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/export.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>
33 #include <linux/gfp.h>
37 /* How many pages do we try to swap or page in/out together? */
40 static DEFINE_PER_CPU(struct pagevec[NR_LRU_LISTS], lru_add_pvecs);
41 static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
42 static DEFINE_PER_CPU(struct pagevec, lru_deactivate_pvecs);
45 * This path almost never happens for VM activity - pages are normally
46 * freed via pagevecs. But it gets used by networking.
48 static void __page_cache_release(struct page *page)
52 struct zone *zone = page_zone(page);
54 spin_lock_irqsave(&zone->lru_lock, flags);
55 VM_BUG_ON(!PageLRU(page));
57 del_page_from_lru(zone, page);
58 spin_unlock_irqrestore(&zone->lru_lock, flags);
62 static void __put_single_page(struct page *page)
64 __page_cache_release(page);
65 free_hot_cold_page(page, 0);
68 static void __put_compound_page(struct page *page)
70 compound_page_dtor *dtor;
72 __page_cache_release(page);
73 dtor = get_compound_page_dtor(page);
77 static void put_compound_page(struct page *page)
79 if (unlikely(PageTail(page))) {
80 /* __split_huge_page_refcount can run under us */
81 struct page *page_head = compound_trans_head(page);
83 if (likely(page != page_head &&
84 get_page_unless_zero(page_head))) {
87 * page_head wasn't a dangling pointer but it
88 * may not be a head page anymore by the time
89 * we obtain the lock. That is ok as long as it
90 * can't be freed from under us.
92 flags = compound_lock_irqsave(page_head);
93 if (unlikely(!PageTail(page))) {
94 /* __split_huge_page_refcount run before us */
95 compound_unlock_irqrestore(page_head, flags);
96 VM_BUG_ON(PageHead(page_head));
97 if (put_page_testzero(page_head))
98 __put_single_page(page_head);
100 if (put_page_testzero(page))
101 __put_single_page(page);
104 VM_BUG_ON(page_head != page->first_page);
106 * We can release the refcount taken by
107 * get_page_unless_zero() now that
108 * __split_huge_page_refcount() is blocked on
111 if (put_page_testzero(page_head))
113 /* __split_huge_page_refcount will wait now */
114 VM_BUG_ON(page_mapcount(page) <= 0);
115 atomic_dec(&page->_mapcount);
116 VM_BUG_ON(atomic_read(&page_head->_count) <= 0);
117 VM_BUG_ON(atomic_read(&page->_count) != 0);
118 compound_unlock_irqrestore(page_head, flags);
119 if (put_page_testzero(page_head)) {
120 if (PageHead(page_head))
121 __put_compound_page(page_head);
123 __put_single_page(page_head);
126 /* page_head is a dangling pointer */
127 VM_BUG_ON(PageTail(page));
130 } else if (put_page_testzero(page)) {
132 __put_compound_page(page);
134 __put_single_page(page);
138 void put_page(struct page *page)
140 if (unlikely(PageCompound(page)))
141 put_compound_page(page);
142 else if (put_page_testzero(page))
143 __put_single_page(page);
145 EXPORT_SYMBOL(put_page);
148 * This function is exported but must not be called by anything other
149 * than get_page(). It implements the slow path of get_page().
151 bool __get_page_tail(struct page *page)
154 * This takes care of get_page() if run on a tail page
155 * returned by one of the get_user_pages/follow_page variants.
156 * get_user_pages/follow_page itself doesn't need the compound
157 * lock because it runs __get_page_tail_foll() under the
158 * proper PT lock that already serializes against
163 struct page *page_head = compound_trans_head(page);
165 if (likely(page != page_head && get_page_unless_zero(page_head))) {
167 * page_head wasn't a dangling pointer but it
168 * may not be a head page anymore by the time
169 * we obtain the lock. That is ok as long as it
170 * can't be freed from under us.
172 flags = compound_lock_irqsave(page_head);
173 /* here __split_huge_page_refcount won't run anymore */
174 if (likely(PageTail(page))) {
175 __get_page_tail_foll(page, false);
178 compound_unlock_irqrestore(page_head, flags);
184 EXPORT_SYMBOL(__get_page_tail);
187 * put_pages_list() - release a list of pages
188 * @pages: list of pages threaded on page->lru
190 * Release a list of pages which are strung together on page.lru. Currently
191 * used by read_cache_pages() and related error recovery code.
193 void put_pages_list(struct list_head *pages)
195 while (!list_empty(pages)) {
198 victim = list_entry(pages->prev, struct page, lru);
199 list_del(&victim->lru);
200 page_cache_release(victim);
203 EXPORT_SYMBOL(put_pages_list);
205 static void pagevec_lru_move_fn(struct pagevec *pvec,
206 void (*move_fn)(struct page *page, void *arg),
210 struct zone *zone = NULL;
211 unsigned long flags = 0;
213 for (i = 0; i < pagevec_count(pvec); i++) {
214 struct page *page = pvec->pages[i];
215 struct zone *pagezone = page_zone(page);
217 if (pagezone != zone) {
219 spin_unlock_irqrestore(&zone->lru_lock, flags);
221 spin_lock_irqsave(&zone->lru_lock, flags);
224 (*move_fn)(page, arg);
227 spin_unlock_irqrestore(&zone->lru_lock, flags);
228 release_pages(pvec->pages, pvec->nr, pvec->cold);
229 pagevec_reinit(pvec);
232 static void pagevec_move_tail_fn(struct page *page, void *arg)
236 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
237 enum lru_list lru = page_lru_base_type(page);
238 struct lruvec *lruvec;
240 lruvec = mem_cgroup_lru_move_lists(page_zone(page),
242 list_move_tail(&page->lru, &lruvec->lists[lru]);
248 * pagevec_move_tail() must be called with IRQ disabled.
249 * Otherwise this may cause nasty races.
251 static void pagevec_move_tail(struct pagevec *pvec)
255 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
256 __count_vm_events(PGROTATED, pgmoved);
260 * Writeback is about to end against a page which has been marked for immediate
261 * reclaim. If it still appears to be reclaimable, move it to the tail of the
264 void rotate_reclaimable_page(struct page *page)
266 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
267 !PageUnevictable(page) && PageLRU(page)) {
268 struct pagevec *pvec;
271 page_cache_get(page);
272 local_irq_save(flags);
273 pvec = &__get_cpu_var(lru_rotate_pvecs);
274 if (!pagevec_add(pvec, page))
275 pagevec_move_tail(pvec);
276 local_irq_restore(flags);
280 static void update_page_reclaim_stat(struct zone *zone, struct page *page,
281 int file, int rotated)
283 struct zone_reclaim_stat *reclaim_stat = &zone->reclaim_stat;
284 struct zone_reclaim_stat *memcg_reclaim_stat;
286 memcg_reclaim_stat = mem_cgroup_get_reclaim_stat_from_page(page);
288 reclaim_stat->recent_scanned[file]++;
290 reclaim_stat->recent_rotated[file]++;
292 if (!memcg_reclaim_stat)
295 memcg_reclaim_stat->recent_scanned[file]++;
297 memcg_reclaim_stat->recent_rotated[file]++;
300 static void __activate_page(struct page *page, void *arg)
302 struct zone *zone = page_zone(page);
304 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
305 int file = page_is_file_cache(page);
306 int lru = page_lru_base_type(page);
307 del_page_from_lru_list(zone, page, lru);
311 add_page_to_lru_list(zone, page, lru);
312 __count_vm_event(PGACTIVATE);
314 update_page_reclaim_stat(zone, page, file, 1);
319 static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
321 static void activate_page_drain(int cpu)
323 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
325 if (pagevec_count(pvec))
326 pagevec_lru_move_fn(pvec, __activate_page, NULL);
329 void activate_page(struct page *page)
331 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
332 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
334 page_cache_get(page);
335 if (!pagevec_add(pvec, page))
336 pagevec_lru_move_fn(pvec, __activate_page, NULL);
337 put_cpu_var(activate_page_pvecs);
342 static inline void activate_page_drain(int cpu)
346 void activate_page(struct page *page)
348 struct zone *zone = page_zone(page);
350 spin_lock_irq(&zone->lru_lock);
351 __activate_page(page, NULL);
352 spin_unlock_irq(&zone->lru_lock);
357 * Mark a page as having seen activity.
359 * inactive,unreferenced -> inactive,referenced
360 * inactive,referenced -> active,unreferenced
361 * active,unreferenced -> active,referenced
363 void mark_page_accessed(struct page *page)
365 if (!PageActive(page) && !PageUnevictable(page) &&
366 PageReferenced(page) && PageLRU(page)) {
368 ClearPageReferenced(page);
369 } else if (!PageReferenced(page)) {
370 SetPageReferenced(page);
374 EXPORT_SYMBOL(mark_page_accessed);
376 void __lru_cache_add(struct page *page, enum lru_list lru)
378 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru];
380 page_cache_get(page);
381 if (!pagevec_add(pvec, page))
382 ____pagevec_lru_add(pvec, lru);
383 put_cpu_var(lru_add_pvecs);
385 EXPORT_SYMBOL(__lru_cache_add);
388 * lru_cache_add_lru - add a page to a page list
389 * @page: the page to be added to the LRU.
390 * @lru: the LRU list to which the page is added.
392 void lru_cache_add_lru(struct page *page, enum lru_list lru)
394 if (PageActive(page)) {
395 VM_BUG_ON(PageUnevictable(page));
396 ClearPageActive(page);
397 } else if (PageUnevictable(page)) {
398 VM_BUG_ON(PageActive(page));
399 ClearPageUnevictable(page);
402 VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page));
403 __lru_cache_add(page, lru);
407 * add_page_to_unevictable_list - add a page to the unevictable list
408 * @page: the page to be added to the unevictable list
410 * Add page directly to its zone's unevictable list. To avoid races with
411 * tasks that might be making the page evictable, through eg. munlock,
412 * munmap or exit, while it's not on the lru, we want to add the page
413 * while it's locked or otherwise "invisible" to other tasks. This is
414 * difficult to do when using the pagevec cache, so bypass that.
416 void add_page_to_unevictable_list(struct page *page)
418 struct zone *zone = page_zone(page);
420 spin_lock_irq(&zone->lru_lock);
421 SetPageUnevictable(page);
423 add_page_to_lru_list(zone, page, LRU_UNEVICTABLE);
424 spin_unlock_irq(&zone->lru_lock);
428 * If the page can not be invalidated, it is moved to the
429 * inactive list to speed up its reclaim. It is moved to the
430 * head of the list, rather than the tail, to give the flusher
431 * threads some time to write it out, as this is much more
432 * effective than the single-page writeout from reclaim.
434 * If the page isn't page_mapped and dirty/writeback, the page
435 * could reclaim asap using PG_reclaim.
437 * 1. active, mapped page -> none
438 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
439 * 3. inactive, mapped page -> none
440 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
441 * 5. inactive, clean -> inactive, tail
444 * In 4, why it moves inactive's head, the VM expects the page would
445 * be write it out by flusher threads as this is much more effective
446 * than the single-page writeout from reclaim.
448 static void lru_deactivate_fn(struct page *page, void *arg)
452 struct zone *zone = page_zone(page);
457 if (PageUnevictable(page))
460 /* Some processes are using the page */
461 if (page_mapped(page))
464 active = PageActive(page);
466 file = page_is_file_cache(page);
467 lru = page_lru_base_type(page);
468 del_page_from_lru_list(zone, page, lru + active);
469 ClearPageActive(page);
470 ClearPageReferenced(page);
471 add_page_to_lru_list(zone, page, lru);
473 if (PageWriteback(page) || PageDirty(page)) {
475 * PG_reclaim could be raced with end_page_writeback
476 * It can make readahead confusing. But race window
477 * is _really_ small and it's non-critical problem.
479 SetPageReclaim(page);
481 struct lruvec *lruvec;
483 * The page's writeback ends up during pagevec
484 * We moves tha page into tail of inactive.
486 lruvec = mem_cgroup_lru_move_lists(zone, page, lru, lru);
487 list_move_tail(&page->lru, &lruvec->lists[lru]);
488 __count_vm_event(PGROTATED);
492 __count_vm_event(PGDEACTIVATE);
493 update_page_reclaim_stat(zone, page, file, 0);
497 * Drain pages out of the cpu's pagevecs.
498 * Either "cpu" is the current CPU, and preemption has already been
499 * disabled; or "cpu" is being hot-unplugged, and is already dead.
501 static void drain_cpu_pagevecs(int cpu)
503 struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu);
504 struct pagevec *pvec;
508 pvec = &pvecs[lru - LRU_BASE];
509 if (pagevec_count(pvec))
510 ____pagevec_lru_add(pvec, lru);
513 pvec = &per_cpu(lru_rotate_pvecs, cpu);
514 if (pagevec_count(pvec)) {
517 /* No harm done if a racing interrupt already did this */
518 local_irq_save(flags);
519 pagevec_move_tail(pvec);
520 local_irq_restore(flags);
523 pvec = &per_cpu(lru_deactivate_pvecs, cpu);
524 if (pagevec_count(pvec))
525 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
527 activate_page_drain(cpu);
531 * deactivate_page - forcefully deactivate a page
532 * @page: page to deactivate
534 * This function hints the VM that @page is a good reclaim candidate,
535 * for example if its invalidation fails due to the page being dirty
536 * or under writeback.
538 void deactivate_page(struct page *page)
541 * In a workload with many unevictable page such as mprotect, unevictable
542 * page deactivation for accelerating reclaim is pointless.
544 if (PageUnevictable(page))
547 if (likely(get_page_unless_zero(page))) {
548 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
550 if (!pagevec_add(pvec, page))
551 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
552 put_cpu_var(lru_deactivate_pvecs);
556 void lru_add_drain(void)
558 drain_cpu_pagevecs(get_cpu());
562 static void lru_add_drain_per_cpu(struct work_struct *dummy)
568 * Returns 0 for success
570 int lru_add_drain_all(void)
572 return schedule_on_each_cpu(lru_add_drain_per_cpu);
576 * Batched page_cache_release(). Decrement the reference count on all the
577 * passed pages. If it fell to zero then remove the page from the LRU and
580 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
581 * for the remainder of the operation.
583 * The locking in this function is against shrink_inactive_list(): we recheck
584 * the page count inside the lock to see whether shrink_inactive_list()
585 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
588 void release_pages(struct page **pages, int nr, int cold)
591 LIST_HEAD(pages_to_free);
592 struct zone *zone = NULL;
593 unsigned long uninitialized_var(flags);
595 for (i = 0; i < nr; i++) {
596 struct page *page = pages[i];
598 if (unlikely(PageCompound(page))) {
600 spin_unlock_irqrestore(&zone->lru_lock, flags);
603 put_compound_page(page);
607 if (!put_page_testzero(page))
611 struct zone *pagezone = page_zone(page);
613 if (pagezone != zone) {
615 spin_unlock_irqrestore(&zone->lru_lock,
618 spin_lock_irqsave(&zone->lru_lock, flags);
620 VM_BUG_ON(!PageLRU(page));
621 __ClearPageLRU(page);
622 del_page_from_lru(zone, page);
625 list_add(&page->lru, &pages_to_free);
628 spin_unlock_irqrestore(&zone->lru_lock, flags);
630 free_hot_cold_page_list(&pages_to_free, cold);
632 EXPORT_SYMBOL(release_pages);
635 * The pages which we're about to release may be in the deferred lru-addition
636 * queues. That would prevent them from really being freed right now. That's
637 * OK from a correctness point of view but is inefficient - those pages may be
638 * cache-warm and we want to give them back to the page allocator ASAP.
640 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
641 * and __pagevec_lru_add_active() call release_pages() directly to avoid
644 void __pagevec_release(struct pagevec *pvec)
647 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
648 pagevec_reinit(pvec);
651 EXPORT_SYMBOL(__pagevec_release);
653 /* used by __split_huge_page_refcount() */
654 void lru_add_page_tail(struct zone* zone,
655 struct page *page, struct page *page_tail)
661 VM_BUG_ON(!PageHead(page));
662 VM_BUG_ON(PageCompound(page_tail));
663 VM_BUG_ON(PageLRU(page_tail));
664 VM_BUG_ON(!spin_is_locked(&zone->lru_lock));
666 SetPageLRU(page_tail);
668 if (page_evictable(page_tail, NULL)) {
669 struct lruvec *lruvec;
671 if (PageActive(page)) {
672 SetPageActive(page_tail);
674 lru = LRU_ACTIVE_ANON;
677 lru = LRU_INACTIVE_ANON;
679 update_page_reclaim_stat(zone, page_tail, file, active);
680 lruvec = mem_cgroup_lru_add_list(zone, page_tail, lru);
681 if (likely(PageLRU(page)))
682 list_add(&page_tail->lru, page->lru.prev);
684 list_add(&page_tail->lru, &lruvec->lists[lru]);
685 __mod_zone_page_state(zone, NR_LRU_BASE + lru,
686 hpage_nr_pages(page_tail));
688 SetPageUnevictable(page_tail);
689 add_page_to_lru_list(zone, page_tail, LRU_UNEVICTABLE);
693 static void ____pagevec_lru_add_fn(struct page *page, void *arg)
695 enum lru_list lru = (enum lru_list)arg;
696 struct zone *zone = page_zone(page);
697 int file = is_file_lru(lru);
698 int active = is_active_lru(lru);
700 VM_BUG_ON(PageActive(page));
701 VM_BUG_ON(PageUnevictable(page));
702 VM_BUG_ON(PageLRU(page));
707 update_page_reclaim_stat(zone, page, file, active);
708 add_page_to_lru_list(zone, page, lru);
712 * Add the passed pages to the LRU, then drop the caller's refcount
713 * on them. Reinitialises the caller's pagevec.
715 void ____pagevec_lru_add(struct pagevec *pvec, enum lru_list lru)
717 VM_BUG_ON(is_unevictable_lru(lru));
719 pagevec_lru_move_fn(pvec, ____pagevec_lru_add_fn, (void *)lru);
722 EXPORT_SYMBOL(____pagevec_lru_add);
725 * Try to drop buffers from the pages in a pagevec
727 void pagevec_strip(struct pagevec *pvec)
731 for (i = 0; i < pagevec_count(pvec); i++) {
732 struct page *page = pvec->pages[i];
734 if (page_has_private(page) && trylock_page(page)) {
735 if (page_has_private(page))
736 try_to_release_page(page, 0);
743 * pagevec_lookup - gang pagecache lookup
744 * @pvec: Where the resulting pages are placed
745 * @mapping: The address_space to search
746 * @start: The starting page index
747 * @nr_pages: The maximum number of pages
749 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
750 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
751 * reference against the pages in @pvec.
753 * The search returns a group of mapping-contiguous pages with ascending
754 * indexes. There may be holes in the indices due to not-present pages.
756 * pagevec_lookup() returns the number of pages which were found.
758 unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
759 pgoff_t start, unsigned nr_pages)
761 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
762 return pagevec_count(pvec);
765 EXPORT_SYMBOL(pagevec_lookup);
767 unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
768 pgoff_t *index, int tag, unsigned nr_pages)
770 pvec->nr = find_get_pages_tag(mapping, index, tag,
771 nr_pages, pvec->pages);
772 return pagevec_count(pvec);
775 EXPORT_SYMBOL(pagevec_lookup_tag);
778 * Perform any setup for the swap system
780 void __init swap_setup(void)
782 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
785 bdi_init(swapper_space.backing_dev_info);
788 /* Use a smaller cluster for small-memory machines */
794 * Right now other parts of the system means that we
795 * _really_ don't want to cluster much more