1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
9 * This file contains the default values for the operation of the
10 * Linux VM subsystem. Fine-tuning documentation can be found in
11 * Documentation/admin-guide/sysctl/vm.rst.
13 * Swap aging added 23.2.95, Stephen Tweedie.
14 * Buffermem limits added 12.3.98, Rik van Riel.
18 #include <linux/sched.h>
19 #include <linux/kernel_stat.h>
20 #include <linux/swap.h>
21 #include <linux/mman.h>
22 #include <linux/pagemap.h>
23 #include <linux/pagevec.h>
24 #include <linux/init.h>
25 #include <linux/export.h>
26 #include <linux/mm_inline.h>
27 #include <linux/percpu_counter.h>
28 #include <linux/memremap.h>
29 #include <linux/percpu.h>
30 #include <linux/cpu.h>
31 #include <linux/notifier.h>
32 #include <linux/backing-dev.h>
33 #include <linux/memcontrol.h>
34 #include <linux/gfp.h>
35 #include <linux/uio.h>
36 #include <linux/hugetlb.h>
37 #include <linux/page_idle.h>
41 #define CREATE_TRACE_POINTS
42 #include <trace/events/pagemap.h>
44 /* How many pages do we try to swap or page in/out together? */
47 static DEFINE_PER_CPU(struct pagevec, lru_add_pvec);
48 static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
49 static DEFINE_PER_CPU(struct pagevec, lru_deactivate_file_pvecs);
50 static DEFINE_PER_CPU(struct pagevec, lru_deactivate_pvecs);
51 static DEFINE_PER_CPU(struct pagevec, lru_lazyfree_pvecs);
53 static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
57 * This path almost never happens for VM activity - pages are normally
58 * freed via pagevecs. But it gets used by networking.
60 static void __page_cache_release(struct page *page)
63 pg_data_t *pgdat = page_pgdat(page);
64 struct lruvec *lruvec;
67 spin_lock_irqsave(&pgdat->lru_lock, flags);
68 lruvec = mem_cgroup_page_lruvec(page, pgdat);
69 VM_BUG_ON_PAGE(!PageLRU(page), page);
71 del_page_from_lru_list(page, lruvec, page_off_lru(page));
72 spin_unlock_irqrestore(&pgdat->lru_lock, flags);
74 __ClearPageWaiters(page);
77 static void __put_single_page(struct page *page)
79 __page_cache_release(page);
80 mem_cgroup_uncharge(page);
81 free_unref_page(page);
84 static void __put_compound_page(struct page *page)
86 compound_page_dtor *dtor;
89 * __page_cache_release() is supposed to be called for thp, not for
90 * hugetlb. This is because hugetlb page does never have PageLRU set
91 * (it's never listed to any LRU lists) and no memcg routines should
92 * be called for hugetlb (it has a separate hugetlb_cgroup.)
95 __page_cache_release(page);
96 dtor = get_compound_page_dtor(page);
100 void __put_page(struct page *page)
102 if (is_zone_device_page(page)) {
103 put_dev_pagemap(page->pgmap);
106 * The page belongs to the device that created pgmap. Do
107 * not return it to page allocator.
112 if (unlikely(PageCompound(page)))
113 __put_compound_page(page);
115 __put_single_page(page);
117 EXPORT_SYMBOL(__put_page);
120 * put_pages_list() - release a list of pages
121 * @pages: list of pages threaded on page->lru
123 * Release a list of pages which are strung together on page.lru. Currently
124 * used by read_cache_pages() and related error recovery code.
126 void put_pages_list(struct list_head *pages)
128 while (!list_empty(pages)) {
131 victim = lru_to_page(pages);
132 list_del(&victim->lru);
136 EXPORT_SYMBOL(put_pages_list);
139 * get_kernel_pages() - pin kernel pages in memory
140 * @kiov: An array of struct kvec structures
141 * @nr_segs: number of segments to pin
142 * @write: pinning for read/write, currently ignored
143 * @pages: array that receives pointers to the pages pinned.
144 * Should be at least nr_segs long.
146 * Returns number of pages pinned. This may be fewer than the number
147 * requested. If nr_pages is 0 or negative, returns 0. If no pages
148 * were pinned, returns -errno. Each page returned must be released
149 * with a put_page() call when it is finished with.
151 int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write,
156 for (seg = 0; seg < nr_segs; seg++) {
157 if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE))
160 pages[seg] = kmap_to_page(kiov[seg].iov_base);
161 get_page(pages[seg]);
166 EXPORT_SYMBOL_GPL(get_kernel_pages);
169 * get_kernel_page() - pin a kernel page in memory
170 * @start: starting kernel address
171 * @write: pinning for read/write, currently ignored
172 * @pages: array that receives pointer to the page pinned.
173 * Must be at least nr_segs long.
175 * Returns 1 if page is pinned. If the page was not pinned, returns
176 * -errno. The page returned must be released with a put_page() call
177 * when it is finished with.
179 int get_kernel_page(unsigned long start, int write, struct page **pages)
181 const struct kvec kiov = {
182 .iov_base = (void *)start,
186 return get_kernel_pages(&kiov, 1, write, pages);
188 EXPORT_SYMBOL_GPL(get_kernel_page);
190 static void pagevec_lru_move_fn(struct pagevec *pvec,
191 void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg),
195 struct pglist_data *pgdat = NULL;
196 struct lruvec *lruvec;
197 unsigned long flags = 0;
199 for (i = 0; i < pagevec_count(pvec); i++) {
200 struct page *page = pvec->pages[i];
201 struct pglist_data *pagepgdat = page_pgdat(page);
203 if (pagepgdat != pgdat) {
205 spin_unlock_irqrestore(&pgdat->lru_lock, flags);
207 spin_lock_irqsave(&pgdat->lru_lock, flags);
210 lruvec = mem_cgroup_page_lruvec(page, pgdat);
211 (*move_fn)(page, lruvec, arg);
214 spin_unlock_irqrestore(&pgdat->lru_lock, flags);
215 release_pages(pvec->pages, pvec->nr);
216 pagevec_reinit(pvec);
219 static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec,
224 if (PageLRU(page) && !PageUnevictable(page)) {
225 del_page_from_lru_list(page, lruvec, page_lru(page));
226 ClearPageActive(page);
227 add_page_to_lru_list_tail(page, lruvec, page_lru(page));
233 * pagevec_move_tail() must be called with IRQ disabled.
234 * Otherwise this may cause nasty races.
236 static void pagevec_move_tail(struct pagevec *pvec)
240 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
241 __count_vm_events(PGROTATED, pgmoved);
245 * Writeback is about to end against a page which has been marked for immediate
246 * reclaim. If it still appears to be reclaimable, move it to the tail of the
249 void rotate_reclaimable_page(struct page *page)
251 if (!PageLocked(page) && !PageDirty(page) &&
252 !PageUnevictable(page) && PageLRU(page)) {
253 struct pagevec *pvec;
257 local_irq_save(flags);
258 pvec = this_cpu_ptr(&lru_rotate_pvecs);
259 if (!pagevec_add(pvec, page) || PageCompound(page))
260 pagevec_move_tail(pvec);
261 local_irq_restore(flags);
265 static void update_page_reclaim_stat(struct lruvec *lruvec,
266 int file, int rotated)
268 struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
270 reclaim_stat->recent_scanned[file]++;
272 reclaim_stat->recent_rotated[file]++;
275 static void __activate_page(struct page *page, struct lruvec *lruvec,
278 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
279 int file = page_is_file_cache(page);
280 int lru = page_lru_base_type(page);
282 del_page_from_lru_list(page, lruvec, lru);
285 add_page_to_lru_list(page, lruvec, lru);
286 trace_mm_lru_activate(page);
288 __count_vm_event(PGACTIVATE);
289 update_page_reclaim_stat(lruvec, file, 1);
294 static void activate_page_drain(int cpu)
296 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
298 if (pagevec_count(pvec))
299 pagevec_lru_move_fn(pvec, __activate_page, NULL);
302 static bool need_activate_page_drain(int cpu)
304 return pagevec_count(&per_cpu(activate_page_pvecs, cpu)) != 0;
307 void activate_page(struct page *page)
309 page = compound_head(page);
310 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
311 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
314 if (!pagevec_add(pvec, page) || PageCompound(page))
315 pagevec_lru_move_fn(pvec, __activate_page, NULL);
316 put_cpu_var(activate_page_pvecs);
321 static inline void activate_page_drain(int cpu)
325 void activate_page(struct page *page)
327 pg_data_t *pgdat = page_pgdat(page);
329 page = compound_head(page);
330 spin_lock_irq(&pgdat->lru_lock);
331 __activate_page(page, mem_cgroup_page_lruvec(page, pgdat), NULL);
332 spin_unlock_irq(&pgdat->lru_lock);
336 static void __lru_cache_activate_page(struct page *page)
338 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
342 * Search backwards on the optimistic assumption that the page being
343 * activated has just been added to this pagevec. Note that only
344 * the local pagevec is examined as a !PageLRU page could be in the
345 * process of being released, reclaimed, migrated or on a remote
346 * pagevec that is currently being drained. Furthermore, marking
347 * a remote pagevec's page PageActive potentially hits a race where
348 * a page is marked PageActive just after it is added to the inactive
349 * list causing accounting errors and BUG_ON checks to trigger.
351 for (i = pagevec_count(pvec) - 1; i >= 0; i--) {
352 struct page *pagevec_page = pvec->pages[i];
354 if (pagevec_page == page) {
360 put_cpu_var(lru_add_pvec);
363 #ifdef CONFIG_LRU_GEN
364 static void page_inc_refs(struct page *page)
366 unsigned long new_flags, old_flags = READ_ONCE(page->flags);
368 if (PageUnevictable(page))
371 if (!PageReferenced(page)) {
372 SetPageReferenced(page);
376 if (!PageWorkingset(page)) {
377 SetPageWorkingset(page);
381 /* see the comment on MAX_NR_TIERS */
383 new_flags = old_flags & LRU_REFS_MASK;
384 if (new_flags == LRU_REFS_MASK)
387 new_flags += BIT(LRU_REFS_PGOFF);
388 new_flags |= old_flags & ~LRU_REFS_MASK;
389 } while (!try_cmpxchg(&page->flags, &old_flags, new_flags));
392 static void page_inc_refs(struct page *page)
395 #endif /* CONFIG_LRU_GEN */
398 * Mark a page as having seen activity.
400 * inactive,unreferenced -> inactive,referenced
401 * inactive,referenced -> active,unreferenced
402 * active,unreferenced -> active,referenced
404 * When a newly allocated page is not yet visible, so safe for non-atomic ops,
405 * __SetPageReferenced(page) may be substituted for mark_page_accessed(page).
407 void mark_page_accessed(struct page *page)
409 page = compound_head(page);
411 if (lru_gen_enabled()) {
416 if (!PageActive(page) && !PageUnevictable(page) &&
417 PageReferenced(page)) {
419 * If the page is on the LRU, queue it for activation via
420 * activate_page_pvecs. Otherwise, assume the page is on a
421 * pagevec, mark it active and it'll be moved to the active
422 * LRU on the next drain.
427 __lru_cache_activate_page(page);
428 ClearPageReferenced(page);
429 if (page_is_file_cache(page))
430 workingset_activation(page);
431 } else if (!PageReferenced(page)) {
432 SetPageReferenced(page);
434 if (page_is_idle(page))
435 clear_page_idle(page);
437 EXPORT_SYMBOL(mark_page_accessed);
439 static void __lru_cache_add(struct page *page)
441 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
444 if (!pagevec_add(pvec, page) || PageCompound(page))
445 __pagevec_lru_add(pvec);
446 put_cpu_var(lru_add_pvec);
450 * lru_cache_add_anon - add a page to the page lists
451 * @page: the page to add
453 void lru_cache_add_anon(struct page *page)
455 if (PageActive(page))
456 ClearPageActive(page);
457 __lru_cache_add(page);
460 void lru_cache_add_file(struct page *page)
462 if (PageActive(page))
463 ClearPageActive(page);
464 __lru_cache_add(page);
466 EXPORT_SYMBOL(lru_cache_add_file);
469 * lru_cache_add - add a page to a page list
470 * @page: the page to be added to the LRU.
472 * Queue the page for addition to the LRU via pagevec. The decision on whether
473 * to add the page to the [in]active [file|anon] list is deferred until the
474 * pagevec is drained. This gives a chance for the caller of lru_cache_add()
475 * have the page added to the active list using mark_page_accessed().
477 void lru_cache_add(struct page *page)
479 VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page);
480 VM_BUG_ON_PAGE(PageLRU(page), page);
481 __lru_cache_add(page);
482 /* see the comment in lru_gen_add_page() */
483 if (lru_gen_enabled() && !PageUnevictable(page) &&
484 lru_gen_in_fault() && !(current->flags & PF_MEMALLOC))
490 * lru_cache_add_active_or_unevictable
491 * @page: the page to be added to LRU
492 * @vma: vma in which page is mapped for determining reclaimability
494 * Place @page on the active or unevictable LRU list, depending on its
495 * evictability. Note that if the page is not evictable, it goes
496 * directly back onto it's zone's unevictable list, it does NOT use a
499 void lru_cache_add_active_or_unevictable(struct page *page,
500 struct vm_area_struct *vma)
502 VM_BUG_ON_PAGE(PageLRU(page), page);
504 if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED))
506 else if (!TestSetPageMlocked(page)) {
508 * We use the irq-unsafe __mod_zone_page_stat because this
509 * counter is not modified from interrupt context, and the pte
510 * lock is held(spinlock), which implies preemption disabled.
512 __mod_zone_page_state(page_zone(page), NR_MLOCK,
513 hpage_nr_pages(page));
514 count_vm_event(UNEVICTABLE_PGMLOCKED);
520 * If the page can not be invalidated, it is moved to the
521 * inactive list to speed up its reclaim. It is moved to the
522 * head of the list, rather than the tail, to give the flusher
523 * threads some time to write it out, as this is much more
524 * effective than the single-page writeout from reclaim.
526 * If the page isn't page_mapped and dirty/writeback, the page
527 * could reclaim asap using PG_reclaim.
529 * 1. active, mapped page -> none
530 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
531 * 3. inactive, mapped page -> none
532 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
533 * 5. inactive, clean -> inactive, tail
536 * In 4, why it moves inactive's head, the VM expects the page would
537 * be write it out by flusher threads as this is much more effective
538 * than the single-page writeout from reclaim.
540 static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec,
549 if (PageUnevictable(page))
552 /* Some processes are using the page */
553 if (page_mapped(page))
556 active = PageActive(page);
557 file = page_is_file_cache(page);
558 lru = page_lru_base_type(page);
560 del_page_from_lru_list(page, lruvec, lru + active);
561 ClearPageActive(page);
562 ClearPageReferenced(page);
564 if (PageWriteback(page) || PageDirty(page)) {
566 * PG_reclaim could be raced with end_page_writeback
567 * It can make readahead confusing. But race window
568 * is _really_ small and it's non-critical problem.
570 add_page_to_lru_list(page, lruvec, lru);
571 SetPageReclaim(page);
574 * The page's writeback ends up during pagevec
575 * We moves tha page into tail of inactive.
577 add_page_to_lru_list_tail(page, lruvec, lru);
578 __count_vm_event(PGROTATED);
582 __count_vm_event(PGDEACTIVATE);
583 update_page_reclaim_stat(lruvec, file, 0);
586 static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec,
589 if (!PageUnevictable(page) && (PageLRU(page) && (PageActive(page) || lru_gen_enabled()))) {
590 int file = page_is_file_cache(page);
591 int lru = page_lru_base_type(page);
593 del_page_from_lru_list(page, lruvec, lru + LRU_ACTIVE);
594 ClearPageActive(page);
595 ClearPageReferenced(page);
596 add_page_to_lru_list(page, lruvec, lru);
598 __count_vm_events(PGDEACTIVATE, hpage_nr_pages(page));
599 update_page_reclaim_stat(lruvec, file, 0);
603 static void lru_lazyfree_fn(struct page *page, struct lruvec *lruvec,
606 if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) &&
607 !PageSwapCache(page) && !PageUnevictable(page)) {
608 bool active = PageActive(page);
610 del_page_from_lru_list(page, lruvec,
611 LRU_INACTIVE_ANON + active);
612 ClearPageActive(page);
613 ClearPageReferenced(page);
615 * lazyfree pages are clean anonymous pages. They have
616 * SwapBacked flag cleared to distinguish normal anonymous
619 ClearPageSwapBacked(page);
620 add_page_to_lru_list(page, lruvec, LRU_INACTIVE_FILE);
622 __count_vm_events(PGLAZYFREE, hpage_nr_pages(page));
623 count_memcg_page_event(page, PGLAZYFREE);
624 update_page_reclaim_stat(lruvec, 1, 0);
629 * Drain pages out of the cpu's pagevecs.
630 * Either "cpu" is the current CPU, and preemption has already been
631 * disabled; or "cpu" is being hot-unplugged, and is already dead.
633 void lru_add_drain_cpu(int cpu)
635 struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu);
637 if (pagevec_count(pvec))
638 __pagevec_lru_add(pvec);
640 pvec = &per_cpu(lru_rotate_pvecs, cpu);
641 if (pagevec_count(pvec)) {
644 /* No harm done if a racing interrupt already did this */
645 local_irq_save(flags);
646 pagevec_move_tail(pvec);
647 local_irq_restore(flags);
650 pvec = &per_cpu(lru_deactivate_file_pvecs, cpu);
651 if (pagevec_count(pvec))
652 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
654 pvec = &per_cpu(lru_deactivate_pvecs, cpu);
655 if (pagevec_count(pvec))
656 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
658 pvec = &per_cpu(lru_lazyfree_pvecs, cpu);
659 if (pagevec_count(pvec))
660 pagevec_lru_move_fn(pvec, lru_lazyfree_fn, NULL);
662 activate_page_drain(cpu);
666 * deactivate_file_page - forcefully deactivate a file page
667 * @page: page to deactivate
669 * This function hints the VM that @page is a good reclaim candidate,
670 * for example if its invalidation fails due to the page being dirty
671 * or under writeback.
673 void deactivate_file_page(struct page *page)
676 * In a workload with many unevictable page such as mprotect,
677 * unevictable page deactivation for accelerating reclaim is pointless.
679 if (PageUnevictable(page))
682 if (likely(get_page_unless_zero(page))) {
683 struct pagevec *pvec = &get_cpu_var(lru_deactivate_file_pvecs);
685 if (!pagevec_add(pvec, page) || PageCompound(page))
686 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
687 put_cpu_var(lru_deactivate_file_pvecs);
692 * deactivate_page - deactivate a page
693 * @page: page to deactivate
695 * deactivate_page() moves @page to the inactive list if @page was on the active
696 * list and was not an unevictable page. This is done to accelerate the reclaim
699 void deactivate_page(struct page *page)
701 if (PageLRU(page) && !PageUnevictable(page) &&
702 (PageActive(page) || lru_gen_enabled())) {
703 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
706 if (!pagevec_add(pvec, page) || PageCompound(page))
707 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
708 put_cpu_var(lru_deactivate_pvecs);
713 * mark_page_lazyfree - make an anon page lazyfree
714 * @page: page to deactivate
716 * mark_page_lazyfree() moves @page to the inactive file list.
717 * This is done to accelerate the reclaim of @page.
719 void mark_page_lazyfree(struct page *page)
721 if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) &&
722 !PageSwapCache(page) && !PageUnevictable(page)) {
723 struct pagevec *pvec = &get_cpu_var(lru_lazyfree_pvecs);
726 if (!pagevec_add(pvec, page) || PageCompound(page))
727 pagevec_lru_move_fn(pvec, lru_lazyfree_fn, NULL);
728 put_cpu_var(lru_lazyfree_pvecs);
732 void lru_add_drain(void)
734 lru_add_drain_cpu(get_cpu());
740 static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
742 static void lru_add_drain_per_cpu(struct work_struct *dummy)
748 * Doesn't need any cpu hotplug locking because we do rely on per-cpu
749 * kworkers being shut down before our page_alloc_cpu_dead callback is
750 * executed on the offlined cpu.
751 * Calling this function with cpu hotplug locks held can actually lead
752 * to obscure indirect dependencies via WQ context.
754 void lru_add_drain_all(void)
756 static DEFINE_MUTEX(lock);
757 static struct cpumask has_work;
761 * Make sure nobody triggers this path before mm_percpu_wq is fully
764 if (WARN_ON(!mm_percpu_wq))
768 cpumask_clear(&has_work);
770 for_each_online_cpu(cpu) {
771 struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
773 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
774 pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
775 pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) ||
776 pagevec_count(&per_cpu(lru_deactivate_pvecs, cpu)) ||
777 pagevec_count(&per_cpu(lru_lazyfree_pvecs, cpu)) ||
778 need_activate_page_drain(cpu)) {
779 INIT_WORK(work, lru_add_drain_per_cpu);
780 queue_work_on(cpu, mm_percpu_wq, work);
781 cpumask_set_cpu(cpu, &has_work);
785 for_each_cpu(cpu, &has_work)
786 flush_work(&per_cpu(lru_add_drain_work, cpu));
791 void lru_add_drain_all(void)
798 * release_pages - batched put_page()
799 * @pages: array of pages to release
800 * @nr: number of pages
802 * Decrement the reference count on all the pages in @pages. If it
803 * fell to zero, remove the page from the LRU and free it.
805 void release_pages(struct page **pages, int nr)
808 LIST_HEAD(pages_to_free);
809 struct pglist_data *locked_pgdat = NULL;
810 struct lruvec *lruvec;
811 unsigned long uninitialized_var(flags);
812 unsigned int uninitialized_var(lock_batch);
814 for (i = 0; i < nr; i++) {
815 struct page *page = pages[i];
818 * Make sure the IRQ-safe lock-holding time does not get
819 * excessive with a continuous string of pages from the
820 * same pgdat. The lock is held only if pgdat != NULL.
822 if (locked_pgdat && ++lock_batch == SWAP_CLUSTER_MAX) {
823 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
827 if (is_huge_zero_page(page))
830 if (is_zone_device_page(page)) {
832 spin_unlock_irqrestore(&locked_pgdat->lru_lock,
837 * ZONE_DEVICE pages that return 'false' from
838 * put_devmap_managed_page() do not require special
839 * processing, and instead, expect a call to
840 * put_page_testzero().
842 if (put_devmap_managed_page(page))
846 page = compound_head(page);
847 if (!put_page_testzero(page))
850 if (PageCompound(page)) {
852 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
855 __put_compound_page(page);
860 struct pglist_data *pgdat = page_pgdat(page);
862 if (pgdat != locked_pgdat) {
864 spin_unlock_irqrestore(&locked_pgdat->lru_lock,
867 locked_pgdat = pgdat;
868 spin_lock_irqsave(&locked_pgdat->lru_lock, flags);
871 lruvec = mem_cgroup_page_lruvec(page, locked_pgdat);
872 VM_BUG_ON_PAGE(!PageLRU(page), page);
873 __ClearPageLRU(page);
874 del_page_from_lru_list(page, lruvec, page_off_lru(page));
877 /* Clear Active bit in case of parallel mark_page_accessed */
878 __ClearPageActive(page);
879 __ClearPageWaiters(page);
881 list_add(&page->lru, &pages_to_free);
884 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
886 mem_cgroup_uncharge_list(&pages_to_free);
887 free_unref_page_list(&pages_to_free);
889 EXPORT_SYMBOL(release_pages);
892 * The pages which we're about to release may be in the deferred lru-addition
893 * queues. That would prevent them from really being freed right now. That's
894 * OK from a correctness point of view but is inefficient - those pages may be
895 * cache-warm and we want to give them back to the page allocator ASAP.
897 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
898 * and __pagevec_lru_add_active() call release_pages() directly to avoid
901 void __pagevec_release(struct pagevec *pvec)
903 if (!pvec->percpu_pvec_drained) {
905 pvec->percpu_pvec_drained = true;
907 release_pages(pvec->pages, pagevec_count(pvec));
908 pagevec_reinit(pvec);
910 EXPORT_SYMBOL(__pagevec_release);
912 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
913 /* used by __split_huge_page_refcount() */
914 void lru_add_page_tail(struct page *page, struct page *page_tail,
915 struct lruvec *lruvec, struct list_head *list)
919 VM_BUG_ON_PAGE(!PageHead(page), page);
920 VM_BUG_ON_PAGE(PageCompound(page_tail), page);
921 VM_BUG_ON_PAGE(PageLRU(page_tail), page);
922 lockdep_assert_held(&lruvec_pgdat(lruvec)->lru_lock);
925 SetPageLRU(page_tail);
927 if (likely(PageLRU(page)))
928 list_add_tail(&page_tail->lru, &page->lru);
930 /* page reclaim is reclaiming a huge page */
932 list_add_tail(&page_tail->lru, list);
935 * Head page has not yet been counted, as an hpage,
936 * so we must account for each subpage individually.
938 * Put page_tail on the list at the correct position
939 * so they all end up in order.
941 add_page_to_lru_list_tail(page_tail, lruvec,
942 page_lru(page_tail));
945 if (!PageUnevictable(page))
946 update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
948 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
950 static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
954 int was_unevictable = TestClearPageUnevictable(page);
956 VM_BUG_ON_PAGE(PageLRU(page), page);
960 * Page becomes evictable in two ways:
961 * 1) Within LRU lock [munlock_vma_page() and __munlock_pagevec()].
962 * 2) Before acquiring LRU lock to put the page to correct LRU and then
963 * a) do PageLRU check with lock [check_move_unevictable_pages]
964 * b) do PageLRU check before lock [clear_page_mlock]
966 * (1) & (2a) are ok as LRU lock will serialize them. For (2b), we need
967 * following strict ordering:
969 * #0: __pagevec_lru_add_fn #1: clear_page_mlock
971 * SetPageLRU() TestClearPageMlocked()
972 * smp_mb() // explicit ordering // above provides strict
974 * PageMlocked() PageLRU()
977 * if '#1' does not observe setting of PG_lru by '#0' and fails
978 * isolation, the explicit barrier will make sure that page_evictable
979 * check will put the page in correct LRU. Without smp_mb(), SetPageLRU
980 * can be reordered after PageMlocked check and can make '#1' to fail
981 * the isolation of the page whose Mlocked bit is cleared (#0 is also
982 * looking at the same page) and the evictable page will be stranded
983 * in an unevictable LRU.
987 if (page_evictable(page)) {
988 lru = page_lru(page);
989 update_page_reclaim_stat(lruvec, page_is_file_cache(page),
992 count_vm_event(UNEVICTABLE_PGRESCUED);
994 lru = LRU_UNEVICTABLE;
995 ClearPageActive(page);
996 SetPageUnevictable(page);
997 if (!was_unevictable)
998 count_vm_event(UNEVICTABLE_PGCULLED);
1001 add_page_to_lru_list(page, lruvec, lru);
1002 trace_mm_lru_insertion(page, lru);
1006 * Add the passed pages to the LRU, then drop the caller's refcount
1007 * on them. Reinitialises the caller's pagevec.
1009 void __pagevec_lru_add(struct pagevec *pvec)
1011 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
1013 EXPORT_SYMBOL(__pagevec_lru_add);
1016 * pagevec_lookup_entries - gang pagecache lookup
1017 * @pvec: Where the resulting entries are placed
1018 * @mapping: The address_space to search
1019 * @start: The starting entry index
1020 * @nr_entries: The maximum number of pages
1021 * @indices: The cache indices corresponding to the entries in @pvec
1023 * pagevec_lookup_entries() will search for and return a group of up
1024 * to @nr_pages pages and shadow entries in the mapping. All
1025 * entries are placed in @pvec. pagevec_lookup_entries() takes a
1026 * reference against actual pages in @pvec.
1028 * The search returns a group of mapping-contiguous entries with
1029 * ascending indexes. There may be holes in the indices due to
1030 * not-present entries.
1032 * pagevec_lookup_entries() returns the number of entries which were
1035 unsigned pagevec_lookup_entries(struct pagevec *pvec,
1036 struct address_space *mapping,
1037 pgoff_t start, unsigned nr_entries,
1040 pvec->nr = find_get_entries(mapping, start, nr_entries,
1041 pvec->pages, indices);
1042 return pagevec_count(pvec);
1046 * pagevec_remove_exceptionals - pagevec exceptionals pruning
1047 * @pvec: The pagevec to prune
1049 * pagevec_lookup_entries() fills both pages and exceptional radix
1050 * tree entries into the pagevec. This function prunes all
1051 * exceptionals from @pvec without leaving holes, so that it can be
1052 * passed on to page-only pagevec operations.
1054 void pagevec_remove_exceptionals(struct pagevec *pvec)
1058 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
1059 struct page *page = pvec->pages[i];
1060 if (!xa_is_value(page))
1061 pvec->pages[j++] = page;
1067 * pagevec_lookup_range - gang pagecache lookup
1068 * @pvec: Where the resulting pages are placed
1069 * @mapping: The address_space to search
1070 * @start: The starting page index
1071 * @end: The final page index
1073 * pagevec_lookup_range() will search for & return a group of up to PAGEVEC_SIZE
1074 * pages in the mapping starting from index @start and upto index @end
1075 * (inclusive). The pages are placed in @pvec. pagevec_lookup() takes a
1076 * reference against the pages in @pvec.
1078 * The search returns a group of mapping-contiguous pages with ascending
1079 * indexes. There may be holes in the indices due to not-present pages. We
1080 * also update @start to index the next page for the traversal.
1082 * pagevec_lookup_range() returns the number of pages which were found. If this
1083 * number is smaller than PAGEVEC_SIZE, the end of specified range has been
1086 unsigned pagevec_lookup_range(struct pagevec *pvec,
1087 struct address_space *mapping, pgoff_t *start, pgoff_t end)
1089 pvec->nr = find_get_pages_range(mapping, start, end, PAGEVEC_SIZE,
1091 return pagevec_count(pvec);
1093 EXPORT_SYMBOL(pagevec_lookup_range);
1095 unsigned pagevec_lookup_range_tag(struct pagevec *pvec,
1096 struct address_space *mapping, pgoff_t *index, pgoff_t end,
1099 pvec->nr = find_get_pages_range_tag(mapping, index, end, tag,
1100 PAGEVEC_SIZE, pvec->pages);
1101 return pagevec_count(pvec);
1103 EXPORT_SYMBOL(pagevec_lookup_range_tag);
1105 unsigned pagevec_lookup_range_nr_tag(struct pagevec *pvec,
1106 struct address_space *mapping, pgoff_t *index, pgoff_t end,
1107 xa_mark_t tag, unsigned max_pages)
1109 pvec->nr = find_get_pages_range_tag(mapping, index, end, tag,
1110 min_t(unsigned int, max_pages, PAGEVEC_SIZE), pvec->pages);
1111 return pagevec_count(pvec);
1113 EXPORT_SYMBOL(pagevec_lookup_range_nr_tag);
1115 * Perform any setup for the swap system
1117 void __init swap_setup(void)
1119 unsigned long megs = totalram_pages() >> (20 - PAGE_SHIFT);
1121 /* Use a smaller cluster for small-memory machines */
1127 * Right now other parts of the system means that we
1128 * _really_ don't want to cluster much more