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/locallock.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);
50 static DEFINE_LOCAL_IRQ_LOCK(rotate_lock);
51 DEFINE_LOCAL_IRQ_LOCK(swapvec_lock);
54 * This path almost never happens for VM activity - pages are normally
55 * freed via pagevecs. But it gets used by networking.
57 static void __page_cache_release(struct page *page)
60 struct zone *zone = page_zone(page);
61 struct lruvec *lruvec;
64 spin_lock_irqsave(&zone->lru_lock, flags);
65 lruvec = mem_cgroup_page_lruvec(page, zone);
66 VM_BUG_ON_PAGE(!PageLRU(page), page);
68 del_page_from_lru_list(page, lruvec, page_off_lru(page));
69 spin_unlock_irqrestore(&zone->lru_lock, flags);
71 mem_cgroup_uncharge(page);
74 static void __put_single_page(struct page *page)
76 __page_cache_release(page);
77 free_hot_cold_page(page, false);
80 static void __put_compound_page(struct page *page)
82 compound_page_dtor *dtor;
85 * __page_cache_release() is supposed to be called for thp, not for
86 * hugetlb. This is because hugetlb page does never have PageLRU set
87 * (it's never listed to any LRU lists) and no memcg routines should
88 * be called for hugetlb (it has a separate hugetlb_cgroup.)
91 __page_cache_release(page);
92 dtor = get_compound_page_dtor(page);
97 * Two special cases here: we could avoid taking compound_lock_irqsave
98 * and could skip the tail refcounting(in _mapcount).
102 * PageHeadHuge will remain true until the compound page
103 * is released and enters the buddy allocator, and it could
104 * not be split by __split_huge_page_refcount().
106 * So if we see PageHeadHuge set, and we have the tail page pin,
107 * then we could safely put head page.
111 * PG_slab is cleared before the slab frees the head page, and
112 * tail pin cannot be the last reference left on the head page,
113 * because the slab code is free to reuse the compound page
114 * after a kfree/kmem_cache_free without having to check if
115 * there's any tail pin left. In turn all tail pinsmust be always
116 * released while the head is still pinned by the slab code
117 * and so we know PG_slab will be still set too.
119 * So if we see PageSlab set, and we have the tail page pin,
120 * then we could safely put head page.
122 static __always_inline
123 void put_unrefcounted_compound_page(struct page *page_head, struct page *page)
126 * If @page is a THP tail, we must read the tail page
127 * flags after the head page flags. The
128 * __split_huge_page_refcount side enforces write memory barriers
129 * between clearing PageTail and before the head page
130 * can be freed and reallocated.
133 if (likely(PageTail(page))) {
135 * __split_huge_page_refcount cannot race
136 * here, see the comment above this function.
138 VM_BUG_ON_PAGE(!PageHead(page_head), page_head);
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 != compound_head(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(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);
631 if (page_is_idle(page))
632 clear_page_idle(page);
634 EXPORT_SYMBOL(mark_page_accessed);
636 static void __lru_cache_add(struct page *page)
638 struct pagevec *pvec = &get_locked_var(swapvec_lock, lru_add_pvec);
640 page_cache_get(page);
641 if (!pagevec_space(pvec))
642 __pagevec_lru_add(pvec);
643 pagevec_add(pvec, page);
644 put_locked_var(swapvec_lock, lru_add_pvec);
648 * lru_cache_add: add a page to the page lists
649 * @page: the page to add
651 void lru_cache_add_anon(struct page *page)
653 if (PageActive(page))
654 ClearPageActive(page);
655 __lru_cache_add(page);
658 void lru_cache_add_file(struct page *page)
660 if (PageActive(page))
661 ClearPageActive(page);
662 __lru_cache_add(page);
664 EXPORT_SYMBOL(lru_cache_add_file);
667 * lru_cache_add - add a page to a page list
668 * @page: the page to be added to the LRU.
670 * Queue the page for addition to the LRU via pagevec. The decision on whether
671 * to add the page to the [in]active [file|anon] list is deferred until the
672 * pagevec is drained. This gives a chance for the caller of lru_cache_add()
673 * have the page added to the active list using mark_page_accessed().
675 void lru_cache_add(struct page *page)
677 VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page);
678 VM_BUG_ON_PAGE(PageLRU(page), page);
679 __lru_cache_add(page);
683 * add_page_to_unevictable_list - add a page to the unevictable list
684 * @page: the page to be added to the unevictable list
686 * Add page directly to its zone's unevictable list. To avoid races with
687 * tasks that might be making the page evictable, through eg. munlock,
688 * munmap or exit, while it's not on the lru, we want to add the page
689 * while it's locked or otherwise "invisible" to other tasks. This is
690 * difficult to do when using the pagevec cache, so bypass that.
692 void add_page_to_unevictable_list(struct page *page)
694 struct zone *zone = page_zone(page);
695 struct lruvec *lruvec;
697 spin_lock_irq(&zone->lru_lock);
698 lruvec = mem_cgroup_page_lruvec(page, zone);
699 ClearPageActive(page);
700 SetPageUnevictable(page);
702 add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE);
703 spin_unlock_irq(&zone->lru_lock);
707 * lru_cache_add_active_or_unevictable
708 * @page: the page to be added to LRU
709 * @vma: vma in which page is mapped for determining reclaimability
711 * Place @page on the active or unevictable LRU list, depending on its
712 * evictability. Note that if the page is not evictable, it goes
713 * directly back onto it's zone's unevictable list, it does NOT use a
716 void lru_cache_add_active_or_unevictable(struct page *page,
717 struct vm_area_struct *vma)
719 VM_BUG_ON_PAGE(PageLRU(page), page);
721 if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED)) {
727 if (!TestSetPageMlocked(page)) {
729 * We use the irq-unsafe __mod_zone_page_stat because this
730 * counter is not modified from interrupt context, and the pte
731 * lock is held(spinlock), which implies preemption disabled.
733 __mod_zone_page_state(page_zone(page), NR_MLOCK,
734 hpage_nr_pages(page));
735 count_vm_event(UNEVICTABLE_PGMLOCKED);
737 add_page_to_unevictable_list(page);
741 * If the page can not be invalidated, it is moved to the
742 * inactive list to speed up its reclaim. It is moved to the
743 * head of the list, rather than the tail, to give the flusher
744 * threads some time to write it out, as this is much more
745 * effective than the single-page writeout from reclaim.
747 * If the page isn't page_mapped and dirty/writeback, the page
748 * could reclaim asap using PG_reclaim.
750 * 1. active, mapped page -> none
751 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
752 * 3. inactive, mapped page -> none
753 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
754 * 5. inactive, clean -> inactive, tail
757 * In 4, why it moves inactive's head, the VM expects the page would
758 * be write it out by flusher threads as this is much more effective
759 * than the single-page writeout from reclaim.
761 static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec,
770 if (PageUnevictable(page))
773 /* Some processes are using the page */
774 if (page_mapped(page))
777 active = PageActive(page);
778 file = page_is_file_cache(page);
779 lru = page_lru_base_type(page);
781 del_page_from_lru_list(page, lruvec, lru + active);
782 ClearPageActive(page);
783 ClearPageReferenced(page);
784 add_page_to_lru_list(page, lruvec, lru);
786 if (PageWriteback(page) || PageDirty(page)) {
788 * PG_reclaim could be raced with end_page_writeback
789 * It can make readahead confusing. But race window
790 * is _really_ small and it's non-critical problem.
792 SetPageReclaim(page);
795 * The page's writeback ends up during pagevec
796 * We moves tha page into tail of inactive.
798 list_move_tail(&page->lru, &lruvec->lists[lru]);
799 __count_vm_event(PGROTATED);
803 __count_vm_event(PGDEACTIVATE);
804 update_page_reclaim_stat(lruvec, file, 0);
808 * Drain pages out of the cpu's pagevecs.
809 * Either "cpu" is the current CPU, and preemption has already been
810 * disabled; or "cpu" is being hot-unplugged, and is already dead.
812 void lru_add_drain_cpu(int cpu)
814 struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu);
816 if (pagevec_count(pvec))
817 __pagevec_lru_add(pvec);
819 pvec = &per_cpu(lru_rotate_pvecs, cpu);
820 if (pagevec_count(pvec)) {
823 /* No harm done if a racing interrupt already did this */
824 #ifdef CONFIG_PREEMPT_RT_BASE
825 local_lock_irqsave_on(rotate_lock, flags, cpu);
826 pagevec_move_tail(pvec);
827 local_unlock_irqrestore_on(rotate_lock, flags, cpu);
829 local_lock_irqsave(rotate_lock, flags);
830 pagevec_move_tail(pvec);
831 local_unlock_irqrestore(rotate_lock, flags);
835 pvec = &per_cpu(lru_deactivate_file_pvecs, cpu);
836 if (pagevec_count(pvec))
837 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
839 activate_page_drain(cpu);
843 * deactivate_file_page - forcefully deactivate a file page
844 * @page: page to deactivate
846 * This function hints the VM that @page is a good reclaim candidate,
847 * for example if its invalidation fails due to the page being dirty
848 * or under writeback.
850 void deactivate_file_page(struct page *page)
853 * In a workload with many unevictable page such as mprotect,
854 * unevictable page deactivation for accelerating reclaim is pointless.
856 if (PageUnevictable(page))
859 if (likely(get_page_unless_zero(page))) {
860 struct pagevec *pvec = &get_locked_var(swapvec_lock,
861 lru_deactivate_file_pvecs);
863 if (!pagevec_add(pvec, page))
864 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
865 put_locked_var(swapvec_lock, lru_deactivate_file_pvecs);
869 void lru_add_drain(void)
871 lru_add_drain_cpu(local_lock_cpu(swapvec_lock));
872 local_unlock_cpu(swapvec_lock);
876 #ifdef CONFIG_PREEMPT_RT_BASE
877 static inline void remote_lru_add_drain(int cpu, struct cpumask *has_work)
879 local_lock_on(swapvec_lock, cpu);
880 lru_add_drain_cpu(cpu);
881 local_unlock_on(swapvec_lock, cpu);
886 static void lru_add_drain_per_cpu(struct work_struct *dummy)
891 static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
892 static inline void remote_lru_add_drain(int cpu, struct cpumask *has_work)
894 struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
896 INIT_WORK(work, lru_add_drain_per_cpu);
897 schedule_work_on(cpu, work);
898 cpumask_set_cpu(cpu, has_work);
902 void lru_add_drain_all(void)
904 static DEFINE_MUTEX(lock);
905 static struct cpumask has_work;
910 cpumask_clear(&has_work);
912 for_each_online_cpu(cpu) {
913 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
914 pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
915 pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) ||
916 need_activate_page_drain(cpu))
917 remote_lru_add_drain(cpu, &has_work);
920 #ifndef CONFIG_PREEMPT_RT_BASE
921 for_each_cpu(cpu, &has_work)
922 flush_work(&per_cpu(lru_add_drain_work, cpu));
930 * release_pages - batched page_cache_release()
931 * @pages: array of pages to release
932 * @nr: number of pages
933 * @cold: whether the pages are cache cold
935 * Decrement the reference count on all the pages in @pages. If it
936 * fell to zero, remove the page from the LRU and free it.
938 void release_pages(struct page **pages, int nr, bool cold)
941 LIST_HEAD(pages_to_free);
942 struct zone *zone = NULL;
943 struct lruvec *lruvec;
944 unsigned long uninitialized_var(flags);
945 unsigned int uninitialized_var(lock_batch);
947 for (i = 0; i < nr; i++) {
948 struct page *page = pages[i];
950 if (unlikely(PageCompound(page))) {
952 spin_unlock_irqrestore(&zone->lru_lock, flags);
955 put_compound_page(page);
960 * Make sure the IRQ-safe lock-holding time does not get
961 * excessive with a continuous string of pages from the
962 * same zone. The lock is held only if zone != NULL.
964 if (zone && ++lock_batch == SWAP_CLUSTER_MAX) {
965 spin_unlock_irqrestore(&zone->lru_lock, flags);
969 if (!put_page_testzero(page))
973 struct zone *pagezone = page_zone(page);
975 if (pagezone != zone) {
977 spin_unlock_irqrestore(&zone->lru_lock,
981 spin_lock_irqsave(&zone->lru_lock, flags);
984 lruvec = mem_cgroup_page_lruvec(page, zone);
985 VM_BUG_ON_PAGE(!PageLRU(page), page);
986 __ClearPageLRU(page);
987 del_page_from_lru_list(page, lruvec, page_off_lru(page));
990 /* Clear Active bit in case of parallel mark_page_accessed */
991 __ClearPageActive(page);
993 list_add(&page->lru, &pages_to_free);
996 spin_unlock_irqrestore(&zone->lru_lock, flags);
998 mem_cgroup_uncharge_list(&pages_to_free);
999 free_hot_cold_page_list(&pages_to_free, cold);
1001 EXPORT_SYMBOL(release_pages);
1004 * The pages which we're about to release may be in the deferred lru-addition
1005 * queues. That would prevent them from really being freed right now. That's
1006 * OK from a correctness point of view but is inefficient - those pages may be
1007 * cache-warm and we want to give them back to the page allocator ASAP.
1009 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
1010 * and __pagevec_lru_add_active() call release_pages() directly to avoid
1013 void __pagevec_release(struct pagevec *pvec)
1016 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
1017 pagevec_reinit(pvec);
1019 EXPORT_SYMBOL(__pagevec_release);
1021 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1022 /* used by __split_huge_page_refcount() */
1023 void lru_add_page_tail(struct page *page, struct page *page_tail,
1024 struct lruvec *lruvec, struct list_head *list)
1028 VM_BUG_ON_PAGE(!PageHead(page), page);
1029 VM_BUG_ON_PAGE(PageCompound(page_tail), page);
1030 VM_BUG_ON_PAGE(PageLRU(page_tail), page);
1031 VM_BUG_ON(NR_CPUS != 1 &&
1032 !spin_is_locked(&lruvec_zone(lruvec)->lru_lock));
1035 SetPageLRU(page_tail);
1037 if (likely(PageLRU(page)))
1038 list_add_tail(&page_tail->lru, &page->lru);
1040 /* page reclaim is reclaiming a huge page */
1041 get_page(page_tail);
1042 list_add_tail(&page_tail->lru, list);
1044 struct list_head *list_head;
1046 * Head page has not yet been counted, as an hpage,
1047 * so we must account for each subpage individually.
1049 * Use the standard add function to put page_tail on the list,
1050 * but then correct its position so they all end up in order.
1052 add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail));
1053 list_head = page_tail->lru.prev;
1054 list_move_tail(&page_tail->lru, list_head);
1057 if (!PageUnevictable(page))
1058 update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
1060 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1062 static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
1065 int file = page_is_file_cache(page);
1066 int active = PageActive(page);
1067 enum lru_list lru = page_lru(page);
1069 VM_BUG_ON_PAGE(PageLRU(page), page);
1072 add_page_to_lru_list(page, lruvec, lru);
1073 update_page_reclaim_stat(lruvec, file, active);
1074 trace_mm_lru_insertion(page, lru);
1078 * Add the passed pages to the LRU, then drop the caller's refcount
1079 * on them. Reinitialises the caller's pagevec.
1081 void __pagevec_lru_add(struct pagevec *pvec)
1083 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
1085 EXPORT_SYMBOL(__pagevec_lru_add);
1088 * pagevec_lookup_entries - gang pagecache lookup
1089 * @pvec: Where the resulting entries are placed
1090 * @mapping: The address_space to search
1091 * @start: The starting entry index
1092 * @nr_entries: The maximum number of entries
1093 * @indices: The cache indices corresponding to the entries in @pvec
1095 * pagevec_lookup_entries() will search for and return a group of up
1096 * to @nr_entries pages and shadow entries in the mapping. All
1097 * entries are placed in @pvec. pagevec_lookup_entries() takes a
1098 * reference against actual pages in @pvec.
1100 * The search returns a group of mapping-contiguous entries with
1101 * ascending indexes. There may be holes in the indices due to
1102 * not-present entries.
1104 * pagevec_lookup_entries() returns the number of entries which were
1107 unsigned pagevec_lookup_entries(struct pagevec *pvec,
1108 struct address_space *mapping,
1109 pgoff_t start, unsigned nr_pages,
1112 pvec->nr = find_get_entries(mapping, start, nr_pages,
1113 pvec->pages, indices);
1114 return pagevec_count(pvec);
1118 * pagevec_remove_exceptionals - pagevec exceptionals pruning
1119 * @pvec: The pagevec to prune
1121 * pagevec_lookup_entries() fills both pages and exceptional radix
1122 * tree entries into the pagevec. This function prunes all
1123 * exceptionals from @pvec without leaving holes, so that it can be
1124 * passed on to page-only pagevec operations.
1126 void pagevec_remove_exceptionals(struct pagevec *pvec)
1130 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
1131 struct page *page = pvec->pages[i];
1132 if (!radix_tree_exceptional_entry(page))
1133 pvec->pages[j++] = page;
1139 * pagevec_lookup - gang pagecache lookup
1140 * @pvec: Where the resulting pages are placed
1141 * @mapping: The address_space to search
1142 * @start: The starting page index
1143 * @nr_pages: The maximum number of pages
1145 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
1146 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
1147 * reference against the pages in @pvec.
1149 * The search returns a group of mapping-contiguous pages with ascending
1150 * indexes. There may be holes in the indices due to not-present pages.
1152 * pagevec_lookup() returns the number of pages which were found.
1154 unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
1155 pgoff_t start, unsigned nr_pages)
1157 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
1158 return pagevec_count(pvec);
1160 EXPORT_SYMBOL(pagevec_lookup);
1162 unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
1163 pgoff_t *index, int tag, unsigned nr_pages)
1165 pvec->nr = find_get_pages_tag(mapping, index, tag,
1166 nr_pages, pvec->pages);
1167 return pagevec_count(pvec);
1169 EXPORT_SYMBOL(pagevec_lookup_tag);
1172 * Perform any setup for the swap system
1174 void __init swap_setup(void)
1176 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
1180 for (i = 0; i < MAX_SWAPFILES; i++)
1181 spin_lock_init(&swapper_spaces[i].tree_lock);
1184 /* Use a smaller cluster for small-memory machines */
1190 * Right now other parts of the system means that we
1191 * _really_ don't want to cluster much more