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>
38 #include <linux/local_lock.h>
42 #define CREATE_TRACE_POINTS
43 #include <trace/events/pagemap.h>
45 /* How many pages do we try to swap or page in/out together? */
48 /* Protecting only lru_rotate.pvec which requires disabling interrupts */
53 static DEFINE_PER_CPU(struct lru_rotate, lru_rotate) = {
54 .lock = INIT_LOCAL_LOCK(lock),
58 * The following struct pagevec are grouped together because they are protected
59 * by disabling preemption (and interrupts remain enabled).
63 struct pagevec lru_add;
64 struct pagevec lru_deactivate_file;
65 struct pagevec lru_deactivate;
66 struct pagevec lru_lazyfree;
68 struct pagevec activate_page;
71 static DEFINE_PER_CPU(struct lru_pvecs, lru_pvecs) = {
72 .lock = INIT_LOCAL_LOCK(lock),
76 * This path almost never happens for VM activity - pages are normally
77 * freed via pagevecs. But it gets used by networking.
79 static void __page_cache_release(struct page *page)
82 pg_data_t *pgdat = page_pgdat(page);
83 struct lruvec *lruvec;
86 spin_lock_irqsave(&pgdat->lru_lock, flags);
87 lruvec = mem_cgroup_page_lruvec(page, pgdat);
88 VM_BUG_ON_PAGE(!PageLRU(page), page);
90 del_page_from_lru_list(page, lruvec, page_off_lru(page));
91 spin_unlock_irqrestore(&pgdat->lru_lock, flags);
93 __ClearPageWaiters(page);
96 static void __put_single_page(struct page *page)
98 __page_cache_release(page);
99 mem_cgroup_uncharge(page);
100 free_unref_page(page);
103 static void __put_compound_page(struct page *page)
106 * __page_cache_release() is supposed to be called for thp, not for
107 * hugetlb. This is because hugetlb page does never have PageLRU set
108 * (it's never listed to any LRU lists) and no memcg routines should
109 * be called for hugetlb (it has a separate hugetlb_cgroup.)
112 __page_cache_release(page);
113 destroy_compound_page(page);
116 void __put_page(struct page *page)
118 if (is_zone_device_page(page)) {
119 put_dev_pagemap(page->pgmap);
122 * The page belongs to the device that created pgmap. Do
123 * not return it to page allocator.
128 if (unlikely(PageCompound(page)))
129 __put_compound_page(page);
131 __put_single_page(page);
133 EXPORT_SYMBOL(__put_page);
136 * put_pages_list() - release a list of pages
137 * @pages: list of pages threaded on page->lru
139 * Release a list of pages which are strung together on page.lru. Currently
140 * used by read_cache_pages() and related error recovery code.
142 void put_pages_list(struct list_head *pages)
144 while (!list_empty(pages)) {
147 victim = lru_to_page(pages);
148 list_del(&victim->lru);
152 EXPORT_SYMBOL(put_pages_list);
155 * get_kernel_pages() - pin kernel pages in memory
156 * @kiov: An array of struct kvec structures
157 * @nr_segs: number of segments to pin
158 * @write: pinning for read/write, currently ignored
159 * @pages: array that receives pointers to the pages pinned.
160 * Should be at least nr_segs long.
162 * Returns number of pages pinned. This may be fewer than the number
163 * requested. If nr_pages is 0 or negative, returns 0. If no pages
164 * were pinned, returns -errno. Each page returned must be released
165 * with a put_page() call when it is finished with.
167 int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write,
172 for (seg = 0; seg < nr_segs; seg++) {
173 if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE))
176 pages[seg] = kmap_to_page(kiov[seg].iov_base);
177 get_page(pages[seg]);
182 EXPORT_SYMBOL_GPL(get_kernel_pages);
185 * get_kernel_page() - pin a kernel page in memory
186 * @start: starting kernel address
187 * @write: pinning for read/write, currently ignored
188 * @pages: array that receives pointer to the page pinned.
189 * Must be at least nr_segs long.
191 * Returns 1 if page is pinned. If the page was not pinned, returns
192 * -errno. The page returned must be released with a put_page() call
193 * when it is finished with.
195 int get_kernel_page(unsigned long start, int write, struct page **pages)
197 const struct kvec kiov = {
198 .iov_base = (void *)start,
202 return get_kernel_pages(&kiov, 1, write, pages);
204 EXPORT_SYMBOL_GPL(get_kernel_page);
206 static void pagevec_lru_move_fn(struct pagevec *pvec,
207 void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg),
211 struct pglist_data *pgdat = NULL;
212 struct lruvec *lruvec;
213 unsigned long flags = 0;
215 for (i = 0; i < pagevec_count(pvec); i++) {
216 struct page *page = pvec->pages[i];
217 struct pglist_data *pagepgdat = page_pgdat(page);
219 if (pagepgdat != pgdat) {
221 spin_unlock_irqrestore(&pgdat->lru_lock, flags);
223 spin_lock_irqsave(&pgdat->lru_lock, flags);
226 lruvec = mem_cgroup_page_lruvec(page, pgdat);
227 (*move_fn)(page, lruvec, arg);
230 spin_unlock_irqrestore(&pgdat->lru_lock, flags);
231 release_pages(pvec->pages, pvec->nr);
232 pagevec_reinit(pvec);
235 static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec,
240 if (PageLRU(page) && !PageUnevictable(page)) {
241 del_page_from_lru_list(page, lruvec, page_lru(page));
242 ClearPageActive(page);
243 add_page_to_lru_list_tail(page, lruvec, page_lru(page));
249 * pagevec_move_tail() must be called with IRQ disabled.
250 * Otherwise this may cause nasty races.
252 static void pagevec_move_tail(struct pagevec *pvec)
256 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
257 __count_vm_events(PGROTATED, pgmoved);
261 * Writeback is about to end against a page which has been marked for immediate
262 * reclaim. If it still appears to be reclaimable, move it to the tail of the
265 void rotate_reclaimable_page(struct page *page)
267 if (!PageLocked(page) && !PageDirty(page) &&
268 !PageUnevictable(page) && PageLRU(page)) {
269 struct pagevec *pvec;
273 local_lock_irqsave(&lru_rotate.lock, flags);
274 pvec = this_cpu_ptr(&lru_rotate.pvec);
275 if (!pagevec_add(pvec, page) || PageCompound(page))
276 pagevec_move_tail(pvec);
277 local_unlock_irqrestore(&lru_rotate.lock, flags);
281 static void update_page_reclaim_stat(struct lruvec *lruvec,
282 int file, int rotated,
283 unsigned int nr_pages)
285 struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
287 reclaim_stat->recent_scanned[file] += nr_pages;
289 reclaim_stat->recent_rotated[file] += nr_pages;
292 static void __activate_page(struct page *page, struct lruvec *lruvec,
295 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
296 int lru = page_lru_base_type(page);
298 del_page_from_lru_list(page, lruvec, lru);
301 add_page_to_lru_list(page, lruvec, lru);
302 trace_mm_lru_activate(page);
304 __count_vm_event(PGACTIVATE);
309 static void activate_page_drain(int cpu)
311 struct pagevec *pvec = &per_cpu(lru_pvecs.activate_page, cpu);
313 if (pagevec_count(pvec))
314 pagevec_lru_move_fn(pvec, __activate_page, NULL);
317 static bool need_activate_page_drain(int cpu)
319 return pagevec_count(&per_cpu(lru_pvecs.activate_page, cpu)) != 0;
322 void activate_page(struct page *page)
324 page = compound_head(page);
325 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
326 struct pagevec *pvec;
328 local_lock(&lru_pvecs.lock);
329 pvec = this_cpu_ptr(&lru_pvecs.activate_page);
331 if (!pagevec_add(pvec, page) || PageCompound(page))
332 pagevec_lru_move_fn(pvec, __activate_page, NULL);
333 local_unlock(&lru_pvecs.lock);
338 static inline void activate_page_drain(int cpu)
342 void activate_page(struct page *page)
344 pg_data_t *pgdat = page_pgdat(page);
346 page = compound_head(page);
347 spin_lock_irq(&pgdat->lru_lock);
348 __activate_page(page, mem_cgroup_page_lruvec(page, pgdat), NULL);
349 spin_unlock_irq(&pgdat->lru_lock);
353 static void __lru_cache_activate_page(struct page *page)
355 struct pagevec *pvec;
358 local_lock(&lru_pvecs.lock);
359 pvec = this_cpu_ptr(&lru_pvecs.lru_add);
362 * Search backwards on the optimistic assumption that the page being
363 * activated has just been added to this pagevec. Note that only
364 * the local pagevec is examined as a !PageLRU page could be in the
365 * process of being released, reclaimed, migrated or on a remote
366 * pagevec that is currently being drained. Furthermore, marking
367 * a remote pagevec's page PageActive potentially hits a race where
368 * a page is marked PageActive just after it is added to the inactive
369 * list causing accounting errors and BUG_ON checks to trigger.
371 for (i = pagevec_count(pvec) - 1; i >= 0; i--) {
372 struct page *pagevec_page = pvec->pages[i];
374 if (pagevec_page == page) {
380 local_unlock(&lru_pvecs.lock);
384 * Mark a page as having seen activity.
386 * inactive,unreferenced -> inactive,referenced
387 * inactive,referenced -> active,unreferenced
388 * active,unreferenced -> active,referenced
390 * When a newly allocated page is not yet visible, so safe for non-atomic ops,
391 * __SetPageReferenced(page) may be substituted for mark_page_accessed(page).
393 void mark_page_accessed(struct page *page)
395 page = compound_head(page);
397 if (!PageReferenced(page)) {
398 SetPageReferenced(page);
399 } else if (PageUnevictable(page)) {
401 * Unevictable pages are on the "LRU_UNEVICTABLE" list. But,
402 * this list is never rotated or maintained, so marking an
403 * evictable page accessed has no effect.
405 } else if (!PageActive(page)) {
407 * If the page is on the LRU, queue it for activation via
408 * lru_pvecs.activate_page. Otherwise, assume the page is on a
409 * pagevec, mark it active and it'll be moved to the active
410 * LRU on the next drain.
415 __lru_cache_activate_page(page);
416 ClearPageReferenced(page);
417 if (page_is_file_lru(page))
418 workingset_activation(page);
420 if (page_is_idle(page))
421 clear_page_idle(page);
423 EXPORT_SYMBOL(mark_page_accessed);
426 * lru_cache_add - add a page to a page list
427 * @page: the page to be added to the LRU.
429 * Queue the page for addition to the LRU via pagevec. The decision on whether
430 * to add the page to the [in]active [file|anon] list is deferred until the
431 * pagevec is drained. This gives a chance for the caller of lru_cache_add()
432 * have the page added to the active list using mark_page_accessed().
434 void lru_cache_add(struct page *page)
436 struct pagevec *pvec;
438 VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page);
439 VM_BUG_ON_PAGE(PageLRU(page), page);
442 local_lock(&lru_pvecs.lock);
443 pvec = this_cpu_ptr(&lru_pvecs.lru_add);
444 if (!pagevec_add(pvec, page) || PageCompound(page))
445 __pagevec_lru_add(pvec);
446 local_unlock(&lru_pvecs.lock);
448 EXPORT_SYMBOL(lru_cache_add);
451 * lru_cache_add_active_or_unevictable
452 * @page: the page to be added to LRU
453 * @vma: vma in which page is mapped for determining reclaimability
455 * Place @page on the active or unevictable LRU list, depending on its
456 * evictability. Note that if the page is not evictable, it goes
457 * directly back onto it's zone's unevictable list, it does NOT use a
460 void lru_cache_add_active_or_unevictable(struct page *page,
461 struct vm_area_struct *vma)
463 VM_BUG_ON_PAGE(PageLRU(page), page);
465 if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED))
467 else if (!TestSetPageMlocked(page)) {
469 * We use the irq-unsafe __mod_zone_page_stat because this
470 * counter is not modified from interrupt context, and the pte
471 * lock is held(spinlock), which implies preemption disabled.
473 __mod_zone_page_state(page_zone(page), NR_MLOCK,
474 hpage_nr_pages(page));
475 count_vm_event(UNEVICTABLE_PGMLOCKED);
481 * If the page can not be invalidated, it is moved to the
482 * inactive list to speed up its reclaim. It is moved to the
483 * head of the list, rather than the tail, to give the flusher
484 * threads some time to write it out, as this is much more
485 * effective than the single-page writeout from reclaim.
487 * If the page isn't page_mapped and dirty/writeback, the page
488 * could reclaim asap using PG_reclaim.
490 * 1. active, mapped page -> none
491 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
492 * 3. inactive, mapped page -> none
493 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
494 * 5. inactive, clean -> inactive, tail
497 * In 4, why it moves inactive's head, the VM expects the page would
498 * be write it out by flusher threads as this is much more effective
499 * than the single-page writeout from reclaim.
501 static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec,
510 if (PageUnevictable(page))
513 /* Some processes are using the page */
514 if (page_mapped(page))
517 active = PageActive(page);
518 file = page_is_file_lru(page);
519 lru = page_lru_base_type(page);
521 del_page_from_lru_list(page, lruvec, lru + active);
522 ClearPageActive(page);
523 ClearPageReferenced(page);
525 if (PageWriteback(page) || PageDirty(page)) {
527 * PG_reclaim could be raced with end_page_writeback
528 * It can make readahead confusing. But race window
529 * is _really_ small and it's non-critical problem.
531 add_page_to_lru_list(page, lruvec, lru);
532 SetPageReclaim(page);
535 * The page's writeback ends up during pagevec
536 * We moves tha page into tail of inactive.
538 add_page_to_lru_list_tail(page, lruvec, lru);
539 __count_vm_event(PGROTATED);
543 __count_vm_event(PGDEACTIVATE);
544 update_page_reclaim_stat(lruvec, file, 0, hpage_nr_pages(page));
547 static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec,
550 if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) {
551 int file = page_is_file_lru(page);
552 int lru = page_lru_base_type(page);
554 del_page_from_lru_list(page, lruvec, lru + LRU_ACTIVE);
555 ClearPageActive(page);
556 ClearPageReferenced(page);
557 add_page_to_lru_list(page, lruvec, lru);
559 __count_vm_events(PGDEACTIVATE, hpage_nr_pages(page));
560 update_page_reclaim_stat(lruvec, file, 0, hpage_nr_pages(page));
564 static void lru_lazyfree_fn(struct page *page, struct lruvec *lruvec,
567 if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) &&
568 !PageSwapCache(page) && !PageUnevictable(page)) {
569 bool active = PageActive(page);
571 del_page_from_lru_list(page, lruvec,
572 LRU_INACTIVE_ANON + active);
573 ClearPageActive(page);
574 ClearPageReferenced(page);
576 * Lazyfree pages are clean anonymous pages. They have
577 * PG_swapbacked flag cleared, to distinguish them from normal
580 ClearPageSwapBacked(page);
581 add_page_to_lru_list(page, lruvec, LRU_INACTIVE_FILE);
583 __count_vm_events(PGLAZYFREE, hpage_nr_pages(page));
584 count_memcg_page_event(page, PGLAZYFREE);
585 update_page_reclaim_stat(lruvec, 1, 0, hpage_nr_pages(page));
590 * Drain pages out of the cpu's pagevecs.
591 * Either "cpu" is the current CPU, and preemption has already been
592 * disabled; or "cpu" is being hot-unplugged, and is already dead.
594 void lru_add_drain_cpu(int cpu)
596 struct pagevec *pvec = &per_cpu(lru_pvecs.lru_add, cpu);
598 if (pagevec_count(pvec))
599 __pagevec_lru_add(pvec);
601 pvec = &per_cpu(lru_rotate.pvec, cpu);
602 if (pagevec_count(pvec)) {
605 /* No harm done if a racing interrupt already did this */
606 local_lock_irqsave(&lru_rotate.lock, flags);
607 pagevec_move_tail(pvec);
608 local_unlock_irqrestore(&lru_rotate.lock, flags);
611 pvec = &per_cpu(lru_pvecs.lru_deactivate_file, cpu);
612 if (pagevec_count(pvec))
613 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
615 pvec = &per_cpu(lru_pvecs.lru_deactivate, cpu);
616 if (pagevec_count(pvec))
617 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
619 pvec = &per_cpu(lru_pvecs.lru_lazyfree, cpu);
620 if (pagevec_count(pvec))
621 pagevec_lru_move_fn(pvec, lru_lazyfree_fn, NULL);
623 activate_page_drain(cpu);
627 * deactivate_file_page - forcefully deactivate a file page
628 * @page: page to deactivate
630 * This function hints the VM that @page is a good reclaim candidate,
631 * for example if its invalidation fails due to the page being dirty
632 * or under writeback.
634 void deactivate_file_page(struct page *page)
637 * In a workload with many unevictable page such as mprotect,
638 * unevictable page deactivation for accelerating reclaim is pointless.
640 if (PageUnevictable(page))
643 if (likely(get_page_unless_zero(page))) {
644 struct pagevec *pvec;
646 local_lock(&lru_pvecs.lock);
647 pvec = this_cpu_ptr(&lru_pvecs.lru_deactivate_file);
649 if (!pagevec_add(pvec, page) || PageCompound(page))
650 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
651 local_unlock(&lru_pvecs.lock);
656 * deactivate_page - deactivate a page
657 * @page: page to deactivate
659 * deactivate_page() moves @page to the inactive list if @page was on the active
660 * list and was not an unevictable page. This is done to accelerate the reclaim
663 void deactivate_page(struct page *page)
665 if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) {
666 struct pagevec *pvec;
668 local_lock(&lru_pvecs.lock);
669 pvec = this_cpu_ptr(&lru_pvecs.lru_deactivate);
671 if (!pagevec_add(pvec, page) || PageCompound(page))
672 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
673 local_unlock(&lru_pvecs.lock);
678 * mark_page_lazyfree - make an anon page lazyfree
679 * @page: page to deactivate
681 * mark_page_lazyfree() moves @page to the inactive file list.
682 * This is done to accelerate the reclaim of @page.
684 void mark_page_lazyfree(struct page *page)
686 if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) &&
687 !PageSwapCache(page) && !PageUnevictable(page)) {
688 struct pagevec *pvec;
690 local_lock(&lru_pvecs.lock);
691 pvec = this_cpu_ptr(&lru_pvecs.lru_lazyfree);
693 if (!pagevec_add(pvec, page) || PageCompound(page))
694 pagevec_lru_move_fn(pvec, lru_lazyfree_fn, NULL);
695 local_unlock(&lru_pvecs.lock);
699 void lru_add_drain(void)
701 local_lock(&lru_pvecs.lock);
702 lru_add_drain_cpu(smp_processor_id());
703 local_unlock(&lru_pvecs.lock);
706 void lru_add_drain_cpu_zone(struct zone *zone)
708 local_lock(&lru_pvecs.lock);
709 lru_add_drain_cpu(smp_processor_id());
710 drain_local_pages(zone);
711 local_unlock(&lru_pvecs.lock);
716 static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
718 static void lru_add_drain_per_cpu(struct work_struct *dummy)
724 * Doesn't need any cpu hotplug locking because we do rely on per-cpu
725 * kworkers being shut down before our page_alloc_cpu_dead callback is
726 * executed on the offlined cpu.
727 * Calling this function with cpu hotplug locks held can actually lead
728 * to obscure indirect dependencies via WQ context.
730 void lru_add_drain_all(void)
732 static seqcount_t seqcount = SEQCNT_ZERO(seqcount);
733 static DEFINE_MUTEX(lock);
734 static struct cpumask has_work;
738 * Make sure nobody triggers this path before mm_percpu_wq is fully
741 if (WARN_ON(!mm_percpu_wq))
744 seq = raw_read_seqcount_latch(&seqcount);
749 * Piggyback on drain started and finished while we waited for lock:
750 * all pages pended at the time of our enter were drained from vectors.
752 if (__read_seqcount_retry(&seqcount, seq))
755 raw_write_seqcount_latch(&seqcount);
757 cpumask_clear(&has_work);
759 for_each_online_cpu(cpu) {
760 struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
762 if (pagevec_count(&per_cpu(lru_pvecs.lru_add, cpu)) ||
763 pagevec_count(&per_cpu(lru_rotate.pvec, cpu)) ||
764 pagevec_count(&per_cpu(lru_pvecs.lru_deactivate_file, cpu)) ||
765 pagevec_count(&per_cpu(lru_pvecs.lru_deactivate, cpu)) ||
766 pagevec_count(&per_cpu(lru_pvecs.lru_lazyfree, cpu)) ||
767 need_activate_page_drain(cpu)) {
768 INIT_WORK(work, lru_add_drain_per_cpu);
769 queue_work_on(cpu, mm_percpu_wq, work);
770 cpumask_set_cpu(cpu, &has_work);
774 for_each_cpu(cpu, &has_work)
775 flush_work(&per_cpu(lru_add_drain_work, cpu));
781 void lru_add_drain_all(void)
788 * release_pages - batched put_page()
789 * @pages: array of pages to release
790 * @nr: number of pages
792 * Decrement the reference count on all the pages in @pages. If it
793 * fell to zero, remove the page from the LRU and free it.
795 void release_pages(struct page **pages, int nr)
798 LIST_HEAD(pages_to_free);
799 struct pglist_data *locked_pgdat = NULL;
800 struct lruvec *lruvec;
801 unsigned long uninitialized_var(flags);
802 unsigned int uninitialized_var(lock_batch);
804 for (i = 0; i < nr; i++) {
805 struct page *page = pages[i];
808 * Make sure the IRQ-safe lock-holding time does not get
809 * excessive with a continuous string of pages from the
810 * same pgdat. The lock is held only if pgdat != NULL.
812 if (locked_pgdat && ++lock_batch == SWAP_CLUSTER_MAX) {
813 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
817 if (is_huge_zero_page(page))
820 if (is_zone_device_page(page)) {
822 spin_unlock_irqrestore(&locked_pgdat->lru_lock,
827 * ZONE_DEVICE pages that return 'false' from
828 * put_devmap_managed_page() do not require special
829 * processing, and instead, expect a call to
830 * put_page_testzero().
832 if (page_is_devmap_managed(page)) {
833 put_devmap_managed_page(page);
838 page = compound_head(page);
839 if (!put_page_testzero(page))
842 if (PageCompound(page)) {
844 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
847 __put_compound_page(page);
852 struct pglist_data *pgdat = page_pgdat(page);
854 if (pgdat != locked_pgdat) {
856 spin_unlock_irqrestore(&locked_pgdat->lru_lock,
859 locked_pgdat = pgdat;
860 spin_lock_irqsave(&locked_pgdat->lru_lock, flags);
863 lruvec = mem_cgroup_page_lruvec(page, locked_pgdat);
864 VM_BUG_ON_PAGE(!PageLRU(page), page);
865 __ClearPageLRU(page);
866 del_page_from_lru_list(page, lruvec, page_off_lru(page));
869 /* Clear Active bit in case of parallel mark_page_accessed */
870 __ClearPageActive(page);
871 __ClearPageWaiters(page);
873 list_add(&page->lru, &pages_to_free);
876 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
878 mem_cgroup_uncharge_list(&pages_to_free);
879 free_unref_page_list(&pages_to_free);
881 EXPORT_SYMBOL(release_pages);
884 * The pages which we're about to release may be in the deferred lru-addition
885 * queues. That would prevent them from really being freed right now. That's
886 * OK from a correctness point of view but is inefficient - those pages may be
887 * cache-warm and we want to give them back to the page allocator ASAP.
889 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
890 * and __pagevec_lru_add_active() call release_pages() directly to avoid
893 void __pagevec_release(struct pagevec *pvec)
895 if (!pvec->percpu_pvec_drained) {
897 pvec->percpu_pvec_drained = true;
899 release_pages(pvec->pages, pagevec_count(pvec));
900 pagevec_reinit(pvec);
902 EXPORT_SYMBOL(__pagevec_release);
904 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
905 /* used by __split_huge_page_refcount() */
906 void lru_add_page_tail(struct page *page, struct page *page_tail,
907 struct lruvec *lruvec, struct list_head *list)
909 VM_BUG_ON_PAGE(!PageHead(page), page);
910 VM_BUG_ON_PAGE(PageCompound(page_tail), page);
911 VM_BUG_ON_PAGE(PageLRU(page_tail), page);
912 lockdep_assert_held(&lruvec_pgdat(lruvec)->lru_lock);
915 SetPageLRU(page_tail);
917 if (likely(PageLRU(page)))
918 list_add_tail(&page_tail->lru, &page->lru);
920 /* page reclaim is reclaiming a huge page */
922 list_add_tail(&page_tail->lru, list);
925 * Head page has not yet been counted, as an hpage,
926 * so we must account for each subpage individually.
928 * Put page_tail on the list at the correct position
929 * so they all end up in order.
931 add_page_to_lru_list_tail(page_tail, lruvec,
932 page_lru(page_tail));
935 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
937 static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
941 int was_unevictable = TestClearPageUnevictable(page);
943 VM_BUG_ON_PAGE(PageLRU(page), page);
946 * Page becomes evictable in two ways:
947 * 1) Within LRU lock [munlock_vma_page() and __munlock_pagevec()].
948 * 2) Before acquiring LRU lock to put the page to correct LRU and then
949 * a) do PageLRU check with lock [check_move_unevictable_pages]
950 * b) do PageLRU check before lock [clear_page_mlock]
952 * (1) & (2a) are ok as LRU lock will serialize them. For (2b), we need
953 * following strict ordering:
955 * #0: __pagevec_lru_add_fn #1: clear_page_mlock
957 * SetPageLRU() TestClearPageMlocked()
958 * smp_mb() // explicit ordering // above provides strict
960 * PageMlocked() PageLRU()
963 * if '#1' does not observe setting of PG_lru by '#0' and fails
964 * isolation, the explicit barrier will make sure that page_evictable
965 * check will put the page in correct LRU. Without smp_mb(), SetPageLRU
966 * can be reordered after PageMlocked check and can make '#1' to fail
967 * the isolation of the page whose Mlocked bit is cleared (#0 is also
968 * looking at the same page) and the evictable page will be stranded
969 * in an unevictable LRU.
972 smp_mb__after_atomic();
974 if (page_evictable(page)) {
975 lru = page_lru(page);
977 count_vm_event(UNEVICTABLE_PGRESCUED);
979 lru = LRU_UNEVICTABLE;
980 ClearPageActive(page);
981 SetPageUnevictable(page);
982 if (!was_unevictable)
983 count_vm_event(UNEVICTABLE_PGCULLED);
986 add_page_to_lru_list(page, lruvec, lru);
987 trace_mm_lru_insertion(page, lru);
991 * Add the passed pages to the LRU, then drop the caller's refcount
992 * on them. Reinitialises the caller's pagevec.
994 void __pagevec_lru_add(struct pagevec *pvec)
996 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
1000 * pagevec_lookup_entries - gang pagecache lookup
1001 * @pvec: Where the resulting entries are placed
1002 * @mapping: The address_space to search
1003 * @start: The starting entry index
1004 * @nr_entries: The maximum number of pages
1005 * @indices: The cache indices corresponding to the entries in @pvec
1007 * pagevec_lookup_entries() will search for and return a group of up
1008 * to @nr_pages pages and shadow entries in the mapping. All
1009 * entries are placed in @pvec. pagevec_lookup_entries() takes a
1010 * reference against actual pages in @pvec.
1012 * The search returns a group of mapping-contiguous entries with
1013 * ascending indexes. There may be holes in the indices due to
1014 * not-present entries.
1016 * Only one subpage of a Transparent Huge Page is returned in one call:
1017 * allowing truncate_inode_pages_range() to evict the whole THP without
1018 * cycling through a pagevec of extra references.
1020 * pagevec_lookup_entries() returns the number of entries which were
1023 unsigned pagevec_lookup_entries(struct pagevec *pvec,
1024 struct address_space *mapping,
1025 pgoff_t start, unsigned nr_entries,
1028 pvec->nr = find_get_entries(mapping, start, nr_entries,
1029 pvec->pages, indices);
1030 return pagevec_count(pvec);
1034 * pagevec_remove_exceptionals - pagevec exceptionals pruning
1035 * @pvec: The pagevec to prune
1037 * pagevec_lookup_entries() fills both pages and exceptional radix
1038 * tree entries into the pagevec. This function prunes all
1039 * exceptionals from @pvec without leaving holes, so that it can be
1040 * passed on to page-only pagevec operations.
1042 void pagevec_remove_exceptionals(struct pagevec *pvec)
1046 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
1047 struct page *page = pvec->pages[i];
1048 if (!xa_is_value(page))
1049 pvec->pages[j++] = page;
1055 * pagevec_lookup_range - gang pagecache lookup
1056 * @pvec: Where the resulting pages are placed
1057 * @mapping: The address_space to search
1058 * @start: The starting page index
1059 * @end: The final page index
1061 * pagevec_lookup_range() will search for & return a group of up to PAGEVEC_SIZE
1062 * pages in the mapping starting from index @start and upto index @end
1063 * (inclusive). The pages are placed in @pvec. pagevec_lookup() takes a
1064 * reference against the pages in @pvec.
1066 * The search returns a group of mapping-contiguous pages with ascending
1067 * indexes. There may be holes in the indices due to not-present pages. We
1068 * also update @start to index the next page for the traversal.
1070 * pagevec_lookup_range() returns the number of pages which were found. If this
1071 * number is smaller than PAGEVEC_SIZE, the end of specified range has been
1074 unsigned pagevec_lookup_range(struct pagevec *pvec,
1075 struct address_space *mapping, pgoff_t *start, pgoff_t end)
1077 pvec->nr = find_get_pages_range(mapping, start, end, PAGEVEC_SIZE,
1079 return pagevec_count(pvec);
1081 EXPORT_SYMBOL(pagevec_lookup_range);
1083 unsigned pagevec_lookup_range_tag(struct pagevec *pvec,
1084 struct address_space *mapping, pgoff_t *index, pgoff_t end,
1087 pvec->nr = find_get_pages_range_tag(mapping, index, end, tag,
1088 PAGEVEC_SIZE, pvec->pages);
1089 return pagevec_count(pvec);
1091 EXPORT_SYMBOL(pagevec_lookup_range_tag);
1093 unsigned pagevec_lookup_range_nr_tag(struct pagevec *pvec,
1094 struct address_space *mapping, pgoff_t *index, pgoff_t end,
1095 xa_mark_t tag, unsigned max_pages)
1097 pvec->nr = find_get_pages_range_tag(mapping, index, end, tag,
1098 min_t(unsigned int, max_pages, PAGEVEC_SIZE), pvec->pages);
1099 return pagevec_count(pvec);
1101 EXPORT_SYMBOL(pagevec_lookup_range_nr_tag);
1103 * Perform any setup for the swap system
1105 void __init swap_setup(void)
1107 unsigned long megs = totalram_pages() >> (20 - PAGE_SHIFT);
1109 /* Use a smaller cluster for small-memory machines */
1115 * Right now other parts of the system means that we
1116 * _really_ don't want to cluster much more
1120 #ifdef CONFIG_DEV_PAGEMAP_OPS
1121 void put_devmap_managed_page(struct page *page)
1125 if (WARN_ON_ONCE(!page_is_devmap_managed(page)))
1128 count = page_ref_dec_return(page);
1131 * devmap page refcounts are 1-based, rather than 0-based: if
1132 * refcount is 1, then the page is free and the refcount is
1133 * stable because nobody holds a reference on the page.
1136 free_devmap_managed_page(page);
1140 EXPORT_SYMBOL(put_devmap_managed_page);