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