Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/mm/swap.c |
| 3 | * |
| 4 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds |
| 5 | */ |
| 6 | |
| 7 | /* |
Simon Arlott | 183ff22 | 2007-10-20 01:27:18 +0200 | [diff] [blame] | 8 | * This file contains the default values for the operation of the |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 9 | * Linux VM subsystem. Fine-tuning documentation can be found in |
| 10 | * Documentation/sysctl/vm.txt. |
| 11 | * Started 18.12.91 |
| 12 | * Swap aging added 23.2.95, Stephen Tweedie. |
| 13 | * Buffermem limits added 12.3.98, Rik van Riel. |
| 14 | */ |
| 15 | |
| 16 | #include <linux/mm.h> |
| 17 | #include <linux/sched.h> |
| 18 | #include <linux/kernel_stat.h> |
| 19 | #include <linux/swap.h> |
| 20 | #include <linux/mman.h> |
| 21 | #include <linux/pagemap.h> |
| 22 | #include <linux/pagevec.h> |
| 23 | #include <linux/init.h> |
Paul Gortmaker | b95f1b31 | 2011-10-16 02:01:52 -0400 | [diff] [blame] | 24 | #include <linux/export.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 25 | #include <linux/mm_inline.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 26 | #include <linux/percpu_counter.h> |
Dan Williams | 3565fce | 2016-01-15 16:56:55 -0800 | [diff] [blame] | 27 | #include <linux/memremap.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 28 | #include <linux/percpu.h> |
| 29 | #include <linux/cpu.h> |
| 30 | #include <linux/notifier.h> |
Peter Zijlstra | e0bf68d | 2007-10-16 23:25:46 -0700 | [diff] [blame] | 31 | #include <linux/backing-dev.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 32 | #include <linux/memcontrol.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 33 | #include <linux/gfp.h> |
Kent Overstreet | a27bb33 | 2013-05-07 16:19:08 -0700 | [diff] [blame] | 34 | #include <linux/uio.h> |
Naoya Horiguchi | 822fc61 | 2015-04-15 16:14:35 -0700 | [diff] [blame] | 35 | #include <linux/hugetlb.h> |
Vladimir Davydov | 33c3fc7 | 2015-09-09 15:35:45 -0700 | [diff] [blame] | 36 | #include <linux/page_idle.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 37 | |
Lee Schermerhorn | 64d6519 | 2008-10-18 20:26:52 -0700 | [diff] [blame] | 38 | #include "internal.h" |
| 39 | |
Mel Gorman | c6286c9 | 2013-07-03 15:02:26 -0700 | [diff] [blame] | 40 | #define CREATE_TRACE_POINTS |
| 41 | #include <trace/events/pagemap.h> |
| 42 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 43 | /* How many pages do we try to swap or page in/out together? */ |
| 44 | int page_cluster; |
| 45 | |
Mel Gorman | 13f7f78 | 2013-07-03 15:02:28 -0700 | [diff] [blame] | 46 | static DEFINE_PER_CPU(struct pagevec, lru_add_pvec); |
Vegard Nossum | f84f9504 | 2008-07-23 21:28:14 -0700 | [diff] [blame] | 47 | static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs); |
Minchan Kim | cc5993b | 2015-04-15 16:13:26 -0700 | [diff] [blame] | 48 | static DEFINE_PER_CPU(struct pagevec, lru_deactivate_file_pvecs); |
Minchan Kim | 10853a0 | 2016-01-15 16:55:11 -0800 | [diff] [blame] | 49 | static DEFINE_PER_CPU(struct pagevec, lru_deactivate_pvecs); |
Ming Li | a4a921a | 2016-05-20 16:57:56 -0700 | [diff] [blame] | 50 | #ifdef CONFIG_SMP |
| 51 | static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs); |
| 52 | #endif |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 53 | |
Adrian Bunk | b221385 | 2006-09-25 23:31:02 -0700 | [diff] [blame] | 54 | /* |
| 55 | * This path almost never happens for VM activity - pages are normally |
| 56 | * freed via pagevecs. But it gets used by networking. |
| 57 | */ |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 58 | static void __page_cache_release(struct page *page) |
Adrian Bunk | b221385 | 2006-09-25 23:31:02 -0700 | [diff] [blame] | 59 | { |
| 60 | if (PageLRU(page)) { |
Adrian Bunk | b221385 | 2006-09-25 23:31:02 -0700 | [diff] [blame] | 61 | struct zone *zone = page_zone(page); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 62 | struct lruvec *lruvec; |
| 63 | unsigned long flags; |
Adrian Bunk | b221385 | 2006-09-25 23:31:02 -0700 | [diff] [blame] | 64 | |
Mel Gorman | a52633d | 2016-07-28 15:45:28 -0700 | [diff] [blame] | 65 | spin_lock_irqsave(zone_lru_lock(zone), flags); |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 66 | lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat); |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 67 | VM_BUG_ON_PAGE(!PageLRU(page), page); |
Adrian Bunk | b221385 | 2006-09-25 23:31:02 -0700 | [diff] [blame] | 68 | __ClearPageLRU(page); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 69 | del_page_from_lru_list(page, lruvec, page_off_lru(page)); |
Mel Gorman | a52633d | 2016-07-28 15:45:28 -0700 | [diff] [blame] | 70 | spin_unlock_irqrestore(zone_lru_lock(zone), flags); |
Adrian Bunk | b221385 | 2006-09-25 23:31:02 -0700 | [diff] [blame] | 71 | } |
Nicholas Piggin | 49e19db | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 72 | __ClearPageWaiters(page); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 73 | mem_cgroup_uncharge(page); |
Andrea Arcangeli | 9180706 | 2011-01-13 15:46:32 -0800 | [diff] [blame] | 74 | } |
| 75 | |
| 76 | static void __put_single_page(struct page *page) |
| 77 | { |
| 78 | __page_cache_release(page); |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 79 | free_hot_cold_page(page, false); |
Adrian Bunk | b221385 | 2006-09-25 23:31:02 -0700 | [diff] [blame] | 80 | } |
| 81 | |
Andrea Arcangeli | 9180706 | 2011-01-13 15:46:32 -0800 | [diff] [blame] | 82 | static void __put_compound_page(struct page *page) |
| 83 | { |
| 84 | compound_page_dtor *dtor; |
| 85 | |
Naoya Horiguchi | 822fc61 | 2015-04-15 16:14:35 -0700 | [diff] [blame] | 86 | /* |
| 87 | * __page_cache_release() is supposed to be called for thp, not for |
| 88 | * hugetlb. This is because hugetlb page does never have PageLRU set |
| 89 | * (it's never listed to any LRU lists) and no memcg routines should |
| 90 | * be called for hugetlb (it has a separate hugetlb_cgroup.) |
| 91 | */ |
| 92 | if (!PageHuge(page)) |
| 93 | __page_cache_release(page); |
Andrea Arcangeli | 9180706 | 2011-01-13 15:46:32 -0800 | [diff] [blame] | 94 | dtor = get_compound_page_dtor(page); |
| 95 | (*dtor)(page); |
| 96 | } |
| 97 | |
Kirill A. Shutemov | ddc58f2 | 2016-01-15 16:52:56 -0800 | [diff] [blame] | 98 | void __put_page(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 99 | { |
Nick Piggin | 8519fb3 | 2006-02-07 12:58:52 -0800 | [diff] [blame] | 100 | if (unlikely(PageCompound(page))) |
Kirill A. Shutemov | ddc58f2 | 2016-01-15 16:52:56 -0800 | [diff] [blame] | 101 | __put_compound_page(page); |
| 102 | else |
Andrea Arcangeli | 9180706 | 2011-01-13 15:46:32 -0800 | [diff] [blame] | 103 | __put_single_page(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 104 | } |
Kirill A. Shutemov | ddc58f2 | 2016-01-15 16:52:56 -0800 | [diff] [blame] | 105 | EXPORT_SYMBOL(__put_page); |
Andrea Arcangeli | 70b50f9 | 2011-11-02 13:36:59 -0700 | [diff] [blame] | 106 | |
Alexander Zarochentsev | 1d7ea73 | 2006-08-13 23:24:27 -0700 | [diff] [blame] | 107 | /** |
Randy Dunlap | 7682486 | 2008-03-19 17:00:40 -0700 | [diff] [blame] | 108 | * put_pages_list() - release a list of pages |
| 109 | * @pages: list of pages threaded on page->lru |
Alexander Zarochentsev | 1d7ea73 | 2006-08-13 23:24:27 -0700 | [diff] [blame] | 110 | * |
| 111 | * Release a list of pages which are strung together on page.lru. Currently |
| 112 | * used by read_cache_pages() and related error recovery code. |
Alexander Zarochentsev | 1d7ea73 | 2006-08-13 23:24:27 -0700 | [diff] [blame] | 113 | */ |
| 114 | void put_pages_list(struct list_head *pages) |
| 115 | { |
| 116 | while (!list_empty(pages)) { |
| 117 | struct page *victim; |
| 118 | |
| 119 | victim = list_entry(pages->prev, struct page, lru); |
| 120 | list_del(&victim->lru); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 121 | put_page(victim); |
Alexander Zarochentsev | 1d7ea73 | 2006-08-13 23:24:27 -0700 | [diff] [blame] | 122 | } |
| 123 | } |
| 124 | EXPORT_SYMBOL(put_pages_list); |
| 125 | |
Mel Gorman | 18022c5 | 2012-07-31 16:44:51 -0700 | [diff] [blame] | 126 | /* |
| 127 | * get_kernel_pages() - pin kernel pages in memory |
| 128 | * @kiov: An array of struct kvec structures |
| 129 | * @nr_segs: number of segments to pin |
| 130 | * @write: pinning for read/write, currently ignored |
| 131 | * @pages: array that receives pointers to the pages pinned. |
| 132 | * Should be at least nr_segs long. |
| 133 | * |
| 134 | * Returns number of pages pinned. This may be fewer than the number |
| 135 | * requested. If nr_pages is 0 or negative, returns 0. If no pages |
| 136 | * were pinned, returns -errno. Each page returned must be released |
| 137 | * with a put_page() call when it is finished with. |
| 138 | */ |
| 139 | int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write, |
| 140 | struct page **pages) |
| 141 | { |
| 142 | int seg; |
| 143 | |
| 144 | for (seg = 0; seg < nr_segs; seg++) { |
| 145 | if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE)) |
| 146 | return seg; |
| 147 | |
Mel Gorman | 5a17811 | 2012-07-31 16:45:02 -0700 | [diff] [blame] | 148 | pages[seg] = kmap_to_page(kiov[seg].iov_base); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 149 | get_page(pages[seg]); |
Mel Gorman | 18022c5 | 2012-07-31 16:44:51 -0700 | [diff] [blame] | 150 | } |
| 151 | |
| 152 | return seg; |
| 153 | } |
| 154 | EXPORT_SYMBOL_GPL(get_kernel_pages); |
| 155 | |
| 156 | /* |
| 157 | * get_kernel_page() - pin a kernel page in memory |
| 158 | * @start: starting kernel address |
| 159 | * @write: pinning for read/write, currently ignored |
| 160 | * @pages: array that receives pointer to the page pinned. |
| 161 | * Must be at least nr_segs long. |
| 162 | * |
| 163 | * Returns 1 if page is pinned. If the page was not pinned, returns |
| 164 | * -errno. The page returned must be released with a put_page() call |
| 165 | * when it is finished with. |
| 166 | */ |
| 167 | int get_kernel_page(unsigned long start, int write, struct page **pages) |
| 168 | { |
| 169 | const struct kvec kiov = { |
| 170 | .iov_base = (void *)start, |
| 171 | .iov_len = PAGE_SIZE |
| 172 | }; |
| 173 | |
| 174 | return get_kernel_pages(&kiov, 1, write, pages); |
| 175 | } |
| 176 | EXPORT_SYMBOL_GPL(get_kernel_page); |
| 177 | |
Shaohua Li | 3dd7ae8 | 2011-03-22 16:33:45 -0700 | [diff] [blame] | 178 | static void pagevec_lru_move_fn(struct pagevec *pvec, |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 179 | void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg), |
| 180 | void *arg) |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 181 | { |
| 182 | int i; |
Mel Gorman | 68eb073 | 2016-07-28 15:47:11 -0700 | [diff] [blame] | 183 | struct pglist_data *pgdat = NULL; |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 184 | struct lruvec *lruvec; |
Shaohua Li | 3dd7ae8 | 2011-03-22 16:33:45 -0700 | [diff] [blame] | 185 | unsigned long flags = 0; |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 186 | |
| 187 | for (i = 0; i < pagevec_count(pvec); i++) { |
| 188 | struct page *page = pvec->pages[i]; |
Mel Gorman | 68eb073 | 2016-07-28 15:47:11 -0700 | [diff] [blame] | 189 | struct pglist_data *pagepgdat = page_pgdat(page); |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 190 | |
Mel Gorman | 68eb073 | 2016-07-28 15:47:11 -0700 | [diff] [blame] | 191 | if (pagepgdat != pgdat) { |
| 192 | if (pgdat) |
| 193 | spin_unlock_irqrestore(&pgdat->lru_lock, flags); |
| 194 | pgdat = pagepgdat; |
| 195 | spin_lock_irqsave(&pgdat->lru_lock, flags); |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 196 | } |
Shaohua Li | 3dd7ae8 | 2011-03-22 16:33:45 -0700 | [diff] [blame] | 197 | |
Mel Gorman | 68eb073 | 2016-07-28 15:47:11 -0700 | [diff] [blame] | 198 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 199 | (*move_fn)(page, lruvec, arg); |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 200 | } |
Mel Gorman | 68eb073 | 2016-07-28 15:47:11 -0700 | [diff] [blame] | 201 | if (pgdat) |
| 202 | spin_unlock_irqrestore(&pgdat->lru_lock, flags); |
Linus Torvalds | 83896fb | 2011-01-17 14:42:34 -0800 | [diff] [blame] | 203 | release_pages(pvec->pages, pvec->nr, pvec->cold); |
| 204 | pagevec_reinit(pvec); |
Shaohua Li | d8505de | 2011-01-13 15:47:33 -0800 | [diff] [blame] | 205 | } |
| 206 | |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 207 | static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec, |
| 208 | void *arg) |
Shaohua Li | 3dd7ae8 | 2011-03-22 16:33:45 -0700 | [diff] [blame] | 209 | { |
| 210 | int *pgmoved = arg; |
Shaohua Li | 3dd7ae8 | 2011-03-22 16:33:45 -0700 | [diff] [blame] | 211 | |
Johannes Weiner | 6b25f41 | 2017-02-24 14:56:23 -0800 | [diff] [blame] | 212 | if (PageLRU(page) && !PageUnevictable(page)) { |
| 213 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
| 214 | ClearPageActive(page); |
| 215 | add_page_to_lru_list_tail(page, lruvec, page_lru(page)); |
Shaohua Li | 3dd7ae8 | 2011-03-22 16:33:45 -0700 | [diff] [blame] | 216 | (*pgmoved)++; |
| 217 | } |
| 218 | } |
| 219 | |
| 220 | /* |
| 221 | * pagevec_move_tail() must be called with IRQ disabled. |
| 222 | * Otherwise this may cause nasty races. |
| 223 | */ |
| 224 | static void pagevec_move_tail(struct pagevec *pvec) |
| 225 | { |
| 226 | int pgmoved = 0; |
| 227 | |
| 228 | pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved); |
| 229 | __count_vm_events(PGROTATED, pgmoved); |
| 230 | } |
| 231 | |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 232 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 233 | * Writeback is about to end against a page which has been marked for immediate |
| 234 | * reclaim. If it still appears to be reclaimable, move it to the tail of the |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 235 | * inactive list. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 236 | */ |
Shaohua Li | 3dd7ae8 | 2011-03-22 16:33:45 -0700 | [diff] [blame] | 237 | void rotate_reclaimable_page(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 238 | { |
Johannes Weiner | 6b25f41 | 2017-02-24 14:56:23 -0800 | [diff] [blame] | 239 | if (!PageLocked(page) && !PageDirty(page) && |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 240 | !PageUnevictable(page) && PageLRU(page)) { |
Miklos Szeredi | ac6aadb | 2008-04-28 02:12:38 -0700 | [diff] [blame] | 241 | struct pagevec *pvec; |
| 242 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 243 | |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 244 | get_page(page); |
Miklos Szeredi | ac6aadb | 2008-04-28 02:12:38 -0700 | [diff] [blame] | 245 | local_irq_save(flags); |
Christoph Lameter | 7c8e018 | 2014-06-04 16:07:56 -0700 | [diff] [blame] | 246 | pvec = this_cpu_ptr(&lru_rotate_pvecs); |
Lukasz Odzioba | 8f18227 | 2016-06-24 14:50:01 -0700 | [diff] [blame] | 247 | if (!pagevec_add(pvec, page) || PageCompound(page)) |
Miklos Szeredi | ac6aadb | 2008-04-28 02:12:38 -0700 | [diff] [blame] | 248 | pagevec_move_tail(pvec); |
| 249 | local_irq_restore(flags); |
| 250 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 251 | } |
| 252 | |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 253 | static void update_page_reclaim_stat(struct lruvec *lruvec, |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 254 | int file, int rotated) |
| 255 | { |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 256 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 257 | |
| 258 | reclaim_stat->recent_scanned[file]++; |
| 259 | if (rotated) |
| 260 | reclaim_stat->recent_rotated[file]++; |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 261 | } |
| 262 | |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 263 | static void __activate_page(struct page *page, struct lruvec *lruvec, |
| 264 | void *arg) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 265 | { |
Linus Torvalds | 7a60857 | 2011-01-17 14:42:19 -0800 | [diff] [blame] | 266 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { |
| 267 | int file = page_is_file_cache(page); |
| 268 | int lru = page_lru_base_type(page); |
Linus Torvalds | 7a60857 | 2011-01-17 14:42:19 -0800 | [diff] [blame] | 269 | |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 270 | del_page_from_lru_list(page, lruvec, lru); |
Linus Torvalds | 7a60857 | 2011-01-17 14:42:19 -0800 | [diff] [blame] | 271 | SetPageActive(page); |
| 272 | lru += LRU_ACTIVE; |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 273 | add_page_to_lru_list(page, lruvec, lru); |
Mel Gorman | 24b7e58 | 2014-08-06 16:07:11 -0700 | [diff] [blame] | 274 | trace_mm_lru_activate(page); |
Linus Torvalds | 7a60857 | 2011-01-17 14:42:19 -0800 | [diff] [blame] | 275 | |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 276 | __count_vm_event(PGACTIVATE); |
| 277 | update_page_reclaim_stat(lruvec, file, 1); |
Linus Torvalds | 7a60857 | 2011-01-17 14:42:19 -0800 | [diff] [blame] | 278 | } |
Shaohua Li | eb709b0 | 2011-05-24 17:12:55 -0700 | [diff] [blame] | 279 | } |
| 280 | |
| 281 | #ifdef CONFIG_SMP |
Shaohua Li | eb709b0 | 2011-05-24 17:12:55 -0700 | [diff] [blame] | 282 | static void activate_page_drain(int cpu) |
| 283 | { |
| 284 | struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu); |
| 285 | |
| 286 | if (pagevec_count(pvec)) |
| 287 | pagevec_lru_move_fn(pvec, __activate_page, NULL); |
| 288 | } |
| 289 | |
Chris Metcalf | 5fbc461 | 2013-09-12 15:13:55 -0700 | [diff] [blame] | 290 | static bool need_activate_page_drain(int cpu) |
| 291 | { |
| 292 | return pagevec_count(&per_cpu(activate_page_pvecs, cpu)) != 0; |
| 293 | } |
| 294 | |
Shaohua Li | eb709b0 | 2011-05-24 17:12:55 -0700 | [diff] [blame] | 295 | void activate_page(struct page *page) |
| 296 | { |
Kirill A. Shutemov | 800d8c6 | 2016-07-26 15:26:18 -0700 | [diff] [blame] | 297 | page = compound_head(page); |
Shaohua Li | eb709b0 | 2011-05-24 17:12:55 -0700 | [diff] [blame] | 298 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { |
| 299 | struct pagevec *pvec = &get_cpu_var(activate_page_pvecs); |
| 300 | |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 301 | get_page(page); |
Lukasz Odzioba | 8f18227 | 2016-06-24 14:50:01 -0700 | [diff] [blame] | 302 | if (!pagevec_add(pvec, page) || PageCompound(page)) |
Shaohua Li | eb709b0 | 2011-05-24 17:12:55 -0700 | [diff] [blame] | 303 | pagevec_lru_move_fn(pvec, __activate_page, NULL); |
| 304 | put_cpu_var(activate_page_pvecs); |
| 305 | } |
| 306 | } |
| 307 | |
| 308 | #else |
| 309 | static inline void activate_page_drain(int cpu) |
| 310 | { |
| 311 | } |
| 312 | |
Chris Metcalf | 5fbc461 | 2013-09-12 15:13:55 -0700 | [diff] [blame] | 313 | static bool need_activate_page_drain(int cpu) |
| 314 | { |
| 315 | return false; |
| 316 | } |
| 317 | |
Shaohua Li | eb709b0 | 2011-05-24 17:12:55 -0700 | [diff] [blame] | 318 | void activate_page(struct page *page) |
| 319 | { |
| 320 | struct zone *zone = page_zone(page); |
| 321 | |
Kirill A. Shutemov | 800d8c6 | 2016-07-26 15:26:18 -0700 | [diff] [blame] | 322 | page = compound_head(page); |
Mel Gorman | a52633d | 2016-07-28 15:45:28 -0700 | [diff] [blame] | 323 | spin_lock_irq(zone_lru_lock(zone)); |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 324 | __activate_page(page, mem_cgroup_page_lruvec(page, zone->zone_pgdat), NULL); |
Mel Gorman | a52633d | 2016-07-28 15:45:28 -0700 | [diff] [blame] | 325 | spin_unlock_irq(zone_lru_lock(zone)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 326 | } |
Shaohua Li | eb709b0 | 2011-05-24 17:12:55 -0700 | [diff] [blame] | 327 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 328 | |
Mel Gorman | 059285a | 2013-07-03 15:02:30 -0700 | [diff] [blame] | 329 | static void __lru_cache_activate_page(struct page *page) |
| 330 | { |
| 331 | struct pagevec *pvec = &get_cpu_var(lru_add_pvec); |
| 332 | int i; |
| 333 | |
| 334 | /* |
| 335 | * Search backwards on the optimistic assumption that the page being |
| 336 | * activated has just been added to this pagevec. Note that only |
| 337 | * the local pagevec is examined as a !PageLRU page could be in the |
| 338 | * process of being released, reclaimed, migrated or on a remote |
| 339 | * pagevec that is currently being drained. Furthermore, marking |
| 340 | * a remote pagevec's page PageActive potentially hits a race where |
| 341 | * a page is marked PageActive just after it is added to the inactive |
| 342 | * list causing accounting errors and BUG_ON checks to trigger. |
| 343 | */ |
| 344 | for (i = pagevec_count(pvec) - 1; i >= 0; i--) { |
| 345 | struct page *pagevec_page = pvec->pages[i]; |
| 346 | |
| 347 | if (pagevec_page == page) { |
| 348 | SetPageActive(page); |
| 349 | break; |
| 350 | } |
| 351 | } |
| 352 | |
| 353 | put_cpu_var(lru_add_pvec); |
| 354 | } |
| 355 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 356 | /* |
| 357 | * Mark a page as having seen activity. |
| 358 | * |
| 359 | * inactive,unreferenced -> inactive,referenced |
| 360 | * inactive,referenced -> active,unreferenced |
| 361 | * active,unreferenced -> active,referenced |
Hugh Dickins | eb39d61 | 2014-08-06 16:06:43 -0700 | [diff] [blame] | 362 | * |
| 363 | * When a newly allocated page is not yet visible, so safe for non-atomic ops, |
| 364 | * __SetPageReferenced(page) may be substituted for mark_page_accessed(page). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 365 | */ |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 366 | void mark_page_accessed(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 367 | { |
Kirill A. Shutemov | e90309c | 2016-01-15 16:54:33 -0800 | [diff] [blame] | 368 | page = compound_head(page); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 369 | if (!PageActive(page) && !PageUnevictable(page) && |
Mel Gorman | 059285a | 2013-07-03 15:02:30 -0700 | [diff] [blame] | 370 | PageReferenced(page)) { |
| 371 | |
| 372 | /* |
| 373 | * If the page is on the LRU, queue it for activation via |
| 374 | * activate_page_pvecs. Otherwise, assume the page is on a |
| 375 | * pagevec, mark it active and it'll be moved to the active |
| 376 | * LRU on the next drain. |
| 377 | */ |
| 378 | if (PageLRU(page)) |
| 379 | activate_page(page); |
| 380 | else |
| 381 | __lru_cache_activate_page(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 382 | ClearPageReferenced(page); |
Johannes Weiner | a528910 | 2014-04-03 14:47:51 -0700 | [diff] [blame] | 383 | if (page_is_file_cache(page)) |
| 384 | workingset_activation(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 385 | } else if (!PageReferenced(page)) { |
| 386 | SetPageReferenced(page); |
| 387 | } |
Vladimir Davydov | 33c3fc7 | 2015-09-09 15:35:45 -0700 | [diff] [blame] | 388 | if (page_is_idle(page)) |
| 389 | clear_page_idle(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 390 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 391 | EXPORT_SYMBOL(mark_page_accessed); |
| 392 | |
Jianyu Zhan | 2329d37 | 2014-06-04 16:07:31 -0700 | [diff] [blame] | 393 | static void __lru_cache_add(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 394 | { |
Mel Gorman | 13f7f78 | 2013-07-03 15:02:28 -0700 | [diff] [blame] | 395 | struct pagevec *pvec = &get_cpu_var(lru_add_pvec); |
| 396 | |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 397 | get_page(page); |
Lukasz Odzioba | 8f18227 | 2016-06-24 14:50:01 -0700 | [diff] [blame] | 398 | if (!pagevec_add(pvec, page) || PageCompound(page)) |
Mel Gorman | a0b8cab3 | 2013-07-03 15:02:32 -0700 | [diff] [blame] | 399 | __pagevec_lru_add(pvec); |
Mel Gorman | 13f7f78 | 2013-07-03 15:02:28 -0700 | [diff] [blame] | 400 | put_cpu_var(lru_add_pvec); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 401 | } |
Jianyu Zhan | 2329d37 | 2014-06-04 16:07:31 -0700 | [diff] [blame] | 402 | |
| 403 | /** |
| 404 | * lru_cache_add: add a page to the page lists |
| 405 | * @page: the page to add |
| 406 | */ |
| 407 | void lru_cache_add_anon(struct page *page) |
| 408 | { |
Mel Gorman | 6fb81a1 | 2014-06-04 16:10:28 -0700 | [diff] [blame] | 409 | if (PageActive(page)) |
| 410 | ClearPageActive(page); |
Jianyu Zhan | 2329d37 | 2014-06-04 16:07:31 -0700 | [diff] [blame] | 411 | __lru_cache_add(page); |
| 412 | } |
| 413 | |
| 414 | void lru_cache_add_file(struct page *page) |
| 415 | { |
Mel Gorman | 6fb81a1 | 2014-06-04 16:10:28 -0700 | [diff] [blame] | 416 | if (PageActive(page)) |
| 417 | ClearPageActive(page); |
Jianyu Zhan | 2329d37 | 2014-06-04 16:07:31 -0700 | [diff] [blame] | 418 | __lru_cache_add(page); |
| 419 | } |
| 420 | EXPORT_SYMBOL(lru_cache_add_file); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 421 | |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 422 | /** |
Mel Gorman | c53954a | 2013-07-03 15:02:34 -0700 | [diff] [blame] | 423 | * lru_cache_add - add a page to a page list |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 424 | * @page: the page to be added to the LRU. |
Jianyu Zhan | 2329d37 | 2014-06-04 16:07:31 -0700 | [diff] [blame] | 425 | * |
| 426 | * Queue the page for addition to the LRU via pagevec. The decision on whether |
| 427 | * to add the page to the [in]active [file|anon] list is deferred until the |
| 428 | * pagevec is drained. This gives a chance for the caller of lru_cache_add() |
| 429 | * have the page added to the active list using mark_page_accessed(). |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 430 | */ |
Mel Gorman | c53954a | 2013-07-03 15:02:34 -0700 | [diff] [blame] | 431 | void lru_cache_add(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 432 | { |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 433 | VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page); |
| 434 | VM_BUG_ON_PAGE(PageLRU(page), page); |
Mel Gorman | c53954a | 2013-07-03 15:02:34 -0700 | [diff] [blame] | 435 | __lru_cache_add(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 436 | } |
| 437 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 438 | /** |
| 439 | * add_page_to_unevictable_list - add a page to the unevictable list |
| 440 | * @page: the page to be added to the unevictable list |
| 441 | * |
| 442 | * Add page directly to its zone's unevictable list. To avoid races with |
| 443 | * tasks that might be making the page evictable, through eg. munlock, |
| 444 | * munmap or exit, while it's not on the lru, we want to add the page |
| 445 | * while it's locked or otherwise "invisible" to other tasks. This is |
| 446 | * difficult to do when using the pagevec cache, so bypass that. |
| 447 | */ |
| 448 | void add_page_to_unevictable_list(struct page *page) |
| 449 | { |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 450 | struct pglist_data *pgdat = page_pgdat(page); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 451 | struct lruvec *lruvec; |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 452 | |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 453 | spin_lock_irq(&pgdat->lru_lock); |
| 454 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
Naoya Horiguchi | ef2a2cb | 2013-07-31 13:53:37 -0700 | [diff] [blame] | 455 | ClearPageActive(page); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 456 | SetPageUnevictable(page); |
| 457 | SetPageLRU(page); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 458 | add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE); |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 459 | spin_unlock_irq(&pgdat->lru_lock); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 460 | } |
| 461 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 462 | /** |
| 463 | * lru_cache_add_active_or_unevictable |
| 464 | * @page: the page to be added to LRU |
| 465 | * @vma: vma in which page is mapped for determining reclaimability |
| 466 | * |
| 467 | * Place @page on the active or unevictable LRU list, depending on its |
| 468 | * evictability. Note that if the page is not evictable, it goes |
| 469 | * directly back onto it's zone's unevictable list, it does NOT use a |
| 470 | * per cpu pagevec. |
| 471 | */ |
Laurent Dufour | 62cd1d9 | 2018-04-17 16:33:20 +0200 | [diff] [blame] | 472 | void __lru_cache_add_active_or_unevictable(struct page *page, |
| 473 | unsigned long vma_flags) |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 474 | { |
| 475 | VM_BUG_ON_PAGE(PageLRU(page), page); |
| 476 | |
Laurent Dufour | 62cd1d9 | 2018-04-17 16:33:20 +0200 | [diff] [blame] | 477 | if (likely((vma_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED)) { |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 478 | SetPageActive(page); |
| 479 | lru_cache_add(page); |
| 480 | return; |
| 481 | } |
| 482 | |
| 483 | if (!TestSetPageMlocked(page)) { |
| 484 | /* |
| 485 | * We use the irq-unsafe __mod_zone_page_stat because this |
| 486 | * counter is not modified from interrupt context, and the pte |
| 487 | * lock is held(spinlock), which implies preemption disabled. |
| 488 | */ |
| 489 | __mod_zone_page_state(page_zone(page), NR_MLOCK, |
| 490 | hpage_nr_pages(page)); |
| 491 | count_vm_event(UNEVICTABLE_PGMLOCKED); |
| 492 | } |
| 493 | add_page_to_unevictable_list(page); |
| 494 | } |
| 495 | |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 496 | /* |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 497 | * If the page can not be invalidated, it is moved to the |
| 498 | * inactive list to speed up its reclaim. It is moved to the |
| 499 | * head of the list, rather than the tail, to give the flusher |
| 500 | * threads some time to write it out, as this is much more |
| 501 | * effective than the single-page writeout from reclaim. |
Minchan Kim | 278df9f | 2011-03-22 16:32:54 -0700 | [diff] [blame] | 502 | * |
| 503 | * If the page isn't page_mapped and dirty/writeback, the page |
| 504 | * could reclaim asap using PG_reclaim. |
| 505 | * |
| 506 | * 1. active, mapped page -> none |
| 507 | * 2. active, dirty/writeback page -> inactive, head, PG_reclaim |
| 508 | * 3. inactive, mapped page -> none |
| 509 | * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim |
| 510 | * 5. inactive, clean -> inactive, tail |
| 511 | * 6. Others -> none |
| 512 | * |
| 513 | * In 4, why it moves inactive's head, the VM expects the page would |
| 514 | * be write it out by flusher threads as this is much more effective |
| 515 | * than the single-page writeout from reclaim. |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 516 | */ |
Minchan Kim | cc5993b | 2015-04-15 16:13:26 -0700 | [diff] [blame] | 517 | static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec, |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 518 | void *arg) |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 519 | { |
| 520 | int lru, file; |
Minchan Kim | 278df9f | 2011-03-22 16:32:54 -0700 | [diff] [blame] | 521 | bool active; |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 522 | |
Minchan Kim | 278df9f | 2011-03-22 16:32:54 -0700 | [diff] [blame] | 523 | if (!PageLRU(page)) |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 524 | return; |
| 525 | |
Minchan Kim | bad49d9 | 2011-05-11 15:13:30 -0700 | [diff] [blame] | 526 | if (PageUnevictable(page)) |
| 527 | return; |
| 528 | |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 529 | /* Some processes are using the page */ |
| 530 | if (page_mapped(page)) |
| 531 | return; |
| 532 | |
Minchan Kim | 278df9f | 2011-03-22 16:32:54 -0700 | [diff] [blame] | 533 | active = PageActive(page); |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 534 | file = page_is_file_cache(page); |
| 535 | lru = page_lru_base_type(page); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 536 | |
| 537 | del_page_from_lru_list(page, lruvec, lru + active); |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 538 | ClearPageActive(page); |
| 539 | ClearPageReferenced(page); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 540 | add_page_to_lru_list(page, lruvec, lru); |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 541 | |
Minchan Kim | 278df9f | 2011-03-22 16:32:54 -0700 | [diff] [blame] | 542 | if (PageWriteback(page) || PageDirty(page)) { |
| 543 | /* |
| 544 | * PG_reclaim could be raced with end_page_writeback |
| 545 | * It can make readahead confusing. But race window |
| 546 | * is _really_ small and it's non-critical problem. |
| 547 | */ |
| 548 | SetPageReclaim(page); |
| 549 | } else { |
| 550 | /* |
| 551 | * The page's writeback ends up during pagevec |
| 552 | * We moves tha page into tail of inactive. |
| 553 | */ |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 554 | list_move_tail(&page->lru, &lruvec->lists[lru]); |
Minchan Kim | 278df9f | 2011-03-22 16:32:54 -0700 | [diff] [blame] | 555 | __count_vm_event(PGROTATED); |
| 556 | } |
| 557 | |
| 558 | if (active) |
| 559 | __count_vm_event(PGDEACTIVATE); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 560 | update_page_reclaim_stat(lruvec, file, 0); |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 561 | } |
| 562 | |
Minchan Kim | 10853a0 | 2016-01-15 16:55:11 -0800 | [diff] [blame] | 563 | |
| 564 | static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec, |
| 565 | void *arg) |
| 566 | { |
| 567 | if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) { |
| 568 | int file = page_is_file_cache(page); |
| 569 | int lru = page_lru_base_type(page); |
| 570 | |
| 571 | del_page_from_lru_list(page, lruvec, lru + LRU_ACTIVE); |
| 572 | ClearPageActive(page); |
| 573 | ClearPageReferenced(page); |
| 574 | add_page_to_lru_list(page, lruvec, lru); |
| 575 | |
| 576 | __count_vm_event(PGDEACTIVATE); |
| 577 | update_page_reclaim_stat(lruvec, file, 0); |
| 578 | } |
| 579 | } |
| 580 | |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 581 | /* |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 582 | * Drain pages out of the cpu's pagevecs. |
| 583 | * Either "cpu" is the current CPU, and preemption has already been |
| 584 | * disabled; or "cpu" is being hot-unplugged, and is already dead. |
| 585 | */ |
Konstantin Khlebnikov | f0cb3c7 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 586 | void lru_add_drain_cpu(int cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 587 | { |
Mel Gorman | 13f7f78 | 2013-07-03 15:02:28 -0700 | [diff] [blame] | 588 | struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 589 | |
Mel Gorman | 13f7f78 | 2013-07-03 15:02:28 -0700 | [diff] [blame] | 590 | if (pagevec_count(pvec)) |
Mel Gorman | a0b8cab3 | 2013-07-03 15:02:32 -0700 | [diff] [blame] | 591 | __pagevec_lru_add(pvec); |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 592 | |
| 593 | pvec = &per_cpu(lru_rotate_pvecs, cpu); |
| 594 | if (pagevec_count(pvec)) { |
| 595 | unsigned long flags; |
| 596 | |
| 597 | /* No harm done if a racing interrupt already did this */ |
| 598 | local_irq_save(flags); |
| 599 | pagevec_move_tail(pvec); |
| 600 | local_irq_restore(flags); |
| 601 | } |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 602 | |
Minchan Kim | cc5993b | 2015-04-15 16:13:26 -0700 | [diff] [blame] | 603 | pvec = &per_cpu(lru_deactivate_file_pvecs, cpu); |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 604 | if (pagevec_count(pvec)) |
Minchan Kim | cc5993b | 2015-04-15 16:13:26 -0700 | [diff] [blame] | 605 | pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL); |
Shaohua Li | eb709b0 | 2011-05-24 17:12:55 -0700 | [diff] [blame] | 606 | |
Minchan Kim | 10853a0 | 2016-01-15 16:55:11 -0800 | [diff] [blame] | 607 | pvec = &per_cpu(lru_deactivate_pvecs, cpu); |
| 608 | if (pagevec_count(pvec)) |
| 609 | pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL); |
| 610 | |
Shaohua Li | eb709b0 | 2011-05-24 17:12:55 -0700 | [diff] [blame] | 611 | activate_page_drain(cpu); |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 612 | } |
| 613 | |
| 614 | /** |
Minchan Kim | cc5993b | 2015-04-15 16:13:26 -0700 | [diff] [blame] | 615 | * deactivate_file_page - forcefully deactivate a file page |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 616 | * @page: page to deactivate |
| 617 | * |
| 618 | * This function hints the VM that @page is a good reclaim candidate, |
| 619 | * for example if its invalidation fails due to the page being dirty |
| 620 | * or under writeback. |
| 621 | */ |
Minchan Kim | cc5993b | 2015-04-15 16:13:26 -0700 | [diff] [blame] | 622 | void deactivate_file_page(struct page *page) |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 623 | { |
Minchan Kim | 821ed6b | 2011-05-24 17:12:31 -0700 | [diff] [blame] | 624 | /* |
Minchan Kim | cc5993b | 2015-04-15 16:13:26 -0700 | [diff] [blame] | 625 | * In a workload with many unevictable page such as mprotect, |
| 626 | * unevictable page deactivation for accelerating reclaim is pointless. |
Minchan Kim | 821ed6b | 2011-05-24 17:12:31 -0700 | [diff] [blame] | 627 | */ |
| 628 | if (PageUnevictable(page)) |
| 629 | return; |
| 630 | |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 631 | if (likely(get_page_unless_zero(page))) { |
Minchan Kim | cc5993b | 2015-04-15 16:13:26 -0700 | [diff] [blame] | 632 | struct pagevec *pvec = &get_cpu_var(lru_deactivate_file_pvecs); |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 633 | |
Lukasz Odzioba | 8f18227 | 2016-06-24 14:50:01 -0700 | [diff] [blame] | 634 | if (!pagevec_add(pvec, page) || PageCompound(page)) |
Minchan Kim | cc5993b | 2015-04-15 16:13:26 -0700 | [diff] [blame] | 635 | pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL); |
| 636 | put_cpu_var(lru_deactivate_file_pvecs); |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 637 | } |
Andrew Morton | 80bfed9 | 2006-01-06 00:11:14 -0800 | [diff] [blame] | 638 | } |
| 639 | |
Minchan Kim | 10853a0 | 2016-01-15 16:55:11 -0800 | [diff] [blame] | 640 | /** |
| 641 | * deactivate_page - deactivate a page |
| 642 | * @page: page to deactivate |
| 643 | * |
| 644 | * deactivate_page() moves @page to the inactive list if @page was on the active |
| 645 | * list and was not an unevictable page. This is done to accelerate the reclaim |
| 646 | * of @page. |
| 647 | */ |
| 648 | void deactivate_page(struct page *page) |
| 649 | { |
| 650 | if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) { |
| 651 | struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs); |
| 652 | |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 653 | get_page(page); |
Lukasz Odzioba | 8f18227 | 2016-06-24 14:50:01 -0700 | [diff] [blame] | 654 | if (!pagevec_add(pvec, page) || PageCompound(page)) |
Minchan Kim | 10853a0 | 2016-01-15 16:55:11 -0800 | [diff] [blame] | 655 | pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL); |
| 656 | put_cpu_var(lru_deactivate_pvecs); |
| 657 | } |
| 658 | } |
| 659 | |
Andrew Morton | 80bfed9 | 2006-01-06 00:11:14 -0800 | [diff] [blame] | 660 | void lru_add_drain(void) |
| 661 | { |
Konstantin Khlebnikov | f0cb3c7 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 662 | lru_add_drain_cpu(get_cpu()); |
Andrew Morton | 80bfed9 | 2006-01-06 00:11:14 -0800 | [diff] [blame] | 663 | put_cpu(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 664 | } |
| 665 | |
David Howells | c402895 | 2006-11-22 14:57:56 +0000 | [diff] [blame] | 666 | static void lru_add_drain_per_cpu(struct work_struct *dummy) |
Nick Piggin | 053837f | 2006-01-18 17:42:27 -0800 | [diff] [blame] | 667 | { |
| 668 | lru_add_drain(); |
| 669 | } |
| 670 | |
Chris Metcalf | 5fbc461 | 2013-09-12 15:13:55 -0700 | [diff] [blame] | 671 | static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work); |
| 672 | |
Wang Sheng-Hui | f3a932ba | 2016-06-08 15:33:56 -0700 | [diff] [blame] | 673 | /* |
| 674 | * lru_add_drain_wq is used to do lru_add_drain_all() from a WQ_MEM_RECLAIM |
| 675 | * workqueue, aiding in getting memory freed. |
| 676 | */ |
| 677 | static struct workqueue_struct *lru_add_drain_wq; |
| 678 | |
| 679 | static int __init lru_init(void) |
| 680 | { |
| 681 | lru_add_drain_wq = alloc_workqueue("lru-add-drain", WQ_MEM_RECLAIM, 0); |
| 682 | |
| 683 | if (WARN(!lru_add_drain_wq, |
| 684 | "Failed to create workqueue lru_add_drain_wq")) |
| 685 | return -ENOMEM; |
| 686 | |
| 687 | return 0; |
| 688 | } |
| 689 | early_initcall(lru_init); |
| 690 | |
Chris Metcalf | 5fbc461 | 2013-09-12 15:13:55 -0700 | [diff] [blame] | 691 | void lru_add_drain_all(void) |
Nick Piggin | 053837f | 2006-01-18 17:42:27 -0800 | [diff] [blame] | 692 | { |
Chris Metcalf | 5fbc461 | 2013-09-12 15:13:55 -0700 | [diff] [blame] | 693 | static DEFINE_MUTEX(lock); |
| 694 | static struct cpumask has_work; |
| 695 | int cpu; |
| 696 | |
| 697 | mutex_lock(&lock); |
| 698 | get_online_cpus(); |
| 699 | cpumask_clear(&has_work); |
| 700 | |
| 701 | for_each_online_cpu(cpu) { |
| 702 | struct work_struct *work = &per_cpu(lru_add_drain_work, cpu); |
| 703 | |
| 704 | if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) || |
| 705 | pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) || |
Minchan Kim | cc5993b | 2015-04-15 16:13:26 -0700 | [diff] [blame] | 706 | pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) || |
Minchan Kim | 10853a0 | 2016-01-15 16:55:11 -0800 | [diff] [blame] | 707 | pagevec_count(&per_cpu(lru_deactivate_pvecs, cpu)) || |
Chris Metcalf | 5fbc461 | 2013-09-12 15:13:55 -0700 | [diff] [blame] | 708 | need_activate_page_drain(cpu)) { |
| 709 | INIT_WORK(work, lru_add_drain_per_cpu); |
Wang Sheng-Hui | f3a932ba | 2016-06-08 15:33:56 -0700 | [diff] [blame] | 710 | queue_work_on(cpu, lru_add_drain_wq, work); |
Chris Metcalf | 5fbc461 | 2013-09-12 15:13:55 -0700 | [diff] [blame] | 711 | cpumask_set_cpu(cpu, &has_work); |
| 712 | } |
| 713 | } |
| 714 | |
| 715 | for_each_cpu(cpu, &has_work) |
| 716 | flush_work(&per_cpu(lru_add_drain_work, cpu)); |
| 717 | |
| 718 | put_online_cpus(); |
| 719 | mutex_unlock(&lock); |
Nick Piggin | 053837f | 2006-01-18 17:42:27 -0800 | [diff] [blame] | 720 | } |
| 721 | |
Michal Hocko | aabfb57 | 2014-10-09 15:28:52 -0700 | [diff] [blame] | 722 | /** |
Kirill A. Shutemov | ea1754a | 2016-04-01 15:29:48 +0300 | [diff] [blame] | 723 | * release_pages - batched put_page() |
Michal Hocko | aabfb57 | 2014-10-09 15:28:52 -0700 | [diff] [blame] | 724 | * @pages: array of pages to release |
| 725 | * @nr: number of pages |
| 726 | * @cold: whether the pages are cache cold |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 727 | * |
Michal Hocko | aabfb57 | 2014-10-09 15:28:52 -0700 | [diff] [blame] | 728 | * Decrement the reference count on all the pages in @pages. If it |
| 729 | * fell to zero, remove the page from the LRU and free it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 730 | */ |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 731 | void release_pages(struct page **pages, int nr, bool cold) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 732 | { |
| 733 | int i; |
Konstantin Khlebnikov | cc59850 | 2012-01-10 15:07:04 -0800 | [diff] [blame] | 734 | LIST_HEAD(pages_to_free); |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 735 | struct pglist_data *locked_pgdat = NULL; |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 736 | struct lruvec *lruvec; |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 737 | unsigned long uninitialized_var(flags); |
Michal Hocko | aabfb57 | 2014-10-09 15:28:52 -0700 | [diff] [blame] | 738 | unsigned int uninitialized_var(lock_batch); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 739 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 740 | for (i = 0; i < nr; i++) { |
| 741 | struct page *page = pages[i]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 742 | |
Michal Hocko | aabfb57 | 2014-10-09 15:28:52 -0700 | [diff] [blame] | 743 | /* |
| 744 | * Make sure the IRQ-safe lock-holding time does not get |
| 745 | * excessive with a continuous string of pages from the |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 746 | * same pgdat. The lock is held only if pgdat != NULL. |
Michal Hocko | aabfb57 | 2014-10-09 15:28:52 -0700 | [diff] [blame] | 747 | */ |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 748 | if (locked_pgdat && ++lock_batch == SWAP_CLUSTER_MAX) { |
| 749 | spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags); |
| 750 | locked_pgdat = NULL; |
Michal Hocko | aabfb57 | 2014-10-09 15:28:52 -0700 | [diff] [blame] | 751 | } |
| 752 | |
Aaron Lu | 6fcb52a | 2016-10-07 17:00:08 -0700 | [diff] [blame] | 753 | if (is_huge_zero_page(page)) |
Kirill A. Shutemov | aa88b68 | 2016-04-28 16:18:27 -0700 | [diff] [blame] | 754 | continue; |
Kirill A. Shutemov | aa88b68 | 2016-04-28 16:18:27 -0700 | [diff] [blame] | 755 | |
Kirill A. Shutemov | ddc58f2 | 2016-01-15 16:52:56 -0800 | [diff] [blame] | 756 | page = compound_head(page); |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 757 | if (!put_page_testzero(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 758 | continue; |
| 759 | |
Kirill A. Shutemov | ddc58f2 | 2016-01-15 16:52:56 -0800 | [diff] [blame] | 760 | if (PageCompound(page)) { |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 761 | if (locked_pgdat) { |
| 762 | spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags); |
| 763 | locked_pgdat = NULL; |
Kirill A. Shutemov | ddc58f2 | 2016-01-15 16:52:56 -0800 | [diff] [blame] | 764 | } |
| 765 | __put_compound_page(page); |
| 766 | continue; |
| 767 | } |
| 768 | |
Nick Piggin | 46453a6 | 2006-03-22 00:07:58 -0800 | [diff] [blame] | 769 | if (PageLRU(page)) { |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 770 | struct pglist_data *pgdat = page_pgdat(page); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 771 | |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 772 | if (pgdat != locked_pgdat) { |
| 773 | if (locked_pgdat) |
| 774 | spin_unlock_irqrestore(&locked_pgdat->lru_lock, |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 775 | flags); |
Michal Hocko | aabfb57 | 2014-10-09 15:28:52 -0700 | [diff] [blame] | 776 | lock_batch = 0; |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 777 | locked_pgdat = pgdat; |
| 778 | spin_lock_irqsave(&locked_pgdat->lru_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 779 | } |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 780 | |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 781 | lruvec = mem_cgroup_page_lruvec(page, locked_pgdat); |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 782 | VM_BUG_ON_PAGE(!PageLRU(page), page); |
Nick Piggin | 6745391 | 2006-03-22 00:08:00 -0800 | [diff] [blame] | 783 | __ClearPageLRU(page); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 784 | del_page_from_lru_list(page, lruvec, page_off_lru(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 785 | } |
Nick Piggin | 46453a6 | 2006-03-22 00:07:58 -0800 | [diff] [blame] | 786 | |
Mel Gorman | c53954a | 2013-07-03 15:02:34 -0700 | [diff] [blame] | 787 | /* Clear Active bit in case of parallel mark_page_accessed */ |
Mel Gorman | e3741b5 | 2014-06-04 16:10:26 -0700 | [diff] [blame] | 788 | __ClearPageActive(page); |
Nicholas Piggin | 49e19db | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 789 | __ClearPageWaiters(page); |
Mel Gorman | c53954a | 2013-07-03 15:02:34 -0700 | [diff] [blame] | 790 | |
Konstantin Khlebnikov | cc59850 | 2012-01-10 15:07:04 -0800 | [diff] [blame] | 791 | list_add(&page->lru, &pages_to_free); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 792 | } |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 793 | if (locked_pgdat) |
| 794 | spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 795 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 796 | mem_cgroup_uncharge_list(&pages_to_free); |
Konstantin Khlebnikov | cc59850 | 2012-01-10 15:07:04 -0800 | [diff] [blame] | 797 | free_hot_cold_page_list(&pages_to_free, cold); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 798 | } |
Miklos Szeredi | 0be8557 | 2010-10-27 15:34:46 -0700 | [diff] [blame] | 799 | EXPORT_SYMBOL(release_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 800 | |
| 801 | /* |
| 802 | * The pages which we're about to release may be in the deferred lru-addition |
| 803 | * queues. That would prevent them from really being freed right now. That's |
| 804 | * OK from a correctness point of view but is inefficient - those pages may be |
| 805 | * cache-warm and we want to give them back to the page allocator ASAP. |
| 806 | * |
| 807 | * So __pagevec_release() will drain those queues here. __pagevec_lru_add() |
| 808 | * and __pagevec_lru_add_active() call release_pages() directly to avoid |
| 809 | * mutual recursion. |
| 810 | */ |
| 811 | void __pagevec_release(struct pagevec *pvec) |
| 812 | { |
| 813 | lru_add_drain(); |
| 814 | release_pages(pvec->pages, pagevec_count(pvec), pvec->cold); |
| 815 | pagevec_reinit(pvec); |
| 816 | } |
Steve French | 7f28570 | 2005-11-01 10:22:55 -0800 | [diff] [blame] | 817 | EXPORT_SYMBOL(__pagevec_release); |
| 818 | |
Hugh Dickins | 12d2710 | 2012-01-12 17:19:52 -0800 | [diff] [blame] | 819 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 820 | /* used by __split_huge_page_refcount() */ |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 821 | void lru_add_page_tail(struct page *page, struct page *page_tail, |
Shaohua Li | 5bc7b8a | 2013-04-29 15:08:36 -0700 | [diff] [blame] | 822 | struct lruvec *lruvec, struct list_head *list) |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 823 | { |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 824 | const int file = 0; |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 825 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 826 | VM_BUG_ON_PAGE(!PageHead(page), page); |
| 827 | VM_BUG_ON_PAGE(PageCompound(page_tail), page); |
| 828 | VM_BUG_ON_PAGE(PageLRU(page_tail), page); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 829 | VM_BUG_ON(NR_CPUS != 1 && |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 830 | !spin_is_locked(&lruvec_pgdat(lruvec)->lru_lock)); |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 831 | |
Shaohua Li | 5bc7b8a | 2013-04-29 15:08:36 -0700 | [diff] [blame] | 832 | if (!list) |
| 833 | SetPageLRU(page_tail); |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 834 | |
Hugh Dickins | 12d2710 | 2012-01-12 17:19:52 -0800 | [diff] [blame] | 835 | if (likely(PageLRU(page))) |
| 836 | list_add_tail(&page_tail->lru, &page->lru); |
Shaohua Li | 5bc7b8a | 2013-04-29 15:08:36 -0700 | [diff] [blame] | 837 | else if (list) { |
| 838 | /* page reclaim is reclaiming a huge page */ |
| 839 | get_page(page_tail); |
| 840 | list_add_tail(&page_tail->lru, list); |
| 841 | } else { |
Hugh Dickins | 12d2710 | 2012-01-12 17:19:52 -0800 | [diff] [blame] | 842 | struct list_head *list_head; |
| 843 | /* |
| 844 | * Head page has not yet been counted, as an hpage, |
| 845 | * so we must account for each subpage individually. |
| 846 | * |
| 847 | * Use the standard add function to put page_tail on the list, |
| 848 | * but then correct its position so they all end up in order. |
| 849 | */ |
Kirill A. Shutemov | e180cf8 | 2013-07-31 13:53:39 -0700 | [diff] [blame] | 850 | add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail)); |
Hugh Dickins | 12d2710 | 2012-01-12 17:19:52 -0800 | [diff] [blame] | 851 | list_head = page_tail->lru.prev; |
| 852 | list_move_tail(&page_tail->lru, list_head); |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 853 | } |
Hugh Dickins | 7512102 | 2012-03-05 14:59:18 -0800 | [diff] [blame] | 854 | |
| 855 | if (!PageUnevictable(page)) |
Kirill A. Shutemov | e180cf8 | 2013-07-31 13:53:39 -0700 | [diff] [blame] | 856 | update_page_reclaim_stat(lruvec, file, PageActive(page_tail)); |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 857 | } |
Hugh Dickins | 12d2710 | 2012-01-12 17:19:52 -0800 | [diff] [blame] | 858 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 859 | |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 860 | static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec, |
| 861 | void *arg) |
Shaohua Li | 3dd7ae8 | 2011-03-22 16:33:45 -0700 | [diff] [blame] | 862 | { |
Mel Gorman | 13f7f78 | 2013-07-03 15:02:28 -0700 | [diff] [blame] | 863 | int file = page_is_file_cache(page); |
| 864 | int active = PageActive(page); |
| 865 | enum lru_list lru = page_lru(page); |
Shaohua Li | 3dd7ae8 | 2011-03-22 16:33:45 -0700 | [diff] [blame] | 866 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 867 | VM_BUG_ON_PAGE(PageLRU(page), page); |
Shaohua Li | 3dd7ae8 | 2011-03-22 16:33:45 -0700 | [diff] [blame] | 868 | |
| 869 | SetPageLRU(page); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 870 | add_page_to_lru_list(page, lruvec, lru); |
| 871 | update_page_reclaim_stat(lruvec, file, active); |
Mel Gorman | 24b7e58 | 2014-08-06 16:07:11 -0700 | [diff] [blame] | 872 | trace_mm_lru_insertion(page, lru); |
Shaohua Li | 3dd7ae8 | 2011-03-22 16:33:45 -0700 | [diff] [blame] | 873 | } |
| 874 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 875 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 876 | * Add the passed pages to the LRU, then drop the caller's refcount |
| 877 | * on them. Reinitialises the caller's pagevec. |
| 878 | */ |
Mel Gorman | a0b8cab3 | 2013-07-03 15:02:32 -0700 | [diff] [blame] | 879 | void __pagevec_lru_add(struct pagevec *pvec) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 880 | { |
Mel Gorman | a0b8cab3 | 2013-07-03 15:02:32 -0700 | [diff] [blame] | 881 | pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 882 | } |
Hugh Dickins | 5095ae83 | 2012-01-12 17:19:58 -0800 | [diff] [blame] | 883 | EXPORT_SYMBOL(__pagevec_lru_add); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 884 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 885 | /** |
Johannes Weiner | 0cd6144 | 2014-04-03 14:47:46 -0700 | [diff] [blame] | 886 | * pagevec_lookup_entries - gang pagecache lookup |
| 887 | * @pvec: Where the resulting entries are placed |
| 888 | * @mapping: The address_space to search |
| 889 | * @start: The starting entry index |
| 890 | * @nr_entries: The maximum number of entries |
| 891 | * @indices: The cache indices corresponding to the entries in @pvec |
| 892 | * |
| 893 | * pagevec_lookup_entries() will search for and return a group of up |
| 894 | * to @nr_entries pages and shadow entries in the mapping. All |
| 895 | * entries are placed in @pvec. pagevec_lookup_entries() takes a |
| 896 | * reference against actual pages in @pvec. |
| 897 | * |
| 898 | * The search returns a group of mapping-contiguous entries with |
| 899 | * ascending indexes. There may be holes in the indices due to |
| 900 | * not-present entries. |
| 901 | * |
| 902 | * pagevec_lookup_entries() returns the number of entries which were |
| 903 | * found. |
| 904 | */ |
| 905 | unsigned pagevec_lookup_entries(struct pagevec *pvec, |
| 906 | struct address_space *mapping, |
| 907 | pgoff_t start, unsigned nr_pages, |
| 908 | pgoff_t *indices) |
| 909 | { |
| 910 | pvec->nr = find_get_entries(mapping, start, nr_pages, |
| 911 | pvec->pages, indices); |
| 912 | return pagevec_count(pvec); |
| 913 | } |
| 914 | |
| 915 | /** |
| 916 | * pagevec_remove_exceptionals - pagevec exceptionals pruning |
| 917 | * @pvec: The pagevec to prune |
| 918 | * |
| 919 | * pagevec_lookup_entries() fills both pages and exceptional radix |
| 920 | * tree entries into the pagevec. This function prunes all |
| 921 | * exceptionals from @pvec without leaving holes, so that it can be |
| 922 | * passed on to page-only pagevec operations. |
| 923 | */ |
| 924 | void pagevec_remove_exceptionals(struct pagevec *pvec) |
| 925 | { |
| 926 | int i, j; |
| 927 | |
| 928 | for (i = 0, j = 0; i < pagevec_count(pvec); i++) { |
| 929 | struct page *page = pvec->pages[i]; |
| 930 | if (!radix_tree_exceptional_entry(page)) |
| 931 | pvec->pages[j++] = page; |
| 932 | } |
| 933 | pvec->nr = j; |
| 934 | } |
| 935 | |
| 936 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 937 | * pagevec_lookup - gang pagecache lookup |
| 938 | * @pvec: Where the resulting pages are placed |
| 939 | * @mapping: The address_space to search |
| 940 | * @start: The starting page index |
| 941 | * @nr_pages: The maximum number of pages |
| 942 | * |
| 943 | * pagevec_lookup() will search for and return a group of up to @nr_pages pages |
| 944 | * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a |
| 945 | * reference against the pages in @pvec. |
| 946 | * |
| 947 | * The search returns a group of mapping-contiguous pages with ascending |
| 948 | * indexes. There may be holes in the indices due to not-present pages. |
| 949 | * |
| 950 | * pagevec_lookup() returns the number of pages which were found. |
| 951 | */ |
| 952 | unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping, |
| 953 | pgoff_t start, unsigned nr_pages) |
| 954 | { |
| 955 | pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages); |
| 956 | return pagevec_count(pvec); |
| 957 | } |
Christoph Hellwig | 78539fd | 2006-01-11 20:47:41 +1100 | [diff] [blame] | 958 | EXPORT_SYMBOL(pagevec_lookup); |
| 959 | |
Jan Kara | fb296a2 | 2017-11-15 17:34:33 -0800 | [diff] [blame] | 960 | unsigned pagevec_lookup_range_tag(struct pagevec *pvec, |
| 961 | struct address_space *mapping, pgoff_t *index, pgoff_t end, |
Jan Kara | 4d93a43 | 2017-11-15 17:35:19 -0800 | [diff] [blame] | 962 | int tag) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 963 | { |
Jan Kara | fb296a2 | 2017-11-15 17:34:33 -0800 | [diff] [blame] | 964 | pvec->nr = find_get_pages_range_tag(mapping, index, end, tag, |
Jan Kara | 4d93a43 | 2017-11-15 17:35:19 -0800 | [diff] [blame] | 965 | PAGEVEC_SIZE, pvec->pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 966 | return pagevec_count(pvec); |
| 967 | } |
Jan Kara | fb296a2 | 2017-11-15 17:34:33 -0800 | [diff] [blame] | 968 | EXPORT_SYMBOL(pagevec_lookup_range_tag); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 969 | |
Jan Kara | 565d744 | 2017-11-15 17:35:12 -0800 | [diff] [blame] | 970 | unsigned pagevec_lookup_range_nr_tag(struct pagevec *pvec, |
| 971 | struct address_space *mapping, pgoff_t *index, pgoff_t end, |
| 972 | int tag, unsigned max_pages) |
| 973 | { |
| 974 | pvec->nr = find_get_pages_range_tag(mapping, index, end, tag, |
| 975 | min_t(unsigned int, max_pages, PAGEVEC_SIZE), pvec->pages); |
| 976 | return pagevec_count(pvec); |
| 977 | } |
| 978 | EXPORT_SYMBOL(pagevec_lookup_range_nr_tag); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 979 | /* |
| 980 | * Perform any setup for the swap system |
| 981 | */ |
| 982 | void __init swap_setup(void) |
| 983 | { |
Jan Beulich | 4481374 | 2009-09-21 17:03:05 -0700 | [diff] [blame] | 984 | unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT); |
Peter Zijlstra | e0bf68d | 2007-10-16 23:25:46 -0700 | [diff] [blame] | 985 | #ifdef CONFIG_SWAP |
Shaohua Li | 33806f0 | 2013-02-22 16:34:37 -0800 | [diff] [blame] | 986 | int i; |
| 987 | |
Kirill A. Shutemov | 27ba064 | 2015-02-10 14:09:59 -0800 | [diff] [blame] | 988 | for (i = 0; i < MAX_SWAPFILES; i++) |
Shaohua Li | 33806f0 | 2013-02-22 16:34:37 -0800 | [diff] [blame] | 989 | spin_lock_init(&swapper_spaces[i].tree_lock); |
Peter Zijlstra | e0bf68d | 2007-10-16 23:25:46 -0700 | [diff] [blame] | 990 | #endif |
| 991 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 992 | /* Use a smaller cluster for small-memory machines */ |
| 993 | if (megs < 16) |
| 994 | page_cluster = 2; |
| 995 | else |
| 996 | page_cluster = 3; |
| 997 | /* |
| 998 | * Right now other parts of the system means that we |
| 999 | * _really_ don't want to cluster much more |
| 1000 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1001 | } |