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);
364 * Mark a page as having seen activity.
366 * inactive,unreferenced -> inactive,referenced
367 * inactive,referenced -> active,unreferenced
368 * active,unreferenced -> active,referenced
370 * When a newly allocated page is not yet visible, so safe for non-atomic ops,
371 * __SetPageReferenced(page) may be substituted for mark_page_accessed(page).
373 void mark_page_accessed(struct page *page)
375 page = compound_head(page);
377 if (!PageReferenced(page)) {
378 SetPageReferenced(page);
379 } else if (PageUnevictable(page)) {
381 * Unevictable pages are on the "LRU_UNEVICTABLE" list. But,
382 * this list is never rotated or maintained, so marking an
383 * evictable page accessed has no effect.
385 } else if (!PageActive(page)) {
387 * If the page is on the LRU, queue it for activation via
388 * activate_page_pvecs. Otherwise, assume the page is on a
389 * pagevec, mark it active and it'll be moved to the active
390 * LRU on the next drain.
395 __lru_cache_activate_page(page);
396 ClearPageReferenced(page);
397 if (page_is_file_cache(page))
398 workingset_activation(page);
400 if (page_is_idle(page))
401 clear_page_idle(page);
403 EXPORT_SYMBOL(mark_page_accessed);
405 static void __lru_cache_add(struct page *page)
407 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
410 if (!pagevec_add(pvec, page) || PageCompound(page))
411 __pagevec_lru_add(pvec);
412 put_cpu_var(lru_add_pvec);
416 * lru_cache_add_anon - add a page to the page lists
417 * @page: the page to add
419 void lru_cache_add_anon(struct page *page)
421 if (PageActive(page))
422 ClearPageActive(page);
423 __lru_cache_add(page);
426 void lru_cache_add_file(struct page *page)
428 if (PageActive(page))
429 ClearPageActive(page);
430 __lru_cache_add(page);
432 EXPORT_SYMBOL(lru_cache_add_file);
435 * lru_cache_add - add a page to a page list
436 * @page: the page to be added to the LRU.
438 * Queue the page for addition to the LRU via pagevec. The decision on whether
439 * to add the page to the [in]active [file|anon] list is deferred until the
440 * pagevec is drained. This gives a chance for the caller of lru_cache_add()
441 * have the page added to the active list using mark_page_accessed().
443 void lru_cache_add(struct page *page)
445 VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page);
446 VM_BUG_ON_PAGE(PageLRU(page), page);
447 __lru_cache_add(page);
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_cache(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);
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_cache(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);
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 * SwapBacked flag cleared to distinguish normal anonymous
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);
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_add_pvec, cpu);
598 if (pagevec_count(pvec))
599 __pagevec_lru_add(pvec);
601 pvec = &per_cpu(lru_rotate_pvecs, cpu);
602 if (pagevec_count(pvec)) {
605 /* No harm done if a racing interrupt already did this */
606 local_irq_save(flags);
607 pagevec_move_tail(pvec);
608 local_irq_restore(flags);
611 pvec = &per_cpu(lru_deactivate_file_pvecs, cpu);
612 if (pagevec_count(pvec))
613 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
615 pvec = &per_cpu(lru_deactivate_pvecs, cpu);
616 if (pagevec_count(pvec))
617 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
619 pvec = &per_cpu(lru_lazyfree_pvecs, 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 = &get_cpu_var(lru_deactivate_file_pvecs);
646 if (!pagevec_add(pvec, page) || PageCompound(page))
647 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
648 put_cpu_var(lru_deactivate_file_pvecs);
653 * deactivate_page - deactivate a page
654 * @page: page to deactivate
656 * deactivate_page() moves @page to the inactive list if @page was on the active
657 * list and was not an unevictable page. This is done to accelerate the reclaim
660 void deactivate_page(struct page *page)
662 if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) {
663 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
666 if (!pagevec_add(pvec, page) || PageCompound(page))
667 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
668 put_cpu_var(lru_deactivate_pvecs);
673 * mark_page_lazyfree - make an anon page lazyfree
674 * @page: page to deactivate
676 * mark_page_lazyfree() moves @page to the inactive file list.
677 * This is done to accelerate the reclaim of @page.
679 void mark_page_lazyfree(struct page *page)
681 if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) &&
682 !PageSwapCache(page) && !PageUnevictable(page)) {
683 struct pagevec *pvec = &get_cpu_var(lru_lazyfree_pvecs);
686 if (!pagevec_add(pvec, page) || PageCompound(page))
687 pagevec_lru_move_fn(pvec, lru_lazyfree_fn, NULL);
688 put_cpu_var(lru_lazyfree_pvecs);
692 void lru_add_drain(void)
694 lru_add_drain_cpu(get_cpu());
700 static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
702 static void lru_add_drain_per_cpu(struct work_struct *dummy)
708 * Doesn't need any cpu hotplug locking because we do rely on per-cpu
709 * kworkers being shut down before our page_alloc_cpu_dead callback is
710 * executed on the offlined cpu.
711 * Calling this function with cpu hotplug locks held can actually lead
712 * to obscure indirect dependencies via WQ context.
714 void lru_add_drain_all(void)
716 static seqcount_t seqcount = SEQCNT_ZERO(seqcount);
717 static DEFINE_MUTEX(lock);
718 static struct cpumask has_work;
722 * Make sure nobody triggers this path before mm_percpu_wq is fully
725 if (WARN_ON(!mm_percpu_wq))
728 seq = raw_read_seqcount_latch(&seqcount);
733 * Piggyback on drain started and finished while we waited for lock:
734 * all pages pended at the time of our enter were drained from vectors.
736 if (__read_seqcount_retry(&seqcount, seq))
739 raw_write_seqcount_latch(&seqcount);
741 cpumask_clear(&has_work);
743 for_each_online_cpu(cpu) {
744 struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
746 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
747 pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
748 pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) ||
749 pagevec_count(&per_cpu(lru_deactivate_pvecs, cpu)) ||
750 pagevec_count(&per_cpu(lru_lazyfree_pvecs, cpu)) ||
751 need_activate_page_drain(cpu)) {
752 INIT_WORK(work, lru_add_drain_per_cpu);
753 queue_work_on(cpu, mm_percpu_wq, work);
754 cpumask_set_cpu(cpu, &has_work);
758 for_each_cpu(cpu, &has_work)
759 flush_work(&per_cpu(lru_add_drain_work, cpu));
765 void lru_add_drain_all(void)
772 * release_pages - batched put_page()
773 * @pages: array of pages to release
774 * @nr: number of pages
776 * Decrement the reference count on all the pages in @pages. If it
777 * fell to zero, remove the page from the LRU and free it.
779 void release_pages(struct page **pages, int nr)
782 LIST_HEAD(pages_to_free);
783 struct pglist_data *locked_pgdat = NULL;
784 struct lruvec *lruvec;
785 unsigned long uninitialized_var(flags);
786 unsigned int uninitialized_var(lock_batch);
788 for (i = 0; i < nr; i++) {
789 struct page *page = pages[i];
792 * Make sure the IRQ-safe lock-holding time does not get
793 * excessive with a continuous string of pages from the
794 * same pgdat. The lock is held only if pgdat != NULL.
796 if (locked_pgdat && ++lock_batch == SWAP_CLUSTER_MAX) {
797 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
801 if (is_huge_zero_page(page))
804 if (is_zone_device_page(page)) {
806 spin_unlock_irqrestore(&locked_pgdat->lru_lock,
811 * ZONE_DEVICE pages that return 'false' from
812 * put_devmap_managed_page() do not require special
813 * processing, and instead, expect a call to
814 * put_page_testzero().
816 if (put_devmap_managed_page(page))
820 page = compound_head(page);
821 if (!put_page_testzero(page))
824 if (PageCompound(page)) {
826 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
829 __put_compound_page(page);
834 struct pglist_data *pgdat = page_pgdat(page);
836 if (pgdat != locked_pgdat) {
838 spin_unlock_irqrestore(&locked_pgdat->lru_lock,
841 locked_pgdat = pgdat;
842 spin_lock_irqsave(&locked_pgdat->lru_lock, flags);
845 lruvec = mem_cgroup_page_lruvec(page, locked_pgdat);
846 VM_BUG_ON_PAGE(!PageLRU(page), page);
847 __ClearPageLRU(page);
848 del_page_from_lru_list(page, lruvec, page_off_lru(page));
851 /* Clear Active bit in case of parallel mark_page_accessed */
852 __ClearPageActive(page);
853 __ClearPageWaiters(page);
855 list_add(&page->lru, &pages_to_free);
858 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
860 mem_cgroup_uncharge_list(&pages_to_free);
861 free_unref_page_list(&pages_to_free);
863 EXPORT_SYMBOL(release_pages);
866 * The pages which we're about to release may be in the deferred lru-addition
867 * queues. That would prevent them from really being freed right now. That's
868 * OK from a correctness point of view but is inefficient - those pages may be
869 * cache-warm and we want to give them back to the page allocator ASAP.
871 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
872 * and __pagevec_lru_add_active() call release_pages() directly to avoid
875 void __pagevec_release(struct pagevec *pvec)
877 if (!pvec->percpu_pvec_drained) {
879 pvec->percpu_pvec_drained = true;
881 release_pages(pvec->pages, pagevec_count(pvec));
882 pagevec_reinit(pvec);
884 EXPORT_SYMBOL(__pagevec_release);
886 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
887 /* used by __split_huge_page_refcount() */
888 void lru_add_page_tail(struct page *page, struct page *page_tail,
889 struct lruvec *lruvec, struct list_head *list)
893 VM_BUG_ON_PAGE(!PageHead(page), page);
894 VM_BUG_ON_PAGE(PageCompound(page_tail), page);
895 VM_BUG_ON_PAGE(PageLRU(page_tail), page);
896 lockdep_assert_held(&lruvec_pgdat(lruvec)->lru_lock);
899 SetPageLRU(page_tail);
901 if (likely(PageLRU(page)))
902 list_add_tail(&page_tail->lru, &page->lru);
904 /* page reclaim is reclaiming a huge page */
906 list_add_tail(&page_tail->lru, list);
909 * Head page has not yet been counted, as an hpage,
910 * so we must account for each subpage individually.
912 * Put page_tail on the list at the correct position
913 * so they all end up in order.
915 add_page_to_lru_list_tail(page_tail, lruvec,
916 page_lru(page_tail));
919 if (!PageUnevictable(page))
920 update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
922 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
924 static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
928 int was_unevictable = TestClearPageUnevictable(page);
930 VM_BUG_ON_PAGE(PageLRU(page), page);
934 * Page becomes evictable in two ways:
935 * 1) Within LRU lock [munlock_vma_page() and __munlock_pagevec()].
936 * 2) Before acquiring LRU lock to put the page to correct LRU and then
937 * a) do PageLRU check with lock [check_move_unevictable_pages]
938 * b) do PageLRU check before lock [clear_page_mlock]
940 * (1) & (2a) are ok as LRU lock will serialize them. For (2b), we need
941 * following strict ordering:
943 * #0: __pagevec_lru_add_fn #1: clear_page_mlock
945 * SetPageLRU() TestClearPageMlocked()
946 * smp_mb() // explicit ordering // above provides strict
948 * PageMlocked() PageLRU()
951 * if '#1' does not observe setting of PG_lru by '#0' and fails
952 * isolation, the explicit barrier will make sure that page_evictable
953 * check will put the page in correct LRU. Without smp_mb(), SetPageLRU
954 * can be reordered after PageMlocked check and can make '#1' to fail
955 * the isolation of the page whose Mlocked bit is cleared (#0 is also
956 * looking at the same page) and the evictable page will be stranded
957 * in an unevictable LRU.
961 if (page_evictable(page)) {
962 lru = page_lru(page);
963 update_page_reclaim_stat(lruvec, page_is_file_cache(page),
966 count_vm_event(UNEVICTABLE_PGRESCUED);
968 lru = LRU_UNEVICTABLE;
969 ClearPageActive(page);
970 SetPageUnevictable(page);
971 if (!was_unevictable)
972 count_vm_event(UNEVICTABLE_PGCULLED);
975 add_page_to_lru_list(page, lruvec, lru);
976 trace_mm_lru_insertion(page, lru);
980 * Add the passed pages to the LRU, then drop the caller's refcount
981 * on them. Reinitialises the caller's pagevec.
983 void __pagevec_lru_add(struct pagevec *pvec)
985 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
987 EXPORT_SYMBOL(__pagevec_lru_add);
990 * pagevec_lookup_entries - gang pagecache lookup
991 * @pvec: Where the resulting entries are placed
992 * @mapping: The address_space to search
993 * @start: The starting entry index
994 * @nr_entries: The maximum number of pages
995 * @indices: The cache indices corresponding to the entries in @pvec
997 * pagevec_lookup_entries() will search for and return a group of up
998 * to @nr_pages pages and shadow entries in the mapping. All
999 * entries are placed in @pvec. pagevec_lookup_entries() takes a
1000 * reference against actual pages in @pvec.
1002 * The search returns a group of mapping-contiguous entries with
1003 * ascending indexes. There may be holes in the indices due to
1004 * not-present entries.
1006 * pagevec_lookup_entries() returns the number of entries which were
1009 unsigned pagevec_lookup_entries(struct pagevec *pvec,
1010 struct address_space *mapping,
1011 pgoff_t start, unsigned nr_entries,
1014 pvec->nr = find_get_entries(mapping, start, nr_entries,
1015 pvec->pages, indices);
1016 return pagevec_count(pvec);
1020 * pagevec_remove_exceptionals - pagevec exceptionals pruning
1021 * @pvec: The pagevec to prune
1023 * pagevec_lookup_entries() fills both pages and exceptional radix
1024 * tree entries into the pagevec. This function prunes all
1025 * exceptionals from @pvec without leaving holes, so that it can be
1026 * passed on to page-only pagevec operations.
1028 void pagevec_remove_exceptionals(struct pagevec *pvec)
1032 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
1033 struct page *page = pvec->pages[i];
1034 if (!xa_is_value(page))
1035 pvec->pages[j++] = page;
1041 * pagevec_lookup_range - gang pagecache lookup
1042 * @pvec: Where the resulting pages are placed
1043 * @mapping: The address_space to search
1044 * @start: The starting page index
1045 * @end: The final page index
1047 * pagevec_lookup_range() will search for & return a group of up to PAGEVEC_SIZE
1048 * pages in the mapping starting from index @start and upto index @end
1049 * (inclusive). The pages are placed in @pvec. pagevec_lookup() takes a
1050 * reference against the pages in @pvec.
1052 * The search returns a group of mapping-contiguous pages with ascending
1053 * indexes. There may be holes in the indices due to not-present pages. We
1054 * also update @start to index the next page for the traversal.
1056 * pagevec_lookup_range() returns the number of pages which were found. If this
1057 * number is smaller than PAGEVEC_SIZE, the end of specified range has been
1060 unsigned pagevec_lookup_range(struct pagevec *pvec,
1061 struct address_space *mapping, pgoff_t *start, pgoff_t end)
1063 pvec->nr = find_get_pages_range(mapping, start, end, PAGEVEC_SIZE,
1065 return pagevec_count(pvec);
1067 EXPORT_SYMBOL(pagevec_lookup_range);
1069 unsigned pagevec_lookup_range_tag(struct pagevec *pvec,
1070 struct address_space *mapping, pgoff_t *index, pgoff_t end,
1073 pvec->nr = find_get_pages_range_tag(mapping, index, end, tag,
1074 PAGEVEC_SIZE, pvec->pages);
1075 return pagevec_count(pvec);
1077 EXPORT_SYMBOL(pagevec_lookup_range_tag);
1079 unsigned pagevec_lookup_range_nr_tag(struct pagevec *pvec,
1080 struct address_space *mapping, pgoff_t *index, pgoff_t end,
1081 xa_mark_t tag, unsigned max_pages)
1083 pvec->nr = find_get_pages_range_tag(mapping, index, end, tag,
1084 min_t(unsigned int, max_pages, PAGEVEC_SIZE), pvec->pages);
1085 return pagevec_count(pvec);
1087 EXPORT_SYMBOL(pagevec_lookup_range_nr_tag);
1089 * Perform any setup for the swap system
1091 void __init swap_setup(void)
1093 unsigned long megs = totalram_pages() >> (20 - PAGE_SHIFT);
1095 /* Use a smaller cluster for small-memory machines */
1101 * Right now other parts of the system means that we
1102 * _really_ don't want to cluster much more