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_lazyfree_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 __ClearPageWaiters(page);
73 mem_cgroup_uncharge(page);
76 static void __put_single_page(struct page *page)
78 __page_cache_release(page);
79 free_hot_cold_page(page, false);
82 static void __put_compound_page(struct page *page)
84 compound_page_dtor *dtor;
87 * __page_cache_release() is supposed to be called for thp, not for
88 * hugetlb. This is because hugetlb page does never have PageLRU set
89 * (it's never listed to any LRU lists) and no memcg routines should
90 * be called for hugetlb (it has a separate hugetlb_cgroup.)
93 __page_cache_release(page);
94 dtor = get_compound_page_dtor(page);
98 void __put_page(struct page *page)
100 if (is_zone_device_page(page)) {
101 put_dev_pagemap(page->pgmap);
104 * The page belongs to the device that created pgmap. Do
105 * not return it to page allocator.
110 if (unlikely(PageCompound(page)))
111 __put_compound_page(page);
113 __put_single_page(page);
115 EXPORT_SYMBOL(__put_page);
118 * put_pages_list() - release a list of pages
119 * @pages: list of pages threaded on page->lru
121 * Release a list of pages which are strung together on page.lru. Currently
122 * used by read_cache_pages() and related error recovery code.
124 void put_pages_list(struct list_head *pages)
126 while (!list_empty(pages)) {
129 victim = list_entry(pages->prev, struct page, lru);
130 list_del(&victim->lru);
134 EXPORT_SYMBOL(put_pages_list);
137 * get_kernel_pages() - pin kernel pages in memory
138 * @kiov: An array of struct kvec structures
139 * @nr_segs: number of segments to pin
140 * @write: pinning for read/write, currently ignored
141 * @pages: array that receives pointers to the pages pinned.
142 * Should be at least nr_segs long.
144 * Returns number of pages pinned. This may be fewer than the number
145 * requested. If nr_pages is 0 or negative, returns 0. If no pages
146 * were pinned, returns -errno. Each page returned must be released
147 * with a put_page() call when it is finished with.
149 int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write,
154 for (seg = 0; seg < nr_segs; seg++) {
155 if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE))
158 pages[seg] = kmap_to_page(kiov[seg].iov_base);
159 get_page(pages[seg]);
164 EXPORT_SYMBOL_GPL(get_kernel_pages);
167 * get_kernel_page() - pin a kernel page in memory
168 * @start: starting kernel address
169 * @write: pinning for read/write, currently ignored
170 * @pages: array that receives pointer to the page pinned.
171 * Must be at least nr_segs long.
173 * Returns 1 if page is pinned. If the page was not pinned, returns
174 * -errno. The page returned must be released with a put_page() call
175 * when it is finished with.
177 int get_kernel_page(unsigned long start, int write, struct page **pages)
179 const struct kvec kiov = {
180 .iov_base = (void *)start,
184 return get_kernel_pages(&kiov, 1, write, pages);
186 EXPORT_SYMBOL_GPL(get_kernel_page);
188 static void pagevec_lru_move_fn(struct pagevec *pvec,
189 void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg),
193 struct pglist_data *pgdat = NULL;
194 struct lruvec *lruvec;
195 unsigned long flags = 0;
197 for (i = 0; i < pagevec_count(pvec); i++) {
198 struct page *page = pvec->pages[i];
199 struct pglist_data *pagepgdat = page_pgdat(page);
201 if (pagepgdat != pgdat) {
203 spin_unlock_irqrestore(&pgdat->lru_lock, flags);
205 spin_lock_irqsave(&pgdat->lru_lock, flags);
208 lruvec = mem_cgroup_page_lruvec(page, pgdat);
209 (*move_fn)(page, lruvec, arg);
212 spin_unlock_irqrestore(&pgdat->lru_lock, flags);
213 release_pages(pvec->pages, pvec->nr, pvec->cold);
214 pagevec_reinit(pvec);
217 static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec,
222 if (PageLRU(page) && !PageUnevictable(page)) {
223 del_page_from_lru_list(page, lruvec, page_lru(page));
224 ClearPageActive(page);
225 add_page_to_lru_list_tail(page, lruvec, page_lru(page));
231 * pagevec_move_tail() must be called with IRQ disabled.
232 * Otherwise this may cause nasty races.
234 static void pagevec_move_tail(struct pagevec *pvec)
238 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
239 __count_vm_events(PGROTATED, pgmoved);
243 * Writeback is about to end against a page which has been marked for immediate
244 * reclaim. If it still appears to be reclaimable, move it to the tail of the
247 void rotate_reclaimable_page(struct page *page)
249 if (!PageLocked(page) && !PageDirty(page) &&
250 !PageUnevictable(page) && PageLRU(page)) {
251 struct pagevec *pvec;
255 local_irq_save(flags);
256 pvec = this_cpu_ptr(&lru_rotate_pvecs);
257 if (!pagevec_add(pvec, page) || PageCompound(page))
258 pagevec_move_tail(pvec);
259 local_irq_restore(flags);
263 static void update_page_reclaim_stat(struct lruvec *lruvec,
264 int file, int rotated)
266 struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
268 reclaim_stat->recent_scanned[file]++;
270 reclaim_stat->recent_rotated[file]++;
273 static void __activate_page(struct page *page, struct lruvec *lruvec,
276 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
277 int file = page_is_file_cache(page);
278 int lru = page_lru_base_type(page);
280 del_page_from_lru_list(page, lruvec, lru);
283 add_page_to_lru_list(page, lruvec, lru);
284 trace_mm_lru_activate(page);
286 __count_vm_event(PGACTIVATE);
287 update_page_reclaim_stat(lruvec, file, 1);
292 static void activate_page_drain(int cpu)
294 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
296 if (pagevec_count(pvec))
297 pagevec_lru_move_fn(pvec, __activate_page, NULL);
300 static bool need_activate_page_drain(int cpu)
302 return pagevec_count(&per_cpu(activate_page_pvecs, cpu)) != 0;
305 void activate_page(struct page *page)
307 page = compound_head(page);
308 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
309 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
312 if (!pagevec_add(pvec, page) || PageCompound(page))
313 pagevec_lru_move_fn(pvec, __activate_page, NULL);
314 put_cpu_var(activate_page_pvecs);
319 static inline void activate_page_drain(int cpu)
323 static bool need_activate_page_drain(int cpu)
328 void activate_page(struct page *page)
330 struct zone *zone = page_zone(page);
332 page = compound_head(page);
333 spin_lock_irq(zone_lru_lock(zone));
334 __activate_page(page, mem_cgroup_page_lruvec(page, zone->zone_pgdat), NULL);
335 spin_unlock_irq(zone_lru_lock(zone));
339 static void __lru_cache_activate_page(struct page *page)
341 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
345 * Search backwards on the optimistic assumption that the page being
346 * activated has just been added to this pagevec. Note that only
347 * the local pagevec is examined as a !PageLRU page could be in the
348 * process of being released, reclaimed, migrated or on a remote
349 * pagevec that is currently being drained. Furthermore, marking
350 * a remote pagevec's page PageActive potentially hits a race where
351 * a page is marked PageActive just after it is added to the inactive
352 * list causing accounting errors and BUG_ON checks to trigger.
354 for (i = pagevec_count(pvec) - 1; i >= 0; i--) {
355 struct page *pagevec_page = pvec->pages[i];
357 if (pagevec_page == page) {
363 put_cpu_var(lru_add_pvec);
367 * Mark a page as having seen activity.
369 * inactive,unreferenced -> inactive,referenced
370 * inactive,referenced -> active,unreferenced
371 * active,unreferenced -> active,referenced
373 * When a newly allocated page is not yet visible, so safe for non-atomic ops,
374 * __SetPageReferenced(page) may be substituted for mark_page_accessed(page).
376 void mark_page_accessed(struct page *page)
378 page = compound_head(page);
379 if (!PageActive(page) && !PageUnevictable(page) &&
380 PageReferenced(page)) {
383 * If the page is on the LRU, queue it for activation via
384 * activate_page_pvecs. Otherwise, assume the page is on a
385 * pagevec, mark it active and it'll be moved to the active
386 * LRU on the next drain.
391 __lru_cache_activate_page(page);
392 ClearPageReferenced(page);
393 if (page_is_file_cache(page))
394 workingset_activation(page);
395 } else if (!PageReferenced(page)) {
396 SetPageReferenced(page);
398 if (page_is_idle(page))
399 clear_page_idle(page);
401 EXPORT_SYMBOL(mark_page_accessed);
403 static void __lru_cache_add(struct page *page)
405 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
408 if (!pagevec_add(pvec, page) || PageCompound(page))
409 __pagevec_lru_add(pvec);
410 put_cpu_var(lru_add_pvec);
414 * lru_cache_add: add a page to the page lists
415 * @page: the page to add
417 void lru_cache_add_anon(struct page *page)
419 if (PageActive(page))
420 ClearPageActive(page);
421 __lru_cache_add(page);
424 void lru_cache_add_file(struct page *page)
426 if (PageActive(page))
427 ClearPageActive(page);
428 __lru_cache_add(page);
430 EXPORT_SYMBOL(lru_cache_add_file);
433 * lru_cache_add - add a page to a page list
434 * @page: the page to be added to the LRU.
436 * Queue the page for addition to the LRU via pagevec. The decision on whether
437 * to add the page to the [in]active [file|anon] list is deferred until the
438 * pagevec is drained. This gives a chance for the caller of lru_cache_add()
439 * have the page added to the active list using mark_page_accessed().
441 void lru_cache_add(struct page *page)
443 VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page);
444 VM_BUG_ON_PAGE(PageLRU(page), page);
445 __lru_cache_add(page);
449 * add_page_to_unevictable_list - add a page to the unevictable list
450 * @page: the page to be added to the unevictable list
452 * Add page directly to its zone's unevictable list. To avoid races with
453 * tasks that might be making the page evictable, through eg. munlock,
454 * munmap or exit, while it's not on the lru, we want to add the page
455 * while it's locked or otherwise "invisible" to other tasks. This is
456 * difficult to do when using the pagevec cache, so bypass that.
458 void add_page_to_unevictable_list(struct page *page)
460 struct pglist_data *pgdat = page_pgdat(page);
461 struct lruvec *lruvec;
463 spin_lock_irq(&pgdat->lru_lock);
464 lruvec = mem_cgroup_page_lruvec(page, pgdat);
465 ClearPageActive(page);
466 SetPageUnevictable(page);
468 add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE);
469 spin_unlock_irq(&pgdat->lru_lock);
473 * lru_cache_add_active_or_unevictable
474 * @page: the page to be added to LRU
475 * @vma: vma in which page is mapped for determining reclaimability
477 * Place @page on the active or unevictable LRU list, depending on its
478 * evictability. Note that if the page is not evictable, it goes
479 * directly back onto it's zone's unevictable list, it does NOT use a
482 void lru_cache_add_active_or_unevictable(struct page *page,
483 struct vm_area_struct *vma)
485 VM_BUG_ON_PAGE(PageLRU(page), page);
487 if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED)) {
493 if (!TestSetPageMlocked(page)) {
495 * We use the irq-unsafe __mod_zone_page_stat because this
496 * counter is not modified from interrupt context, and the pte
497 * lock is held(spinlock), which implies preemption disabled.
499 __mod_zone_page_state(page_zone(page), NR_MLOCK,
500 hpage_nr_pages(page));
501 count_vm_event(UNEVICTABLE_PGMLOCKED);
503 add_page_to_unevictable_list(page);
507 * If the page can not be invalidated, it is moved to the
508 * inactive list to speed up its reclaim. It is moved to the
509 * head of the list, rather than the tail, to give the flusher
510 * threads some time to write it out, as this is much more
511 * effective than the single-page writeout from reclaim.
513 * If the page isn't page_mapped and dirty/writeback, the page
514 * could reclaim asap using PG_reclaim.
516 * 1. active, mapped page -> none
517 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
518 * 3. inactive, mapped page -> none
519 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
520 * 5. inactive, clean -> inactive, tail
523 * In 4, why it moves inactive's head, the VM expects the page would
524 * be write it out by flusher threads as this is much more effective
525 * than the single-page writeout from reclaim.
527 static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec,
536 if (PageUnevictable(page))
539 /* Some processes are using the page */
540 if (page_mapped(page))
543 active = PageActive(page);
544 file = page_is_file_cache(page);
545 lru = page_lru_base_type(page);
547 del_page_from_lru_list(page, lruvec, lru + active);
548 ClearPageActive(page);
549 ClearPageReferenced(page);
550 add_page_to_lru_list(page, lruvec, lru);
552 if (PageWriteback(page) || PageDirty(page)) {
554 * PG_reclaim could be raced with end_page_writeback
555 * It can make readahead confusing. But race window
556 * is _really_ small and it's non-critical problem.
558 SetPageReclaim(page);
561 * The page's writeback ends up during pagevec
562 * We moves tha page into tail of inactive.
564 list_move_tail(&page->lru, &lruvec->lists[lru]);
565 __count_vm_event(PGROTATED);
569 __count_vm_event(PGDEACTIVATE);
570 update_page_reclaim_stat(lruvec, file, 0);
574 static void lru_lazyfree_fn(struct page *page, struct lruvec *lruvec,
577 if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) &&
578 !PageUnevictable(page)) {
579 bool active = PageActive(page);
581 del_page_from_lru_list(page, lruvec,
582 LRU_INACTIVE_ANON + active);
583 ClearPageActive(page);
584 ClearPageReferenced(page);
586 * lazyfree pages are clean anonymous pages. They have
587 * SwapBacked flag cleared to distinguish normal anonymous
590 ClearPageSwapBacked(page);
591 add_page_to_lru_list(page, lruvec, LRU_INACTIVE_FILE);
593 __count_vm_events(PGLAZYFREE, hpage_nr_pages(page));
594 count_memcg_page_event(page, PGLAZYFREE);
595 update_page_reclaim_stat(lruvec, 1, 0);
600 * Drain pages out of the cpu's pagevecs.
601 * Either "cpu" is the current CPU, and preemption has already been
602 * disabled; or "cpu" is being hot-unplugged, and is already dead.
604 void lru_add_drain_cpu(int cpu)
606 struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu);
608 if (pagevec_count(pvec))
609 __pagevec_lru_add(pvec);
611 pvec = &per_cpu(lru_rotate_pvecs, cpu);
612 if (pagevec_count(pvec)) {
615 /* No harm done if a racing interrupt already did this */
616 local_irq_save(flags);
617 pagevec_move_tail(pvec);
618 local_irq_restore(flags);
621 pvec = &per_cpu(lru_deactivate_file_pvecs, cpu);
622 if (pagevec_count(pvec))
623 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
625 pvec = &per_cpu(lru_lazyfree_pvecs, cpu);
626 if (pagevec_count(pvec))
627 pagevec_lru_move_fn(pvec, lru_lazyfree_fn, NULL);
629 activate_page_drain(cpu);
633 * deactivate_file_page - forcefully deactivate a file page
634 * @page: page to deactivate
636 * This function hints the VM that @page is a good reclaim candidate,
637 * for example if its invalidation fails due to the page being dirty
638 * or under writeback.
640 void deactivate_file_page(struct page *page)
643 * In a workload with many unevictable page such as mprotect,
644 * unevictable page deactivation for accelerating reclaim is pointless.
646 if (PageUnevictable(page))
649 if (likely(get_page_unless_zero(page))) {
650 struct pagevec *pvec = &get_cpu_var(lru_deactivate_file_pvecs);
652 if (!pagevec_add(pvec, page) || PageCompound(page))
653 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
654 put_cpu_var(lru_deactivate_file_pvecs);
659 * mark_page_lazyfree - make an anon page lazyfree
660 * @page: page to deactivate
662 * mark_page_lazyfree() moves @page to the inactive file list.
663 * This is done to accelerate the reclaim of @page.
665 void mark_page_lazyfree(struct page *page)
667 if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) &&
668 !PageUnevictable(page)) {
669 struct pagevec *pvec = &get_cpu_var(lru_lazyfree_pvecs);
672 if (!pagevec_add(pvec, page) || PageCompound(page))
673 pagevec_lru_move_fn(pvec, lru_lazyfree_fn, NULL);
674 put_cpu_var(lru_lazyfree_pvecs);
678 void lru_add_drain(void)
680 lru_add_drain_cpu(get_cpu());
684 static void lru_add_drain_per_cpu(struct work_struct *dummy)
689 static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
691 void lru_add_drain_all_cpuslocked(void)
693 static DEFINE_MUTEX(lock);
694 static struct cpumask has_work;
698 * Make sure nobody triggers this path before mm_percpu_wq is fully
701 if (WARN_ON(!mm_percpu_wq))
705 cpumask_clear(&has_work);
707 for_each_online_cpu(cpu) {
708 struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
710 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
711 pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
712 pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) ||
713 pagevec_count(&per_cpu(lru_lazyfree_pvecs, cpu)) ||
714 need_activate_page_drain(cpu)) {
715 INIT_WORK(work, lru_add_drain_per_cpu);
716 queue_work_on(cpu, mm_percpu_wq, work);
717 cpumask_set_cpu(cpu, &has_work);
721 for_each_cpu(cpu, &has_work)
722 flush_work(&per_cpu(lru_add_drain_work, cpu));
727 void lru_add_drain_all(void)
730 lru_add_drain_all_cpuslocked();
735 * release_pages - batched put_page()
736 * @pages: array of pages to release
737 * @nr: number of pages
738 * @cold: whether the pages are cache cold
740 * Decrement the reference count on all the pages in @pages. If it
741 * fell to zero, remove the page from the LRU and free it.
743 void release_pages(struct page **pages, int nr, bool cold)
746 LIST_HEAD(pages_to_free);
747 struct pglist_data *locked_pgdat = NULL;
748 struct lruvec *lruvec;
749 unsigned long uninitialized_var(flags);
750 unsigned int uninitialized_var(lock_batch);
752 for (i = 0; i < nr; i++) {
753 struct page *page = pages[i];
756 * Make sure the IRQ-safe lock-holding time does not get
757 * excessive with a continuous string of pages from the
758 * same pgdat. The lock is held only if pgdat != NULL.
760 if (locked_pgdat && ++lock_batch == SWAP_CLUSTER_MAX) {
761 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
765 if (is_huge_zero_page(page))
768 /* Device public page can not be huge page */
769 if (is_device_public_page(page)) {
771 spin_unlock_irqrestore(&locked_pgdat->lru_lock,
775 put_zone_device_private_or_public_page(page);
779 page = compound_head(page);
780 if (!put_page_testzero(page))
783 if (PageCompound(page)) {
785 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
788 __put_compound_page(page);
793 struct pglist_data *pgdat = page_pgdat(page);
795 if (pgdat != locked_pgdat) {
797 spin_unlock_irqrestore(&locked_pgdat->lru_lock,
800 locked_pgdat = pgdat;
801 spin_lock_irqsave(&locked_pgdat->lru_lock, flags);
804 lruvec = mem_cgroup_page_lruvec(page, locked_pgdat);
805 VM_BUG_ON_PAGE(!PageLRU(page), page);
806 __ClearPageLRU(page);
807 del_page_from_lru_list(page, lruvec, page_off_lru(page));
810 /* Clear Active bit in case of parallel mark_page_accessed */
811 __ClearPageActive(page);
812 __ClearPageWaiters(page);
814 list_add(&page->lru, &pages_to_free);
817 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
819 mem_cgroup_uncharge_list(&pages_to_free);
820 free_hot_cold_page_list(&pages_to_free, cold);
822 EXPORT_SYMBOL(release_pages);
825 * The pages which we're about to release may be in the deferred lru-addition
826 * queues. That would prevent them from really being freed right now. That's
827 * OK from a correctness point of view but is inefficient - those pages may be
828 * cache-warm and we want to give them back to the page allocator ASAP.
830 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
831 * and __pagevec_lru_add_active() call release_pages() directly to avoid
834 void __pagevec_release(struct pagevec *pvec)
837 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
838 pagevec_reinit(pvec);
840 EXPORT_SYMBOL(__pagevec_release);
842 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
843 /* used by __split_huge_page_refcount() */
844 void lru_add_page_tail(struct page *page, struct page *page_tail,
845 struct lruvec *lruvec, struct list_head *list)
849 VM_BUG_ON_PAGE(!PageHead(page), page);
850 VM_BUG_ON_PAGE(PageCompound(page_tail), page);
851 VM_BUG_ON_PAGE(PageLRU(page_tail), page);
852 VM_BUG_ON(NR_CPUS != 1 &&
853 !spin_is_locked(&lruvec_pgdat(lruvec)->lru_lock));
856 SetPageLRU(page_tail);
858 if (likely(PageLRU(page)))
859 list_add_tail(&page_tail->lru, &page->lru);
861 /* page reclaim is reclaiming a huge page */
863 list_add_tail(&page_tail->lru, list);
865 struct list_head *list_head;
867 * Head page has not yet been counted, as an hpage,
868 * so we must account for each subpage individually.
870 * Use the standard add function to put page_tail on the list,
871 * but then correct its position so they all end up in order.
873 add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail));
874 list_head = page_tail->lru.prev;
875 list_move_tail(&page_tail->lru, list_head);
878 if (!PageUnevictable(page))
879 update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
881 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
883 static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
886 int file = page_is_file_cache(page);
887 int active = PageActive(page);
888 enum lru_list lru = page_lru(page);
890 VM_BUG_ON_PAGE(PageLRU(page), page);
893 add_page_to_lru_list(page, lruvec, lru);
894 update_page_reclaim_stat(lruvec, file, active);
895 trace_mm_lru_insertion(page, lru);
899 * Add the passed pages to the LRU, then drop the caller's refcount
900 * on them. Reinitialises the caller's pagevec.
902 void __pagevec_lru_add(struct pagevec *pvec)
904 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
906 EXPORT_SYMBOL(__pagevec_lru_add);
909 * pagevec_lookup_entries - gang pagecache lookup
910 * @pvec: Where the resulting entries are placed
911 * @mapping: The address_space to search
912 * @start: The starting entry index
913 * @nr_entries: The maximum number of entries
914 * @indices: The cache indices corresponding to the entries in @pvec
916 * pagevec_lookup_entries() will search for and return a group of up
917 * to @nr_entries pages and shadow entries in the mapping. All
918 * entries are placed in @pvec. pagevec_lookup_entries() takes a
919 * reference against actual pages in @pvec.
921 * The search returns a group of mapping-contiguous entries with
922 * ascending indexes. There may be holes in the indices due to
923 * not-present entries.
925 * pagevec_lookup_entries() returns the number of entries which were
928 unsigned pagevec_lookup_entries(struct pagevec *pvec,
929 struct address_space *mapping,
930 pgoff_t start, unsigned nr_pages,
933 pvec->nr = find_get_entries(mapping, start, nr_pages,
934 pvec->pages, indices);
935 return pagevec_count(pvec);
939 * pagevec_remove_exceptionals - pagevec exceptionals pruning
940 * @pvec: The pagevec to prune
942 * pagevec_lookup_entries() fills both pages and exceptional radix
943 * tree entries into the pagevec. This function prunes all
944 * exceptionals from @pvec without leaving holes, so that it can be
945 * passed on to page-only pagevec operations.
947 void pagevec_remove_exceptionals(struct pagevec *pvec)
951 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
952 struct page *page = pvec->pages[i];
953 if (!radix_tree_exceptional_entry(page))
954 pvec->pages[j++] = page;
960 * pagevec_lookup_range - gang pagecache lookup
961 * @pvec: Where the resulting pages are placed
962 * @mapping: The address_space to search
963 * @start: The starting page index
964 * @end: The final page index
965 * @nr_pages: The maximum number of pages
967 * pagevec_lookup_range() will search for and return a group of up to @nr_pages
968 * pages in the mapping starting from index @start and upto index @end
969 * (inclusive). The pages are placed in @pvec. pagevec_lookup() takes a
970 * reference against the pages in @pvec.
972 * The search returns a group of mapping-contiguous pages with ascending
973 * indexes. There may be holes in the indices due to not-present pages. We
974 * also update @start to index the next page for the traversal.
976 * pagevec_lookup_range() returns the number of pages which were found. If this
977 * number is smaller than @nr_pages, the end of specified range has been
980 unsigned pagevec_lookup_range(struct pagevec *pvec,
981 struct address_space *mapping, pgoff_t *start, pgoff_t end)
983 pvec->nr = find_get_pages_range(mapping, start, end, PAGEVEC_SIZE,
985 return pagevec_count(pvec);
987 EXPORT_SYMBOL(pagevec_lookup_range);
989 unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
990 pgoff_t *index, int tag, unsigned nr_pages)
992 pvec->nr = find_get_pages_tag(mapping, index, tag,
993 nr_pages, pvec->pages);
994 return pagevec_count(pvec);
996 EXPORT_SYMBOL(pagevec_lookup_tag);
999 * Perform any setup for the swap system
1001 void __init swap_setup(void)
1003 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
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