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 | |
| 34 | /* How many pages do we try to swap or page in/out together? */ |
| 35 | int page_cluster; |
| 36 | |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 37 | static DEFINE_PER_CPU(struct pagevec[NR_LRU_LISTS], lru_add_pvecs); |
Vegard Nossum | f84f9504 | 2008-07-23 21:28:14 -0700 | [diff] [blame] | 38 | static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs); |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 39 | |
Adrian Bunk | b221385 | 2006-09-25 23:31:02 -0700 | [diff] [blame] | 40 | /* |
| 41 | * This path almost never happens for VM activity - pages are normally |
| 42 | * freed via pagevecs. But it gets used by networking. |
| 43 | */ |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 44 | static void __page_cache_release(struct page *page) |
Adrian Bunk | b221385 | 2006-09-25 23:31:02 -0700 | [diff] [blame] | 45 | { |
| 46 | if (PageLRU(page)) { |
| 47 | unsigned long flags; |
| 48 | struct zone *zone = page_zone(page); |
| 49 | |
| 50 | spin_lock_irqsave(&zone->lru_lock, flags); |
| 51 | VM_BUG_ON(!PageLRU(page)); |
| 52 | __ClearPageLRU(page); |
| 53 | del_page_from_lru(zone, page); |
| 54 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
| 55 | } |
| 56 | free_hot_page(page); |
| 57 | } |
| 58 | |
Nick Piggin | 8519fb3 | 2006-02-07 12:58:52 -0800 | [diff] [blame] | 59 | static void put_compound_page(struct page *page) |
| 60 | { |
Christoph Lameter | d85f338 | 2007-05-06 14:49:39 -0700 | [diff] [blame] | 61 | page = compound_head(page); |
Nick Piggin | 8519fb3 | 2006-02-07 12:58:52 -0800 | [diff] [blame] | 62 | if (put_page_testzero(page)) { |
Andy Whitcroft | 33f2ef8 | 2006-12-06 20:33:32 -0800 | [diff] [blame] | 63 | compound_page_dtor *dtor; |
Nick Piggin | 8519fb3 | 2006-02-07 12:58:52 -0800 | [diff] [blame] | 64 | |
Andy Whitcroft | 33f2ef8 | 2006-12-06 20:33:32 -0800 | [diff] [blame] | 65 | dtor = get_compound_page_dtor(page); |
Nick Piggin | 8519fb3 | 2006-02-07 12:58:52 -0800 | [diff] [blame] | 66 | (*dtor)(page); |
| 67 | } |
| 68 | } |
| 69 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 70 | void put_page(struct page *page) |
| 71 | { |
Nick Piggin | 8519fb3 | 2006-02-07 12:58:52 -0800 | [diff] [blame] | 72 | if (unlikely(PageCompound(page))) |
| 73 | put_compound_page(page); |
| 74 | else if (put_page_testzero(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 75 | __page_cache_release(page); |
| 76 | } |
| 77 | EXPORT_SYMBOL(put_page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 78 | |
Alexander Zarochentsev | 1d7ea73 | 2006-08-13 23:24:27 -0700 | [diff] [blame] | 79 | /** |
Randy Dunlap | 7682486 | 2008-03-19 17:00:40 -0700 | [diff] [blame] | 80 | * put_pages_list() - release a list of pages |
| 81 | * @pages: list of pages threaded on page->lru |
Alexander Zarochentsev | 1d7ea73 | 2006-08-13 23:24:27 -0700 | [diff] [blame] | 82 | * |
| 83 | * Release a list of pages which are strung together on page.lru. Currently |
| 84 | * used by read_cache_pages() and related error recovery code. |
Alexander Zarochentsev | 1d7ea73 | 2006-08-13 23:24:27 -0700 | [diff] [blame] | 85 | */ |
| 86 | void put_pages_list(struct list_head *pages) |
| 87 | { |
| 88 | while (!list_empty(pages)) { |
| 89 | struct page *victim; |
| 90 | |
| 91 | victim = list_entry(pages->prev, struct page, lru); |
| 92 | list_del(&victim->lru); |
| 93 | page_cache_release(victim); |
| 94 | } |
| 95 | } |
| 96 | EXPORT_SYMBOL(put_pages_list); |
| 97 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 98 | /* |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 99 | * pagevec_move_tail() must be called with IRQ disabled. |
| 100 | * Otherwise this may cause nasty races. |
| 101 | */ |
| 102 | static void pagevec_move_tail(struct pagevec *pvec) |
| 103 | { |
| 104 | int i; |
| 105 | int pgmoved = 0; |
| 106 | struct zone *zone = NULL; |
| 107 | |
| 108 | for (i = 0; i < pagevec_count(pvec); i++) { |
| 109 | struct page *page = pvec->pages[i]; |
| 110 | struct zone *pagezone = page_zone(page); |
| 111 | |
| 112 | if (pagezone != zone) { |
| 113 | if (zone) |
| 114 | spin_unlock(&zone->lru_lock); |
| 115 | zone = pagezone; |
| 116 | spin_lock(&zone->lru_lock); |
| 117 | } |
| 118 | if (PageLRU(page) && !PageActive(page)) { |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 119 | list_move_tail(&page->lru, &zone->lru[LRU_INACTIVE].list); |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 120 | pgmoved++; |
| 121 | } |
| 122 | } |
| 123 | if (zone) |
| 124 | spin_unlock(&zone->lru_lock); |
| 125 | __count_vm_events(PGROTATED, pgmoved); |
| 126 | release_pages(pvec->pages, pvec->nr, pvec->cold); |
| 127 | pagevec_reinit(pvec); |
| 128 | } |
| 129 | |
| 130 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 131 | * Writeback is about to end against a page which has been marked for immediate |
| 132 | * 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] | 133 | * inactive list. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 134 | */ |
Miklos Szeredi | ac6aadb | 2008-04-28 02:12:38 -0700 | [diff] [blame] | 135 | void rotate_reclaimable_page(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 136 | { |
Miklos Szeredi | ac6aadb | 2008-04-28 02:12:38 -0700 | [diff] [blame] | 137 | if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) && |
| 138 | PageLRU(page)) { |
| 139 | struct pagevec *pvec; |
| 140 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 141 | |
Miklos Szeredi | ac6aadb | 2008-04-28 02:12:38 -0700 | [diff] [blame] | 142 | page_cache_get(page); |
| 143 | local_irq_save(flags); |
| 144 | pvec = &__get_cpu_var(lru_rotate_pvecs); |
| 145 | if (!pagevec_add(pvec, page)) |
| 146 | pagevec_move_tail(pvec); |
| 147 | local_irq_restore(flags); |
| 148 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 149 | } |
| 150 | |
| 151 | /* |
| 152 | * FIXME: speed this up? |
| 153 | */ |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 154 | void activate_page(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 155 | { |
| 156 | struct zone *zone = page_zone(page); |
| 157 | |
| 158 | spin_lock_irq(&zone->lru_lock); |
| 159 | if (PageLRU(page) && !PageActive(page)) { |
| 160 | del_page_from_inactive_list(zone, page); |
| 161 | SetPageActive(page); |
| 162 | add_page_to_active_list(zone, page); |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 163 | __count_vm_event(PGACTIVATE); |
Hugh Dickins | 427d541 | 2008-03-04 14:29:03 -0800 | [diff] [blame] | 164 | mem_cgroup_move_lists(page, true); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 165 | } |
| 166 | spin_unlock_irq(&zone->lru_lock); |
| 167 | } |
| 168 | |
| 169 | /* |
| 170 | * Mark a page as having seen activity. |
| 171 | * |
| 172 | * inactive,unreferenced -> inactive,referenced |
| 173 | * inactive,referenced -> active,unreferenced |
| 174 | * active,unreferenced -> active,referenced |
| 175 | */ |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 176 | void mark_page_accessed(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 177 | { |
| 178 | if (!PageActive(page) && PageReferenced(page) && PageLRU(page)) { |
| 179 | activate_page(page); |
| 180 | ClearPageReferenced(page); |
| 181 | } else if (!PageReferenced(page)) { |
| 182 | SetPageReferenced(page); |
| 183 | } |
| 184 | } |
| 185 | |
| 186 | EXPORT_SYMBOL(mark_page_accessed); |
| 187 | |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 188 | void __lru_cache_add(struct page *page, enum lru_list lru) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 189 | { |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 190 | struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 191 | |
| 192 | page_cache_get(page); |
| 193 | if (!pagevec_add(pvec, page)) |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 194 | ____pagevec_lru_add(pvec, lru); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 195 | put_cpu_var(lru_add_pvecs); |
| 196 | } |
| 197 | |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 198 | /** |
| 199 | * lru_cache_add_lru - add a page to a page list |
| 200 | * @page: the page to be added to the LRU. |
| 201 | * @lru: the LRU list to which the page is added. |
| 202 | */ |
| 203 | void lru_cache_add_lru(struct page *page, enum lru_list lru) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 204 | { |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 205 | if (PageActive(page)) { |
| 206 | ClearPageActive(page); |
| 207 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 208 | |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 209 | VM_BUG_ON(PageLRU(page) || PageActive(page)); |
| 210 | __lru_cache_add(page, lru); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 211 | } |
| 212 | |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 213 | /* |
| 214 | * Drain pages out of the cpu's pagevecs. |
| 215 | * Either "cpu" is the current CPU, and preemption has already been |
| 216 | * disabled; or "cpu" is being hot-unplugged, and is already dead. |
| 217 | */ |
| 218 | static void drain_cpu_pagevecs(int cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 219 | { |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 220 | struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu); |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 221 | struct pagevec *pvec; |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 222 | int lru; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 223 | |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 224 | for_each_lru(lru) { |
| 225 | pvec = &pvecs[lru - LRU_BASE]; |
| 226 | if (pagevec_count(pvec)) |
| 227 | ____pagevec_lru_add(pvec, lru); |
| 228 | } |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 229 | |
| 230 | pvec = &per_cpu(lru_rotate_pvecs, cpu); |
| 231 | if (pagevec_count(pvec)) { |
| 232 | unsigned long flags; |
| 233 | |
| 234 | /* No harm done if a racing interrupt already did this */ |
| 235 | local_irq_save(flags); |
| 236 | pagevec_move_tail(pvec); |
| 237 | local_irq_restore(flags); |
| 238 | } |
Andrew Morton | 80bfed9 | 2006-01-06 00:11:14 -0800 | [diff] [blame] | 239 | } |
| 240 | |
| 241 | void lru_add_drain(void) |
| 242 | { |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 243 | drain_cpu_pagevecs(get_cpu()); |
Andrew Morton | 80bfed9 | 2006-01-06 00:11:14 -0800 | [diff] [blame] | 244 | put_cpu(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 245 | } |
| 246 | |
Nick Piggin | 053837f | 2006-01-18 17:42:27 -0800 | [diff] [blame] | 247 | #ifdef CONFIG_NUMA |
David Howells | c402895 | 2006-11-22 14:57:56 +0000 | [diff] [blame] | 248 | static void lru_add_drain_per_cpu(struct work_struct *dummy) |
Nick Piggin | 053837f | 2006-01-18 17:42:27 -0800 | [diff] [blame] | 249 | { |
| 250 | lru_add_drain(); |
| 251 | } |
| 252 | |
| 253 | /* |
| 254 | * Returns 0 for success |
| 255 | */ |
| 256 | int lru_add_drain_all(void) |
| 257 | { |
David Howells | c402895 | 2006-11-22 14:57:56 +0000 | [diff] [blame] | 258 | return schedule_on_each_cpu(lru_add_drain_per_cpu); |
Nick Piggin | 053837f | 2006-01-18 17:42:27 -0800 | [diff] [blame] | 259 | } |
| 260 | |
| 261 | #else |
| 262 | |
| 263 | /* |
| 264 | * Returns 0 for success |
| 265 | */ |
| 266 | int lru_add_drain_all(void) |
| 267 | { |
| 268 | lru_add_drain(); |
| 269 | return 0; |
| 270 | } |
| 271 | #endif |
| 272 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 273 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 274 | * Batched page_cache_release(). Decrement the reference count on all the |
| 275 | * passed pages. If it fell to zero then remove the page from the LRU and |
| 276 | * free it. |
| 277 | * |
| 278 | * Avoid taking zone->lru_lock if possible, but if it is taken, retain it |
| 279 | * for the remainder of the operation. |
| 280 | * |
Fernando Luis Vazquez Cao | ab33dc0 | 2008-07-29 22:33:40 -0700 | [diff] [blame] | 281 | * The locking in this function is against shrink_inactive_list(): we recheck |
| 282 | * the page count inside the lock to see whether shrink_inactive_list() |
| 283 | * grabbed the page via the LRU. If it did, give up: shrink_inactive_list() |
| 284 | * will free it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 285 | */ |
| 286 | void release_pages(struct page **pages, int nr, int cold) |
| 287 | { |
| 288 | int i; |
| 289 | struct pagevec pages_to_free; |
| 290 | struct zone *zone = NULL; |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 291 | unsigned long uninitialized_var(flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 292 | |
| 293 | pagevec_init(&pages_to_free, cold); |
| 294 | for (i = 0; i < nr; i++) { |
| 295 | struct page *page = pages[i]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 296 | |
Nick Piggin | 8519fb3 | 2006-02-07 12:58:52 -0800 | [diff] [blame] | 297 | if (unlikely(PageCompound(page))) { |
| 298 | if (zone) { |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 299 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
Nick Piggin | 8519fb3 | 2006-02-07 12:58:52 -0800 | [diff] [blame] | 300 | zone = NULL; |
| 301 | } |
| 302 | put_compound_page(page); |
| 303 | continue; |
| 304 | } |
| 305 | |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 306 | if (!put_page_testzero(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 307 | continue; |
| 308 | |
Nick Piggin | 46453a6 | 2006-03-22 00:07:58 -0800 | [diff] [blame] | 309 | if (PageLRU(page)) { |
| 310 | struct zone *pagezone = page_zone(page); |
| 311 | if (pagezone != zone) { |
| 312 | if (zone) |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 313 | spin_unlock_irqrestore(&zone->lru_lock, |
| 314 | flags); |
Nick Piggin | 46453a6 | 2006-03-22 00:07:58 -0800 | [diff] [blame] | 315 | zone = pagezone; |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 316 | spin_lock_irqsave(&zone->lru_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 317 | } |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 318 | VM_BUG_ON(!PageLRU(page)); |
Nick Piggin | 6745391 | 2006-03-22 00:08:00 -0800 | [diff] [blame] | 319 | __ClearPageLRU(page); |
Nick Piggin | 46453a6 | 2006-03-22 00:07:58 -0800 | [diff] [blame] | 320 | del_page_from_lru(zone, page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 321 | } |
Nick Piggin | 46453a6 | 2006-03-22 00:07:58 -0800 | [diff] [blame] | 322 | |
| 323 | if (!pagevec_add(&pages_to_free, page)) { |
| 324 | if (zone) { |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 325 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
Nick Piggin | 46453a6 | 2006-03-22 00:07:58 -0800 | [diff] [blame] | 326 | zone = NULL; |
| 327 | } |
| 328 | __pagevec_free(&pages_to_free); |
| 329 | pagevec_reinit(&pages_to_free); |
| 330 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 331 | } |
| 332 | if (zone) |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 333 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 334 | |
| 335 | pagevec_free(&pages_to_free); |
| 336 | } |
| 337 | |
| 338 | /* |
| 339 | * The pages which we're about to release may be in the deferred lru-addition |
| 340 | * queues. That would prevent them from really being freed right now. That's |
| 341 | * OK from a correctness point of view but is inefficient - those pages may be |
| 342 | * cache-warm and we want to give them back to the page allocator ASAP. |
| 343 | * |
| 344 | * So __pagevec_release() will drain those queues here. __pagevec_lru_add() |
| 345 | * and __pagevec_lru_add_active() call release_pages() directly to avoid |
| 346 | * mutual recursion. |
| 347 | */ |
| 348 | void __pagevec_release(struct pagevec *pvec) |
| 349 | { |
| 350 | lru_add_drain(); |
| 351 | release_pages(pvec->pages, pagevec_count(pvec), pvec->cold); |
| 352 | pagevec_reinit(pvec); |
| 353 | } |
| 354 | |
Steve French | 7f28570 | 2005-11-01 10:22:55 -0800 | [diff] [blame] | 355 | EXPORT_SYMBOL(__pagevec_release); |
| 356 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 357 | /* |
| 358 | * pagevec_release() for pages which are known to not be on the LRU |
| 359 | * |
| 360 | * This function reinitialises the caller's pagevec. |
| 361 | */ |
| 362 | void __pagevec_release_nonlru(struct pagevec *pvec) |
| 363 | { |
| 364 | int i; |
| 365 | struct pagevec pages_to_free; |
| 366 | |
| 367 | pagevec_init(&pages_to_free, pvec->cold); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 368 | for (i = 0; i < pagevec_count(pvec); i++) { |
| 369 | struct page *page = pvec->pages[i]; |
| 370 | |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 371 | VM_BUG_ON(PageLRU(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 372 | if (put_page_testzero(page)) |
| 373 | pagevec_add(&pages_to_free, page); |
| 374 | } |
| 375 | pagevec_free(&pages_to_free); |
| 376 | pagevec_reinit(pvec); |
| 377 | } |
| 378 | |
| 379 | /* |
| 380 | * Add the passed pages to the LRU, then drop the caller's refcount |
| 381 | * on them. Reinitialises the caller's pagevec. |
| 382 | */ |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 383 | void ____pagevec_lru_add(struct pagevec *pvec, enum lru_list lru) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 384 | { |
| 385 | int i; |
| 386 | struct zone *zone = NULL; |
| 387 | |
| 388 | for (i = 0; i < pagevec_count(pvec); i++) { |
| 389 | struct page *page = pvec->pages[i]; |
| 390 | struct zone *pagezone = page_zone(page); |
| 391 | |
| 392 | if (pagezone != zone) { |
| 393 | if (zone) |
| 394 | spin_unlock_irq(&zone->lru_lock); |
| 395 | zone = pagezone; |
| 396 | spin_lock_irq(&zone->lru_lock); |
| 397 | } |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 398 | VM_BUG_ON(PageLRU(page)); |
Nick Piggin | 8d438f9 | 2006-03-22 00:07:59 -0800 | [diff] [blame] | 399 | SetPageLRU(page); |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 400 | if (is_active_lru(lru)) |
| 401 | SetPageActive(page); |
| 402 | add_page_to_lru_list(zone, page, lru); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 403 | } |
| 404 | if (zone) |
| 405 | spin_unlock_irq(&zone->lru_lock); |
| 406 | release_pages(pvec->pages, pvec->nr, pvec->cold); |
| 407 | pagevec_reinit(pvec); |
| 408 | } |
| 409 | |
KOSAKI Motohiro | f04e9eb | 2008-10-18 20:26:19 -0700 | [diff] [blame] | 410 | EXPORT_SYMBOL(____pagevec_lru_add); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 411 | |
| 412 | /* |
| 413 | * Try to drop buffers from the pages in a pagevec |
| 414 | */ |
| 415 | void pagevec_strip(struct pagevec *pvec) |
| 416 | { |
| 417 | int i; |
| 418 | |
| 419 | for (i = 0; i < pagevec_count(pvec); i++) { |
| 420 | struct page *page = pvec->pages[i]; |
| 421 | |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 422 | if (PagePrivate(page) && trylock_page(page)) { |
Christoph Lameter | 5b40dc7 | 2006-03-16 23:04:07 -0800 | [diff] [blame] | 423 | if (PagePrivate(page)) |
| 424 | try_to_release_page(page, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 425 | unlock_page(page); |
| 426 | } |
| 427 | } |
| 428 | } |
| 429 | |
| 430 | /** |
Rik van Riel | 68a22394 | 2008-10-18 20:26:23 -0700 | [diff] [blame] | 431 | * pagevec_swap_free - try to free swap space from the pages in a pagevec |
| 432 | * @pvec: pagevec with swapcache pages to free the swap space of |
| 433 | * |
| 434 | * The caller needs to hold an extra reference to each page and |
| 435 | * not hold the page lock on the pages. This function uses a |
| 436 | * trylock on the page lock so it may not always free the swap |
| 437 | * space associated with a page. |
| 438 | */ |
| 439 | void pagevec_swap_free(struct pagevec *pvec) |
| 440 | { |
| 441 | int i; |
| 442 | |
| 443 | for (i = 0; i < pagevec_count(pvec); i++) { |
| 444 | struct page *page = pvec->pages[i]; |
| 445 | |
| 446 | if (PageSwapCache(page) && trylock_page(page)) { |
| 447 | if (PageSwapCache(page)) |
| 448 | remove_exclusive_swap_page_ref(page); |
| 449 | unlock_page(page); |
| 450 | } |
| 451 | } |
| 452 | } |
| 453 | |
| 454 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 455 | * pagevec_lookup - gang pagecache lookup |
| 456 | * @pvec: Where the resulting pages are placed |
| 457 | * @mapping: The address_space to search |
| 458 | * @start: The starting page index |
| 459 | * @nr_pages: The maximum number of pages |
| 460 | * |
| 461 | * pagevec_lookup() will search for and return a group of up to @nr_pages pages |
| 462 | * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a |
| 463 | * reference against the pages in @pvec. |
| 464 | * |
| 465 | * The search returns a group of mapping-contiguous pages with ascending |
| 466 | * indexes. There may be holes in the indices due to not-present pages. |
| 467 | * |
| 468 | * pagevec_lookup() returns the number of pages which were found. |
| 469 | */ |
| 470 | unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping, |
| 471 | pgoff_t start, unsigned nr_pages) |
| 472 | { |
| 473 | pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages); |
| 474 | return pagevec_count(pvec); |
| 475 | } |
| 476 | |
Christoph Hellwig | 78539fd | 2006-01-11 20:47:41 +1100 | [diff] [blame] | 477 | EXPORT_SYMBOL(pagevec_lookup); |
| 478 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 479 | unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping, |
| 480 | pgoff_t *index, int tag, unsigned nr_pages) |
| 481 | { |
| 482 | pvec->nr = find_get_pages_tag(mapping, index, tag, |
| 483 | nr_pages, pvec->pages); |
| 484 | return pagevec_count(pvec); |
| 485 | } |
| 486 | |
Steve French | 7f28570 | 2005-11-01 10:22:55 -0800 | [diff] [blame] | 487 | EXPORT_SYMBOL(pagevec_lookup_tag); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 488 | |
| 489 | #ifdef CONFIG_SMP |
| 490 | /* |
| 491 | * We tolerate a little inaccuracy to avoid ping-ponging the counter between |
| 492 | * CPUs |
| 493 | */ |
| 494 | #define ACCT_THRESHOLD max(16, NR_CPUS * 2) |
| 495 | |
Vegard Nossum | f84f9504 | 2008-07-23 21:28:14 -0700 | [diff] [blame] | 496 | static DEFINE_PER_CPU(long, committed_space); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 497 | |
| 498 | void vm_acct_memory(long pages) |
| 499 | { |
| 500 | long *local; |
| 501 | |
| 502 | preempt_disable(); |
| 503 | local = &__get_cpu_var(committed_space); |
| 504 | *local += pages; |
| 505 | if (*local > ACCT_THRESHOLD || *local < -ACCT_THRESHOLD) { |
Alan Cox | 80119ef | 2008-05-23 13:04:31 -0700 | [diff] [blame] | 506 | atomic_long_add(*local, &vm_committed_space); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 507 | *local = 0; |
| 508 | } |
| 509 | preempt_enable(); |
| 510 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 511 | |
| 512 | #ifdef CONFIG_HOTPLUG_CPU |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 513 | |
| 514 | /* Drop the CPU's cached committed space back into the central pool. */ |
| 515 | static int cpu_swap_callback(struct notifier_block *nfb, |
| 516 | unsigned long action, |
| 517 | void *hcpu) |
| 518 | { |
| 519 | long *committed; |
| 520 | |
| 521 | committed = &per_cpu(committed_space, (long)hcpu); |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 522 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
Alan Cox | 80119ef | 2008-05-23 13:04:31 -0700 | [diff] [blame] | 523 | atomic_long_add(*committed, &vm_committed_space); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 524 | *committed = 0; |
Hisashi Hifumi | 902aaed | 2007-10-16 01:24:52 -0700 | [diff] [blame] | 525 | drain_cpu_pagevecs((long)hcpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 526 | } |
| 527 | return NOTIFY_OK; |
| 528 | } |
| 529 | #endif /* CONFIG_HOTPLUG_CPU */ |
| 530 | #endif /* CONFIG_SMP */ |
| 531 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 532 | /* |
| 533 | * Perform any setup for the swap system |
| 534 | */ |
| 535 | void __init swap_setup(void) |
| 536 | { |
| 537 | unsigned long megs = num_physpages >> (20 - PAGE_SHIFT); |
| 538 | |
Peter Zijlstra | e0bf68d | 2007-10-16 23:25:46 -0700 | [diff] [blame] | 539 | #ifdef CONFIG_SWAP |
| 540 | bdi_init(swapper_space.backing_dev_info); |
| 541 | #endif |
| 542 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 543 | /* Use a smaller cluster for small-memory machines */ |
| 544 | if (megs < 16) |
| 545 | page_cluster = 2; |
| 546 | else |
| 547 | page_cluster = 3; |
| 548 | /* |
| 549 | * Right now other parts of the system means that we |
| 550 | * _really_ don't want to cluster much more |
| 551 | */ |
Ingo Molnar | 0231606 | 2006-12-06 20:38:17 -0800 | [diff] [blame] | 552 | #ifdef CONFIG_HOTPLUG_CPU |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 553 | hotcpu_notifier(cpu_swap_callback, 0); |
Ingo Molnar | 0231606 | 2006-12-06 20:38:17 -0800 | [diff] [blame] | 554 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 555 | } |