swap.c revision 43fac94dd62667c83dd2daa5b7ac548512af780a
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/percpu_counter.h> 28#include <linux/percpu.h> 29#include <linux/cpu.h> 30#include <linux/notifier.h> 31 32/* How many pages do we try to swap or page in/out together? */ 33int page_cluster; 34 35/* 36 * This path almost never happens for VM activity - pages are normally 37 * freed via pagevecs. But it gets used by networking. 38 */ 39static void fastcall __page_cache_release(struct page *page) 40{ 41 if (PageLRU(page)) { 42 unsigned long flags; 43 struct zone *zone = page_zone(page); 44 45 spin_lock_irqsave(&zone->lru_lock, flags); 46 VM_BUG_ON(!PageLRU(page)); 47 __ClearPageLRU(page); 48 del_page_from_lru(zone, page); 49 spin_unlock_irqrestore(&zone->lru_lock, flags); 50 } 51 free_hot_page(page); 52} 53 54static void put_compound_page(struct page *page) 55{ 56 page = compound_head(page); 57 if (put_page_testzero(page)) { 58 compound_page_dtor *dtor; 59 60 dtor = get_compound_page_dtor(page); 61 (*dtor)(page); 62 } 63} 64 65void put_page(struct page *page) 66{ 67 if (unlikely(PageCompound(page))) 68 put_compound_page(page); 69 else if (put_page_testzero(page)) 70 __page_cache_release(page); 71} 72EXPORT_SYMBOL(put_page); 73 74/** 75 * put_pages_list(): release a list of pages 76 * 77 * Release a list of pages which are strung together on page.lru. Currently 78 * used by read_cache_pages() and related error recovery code. 79 * 80 * @pages: list of pages threaded on page->lru 81 */ 82void put_pages_list(struct list_head *pages) 83{ 84 while (!list_empty(pages)) { 85 struct page *victim; 86 87 victim = list_entry(pages->prev, struct page, lru); 88 list_del(&victim->lru); 89 page_cache_release(victim); 90 } 91} 92EXPORT_SYMBOL(put_pages_list); 93 94/* 95 * Writeback is about to end against a page which has been marked for immediate 96 * reclaim. If it still appears to be reclaimable, move it to the tail of the 97 * inactive list. The page still has PageWriteback set, which will pin it. 98 * 99 * We don't expect many pages to come through here, so don't bother batching 100 * things up. 101 * 102 * To avoid placing the page at the tail of the LRU while PG_writeback is still 103 * set, this function will clear PG_writeback before performing the page 104 * motion. Do that inside the lru lock because once PG_writeback is cleared 105 * we may not touch the page. 106 * 107 * Returns zero if it cleared PG_writeback. 108 */ 109int rotate_reclaimable_page(struct page *page) 110{ 111 struct zone *zone; 112 unsigned long flags; 113 114 if (PageLocked(page)) 115 return 1; 116 if (PageDirty(page)) 117 return 1; 118 if (PageActive(page)) 119 return 1; 120 if (!PageLRU(page)) 121 return 1; 122 123 zone = page_zone(page); 124 spin_lock_irqsave(&zone->lru_lock, flags); 125 if (PageLRU(page) && !PageActive(page)) { 126 list_move_tail(&page->lru, &zone->inactive_list); 127 __count_vm_event(PGROTATED); 128 } 129 if (!test_clear_page_writeback(page)) 130 BUG(); 131 spin_unlock_irqrestore(&zone->lru_lock, flags); 132 return 0; 133} 134 135/* 136 * FIXME: speed this up? 137 */ 138void fastcall activate_page(struct page *page) 139{ 140 struct zone *zone = page_zone(page); 141 142 spin_lock_irq(&zone->lru_lock); 143 if (PageLRU(page) && !PageActive(page)) { 144 del_page_from_inactive_list(zone, page); 145 SetPageActive(page); 146 add_page_to_active_list(zone, page); 147 __count_vm_event(PGACTIVATE); 148 } 149 spin_unlock_irq(&zone->lru_lock); 150} 151 152/* 153 * Mark a page as having seen activity. 154 * 155 * inactive,unreferenced -> inactive,referenced 156 * inactive,referenced -> active,unreferenced 157 * active,unreferenced -> active,referenced 158 */ 159void fastcall mark_page_accessed(struct page *page) 160{ 161 if (!PageActive(page) && PageReferenced(page) && PageLRU(page)) { 162 activate_page(page); 163 ClearPageReferenced(page); 164 } else if (!PageReferenced(page)) { 165 SetPageReferenced(page); 166 } 167} 168 169EXPORT_SYMBOL(mark_page_accessed); 170 171/** 172 * lru_cache_add: add a page to the page lists 173 * @page: the page to add 174 */ 175static DEFINE_PER_CPU(struct pagevec, lru_add_pvecs) = { 0, }; 176static DEFINE_PER_CPU(struct pagevec, lru_add_active_pvecs) = { 0, }; 177 178void fastcall lru_cache_add(struct page *page) 179{ 180 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs); 181 182 page_cache_get(page); 183 if (!pagevec_add(pvec, page)) 184 __pagevec_lru_add(pvec); 185 put_cpu_var(lru_add_pvecs); 186} 187 188void fastcall lru_cache_add_active(struct page *page) 189{ 190 struct pagevec *pvec = &get_cpu_var(lru_add_active_pvecs); 191 192 page_cache_get(page); 193 if (!pagevec_add(pvec, page)) 194 __pagevec_lru_add_active(pvec); 195 put_cpu_var(lru_add_active_pvecs); 196} 197 198static void __lru_add_drain(int cpu) 199{ 200 struct pagevec *pvec = &per_cpu(lru_add_pvecs, cpu); 201 202 /* CPU is dead, so no locking needed. */ 203 if (pagevec_count(pvec)) 204 __pagevec_lru_add(pvec); 205 pvec = &per_cpu(lru_add_active_pvecs, cpu); 206 if (pagevec_count(pvec)) 207 __pagevec_lru_add_active(pvec); 208} 209 210void lru_add_drain(void) 211{ 212 __lru_add_drain(get_cpu()); 213 put_cpu(); 214} 215 216#ifdef CONFIG_NUMA 217static void lru_add_drain_per_cpu(struct work_struct *dummy) 218{ 219 lru_add_drain(); 220} 221 222/* 223 * Returns 0 for success 224 */ 225int lru_add_drain_all(void) 226{ 227 return schedule_on_each_cpu(lru_add_drain_per_cpu); 228} 229 230#else 231 232/* 233 * Returns 0 for success 234 */ 235int lru_add_drain_all(void) 236{ 237 lru_add_drain(); 238 return 0; 239} 240#endif 241 242/* 243 * Batched page_cache_release(). Decrement the reference count on all the 244 * passed pages. If it fell to zero then remove the page from the LRU and 245 * free it. 246 * 247 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it 248 * for the remainder of the operation. 249 * 250 * The locking in this function is against shrink_cache(): we recheck the 251 * page count inside the lock to see whether shrink_cache grabbed the page 252 * via the LRU. If it did, give up: shrink_cache will free it. 253 */ 254void release_pages(struct page **pages, int nr, int cold) 255{ 256 int i; 257 struct pagevec pages_to_free; 258 struct zone *zone = NULL; 259 260 pagevec_init(&pages_to_free, cold); 261 for (i = 0; i < nr; i++) { 262 struct page *page = pages[i]; 263 264 if (unlikely(PageCompound(page))) { 265 if (zone) { 266 spin_unlock_irq(&zone->lru_lock); 267 zone = NULL; 268 } 269 put_compound_page(page); 270 continue; 271 } 272 273 if (!put_page_testzero(page)) 274 continue; 275 276 if (PageLRU(page)) { 277 struct zone *pagezone = page_zone(page); 278 if (pagezone != zone) { 279 if (zone) 280 spin_unlock_irq(&zone->lru_lock); 281 zone = pagezone; 282 spin_lock_irq(&zone->lru_lock); 283 } 284 VM_BUG_ON(!PageLRU(page)); 285 __ClearPageLRU(page); 286 del_page_from_lru(zone, page); 287 } 288 289 if (!pagevec_add(&pages_to_free, page)) { 290 if (zone) { 291 spin_unlock_irq(&zone->lru_lock); 292 zone = NULL; 293 } 294 __pagevec_free(&pages_to_free); 295 pagevec_reinit(&pages_to_free); 296 } 297 } 298 if (zone) 299 spin_unlock_irq(&zone->lru_lock); 300 301 pagevec_free(&pages_to_free); 302} 303 304/* 305 * The pages which we're about to release may be in the deferred lru-addition 306 * queues. That would prevent them from really being freed right now. That's 307 * OK from a correctness point of view but is inefficient - those pages may be 308 * cache-warm and we want to give them back to the page allocator ASAP. 309 * 310 * So __pagevec_release() will drain those queues here. __pagevec_lru_add() 311 * and __pagevec_lru_add_active() call release_pages() directly to avoid 312 * mutual recursion. 313 */ 314void __pagevec_release(struct pagevec *pvec) 315{ 316 lru_add_drain(); 317 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold); 318 pagevec_reinit(pvec); 319} 320 321EXPORT_SYMBOL(__pagevec_release); 322 323/* 324 * pagevec_release() for pages which are known to not be on the LRU 325 * 326 * This function reinitialises the caller's pagevec. 327 */ 328void __pagevec_release_nonlru(struct pagevec *pvec) 329{ 330 int i; 331 struct pagevec pages_to_free; 332 333 pagevec_init(&pages_to_free, pvec->cold); 334 for (i = 0; i < pagevec_count(pvec); i++) { 335 struct page *page = pvec->pages[i]; 336 337 VM_BUG_ON(PageLRU(page)); 338 if (put_page_testzero(page)) 339 pagevec_add(&pages_to_free, page); 340 } 341 pagevec_free(&pages_to_free); 342 pagevec_reinit(pvec); 343} 344 345/* 346 * Add the passed pages to the LRU, then drop the caller's refcount 347 * on them. Reinitialises the caller's pagevec. 348 */ 349void __pagevec_lru_add(struct pagevec *pvec) 350{ 351 int i; 352 struct zone *zone = NULL; 353 354 for (i = 0; i < pagevec_count(pvec); i++) { 355 struct page *page = pvec->pages[i]; 356 struct zone *pagezone = page_zone(page); 357 358 if (pagezone != zone) { 359 if (zone) 360 spin_unlock_irq(&zone->lru_lock); 361 zone = pagezone; 362 spin_lock_irq(&zone->lru_lock); 363 } 364 VM_BUG_ON(PageLRU(page)); 365 SetPageLRU(page); 366 add_page_to_inactive_list(zone, page); 367 } 368 if (zone) 369 spin_unlock_irq(&zone->lru_lock); 370 release_pages(pvec->pages, pvec->nr, pvec->cold); 371 pagevec_reinit(pvec); 372} 373 374EXPORT_SYMBOL(__pagevec_lru_add); 375 376void __pagevec_lru_add_active(struct pagevec *pvec) 377{ 378 int i; 379 struct zone *zone = NULL; 380 381 for (i = 0; i < pagevec_count(pvec); i++) { 382 struct page *page = pvec->pages[i]; 383 struct zone *pagezone = page_zone(page); 384 385 if (pagezone != zone) { 386 if (zone) 387 spin_unlock_irq(&zone->lru_lock); 388 zone = pagezone; 389 spin_lock_irq(&zone->lru_lock); 390 } 391 VM_BUG_ON(PageLRU(page)); 392 SetPageLRU(page); 393 VM_BUG_ON(PageActive(page)); 394 SetPageActive(page); 395 add_page_to_active_list(zone, page); 396 } 397 if (zone) 398 spin_unlock_irq(&zone->lru_lock); 399 release_pages(pvec->pages, pvec->nr, pvec->cold); 400 pagevec_reinit(pvec); 401} 402 403/* 404 * Try to drop buffers from the pages in a pagevec 405 */ 406void pagevec_strip(struct pagevec *pvec) 407{ 408 int i; 409 410 for (i = 0; i < pagevec_count(pvec); i++) { 411 struct page *page = pvec->pages[i]; 412 413 if (PagePrivate(page) && !TestSetPageLocked(page)) { 414 if (PagePrivate(page)) 415 try_to_release_page(page, 0); 416 unlock_page(page); 417 } 418 } 419} 420 421/** 422 * pagevec_lookup - gang pagecache lookup 423 * @pvec: Where the resulting pages are placed 424 * @mapping: The address_space to search 425 * @start: The starting page index 426 * @nr_pages: The maximum number of pages 427 * 428 * pagevec_lookup() will search for and return a group of up to @nr_pages pages 429 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a 430 * reference against the pages in @pvec. 431 * 432 * The search returns a group of mapping-contiguous pages with ascending 433 * indexes. There may be holes in the indices due to not-present pages. 434 * 435 * pagevec_lookup() returns the number of pages which were found. 436 */ 437unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping, 438 pgoff_t start, unsigned nr_pages) 439{ 440 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages); 441 return pagevec_count(pvec); 442} 443 444EXPORT_SYMBOL(pagevec_lookup); 445 446unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping, 447 pgoff_t *index, int tag, unsigned nr_pages) 448{ 449 pvec->nr = find_get_pages_tag(mapping, index, tag, 450 nr_pages, pvec->pages); 451 return pagevec_count(pvec); 452} 453 454EXPORT_SYMBOL(pagevec_lookup_tag); 455 456#ifdef CONFIG_SMP 457/* 458 * We tolerate a little inaccuracy to avoid ping-ponging the counter between 459 * CPUs 460 */ 461#define ACCT_THRESHOLD max(16, NR_CPUS * 2) 462 463static DEFINE_PER_CPU(long, committed_space) = 0; 464 465void vm_acct_memory(long pages) 466{ 467 long *local; 468 469 preempt_disable(); 470 local = &__get_cpu_var(committed_space); 471 *local += pages; 472 if (*local > ACCT_THRESHOLD || *local < -ACCT_THRESHOLD) { 473 atomic_add(*local, &vm_committed_space); 474 *local = 0; 475 } 476 preempt_enable(); 477} 478 479#ifdef CONFIG_HOTPLUG_CPU 480 481/* Drop the CPU's cached committed space back into the central pool. */ 482static int cpu_swap_callback(struct notifier_block *nfb, 483 unsigned long action, 484 void *hcpu) 485{ 486 long *committed; 487 488 committed = &per_cpu(committed_space, (long)hcpu); 489 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { 490 atomic_add(*committed, &vm_committed_space); 491 *committed = 0; 492 __lru_add_drain((long)hcpu); 493 } 494 return NOTIFY_OK; 495} 496#endif /* CONFIG_HOTPLUG_CPU */ 497#endif /* CONFIG_SMP */ 498 499/* 500 * Perform any setup for the swap system 501 */ 502void __init swap_setup(void) 503{ 504 unsigned long megs = num_physpages >> (20 - PAGE_SHIFT); 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#ifdef CONFIG_HOTPLUG_CPU 516 hotcpu_notifier(cpu_swap_callback, 0); 517#endif 518} 519