1457c8996SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * linux/mm/swap.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds 61da177e4SLinus Torvalds */ 71da177e4SLinus Torvalds 81da177e4SLinus Torvalds /* 9183ff22bSSimon Arlott * This file contains the default values for the operation of the 101da177e4SLinus Torvalds * Linux VM subsystem. Fine-tuning documentation can be found in 1157043247SMauro Carvalho Chehab * Documentation/admin-guide/sysctl/vm.rst. 121da177e4SLinus Torvalds * Started 18.12.91 131da177e4SLinus Torvalds * Swap aging added 23.2.95, Stephen Tweedie. 141da177e4SLinus Torvalds * Buffermem limits added 12.3.98, Rik van Riel. 151da177e4SLinus Torvalds */ 161da177e4SLinus Torvalds 171da177e4SLinus Torvalds #include <linux/mm.h> 181da177e4SLinus Torvalds #include <linux/sched.h> 191da177e4SLinus Torvalds #include <linux/kernel_stat.h> 201da177e4SLinus Torvalds #include <linux/swap.h> 211da177e4SLinus Torvalds #include <linux/mman.h> 221da177e4SLinus Torvalds #include <linux/pagemap.h> 231da177e4SLinus Torvalds #include <linux/pagevec.h> 241da177e4SLinus Torvalds #include <linux/init.h> 25b95f1b31SPaul Gortmaker #include <linux/export.h> 261da177e4SLinus Torvalds #include <linux/mm_inline.h> 271da177e4SLinus Torvalds #include <linux/percpu_counter.h> 283565fce3SDan Williams #include <linux/memremap.h> 291da177e4SLinus Torvalds #include <linux/percpu.h> 301da177e4SLinus Torvalds #include <linux/cpu.h> 311da177e4SLinus Torvalds #include <linux/notifier.h> 32e0bf68ddSPeter Zijlstra #include <linux/backing-dev.h> 3366e1707bSBalbir Singh #include <linux/memcontrol.h> 345a0e3ad6STejun Heo #include <linux/gfp.h> 35a27bb332SKent Overstreet #include <linux/uio.h> 36822fc613SNaoya Horiguchi #include <linux/hugetlb.h> 3733c3fc71SVladimir Davydov #include <linux/page_idle.h> 38b01b2141SIngo Molnar #include <linux/local_lock.h> 398cc621d2SMinchan Kim #include <linux/buffer_head.h> 401da177e4SLinus Torvalds 4164d6519dSLee Schermerhorn #include "internal.h" 4264d6519dSLee Schermerhorn 43c6286c98SMel Gorman #define CREATE_TRACE_POINTS 44c6286c98SMel Gorman #include <trace/events/pagemap.h> 45c6286c98SMel Gorman 461da177e4SLinus Torvalds /* How many pages do we try to swap or page in/out together? */ 471da177e4SLinus Torvalds int page_cluster; 481da177e4SLinus Torvalds 49b01b2141SIngo Molnar /* Protecting only lru_rotate.pvec which requires disabling interrupts */ 50b01b2141SIngo Molnar struct lru_rotate { 51b01b2141SIngo Molnar local_lock_t lock; 52b01b2141SIngo Molnar struct pagevec pvec; 53b01b2141SIngo Molnar }; 54b01b2141SIngo Molnar static DEFINE_PER_CPU(struct lru_rotate, lru_rotate) = { 55b01b2141SIngo Molnar .lock = INIT_LOCAL_LOCK(lock), 56b01b2141SIngo Molnar }; 57b01b2141SIngo Molnar 58b01b2141SIngo Molnar /* 59b01b2141SIngo Molnar * The following struct pagevec are grouped together because they are protected 60b01b2141SIngo Molnar * by disabling preemption (and interrupts remain enabled). 61b01b2141SIngo Molnar */ 62b01b2141SIngo Molnar struct lru_pvecs { 63b01b2141SIngo Molnar local_lock_t lock; 64b01b2141SIngo Molnar struct pagevec lru_add; 65b01b2141SIngo Molnar struct pagevec lru_deactivate_file; 66b01b2141SIngo Molnar struct pagevec lru_deactivate; 67b01b2141SIngo Molnar struct pagevec lru_lazyfree; 68a4a921aaSMing Li #ifdef CONFIG_SMP 69b01b2141SIngo Molnar struct pagevec activate_page; 70a4a921aaSMing Li #endif 71b01b2141SIngo Molnar }; 72b01b2141SIngo Molnar static DEFINE_PER_CPU(struct lru_pvecs, lru_pvecs) = { 73b01b2141SIngo Molnar .lock = INIT_LOCAL_LOCK(lock), 74b01b2141SIngo Molnar }; 75902aaed0SHisashi Hifumi 76b221385bSAdrian Bunk /* 77b109b870SHugh Dickins * This path almost never happens for VM activity - pages are normally freed 78b109b870SHugh Dickins * via pagevecs. But it gets used by networking - and for compound pages. 79b221385bSAdrian Bunk */ 80920c7a5dSHarvey Harrison static void __page_cache_release(struct page *page) 81b221385bSAdrian Bunk { 82b221385bSAdrian Bunk if (PageLRU(page)) { 83e809c3feSMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 84fa9add64SHugh Dickins struct lruvec *lruvec; 85fa9add64SHugh Dickins unsigned long flags; 86b221385bSAdrian Bunk 87e809c3feSMatthew Wilcox (Oracle) lruvec = folio_lruvec_lock_irqsave(folio, &flags); 8846ae6b2cSYu Zhao del_page_from_lru_list(page, lruvec); 8987560179SYu Zhao __clear_page_lru_flags(page); 906168d0daSAlex Shi unlock_page_lruvec_irqrestore(lruvec, flags); 91b221385bSAdrian Bunk } 92b109b870SHugh Dickins /* See comment on PageMlocked in release_pages() */ 93b109b870SHugh Dickins if (unlikely(PageMlocked(page))) { 94b109b870SHugh Dickins int nr_pages = thp_nr_pages(page); 95b109b870SHugh Dickins 96b109b870SHugh Dickins __ClearPageMlocked(page); 97b109b870SHugh Dickins mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages); 98b109b870SHugh Dickins count_vm_events(UNEVICTABLE_PGCLEARED, nr_pages); 99b109b870SHugh Dickins } 10091807063SAndrea Arcangeli } 10191807063SAndrea Arcangeli 10291807063SAndrea Arcangeli static void __put_single_page(struct page *page) 10391807063SAndrea Arcangeli { 10491807063SAndrea Arcangeli __page_cache_release(page); 105bbc6b703SMatthew Wilcox (Oracle) mem_cgroup_uncharge(page_folio(page)); 10644042b44SMel Gorman free_unref_page(page, 0); 107b221385bSAdrian Bunk } 108b221385bSAdrian Bunk 10991807063SAndrea Arcangeli static void __put_compound_page(struct page *page) 11091807063SAndrea Arcangeli { 111822fc613SNaoya Horiguchi /* 112822fc613SNaoya Horiguchi * __page_cache_release() is supposed to be called for thp, not for 113822fc613SNaoya Horiguchi * hugetlb. This is because hugetlb page does never have PageLRU set 114822fc613SNaoya Horiguchi * (it's never listed to any LRU lists) and no memcg routines should 115822fc613SNaoya Horiguchi * be called for hugetlb (it has a separate hugetlb_cgroup.) 116822fc613SNaoya Horiguchi */ 117822fc613SNaoya Horiguchi if (!PageHuge(page)) 11891807063SAndrea Arcangeli __page_cache_release(page); 119ff45fc3cSMatthew Wilcox (Oracle) destroy_compound_page(page); 12091807063SAndrea Arcangeli } 12191807063SAndrea Arcangeli 122ddc58f27SKirill A. Shutemov void __put_page(struct page *page) 123c747ce79SJianyu Zhan { 12427674ef6SChristoph Hellwig if (unlikely(is_zone_device_page(page))) 12527674ef6SChristoph Hellwig free_zone_device_page(page); 12627674ef6SChristoph Hellwig else if (unlikely(PageCompound(page))) 12726296ad2SAndrew Morton __put_compound_page(page); 12826296ad2SAndrew Morton else 12926296ad2SAndrew Morton __put_single_page(page); 13026296ad2SAndrew Morton } 131ddc58f27SKirill A. Shutemov EXPORT_SYMBOL(__put_page); 13270b50f94SAndrea Arcangeli 1331d7ea732SAlexander Zarochentsev /** 1347682486bSRandy Dunlap * put_pages_list() - release a list of pages 1357682486bSRandy Dunlap * @pages: list of pages threaded on page->lru 1361d7ea732SAlexander Zarochentsev * 137988c69f1SMatthew Wilcox (Oracle) * Release a list of pages which are strung together on page.lru. 1381d7ea732SAlexander Zarochentsev */ 1391d7ea732SAlexander Zarochentsev void put_pages_list(struct list_head *pages) 1401d7ea732SAlexander Zarochentsev { 141988c69f1SMatthew Wilcox (Oracle) struct page *page, *next; 1421d7ea732SAlexander Zarochentsev 143988c69f1SMatthew Wilcox (Oracle) list_for_each_entry_safe(page, next, pages, lru) { 144988c69f1SMatthew Wilcox (Oracle) if (!put_page_testzero(page)) { 145988c69f1SMatthew Wilcox (Oracle) list_del(&page->lru); 146988c69f1SMatthew Wilcox (Oracle) continue; 1471d7ea732SAlexander Zarochentsev } 148988c69f1SMatthew Wilcox (Oracle) if (PageHead(page)) { 149988c69f1SMatthew Wilcox (Oracle) list_del(&page->lru); 150988c69f1SMatthew Wilcox (Oracle) __put_compound_page(page); 151988c69f1SMatthew Wilcox (Oracle) continue; 152988c69f1SMatthew Wilcox (Oracle) } 153988c69f1SMatthew Wilcox (Oracle) /* Cannot be PageLRU because it's passed to us using the lru */ 154988c69f1SMatthew Wilcox (Oracle) } 155988c69f1SMatthew Wilcox (Oracle) 156988c69f1SMatthew Wilcox (Oracle) free_unref_page_list(pages); 1573cd018b4SMatthew Wilcox INIT_LIST_HEAD(pages); 1581d7ea732SAlexander Zarochentsev } 1591d7ea732SAlexander Zarochentsev EXPORT_SYMBOL(put_pages_list); 1601d7ea732SAlexander Zarochentsev 16118022c5dSMel Gorman /* 16218022c5dSMel Gorman * get_kernel_pages() - pin kernel pages in memory 16318022c5dSMel Gorman * @kiov: An array of struct kvec structures 16418022c5dSMel Gorman * @nr_segs: number of segments to pin 16518022c5dSMel Gorman * @write: pinning for read/write, currently ignored 16618022c5dSMel Gorman * @pages: array that receives pointers to the pages pinned. 16718022c5dSMel Gorman * Should be at least nr_segs long. 16818022c5dSMel Gorman * 169133d2743SMiaohe Lin * Returns number of pages pinned. This may be fewer than the number requested. 170133d2743SMiaohe Lin * If nr_segs is 0 or negative, returns 0. If no pages were pinned, returns 0. 171133d2743SMiaohe Lin * Each page returned must be released with a put_page() call when it is 172133d2743SMiaohe Lin * finished with. 17318022c5dSMel Gorman */ 17418022c5dSMel Gorman int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write, 17518022c5dSMel Gorman struct page **pages) 17618022c5dSMel Gorman { 17718022c5dSMel Gorman int seg; 17818022c5dSMel Gorman 17918022c5dSMel Gorman for (seg = 0; seg < nr_segs; seg++) { 18018022c5dSMel Gorman if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE)) 18118022c5dSMel Gorman return seg; 18218022c5dSMel Gorman 1835a178119SMel Gorman pages[seg] = kmap_to_page(kiov[seg].iov_base); 18409cbfeafSKirill A. Shutemov get_page(pages[seg]); 18518022c5dSMel Gorman } 18618022c5dSMel Gorman 18718022c5dSMel Gorman return seg; 18818022c5dSMel Gorman } 18918022c5dSMel Gorman EXPORT_SYMBOL_GPL(get_kernel_pages); 19018022c5dSMel Gorman 1913dd7ae8eSShaohua Li static void pagevec_lru_move_fn(struct pagevec *pvec, 192c7c7b80cSAlex Shi void (*move_fn)(struct page *page, struct lruvec *lruvec)) 193902aaed0SHisashi Hifumi { 194902aaed0SHisashi Hifumi int i; 1956168d0daSAlex Shi struct lruvec *lruvec = NULL; 1963dd7ae8eSShaohua Li unsigned long flags = 0; 197902aaed0SHisashi Hifumi 198902aaed0SHisashi Hifumi for (i = 0; i < pagevec_count(pvec); i++) { 199902aaed0SHisashi Hifumi struct page *page = pvec->pages[i]; 2000de340cbSMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 2013dd7ae8eSShaohua Li 202fc574c23SAlex Shi /* block memcg migration during page moving between lru */ 203fc574c23SAlex Shi if (!TestClearPageLRU(page)) 204fc574c23SAlex Shi continue; 205fc574c23SAlex Shi 2060de340cbSMatthew Wilcox (Oracle) lruvec = folio_lruvec_relock_irqsave(folio, lruvec, &flags); 207c7c7b80cSAlex Shi (*move_fn)(page, lruvec); 208fc574c23SAlex Shi 209fc574c23SAlex Shi SetPageLRU(page); 2103dd7ae8eSShaohua Li } 2116168d0daSAlex Shi if (lruvec) 2126168d0daSAlex Shi unlock_page_lruvec_irqrestore(lruvec, flags); 213c6f92f9fSMel Gorman release_pages(pvec->pages, pvec->nr); 2143dd7ae8eSShaohua Li pagevec_reinit(pvec); 2153dd7ae8eSShaohua Li } 2163dd7ae8eSShaohua Li 217c7c7b80cSAlex Shi static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec) 2183dd7ae8eSShaohua Li { 219575ced1cSMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 220575ced1cSMatthew Wilcox (Oracle) 221575ced1cSMatthew Wilcox (Oracle) if (!folio_test_unevictable(folio)) { 222575ced1cSMatthew Wilcox (Oracle) lruvec_del_folio(lruvec, folio); 223575ced1cSMatthew Wilcox (Oracle) folio_clear_active(folio); 224575ced1cSMatthew Wilcox (Oracle) lruvec_add_folio_tail(lruvec, folio); 225575ced1cSMatthew Wilcox (Oracle) __count_vm_events(PGROTATED, folio_nr_pages(folio)); 226902aaed0SHisashi Hifumi } 227902aaed0SHisashi Hifumi } 2283dd7ae8eSShaohua Li 229d479960eSMinchan Kim /* return true if pagevec needs to drain */ 230d479960eSMinchan Kim static bool pagevec_add_and_need_flush(struct pagevec *pvec, struct page *page) 231d479960eSMinchan Kim { 232d479960eSMinchan Kim bool ret = false; 233d479960eSMinchan Kim 234d479960eSMinchan Kim if (!pagevec_add(pvec, page) || PageCompound(page) || 235d479960eSMinchan Kim lru_cache_disabled()) 236d479960eSMinchan Kim ret = true; 237d479960eSMinchan Kim 238d479960eSMinchan Kim return ret; 239d479960eSMinchan Kim } 240d479960eSMinchan Kim 2413dd7ae8eSShaohua Li /* 242575ced1cSMatthew Wilcox (Oracle) * Writeback is about to end against a folio which has been marked for 243575ced1cSMatthew Wilcox (Oracle) * immediate reclaim. If it still appears to be reclaimable, move it 244575ced1cSMatthew Wilcox (Oracle) * to the tail of the inactive list. 245c7c7b80cSAlex Shi * 246575ced1cSMatthew Wilcox (Oracle) * folio_rotate_reclaimable() must disable IRQs, to prevent nasty races. 2471da177e4SLinus Torvalds */ 248575ced1cSMatthew Wilcox (Oracle) void folio_rotate_reclaimable(struct folio *folio) 2491da177e4SLinus Torvalds { 250575ced1cSMatthew Wilcox (Oracle) if (!folio_test_locked(folio) && !folio_test_dirty(folio) && 251575ced1cSMatthew Wilcox (Oracle) !folio_test_unevictable(folio) && folio_test_lru(folio)) { 252902aaed0SHisashi Hifumi struct pagevec *pvec; 2531da177e4SLinus Torvalds unsigned long flags; 2541da177e4SLinus Torvalds 255575ced1cSMatthew Wilcox (Oracle) folio_get(folio); 256b01b2141SIngo Molnar local_lock_irqsave(&lru_rotate.lock, flags); 257b01b2141SIngo Molnar pvec = this_cpu_ptr(&lru_rotate.pvec); 258575ced1cSMatthew Wilcox (Oracle) if (pagevec_add_and_need_flush(pvec, &folio->page)) 259c7c7b80cSAlex Shi pagevec_lru_move_fn(pvec, pagevec_move_tail_fn); 260b01b2141SIngo Molnar local_unlock_irqrestore(&lru_rotate.lock, flags); 261ac6aadb2SMiklos Szeredi } 2621da177e4SLinus Torvalds } 2631da177e4SLinus Torvalds 26496f8bf4fSJohannes Weiner void lru_note_cost(struct lruvec *lruvec, bool file, unsigned int nr_pages) 2653e2f41f1SKOSAKI Motohiro { 2667cf111bcSJohannes Weiner do { 2677cf111bcSJohannes Weiner unsigned long lrusize; 2687cf111bcSJohannes Weiner 2696168d0daSAlex Shi /* 2706168d0daSAlex Shi * Hold lruvec->lru_lock is safe here, since 2716168d0daSAlex Shi * 1) The pinned lruvec in reclaim, or 2726168d0daSAlex Shi * 2) From a pre-LRU page during refault (which also holds the 2736168d0daSAlex Shi * rcu lock, so would be safe even if the page was on the LRU 2746168d0daSAlex Shi * and could move simultaneously to a new lruvec). 2756168d0daSAlex Shi */ 2766168d0daSAlex Shi spin_lock_irq(&lruvec->lru_lock); 2777cf111bcSJohannes Weiner /* Record cost event */ 27896f8bf4fSJohannes Weiner if (file) 27996f8bf4fSJohannes Weiner lruvec->file_cost += nr_pages; 2801431d4d1SJohannes Weiner else 28196f8bf4fSJohannes Weiner lruvec->anon_cost += nr_pages; 2827cf111bcSJohannes Weiner 2837cf111bcSJohannes Weiner /* 2847cf111bcSJohannes Weiner * Decay previous events 2857cf111bcSJohannes Weiner * 2867cf111bcSJohannes Weiner * Because workloads change over time (and to avoid 2877cf111bcSJohannes Weiner * overflow) we keep these statistics as a floating 2887cf111bcSJohannes Weiner * average, which ends up weighing recent refaults 2897cf111bcSJohannes Weiner * more than old ones. 2907cf111bcSJohannes Weiner */ 2917cf111bcSJohannes Weiner lrusize = lruvec_page_state(lruvec, NR_INACTIVE_ANON) + 2927cf111bcSJohannes Weiner lruvec_page_state(lruvec, NR_ACTIVE_ANON) + 2937cf111bcSJohannes Weiner lruvec_page_state(lruvec, NR_INACTIVE_FILE) + 2947cf111bcSJohannes Weiner lruvec_page_state(lruvec, NR_ACTIVE_FILE); 2957cf111bcSJohannes Weiner 2967cf111bcSJohannes Weiner if (lruvec->file_cost + lruvec->anon_cost > lrusize / 4) { 2977cf111bcSJohannes Weiner lruvec->file_cost /= 2; 2987cf111bcSJohannes Weiner lruvec->anon_cost /= 2; 2997cf111bcSJohannes Weiner } 3006168d0daSAlex Shi spin_unlock_irq(&lruvec->lru_lock); 3017cf111bcSJohannes Weiner } while ((lruvec = parent_lruvec(lruvec))); 3023e2f41f1SKOSAKI Motohiro } 3033e2f41f1SKOSAKI Motohiro 3040995d7e5SMatthew Wilcox (Oracle) void lru_note_cost_folio(struct folio *folio) 30596f8bf4fSJohannes Weiner { 3060995d7e5SMatthew Wilcox (Oracle) lru_note_cost(folio_lruvec(folio), folio_is_file_lru(folio), 3070995d7e5SMatthew Wilcox (Oracle) folio_nr_pages(folio)); 30896f8bf4fSJohannes Weiner } 30996f8bf4fSJohannes Weiner 310f2d27392SMatthew Wilcox (Oracle) static void __folio_activate(struct folio *folio, struct lruvec *lruvec) 311744ed144SShaohua Li { 312f2d27392SMatthew Wilcox (Oracle) if (!folio_test_active(folio) && !folio_test_unevictable(folio)) { 313f2d27392SMatthew Wilcox (Oracle) long nr_pages = folio_nr_pages(folio); 314744ed144SShaohua Li 315f2d27392SMatthew Wilcox (Oracle) lruvec_del_folio(lruvec, folio); 316f2d27392SMatthew Wilcox (Oracle) folio_set_active(folio); 317f2d27392SMatthew Wilcox (Oracle) lruvec_add_folio(lruvec, folio); 318f2d27392SMatthew Wilcox (Oracle) trace_mm_lru_activate(folio); 3197a608572SLinus Torvalds 32021e330fcSShakeel Butt __count_vm_events(PGACTIVATE, nr_pages); 32121e330fcSShakeel Butt __count_memcg_events(lruvec_memcg(lruvec), PGACTIVATE, 32221e330fcSShakeel Butt nr_pages); 323744ed144SShaohua Li } 324eb709b0dSShaohua Li } 325eb709b0dSShaohua Li 326eb709b0dSShaohua Li #ifdef CONFIG_SMP 327f2d27392SMatthew Wilcox (Oracle) static void __activate_page(struct page *page, struct lruvec *lruvec) 328f2d27392SMatthew Wilcox (Oracle) { 329f2d27392SMatthew Wilcox (Oracle) return __folio_activate(page_folio(page), lruvec); 330f2d27392SMatthew Wilcox (Oracle) } 331f2d27392SMatthew Wilcox (Oracle) 332eb709b0dSShaohua Li static void activate_page_drain(int cpu) 333eb709b0dSShaohua Li { 334b01b2141SIngo Molnar struct pagevec *pvec = &per_cpu(lru_pvecs.activate_page, cpu); 335eb709b0dSShaohua Li 336eb709b0dSShaohua Li if (pagevec_count(pvec)) 337c7c7b80cSAlex Shi pagevec_lru_move_fn(pvec, __activate_page); 338eb709b0dSShaohua Li } 339eb709b0dSShaohua Li 3405fbc4616SChris Metcalf static bool need_activate_page_drain(int cpu) 3415fbc4616SChris Metcalf { 342b01b2141SIngo Molnar return pagevec_count(&per_cpu(lru_pvecs.activate_page, cpu)) != 0; 3435fbc4616SChris Metcalf } 3445fbc4616SChris Metcalf 345f2d27392SMatthew Wilcox (Oracle) static void folio_activate(struct folio *folio) 346eb709b0dSShaohua Li { 347f2d27392SMatthew Wilcox (Oracle) if (folio_test_lru(folio) && !folio_test_active(folio) && 348f2d27392SMatthew Wilcox (Oracle) !folio_test_unevictable(folio)) { 349b01b2141SIngo Molnar struct pagevec *pvec; 350eb709b0dSShaohua Li 351f2d27392SMatthew Wilcox (Oracle) folio_get(folio); 352b01b2141SIngo Molnar local_lock(&lru_pvecs.lock); 353b01b2141SIngo Molnar pvec = this_cpu_ptr(&lru_pvecs.activate_page); 354f2d27392SMatthew Wilcox (Oracle) if (pagevec_add_and_need_flush(pvec, &folio->page)) 355c7c7b80cSAlex Shi pagevec_lru_move_fn(pvec, __activate_page); 356b01b2141SIngo Molnar local_unlock(&lru_pvecs.lock); 357eb709b0dSShaohua Li } 358eb709b0dSShaohua Li } 359eb709b0dSShaohua Li 360eb709b0dSShaohua Li #else 361eb709b0dSShaohua Li static inline void activate_page_drain(int cpu) 362eb709b0dSShaohua Li { 363eb709b0dSShaohua Li } 364eb709b0dSShaohua Li 365f2d27392SMatthew Wilcox (Oracle) static void folio_activate(struct folio *folio) 366eb709b0dSShaohua Li { 3676168d0daSAlex Shi struct lruvec *lruvec; 368eb709b0dSShaohua Li 369f2d27392SMatthew Wilcox (Oracle) if (folio_test_clear_lru(folio)) { 370e809c3feSMatthew Wilcox (Oracle) lruvec = folio_lruvec_lock_irq(folio); 371f2d27392SMatthew Wilcox (Oracle) __folio_activate(folio, lruvec); 3726168d0daSAlex Shi unlock_page_lruvec_irq(lruvec); 373f2d27392SMatthew Wilcox (Oracle) folio_set_lru(folio); 3746168d0daSAlex Shi } 3751da177e4SLinus Torvalds } 376eb709b0dSShaohua Li #endif 3771da177e4SLinus Torvalds 37876580b65SMatthew Wilcox (Oracle) static void __lru_cache_activate_folio(struct folio *folio) 379059285a2SMel Gorman { 380b01b2141SIngo Molnar struct pagevec *pvec; 381059285a2SMel Gorman int i; 382059285a2SMel Gorman 383b01b2141SIngo Molnar local_lock(&lru_pvecs.lock); 384b01b2141SIngo Molnar pvec = this_cpu_ptr(&lru_pvecs.lru_add); 385b01b2141SIngo Molnar 386059285a2SMel Gorman /* 387059285a2SMel Gorman * Search backwards on the optimistic assumption that the page being 388059285a2SMel Gorman * activated has just been added to this pagevec. Note that only 389059285a2SMel Gorman * the local pagevec is examined as a !PageLRU page could be in the 390059285a2SMel Gorman * process of being released, reclaimed, migrated or on a remote 391059285a2SMel Gorman * pagevec that is currently being drained. Furthermore, marking 392059285a2SMel Gorman * a remote pagevec's page PageActive potentially hits a race where 393059285a2SMel Gorman * a page is marked PageActive just after it is added to the inactive 394059285a2SMel Gorman * list causing accounting errors and BUG_ON checks to trigger. 395059285a2SMel Gorman */ 396059285a2SMel Gorman for (i = pagevec_count(pvec) - 1; i >= 0; i--) { 397059285a2SMel Gorman struct page *pagevec_page = pvec->pages[i]; 398059285a2SMel Gorman 39976580b65SMatthew Wilcox (Oracle) if (pagevec_page == &folio->page) { 40076580b65SMatthew Wilcox (Oracle) folio_set_active(folio); 401059285a2SMel Gorman break; 402059285a2SMel Gorman } 403059285a2SMel Gorman } 404059285a2SMel Gorman 405b01b2141SIngo Molnar local_unlock(&lru_pvecs.lock); 406059285a2SMel Gorman } 407059285a2SMel Gorman 4081da177e4SLinus Torvalds /* 4091da177e4SLinus Torvalds * Mark a page as having seen activity. 4101da177e4SLinus Torvalds * 4111da177e4SLinus Torvalds * inactive,unreferenced -> inactive,referenced 4121da177e4SLinus Torvalds * inactive,referenced -> active,unreferenced 4131da177e4SLinus Torvalds * active,unreferenced -> active,referenced 414eb39d618SHugh Dickins * 415eb39d618SHugh Dickins * When a newly allocated page is not yet visible, so safe for non-atomic ops, 416eb39d618SHugh Dickins * __SetPageReferenced(page) may be substituted for mark_page_accessed(page). 4171da177e4SLinus Torvalds */ 41876580b65SMatthew Wilcox (Oracle) void folio_mark_accessed(struct folio *folio) 4191da177e4SLinus Torvalds { 42076580b65SMatthew Wilcox (Oracle) if (!folio_test_referenced(folio)) { 42176580b65SMatthew Wilcox (Oracle) folio_set_referenced(folio); 42276580b65SMatthew Wilcox (Oracle) } else if (folio_test_unevictable(folio)) { 423a1100a74SFengguang Wu /* 424a1100a74SFengguang Wu * Unevictable pages are on the "LRU_UNEVICTABLE" list. But, 425a1100a74SFengguang Wu * this list is never rotated or maintained, so marking an 426914c32e4SBang Li * unevictable page accessed has no effect. 427a1100a74SFengguang Wu */ 42876580b65SMatthew Wilcox (Oracle) } else if (!folio_test_active(folio)) { 429059285a2SMel Gorman /* 430059285a2SMel Gorman * If the page is on the LRU, queue it for activation via 431b01b2141SIngo Molnar * lru_pvecs.activate_page. Otherwise, assume the page is on a 432059285a2SMel Gorman * pagevec, mark it active and it'll be moved to the active 433059285a2SMel Gorman * LRU on the next drain. 434059285a2SMel Gorman */ 43576580b65SMatthew Wilcox (Oracle) if (folio_test_lru(folio)) 43676580b65SMatthew Wilcox (Oracle) folio_activate(folio); 437059285a2SMel Gorman else 43876580b65SMatthew Wilcox (Oracle) __lru_cache_activate_folio(folio); 43976580b65SMatthew Wilcox (Oracle) folio_clear_referenced(folio); 44076580b65SMatthew Wilcox (Oracle) workingset_activation(folio); 4411da177e4SLinus Torvalds } 44276580b65SMatthew Wilcox (Oracle) if (folio_test_idle(folio)) 44376580b65SMatthew Wilcox (Oracle) folio_clear_idle(folio); 4441da177e4SLinus Torvalds } 44576580b65SMatthew Wilcox (Oracle) EXPORT_SYMBOL(folio_mark_accessed); 4461da177e4SLinus Torvalds 447f04e9ebbSKOSAKI Motohiro /** 4480d31125dSMatthew Wilcox (Oracle) * folio_add_lru - Add a folio to an LRU list. 4490d31125dSMatthew Wilcox (Oracle) * @folio: The folio to be added to the LRU. 4502329d375SJianyu Zhan * 4510d31125dSMatthew Wilcox (Oracle) * Queue the folio for addition to the LRU. The decision on whether 4522329d375SJianyu Zhan * to add the page to the [in]active [file|anon] list is deferred until the 4530d31125dSMatthew Wilcox (Oracle) * pagevec is drained. This gives a chance for the caller of folio_add_lru() 4540d31125dSMatthew Wilcox (Oracle) * have the folio added to the active list using folio_mark_accessed(). 455f04e9ebbSKOSAKI Motohiro */ 4560d31125dSMatthew Wilcox (Oracle) void folio_add_lru(struct folio *folio) 4571da177e4SLinus Torvalds { 4586058eaecSJohannes Weiner struct pagevec *pvec; 4596058eaecSJohannes Weiner 4600d31125dSMatthew Wilcox (Oracle) VM_BUG_ON_FOLIO(folio_test_active(folio) && folio_test_unevictable(folio), folio); 4610d31125dSMatthew Wilcox (Oracle) VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); 4626058eaecSJohannes Weiner 4630d31125dSMatthew Wilcox (Oracle) folio_get(folio); 4646058eaecSJohannes Weiner local_lock(&lru_pvecs.lock); 4656058eaecSJohannes Weiner pvec = this_cpu_ptr(&lru_pvecs.lru_add); 4660d31125dSMatthew Wilcox (Oracle) if (pagevec_add_and_need_flush(pvec, &folio->page)) 4676058eaecSJohannes Weiner __pagevec_lru_add(pvec); 4686058eaecSJohannes Weiner local_unlock(&lru_pvecs.lock); 4691da177e4SLinus Torvalds } 4700d31125dSMatthew Wilcox (Oracle) EXPORT_SYMBOL(folio_add_lru); 4711da177e4SLinus Torvalds 472894bc310SLee Schermerhorn /** 473b518154eSJoonsoo Kim * lru_cache_add_inactive_or_unevictable 47400501b53SJohannes Weiner * @page: the page to be added to LRU 47500501b53SJohannes Weiner * @vma: vma in which page is mapped for determining reclaimability 47600501b53SJohannes Weiner * 477b518154eSJoonsoo Kim * Place @page on the inactive or unevictable LRU list, depending on its 47812eab428SMiaohe Lin * evictability. 47900501b53SJohannes Weiner */ 480b518154eSJoonsoo Kim void lru_cache_add_inactive_or_unevictable(struct page *page, 48100501b53SJohannes Weiner struct vm_area_struct *vma) 48200501b53SJohannes Weiner { 48300501b53SJohannes Weiner VM_BUG_ON_PAGE(PageLRU(page), page); 48400501b53SJohannes Weiner 4852fbb0c10SHugh Dickins if (unlikely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) == VM_LOCKED)) 4862fbb0c10SHugh Dickins mlock_new_page(page); 4872fbb0c10SHugh Dickins else 4889c4e6b1aSShakeel Butt lru_cache_add(page); 48900501b53SJohannes Weiner } 49000501b53SJohannes Weiner 491902aaed0SHisashi Hifumi /* 49231560180SMinchan Kim * If the page can not be invalidated, it is moved to the 49331560180SMinchan Kim * inactive list to speed up its reclaim. It is moved to the 49431560180SMinchan Kim * head of the list, rather than the tail, to give the flusher 49531560180SMinchan Kim * threads some time to write it out, as this is much more 49631560180SMinchan Kim * effective than the single-page writeout from reclaim. 497278df9f4SMinchan Kim * 498278df9f4SMinchan Kim * If the page isn't page_mapped and dirty/writeback, the page 499278df9f4SMinchan Kim * could reclaim asap using PG_reclaim. 500278df9f4SMinchan Kim * 501278df9f4SMinchan Kim * 1. active, mapped page -> none 502278df9f4SMinchan Kim * 2. active, dirty/writeback page -> inactive, head, PG_reclaim 503278df9f4SMinchan Kim * 3. inactive, mapped page -> none 504278df9f4SMinchan Kim * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim 505278df9f4SMinchan Kim * 5. inactive, clean -> inactive, tail 506278df9f4SMinchan Kim * 6. Others -> none 507278df9f4SMinchan Kim * 508278df9f4SMinchan Kim * In 4, why it moves inactive's head, the VM expects the page would 509278df9f4SMinchan Kim * be write it out by flusher threads as this is much more effective 510278df9f4SMinchan Kim * than the single-page writeout from reclaim. 51131560180SMinchan Kim */ 512c7c7b80cSAlex Shi static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec) 51331560180SMinchan Kim { 51446ae6b2cSYu Zhao bool active = PageActive(page); 5156c357848SMatthew Wilcox (Oracle) int nr_pages = thp_nr_pages(page); 51631560180SMinchan Kim 517bad49d9cSMinchan Kim if (PageUnevictable(page)) 518bad49d9cSMinchan Kim return; 519bad49d9cSMinchan Kim 52031560180SMinchan Kim /* Some processes are using the page */ 52131560180SMinchan Kim if (page_mapped(page)) 52231560180SMinchan Kim return; 52331560180SMinchan Kim 52446ae6b2cSYu Zhao del_page_from_lru_list(page, lruvec); 52531560180SMinchan Kim ClearPageActive(page); 52631560180SMinchan Kim ClearPageReferenced(page); 52731560180SMinchan Kim 528278df9f4SMinchan Kim if (PageWriteback(page) || PageDirty(page)) { 529278df9f4SMinchan Kim /* 530278df9f4SMinchan Kim * PG_reclaim could be raced with end_page_writeback 531278df9f4SMinchan Kim * It can make readahead confusing. But race window 532278df9f4SMinchan Kim * is _really_ small and it's non-critical problem. 533278df9f4SMinchan Kim */ 5343a9c9788SYu Zhao add_page_to_lru_list(page, lruvec); 535278df9f4SMinchan Kim SetPageReclaim(page); 536278df9f4SMinchan Kim } else { 537278df9f4SMinchan Kim /* 538278df9f4SMinchan Kim * The page's writeback ends up during pagevec 539c4ffefd1SHyeonggon Yoo * We move that page into tail of inactive. 540278df9f4SMinchan Kim */ 5413a9c9788SYu Zhao add_page_to_lru_list_tail(page, lruvec); 5425d91f31fSShakeel Butt __count_vm_events(PGROTATED, nr_pages); 543278df9f4SMinchan Kim } 544278df9f4SMinchan Kim 54521e330fcSShakeel Butt if (active) { 5465d91f31fSShakeel Butt __count_vm_events(PGDEACTIVATE, nr_pages); 54721e330fcSShakeel Butt __count_memcg_events(lruvec_memcg(lruvec), PGDEACTIVATE, 54821e330fcSShakeel Butt nr_pages); 54921e330fcSShakeel Butt } 55031560180SMinchan Kim } 55131560180SMinchan Kim 552c7c7b80cSAlex Shi static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec) 5539c276cc6SMinchan Kim { 554fc574c23SAlex Shi if (PageActive(page) && !PageUnevictable(page)) { 5556c357848SMatthew Wilcox (Oracle) int nr_pages = thp_nr_pages(page); 5569c276cc6SMinchan Kim 55746ae6b2cSYu Zhao del_page_from_lru_list(page, lruvec); 5589c276cc6SMinchan Kim ClearPageActive(page); 5599c276cc6SMinchan Kim ClearPageReferenced(page); 5603a9c9788SYu Zhao add_page_to_lru_list(page, lruvec); 5619c276cc6SMinchan Kim 56221e330fcSShakeel Butt __count_vm_events(PGDEACTIVATE, nr_pages); 56321e330fcSShakeel Butt __count_memcg_events(lruvec_memcg(lruvec), PGDEACTIVATE, 56421e330fcSShakeel Butt nr_pages); 5659c276cc6SMinchan Kim } 5669c276cc6SMinchan Kim } 56710853a03SMinchan Kim 568c7c7b80cSAlex Shi static void lru_lazyfree_fn(struct page *page, struct lruvec *lruvec) 56910853a03SMinchan Kim { 570fc574c23SAlex Shi if (PageAnon(page) && PageSwapBacked(page) && 57124c92eb7SShaohua Li !PageSwapCache(page) && !PageUnevictable(page)) { 5726c357848SMatthew Wilcox (Oracle) int nr_pages = thp_nr_pages(page); 57310853a03SMinchan Kim 57446ae6b2cSYu Zhao del_page_from_lru_list(page, lruvec); 57510853a03SMinchan Kim ClearPageActive(page); 57610853a03SMinchan Kim ClearPageReferenced(page); 577f7ad2a6cSShaohua Li /* 5789de4f22aSHuang Ying * Lazyfree pages are clean anonymous pages. They have 5799de4f22aSHuang Ying * PG_swapbacked flag cleared, to distinguish them from normal 5809de4f22aSHuang Ying * anonymous pages 581f7ad2a6cSShaohua Li */ 582f7ad2a6cSShaohua Li ClearPageSwapBacked(page); 5833a9c9788SYu Zhao add_page_to_lru_list(page, lruvec); 58410853a03SMinchan Kim 58521e330fcSShakeel Butt __count_vm_events(PGLAZYFREE, nr_pages); 58621e330fcSShakeel Butt __count_memcg_events(lruvec_memcg(lruvec), PGLAZYFREE, 58721e330fcSShakeel Butt nr_pages); 58810853a03SMinchan Kim } 58910853a03SMinchan Kim } 59010853a03SMinchan Kim 59131560180SMinchan Kim /* 592902aaed0SHisashi Hifumi * Drain pages out of the cpu's pagevecs. 593902aaed0SHisashi Hifumi * Either "cpu" is the current CPU, and preemption has already been 594902aaed0SHisashi Hifumi * disabled; or "cpu" is being hot-unplugged, and is already dead. 595902aaed0SHisashi Hifumi */ 596f0cb3c76SKonstantin Khlebnikov void lru_add_drain_cpu(int cpu) 5971da177e4SLinus Torvalds { 598b01b2141SIngo Molnar struct pagevec *pvec = &per_cpu(lru_pvecs.lru_add, cpu); 5991da177e4SLinus Torvalds 6001da177e4SLinus Torvalds if (pagevec_count(pvec)) 601a0b8cab3SMel Gorman __pagevec_lru_add(pvec); 602902aaed0SHisashi Hifumi 603b01b2141SIngo Molnar pvec = &per_cpu(lru_rotate.pvec, cpu); 6047e0cc01eSQian Cai /* Disabling interrupts below acts as a compiler barrier. */ 6057e0cc01eSQian Cai if (data_race(pagevec_count(pvec))) { 606902aaed0SHisashi Hifumi unsigned long flags; 607902aaed0SHisashi Hifumi 608902aaed0SHisashi Hifumi /* No harm done if a racing interrupt already did this */ 609b01b2141SIngo Molnar local_lock_irqsave(&lru_rotate.lock, flags); 610c7c7b80cSAlex Shi pagevec_lru_move_fn(pvec, pagevec_move_tail_fn); 611b01b2141SIngo Molnar local_unlock_irqrestore(&lru_rotate.lock, flags); 612902aaed0SHisashi Hifumi } 61331560180SMinchan Kim 614b01b2141SIngo Molnar pvec = &per_cpu(lru_pvecs.lru_deactivate_file, cpu); 61531560180SMinchan Kim if (pagevec_count(pvec)) 616c7c7b80cSAlex Shi pagevec_lru_move_fn(pvec, lru_deactivate_file_fn); 617eb709b0dSShaohua Li 618b01b2141SIngo Molnar pvec = &per_cpu(lru_pvecs.lru_deactivate, cpu); 6199c276cc6SMinchan Kim if (pagevec_count(pvec)) 620c7c7b80cSAlex Shi pagevec_lru_move_fn(pvec, lru_deactivate_fn); 6219c276cc6SMinchan Kim 622b01b2141SIngo Molnar pvec = &per_cpu(lru_pvecs.lru_lazyfree, cpu); 62310853a03SMinchan Kim if (pagevec_count(pvec)) 624c7c7b80cSAlex Shi pagevec_lru_move_fn(pvec, lru_lazyfree_fn); 62510853a03SMinchan Kim 626eb709b0dSShaohua Li activate_page_drain(cpu); 62731560180SMinchan Kim } 62831560180SMinchan Kim 62931560180SMinchan Kim /** 630261b6840SMatthew Wilcox (Oracle) * deactivate_file_folio() - Forcefully deactivate a file folio. 631261b6840SMatthew Wilcox (Oracle) * @folio: Folio to deactivate. 63231560180SMinchan Kim * 633261b6840SMatthew Wilcox (Oracle) * This function hints to the VM that @folio is a good reclaim candidate, 634261b6840SMatthew Wilcox (Oracle) * for example if its invalidation fails due to the folio being dirty 63531560180SMinchan Kim * or under writeback. 636261b6840SMatthew Wilcox (Oracle) * 637261b6840SMatthew Wilcox (Oracle) * Context: Caller holds a reference on the page. 63831560180SMinchan Kim */ 639261b6840SMatthew Wilcox (Oracle) void deactivate_file_folio(struct folio *folio) 64031560180SMinchan Kim { 641b01b2141SIngo Molnar struct pagevec *pvec; 642b01b2141SIngo Molnar 643261b6840SMatthew Wilcox (Oracle) /* 644261b6840SMatthew Wilcox (Oracle) * In a workload with many unevictable pages such as mprotect, 645261b6840SMatthew Wilcox (Oracle) * unevictable folio deactivation for accelerating reclaim is pointless. 646261b6840SMatthew Wilcox (Oracle) */ 647261b6840SMatthew Wilcox (Oracle) if (folio_test_unevictable(folio)) 648261b6840SMatthew Wilcox (Oracle) return; 649261b6840SMatthew Wilcox (Oracle) 650261b6840SMatthew Wilcox (Oracle) folio_get(folio); 651b01b2141SIngo Molnar local_lock(&lru_pvecs.lock); 652b01b2141SIngo Molnar pvec = this_cpu_ptr(&lru_pvecs.lru_deactivate_file); 65331560180SMinchan Kim 654261b6840SMatthew Wilcox (Oracle) if (pagevec_add_and_need_flush(pvec, &folio->page)) 655c7c7b80cSAlex Shi pagevec_lru_move_fn(pvec, lru_deactivate_file_fn); 656b01b2141SIngo Molnar local_unlock(&lru_pvecs.lock); 65731560180SMinchan Kim } 65880bfed90SAndrew Morton 6599c276cc6SMinchan Kim /* 6609c276cc6SMinchan Kim * deactivate_page - deactivate a page 6619c276cc6SMinchan Kim * @page: page to deactivate 6629c276cc6SMinchan Kim * 6639c276cc6SMinchan Kim * deactivate_page() moves @page to the inactive list if @page was on the active 6649c276cc6SMinchan Kim * list and was not an unevictable page. This is done to accelerate the reclaim 6659c276cc6SMinchan Kim * of @page. 6669c276cc6SMinchan Kim */ 6679c276cc6SMinchan Kim void deactivate_page(struct page *page) 6689c276cc6SMinchan Kim { 6699c276cc6SMinchan Kim if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) { 670b01b2141SIngo Molnar struct pagevec *pvec; 6719c276cc6SMinchan Kim 672b01b2141SIngo Molnar local_lock(&lru_pvecs.lock); 673b01b2141SIngo Molnar pvec = this_cpu_ptr(&lru_pvecs.lru_deactivate); 6749c276cc6SMinchan Kim get_page(page); 675d479960eSMinchan Kim if (pagevec_add_and_need_flush(pvec, page)) 676c7c7b80cSAlex Shi pagevec_lru_move_fn(pvec, lru_deactivate_fn); 677b01b2141SIngo Molnar local_unlock(&lru_pvecs.lock); 6789c276cc6SMinchan Kim } 6799c276cc6SMinchan Kim } 6809c276cc6SMinchan Kim 68110853a03SMinchan Kim /** 682f7ad2a6cSShaohua Li * mark_page_lazyfree - make an anon page lazyfree 68310853a03SMinchan Kim * @page: page to deactivate 68410853a03SMinchan Kim * 685f7ad2a6cSShaohua Li * mark_page_lazyfree() moves @page to the inactive file list. 686f7ad2a6cSShaohua Li * This is done to accelerate the reclaim of @page. 68710853a03SMinchan Kim */ 688f7ad2a6cSShaohua Li void mark_page_lazyfree(struct page *page) 68910853a03SMinchan Kim { 690f7ad2a6cSShaohua Li if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) && 69124c92eb7SShaohua Li !PageSwapCache(page) && !PageUnevictable(page)) { 692b01b2141SIngo Molnar struct pagevec *pvec; 69310853a03SMinchan Kim 694b01b2141SIngo Molnar local_lock(&lru_pvecs.lock); 695b01b2141SIngo Molnar pvec = this_cpu_ptr(&lru_pvecs.lru_lazyfree); 69609cbfeafSKirill A. Shutemov get_page(page); 697d479960eSMinchan Kim if (pagevec_add_and_need_flush(pvec, page)) 698c7c7b80cSAlex Shi pagevec_lru_move_fn(pvec, lru_lazyfree_fn); 699b01b2141SIngo Molnar local_unlock(&lru_pvecs.lock); 70010853a03SMinchan Kim } 70110853a03SMinchan Kim } 70210853a03SMinchan Kim 70380bfed90SAndrew Morton void lru_add_drain(void) 70480bfed90SAndrew Morton { 705b01b2141SIngo Molnar local_lock(&lru_pvecs.lock); 706b01b2141SIngo Molnar lru_add_drain_cpu(smp_processor_id()); 707b01b2141SIngo Molnar local_unlock(&lru_pvecs.lock); 708adb11e78SSebastian Andrzej Siewior mlock_page_drain_local(); 709b01b2141SIngo Molnar } 710b01b2141SIngo Molnar 711243418e3SMinchan Kim /* 712243418e3SMinchan Kim * It's called from per-cpu workqueue context in SMP case so 713243418e3SMinchan Kim * lru_add_drain_cpu and invalidate_bh_lrus_cpu should run on 714243418e3SMinchan Kim * the same cpu. It shouldn't be a problem in !SMP case since 715243418e3SMinchan Kim * the core is only one and the locks will disable preemption. 716243418e3SMinchan Kim */ 717243418e3SMinchan Kim static void lru_add_and_bh_lrus_drain(void) 718243418e3SMinchan Kim { 719243418e3SMinchan Kim local_lock(&lru_pvecs.lock); 720243418e3SMinchan Kim lru_add_drain_cpu(smp_processor_id()); 721243418e3SMinchan Kim local_unlock(&lru_pvecs.lock); 722243418e3SMinchan Kim invalidate_bh_lrus_cpu(); 723adb11e78SSebastian Andrzej Siewior mlock_page_drain_local(); 724243418e3SMinchan Kim } 725243418e3SMinchan Kim 726b01b2141SIngo Molnar void lru_add_drain_cpu_zone(struct zone *zone) 727b01b2141SIngo Molnar { 728b01b2141SIngo Molnar local_lock(&lru_pvecs.lock); 729b01b2141SIngo Molnar lru_add_drain_cpu(smp_processor_id()); 730b01b2141SIngo Molnar drain_local_pages(zone); 731b01b2141SIngo Molnar local_unlock(&lru_pvecs.lock); 732adb11e78SSebastian Andrzej Siewior mlock_page_drain_local(); 7331da177e4SLinus Torvalds } 7341da177e4SLinus Torvalds 7356ea183d6SMichal Hocko #ifdef CONFIG_SMP 7366ea183d6SMichal Hocko 7376ea183d6SMichal Hocko static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work); 7386ea183d6SMichal Hocko 739c4028958SDavid Howells static void lru_add_drain_per_cpu(struct work_struct *dummy) 740053837fcSNick Piggin { 741243418e3SMinchan Kim lru_add_and_bh_lrus_drain(); 742053837fcSNick Piggin } 743053837fcSNick Piggin 7449852a721SMichal Hocko /* 7459852a721SMichal Hocko * Doesn't need any cpu hotplug locking because we do rely on per-cpu 7469852a721SMichal Hocko * kworkers being shut down before our page_alloc_cpu_dead callback is 7479852a721SMichal Hocko * executed on the offlined cpu. 7489852a721SMichal Hocko * Calling this function with cpu hotplug locks held can actually lead 7499852a721SMichal Hocko * to obscure indirect dependencies via WQ context. 7509852a721SMichal Hocko */ 7513db3264dSMiaohe Lin static inline void __lru_add_drain_all(bool force_all_cpus) 752053837fcSNick Piggin { 7536446a513SAhmed S. Darwish /* 7546446a513SAhmed S. Darwish * lru_drain_gen - Global pages generation number 7556446a513SAhmed S. Darwish * 7566446a513SAhmed S. Darwish * (A) Definition: global lru_drain_gen = x implies that all generations 7576446a513SAhmed S. Darwish * 0 < n <= x are already *scheduled* for draining. 7586446a513SAhmed S. Darwish * 7596446a513SAhmed S. Darwish * This is an optimization for the highly-contended use case where a 7606446a513SAhmed S. Darwish * user space workload keeps constantly generating a flow of pages for 7616446a513SAhmed S. Darwish * each CPU. 7626446a513SAhmed S. Darwish */ 7636446a513SAhmed S. Darwish static unsigned int lru_drain_gen; 7645fbc4616SChris Metcalf static struct cpumask has_work; 7656446a513SAhmed S. Darwish static DEFINE_MUTEX(lock); 7666446a513SAhmed S. Darwish unsigned cpu, this_gen; 7675fbc4616SChris Metcalf 768ce612879SMichal Hocko /* 769ce612879SMichal Hocko * Make sure nobody triggers this path before mm_percpu_wq is fully 770ce612879SMichal Hocko * initialized. 771ce612879SMichal Hocko */ 772ce612879SMichal Hocko if (WARN_ON(!mm_percpu_wq)) 773ce612879SMichal Hocko return; 774ce612879SMichal Hocko 7756446a513SAhmed S. Darwish /* 7766446a513SAhmed S. Darwish * Guarantee pagevec counter stores visible by this CPU are visible to 7776446a513SAhmed S. Darwish * other CPUs before loading the current drain generation. 7786446a513SAhmed S. Darwish */ 7796446a513SAhmed S. Darwish smp_mb(); 7806446a513SAhmed S. Darwish 7816446a513SAhmed S. Darwish /* 7826446a513SAhmed S. Darwish * (B) Locally cache global LRU draining generation number 7836446a513SAhmed S. Darwish * 7846446a513SAhmed S. Darwish * The read barrier ensures that the counter is loaded before the mutex 7856446a513SAhmed S. Darwish * is taken. It pairs with smp_mb() inside the mutex critical section 7866446a513SAhmed S. Darwish * at (D). 7876446a513SAhmed S. Darwish */ 7886446a513SAhmed S. Darwish this_gen = smp_load_acquire(&lru_drain_gen); 789eef1a429SKonstantin Khlebnikov 7905fbc4616SChris Metcalf mutex_lock(&lock); 791eef1a429SKonstantin Khlebnikov 792eef1a429SKonstantin Khlebnikov /* 7936446a513SAhmed S. Darwish * (C) Exit the draining operation if a newer generation, from another 7946446a513SAhmed S. Darwish * lru_add_drain_all(), was already scheduled for draining. Check (A). 795eef1a429SKonstantin Khlebnikov */ 796d479960eSMinchan Kim if (unlikely(this_gen != lru_drain_gen && !force_all_cpus)) 797eef1a429SKonstantin Khlebnikov goto done; 798eef1a429SKonstantin Khlebnikov 7996446a513SAhmed S. Darwish /* 8006446a513SAhmed S. Darwish * (D) Increment global generation number 8016446a513SAhmed S. Darwish * 8026446a513SAhmed S. Darwish * Pairs with smp_load_acquire() at (B), outside of the critical 8036446a513SAhmed S. Darwish * section. Use a full memory barrier to guarantee that the new global 8046446a513SAhmed S. Darwish * drain generation number is stored before loading pagevec counters. 8056446a513SAhmed S. Darwish * 8066446a513SAhmed S. Darwish * This pairing must be done here, before the for_each_online_cpu loop 8076446a513SAhmed S. Darwish * below which drains the page vectors. 8086446a513SAhmed S. Darwish * 8096446a513SAhmed S. Darwish * Let x, y, and z represent some system CPU numbers, where x < y < z. 810cb152a1aSShijie Luo * Assume CPU #z is in the middle of the for_each_online_cpu loop 8116446a513SAhmed S. Darwish * below and has already reached CPU #y's per-cpu data. CPU #x comes 8126446a513SAhmed S. Darwish * along, adds some pages to its per-cpu vectors, then calls 8136446a513SAhmed S. Darwish * lru_add_drain_all(). 8146446a513SAhmed S. Darwish * 8156446a513SAhmed S. Darwish * If the paired barrier is done at any later step, e.g. after the 8166446a513SAhmed S. Darwish * loop, CPU #x will just exit at (C) and miss flushing out all of its 8176446a513SAhmed S. Darwish * added pages. 8186446a513SAhmed S. Darwish */ 8196446a513SAhmed S. Darwish WRITE_ONCE(lru_drain_gen, lru_drain_gen + 1); 8206446a513SAhmed S. Darwish smp_mb(); 821eef1a429SKonstantin Khlebnikov 8225fbc4616SChris Metcalf cpumask_clear(&has_work); 8235fbc4616SChris Metcalf for_each_online_cpu(cpu) { 8245fbc4616SChris Metcalf struct work_struct *work = &per_cpu(lru_add_drain_work, cpu); 8255fbc4616SChris Metcalf 826ff042f4aSMarcelo Tosatti if (pagevec_count(&per_cpu(lru_pvecs.lru_add, cpu)) || 8277e0cc01eSQian Cai data_race(pagevec_count(&per_cpu(lru_rotate.pvec, cpu))) || 828b01b2141SIngo Molnar pagevec_count(&per_cpu(lru_pvecs.lru_deactivate_file, cpu)) || 829b01b2141SIngo Molnar pagevec_count(&per_cpu(lru_pvecs.lru_deactivate, cpu)) || 830b01b2141SIngo Molnar pagevec_count(&per_cpu(lru_pvecs.lru_lazyfree, cpu)) || 8318cc621d2SMinchan Kim need_activate_page_drain(cpu) || 8322fbb0c10SHugh Dickins need_mlock_page_drain(cpu) || 8338cc621d2SMinchan Kim has_bh_in_lru(cpu, NULL)) { 8345fbc4616SChris Metcalf INIT_WORK(work, lru_add_drain_per_cpu); 835ce612879SMichal Hocko queue_work_on(cpu, mm_percpu_wq, work); 8366446a513SAhmed S. Darwish __cpumask_set_cpu(cpu, &has_work); 8375fbc4616SChris Metcalf } 8385fbc4616SChris Metcalf } 8395fbc4616SChris Metcalf 8405fbc4616SChris Metcalf for_each_cpu(cpu, &has_work) 8415fbc4616SChris Metcalf flush_work(&per_cpu(lru_add_drain_work, cpu)); 8425fbc4616SChris Metcalf 843eef1a429SKonstantin Khlebnikov done: 8445fbc4616SChris Metcalf mutex_unlock(&lock); 845053837fcSNick Piggin } 846d479960eSMinchan Kim 847d479960eSMinchan Kim void lru_add_drain_all(void) 848d479960eSMinchan Kim { 849d479960eSMinchan Kim __lru_add_drain_all(false); 850d479960eSMinchan Kim } 8516ea183d6SMichal Hocko #else 8526ea183d6SMichal Hocko void lru_add_drain_all(void) 8536ea183d6SMichal Hocko { 8546ea183d6SMichal Hocko lru_add_drain(); 8556ea183d6SMichal Hocko } 8566446a513SAhmed S. Darwish #endif /* CONFIG_SMP */ 857053837fcSNick Piggin 858d479960eSMinchan Kim atomic_t lru_disable_count = ATOMIC_INIT(0); 859d479960eSMinchan Kim 860d479960eSMinchan Kim /* 861d479960eSMinchan Kim * lru_cache_disable() needs to be called before we start compiling 862d479960eSMinchan Kim * a list of pages to be migrated using isolate_lru_page(). 863d479960eSMinchan Kim * It drains pages on LRU cache and then disable on all cpus until 864d479960eSMinchan Kim * lru_cache_enable is called. 865d479960eSMinchan Kim * 866d479960eSMinchan Kim * Must be paired with a call to lru_cache_enable(). 867d479960eSMinchan Kim */ 868d479960eSMinchan Kim void lru_cache_disable(void) 869d479960eSMinchan Kim { 870d479960eSMinchan Kim atomic_inc(&lru_disable_count); 871d479960eSMinchan Kim /* 872ff042f4aSMarcelo Tosatti * Readers of lru_disable_count are protected by either disabling 873ff042f4aSMarcelo Tosatti * preemption or rcu_read_lock: 874ff042f4aSMarcelo Tosatti * 875ff042f4aSMarcelo Tosatti * preempt_disable, local_irq_disable [bh_lru_lock()] 876ff042f4aSMarcelo Tosatti * rcu_read_lock [rt_spin_lock CONFIG_PREEMPT_RT] 877ff042f4aSMarcelo Tosatti * preempt_disable [local_lock !CONFIG_PREEMPT_RT] 878ff042f4aSMarcelo Tosatti * 879ff042f4aSMarcelo Tosatti * Since v5.1 kernel, synchronize_rcu() is guaranteed to wait on 880ff042f4aSMarcelo Tosatti * preempt_disable() regions of code. So any CPU which sees 881ff042f4aSMarcelo Tosatti * lru_disable_count = 0 will have exited the critical 882ff042f4aSMarcelo Tosatti * section when synchronize_rcu() returns. 883d479960eSMinchan Kim */ 884*31733463SMarcelo Tosatti synchronize_rcu_expedited(); 885ff042f4aSMarcelo Tosatti #ifdef CONFIG_SMP 886d479960eSMinchan Kim __lru_add_drain_all(true); 887d479960eSMinchan Kim #else 888243418e3SMinchan Kim lru_add_and_bh_lrus_drain(); 889d479960eSMinchan Kim #endif 890d479960eSMinchan Kim } 891d479960eSMinchan Kim 892aabfb572SMichal Hocko /** 893ea1754a0SKirill A. Shutemov * release_pages - batched put_page() 894aabfb572SMichal Hocko * @pages: array of pages to release 895aabfb572SMichal Hocko * @nr: number of pages 8961da177e4SLinus Torvalds * 897aabfb572SMichal Hocko * Decrement the reference count on all the pages in @pages. If it 898aabfb572SMichal Hocko * fell to zero, remove the page from the LRU and free it. 8991da177e4SLinus Torvalds */ 900c6f92f9fSMel Gorman void release_pages(struct page **pages, int nr) 9011da177e4SLinus Torvalds { 9021da177e4SLinus Torvalds int i; 903cc59850eSKonstantin Khlebnikov LIST_HEAD(pages_to_free); 9046168d0daSAlex Shi struct lruvec *lruvec = NULL; 9050de340cbSMatthew Wilcox (Oracle) unsigned long flags = 0; 9063f649ab7SKees Cook unsigned int lock_batch; 9071da177e4SLinus Torvalds 9081da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 9091da177e4SLinus Torvalds struct page *page = pages[i]; 9100de340cbSMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 9111da177e4SLinus Torvalds 912aabfb572SMichal Hocko /* 913aabfb572SMichal Hocko * Make sure the IRQ-safe lock-holding time does not get 914aabfb572SMichal Hocko * excessive with a continuous string of pages from the 9156168d0daSAlex Shi * same lruvec. The lock is held only if lruvec != NULL. 916aabfb572SMichal Hocko */ 9176168d0daSAlex Shi if (lruvec && ++lock_batch == SWAP_CLUSTER_MAX) { 9186168d0daSAlex Shi unlock_page_lruvec_irqrestore(lruvec, flags); 9196168d0daSAlex Shi lruvec = NULL; 920aabfb572SMichal Hocko } 921aabfb572SMichal Hocko 9220de340cbSMatthew Wilcox (Oracle) page = &folio->page; 9236fcb52a5SAaron Lu if (is_huge_zero_page(page)) 924aa88b68cSKirill A. Shutemov continue; 925aa88b68cSKirill A. Shutemov 926c5d6c45eSIra Weiny if (is_zone_device_page(page)) { 9276168d0daSAlex Shi if (lruvec) { 9286168d0daSAlex Shi unlock_page_lruvec_irqrestore(lruvec, flags); 9296168d0daSAlex Shi lruvec = NULL; 930df6ad698SJérôme Glisse } 93189574945SChristoph Hellwig if (put_devmap_managed_page(page)) 932df6ad698SJérôme Glisse continue; 93343fbdeb3SRalph Campbell if (put_page_testzero(page)) 93427674ef6SChristoph Hellwig free_zone_device_page(page); 93543fbdeb3SRalph Campbell continue; 93607d80269SJohn Hubbard } 937df6ad698SJérôme Glisse 938b5810039SNick Piggin if (!put_page_testzero(page)) 9391da177e4SLinus Torvalds continue; 9401da177e4SLinus Torvalds 941ddc58f27SKirill A. Shutemov if (PageCompound(page)) { 9426168d0daSAlex Shi if (lruvec) { 9436168d0daSAlex Shi unlock_page_lruvec_irqrestore(lruvec, flags); 9446168d0daSAlex Shi lruvec = NULL; 945ddc58f27SKirill A. Shutemov } 946ddc58f27SKirill A. Shutemov __put_compound_page(page); 947ddc58f27SKirill A. Shutemov continue; 948ddc58f27SKirill A. Shutemov } 949ddc58f27SKirill A. Shutemov 95046453a6eSNick Piggin if (PageLRU(page)) { 9512a5e4e34SAlexander Duyck struct lruvec *prev_lruvec = lruvec; 952894bc310SLee Schermerhorn 9530de340cbSMatthew Wilcox (Oracle) lruvec = folio_lruvec_relock_irqsave(folio, lruvec, 9542a5e4e34SAlexander Duyck &flags); 9552a5e4e34SAlexander Duyck if (prev_lruvec != lruvec) 956aabfb572SMichal Hocko lock_batch = 0; 957fa9add64SHugh Dickins 95846ae6b2cSYu Zhao del_page_from_lru_list(page, lruvec); 95987560179SYu Zhao __clear_page_lru_flags(page); 96046453a6eSNick Piggin } 96146453a6eSNick Piggin 962b109b870SHugh Dickins /* 963b109b870SHugh Dickins * In rare cases, when truncation or holepunching raced with 964b109b870SHugh Dickins * munlock after VM_LOCKED was cleared, Mlocked may still be 965b109b870SHugh Dickins * found set here. This does not indicate a problem, unless 966b109b870SHugh Dickins * "unevictable_pgs_cleared" appears worryingly large. 967b109b870SHugh Dickins */ 968b109b870SHugh Dickins if (unlikely(PageMlocked(page))) { 969b109b870SHugh Dickins __ClearPageMlocked(page); 970b109b870SHugh Dickins dec_zone_page_state(page, NR_MLOCK); 971b109b870SHugh Dickins count_vm_event(UNEVICTABLE_PGCLEARED); 972b109b870SHugh Dickins } 973b109b870SHugh Dickins 974cc59850eSKonstantin Khlebnikov list_add(&page->lru, &pages_to_free); 9751da177e4SLinus Torvalds } 9766168d0daSAlex Shi if (lruvec) 9776168d0daSAlex Shi unlock_page_lruvec_irqrestore(lruvec, flags); 9781da177e4SLinus Torvalds 979747db954SJohannes Weiner mem_cgroup_uncharge_list(&pages_to_free); 9802d4894b5SMel Gorman free_unref_page_list(&pages_to_free); 9811da177e4SLinus Torvalds } 9820be8557bSMiklos Szeredi EXPORT_SYMBOL(release_pages); 9831da177e4SLinus Torvalds 9841da177e4SLinus Torvalds /* 9851da177e4SLinus Torvalds * The pages which we're about to release may be in the deferred lru-addition 9861da177e4SLinus Torvalds * queues. That would prevent them from really being freed right now. That's 9871da177e4SLinus Torvalds * OK from a correctness point of view but is inefficient - those pages may be 9881da177e4SLinus Torvalds * cache-warm and we want to give them back to the page allocator ASAP. 9891da177e4SLinus Torvalds * 9901da177e4SLinus Torvalds * So __pagevec_release() will drain those queues here. __pagevec_lru_add() 9911da177e4SLinus Torvalds * and __pagevec_lru_add_active() call release_pages() directly to avoid 9921da177e4SLinus Torvalds * mutual recursion. 9931da177e4SLinus Torvalds */ 9941da177e4SLinus Torvalds void __pagevec_release(struct pagevec *pvec) 9951da177e4SLinus Torvalds { 9967f0b5fb9SMel Gorman if (!pvec->percpu_pvec_drained) { 9971da177e4SLinus Torvalds lru_add_drain(); 9987f0b5fb9SMel Gorman pvec->percpu_pvec_drained = true; 999d9ed0d08SMel Gorman } 1000c6f92f9fSMel Gorman release_pages(pvec->pages, pagevec_count(pvec)); 10011da177e4SLinus Torvalds pagevec_reinit(pvec); 10021da177e4SLinus Torvalds } 10037f285701SSteve French EXPORT_SYMBOL(__pagevec_release); 10047f285701SSteve French 1005934387c9SMatthew Wilcox (Oracle) static void __pagevec_lru_add_fn(struct folio *folio, struct lruvec *lruvec) 10063dd7ae8eSShaohua Li { 1007934387c9SMatthew Wilcox (Oracle) int was_unevictable = folio_test_clear_unevictable(folio); 1008934387c9SMatthew Wilcox (Oracle) long nr_pages = folio_nr_pages(folio); 10093dd7ae8eSShaohua Li 1010934387c9SMatthew Wilcox (Oracle) VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); 10113dd7ae8eSShaohua Li 1012934387c9SMatthew Wilcox (Oracle) folio_set_lru(folio); 10132262ace6SHugh Dickins /* 10142262ace6SHugh Dickins * Is an smp_mb__after_atomic() still required here, before 10152262ace6SHugh Dickins * folio_evictable() tests PageMlocked, to rule out the possibility 10162262ace6SHugh Dickins * of stranding an evictable folio on an unevictable LRU? I think 10172fbb0c10SHugh Dickins * not, because __munlock_page() only clears PageMlocked while the LRU 10182262ace6SHugh Dickins * lock is held. 10192262ace6SHugh Dickins * 10202262ace6SHugh Dickins * (That is not true of __page_cache_release(), and not necessarily 10212262ace6SHugh Dickins * true of release_pages(): but those only clear PageMlocked after 10222262ace6SHugh Dickins * put_page_testzero() has excluded any other users of the page.) 10232262ace6SHugh Dickins */ 1024934387c9SMatthew Wilcox (Oracle) if (folio_evictable(folio)) { 10259c4e6b1aSShakeel Butt if (was_unevictable) 10265d91f31fSShakeel Butt __count_vm_events(UNEVICTABLE_PGRESCUED, nr_pages); 10279c4e6b1aSShakeel Butt } else { 1028934387c9SMatthew Wilcox (Oracle) folio_clear_active(folio); 1029934387c9SMatthew Wilcox (Oracle) folio_set_unevictable(folio); 10302fbb0c10SHugh Dickins /* 10312fbb0c10SHugh Dickins * folio->mlock_count = !!folio_test_mlocked(folio)? 10322fbb0c10SHugh Dickins * But that leaves __mlock_page() in doubt whether another 10332fbb0c10SHugh Dickins * actor has already counted the mlock or not. Err on the 10342fbb0c10SHugh Dickins * safe side, underestimate, let page reclaim fix it, rather 10352fbb0c10SHugh Dickins * than leaving a page on the unevictable LRU indefinitely. 10362fbb0c10SHugh Dickins */ 10372fbb0c10SHugh Dickins folio->mlock_count = 0; 10389c4e6b1aSShakeel Butt if (!was_unevictable) 10395d91f31fSShakeel Butt __count_vm_events(UNEVICTABLE_PGCULLED, nr_pages); 10409c4e6b1aSShakeel Butt } 10419c4e6b1aSShakeel Butt 1042934387c9SMatthew Wilcox (Oracle) lruvec_add_folio(lruvec, folio); 1043934387c9SMatthew Wilcox (Oracle) trace_mm_lru_insertion(folio); 10443dd7ae8eSShaohua Li } 10453dd7ae8eSShaohua Li 10461da177e4SLinus Torvalds /* 10471da177e4SLinus Torvalds * Add the passed pages to the LRU, then drop the caller's refcount 10481da177e4SLinus Torvalds * on them. Reinitialises the caller's pagevec. 10491da177e4SLinus Torvalds */ 1050a0b8cab3SMel Gorman void __pagevec_lru_add(struct pagevec *pvec) 10511da177e4SLinus Torvalds { 1052fc574c23SAlex Shi int i; 10536168d0daSAlex Shi struct lruvec *lruvec = NULL; 1054fc574c23SAlex Shi unsigned long flags = 0; 1055fc574c23SAlex Shi 1056fc574c23SAlex Shi for (i = 0; i < pagevec_count(pvec); i++) { 1057934387c9SMatthew Wilcox (Oracle) struct folio *folio = page_folio(pvec->pages[i]); 1058fc574c23SAlex Shi 10590de340cbSMatthew Wilcox (Oracle) lruvec = folio_lruvec_relock_irqsave(folio, lruvec, &flags); 1060934387c9SMatthew Wilcox (Oracle) __pagevec_lru_add_fn(folio, lruvec); 1061fc574c23SAlex Shi } 10626168d0daSAlex Shi if (lruvec) 10636168d0daSAlex Shi unlock_page_lruvec_irqrestore(lruvec, flags); 1064fc574c23SAlex Shi release_pages(pvec->pages, pvec->nr); 1065fc574c23SAlex Shi pagevec_reinit(pvec); 10661da177e4SLinus Torvalds } 1067f04e9ebbSKOSAKI Motohiro 10681da177e4SLinus Torvalds /** 10691613fac9SMatthew Wilcox (Oracle) * folio_batch_remove_exceptionals() - Prune non-folios from a batch. 10701613fac9SMatthew Wilcox (Oracle) * @fbatch: The batch to prune 10710cd6144aSJohannes Weiner * 10721613fac9SMatthew Wilcox (Oracle) * find_get_entries() fills a batch with both folios and shadow/swap/DAX 10731613fac9SMatthew Wilcox (Oracle) * entries. This function prunes all the non-folio entries from @fbatch 10741613fac9SMatthew Wilcox (Oracle) * without leaving holes, so that it can be passed on to folio-only batch 10751613fac9SMatthew Wilcox (Oracle) * operations. 10760cd6144aSJohannes Weiner */ 10771613fac9SMatthew Wilcox (Oracle) void folio_batch_remove_exceptionals(struct folio_batch *fbatch) 10780cd6144aSJohannes Weiner { 10791613fac9SMatthew Wilcox (Oracle) unsigned int i, j; 10800cd6144aSJohannes Weiner 10811613fac9SMatthew Wilcox (Oracle) for (i = 0, j = 0; i < folio_batch_count(fbatch); i++) { 10821613fac9SMatthew Wilcox (Oracle) struct folio *folio = fbatch->folios[i]; 10831613fac9SMatthew Wilcox (Oracle) if (!xa_is_value(folio)) 10841613fac9SMatthew Wilcox (Oracle) fbatch->folios[j++] = folio; 10850cd6144aSJohannes Weiner } 10861613fac9SMatthew Wilcox (Oracle) fbatch->nr = j; 10870cd6144aSJohannes Weiner } 10880cd6144aSJohannes Weiner 10890cd6144aSJohannes Weiner /** 1090b947cee4SJan Kara * pagevec_lookup_range - gang pagecache lookup 10911da177e4SLinus Torvalds * @pvec: Where the resulting pages are placed 10921da177e4SLinus Torvalds * @mapping: The address_space to search 10931da177e4SLinus Torvalds * @start: The starting page index 1094b947cee4SJan Kara * @end: The final page index 10951da177e4SLinus Torvalds * 1096e02a9f04SRandy Dunlap * pagevec_lookup_range() will search for & return a group of up to PAGEVEC_SIZE 1097b947cee4SJan Kara * pages in the mapping starting from index @start and upto index @end 1098b947cee4SJan Kara * (inclusive). The pages are placed in @pvec. pagevec_lookup() takes a 10991da177e4SLinus Torvalds * reference against the pages in @pvec. 11001da177e4SLinus Torvalds * 11011da177e4SLinus Torvalds * The search returns a group of mapping-contiguous pages with ascending 1102d72dc8a2SJan Kara * indexes. There may be holes in the indices due to not-present pages. We 1103d72dc8a2SJan Kara * also update @start to index the next page for the traversal. 11041da177e4SLinus Torvalds * 1105b947cee4SJan Kara * pagevec_lookup_range() returns the number of pages which were found. If this 1106e02a9f04SRandy Dunlap * number is smaller than PAGEVEC_SIZE, the end of specified range has been 1107b947cee4SJan Kara * reached. 11081da177e4SLinus Torvalds */ 1109b947cee4SJan Kara unsigned pagevec_lookup_range(struct pagevec *pvec, 1110397162ffSJan Kara struct address_space *mapping, pgoff_t *start, pgoff_t end) 11111da177e4SLinus Torvalds { 1112397162ffSJan Kara pvec->nr = find_get_pages_range(mapping, start, end, PAGEVEC_SIZE, 1113b947cee4SJan Kara pvec->pages); 11141da177e4SLinus Torvalds return pagevec_count(pvec); 11151da177e4SLinus Torvalds } 1116b947cee4SJan Kara EXPORT_SYMBOL(pagevec_lookup_range); 111778539fdfSChristoph Hellwig 111872b045aeSJan Kara unsigned pagevec_lookup_range_tag(struct pagevec *pvec, 111972b045aeSJan Kara struct address_space *mapping, pgoff_t *index, pgoff_t end, 112010bbd235SMatthew Wilcox xa_mark_t tag) 11211da177e4SLinus Torvalds { 112272b045aeSJan Kara pvec->nr = find_get_pages_range_tag(mapping, index, end, tag, 112367fd707fSJan Kara PAGEVEC_SIZE, pvec->pages); 11241da177e4SLinus Torvalds return pagevec_count(pvec); 11251da177e4SLinus Torvalds } 112672b045aeSJan Kara EXPORT_SYMBOL(pagevec_lookup_range_tag); 11271da177e4SLinus Torvalds 11281da177e4SLinus Torvalds /* 11291da177e4SLinus Torvalds * Perform any setup for the swap system 11301da177e4SLinus Torvalds */ 11311da177e4SLinus Torvalds void __init swap_setup(void) 11321da177e4SLinus Torvalds { 1133ca79b0c2SArun KS unsigned long megs = totalram_pages() >> (20 - PAGE_SHIFT); 1134e0bf68ddSPeter Zijlstra 11351da177e4SLinus Torvalds /* Use a smaller cluster for small-memory machines */ 11361da177e4SLinus Torvalds if (megs < 16) 11371da177e4SLinus Torvalds page_cluster = 2; 11381da177e4SLinus Torvalds else 11391da177e4SLinus Torvalds page_cluster = 3; 11401da177e4SLinus Torvalds /* 11411da177e4SLinus Torvalds * Right now other parts of the system means that we 11421da177e4SLinus Torvalds * _really_ don't want to cluster much more 11431da177e4SLinus Torvalds */ 11441da177e4SLinus Torvalds } 1145