11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/mm/swap.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds 51da177e4SLinus Torvalds */ 61da177e4SLinus Torvalds 71da177e4SLinus Torvalds /* 8183ff22bSSimon Arlott * This file contains the default values for the operation of the 91da177e4SLinus Torvalds * Linux VM subsystem. Fine-tuning documentation can be found in 101da177e4SLinus Torvalds * Documentation/sysctl/vm.txt. 111da177e4SLinus Torvalds * Started 18.12.91 121da177e4SLinus Torvalds * Swap aging added 23.2.95, Stephen Tweedie. 131da177e4SLinus Torvalds * Buffermem limits added 12.3.98, Rik van Riel. 141da177e4SLinus Torvalds */ 151da177e4SLinus Torvalds 161da177e4SLinus Torvalds #include <linux/mm.h> 171da177e4SLinus Torvalds #include <linux/sched.h> 181da177e4SLinus Torvalds #include <linux/kernel_stat.h> 191da177e4SLinus Torvalds #include <linux/swap.h> 201da177e4SLinus Torvalds #include <linux/mman.h> 211da177e4SLinus Torvalds #include <linux/pagemap.h> 221da177e4SLinus Torvalds #include <linux/pagevec.h> 231da177e4SLinus Torvalds #include <linux/init.h> 24b95f1b31SPaul Gortmaker #include <linux/export.h> 251da177e4SLinus Torvalds #include <linux/mm_inline.h> 261da177e4SLinus Torvalds #include <linux/percpu_counter.h> 271da177e4SLinus Torvalds #include <linux/percpu.h> 281da177e4SLinus Torvalds #include <linux/cpu.h> 291da177e4SLinus Torvalds #include <linux/notifier.h> 30e0bf68ddSPeter Zijlstra #include <linux/backing-dev.h> 3166e1707bSBalbir Singh #include <linux/memcontrol.h> 325a0e3ad6STejun Heo #include <linux/gfp.h> 33a27bb332SKent Overstreet #include <linux/uio.h> 341da177e4SLinus Torvalds 3564d6519dSLee Schermerhorn #include "internal.h" 3664d6519dSLee Schermerhorn 37c6286c98SMel Gorman #define CREATE_TRACE_POINTS 38c6286c98SMel Gorman #include <trace/events/pagemap.h> 39c6286c98SMel Gorman 401da177e4SLinus Torvalds /* How many pages do we try to swap or page in/out together? */ 411da177e4SLinus Torvalds int page_cluster; 421da177e4SLinus Torvalds 4313f7f789SMel Gorman static DEFINE_PER_CPU(struct pagevec, lru_add_pvec); 44f84f9504SVegard Nossum static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs); 4531560180SMinchan Kim static DEFINE_PER_CPU(struct pagevec, lru_deactivate_pvecs); 46902aaed0SHisashi Hifumi 47b221385bSAdrian Bunk /* 48b221385bSAdrian Bunk * This path almost never happens for VM activity - pages are normally 49b221385bSAdrian Bunk * freed via pagevecs. But it gets used by networking. 50b221385bSAdrian Bunk */ 51920c7a5dSHarvey Harrison static void __page_cache_release(struct page *page) 52b221385bSAdrian Bunk { 53b221385bSAdrian Bunk if (PageLRU(page)) { 54b221385bSAdrian Bunk struct zone *zone = page_zone(page); 55fa9add64SHugh Dickins struct lruvec *lruvec; 56fa9add64SHugh Dickins unsigned long flags; 57b221385bSAdrian Bunk 58b221385bSAdrian Bunk spin_lock_irqsave(&zone->lru_lock, flags); 59fa9add64SHugh Dickins lruvec = mem_cgroup_page_lruvec(page, zone); 60b221385bSAdrian Bunk VM_BUG_ON(!PageLRU(page)); 61b221385bSAdrian Bunk __ClearPageLRU(page); 62fa9add64SHugh Dickins del_page_from_lru_list(page, lruvec, page_off_lru(page)); 63b221385bSAdrian Bunk spin_unlock_irqrestore(&zone->lru_lock, flags); 64b221385bSAdrian Bunk } 6591807063SAndrea Arcangeli } 6691807063SAndrea Arcangeli 6791807063SAndrea Arcangeli static void __put_single_page(struct page *page) 6891807063SAndrea Arcangeli { 6991807063SAndrea Arcangeli __page_cache_release(page); 70fc91668eSLi Hong free_hot_cold_page(page, 0); 71b221385bSAdrian Bunk } 72b221385bSAdrian Bunk 7391807063SAndrea Arcangeli static void __put_compound_page(struct page *page) 7491807063SAndrea Arcangeli { 7591807063SAndrea Arcangeli compound_page_dtor *dtor; 7691807063SAndrea Arcangeli 7791807063SAndrea Arcangeli __page_cache_release(page); 7891807063SAndrea Arcangeli dtor = get_compound_page_dtor(page); 7991807063SAndrea Arcangeli (*dtor)(page); 8091807063SAndrea Arcangeli } 8191807063SAndrea Arcangeli 828519fb30SNick Piggin static void put_compound_page(struct page *page) 831da177e4SLinus Torvalds { 8491807063SAndrea Arcangeli if (unlikely(PageTail(page))) { 8591807063SAndrea Arcangeli /* __split_huge_page_refcount can run under us */ 8670b50f94SAndrea Arcangeli struct page *page_head = compound_trans_head(page); 8770b50f94SAndrea Arcangeli 8870b50f94SAndrea Arcangeli if (likely(page != page_head && 8970b50f94SAndrea Arcangeli get_page_unless_zero(page_head))) { 9091807063SAndrea Arcangeli unsigned long flags; 915bf5f03cSPravin B Shelar 925bf5f03cSPravin B Shelar /* 935bf5f03cSPravin B Shelar * THP can not break up slab pages so avoid taking 945bf5f03cSPravin B Shelar * compound_lock(). Slab performs non-atomic bit ops 955bf5f03cSPravin B Shelar * on page->flags for better performance. In particular 965bf5f03cSPravin B Shelar * slab_unlock() in slub used to be a hot path. It is 975bf5f03cSPravin B Shelar * still hot on arches that do not support 985bf5f03cSPravin B Shelar * this_cpu_cmpxchg_double(). 995bf5f03cSPravin B Shelar */ 1005bf5f03cSPravin B Shelar if (PageSlab(page_head)) { 1015bf5f03cSPravin B Shelar if (PageTail(page)) { 1025bf5f03cSPravin B Shelar if (put_page_testzero(page_head)) 1035bf5f03cSPravin B Shelar VM_BUG_ON(1); 1045bf5f03cSPravin B Shelar 1055bf5f03cSPravin B Shelar atomic_dec(&page->_mapcount); 1065bf5f03cSPravin B Shelar goto skip_lock_tail; 1075bf5f03cSPravin B Shelar } else 1085bf5f03cSPravin B Shelar goto skip_lock; 1095bf5f03cSPravin B Shelar } 11091807063SAndrea Arcangeli /* 11170b50f94SAndrea Arcangeli * page_head wasn't a dangling pointer but it 11270b50f94SAndrea Arcangeli * may not be a head page anymore by the time 11370b50f94SAndrea Arcangeli * we obtain the lock. That is ok as long as it 11470b50f94SAndrea Arcangeli * can't be freed from under us. 11591807063SAndrea Arcangeli */ 11691807063SAndrea Arcangeli flags = compound_lock_irqsave(page_head); 11791807063SAndrea Arcangeli if (unlikely(!PageTail(page))) { 11891807063SAndrea Arcangeli /* __split_huge_page_refcount run before us */ 11991807063SAndrea Arcangeli compound_unlock_irqrestore(page_head, flags); 1205bf5f03cSPravin B Shelar skip_lock: 12191807063SAndrea Arcangeli if (put_page_testzero(page_head)) 12291807063SAndrea Arcangeli __put_single_page(page_head); 12391807063SAndrea Arcangeli out_put_single: 12491807063SAndrea Arcangeli if (put_page_testzero(page)) 12591807063SAndrea Arcangeli __put_single_page(page); 12691807063SAndrea Arcangeli return; 12791807063SAndrea Arcangeli } 12891807063SAndrea Arcangeli VM_BUG_ON(page_head != page->first_page); 12991807063SAndrea Arcangeli /* 13091807063SAndrea Arcangeli * We can release the refcount taken by 13170b50f94SAndrea Arcangeli * get_page_unless_zero() now that 13270b50f94SAndrea Arcangeli * __split_huge_page_refcount() is blocked on 13370b50f94SAndrea Arcangeli * the compound_lock. 13491807063SAndrea Arcangeli */ 13591807063SAndrea Arcangeli if (put_page_testzero(page_head)) 13691807063SAndrea Arcangeli VM_BUG_ON(1); 13791807063SAndrea Arcangeli /* __split_huge_page_refcount will wait now */ 13870b50f94SAndrea Arcangeli VM_BUG_ON(page_mapcount(page) <= 0); 13970b50f94SAndrea Arcangeli atomic_dec(&page->_mapcount); 14091807063SAndrea Arcangeli VM_BUG_ON(atomic_read(&page_head->_count) <= 0); 14170b50f94SAndrea Arcangeli VM_BUG_ON(atomic_read(&page->_count) != 0); 14291807063SAndrea Arcangeli compound_unlock_irqrestore(page_head, flags); 1435bf5f03cSPravin B Shelar 1445bf5f03cSPravin B Shelar skip_lock_tail: 145a95a82e9SAndrea Arcangeli if (put_page_testzero(page_head)) { 146a95a82e9SAndrea Arcangeli if (PageHead(page_head)) 14791807063SAndrea Arcangeli __put_compound_page(page_head); 148a95a82e9SAndrea Arcangeli else 149a95a82e9SAndrea Arcangeli __put_single_page(page_head); 150a95a82e9SAndrea Arcangeli } 15191807063SAndrea Arcangeli } else { 15291807063SAndrea Arcangeli /* page_head is a dangling pointer */ 15391807063SAndrea Arcangeli VM_BUG_ON(PageTail(page)); 15491807063SAndrea Arcangeli goto out_put_single; 15591807063SAndrea Arcangeli } 15691807063SAndrea Arcangeli } else if (put_page_testzero(page)) { 15791807063SAndrea Arcangeli if (PageHead(page)) 15891807063SAndrea Arcangeli __put_compound_page(page); 15991807063SAndrea Arcangeli else 16091807063SAndrea Arcangeli __put_single_page(page); 1611da177e4SLinus Torvalds } 1621da177e4SLinus Torvalds } 1638519fb30SNick Piggin 1648519fb30SNick Piggin void put_page(struct page *page) 1658519fb30SNick Piggin { 1668519fb30SNick Piggin if (unlikely(PageCompound(page))) 1678519fb30SNick Piggin put_compound_page(page); 1688519fb30SNick Piggin else if (put_page_testzero(page)) 16991807063SAndrea Arcangeli __put_single_page(page); 1701da177e4SLinus Torvalds } 1711da177e4SLinus Torvalds EXPORT_SYMBOL(put_page); 1721da177e4SLinus Torvalds 17370b50f94SAndrea Arcangeli /* 17470b50f94SAndrea Arcangeli * This function is exported but must not be called by anything other 17570b50f94SAndrea Arcangeli * than get_page(). It implements the slow path of get_page(). 17670b50f94SAndrea Arcangeli */ 17770b50f94SAndrea Arcangeli bool __get_page_tail(struct page *page) 17870b50f94SAndrea Arcangeli { 17970b50f94SAndrea Arcangeli /* 18070b50f94SAndrea Arcangeli * This takes care of get_page() if run on a tail page 18170b50f94SAndrea Arcangeli * returned by one of the get_user_pages/follow_page variants. 18270b50f94SAndrea Arcangeli * get_user_pages/follow_page itself doesn't need the compound 18370b50f94SAndrea Arcangeli * lock because it runs __get_page_tail_foll() under the 18470b50f94SAndrea Arcangeli * proper PT lock that already serializes against 18570b50f94SAndrea Arcangeli * split_huge_page(). 18670b50f94SAndrea Arcangeli */ 18770b50f94SAndrea Arcangeli unsigned long flags; 18870b50f94SAndrea Arcangeli bool got = false; 18970b50f94SAndrea Arcangeli struct page *page_head = compound_trans_head(page); 19070b50f94SAndrea Arcangeli 19170b50f94SAndrea Arcangeli if (likely(page != page_head && get_page_unless_zero(page_head))) { 1925bf5f03cSPravin B Shelar 1935bf5f03cSPravin B Shelar /* Ref to put_compound_page() comment. */ 1945bf5f03cSPravin B Shelar if (PageSlab(page_head)) { 1955bf5f03cSPravin B Shelar if (likely(PageTail(page))) { 1965bf5f03cSPravin B Shelar __get_page_tail_foll(page, false); 1975bf5f03cSPravin B Shelar return true; 1985bf5f03cSPravin B Shelar } else { 1995bf5f03cSPravin B Shelar put_page(page_head); 2005bf5f03cSPravin B Shelar return false; 2015bf5f03cSPravin B Shelar } 2025bf5f03cSPravin B Shelar } 2035bf5f03cSPravin B Shelar 20470b50f94SAndrea Arcangeli /* 20570b50f94SAndrea Arcangeli * page_head wasn't a dangling pointer but it 20670b50f94SAndrea Arcangeli * may not be a head page anymore by the time 20770b50f94SAndrea Arcangeli * we obtain the lock. That is ok as long as it 20870b50f94SAndrea Arcangeli * can't be freed from under us. 20970b50f94SAndrea Arcangeli */ 21070b50f94SAndrea Arcangeli flags = compound_lock_irqsave(page_head); 21170b50f94SAndrea Arcangeli /* here __split_huge_page_refcount won't run anymore */ 21270b50f94SAndrea Arcangeli if (likely(PageTail(page))) { 21370b50f94SAndrea Arcangeli __get_page_tail_foll(page, false); 21470b50f94SAndrea Arcangeli got = true; 21570b50f94SAndrea Arcangeli } 21670b50f94SAndrea Arcangeli compound_unlock_irqrestore(page_head, flags); 21770b50f94SAndrea Arcangeli if (unlikely(!got)) 21870b50f94SAndrea Arcangeli put_page(page_head); 21970b50f94SAndrea Arcangeli } 22070b50f94SAndrea Arcangeli return got; 22170b50f94SAndrea Arcangeli } 22270b50f94SAndrea Arcangeli EXPORT_SYMBOL(__get_page_tail); 22370b50f94SAndrea Arcangeli 2241d7ea732SAlexander Zarochentsev /** 2257682486bSRandy Dunlap * put_pages_list() - release a list of pages 2267682486bSRandy Dunlap * @pages: list of pages threaded on page->lru 2271d7ea732SAlexander Zarochentsev * 2281d7ea732SAlexander Zarochentsev * Release a list of pages which are strung together on page.lru. Currently 2291d7ea732SAlexander Zarochentsev * used by read_cache_pages() and related error recovery code. 2301d7ea732SAlexander Zarochentsev */ 2311d7ea732SAlexander Zarochentsev void put_pages_list(struct list_head *pages) 2321d7ea732SAlexander Zarochentsev { 2331d7ea732SAlexander Zarochentsev while (!list_empty(pages)) { 2341d7ea732SAlexander Zarochentsev struct page *victim; 2351d7ea732SAlexander Zarochentsev 2361d7ea732SAlexander Zarochentsev victim = list_entry(pages->prev, struct page, lru); 2371d7ea732SAlexander Zarochentsev list_del(&victim->lru); 2381d7ea732SAlexander Zarochentsev page_cache_release(victim); 2391d7ea732SAlexander Zarochentsev } 2401d7ea732SAlexander Zarochentsev } 2411d7ea732SAlexander Zarochentsev EXPORT_SYMBOL(put_pages_list); 2421d7ea732SAlexander Zarochentsev 24318022c5dSMel Gorman /* 24418022c5dSMel Gorman * get_kernel_pages() - pin kernel pages in memory 24518022c5dSMel Gorman * @kiov: An array of struct kvec structures 24618022c5dSMel Gorman * @nr_segs: number of segments to pin 24718022c5dSMel Gorman * @write: pinning for read/write, currently ignored 24818022c5dSMel Gorman * @pages: array that receives pointers to the pages pinned. 24918022c5dSMel Gorman * Should be at least nr_segs long. 25018022c5dSMel Gorman * 25118022c5dSMel Gorman * Returns number of pages pinned. This may be fewer than the number 25218022c5dSMel Gorman * requested. If nr_pages is 0 or negative, returns 0. If no pages 25318022c5dSMel Gorman * were pinned, returns -errno. Each page returned must be released 25418022c5dSMel Gorman * with a put_page() call when it is finished with. 25518022c5dSMel Gorman */ 25618022c5dSMel Gorman int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write, 25718022c5dSMel Gorman struct page **pages) 25818022c5dSMel Gorman { 25918022c5dSMel Gorman int seg; 26018022c5dSMel Gorman 26118022c5dSMel Gorman for (seg = 0; seg < nr_segs; seg++) { 26218022c5dSMel Gorman if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE)) 26318022c5dSMel Gorman return seg; 26418022c5dSMel Gorman 2655a178119SMel Gorman pages[seg] = kmap_to_page(kiov[seg].iov_base); 26618022c5dSMel Gorman page_cache_get(pages[seg]); 26718022c5dSMel Gorman } 26818022c5dSMel Gorman 26918022c5dSMel Gorman return seg; 27018022c5dSMel Gorman } 27118022c5dSMel Gorman EXPORT_SYMBOL_GPL(get_kernel_pages); 27218022c5dSMel Gorman 27318022c5dSMel Gorman /* 27418022c5dSMel Gorman * get_kernel_page() - pin a kernel page in memory 27518022c5dSMel Gorman * @start: starting kernel address 27618022c5dSMel Gorman * @write: pinning for read/write, currently ignored 27718022c5dSMel Gorman * @pages: array that receives pointer to the page pinned. 27818022c5dSMel Gorman * Must be at least nr_segs long. 27918022c5dSMel Gorman * 28018022c5dSMel Gorman * Returns 1 if page is pinned. If the page was not pinned, returns 28118022c5dSMel Gorman * -errno. The page returned must be released with a put_page() call 28218022c5dSMel Gorman * when it is finished with. 28318022c5dSMel Gorman */ 28418022c5dSMel Gorman int get_kernel_page(unsigned long start, int write, struct page **pages) 28518022c5dSMel Gorman { 28618022c5dSMel Gorman const struct kvec kiov = { 28718022c5dSMel Gorman .iov_base = (void *)start, 28818022c5dSMel Gorman .iov_len = PAGE_SIZE 28918022c5dSMel Gorman }; 29018022c5dSMel Gorman 29118022c5dSMel Gorman return get_kernel_pages(&kiov, 1, write, pages); 29218022c5dSMel Gorman } 29318022c5dSMel Gorman EXPORT_SYMBOL_GPL(get_kernel_page); 29418022c5dSMel Gorman 2953dd7ae8eSShaohua Li static void pagevec_lru_move_fn(struct pagevec *pvec, 296fa9add64SHugh Dickins void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg), 2973dd7ae8eSShaohua Li void *arg) 298902aaed0SHisashi Hifumi { 299902aaed0SHisashi Hifumi int i; 300902aaed0SHisashi Hifumi struct zone *zone = NULL; 301fa9add64SHugh Dickins struct lruvec *lruvec; 3023dd7ae8eSShaohua Li unsigned long flags = 0; 303902aaed0SHisashi Hifumi 304902aaed0SHisashi Hifumi for (i = 0; i < pagevec_count(pvec); i++) { 305902aaed0SHisashi Hifumi struct page *page = pvec->pages[i]; 306902aaed0SHisashi Hifumi struct zone *pagezone = page_zone(page); 307902aaed0SHisashi Hifumi 308902aaed0SHisashi Hifumi if (pagezone != zone) { 309902aaed0SHisashi Hifumi if (zone) 3103dd7ae8eSShaohua Li spin_unlock_irqrestore(&zone->lru_lock, flags); 311902aaed0SHisashi Hifumi zone = pagezone; 3123dd7ae8eSShaohua Li spin_lock_irqsave(&zone->lru_lock, flags); 313902aaed0SHisashi Hifumi } 3143dd7ae8eSShaohua Li 315fa9add64SHugh Dickins lruvec = mem_cgroup_page_lruvec(page, zone); 316fa9add64SHugh Dickins (*move_fn)(page, lruvec, arg); 3173dd7ae8eSShaohua Li } 3183dd7ae8eSShaohua Li if (zone) 3193dd7ae8eSShaohua Li spin_unlock_irqrestore(&zone->lru_lock, flags); 3203dd7ae8eSShaohua Li release_pages(pvec->pages, pvec->nr, pvec->cold); 3213dd7ae8eSShaohua Li pagevec_reinit(pvec); 3223dd7ae8eSShaohua Li } 3233dd7ae8eSShaohua Li 324fa9add64SHugh Dickins static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec, 325fa9add64SHugh Dickins void *arg) 3263dd7ae8eSShaohua Li { 3273dd7ae8eSShaohua Li int *pgmoved = arg; 3283dd7ae8eSShaohua Li 329894bc310SLee Schermerhorn if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { 3303f58a829SMinchan Kim enum lru_list lru = page_lru_base_type(page); 331925b7673SJohannes Weiner list_move_tail(&page->lru, &lruvec->lists[lru]); 3323dd7ae8eSShaohua Li (*pgmoved)++; 333902aaed0SHisashi Hifumi } 334902aaed0SHisashi Hifumi } 3353dd7ae8eSShaohua Li 3363dd7ae8eSShaohua Li /* 3373dd7ae8eSShaohua Li * pagevec_move_tail() must be called with IRQ disabled. 3383dd7ae8eSShaohua Li * Otherwise this may cause nasty races. 3393dd7ae8eSShaohua Li */ 3403dd7ae8eSShaohua Li static void pagevec_move_tail(struct pagevec *pvec) 3413dd7ae8eSShaohua Li { 3423dd7ae8eSShaohua Li int pgmoved = 0; 3433dd7ae8eSShaohua Li 3443dd7ae8eSShaohua Li pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved); 345902aaed0SHisashi Hifumi __count_vm_events(PGROTATED, pgmoved); 346902aaed0SHisashi Hifumi } 347902aaed0SHisashi Hifumi 348902aaed0SHisashi Hifumi /* 3491da177e4SLinus Torvalds * Writeback is about to end against a page which has been marked for immediate 3501da177e4SLinus Torvalds * reclaim. If it still appears to be reclaimable, move it to the tail of the 351902aaed0SHisashi Hifumi * inactive list. 3521da177e4SLinus Torvalds */ 353ac6aadb2SMiklos Szeredi void rotate_reclaimable_page(struct page *page) 3541da177e4SLinus Torvalds { 355ac6aadb2SMiklos Szeredi if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) && 356894bc310SLee Schermerhorn !PageUnevictable(page) && PageLRU(page)) { 357902aaed0SHisashi Hifumi struct pagevec *pvec; 3581da177e4SLinus Torvalds unsigned long flags; 3591da177e4SLinus Torvalds 360902aaed0SHisashi Hifumi page_cache_get(page); 361902aaed0SHisashi Hifumi local_irq_save(flags); 362902aaed0SHisashi Hifumi pvec = &__get_cpu_var(lru_rotate_pvecs); 363902aaed0SHisashi Hifumi if (!pagevec_add(pvec, page)) 364902aaed0SHisashi Hifumi pagevec_move_tail(pvec); 365902aaed0SHisashi Hifumi local_irq_restore(flags); 366ac6aadb2SMiklos Szeredi } 3671da177e4SLinus Torvalds } 3681da177e4SLinus Torvalds 369fa9add64SHugh Dickins static void update_page_reclaim_stat(struct lruvec *lruvec, 3703e2f41f1SKOSAKI Motohiro int file, int rotated) 3713e2f41f1SKOSAKI Motohiro { 372fa9add64SHugh Dickins struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; 3733e2f41f1SKOSAKI Motohiro 3743e2f41f1SKOSAKI Motohiro reclaim_stat->recent_scanned[file]++; 3753e2f41f1SKOSAKI Motohiro if (rotated) 3763e2f41f1SKOSAKI Motohiro reclaim_stat->recent_rotated[file]++; 3773e2f41f1SKOSAKI Motohiro } 3783e2f41f1SKOSAKI Motohiro 379fa9add64SHugh Dickins static void __activate_page(struct page *page, struct lruvec *lruvec, 380fa9add64SHugh Dickins void *arg) 381744ed144SShaohua Li { 3827a608572SLinus Torvalds if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { 383744ed144SShaohua Li int file = page_is_file_cache(page); 384744ed144SShaohua Li int lru = page_lru_base_type(page); 385744ed144SShaohua Li 386fa9add64SHugh Dickins del_page_from_lru_list(page, lruvec, lru); 387744ed144SShaohua Li SetPageActive(page); 388744ed144SShaohua Li lru += LRU_ACTIVE; 389fa9add64SHugh Dickins add_page_to_lru_list(page, lruvec, lru); 390c6286c98SMel Gorman trace_mm_lru_activate(page, page_to_pfn(page)); 3917a608572SLinus Torvalds 392fa9add64SHugh Dickins __count_vm_event(PGACTIVATE); 393fa9add64SHugh Dickins update_page_reclaim_stat(lruvec, file, 1); 394744ed144SShaohua Li } 395eb709b0dSShaohua Li } 396eb709b0dSShaohua Li 397eb709b0dSShaohua Li #ifdef CONFIG_SMP 398eb709b0dSShaohua Li static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs); 399eb709b0dSShaohua Li 400eb709b0dSShaohua Li static void activate_page_drain(int cpu) 401eb709b0dSShaohua Li { 402eb709b0dSShaohua Li struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu); 403eb709b0dSShaohua Li 404eb709b0dSShaohua Li if (pagevec_count(pvec)) 405eb709b0dSShaohua Li pagevec_lru_move_fn(pvec, __activate_page, NULL); 406eb709b0dSShaohua Li } 407eb709b0dSShaohua Li 408eb709b0dSShaohua Li void activate_page(struct page *page) 409eb709b0dSShaohua Li { 410eb709b0dSShaohua Li if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { 411eb709b0dSShaohua Li struct pagevec *pvec = &get_cpu_var(activate_page_pvecs); 412eb709b0dSShaohua Li 413eb709b0dSShaohua Li page_cache_get(page); 414eb709b0dSShaohua Li if (!pagevec_add(pvec, page)) 415eb709b0dSShaohua Li pagevec_lru_move_fn(pvec, __activate_page, NULL); 416eb709b0dSShaohua Li put_cpu_var(activate_page_pvecs); 417eb709b0dSShaohua Li } 418eb709b0dSShaohua Li } 419eb709b0dSShaohua Li 420eb709b0dSShaohua Li #else 421eb709b0dSShaohua Li static inline void activate_page_drain(int cpu) 422eb709b0dSShaohua Li { 423eb709b0dSShaohua Li } 424eb709b0dSShaohua Li 425eb709b0dSShaohua Li void activate_page(struct page *page) 426eb709b0dSShaohua Li { 427eb709b0dSShaohua Li struct zone *zone = page_zone(page); 428eb709b0dSShaohua Li 429eb709b0dSShaohua Li spin_lock_irq(&zone->lru_lock); 430fa9add64SHugh Dickins __activate_page(page, mem_cgroup_page_lruvec(page, zone), NULL); 4311da177e4SLinus Torvalds spin_unlock_irq(&zone->lru_lock); 4321da177e4SLinus Torvalds } 433eb709b0dSShaohua Li #endif 4341da177e4SLinus Torvalds 435*059285a2SMel Gorman static void __lru_cache_activate_page(struct page *page) 436*059285a2SMel Gorman { 437*059285a2SMel Gorman struct pagevec *pvec = &get_cpu_var(lru_add_pvec); 438*059285a2SMel Gorman int i; 439*059285a2SMel Gorman 440*059285a2SMel Gorman /* 441*059285a2SMel Gorman * Search backwards on the optimistic assumption that the page being 442*059285a2SMel Gorman * activated has just been added to this pagevec. Note that only 443*059285a2SMel Gorman * the local pagevec is examined as a !PageLRU page could be in the 444*059285a2SMel Gorman * process of being released, reclaimed, migrated or on a remote 445*059285a2SMel Gorman * pagevec that is currently being drained. Furthermore, marking 446*059285a2SMel Gorman * a remote pagevec's page PageActive potentially hits a race where 447*059285a2SMel Gorman * a page is marked PageActive just after it is added to the inactive 448*059285a2SMel Gorman * list causing accounting errors and BUG_ON checks to trigger. 449*059285a2SMel Gorman */ 450*059285a2SMel Gorman for (i = pagevec_count(pvec) - 1; i >= 0; i--) { 451*059285a2SMel Gorman struct page *pagevec_page = pvec->pages[i]; 452*059285a2SMel Gorman 453*059285a2SMel Gorman if (pagevec_page == page) { 454*059285a2SMel Gorman SetPageActive(page); 455*059285a2SMel Gorman break; 456*059285a2SMel Gorman } 457*059285a2SMel Gorman } 458*059285a2SMel Gorman 459*059285a2SMel Gorman put_cpu_var(lru_add_pvec); 460*059285a2SMel Gorman } 461*059285a2SMel Gorman 4621da177e4SLinus Torvalds /* 4631da177e4SLinus Torvalds * Mark a page as having seen activity. 4641da177e4SLinus Torvalds * 4651da177e4SLinus Torvalds * inactive,unreferenced -> inactive,referenced 4661da177e4SLinus Torvalds * inactive,referenced -> active,unreferenced 4671da177e4SLinus Torvalds * active,unreferenced -> active,referenced 4681da177e4SLinus Torvalds */ 469920c7a5dSHarvey Harrison void mark_page_accessed(struct page *page) 4701da177e4SLinus Torvalds { 471894bc310SLee Schermerhorn if (!PageActive(page) && !PageUnevictable(page) && 472*059285a2SMel Gorman PageReferenced(page)) { 473*059285a2SMel Gorman 474*059285a2SMel Gorman /* 475*059285a2SMel Gorman * If the page is on the LRU, queue it for activation via 476*059285a2SMel Gorman * activate_page_pvecs. Otherwise, assume the page is on a 477*059285a2SMel Gorman * pagevec, mark it active and it'll be moved to the active 478*059285a2SMel Gorman * LRU on the next drain. 479*059285a2SMel Gorman */ 480*059285a2SMel Gorman if (PageLRU(page)) 4811da177e4SLinus Torvalds activate_page(page); 482*059285a2SMel Gorman else 483*059285a2SMel Gorman __lru_cache_activate_page(page); 4841da177e4SLinus Torvalds ClearPageReferenced(page); 4851da177e4SLinus Torvalds } else if (!PageReferenced(page)) { 4861da177e4SLinus Torvalds SetPageReferenced(page); 4871da177e4SLinus Torvalds } 4881da177e4SLinus Torvalds } 4891da177e4SLinus Torvalds EXPORT_SYMBOL(mark_page_accessed); 4901da177e4SLinus Torvalds 491d741c9cdSRobin Dong /* 49213f7f789SMel Gorman * Queue the page for addition to the LRU via pagevec. The decision on whether 49313f7f789SMel Gorman * to add the page to the [in]active [file|anon] list is deferred until the 49413f7f789SMel Gorman * pagevec is drained. This gives a chance for the caller of __lru_cache_add() 49513f7f789SMel Gorman * have the page added to the active list using mark_page_accessed(). 496d741c9cdSRobin Dong */ 497f04e9ebbSKOSAKI Motohiro void __lru_cache_add(struct page *page, enum lru_list lru) 4981da177e4SLinus Torvalds { 49913f7f789SMel Gorman struct pagevec *pvec = &get_cpu_var(lru_add_pvec); 50013f7f789SMel Gorman 50113f7f789SMel Gorman if (is_active_lru(lru)) 50213f7f789SMel Gorman SetPageActive(page); 50313f7f789SMel Gorman else 50413f7f789SMel Gorman ClearPageActive(page); 5051da177e4SLinus Torvalds 5061da177e4SLinus Torvalds page_cache_get(page); 507d741c9cdSRobin Dong if (!pagevec_space(pvec)) 5085095ae83SHugh Dickins __pagevec_lru_add(pvec, lru); 509d741c9cdSRobin Dong pagevec_add(pvec, page); 51013f7f789SMel Gorman put_cpu_var(lru_add_pvec); 5111da177e4SLinus Torvalds } 51247846b06SMiklos Szeredi EXPORT_SYMBOL(__lru_cache_add); 5131da177e4SLinus Torvalds 514f04e9ebbSKOSAKI Motohiro /** 515f04e9ebbSKOSAKI Motohiro * lru_cache_add_lru - add a page to a page list 516f04e9ebbSKOSAKI Motohiro * @page: the page to be added to the LRU. 517f04e9ebbSKOSAKI Motohiro * @lru: the LRU list to which the page is added. 518f04e9ebbSKOSAKI Motohiro */ 519f04e9ebbSKOSAKI Motohiro void lru_cache_add_lru(struct page *page, enum lru_list lru) 5201da177e4SLinus Torvalds { 521f04e9ebbSKOSAKI Motohiro if (PageActive(page)) { 522894bc310SLee Schermerhorn VM_BUG_ON(PageUnevictable(page)); 523894bc310SLee Schermerhorn } else if (PageUnevictable(page)) { 524894bc310SLee Schermerhorn VM_BUG_ON(PageActive(page)); 525f04e9ebbSKOSAKI Motohiro } 5261da177e4SLinus Torvalds 52713f7f789SMel Gorman VM_BUG_ON(PageLRU(page)); 528f04e9ebbSKOSAKI Motohiro __lru_cache_add(page, lru); 5291da177e4SLinus Torvalds } 5301da177e4SLinus Torvalds 531894bc310SLee Schermerhorn /** 532894bc310SLee Schermerhorn * add_page_to_unevictable_list - add a page to the unevictable list 533894bc310SLee Schermerhorn * @page: the page to be added to the unevictable list 534894bc310SLee Schermerhorn * 535894bc310SLee Schermerhorn * Add page directly to its zone's unevictable list. To avoid races with 536894bc310SLee Schermerhorn * tasks that might be making the page evictable, through eg. munlock, 537894bc310SLee Schermerhorn * munmap or exit, while it's not on the lru, we want to add the page 538894bc310SLee Schermerhorn * while it's locked or otherwise "invisible" to other tasks. This is 539894bc310SLee Schermerhorn * difficult to do when using the pagevec cache, so bypass that. 540894bc310SLee Schermerhorn */ 541894bc310SLee Schermerhorn void add_page_to_unevictable_list(struct page *page) 542894bc310SLee Schermerhorn { 543894bc310SLee Schermerhorn struct zone *zone = page_zone(page); 544fa9add64SHugh Dickins struct lruvec *lruvec; 545894bc310SLee Schermerhorn 546894bc310SLee Schermerhorn spin_lock_irq(&zone->lru_lock); 547fa9add64SHugh Dickins lruvec = mem_cgroup_page_lruvec(page, zone); 548894bc310SLee Schermerhorn SetPageUnevictable(page); 549894bc310SLee Schermerhorn SetPageLRU(page); 550fa9add64SHugh Dickins add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE); 551894bc310SLee Schermerhorn spin_unlock_irq(&zone->lru_lock); 552894bc310SLee Schermerhorn } 553894bc310SLee Schermerhorn 554902aaed0SHisashi Hifumi /* 55531560180SMinchan Kim * If the page can not be invalidated, it is moved to the 55631560180SMinchan Kim * inactive list to speed up its reclaim. It is moved to the 55731560180SMinchan Kim * head of the list, rather than the tail, to give the flusher 55831560180SMinchan Kim * threads some time to write it out, as this is much more 55931560180SMinchan Kim * effective than the single-page writeout from reclaim. 560278df9f4SMinchan Kim * 561278df9f4SMinchan Kim * If the page isn't page_mapped and dirty/writeback, the page 562278df9f4SMinchan Kim * could reclaim asap using PG_reclaim. 563278df9f4SMinchan Kim * 564278df9f4SMinchan Kim * 1. active, mapped page -> none 565278df9f4SMinchan Kim * 2. active, dirty/writeback page -> inactive, head, PG_reclaim 566278df9f4SMinchan Kim * 3. inactive, mapped page -> none 567278df9f4SMinchan Kim * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim 568278df9f4SMinchan Kim * 5. inactive, clean -> inactive, tail 569278df9f4SMinchan Kim * 6. Others -> none 570278df9f4SMinchan Kim * 571278df9f4SMinchan Kim * In 4, why it moves inactive's head, the VM expects the page would 572278df9f4SMinchan Kim * be write it out by flusher threads as this is much more effective 573278df9f4SMinchan Kim * than the single-page writeout from reclaim. 57431560180SMinchan Kim */ 575fa9add64SHugh Dickins static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec, 576fa9add64SHugh Dickins void *arg) 57731560180SMinchan Kim { 57831560180SMinchan Kim int lru, file; 579278df9f4SMinchan Kim bool active; 58031560180SMinchan Kim 581278df9f4SMinchan Kim if (!PageLRU(page)) 58231560180SMinchan Kim return; 58331560180SMinchan Kim 584bad49d9cSMinchan Kim if (PageUnevictable(page)) 585bad49d9cSMinchan Kim return; 586bad49d9cSMinchan Kim 58731560180SMinchan Kim /* Some processes are using the page */ 58831560180SMinchan Kim if (page_mapped(page)) 58931560180SMinchan Kim return; 59031560180SMinchan Kim 591278df9f4SMinchan Kim active = PageActive(page); 59231560180SMinchan Kim file = page_is_file_cache(page); 59331560180SMinchan Kim lru = page_lru_base_type(page); 594fa9add64SHugh Dickins 595fa9add64SHugh Dickins del_page_from_lru_list(page, lruvec, lru + active); 59631560180SMinchan Kim ClearPageActive(page); 59731560180SMinchan Kim ClearPageReferenced(page); 598fa9add64SHugh Dickins add_page_to_lru_list(page, lruvec, lru); 59931560180SMinchan Kim 600278df9f4SMinchan Kim if (PageWriteback(page) || PageDirty(page)) { 601278df9f4SMinchan Kim /* 602278df9f4SMinchan Kim * PG_reclaim could be raced with end_page_writeback 603278df9f4SMinchan Kim * It can make readahead confusing. But race window 604278df9f4SMinchan Kim * is _really_ small and it's non-critical problem. 605278df9f4SMinchan Kim */ 606278df9f4SMinchan Kim SetPageReclaim(page); 607278df9f4SMinchan Kim } else { 608278df9f4SMinchan Kim /* 609278df9f4SMinchan Kim * The page's writeback ends up during pagevec 610278df9f4SMinchan Kim * We moves tha page into tail of inactive. 611278df9f4SMinchan Kim */ 612925b7673SJohannes Weiner list_move_tail(&page->lru, &lruvec->lists[lru]); 613278df9f4SMinchan Kim __count_vm_event(PGROTATED); 614278df9f4SMinchan Kim } 615278df9f4SMinchan Kim 616278df9f4SMinchan Kim if (active) 617278df9f4SMinchan Kim __count_vm_event(PGDEACTIVATE); 618fa9add64SHugh Dickins update_page_reclaim_stat(lruvec, file, 0); 61931560180SMinchan Kim } 62031560180SMinchan Kim 62131560180SMinchan Kim /* 622902aaed0SHisashi Hifumi * Drain pages out of the cpu's pagevecs. 623902aaed0SHisashi Hifumi * Either "cpu" is the current CPU, and preemption has already been 624902aaed0SHisashi Hifumi * disabled; or "cpu" is being hot-unplugged, and is already dead. 625902aaed0SHisashi Hifumi */ 626f0cb3c76SKonstantin Khlebnikov void lru_add_drain_cpu(int cpu) 6271da177e4SLinus Torvalds { 62813f7f789SMel Gorman struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu); 6291da177e4SLinus Torvalds 6301da177e4SLinus Torvalds if (pagevec_count(pvec)) 63113f7f789SMel Gorman __pagevec_lru_add(pvec, NR_LRU_LISTS); 632902aaed0SHisashi Hifumi 633902aaed0SHisashi Hifumi pvec = &per_cpu(lru_rotate_pvecs, cpu); 634902aaed0SHisashi Hifumi if (pagevec_count(pvec)) { 635902aaed0SHisashi Hifumi unsigned long flags; 636902aaed0SHisashi Hifumi 637902aaed0SHisashi Hifumi /* No harm done if a racing interrupt already did this */ 638902aaed0SHisashi Hifumi local_irq_save(flags); 639902aaed0SHisashi Hifumi pagevec_move_tail(pvec); 640902aaed0SHisashi Hifumi local_irq_restore(flags); 641902aaed0SHisashi Hifumi } 64231560180SMinchan Kim 64331560180SMinchan Kim pvec = &per_cpu(lru_deactivate_pvecs, cpu); 64431560180SMinchan Kim if (pagevec_count(pvec)) 6453dd7ae8eSShaohua Li pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL); 646eb709b0dSShaohua Li 647eb709b0dSShaohua Li activate_page_drain(cpu); 64831560180SMinchan Kim } 64931560180SMinchan Kim 65031560180SMinchan Kim /** 65131560180SMinchan Kim * deactivate_page - forcefully deactivate a page 65231560180SMinchan Kim * @page: page to deactivate 65331560180SMinchan Kim * 65431560180SMinchan Kim * This function hints the VM that @page is a good reclaim candidate, 65531560180SMinchan Kim * for example if its invalidation fails due to the page being dirty 65631560180SMinchan Kim * or under writeback. 65731560180SMinchan Kim */ 65831560180SMinchan Kim void deactivate_page(struct page *page) 65931560180SMinchan Kim { 660821ed6bbSMinchan Kim /* 661821ed6bbSMinchan Kim * In a workload with many unevictable page such as mprotect, unevictable 662821ed6bbSMinchan Kim * page deactivation for accelerating reclaim is pointless. 663821ed6bbSMinchan Kim */ 664821ed6bbSMinchan Kim if (PageUnevictable(page)) 665821ed6bbSMinchan Kim return; 666821ed6bbSMinchan Kim 66731560180SMinchan Kim if (likely(get_page_unless_zero(page))) { 66831560180SMinchan Kim struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs); 66931560180SMinchan Kim 67031560180SMinchan Kim if (!pagevec_add(pvec, page)) 6713dd7ae8eSShaohua Li pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL); 67231560180SMinchan Kim put_cpu_var(lru_deactivate_pvecs); 67331560180SMinchan Kim } 67480bfed90SAndrew Morton } 67580bfed90SAndrew Morton 67680bfed90SAndrew Morton void lru_add_drain(void) 67780bfed90SAndrew Morton { 678f0cb3c76SKonstantin Khlebnikov lru_add_drain_cpu(get_cpu()); 67980bfed90SAndrew Morton put_cpu(); 6801da177e4SLinus Torvalds } 6811da177e4SLinus Torvalds 682c4028958SDavid Howells static void lru_add_drain_per_cpu(struct work_struct *dummy) 683053837fcSNick Piggin { 684053837fcSNick Piggin lru_add_drain(); 685053837fcSNick Piggin } 686053837fcSNick Piggin 687053837fcSNick Piggin /* 688053837fcSNick Piggin * Returns 0 for success 689053837fcSNick Piggin */ 690053837fcSNick Piggin int lru_add_drain_all(void) 691053837fcSNick Piggin { 692c4028958SDavid Howells return schedule_on_each_cpu(lru_add_drain_per_cpu); 693053837fcSNick Piggin } 694053837fcSNick Piggin 6951da177e4SLinus Torvalds /* 6961da177e4SLinus Torvalds * Batched page_cache_release(). Decrement the reference count on all the 6971da177e4SLinus Torvalds * passed pages. If it fell to zero then remove the page from the LRU and 6981da177e4SLinus Torvalds * free it. 6991da177e4SLinus Torvalds * 7001da177e4SLinus Torvalds * Avoid taking zone->lru_lock if possible, but if it is taken, retain it 7011da177e4SLinus Torvalds * for the remainder of the operation. 7021da177e4SLinus Torvalds * 703ab33dc09SFernando Luis Vazquez Cao * The locking in this function is against shrink_inactive_list(): we recheck 704ab33dc09SFernando Luis Vazquez Cao * the page count inside the lock to see whether shrink_inactive_list() 705ab33dc09SFernando Luis Vazquez Cao * grabbed the page via the LRU. If it did, give up: shrink_inactive_list() 706ab33dc09SFernando Luis Vazquez Cao * will free it. 7071da177e4SLinus Torvalds */ 7081da177e4SLinus Torvalds void release_pages(struct page **pages, int nr, int cold) 7091da177e4SLinus Torvalds { 7101da177e4SLinus Torvalds int i; 711cc59850eSKonstantin Khlebnikov LIST_HEAD(pages_to_free); 7121da177e4SLinus Torvalds struct zone *zone = NULL; 713fa9add64SHugh Dickins struct lruvec *lruvec; 714902aaed0SHisashi Hifumi unsigned long uninitialized_var(flags); 7151da177e4SLinus Torvalds 7161da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 7171da177e4SLinus Torvalds struct page *page = pages[i]; 7181da177e4SLinus Torvalds 7198519fb30SNick Piggin if (unlikely(PageCompound(page))) { 7208519fb30SNick Piggin if (zone) { 721902aaed0SHisashi Hifumi spin_unlock_irqrestore(&zone->lru_lock, flags); 7228519fb30SNick Piggin zone = NULL; 7238519fb30SNick Piggin } 7248519fb30SNick Piggin put_compound_page(page); 7258519fb30SNick Piggin continue; 7268519fb30SNick Piggin } 7278519fb30SNick Piggin 728b5810039SNick Piggin if (!put_page_testzero(page)) 7291da177e4SLinus Torvalds continue; 7301da177e4SLinus Torvalds 73146453a6eSNick Piggin if (PageLRU(page)) { 73246453a6eSNick Piggin struct zone *pagezone = page_zone(page); 733894bc310SLee Schermerhorn 7341da177e4SLinus Torvalds if (pagezone != zone) { 7351da177e4SLinus Torvalds if (zone) 736902aaed0SHisashi Hifumi spin_unlock_irqrestore(&zone->lru_lock, 737902aaed0SHisashi Hifumi flags); 7381da177e4SLinus Torvalds zone = pagezone; 739902aaed0SHisashi Hifumi spin_lock_irqsave(&zone->lru_lock, flags); 7401da177e4SLinus Torvalds } 741fa9add64SHugh Dickins 742fa9add64SHugh Dickins lruvec = mem_cgroup_page_lruvec(page, zone); 743725d704eSNick Piggin VM_BUG_ON(!PageLRU(page)); 74467453911SNick Piggin __ClearPageLRU(page); 745fa9add64SHugh Dickins del_page_from_lru_list(page, lruvec, page_off_lru(page)); 74646453a6eSNick Piggin } 74746453a6eSNick Piggin 748cc59850eSKonstantin Khlebnikov list_add(&page->lru, &pages_to_free); 7491da177e4SLinus Torvalds } 7501da177e4SLinus Torvalds if (zone) 751902aaed0SHisashi Hifumi spin_unlock_irqrestore(&zone->lru_lock, flags); 7521da177e4SLinus Torvalds 753cc59850eSKonstantin Khlebnikov free_hot_cold_page_list(&pages_to_free, cold); 7541da177e4SLinus Torvalds } 7550be8557bSMiklos Szeredi EXPORT_SYMBOL(release_pages); 7561da177e4SLinus Torvalds 7571da177e4SLinus Torvalds /* 7581da177e4SLinus Torvalds * The pages which we're about to release may be in the deferred lru-addition 7591da177e4SLinus Torvalds * queues. That would prevent them from really being freed right now. That's 7601da177e4SLinus Torvalds * OK from a correctness point of view but is inefficient - those pages may be 7611da177e4SLinus Torvalds * cache-warm and we want to give them back to the page allocator ASAP. 7621da177e4SLinus Torvalds * 7631da177e4SLinus Torvalds * So __pagevec_release() will drain those queues here. __pagevec_lru_add() 7641da177e4SLinus Torvalds * and __pagevec_lru_add_active() call release_pages() directly to avoid 7651da177e4SLinus Torvalds * mutual recursion. 7661da177e4SLinus Torvalds */ 7671da177e4SLinus Torvalds void __pagevec_release(struct pagevec *pvec) 7681da177e4SLinus Torvalds { 7691da177e4SLinus Torvalds lru_add_drain(); 7701da177e4SLinus Torvalds release_pages(pvec->pages, pagevec_count(pvec), pvec->cold); 7711da177e4SLinus Torvalds pagevec_reinit(pvec); 7721da177e4SLinus Torvalds } 7737f285701SSteve French EXPORT_SYMBOL(__pagevec_release); 7747f285701SSteve French 77512d27107SHugh Dickins #ifdef CONFIG_TRANSPARENT_HUGEPAGE 77671e3aac0SAndrea Arcangeli /* used by __split_huge_page_refcount() */ 777fa9add64SHugh Dickins void lru_add_page_tail(struct page *page, struct page *page_tail, 7785bc7b8acSShaohua Li struct lruvec *lruvec, struct list_head *list) 77971e3aac0SAndrea Arcangeli { 7807512102cSHugh Dickins int uninitialized_var(active); 78171e3aac0SAndrea Arcangeli enum lru_list lru; 78271e3aac0SAndrea Arcangeli const int file = 0; 78371e3aac0SAndrea Arcangeli 78471e3aac0SAndrea Arcangeli VM_BUG_ON(!PageHead(page)); 78571e3aac0SAndrea Arcangeli VM_BUG_ON(PageCompound(page_tail)); 78671e3aac0SAndrea Arcangeli VM_BUG_ON(PageLRU(page_tail)); 787fa9add64SHugh Dickins VM_BUG_ON(NR_CPUS != 1 && 788fa9add64SHugh Dickins !spin_is_locked(&lruvec_zone(lruvec)->lru_lock)); 78971e3aac0SAndrea Arcangeli 7905bc7b8acSShaohua Li if (!list) 79171e3aac0SAndrea Arcangeli SetPageLRU(page_tail); 79271e3aac0SAndrea Arcangeli 79339b5f29aSHugh Dickins if (page_evictable(page_tail)) { 79471e3aac0SAndrea Arcangeli if (PageActive(page)) { 79571e3aac0SAndrea Arcangeli SetPageActive(page_tail); 79671e3aac0SAndrea Arcangeli active = 1; 79771e3aac0SAndrea Arcangeli lru = LRU_ACTIVE_ANON; 79871e3aac0SAndrea Arcangeli } else { 79971e3aac0SAndrea Arcangeli active = 0; 80071e3aac0SAndrea Arcangeli lru = LRU_INACTIVE_ANON; 80171e3aac0SAndrea Arcangeli } 80271e3aac0SAndrea Arcangeli } else { 80371e3aac0SAndrea Arcangeli SetPageUnevictable(page_tail); 80412d27107SHugh Dickins lru = LRU_UNEVICTABLE; 80512d27107SHugh Dickins } 80612d27107SHugh Dickins 80712d27107SHugh Dickins if (likely(PageLRU(page))) 80812d27107SHugh Dickins list_add_tail(&page_tail->lru, &page->lru); 8095bc7b8acSShaohua Li else if (list) { 8105bc7b8acSShaohua Li /* page reclaim is reclaiming a huge page */ 8115bc7b8acSShaohua Li get_page(page_tail); 8125bc7b8acSShaohua Li list_add_tail(&page_tail->lru, list); 8135bc7b8acSShaohua Li } else { 81412d27107SHugh Dickins struct list_head *list_head; 81512d27107SHugh Dickins /* 81612d27107SHugh Dickins * Head page has not yet been counted, as an hpage, 81712d27107SHugh Dickins * so we must account for each subpage individually. 81812d27107SHugh Dickins * 81912d27107SHugh Dickins * Use the standard add function to put page_tail on the list, 82012d27107SHugh Dickins * but then correct its position so they all end up in order. 82112d27107SHugh Dickins */ 822fa9add64SHugh Dickins add_page_to_lru_list(page_tail, lruvec, lru); 82312d27107SHugh Dickins list_head = page_tail->lru.prev; 82412d27107SHugh Dickins list_move_tail(&page_tail->lru, list_head); 82571e3aac0SAndrea Arcangeli } 8267512102cSHugh Dickins 8277512102cSHugh Dickins if (!PageUnevictable(page)) 828fa9add64SHugh Dickins update_page_reclaim_stat(lruvec, file, active); 82971e3aac0SAndrea Arcangeli } 83012d27107SHugh Dickins #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 83171e3aac0SAndrea Arcangeli 832fa9add64SHugh Dickins static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec, 833fa9add64SHugh Dickins void *arg) 8343dd7ae8eSShaohua Li { 83513f7f789SMel Gorman enum lru_list requested_lru = (enum lru_list)arg; 83613f7f789SMel Gorman int file = page_is_file_cache(page); 83713f7f789SMel Gorman int active = PageActive(page); 83813f7f789SMel Gorman enum lru_list lru = page_lru(page); 8393dd7ae8eSShaohua Li 84013f7f789SMel Gorman WARN_ON_ONCE(requested_lru < NR_LRU_LISTS && requested_lru != lru); 8413dd7ae8eSShaohua Li VM_BUG_ON(PageUnevictable(page)); 8423dd7ae8eSShaohua Li VM_BUG_ON(PageLRU(page)); 8433dd7ae8eSShaohua Li 8443dd7ae8eSShaohua Li SetPageLRU(page); 845fa9add64SHugh Dickins add_page_to_lru_list(page, lruvec, lru); 846fa9add64SHugh Dickins update_page_reclaim_stat(lruvec, file, active); 847c6286c98SMel Gorman trace_mm_lru_insertion(page, page_to_pfn(page), lru, trace_pagemap_flags(page)); 8483dd7ae8eSShaohua Li } 8493dd7ae8eSShaohua Li 8501da177e4SLinus Torvalds /* 8511da177e4SLinus Torvalds * Add the passed pages to the LRU, then drop the caller's refcount 8521da177e4SLinus Torvalds * on them. Reinitialises the caller's pagevec. 8531da177e4SLinus Torvalds */ 8545095ae83SHugh Dickins void __pagevec_lru_add(struct pagevec *pvec, enum lru_list lru) 8551da177e4SLinus Torvalds { 856894bc310SLee Schermerhorn VM_BUG_ON(is_unevictable_lru(lru)); 8571da177e4SLinus Torvalds 8585095ae83SHugh Dickins pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, (void *)lru); 8591da177e4SLinus Torvalds } 8605095ae83SHugh Dickins EXPORT_SYMBOL(__pagevec_lru_add); 861f04e9ebbSKOSAKI Motohiro 8621da177e4SLinus Torvalds /** 8631da177e4SLinus Torvalds * pagevec_lookup - gang pagecache lookup 8641da177e4SLinus Torvalds * @pvec: Where the resulting pages are placed 8651da177e4SLinus Torvalds * @mapping: The address_space to search 8661da177e4SLinus Torvalds * @start: The starting page index 8671da177e4SLinus Torvalds * @nr_pages: The maximum number of pages 8681da177e4SLinus Torvalds * 8691da177e4SLinus Torvalds * pagevec_lookup() will search for and return a group of up to @nr_pages pages 8701da177e4SLinus Torvalds * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a 8711da177e4SLinus Torvalds * reference against the pages in @pvec. 8721da177e4SLinus Torvalds * 8731da177e4SLinus Torvalds * The search returns a group of mapping-contiguous pages with ascending 8741da177e4SLinus Torvalds * indexes. There may be holes in the indices due to not-present pages. 8751da177e4SLinus Torvalds * 8761da177e4SLinus Torvalds * pagevec_lookup() returns the number of pages which were found. 8771da177e4SLinus Torvalds */ 8781da177e4SLinus Torvalds unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping, 8791da177e4SLinus Torvalds pgoff_t start, unsigned nr_pages) 8801da177e4SLinus Torvalds { 8811da177e4SLinus Torvalds pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages); 8821da177e4SLinus Torvalds return pagevec_count(pvec); 8831da177e4SLinus Torvalds } 88478539fdfSChristoph Hellwig EXPORT_SYMBOL(pagevec_lookup); 88578539fdfSChristoph Hellwig 8861da177e4SLinus Torvalds unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping, 8871da177e4SLinus Torvalds pgoff_t *index, int tag, unsigned nr_pages) 8881da177e4SLinus Torvalds { 8891da177e4SLinus Torvalds pvec->nr = find_get_pages_tag(mapping, index, tag, 8901da177e4SLinus Torvalds nr_pages, pvec->pages); 8911da177e4SLinus Torvalds return pagevec_count(pvec); 8921da177e4SLinus Torvalds } 8937f285701SSteve French EXPORT_SYMBOL(pagevec_lookup_tag); 8941da177e4SLinus Torvalds 8951da177e4SLinus Torvalds /* 8961da177e4SLinus Torvalds * Perform any setup for the swap system 8971da177e4SLinus Torvalds */ 8981da177e4SLinus Torvalds void __init swap_setup(void) 8991da177e4SLinus Torvalds { 9004481374cSJan Beulich unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT); 901e0bf68ddSPeter Zijlstra #ifdef CONFIG_SWAP 90233806f06SShaohua Li int i; 90333806f06SShaohua Li 90433806f06SShaohua Li bdi_init(swapper_spaces[0].backing_dev_info); 90533806f06SShaohua Li for (i = 0; i < MAX_SWAPFILES; i++) { 90633806f06SShaohua Li spin_lock_init(&swapper_spaces[i].tree_lock); 90733806f06SShaohua Li INIT_LIST_HEAD(&swapper_spaces[i].i_mmap_nonlinear); 90833806f06SShaohua Li } 909e0bf68ddSPeter Zijlstra #endif 910e0bf68ddSPeter Zijlstra 9111da177e4SLinus Torvalds /* Use a smaller cluster for small-memory machines */ 9121da177e4SLinus Torvalds if (megs < 16) 9131da177e4SLinus Torvalds page_cluster = 2; 9141da177e4SLinus Torvalds else 9151da177e4SLinus Torvalds page_cluster = 3; 9161da177e4SLinus Torvalds /* 9171da177e4SLinus Torvalds * Right now other parts of the system means that we 9181da177e4SLinus Torvalds * _really_ don't want to cluster much more 9191da177e4SLinus Torvalds */ 9201da177e4SLinus Torvalds } 921