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/percpu_counter.h>
27 #include <linux/percpu.h>
28 #include <linux/cpu.h>
29 #include <linux/notifier.h>
30 #include <linux/backing-dev.h>
31 #include <linux/memcontrol.h>
32 #include <linux/gfp.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);
41 static DEFINE_PER_CPU(struct pagevec, lru_deactivate_pvecs);
44 * This path almost never happens for VM activity - pages are normally
45 * freed via pagevecs. But it gets used by networking.
47 static void __page_cache_release(struct page *page)
50 struct zone *zone = page_zone(page);
51 struct lruvec *lruvec;
54 spin_lock_irqsave(&zone->lru_lock, flags);
55 lruvec = mem_cgroup_page_lruvec(page, zone);
56 VM_BUG_ON(!PageLRU(page));
58 del_page_from_lru_list(page, lruvec, page_off_lru(page));
59 spin_unlock_irqrestore(&zone->lru_lock, flags);
63 static void __put_single_page(struct page *page)
65 __page_cache_release(page);
66 free_hot_cold_page(page, 0);
69 static void __put_compound_page(struct page *page)
71 compound_page_dtor *dtor;
73 __page_cache_release(page);
74 dtor = get_compound_page_dtor(page);
78 static void put_compound_page(struct page *page)
80 if (unlikely(PageTail(page))) {
81 /* __split_huge_page_refcount can run under us */
82 struct page *page_head = compound_trans_head(page);
84 if (likely(page != page_head &&
85 get_page_unless_zero(page_head))) {
89 * THP can not break up slab pages so avoid taking
90 * compound_lock(). Slab performs non-atomic bit ops
91 * on page->flags for better performance. In particular
92 * slab_unlock() in slub used to be a hot path. It is
93 * still hot on arches that do not support
94 * this_cpu_cmpxchg_double().
96 if (PageSlab(page_head)) {
98 if (put_page_testzero(page_head))
101 atomic_dec(&page->_mapcount);
107 * page_head wasn't a dangling pointer but it
108 * may not be a head page anymore by the time
109 * we obtain the lock. That is ok as long as it
110 * can't be freed from under us.
112 flags = compound_lock_irqsave(page_head);
113 if (unlikely(!PageTail(page))) {
114 /* __split_huge_page_refcount run before us */
115 compound_unlock_irqrestore(page_head, flags);
117 if (put_page_testzero(page_head))
118 __put_single_page(page_head);
120 if (put_page_testzero(page))
121 __put_single_page(page);
124 VM_BUG_ON(page_head != page->first_page);
126 * We can release the refcount taken by
127 * get_page_unless_zero() now that
128 * __split_huge_page_refcount() is blocked on
131 if (put_page_testzero(page_head))
133 /* __split_huge_page_refcount will wait now */
134 VM_BUG_ON(page_mapcount(page) <= 0);
135 atomic_dec(&page->_mapcount);
136 VM_BUG_ON(atomic_read(&page_head->_count) <= 0);
137 VM_BUG_ON(atomic_read(&page->_count) != 0);
138 compound_unlock_irqrestore(page_head, flags);
141 if (put_page_testzero(page_head)) {
142 if (PageHead(page_head))
143 __put_compound_page(page_head);
145 __put_single_page(page_head);
148 /* page_head is a dangling pointer */
149 VM_BUG_ON(PageTail(page));
152 } else if (put_page_testzero(page)) {
154 __put_compound_page(page);
156 __put_single_page(page);
160 void put_page(struct page *page)
162 if (unlikely(PageCompound(page)))
163 put_compound_page(page);
164 else if (put_page_testzero(page))
165 __put_single_page(page);
167 EXPORT_SYMBOL(put_page);
170 * This function is exported but must not be called by anything other
171 * than get_page(). It implements the slow path of get_page().
173 bool __get_page_tail(struct page *page)
176 * This takes care of get_page() if run on a tail page
177 * returned by one of the get_user_pages/follow_page variants.
178 * get_user_pages/follow_page itself doesn't need the compound
179 * lock because it runs __get_page_tail_foll() under the
180 * proper PT lock that already serializes against
185 struct page *page_head = compound_trans_head(page);
187 if (likely(page != page_head && get_page_unless_zero(page_head))) {
189 /* Ref to put_compound_page() comment. */
190 if (PageSlab(page_head)) {
191 if (likely(PageTail(page))) {
192 __get_page_tail_foll(page, false);
201 * page_head wasn't a dangling pointer but it
202 * may not be a head page anymore by the time
203 * we obtain the lock. That is ok as long as it
204 * can't be freed from under us.
206 flags = compound_lock_irqsave(page_head);
207 /* here __split_huge_page_refcount won't run anymore */
208 if (likely(PageTail(page))) {
209 __get_page_tail_foll(page, false);
212 compound_unlock_irqrestore(page_head, flags);
218 EXPORT_SYMBOL(__get_page_tail);
221 * put_pages_list() - release a list of pages
222 * @pages: list of pages threaded on page->lru
224 * Release a list of pages which are strung together on page.lru. Currently
225 * used by read_cache_pages() and related error recovery code.
227 void put_pages_list(struct list_head *pages)
229 while (!list_empty(pages)) {
232 victim = list_entry(pages->prev, struct page, lru);
233 list_del(&victim->lru);
234 page_cache_release(victim);
237 EXPORT_SYMBOL(put_pages_list);
240 * get_kernel_pages() - pin kernel pages in memory
241 * @kiov: An array of struct kvec structures
242 * @nr_segs: number of segments to pin
243 * @write: pinning for read/write, currently ignored
244 * @pages: array that receives pointers to the pages pinned.
245 * Should be at least nr_segs long.
247 * Returns number of pages pinned. This may be fewer than the number
248 * requested. If nr_pages is 0 or negative, returns 0. If no pages
249 * were pinned, returns -errno. Each page returned must be released
250 * with a put_page() call when it is finished with.
252 int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write,
257 for (seg = 0; seg < nr_segs; seg++) {
258 if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE))
261 /* virt_to_page sanity checks the PFN */
262 pages[seg] = virt_to_page(kiov[seg].iov_base);
263 page_cache_get(pages[seg]);
268 EXPORT_SYMBOL_GPL(get_kernel_pages);
271 * get_kernel_page() - pin a kernel page in memory
272 * @start: starting kernel address
273 * @write: pinning for read/write, currently ignored
274 * @pages: array that receives pointer to the page pinned.
275 * Must be at least nr_segs long.
277 * Returns 1 if page is pinned. If the page was not pinned, returns
278 * -errno. The page returned must be released with a put_page() call
279 * when it is finished with.
281 int get_kernel_page(unsigned long start, int write, struct page **pages)
283 const struct kvec kiov = {
284 .iov_base = (void *)start,
288 return get_kernel_pages(&kiov, 1, write, pages);
290 EXPORT_SYMBOL_GPL(get_kernel_page);
292 static void pagevec_lru_move_fn(struct pagevec *pvec,
293 void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg),
297 struct zone *zone = NULL;
298 struct lruvec *lruvec;
299 unsigned long flags = 0;
301 for (i = 0; i < pagevec_count(pvec); i++) {
302 struct page *page = pvec->pages[i];
303 struct zone *pagezone = page_zone(page);
305 if (pagezone != zone) {
307 spin_unlock_irqrestore(&zone->lru_lock, flags);
309 spin_lock_irqsave(&zone->lru_lock, flags);
312 lruvec = mem_cgroup_page_lruvec(page, zone);
313 (*move_fn)(page, lruvec, arg);
316 spin_unlock_irqrestore(&zone->lru_lock, flags);
317 release_pages(pvec->pages, pvec->nr, pvec->cold);
318 pagevec_reinit(pvec);
321 static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec,
326 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
327 enum lru_list lru = page_lru_base_type(page);
328 list_move_tail(&page->lru, &lruvec->lists[lru]);
334 * pagevec_move_tail() must be called with IRQ disabled.
335 * Otherwise this may cause nasty races.
337 static void pagevec_move_tail(struct pagevec *pvec)
341 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
342 __count_vm_events(PGROTATED, pgmoved);
346 * Writeback is about to end against a page which has been marked for immediate
347 * reclaim. If it still appears to be reclaimable, move it to the tail of the
350 void rotate_reclaimable_page(struct page *page)
352 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
353 !PageUnevictable(page) && PageLRU(page)) {
354 struct pagevec *pvec;
357 page_cache_get(page);
358 local_irq_save(flags);
359 pvec = &__get_cpu_var(lru_rotate_pvecs);
360 if (!pagevec_add(pvec, page))
361 pagevec_move_tail(pvec);
362 local_irq_restore(flags);
366 static void update_page_reclaim_stat(struct lruvec *lruvec,
367 int file, int rotated)
369 struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
371 reclaim_stat->recent_scanned[file]++;
373 reclaim_stat->recent_rotated[file]++;
376 static void __activate_page(struct page *page, struct lruvec *lruvec,
379 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
380 int file = page_is_file_cache(page);
381 int lru = page_lru_base_type(page);
383 del_page_from_lru_list(page, lruvec, lru);
386 add_page_to_lru_list(page, lruvec, lru);
388 __count_vm_event(PGACTIVATE);
389 update_page_reclaim_stat(lruvec, file, 1);
394 static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
396 static void activate_page_drain(int cpu)
398 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
400 if (pagevec_count(pvec))
401 pagevec_lru_move_fn(pvec, __activate_page, NULL);
404 void activate_page(struct page *page)
406 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
407 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
409 page_cache_get(page);
410 if (!pagevec_add(pvec, page))
411 pagevec_lru_move_fn(pvec, __activate_page, NULL);
412 put_cpu_var(activate_page_pvecs);
417 static inline void activate_page_drain(int cpu)
421 void activate_page(struct page *page)
423 struct zone *zone = page_zone(page);
425 spin_lock_irq(&zone->lru_lock);
426 __activate_page(page, mem_cgroup_page_lruvec(page, zone), NULL);
427 spin_unlock_irq(&zone->lru_lock);
432 * Mark a page as having seen activity.
434 * inactive,unreferenced -> inactive,referenced
435 * inactive,referenced -> active,unreferenced
436 * active,unreferenced -> active,referenced
438 void mark_page_accessed(struct page *page)
440 if (!PageActive(page) && !PageUnevictable(page) &&
441 PageReferenced(page) && PageLRU(page)) {
443 ClearPageReferenced(page);
444 } else if (!PageReferenced(page)) {
445 SetPageReferenced(page);
448 EXPORT_SYMBOL(mark_page_accessed);
450 void __lru_cache_add(struct page *page, enum lru_list lru)
452 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru];
454 page_cache_get(page);
455 if (!pagevec_add(pvec, page))
456 __pagevec_lru_add(pvec, lru);
457 put_cpu_var(lru_add_pvecs);
459 EXPORT_SYMBOL(__lru_cache_add);
462 * lru_cache_add_lru - add a page to a page list
463 * @page: the page to be added to the LRU.
464 * @lru: the LRU list to which the page is added.
466 void lru_cache_add_lru(struct page *page, enum lru_list lru)
468 if (PageActive(page)) {
469 VM_BUG_ON(PageUnevictable(page));
470 ClearPageActive(page);
471 } else if (PageUnevictable(page)) {
472 VM_BUG_ON(PageActive(page));
473 ClearPageUnevictable(page);
476 VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page));
477 __lru_cache_add(page, lru);
481 * add_page_to_unevictable_list - add a page to the unevictable list
482 * @page: the page to be added to the unevictable list
484 * Add page directly to its zone's unevictable list. To avoid races with
485 * tasks that might be making the page evictable, through eg. munlock,
486 * munmap or exit, while it's not on the lru, we want to add the page
487 * while it's locked or otherwise "invisible" to other tasks. This is
488 * difficult to do when using the pagevec cache, so bypass that.
490 void add_page_to_unevictable_list(struct page *page)
492 struct zone *zone = page_zone(page);
493 struct lruvec *lruvec;
495 spin_lock_irq(&zone->lru_lock);
496 lruvec = mem_cgroup_page_lruvec(page, zone);
497 SetPageUnevictable(page);
499 add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE);
500 spin_unlock_irq(&zone->lru_lock);
504 * If the page can not be invalidated, it is moved to the
505 * inactive list to speed up its reclaim. It is moved to the
506 * head of the list, rather than the tail, to give the flusher
507 * threads some time to write it out, as this is much more
508 * effective than the single-page writeout from reclaim.
510 * If the page isn't page_mapped and dirty/writeback, the page
511 * could reclaim asap using PG_reclaim.
513 * 1. active, mapped page -> none
514 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
515 * 3. inactive, mapped page -> none
516 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
517 * 5. inactive, clean -> inactive, tail
520 * In 4, why it moves inactive's head, the VM expects the page would
521 * be write it out by flusher threads as this is much more effective
522 * than the single-page writeout from reclaim.
524 static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec,
533 if (PageUnevictable(page))
536 /* Some processes are using the page */
537 if (page_mapped(page))
540 active = PageActive(page);
541 file = page_is_file_cache(page);
542 lru = page_lru_base_type(page);
544 del_page_from_lru_list(page, lruvec, lru + active);
545 ClearPageActive(page);
546 ClearPageReferenced(page);
547 add_page_to_lru_list(page, lruvec, lru);
549 if (PageWriteback(page) || PageDirty(page)) {
551 * PG_reclaim could be raced with end_page_writeback
552 * It can make readahead confusing. But race window
553 * is _really_ small and it's non-critical problem.
555 SetPageReclaim(page);
558 * The page's writeback ends up during pagevec
559 * We moves tha page into tail of inactive.
561 list_move_tail(&page->lru, &lruvec->lists[lru]);
562 __count_vm_event(PGROTATED);
566 __count_vm_event(PGDEACTIVATE);
567 update_page_reclaim_stat(lruvec, file, 0);
571 * Drain pages out of the cpu's pagevecs.
572 * Either "cpu" is the current CPU, and preemption has already been
573 * disabled; or "cpu" is being hot-unplugged, and is already dead.
575 void lru_add_drain_cpu(int cpu)
577 struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu);
578 struct pagevec *pvec;
582 pvec = &pvecs[lru - LRU_BASE];
583 if (pagevec_count(pvec))
584 __pagevec_lru_add(pvec, lru);
587 pvec = &per_cpu(lru_rotate_pvecs, cpu);
588 if (pagevec_count(pvec)) {
591 /* No harm done if a racing interrupt already did this */
592 local_irq_save(flags);
593 pagevec_move_tail(pvec);
594 local_irq_restore(flags);
597 pvec = &per_cpu(lru_deactivate_pvecs, cpu);
598 if (pagevec_count(pvec))
599 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
601 activate_page_drain(cpu);
605 * deactivate_page - forcefully deactivate a page
606 * @page: page to deactivate
608 * This function hints the VM that @page is a good reclaim candidate,
609 * for example if its invalidation fails due to the page being dirty
610 * or under writeback.
612 void deactivate_page(struct page *page)
615 * In a workload with many unevictable page such as mprotect, unevictable
616 * page deactivation for accelerating reclaim is pointless.
618 if (PageUnevictable(page))
621 if (likely(get_page_unless_zero(page))) {
622 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
624 if (!pagevec_add(pvec, page))
625 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
626 put_cpu_var(lru_deactivate_pvecs);
630 void lru_add_drain(void)
632 lru_add_drain_cpu(get_cpu());
636 static void lru_add_drain_per_cpu(struct work_struct *dummy)
642 * Returns 0 for success
644 int lru_add_drain_all(void)
646 return schedule_on_each_cpu(lru_add_drain_per_cpu);
650 * Batched page_cache_release(). Decrement the reference count on all the
651 * passed pages. If it fell to zero then remove the page from the LRU and
654 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
655 * for the remainder of the operation.
657 * The locking in this function is against shrink_inactive_list(): we recheck
658 * the page count inside the lock to see whether shrink_inactive_list()
659 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
662 void release_pages(struct page **pages, int nr, int cold)
665 LIST_HEAD(pages_to_free);
666 struct zone *zone = NULL;
667 struct lruvec *lruvec;
668 unsigned long uninitialized_var(flags);
670 for (i = 0; i < nr; i++) {
671 struct page *page = pages[i];
673 if (unlikely(PageCompound(page))) {
675 spin_unlock_irqrestore(&zone->lru_lock, flags);
678 put_compound_page(page);
682 if (!put_page_testzero(page))
686 struct zone *pagezone = page_zone(page);
688 if (pagezone != zone) {
690 spin_unlock_irqrestore(&zone->lru_lock,
693 spin_lock_irqsave(&zone->lru_lock, flags);
696 lruvec = mem_cgroup_page_lruvec(page, zone);
697 VM_BUG_ON(!PageLRU(page));
698 __ClearPageLRU(page);
699 del_page_from_lru_list(page, lruvec, page_off_lru(page));
702 list_add(&page->lru, &pages_to_free);
705 spin_unlock_irqrestore(&zone->lru_lock, flags);
707 free_hot_cold_page_list(&pages_to_free, cold);
709 EXPORT_SYMBOL(release_pages);
712 * The pages which we're about to release may be in the deferred lru-addition
713 * queues. That would prevent them from really being freed right now. That's
714 * OK from a correctness point of view but is inefficient - those pages may be
715 * cache-warm and we want to give them back to the page allocator ASAP.
717 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
718 * and __pagevec_lru_add_active() call release_pages() directly to avoid
721 void __pagevec_release(struct pagevec *pvec)
724 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
725 pagevec_reinit(pvec);
727 EXPORT_SYMBOL(__pagevec_release);
729 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
730 /* used by __split_huge_page_refcount() */
731 void lru_add_page_tail(struct page *page, struct page *page_tail,
732 struct lruvec *lruvec)
734 int uninitialized_var(active);
738 VM_BUG_ON(!PageHead(page));
739 VM_BUG_ON(PageCompound(page_tail));
740 VM_BUG_ON(PageLRU(page_tail));
741 VM_BUG_ON(NR_CPUS != 1 &&
742 !spin_is_locked(&lruvec_zone(lruvec)->lru_lock));
744 SetPageLRU(page_tail);
746 if (page_evictable(page_tail, NULL)) {
747 if (PageActive(page)) {
748 SetPageActive(page_tail);
750 lru = LRU_ACTIVE_ANON;
753 lru = LRU_INACTIVE_ANON;
756 SetPageUnevictable(page_tail);
757 lru = LRU_UNEVICTABLE;
760 if (likely(PageLRU(page)))
761 list_add_tail(&page_tail->lru, &page->lru);
763 struct list_head *list_head;
765 * Head page has not yet been counted, as an hpage,
766 * so we must account for each subpage individually.
768 * Use the standard add function to put page_tail on the list,
769 * but then correct its position so they all end up in order.
771 add_page_to_lru_list(page_tail, lruvec, lru);
772 list_head = page_tail->lru.prev;
773 list_move_tail(&page_tail->lru, list_head);
776 if (!PageUnevictable(page))
777 update_page_reclaim_stat(lruvec, file, active);
779 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
781 static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
784 enum lru_list lru = (enum lru_list)arg;
785 int file = is_file_lru(lru);
786 int active = is_active_lru(lru);
788 VM_BUG_ON(PageActive(page));
789 VM_BUG_ON(PageUnevictable(page));
790 VM_BUG_ON(PageLRU(page));
795 add_page_to_lru_list(page, lruvec, lru);
796 update_page_reclaim_stat(lruvec, file, active);
800 * Add the passed pages to the LRU, then drop the caller's refcount
801 * on them. Reinitialises the caller's pagevec.
803 void __pagevec_lru_add(struct pagevec *pvec, enum lru_list lru)
805 VM_BUG_ON(is_unevictable_lru(lru));
807 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, (void *)lru);
809 EXPORT_SYMBOL(__pagevec_lru_add);
812 * pagevec_lookup - gang pagecache lookup
813 * @pvec: Where the resulting pages are placed
814 * @mapping: The address_space to search
815 * @start: The starting page index
816 * @nr_pages: The maximum number of pages
818 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
819 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
820 * reference against the pages in @pvec.
822 * The search returns a group of mapping-contiguous pages with ascending
823 * indexes. There may be holes in the indices due to not-present pages.
825 * pagevec_lookup() returns the number of pages which were found.
827 unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
828 pgoff_t start, unsigned nr_pages)
830 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
831 return pagevec_count(pvec);
833 EXPORT_SYMBOL(pagevec_lookup);
835 unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
836 pgoff_t *index, int tag, unsigned nr_pages)
838 pvec->nr = find_get_pages_tag(mapping, index, tag,
839 nr_pages, pvec->pages);
840 return pagevec_count(pvec);
842 EXPORT_SYMBOL(pagevec_lookup_tag);
845 * Perform any setup for the swap system
847 void __init swap_setup(void)
849 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
852 bdi_init(swapper_space.backing_dev_info);
855 /* Use a smaller cluster for small-memory machines */
861 * Right now other parts of the system means that we
862 * _really_ don't want to cluster much more