4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
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.
12 * Swap aging added 23.2.95, Stephen Tweedie.
13 * Buffermem limits added 12.3.98, Rik van Riel.
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>
36 /* How many pages do we try to swap or page in/out together? */
39 static DEFINE_PER_CPU(struct pagevec[NR_LRU_LISTS], lru_add_pvecs);
40 static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
43 * This path almost never happens for VM activity - pages are normally
44 * freed via pagevecs. But it gets used by networking.
46 static void __page_cache_release(struct page *page)
50 struct zone *zone = page_zone(page);
52 spin_lock_irqsave(&zone->lru_lock, flags);
53 VM_BUG_ON(!PageLRU(page));
55 del_page_from_lru(zone, page);
56 spin_unlock_irqrestore(&zone->lru_lock, flags);
61 static void put_compound_page(struct page *page)
63 page = compound_head(page);
64 if (put_page_testzero(page)) {
65 compound_page_dtor *dtor;
67 dtor = get_compound_page_dtor(page);
72 void put_page(struct page *page)
74 if (unlikely(PageCompound(page)))
75 put_compound_page(page);
76 else if (put_page_testzero(page))
77 __page_cache_release(page);
79 EXPORT_SYMBOL(put_page);
82 * put_pages_list() - release a list of pages
83 * @pages: list of pages threaded on page->lru
85 * Release a list of pages which are strung together on page.lru. Currently
86 * used by read_cache_pages() and related error recovery code.
88 void put_pages_list(struct list_head *pages)
90 while (!list_empty(pages)) {
93 victim = list_entry(pages->prev, struct page, lru);
94 list_del(&victim->lru);
95 page_cache_release(victim);
98 EXPORT_SYMBOL(put_pages_list);
101 * pagevec_move_tail() must be called with IRQ disabled.
102 * Otherwise this may cause nasty races.
104 static void pagevec_move_tail(struct pagevec *pvec)
108 struct zone *zone = NULL;
110 for (i = 0; i < pagevec_count(pvec); i++) {
111 struct page *page = pvec->pages[i];
112 struct zone *pagezone = page_zone(page);
114 if (pagezone != zone) {
116 spin_unlock(&zone->lru_lock);
118 spin_lock(&zone->lru_lock);
120 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
121 int lru = page_is_file_cache(page);
122 list_move_tail(&page->lru, &zone->lru[lru].list);
127 spin_unlock(&zone->lru_lock);
128 __count_vm_events(PGROTATED, pgmoved);
129 release_pages(pvec->pages, pvec->nr, pvec->cold);
130 pagevec_reinit(pvec);
134 * Writeback is about to end against a page which has been marked for immediate
135 * reclaim. If it still appears to be reclaimable, move it to the tail of the
138 void rotate_reclaimable_page(struct page *page)
140 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
141 !PageUnevictable(page) && PageLRU(page)) {
142 struct pagevec *pvec;
145 page_cache_get(page);
146 local_irq_save(flags);
147 pvec = &__get_cpu_var(lru_rotate_pvecs);
148 if (!pagevec_add(pvec, page))
149 pagevec_move_tail(pvec);
150 local_irq_restore(flags);
155 * FIXME: speed this up?
157 void activate_page(struct page *page)
159 struct zone *zone = page_zone(page);
161 spin_lock_irq(&zone->lru_lock);
162 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
163 int file = page_is_file_cache(page);
164 int lru = LRU_BASE + file;
165 del_page_from_lru_list(zone, page, lru);
169 add_page_to_lru_list(zone, page, lru);
170 __count_vm_event(PGACTIVATE);
171 mem_cgroup_move_lists(page, lru);
173 zone->recent_rotated[!!file]++;
174 zone->recent_scanned[!!file]++;
176 spin_unlock_irq(&zone->lru_lock);
180 * Mark a page as having seen activity.
182 * inactive,unreferenced -> inactive,referenced
183 * inactive,referenced -> active,unreferenced
184 * active,unreferenced -> active,referenced
186 void mark_page_accessed(struct page *page)
188 if (!PageActive(page) && !PageUnevictable(page) &&
189 PageReferenced(page) && PageLRU(page)) {
191 ClearPageReferenced(page);
192 } else if (!PageReferenced(page)) {
193 SetPageReferenced(page);
197 EXPORT_SYMBOL(mark_page_accessed);
199 void __lru_cache_add(struct page *page, enum lru_list lru)
201 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru];
203 page_cache_get(page);
204 if (!pagevec_add(pvec, page))
205 ____pagevec_lru_add(pvec, lru);
206 put_cpu_var(lru_add_pvecs);
210 * lru_cache_add_lru - add a page to a page list
211 * @page: the page to be added to the LRU.
212 * @lru: the LRU list to which the page is added.
214 void lru_cache_add_lru(struct page *page, enum lru_list lru)
216 if (PageActive(page)) {
217 VM_BUG_ON(PageUnevictable(page));
218 ClearPageActive(page);
219 } else if (PageUnevictable(page)) {
220 VM_BUG_ON(PageActive(page));
221 ClearPageUnevictable(page);
224 VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page));
225 __lru_cache_add(page, lru);
229 * add_page_to_unevictable_list - add a page to the unevictable list
230 * @page: the page to be added to the unevictable list
232 * Add page directly to its zone's unevictable list. To avoid races with
233 * tasks that might be making the page evictable, through eg. munlock,
234 * munmap or exit, while it's not on the lru, we want to add the page
235 * while it's locked or otherwise "invisible" to other tasks. This is
236 * difficult to do when using the pagevec cache, so bypass that.
238 void add_page_to_unevictable_list(struct page *page)
240 struct zone *zone = page_zone(page);
242 spin_lock_irq(&zone->lru_lock);
243 SetPageUnevictable(page);
245 add_page_to_lru_list(zone, page, LRU_UNEVICTABLE);
246 spin_unlock_irq(&zone->lru_lock);
250 * Drain pages out of the cpu's pagevecs.
251 * Either "cpu" is the current CPU, and preemption has already been
252 * disabled; or "cpu" is being hot-unplugged, and is already dead.
254 static void drain_cpu_pagevecs(int cpu)
256 struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu);
257 struct pagevec *pvec;
261 pvec = &pvecs[lru - LRU_BASE];
262 if (pagevec_count(pvec))
263 ____pagevec_lru_add(pvec, lru);
266 pvec = &per_cpu(lru_rotate_pvecs, cpu);
267 if (pagevec_count(pvec)) {
270 /* No harm done if a racing interrupt already did this */
271 local_irq_save(flags);
272 pagevec_move_tail(pvec);
273 local_irq_restore(flags);
277 void lru_add_drain(void)
279 drain_cpu_pagevecs(get_cpu());
283 static void lru_add_drain_per_cpu(struct work_struct *dummy)
289 * Returns 0 for success
291 int lru_add_drain_all(void)
293 return schedule_on_each_cpu(lru_add_drain_per_cpu);
297 * Batched page_cache_release(). Decrement the reference count on all the
298 * passed pages. If it fell to zero then remove the page from the LRU and
301 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
302 * for the remainder of the operation.
304 * The locking in this function is against shrink_inactive_list(): we recheck
305 * the page count inside the lock to see whether shrink_inactive_list()
306 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
309 void release_pages(struct page **pages, int nr, int cold)
312 struct pagevec pages_to_free;
313 struct zone *zone = NULL;
314 unsigned long uninitialized_var(flags);
316 pagevec_init(&pages_to_free, cold);
317 for (i = 0; i < nr; i++) {
318 struct page *page = pages[i];
320 if (unlikely(PageCompound(page))) {
322 spin_unlock_irqrestore(&zone->lru_lock, flags);
325 put_compound_page(page);
329 if (!put_page_testzero(page))
333 struct zone *pagezone = page_zone(page);
335 if (pagezone != zone) {
337 spin_unlock_irqrestore(&zone->lru_lock,
340 spin_lock_irqsave(&zone->lru_lock, flags);
342 VM_BUG_ON(!PageLRU(page));
343 __ClearPageLRU(page);
344 del_page_from_lru(zone, page);
347 if (!pagevec_add(&pages_to_free, page)) {
349 spin_unlock_irqrestore(&zone->lru_lock, flags);
352 __pagevec_free(&pages_to_free);
353 pagevec_reinit(&pages_to_free);
357 spin_unlock_irqrestore(&zone->lru_lock, flags);
359 pagevec_free(&pages_to_free);
363 * The pages which we're about to release may be in the deferred lru-addition
364 * queues. That would prevent them from really being freed right now. That's
365 * OK from a correctness point of view but is inefficient - those pages may be
366 * cache-warm and we want to give them back to the page allocator ASAP.
368 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
369 * and __pagevec_lru_add_active() call release_pages() directly to avoid
372 void __pagevec_release(struct pagevec *pvec)
375 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
376 pagevec_reinit(pvec);
379 EXPORT_SYMBOL(__pagevec_release);
382 * Add the passed pages to the LRU, then drop the caller's refcount
383 * on them. Reinitialises the caller's pagevec.
385 void ____pagevec_lru_add(struct pagevec *pvec, enum lru_list lru)
388 struct zone *zone = NULL;
389 VM_BUG_ON(is_unevictable_lru(lru));
391 for (i = 0; i < pagevec_count(pvec); i++) {
392 struct page *page = pvec->pages[i];
393 struct zone *pagezone = page_zone(page);
396 if (pagezone != zone) {
398 spin_unlock_irq(&zone->lru_lock);
400 spin_lock_irq(&zone->lru_lock);
402 VM_BUG_ON(PageActive(page));
403 VM_BUG_ON(PageUnevictable(page));
404 VM_BUG_ON(PageLRU(page));
406 file = is_file_lru(lru);
407 zone->recent_scanned[file]++;
408 if (is_active_lru(lru)) {
410 zone->recent_rotated[file]++;
412 add_page_to_lru_list(zone, page, lru);
415 spin_unlock_irq(&zone->lru_lock);
416 release_pages(pvec->pages, pvec->nr, pvec->cold);
417 pagevec_reinit(pvec);
420 EXPORT_SYMBOL(____pagevec_lru_add);
423 * Try to drop buffers from the pages in a pagevec
425 void pagevec_strip(struct pagevec *pvec)
429 for (i = 0; i < pagevec_count(pvec); i++) {
430 struct page *page = pvec->pages[i];
432 if (PagePrivate(page) && trylock_page(page)) {
433 if (PagePrivate(page))
434 try_to_release_page(page, 0);
441 * pagevec_swap_free - try to free swap space from the pages in a pagevec
442 * @pvec: pagevec with swapcache pages to free the swap space of
444 * The caller needs to hold an extra reference to each page and
445 * not hold the page lock on the pages. This function uses a
446 * trylock on the page lock so it may not always free the swap
447 * space associated with a page.
449 void pagevec_swap_free(struct pagevec *pvec)
453 for (i = 0; i < pagevec_count(pvec); i++) {
454 struct page *page = pvec->pages[i];
456 if (PageSwapCache(page) && trylock_page(page)) {
457 try_to_free_swap(page);
464 * pagevec_lookup - gang pagecache lookup
465 * @pvec: Where the resulting pages are placed
466 * @mapping: The address_space to search
467 * @start: The starting page index
468 * @nr_pages: The maximum number of pages
470 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
471 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
472 * reference against the pages in @pvec.
474 * The search returns a group of mapping-contiguous pages with ascending
475 * indexes. There may be holes in the indices due to not-present pages.
477 * pagevec_lookup() returns the number of pages which were found.
479 unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
480 pgoff_t start, unsigned nr_pages)
482 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
483 return pagevec_count(pvec);
486 EXPORT_SYMBOL(pagevec_lookup);
488 unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
489 pgoff_t *index, int tag, unsigned nr_pages)
491 pvec->nr = find_get_pages_tag(mapping, index, tag,
492 nr_pages, pvec->pages);
493 return pagevec_count(pvec);
496 EXPORT_SYMBOL(pagevec_lookup_tag);
500 * We tolerate a little inaccuracy to avoid ping-ponging the counter between
503 #define ACCT_THRESHOLD max(16, NR_CPUS * 2)
505 static DEFINE_PER_CPU(long, committed_space);
507 void vm_acct_memory(long pages)
512 local = &__get_cpu_var(committed_space);
514 if (*local > ACCT_THRESHOLD || *local < -ACCT_THRESHOLD) {
515 atomic_long_add(*local, &vm_committed_space);
521 #ifdef CONFIG_HOTPLUG_CPU
523 /* Drop the CPU's cached committed space back into the central pool. */
524 static int cpu_swap_callback(struct notifier_block *nfb,
525 unsigned long action,
530 committed = &per_cpu(committed_space, (long)hcpu);
531 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
532 atomic_long_add(*committed, &vm_committed_space);
534 drain_cpu_pagevecs((long)hcpu);
538 #endif /* CONFIG_HOTPLUG_CPU */
539 #endif /* CONFIG_SMP */
542 * Perform any setup for the swap system
544 void __init swap_setup(void)
546 unsigned long megs = num_physpages >> (20 - PAGE_SHIFT);
549 bdi_init(swapper_space.backing_dev_info);
552 /* Use a smaller cluster for small-memory machines */
558 * Right now other parts of the system means that we
559 * _really_ don't want to cluster much more
561 #ifdef CONFIG_HOTPLUG_CPU
562 hotcpu_notifier(cpu_swap_callback, 0);