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