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> |
| 24 | #include <linux/module.h> |
| 25 | #include <linux/mm_inline.h> |
| 26 | #include <linux/buffer_head.h> /* for try_to_release_page() */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 27 | #include <linux/percpu_counter.h> |
| 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> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 33 | |
Lee Schermerhorn | 64d6519 | 2008-10-18 20:26:52 -0700 | [diff] [blame] | 34 | #include "internal.h" |
| 35 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 36 | /* How many pages do we try to swap or page in/out together? */ |
| 37 | int page_cluster; |
| 38 | |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 39 | static DEFINE_PER_CPU(struct pagevec[NR_LRU_LISTS], lru_add_pvecs); |
Vegard Nossum | f84f9504 | 2008-07-23 21:28:14 -0700 | [diff] [blame] | 40 | static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs); |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 41 | |
Adrian Bunk | b221385 | 2006-09-25 23:31:02 -0700 | [diff] [blame] | 42 | /* |
| 43 | * This path almost never happens for VM activity - pages are normally |
| 44 | * freed via pagevecs. But it gets used by networking. |
| 45 | */ |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 46 | static void __page_cache_release(struct page *page) |
Adrian Bunk | b221385 | 2006-09-25 23:31:02 -0700 | [diff] [blame] | 47 | { |
| 48 | if (PageLRU(page)) { |
| 49 | unsigned long flags; |
| 50 | struct zone *zone = page_zone(page); |
| 51 | |
| 52 | spin_lock_irqsave(&zone->lru_lock, flags); |
| 53 | VM_BUG_ON(!PageLRU(page)); |
| 54 | __ClearPageLRU(page); |
| 55 | del_page_from_lru(zone, page); |
| 56 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
| 57 | } |
| 58 | free_hot_page(page); |
| 59 | } |
| 60 | |
Nick Piggin | 8519fb3 | 2006-02-07 12:58:52 -0800 | [diff] [blame] | 61 | static void put_compound_page(struct page *page) |
| 62 | { |
Christoph Lameter | d85f338 | 2007-05-06 14:49:39 -0700 | [diff] [blame] | 63 | page = compound_head(page); |
Nick Piggin | 8519fb3 | 2006-02-07 12:58:52 -0800 | [diff] [blame] | 64 | if (put_page_testzero(page)) { |
Andy Whitcroft | 33f2ef8 | 2006-12-06 20:33:32 -0800 | [diff] [blame] | 65 | compound_page_dtor *dtor; |
Nick Piggin | 8519fb3 | 2006-02-07 12:58:52 -0800 | [diff] [blame] | 66 | |
Andy Whitcroft | 33f2ef8 | 2006-12-06 20:33:32 -0800 | [diff] [blame] | 67 | dtor = get_compound_page_dtor(page); |
Nick Piggin | 8519fb3 | 2006-02-07 12:58:52 -0800 | [diff] [blame] | 68 | (*dtor)(page); |
| 69 | } |
| 70 | } |
| 71 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 72 | void put_page(struct page *page) |
| 73 | { |
Nick Piggin | 8519fb3 | 2006-02-07 12:58:52 -0800 | [diff] [blame] | 74 | if (unlikely(PageCompound(page))) |
| 75 | put_compound_page(page); |
| 76 | else if (put_page_testzero(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 77 | __page_cache_release(page); |
| 78 | } |
| 79 | EXPORT_SYMBOL(put_page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 80 | |
Alexander Zarochentsev | 1d7ea73 | 2006-08-13 23:24:27 -0700 | [diff] [blame] | 81 | /** |
Randy Dunlap | 7682486 | 2008-03-19 17:00:40 -0700 | [diff] [blame] | 82 | * put_pages_list() - release a list of pages |
| 83 | * @pages: list of pages threaded on page->lru |
Alexander Zarochentsev | 1d7ea73 | 2006-08-13 23:24:27 -0700 | [diff] [blame] | 84 | * |
| 85 | * Release a list of pages which are strung together on page.lru. Currently |
| 86 | * used by read_cache_pages() and related error recovery code. |
Alexander Zarochentsev | 1d7ea73 | 2006-08-13 23:24:27 -0700 | [diff] [blame] | 87 | */ |
| 88 | void put_pages_list(struct list_head *pages) |
| 89 | { |
| 90 | while (!list_empty(pages)) { |
| 91 | struct page *victim; |
| 92 | |
| 93 | victim = list_entry(pages->prev, struct page, lru); |
| 94 | list_del(&victim->lru); |
| 95 | page_cache_release(victim); |
| 96 | } |
| 97 | } |
| 98 | EXPORT_SYMBOL(put_pages_list); |
| 99 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 100 | /* |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 101 | * pagevec_move_tail() must be called with IRQ disabled. |
| 102 | * Otherwise this may cause nasty races. |
| 103 | */ |
| 104 | static void pagevec_move_tail(struct pagevec *pvec) |
| 105 | { |
| 106 | int i; |
| 107 | int pgmoved = 0; |
| 108 | struct zone *zone = NULL; |
| 109 | |
| 110 | for (i = 0; i < pagevec_count(pvec); i++) { |
| 111 | struct page *page = pvec->pages[i]; |
| 112 | struct zone *pagezone = page_zone(page); |
| 113 | |
| 114 | if (pagezone != zone) { |
| 115 | if (zone) |
| 116 | spin_unlock(&zone->lru_lock); |
| 117 | zone = pagezone; |
| 118 | spin_lock(&zone->lru_lock); |
| 119 | } |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 120 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 121 | int lru = page_is_file_cache(page); |
| 122 | list_move_tail(&page->lru, &zone->lru[lru].list); |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 123 | pgmoved++; |
| 124 | } |
| 125 | } |
| 126 | if (zone) |
| 127 | spin_unlock(&zone->lru_lock); |
| 128 | __count_vm_events(PGROTATED, pgmoved); |
| 129 | release_pages(pvec->pages, pvec->nr, pvec->cold); |
| 130 | pagevec_reinit(pvec); |
| 131 | } |
| 132 | |
| 133 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 134 | * Writeback is about to end against a page which has been marked for immediate |
| 135 | * 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] | 136 | * inactive list. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 137 | */ |
Miklos Szeredi | ac6aadb | 2008-04-28 02:12:38 -0700 | [diff] [blame] | 138 | void rotate_reclaimable_page(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 139 | { |
Miklos Szeredi | ac6aadb | 2008-04-28 02:12:38 -0700 | [diff] [blame] | 140 | if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) && |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 141 | !PageUnevictable(page) && PageLRU(page)) { |
Miklos Szeredi | ac6aadb | 2008-04-28 02:12:38 -0700 | [diff] [blame] | 142 | struct pagevec *pvec; |
| 143 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 144 | |
Miklos Szeredi | ac6aadb | 2008-04-28 02:12:38 -0700 | [diff] [blame] | 145 | page_cache_get(page); |
| 146 | local_irq_save(flags); |
| 147 | pvec = &__get_cpu_var(lru_rotate_pvecs); |
| 148 | if (!pagevec_add(pvec, page)) |
| 149 | pagevec_move_tail(pvec); |
| 150 | local_irq_restore(flags); |
| 151 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 152 | } |
| 153 | |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 154 | static void update_page_reclaim_stat(struct zone *zone, struct page *page, |
| 155 | int file, int rotated) |
| 156 | { |
| 157 | struct zone_reclaim_stat *reclaim_stat = &zone->reclaim_stat; |
| 158 | struct zone_reclaim_stat *memcg_reclaim_stat; |
| 159 | |
| 160 | memcg_reclaim_stat = mem_cgroup_get_reclaim_stat_from_page(page); |
| 161 | |
| 162 | reclaim_stat->recent_scanned[file]++; |
| 163 | if (rotated) |
| 164 | reclaim_stat->recent_rotated[file]++; |
| 165 | |
| 166 | if (!memcg_reclaim_stat) |
| 167 | return; |
| 168 | |
| 169 | memcg_reclaim_stat->recent_scanned[file]++; |
| 170 | if (rotated) |
| 171 | memcg_reclaim_stat->recent_rotated[file]++; |
| 172 | } |
| 173 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 174 | /* |
| 175 | * FIXME: speed this up? |
| 176 | */ |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 177 | void activate_page(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 178 | { |
| 179 | struct zone *zone = page_zone(page); |
| 180 | |
| 181 | spin_lock_irq(&zone->lru_lock); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 182 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 183 | int file = page_is_file_cache(page); |
| 184 | int lru = LRU_BASE + file; |
| 185 | del_page_from_lru_list(zone, page, lru); |
| 186 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 187 | SetPageActive(page); |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 188 | lru += LRU_ACTIVE; |
| 189 | add_page_to_lru_list(zone, page, lru); |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 190 | __count_vm_event(PGACTIVATE); |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 191 | |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 192 | update_page_reclaim_stat(zone, page, !!file, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 193 | } |
| 194 | spin_unlock_irq(&zone->lru_lock); |
| 195 | } |
| 196 | |
| 197 | /* |
| 198 | * Mark a page as having seen activity. |
| 199 | * |
| 200 | * inactive,unreferenced -> inactive,referenced |
| 201 | * inactive,referenced -> active,unreferenced |
| 202 | * active,unreferenced -> active,referenced |
| 203 | */ |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 204 | void mark_page_accessed(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 205 | { |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 206 | if (!PageActive(page) && !PageUnevictable(page) && |
| 207 | PageReferenced(page) && PageLRU(page)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 208 | activate_page(page); |
| 209 | ClearPageReferenced(page); |
| 210 | } else if (!PageReferenced(page)) { |
| 211 | SetPageReferenced(page); |
| 212 | } |
| 213 | } |
| 214 | |
| 215 | EXPORT_SYMBOL(mark_page_accessed); |
| 216 | |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 217 | void __lru_cache_add(struct page *page, enum lru_list lru) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 218 | { |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 219 | struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 220 | |
| 221 | page_cache_get(page); |
| 222 | if (!pagevec_add(pvec, page)) |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 223 | ____pagevec_lru_add(pvec, lru); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 224 | put_cpu_var(lru_add_pvecs); |
| 225 | } |
| 226 | |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 227 | /** |
| 228 | * lru_cache_add_lru - add a page to a page list |
| 229 | * @page: the page to be added to the LRU. |
| 230 | * @lru: the LRU list to which the page is added. |
| 231 | */ |
| 232 | void lru_cache_add_lru(struct page *page, enum lru_list lru) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 233 | { |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 234 | if (PageActive(page)) { |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 235 | VM_BUG_ON(PageUnevictable(page)); |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 236 | ClearPageActive(page); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 237 | } else if (PageUnevictable(page)) { |
| 238 | VM_BUG_ON(PageActive(page)); |
| 239 | ClearPageUnevictable(page); |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 240 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 241 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 242 | VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page)); |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 243 | __lru_cache_add(page, lru); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 244 | } |
| 245 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 246 | /** |
| 247 | * add_page_to_unevictable_list - add a page to the unevictable list |
| 248 | * @page: the page to be added to the unevictable list |
| 249 | * |
| 250 | * Add page directly to its zone's unevictable list. To avoid races with |
| 251 | * tasks that might be making the page evictable, through eg. munlock, |
| 252 | * munmap or exit, while it's not on the lru, we want to add the page |
| 253 | * while it's locked or otherwise "invisible" to other tasks. This is |
| 254 | * difficult to do when using the pagevec cache, so bypass that. |
| 255 | */ |
| 256 | void add_page_to_unevictable_list(struct page *page) |
| 257 | { |
| 258 | struct zone *zone = page_zone(page); |
| 259 | |
| 260 | spin_lock_irq(&zone->lru_lock); |
| 261 | SetPageUnevictable(page); |
| 262 | SetPageLRU(page); |
| 263 | add_page_to_lru_list(zone, page, LRU_UNEVICTABLE); |
| 264 | spin_unlock_irq(&zone->lru_lock); |
| 265 | } |
| 266 | |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 267 | /* |
| 268 | * Drain pages out of the cpu's pagevecs. |
| 269 | * Either "cpu" is the current CPU, and preemption has already been |
| 270 | * disabled; or "cpu" is being hot-unplugged, and is already dead. |
| 271 | */ |
| 272 | static void drain_cpu_pagevecs(int cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 273 | { |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 274 | struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu); |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 275 | struct pagevec *pvec; |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 276 | int lru; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 277 | |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 278 | for_each_lru(lru) { |
| 279 | pvec = &pvecs[lru - LRU_BASE]; |
| 280 | if (pagevec_count(pvec)) |
| 281 | ____pagevec_lru_add(pvec, lru); |
| 282 | } |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 283 | |
| 284 | pvec = &per_cpu(lru_rotate_pvecs, cpu); |
| 285 | if (pagevec_count(pvec)) { |
| 286 | unsigned long flags; |
| 287 | |
| 288 | /* No harm done if a racing interrupt already did this */ |
| 289 | local_irq_save(flags); |
| 290 | pagevec_move_tail(pvec); |
| 291 | local_irq_restore(flags); |
| 292 | } |
Andrew Morton | 80bfed9 | 2006-01-06 00:11:14 -0800 | [diff] [blame] | 293 | } |
| 294 | |
| 295 | void lru_add_drain(void) |
| 296 | { |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 297 | drain_cpu_pagevecs(get_cpu()); |
Andrew Morton | 80bfed9 | 2006-01-06 00:11:14 -0800 | [diff] [blame] | 298 | put_cpu(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 299 | } |
| 300 | |
David Howells | c402895 | 2006-11-22 14:57:56 +0000 | [diff] [blame] | 301 | static void lru_add_drain_per_cpu(struct work_struct *dummy) |
Nick Piggin | 053837f | 2006-01-18 17:42:27 -0800 | [diff] [blame] | 302 | { |
| 303 | lru_add_drain(); |
| 304 | } |
| 305 | |
| 306 | /* |
| 307 | * Returns 0 for success |
| 308 | */ |
| 309 | int lru_add_drain_all(void) |
| 310 | { |
David Howells | c402895 | 2006-11-22 14:57:56 +0000 | [diff] [blame] | 311 | return schedule_on_each_cpu(lru_add_drain_per_cpu); |
Nick Piggin | 053837f | 2006-01-18 17:42:27 -0800 | [diff] [blame] | 312 | } |
| 313 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 314 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 315 | * Batched page_cache_release(). Decrement the reference count on all the |
| 316 | * passed pages. If it fell to zero then remove the page from the LRU and |
| 317 | * free it. |
| 318 | * |
| 319 | * Avoid taking zone->lru_lock if possible, but if it is taken, retain it |
| 320 | * for the remainder of the operation. |
| 321 | * |
Fernando Luis Vazquez Cao | ab33dc0 | 2008-07-29 22:33:40 -0700 | [diff] [blame] | 322 | * The locking in this function is against shrink_inactive_list(): we recheck |
| 323 | * the page count inside the lock to see whether shrink_inactive_list() |
| 324 | * grabbed the page via the LRU. If it did, give up: shrink_inactive_list() |
| 325 | * will free it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 326 | */ |
| 327 | void release_pages(struct page **pages, int nr, int cold) |
| 328 | { |
| 329 | int i; |
| 330 | struct pagevec pages_to_free; |
| 331 | struct zone *zone = NULL; |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 332 | unsigned long uninitialized_var(flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 333 | |
| 334 | pagevec_init(&pages_to_free, cold); |
| 335 | for (i = 0; i < nr; i++) { |
| 336 | struct page *page = pages[i]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 337 | |
Nick Piggin | 8519fb3 | 2006-02-07 12:58:52 -0800 | [diff] [blame] | 338 | if (unlikely(PageCompound(page))) { |
| 339 | if (zone) { |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 340 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
Nick Piggin | 8519fb3 | 2006-02-07 12:58:52 -0800 | [diff] [blame] | 341 | zone = NULL; |
| 342 | } |
| 343 | put_compound_page(page); |
| 344 | continue; |
| 345 | } |
| 346 | |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 347 | if (!put_page_testzero(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 348 | continue; |
| 349 | |
Nick Piggin | 46453a6 | 2006-03-22 00:07:58 -0800 | [diff] [blame] | 350 | if (PageLRU(page)) { |
| 351 | struct zone *pagezone = page_zone(page); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 352 | |
Nick Piggin | 46453a6 | 2006-03-22 00:07:58 -0800 | [diff] [blame] | 353 | if (pagezone != zone) { |
| 354 | if (zone) |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 355 | spin_unlock_irqrestore(&zone->lru_lock, |
| 356 | flags); |
Nick Piggin | 46453a6 | 2006-03-22 00:07:58 -0800 | [diff] [blame] | 357 | zone = pagezone; |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 358 | spin_lock_irqsave(&zone->lru_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 359 | } |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 360 | VM_BUG_ON(!PageLRU(page)); |
Nick Piggin | 6745391 | 2006-03-22 00:08:00 -0800 | [diff] [blame] | 361 | __ClearPageLRU(page); |
Nick Piggin | 46453a6 | 2006-03-22 00:07:58 -0800 | [diff] [blame] | 362 | del_page_from_lru(zone, page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 363 | } |
Nick Piggin | 46453a6 | 2006-03-22 00:07:58 -0800 | [diff] [blame] | 364 | |
| 365 | if (!pagevec_add(&pages_to_free, page)) { |
| 366 | if (zone) { |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 367 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
Nick Piggin | 46453a6 | 2006-03-22 00:07:58 -0800 | [diff] [blame] | 368 | zone = NULL; |
| 369 | } |
| 370 | __pagevec_free(&pages_to_free); |
| 371 | pagevec_reinit(&pages_to_free); |
| 372 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 373 | } |
| 374 | if (zone) |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 375 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 376 | |
| 377 | pagevec_free(&pages_to_free); |
| 378 | } |
| 379 | |
| 380 | /* |
| 381 | * The pages which we're about to release may be in the deferred lru-addition |
| 382 | * queues. That would prevent them from really being freed right now. That's |
| 383 | * OK from a correctness point of view but is inefficient - those pages may be |
| 384 | * cache-warm and we want to give them back to the page allocator ASAP. |
| 385 | * |
| 386 | * So __pagevec_release() will drain those queues here. __pagevec_lru_add() |
| 387 | * and __pagevec_lru_add_active() call release_pages() directly to avoid |
| 388 | * mutual recursion. |
| 389 | */ |
| 390 | void __pagevec_release(struct pagevec *pvec) |
| 391 | { |
| 392 | lru_add_drain(); |
| 393 | release_pages(pvec->pages, pagevec_count(pvec), pvec->cold); |
| 394 | pagevec_reinit(pvec); |
| 395 | } |
| 396 | |
Steve French | 7f28570 | 2005-11-01 10:22:55 -0800 | [diff] [blame] | 397 | EXPORT_SYMBOL(__pagevec_release); |
| 398 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 399 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 400 | * Add the passed pages to the LRU, then drop the caller's refcount |
| 401 | * on them. Reinitialises the caller's pagevec. |
| 402 | */ |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 403 | void ____pagevec_lru_add(struct pagevec *pvec, enum lru_list lru) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 404 | { |
| 405 | int i; |
| 406 | struct zone *zone = NULL; |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 407 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 408 | VM_BUG_ON(is_unevictable_lru(lru)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 409 | |
| 410 | for (i = 0; i < pagevec_count(pvec); i++) { |
| 411 | struct page *page = pvec->pages[i]; |
| 412 | struct zone *pagezone = page_zone(page); |
Rik van Riel | 9ff473b | 2008-12-02 10:31:52 -0800 | [diff] [blame] | 413 | int file; |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 414 | int active; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 415 | |
| 416 | if (pagezone != zone) { |
| 417 | if (zone) |
| 418 | spin_unlock_irq(&zone->lru_lock); |
| 419 | zone = pagezone; |
| 420 | spin_lock_irq(&zone->lru_lock); |
| 421 | } |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 422 | VM_BUG_ON(PageActive(page)); |
| 423 | VM_BUG_ON(PageUnevictable(page)); |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 424 | VM_BUG_ON(PageLRU(page)); |
Nick Piggin | 8d438f9 | 2006-03-22 00:07:59 -0800 | [diff] [blame] | 425 | SetPageLRU(page); |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 426 | active = is_active_lru(lru); |
Rik van Riel | 9ff473b | 2008-12-02 10:31:52 -0800 | [diff] [blame] | 427 | file = is_file_lru(lru); |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 428 | if (active) |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 429 | SetPageActive(page); |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 430 | update_page_reclaim_stat(zone, page, file, active); |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 431 | add_page_to_lru_list(zone, page, lru); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 432 | } |
| 433 | if (zone) |
| 434 | spin_unlock_irq(&zone->lru_lock); |
| 435 | release_pages(pvec->pages, pvec->nr, pvec->cold); |
| 436 | pagevec_reinit(pvec); |
| 437 | } |
| 438 | |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 439 | EXPORT_SYMBOL(____pagevec_lru_add); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 440 | |
| 441 | /* |
| 442 | * Try to drop buffers from the pages in a pagevec |
| 443 | */ |
| 444 | void pagevec_strip(struct pagevec *pvec) |
| 445 | { |
| 446 | int i; |
| 447 | |
| 448 | for (i = 0; i < pagevec_count(pvec); i++) { |
| 449 | struct page *page = pvec->pages[i]; |
| 450 | |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 451 | if (PagePrivate(page) && trylock_page(page)) { |
Christoph Lameter | 5b40dc7 | 2006-03-16 23:04:07 -0800 | [diff] [blame] | 452 | if (PagePrivate(page)) |
| 453 | try_to_release_page(page, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 454 | unlock_page(page); |
| 455 | } |
| 456 | } |
| 457 | } |
| 458 | |
| 459 | /** |
| 460 | * pagevec_lookup - gang pagecache lookup |
| 461 | * @pvec: Where the resulting pages are placed |
| 462 | * @mapping: The address_space to search |
| 463 | * @start: The starting page index |
| 464 | * @nr_pages: The maximum number of pages |
| 465 | * |
| 466 | * pagevec_lookup() will search for and return a group of up to @nr_pages pages |
| 467 | * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a |
| 468 | * reference against the pages in @pvec. |
| 469 | * |
| 470 | * The search returns a group of mapping-contiguous pages with ascending |
| 471 | * indexes. There may be holes in the indices due to not-present pages. |
| 472 | * |
| 473 | * pagevec_lookup() returns the number of pages which were found. |
| 474 | */ |
| 475 | unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping, |
| 476 | pgoff_t start, unsigned nr_pages) |
| 477 | { |
| 478 | pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages); |
| 479 | return pagevec_count(pvec); |
| 480 | } |
| 481 | |
Christoph Hellwig | 78539fd | 2006-01-11 20:47:41 +1100 | [diff] [blame] | 482 | EXPORT_SYMBOL(pagevec_lookup); |
| 483 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 484 | unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping, |
| 485 | pgoff_t *index, int tag, unsigned nr_pages) |
| 486 | { |
| 487 | pvec->nr = find_get_pages_tag(mapping, index, tag, |
| 488 | nr_pages, pvec->pages); |
| 489 | return pagevec_count(pvec); |
| 490 | } |
| 491 | |
Steve French | 7f28570 | 2005-11-01 10:22:55 -0800 | [diff] [blame] | 492 | EXPORT_SYMBOL(pagevec_lookup_tag); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 493 | |
| 494 | #ifdef CONFIG_SMP |
| 495 | /* |
| 496 | * We tolerate a little inaccuracy to avoid ping-ponging the counter between |
| 497 | * CPUs |
| 498 | */ |
| 499 | #define ACCT_THRESHOLD max(16, NR_CPUS * 2) |
| 500 | |
Vegard Nossum | f84f9504 | 2008-07-23 21:28:14 -0700 | [diff] [blame] | 501 | static DEFINE_PER_CPU(long, committed_space); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 502 | |
| 503 | void vm_acct_memory(long pages) |
| 504 | { |
| 505 | long *local; |
| 506 | |
| 507 | preempt_disable(); |
| 508 | local = &__get_cpu_var(committed_space); |
| 509 | *local += pages; |
| 510 | if (*local > ACCT_THRESHOLD || *local < -ACCT_THRESHOLD) { |
Alan Cox | 80119ef | 2008-05-23 13:04:31 -0700 | [diff] [blame] | 511 | atomic_long_add(*local, &vm_committed_space); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 512 | *local = 0; |
| 513 | } |
| 514 | preempt_enable(); |
| 515 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 516 | |
| 517 | #ifdef CONFIG_HOTPLUG_CPU |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 518 | |
| 519 | /* Drop the CPU's cached committed space back into the central pool. */ |
| 520 | static int cpu_swap_callback(struct notifier_block *nfb, |
| 521 | unsigned long action, |
| 522 | void *hcpu) |
| 523 | { |
| 524 | long *committed; |
| 525 | |
| 526 | committed = &per_cpu(committed_space, (long)hcpu); |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 527 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
Alan Cox | 80119ef | 2008-05-23 13:04:31 -0700 | [diff] [blame] | 528 | atomic_long_add(*committed, &vm_committed_space); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 529 | *committed = 0; |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 530 | drain_cpu_pagevecs((long)hcpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 531 | } |
| 532 | return NOTIFY_OK; |
| 533 | } |
| 534 | #endif /* CONFIG_HOTPLUG_CPU */ |
| 535 | #endif /* CONFIG_SMP */ |
| 536 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 537 | /* |
| 538 | * Perform any setup for the swap system |
| 539 | */ |
| 540 | void __init swap_setup(void) |
| 541 | { |
| 542 | unsigned long megs = num_physpages >> (20 - PAGE_SHIFT); |
| 543 | |
Peter Zijlstra | e0bf68d | 2007-10-16 23:25:46 -0700 | [diff] [blame] | 544 | #ifdef CONFIG_SWAP |
| 545 | bdi_init(swapper_space.backing_dev_info); |
| 546 | #endif |
| 547 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 548 | /* Use a smaller cluster for small-memory machines */ |
| 549 | if (megs < 16) |
| 550 | page_cluster = 2; |
| 551 | else |
| 552 | page_cluster = 3; |
| 553 | /* |
| 554 | * Right now other parts of the system means that we |
| 555 | * _really_ don't want to cluster much more |
| 556 | */ |
Ingo Molnar | 0231606 | 2006-12-06 20:38:17 -0800 | [diff] [blame] | 557 | #ifdef CONFIG_HOTPLUG_CPU |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 558 | hotcpu_notifier(cpu_swap_callback, 0); |
Ingo Molnar | 0231606 | 2006-12-06 20:38:17 -0800 | [diff] [blame] | 559 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 560 | } |