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 11*57043247SMauro 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> 381da177e4SLinus Torvalds 3964d6519dSLee Schermerhorn #include "internal.h" 4064d6519dSLee Schermerhorn 41c6286c98SMel Gorman #define CREATE_TRACE_POINTS 42c6286c98SMel Gorman #include <trace/events/pagemap.h> 43c6286c98SMel Gorman 441da177e4SLinus Torvalds /* How many pages do we try to swap or page in/out together? */ 451da177e4SLinus Torvalds int page_cluster; 461da177e4SLinus Torvalds 4713f7f789SMel Gorman static DEFINE_PER_CPU(struct pagevec, lru_add_pvec); 48f84f9504SVegard Nossum static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs); 49cc5993bdSMinchan Kim static DEFINE_PER_CPU(struct pagevec, lru_deactivate_file_pvecs); 50f7ad2a6cSShaohua Li static DEFINE_PER_CPU(struct pagevec, lru_lazyfree_pvecs); 51a4a921aaSMing Li #ifdef CONFIG_SMP 52a4a921aaSMing Li static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs); 53a4a921aaSMing Li #endif 54902aaed0SHisashi Hifumi 55b221385bSAdrian Bunk /* 56b221385bSAdrian Bunk * This path almost never happens for VM activity - pages are normally 57b221385bSAdrian Bunk * freed via pagevecs. But it gets used by networking. 58b221385bSAdrian Bunk */ 59920c7a5dSHarvey Harrison static void __page_cache_release(struct page *page) 60b221385bSAdrian Bunk { 61b221385bSAdrian Bunk if (PageLRU(page)) { 62f4b7e272SAndrey Ryabinin pg_data_t *pgdat = page_pgdat(page); 63fa9add64SHugh Dickins struct lruvec *lruvec; 64fa9add64SHugh Dickins unsigned long flags; 65b221385bSAdrian Bunk 66f4b7e272SAndrey Ryabinin spin_lock_irqsave(&pgdat->lru_lock, flags); 67f4b7e272SAndrey Ryabinin lruvec = mem_cgroup_page_lruvec(page, pgdat); 68309381feSSasha Levin VM_BUG_ON_PAGE(!PageLRU(page), page); 69b221385bSAdrian Bunk __ClearPageLRU(page); 70fa9add64SHugh Dickins del_page_from_lru_list(page, lruvec, page_off_lru(page)); 71f4b7e272SAndrey Ryabinin spin_unlock_irqrestore(&pgdat->lru_lock, flags); 72b221385bSAdrian Bunk } 7362906027SNicholas Piggin __ClearPageWaiters(page); 740a31bc97SJohannes Weiner mem_cgroup_uncharge(page); 7591807063SAndrea Arcangeli } 7691807063SAndrea Arcangeli 7791807063SAndrea Arcangeli static void __put_single_page(struct page *page) 7891807063SAndrea Arcangeli { 7991807063SAndrea Arcangeli __page_cache_release(page); 802d4894b5SMel Gorman free_unref_page(page); 81b221385bSAdrian Bunk } 82b221385bSAdrian Bunk 8391807063SAndrea Arcangeli static void __put_compound_page(struct page *page) 8491807063SAndrea Arcangeli { 8591807063SAndrea Arcangeli compound_page_dtor *dtor; 8691807063SAndrea Arcangeli 87822fc613SNaoya Horiguchi /* 88822fc613SNaoya Horiguchi * __page_cache_release() is supposed to be called for thp, not for 89822fc613SNaoya Horiguchi * hugetlb. This is because hugetlb page does never have PageLRU set 90822fc613SNaoya Horiguchi * (it's never listed to any LRU lists) and no memcg routines should 91822fc613SNaoya Horiguchi * be called for hugetlb (it has a separate hugetlb_cgroup.) 92822fc613SNaoya Horiguchi */ 93822fc613SNaoya Horiguchi if (!PageHuge(page)) 9491807063SAndrea Arcangeli __page_cache_release(page); 9591807063SAndrea Arcangeli dtor = get_compound_page_dtor(page); 9691807063SAndrea Arcangeli (*dtor)(page); 9791807063SAndrea Arcangeli } 9891807063SAndrea Arcangeli 99ddc58f27SKirill A. Shutemov void __put_page(struct page *page) 100c747ce79SJianyu Zhan { 10171389703SDan Williams if (is_zone_device_page(page)) { 10271389703SDan Williams put_dev_pagemap(page->pgmap); 10371389703SDan Williams 10471389703SDan Williams /* 10571389703SDan Williams * The page belongs to the device that created pgmap. Do 10671389703SDan Williams * not return it to page allocator. 10771389703SDan Williams */ 10871389703SDan Williams return; 10971389703SDan Williams } 11071389703SDan Williams 111ddc58f27SKirill A. Shutemov if (unlikely(PageCompound(page))) 11226296ad2SAndrew Morton __put_compound_page(page); 11326296ad2SAndrew Morton else 11426296ad2SAndrew Morton __put_single_page(page); 11526296ad2SAndrew Morton } 116ddc58f27SKirill A. Shutemov EXPORT_SYMBOL(__put_page); 11770b50f94SAndrea Arcangeli 1181d7ea732SAlexander Zarochentsev /** 1197682486bSRandy Dunlap * put_pages_list() - release a list of pages 1207682486bSRandy Dunlap * @pages: list of pages threaded on page->lru 1211d7ea732SAlexander Zarochentsev * 1221d7ea732SAlexander Zarochentsev * Release a list of pages which are strung together on page.lru. Currently 1231d7ea732SAlexander Zarochentsev * used by read_cache_pages() and related error recovery code. 1241d7ea732SAlexander Zarochentsev */ 1251d7ea732SAlexander Zarochentsev void put_pages_list(struct list_head *pages) 1261d7ea732SAlexander Zarochentsev { 1271d7ea732SAlexander Zarochentsev while (!list_empty(pages)) { 1281d7ea732SAlexander Zarochentsev struct page *victim; 1291d7ea732SAlexander Zarochentsev 130f86196eaSNikolay Borisov victim = lru_to_page(pages); 1311d7ea732SAlexander Zarochentsev list_del(&victim->lru); 13209cbfeafSKirill A. Shutemov put_page(victim); 1331d7ea732SAlexander Zarochentsev } 1341d7ea732SAlexander Zarochentsev } 1351d7ea732SAlexander Zarochentsev EXPORT_SYMBOL(put_pages_list); 1361d7ea732SAlexander Zarochentsev 13718022c5dSMel Gorman /* 13818022c5dSMel Gorman * get_kernel_pages() - pin kernel pages in memory 13918022c5dSMel Gorman * @kiov: An array of struct kvec structures 14018022c5dSMel Gorman * @nr_segs: number of segments to pin 14118022c5dSMel Gorman * @write: pinning for read/write, currently ignored 14218022c5dSMel Gorman * @pages: array that receives pointers to the pages pinned. 14318022c5dSMel Gorman * Should be at least nr_segs long. 14418022c5dSMel Gorman * 14518022c5dSMel Gorman * Returns number of pages pinned. This may be fewer than the number 14618022c5dSMel Gorman * requested. If nr_pages is 0 or negative, returns 0. If no pages 14718022c5dSMel Gorman * were pinned, returns -errno. Each page returned must be released 14818022c5dSMel Gorman * with a put_page() call when it is finished with. 14918022c5dSMel Gorman */ 15018022c5dSMel Gorman int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write, 15118022c5dSMel Gorman struct page **pages) 15218022c5dSMel Gorman { 15318022c5dSMel Gorman int seg; 15418022c5dSMel Gorman 15518022c5dSMel Gorman for (seg = 0; seg < nr_segs; seg++) { 15618022c5dSMel Gorman if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE)) 15718022c5dSMel Gorman return seg; 15818022c5dSMel Gorman 1595a178119SMel Gorman pages[seg] = kmap_to_page(kiov[seg].iov_base); 16009cbfeafSKirill A. Shutemov get_page(pages[seg]); 16118022c5dSMel Gorman } 16218022c5dSMel Gorman 16318022c5dSMel Gorman return seg; 16418022c5dSMel Gorman } 16518022c5dSMel Gorman EXPORT_SYMBOL_GPL(get_kernel_pages); 16618022c5dSMel Gorman 16718022c5dSMel Gorman /* 16818022c5dSMel Gorman * get_kernel_page() - pin a kernel page in memory 16918022c5dSMel Gorman * @start: starting kernel address 17018022c5dSMel Gorman * @write: pinning for read/write, currently ignored 17118022c5dSMel Gorman * @pages: array that receives pointer to the page pinned. 17218022c5dSMel Gorman * Must be at least nr_segs long. 17318022c5dSMel Gorman * 17418022c5dSMel Gorman * Returns 1 if page is pinned. If the page was not pinned, returns 17518022c5dSMel Gorman * -errno. The page returned must be released with a put_page() call 17618022c5dSMel Gorman * when it is finished with. 17718022c5dSMel Gorman */ 17818022c5dSMel Gorman int get_kernel_page(unsigned long start, int write, struct page **pages) 17918022c5dSMel Gorman { 18018022c5dSMel Gorman const struct kvec kiov = { 18118022c5dSMel Gorman .iov_base = (void *)start, 18218022c5dSMel Gorman .iov_len = PAGE_SIZE 18318022c5dSMel Gorman }; 18418022c5dSMel Gorman 18518022c5dSMel Gorman return get_kernel_pages(&kiov, 1, write, pages); 18618022c5dSMel Gorman } 18718022c5dSMel Gorman EXPORT_SYMBOL_GPL(get_kernel_page); 18818022c5dSMel Gorman 1893dd7ae8eSShaohua Li static void pagevec_lru_move_fn(struct pagevec *pvec, 190fa9add64SHugh Dickins void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg), 1913dd7ae8eSShaohua Li void *arg) 192902aaed0SHisashi Hifumi { 193902aaed0SHisashi Hifumi int i; 19468eb0731SMel Gorman struct pglist_data *pgdat = NULL; 195fa9add64SHugh Dickins struct lruvec *lruvec; 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]; 20068eb0731SMel Gorman struct pglist_data *pagepgdat = page_pgdat(page); 201902aaed0SHisashi Hifumi 20268eb0731SMel Gorman if (pagepgdat != pgdat) { 20368eb0731SMel Gorman if (pgdat) 20468eb0731SMel Gorman spin_unlock_irqrestore(&pgdat->lru_lock, flags); 20568eb0731SMel Gorman pgdat = pagepgdat; 20668eb0731SMel Gorman spin_lock_irqsave(&pgdat->lru_lock, flags); 207902aaed0SHisashi Hifumi } 2083dd7ae8eSShaohua Li 20968eb0731SMel Gorman lruvec = mem_cgroup_page_lruvec(page, pgdat); 210fa9add64SHugh Dickins (*move_fn)(page, lruvec, arg); 2113dd7ae8eSShaohua Li } 21268eb0731SMel Gorman if (pgdat) 21368eb0731SMel Gorman spin_unlock_irqrestore(&pgdat->lru_lock, flags); 214c6f92f9fSMel Gorman release_pages(pvec->pages, pvec->nr); 2153dd7ae8eSShaohua Li pagevec_reinit(pvec); 2163dd7ae8eSShaohua Li } 2173dd7ae8eSShaohua Li 218fa9add64SHugh Dickins static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec, 219fa9add64SHugh Dickins void *arg) 2203dd7ae8eSShaohua Li { 2213dd7ae8eSShaohua Li int *pgmoved = arg; 2223dd7ae8eSShaohua Li 223c55e8d03SJohannes Weiner if (PageLRU(page) && !PageUnevictable(page)) { 224c55e8d03SJohannes Weiner del_page_from_lru_list(page, lruvec, page_lru(page)); 225c55e8d03SJohannes Weiner ClearPageActive(page); 226c55e8d03SJohannes Weiner add_page_to_lru_list_tail(page, lruvec, page_lru(page)); 2273dd7ae8eSShaohua Li (*pgmoved)++; 228902aaed0SHisashi Hifumi } 229902aaed0SHisashi Hifumi } 2303dd7ae8eSShaohua Li 2313dd7ae8eSShaohua Li /* 2323dd7ae8eSShaohua Li * pagevec_move_tail() must be called with IRQ disabled. 2333dd7ae8eSShaohua Li * Otherwise this may cause nasty races. 2343dd7ae8eSShaohua Li */ 2353dd7ae8eSShaohua Li static void pagevec_move_tail(struct pagevec *pvec) 2363dd7ae8eSShaohua Li { 2373dd7ae8eSShaohua Li int pgmoved = 0; 2383dd7ae8eSShaohua Li 2393dd7ae8eSShaohua Li pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved); 240902aaed0SHisashi Hifumi __count_vm_events(PGROTATED, pgmoved); 241902aaed0SHisashi Hifumi } 242902aaed0SHisashi Hifumi 243902aaed0SHisashi Hifumi /* 2441da177e4SLinus Torvalds * Writeback is about to end against a page which has been marked for immediate 2451da177e4SLinus Torvalds * reclaim. If it still appears to be reclaimable, move it to the tail of the 246902aaed0SHisashi Hifumi * inactive list. 2471da177e4SLinus Torvalds */ 248ac6aadb2SMiklos Szeredi void rotate_reclaimable_page(struct page *page) 2491da177e4SLinus Torvalds { 250c55e8d03SJohannes Weiner if (!PageLocked(page) && !PageDirty(page) && 251894bc310SLee Schermerhorn !PageUnevictable(page) && PageLRU(page)) { 252902aaed0SHisashi Hifumi struct pagevec *pvec; 2531da177e4SLinus Torvalds unsigned long flags; 2541da177e4SLinus Torvalds 25509cbfeafSKirill A. Shutemov get_page(page); 256902aaed0SHisashi Hifumi local_irq_save(flags); 2577c8e0181SChristoph Lameter pvec = this_cpu_ptr(&lru_rotate_pvecs); 2588f182270SLukasz Odzioba if (!pagevec_add(pvec, page) || PageCompound(page)) 259902aaed0SHisashi Hifumi pagevec_move_tail(pvec); 260902aaed0SHisashi Hifumi local_irq_restore(flags); 261ac6aadb2SMiklos Szeredi } 2621da177e4SLinus Torvalds } 2631da177e4SLinus Torvalds 264fa9add64SHugh Dickins static void update_page_reclaim_stat(struct lruvec *lruvec, 2653e2f41f1SKOSAKI Motohiro int file, int rotated) 2663e2f41f1SKOSAKI Motohiro { 267fa9add64SHugh Dickins struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; 2683e2f41f1SKOSAKI Motohiro 2693e2f41f1SKOSAKI Motohiro reclaim_stat->recent_scanned[file]++; 2703e2f41f1SKOSAKI Motohiro if (rotated) 2713e2f41f1SKOSAKI Motohiro reclaim_stat->recent_rotated[file]++; 2723e2f41f1SKOSAKI Motohiro } 2733e2f41f1SKOSAKI Motohiro 274fa9add64SHugh Dickins static void __activate_page(struct page *page, struct lruvec *lruvec, 275fa9add64SHugh Dickins void *arg) 276744ed144SShaohua Li { 2777a608572SLinus Torvalds if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { 278744ed144SShaohua Li int file = page_is_file_cache(page); 279744ed144SShaohua Li int lru = page_lru_base_type(page); 280744ed144SShaohua Li 281fa9add64SHugh Dickins del_page_from_lru_list(page, lruvec, lru); 282744ed144SShaohua Li SetPageActive(page); 283744ed144SShaohua Li lru += LRU_ACTIVE; 284fa9add64SHugh Dickins add_page_to_lru_list(page, lruvec, lru); 28524b7e581SMel Gorman trace_mm_lru_activate(page); 2867a608572SLinus Torvalds 287fa9add64SHugh Dickins __count_vm_event(PGACTIVATE); 288fa9add64SHugh Dickins update_page_reclaim_stat(lruvec, file, 1); 289744ed144SShaohua Li } 290eb709b0dSShaohua Li } 291eb709b0dSShaohua Li 292eb709b0dSShaohua Li #ifdef CONFIG_SMP 293eb709b0dSShaohua Li static void activate_page_drain(int cpu) 294eb709b0dSShaohua Li { 295eb709b0dSShaohua Li struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu); 296eb709b0dSShaohua Li 297eb709b0dSShaohua Li if (pagevec_count(pvec)) 298eb709b0dSShaohua Li pagevec_lru_move_fn(pvec, __activate_page, NULL); 299eb709b0dSShaohua Li } 300eb709b0dSShaohua Li 3015fbc4616SChris Metcalf static bool need_activate_page_drain(int cpu) 3025fbc4616SChris Metcalf { 3035fbc4616SChris Metcalf return pagevec_count(&per_cpu(activate_page_pvecs, cpu)) != 0; 3045fbc4616SChris Metcalf } 3055fbc4616SChris Metcalf 306eb709b0dSShaohua Li void activate_page(struct page *page) 307eb709b0dSShaohua Li { 308800d8c63SKirill A. Shutemov page = compound_head(page); 309eb709b0dSShaohua Li if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { 310eb709b0dSShaohua Li struct pagevec *pvec = &get_cpu_var(activate_page_pvecs); 311eb709b0dSShaohua Li 31209cbfeafSKirill A. Shutemov get_page(page); 3138f182270SLukasz Odzioba if (!pagevec_add(pvec, page) || PageCompound(page)) 314eb709b0dSShaohua Li pagevec_lru_move_fn(pvec, __activate_page, NULL); 315eb709b0dSShaohua Li put_cpu_var(activate_page_pvecs); 316eb709b0dSShaohua Li } 317eb709b0dSShaohua Li } 318eb709b0dSShaohua Li 319eb709b0dSShaohua Li #else 320eb709b0dSShaohua Li static inline void activate_page_drain(int cpu) 321eb709b0dSShaohua Li { 322eb709b0dSShaohua Li } 323eb709b0dSShaohua Li 324eb709b0dSShaohua Li void activate_page(struct page *page) 325eb709b0dSShaohua Li { 326f4b7e272SAndrey Ryabinin pg_data_t *pgdat = page_pgdat(page); 327eb709b0dSShaohua Li 328800d8c63SKirill A. Shutemov page = compound_head(page); 329f4b7e272SAndrey Ryabinin spin_lock_irq(&pgdat->lru_lock); 330f4b7e272SAndrey Ryabinin __activate_page(page, mem_cgroup_page_lruvec(page, pgdat), NULL); 331f4b7e272SAndrey Ryabinin spin_unlock_irq(&pgdat->lru_lock); 3321da177e4SLinus Torvalds } 333eb709b0dSShaohua Li #endif 3341da177e4SLinus Torvalds 335059285a2SMel Gorman static void __lru_cache_activate_page(struct page *page) 336059285a2SMel Gorman { 337059285a2SMel Gorman struct pagevec *pvec = &get_cpu_var(lru_add_pvec); 338059285a2SMel Gorman int i; 339059285a2SMel Gorman 340059285a2SMel Gorman /* 341059285a2SMel Gorman * Search backwards on the optimistic assumption that the page being 342059285a2SMel Gorman * activated has just been added to this pagevec. Note that only 343059285a2SMel Gorman * the local pagevec is examined as a !PageLRU page could be in the 344059285a2SMel Gorman * process of being released, reclaimed, migrated or on a remote 345059285a2SMel Gorman * pagevec that is currently being drained. Furthermore, marking 346059285a2SMel Gorman * a remote pagevec's page PageActive potentially hits a race where 347059285a2SMel Gorman * a page is marked PageActive just after it is added to the inactive 348059285a2SMel Gorman * list causing accounting errors and BUG_ON checks to trigger. 349059285a2SMel Gorman */ 350059285a2SMel Gorman for (i = pagevec_count(pvec) - 1; i >= 0; i--) { 351059285a2SMel Gorman struct page *pagevec_page = pvec->pages[i]; 352059285a2SMel Gorman 353059285a2SMel Gorman if (pagevec_page == page) { 354059285a2SMel Gorman SetPageActive(page); 355059285a2SMel Gorman break; 356059285a2SMel Gorman } 357059285a2SMel Gorman } 358059285a2SMel Gorman 359059285a2SMel Gorman put_cpu_var(lru_add_pvec); 360059285a2SMel Gorman } 361059285a2SMel Gorman 3621da177e4SLinus Torvalds /* 3631da177e4SLinus Torvalds * Mark a page as having seen activity. 3641da177e4SLinus Torvalds * 3651da177e4SLinus Torvalds * inactive,unreferenced -> inactive,referenced 3661da177e4SLinus Torvalds * inactive,referenced -> active,unreferenced 3671da177e4SLinus Torvalds * active,unreferenced -> active,referenced 368eb39d618SHugh Dickins * 369eb39d618SHugh Dickins * When a newly allocated page is not yet visible, so safe for non-atomic ops, 370eb39d618SHugh Dickins * __SetPageReferenced(page) may be substituted for mark_page_accessed(page). 3711da177e4SLinus Torvalds */ 372920c7a5dSHarvey Harrison void mark_page_accessed(struct page *page) 3731da177e4SLinus Torvalds { 374e90309c9SKirill A. Shutemov page = compound_head(page); 375894bc310SLee Schermerhorn if (!PageActive(page) && !PageUnevictable(page) && 376059285a2SMel Gorman PageReferenced(page)) { 377059285a2SMel Gorman 378059285a2SMel Gorman /* 379059285a2SMel Gorman * If the page is on the LRU, queue it for activation via 380059285a2SMel Gorman * activate_page_pvecs. Otherwise, assume the page is on a 381059285a2SMel Gorman * pagevec, mark it active and it'll be moved to the active 382059285a2SMel Gorman * LRU on the next drain. 383059285a2SMel Gorman */ 384059285a2SMel Gorman if (PageLRU(page)) 3851da177e4SLinus Torvalds activate_page(page); 386059285a2SMel Gorman else 387059285a2SMel Gorman __lru_cache_activate_page(page); 3881da177e4SLinus Torvalds ClearPageReferenced(page); 389a528910eSJohannes Weiner if (page_is_file_cache(page)) 390a528910eSJohannes Weiner workingset_activation(page); 3911da177e4SLinus Torvalds } else if (!PageReferenced(page)) { 3921da177e4SLinus Torvalds SetPageReferenced(page); 3931da177e4SLinus Torvalds } 39433c3fc71SVladimir Davydov if (page_is_idle(page)) 39533c3fc71SVladimir Davydov clear_page_idle(page); 3961da177e4SLinus Torvalds } 3971da177e4SLinus Torvalds EXPORT_SYMBOL(mark_page_accessed); 3981da177e4SLinus Torvalds 3992329d375SJianyu Zhan static void __lru_cache_add(struct page *page) 4001da177e4SLinus Torvalds { 40113f7f789SMel Gorman struct pagevec *pvec = &get_cpu_var(lru_add_pvec); 40213f7f789SMel Gorman 40309cbfeafSKirill A. Shutemov get_page(page); 4048f182270SLukasz Odzioba if (!pagevec_add(pvec, page) || PageCompound(page)) 405a0b8cab3SMel Gorman __pagevec_lru_add(pvec); 40613f7f789SMel Gorman put_cpu_var(lru_add_pvec); 4071da177e4SLinus Torvalds } 4082329d375SJianyu Zhan 4092329d375SJianyu Zhan /** 410e02a9f04SRandy Dunlap * lru_cache_add_anon - add a page to the page lists 4112329d375SJianyu Zhan * @page: the page to add 4122329d375SJianyu Zhan */ 4132329d375SJianyu Zhan void lru_cache_add_anon(struct page *page) 4142329d375SJianyu Zhan { 4156fb81a17SMel Gorman if (PageActive(page)) 4162329d375SJianyu Zhan ClearPageActive(page); 4172329d375SJianyu Zhan __lru_cache_add(page); 4182329d375SJianyu Zhan } 4192329d375SJianyu Zhan 4202329d375SJianyu Zhan void lru_cache_add_file(struct page *page) 4212329d375SJianyu Zhan { 4226fb81a17SMel Gorman if (PageActive(page)) 4232329d375SJianyu Zhan ClearPageActive(page); 4242329d375SJianyu Zhan __lru_cache_add(page); 4252329d375SJianyu Zhan } 4262329d375SJianyu Zhan EXPORT_SYMBOL(lru_cache_add_file); 4271da177e4SLinus Torvalds 428f04e9ebbSKOSAKI Motohiro /** 429c53954a0SMel Gorman * lru_cache_add - add a page to a page list 430f04e9ebbSKOSAKI Motohiro * @page: the page to be added to the LRU. 4312329d375SJianyu Zhan * 4322329d375SJianyu Zhan * Queue the page for addition to the LRU via pagevec. The decision on whether 4332329d375SJianyu Zhan * to add the page to the [in]active [file|anon] list is deferred until the 4342329d375SJianyu Zhan * pagevec is drained. This gives a chance for the caller of lru_cache_add() 4352329d375SJianyu Zhan * have the page added to the active list using mark_page_accessed(). 436f04e9ebbSKOSAKI Motohiro */ 437c53954a0SMel Gorman void lru_cache_add(struct page *page) 4381da177e4SLinus Torvalds { 439309381feSSasha Levin VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page); 440309381feSSasha Levin VM_BUG_ON_PAGE(PageLRU(page), page); 441c53954a0SMel Gorman __lru_cache_add(page); 4421da177e4SLinus Torvalds } 4431da177e4SLinus Torvalds 444894bc310SLee Schermerhorn /** 44500501b53SJohannes Weiner * lru_cache_add_active_or_unevictable 44600501b53SJohannes Weiner * @page: the page to be added to LRU 44700501b53SJohannes Weiner * @vma: vma in which page is mapped for determining reclaimability 44800501b53SJohannes Weiner * 44900501b53SJohannes Weiner * Place @page on the active or unevictable LRU list, depending on its 45000501b53SJohannes Weiner * evictability. Note that if the page is not evictable, it goes 45100501b53SJohannes Weiner * directly back onto it's zone's unevictable list, it does NOT use a 45200501b53SJohannes Weiner * per cpu pagevec. 45300501b53SJohannes Weiner */ 45400501b53SJohannes Weiner void lru_cache_add_active_or_unevictable(struct page *page, 45500501b53SJohannes Weiner struct vm_area_struct *vma) 45600501b53SJohannes Weiner { 45700501b53SJohannes Weiner VM_BUG_ON_PAGE(PageLRU(page), page); 45800501b53SJohannes Weiner 4599c4e6b1aSShakeel Butt if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED)) 46000501b53SJohannes Weiner SetPageActive(page); 4619c4e6b1aSShakeel Butt else if (!TestSetPageMlocked(page)) { 46200501b53SJohannes Weiner /* 46300501b53SJohannes Weiner * We use the irq-unsafe __mod_zone_page_stat because this 46400501b53SJohannes Weiner * counter is not modified from interrupt context, and the pte 46500501b53SJohannes Weiner * lock is held(spinlock), which implies preemption disabled. 46600501b53SJohannes Weiner */ 46700501b53SJohannes Weiner __mod_zone_page_state(page_zone(page), NR_MLOCK, 46800501b53SJohannes Weiner hpage_nr_pages(page)); 46900501b53SJohannes Weiner count_vm_event(UNEVICTABLE_PGMLOCKED); 47000501b53SJohannes Weiner } 4719c4e6b1aSShakeel Butt lru_cache_add(page); 47200501b53SJohannes Weiner } 47300501b53SJohannes Weiner 474902aaed0SHisashi Hifumi /* 47531560180SMinchan Kim * If the page can not be invalidated, it is moved to the 47631560180SMinchan Kim * inactive list to speed up its reclaim. It is moved to the 47731560180SMinchan Kim * head of the list, rather than the tail, to give the flusher 47831560180SMinchan Kim * threads some time to write it out, as this is much more 47931560180SMinchan Kim * effective than the single-page writeout from reclaim. 480278df9f4SMinchan Kim * 481278df9f4SMinchan Kim * If the page isn't page_mapped and dirty/writeback, the page 482278df9f4SMinchan Kim * could reclaim asap using PG_reclaim. 483278df9f4SMinchan Kim * 484278df9f4SMinchan Kim * 1. active, mapped page -> none 485278df9f4SMinchan Kim * 2. active, dirty/writeback page -> inactive, head, PG_reclaim 486278df9f4SMinchan Kim * 3. inactive, mapped page -> none 487278df9f4SMinchan Kim * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim 488278df9f4SMinchan Kim * 5. inactive, clean -> inactive, tail 489278df9f4SMinchan Kim * 6. Others -> none 490278df9f4SMinchan Kim * 491278df9f4SMinchan Kim * In 4, why it moves inactive's head, the VM expects the page would 492278df9f4SMinchan Kim * be write it out by flusher threads as this is much more effective 493278df9f4SMinchan Kim * than the single-page writeout from reclaim. 49431560180SMinchan Kim */ 495cc5993bdSMinchan Kim static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec, 496fa9add64SHugh Dickins void *arg) 49731560180SMinchan Kim { 49831560180SMinchan Kim int lru, file; 499278df9f4SMinchan Kim bool active; 50031560180SMinchan Kim 501278df9f4SMinchan Kim if (!PageLRU(page)) 50231560180SMinchan Kim return; 50331560180SMinchan Kim 504bad49d9cSMinchan Kim if (PageUnevictable(page)) 505bad49d9cSMinchan Kim return; 506bad49d9cSMinchan Kim 50731560180SMinchan Kim /* Some processes are using the page */ 50831560180SMinchan Kim if (page_mapped(page)) 50931560180SMinchan Kim return; 51031560180SMinchan Kim 511278df9f4SMinchan Kim active = PageActive(page); 51231560180SMinchan Kim file = page_is_file_cache(page); 51331560180SMinchan Kim lru = page_lru_base_type(page); 514fa9add64SHugh Dickins 515fa9add64SHugh Dickins del_page_from_lru_list(page, lruvec, lru + active); 51631560180SMinchan Kim ClearPageActive(page); 51731560180SMinchan Kim ClearPageReferenced(page); 518fa9add64SHugh Dickins add_page_to_lru_list(page, lruvec, lru); 51931560180SMinchan Kim 520278df9f4SMinchan Kim if (PageWriteback(page) || PageDirty(page)) { 521278df9f4SMinchan Kim /* 522278df9f4SMinchan Kim * PG_reclaim could be raced with end_page_writeback 523278df9f4SMinchan Kim * It can make readahead confusing. But race window 524278df9f4SMinchan Kim * is _really_ small and it's non-critical problem. 525278df9f4SMinchan Kim */ 526278df9f4SMinchan Kim SetPageReclaim(page); 527278df9f4SMinchan Kim } else { 528278df9f4SMinchan Kim /* 529278df9f4SMinchan Kim * The page's writeback ends up during pagevec 530278df9f4SMinchan Kim * We moves tha page into tail of inactive. 531278df9f4SMinchan Kim */ 532925b7673SJohannes Weiner list_move_tail(&page->lru, &lruvec->lists[lru]); 533278df9f4SMinchan Kim __count_vm_event(PGROTATED); 534278df9f4SMinchan Kim } 535278df9f4SMinchan Kim 536278df9f4SMinchan Kim if (active) 537278df9f4SMinchan Kim __count_vm_event(PGDEACTIVATE); 538fa9add64SHugh Dickins update_page_reclaim_stat(lruvec, file, 0); 53931560180SMinchan Kim } 54031560180SMinchan Kim 54110853a03SMinchan Kim 542f7ad2a6cSShaohua Li static void lru_lazyfree_fn(struct page *page, struct lruvec *lruvec, 54310853a03SMinchan Kim void *arg) 54410853a03SMinchan Kim { 545f7ad2a6cSShaohua Li if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) && 54624c92eb7SShaohua Li !PageSwapCache(page) && !PageUnevictable(page)) { 547f7ad2a6cSShaohua Li bool active = PageActive(page); 54810853a03SMinchan Kim 549f7ad2a6cSShaohua Li del_page_from_lru_list(page, lruvec, 550f7ad2a6cSShaohua Li LRU_INACTIVE_ANON + active); 55110853a03SMinchan Kim ClearPageActive(page); 55210853a03SMinchan Kim ClearPageReferenced(page); 553f7ad2a6cSShaohua Li /* 554f7ad2a6cSShaohua Li * lazyfree pages are clean anonymous pages. They have 555f7ad2a6cSShaohua Li * SwapBacked flag cleared to distinguish normal anonymous 556f7ad2a6cSShaohua Li * pages 557f7ad2a6cSShaohua Li */ 558f7ad2a6cSShaohua Li ClearPageSwapBacked(page); 559f7ad2a6cSShaohua Li add_page_to_lru_list(page, lruvec, LRU_INACTIVE_FILE); 56010853a03SMinchan Kim 561f7ad2a6cSShaohua Li __count_vm_events(PGLAZYFREE, hpage_nr_pages(page)); 5622262185cSRoman Gushchin count_memcg_page_event(page, PGLAZYFREE); 563f7ad2a6cSShaohua Li update_page_reclaim_stat(lruvec, 1, 0); 56410853a03SMinchan Kim } 56510853a03SMinchan Kim } 56610853a03SMinchan Kim 56731560180SMinchan Kim /* 568902aaed0SHisashi Hifumi * Drain pages out of the cpu's pagevecs. 569902aaed0SHisashi Hifumi * Either "cpu" is the current CPU, and preemption has already been 570902aaed0SHisashi Hifumi * disabled; or "cpu" is being hot-unplugged, and is already dead. 571902aaed0SHisashi Hifumi */ 572f0cb3c76SKonstantin Khlebnikov void lru_add_drain_cpu(int cpu) 5731da177e4SLinus Torvalds { 57413f7f789SMel Gorman struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu); 5751da177e4SLinus Torvalds 5761da177e4SLinus Torvalds if (pagevec_count(pvec)) 577a0b8cab3SMel Gorman __pagevec_lru_add(pvec); 578902aaed0SHisashi Hifumi 579902aaed0SHisashi Hifumi pvec = &per_cpu(lru_rotate_pvecs, cpu); 580902aaed0SHisashi Hifumi if (pagevec_count(pvec)) { 581902aaed0SHisashi Hifumi unsigned long flags; 582902aaed0SHisashi Hifumi 583902aaed0SHisashi Hifumi /* No harm done if a racing interrupt already did this */ 584902aaed0SHisashi Hifumi local_irq_save(flags); 585902aaed0SHisashi Hifumi pagevec_move_tail(pvec); 586902aaed0SHisashi Hifumi local_irq_restore(flags); 587902aaed0SHisashi Hifumi } 58831560180SMinchan Kim 589cc5993bdSMinchan Kim pvec = &per_cpu(lru_deactivate_file_pvecs, cpu); 59031560180SMinchan Kim if (pagevec_count(pvec)) 591cc5993bdSMinchan Kim pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL); 592eb709b0dSShaohua Li 593f7ad2a6cSShaohua Li pvec = &per_cpu(lru_lazyfree_pvecs, cpu); 59410853a03SMinchan Kim if (pagevec_count(pvec)) 595f7ad2a6cSShaohua Li pagevec_lru_move_fn(pvec, lru_lazyfree_fn, NULL); 59610853a03SMinchan Kim 597eb709b0dSShaohua Li activate_page_drain(cpu); 59831560180SMinchan Kim } 59931560180SMinchan Kim 60031560180SMinchan Kim /** 601cc5993bdSMinchan Kim * deactivate_file_page - forcefully deactivate a file page 60231560180SMinchan Kim * @page: page to deactivate 60331560180SMinchan Kim * 60431560180SMinchan Kim * This function hints the VM that @page is a good reclaim candidate, 60531560180SMinchan Kim * for example if its invalidation fails due to the page being dirty 60631560180SMinchan Kim * or under writeback. 60731560180SMinchan Kim */ 608cc5993bdSMinchan Kim void deactivate_file_page(struct page *page) 60931560180SMinchan Kim { 610821ed6bbSMinchan Kim /* 611cc5993bdSMinchan Kim * In a workload with many unevictable page such as mprotect, 612cc5993bdSMinchan Kim * unevictable page deactivation for accelerating reclaim is pointless. 613821ed6bbSMinchan Kim */ 614821ed6bbSMinchan Kim if (PageUnevictable(page)) 615821ed6bbSMinchan Kim return; 616821ed6bbSMinchan Kim 61731560180SMinchan Kim if (likely(get_page_unless_zero(page))) { 618cc5993bdSMinchan Kim struct pagevec *pvec = &get_cpu_var(lru_deactivate_file_pvecs); 61931560180SMinchan Kim 6208f182270SLukasz Odzioba if (!pagevec_add(pvec, page) || PageCompound(page)) 621cc5993bdSMinchan Kim pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL); 622cc5993bdSMinchan Kim put_cpu_var(lru_deactivate_file_pvecs); 62331560180SMinchan Kim } 62480bfed90SAndrew Morton } 62580bfed90SAndrew Morton 62610853a03SMinchan Kim /** 627f7ad2a6cSShaohua Li * mark_page_lazyfree - make an anon page lazyfree 62810853a03SMinchan Kim * @page: page to deactivate 62910853a03SMinchan Kim * 630f7ad2a6cSShaohua Li * mark_page_lazyfree() moves @page to the inactive file list. 631f7ad2a6cSShaohua Li * This is done to accelerate the reclaim of @page. 63210853a03SMinchan Kim */ 633f7ad2a6cSShaohua Li void mark_page_lazyfree(struct page *page) 63410853a03SMinchan Kim { 635f7ad2a6cSShaohua Li if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) && 63624c92eb7SShaohua Li !PageSwapCache(page) && !PageUnevictable(page)) { 637f7ad2a6cSShaohua Li struct pagevec *pvec = &get_cpu_var(lru_lazyfree_pvecs); 63810853a03SMinchan Kim 63909cbfeafSKirill A. Shutemov get_page(page); 6408f182270SLukasz Odzioba if (!pagevec_add(pvec, page) || PageCompound(page)) 641f7ad2a6cSShaohua Li pagevec_lru_move_fn(pvec, lru_lazyfree_fn, NULL); 642f7ad2a6cSShaohua Li put_cpu_var(lru_lazyfree_pvecs); 64310853a03SMinchan Kim } 64410853a03SMinchan Kim } 64510853a03SMinchan Kim 64680bfed90SAndrew Morton void lru_add_drain(void) 64780bfed90SAndrew Morton { 648f0cb3c76SKonstantin Khlebnikov lru_add_drain_cpu(get_cpu()); 64980bfed90SAndrew Morton put_cpu(); 6501da177e4SLinus Torvalds } 6511da177e4SLinus Torvalds 6526ea183d6SMichal Hocko #ifdef CONFIG_SMP 6536ea183d6SMichal Hocko 6546ea183d6SMichal Hocko static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work); 6556ea183d6SMichal Hocko 656c4028958SDavid Howells static void lru_add_drain_per_cpu(struct work_struct *dummy) 657053837fcSNick Piggin { 658053837fcSNick Piggin lru_add_drain(); 659053837fcSNick Piggin } 660053837fcSNick Piggin 6619852a721SMichal Hocko /* 6629852a721SMichal Hocko * Doesn't need any cpu hotplug locking because we do rely on per-cpu 6639852a721SMichal Hocko * kworkers being shut down before our page_alloc_cpu_dead callback is 6649852a721SMichal Hocko * executed on the offlined cpu. 6659852a721SMichal Hocko * Calling this function with cpu hotplug locks held can actually lead 6669852a721SMichal Hocko * to obscure indirect dependencies via WQ context. 6679852a721SMichal Hocko */ 6689852a721SMichal Hocko void lru_add_drain_all(void) 669053837fcSNick Piggin { 6705fbc4616SChris Metcalf static DEFINE_MUTEX(lock); 6715fbc4616SChris Metcalf static struct cpumask has_work; 6725fbc4616SChris Metcalf int cpu; 6735fbc4616SChris Metcalf 674ce612879SMichal Hocko /* 675ce612879SMichal Hocko * Make sure nobody triggers this path before mm_percpu_wq is fully 676ce612879SMichal Hocko * initialized. 677ce612879SMichal Hocko */ 678ce612879SMichal Hocko if (WARN_ON(!mm_percpu_wq)) 679ce612879SMichal Hocko return; 680ce612879SMichal Hocko 6815fbc4616SChris Metcalf mutex_lock(&lock); 6825fbc4616SChris Metcalf cpumask_clear(&has_work); 6835fbc4616SChris Metcalf 6845fbc4616SChris Metcalf for_each_online_cpu(cpu) { 6855fbc4616SChris Metcalf struct work_struct *work = &per_cpu(lru_add_drain_work, cpu); 6865fbc4616SChris Metcalf 6875fbc4616SChris Metcalf if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) || 6885fbc4616SChris Metcalf pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) || 689cc5993bdSMinchan Kim pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) || 690f7ad2a6cSShaohua Li pagevec_count(&per_cpu(lru_lazyfree_pvecs, cpu)) || 6915fbc4616SChris Metcalf need_activate_page_drain(cpu)) { 6925fbc4616SChris Metcalf INIT_WORK(work, lru_add_drain_per_cpu); 693ce612879SMichal Hocko queue_work_on(cpu, mm_percpu_wq, work); 6945fbc4616SChris Metcalf cpumask_set_cpu(cpu, &has_work); 6955fbc4616SChris Metcalf } 6965fbc4616SChris Metcalf } 6975fbc4616SChris Metcalf 6985fbc4616SChris Metcalf for_each_cpu(cpu, &has_work) 6995fbc4616SChris Metcalf flush_work(&per_cpu(lru_add_drain_work, cpu)); 7005fbc4616SChris Metcalf 7015fbc4616SChris Metcalf mutex_unlock(&lock); 702053837fcSNick Piggin } 7036ea183d6SMichal Hocko #else 7046ea183d6SMichal Hocko void lru_add_drain_all(void) 7056ea183d6SMichal Hocko { 7066ea183d6SMichal Hocko lru_add_drain(); 7076ea183d6SMichal Hocko } 7086ea183d6SMichal Hocko #endif 709053837fcSNick Piggin 710aabfb572SMichal Hocko /** 711ea1754a0SKirill A. Shutemov * release_pages - batched put_page() 712aabfb572SMichal Hocko * @pages: array of pages to release 713aabfb572SMichal Hocko * @nr: number of pages 7141da177e4SLinus Torvalds * 715aabfb572SMichal Hocko * Decrement the reference count on all the pages in @pages. If it 716aabfb572SMichal Hocko * fell to zero, remove the page from the LRU and free it. 7171da177e4SLinus Torvalds */ 718c6f92f9fSMel Gorman void release_pages(struct page **pages, int nr) 7191da177e4SLinus Torvalds { 7201da177e4SLinus Torvalds int i; 721cc59850eSKonstantin Khlebnikov LIST_HEAD(pages_to_free); 722599d0c95SMel Gorman struct pglist_data *locked_pgdat = NULL; 723fa9add64SHugh Dickins struct lruvec *lruvec; 724902aaed0SHisashi Hifumi unsigned long uninitialized_var(flags); 725aabfb572SMichal Hocko unsigned int uninitialized_var(lock_batch); 7261da177e4SLinus Torvalds 7271da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 7281da177e4SLinus Torvalds struct page *page = pages[i]; 7291da177e4SLinus Torvalds 730aabfb572SMichal Hocko /* 731aabfb572SMichal Hocko * Make sure the IRQ-safe lock-holding time does not get 732aabfb572SMichal Hocko * excessive with a continuous string of pages from the 733599d0c95SMel Gorman * same pgdat. The lock is held only if pgdat != NULL. 734aabfb572SMichal Hocko */ 735599d0c95SMel Gorman if (locked_pgdat && ++lock_batch == SWAP_CLUSTER_MAX) { 736599d0c95SMel Gorman spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags); 737599d0c95SMel Gorman locked_pgdat = NULL; 738aabfb572SMichal Hocko } 739aabfb572SMichal Hocko 7406fcb52a5SAaron Lu if (is_huge_zero_page(page)) 741aa88b68cSKirill A. Shutemov continue; 742aa88b68cSKirill A. Shutemov 743c5d6c45eSIra Weiny if (is_zone_device_page(page)) { 744df6ad698SJérôme Glisse if (locked_pgdat) { 745df6ad698SJérôme Glisse spin_unlock_irqrestore(&locked_pgdat->lru_lock, 746df6ad698SJérôme Glisse flags); 747df6ad698SJérôme Glisse locked_pgdat = NULL; 748df6ad698SJérôme Glisse } 749c5d6c45eSIra Weiny /* 750c5d6c45eSIra Weiny * ZONE_DEVICE pages that return 'false' from 751c5d6c45eSIra Weiny * put_devmap_managed_page() do not require special 752c5d6c45eSIra Weiny * processing, and instead, expect a call to 753c5d6c45eSIra Weiny * put_page_testzero(). 754c5d6c45eSIra Weiny */ 755c5d6c45eSIra Weiny if (put_devmap_managed_page(page)) 756df6ad698SJérôme Glisse continue; 757df6ad698SJérôme Glisse } 758df6ad698SJérôme Glisse 759ddc58f27SKirill A. Shutemov page = compound_head(page); 760b5810039SNick Piggin if (!put_page_testzero(page)) 7611da177e4SLinus Torvalds continue; 7621da177e4SLinus Torvalds 763ddc58f27SKirill A. Shutemov if (PageCompound(page)) { 764599d0c95SMel Gorman if (locked_pgdat) { 765599d0c95SMel Gorman spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags); 766599d0c95SMel Gorman locked_pgdat = NULL; 767ddc58f27SKirill A. Shutemov } 768ddc58f27SKirill A. Shutemov __put_compound_page(page); 769ddc58f27SKirill A. Shutemov continue; 770ddc58f27SKirill A. Shutemov } 771ddc58f27SKirill A. Shutemov 77246453a6eSNick Piggin if (PageLRU(page)) { 773599d0c95SMel Gorman struct pglist_data *pgdat = page_pgdat(page); 774894bc310SLee Schermerhorn 775599d0c95SMel Gorman if (pgdat != locked_pgdat) { 776599d0c95SMel Gorman if (locked_pgdat) 777599d0c95SMel Gorman spin_unlock_irqrestore(&locked_pgdat->lru_lock, 778902aaed0SHisashi Hifumi flags); 779aabfb572SMichal Hocko lock_batch = 0; 780599d0c95SMel Gorman locked_pgdat = pgdat; 781599d0c95SMel Gorman spin_lock_irqsave(&locked_pgdat->lru_lock, flags); 7821da177e4SLinus Torvalds } 783fa9add64SHugh Dickins 784599d0c95SMel Gorman lruvec = mem_cgroup_page_lruvec(page, locked_pgdat); 785309381feSSasha Levin VM_BUG_ON_PAGE(!PageLRU(page), page); 78667453911SNick Piggin __ClearPageLRU(page); 787fa9add64SHugh Dickins del_page_from_lru_list(page, lruvec, page_off_lru(page)); 78846453a6eSNick Piggin } 78946453a6eSNick Piggin 790c53954a0SMel Gorman /* Clear Active bit in case of parallel mark_page_accessed */ 791e3741b50SMel Gorman __ClearPageActive(page); 79262906027SNicholas Piggin __ClearPageWaiters(page); 793c53954a0SMel Gorman 794cc59850eSKonstantin Khlebnikov list_add(&page->lru, &pages_to_free); 7951da177e4SLinus Torvalds } 796599d0c95SMel Gorman if (locked_pgdat) 797599d0c95SMel Gorman spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags); 7981da177e4SLinus Torvalds 799747db954SJohannes Weiner mem_cgroup_uncharge_list(&pages_to_free); 8002d4894b5SMel Gorman free_unref_page_list(&pages_to_free); 8011da177e4SLinus Torvalds } 8020be8557bSMiklos Szeredi EXPORT_SYMBOL(release_pages); 8031da177e4SLinus Torvalds 8041da177e4SLinus Torvalds /* 8051da177e4SLinus Torvalds * The pages which we're about to release may be in the deferred lru-addition 8061da177e4SLinus Torvalds * queues. That would prevent them from really being freed right now. That's 8071da177e4SLinus Torvalds * OK from a correctness point of view but is inefficient - those pages may be 8081da177e4SLinus Torvalds * cache-warm and we want to give them back to the page allocator ASAP. 8091da177e4SLinus Torvalds * 8101da177e4SLinus Torvalds * So __pagevec_release() will drain those queues here. __pagevec_lru_add() 8111da177e4SLinus Torvalds * and __pagevec_lru_add_active() call release_pages() directly to avoid 8121da177e4SLinus Torvalds * mutual recursion. 8131da177e4SLinus Torvalds */ 8141da177e4SLinus Torvalds void __pagevec_release(struct pagevec *pvec) 8151da177e4SLinus Torvalds { 8167f0b5fb9SMel Gorman if (!pvec->percpu_pvec_drained) { 8171da177e4SLinus Torvalds lru_add_drain(); 8187f0b5fb9SMel Gorman pvec->percpu_pvec_drained = true; 819d9ed0d08SMel Gorman } 820c6f92f9fSMel Gorman release_pages(pvec->pages, pagevec_count(pvec)); 8211da177e4SLinus Torvalds pagevec_reinit(pvec); 8221da177e4SLinus Torvalds } 8237f285701SSteve French EXPORT_SYMBOL(__pagevec_release); 8247f285701SSteve French 82512d27107SHugh Dickins #ifdef CONFIG_TRANSPARENT_HUGEPAGE 82671e3aac0SAndrea Arcangeli /* used by __split_huge_page_refcount() */ 827fa9add64SHugh Dickins void lru_add_page_tail(struct page *page, struct page *page_tail, 8285bc7b8acSShaohua Li struct lruvec *lruvec, struct list_head *list) 82971e3aac0SAndrea Arcangeli { 83071e3aac0SAndrea Arcangeli const int file = 0; 83171e3aac0SAndrea Arcangeli 832309381feSSasha Levin VM_BUG_ON_PAGE(!PageHead(page), page); 833309381feSSasha Levin VM_BUG_ON_PAGE(PageCompound(page_tail), page); 834309381feSSasha Levin VM_BUG_ON_PAGE(PageLRU(page_tail), page); 83535f3aa39SLance Roy lockdep_assert_held(&lruvec_pgdat(lruvec)->lru_lock); 83671e3aac0SAndrea Arcangeli 8375bc7b8acSShaohua Li if (!list) 83871e3aac0SAndrea Arcangeli SetPageLRU(page_tail); 83971e3aac0SAndrea Arcangeli 84012d27107SHugh Dickins if (likely(PageLRU(page))) 84112d27107SHugh Dickins list_add_tail(&page_tail->lru, &page->lru); 8425bc7b8acSShaohua Li else if (list) { 8435bc7b8acSShaohua Li /* page reclaim is reclaiming a huge page */ 8445bc7b8acSShaohua Li get_page(page_tail); 8455bc7b8acSShaohua Li list_add_tail(&page_tail->lru, list); 8465bc7b8acSShaohua Li } else { 84712d27107SHugh Dickins struct list_head *list_head; 84812d27107SHugh Dickins /* 84912d27107SHugh Dickins * Head page has not yet been counted, as an hpage, 85012d27107SHugh Dickins * so we must account for each subpage individually. 85112d27107SHugh Dickins * 85212d27107SHugh Dickins * Use the standard add function to put page_tail on the list, 85312d27107SHugh Dickins * but then correct its position so they all end up in order. 85412d27107SHugh Dickins */ 855e180cf80SKirill A. Shutemov add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail)); 85612d27107SHugh Dickins list_head = page_tail->lru.prev; 85712d27107SHugh Dickins list_move_tail(&page_tail->lru, list_head); 85871e3aac0SAndrea Arcangeli } 8597512102cSHugh Dickins 8607512102cSHugh Dickins if (!PageUnevictable(page)) 861e180cf80SKirill A. Shutemov update_page_reclaim_stat(lruvec, file, PageActive(page_tail)); 86271e3aac0SAndrea Arcangeli } 86312d27107SHugh Dickins #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 86471e3aac0SAndrea Arcangeli 865fa9add64SHugh Dickins static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec, 866fa9add64SHugh Dickins void *arg) 8673dd7ae8eSShaohua Li { 8689c4e6b1aSShakeel Butt enum lru_list lru; 8699c4e6b1aSShakeel Butt int was_unevictable = TestClearPageUnevictable(page); 8703dd7ae8eSShaohua Li 871309381feSSasha Levin VM_BUG_ON_PAGE(PageLRU(page), page); 8723dd7ae8eSShaohua Li 8733dd7ae8eSShaohua Li SetPageLRU(page); 8749c4e6b1aSShakeel Butt /* 8759c4e6b1aSShakeel Butt * Page becomes evictable in two ways: 876dae966dcSPeng Fan * 1) Within LRU lock [munlock_vma_page() and __munlock_pagevec()]. 8779c4e6b1aSShakeel Butt * 2) Before acquiring LRU lock to put the page to correct LRU and then 8789c4e6b1aSShakeel Butt * a) do PageLRU check with lock [check_move_unevictable_pages] 8799c4e6b1aSShakeel Butt * b) do PageLRU check before lock [clear_page_mlock] 8809c4e6b1aSShakeel Butt * 8819c4e6b1aSShakeel Butt * (1) & (2a) are ok as LRU lock will serialize them. For (2b), we need 8829c4e6b1aSShakeel Butt * following strict ordering: 8839c4e6b1aSShakeel Butt * 8849c4e6b1aSShakeel Butt * #0: __pagevec_lru_add_fn #1: clear_page_mlock 8859c4e6b1aSShakeel Butt * 8869c4e6b1aSShakeel Butt * SetPageLRU() TestClearPageMlocked() 8879c4e6b1aSShakeel Butt * smp_mb() // explicit ordering // above provides strict 8889c4e6b1aSShakeel Butt * // ordering 8899c4e6b1aSShakeel Butt * PageMlocked() PageLRU() 8909c4e6b1aSShakeel Butt * 8919c4e6b1aSShakeel Butt * 8929c4e6b1aSShakeel Butt * if '#1' does not observe setting of PG_lru by '#0' and fails 8939c4e6b1aSShakeel Butt * isolation, the explicit barrier will make sure that page_evictable 8949c4e6b1aSShakeel Butt * check will put the page in correct LRU. Without smp_mb(), SetPageLRU 8959c4e6b1aSShakeel Butt * can be reordered after PageMlocked check and can make '#1' to fail 8969c4e6b1aSShakeel Butt * the isolation of the page whose Mlocked bit is cleared (#0 is also 8979c4e6b1aSShakeel Butt * looking at the same page) and the evictable page will be stranded 8989c4e6b1aSShakeel Butt * in an unevictable LRU. 8999c4e6b1aSShakeel Butt */ 9009c4e6b1aSShakeel Butt smp_mb(); 9019c4e6b1aSShakeel Butt 9029c4e6b1aSShakeel Butt if (page_evictable(page)) { 9039c4e6b1aSShakeel Butt lru = page_lru(page); 9049c4e6b1aSShakeel Butt update_page_reclaim_stat(lruvec, page_is_file_cache(page), 9059c4e6b1aSShakeel Butt PageActive(page)); 9069c4e6b1aSShakeel Butt if (was_unevictable) 9079c4e6b1aSShakeel Butt count_vm_event(UNEVICTABLE_PGRESCUED); 9089c4e6b1aSShakeel Butt } else { 9099c4e6b1aSShakeel Butt lru = LRU_UNEVICTABLE; 9109c4e6b1aSShakeel Butt ClearPageActive(page); 9119c4e6b1aSShakeel Butt SetPageUnevictable(page); 9129c4e6b1aSShakeel Butt if (!was_unevictable) 9139c4e6b1aSShakeel Butt count_vm_event(UNEVICTABLE_PGCULLED); 9149c4e6b1aSShakeel Butt } 9159c4e6b1aSShakeel Butt 916fa9add64SHugh Dickins add_page_to_lru_list(page, lruvec, lru); 91724b7e581SMel Gorman trace_mm_lru_insertion(page, lru); 9183dd7ae8eSShaohua Li } 9193dd7ae8eSShaohua Li 9201da177e4SLinus Torvalds /* 9211da177e4SLinus Torvalds * Add the passed pages to the LRU, then drop the caller's refcount 9221da177e4SLinus Torvalds * on them. Reinitialises the caller's pagevec. 9231da177e4SLinus Torvalds */ 924a0b8cab3SMel Gorman void __pagevec_lru_add(struct pagevec *pvec) 9251da177e4SLinus Torvalds { 926a0b8cab3SMel Gorman pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL); 9271da177e4SLinus Torvalds } 9285095ae83SHugh Dickins EXPORT_SYMBOL(__pagevec_lru_add); 929f04e9ebbSKOSAKI Motohiro 9301da177e4SLinus Torvalds /** 9310cd6144aSJohannes Weiner * pagevec_lookup_entries - gang pagecache lookup 9320cd6144aSJohannes Weiner * @pvec: Where the resulting entries are placed 9330cd6144aSJohannes Weiner * @mapping: The address_space to search 9340cd6144aSJohannes Weiner * @start: The starting entry index 935cb6f0f34SMike Rapoport * @nr_entries: The maximum number of pages 9360cd6144aSJohannes Weiner * @indices: The cache indices corresponding to the entries in @pvec 9370cd6144aSJohannes Weiner * 9380cd6144aSJohannes Weiner * pagevec_lookup_entries() will search for and return a group of up 939f144c390SMike Rapoport * to @nr_pages pages and shadow entries in the mapping. All 9400cd6144aSJohannes Weiner * entries are placed in @pvec. pagevec_lookup_entries() takes a 9410cd6144aSJohannes Weiner * reference against actual pages in @pvec. 9420cd6144aSJohannes Weiner * 9430cd6144aSJohannes Weiner * The search returns a group of mapping-contiguous entries with 9440cd6144aSJohannes Weiner * ascending indexes. There may be holes in the indices due to 9450cd6144aSJohannes Weiner * not-present entries. 9460cd6144aSJohannes Weiner * 9470cd6144aSJohannes Weiner * pagevec_lookup_entries() returns the number of entries which were 9480cd6144aSJohannes Weiner * found. 9490cd6144aSJohannes Weiner */ 9500cd6144aSJohannes Weiner unsigned pagevec_lookup_entries(struct pagevec *pvec, 9510cd6144aSJohannes Weiner struct address_space *mapping, 952e02a9f04SRandy Dunlap pgoff_t start, unsigned nr_entries, 9530cd6144aSJohannes Weiner pgoff_t *indices) 9540cd6144aSJohannes Weiner { 955e02a9f04SRandy Dunlap pvec->nr = find_get_entries(mapping, start, nr_entries, 9560cd6144aSJohannes Weiner pvec->pages, indices); 9570cd6144aSJohannes Weiner return pagevec_count(pvec); 9580cd6144aSJohannes Weiner } 9590cd6144aSJohannes Weiner 9600cd6144aSJohannes Weiner /** 9610cd6144aSJohannes Weiner * pagevec_remove_exceptionals - pagevec exceptionals pruning 9620cd6144aSJohannes Weiner * @pvec: The pagevec to prune 9630cd6144aSJohannes Weiner * 9640cd6144aSJohannes Weiner * pagevec_lookup_entries() fills both pages and exceptional radix 9650cd6144aSJohannes Weiner * tree entries into the pagevec. This function prunes all 9660cd6144aSJohannes Weiner * exceptionals from @pvec without leaving holes, so that it can be 9670cd6144aSJohannes Weiner * passed on to page-only pagevec operations. 9680cd6144aSJohannes Weiner */ 9690cd6144aSJohannes Weiner void pagevec_remove_exceptionals(struct pagevec *pvec) 9700cd6144aSJohannes Weiner { 9710cd6144aSJohannes Weiner int i, j; 9720cd6144aSJohannes Weiner 9730cd6144aSJohannes Weiner for (i = 0, j = 0; i < pagevec_count(pvec); i++) { 9740cd6144aSJohannes Weiner struct page *page = pvec->pages[i]; 9753159f943SMatthew Wilcox if (!xa_is_value(page)) 9760cd6144aSJohannes Weiner pvec->pages[j++] = page; 9770cd6144aSJohannes Weiner } 9780cd6144aSJohannes Weiner pvec->nr = j; 9790cd6144aSJohannes Weiner } 9800cd6144aSJohannes Weiner 9810cd6144aSJohannes Weiner /** 982b947cee4SJan Kara * pagevec_lookup_range - gang pagecache lookup 9831da177e4SLinus Torvalds * @pvec: Where the resulting pages are placed 9841da177e4SLinus Torvalds * @mapping: The address_space to search 9851da177e4SLinus Torvalds * @start: The starting page index 986b947cee4SJan Kara * @end: The final page index 9871da177e4SLinus Torvalds * 988e02a9f04SRandy Dunlap * pagevec_lookup_range() will search for & return a group of up to PAGEVEC_SIZE 989b947cee4SJan Kara * pages in the mapping starting from index @start and upto index @end 990b947cee4SJan Kara * (inclusive). The pages are placed in @pvec. pagevec_lookup() takes a 9911da177e4SLinus Torvalds * reference against the pages in @pvec. 9921da177e4SLinus Torvalds * 9931da177e4SLinus Torvalds * The search returns a group of mapping-contiguous pages with ascending 994d72dc8a2SJan Kara * indexes. There may be holes in the indices due to not-present pages. We 995d72dc8a2SJan Kara * also update @start to index the next page for the traversal. 9961da177e4SLinus Torvalds * 997b947cee4SJan Kara * pagevec_lookup_range() returns the number of pages which were found. If this 998e02a9f04SRandy Dunlap * number is smaller than PAGEVEC_SIZE, the end of specified range has been 999b947cee4SJan Kara * reached. 10001da177e4SLinus Torvalds */ 1001b947cee4SJan Kara unsigned pagevec_lookup_range(struct pagevec *pvec, 1002397162ffSJan Kara struct address_space *mapping, pgoff_t *start, pgoff_t end) 10031da177e4SLinus Torvalds { 1004397162ffSJan Kara pvec->nr = find_get_pages_range(mapping, start, end, PAGEVEC_SIZE, 1005b947cee4SJan Kara pvec->pages); 10061da177e4SLinus Torvalds return pagevec_count(pvec); 10071da177e4SLinus Torvalds } 1008b947cee4SJan Kara EXPORT_SYMBOL(pagevec_lookup_range); 100978539fdfSChristoph Hellwig 101072b045aeSJan Kara unsigned pagevec_lookup_range_tag(struct pagevec *pvec, 101172b045aeSJan Kara struct address_space *mapping, pgoff_t *index, pgoff_t end, 101210bbd235SMatthew Wilcox xa_mark_t tag) 10131da177e4SLinus Torvalds { 101472b045aeSJan Kara pvec->nr = find_get_pages_range_tag(mapping, index, end, tag, 101567fd707fSJan Kara PAGEVEC_SIZE, pvec->pages); 10161da177e4SLinus Torvalds return pagevec_count(pvec); 10171da177e4SLinus Torvalds } 101872b045aeSJan Kara EXPORT_SYMBOL(pagevec_lookup_range_tag); 10191da177e4SLinus Torvalds 102093d3b714SJan Kara unsigned pagevec_lookup_range_nr_tag(struct pagevec *pvec, 102193d3b714SJan Kara struct address_space *mapping, pgoff_t *index, pgoff_t end, 102210bbd235SMatthew Wilcox xa_mark_t tag, unsigned max_pages) 102393d3b714SJan Kara { 102493d3b714SJan Kara pvec->nr = find_get_pages_range_tag(mapping, index, end, tag, 102593d3b714SJan Kara min_t(unsigned int, max_pages, PAGEVEC_SIZE), pvec->pages); 102693d3b714SJan Kara return pagevec_count(pvec); 102793d3b714SJan Kara } 102893d3b714SJan Kara EXPORT_SYMBOL(pagevec_lookup_range_nr_tag); 10291da177e4SLinus Torvalds /* 10301da177e4SLinus Torvalds * Perform any setup for the swap system 10311da177e4SLinus Torvalds */ 10321da177e4SLinus Torvalds void __init swap_setup(void) 10331da177e4SLinus Torvalds { 1034ca79b0c2SArun KS unsigned long megs = totalram_pages() >> (20 - PAGE_SHIFT); 1035e0bf68ddSPeter Zijlstra 10361da177e4SLinus Torvalds /* Use a smaller cluster for small-memory machines */ 10371da177e4SLinus Torvalds if (megs < 16) 10381da177e4SLinus Torvalds page_cluster = 2; 10391da177e4SLinus Torvalds else 10401da177e4SLinus Torvalds page_cluster = 3; 10411da177e4SLinus Torvalds /* 10421da177e4SLinus Torvalds * Right now other parts of the system means that we 10431da177e4SLinus Torvalds * _really_ don't want to cluster much more 10441da177e4SLinus Torvalds */ 10451da177e4SLinus Torvalds } 1046