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/memremap.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>
34 #include <linux/uio.h>
35 #include <linux/hugetlb.h>
36 #include <linux/page_idle.h>
40 #define CREATE_TRACE_POINTS
41 #include <trace/events/pagemap.h>
43 /* How many pages do we try to swap or page in/out together? */
46 static DEFINE_PER_CPU(struct pagevec, lru_add_pvec);
47 static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
48 static DEFINE_PER_CPU(struct pagevec, lru_deactivate_file_pvecs);
49 static DEFINE_PER_CPU(struct pagevec, lru_deactivate_pvecs);
51 static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
55 * This path almost never happens for VM activity - pages are normally
56 * freed via pagevecs. But it gets used by networking.
58 static void __page_cache_release(struct page *page)
61 struct zone *zone = page_zone(page);
62 struct lruvec *lruvec;
65 spin_lock_irqsave(&zone->lru_lock, flags);
66 lruvec = mem_cgroup_page_lruvec(page, zone);
67 VM_BUG_ON_PAGE(!PageLRU(page), page);
69 del_page_from_lru_list(page, lruvec, page_off_lru(page));
70 spin_unlock_irqrestore(&zone->lru_lock, flags);
72 mem_cgroup_uncharge(page);
75 static void __put_single_page(struct page *page)
77 __page_cache_release(page);
78 free_hot_cold_page(page, false);
81 static void __put_compound_page(struct page *page)
83 compound_page_dtor *dtor;
86 * __page_cache_release() is supposed to be called for thp, not for
87 * hugetlb. This is because hugetlb page does never have PageLRU set
88 * (it's never listed to any LRU lists) and no memcg routines should
89 * be called for hugetlb (it has a separate hugetlb_cgroup.)
92 __page_cache_release(page);
93 dtor = get_compound_page_dtor(page);
97 void __put_page(struct page *page)
99 if (unlikely(PageCompound(page)))
100 __put_compound_page(page);
102 __put_single_page(page);
104 EXPORT_SYMBOL(__put_page);
107 * put_pages_list() - release a list of pages
108 * @pages: list of pages threaded on page->lru
110 * Release a list of pages which are strung together on page.lru. Currently
111 * used by read_cache_pages() and related error recovery code.
113 void put_pages_list(struct list_head *pages)
115 while (!list_empty(pages)) {
118 victim = list_entry(pages->prev, struct page, lru);
119 list_del(&victim->lru);
123 EXPORT_SYMBOL(put_pages_list);
126 * get_kernel_pages() - pin kernel pages in memory
127 * @kiov: An array of struct kvec structures
128 * @nr_segs: number of segments to pin
129 * @write: pinning for read/write, currently ignored
130 * @pages: array that receives pointers to the pages pinned.
131 * Should be at least nr_segs long.
133 * Returns number of pages pinned. This may be fewer than the number
134 * requested. If nr_pages is 0 or negative, returns 0. If no pages
135 * were pinned, returns -errno. Each page returned must be released
136 * with a put_page() call when it is finished with.
138 int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write,
143 for (seg = 0; seg < nr_segs; seg++) {
144 if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE))
147 pages[seg] = kmap_to_page(kiov[seg].iov_base);
148 get_page(pages[seg]);
153 EXPORT_SYMBOL_GPL(get_kernel_pages);
156 * get_kernel_page() - pin a kernel page in memory
157 * @start: starting kernel address
158 * @write: pinning for read/write, currently ignored
159 * @pages: array that receives pointer to the page pinned.
160 * Must be at least nr_segs long.
162 * Returns 1 if page is pinned. If the page was not pinned, returns
163 * -errno. The page returned must be released with a put_page() call
164 * when it is finished with.
166 int get_kernel_page(unsigned long start, int write, struct page **pages)
168 const struct kvec kiov = {
169 .iov_base = (void *)start,
173 return get_kernel_pages(&kiov, 1, write, pages);
175 EXPORT_SYMBOL_GPL(get_kernel_page);
177 static void pagevec_lru_move_fn(struct pagevec *pvec,
178 void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg),
182 struct zone *zone = NULL;
183 struct lruvec *lruvec;
184 unsigned long flags = 0;
186 for (i = 0; i < pagevec_count(pvec); i++) {
187 struct page *page = pvec->pages[i];
188 struct zone *pagezone = page_zone(page);
190 if (pagezone != zone) {
192 spin_unlock_irqrestore(&zone->lru_lock, flags);
194 spin_lock_irqsave(&zone->lru_lock, flags);
197 lruvec = mem_cgroup_page_lruvec(page, zone);
198 (*move_fn)(page, lruvec, arg);
201 spin_unlock_irqrestore(&zone->lru_lock, flags);
202 release_pages(pvec->pages, pvec->nr, pvec->cold);
203 pagevec_reinit(pvec);
206 static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec,
211 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
212 enum lru_list lru = page_lru_base_type(page);
213 list_move_tail(&page->lru, &lruvec->lists[lru]);
219 * pagevec_move_tail() must be called with IRQ disabled.
220 * Otherwise this may cause nasty races.
222 static void pagevec_move_tail(struct pagevec *pvec)
226 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
227 __count_vm_events(PGROTATED, pgmoved);
231 * Writeback is about to end against a page which has been marked for immediate
232 * reclaim. If it still appears to be reclaimable, move it to the tail of the
235 void rotate_reclaimable_page(struct page *page)
237 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
238 !PageUnevictable(page) && PageLRU(page)) {
239 struct pagevec *pvec;
243 local_irq_save(flags);
244 pvec = this_cpu_ptr(&lru_rotate_pvecs);
245 if (!pagevec_add(pvec, page) || PageCompound(page))
246 pagevec_move_tail(pvec);
247 local_irq_restore(flags);
251 static void update_page_reclaim_stat(struct lruvec *lruvec,
252 int file, int rotated)
254 struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
256 reclaim_stat->recent_scanned[file]++;
258 reclaim_stat->recent_rotated[file]++;
261 static void __activate_page(struct page *page, struct lruvec *lruvec,
264 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
265 int file = page_is_file_cache(page);
266 int lru = page_lru_base_type(page);
268 del_page_from_lru_list(page, lruvec, lru);
271 add_page_to_lru_list(page, lruvec, lru);
272 trace_mm_lru_activate(page);
274 __count_vm_event(PGACTIVATE);
275 update_page_reclaim_stat(lruvec, file, 1);
280 static void activate_page_drain(int cpu)
282 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
284 if (pagevec_count(pvec))
285 pagevec_lru_move_fn(pvec, __activate_page, NULL);
288 static bool need_activate_page_drain(int cpu)
290 return pagevec_count(&per_cpu(activate_page_pvecs, cpu)) != 0;
293 void activate_page(struct page *page)
295 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
296 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
299 if (!pagevec_add(pvec, page) || PageCompound(page))
300 pagevec_lru_move_fn(pvec, __activate_page, NULL);
301 put_cpu_var(activate_page_pvecs);
306 static inline void activate_page_drain(int cpu)
310 static bool need_activate_page_drain(int cpu)
315 void activate_page(struct page *page)
317 struct zone *zone = page_zone(page);
319 spin_lock_irq(&zone->lru_lock);
320 __activate_page(page, mem_cgroup_page_lruvec(page, zone), NULL);
321 spin_unlock_irq(&zone->lru_lock);
325 static void __lru_cache_activate_page(struct page *page)
327 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
331 * Search backwards on the optimistic assumption that the page being
332 * activated has just been added to this pagevec. Note that only
333 * the local pagevec is examined as a !PageLRU page could be in the
334 * process of being released, reclaimed, migrated or on a remote
335 * pagevec that is currently being drained. Furthermore, marking
336 * a remote pagevec's page PageActive potentially hits a race where
337 * a page is marked PageActive just after it is added to the inactive
338 * list causing accounting errors and BUG_ON checks to trigger.
340 for (i = pagevec_count(pvec) - 1; i >= 0; i--) {
341 struct page *pagevec_page = pvec->pages[i];
343 if (pagevec_page == page) {
349 put_cpu_var(lru_add_pvec);
353 * Mark a page as having seen activity.
355 * inactive,unreferenced -> inactive,referenced
356 * inactive,referenced -> active,unreferenced
357 * active,unreferenced -> active,referenced
359 * When a newly allocated page is not yet visible, so safe for non-atomic ops,
360 * __SetPageReferenced(page) may be substituted for mark_page_accessed(page).
362 void mark_page_accessed(struct page *page)
364 page = compound_head(page);
365 if (!PageActive(page) && !PageUnevictable(page) &&
366 PageReferenced(page)) {
369 * If the page is on the LRU, queue it for activation via
370 * activate_page_pvecs. Otherwise, assume the page is on a
371 * pagevec, mark it active and it'll be moved to the active
372 * LRU on the next drain.
377 __lru_cache_activate_page(page);
378 ClearPageReferenced(page);
379 if (page_is_file_cache(page))
380 workingset_activation(page);
381 } else if (!PageReferenced(page)) {
382 SetPageReferenced(page);
384 if (page_is_idle(page))
385 clear_page_idle(page);
387 EXPORT_SYMBOL(mark_page_accessed);
389 static void __lru_cache_add(struct page *page)
391 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
394 if (!pagevec_add(pvec, page) || PageCompound(page))
395 __pagevec_lru_add(pvec);
396 put_cpu_var(lru_add_pvec);
400 * lru_cache_add: add a page to the page lists
401 * @page: the page to add
403 void lru_cache_add_anon(struct page *page)
405 if (PageActive(page))
406 ClearPageActive(page);
407 __lru_cache_add(page);
410 void lru_cache_add_file(struct page *page)
412 if (PageActive(page))
413 ClearPageActive(page);
414 __lru_cache_add(page);
416 EXPORT_SYMBOL(lru_cache_add_file);
419 * lru_cache_add - add a page to a page list
420 * @page: the page to be added to the LRU.
422 * Queue the page for addition to the LRU via pagevec. The decision on whether
423 * to add the page to the [in]active [file|anon] list is deferred until the
424 * pagevec is drained. This gives a chance for the caller of lru_cache_add()
425 * have the page added to the active list using mark_page_accessed().
427 void lru_cache_add(struct page *page)
429 VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page);
430 VM_BUG_ON_PAGE(PageLRU(page), page);
431 __lru_cache_add(page);
435 * add_page_to_unevictable_list - add a page to the unevictable list
436 * @page: the page to be added to the unevictable list
438 * Add page directly to its zone's unevictable list. To avoid races with
439 * tasks that might be making the page evictable, through eg. munlock,
440 * munmap or exit, while it's not on the lru, we want to add the page
441 * while it's locked or otherwise "invisible" to other tasks. This is
442 * difficult to do when using the pagevec cache, so bypass that.
444 void add_page_to_unevictable_list(struct page *page)
446 struct zone *zone = page_zone(page);
447 struct lruvec *lruvec;
449 spin_lock_irq(&zone->lru_lock);
450 lruvec = mem_cgroup_page_lruvec(page, zone);
451 ClearPageActive(page);
452 SetPageUnevictable(page);
454 add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE);
455 spin_unlock_irq(&zone->lru_lock);
459 * lru_cache_add_active_or_unevictable
460 * @page: the page to be added to LRU
461 * @vma: vma in which page is mapped for determining reclaimability
463 * Place @page on the active or unevictable LRU list, depending on its
464 * evictability. Note that if the page is not evictable, it goes
465 * directly back onto it's zone's unevictable list, it does NOT use a
468 void lru_cache_add_active_or_unevictable(struct page *page,
469 struct vm_area_struct *vma)
471 VM_BUG_ON_PAGE(PageLRU(page), page);
473 if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED)) {
479 if (!TestSetPageMlocked(page)) {
481 * We use the irq-unsafe __mod_zone_page_stat because this
482 * counter is not modified from interrupt context, and the pte
483 * lock is held(spinlock), which implies preemption disabled.
485 __mod_zone_page_state(page_zone(page), NR_MLOCK,
486 hpage_nr_pages(page));
487 count_vm_event(UNEVICTABLE_PGMLOCKED);
489 add_page_to_unevictable_list(page);
493 * If the page can not be invalidated, it is moved to the
494 * inactive list to speed up its reclaim. It is moved to the
495 * head of the list, rather than the tail, to give the flusher
496 * threads some time to write it out, as this is much more
497 * effective than the single-page writeout from reclaim.
499 * If the page isn't page_mapped and dirty/writeback, the page
500 * could reclaim asap using PG_reclaim.
502 * 1. active, mapped page -> none
503 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
504 * 3. inactive, mapped page -> none
505 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
506 * 5. inactive, clean -> inactive, tail
509 * In 4, why it moves inactive's head, the VM expects the page would
510 * be write it out by flusher threads as this is much more effective
511 * than the single-page writeout from reclaim.
513 static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec,
522 if (PageUnevictable(page))
525 /* Some processes are using the page */
526 if (page_mapped(page))
529 active = PageActive(page);
530 file = page_is_file_cache(page);
531 lru = page_lru_base_type(page);
533 del_page_from_lru_list(page, lruvec, lru + active);
534 ClearPageActive(page);
535 ClearPageReferenced(page);
536 add_page_to_lru_list(page, lruvec, lru);
538 if (PageWriteback(page) || PageDirty(page)) {
540 * PG_reclaim could be raced with end_page_writeback
541 * It can make readahead confusing. But race window
542 * is _really_ small and it's non-critical problem.
544 SetPageReclaim(page);
547 * The page's writeback ends up during pagevec
548 * We moves tha page into tail of inactive.
550 list_move_tail(&page->lru, &lruvec->lists[lru]);
551 __count_vm_event(PGROTATED);
555 __count_vm_event(PGDEACTIVATE);
556 update_page_reclaim_stat(lruvec, file, 0);
560 static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec,
563 if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) {
564 int file = page_is_file_cache(page);
565 int lru = page_lru_base_type(page);
567 del_page_from_lru_list(page, lruvec, lru + LRU_ACTIVE);
568 ClearPageActive(page);
569 ClearPageReferenced(page);
570 add_page_to_lru_list(page, lruvec, lru);
572 __count_vm_event(PGDEACTIVATE);
573 update_page_reclaim_stat(lruvec, file, 0);
578 * Drain pages out of the cpu's pagevecs.
579 * Either "cpu" is the current CPU, and preemption has already been
580 * disabled; or "cpu" is being hot-unplugged, and is already dead.
582 void lru_add_drain_cpu(int cpu)
584 struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu);
586 if (pagevec_count(pvec))
587 __pagevec_lru_add(pvec);
589 pvec = &per_cpu(lru_rotate_pvecs, cpu);
590 if (pagevec_count(pvec)) {
593 /* No harm done if a racing interrupt already did this */
594 local_irq_save(flags);
595 pagevec_move_tail(pvec);
596 local_irq_restore(flags);
599 pvec = &per_cpu(lru_deactivate_file_pvecs, cpu);
600 if (pagevec_count(pvec))
601 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
603 pvec = &per_cpu(lru_deactivate_pvecs, cpu);
604 if (pagevec_count(pvec))
605 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
607 activate_page_drain(cpu);
611 * deactivate_file_page - forcefully deactivate a file page
612 * @page: page to deactivate
614 * This function hints the VM that @page is a good reclaim candidate,
615 * for example if its invalidation fails due to the page being dirty
616 * or under writeback.
618 void deactivate_file_page(struct page *page)
621 * In a workload with many unevictable page such as mprotect,
622 * unevictable page deactivation for accelerating reclaim is pointless.
624 if (PageUnevictable(page))
627 if (likely(get_page_unless_zero(page))) {
628 struct pagevec *pvec = &get_cpu_var(lru_deactivate_file_pvecs);
630 if (!pagevec_add(pvec, page) || PageCompound(page))
631 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
632 put_cpu_var(lru_deactivate_file_pvecs);
637 * deactivate_page - deactivate a page
638 * @page: page to deactivate
640 * deactivate_page() moves @page to the inactive list if @page was on the active
641 * list and was not an unevictable page. This is done to accelerate the reclaim
644 void deactivate_page(struct page *page)
646 if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) {
647 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
650 if (!pagevec_add(pvec, page) || PageCompound(page))
651 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
652 put_cpu_var(lru_deactivate_pvecs);
656 void lru_add_drain(void)
658 lru_add_drain_cpu(get_cpu());
662 static void lru_add_drain_per_cpu(struct work_struct *dummy)
667 static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
670 * lru_add_drain_wq is used to do lru_add_drain_all() from a WQ_MEM_RECLAIM
671 * workqueue, aiding in getting memory freed.
673 static struct workqueue_struct *lru_add_drain_wq;
675 static int __init lru_init(void)
677 lru_add_drain_wq = alloc_workqueue("lru-add-drain", WQ_MEM_RECLAIM, 0);
679 if (WARN(!lru_add_drain_wq,
680 "Failed to create workqueue lru_add_drain_wq"))
685 early_initcall(lru_init);
687 void lru_add_drain_all(void)
689 static DEFINE_MUTEX(lock);
690 static struct cpumask has_work;
695 cpumask_clear(&has_work);
697 for_each_online_cpu(cpu) {
698 struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
700 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
701 pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
702 pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) ||
703 pagevec_count(&per_cpu(lru_deactivate_pvecs, cpu)) ||
704 need_activate_page_drain(cpu)) {
705 INIT_WORK(work, lru_add_drain_per_cpu);
706 queue_work_on(cpu, lru_add_drain_wq, work);
707 cpumask_set_cpu(cpu, &has_work);
711 for_each_cpu(cpu, &has_work)
712 flush_work(&per_cpu(lru_add_drain_work, cpu));
719 * release_pages - batched put_page()
720 * @pages: array of pages to release
721 * @nr: number of pages
722 * @cold: whether the pages are cache cold
724 * Decrement the reference count on all the pages in @pages. If it
725 * fell to zero, remove the page from the LRU and free it.
727 void release_pages(struct page **pages, int nr, bool cold)
730 LIST_HEAD(pages_to_free);
731 struct zone *zone = NULL;
732 struct lruvec *lruvec;
733 unsigned long uninitialized_var(flags);
734 unsigned int uninitialized_var(lock_batch);
736 for (i = 0; i < nr; i++) {
737 struct page *page = pages[i];
740 * Make sure the IRQ-safe lock-holding time does not get
741 * excessive with a continuous string of pages from the
742 * same zone. The lock is held only if zone != NULL.
744 if (zone && ++lock_batch == SWAP_CLUSTER_MAX) {
745 spin_unlock_irqrestore(&zone->lru_lock, flags);
749 if (is_huge_zero_page(page)) {
750 put_huge_zero_page();
754 page = compound_head(page);
755 if (!put_page_testzero(page))
758 if (PageCompound(page)) {
760 spin_unlock_irqrestore(&zone->lru_lock, flags);
763 __put_compound_page(page);
768 struct zone *pagezone = page_zone(page);
770 if (pagezone != zone) {
772 spin_unlock_irqrestore(&zone->lru_lock,
776 spin_lock_irqsave(&zone->lru_lock, flags);
779 lruvec = mem_cgroup_page_lruvec(page, zone);
780 VM_BUG_ON_PAGE(!PageLRU(page), page);
781 __ClearPageLRU(page);
782 del_page_from_lru_list(page, lruvec, page_off_lru(page));
785 /* Clear Active bit in case of parallel mark_page_accessed */
786 __ClearPageActive(page);
788 list_add(&page->lru, &pages_to_free);
791 spin_unlock_irqrestore(&zone->lru_lock, flags);
793 mem_cgroup_uncharge_list(&pages_to_free);
794 free_hot_cold_page_list(&pages_to_free, cold);
796 EXPORT_SYMBOL(release_pages);
799 * The pages which we're about to release may be in the deferred lru-addition
800 * queues. That would prevent them from really being freed right now. That's
801 * OK from a correctness point of view but is inefficient - those pages may be
802 * cache-warm and we want to give them back to the page allocator ASAP.
804 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
805 * and __pagevec_lru_add_active() call release_pages() directly to avoid
808 void __pagevec_release(struct pagevec *pvec)
811 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
812 pagevec_reinit(pvec);
814 EXPORT_SYMBOL(__pagevec_release);
816 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
817 /* used by __split_huge_page_refcount() */
818 void lru_add_page_tail(struct page *page, struct page *page_tail,
819 struct lruvec *lruvec, struct list_head *list)
823 VM_BUG_ON_PAGE(!PageHead(page), page);
824 VM_BUG_ON_PAGE(PageCompound(page_tail), page);
825 VM_BUG_ON_PAGE(PageLRU(page_tail), page);
826 VM_BUG_ON(NR_CPUS != 1 &&
827 !spin_is_locked(&lruvec_zone(lruvec)->lru_lock));
830 SetPageLRU(page_tail);
832 if (likely(PageLRU(page)))
833 list_add_tail(&page_tail->lru, &page->lru);
835 /* page reclaim is reclaiming a huge page */
837 list_add_tail(&page_tail->lru, list);
839 struct list_head *list_head;
841 * Head page has not yet been counted, as an hpage,
842 * so we must account for each subpage individually.
844 * Use the standard add function to put page_tail on the list,
845 * but then correct its position so they all end up in order.
847 add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail));
848 list_head = page_tail->lru.prev;
849 list_move_tail(&page_tail->lru, list_head);
852 if (!PageUnevictable(page))
853 update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
855 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
857 static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
860 int file = page_is_file_cache(page);
861 int active = PageActive(page);
862 enum lru_list lru = page_lru(page);
864 VM_BUG_ON_PAGE(PageLRU(page), page);
867 add_page_to_lru_list(page, lruvec, lru);
868 update_page_reclaim_stat(lruvec, file, active);
869 trace_mm_lru_insertion(page, lru);
873 * Add the passed pages to the LRU, then drop the caller's refcount
874 * on them. Reinitialises the caller's pagevec.
876 void __pagevec_lru_add(struct pagevec *pvec)
878 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
880 EXPORT_SYMBOL(__pagevec_lru_add);
883 * pagevec_lookup_entries - gang pagecache lookup
884 * @pvec: Where the resulting entries are placed
885 * @mapping: The address_space to search
886 * @start: The starting entry index
887 * @nr_entries: The maximum number of entries
888 * @indices: The cache indices corresponding to the entries in @pvec
890 * pagevec_lookup_entries() will search for and return a group of up
891 * to @nr_entries pages and shadow entries in the mapping. All
892 * entries are placed in @pvec. pagevec_lookup_entries() takes a
893 * reference against actual pages in @pvec.
895 * The search returns a group of mapping-contiguous entries with
896 * ascending indexes. There may be holes in the indices due to
897 * not-present entries.
899 * pagevec_lookup_entries() returns the number of entries which were
902 unsigned pagevec_lookup_entries(struct pagevec *pvec,
903 struct address_space *mapping,
904 pgoff_t start, unsigned nr_pages,
907 pvec->nr = find_get_entries(mapping, start, nr_pages,
908 pvec->pages, indices);
909 return pagevec_count(pvec);
913 * pagevec_remove_exceptionals - pagevec exceptionals pruning
914 * @pvec: The pagevec to prune
916 * pagevec_lookup_entries() fills both pages and exceptional radix
917 * tree entries into the pagevec. This function prunes all
918 * exceptionals from @pvec without leaving holes, so that it can be
919 * passed on to page-only pagevec operations.
921 void pagevec_remove_exceptionals(struct pagevec *pvec)
925 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
926 struct page *page = pvec->pages[i];
927 if (!radix_tree_exceptional_entry(page))
928 pvec->pages[j++] = page;
934 * pagevec_lookup - gang pagecache lookup
935 * @pvec: Where the resulting pages are placed
936 * @mapping: The address_space to search
937 * @start: The starting page index
938 * @nr_pages: The maximum number of pages
940 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
941 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
942 * reference against the pages in @pvec.
944 * The search returns a group of mapping-contiguous pages with ascending
945 * indexes. There may be holes in the indices due to not-present pages.
947 * pagevec_lookup() returns the number of pages which were found.
949 unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
950 pgoff_t start, unsigned nr_pages)
952 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
953 return pagevec_count(pvec);
955 EXPORT_SYMBOL(pagevec_lookup);
957 unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
958 pgoff_t *index, int tag, unsigned nr_pages)
960 pvec->nr = find_get_pages_tag(mapping, index, tag,
961 nr_pages, pvec->pages);
962 return pagevec_count(pvec);
964 EXPORT_SYMBOL(pagevec_lookup_tag);
967 * Perform any setup for the swap system
969 void __init swap_setup(void)
971 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
975 for (i = 0; i < MAX_SWAPFILES; i++)
976 spin_lock_init(&swapper_spaces[i].tree_lock);
979 /* Use a smaller cluster for small-memory machines */
985 * Right now other parts of the system means that we
986 * _really_ don't want to cluster much more