4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * This file contains the default values for the operation of the
9 * Linux VM subsystem. Fine-tuning documentation can be found in
10 * Documentation/sysctl/vm.txt.
12 * Swap aging added 23.2.95, Stephen Tweedie.
13 * Buffermem limits added 12.3.98, Rik van Riel.
17 #include <linux/sched.h>
18 #include <linux/kernel_stat.h>
19 #include <linux/swap.h>
20 #include <linux/mman.h>
21 #include <linux/pagemap.h>
22 #include <linux/pagevec.h>
23 #include <linux/init.h>
24 #include <linux/export.h>
25 #include <linux/mm_inline.h>
26 #include <linux/percpu_counter.h>
27 #include <linux/percpu.h>
28 #include <linux/cpu.h>
29 #include <linux/notifier.h>
30 #include <linux/backing-dev.h>
31 #include <linux/memcontrol.h>
32 #include <linux/gfp.h>
33 #include <linux/uio.h>
34 #include <linux/hugetlb.h>
38 #define CREATE_TRACE_POINTS
39 #include <trace/events/pagemap.h>
41 /* How many pages do we try to swap or page in/out together? */
44 static DEFINE_PER_CPU(struct pagevec, lru_add_pvec);
45 static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
46 static DEFINE_PER_CPU(struct pagevec, lru_deactivate_file_pvecs);
49 * This path almost never happens for VM activity - pages are normally
50 * freed via pagevecs. But it gets used by networking.
52 static void __page_cache_release(struct page *page)
55 struct zone *zone = page_zone(page);
56 struct lruvec *lruvec;
59 spin_lock_irqsave(&zone->lru_lock, flags);
60 lruvec = mem_cgroup_page_lruvec(page, zone);
61 VM_BUG_ON_PAGE(!PageLRU(page), page);
63 del_page_from_lru_list(page, lruvec, page_off_lru(page));
64 spin_unlock_irqrestore(&zone->lru_lock, flags);
66 mem_cgroup_uncharge(page);
69 static void __put_single_page(struct page *page)
71 __page_cache_release(page);
72 free_hot_cold_page(page, false);
75 static void __put_compound_page(struct page *page)
77 compound_page_dtor *dtor;
80 * __page_cache_release() is supposed to be called for thp, not for
81 * hugetlb. This is because hugetlb page does never have PageLRU set
82 * (it's never listed to any LRU lists) and no memcg routines should
83 * be called for hugetlb (it has a separate hugetlb_cgroup.)
86 __page_cache_release(page);
87 dtor = get_compound_page_dtor(page);
92 * Two special cases here: we could avoid taking compound_lock_irqsave
93 * and could skip the tail refcounting(in _mapcount).
97 * PageHeadHuge will remain true until the compound page
98 * is released and enters the buddy allocator, and it could
99 * not be split by __split_huge_page_refcount().
101 * So if we see PageHeadHuge set, and we have the tail page pin,
102 * then we could safely put head page.
106 * PG_slab is cleared before the slab frees the head page, and
107 * tail pin cannot be the last reference left on the head page,
108 * because the slab code is free to reuse the compound page
109 * after a kfree/kmem_cache_free without having to check if
110 * there's any tail pin left. In turn all tail pinsmust be always
111 * released while the head is still pinned by the slab code
112 * and so we know PG_slab will be still set too.
114 * So if we see PageSlab set, and we have the tail page pin,
115 * then we could safely put head page.
117 static __always_inline
118 void put_unrefcounted_compound_page(struct page *page_head, struct page *page)
121 * If @page is a THP tail, we must read the tail page
122 * flags after the head page flags. The
123 * __split_huge_page_refcount side enforces write memory barriers
124 * between clearing PageTail and before the head page
125 * can be freed and reallocated.
128 if (likely(PageTail(page))) {
130 * __split_huge_page_refcount cannot race
131 * here, see the comment above this function.
133 VM_BUG_ON_PAGE(!PageHead(page_head), page_head);
134 if (put_page_testzero(page_head)) {
136 * If this is the tail of a slab THP page,
137 * the tail pin must not be the last reference
138 * held on the page, because the PG_slab cannot
139 * be cleared before all tail pins (which skips
140 * the _mapcount tail refcounting) have been
143 * If this is the tail of a hugetlbfs page,
144 * the tail pin may be the last reference on
145 * the page instead, because PageHeadHuge will
146 * not go away until the compound page enters
147 * the buddy allocator.
149 VM_BUG_ON_PAGE(PageSlab(page_head), page_head);
150 __put_compound_page(page_head);
154 * __split_huge_page_refcount run before us,
155 * @page was a THP tail. The split @page_head
156 * has been freed and reallocated as slab or
157 * hugetlbfs page of smaller order (only
158 * possible if reallocated as slab on x86).
160 if (put_page_testzero(page))
161 __put_single_page(page);
164 static __always_inline
165 void put_refcounted_compound_page(struct page *page_head, struct page *page)
167 if (likely(page != page_head && get_page_unless_zero(page_head))) {
171 * @page_head wasn't a dangling pointer but it may not
172 * be a head page anymore by the time we obtain the
173 * lock. That is ok as long as it can't be freed from
176 flags = compound_lock_irqsave(page_head);
177 if (unlikely(!PageTail(page))) {
178 /* __split_huge_page_refcount run before us */
179 compound_unlock_irqrestore(page_head, flags);
180 if (put_page_testzero(page_head)) {
182 * The @page_head may have been freed
183 * and reallocated as a compound page
184 * of smaller order and then freed
185 * again. All we know is that it
186 * cannot have become: a THP page, a
187 * compound page of higher order, a
188 * tail page. That is because we
189 * still hold the refcount of the
190 * split THP tail and page_head was
191 * the THP head before the split.
193 if (PageHead(page_head))
194 __put_compound_page(page_head);
196 __put_single_page(page_head);
199 if (put_page_testzero(page))
200 __put_single_page(page);
203 VM_BUG_ON_PAGE(page_head != page->first_page, page);
205 * We can release the refcount taken by
206 * get_page_unless_zero() now that
207 * __split_huge_page_refcount() is blocked on the
210 if (put_page_testzero(page_head))
211 VM_BUG_ON_PAGE(1, page_head);
212 /* __split_huge_page_refcount will wait now */
213 VM_BUG_ON_PAGE(page_mapcount(page) <= 0, page);
214 atomic_dec(&page->_mapcount);
215 VM_BUG_ON_PAGE(atomic_read(&page_head->_count) <= 0, page_head);
216 VM_BUG_ON_PAGE(atomic_read(&page->_count) != 0, page);
217 compound_unlock_irqrestore(page_head, flags);
219 if (put_page_testzero(page_head)) {
220 if (PageHead(page_head))
221 __put_compound_page(page_head);
223 __put_single_page(page_head);
226 /* @page_head is a dangling pointer */
227 VM_BUG_ON_PAGE(PageTail(page), page);
232 static void put_compound_page(struct page *page)
234 struct page *page_head;
237 * We see the PageCompound set and PageTail not set, so @page maybe:
238 * 1. hugetlbfs head page, or
241 if (likely(!PageTail(page))) {
242 if (put_page_testzero(page)) {
244 * By the time all refcounts have been released
245 * split_huge_page cannot run anymore from under us.
248 __put_compound_page(page);
250 __put_single_page(page);
256 * We see the PageCompound set and PageTail set, so @page maybe:
257 * 1. a tail hugetlbfs page, or
258 * 2. a tail THP page, or
259 * 3. a split THP page.
261 * Case 3 is possible, as we may race with
262 * __split_huge_page_refcount tearing down a THP page.
264 page_head = compound_head_by_tail(page);
265 if (!__compound_tail_refcounted(page_head))
266 put_unrefcounted_compound_page(page_head, page);
268 put_refcounted_compound_page(page_head, page);
271 void put_page(struct page *page)
273 if (unlikely(PageCompound(page)))
274 put_compound_page(page);
275 else if (put_page_testzero(page))
276 __put_single_page(page);
278 EXPORT_SYMBOL(put_page);
281 * This function is exported but must not be called by anything other
282 * than get_page(). It implements the slow path of get_page().
284 bool __get_page_tail(struct page *page)
287 * This takes care of get_page() if run on a tail page
288 * returned by one of the get_user_pages/follow_page variants.
289 * get_user_pages/follow_page itself doesn't need the compound
290 * lock because it runs __get_page_tail_foll() under the
291 * proper PT lock that already serializes against
296 struct page *page_head = compound_head(page);
298 /* Ref to put_compound_page() comment. */
299 if (!__compound_tail_refcounted(page_head)) {
301 if (likely(PageTail(page))) {
303 * This is a hugetlbfs page or a slab
304 * page. __split_huge_page_refcount
307 VM_BUG_ON_PAGE(!PageHead(page_head), page_head);
308 __get_page_tail_foll(page, true);
312 * __split_huge_page_refcount run
313 * before us, "page" was a THP
314 * tail. The split page_head has been
315 * freed and reallocated as slab or
316 * hugetlbfs page of smaller order
317 * (only possible if reallocated as
325 if (likely(page != page_head && get_page_unless_zero(page_head))) {
327 * page_head wasn't a dangling pointer but it
328 * may not be a head page anymore by the time
329 * we obtain the lock. That is ok as long as it
330 * can't be freed from under us.
332 flags = compound_lock_irqsave(page_head);
333 /* here __split_huge_page_refcount won't run anymore */
334 if (likely(PageTail(page))) {
335 __get_page_tail_foll(page, false);
338 compound_unlock_irqrestore(page_head, flags);
344 EXPORT_SYMBOL(__get_page_tail);
347 * put_pages_list() - release a list of pages
348 * @pages: list of pages threaded on page->lru
350 * Release a list of pages which are strung together on page.lru. Currently
351 * used by read_cache_pages() and related error recovery code.
353 void put_pages_list(struct list_head *pages)
355 while (!list_empty(pages)) {
358 victim = list_entry(pages->prev, struct page, lru);
359 list_del(&victim->lru);
360 page_cache_release(victim);
363 EXPORT_SYMBOL(put_pages_list);
366 * get_kernel_pages() - pin kernel pages in memory
367 * @kiov: An array of struct kvec structures
368 * @nr_segs: number of segments to pin
369 * @write: pinning for read/write, currently ignored
370 * @pages: array that receives pointers to the pages pinned.
371 * Should be at least nr_segs long.
373 * Returns number of pages pinned. This may be fewer than the number
374 * requested. If nr_pages is 0 or negative, returns 0. If no pages
375 * were pinned, returns -errno. Each page returned must be released
376 * with a put_page() call when it is finished with.
378 int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write,
383 for (seg = 0; seg < nr_segs; seg++) {
384 if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE))
387 pages[seg] = kmap_to_page(kiov[seg].iov_base);
388 page_cache_get(pages[seg]);
393 EXPORT_SYMBOL_GPL(get_kernel_pages);
396 * get_kernel_page() - pin a kernel page in memory
397 * @start: starting kernel address
398 * @write: pinning for read/write, currently ignored
399 * @pages: array that receives pointer to the page pinned.
400 * Must be at least nr_segs long.
402 * Returns 1 if page is pinned. If the page was not pinned, returns
403 * -errno. The page returned must be released with a put_page() call
404 * when it is finished with.
406 int get_kernel_page(unsigned long start, int write, struct page **pages)
408 const struct kvec kiov = {
409 .iov_base = (void *)start,
413 return get_kernel_pages(&kiov, 1, write, pages);
415 EXPORT_SYMBOL_GPL(get_kernel_page);
417 static void pagevec_lru_move_fn(struct pagevec *pvec,
418 void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg),
422 struct zone *zone = NULL;
423 struct lruvec *lruvec;
424 unsigned long flags = 0;
426 for (i = 0; i < pagevec_count(pvec); i++) {
427 struct page *page = pvec->pages[i];
428 struct zone *pagezone = page_zone(page);
430 if (pagezone != zone) {
432 spin_unlock_irqrestore(&zone->lru_lock, flags);
434 spin_lock_irqsave(&zone->lru_lock, flags);
437 lruvec = mem_cgroup_page_lruvec(page, zone);
438 (*move_fn)(page, lruvec, arg);
441 spin_unlock_irqrestore(&zone->lru_lock, flags);
442 release_pages(pvec->pages, pvec->nr, pvec->cold);
443 pagevec_reinit(pvec);
446 static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec,
451 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
452 enum lru_list lru = page_lru_base_type(page);
453 list_move_tail(&page->lru, &lruvec->lists[lru]);
459 * pagevec_move_tail() must be called with IRQ disabled.
460 * Otherwise this may cause nasty races.
462 static void pagevec_move_tail(struct pagevec *pvec)
466 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
467 __count_vm_events(PGROTATED, pgmoved);
471 * Writeback is about to end against a page which has been marked for immediate
472 * reclaim. If it still appears to be reclaimable, move it to the tail of the
475 void rotate_reclaimable_page(struct page *page)
477 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
478 !PageUnevictable(page) && PageLRU(page)) {
479 struct pagevec *pvec;
482 page_cache_get(page);
483 local_irq_save(flags);
484 pvec = this_cpu_ptr(&lru_rotate_pvecs);
485 if (!pagevec_add(pvec, page))
486 pagevec_move_tail(pvec);
487 local_irq_restore(flags);
491 static void update_page_reclaim_stat(struct lruvec *lruvec,
492 int file, int rotated)
494 struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
496 reclaim_stat->recent_scanned[file]++;
498 reclaim_stat->recent_rotated[file]++;
501 static void __activate_page(struct page *page, struct lruvec *lruvec,
504 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
505 int file = page_is_file_cache(page);
506 int lru = page_lru_base_type(page);
508 del_page_from_lru_list(page, lruvec, lru);
511 add_page_to_lru_list(page, lruvec, lru);
512 trace_mm_lru_activate(page);
514 __count_vm_event(PGACTIVATE);
515 update_page_reclaim_stat(lruvec, file, 1);
520 static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
522 static void activate_page_drain(int cpu)
524 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
526 if (pagevec_count(pvec))
527 pagevec_lru_move_fn(pvec, __activate_page, NULL);
530 static bool need_activate_page_drain(int cpu)
532 return pagevec_count(&per_cpu(activate_page_pvecs, cpu)) != 0;
535 void activate_page(struct page *page)
537 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
538 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
540 page_cache_get(page);
541 if (!pagevec_add(pvec, page))
542 pagevec_lru_move_fn(pvec, __activate_page, NULL);
543 put_cpu_var(activate_page_pvecs);
548 static inline void activate_page_drain(int cpu)
552 static bool need_activate_page_drain(int cpu)
557 void activate_page(struct page *page)
559 struct zone *zone = page_zone(page);
561 spin_lock_irq(&zone->lru_lock);
562 __activate_page(page, mem_cgroup_page_lruvec(page, zone), NULL);
563 spin_unlock_irq(&zone->lru_lock);
567 static void __lru_cache_activate_page(struct page *page)
569 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
573 * Search backwards on the optimistic assumption that the page being
574 * activated has just been added to this pagevec. Note that only
575 * the local pagevec is examined as a !PageLRU page could be in the
576 * process of being released, reclaimed, migrated or on a remote
577 * pagevec that is currently being drained. Furthermore, marking
578 * a remote pagevec's page PageActive potentially hits a race where
579 * a page is marked PageActive just after it is added to the inactive
580 * list causing accounting errors and BUG_ON checks to trigger.
582 for (i = pagevec_count(pvec) - 1; i >= 0; i--) {
583 struct page *pagevec_page = pvec->pages[i];
585 if (pagevec_page == page) {
591 put_cpu_var(lru_add_pvec);
595 * Mark a page as having seen activity.
597 * inactive,unreferenced -> inactive,referenced
598 * inactive,referenced -> active,unreferenced
599 * active,unreferenced -> active,referenced
601 * When a newly allocated page is not yet visible, so safe for non-atomic ops,
602 * __SetPageReferenced(page) may be substituted for mark_page_accessed(page).
604 void mark_page_accessed(struct page *page)
606 if (!PageActive(page) && !PageUnevictable(page) &&
607 PageReferenced(page)) {
610 * If the page is on the LRU, queue it for activation via
611 * activate_page_pvecs. Otherwise, assume the page is on a
612 * pagevec, mark it active and it'll be moved to the active
613 * LRU on the next drain.
618 __lru_cache_activate_page(page);
619 ClearPageReferenced(page);
620 if (page_is_file_cache(page))
621 workingset_activation(page);
622 } else if (!PageReferenced(page)) {
623 SetPageReferenced(page);
626 EXPORT_SYMBOL(mark_page_accessed);
628 static void __lru_cache_add(struct page *page)
630 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
632 page_cache_get(page);
633 if (!pagevec_space(pvec))
634 __pagevec_lru_add(pvec);
635 pagevec_add(pvec, page);
636 put_cpu_var(lru_add_pvec);
640 * lru_cache_add: add a page to the page lists
641 * @page: the page to add
643 void lru_cache_add_anon(struct page *page)
645 if (PageActive(page))
646 ClearPageActive(page);
647 __lru_cache_add(page);
650 void lru_cache_add_file(struct page *page)
652 if (PageActive(page))
653 ClearPageActive(page);
654 __lru_cache_add(page);
656 EXPORT_SYMBOL(lru_cache_add_file);
659 * lru_cache_add - add a page to a page list
660 * @page: the page to be added to the LRU.
662 * Queue the page for addition to the LRU via pagevec. The decision on whether
663 * to add the page to the [in]active [file|anon] list is deferred until the
664 * pagevec is drained. This gives a chance for the caller of lru_cache_add()
665 * have the page added to the active list using mark_page_accessed().
667 void lru_cache_add(struct page *page)
669 VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page);
670 VM_BUG_ON_PAGE(PageLRU(page), page);
671 __lru_cache_add(page);
675 * add_page_to_unevictable_list - add a page to the unevictable list
676 * @page: the page to be added to the unevictable list
678 * Add page directly to its zone's unevictable list. To avoid races with
679 * tasks that might be making the page evictable, through eg. munlock,
680 * munmap or exit, while it's not on the lru, we want to add the page
681 * while it's locked or otherwise "invisible" to other tasks. This is
682 * difficult to do when using the pagevec cache, so bypass that.
684 void add_page_to_unevictable_list(struct page *page)
686 struct zone *zone = page_zone(page);
687 struct lruvec *lruvec;
689 spin_lock_irq(&zone->lru_lock);
690 lruvec = mem_cgroup_page_lruvec(page, zone);
691 ClearPageActive(page);
692 SetPageUnevictable(page);
694 add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE);
695 spin_unlock_irq(&zone->lru_lock);
699 * lru_cache_add_active_or_unevictable
700 * @page: the page to be added to LRU
701 * @vma: vma in which page is mapped for determining reclaimability
703 * Place @page on the active or unevictable LRU list, depending on its
704 * evictability. Note that if the page is not evictable, it goes
705 * directly back onto it's zone's unevictable list, it does NOT use a
708 void lru_cache_add_active_or_unevictable(struct page *page,
709 struct vm_area_struct *vma)
711 VM_BUG_ON_PAGE(PageLRU(page), page);
713 if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED)) {
719 if (!TestSetPageMlocked(page)) {
721 * We use the irq-unsafe __mod_zone_page_stat because this
722 * counter is not modified from interrupt context, and the pte
723 * lock is held(spinlock), which implies preemption disabled.
725 __mod_zone_page_state(page_zone(page), NR_MLOCK,
726 hpage_nr_pages(page));
727 count_vm_event(UNEVICTABLE_PGMLOCKED);
729 add_page_to_unevictable_list(page);
733 * If the page can not be invalidated, it is moved to the
734 * inactive list to speed up its reclaim. It is moved to the
735 * head of the list, rather than the tail, to give the flusher
736 * threads some time to write it out, as this is much more
737 * effective than the single-page writeout from reclaim.
739 * If the page isn't page_mapped and dirty/writeback, the page
740 * could reclaim asap using PG_reclaim.
742 * 1. active, mapped page -> none
743 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
744 * 3. inactive, mapped page -> none
745 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
746 * 5. inactive, clean -> inactive, tail
749 * In 4, why it moves inactive's head, the VM expects the page would
750 * be write it out by flusher threads as this is much more effective
751 * than the single-page writeout from reclaim.
753 static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec,
762 if (PageUnevictable(page))
765 /* Some processes are using the page */
766 if (page_mapped(page))
769 active = PageActive(page);
770 file = page_is_file_cache(page);
771 lru = page_lru_base_type(page);
773 del_page_from_lru_list(page, lruvec, lru + active);
774 ClearPageActive(page);
775 ClearPageReferenced(page);
776 add_page_to_lru_list(page, lruvec, lru);
778 if (PageWriteback(page) || PageDirty(page)) {
780 * PG_reclaim could be raced with end_page_writeback
781 * It can make readahead confusing. But race window
782 * is _really_ small and it's non-critical problem.
784 SetPageReclaim(page);
787 * The page's writeback ends up during pagevec
788 * We moves tha page into tail of inactive.
790 list_move_tail(&page->lru, &lruvec->lists[lru]);
791 __count_vm_event(PGROTATED);
795 __count_vm_event(PGDEACTIVATE);
796 update_page_reclaim_stat(lruvec, file, 0);
800 * Drain pages out of the cpu's pagevecs.
801 * Either "cpu" is the current CPU, and preemption has already been
802 * disabled; or "cpu" is being hot-unplugged, and is already dead.
804 void lru_add_drain_cpu(int cpu)
806 struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu);
808 if (pagevec_count(pvec))
809 __pagevec_lru_add(pvec);
811 pvec = &per_cpu(lru_rotate_pvecs, cpu);
812 if (pagevec_count(pvec)) {
815 /* No harm done if a racing interrupt already did this */
816 local_irq_save(flags);
817 pagevec_move_tail(pvec);
818 local_irq_restore(flags);
821 pvec = &per_cpu(lru_deactivate_file_pvecs, cpu);
822 if (pagevec_count(pvec))
823 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
825 activate_page_drain(cpu);
829 * deactivate_file_page - forcefully deactivate a file page
830 * @page: page to deactivate
832 * This function hints the VM that @page is a good reclaim candidate,
833 * for example if its invalidation fails due to the page being dirty
834 * or under writeback.
836 void deactivate_file_page(struct page *page)
839 * In a workload with many unevictable page such as mprotect,
840 * unevictable page deactivation for accelerating reclaim is pointless.
842 if (PageUnevictable(page))
845 if (likely(get_page_unless_zero(page))) {
846 struct pagevec *pvec = &get_cpu_var(lru_deactivate_file_pvecs);
848 if (!pagevec_add(pvec, page))
849 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
850 put_cpu_var(lru_deactivate_file_pvecs);
854 void lru_add_drain(void)
856 lru_add_drain_cpu(get_cpu());
860 static void lru_add_drain_per_cpu(struct work_struct *dummy)
865 static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
867 void lru_add_drain_all(void)
869 static DEFINE_MUTEX(lock);
870 static struct cpumask has_work;
875 cpumask_clear(&has_work);
877 for_each_online_cpu(cpu) {
878 struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
880 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
881 pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
882 pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) ||
883 need_activate_page_drain(cpu)) {
884 INIT_WORK(work, lru_add_drain_per_cpu);
885 schedule_work_on(cpu, work);
886 cpumask_set_cpu(cpu, &has_work);
890 for_each_cpu(cpu, &has_work)
891 flush_work(&per_cpu(lru_add_drain_work, cpu));
898 * release_pages - batched page_cache_release()
899 * @pages: array of pages to release
900 * @nr: number of pages
901 * @cold: whether the pages are cache cold
903 * Decrement the reference count on all the pages in @pages. If it
904 * fell to zero, remove the page from the LRU and free it.
906 void release_pages(struct page **pages, int nr, bool cold)
909 LIST_HEAD(pages_to_free);
910 struct zone *zone = NULL;
911 struct lruvec *lruvec;
912 unsigned long uninitialized_var(flags);
913 unsigned int uninitialized_var(lock_batch);
915 for (i = 0; i < nr; i++) {
916 struct page *page = pages[i];
918 if (unlikely(PageCompound(page))) {
920 spin_unlock_irqrestore(&zone->lru_lock, flags);
923 put_compound_page(page);
928 * Make sure the IRQ-safe lock-holding time does not get
929 * excessive with a continuous string of pages from the
930 * same zone. The lock is held only if zone != NULL.
932 if (zone && ++lock_batch == SWAP_CLUSTER_MAX) {
933 spin_unlock_irqrestore(&zone->lru_lock, flags);
937 if (!put_page_testzero(page))
941 struct zone *pagezone = page_zone(page);
943 if (pagezone != zone) {
945 spin_unlock_irqrestore(&zone->lru_lock,
949 spin_lock_irqsave(&zone->lru_lock, flags);
952 lruvec = mem_cgroup_page_lruvec(page, zone);
953 VM_BUG_ON_PAGE(!PageLRU(page), page);
954 __ClearPageLRU(page);
955 del_page_from_lru_list(page, lruvec, page_off_lru(page));
958 /* Clear Active bit in case of parallel mark_page_accessed */
959 __ClearPageActive(page);
961 list_add(&page->lru, &pages_to_free);
964 spin_unlock_irqrestore(&zone->lru_lock, flags);
966 mem_cgroup_uncharge_list(&pages_to_free);
967 free_hot_cold_page_list(&pages_to_free, cold);
969 EXPORT_SYMBOL(release_pages);
972 * The pages which we're about to release may be in the deferred lru-addition
973 * queues. That would prevent them from really being freed right now. That's
974 * OK from a correctness point of view but is inefficient - those pages may be
975 * cache-warm and we want to give them back to the page allocator ASAP.
977 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
978 * and __pagevec_lru_add_active() call release_pages() directly to avoid
981 void __pagevec_release(struct pagevec *pvec)
984 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
985 pagevec_reinit(pvec);
987 EXPORT_SYMBOL(__pagevec_release);
989 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
990 /* used by __split_huge_page_refcount() */
991 void lru_add_page_tail(struct page *page, struct page *page_tail,
992 struct lruvec *lruvec, struct list_head *list)
996 VM_BUG_ON_PAGE(!PageHead(page), page);
997 VM_BUG_ON_PAGE(PageCompound(page_tail), page);
998 VM_BUG_ON_PAGE(PageLRU(page_tail), page);
999 VM_BUG_ON(NR_CPUS != 1 &&
1000 !spin_is_locked(&lruvec_zone(lruvec)->lru_lock));
1003 SetPageLRU(page_tail);
1005 if (likely(PageLRU(page)))
1006 list_add_tail(&page_tail->lru, &page->lru);
1008 /* page reclaim is reclaiming a huge page */
1009 get_page(page_tail);
1010 list_add_tail(&page_tail->lru, list);
1012 struct list_head *list_head;
1014 * Head page has not yet been counted, as an hpage,
1015 * so we must account for each subpage individually.
1017 * Use the standard add function to put page_tail on the list,
1018 * but then correct its position so they all end up in order.
1020 add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail));
1021 list_head = page_tail->lru.prev;
1022 list_move_tail(&page_tail->lru, list_head);
1025 if (!PageUnevictable(page))
1026 update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
1028 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1030 static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
1033 int file = page_is_file_cache(page);
1034 int active = PageActive(page);
1035 enum lru_list lru = page_lru(page);
1037 VM_BUG_ON_PAGE(PageLRU(page), page);
1040 add_page_to_lru_list(page, lruvec, lru);
1041 update_page_reclaim_stat(lruvec, file, active);
1042 trace_mm_lru_insertion(page, lru);
1046 * Add the passed pages to the LRU, then drop the caller's refcount
1047 * on them. Reinitialises the caller's pagevec.
1049 void __pagevec_lru_add(struct pagevec *pvec)
1051 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
1053 EXPORT_SYMBOL(__pagevec_lru_add);
1056 * pagevec_lookup_entries - gang pagecache lookup
1057 * @pvec: Where the resulting entries are placed
1058 * @mapping: The address_space to search
1059 * @start: The starting entry index
1060 * @nr_entries: The maximum number of entries
1061 * @indices: The cache indices corresponding to the entries in @pvec
1063 * pagevec_lookup_entries() will search for and return a group of up
1064 * to @nr_entries pages and shadow entries in the mapping. All
1065 * entries are placed in @pvec. pagevec_lookup_entries() takes a
1066 * reference against actual pages in @pvec.
1068 * The search returns a group of mapping-contiguous entries with
1069 * ascending indexes. There may be holes in the indices due to
1070 * not-present entries.
1072 * pagevec_lookup_entries() returns the number of entries which were
1075 unsigned pagevec_lookup_entries(struct pagevec *pvec,
1076 struct address_space *mapping,
1077 pgoff_t start, unsigned nr_pages,
1080 pvec->nr = find_get_entries(mapping, start, nr_pages,
1081 pvec->pages, indices);
1082 return pagevec_count(pvec);
1086 * pagevec_remove_exceptionals - pagevec exceptionals pruning
1087 * @pvec: The pagevec to prune
1089 * pagevec_lookup_entries() fills both pages and exceptional radix
1090 * tree entries into the pagevec. This function prunes all
1091 * exceptionals from @pvec without leaving holes, so that it can be
1092 * passed on to page-only pagevec operations.
1094 void pagevec_remove_exceptionals(struct pagevec *pvec)
1098 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
1099 struct page *page = pvec->pages[i];
1100 if (!radix_tree_exceptional_entry(page))
1101 pvec->pages[j++] = page;
1107 * pagevec_lookup - gang pagecache lookup
1108 * @pvec: Where the resulting pages are placed
1109 * @mapping: The address_space to search
1110 * @start: The starting page index
1111 * @nr_pages: The maximum number of pages
1113 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
1114 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
1115 * reference against the pages in @pvec.
1117 * The search returns a group of mapping-contiguous pages with ascending
1118 * indexes. There may be holes in the indices due to not-present pages.
1120 * pagevec_lookup() returns the number of pages which were found.
1122 unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
1123 pgoff_t start, unsigned nr_pages)
1125 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
1126 return pagevec_count(pvec);
1128 EXPORT_SYMBOL(pagevec_lookup);
1130 unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
1131 pgoff_t *index, int tag, unsigned nr_pages)
1133 pvec->nr = find_get_pages_tag(mapping, index, tag,
1134 nr_pages, pvec->pages);
1135 return pagevec_count(pvec);
1137 EXPORT_SYMBOL(pagevec_lookup_tag);
1140 * Perform any setup for the swap system
1142 void __init swap_setup(void)
1144 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
1148 for (i = 0; i < MAX_SWAPFILES; i++)
1149 spin_lock_init(&swapper_spaces[i].tree_lock);
1152 /* Use a smaller cluster for small-memory machines */
1158 * Right now other parts of the system means that we
1159 * _really_ don't want to cluster much more