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>
35 #include <linux/locallock.h>
39 #define CREATE_TRACE_POINTS
40 #include <trace/events/pagemap.h>
42 /* How many pages do we try to swap or page in/out together? */
45 static DEFINE_PER_CPU(struct pagevec, lru_add_pvec);
46 static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
47 static DEFINE_PER_CPU(struct pagevec, lru_deactivate_file_pvecs);
49 static DEFINE_LOCAL_IRQ_LOCK(rotate_lock);
50 DEFINE_LOCAL_IRQ_LOCK(swapvec_lock);
53 * This path almost never happens for VM activity - pages are normally
54 * freed via pagevecs. But it gets used by networking.
56 static void __page_cache_release(struct page *page)
59 struct zone *zone = page_zone(page);
60 struct lruvec *lruvec;
63 spin_lock_irqsave(&zone->lru_lock, flags);
64 lruvec = mem_cgroup_page_lruvec(page, zone);
65 VM_BUG_ON_PAGE(!PageLRU(page), page);
67 del_page_from_lru_list(page, lruvec, page_off_lru(page));
68 spin_unlock_irqrestore(&zone->lru_lock, flags);
70 mem_cgroup_uncharge(page);
73 static void __put_single_page(struct page *page)
75 __page_cache_release(page);
76 free_hot_cold_page(page, false);
79 static void __put_compound_page(struct page *page)
81 compound_page_dtor *dtor;
84 * __page_cache_release() is supposed to be called for thp, not for
85 * hugetlb. This is because hugetlb page does never have PageLRU set
86 * (it's never listed to any LRU lists) and no memcg routines should
87 * be called for hugetlb (it has a separate hugetlb_cgroup.)
90 __page_cache_release(page);
91 dtor = get_compound_page_dtor(page);
96 * Two special cases here: we could avoid taking compound_lock_irqsave
97 * and could skip the tail refcounting(in _mapcount).
101 * PageHeadHuge will remain true until the compound page
102 * is released and enters the buddy allocator, and it could
103 * not be split by __split_huge_page_refcount().
105 * So if we see PageHeadHuge set, and we have the tail page pin,
106 * then we could safely put head page.
110 * PG_slab is cleared before the slab frees the head page, and
111 * tail pin cannot be the last reference left on the head page,
112 * because the slab code is free to reuse the compound page
113 * after a kfree/kmem_cache_free without having to check if
114 * there's any tail pin left. In turn all tail pinsmust be always
115 * released while the head is still pinned by the slab code
116 * and so we know PG_slab will be still set too.
118 * So if we see PageSlab set, and we have the tail page pin,
119 * then we could safely put head page.
121 static __always_inline
122 void put_unrefcounted_compound_page(struct page *page_head, struct page *page)
125 * If @page is a THP tail, we must read the tail page
126 * flags after the head page flags. The
127 * __split_huge_page_refcount side enforces write memory barriers
128 * between clearing PageTail and before the head page
129 * can be freed and reallocated.
132 if (likely(PageTail(page))) {
134 * __split_huge_page_refcount cannot race
135 * here, see the comment above this function.
137 VM_BUG_ON_PAGE(!PageHead(page_head), page_head);
138 VM_BUG_ON_PAGE(page_mapcount(page) != 0, page);
139 if (put_page_testzero(page_head)) {
141 * If this is the tail of a slab THP page,
142 * the tail pin must not be the last reference
143 * held on the page, because the PG_slab cannot
144 * be cleared before all tail pins (which skips
145 * the _mapcount tail refcounting) have been
148 * If this is the tail of a hugetlbfs page,
149 * the tail pin may be the last reference on
150 * the page instead, because PageHeadHuge will
151 * not go away until the compound page enters
152 * the buddy allocator.
154 VM_BUG_ON_PAGE(PageSlab(page_head), page_head);
155 __put_compound_page(page_head);
159 * __split_huge_page_refcount run before us,
160 * @page was a THP tail. The split @page_head
161 * has been freed and reallocated as slab or
162 * hugetlbfs page of smaller order (only
163 * possible if reallocated as slab on x86).
165 if (put_page_testzero(page))
166 __put_single_page(page);
169 static __always_inline
170 void put_refcounted_compound_page(struct page *page_head, struct page *page)
172 if (likely(page != page_head && get_page_unless_zero(page_head))) {
176 * @page_head wasn't a dangling pointer but it may not
177 * be a head page anymore by the time we obtain the
178 * lock. That is ok as long as it can't be freed from
181 flags = compound_lock_irqsave(page_head);
182 if (unlikely(!PageTail(page))) {
183 /* __split_huge_page_refcount run before us */
184 compound_unlock_irqrestore(page_head, flags);
185 if (put_page_testzero(page_head)) {
187 * The @page_head may have been freed
188 * and reallocated as a compound page
189 * of smaller order and then freed
190 * again. All we know is that it
191 * cannot have become: a THP page, a
192 * compound page of higher order, a
193 * tail page. That is because we
194 * still hold the refcount of the
195 * split THP tail and page_head was
196 * the THP head before the split.
198 if (PageHead(page_head))
199 __put_compound_page(page_head);
201 __put_single_page(page_head);
204 if (put_page_testzero(page))
205 __put_single_page(page);
208 VM_BUG_ON_PAGE(page_head != page->first_page, page);
210 * We can release the refcount taken by
211 * get_page_unless_zero() now that
212 * __split_huge_page_refcount() is blocked on the
215 if (put_page_testzero(page_head))
216 VM_BUG_ON_PAGE(1, page_head);
217 /* __split_huge_page_refcount will wait now */
218 VM_BUG_ON_PAGE(page_mapcount(page) <= 0, page);
219 atomic_dec(&page->_mapcount);
220 VM_BUG_ON_PAGE(atomic_read(&page_head->_count) <= 0, page_head);
221 VM_BUG_ON_PAGE(atomic_read(&page->_count) != 0, page);
222 compound_unlock_irqrestore(page_head, flags);
224 if (put_page_testzero(page_head)) {
225 if (PageHead(page_head))
226 __put_compound_page(page_head);
228 __put_single_page(page_head);
231 /* @page_head is a dangling pointer */
232 VM_BUG_ON_PAGE(PageTail(page), page);
237 static void put_compound_page(struct page *page)
239 struct page *page_head;
242 * We see the PageCompound set and PageTail not set, so @page maybe:
243 * 1. hugetlbfs head page, or
246 if (likely(!PageTail(page))) {
247 if (put_page_testzero(page)) {
249 * By the time all refcounts have been released
250 * split_huge_page cannot run anymore from under us.
253 __put_compound_page(page);
255 __put_single_page(page);
261 * We see the PageCompound set and PageTail set, so @page maybe:
262 * 1. a tail hugetlbfs page, or
263 * 2. a tail THP page, or
264 * 3. a split THP page.
266 * Case 3 is possible, as we may race with
267 * __split_huge_page_refcount tearing down a THP page.
269 page_head = compound_head_by_tail(page);
270 if (!__compound_tail_refcounted(page_head))
271 put_unrefcounted_compound_page(page_head, page);
273 put_refcounted_compound_page(page_head, page);
276 void put_page(struct page *page)
278 if (unlikely(PageCompound(page)))
279 put_compound_page(page);
280 else if (put_page_testzero(page))
281 __put_single_page(page);
283 EXPORT_SYMBOL(put_page);
286 * This function is exported but must not be called by anything other
287 * than get_page(). It implements the slow path of get_page().
289 bool __get_page_tail(struct page *page)
292 * This takes care of get_page() if run on a tail page
293 * returned by one of the get_user_pages/follow_page variants.
294 * get_user_pages/follow_page itself doesn't need the compound
295 * lock because it runs __get_page_tail_foll() under the
296 * proper PT lock that already serializes against
301 struct page *page_head = compound_head(page);
303 /* Ref to put_compound_page() comment. */
304 if (!__compound_tail_refcounted(page_head)) {
306 if (likely(PageTail(page))) {
308 * This is a hugetlbfs page or a slab
309 * page. __split_huge_page_refcount
312 VM_BUG_ON_PAGE(!PageHead(page_head), page_head);
313 __get_page_tail_foll(page, true);
317 * __split_huge_page_refcount run
318 * before us, "page" was a THP
319 * tail. The split page_head has been
320 * freed and reallocated as slab or
321 * hugetlbfs page of smaller order
322 * (only possible if reallocated as
330 if (likely(page != page_head && get_page_unless_zero(page_head))) {
332 * page_head wasn't a dangling pointer but it
333 * may not be a head page anymore by the time
334 * we obtain the lock. That is ok as long as it
335 * can't be freed from under us.
337 flags = compound_lock_irqsave(page_head);
338 /* here __split_huge_page_refcount won't run anymore */
339 if (likely(PageTail(page))) {
340 __get_page_tail_foll(page, false);
343 compound_unlock_irqrestore(page_head, flags);
349 EXPORT_SYMBOL(__get_page_tail);
352 * put_pages_list() - release a list of pages
353 * @pages: list of pages threaded on page->lru
355 * Release a list of pages which are strung together on page.lru. Currently
356 * used by read_cache_pages() and related error recovery code.
358 void put_pages_list(struct list_head *pages)
360 while (!list_empty(pages)) {
363 victim = list_entry(pages->prev, struct page, lru);
364 list_del(&victim->lru);
365 page_cache_release(victim);
368 EXPORT_SYMBOL(put_pages_list);
371 * get_kernel_pages() - pin kernel pages in memory
372 * @kiov: An array of struct kvec structures
373 * @nr_segs: number of segments to pin
374 * @write: pinning for read/write, currently ignored
375 * @pages: array that receives pointers to the pages pinned.
376 * Should be at least nr_segs long.
378 * Returns number of pages pinned. This may be fewer than the number
379 * requested. If nr_pages is 0 or negative, returns 0. If no pages
380 * were pinned, returns -errno. Each page returned must be released
381 * with a put_page() call when it is finished with.
383 int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write,
388 for (seg = 0; seg < nr_segs; seg++) {
389 if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE))
392 pages[seg] = kmap_to_page(kiov[seg].iov_base);
393 page_cache_get(pages[seg]);
398 EXPORT_SYMBOL_GPL(get_kernel_pages);
401 * get_kernel_page() - pin a kernel page in memory
402 * @start: starting kernel address
403 * @write: pinning for read/write, currently ignored
404 * @pages: array that receives pointer to the page pinned.
405 * Must be at least nr_segs long.
407 * Returns 1 if page is pinned. If the page was not pinned, returns
408 * -errno. The page returned must be released with a put_page() call
409 * when it is finished with.
411 int get_kernel_page(unsigned long start, int write, struct page **pages)
413 const struct kvec kiov = {
414 .iov_base = (void *)start,
418 return get_kernel_pages(&kiov, 1, write, pages);
420 EXPORT_SYMBOL_GPL(get_kernel_page);
422 static void pagevec_lru_move_fn(struct pagevec *pvec,
423 void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg),
427 struct zone *zone = NULL;
428 struct lruvec *lruvec;
429 unsigned long flags = 0;
431 for (i = 0; i < pagevec_count(pvec); i++) {
432 struct page *page = pvec->pages[i];
433 struct zone *pagezone = page_zone(page);
435 if (pagezone != zone) {
437 spin_unlock_irqrestore(&zone->lru_lock, flags);
439 spin_lock_irqsave(&zone->lru_lock, flags);
442 lruvec = mem_cgroup_page_lruvec(page, zone);
443 (*move_fn)(page, lruvec, arg);
446 spin_unlock_irqrestore(&zone->lru_lock, flags);
447 release_pages(pvec->pages, pvec->nr, pvec->cold);
448 pagevec_reinit(pvec);
451 static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec,
456 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
457 enum lru_list lru = page_lru_base_type(page);
458 list_move_tail(&page->lru, &lruvec->lists[lru]);
464 * pagevec_move_tail() must be called with IRQ disabled.
465 * Otherwise this may cause nasty races.
467 static void pagevec_move_tail(struct pagevec *pvec)
471 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
472 __count_vm_events(PGROTATED, pgmoved);
476 * Writeback is about to end against a page which has been marked for immediate
477 * reclaim. If it still appears to be reclaimable, move it to the tail of the
480 void rotate_reclaimable_page(struct page *page)
482 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
483 !PageUnevictable(page) && PageLRU(page)) {
484 struct pagevec *pvec;
487 page_cache_get(page);
488 local_lock_irqsave(rotate_lock, flags);
489 pvec = this_cpu_ptr(&lru_rotate_pvecs);
490 if (!pagevec_add(pvec, page))
491 pagevec_move_tail(pvec);
492 local_unlock_irqrestore(rotate_lock, flags);
496 static void update_page_reclaim_stat(struct lruvec *lruvec,
497 int file, int rotated)
499 struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
501 reclaim_stat->recent_scanned[file]++;
503 reclaim_stat->recent_rotated[file]++;
506 static void __activate_page(struct page *page, struct lruvec *lruvec,
509 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
510 int file = page_is_file_cache(page);
511 int lru = page_lru_base_type(page);
513 del_page_from_lru_list(page, lruvec, lru);
516 add_page_to_lru_list(page, lruvec, lru);
517 trace_mm_lru_activate(page);
519 __count_vm_event(PGACTIVATE);
520 update_page_reclaim_stat(lruvec, file, 1);
525 static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
527 static void activate_page_drain(int cpu)
529 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
531 if (pagevec_count(pvec))
532 pagevec_lru_move_fn(pvec, __activate_page, NULL);
535 static bool need_activate_page_drain(int cpu)
537 return pagevec_count(&per_cpu(activate_page_pvecs, cpu)) != 0;
540 void activate_page(struct page *page)
542 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
543 struct pagevec *pvec = &get_locked_var(swapvec_lock,
544 activate_page_pvecs);
546 page_cache_get(page);
547 if (!pagevec_add(pvec, page))
548 pagevec_lru_move_fn(pvec, __activate_page, NULL);
549 put_locked_var(swapvec_lock, activate_page_pvecs);
554 static inline void activate_page_drain(int cpu)
558 static bool need_activate_page_drain(int cpu)
563 void activate_page(struct page *page)
565 struct zone *zone = page_zone(page);
567 spin_lock_irq(&zone->lru_lock);
568 __activate_page(page, mem_cgroup_page_lruvec(page, zone), NULL);
569 spin_unlock_irq(&zone->lru_lock);
573 static void __lru_cache_activate_page(struct page *page)
575 struct pagevec *pvec = &get_locked_var(swapvec_lock, lru_add_pvec);
579 * Search backwards on the optimistic assumption that the page being
580 * activated has just been added to this pagevec. Note that only
581 * the local pagevec is examined as a !PageLRU page could be in the
582 * process of being released, reclaimed, migrated or on a remote
583 * pagevec that is currently being drained. Furthermore, marking
584 * a remote pagevec's page PageActive potentially hits a race where
585 * a page is marked PageActive just after it is added to the inactive
586 * list causing accounting errors and BUG_ON checks to trigger.
588 for (i = pagevec_count(pvec) - 1; i >= 0; i--) {
589 struct page *pagevec_page = pvec->pages[i];
591 if (pagevec_page == page) {
597 put_locked_var(swapvec_lock, lru_add_pvec);
601 * Mark a page as having seen activity.
603 * inactive,unreferenced -> inactive,referenced
604 * inactive,referenced -> active,unreferenced
605 * active,unreferenced -> active,referenced
607 * When a newly allocated page is not yet visible, so safe for non-atomic ops,
608 * __SetPageReferenced(page) may be substituted for mark_page_accessed(page).
610 void mark_page_accessed(struct page *page)
612 if (!PageActive(page) && !PageUnevictable(page) &&
613 PageReferenced(page)) {
616 * If the page is on the LRU, queue it for activation via
617 * activate_page_pvecs. Otherwise, assume the page is on a
618 * pagevec, mark it active and it'll be moved to the active
619 * LRU on the next drain.
624 __lru_cache_activate_page(page);
625 ClearPageReferenced(page);
626 if (page_is_file_cache(page))
627 workingset_activation(page);
628 } else if (!PageReferenced(page)) {
629 SetPageReferenced(page);
632 EXPORT_SYMBOL(mark_page_accessed);
634 static void __lru_cache_add(struct page *page)
636 struct pagevec *pvec = &get_locked_var(swapvec_lock, lru_add_pvec);
638 page_cache_get(page);
639 if (!pagevec_space(pvec))
640 __pagevec_lru_add(pvec);
641 pagevec_add(pvec, page);
642 put_locked_var(swapvec_lock, lru_add_pvec);
646 * lru_cache_add: add a page to the page lists
647 * @page: the page to add
649 void lru_cache_add_anon(struct page *page)
651 if (PageActive(page))
652 ClearPageActive(page);
653 __lru_cache_add(page);
656 void lru_cache_add_file(struct page *page)
658 if (PageActive(page))
659 ClearPageActive(page);
660 __lru_cache_add(page);
662 EXPORT_SYMBOL(lru_cache_add_file);
665 * lru_cache_add - add a page to a page list
666 * @page: the page to be added to the LRU.
668 * Queue the page for addition to the LRU via pagevec. The decision on whether
669 * to add the page to the [in]active [file|anon] list is deferred until the
670 * pagevec is drained. This gives a chance for the caller of lru_cache_add()
671 * have the page added to the active list using mark_page_accessed().
673 void lru_cache_add(struct page *page)
675 VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page);
676 VM_BUG_ON_PAGE(PageLRU(page), page);
677 __lru_cache_add(page);
681 * add_page_to_unevictable_list - add a page to the unevictable list
682 * @page: the page to be added to the unevictable list
684 * Add page directly to its zone's unevictable list. To avoid races with
685 * tasks that might be making the page evictable, through eg. munlock,
686 * munmap or exit, while it's not on the lru, we want to add the page
687 * while it's locked or otherwise "invisible" to other tasks. This is
688 * difficult to do when using the pagevec cache, so bypass that.
690 void add_page_to_unevictable_list(struct page *page)
692 struct zone *zone = page_zone(page);
693 struct lruvec *lruvec;
695 spin_lock_irq(&zone->lru_lock);
696 lruvec = mem_cgroup_page_lruvec(page, zone);
697 ClearPageActive(page);
698 SetPageUnevictable(page);
700 add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE);
701 spin_unlock_irq(&zone->lru_lock);
705 * lru_cache_add_active_or_unevictable
706 * @page: the page to be added to LRU
707 * @vma: vma in which page is mapped for determining reclaimability
709 * Place @page on the active or unevictable LRU list, depending on its
710 * evictability. Note that if the page is not evictable, it goes
711 * directly back onto it's zone's unevictable list, it does NOT use a
714 void lru_cache_add_active_or_unevictable(struct page *page,
715 struct vm_area_struct *vma)
717 VM_BUG_ON_PAGE(PageLRU(page), page);
719 if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED)) {
725 if (!TestSetPageMlocked(page)) {
727 * We use the irq-unsafe __mod_zone_page_stat because this
728 * counter is not modified from interrupt context, and the pte
729 * lock is held(spinlock), which implies preemption disabled.
731 __mod_zone_page_state(page_zone(page), NR_MLOCK,
732 hpage_nr_pages(page));
733 count_vm_event(UNEVICTABLE_PGMLOCKED);
735 add_page_to_unevictable_list(page);
739 * If the page can not be invalidated, it is moved to the
740 * inactive list to speed up its reclaim. It is moved to the
741 * head of the list, rather than the tail, to give the flusher
742 * threads some time to write it out, as this is much more
743 * effective than the single-page writeout from reclaim.
745 * If the page isn't page_mapped and dirty/writeback, the page
746 * could reclaim asap using PG_reclaim.
748 * 1. active, mapped page -> none
749 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
750 * 3. inactive, mapped page -> none
751 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
752 * 5. inactive, clean -> inactive, tail
755 * In 4, why it moves inactive's head, the VM expects the page would
756 * be write it out by flusher threads as this is much more effective
757 * than the single-page writeout from reclaim.
759 static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec,
768 if (PageUnevictable(page))
771 /* Some processes are using the page */
772 if (page_mapped(page))
775 active = PageActive(page);
776 file = page_is_file_cache(page);
777 lru = page_lru_base_type(page);
779 del_page_from_lru_list(page, lruvec, lru + active);
780 ClearPageActive(page);
781 ClearPageReferenced(page);
782 add_page_to_lru_list(page, lruvec, lru);
784 if (PageWriteback(page) || PageDirty(page)) {
786 * PG_reclaim could be raced with end_page_writeback
787 * It can make readahead confusing. But race window
788 * is _really_ small and it's non-critical problem.
790 SetPageReclaim(page);
793 * The page's writeback ends up during pagevec
794 * We moves tha page into tail of inactive.
796 list_move_tail(&page->lru, &lruvec->lists[lru]);
797 __count_vm_event(PGROTATED);
801 __count_vm_event(PGDEACTIVATE);
802 update_page_reclaim_stat(lruvec, file, 0);
806 * Drain pages out of the cpu's pagevecs.
807 * Either "cpu" is the current CPU, and preemption has already been
808 * disabled; or "cpu" is being hot-unplugged, and is already dead.
810 void lru_add_drain_cpu(int cpu)
812 struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu);
814 if (pagevec_count(pvec))
815 __pagevec_lru_add(pvec);
817 pvec = &per_cpu(lru_rotate_pvecs, cpu);
818 if (pagevec_count(pvec)) {
821 /* No harm done if a racing interrupt already did this */
822 local_lock_irqsave(rotate_lock, flags);
823 pagevec_move_tail(pvec);
824 local_unlock_irqrestore(rotate_lock, flags);
827 pvec = &per_cpu(lru_deactivate_file_pvecs, cpu);
828 if (pagevec_count(pvec))
829 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
831 activate_page_drain(cpu);
835 * deactivate_file_page - forcefully deactivate a file page
836 * @page: page to deactivate
838 * This function hints the VM that @page is a good reclaim candidate,
839 * for example if its invalidation fails due to the page being dirty
840 * or under writeback.
842 void deactivate_file_page(struct page *page)
845 * In a workload with many unevictable page such as mprotect,
846 * unevictable page deactivation for accelerating reclaim is pointless.
848 if (PageUnevictable(page))
851 if (likely(get_page_unless_zero(page))) {
852 struct pagevec *pvec = &get_locked_var(swapvec_lock,
853 lru_deactivate_file_pvecs);
855 if (!pagevec_add(pvec, page))
856 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
857 put_locked_var(swapvec_lock, lru_deactivate_file_pvecs);
861 void lru_add_drain(void)
863 lru_add_drain_cpu(local_lock_cpu(swapvec_lock));
864 local_unlock_cpu(swapvec_lock);
867 static void lru_add_drain_per_cpu(struct work_struct *dummy)
872 static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
874 void lru_add_drain_all(void)
876 static DEFINE_MUTEX(lock);
877 static struct cpumask has_work;
882 cpumask_clear(&has_work);
884 for_each_online_cpu(cpu) {
885 struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
887 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
888 pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
889 pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) ||
890 need_activate_page_drain(cpu)) {
891 INIT_WORK(work, lru_add_drain_per_cpu);
892 schedule_work_on(cpu, work);
893 cpumask_set_cpu(cpu, &has_work);
897 for_each_cpu(cpu, &has_work)
898 flush_work(&per_cpu(lru_add_drain_work, cpu));
905 * release_pages - batched page_cache_release()
906 * @pages: array of pages to release
907 * @nr: number of pages
908 * @cold: whether the pages are cache cold
910 * Decrement the reference count on all the pages in @pages. If it
911 * fell to zero, remove the page from the LRU and free it.
913 void release_pages(struct page **pages, int nr, bool cold)
916 LIST_HEAD(pages_to_free);
917 struct zone *zone = NULL;
918 struct lruvec *lruvec;
919 unsigned long uninitialized_var(flags);
920 unsigned int uninitialized_var(lock_batch);
922 for (i = 0; i < nr; i++) {
923 struct page *page = pages[i];
925 if (unlikely(PageCompound(page))) {
927 spin_unlock_irqrestore(&zone->lru_lock, flags);
930 put_compound_page(page);
935 * Make sure the IRQ-safe lock-holding time does not get
936 * excessive with a continuous string of pages from the
937 * same zone. The lock is held only if zone != NULL.
939 if (zone && ++lock_batch == SWAP_CLUSTER_MAX) {
940 spin_unlock_irqrestore(&zone->lru_lock, flags);
944 if (!put_page_testzero(page))
948 struct zone *pagezone = page_zone(page);
950 if (pagezone != zone) {
952 spin_unlock_irqrestore(&zone->lru_lock,
956 spin_lock_irqsave(&zone->lru_lock, flags);
959 lruvec = mem_cgroup_page_lruvec(page, zone);
960 VM_BUG_ON_PAGE(!PageLRU(page), page);
961 __ClearPageLRU(page);
962 del_page_from_lru_list(page, lruvec, page_off_lru(page));
965 /* Clear Active bit in case of parallel mark_page_accessed */
966 __ClearPageActive(page);
968 list_add(&page->lru, &pages_to_free);
971 spin_unlock_irqrestore(&zone->lru_lock, flags);
973 mem_cgroup_uncharge_list(&pages_to_free);
974 free_hot_cold_page_list(&pages_to_free, cold);
976 EXPORT_SYMBOL(release_pages);
979 * The pages which we're about to release may be in the deferred lru-addition
980 * queues. That would prevent them from really being freed right now. That's
981 * OK from a correctness point of view but is inefficient - those pages may be
982 * cache-warm and we want to give them back to the page allocator ASAP.
984 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
985 * and __pagevec_lru_add_active() call release_pages() directly to avoid
988 void __pagevec_release(struct pagevec *pvec)
991 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
992 pagevec_reinit(pvec);
994 EXPORT_SYMBOL(__pagevec_release);
996 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
997 /* used by __split_huge_page_refcount() */
998 void lru_add_page_tail(struct page *page, struct page *page_tail,
999 struct lruvec *lruvec, struct list_head *list)
1003 VM_BUG_ON_PAGE(!PageHead(page), page);
1004 VM_BUG_ON_PAGE(PageCompound(page_tail), page);
1005 VM_BUG_ON_PAGE(PageLRU(page_tail), page);
1006 VM_BUG_ON(NR_CPUS != 1 &&
1007 !spin_is_locked(&lruvec_zone(lruvec)->lru_lock));
1010 SetPageLRU(page_tail);
1012 if (likely(PageLRU(page)))
1013 list_add_tail(&page_tail->lru, &page->lru);
1015 /* page reclaim is reclaiming a huge page */
1016 get_page(page_tail);
1017 list_add_tail(&page_tail->lru, list);
1019 struct list_head *list_head;
1021 * Head page has not yet been counted, as an hpage,
1022 * so we must account for each subpage individually.
1024 * Use the standard add function to put page_tail on the list,
1025 * but then correct its position so they all end up in order.
1027 add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail));
1028 list_head = page_tail->lru.prev;
1029 list_move_tail(&page_tail->lru, list_head);
1032 if (!PageUnevictable(page))
1033 update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
1035 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1037 static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
1040 int file = page_is_file_cache(page);
1041 int active = PageActive(page);
1042 enum lru_list lru = page_lru(page);
1044 VM_BUG_ON_PAGE(PageLRU(page), page);
1047 add_page_to_lru_list(page, lruvec, lru);
1048 update_page_reclaim_stat(lruvec, file, active);
1049 trace_mm_lru_insertion(page, lru);
1053 * Add the passed pages to the LRU, then drop the caller's refcount
1054 * on them. Reinitialises the caller's pagevec.
1056 void __pagevec_lru_add(struct pagevec *pvec)
1058 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
1060 EXPORT_SYMBOL(__pagevec_lru_add);
1063 * pagevec_lookup_entries - gang pagecache lookup
1064 * @pvec: Where the resulting entries are placed
1065 * @mapping: The address_space to search
1066 * @start: The starting entry index
1067 * @nr_entries: The maximum number of entries
1068 * @indices: The cache indices corresponding to the entries in @pvec
1070 * pagevec_lookup_entries() will search for and return a group of up
1071 * to @nr_entries pages and shadow entries in the mapping. All
1072 * entries are placed in @pvec. pagevec_lookup_entries() takes a
1073 * reference against actual pages in @pvec.
1075 * The search returns a group of mapping-contiguous entries with
1076 * ascending indexes. There may be holes in the indices due to
1077 * not-present entries.
1079 * pagevec_lookup_entries() returns the number of entries which were
1082 unsigned pagevec_lookup_entries(struct pagevec *pvec,
1083 struct address_space *mapping,
1084 pgoff_t start, unsigned nr_pages,
1087 pvec->nr = find_get_entries(mapping, start, nr_pages,
1088 pvec->pages, indices);
1089 return pagevec_count(pvec);
1093 * pagevec_remove_exceptionals - pagevec exceptionals pruning
1094 * @pvec: The pagevec to prune
1096 * pagevec_lookup_entries() fills both pages and exceptional radix
1097 * tree entries into the pagevec. This function prunes all
1098 * exceptionals from @pvec without leaving holes, so that it can be
1099 * passed on to page-only pagevec operations.
1101 void pagevec_remove_exceptionals(struct pagevec *pvec)
1105 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
1106 struct page *page = pvec->pages[i];
1107 if (!radix_tree_exceptional_entry(page))
1108 pvec->pages[j++] = page;
1114 * pagevec_lookup - gang pagecache lookup
1115 * @pvec: Where the resulting pages are placed
1116 * @mapping: The address_space to search
1117 * @start: The starting page index
1118 * @nr_pages: The maximum number of pages
1120 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
1121 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
1122 * reference against the pages in @pvec.
1124 * The search returns a group of mapping-contiguous pages with ascending
1125 * indexes. There may be holes in the indices due to not-present pages.
1127 * pagevec_lookup() returns the number of pages which were found.
1129 unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
1130 pgoff_t start, unsigned nr_pages)
1132 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
1133 return pagevec_count(pvec);
1135 EXPORT_SYMBOL(pagevec_lookup);
1137 unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
1138 pgoff_t *index, int tag, unsigned nr_pages)
1140 pvec->nr = find_get_pages_tag(mapping, index, tag,
1141 nr_pages, pvec->pages);
1142 return pagevec_count(pvec);
1144 EXPORT_SYMBOL(pagevec_lookup_tag);
1147 * Perform any setup for the swap system
1149 void __init swap_setup(void)
1151 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
1155 for (i = 0; i < MAX_SWAPFILES; i++)
1156 spin_lock_init(&swapper_spaces[i].tree_lock);
1159 /* Use a smaller cluster for small-memory machines */
1165 * Right now other parts of the system means that we
1166 * _really_ don't want to cluster much more