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 46ea0ffd0cSKairui Song /* How many pages do we try to swap or page in/out together? As a power of 2 */ 471da177e4SLinus Torvalds int page_cluster; 48ea0ffd0cSKairui Song const int page_cluster_max = 31; 491da177e4SLinus Torvalds 50c2bc1681SMatthew Wilcox (Oracle) /* Protecting only lru_rotate.fbatch which requires disabling interrupts */ 51b01b2141SIngo Molnar struct lru_rotate { 52b01b2141SIngo Molnar local_lock_t lock; 53c2bc1681SMatthew Wilcox (Oracle) struct folio_batch fbatch; 54b01b2141SIngo Molnar }; 55b01b2141SIngo Molnar static DEFINE_PER_CPU(struct lru_rotate, lru_rotate) = { 56b01b2141SIngo Molnar .lock = INIT_LOCAL_LOCK(lock), 57b01b2141SIngo Molnar }; 58b01b2141SIngo Molnar 59b01b2141SIngo Molnar /* 6082ac64d8SMatthew Wilcox (Oracle) * The following folio batches are grouped together because they are protected 61b01b2141SIngo Molnar * by disabling preemption (and interrupts remain enabled). 62b01b2141SIngo Molnar */ 6382ac64d8SMatthew Wilcox (Oracle) struct cpu_fbatches { 64b01b2141SIngo Molnar local_lock_t lock; 6570dea534SMatthew Wilcox (Oracle) struct folio_batch lru_add; 667a3dbfe8SMatthew Wilcox (Oracle) struct folio_batch lru_deactivate_file; 6785cd7791SMatthew Wilcox (Oracle) struct folio_batch lru_deactivate; 68cec394baSMatthew Wilcox (Oracle) struct folio_batch lru_lazyfree; 69a4a921aaSMing Li #ifdef CONFIG_SMP 703a44610bSMatthew Wilcox (Oracle) struct folio_batch activate; 71a4a921aaSMing Li #endif 72b01b2141SIngo Molnar }; 7382ac64d8SMatthew Wilcox (Oracle) static DEFINE_PER_CPU(struct cpu_fbatches, cpu_fbatches) = { 74b01b2141SIngo Molnar .lock = INIT_LOCAL_LOCK(lock), 75b01b2141SIngo Molnar }; 76902aaed0SHisashi Hifumi 77b221385bSAdrian Bunk /* 78b109b870SHugh Dickins * This path almost never happens for VM activity - pages are normally freed 79b109b870SHugh Dickins * via pagevecs. But it gets used by networking - and for compound pages. 80b221385bSAdrian Bunk */ 81188e8caeSMatthew Wilcox (Oracle) static void __page_cache_release(struct folio *folio) 82b221385bSAdrian Bunk { 83188e8caeSMatthew Wilcox (Oracle) if (folio_test_lru(folio)) { 84fa9add64SHugh Dickins struct lruvec *lruvec; 85fa9add64SHugh Dickins unsigned long flags; 86b221385bSAdrian Bunk 87e809c3feSMatthew Wilcox (Oracle) lruvec = folio_lruvec_lock_irqsave(folio, &flags); 88188e8caeSMatthew Wilcox (Oracle) lruvec_del_folio(lruvec, folio); 89188e8caeSMatthew Wilcox (Oracle) __folio_clear_lru_flags(folio); 906168d0daSAlex Shi unlock_page_lruvec_irqrestore(lruvec, flags); 91b221385bSAdrian Bunk } 92188e8caeSMatthew Wilcox (Oracle) /* See comment on folio_test_mlocked in release_pages() */ 93188e8caeSMatthew Wilcox (Oracle) if (unlikely(folio_test_mlocked(folio))) { 94188e8caeSMatthew Wilcox (Oracle) long nr_pages = folio_nr_pages(folio); 95b109b870SHugh Dickins 96188e8caeSMatthew Wilcox (Oracle) __folio_clear_mlocked(folio); 97188e8caeSMatthew Wilcox (Oracle) zone_stat_mod_folio(folio, NR_MLOCK, -nr_pages); 98b109b870SHugh Dickins count_vm_events(UNEVICTABLE_PGCLEARED, nr_pages); 99b109b870SHugh Dickins } 10091807063SAndrea Arcangeli } 10191807063SAndrea Arcangeli 10283d99659SMatthew Wilcox (Oracle) static void __folio_put_small(struct folio *folio) 10391807063SAndrea Arcangeli { 104188e8caeSMatthew Wilcox (Oracle) __page_cache_release(folio); 10583d99659SMatthew Wilcox (Oracle) mem_cgroup_uncharge(folio); 10683d99659SMatthew Wilcox (Oracle) free_unref_page(&folio->page, 0); 107b221385bSAdrian Bunk } 108b221385bSAdrian Bunk 1095ef82fe7SMatthew Wilcox (Oracle) static void __folio_put_large(struct folio *folio) 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 */ 1175ef82fe7SMatthew Wilcox (Oracle) if (!folio_test_hugetlb(folio)) 118188e8caeSMatthew Wilcox (Oracle) __page_cache_release(folio); 1195375336cSMatthew Wilcox (Oracle) destroy_large_folio(folio); 12091807063SAndrea Arcangeli } 12191807063SAndrea Arcangeli 1228d29c703SMatthew Wilcox (Oracle) void __folio_put(struct folio *folio) 123c747ce79SJianyu Zhan { 1248d29c703SMatthew Wilcox (Oracle) if (unlikely(folio_is_zone_device(folio))) 1258d29c703SMatthew Wilcox (Oracle) free_zone_device_page(&folio->page); 1268d29c703SMatthew Wilcox (Oracle) else if (unlikely(folio_test_large(folio))) 1275ef82fe7SMatthew Wilcox (Oracle) __folio_put_large(folio); 12826296ad2SAndrew Morton else 12983d99659SMatthew Wilcox (Oracle) __folio_put_small(folio); 13026296ad2SAndrew Morton } 1318d29c703SMatthew Wilcox (Oracle) EXPORT_SYMBOL(__folio_put); 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 { 1412f58e5deSMatthew Wilcox (Oracle) struct folio *folio, *next; 1421d7ea732SAlexander Zarochentsev 1432f58e5deSMatthew Wilcox (Oracle) list_for_each_entry_safe(folio, next, pages, lru) { 1442f58e5deSMatthew Wilcox (Oracle) if (!folio_put_testzero(folio)) { 1452f58e5deSMatthew Wilcox (Oracle) list_del(&folio->lru); 146988c69f1SMatthew Wilcox (Oracle) continue; 1471d7ea732SAlexander Zarochentsev } 1482f58e5deSMatthew Wilcox (Oracle) if (folio_test_large(folio)) { 1492f58e5deSMatthew Wilcox (Oracle) list_del(&folio->lru); 1505ef82fe7SMatthew Wilcox (Oracle) __folio_put_large(folio); 151988c69f1SMatthew Wilcox (Oracle) continue; 152988c69f1SMatthew Wilcox (Oracle) } 1532f58e5deSMatthew Wilcox (Oracle) /* LRU flag must be clear because it's passed 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 191c2bc1681SMatthew Wilcox (Oracle) typedef void (*move_fn_t)(struct lruvec *lruvec, struct folio *folio); 192c2bc1681SMatthew Wilcox (Oracle) 19370dea534SMatthew Wilcox (Oracle) static void lru_add_fn(struct lruvec *lruvec, struct folio *folio) 1947d80dd09SMatthew Wilcox (Oracle) { 1957d80dd09SMatthew Wilcox (Oracle) int was_unevictable = folio_test_clear_unevictable(folio); 1967d80dd09SMatthew Wilcox (Oracle) long nr_pages = folio_nr_pages(folio); 1977d80dd09SMatthew Wilcox (Oracle) 1987d80dd09SMatthew Wilcox (Oracle) VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); 1997d80dd09SMatthew Wilcox (Oracle) 2007d80dd09SMatthew Wilcox (Oracle) /* 2017d80dd09SMatthew Wilcox (Oracle) * Is an smp_mb__after_atomic() still required here, before 202188e8caeSMatthew Wilcox (Oracle) * folio_evictable() tests the mlocked flag, to rule out the possibility 2037d80dd09SMatthew Wilcox (Oracle) * of stranding an evictable folio on an unevictable LRU? I think 204188e8caeSMatthew Wilcox (Oracle) * not, because __munlock_page() only clears the mlocked flag 205188e8caeSMatthew Wilcox (Oracle) * while the LRU lock is held. 2067d80dd09SMatthew Wilcox (Oracle) * 2077d80dd09SMatthew Wilcox (Oracle) * (That is not true of __page_cache_release(), and not necessarily 208188e8caeSMatthew Wilcox (Oracle) * true of release_pages(): but those only clear the mlocked flag after 209188e8caeSMatthew Wilcox (Oracle) * folio_put_testzero() has excluded any other users of the folio.) 2107d80dd09SMatthew Wilcox (Oracle) */ 2117d80dd09SMatthew Wilcox (Oracle) if (folio_evictable(folio)) { 2127d80dd09SMatthew Wilcox (Oracle) if (was_unevictable) 2137d80dd09SMatthew Wilcox (Oracle) __count_vm_events(UNEVICTABLE_PGRESCUED, nr_pages); 2147d80dd09SMatthew Wilcox (Oracle) } else { 2157d80dd09SMatthew Wilcox (Oracle) folio_clear_active(folio); 2167d80dd09SMatthew Wilcox (Oracle) folio_set_unevictable(folio); 2177d80dd09SMatthew Wilcox (Oracle) /* 2187d80dd09SMatthew Wilcox (Oracle) * folio->mlock_count = !!folio_test_mlocked(folio)? 2197d80dd09SMatthew Wilcox (Oracle) * But that leaves __mlock_page() in doubt whether another 2207d80dd09SMatthew Wilcox (Oracle) * actor has already counted the mlock or not. Err on the 2217d80dd09SMatthew Wilcox (Oracle) * safe side, underestimate, let page reclaim fix it, rather 2227d80dd09SMatthew Wilcox (Oracle) * than leaving a page on the unevictable LRU indefinitely. 2237d80dd09SMatthew Wilcox (Oracle) */ 2247d80dd09SMatthew Wilcox (Oracle) folio->mlock_count = 0; 2257d80dd09SMatthew Wilcox (Oracle) if (!was_unevictable) 2267d80dd09SMatthew Wilcox (Oracle) __count_vm_events(UNEVICTABLE_PGCULLED, nr_pages); 2277d80dd09SMatthew Wilcox (Oracle) } 2287d80dd09SMatthew Wilcox (Oracle) 2297d80dd09SMatthew Wilcox (Oracle) lruvec_add_folio(lruvec, folio); 2307d80dd09SMatthew Wilcox (Oracle) trace_mm_lru_insertion(folio); 2317d80dd09SMatthew Wilcox (Oracle) } 2327d80dd09SMatthew Wilcox (Oracle) 233c2bc1681SMatthew Wilcox (Oracle) static void folio_batch_move_lru(struct folio_batch *fbatch, move_fn_t move_fn) 234902aaed0SHisashi Hifumi { 235902aaed0SHisashi Hifumi int i; 2366168d0daSAlex Shi struct lruvec *lruvec = NULL; 2373dd7ae8eSShaohua Li unsigned long flags = 0; 238902aaed0SHisashi Hifumi 239c2bc1681SMatthew Wilcox (Oracle) for (i = 0; i < folio_batch_count(fbatch); i++) { 240c2bc1681SMatthew Wilcox (Oracle) struct folio *folio = fbatch->folios[i]; 2413dd7ae8eSShaohua Li 242c2bc1681SMatthew Wilcox (Oracle) /* block memcg migration while the folio moves between lru */ 24370dea534SMatthew Wilcox (Oracle) if (move_fn != lru_add_fn && !folio_test_clear_lru(folio)) 244fc574c23SAlex Shi continue; 245fc574c23SAlex Shi 2460de340cbSMatthew Wilcox (Oracle) lruvec = folio_lruvec_relock_irqsave(folio, lruvec, &flags); 247c2bc1681SMatthew Wilcox (Oracle) move_fn(lruvec, folio); 248fc574c23SAlex Shi 249c2bc1681SMatthew Wilcox (Oracle) folio_set_lru(folio); 2503dd7ae8eSShaohua Li } 251c2bc1681SMatthew Wilcox (Oracle) 2526168d0daSAlex Shi if (lruvec) 2536168d0daSAlex Shi unlock_page_lruvec_irqrestore(lruvec, flags); 254c2bc1681SMatthew Wilcox (Oracle) folios_put(fbatch->folios, folio_batch_count(fbatch)); 255c2bc1681SMatthew Wilcox (Oracle) folio_batch_init(fbatch); 2563dd7ae8eSShaohua Li } 2573dd7ae8eSShaohua Li 258c2bc1681SMatthew Wilcox (Oracle) static void folio_batch_add_and_move(struct folio_batch *fbatch, 259c2bc1681SMatthew Wilcox (Oracle) struct folio *folio, move_fn_t move_fn) 2603dd7ae8eSShaohua Li { 261c2bc1681SMatthew Wilcox (Oracle) if (folio_batch_add(fbatch, folio) && !folio_test_large(folio) && 262c2bc1681SMatthew Wilcox (Oracle) !lru_cache_disabled()) 263c2bc1681SMatthew Wilcox (Oracle) return; 264c2bc1681SMatthew Wilcox (Oracle) folio_batch_move_lru(fbatch, move_fn); 265c2bc1681SMatthew Wilcox (Oracle) } 266575ced1cSMatthew Wilcox (Oracle) 267c2bc1681SMatthew Wilcox (Oracle) static void lru_move_tail_fn(struct lruvec *lruvec, struct folio *folio) 268c2bc1681SMatthew Wilcox (Oracle) { 269575ced1cSMatthew Wilcox (Oracle) if (!folio_test_unevictable(folio)) { 270575ced1cSMatthew Wilcox (Oracle) lruvec_del_folio(lruvec, folio); 271575ced1cSMatthew Wilcox (Oracle) folio_clear_active(folio); 272575ced1cSMatthew Wilcox (Oracle) lruvec_add_folio_tail(lruvec, folio); 273575ced1cSMatthew Wilcox (Oracle) __count_vm_events(PGROTATED, folio_nr_pages(folio)); 274902aaed0SHisashi Hifumi } 275902aaed0SHisashi Hifumi } 2763dd7ae8eSShaohua Li 2773dd7ae8eSShaohua Li /* 278575ced1cSMatthew Wilcox (Oracle) * Writeback is about to end against a folio which has been marked for 279575ced1cSMatthew Wilcox (Oracle) * immediate reclaim. If it still appears to be reclaimable, move it 280575ced1cSMatthew Wilcox (Oracle) * to the tail of the inactive list. 281c7c7b80cSAlex Shi * 282575ced1cSMatthew Wilcox (Oracle) * folio_rotate_reclaimable() must disable IRQs, to prevent nasty races. 2831da177e4SLinus Torvalds */ 284575ced1cSMatthew Wilcox (Oracle) void folio_rotate_reclaimable(struct folio *folio) 2851da177e4SLinus Torvalds { 286575ced1cSMatthew Wilcox (Oracle) if (!folio_test_locked(folio) && !folio_test_dirty(folio) && 287575ced1cSMatthew Wilcox (Oracle) !folio_test_unevictable(folio) && folio_test_lru(folio)) { 288c2bc1681SMatthew Wilcox (Oracle) struct folio_batch *fbatch; 2891da177e4SLinus Torvalds unsigned long flags; 2901da177e4SLinus Torvalds 291575ced1cSMatthew Wilcox (Oracle) folio_get(folio); 292b01b2141SIngo Molnar local_lock_irqsave(&lru_rotate.lock, flags); 293c2bc1681SMatthew Wilcox (Oracle) fbatch = this_cpu_ptr(&lru_rotate.fbatch); 294c2bc1681SMatthew Wilcox (Oracle) folio_batch_add_and_move(fbatch, folio, lru_move_tail_fn); 295b01b2141SIngo Molnar local_unlock_irqrestore(&lru_rotate.lock, flags); 296ac6aadb2SMiklos Szeredi } 2971da177e4SLinus Torvalds } 2981da177e4SLinus Torvalds 2990538a82cSJohannes Weiner void lru_note_cost(struct lruvec *lruvec, bool file, 3000538a82cSJohannes Weiner unsigned int nr_io, unsigned int nr_rotated) 3013e2f41f1SKOSAKI Motohiro { 3020538a82cSJohannes Weiner unsigned long cost; 3030538a82cSJohannes Weiner 3040538a82cSJohannes Weiner /* 3050538a82cSJohannes Weiner * Reflect the relative cost of incurring IO and spending CPU 3060538a82cSJohannes Weiner * time on rotations. This doesn't attempt to make a precise 3070538a82cSJohannes Weiner * comparison, it just says: if reloads are about comparable 3080538a82cSJohannes Weiner * between the LRU lists, or rotations are overwhelmingly 3090538a82cSJohannes Weiner * different between them, adjust scan balance for CPU work. 3100538a82cSJohannes Weiner */ 3110538a82cSJohannes Weiner cost = nr_io * SWAP_CLUSTER_MAX + nr_rotated; 3120538a82cSJohannes Weiner 3137cf111bcSJohannes Weiner do { 3147cf111bcSJohannes Weiner unsigned long lrusize; 3157cf111bcSJohannes Weiner 3166168d0daSAlex Shi /* 3176168d0daSAlex Shi * Hold lruvec->lru_lock is safe here, since 3186168d0daSAlex Shi * 1) The pinned lruvec in reclaim, or 3196168d0daSAlex Shi * 2) From a pre-LRU page during refault (which also holds the 3206168d0daSAlex Shi * rcu lock, so would be safe even if the page was on the LRU 3216168d0daSAlex Shi * and could move simultaneously to a new lruvec). 3226168d0daSAlex Shi */ 3236168d0daSAlex Shi spin_lock_irq(&lruvec->lru_lock); 3247cf111bcSJohannes Weiner /* Record cost event */ 32596f8bf4fSJohannes Weiner if (file) 3260538a82cSJohannes Weiner lruvec->file_cost += cost; 3271431d4d1SJohannes Weiner else 3280538a82cSJohannes Weiner lruvec->anon_cost += cost; 3297cf111bcSJohannes Weiner 3307cf111bcSJohannes Weiner /* 3317cf111bcSJohannes Weiner * Decay previous events 3327cf111bcSJohannes Weiner * 3337cf111bcSJohannes Weiner * Because workloads change over time (and to avoid 3347cf111bcSJohannes Weiner * overflow) we keep these statistics as a floating 3357cf111bcSJohannes Weiner * average, which ends up weighing recent refaults 3367cf111bcSJohannes Weiner * more than old ones. 3377cf111bcSJohannes Weiner */ 3387cf111bcSJohannes Weiner lrusize = lruvec_page_state(lruvec, NR_INACTIVE_ANON) + 3397cf111bcSJohannes Weiner lruvec_page_state(lruvec, NR_ACTIVE_ANON) + 3407cf111bcSJohannes Weiner lruvec_page_state(lruvec, NR_INACTIVE_FILE) + 3417cf111bcSJohannes Weiner lruvec_page_state(lruvec, NR_ACTIVE_FILE); 3427cf111bcSJohannes Weiner 3437cf111bcSJohannes Weiner if (lruvec->file_cost + lruvec->anon_cost > lrusize / 4) { 3447cf111bcSJohannes Weiner lruvec->file_cost /= 2; 3457cf111bcSJohannes Weiner lruvec->anon_cost /= 2; 3467cf111bcSJohannes Weiner } 3476168d0daSAlex Shi spin_unlock_irq(&lruvec->lru_lock); 3487cf111bcSJohannes Weiner } while ((lruvec = parent_lruvec(lruvec))); 3493e2f41f1SKOSAKI Motohiro } 3503e2f41f1SKOSAKI Motohiro 3510538a82cSJohannes Weiner void lru_note_cost_refault(struct folio *folio) 35296f8bf4fSJohannes Weiner { 3530995d7e5SMatthew Wilcox (Oracle) lru_note_cost(folio_lruvec(folio), folio_is_file_lru(folio), 3540538a82cSJohannes Weiner folio_nr_pages(folio), 0); 35596f8bf4fSJohannes Weiner } 35696f8bf4fSJohannes Weiner 3573a44610bSMatthew Wilcox (Oracle) static void folio_activate_fn(struct lruvec *lruvec, struct folio *folio) 358744ed144SShaohua Li { 359f2d27392SMatthew Wilcox (Oracle) if (!folio_test_active(folio) && !folio_test_unevictable(folio)) { 360f2d27392SMatthew Wilcox (Oracle) long nr_pages = folio_nr_pages(folio); 361744ed144SShaohua Li 362f2d27392SMatthew Wilcox (Oracle) lruvec_del_folio(lruvec, folio); 363f2d27392SMatthew Wilcox (Oracle) folio_set_active(folio); 364f2d27392SMatthew Wilcox (Oracle) lruvec_add_folio(lruvec, folio); 365f2d27392SMatthew Wilcox (Oracle) trace_mm_lru_activate(folio); 3667a608572SLinus Torvalds 36721e330fcSShakeel Butt __count_vm_events(PGACTIVATE, nr_pages); 36821e330fcSShakeel Butt __count_memcg_events(lruvec_memcg(lruvec), PGACTIVATE, 36921e330fcSShakeel Butt nr_pages); 370744ed144SShaohua Li } 371eb709b0dSShaohua Li } 372eb709b0dSShaohua Li 373eb709b0dSShaohua Li #ifdef CONFIG_SMP 3743a44610bSMatthew Wilcox (Oracle) static void folio_activate_drain(int cpu) 375f2d27392SMatthew Wilcox (Oracle) { 37682ac64d8SMatthew Wilcox (Oracle) struct folio_batch *fbatch = &per_cpu(cpu_fbatches.activate, cpu); 377f2d27392SMatthew Wilcox (Oracle) 3783a44610bSMatthew Wilcox (Oracle) if (folio_batch_count(fbatch)) 3793a44610bSMatthew Wilcox (Oracle) folio_batch_move_lru(fbatch, folio_activate_fn); 3805fbc4616SChris Metcalf } 3815fbc4616SChris Metcalf 382018ee47fSYu Zhao void folio_activate(struct folio *folio) 383eb709b0dSShaohua Li { 384f2d27392SMatthew Wilcox (Oracle) if (folio_test_lru(folio) && !folio_test_active(folio) && 385f2d27392SMatthew Wilcox (Oracle) !folio_test_unevictable(folio)) { 3863a44610bSMatthew Wilcox (Oracle) struct folio_batch *fbatch; 387eb709b0dSShaohua Li 388f2d27392SMatthew Wilcox (Oracle) folio_get(folio); 38982ac64d8SMatthew Wilcox (Oracle) local_lock(&cpu_fbatches.lock); 39082ac64d8SMatthew Wilcox (Oracle) fbatch = this_cpu_ptr(&cpu_fbatches.activate); 3913a44610bSMatthew Wilcox (Oracle) folio_batch_add_and_move(fbatch, folio, folio_activate_fn); 39282ac64d8SMatthew Wilcox (Oracle) local_unlock(&cpu_fbatches.lock); 393eb709b0dSShaohua Li } 394eb709b0dSShaohua Li } 395eb709b0dSShaohua Li 396eb709b0dSShaohua Li #else 3973a44610bSMatthew Wilcox (Oracle) static inline void folio_activate_drain(int cpu) 398eb709b0dSShaohua Li { 399eb709b0dSShaohua Li } 400eb709b0dSShaohua Li 401018ee47fSYu Zhao void folio_activate(struct folio *folio) 402eb709b0dSShaohua Li { 4036168d0daSAlex Shi struct lruvec *lruvec; 404eb709b0dSShaohua Li 405f2d27392SMatthew Wilcox (Oracle) if (folio_test_clear_lru(folio)) { 406e809c3feSMatthew Wilcox (Oracle) lruvec = folio_lruvec_lock_irq(folio); 4073a44610bSMatthew Wilcox (Oracle) folio_activate_fn(lruvec, folio); 4086168d0daSAlex Shi unlock_page_lruvec_irq(lruvec); 409f2d27392SMatthew Wilcox (Oracle) folio_set_lru(folio); 4106168d0daSAlex Shi } 4111da177e4SLinus Torvalds } 412eb709b0dSShaohua Li #endif 4131da177e4SLinus Torvalds 41476580b65SMatthew Wilcox (Oracle) static void __lru_cache_activate_folio(struct folio *folio) 415059285a2SMel Gorman { 41670dea534SMatthew Wilcox (Oracle) struct folio_batch *fbatch; 417059285a2SMel Gorman int i; 418059285a2SMel Gorman 41982ac64d8SMatthew Wilcox (Oracle) local_lock(&cpu_fbatches.lock); 42082ac64d8SMatthew Wilcox (Oracle) fbatch = this_cpu_ptr(&cpu_fbatches.lru_add); 421b01b2141SIngo Molnar 422059285a2SMel Gorman /* 42370dea534SMatthew Wilcox (Oracle) * Search backwards on the optimistic assumption that the folio being 42470dea534SMatthew Wilcox (Oracle) * activated has just been added to this batch. Note that only 42570dea534SMatthew Wilcox (Oracle) * the local batch is examined as a !LRU folio could be in the 426059285a2SMel Gorman * process of being released, reclaimed, migrated or on a remote 42770dea534SMatthew Wilcox (Oracle) * batch that is currently being drained. Furthermore, marking 42870dea534SMatthew Wilcox (Oracle) * a remote batch's folio active potentially hits a race where 42970dea534SMatthew Wilcox (Oracle) * a folio is marked active just after it is added to the inactive 430059285a2SMel Gorman * list causing accounting errors and BUG_ON checks to trigger. 431059285a2SMel Gorman */ 43270dea534SMatthew Wilcox (Oracle) for (i = folio_batch_count(fbatch) - 1; i >= 0; i--) { 43370dea534SMatthew Wilcox (Oracle) struct folio *batch_folio = fbatch->folios[i]; 434059285a2SMel Gorman 43570dea534SMatthew Wilcox (Oracle) if (batch_folio == folio) { 43676580b65SMatthew Wilcox (Oracle) folio_set_active(folio); 437059285a2SMel Gorman break; 438059285a2SMel Gorman } 439059285a2SMel Gorman } 440059285a2SMel Gorman 44182ac64d8SMatthew Wilcox (Oracle) local_unlock(&cpu_fbatches.lock); 442059285a2SMel Gorman } 443059285a2SMel Gorman 444ac35a490SYu Zhao #ifdef CONFIG_LRU_GEN 445ac35a490SYu Zhao static void folio_inc_refs(struct folio *folio) 446ac35a490SYu Zhao { 447ac35a490SYu Zhao unsigned long new_flags, old_flags = READ_ONCE(folio->flags); 448ac35a490SYu Zhao 449ac35a490SYu Zhao if (folio_test_unevictable(folio)) 450ac35a490SYu Zhao return; 451ac35a490SYu Zhao 452ac35a490SYu Zhao if (!folio_test_referenced(folio)) { 453ac35a490SYu Zhao folio_set_referenced(folio); 454ac35a490SYu Zhao return; 455ac35a490SYu Zhao } 456ac35a490SYu Zhao 457ac35a490SYu Zhao if (!folio_test_workingset(folio)) { 458ac35a490SYu Zhao folio_set_workingset(folio); 459ac35a490SYu Zhao return; 460ac35a490SYu Zhao } 461ac35a490SYu Zhao 462ac35a490SYu Zhao /* see the comment on MAX_NR_TIERS */ 463ac35a490SYu Zhao do { 464ac35a490SYu Zhao new_flags = old_flags & LRU_REFS_MASK; 465ac35a490SYu Zhao if (new_flags == LRU_REFS_MASK) 466ac35a490SYu Zhao break; 467ac35a490SYu Zhao 468ac35a490SYu Zhao new_flags += BIT(LRU_REFS_PGOFF); 469ac35a490SYu Zhao new_flags |= old_flags & ~LRU_REFS_MASK; 470ac35a490SYu Zhao } while (!try_cmpxchg(&folio->flags, &old_flags, new_flags)); 471ac35a490SYu Zhao } 472ac35a490SYu Zhao #else 473ac35a490SYu Zhao static void folio_inc_refs(struct folio *folio) 474ac35a490SYu Zhao { 475ac35a490SYu Zhao } 476ac35a490SYu Zhao #endif /* CONFIG_LRU_GEN */ 477ac35a490SYu Zhao 4781da177e4SLinus Torvalds /* 4791da177e4SLinus Torvalds * Mark a page as having seen activity. 4801da177e4SLinus Torvalds * 4811da177e4SLinus Torvalds * inactive,unreferenced -> inactive,referenced 4821da177e4SLinus Torvalds * inactive,referenced -> active,unreferenced 4831da177e4SLinus Torvalds * active,unreferenced -> active,referenced 484eb39d618SHugh Dickins * 485eb39d618SHugh Dickins * When a newly allocated page is not yet visible, so safe for non-atomic ops, 486eb39d618SHugh Dickins * __SetPageReferenced(page) may be substituted for mark_page_accessed(page). 4871da177e4SLinus Torvalds */ 48876580b65SMatthew Wilcox (Oracle) void folio_mark_accessed(struct folio *folio) 4891da177e4SLinus Torvalds { 490ac35a490SYu Zhao if (lru_gen_enabled()) { 491ac35a490SYu Zhao folio_inc_refs(folio); 492ac35a490SYu Zhao return; 493ac35a490SYu Zhao } 494ac35a490SYu Zhao 49576580b65SMatthew Wilcox (Oracle) if (!folio_test_referenced(folio)) { 49676580b65SMatthew Wilcox (Oracle) folio_set_referenced(folio); 49776580b65SMatthew Wilcox (Oracle) } else if (folio_test_unevictable(folio)) { 498a1100a74SFengguang Wu /* 499a1100a74SFengguang Wu * Unevictable pages are on the "LRU_UNEVICTABLE" list. But, 500a1100a74SFengguang Wu * this list is never rotated or maintained, so marking an 501914c32e4SBang Li * unevictable page accessed has no effect. 502a1100a74SFengguang Wu */ 50376580b65SMatthew Wilcox (Oracle) } else if (!folio_test_active(folio)) { 504059285a2SMel Gorman /* 5053a44610bSMatthew Wilcox (Oracle) * If the folio is on the LRU, queue it for activation via 50682ac64d8SMatthew Wilcox (Oracle) * cpu_fbatches.activate. Otherwise, assume the folio is in a 5073a44610bSMatthew Wilcox (Oracle) * folio_batch, mark it active and it'll be moved to the active 508059285a2SMel Gorman * LRU on the next drain. 509059285a2SMel Gorman */ 51076580b65SMatthew Wilcox (Oracle) if (folio_test_lru(folio)) 51176580b65SMatthew Wilcox (Oracle) folio_activate(folio); 512059285a2SMel Gorman else 51376580b65SMatthew Wilcox (Oracle) __lru_cache_activate_folio(folio); 51476580b65SMatthew Wilcox (Oracle) folio_clear_referenced(folio); 51576580b65SMatthew Wilcox (Oracle) workingset_activation(folio); 5161da177e4SLinus Torvalds } 51776580b65SMatthew Wilcox (Oracle) if (folio_test_idle(folio)) 51876580b65SMatthew Wilcox (Oracle) folio_clear_idle(folio); 5191da177e4SLinus Torvalds } 52076580b65SMatthew Wilcox (Oracle) EXPORT_SYMBOL(folio_mark_accessed); 5211da177e4SLinus Torvalds 522f04e9ebbSKOSAKI Motohiro /** 5230d31125dSMatthew Wilcox (Oracle) * folio_add_lru - Add a folio to an LRU list. 5240d31125dSMatthew Wilcox (Oracle) * @folio: The folio to be added to the LRU. 5252329d375SJianyu Zhan * 5260d31125dSMatthew Wilcox (Oracle) * Queue the folio for addition to the LRU. The decision on whether 5272329d375SJianyu Zhan * to add the page to the [in]active [file|anon] list is deferred until the 52882ac64d8SMatthew Wilcox (Oracle) * folio_batch is drained. This gives a chance for the caller of folio_add_lru() 5290d31125dSMatthew Wilcox (Oracle) * have the folio added to the active list using folio_mark_accessed(). 530f04e9ebbSKOSAKI Motohiro */ 5310d31125dSMatthew Wilcox (Oracle) void folio_add_lru(struct folio *folio) 5321da177e4SLinus Torvalds { 53370dea534SMatthew Wilcox (Oracle) struct folio_batch *fbatch; 5346058eaecSJohannes Weiner 53570dea534SMatthew Wilcox (Oracle) VM_BUG_ON_FOLIO(folio_test_active(folio) && 53670dea534SMatthew Wilcox (Oracle) folio_test_unevictable(folio), folio); 5370d31125dSMatthew Wilcox (Oracle) VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); 5386058eaecSJohannes Weiner 539ec1c86b2SYu Zhao /* see the comment in lru_gen_add_folio() */ 540ec1c86b2SYu Zhao if (lru_gen_enabled() && !folio_test_unevictable(folio) && 541ec1c86b2SYu Zhao lru_gen_in_fault() && !(current->flags & PF_MEMALLOC)) 542ec1c86b2SYu Zhao folio_set_active(folio); 543ec1c86b2SYu Zhao 5440d31125dSMatthew Wilcox (Oracle) folio_get(folio); 54582ac64d8SMatthew Wilcox (Oracle) local_lock(&cpu_fbatches.lock); 54682ac64d8SMatthew Wilcox (Oracle) fbatch = this_cpu_ptr(&cpu_fbatches.lru_add); 54770dea534SMatthew Wilcox (Oracle) folio_batch_add_and_move(fbatch, folio, lru_add_fn); 54882ac64d8SMatthew Wilcox (Oracle) local_unlock(&cpu_fbatches.lock); 5491da177e4SLinus Torvalds } 5500d31125dSMatthew Wilcox (Oracle) EXPORT_SYMBOL(folio_add_lru); 5511da177e4SLinus Torvalds 552894bc310SLee Schermerhorn /** 553681ecf63SMatthew Wilcox (Oracle) * folio_add_lru_vma() - Add a folio to the appropate LRU list for this VMA. 554681ecf63SMatthew Wilcox (Oracle) * @folio: The folio to be added to the LRU. 555681ecf63SMatthew Wilcox (Oracle) * @vma: VMA in which the folio is mapped. 55600501b53SJohannes Weiner * 557681ecf63SMatthew Wilcox (Oracle) * If the VMA is mlocked, @folio is added to the unevictable list. 558681ecf63SMatthew Wilcox (Oracle) * Otherwise, it is treated the same way as folio_add_lru(). 55900501b53SJohannes Weiner */ 560681ecf63SMatthew Wilcox (Oracle) void folio_add_lru_vma(struct folio *folio, struct vm_area_struct *vma) 56100501b53SJohannes Weiner { 562681ecf63SMatthew Wilcox (Oracle) VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); 56300501b53SJohannes Weiner 5642fbb0c10SHugh Dickins if (unlikely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) == VM_LOCKED)) 565*96f97c43SLorenzo Stoakes mlock_new_folio(folio); 5662fbb0c10SHugh Dickins else 567681ecf63SMatthew Wilcox (Oracle) folio_add_lru(folio); 56800501b53SJohannes Weiner } 56900501b53SJohannes Weiner 570902aaed0SHisashi Hifumi /* 5717a3dbfe8SMatthew Wilcox (Oracle) * If the folio cannot be invalidated, it is moved to the 57231560180SMinchan Kim * inactive list to speed up its reclaim. It is moved to the 57331560180SMinchan Kim * head of the list, rather than the tail, to give the flusher 57431560180SMinchan Kim * threads some time to write it out, as this is much more 57531560180SMinchan Kim * effective than the single-page writeout from reclaim. 576278df9f4SMinchan Kim * 5777a3dbfe8SMatthew Wilcox (Oracle) * If the folio isn't mapped and dirty/writeback, the folio 5787a3dbfe8SMatthew Wilcox (Oracle) * could be reclaimed asap using the reclaim flag. 579278df9f4SMinchan Kim * 5807a3dbfe8SMatthew Wilcox (Oracle) * 1. active, mapped folio -> none 5817a3dbfe8SMatthew Wilcox (Oracle) * 2. active, dirty/writeback folio -> inactive, head, reclaim 5827a3dbfe8SMatthew Wilcox (Oracle) * 3. inactive, mapped folio -> none 5837a3dbfe8SMatthew Wilcox (Oracle) * 4. inactive, dirty/writeback folio -> inactive, head, reclaim 584278df9f4SMinchan Kim * 5. inactive, clean -> inactive, tail 585278df9f4SMinchan Kim * 6. Others -> none 586278df9f4SMinchan Kim * 5877a3dbfe8SMatthew Wilcox (Oracle) * In 4, it moves to the head of the inactive list so the folio is 5887a3dbfe8SMatthew Wilcox (Oracle) * written out by flusher threads as this is much more efficient 589278df9f4SMinchan Kim * than the single-page writeout from reclaim. 59031560180SMinchan Kim */ 5917a3dbfe8SMatthew Wilcox (Oracle) static void lru_deactivate_file_fn(struct lruvec *lruvec, struct folio *folio) 59231560180SMinchan Kim { 5937a3dbfe8SMatthew Wilcox (Oracle) bool active = folio_test_active(folio); 5947a3dbfe8SMatthew Wilcox (Oracle) long nr_pages = folio_nr_pages(folio); 59531560180SMinchan Kim 5967a3dbfe8SMatthew Wilcox (Oracle) if (folio_test_unevictable(folio)) 597bad49d9cSMinchan Kim return; 598bad49d9cSMinchan Kim 5997a3dbfe8SMatthew Wilcox (Oracle) /* Some processes are using the folio */ 6007a3dbfe8SMatthew Wilcox (Oracle) if (folio_mapped(folio)) 60131560180SMinchan Kim return; 60231560180SMinchan Kim 6037a3dbfe8SMatthew Wilcox (Oracle) lruvec_del_folio(lruvec, folio); 6047a3dbfe8SMatthew Wilcox (Oracle) folio_clear_active(folio); 6057a3dbfe8SMatthew Wilcox (Oracle) folio_clear_referenced(folio); 60631560180SMinchan Kim 6077a3dbfe8SMatthew Wilcox (Oracle) if (folio_test_writeback(folio) || folio_test_dirty(folio)) { 608278df9f4SMinchan Kim /* 6097a3dbfe8SMatthew Wilcox (Oracle) * Setting the reclaim flag could race with 6107a3dbfe8SMatthew Wilcox (Oracle) * folio_end_writeback() and confuse readahead. But the 6117a3dbfe8SMatthew Wilcox (Oracle) * race window is _really_ small and it's not a critical 6127a3dbfe8SMatthew Wilcox (Oracle) * problem. 613278df9f4SMinchan Kim */ 6147a3dbfe8SMatthew Wilcox (Oracle) lruvec_add_folio(lruvec, folio); 6157a3dbfe8SMatthew Wilcox (Oracle) folio_set_reclaim(folio); 616278df9f4SMinchan Kim } else { 617278df9f4SMinchan Kim /* 6187a3dbfe8SMatthew Wilcox (Oracle) * The folio's writeback ended while it was in the batch. 6197a3dbfe8SMatthew Wilcox (Oracle) * We move that folio to the tail of the inactive list. 620278df9f4SMinchan Kim */ 6217a3dbfe8SMatthew Wilcox (Oracle) lruvec_add_folio_tail(lruvec, folio); 6225d91f31fSShakeel Butt __count_vm_events(PGROTATED, nr_pages); 623278df9f4SMinchan Kim } 624278df9f4SMinchan Kim 62521e330fcSShakeel Butt if (active) { 6265d91f31fSShakeel Butt __count_vm_events(PGDEACTIVATE, nr_pages); 62721e330fcSShakeel Butt __count_memcg_events(lruvec_memcg(lruvec), PGDEACTIVATE, 62821e330fcSShakeel Butt nr_pages); 62921e330fcSShakeel Butt } 63031560180SMinchan Kim } 63131560180SMinchan Kim 63285cd7791SMatthew Wilcox (Oracle) static void lru_deactivate_fn(struct lruvec *lruvec, struct folio *folio) 6339c276cc6SMinchan Kim { 634ec1c86b2SYu Zhao if (!folio_test_unevictable(folio) && (folio_test_active(folio) || lru_gen_enabled())) { 63585cd7791SMatthew Wilcox (Oracle) long nr_pages = folio_nr_pages(folio); 6369c276cc6SMinchan Kim 63785cd7791SMatthew Wilcox (Oracle) lruvec_del_folio(lruvec, folio); 63885cd7791SMatthew Wilcox (Oracle) folio_clear_active(folio); 63985cd7791SMatthew Wilcox (Oracle) folio_clear_referenced(folio); 64085cd7791SMatthew Wilcox (Oracle) lruvec_add_folio(lruvec, folio); 6419c276cc6SMinchan Kim 64221e330fcSShakeel Butt __count_vm_events(PGDEACTIVATE, nr_pages); 64321e330fcSShakeel Butt __count_memcg_events(lruvec_memcg(lruvec), PGDEACTIVATE, 64421e330fcSShakeel Butt nr_pages); 6459c276cc6SMinchan Kim } 6469c276cc6SMinchan Kim } 64710853a03SMinchan Kim 648cec394baSMatthew Wilcox (Oracle) static void lru_lazyfree_fn(struct lruvec *lruvec, struct folio *folio) 64910853a03SMinchan Kim { 650cec394baSMatthew Wilcox (Oracle) if (folio_test_anon(folio) && folio_test_swapbacked(folio) && 651cec394baSMatthew Wilcox (Oracle) !folio_test_swapcache(folio) && !folio_test_unevictable(folio)) { 652cec394baSMatthew Wilcox (Oracle) long nr_pages = folio_nr_pages(folio); 65310853a03SMinchan Kim 654cec394baSMatthew Wilcox (Oracle) lruvec_del_folio(lruvec, folio); 655cec394baSMatthew Wilcox (Oracle) folio_clear_active(folio); 656cec394baSMatthew Wilcox (Oracle) folio_clear_referenced(folio); 657f7ad2a6cSShaohua Li /* 658cec394baSMatthew Wilcox (Oracle) * Lazyfree folios are clean anonymous folios. They have 659cec394baSMatthew Wilcox (Oracle) * the swapbacked flag cleared, to distinguish them from normal 660cec394baSMatthew Wilcox (Oracle) * anonymous folios 661f7ad2a6cSShaohua Li */ 662cec394baSMatthew Wilcox (Oracle) folio_clear_swapbacked(folio); 663cec394baSMatthew Wilcox (Oracle) lruvec_add_folio(lruvec, folio); 66410853a03SMinchan Kim 66521e330fcSShakeel Butt __count_vm_events(PGLAZYFREE, nr_pages); 66621e330fcSShakeel Butt __count_memcg_events(lruvec_memcg(lruvec), PGLAZYFREE, 66721e330fcSShakeel Butt nr_pages); 66810853a03SMinchan Kim } 66910853a03SMinchan Kim } 67010853a03SMinchan Kim 67131560180SMinchan Kim /* 67282ac64d8SMatthew Wilcox (Oracle) * Drain pages out of the cpu's folio_batch. 673902aaed0SHisashi Hifumi * Either "cpu" is the current CPU, and preemption has already been 674902aaed0SHisashi Hifumi * disabled; or "cpu" is being hot-unplugged, and is already dead. 675902aaed0SHisashi Hifumi */ 676f0cb3c76SKonstantin Khlebnikov void lru_add_drain_cpu(int cpu) 6771da177e4SLinus Torvalds { 678a2d33b5dSMatthew Wilcox (Oracle) struct cpu_fbatches *fbatches = &per_cpu(cpu_fbatches, cpu); 679a2d33b5dSMatthew Wilcox (Oracle) struct folio_batch *fbatch = &fbatches->lru_add; 6801da177e4SLinus Torvalds 68170dea534SMatthew Wilcox (Oracle) if (folio_batch_count(fbatch)) 68270dea534SMatthew Wilcox (Oracle) folio_batch_move_lru(fbatch, lru_add_fn); 683902aaed0SHisashi Hifumi 684c2bc1681SMatthew Wilcox (Oracle) fbatch = &per_cpu(lru_rotate.fbatch, cpu); 6857e0cc01eSQian Cai /* Disabling interrupts below acts as a compiler barrier. */ 686c2bc1681SMatthew Wilcox (Oracle) if (data_race(folio_batch_count(fbatch))) { 687902aaed0SHisashi Hifumi unsigned long flags; 688902aaed0SHisashi Hifumi 689902aaed0SHisashi Hifumi /* No harm done if a racing interrupt already did this */ 690b01b2141SIngo Molnar local_lock_irqsave(&lru_rotate.lock, flags); 691c2bc1681SMatthew Wilcox (Oracle) folio_batch_move_lru(fbatch, lru_move_tail_fn); 692b01b2141SIngo Molnar local_unlock_irqrestore(&lru_rotate.lock, flags); 693902aaed0SHisashi Hifumi } 69431560180SMinchan Kim 695a2d33b5dSMatthew Wilcox (Oracle) fbatch = &fbatches->lru_deactivate_file; 6967a3dbfe8SMatthew Wilcox (Oracle) if (folio_batch_count(fbatch)) 6977a3dbfe8SMatthew Wilcox (Oracle) folio_batch_move_lru(fbatch, lru_deactivate_file_fn); 698eb709b0dSShaohua Li 699a2d33b5dSMatthew Wilcox (Oracle) fbatch = &fbatches->lru_deactivate; 70085cd7791SMatthew Wilcox (Oracle) if (folio_batch_count(fbatch)) 70185cd7791SMatthew Wilcox (Oracle) folio_batch_move_lru(fbatch, lru_deactivate_fn); 7029c276cc6SMinchan Kim 703a2d33b5dSMatthew Wilcox (Oracle) fbatch = &fbatches->lru_lazyfree; 704cec394baSMatthew Wilcox (Oracle) if (folio_batch_count(fbatch)) 705cec394baSMatthew Wilcox (Oracle) folio_batch_move_lru(fbatch, lru_lazyfree_fn); 70610853a03SMinchan Kim 7073a44610bSMatthew Wilcox (Oracle) folio_activate_drain(cpu); 70831560180SMinchan Kim } 70931560180SMinchan Kim 71031560180SMinchan Kim /** 7117a3dbfe8SMatthew Wilcox (Oracle) * deactivate_file_folio() - Deactivate a file folio. 712261b6840SMatthew Wilcox (Oracle) * @folio: Folio to deactivate. 71331560180SMinchan Kim * 714261b6840SMatthew Wilcox (Oracle) * This function hints to the VM that @folio is a good reclaim candidate, 715261b6840SMatthew Wilcox (Oracle) * for example if its invalidation fails due to the folio being dirty 71631560180SMinchan Kim * or under writeback. 717261b6840SMatthew Wilcox (Oracle) * 7187a3dbfe8SMatthew Wilcox (Oracle) * Context: Caller holds a reference on the folio. 71931560180SMinchan Kim */ 720261b6840SMatthew Wilcox (Oracle) void deactivate_file_folio(struct folio *folio) 72131560180SMinchan Kim { 7227a3dbfe8SMatthew Wilcox (Oracle) struct folio_batch *fbatch; 723b01b2141SIngo Molnar 7247a3dbfe8SMatthew Wilcox (Oracle) /* Deactivating an unevictable folio will not accelerate reclaim */ 725261b6840SMatthew Wilcox (Oracle) if (folio_test_unevictable(folio)) 726261b6840SMatthew Wilcox (Oracle) return; 727261b6840SMatthew Wilcox (Oracle) 728261b6840SMatthew Wilcox (Oracle) folio_get(folio); 72982ac64d8SMatthew Wilcox (Oracle) local_lock(&cpu_fbatches.lock); 73082ac64d8SMatthew Wilcox (Oracle) fbatch = this_cpu_ptr(&cpu_fbatches.lru_deactivate_file); 7317a3dbfe8SMatthew Wilcox (Oracle) folio_batch_add_and_move(fbatch, folio, lru_deactivate_file_fn); 73282ac64d8SMatthew Wilcox (Oracle) local_unlock(&cpu_fbatches.lock); 73331560180SMinchan Kim } 73480bfed90SAndrew Morton 7359c276cc6SMinchan Kim /* 7365a9e3474SVishal Moola (Oracle) * folio_deactivate - deactivate a folio 7375a9e3474SVishal Moola (Oracle) * @folio: folio to deactivate 7389c276cc6SMinchan Kim * 7395a9e3474SVishal Moola (Oracle) * folio_deactivate() moves @folio to the inactive list if @folio was on the 7405a9e3474SVishal Moola (Oracle) * active list and was not unevictable. This is done to accelerate the 7415a9e3474SVishal Moola (Oracle) * reclaim of @folio. 7429c276cc6SMinchan Kim */ 7435a9e3474SVishal Moola (Oracle) void folio_deactivate(struct folio *folio) 7449c276cc6SMinchan Kim { 745ec1c86b2SYu Zhao if (folio_test_lru(folio) && !folio_test_unevictable(folio) && 746ec1c86b2SYu Zhao (folio_test_active(folio) || lru_gen_enabled())) { 74785cd7791SMatthew Wilcox (Oracle) struct folio_batch *fbatch; 74885cd7791SMatthew Wilcox (Oracle) 74985cd7791SMatthew Wilcox (Oracle) folio_get(folio); 75082ac64d8SMatthew Wilcox (Oracle) local_lock(&cpu_fbatches.lock); 75182ac64d8SMatthew Wilcox (Oracle) fbatch = this_cpu_ptr(&cpu_fbatches.lru_deactivate); 75285cd7791SMatthew Wilcox (Oracle) folio_batch_add_and_move(fbatch, folio, lru_deactivate_fn); 75382ac64d8SMatthew Wilcox (Oracle) local_unlock(&cpu_fbatches.lock); 7549c276cc6SMinchan Kim } 7559c276cc6SMinchan Kim } 7569c276cc6SMinchan Kim 75710853a03SMinchan Kim /** 7586a6fe9ebSKefeng Wang * folio_mark_lazyfree - make an anon folio lazyfree 7596a6fe9ebSKefeng Wang * @folio: folio to deactivate 76010853a03SMinchan Kim * 7616a6fe9ebSKefeng Wang * folio_mark_lazyfree() moves @folio to the inactive file list. 7626a6fe9ebSKefeng Wang * This is done to accelerate the reclaim of @folio. 76310853a03SMinchan Kim */ 7646a6fe9ebSKefeng Wang void folio_mark_lazyfree(struct folio *folio) 76510853a03SMinchan Kim { 766cec394baSMatthew Wilcox (Oracle) if (folio_test_lru(folio) && folio_test_anon(folio) && 767cec394baSMatthew Wilcox (Oracle) folio_test_swapbacked(folio) && !folio_test_swapcache(folio) && 768cec394baSMatthew Wilcox (Oracle) !folio_test_unevictable(folio)) { 769cec394baSMatthew Wilcox (Oracle) struct folio_batch *fbatch; 770cec394baSMatthew Wilcox (Oracle) 771cec394baSMatthew Wilcox (Oracle) folio_get(folio); 77282ac64d8SMatthew Wilcox (Oracle) local_lock(&cpu_fbatches.lock); 77382ac64d8SMatthew Wilcox (Oracle) fbatch = this_cpu_ptr(&cpu_fbatches.lru_lazyfree); 774cec394baSMatthew Wilcox (Oracle) folio_batch_add_and_move(fbatch, folio, lru_lazyfree_fn); 77582ac64d8SMatthew Wilcox (Oracle) local_unlock(&cpu_fbatches.lock); 77610853a03SMinchan Kim } 77710853a03SMinchan Kim } 77810853a03SMinchan Kim 77980bfed90SAndrew Morton void lru_add_drain(void) 78080bfed90SAndrew Morton { 78182ac64d8SMatthew Wilcox (Oracle) local_lock(&cpu_fbatches.lock); 782b01b2141SIngo Molnar lru_add_drain_cpu(smp_processor_id()); 78382ac64d8SMatthew Wilcox (Oracle) local_unlock(&cpu_fbatches.lock); 784*96f97c43SLorenzo Stoakes mlock_drain_local(); 785b01b2141SIngo Molnar } 786b01b2141SIngo Molnar 787243418e3SMinchan Kim /* 788243418e3SMinchan Kim * It's called from per-cpu workqueue context in SMP case so 789243418e3SMinchan Kim * lru_add_drain_cpu and invalidate_bh_lrus_cpu should run on 790243418e3SMinchan Kim * the same cpu. It shouldn't be a problem in !SMP case since 791243418e3SMinchan Kim * the core is only one and the locks will disable preemption. 792243418e3SMinchan Kim */ 793243418e3SMinchan Kim static void lru_add_and_bh_lrus_drain(void) 794243418e3SMinchan Kim { 79582ac64d8SMatthew Wilcox (Oracle) local_lock(&cpu_fbatches.lock); 796243418e3SMinchan Kim lru_add_drain_cpu(smp_processor_id()); 79782ac64d8SMatthew Wilcox (Oracle) local_unlock(&cpu_fbatches.lock); 798243418e3SMinchan Kim invalidate_bh_lrus_cpu(); 799*96f97c43SLorenzo Stoakes mlock_drain_local(); 800243418e3SMinchan Kim } 801243418e3SMinchan Kim 802b01b2141SIngo Molnar void lru_add_drain_cpu_zone(struct zone *zone) 803b01b2141SIngo Molnar { 80482ac64d8SMatthew Wilcox (Oracle) local_lock(&cpu_fbatches.lock); 805b01b2141SIngo Molnar lru_add_drain_cpu(smp_processor_id()); 806b01b2141SIngo Molnar drain_local_pages(zone); 80782ac64d8SMatthew Wilcox (Oracle) local_unlock(&cpu_fbatches.lock); 808*96f97c43SLorenzo Stoakes mlock_drain_local(); 8091da177e4SLinus Torvalds } 8101da177e4SLinus Torvalds 8116ea183d6SMichal Hocko #ifdef CONFIG_SMP 8126ea183d6SMichal Hocko 8136ea183d6SMichal Hocko static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work); 8146ea183d6SMichal Hocko 815c4028958SDavid Howells static void lru_add_drain_per_cpu(struct work_struct *dummy) 816053837fcSNick Piggin { 817243418e3SMinchan Kim lru_add_and_bh_lrus_drain(); 818053837fcSNick Piggin } 819053837fcSNick Piggin 8204864545aSMatthew Wilcox (Oracle) static bool cpu_needs_drain(unsigned int cpu) 8214864545aSMatthew Wilcox (Oracle) { 8224864545aSMatthew Wilcox (Oracle) struct cpu_fbatches *fbatches = &per_cpu(cpu_fbatches, cpu); 8234864545aSMatthew Wilcox (Oracle) 8244864545aSMatthew Wilcox (Oracle) /* Check these in order of likelihood that they're not zero */ 8254864545aSMatthew Wilcox (Oracle) return folio_batch_count(&fbatches->lru_add) || 8264864545aSMatthew Wilcox (Oracle) data_race(folio_batch_count(&per_cpu(lru_rotate.fbatch, cpu))) || 8274864545aSMatthew Wilcox (Oracle) folio_batch_count(&fbatches->lru_deactivate_file) || 8284864545aSMatthew Wilcox (Oracle) folio_batch_count(&fbatches->lru_deactivate) || 8294864545aSMatthew Wilcox (Oracle) folio_batch_count(&fbatches->lru_lazyfree) || 8304864545aSMatthew Wilcox (Oracle) folio_batch_count(&fbatches->activate) || 831*96f97c43SLorenzo Stoakes need_mlock_drain(cpu) || 8324864545aSMatthew Wilcox (Oracle) has_bh_in_lru(cpu, NULL); 8334864545aSMatthew Wilcox (Oracle) } 8344864545aSMatthew Wilcox (Oracle) 8359852a721SMichal Hocko /* 8369852a721SMichal Hocko * Doesn't need any cpu hotplug locking because we do rely on per-cpu 8379852a721SMichal Hocko * kworkers being shut down before our page_alloc_cpu_dead callback is 8389852a721SMichal Hocko * executed on the offlined cpu. 8399852a721SMichal Hocko * Calling this function with cpu hotplug locks held can actually lead 8409852a721SMichal Hocko * to obscure indirect dependencies via WQ context. 8419852a721SMichal Hocko */ 8423db3264dSMiaohe Lin static inline void __lru_add_drain_all(bool force_all_cpus) 843053837fcSNick Piggin { 8446446a513SAhmed S. Darwish /* 8456446a513SAhmed S. Darwish * lru_drain_gen - Global pages generation number 8466446a513SAhmed S. Darwish * 8476446a513SAhmed S. Darwish * (A) Definition: global lru_drain_gen = x implies that all generations 8486446a513SAhmed S. Darwish * 0 < n <= x are already *scheduled* for draining. 8496446a513SAhmed S. Darwish * 8506446a513SAhmed S. Darwish * This is an optimization for the highly-contended use case where a 8516446a513SAhmed S. Darwish * user space workload keeps constantly generating a flow of pages for 8526446a513SAhmed S. Darwish * each CPU. 8536446a513SAhmed S. Darwish */ 8546446a513SAhmed S. Darwish static unsigned int lru_drain_gen; 8555fbc4616SChris Metcalf static struct cpumask has_work; 8566446a513SAhmed S. Darwish static DEFINE_MUTEX(lock); 8576446a513SAhmed S. Darwish unsigned cpu, this_gen; 8585fbc4616SChris Metcalf 859ce612879SMichal Hocko /* 860ce612879SMichal Hocko * Make sure nobody triggers this path before mm_percpu_wq is fully 861ce612879SMichal Hocko * initialized. 862ce612879SMichal Hocko */ 863ce612879SMichal Hocko if (WARN_ON(!mm_percpu_wq)) 864ce612879SMichal Hocko return; 865ce612879SMichal Hocko 8666446a513SAhmed S. Darwish /* 86782ac64d8SMatthew Wilcox (Oracle) * Guarantee folio_batch counter stores visible by this CPU 86882ac64d8SMatthew Wilcox (Oracle) * are visible to other CPUs before loading the current drain 86982ac64d8SMatthew Wilcox (Oracle) * generation. 8706446a513SAhmed S. Darwish */ 8716446a513SAhmed S. Darwish smp_mb(); 8726446a513SAhmed S. Darwish 8736446a513SAhmed S. Darwish /* 8746446a513SAhmed S. Darwish * (B) Locally cache global LRU draining generation number 8756446a513SAhmed S. Darwish * 8766446a513SAhmed S. Darwish * The read barrier ensures that the counter is loaded before the mutex 8776446a513SAhmed S. Darwish * is taken. It pairs with smp_mb() inside the mutex critical section 8786446a513SAhmed S. Darwish * at (D). 8796446a513SAhmed S. Darwish */ 8806446a513SAhmed S. Darwish this_gen = smp_load_acquire(&lru_drain_gen); 881eef1a429SKonstantin Khlebnikov 8825fbc4616SChris Metcalf mutex_lock(&lock); 883eef1a429SKonstantin Khlebnikov 884eef1a429SKonstantin Khlebnikov /* 8856446a513SAhmed S. Darwish * (C) Exit the draining operation if a newer generation, from another 8866446a513SAhmed S. Darwish * lru_add_drain_all(), was already scheduled for draining. Check (A). 887eef1a429SKonstantin Khlebnikov */ 888d479960eSMinchan Kim if (unlikely(this_gen != lru_drain_gen && !force_all_cpus)) 889eef1a429SKonstantin Khlebnikov goto done; 890eef1a429SKonstantin Khlebnikov 8916446a513SAhmed S. Darwish /* 8926446a513SAhmed S. Darwish * (D) Increment global generation number 8936446a513SAhmed S. Darwish * 8946446a513SAhmed S. Darwish * Pairs with smp_load_acquire() at (B), outside of the critical 89582ac64d8SMatthew Wilcox (Oracle) * section. Use a full memory barrier to guarantee that the 89682ac64d8SMatthew Wilcox (Oracle) * new global drain generation number is stored before loading 89782ac64d8SMatthew Wilcox (Oracle) * folio_batch counters. 8986446a513SAhmed S. Darwish * 8996446a513SAhmed S. Darwish * This pairing must be done here, before the for_each_online_cpu loop 9006446a513SAhmed S. Darwish * below which drains the page vectors. 9016446a513SAhmed S. Darwish * 9026446a513SAhmed S. Darwish * Let x, y, and z represent some system CPU numbers, where x < y < z. 903cb152a1aSShijie Luo * Assume CPU #z is in the middle of the for_each_online_cpu loop 9046446a513SAhmed S. Darwish * below and has already reached CPU #y's per-cpu data. CPU #x comes 9056446a513SAhmed S. Darwish * along, adds some pages to its per-cpu vectors, then calls 9066446a513SAhmed S. Darwish * lru_add_drain_all(). 9076446a513SAhmed S. Darwish * 9086446a513SAhmed S. Darwish * If the paired barrier is done at any later step, e.g. after the 9096446a513SAhmed S. Darwish * loop, CPU #x will just exit at (C) and miss flushing out all of its 9106446a513SAhmed S. Darwish * added pages. 9116446a513SAhmed S. Darwish */ 9126446a513SAhmed S. Darwish WRITE_ONCE(lru_drain_gen, lru_drain_gen + 1); 9136446a513SAhmed S. Darwish smp_mb(); 914eef1a429SKonstantin Khlebnikov 9155fbc4616SChris Metcalf cpumask_clear(&has_work); 9165fbc4616SChris Metcalf for_each_online_cpu(cpu) { 9175fbc4616SChris Metcalf struct work_struct *work = &per_cpu(lru_add_drain_work, cpu); 9185fbc4616SChris Metcalf 9194864545aSMatthew Wilcox (Oracle) if (cpu_needs_drain(cpu)) { 9205fbc4616SChris Metcalf INIT_WORK(work, lru_add_drain_per_cpu); 921ce612879SMichal Hocko queue_work_on(cpu, mm_percpu_wq, work); 9226446a513SAhmed S. Darwish __cpumask_set_cpu(cpu, &has_work); 9235fbc4616SChris Metcalf } 9245fbc4616SChris Metcalf } 9255fbc4616SChris Metcalf 9265fbc4616SChris Metcalf for_each_cpu(cpu, &has_work) 9275fbc4616SChris Metcalf flush_work(&per_cpu(lru_add_drain_work, cpu)); 9285fbc4616SChris Metcalf 929eef1a429SKonstantin Khlebnikov done: 9305fbc4616SChris Metcalf mutex_unlock(&lock); 931053837fcSNick Piggin } 932d479960eSMinchan Kim 933d479960eSMinchan Kim void lru_add_drain_all(void) 934d479960eSMinchan Kim { 935d479960eSMinchan Kim __lru_add_drain_all(false); 936d479960eSMinchan Kim } 9376ea183d6SMichal Hocko #else 9386ea183d6SMichal Hocko void lru_add_drain_all(void) 9396ea183d6SMichal Hocko { 9406ea183d6SMichal Hocko lru_add_drain(); 9416ea183d6SMichal Hocko } 9426446a513SAhmed S. Darwish #endif /* CONFIG_SMP */ 943053837fcSNick Piggin 944d479960eSMinchan Kim atomic_t lru_disable_count = ATOMIC_INIT(0); 945d479960eSMinchan Kim 946d479960eSMinchan Kim /* 947d479960eSMinchan Kim * lru_cache_disable() needs to be called before we start compiling 948d479960eSMinchan Kim * a list of pages to be migrated using isolate_lru_page(). 949d479960eSMinchan Kim * It drains pages on LRU cache and then disable on all cpus until 950d479960eSMinchan Kim * lru_cache_enable is called. 951d479960eSMinchan Kim * 952d479960eSMinchan Kim * Must be paired with a call to lru_cache_enable(). 953d479960eSMinchan Kim */ 954d479960eSMinchan Kim void lru_cache_disable(void) 955d479960eSMinchan Kim { 956d479960eSMinchan Kim atomic_inc(&lru_disable_count); 957d479960eSMinchan Kim /* 958ff042f4aSMarcelo Tosatti * Readers of lru_disable_count are protected by either disabling 959ff042f4aSMarcelo Tosatti * preemption or rcu_read_lock: 960ff042f4aSMarcelo Tosatti * 961ff042f4aSMarcelo Tosatti * preempt_disable, local_irq_disable [bh_lru_lock()] 962ff042f4aSMarcelo Tosatti * rcu_read_lock [rt_spin_lock CONFIG_PREEMPT_RT] 963ff042f4aSMarcelo Tosatti * preempt_disable [local_lock !CONFIG_PREEMPT_RT] 964ff042f4aSMarcelo Tosatti * 965ff042f4aSMarcelo Tosatti * Since v5.1 kernel, synchronize_rcu() is guaranteed to wait on 966ff042f4aSMarcelo Tosatti * preempt_disable() regions of code. So any CPU which sees 967ff042f4aSMarcelo Tosatti * lru_disable_count = 0 will have exited the critical 968ff042f4aSMarcelo Tosatti * section when synchronize_rcu() returns. 969d479960eSMinchan Kim */ 97031733463SMarcelo Tosatti synchronize_rcu_expedited(); 971ff042f4aSMarcelo Tosatti #ifdef CONFIG_SMP 972d479960eSMinchan Kim __lru_add_drain_all(true); 973d479960eSMinchan Kim #else 974243418e3SMinchan Kim lru_add_and_bh_lrus_drain(); 975d479960eSMinchan Kim #endif 976d479960eSMinchan Kim } 977d479960eSMinchan Kim 978aabfb572SMichal Hocko /** 979ea1754a0SKirill A. Shutemov * release_pages - batched put_page() 980449c7967SLinus Torvalds * @arg: array of pages to release 981aabfb572SMichal Hocko * @nr: number of pages 9821da177e4SLinus Torvalds * 983449c7967SLinus Torvalds * Decrement the reference count on all the pages in @arg. If it 984aabfb572SMichal Hocko * fell to zero, remove the page from the LRU and free it. 985449c7967SLinus Torvalds * 986449c7967SLinus Torvalds * Note that the argument can be an array of pages, encoded pages, 987449c7967SLinus Torvalds * or folio pointers. We ignore any encoded bits, and turn any of 988449c7967SLinus Torvalds * them into just a folio that gets free'd. 9891da177e4SLinus Torvalds */ 990449c7967SLinus Torvalds void release_pages(release_pages_arg arg, int nr) 9911da177e4SLinus Torvalds { 9921da177e4SLinus Torvalds int i; 993449c7967SLinus Torvalds struct encoded_page **encoded = arg.encoded_pages; 994cc59850eSKonstantin Khlebnikov LIST_HEAD(pages_to_free); 9956168d0daSAlex Shi struct lruvec *lruvec = NULL; 9960de340cbSMatthew Wilcox (Oracle) unsigned long flags = 0; 9973f649ab7SKees Cook unsigned int lock_batch; 9981da177e4SLinus Torvalds 9991da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 1000449c7967SLinus Torvalds struct folio *folio; 1001449c7967SLinus Torvalds 1002449c7967SLinus Torvalds /* Turn any of the argument types into a folio */ 1003449c7967SLinus Torvalds folio = page_folio(encoded_page_ptr(encoded[i])); 10041da177e4SLinus Torvalds 1005aabfb572SMichal Hocko /* 1006aabfb572SMichal Hocko * Make sure the IRQ-safe lock-holding time does not get 1007aabfb572SMichal Hocko * excessive with a continuous string of pages from the 10086168d0daSAlex Shi * same lruvec. The lock is held only if lruvec != NULL. 1009aabfb572SMichal Hocko */ 10106168d0daSAlex Shi if (lruvec && ++lock_batch == SWAP_CLUSTER_MAX) { 10116168d0daSAlex Shi unlock_page_lruvec_irqrestore(lruvec, flags); 10126168d0daSAlex Shi lruvec = NULL; 1013aabfb572SMichal Hocko } 1014aabfb572SMichal Hocko 1015ab5e653eSMatthew Wilcox (Oracle) if (is_huge_zero_page(&folio->page)) 1016aa88b68cSKirill A. Shutemov continue; 1017aa88b68cSKirill A. Shutemov 1018ab5e653eSMatthew Wilcox (Oracle) if (folio_is_zone_device(folio)) { 10196168d0daSAlex Shi if (lruvec) { 10206168d0daSAlex Shi unlock_page_lruvec_irqrestore(lruvec, flags); 10216168d0daSAlex Shi lruvec = NULL; 1022df6ad698SJérôme Glisse } 1023ab5e653eSMatthew Wilcox (Oracle) if (put_devmap_managed_page(&folio->page)) 1024df6ad698SJérôme Glisse continue; 1025ab5e653eSMatthew Wilcox (Oracle) if (folio_put_testzero(folio)) 1026ab5e653eSMatthew Wilcox (Oracle) free_zone_device_page(&folio->page); 102743fbdeb3SRalph Campbell continue; 102807d80269SJohn Hubbard } 1029df6ad698SJérôme Glisse 1030ab5e653eSMatthew Wilcox (Oracle) if (!folio_put_testzero(folio)) 10311da177e4SLinus Torvalds continue; 10321da177e4SLinus Torvalds 1033ab5e653eSMatthew Wilcox (Oracle) if (folio_test_large(folio)) { 10346168d0daSAlex Shi if (lruvec) { 10356168d0daSAlex Shi unlock_page_lruvec_irqrestore(lruvec, flags); 10366168d0daSAlex Shi lruvec = NULL; 1037ddc58f27SKirill A. Shutemov } 10385ef82fe7SMatthew Wilcox (Oracle) __folio_put_large(folio); 1039ddc58f27SKirill A. Shutemov continue; 1040ddc58f27SKirill A. Shutemov } 1041ddc58f27SKirill A. Shutemov 1042ab5e653eSMatthew Wilcox (Oracle) if (folio_test_lru(folio)) { 10432a5e4e34SAlexander Duyck struct lruvec *prev_lruvec = lruvec; 1044894bc310SLee Schermerhorn 10450de340cbSMatthew Wilcox (Oracle) lruvec = folio_lruvec_relock_irqsave(folio, lruvec, 10462a5e4e34SAlexander Duyck &flags); 10472a5e4e34SAlexander Duyck if (prev_lruvec != lruvec) 1048aabfb572SMichal Hocko lock_batch = 0; 1049fa9add64SHugh Dickins 1050ab5e653eSMatthew Wilcox (Oracle) lruvec_del_folio(lruvec, folio); 1051ab5e653eSMatthew Wilcox (Oracle) __folio_clear_lru_flags(folio); 105246453a6eSNick Piggin } 105346453a6eSNick Piggin 1054b109b870SHugh Dickins /* 1055b109b870SHugh Dickins * In rare cases, when truncation or holepunching raced with 1056b109b870SHugh Dickins * munlock after VM_LOCKED was cleared, Mlocked may still be 1057b109b870SHugh Dickins * found set here. This does not indicate a problem, unless 1058b109b870SHugh Dickins * "unevictable_pgs_cleared" appears worryingly large. 1059b109b870SHugh Dickins */ 1060ab5e653eSMatthew Wilcox (Oracle) if (unlikely(folio_test_mlocked(folio))) { 1061ab5e653eSMatthew Wilcox (Oracle) __folio_clear_mlocked(folio); 1062ab5e653eSMatthew Wilcox (Oracle) zone_stat_sub_folio(folio, NR_MLOCK); 1063b109b870SHugh Dickins count_vm_event(UNEVICTABLE_PGCLEARED); 1064b109b870SHugh Dickins } 1065b109b870SHugh Dickins 1066ab5e653eSMatthew Wilcox (Oracle) list_add(&folio->lru, &pages_to_free); 10671da177e4SLinus Torvalds } 10686168d0daSAlex Shi if (lruvec) 10696168d0daSAlex Shi unlock_page_lruvec_irqrestore(lruvec, flags); 10701da177e4SLinus Torvalds 1071747db954SJohannes Weiner mem_cgroup_uncharge_list(&pages_to_free); 10722d4894b5SMel Gorman free_unref_page_list(&pages_to_free); 10731da177e4SLinus Torvalds } 10740be8557bSMiklos Szeredi EXPORT_SYMBOL(release_pages); 10751da177e4SLinus Torvalds 10761da177e4SLinus Torvalds /* 10771da177e4SLinus Torvalds * The pages which we're about to release may be in the deferred lru-addition 10781da177e4SLinus Torvalds * queues. That would prevent them from really being freed right now. That's 10791da177e4SLinus Torvalds * OK from a correctness point of view but is inefficient - those pages may be 10801da177e4SLinus Torvalds * cache-warm and we want to give them back to the page allocator ASAP. 10811da177e4SLinus Torvalds * 108270dea534SMatthew Wilcox (Oracle) * So __pagevec_release() will drain those queues here. 108370dea534SMatthew Wilcox (Oracle) * folio_batch_move_lru() calls folios_put() directly to avoid 10841da177e4SLinus Torvalds * mutual recursion. 10851da177e4SLinus Torvalds */ 10861da177e4SLinus Torvalds void __pagevec_release(struct pagevec *pvec) 10871da177e4SLinus Torvalds { 10887f0b5fb9SMel Gorman if (!pvec->percpu_pvec_drained) { 10891da177e4SLinus Torvalds lru_add_drain(); 10907f0b5fb9SMel Gorman pvec->percpu_pvec_drained = true; 1091d9ed0d08SMel Gorman } 1092c6f92f9fSMel Gorman release_pages(pvec->pages, pagevec_count(pvec)); 10931da177e4SLinus Torvalds pagevec_reinit(pvec); 10941da177e4SLinus Torvalds } 10957f285701SSteve French EXPORT_SYMBOL(__pagevec_release); 10967f285701SSteve French 10971da177e4SLinus Torvalds /** 10981613fac9SMatthew Wilcox (Oracle) * folio_batch_remove_exceptionals() - Prune non-folios from a batch. 10991613fac9SMatthew Wilcox (Oracle) * @fbatch: The batch to prune 11000cd6144aSJohannes Weiner * 11011613fac9SMatthew Wilcox (Oracle) * find_get_entries() fills a batch with both folios and shadow/swap/DAX 11021613fac9SMatthew Wilcox (Oracle) * entries. This function prunes all the non-folio entries from @fbatch 11031613fac9SMatthew Wilcox (Oracle) * without leaving holes, so that it can be passed on to folio-only batch 11041613fac9SMatthew Wilcox (Oracle) * operations. 11050cd6144aSJohannes Weiner */ 11061613fac9SMatthew Wilcox (Oracle) void folio_batch_remove_exceptionals(struct folio_batch *fbatch) 11070cd6144aSJohannes Weiner { 11081613fac9SMatthew Wilcox (Oracle) unsigned int i, j; 11090cd6144aSJohannes Weiner 11101613fac9SMatthew Wilcox (Oracle) for (i = 0, j = 0; i < folio_batch_count(fbatch); i++) { 11111613fac9SMatthew Wilcox (Oracle) struct folio *folio = fbatch->folios[i]; 11121613fac9SMatthew Wilcox (Oracle) if (!xa_is_value(folio)) 11131613fac9SMatthew Wilcox (Oracle) fbatch->folios[j++] = folio; 11140cd6144aSJohannes Weiner } 11151613fac9SMatthew Wilcox (Oracle) fbatch->nr = j; 11160cd6144aSJohannes Weiner } 11170cd6144aSJohannes Weiner 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