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(zone), flags);
66 lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat);
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(zone), 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 pglist_data *pgdat = 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 pglist_data *pagepgdat = page_pgdat(page);
190 if (pagepgdat != pgdat) {
192 spin_unlock_irqrestore(&pgdat->lru_lock, flags);
194 spin_lock_irqsave(&pgdat->lru_lock, flags);
197 lruvec = mem_cgroup_page_lruvec(page, pgdat);
198 (*move_fn)(page, lruvec, arg);
201 spin_unlock_irqrestore(&pgdat->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 page = compound_head(page);
296 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
297 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
300 if (!pagevec_add(pvec, page) || PageCompound(page))
301 pagevec_lru_move_fn(pvec, __activate_page, NULL);
302 put_cpu_var(activate_page_pvecs);
307 static inline void activate_page_drain(int cpu)
311 static bool need_activate_page_drain(int cpu)
316 void activate_page(struct page *page)
318 struct zone *zone = page_zone(page);
320 page = compound_head(page);
321 spin_lock_irq(zone_lru_lock(zone));
322 __activate_page(page, mem_cgroup_page_lruvec(page, zone->zone_pgdat), NULL);
323 spin_unlock_irq(zone_lru_lock(zone));
327 static void __lru_cache_activate_page(struct page *page)
329 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
333 * Search backwards on the optimistic assumption that the page being
334 * activated has just been added to this pagevec. Note that only
335 * the local pagevec is examined as a !PageLRU page could be in the
336 * process of being released, reclaimed, migrated or on a remote
337 * pagevec that is currently being drained. Furthermore, marking
338 * a remote pagevec's page PageActive potentially hits a race where
339 * a page is marked PageActive just after it is added to the inactive
340 * list causing accounting errors and BUG_ON checks to trigger.
342 for (i = pagevec_count(pvec) - 1; i >= 0; i--) {
343 struct page *pagevec_page = pvec->pages[i];
345 if (pagevec_page == page) {
351 put_cpu_var(lru_add_pvec);
355 * Mark a page as having seen activity.
357 * inactive,unreferenced -> inactive,referenced
358 * inactive,referenced -> active,unreferenced
359 * active,unreferenced -> active,referenced
361 * When a newly allocated page is not yet visible, so safe for non-atomic ops,
362 * __SetPageReferenced(page) may be substituted for mark_page_accessed(page).
364 void mark_page_accessed(struct page *page)
366 page = compound_head(page);
367 if (!PageActive(page) && !PageUnevictable(page) &&
368 PageReferenced(page)) {
371 * If the page is on the LRU, queue it for activation via
372 * activate_page_pvecs. Otherwise, assume the page is on a
373 * pagevec, mark it active and it'll be moved to the active
374 * LRU on the next drain.
379 __lru_cache_activate_page(page);
380 ClearPageReferenced(page);
381 if (page_is_file_cache(page))
382 workingset_activation(page);
383 } else if (!PageReferenced(page)) {
384 SetPageReferenced(page);
386 if (page_is_idle(page))
387 clear_page_idle(page);
389 EXPORT_SYMBOL(mark_page_accessed);
391 static void __lru_cache_add(struct page *page)
393 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
396 if (!pagevec_add(pvec, page) || PageCompound(page))
397 __pagevec_lru_add(pvec);
398 put_cpu_var(lru_add_pvec);
402 * lru_cache_add: add a page to the page lists
403 * @page: the page to add
405 void lru_cache_add_anon(struct page *page)
407 if (PageActive(page))
408 ClearPageActive(page);
409 __lru_cache_add(page);
412 void lru_cache_add_file(struct page *page)
414 if (PageActive(page))
415 ClearPageActive(page);
416 __lru_cache_add(page);
418 EXPORT_SYMBOL(lru_cache_add_file);
421 * lru_cache_add - add a page to a page list
422 * @page: the page to be added to the LRU.
424 * Queue the page for addition to the LRU via pagevec. The decision on whether
425 * to add the page to the [in]active [file|anon] list is deferred until the
426 * pagevec is drained. This gives a chance for the caller of lru_cache_add()
427 * have the page added to the active list using mark_page_accessed().
429 void lru_cache_add(struct page *page)
431 VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page);
432 VM_BUG_ON_PAGE(PageLRU(page), page);
433 __lru_cache_add(page);
437 * add_page_to_unevictable_list - add a page to the unevictable list
438 * @page: the page to be added to the unevictable list
440 * Add page directly to its zone's unevictable list. To avoid races with
441 * tasks that might be making the page evictable, through eg. munlock,
442 * munmap or exit, while it's not on the lru, we want to add the page
443 * while it's locked or otherwise "invisible" to other tasks. This is
444 * difficult to do when using the pagevec cache, so bypass that.
446 void add_page_to_unevictable_list(struct page *page)
448 struct pglist_data *pgdat = page_pgdat(page);
449 struct lruvec *lruvec;
451 spin_lock_irq(&pgdat->lru_lock);
452 lruvec = mem_cgroup_page_lruvec(page, pgdat);
453 ClearPageActive(page);
454 SetPageUnevictable(page);
456 add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE);
457 spin_unlock_irq(&pgdat->lru_lock);
461 * lru_cache_add_active_or_unevictable
462 * @page: the page to be added to LRU
463 * @vma: vma in which page is mapped for determining reclaimability
465 * Place @page on the active or unevictable LRU list, depending on its
466 * evictability. Note that if the page is not evictable, it goes
467 * directly back onto it's zone's unevictable list, it does NOT use a
470 void lru_cache_add_active_or_unevictable(struct page *page,
471 struct vm_area_struct *vma)
473 VM_BUG_ON_PAGE(PageLRU(page), page);
475 if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED)) {
481 if (!TestSetPageMlocked(page)) {
483 * We use the irq-unsafe __mod_zone_page_stat because this
484 * counter is not modified from interrupt context, and the pte
485 * lock is held(spinlock), which implies preemption disabled.
487 __mod_zone_page_state(page_zone(page), NR_MLOCK,
488 hpage_nr_pages(page));
489 count_vm_event(UNEVICTABLE_PGMLOCKED);
491 add_page_to_unevictable_list(page);
495 * If the page can not be invalidated, it is moved to the
496 * inactive list to speed up its reclaim. It is moved to the
497 * head of the list, rather than the tail, to give the flusher
498 * threads some time to write it out, as this is much more
499 * effective than the single-page writeout from reclaim.
501 * If the page isn't page_mapped and dirty/writeback, the page
502 * could reclaim asap using PG_reclaim.
504 * 1. active, mapped page -> none
505 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
506 * 3. inactive, mapped page -> none
507 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
508 * 5. inactive, clean -> inactive, tail
511 * In 4, why it moves inactive's head, the VM expects the page would
512 * be write it out by flusher threads as this is much more effective
513 * than the single-page writeout from reclaim.
515 static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec,
524 if (PageUnevictable(page))
527 /* Some processes are using the page */
528 if (page_mapped(page))
531 active = PageActive(page);
532 file = page_is_file_cache(page);
533 lru = page_lru_base_type(page);
535 del_page_from_lru_list(page, lruvec, lru + active);
536 ClearPageActive(page);
537 ClearPageReferenced(page);
538 add_page_to_lru_list(page, lruvec, lru);
540 if (PageWriteback(page) || PageDirty(page)) {
542 * PG_reclaim could be raced with end_page_writeback
543 * It can make readahead confusing. But race window
544 * is _really_ small and it's non-critical problem.
546 SetPageReclaim(page);
549 * The page's writeback ends up during pagevec
550 * We moves tha page into tail of inactive.
552 list_move_tail(&page->lru, &lruvec->lists[lru]);
553 __count_vm_event(PGROTATED);
557 __count_vm_event(PGDEACTIVATE);
558 update_page_reclaim_stat(lruvec, file, 0);
562 static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec,
565 if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) {
566 int file = page_is_file_cache(page);
567 int lru = page_lru_base_type(page);
569 del_page_from_lru_list(page, lruvec, lru + LRU_ACTIVE);
570 ClearPageActive(page);
571 ClearPageReferenced(page);
572 add_page_to_lru_list(page, lruvec, lru);
574 __count_vm_event(PGDEACTIVATE);
575 update_page_reclaim_stat(lruvec, file, 0);
580 * Drain pages out of the cpu's pagevecs.
581 * Either "cpu" is the current CPU, and preemption has already been
582 * disabled; or "cpu" is being hot-unplugged, and is already dead.
584 void lru_add_drain_cpu(int cpu)
586 struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu);
588 if (pagevec_count(pvec))
589 __pagevec_lru_add(pvec);
591 pvec = &per_cpu(lru_rotate_pvecs, cpu);
592 if (pagevec_count(pvec)) {
595 /* No harm done if a racing interrupt already did this */
596 local_irq_save(flags);
597 pagevec_move_tail(pvec);
598 local_irq_restore(flags);
601 pvec = &per_cpu(lru_deactivate_file_pvecs, cpu);
602 if (pagevec_count(pvec))
603 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
605 pvec = &per_cpu(lru_deactivate_pvecs, cpu);
606 if (pagevec_count(pvec))
607 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
609 activate_page_drain(cpu);
613 * deactivate_file_page - forcefully deactivate a file page
614 * @page: page to deactivate
616 * This function hints the VM that @page is a good reclaim candidate,
617 * for example if its invalidation fails due to the page being dirty
618 * or under writeback.
620 void deactivate_file_page(struct page *page)
623 * In a workload with many unevictable page such as mprotect,
624 * unevictable page deactivation for accelerating reclaim is pointless.
626 if (PageUnevictable(page))
629 if (likely(get_page_unless_zero(page))) {
630 struct pagevec *pvec = &get_cpu_var(lru_deactivate_file_pvecs);
632 if (!pagevec_add(pvec, page) || PageCompound(page))
633 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
634 put_cpu_var(lru_deactivate_file_pvecs);
639 * deactivate_page - deactivate a page
640 * @page: page to deactivate
642 * deactivate_page() moves @page to the inactive list if @page was on the active
643 * list and was not an unevictable page. This is done to accelerate the reclaim
646 void deactivate_page(struct page *page)
648 if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) {
649 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
652 if (!pagevec_add(pvec, page) || PageCompound(page))
653 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
654 put_cpu_var(lru_deactivate_pvecs);
658 void lru_add_drain(void)
660 lru_add_drain_cpu(get_cpu());
664 static void lru_add_drain_per_cpu(struct work_struct *dummy)
669 static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
672 * lru_add_drain_wq is used to do lru_add_drain_all() from a WQ_MEM_RECLAIM
673 * workqueue, aiding in getting memory freed.
675 static struct workqueue_struct *lru_add_drain_wq;
677 static int __init lru_init(void)
679 lru_add_drain_wq = alloc_workqueue("lru-add-drain", WQ_MEM_RECLAIM, 0);
681 if (WARN(!lru_add_drain_wq,
682 "Failed to create workqueue lru_add_drain_wq"))
687 early_initcall(lru_init);
689 void lru_add_drain_all(void)
691 static DEFINE_MUTEX(lock);
692 static struct cpumask has_work;
697 cpumask_clear(&has_work);
699 for_each_online_cpu(cpu) {
700 struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
702 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
703 pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
704 pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) ||
705 pagevec_count(&per_cpu(lru_deactivate_pvecs, cpu)) ||
706 need_activate_page_drain(cpu)) {
707 INIT_WORK(work, lru_add_drain_per_cpu);
708 queue_work_on(cpu, lru_add_drain_wq, work);
709 cpumask_set_cpu(cpu, &has_work);
713 for_each_cpu(cpu, &has_work)
714 flush_work(&per_cpu(lru_add_drain_work, cpu));
721 * release_pages - batched put_page()
722 * @pages: array of pages to release
723 * @nr: number of pages
724 * @cold: whether the pages are cache cold
726 * Decrement the reference count on all the pages in @pages. If it
727 * fell to zero, remove the page from the LRU and free it.
729 void release_pages(struct page **pages, int nr, bool cold)
732 LIST_HEAD(pages_to_free);
733 struct pglist_data *locked_pgdat = NULL;
734 struct lruvec *lruvec;
735 unsigned long uninitialized_var(flags);
736 unsigned int uninitialized_var(lock_batch);
738 for (i = 0; i < nr; i++) {
739 struct page *page = pages[i];
742 * Make sure the IRQ-safe lock-holding time does not get
743 * excessive with a continuous string of pages from the
744 * same pgdat. The lock is held only if pgdat != NULL.
746 if (locked_pgdat && ++lock_batch == SWAP_CLUSTER_MAX) {
747 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
751 if (is_huge_zero_page(page)) {
752 put_huge_zero_page();
756 page = compound_head(page);
757 if (!put_page_testzero(page))
760 if (PageCompound(page)) {
762 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
765 __put_compound_page(page);
770 struct pglist_data *pgdat = page_pgdat(page);
772 if (pgdat != locked_pgdat) {
774 spin_unlock_irqrestore(&locked_pgdat->lru_lock,
777 locked_pgdat = pgdat;
778 spin_lock_irqsave(&locked_pgdat->lru_lock, flags);
781 lruvec = mem_cgroup_page_lruvec(page, locked_pgdat);
782 VM_BUG_ON_PAGE(!PageLRU(page), page);
783 __ClearPageLRU(page);
784 del_page_from_lru_list(page, lruvec, page_off_lru(page));
787 /* Clear Active bit in case of parallel mark_page_accessed */
788 __ClearPageActive(page);
790 list_add(&page->lru, &pages_to_free);
793 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
795 mem_cgroup_uncharge_list(&pages_to_free);
796 free_hot_cold_page_list(&pages_to_free, cold);
798 EXPORT_SYMBOL(release_pages);
801 * The pages which we're about to release may be in the deferred lru-addition
802 * queues. That would prevent them from really being freed right now. That's
803 * OK from a correctness point of view but is inefficient - those pages may be
804 * cache-warm and we want to give them back to the page allocator ASAP.
806 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
807 * and __pagevec_lru_add_active() call release_pages() directly to avoid
810 void __pagevec_release(struct pagevec *pvec)
813 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
814 pagevec_reinit(pvec);
816 EXPORT_SYMBOL(__pagevec_release);
818 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
819 /* used by __split_huge_page_refcount() */
820 void lru_add_page_tail(struct page *page, struct page *page_tail,
821 struct lruvec *lruvec, struct list_head *list)
825 VM_BUG_ON_PAGE(!PageHead(page), page);
826 VM_BUG_ON_PAGE(PageCompound(page_tail), page);
827 VM_BUG_ON_PAGE(PageLRU(page_tail), page);
828 VM_BUG_ON(NR_CPUS != 1 &&
829 !spin_is_locked(&lruvec_pgdat(lruvec)->lru_lock));
832 SetPageLRU(page_tail);
834 if (likely(PageLRU(page)))
835 list_add_tail(&page_tail->lru, &page->lru);
837 /* page reclaim is reclaiming a huge page */
839 list_add_tail(&page_tail->lru, list);
841 struct list_head *list_head;
843 * Head page has not yet been counted, as an hpage,
844 * so we must account for each subpage individually.
846 * Use the standard add function to put page_tail on the list,
847 * but then correct its position so they all end up in order.
849 add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail));
850 list_head = page_tail->lru.prev;
851 list_move_tail(&page_tail->lru, list_head);
854 if (!PageUnevictable(page))
855 update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
857 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
859 static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
862 int file = page_is_file_cache(page);
863 int active = PageActive(page);
864 enum lru_list lru = page_lru(page);
866 VM_BUG_ON_PAGE(PageLRU(page), page);
869 add_page_to_lru_list(page, lruvec, lru);
870 update_page_reclaim_stat(lruvec, file, active);
871 trace_mm_lru_insertion(page, lru);
875 * Add the passed pages to the LRU, then drop the caller's refcount
876 * on them. Reinitialises the caller's pagevec.
878 void __pagevec_lru_add(struct pagevec *pvec)
880 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
882 EXPORT_SYMBOL(__pagevec_lru_add);
885 * pagevec_lookup_entries - gang pagecache lookup
886 * @pvec: Where the resulting entries are placed
887 * @mapping: The address_space to search
888 * @start: The starting entry index
889 * @nr_entries: The maximum number of entries
890 * @indices: The cache indices corresponding to the entries in @pvec
892 * pagevec_lookup_entries() will search for and return a group of up
893 * to @nr_entries pages and shadow entries in the mapping. All
894 * entries are placed in @pvec. pagevec_lookup_entries() takes a
895 * reference against actual pages in @pvec.
897 * The search returns a group of mapping-contiguous entries with
898 * ascending indexes. There may be holes in the indices due to
899 * not-present entries.
901 * pagevec_lookup_entries() returns the number of entries which were
904 unsigned pagevec_lookup_entries(struct pagevec *pvec,
905 struct address_space *mapping,
906 pgoff_t start, unsigned nr_pages,
909 pvec->nr = find_get_entries(mapping, start, nr_pages,
910 pvec->pages, indices);
911 return pagevec_count(pvec);
915 * pagevec_remove_exceptionals - pagevec exceptionals pruning
916 * @pvec: The pagevec to prune
918 * pagevec_lookup_entries() fills both pages and exceptional radix
919 * tree entries into the pagevec. This function prunes all
920 * exceptionals from @pvec without leaving holes, so that it can be
921 * passed on to page-only pagevec operations.
923 void pagevec_remove_exceptionals(struct pagevec *pvec)
927 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
928 struct page *page = pvec->pages[i];
929 if (!radix_tree_exceptional_entry(page))
930 pvec->pages[j++] = page;
936 * pagevec_lookup - gang pagecache lookup
937 * @pvec: Where the resulting pages are placed
938 * @mapping: The address_space to search
939 * @start: The starting page index
940 * @nr_pages: The maximum number of pages
942 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
943 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
944 * reference against the pages in @pvec.
946 * The search returns a group of mapping-contiguous pages with ascending
947 * indexes. There may be holes in the indices due to not-present pages.
949 * pagevec_lookup() returns the number of pages which were found.
951 unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
952 pgoff_t start, unsigned nr_pages)
954 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
955 return pagevec_count(pvec);
957 EXPORT_SYMBOL(pagevec_lookup);
959 unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
960 pgoff_t *index, int tag, unsigned nr_pages)
962 pvec->nr = find_get_pages_tag(mapping, index, tag,
963 nr_pages, pvec->pages);
964 return pagevec_count(pvec);
966 EXPORT_SYMBOL(pagevec_lookup_tag);
969 * Perform any setup for the swap system
971 void __init swap_setup(void)
973 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
977 for (i = 0; i < MAX_SWAPFILES; i++)
978 spin_lock_init(&swapper_spaces[i].tree_lock);
981 /* Use a smaller cluster for small-memory machines */
987 * Right now other parts of the system means that we
988 * _really_ don't want to cluster much more