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)
105 compound_page_dtor *dtor;
108 * __page_cache_release() is supposed to be called for thp, not for
109 * hugetlb. This is because hugetlb page does never have PageLRU set
110 * (it's never listed to any LRU lists) and no memcg routines should
111 * be called for hugetlb (it has a separate hugetlb_cgroup.)
114 __page_cache_release(page);
115 dtor = get_compound_page_dtor(page);
119 void __put_page(struct page *page)
121 if (is_zone_device_page(page)) {
122 put_dev_pagemap(page->pgmap);
125 * The page belongs to the device that created pgmap. Do
126 * not return it to page allocator.
131 if (unlikely(PageCompound(page)))
132 __put_compound_page(page);
134 __put_single_page(page);
136 EXPORT_SYMBOL(__put_page);
139 * put_pages_list() - release a list of pages
140 * @pages: list of pages threaded on page->lru
142 * Release a list of pages which are strung together on page.lru. Currently
143 * used by read_cache_pages() and related error recovery code.
145 void put_pages_list(struct list_head *pages)
147 while (!list_empty(pages)) {
150 victim = lru_to_page(pages);
151 list_del(&victim->lru);
155 EXPORT_SYMBOL(put_pages_list);
158 * get_kernel_pages() - pin kernel pages in memory
159 * @kiov: An array of struct kvec structures
160 * @nr_segs: number of segments to pin
161 * @write: pinning for read/write, currently ignored
162 * @pages: array that receives pointers to the pages pinned.
163 * Should be at least nr_segs long.
165 * Returns number of pages pinned. This may be fewer than the number
166 * requested. If nr_pages is 0 or negative, returns 0. If no pages
167 * were pinned, returns -errno. Each page returned must be released
168 * with a put_page() call when it is finished with.
170 int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write,
175 for (seg = 0; seg < nr_segs; seg++) {
176 if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE))
179 pages[seg] = kmap_to_page(kiov[seg].iov_base);
180 get_page(pages[seg]);
185 EXPORT_SYMBOL_GPL(get_kernel_pages);
188 * get_kernel_page() - pin a kernel page in memory
189 * @start: starting kernel address
190 * @write: pinning for read/write, currently ignored
191 * @pages: array that receives pointer to the page pinned.
192 * Must be at least nr_segs long.
194 * Returns 1 if page is pinned. If the page was not pinned, returns
195 * -errno. The page returned must be released with a put_page() call
196 * when it is finished with.
198 int get_kernel_page(unsigned long start, int write, struct page **pages)
200 const struct kvec kiov = {
201 .iov_base = (void *)start,
205 return get_kernel_pages(&kiov, 1, write, pages);
207 EXPORT_SYMBOL_GPL(get_kernel_page);
209 static void pagevec_lru_move_fn(struct pagevec *pvec,
210 void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg),
214 struct pglist_data *pgdat = NULL;
215 struct lruvec *lruvec;
216 unsigned long flags = 0;
218 for (i = 0; i < pagevec_count(pvec); i++) {
219 struct page *page = pvec->pages[i];
220 struct pglist_data *pagepgdat = page_pgdat(page);
222 if (pagepgdat != pgdat) {
224 spin_unlock_irqrestore(&pgdat->lru_lock, flags);
226 spin_lock_irqsave(&pgdat->lru_lock, flags);
229 lruvec = mem_cgroup_page_lruvec(page, pgdat);
230 (*move_fn)(page, lruvec, arg);
233 spin_unlock_irqrestore(&pgdat->lru_lock, flags);
234 release_pages(pvec->pages, pvec->nr);
235 pagevec_reinit(pvec);
238 static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec,
243 if (PageLRU(page) && !PageUnevictable(page)) {
244 del_page_from_lru_list(page, lruvec, page_lru(page));
245 ClearPageActive(page);
246 add_page_to_lru_list_tail(page, lruvec, page_lru(page));
252 * pagevec_move_tail() must be called with IRQ disabled.
253 * Otherwise this may cause nasty races.
255 static void pagevec_move_tail(struct pagevec *pvec)
259 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
260 __count_vm_events(PGROTATED, pgmoved);
264 * Writeback is about to end against a page which has been marked for immediate
265 * reclaim. If it still appears to be reclaimable, move it to the tail of the
268 void rotate_reclaimable_page(struct page *page)
270 if (!PageLocked(page) && !PageDirty(page) &&
271 !PageUnevictable(page) && PageLRU(page)) {
272 struct pagevec *pvec;
276 local_lock_irqsave(&lru_rotate.lock, flags);
277 pvec = this_cpu_ptr(&lru_rotate.pvec);
278 if (!pagevec_add(pvec, page) || PageCompound(page))
279 pagevec_move_tail(pvec);
280 local_unlock_irqrestore(&lru_rotate.lock, flags);
284 static void update_page_reclaim_stat(struct lruvec *lruvec,
285 int file, int rotated)
287 struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
289 reclaim_stat->recent_scanned[file]++;
291 reclaim_stat->recent_rotated[file]++;
294 static void __activate_page(struct page *page, struct lruvec *lruvec,
297 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
298 int file = page_is_file_lru(page);
299 int lru = page_lru_base_type(page);
301 del_page_from_lru_list(page, lruvec, lru);
304 add_page_to_lru_list(page, lruvec, lru);
305 trace_mm_lru_activate(page);
307 __count_vm_event(PGACTIVATE);
308 update_page_reclaim_stat(lruvec, file, 1);
313 static void activate_page_drain(int cpu)
315 struct pagevec *pvec = &per_cpu(lru_pvecs.activate_page, cpu);
317 if (pagevec_count(pvec))
318 pagevec_lru_move_fn(pvec, __activate_page, NULL);
321 static bool need_activate_page_drain(int cpu)
323 return pagevec_count(&per_cpu(lru_pvecs.activate_page, cpu)) != 0;
326 void activate_page(struct page *page)
328 page = compound_head(page);
329 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
330 struct pagevec *pvec;
332 local_lock(&lru_pvecs.lock);
333 pvec = this_cpu_ptr(&lru_pvecs.activate_page);
335 if (!pagevec_add(pvec, page) || PageCompound(page))
336 pagevec_lru_move_fn(pvec, __activate_page, NULL);
337 local_unlock(&lru_pvecs.lock);
342 static inline void activate_page_drain(int cpu)
346 void activate_page(struct page *page)
348 pg_data_t *pgdat = page_pgdat(page);
350 page = compound_head(page);
351 spin_lock_irq(&pgdat->lru_lock);
352 __activate_page(page, mem_cgroup_page_lruvec(page, pgdat), NULL);
353 spin_unlock_irq(&pgdat->lru_lock);
357 static void __lru_cache_activate_page(struct page *page)
359 struct pagevec *pvec;
362 local_lock(&lru_pvecs.lock);
363 pvec = this_cpu_ptr(&lru_pvecs.lru_add);
366 * Search backwards on the optimistic assumption that the page being
367 * activated has just been added to this pagevec. Note that only
368 * the local pagevec is examined as a !PageLRU page could be in the
369 * process of being released, reclaimed, migrated or on a remote
370 * pagevec that is currently being drained. Furthermore, marking
371 * a remote pagevec's page PageActive potentially hits a race where
372 * a page is marked PageActive just after it is added to the inactive
373 * list causing accounting errors and BUG_ON checks to trigger.
375 for (i = pagevec_count(pvec) - 1; i >= 0; i--) {
376 struct page *pagevec_page = pvec->pages[i];
378 if (pagevec_page == page) {
384 local_unlock(&lru_pvecs.lock);
388 * Mark a page as having seen activity.
390 * inactive,unreferenced -> inactive,referenced
391 * inactive,referenced -> active,unreferenced
392 * active,unreferenced -> active,referenced
394 * When a newly allocated page is not yet visible, so safe for non-atomic ops,
395 * __SetPageReferenced(page) may be substituted for mark_page_accessed(page).
397 void mark_page_accessed(struct page *page)
399 page = compound_head(page);
401 if (!PageReferenced(page)) {
402 SetPageReferenced(page);
403 } else if (PageUnevictable(page)) {
405 * Unevictable pages are on the "LRU_UNEVICTABLE" list. But,
406 * this list is never rotated or maintained, so marking an
407 * evictable page accessed has no effect.
409 } else if (!PageActive(page)) {
411 * If the page is on the LRU, queue it for activation via
412 * lru_pvecs.activate_page. Otherwise, assume the page is on a
413 * pagevec, mark it active and it'll be moved to the active
414 * LRU on the next drain.
419 __lru_cache_activate_page(page);
420 ClearPageReferenced(page);
421 if (page_is_file_lru(page))
422 workingset_activation(page);
424 if (page_is_idle(page))
425 clear_page_idle(page);
427 EXPORT_SYMBOL(mark_page_accessed);
429 static void __lru_cache_add(struct page *page)
431 struct pagevec *pvec;
433 local_lock(&lru_pvecs.lock);
434 pvec = this_cpu_ptr(&lru_pvecs.lru_add);
436 if (!pagevec_add(pvec, page) || PageCompound(page))
437 __pagevec_lru_add(pvec);
438 local_unlock(&lru_pvecs.lock);
442 * lru_cache_add_anon - add a page to the page lists
443 * @page: the page to add
445 void lru_cache_add_anon(struct page *page)
447 if (PageActive(page))
448 ClearPageActive(page);
449 __lru_cache_add(page);
452 void lru_cache_add_file(struct page *page)
454 if (PageActive(page))
455 ClearPageActive(page);
456 __lru_cache_add(page);
458 EXPORT_SYMBOL(lru_cache_add_file);
461 * lru_cache_add - add a page to a page list
462 * @page: the page to be added to the LRU.
464 * Queue the page for addition to the LRU via pagevec. The decision on whether
465 * to add the page to the [in]active [file|anon] list is deferred until the
466 * pagevec is drained. This gives a chance for the caller of lru_cache_add()
467 * have the page added to the active list using mark_page_accessed().
469 void lru_cache_add(struct page *page)
471 VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page);
472 VM_BUG_ON_PAGE(PageLRU(page), page);
473 __lru_cache_add(page);
477 * lru_cache_add_active_or_unevictable
478 * @page: the page to be added to LRU
479 * @vma: vma in which page is mapped for determining reclaimability
481 * Place @page on the active or unevictable LRU list, depending on its
482 * evictability. Note that if the page is not evictable, it goes
483 * directly back onto it's zone's unevictable list, it does NOT use a
486 void lru_cache_add_active_or_unevictable(struct page *page,
487 struct vm_area_struct *vma)
489 VM_BUG_ON_PAGE(PageLRU(page), page);
491 if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED))
493 else if (!TestSetPageMlocked(page)) {
495 * We use the irq-unsafe __mod_zone_page_stat because this
496 * counter is not modified from interrupt context, and the pte
497 * lock is held(spinlock), which implies preemption disabled.
499 __mod_zone_page_state(page_zone(page), NR_MLOCK,
500 hpage_nr_pages(page));
501 count_vm_event(UNEVICTABLE_PGMLOCKED);
507 * If the page can not be invalidated, it is moved to the
508 * inactive list to speed up its reclaim. It is moved to the
509 * head of the list, rather than the tail, to give the flusher
510 * threads some time to write it out, as this is much more
511 * effective than the single-page writeout from reclaim.
513 * If the page isn't page_mapped and dirty/writeback, the page
514 * could reclaim asap using PG_reclaim.
516 * 1. active, mapped page -> none
517 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
518 * 3. inactive, mapped page -> none
519 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
520 * 5. inactive, clean -> inactive, tail
523 * In 4, why it moves inactive's head, the VM expects the page would
524 * be write it out by flusher threads as this is much more effective
525 * than the single-page writeout from reclaim.
527 static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec,
536 if (PageUnevictable(page))
539 /* Some processes are using the page */
540 if (page_mapped(page))
543 active = PageActive(page);
544 file = page_is_file_lru(page);
545 lru = page_lru_base_type(page);
547 del_page_from_lru_list(page, lruvec, lru + active);
548 ClearPageActive(page);
549 ClearPageReferenced(page);
551 if (PageWriteback(page) || PageDirty(page)) {
553 * PG_reclaim could be raced with end_page_writeback
554 * It can make readahead confusing. But race window
555 * is _really_ small and it's non-critical problem.
557 add_page_to_lru_list(page, lruvec, lru);
558 SetPageReclaim(page);
561 * The page's writeback ends up during pagevec
562 * We moves tha page into tail of inactive.
564 add_page_to_lru_list_tail(page, lruvec, lru);
565 __count_vm_event(PGROTATED);
569 __count_vm_event(PGDEACTIVATE);
570 update_page_reclaim_stat(lruvec, file, 0);
573 static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec,
576 if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) {
577 int file = page_is_file_lru(page);
578 int lru = page_lru_base_type(page);
580 del_page_from_lru_list(page, lruvec, lru + LRU_ACTIVE);
581 ClearPageActive(page);
582 ClearPageReferenced(page);
583 add_page_to_lru_list(page, lruvec, lru);
585 __count_vm_events(PGDEACTIVATE, hpage_nr_pages(page));
586 update_page_reclaim_stat(lruvec, file, 0);
590 static void lru_lazyfree_fn(struct page *page, struct lruvec *lruvec,
593 if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) &&
594 !PageSwapCache(page) && !PageUnevictable(page)) {
595 bool active = PageActive(page);
597 del_page_from_lru_list(page, lruvec,
598 LRU_INACTIVE_ANON + active);
599 ClearPageActive(page);
600 ClearPageReferenced(page);
602 * Lazyfree pages are clean anonymous pages. They have
603 * PG_swapbacked flag cleared, to distinguish them from normal
606 ClearPageSwapBacked(page);
607 add_page_to_lru_list(page, lruvec, LRU_INACTIVE_FILE);
609 __count_vm_events(PGLAZYFREE, hpage_nr_pages(page));
610 count_memcg_page_event(page, PGLAZYFREE);
611 update_page_reclaim_stat(lruvec, 1, 0);
616 * Drain pages out of the cpu's pagevecs.
617 * Either "cpu" is the current CPU, and preemption has already been
618 * disabled; or "cpu" is being hot-unplugged, and is already dead.
620 void lru_add_drain_cpu(int cpu)
622 struct pagevec *pvec = &per_cpu(lru_pvecs.lru_add, cpu);
624 if (pagevec_count(pvec))
625 __pagevec_lru_add(pvec);
627 pvec = &per_cpu(lru_rotate.pvec, cpu);
628 if (pagevec_count(pvec)) {
631 /* No harm done if a racing interrupt already did this */
632 local_lock_irqsave(&lru_rotate.lock, flags);
633 pagevec_move_tail(pvec);
634 local_unlock_irqrestore(&lru_rotate.lock, flags);
637 pvec = &per_cpu(lru_pvecs.lru_deactivate_file, cpu);
638 if (pagevec_count(pvec))
639 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
641 pvec = &per_cpu(lru_pvecs.lru_deactivate, cpu);
642 if (pagevec_count(pvec))
643 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
645 pvec = &per_cpu(lru_pvecs.lru_lazyfree, cpu);
646 if (pagevec_count(pvec))
647 pagevec_lru_move_fn(pvec, lru_lazyfree_fn, NULL);
649 activate_page_drain(cpu);
653 * deactivate_file_page - forcefully deactivate a file page
654 * @page: page to deactivate
656 * This function hints the VM that @page is a good reclaim candidate,
657 * for example if its invalidation fails due to the page being dirty
658 * or under writeback.
660 void deactivate_file_page(struct page *page)
663 * In a workload with many unevictable page such as mprotect,
664 * unevictable page deactivation for accelerating reclaim is pointless.
666 if (PageUnevictable(page))
669 if (likely(get_page_unless_zero(page))) {
670 struct pagevec *pvec;
672 local_lock(&lru_pvecs.lock);
673 pvec = this_cpu_ptr(&lru_pvecs.lru_deactivate_file);
675 if (!pagevec_add(pvec, page) || PageCompound(page))
676 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
677 local_unlock(&lru_pvecs.lock);
682 * deactivate_page - deactivate a page
683 * @page: page to deactivate
685 * deactivate_page() moves @page to the inactive list if @page was on the active
686 * list and was not an unevictable page. This is done to accelerate the reclaim
689 void deactivate_page(struct page *page)
691 if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) {
692 struct pagevec *pvec;
694 local_lock(&lru_pvecs.lock);
695 pvec = this_cpu_ptr(&lru_pvecs.lru_deactivate);
697 if (!pagevec_add(pvec, page) || PageCompound(page))
698 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
699 local_unlock(&lru_pvecs.lock);
704 * mark_page_lazyfree - make an anon page lazyfree
705 * @page: page to deactivate
707 * mark_page_lazyfree() moves @page to the inactive file list.
708 * This is done to accelerate the reclaim of @page.
710 void mark_page_lazyfree(struct page *page)
712 if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) &&
713 !PageSwapCache(page) && !PageUnevictable(page)) {
714 struct pagevec *pvec;
716 local_lock(&lru_pvecs.lock);
717 pvec = this_cpu_ptr(&lru_pvecs.lru_lazyfree);
719 if (!pagevec_add(pvec, page) || PageCompound(page))
720 pagevec_lru_move_fn(pvec, lru_lazyfree_fn, NULL);
721 local_unlock(&lru_pvecs.lock);
725 void lru_add_drain(void)
727 local_lock(&lru_pvecs.lock);
728 lru_add_drain_cpu(smp_processor_id());
729 local_unlock(&lru_pvecs.lock);
732 void lru_add_drain_cpu_zone(struct zone *zone)
734 local_lock(&lru_pvecs.lock);
735 lru_add_drain_cpu(smp_processor_id());
736 drain_local_pages(zone);
737 local_unlock(&lru_pvecs.lock);
742 static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
744 static void lru_add_drain_per_cpu(struct work_struct *dummy)
750 * Doesn't need any cpu hotplug locking because we do rely on per-cpu
751 * kworkers being shut down before our page_alloc_cpu_dead callback is
752 * executed on the offlined cpu.
753 * Calling this function with cpu hotplug locks held can actually lead
754 * to obscure indirect dependencies via WQ context.
756 void lru_add_drain_all(void)
758 static seqcount_t seqcount = SEQCNT_ZERO(seqcount);
759 static DEFINE_MUTEX(lock);
760 static struct cpumask has_work;
764 * Make sure nobody triggers this path before mm_percpu_wq is fully
767 if (WARN_ON(!mm_percpu_wq))
770 seq = raw_read_seqcount_latch(&seqcount);
775 * Piggyback on drain started and finished while we waited for lock:
776 * all pages pended at the time of our enter were drained from vectors.
778 if (__read_seqcount_retry(&seqcount, seq))
781 raw_write_seqcount_latch(&seqcount);
783 cpumask_clear(&has_work);
785 for_each_online_cpu(cpu) {
786 struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
788 if (pagevec_count(&per_cpu(lru_pvecs.lru_add, cpu)) ||
789 pagevec_count(&per_cpu(lru_rotate.pvec, cpu)) ||
790 pagevec_count(&per_cpu(lru_pvecs.lru_deactivate_file, cpu)) ||
791 pagevec_count(&per_cpu(lru_pvecs.lru_deactivate, cpu)) ||
792 pagevec_count(&per_cpu(lru_pvecs.lru_lazyfree, cpu)) ||
793 need_activate_page_drain(cpu)) {
794 INIT_WORK(work, lru_add_drain_per_cpu);
795 queue_work_on(cpu, mm_percpu_wq, work);
796 cpumask_set_cpu(cpu, &has_work);
800 for_each_cpu(cpu, &has_work)
801 flush_work(&per_cpu(lru_add_drain_work, cpu));
807 void lru_add_drain_all(void)
814 * release_pages - batched put_page()
815 * @pages: array of pages to release
816 * @nr: number of pages
818 * Decrement the reference count on all the pages in @pages. If it
819 * fell to zero, remove the page from the LRU and free it.
821 void release_pages(struct page **pages, int nr)
824 LIST_HEAD(pages_to_free);
825 struct pglist_data *locked_pgdat = NULL;
826 struct lruvec *lruvec;
827 unsigned long uninitialized_var(flags);
828 unsigned int uninitialized_var(lock_batch);
830 for (i = 0; i < nr; i++) {
831 struct page *page = pages[i];
834 * Make sure the IRQ-safe lock-holding time does not get
835 * excessive with a continuous string of pages from the
836 * same pgdat. The lock is held only if pgdat != NULL.
838 if (locked_pgdat && ++lock_batch == SWAP_CLUSTER_MAX) {
839 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
843 if (is_huge_zero_page(page))
846 if (is_zone_device_page(page)) {
848 spin_unlock_irqrestore(&locked_pgdat->lru_lock,
853 * ZONE_DEVICE pages that return 'false' from
854 * put_devmap_managed_page() do not require special
855 * processing, and instead, expect a call to
856 * put_page_testzero().
858 if (page_is_devmap_managed(page)) {
859 put_devmap_managed_page(page);
864 page = compound_head(page);
865 if (!put_page_testzero(page))
868 if (PageCompound(page)) {
870 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
873 __put_compound_page(page);
878 struct pglist_data *pgdat = page_pgdat(page);
880 if (pgdat != locked_pgdat) {
882 spin_unlock_irqrestore(&locked_pgdat->lru_lock,
885 locked_pgdat = pgdat;
886 spin_lock_irqsave(&locked_pgdat->lru_lock, flags);
889 lruvec = mem_cgroup_page_lruvec(page, locked_pgdat);
890 VM_BUG_ON_PAGE(!PageLRU(page), page);
891 __ClearPageLRU(page);
892 del_page_from_lru_list(page, lruvec, page_off_lru(page));
895 /* Clear Active bit in case of parallel mark_page_accessed */
896 __ClearPageActive(page);
897 __ClearPageWaiters(page);
899 list_add(&page->lru, &pages_to_free);
902 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
904 mem_cgroup_uncharge_list(&pages_to_free);
905 free_unref_page_list(&pages_to_free);
907 EXPORT_SYMBOL(release_pages);
910 * The pages which we're about to release may be in the deferred lru-addition
911 * queues. That would prevent them from really being freed right now. That's
912 * OK from a correctness point of view but is inefficient - those pages may be
913 * cache-warm and we want to give them back to the page allocator ASAP.
915 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
916 * and __pagevec_lru_add_active() call release_pages() directly to avoid
919 void __pagevec_release(struct pagevec *pvec)
921 if (!pvec->percpu_pvec_drained) {
923 pvec->percpu_pvec_drained = true;
925 release_pages(pvec->pages, pagevec_count(pvec));
926 pagevec_reinit(pvec);
928 EXPORT_SYMBOL(__pagevec_release);
930 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
931 /* used by __split_huge_page_refcount() */
932 void lru_add_page_tail(struct page *page, struct page *page_tail,
933 struct lruvec *lruvec, struct list_head *list)
937 VM_BUG_ON_PAGE(!PageHead(page), page);
938 VM_BUG_ON_PAGE(PageCompound(page_tail), page);
939 VM_BUG_ON_PAGE(PageLRU(page_tail), page);
940 lockdep_assert_held(&lruvec_pgdat(lruvec)->lru_lock);
943 SetPageLRU(page_tail);
945 if (likely(PageLRU(page)))
946 list_add_tail(&page_tail->lru, &page->lru);
948 /* page reclaim is reclaiming a huge page */
950 list_add_tail(&page_tail->lru, list);
953 * Head page has not yet been counted, as an hpage,
954 * so we must account for each subpage individually.
956 * Put page_tail on the list at the correct position
957 * so they all end up in order.
959 add_page_to_lru_list_tail(page_tail, lruvec,
960 page_lru(page_tail));
963 if (!PageUnevictable(page))
964 update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
966 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
968 static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
972 int was_unevictable = TestClearPageUnevictable(page);
974 VM_BUG_ON_PAGE(PageLRU(page), page);
977 * Page becomes evictable in two ways:
978 * 1) Within LRU lock [munlock_vma_page() and __munlock_pagevec()].
979 * 2) Before acquiring LRU lock to put the page to correct LRU and then
980 * a) do PageLRU check with lock [check_move_unevictable_pages]
981 * b) do PageLRU check before lock [clear_page_mlock]
983 * (1) & (2a) are ok as LRU lock will serialize them. For (2b), we need
984 * following strict ordering:
986 * #0: __pagevec_lru_add_fn #1: clear_page_mlock
988 * SetPageLRU() TestClearPageMlocked()
989 * smp_mb() // explicit ordering // above provides strict
991 * PageMlocked() PageLRU()
994 * if '#1' does not observe setting of PG_lru by '#0' and fails
995 * isolation, the explicit barrier will make sure that page_evictable
996 * check will put the page in correct LRU. Without smp_mb(), SetPageLRU
997 * can be reordered after PageMlocked check and can make '#1' to fail
998 * the isolation of the page whose Mlocked bit is cleared (#0 is also
999 * looking at the same page) and the evictable page will be stranded
1000 * in an unevictable LRU.
1003 smp_mb__after_atomic();
1005 if (page_evictable(page)) {
1006 lru = page_lru(page);
1007 update_page_reclaim_stat(lruvec, page_is_file_lru(page),
1009 if (was_unevictable)
1010 count_vm_event(UNEVICTABLE_PGRESCUED);
1012 lru = LRU_UNEVICTABLE;
1013 ClearPageActive(page);
1014 SetPageUnevictable(page);
1015 if (!was_unevictable)
1016 count_vm_event(UNEVICTABLE_PGCULLED);
1019 add_page_to_lru_list(page, lruvec, lru);
1020 trace_mm_lru_insertion(page, lru);
1024 * Add the passed pages to the LRU, then drop the caller's refcount
1025 * on them. Reinitialises the caller's pagevec.
1027 void __pagevec_lru_add(struct pagevec *pvec)
1029 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
1033 * pagevec_lookup_entries - gang pagecache lookup
1034 * @pvec: Where the resulting entries are placed
1035 * @mapping: The address_space to search
1036 * @start: The starting entry index
1037 * @nr_entries: The maximum number of pages
1038 * @indices: The cache indices corresponding to the entries in @pvec
1040 * pagevec_lookup_entries() will search for and return a group of up
1041 * to @nr_pages pages and shadow entries in the mapping. All
1042 * entries are placed in @pvec. pagevec_lookup_entries() takes a
1043 * reference against actual pages in @pvec.
1045 * The search returns a group of mapping-contiguous entries with
1046 * ascending indexes. There may be holes in the indices due to
1047 * not-present entries.
1049 * Only one subpage of a Transparent Huge Page is returned in one call:
1050 * allowing truncate_inode_pages_range() to evict the whole THP without
1051 * cycling through a pagevec of extra references.
1053 * pagevec_lookup_entries() returns the number of entries which were
1056 unsigned pagevec_lookup_entries(struct pagevec *pvec,
1057 struct address_space *mapping,
1058 pgoff_t start, unsigned nr_entries,
1061 pvec->nr = find_get_entries(mapping, start, nr_entries,
1062 pvec->pages, indices);
1063 return pagevec_count(pvec);
1067 * pagevec_remove_exceptionals - pagevec exceptionals pruning
1068 * @pvec: The pagevec to prune
1070 * pagevec_lookup_entries() fills both pages and exceptional radix
1071 * tree entries into the pagevec. This function prunes all
1072 * exceptionals from @pvec without leaving holes, so that it can be
1073 * passed on to page-only pagevec operations.
1075 void pagevec_remove_exceptionals(struct pagevec *pvec)
1079 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
1080 struct page *page = pvec->pages[i];
1081 if (!xa_is_value(page))
1082 pvec->pages[j++] = page;
1088 * pagevec_lookup_range - gang pagecache lookup
1089 * @pvec: Where the resulting pages are placed
1090 * @mapping: The address_space to search
1091 * @start: The starting page index
1092 * @end: The final page index
1094 * pagevec_lookup_range() will search for & return a group of up to PAGEVEC_SIZE
1095 * pages in the mapping starting from index @start and upto index @end
1096 * (inclusive). The pages are placed in @pvec. pagevec_lookup() takes a
1097 * reference against the pages in @pvec.
1099 * The search returns a group of mapping-contiguous pages with ascending
1100 * indexes. There may be holes in the indices due to not-present pages. We
1101 * also update @start to index the next page for the traversal.
1103 * pagevec_lookup_range() returns the number of pages which were found. If this
1104 * number is smaller than PAGEVEC_SIZE, the end of specified range has been
1107 unsigned pagevec_lookup_range(struct pagevec *pvec,
1108 struct address_space *mapping, pgoff_t *start, pgoff_t end)
1110 pvec->nr = find_get_pages_range(mapping, start, end, PAGEVEC_SIZE,
1112 return pagevec_count(pvec);
1114 EXPORT_SYMBOL(pagevec_lookup_range);
1116 unsigned pagevec_lookup_range_tag(struct pagevec *pvec,
1117 struct address_space *mapping, pgoff_t *index, pgoff_t end,
1120 pvec->nr = find_get_pages_range_tag(mapping, index, end, tag,
1121 PAGEVEC_SIZE, pvec->pages);
1122 return pagevec_count(pvec);
1124 EXPORT_SYMBOL(pagevec_lookup_range_tag);
1126 unsigned pagevec_lookup_range_nr_tag(struct pagevec *pvec,
1127 struct address_space *mapping, pgoff_t *index, pgoff_t end,
1128 xa_mark_t tag, unsigned max_pages)
1130 pvec->nr = find_get_pages_range_tag(mapping, index, end, tag,
1131 min_t(unsigned int, max_pages, PAGEVEC_SIZE), pvec->pages);
1132 return pagevec_count(pvec);
1134 EXPORT_SYMBOL(pagevec_lookup_range_nr_tag);
1136 * Perform any setup for the swap system
1138 void __init swap_setup(void)
1140 unsigned long megs = totalram_pages() >> (20 - PAGE_SHIFT);
1142 /* Use a smaller cluster for small-memory machines */
1148 * Right now other parts of the system means that we
1149 * _really_ don't want to cluster much more
1153 #ifdef CONFIG_DEV_PAGEMAP_OPS
1154 void put_devmap_managed_page(struct page *page)
1158 if (WARN_ON_ONCE(!page_is_devmap_managed(page)))
1161 count = page_ref_dec_return(page);
1164 * devmap page refcounts are 1-based, rather than 0-based: if
1165 * refcount is 1, then the page is free and the refcount is
1166 * stable because nobody holds a reference on the page.
1169 free_devmap_managed_page(page);
1173 EXPORT_SYMBOL(put_devmap_managed_page);