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>
37 #define CREATE_TRACE_POINTS
38 #include <trace/events/pagemap.h>
40 /* How many pages do we try to swap or page in/out together? */
43 static DEFINE_PER_CPU(struct pagevec, lru_add_pvec);
44 static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
45 static DEFINE_PER_CPU(struct pagevec, lru_deactivate_pvecs);
48 * This path almost never happens for VM activity - pages are normally
49 * freed via pagevecs. But it gets used by networking.
51 static void __page_cache_release(struct page *page)
54 struct zone *zone = page_zone(page);
55 struct lruvec *lruvec;
58 spin_lock_irqsave(&zone->lru_lock, flags);
59 lruvec = mem_cgroup_page_lruvec(page, zone);
60 VM_BUG_ON(!PageLRU(page));
62 del_page_from_lru_list(page, lruvec, page_off_lru(page));
63 spin_unlock_irqrestore(&zone->lru_lock, flags);
67 static void __put_single_page(struct page *page)
69 __page_cache_release(page);
70 free_hot_cold_page(page, 0);
73 static void __put_compound_page(struct page *page)
75 compound_page_dtor *dtor;
77 __page_cache_release(page);
78 dtor = get_compound_page_dtor(page);
82 static void put_compound_page(struct page *page)
84 if (unlikely(PageTail(page))) {
85 /* __split_huge_page_refcount can run under us */
86 struct page *page_head = compound_trans_head(page);
89 * THP can not break up slab pages so avoid taking
90 * compound_lock() and skip the tail page refcounting
91 * (in _mapcount) too. Slab performs non-atomic bit
92 * ops on page->flags for better performance. In
93 * particular slab_unlock() in slub used to be a hot
94 * path. It is still hot on arches that do not support
95 * this_cpu_cmpxchg_double().
97 * If "page" is part of a slab or hugetlbfs page it
98 * cannot be splitted and the head page cannot change
99 * from under us. And if "page" is part of a THP page
100 * under splitting, if the head page pointed by the
101 * THP tail isn't a THP head anymore, we'll find
102 * PageTail clear after smp_rmb() and we'll treat it
105 if (!__compound_tail_refcounted(page_head)) {
107 * If "page" is a THP tail, we must read the tail page
108 * flags after the head page flags. The
109 * split_huge_page side enforces write memory
110 * barriers between clearing PageTail and before the
111 * head page can be freed and reallocated.
114 if (likely(PageTail(page))) {
116 * __split_huge_page_refcount
119 VM_BUG_ON(!PageHead(page_head));
120 VM_BUG_ON(page_mapcount(page) != 0);
121 if (put_page_testzero(page_head)) {
123 * If this is the tail of a
124 * slab compound page, the
125 * tail pin must not be the
126 * last reference held on the
127 * page, because the PG_slab
128 * cannot be cleared before
129 * all tail pins (which skips
131 * refcounting) have been
132 * released. For hugetlbfs the
133 * tail pin may be the last
134 * reference on the page
136 * PageHeadHuge will not go
137 * away until the compound
138 * page enters the buddy
141 VM_BUG_ON(PageSlab(page_head));
142 __put_compound_page(page_head);
147 * __split_huge_page_refcount
148 * run before us, "page" was a
149 * THP tail. The split
150 * page_head has been freed
151 * and reallocated as slab or
152 * hugetlbfs page of smaller
153 * order (only possible if
154 * reallocated as slab on
160 if (likely(page != page_head &&
161 get_page_unless_zero(page_head))) {
165 * page_head wasn't a dangling pointer but it
166 * may not be a head page anymore by the time
167 * we obtain the lock. That is ok as long as it
168 * can't be freed from under us.
170 flags = compound_lock_irqsave(page_head);
171 if (unlikely(!PageTail(page))) {
172 /* __split_huge_page_refcount run before us */
173 compound_unlock_irqrestore(page_head, flags);
174 if (put_page_testzero(page_head)) {
176 * The head page may have been
177 * freed and reallocated as a
178 * compound page of smaller
179 * order and then freed again.
180 * All we know is that it
181 * cannot have become: a THP
182 * page, a compound page of
183 * higher order, a tail page.
184 * That is because we still
185 * hold the refcount of the
187 * page_head was the THP head
190 if (PageHead(page_head))
191 __put_compound_page(page_head);
193 __put_single_page(page_head);
196 if (put_page_testzero(page))
197 __put_single_page(page);
200 VM_BUG_ON(page_head != page->first_page);
202 * We can release the refcount taken by
203 * get_page_unless_zero() now that
204 * __split_huge_page_refcount() is blocked on
207 if (put_page_testzero(page_head))
209 /* __split_huge_page_refcount will wait now */
210 VM_BUG_ON(page_mapcount(page) <= 0);
211 atomic_dec(&page->_mapcount);
212 VM_BUG_ON(atomic_read(&page_head->_count) <= 0);
213 VM_BUG_ON(atomic_read(&page->_count) != 0);
214 compound_unlock_irqrestore(page_head, flags);
216 if (put_page_testzero(page_head)) {
217 if (PageHead(page_head))
218 __put_compound_page(page_head);
220 __put_single_page(page_head);
223 /* page_head is a dangling pointer */
224 VM_BUG_ON(PageTail(page));
227 } else if (put_page_testzero(page)) {
229 __put_compound_page(page);
231 __put_single_page(page);
235 void put_page(struct page *page)
237 if (unlikely(PageCompound(page)))
238 put_compound_page(page);
239 else if (put_page_testzero(page))
240 __put_single_page(page);
242 EXPORT_SYMBOL(put_page);
245 * This function is exported but must not be called by anything other
246 * than get_page(). It implements the slow path of get_page().
248 bool __get_page_tail(struct page *page)
251 * This takes care of get_page() if run on a tail page
252 * returned by one of the get_user_pages/follow_page variants.
253 * get_user_pages/follow_page itself doesn't need the compound
254 * lock because it runs __get_page_tail_foll() under the
255 * proper PT lock that already serializes against
260 struct page *page_head = compound_trans_head(page);
262 /* Ref to put_compound_page() comment. */
263 if (!__compound_tail_refcounted(page_head)) {
265 if (likely(PageTail(page))) {
267 * This is a hugetlbfs page or a slab
268 * page. __split_huge_page_refcount
271 VM_BUG_ON(!PageHead(page_head));
272 __get_page_tail_foll(page, true);
276 * __split_huge_page_refcount run
277 * before us, "page" was a THP
278 * tail. The split page_head has been
279 * freed and reallocated as slab or
280 * hugetlbfs page of smaller order
281 * (only possible if reallocated as
289 if (likely(page != page_head && get_page_unless_zero(page_head))) {
291 * page_head wasn't a dangling pointer but it
292 * may not be a head page anymore by the time
293 * we obtain the lock. That is ok as long as it
294 * can't be freed from under us.
296 flags = compound_lock_irqsave(page_head);
297 /* here __split_huge_page_refcount won't run anymore */
298 if (likely(PageTail(page))) {
299 __get_page_tail_foll(page, false);
302 compound_unlock_irqrestore(page_head, flags);
308 EXPORT_SYMBOL(__get_page_tail);
311 * put_pages_list() - release a list of pages
312 * @pages: list of pages threaded on page->lru
314 * Release a list of pages which are strung together on page.lru. Currently
315 * used by read_cache_pages() and related error recovery code.
317 void put_pages_list(struct list_head *pages)
319 while (!list_empty(pages)) {
322 victim = list_entry(pages->prev, struct page, lru);
323 list_del(&victim->lru);
324 page_cache_release(victim);
327 EXPORT_SYMBOL(put_pages_list);
330 * get_kernel_pages() - pin kernel pages in memory
331 * @kiov: An array of struct kvec structures
332 * @nr_segs: number of segments to pin
333 * @write: pinning for read/write, currently ignored
334 * @pages: array that receives pointers to the pages pinned.
335 * Should be at least nr_segs long.
337 * Returns number of pages pinned. This may be fewer than the number
338 * requested. If nr_pages is 0 or negative, returns 0. If no pages
339 * were pinned, returns -errno. Each page returned must be released
340 * with a put_page() call when it is finished with.
342 int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write,
347 for (seg = 0; seg < nr_segs; seg++) {
348 if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE))
351 pages[seg] = kmap_to_page(kiov[seg].iov_base);
352 page_cache_get(pages[seg]);
357 EXPORT_SYMBOL_GPL(get_kernel_pages);
360 * get_kernel_page() - pin a kernel page in memory
361 * @start: starting kernel address
362 * @write: pinning for read/write, currently ignored
363 * @pages: array that receives pointer to the page pinned.
364 * Must be at least nr_segs long.
366 * Returns 1 if page is pinned. If the page was not pinned, returns
367 * -errno. The page returned must be released with a put_page() call
368 * when it is finished with.
370 int get_kernel_page(unsigned long start, int write, struct page **pages)
372 const struct kvec kiov = {
373 .iov_base = (void *)start,
377 return get_kernel_pages(&kiov, 1, write, pages);
379 EXPORT_SYMBOL_GPL(get_kernel_page);
381 static void pagevec_lru_move_fn(struct pagevec *pvec,
382 void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg),
386 struct zone *zone = NULL;
387 struct lruvec *lruvec;
388 unsigned long flags = 0;
390 for (i = 0; i < pagevec_count(pvec); i++) {
391 struct page *page = pvec->pages[i];
392 struct zone *pagezone = page_zone(page);
394 if (pagezone != zone) {
396 spin_unlock_irqrestore(&zone->lru_lock, flags);
398 spin_lock_irqsave(&zone->lru_lock, flags);
401 lruvec = mem_cgroup_page_lruvec(page, zone);
402 (*move_fn)(page, lruvec, arg);
405 spin_unlock_irqrestore(&zone->lru_lock, flags);
406 release_pages(pvec->pages, pvec->nr, pvec->cold);
407 pagevec_reinit(pvec);
410 static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec,
415 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
416 enum lru_list lru = page_lru_base_type(page);
417 list_move_tail(&page->lru, &lruvec->lists[lru]);
423 * pagevec_move_tail() must be called with IRQ disabled.
424 * Otherwise this may cause nasty races.
426 static void pagevec_move_tail(struct pagevec *pvec)
430 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
431 __count_vm_events(PGROTATED, pgmoved);
435 * Writeback is about to end against a page which has been marked for immediate
436 * reclaim. If it still appears to be reclaimable, move it to the tail of the
439 void rotate_reclaimable_page(struct page *page)
441 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
442 !PageUnevictable(page) && PageLRU(page)) {
443 struct pagevec *pvec;
446 page_cache_get(page);
447 local_irq_save(flags);
448 pvec = &__get_cpu_var(lru_rotate_pvecs);
449 if (!pagevec_add(pvec, page))
450 pagevec_move_tail(pvec);
451 local_irq_restore(flags);
455 static void update_page_reclaim_stat(struct lruvec *lruvec,
456 int file, int rotated)
458 struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
460 reclaim_stat->recent_scanned[file]++;
462 reclaim_stat->recent_rotated[file]++;
465 static void __activate_page(struct page *page, struct lruvec *lruvec,
468 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
469 int file = page_is_file_cache(page);
470 int lru = page_lru_base_type(page);
472 del_page_from_lru_list(page, lruvec, lru);
475 add_page_to_lru_list(page, lruvec, lru);
476 trace_mm_lru_activate(page, page_to_pfn(page));
478 __count_vm_event(PGACTIVATE);
479 update_page_reclaim_stat(lruvec, file, 1);
484 static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
486 static void activate_page_drain(int cpu)
488 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
490 if (pagevec_count(pvec))
491 pagevec_lru_move_fn(pvec, __activate_page, NULL);
494 static bool need_activate_page_drain(int cpu)
496 return pagevec_count(&per_cpu(activate_page_pvecs, cpu)) != 0;
499 void activate_page(struct page *page)
501 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
502 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
504 page_cache_get(page);
505 if (!pagevec_add(pvec, page))
506 pagevec_lru_move_fn(pvec, __activate_page, NULL);
507 put_cpu_var(activate_page_pvecs);
512 static inline void activate_page_drain(int cpu)
516 static bool need_activate_page_drain(int cpu)
521 void activate_page(struct page *page)
523 struct zone *zone = page_zone(page);
525 spin_lock_irq(&zone->lru_lock);
526 __activate_page(page, mem_cgroup_page_lruvec(page, zone), NULL);
527 spin_unlock_irq(&zone->lru_lock);
531 static void __lru_cache_activate_page(struct page *page)
533 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
537 * Search backwards on the optimistic assumption that the page being
538 * activated has just been added to this pagevec. Note that only
539 * the local pagevec is examined as a !PageLRU page could be in the
540 * process of being released, reclaimed, migrated or on a remote
541 * pagevec that is currently being drained. Furthermore, marking
542 * a remote pagevec's page PageActive potentially hits a race where
543 * a page is marked PageActive just after it is added to the inactive
544 * list causing accounting errors and BUG_ON checks to trigger.
546 for (i = pagevec_count(pvec) - 1; i >= 0; i--) {
547 struct page *pagevec_page = pvec->pages[i];
549 if (pagevec_page == page) {
555 put_cpu_var(lru_add_pvec);
559 * Mark a page as having seen activity.
561 * inactive,unreferenced -> inactive,referenced
562 * inactive,referenced -> active,unreferenced
563 * active,unreferenced -> active,referenced
565 void mark_page_accessed(struct page *page)
567 if (!PageActive(page) && !PageUnevictable(page) &&
568 PageReferenced(page)) {
571 * If the page is on the LRU, queue it for activation via
572 * activate_page_pvecs. Otherwise, assume the page is on a
573 * pagevec, mark it active and it'll be moved to the active
574 * LRU on the next drain.
579 __lru_cache_activate_page(page);
580 ClearPageReferenced(page);
581 } else if (!PageReferenced(page)) {
582 SetPageReferenced(page);
585 EXPORT_SYMBOL(mark_page_accessed);
588 * Queue the page for addition to the LRU via pagevec. The decision on whether
589 * to add the page to the [in]active [file|anon] list is deferred until the
590 * pagevec is drained. This gives a chance for the caller of __lru_cache_add()
591 * have the page added to the active list using mark_page_accessed().
593 void __lru_cache_add(struct page *page)
595 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
597 page_cache_get(page);
598 if (!pagevec_space(pvec))
599 __pagevec_lru_add(pvec);
600 pagevec_add(pvec, page);
601 put_cpu_var(lru_add_pvec);
603 EXPORT_SYMBOL(__lru_cache_add);
606 * lru_cache_add - add a page to a page list
607 * @page: the page to be added to the LRU.
609 void lru_cache_add(struct page *page)
611 VM_BUG_ON(PageActive(page) && PageUnevictable(page));
612 VM_BUG_ON(PageLRU(page));
613 __lru_cache_add(page);
617 * add_page_to_unevictable_list - add a page to the unevictable list
618 * @page: the page to be added to the unevictable list
620 * Add page directly to its zone's unevictable list. To avoid races with
621 * tasks that might be making the page evictable, through eg. munlock,
622 * munmap or exit, while it's not on the lru, we want to add the page
623 * while it's locked or otherwise "invisible" to other tasks. This is
624 * difficult to do when using the pagevec cache, so bypass that.
626 void add_page_to_unevictable_list(struct page *page)
628 struct zone *zone = page_zone(page);
629 struct lruvec *lruvec;
631 spin_lock_irq(&zone->lru_lock);
632 lruvec = mem_cgroup_page_lruvec(page, zone);
633 ClearPageActive(page);
634 SetPageUnevictable(page);
636 add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE);
637 spin_unlock_irq(&zone->lru_lock);
641 * If the page can not be invalidated, it is moved to the
642 * inactive list to speed up its reclaim. It is moved to the
643 * head of the list, rather than the tail, to give the flusher
644 * threads some time to write it out, as this is much more
645 * effective than the single-page writeout from reclaim.
647 * If the page isn't page_mapped and dirty/writeback, the page
648 * could reclaim asap using PG_reclaim.
650 * 1. active, mapped page -> none
651 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
652 * 3. inactive, mapped page -> none
653 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
654 * 5. inactive, clean -> inactive, tail
657 * In 4, why it moves inactive's head, the VM expects the page would
658 * be write it out by flusher threads as this is much more effective
659 * than the single-page writeout from reclaim.
661 static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec,
670 if (PageUnevictable(page))
673 /* Some processes are using the page */
674 if (page_mapped(page))
677 active = PageActive(page);
678 file = page_is_file_cache(page);
679 lru = page_lru_base_type(page);
681 del_page_from_lru_list(page, lruvec, lru + active);
682 ClearPageActive(page);
683 ClearPageReferenced(page);
684 add_page_to_lru_list(page, lruvec, lru);
686 if (PageWriteback(page) || PageDirty(page)) {
688 * PG_reclaim could be raced with end_page_writeback
689 * It can make readahead confusing. But race window
690 * is _really_ small and it's non-critical problem.
692 SetPageReclaim(page);
695 * The page's writeback ends up during pagevec
696 * We moves tha page into tail of inactive.
698 list_move_tail(&page->lru, &lruvec->lists[lru]);
699 __count_vm_event(PGROTATED);
703 __count_vm_event(PGDEACTIVATE);
704 update_page_reclaim_stat(lruvec, file, 0);
708 * Drain pages out of the cpu's pagevecs.
709 * Either "cpu" is the current CPU, and preemption has already been
710 * disabled; or "cpu" is being hot-unplugged, and is already dead.
712 void lru_add_drain_cpu(int cpu)
714 struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu);
716 if (pagevec_count(pvec))
717 __pagevec_lru_add(pvec);
719 pvec = &per_cpu(lru_rotate_pvecs, cpu);
720 if (pagevec_count(pvec)) {
723 /* No harm done if a racing interrupt already did this */
724 local_irq_save(flags);
725 pagevec_move_tail(pvec);
726 local_irq_restore(flags);
729 pvec = &per_cpu(lru_deactivate_pvecs, cpu);
730 if (pagevec_count(pvec))
731 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
733 activate_page_drain(cpu);
737 * deactivate_page - forcefully deactivate a page
738 * @page: page to deactivate
740 * This function hints the VM that @page is a good reclaim candidate,
741 * for example if its invalidation fails due to the page being dirty
742 * or under writeback.
744 void deactivate_page(struct page *page)
747 * In a workload with many unevictable page such as mprotect, unevictable
748 * page deactivation for accelerating reclaim is pointless.
750 if (PageUnevictable(page))
753 if (likely(get_page_unless_zero(page))) {
754 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
756 if (!pagevec_add(pvec, page))
757 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
758 put_cpu_var(lru_deactivate_pvecs);
762 void lru_add_drain(void)
764 lru_add_drain_cpu(get_cpu());
768 static void lru_add_drain_per_cpu(struct work_struct *dummy)
773 static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
775 void lru_add_drain_all(void)
777 static DEFINE_MUTEX(lock);
778 static struct cpumask has_work;
783 cpumask_clear(&has_work);
785 for_each_online_cpu(cpu) {
786 struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
788 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
789 pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
790 pagevec_count(&per_cpu(lru_deactivate_pvecs, cpu)) ||
791 need_activate_page_drain(cpu)) {
792 INIT_WORK(work, lru_add_drain_per_cpu);
793 schedule_work_on(cpu, work);
794 cpumask_set_cpu(cpu, &has_work);
798 for_each_cpu(cpu, &has_work)
799 flush_work(&per_cpu(lru_add_drain_work, cpu));
806 * Batched page_cache_release(). Decrement the reference count on all the
807 * passed pages. If it fell to zero then remove the page from the LRU and
810 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
811 * for the remainder of the operation.
813 * The locking in this function is against shrink_inactive_list(): we recheck
814 * the page count inside the lock to see whether shrink_inactive_list()
815 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
818 void release_pages(struct page **pages, int nr, int cold)
821 LIST_HEAD(pages_to_free);
822 struct zone *zone = NULL;
823 struct lruvec *lruvec;
824 unsigned long uninitialized_var(flags);
826 for (i = 0; i < nr; i++) {
827 struct page *page = pages[i];
829 if (unlikely(PageCompound(page))) {
831 spin_unlock_irqrestore(&zone->lru_lock, flags);
834 put_compound_page(page);
838 if (!put_page_testzero(page))
842 struct zone *pagezone = page_zone(page);
844 if (pagezone != zone) {
846 spin_unlock_irqrestore(&zone->lru_lock,
849 spin_lock_irqsave(&zone->lru_lock, flags);
852 lruvec = mem_cgroup_page_lruvec(page, zone);
853 VM_BUG_ON(!PageLRU(page));
854 __ClearPageLRU(page);
855 del_page_from_lru_list(page, lruvec, page_off_lru(page));
858 /* Clear Active bit in case of parallel mark_page_accessed */
859 ClearPageActive(page);
861 list_add(&page->lru, &pages_to_free);
864 spin_unlock_irqrestore(&zone->lru_lock, flags);
866 free_hot_cold_page_list(&pages_to_free, cold);
868 EXPORT_SYMBOL(release_pages);
871 * The pages which we're about to release may be in the deferred lru-addition
872 * queues. That would prevent them from really being freed right now. That's
873 * OK from a correctness point of view but is inefficient - those pages may be
874 * cache-warm and we want to give them back to the page allocator ASAP.
876 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
877 * and __pagevec_lru_add_active() call release_pages() directly to avoid
880 void __pagevec_release(struct pagevec *pvec)
883 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
884 pagevec_reinit(pvec);
886 EXPORT_SYMBOL(__pagevec_release);
888 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
889 /* used by __split_huge_page_refcount() */
890 void lru_add_page_tail(struct page *page, struct page *page_tail,
891 struct lruvec *lruvec, struct list_head *list)
895 VM_BUG_ON(!PageHead(page));
896 VM_BUG_ON(PageCompound(page_tail));
897 VM_BUG_ON(PageLRU(page_tail));
898 VM_BUG_ON(NR_CPUS != 1 &&
899 !spin_is_locked(&lruvec_zone(lruvec)->lru_lock));
902 SetPageLRU(page_tail);
904 if (likely(PageLRU(page)))
905 list_add_tail(&page_tail->lru, &page->lru);
907 /* page reclaim is reclaiming a huge page */
909 list_add_tail(&page_tail->lru, list);
911 struct list_head *list_head;
913 * Head page has not yet been counted, as an hpage,
914 * so we must account for each subpage individually.
916 * Use the standard add function to put page_tail on the list,
917 * but then correct its position so they all end up in order.
919 add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail));
920 list_head = page_tail->lru.prev;
921 list_move_tail(&page_tail->lru, list_head);
924 if (!PageUnevictable(page))
925 update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
927 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
929 static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
932 int file = page_is_file_cache(page);
933 int active = PageActive(page);
934 enum lru_list lru = page_lru(page);
936 VM_BUG_ON(PageLRU(page));
939 add_page_to_lru_list(page, lruvec, lru);
940 update_page_reclaim_stat(lruvec, file, active);
941 trace_mm_lru_insertion(page, page_to_pfn(page), lru, trace_pagemap_flags(page));
945 * Add the passed pages to the LRU, then drop the caller's refcount
946 * on them. Reinitialises the caller's pagevec.
948 void __pagevec_lru_add(struct pagevec *pvec)
950 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
952 EXPORT_SYMBOL(__pagevec_lru_add);
955 * pagevec_lookup - gang pagecache lookup
956 * @pvec: Where the resulting pages are placed
957 * @mapping: The address_space to search
958 * @start: The starting page index
959 * @nr_pages: The maximum number of pages
961 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
962 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
963 * reference against the pages in @pvec.
965 * The search returns a group of mapping-contiguous pages with ascending
966 * indexes. There may be holes in the indices due to not-present pages.
968 * pagevec_lookup() returns the number of pages which were found.
970 unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
971 pgoff_t start, unsigned nr_pages)
973 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
974 return pagevec_count(pvec);
976 EXPORT_SYMBOL(pagevec_lookup);
978 unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
979 pgoff_t *index, int tag, unsigned nr_pages)
981 pvec->nr = find_get_pages_tag(mapping, index, tag,
982 nr_pages, pvec->pages);
983 return pagevec_count(pvec);
985 EXPORT_SYMBOL(pagevec_lookup_tag);
988 * Perform any setup for the swap system
990 void __init swap_setup(void)
992 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
996 if (bdi_init(swapper_spaces[0].backing_dev_info))
997 panic("Failed to init swap bdi");
998 for (i = 0; i < MAX_SWAPFILES; i++) {
999 spin_lock_init(&swapper_spaces[i].tree_lock);
1000 INIT_LIST_HEAD(&swapper_spaces[i].i_mmap_nonlinear);
1004 /* Use a smaller cluster for small-memory machines */
1010 * Right now other parts of the system means that we
1011 * _really_ don't want to cluster much more