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>
33 #include <linux/uio.h>
34 #include <linux/hugetlb.h>
38 #define CREATE_TRACE_POINTS
39 #include <trace/events/pagemap.h>
41 /* How many pages do we try to swap or page in/out together? */
44 static DEFINE_PER_CPU(struct pagevec, lru_add_pvec);
45 static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
46 static DEFINE_PER_CPU(struct pagevec, lru_deactivate_pvecs);
49 * This path almost never happens for VM activity - pages are normally
50 * freed via pagevecs. But it gets used by networking.
52 static void __page_cache_release(struct page *page)
55 struct zone *zone = page_zone(page);
56 struct lruvec *lruvec;
59 spin_lock_irqsave(&zone->lru_lock, flags);
60 lruvec = mem_cgroup_page_lruvec(page, zone);
61 VM_BUG_ON(!PageLRU(page));
63 del_page_from_lru_list(page, lruvec, page_off_lru(page));
64 spin_unlock_irqrestore(&zone->lru_lock, flags);
68 static void __put_single_page(struct page *page)
70 __page_cache_release(page);
71 free_hot_cold_page(page, 0);
74 static void __put_compound_page(struct page *page)
76 compound_page_dtor *dtor;
78 __page_cache_release(page);
79 dtor = get_compound_page_dtor(page);
83 static void put_compound_page(struct page *page)
85 if (unlikely(PageTail(page))) {
86 /* __split_huge_page_refcount can run under us */
87 struct page *page_head = compound_trans_head(page);
90 * THP can not break up slab pages so avoid taking
91 * compound_lock() and skip the tail page refcounting
92 * (in _mapcount) too. Slab performs non-atomic bit
93 * ops on page->flags for better performance. In
94 * particular slab_unlock() in slub used to be a hot
95 * path. It is still hot on arches that do not support
96 * this_cpu_cmpxchg_double().
98 * If "page" is part of a slab or hugetlbfs page it
99 * cannot be splitted and the head page cannot change
100 * from under us. And if "page" is part of a THP page
101 * under splitting, if the head page pointed by the
102 * THP tail isn't a THP head anymore, we'll find
103 * PageTail clear after smp_rmb() and we'll treat it
106 if (!__compound_tail_refcounted(page_head)) {
108 * If "page" is a THP tail, we must read the tail page
109 * flags after the head page flags. The
110 * split_huge_page side enforces write memory
111 * barriers between clearing PageTail and before the
112 * head page can be freed and reallocated.
115 if (likely(PageTail(page))) {
117 * __split_huge_page_refcount
120 VM_BUG_ON(!PageHead(page_head));
121 VM_BUG_ON(page_mapcount(page) != 0);
122 if (put_page_testzero(page_head)) {
124 * If this is the tail of a
125 * slab compound page, the
126 * tail pin must not be the
127 * last reference held on the
128 * page, because the PG_slab
129 * cannot be cleared before
130 * all tail pins (which skips
132 * refcounting) have been
133 * released. For hugetlbfs the
134 * tail pin may be the last
135 * reference on the page
137 * PageHeadHuge will not go
138 * away until the compound
139 * page enters the buddy
142 VM_BUG_ON(PageSlab(page_head));
143 __put_compound_page(page_head);
148 * __split_huge_page_refcount
149 * run before us, "page" was a
150 * THP tail. The split
151 * page_head has been freed
152 * and reallocated as slab or
153 * hugetlbfs page of smaller
154 * order (only possible if
155 * reallocated as slab on
161 if (likely(page != page_head &&
162 get_page_unless_zero(page_head))) {
166 * page_head wasn't a dangling pointer but it
167 * may not be a head page anymore by the time
168 * we obtain the lock. That is ok as long as it
169 * can't be freed from under us.
171 flags = compound_lock_irqsave(page_head);
172 if (unlikely(!PageTail(page))) {
173 /* __split_huge_page_refcount run before us */
174 compound_unlock_irqrestore(page_head, flags);
175 if (put_page_testzero(page_head)) {
177 * The head page may have been
178 * freed and reallocated as a
179 * compound page of smaller
180 * order and then freed again.
181 * All we know is that it
182 * cannot have become: a THP
183 * page, a compound page of
184 * higher order, a tail page.
185 * That is because we still
186 * hold the refcount of the
188 * page_head was the THP head
191 if (PageHead(page_head))
192 __put_compound_page(page_head);
194 __put_single_page(page_head);
197 if (put_page_testzero(page))
198 __put_single_page(page);
201 VM_BUG_ON(page_head != page->first_page);
203 * We can release the refcount taken by
204 * get_page_unless_zero() now that
205 * __split_huge_page_refcount() is blocked on
208 if (put_page_testzero(page_head))
210 /* __split_huge_page_refcount will wait now */
211 VM_BUG_ON(page_mapcount(page) <= 0);
212 atomic_dec(&page->_mapcount);
213 VM_BUG_ON(atomic_read(&page_head->_count) <= 0);
214 VM_BUG_ON(atomic_read(&page->_count) != 0);
215 compound_unlock_irqrestore(page_head, flags);
217 if (put_page_testzero(page_head)) {
218 if (PageHead(page_head))
219 __put_compound_page(page_head);
221 __put_single_page(page_head);
224 /* page_head is a dangling pointer */
225 VM_BUG_ON(PageTail(page));
228 } else if (put_page_testzero(page)) {
230 __put_compound_page(page);
232 __put_single_page(page);
236 void put_page(struct page *page)
238 if (unlikely(PageCompound(page)))
239 put_compound_page(page);
240 else if (put_page_testzero(page))
241 __put_single_page(page);
243 EXPORT_SYMBOL(put_page);
246 * This function is exported but must not be called by anything other
247 * than get_page(). It implements the slow path of get_page().
249 bool __get_page_tail(struct page *page)
252 * This takes care of get_page() if run on a tail page
253 * returned by one of the get_user_pages/follow_page variants.
254 * get_user_pages/follow_page itself doesn't need the compound
255 * lock because it runs __get_page_tail_foll() under the
256 * proper PT lock that already serializes against
261 struct page *page_head = compound_trans_head(page);
263 /* Ref to put_compound_page() comment. */
264 if (PageSlab(page_head) || PageHeadHuge(page_head)) {
266 if (likely(PageTail(page))) {
268 * This is a hugetlbfs page or a slab
269 * page. __split_huge_page_refcount
272 VM_BUG_ON(!PageHead(page_head));
273 __get_page_tail_foll(page, true);
277 * __split_huge_page_refcount run
278 * before us, "page" was a THP
279 * tail. The split page_head has been
280 * freed and reallocated as slab or
281 * hugetlbfs page of smaller order
282 * (only possible if reallocated as
290 if (likely(page != page_head && get_page_unless_zero(page_head))) {
292 * page_head wasn't a dangling pointer but it
293 * may not be a head page anymore by the time
294 * we obtain the lock. That is ok as long as it
295 * can't be freed from under us.
297 flags = compound_lock_irqsave(page_head);
298 /* here __split_huge_page_refcount won't run anymore */
299 if (likely(PageTail(page))) {
300 __get_page_tail_foll(page, false);
303 compound_unlock_irqrestore(page_head, flags);
309 EXPORT_SYMBOL(__get_page_tail);
312 * put_pages_list() - release a list of pages
313 * @pages: list of pages threaded on page->lru
315 * Release a list of pages which are strung together on page.lru. Currently
316 * used by read_cache_pages() and related error recovery code.
318 void put_pages_list(struct list_head *pages)
320 while (!list_empty(pages)) {
323 victim = list_entry(pages->prev, struct page, lru);
324 list_del(&victim->lru);
325 page_cache_release(victim);
328 EXPORT_SYMBOL(put_pages_list);
331 * get_kernel_pages() - pin kernel pages in memory
332 * @kiov: An array of struct kvec structures
333 * @nr_segs: number of segments to pin
334 * @write: pinning for read/write, currently ignored
335 * @pages: array that receives pointers to the pages pinned.
336 * Should be at least nr_segs long.
338 * Returns number of pages pinned. This may be fewer than the number
339 * requested. If nr_pages is 0 or negative, returns 0. If no pages
340 * were pinned, returns -errno. Each page returned must be released
341 * with a put_page() call when it is finished with.
343 int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write,
348 for (seg = 0; seg < nr_segs; seg++) {
349 if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE))
352 pages[seg] = kmap_to_page(kiov[seg].iov_base);
353 page_cache_get(pages[seg]);
358 EXPORT_SYMBOL_GPL(get_kernel_pages);
361 * get_kernel_page() - pin a kernel page in memory
362 * @start: starting kernel address
363 * @write: pinning for read/write, currently ignored
364 * @pages: array that receives pointer to the page pinned.
365 * Must be at least nr_segs long.
367 * Returns 1 if page is pinned. If the page was not pinned, returns
368 * -errno. The page returned must be released with a put_page() call
369 * when it is finished with.
371 int get_kernel_page(unsigned long start, int write, struct page **pages)
373 const struct kvec kiov = {
374 .iov_base = (void *)start,
378 return get_kernel_pages(&kiov, 1, write, pages);
380 EXPORT_SYMBOL_GPL(get_kernel_page);
382 static void pagevec_lru_move_fn(struct pagevec *pvec,
383 void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg),
387 struct zone *zone = NULL;
388 struct lruvec *lruvec;
389 unsigned long flags = 0;
391 for (i = 0; i < pagevec_count(pvec); i++) {
392 struct page *page = pvec->pages[i];
393 struct zone *pagezone = page_zone(page);
395 if (pagezone != zone) {
397 spin_unlock_irqrestore(&zone->lru_lock, flags);
399 spin_lock_irqsave(&zone->lru_lock, flags);
402 lruvec = mem_cgroup_page_lruvec(page, zone);
403 (*move_fn)(page, lruvec, arg);
406 spin_unlock_irqrestore(&zone->lru_lock, flags);
407 release_pages(pvec->pages, pvec->nr, pvec->cold);
408 pagevec_reinit(pvec);
411 static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec,
416 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
417 enum lru_list lru = page_lru_base_type(page);
418 list_move_tail(&page->lru, &lruvec->lists[lru]);
424 * pagevec_move_tail() must be called with IRQ disabled.
425 * Otherwise this may cause nasty races.
427 static void pagevec_move_tail(struct pagevec *pvec)
431 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
432 __count_vm_events(PGROTATED, pgmoved);
436 * Writeback is about to end against a page which has been marked for immediate
437 * reclaim. If it still appears to be reclaimable, move it to the tail of the
440 void rotate_reclaimable_page(struct page *page)
442 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
443 !PageUnevictable(page) && PageLRU(page)) {
444 struct pagevec *pvec;
447 page_cache_get(page);
448 local_irq_save(flags);
449 pvec = &__get_cpu_var(lru_rotate_pvecs);
450 if (!pagevec_add(pvec, page))
451 pagevec_move_tail(pvec);
452 local_irq_restore(flags);
456 static void update_page_reclaim_stat(struct lruvec *lruvec,
457 int file, int rotated)
459 struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
461 reclaim_stat->recent_scanned[file]++;
463 reclaim_stat->recent_rotated[file]++;
466 static void __activate_page(struct page *page, struct lruvec *lruvec,
469 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
470 int file = page_is_file_cache(page);
471 int lru = page_lru_base_type(page);
473 del_page_from_lru_list(page, lruvec, lru);
476 add_page_to_lru_list(page, lruvec, lru);
477 trace_mm_lru_activate(page, page_to_pfn(page));
479 __count_vm_event(PGACTIVATE);
480 update_page_reclaim_stat(lruvec, file, 1);
485 static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
487 static void activate_page_drain(int cpu)
489 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
491 if (pagevec_count(pvec))
492 pagevec_lru_move_fn(pvec, __activate_page, NULL);
495 static bool need_activate_page_drain(int cpu)
497 return pagevec_count(&per_cpu(activate_page_pvecs, cpu)) != 0;
500 void activate_page(struct page *page)
502 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
503 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
505 page_cache_get(page);
506 if (!pagevec_add(pvec, page))
507 pagevec_lru_move_fn(pvec, __activate_page, NULL);
508 put_cpu_var(activate_page_pvecs);
513 static inline void activate_page_drain(int cpu)
517 static bool need_activate_page_drain(int cpu)
522 void activate_page(struct page *page)
524 struct zone *zone = page_zone(page);
526 spin_lock_irq(&zone->lru_lock);
527 __activate_page(page, mem_cgroup_page_lruvec(page, zone), NULL);
528 spin_unlock_irq(&zone->lru_lock);
532 static void __lru_cache_activate_page(struct page *page)
534 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
538 * Search backwards on the optimistic assumption that the page being
539 * activated has just been added to this pagevec. Note that only
540 * the local pagevec is examined as a !PageLRU page could be in the
541 * process of being released, reclaimed, migrated or on a remote
542 * pagevec that is currently being drained. Furthermore, marking
543 * a remote pagevec's page PageActive potentially hits a race where
544 * a page is marked PageActive just after it is added to the inactive
545 * list causing accounting errors and BUG_ON checks to trigger.
547 for (i = pagevec_count(pvec) - 1; i >= 0; i--) {
548 struct page *pagevec_page = pvec->pages[i];
550 if (pagevec_page == page) {
556 put_cpu_var(lru_add_pvec);
560 * Mark a page as having seen activity.
562 * inactive,unreferenced -> inactive,referenced
563 * inactive,referenced -> active,unreferenced
564 * active,unreferenced -> active,referenced
566 void mark_page_accessed(struct page *page)
568 if (!PageActive(page) && !PageUnevictable(page) &&
569 PageReferenced(page)) {
572 * If the page is on the LRU, queue it for activation via
573 * activate_page_pvecs. Otherwise, assume the page is on a
574 * pagevec, mark it active and it'll be moved to the active
575 * LRU on the next drain.
580 __lru_cache_activate_page(page);
581 ClearPageReferenced(page);
582 } else if (!PageReferenced(page)) {
583 SetPageReferenced(page);
586 EXPORT_SYMBOL(mark_page_accessed);
589 * Queue the page for addition to the LRU via pagevec. The decision on whether
590 * to add the page to the [in]active [file|anon] list is deferred until the
591 * pagevec is drained. This gives a chance for the caller of __lru_cache_add()
592 * have the page added to the active list using mark_page_accessed().
594 void __lru_cache_add(struct page *page)
596 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
598 page_cache_get(page);
599 if (!pagevec_space(pvec))
600 __pagevec_lru_add(pvec);
601 pagevec_add(pvec, page);
602 put_cpu_var(lru_add_pvec);
604 EXPORT_SYMBOL(__lru_cache_add);
607 * lru_cache_add - add a page to a page list
608 * @page: the page to be added to the LRU.
610 void lru_cache_add(struct page *page)
612 VM_BUG_ON(PageActive(page) && PageUnevictable(page));
613 VM_BUG_ON(PageLRU(page));
614 __lru_cache_add(page);
618 * add_page_to_unevictable_list - add a page to the unevictable list
619 * @page: the page to be added to the unevictable list
621 * Add page directly to its zone's unevictable list. To avoid races with
622 * tasks that might be making the page evictable, through eg. munlock,
623 * munmap or exit, while it's not on the lru, we want to add the page
624 * while it's locked or otherwise "invisible" to other tasks. This is
625 * difficult to do when using the pagevec cache, so bypass that.
627 void add_page_to_unevictable_list(struct page *page)
629 struct zone *zone = page_zone(page);
630 struct lruvec *lruvec;
632 spin_lock_irq(&zone->lru_lock);
633 lruvec = mem_cgroup_page_lruvec(page, zone);
634 ClearPageActive(page);
635 SetPageUnevictable(page);
637 add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE);
638 spin_unlock_irq(&zone->lru_lock);
642 * If the page can not be invalidated, it is moved to the
643 * inactive list to speed up its reclaim. It is moved to the
644 * head of the list, rather than the tail, to give the flusher
645 * threads some time to write it out, as this is much more
646 * effective than the single-page writeout from reclaim.
648 * If the page isn't page_mapped and dirty/writeback, the page
649 * could reclaim asap using PG_reclaim.
651 * 1. active, mapped page -> none
652 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
653 * 3. inactive, mapped page -> none
654 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
655 * 5. inactive, clean -> inactive, tail
658 * In 4, why it moves inactive's head, the VM expects the page would
659 * be write it out by flusher threads as this is much more effective
660 * than the single-page writeout from reclaim.
662 static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec,
671 if (PageUnevictable(page))
674 /* Some processes are using the page */
675 if (page_mapped(page))
678 active = PageActive(page);
679 file = page_is_file_cache(page);
680 lru = page_lru_base_type(page);
682 del_page_from_lru_list(page, lruvec, lru + active);
683 ClearPageActive(page);
684 ClearPageReferenced(page);
685 add_page_to_lru_list(page, lruvec, lru);
687 if (PageWriteback(page) || PageDirty(page)) {
689 * PG_reclaim could be raced with end_page_writeback
690 * It can make readahead confusing. But race window
691 * is _really_ small and it's non-critical problem.
693 SetPageReclaim(page);
696 * The page's writeback ends up during pagevec
697 * We moves tha page into tail of inactive.
699 list_move_tail(&page->lru, &lruvec->lists[lru]);
700 __count_vm_event(PGROTATED);
704 __count_vm_event(PGDEACTIVATE);
705 update_page_reclaim_stat(lruvec, file, 0);
709 * Drain pages out of the cpu's pagevecs.
710 * Either "cpu" is the current CPU, and preemption has already been
711 * disabled; or "cpu" is being hot-unplugged, and is already dead.
713 void lru_add_drain_cpu(int cpu)
715 struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu);
717 if (pagevec_count(pvec))
718 __pagevec_lru_add(pvec);
720 pvec = &per_cpu(lru_rotate_pvecs, cpu);
721 if (pagevec_count(pvec)) {
724 /* No harm done if a racing interrupt already did this */
725 local_irq_save(flags);
726 pagevec_move_tail(pvec);
727 local_irq_restore(flags);
730 pvec = &per_cpu(lru_deactivate_pvecs, cpu);
731 if (pagevec_count(pvec))
732 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
734 activate_page_drain(cpu);
738 * deactivate_page - forcefully deactivate a page
739 * @page: page to deactivate
741 * This function hints the VM that @page is a good reclaim candidate,
742 * for example if its invalidation fails due to the page being dirty
743 * or under writeback.
745 void deactivate_page(struct page *page)
748 * In a workload with many unevictable page such as mprotect, unevictable
749 * page deactivation for accelerating reclaim is pointless.
751 if (PageUnevictable(page))
754 if (likely(get_page_unless_zero(page))) {
755 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
757 if (!pagevec_add(pvec, page))
758 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
759 put_cpu_var(lru_deactivate_pvecs);
763 void lru_add_drain(void)
765 lru_add_drain_cpu(get_cpu());
769 static void lru_add_drain_per_cpu(struct work_struct *dummy)
774 static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
776 void lru_add_drain_all(void)
778 static DEFINE_MUTEX(lock);
779 static struct cpumask has_work;
784 cpumask_clear(&has_work);
786 for_each_online_cpu(cpu) {
787 struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
789 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
790 pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
791 pagevec_count(&per_cpu(lru_deactivate_pvecs, cpu)) ||
792 need_activate_page_drain(cpu)) {
793 INIT_WORK(work, lru_add_drain_per_cpu);
794 schedule_work_on(cpu, work);
795 cpumask_set_cpu(cpu, &has_work);
799 for_each_cpu(cpu, &has_work)
800 flush_work(&per_cpu(lru_add_drain_work, cpu));
807 * Batched page_cache_release(). Decrement the reference count on all the
808 * passed pages. If it fell to zero then remove the page from the LRU and
811 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
812 * for the remainder of the operation.
814 * The locking in this function is against shrink_inactive_list(): we recheck
815 * the page count inside the lock to see whether shrink_inactive_list()
816 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
819 void release_pages(struct page **pages, int nr, int cold)
822 LIST_HEAD(pages_to_free);
823 struct zone *zone = NULL;
824 struct lruvec *lruvec;
825 unsigned long uninitialized_var(flags);
827 for (i = 0; i < nr; i++) {
828 struct page *page = pages[i];
830 if (unlikely(PageCompound(page))) {
832 spin_unlock_irqrestore(&zone->lru_lock, flags);
835 put_compound_page(page);
839 if (!put_page_testzero(page))
843 struct zone *pagezone = page_zone(page);
845 if (pagezone != zone) {
847 spin_unlock_irqrestore(&zone->lru_lock,
850 spin_lock_irqsave(&zone->lru_lock, flags);
853 lruvec = mem_cgroup_page_lruvec(page, zone);
854 VM_BUG_ON(!PageLRU(page));
855 __ClearPageLRU(page);
856 del_page_from_lru_list(page, lruvec, page_off_lru(page));
859 /* Clear Active bit in case of parallel mark_page_accessed */
860 ClearPageActive(page);
862 list_add(&page->lru, &pages_to_free);
865 spin_unlock_irqrestore(&zone->lru_lock, flags);
867 free_hot_cold_page_list(&pages_to_free, cold);
869 EXPORT_SYMBOL(release_pages);
872 * The pages which we're about to release may be in the deferred lru-addition
873 * queues. That would prevent them from really being freed right now. That's
874 * OK from a correctness point of view but is inefficient - those pages may be
875 * cache-warm and we want to give them back to the page allocator ASAP.
877 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
878 * and __pagevec_lru_add_active() call release_pages() directly to avoid
881 void __pagevec_release(struct pagevec *pvec)
884 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
885 pagevec_reinit(pvec);
887 EXPORT_SYMBOL(__pagevec_release);
889 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
890 /* used by __split_huge_page_refcount() */
891 void lru_add_page_tail(struct page *page, struct page *page_tail,
892 struct lruvec *lruvec, struct list_head *list)
896 VM_BUG_ON(!PageHead(page));
897 VM_BUG_ON(PageCompound(page_tail));
898 VM_BUG_ON(PageLRU(page_tail));
899 VM_BUG_ON(NR_CPUS != 1 &&
900 !spin_is_locked(&lruvec_zone(lruvec)->lru_lock));
903 SetPageLRU(page_tail);
905 if (likely(PageLRU(page)))
906 list_add_tail(&page_tail->lru, &page->lru);
908 /* page reclaim is reclaiming a huge page */
910 list_add_tail(&page_tail->lru, list);
912 struct list_head *list_head;
914 * Head page has not yet been counted, as an hpage,
915 * so we must account for each subpage individually.
917 * Use the standard add function to put page_tail on the list,
918 * but then correct its position so they all end up in order.
920 add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail));
921 list_head = page_tail->lru.prev;
922 list_move_tail(&page_tail->lru, list_head);
925 if (!PageUnevictable(page))
926 update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
928 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
930 static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
933 int file = page_is_file_cache(page);
934 int active = PageActive(page);
935 enum lru_list lru = page_lru(page);
937 VM_BUG_ON(PageLRU(page));
940 add_page_to_lru_list(page, lruvec, lru);
941 update_page_reclaim_stat(lruvec, file, active);
942 trace_mm_lru_insertion(page, page_to_pfn(page), lru, trace_pagemap_flags(page));
946 * Add the passed pages to the LRU, then drop the caller's refcount
947 * on them. Reinitialises the caller's pagevec.
949 void __pagevec_lru_add(struct pagevec *pvec)
951 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
953 EXPORT_SYMBOL(__pagevec_lru_add);
956 * pagevec_lookup - gang pagecache lookup
957 * @pvec: Where the resulting pages are placed
958 * @mapping: The address_space to search
959 * @start: The starting page index
960 * @nr_pages: The maximum number of pages
962 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
963 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
964 * reference against the pages in @pvec.
966 * The search returns a group of mapping-contiguous pages with ascending
967 * indexes. There may be holes in the indices due to not-present pages.
969 * pagevec_lookup() returns the number of pages which were found.
971 unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
972 pgoff_t start, unsigned nr_pages)
974 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
975 return pagevec_count(pvec);
977 EXPORT_SYMBOL(pagevec_lookup);
979 unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
980 pgoff_t *index, int tag, unsigned nr_pages)
982 pvec->nr = find_get_pages_tag(mapping, index, tag,
983 nr_pages, pvec->pages);
984 return pagevec_count(pvec);
986 EXPORT_SYMBOL(pagevec_lookup_tag);
989 * Perform any setup for the swap system
991 void __init swap_setup(void)
993 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
997 if (bdi_init(swapper_spaces[0].backing_dev_info))
998 panic("Failed to init swap bdi");
999 for (i = 0; i < MAX_SWAPFILES; i++) {
1000 spin_lock_init(&swapper_spaces[i].tree_lock);
1001 INIT_LIST_HEAD(&swapper_spaces[i].i_mmap_nonlinear);
1005 /* Use a smaller cluster for small-memory machines */
1011 * Right now other parts of the system means that we
1012 * _really_ don't want to cluster much more