcall lru_cache_add(struct page *page)
 162 {
 163         struct pagevec *pvec = &get_cpu_var(lru_add_pvecs);
 164
 165         page_cache_get(page);
 166         if (!pagevec_add(pvec, page))
 167                 __pagevec_lru_add(pvec);
 168         put_cpu_var(lru_add_pvecs);
 169 }
 170
 171 void fastcall lru_cache_add_active(struct page *page)
 172 {
 173         struct pagevec *pvec = &get_cpu_var(lru_add_active_pvecs);
 174
 175         page_cache_get(page);
 176         if (!pagevec_add(pvec, page))
 177                 __pagevec_lru_add_active(pvec);
 178         put_cpu_var(lru_add_active_pvecs);
 179 }
 180
 181 static void __lru_add_drain(int cpu)
 182 {
 183         struct pagevec *pvec = &per_cpu(lru_add_pvecs, cpu);
 184
 185         /* CPU is dead, so no locking needed. */
 186         if (pagevec_count(pvec))
 187                 __pagevec_lru_add(pvec);
 188         pvec = &per_cpu(lru_add_active_pvecs, cpu);
 189         if (pagevec_count(pvec))
 190                 __pagevec_lru_add_active(pvec);
 191 }
 192
 193 void lru_add_drain(void)
 194 {
 195         __lru_add_drain(get_cpu());
 196         put_cpu();
 197 }
 198
 199 #ifdef CONFIG_NUMA
 200 static void lru_add_drain_per_cpu(void *dummy)
 201 {
 202         lru_add_drain();
 203 }
 204
 205 /*
 206  * Returns 0 for success
 207  */
 208 int lru_add_drain_all(void)
 209 {
 210         return schedule_on_each_cpu(lru_add_drain_per_cpu, NULL);
 211 }
 212
 213 #else
 214
 215 /*
 216  * Returns 0 for success
 217  */
 218 int lru_add_drain_all(void)
 219 {
 220         lru_add_drain();
 221         return 0;
 222 }
 223 #endif
 224
 225 /*
 226  * This path almost never happens for VM activity - pages are normally
 227  * freed via pagevecs.  But it gets used by networking.
 228  */
 229 void fastcall __page_cache_release(struct page *page)
 230 {
 231         if (PageLRU(page)) {
 232                 unsigned long flags;
 233                 struct zone *zone = page_zone(page);
 234
 235                 spin_lock_irqsave(&zone->lru_lock, flags);
 236                 BUG_ON(!PageLRU(page));
 237                 __ClearPageLRU(page);
 238                 del_page_from_lru(zone, page);
 239                 spin_unlock_irqrestore(&zone->lru_lock, flags);
 240         }
 241         free_hot_page(page);
 242 }
 243 EXPORT_SYMBOL(__page_cache_release);
 244
 245 /*
 246  * Batched page_cache_release().  Decrement the reference count on all the
 247  * passed pages.  If it fell to zero then remove the page from the LRU and
 248  * free it.
 249  *
 250  * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
 251  * for the remainder of the operation.
 252  *
 253  * The locking in this function is against shrink_cache(): we recheck the
 254  * page count inside the lock to see whether shrink_cache grabbed the page
 255  * via the LRU.  If it did, give up: shrink_cache will free it.
 256  */
 257 void release_pages(struct page **pages, int nr, int cold)
 258 {
 259         int i;
 260         struct pagevec pages_to_free;
 261         struct zone *zone = NULL;
 262
 263         pagevec_init(&pages_to_free, cold);
 264         for (i = 0; i < nr; i++) {
 265                 struct page *page = pages[i];
 266
 267                 if (unlikely(PageCompound(page))) {
 268                         if (zone) {
 269                                 spin_unlock_irq(&zone->lru_lock);
 270                                 zone = NULL;
 271                         }
 272                         put_compound_page(page);
 273                         continue;
 274                 }
 275
 276                 if (!put_page_testzero(page))
 277                         continue;
 278
 279                 if (PageLRU(page)) {
 280                         struct zone *pagezone = page_zone(page);
 281                         if (pagezone != zone) {
 282                                 if (zone)
 283                                         spin_unlock_irq(&zone->lru_lock);
 284                                 zone = pagezone;
 285                                 spin_lock_irq(&zone->lru_lock);
 286                         }
 287                         BUG_ON(!PageLRU(page));
 288                         __ClearPageLRU(page);
 289                         del_page_from_lru(zone, page);
 290                 }
 291
 292                 if (!pagevec_add(&pages_to_free, page)) {
 293                         if (zone) {
 294                                 spin_unlock_irq(&zone->lru_lock);
 295                                 zone = NULL;
 296                         }
 297                         __pagevec_free(&pages_to_free);
 298                         pagevec_reinit(&pages_to_free);
 299                 }
 300         }
 301         if (zone)
 302                 spin_unlock_irq(&zone->lru_lock);
 303
 304         pagevec_free(&pages_to_free);
 305 }
 306
 307 /*
 308  * The pages which we're about to release may be in the deferred lru-addition
 309  * queues.  That would prevent them from really being freed right now.  That's
 310  * OK from a correctness point of view but is inefficient - those pages may be
 311  * cache-warm and we want to give them back to the page allocator ASAP.
 312  *
 313  * So __pagevec_release() will drain those queues here.  __pagevec_lru_add()
 314  * and __pagevec_lru_add_active() call release_pages() directly to avoid
 315  * mutual recursion.
 316  */
 317 void __pagevec_release(struct pagevec *pvec)
 318 {
 319         lru_add_drain();
 320         release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
 321         pagevec_reinit(pvec);
 322 }
 323
 324 EXPORT_SYMBOL(__pagevec_release);
 325
 326 /*
 327  * pagevec_release() for pages which are known to not be on the LRU
 328  *
 329  * This function reinitialises the caller's pagevec.
 330  */
 331 void __pagevec_release_nonlru(struct pagevec *pvec)
 332 {
 333         int i;
 334         struct pagevec pages_to_free;
 335
 336         pagevec_init(&pages_to_free, pvec->cold);
 337         for (i = 0; i < pagevec_count(pvec); i++) {
 338                 struct page *page = pvec->pages[i];
 339
 340                 BUG_ON(PageLRU(page));
 341                 if (put_page_testzero(page))
 342                         pagevec_add(&pages_to_free, page);
 343         }
 344         pagevec_free(&pages_to_free);
 345         pagevec_reinit(pvec);
 346 }
 347
 348 /*
 349  * Add the passed pages to the LRU, then drop the caller's refcount
 350  * on them.  Reinitialises the caller's pagevec.
 351  */
 352 void __pagevec_lru_add(struct pagevec *pvec)
 353 {
 354         int i;
 355         struct zone *zone = NULL;
 356
 357         for (i = 0; i < pagevec_count(pvec); i++) {
 358                 struct page *page = pvec->pages[i];
 359                 struct zone *pagezone = page_zone(page);
 360
 361                 if (pagezone != zone) {
 362                         if (zone)
 363                                 spin_unlock_irq(&zone->lru_lock);
 364                         zone = pagezone;
 365                         spin_lock_irq(&zone->lru_lock);
 366                 }
 367                 BUG_ON(PageLRU(page));
 368                 SetPageLRU(page);
 369                 add_page_to_inactive_list(zone, page);
 370         }
 371         if (zone)
 372                 spin_unlock_irq(&zone->lru_lock);
 373         release_pages(pvec->pages, pvec->nr, pvec->cold);
 374         pagevec_reinit(pvec);
 375 }
 376
 377 EXPORT_SYMBOL(__pagevec_lru_add);
 378
 379 void __pagevec_lru_add_active(struct pagevec *pvec)
 380 {
 381         int i;
 382         struct zone *zone = NULL;
 383
 384         for (i = 0; i < pagevec_count(pvec); i++) {
 385                 struct page *page = pvec->pages[i];
 386                 struct zone *pagezone = page_zone(page);
 387
 388                 if (pagezone != zone) {
 389                         if (zone)
 390                                 spin_unlock_irq(&zone->lru_lock);
 391                         zone = pagezone;
 392                         spin_lock_irq(&zone->lru_lock);
 393                 }
 394                 BUG_ON(PageLRU(page));
 395                 SetPageLRU(page);
 396                 BUG_ON(PageActive(page));
 397                 SetPageActive(page);
 398                 add_page_to_active_list(zone, page);
 399         }
 400         if (zone)
 401                 spin_unlock_irq(&zone->lru_lock);
 402         release_pages(pvec->pages, pvec->nr, pvec->cold);
 403         pagevec_reinit(pvec);
 404 }
 405
 406 /*
 407  * Try to drop buffers from the pages in a pagevec
 408  */
 409 void pagevec_strip(struct pagevec *pvec)
 410 {
 411         int i;
 412
 413         for (i = 0; i < pagevec_count(pvec); i++) {
 414                 struct page *page = pvec->pages[i];
 415
 416                 if (PagePrivate(page) && !TestSetPageLocked(page)) {
 417                         if (PagePrivate(page))
 418                                 try_to_release_page(page, 0);
 419                         unlock_page(page);
 420                 }
 421         }
 422 }
 423
 424 /**
 425  * pagevec_lookup - gang pagecache lookup
 426  * @pvec:       Where the resulting pages are placed
 427  * @mapping:    The address_space to search
 428  * @start:      The starting page index
 429  * @nr_pages:   The maximum number of pages
 430  *
 431  * pagevec_lookup() will search for and return a group of up to @nr_pages pages
 432  * in the mapping.  The pages are placed in @pvec.  pagevec_lookup() takes a
 433  * reference against the pages in @pvec.
 434  *
 435  * The search returns a group of mapping-contiguous pages with ascending
 436  * indexes.  There may be holes in the indices due to not-present pages.
 437  *
 438  * pagevec_lookup() returns the number of pages which were found.
 439  */
 440 unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
 441                 pgoff_t start, unsigned nr_pages)
 442 {
 443         pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
 444         return pagevec_count(pvec);
 445 }
 446
 447 EXPORT_SYMBOL(pagevec_lookup);
 448
 449 unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
 450                 pgoff_t *index, int tag, unsigned nr_pages)
 451 {
 452         pvec->nr = find_get_pages_tag(mapping, index, tag,
 453                                         nr_pages, pvec->pages);
 454         return pagevec_count(pvec);
 455 }
 456
 457 EXPORT_SYMBOL(pagevec_lookup_tag);
 458
 459 #ifdef CONFIG_SMP
 460 /*
 461  * We tolerate a little inaccuracy to avoid ping-ponging the counter between
 462  * CPUs
 463  */
 464 #define ACCT_THRESHOLD  max(16, NR_CPUS * 2)
 465
 466 static DEFINE_PER_CPU(long, committed_space) = 0;
 467
 468 void vm_acct_memory(long pages)
 469 {
 470         long *local;
 471
 472         preempt_disable();
 473         local = &__get_cpu_var(committed_space);
 474         *local += pages;
 475         if (*local > ACCT_THRESHOLD || *local < -ACCT_THRESHOLD) {
 476                 atomic_add(*local, &vm_committed_space);
 477                 *local = 0;
 478         }
 479         preempt_enable();
 480 }
 481
 482 #ifdef CONFIG_HOTPLUG_CPU
 483
 484 /* Drop the CPU's cached committed space back into the central pool. */
 485 static int cpu_swap_callback(struct notifier_block *nfb,
 486                              unsigned long action,
 487                              void *hcpu)
 488 {
 489         long *committed;
 490
 491         committed = &per_cpu(committed_space, (long)hcpu);
 492         if (action == CPU_DEAD) {
 493                 atomic_add(*committed, &vm_committed_space);
 494                 *committed = 0;
 495                 __lru_add_drain((long)hcpu);
 496         }
 497         return NOTIFY_OK;
 498 }
 499 #endif /* CONFIG_HOTPLUG_CPU */
 500 #endif /* CONFIG_SMP */
 501
 502 /*
 503  * Perform any setup for the swap system
 504  */
 505 void __init swap_setup(void)
 506 {
 507         unsigned long megs = num_physpages >> (20 - PAGE_SHIFT);
 508
 509         /* Use a smaller cluster for small-memory machines */
 510         if (megs < 16)
 511                 page_cluster = 2;
 512         else
 513                 page_cluster = 3;
 514         /*
 515          * Right now other parts of the system means that we
 516          * _really_ don't want to cluster much more
 517          */
 518         hotcpu_notifier(cpu_swap_callback, 0);
 519 }