Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/mm/vmscan.c |
| 3 | * |
| 4 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds |
| 5 | * |
| 6 | * Swap reorganised 29.12.95, Stephen Tweedie. |
| 7 | * kswapd added: 7.1.96 sct |
| 8 | * Removed kswapd_ctl limits, and swap out as many pages as needed |
| 9 | * to bring the system back to freepages.high: 2.4.97, Rik van Riel. |
| 10 | * Zone aware kswapd started 02/00, Kanoj Sarcar (kanoj@sgi.com). |
| 11 | * Multiqueue VM started 5.8.00, Rik van Riel. |
| 12 | */ |
| 13 | |
| 14 | #include <linux/mm.h> |
| 15 | #include <linux/module.h> |
| 16 | #include <linux/slab.h> |
| 17 | #include <linux/kernel_stat.h> |
| 18 | #include <linux/swap.h> |
| 19 | #include <linux/pagemap.h> |
| 20 | #include <linux/init.h> |
| 21 | #include <linux/highmem.h> |
Andrew Morton | e129b5c | 2006-09-27 01:50:00 -0700 | [diff] [blame] | 22 | #include <linux/vmstat.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 23 | #include <linux/file.h> |
| 24 | #include <linux/writeback.h> |
| 25 | #include <linux/blkdev.h> |
| 26 | #include <linux/buffer_head.h> /* for try_to_release_page(), |
| 27 | buffer_heads_over_limit */ |
| 28 | #include <linux/mm_inline.h> |
| 29 | #include <linux/pagevec.h> |
| 30 | #include <linux/backing-dev.h> |
| 31 | #include <linux/rmap.h> |
| 32 | #include <linux/topology.h> |
| 33 | #include <linux/cpu.h> |
| 34 | #include <linux/cpuset.h> |
| 35 | #include <linux/notifier.h> |
| 36 | #include <linux/rwsem.h> |
Rafael J. Wysocki | 248a030 | 2006-03-22 00:09:04 -0800 | [diff] [blame] | 37 | #include <linux/delay.h> |
Yasunori Goto | 3218ae1 | 2006-06-27 02:53:33 -0700 | [diff] [blame] | 38 | #include <linux/kthread.h> |
Nigel Cunningham | 7dfb710 | 2006-12-06 20:34:23 -0800 | [diff] [blame] | 39 | #include <linux/freezer.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 40 | #include <linux/memcontrol.h> |
Keika Kobayashi | 873b477 | 2008-07-25 01:48:52 -0700 | [diff] [blame] | 41 | #include <linux/delayacct.h> |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 42 | #include <linux/sysctl.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 43 | |
| 44 | #include <asm/tlbflush.h> |
| 45 | #include <asm/div64.h> |
| 46 | |
| 47 | #include <linux/swapops.h> |
| 48 | |
Nick Piggin | 0f8053a | 2006-03-22 00:08:33 -0800 | [diff] [blame] | 49 | #include "internal.h" |
| 50 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 51 | struct scan_control { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 52 | /* Incremented by the number of inactive pages that were scanned */ |
| 53 | unsigned long nr_scanned; |
| 54 | |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 55 | /* Number of pages freed so far during a call to shrink_zones() */ |
| 56 | unsigned long nr_reclaimed; |
| 57 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 58 | /* This context's GFP mask */ |
Al Viro | 6daa0e2 | 2005-10-21 03:18:50 -0400 | [diff] [blame] | 59 | gfp_t gfp_mask; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 60 | |
| 61 | int may_writepage; |
| 62 | |
Christoph Lameter | f1fd106 | 2006-01-18 17:42:30 -0800 | [diff] [blame] | 63 | /* Can pages be swapped as part of reclaim? */ |
| 64 | int may_swap; |
| 65 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 66 | /* This context's SWAP_CLUSTER_MAX. If freeing memory for |
| 67 | * suspend, we effectively ignore SWAP_CLUSTER_MAX. |
| 68 | * In this context, it doesn't matter that we scan the |
| 69 | * whole list at once. */ |
| 70 | int swap_cluster_max; |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 71 | |
| 72 | int swappiness; |
Nick Piggin | 408d854 | 2006-09-25 23:31:27 -0700 | [diff] [blame] | 73 | |
| 74 | int all_unreclaimable; |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 75 | |
| 76 | int order; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 77 | |
| 78 | /* Which cgroup do we reclaim from */ |
| 79 | struct mem_cgroup *mem_cgroup; |
| 80 | |
| 81 | /* Pluggable isolate pages callback */ |
| 82 | unsigned long (*isolate_pages)(unsigned long nr, struct list_head *dst, |
| 83 | unsigned long *scanned, int order, int mode, |
| 84 | struct zone *z, struct mem_cgroup *mem_cont, |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 85 | int active, int file); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 86 | }; |
| 87 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 88 | #define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru)) |
| 89 | |
| 90 | #ifdef ARCH_HAS_PREFETCH |
| 91 | #define prefetch_prev_lru_page(_page, _base, _field) \ |
| 92 | do { \ |
| 93 | if ((_page)->lru.prev != _base) { \ |
| 94 | struct page *prev; \ |
| 95 | \ |
| 96 | prev = lru_to_page(&(_page->lru)); \ |
| 97 | prefetch(&prev->_field); \ |
| 98 | } \ |
| 99 | } while (0) |
| 100 | #else |
| 101 | #define prefetch_prev_lru_page(_page, _base, _field) do { } while (0) |
| 102 | #endif |
| 103 | |
| 104 | #ifdef ARCH_HAS_PREFETCHW |
| 105 | #define prefetchw_prev_lru_page(_page, _base, _field) \ |
| 106 | do { \ |
| 107 | if ((_page)->lru.prev != _base) { \ |
| 108 | struct page *prev; \ |
| 109 | \ |
| 110 | prev = lru_to_page(&(_page->lru)); \ |
| 111 | prefetchw(&prev->_field); \ |
| 112 | } \ |
| 113 | } while (0) |
| 114 | #else |
| 115 | #define prefetchw_prev_lru_page(_page, _base, _field) do { } while (0) |
| 116 | #endif |
| 117 | |
| 118 | /* |
| 119 | * From 0 .. 100. Higher means more swappy. |
| 120 | */ |
| 121 | int vm_swappiness = 60; |
Andrew Morton | bd1e22b | 2006-06-23 02:03:47 -0700 | [diff] [blame] | 122 | long vm_total_pages; /* The total number of pages which the VM controls */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 123 | |
| 124 | static LIST_HEAD(shrinker_list); |
| 125 | static DECLARE_RWSEM(shrinker_rwsem); |
| 126 | |
Balbir Singh | 00f0b82 | 2008-03-04 14:28:39 -0800 | [diff] [blame] | 127 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
KAMEZAWA Hiroyuki | 91a4547 | 2008-02-07 00:14:29 -0800 | [diff] [blame] | 128 | #define scan_global_lru(sc) (!(sc)->mem_cgroup) |
| 129 | #else |
| 130 | #define scan_global_lru(sc) (1) |
| 131 | #endif |
| 132 | |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 133 | static struct zone_reclaim_stat *get_reclaim_stat(struct zone *zone, |
| 134 | struct scan_control *sc) |
| 135 | { |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 136 | if (!scan_global_lru(sc)) |
| 137 | return mem_cgroup_get_reclaim_stat(sc->mem_cgroup, zone); |
| 138 | |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 139 | return &zone->reclaim_stat; |
| 140 | } |
| 141 | |
KOSAKI Motohiro | c9f299d | 2009-01-07 18:08:16 -0800 | [diff] [blame] | 142 | static unsigned long zone_nr_pages(struct zone *zone, struct scan_control *sc, |
| 143 | enum lru_list lru) |
| 144 | { |
KOSAKI Motohiro | a3d8e05 | 2009-01-07 18:08:19 -0800 | [diff] [blame] | 145 | if (!scan_global_lru(sc)) |
| 146 | return mem_cgroup_zone_nr_pages(sc->mem_cgroup, zone, lru); |
| 147 | |
KOSAKI Motohiro | c9f299d | 2009-01-07 18:08:16 -0800 | [diff] [blame] | 148 | return zone_page_state(zone, NR_LRU_BASE + lru); |
| 149 | } |
| 150 | |
| 151 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 152 | /* |
| 153 | * Add a shrinker callback to be called from the vm |
| 154 | */ |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 155 | void register_shrinker(struct shrinker *shrinker) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 156 | { |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 157 | shrinker->nr = 0; |
| 158 | down_write(&shrinker_rwsem); |
| 159 | list_add_tail(&shrinker->list, &shrinker_list); |
| 160 | up_write(&shrinker_rwsem); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 161 | } |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 162 | EXPORT_SYMBOL(register_shrinker); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 163 | |
| 164 | /* |
| 165 | * Remove one |
| 166 | */ |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 167 | void unregister_shrinker(struct shrinker *shrinker) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 168 | { |
| 169 | down_write(&shrinker_rwsem); |
| 170 | list_del(&shrinker->list); |
| 171 | up_write(&shrinker_rwsem); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 172 | } |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 173 | EXPORT_SYMBOL(unregister_shrinker); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 174 | |
| 175 | #define SHRINK_BATCH 128 |
| 176 | /* |
| 177 | * Call the shrink functions to age shrinkable caches |
| 178 | * |
| 179 | * Here we assume it costs one seek to replace a lru page and that it also |
| 180 | * takes a seek to recreate a cache object. With this in mind we age equal |
| 181 | * percentages of the lru and ageable caches. This should balance the seeks |
| 182 | * generated by these structures. |
| 183 | * |
Simon Arlott | 183ff22 | 2007-10-20 01:27:18 +0200 | [diff] [blame] | 184 | * If the vm encountered mapped pages on the LRU it increase the pressure on |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 185 | * slab to avoid swapping. |
| 186 | * |
| 187 | * We do weird things to avoid (scanned*seeks*entries) overflowing 32 bits. |
| 188 | * |
| 189 | * `lru_pages' represents the number of on-LRU pages in all the zones which |
| 190 | * are eligible for the caller's allocation attempt. It is used for balancing |
| 191 | * slab reclaim versus page reclaim. |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 192 | * |
| 193 | * Returns the number of slab objects which we shrunk. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 194 | */ |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 195 | unsigned long shrink_slab(unsigned long scanned, gfp_t gfp_mask, |
| 196 | unsigned long lru_pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 197 | { |
| 198 | struct shrinker *shrinker; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 199 | unsigned long ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 200 | |
| 201 | if (scanned == 0) |
| 202 | scanned = SWAP_CLUSTER_MAX; |
| 203 | |
| 204 | if (!down_read_trylock(&shrinker_rwsem)) |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 205 | return 1; /* Assume we'll be able to shrink next time */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 206 | |
| 207 | list_for_each_entry(shrinker, &shrinker_list, list) { |
| 208 | unsigned long long delta; |
| 209 | unsigned long total_scan; |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 210 | unsigned long max_pass = (*shrinker->shrink)(0, gfp_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 211 | |
| 212 | delta = (4 * scanned) / shrinker->seeks; |
Andrea Arcangeli | ea164d7 | 2005-11-28 13:44:15 -0800 | [diff] [blame] | 213 | delta *= max_pass; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 214 | do_div(delta, lru_pages + 1); |
| 215 | shrinker->nr += delta; |
Andrea Arcangeli | ea164d7 | 2005-11-28 13:44:15 -0800 | [diff] [blame] | 216 | if (shrinker->nr < 0) { |
| 217 | printk(KERN_ERR "%s: nr=%ld\n", |
Harvey Harrison | d40cee2 | 2008-04-30 00:55:07 -0700 | [diff] [blame] | 218 | __func__, shrinker->nr); |
Andrea Arcangeli | ea164d7 | 2005-11-28 13:44:15 -0800 | [diff] [blame] | 219 | shrinker->nr = max_pass; |
| 220 | } |
| 221 | |
| 222 | /* |
| 223 | * Avoid risking looping forever due to too large nr value: |
| 224 | * never try to free more than twice the estimate number of |
| 225 | * freeable entries. |
| 226 | */ |
| 227 | if (shrinker->nr > max_pass * 2) |
| 228 | shrinker->nr = max_pass * 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 229 | |
| 230 | total_scan = shrinker->nr; |
| 231 | shrinker->nr = 0; |
| 232 | |
| 233 | while (total_scan >= SHRINK_BATCH) { |
| 234 | long this_scan = SHRINK_BATCH; |
| 235 | int shrink_ret; |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 236 | int nr_before; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 237 | |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 238 | nr_before = (*shrinker->shrink)(0, gfp_mask); |
| 239 | shrink_ret = (*shrinker->shrink)(this_scan, gfp_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 240 | if (shrink_ret == -1) |
| 241 | break; |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 242 | if (shrink_ret < nr_before) |
| 243 | ret += nr_before - shrink_ret; |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 244 | count_vm_events(SLABS_SCANNED, this_scan); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 245 | total_scan -= this_scan; |
| 246 | |
| 247 | cond_resched(); |
| 248 | } |
| 249 | |
| 250 | shrinker->nr += total_scan; |
| 251 | } |
| 252 | up_read(&shrinker_rwsem); |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 253 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 254 | } |
| 255 | |
| 256 | /* Called without lock on whether page is mapped, so answer is unstable */ |
| 257 | static inline int page_mapping_inuse(struct page *page) |
| 258 | { |
| 259 | struct address_space *mapping; |
| 260 | |
| 261 | /* Page is in somebody's page tables. */ |
| 262 | if (page_mapped(page)) |
| 263 | return 1; |
| 264 | |
| 265 | /* Be more reluctant to reclaim swapcache than pagecache */ |
| 266 | if (PageSwapCache(page)) |
| 267 | return 1; |
| 268 | |
| 269 | mapping = page_mapping(page); |
| 270 | if (!mapping) |
| 271 | return 0; |
| 272 | |
| 273 | /* File is mmap'd by somebody? */ |
| 274 | return mapping_mapped(mapping); |
| 275 | } |
| 276 | |
| 277 | static inline int is_page_cache_freeable(struct page *page) |
| 278 | { |
| 279 | return page_count(page) - !!PagePrivate(page) == 2; |
| 280 | } |
| 281 | |
| 282 | static int may_write_to_queue(struct backing_dev_info *bdi) |
| 283 | { |
Christoph Lameter | 930d915 | 2006-01-08 01:00:47 -0800 | [diff] [blame] | 284 | if (current->flags & PF_SWAPWRITE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 285 | return 1; |
| 286 | if (!bdi_write_congested(bdi)) |
| 287 | return 1; |
| 288 | if (bdi == current->backing_dev_info) |
| 289 | return 1; |
| 290 | return 0; |
| 291 | } |
| 292 | |
| 293 | /* |
| 294 | * We detected a synchronous write error writing a page out. Probably |
| 295 | * -ENOSPC. We need to propagate that into the address_space for a subsequent |
| 296 | * fsync(), msync() or close(). |
| 297 | * |
| 298 | * The tricky part is that after writepage we cannot touch the mapping: nothing |
| 299 | * prevents it from being freed up. But we have a ref on the page and once |
| 300 | * that page is locked, the mapping is pinned. |
| 301 | * |
| 302 | * We're allowed to run sleeping lock_page() here because we know the caller has |
| 303 | * __GFP_FS. |
| 304 | */ |
| 305 | static void handle_write_error(struct address_space *mapping, |
| 306 | struct page *page, int error) |
| 307 | { |
| 308 | lock_page(page); |
Guillaume Chazarain | 3e9f45b | 2007-05-08 00:23:25 -0700 | [diff] [blame] | 309 | if (page_mapping(page) == mapping) |
| 310 | mapping_set_error(mapping, error); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 311 | unlock_page(page); |
| 312 | } |
| 313 | |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 314 | /* Request for sync pageout. */ |
| 315 | enum pageout_io { |
| 316 | PAGEOUT_IO_ASYNC, |
| 317 | PAGEOUT_IO_SYNC, |
| 318 | }; |
| 319 | |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 320 | /* possible outcome of pageout() */ |
| 321 | typedef enum { |
| 322 | /* failed to write page out, page is locked */ |
| 323 | PAGE_KEEP, |
| 324 | /* move page to the active list, page is locked */ |
| 325 | PAGE_ACTIVATE, |
| 326 | /* page has been sent to the disk successfully, page is unlocked */ |
| 327 | PAGE_SUCCESS, |
| 328 | /* page is clean and locked */ |
| 329 | PAGE_CLEAN, |
| 330 | } pageout_t; |
| 331 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 332 | /* |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 333 | * pageout is called by shrink_page_list() for each dirty page. |
| 334 | * Calls ->writepage(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 335 | */ |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 336 | static pageout_t pageout(struct page *page, struct address_space *mapping, |
| 337 | enum pageout_io sync_writeback) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 338 | { |
| 339 | /* |
| 340 | * If the page is dirty, only perform writeback if that write |
| 341 | * will be non-blocking. To prevent this allocation from being |
| 342 | * stalled by pagecache activity. But note that there may be |
| 343 | * stalls if we need to run get_block(). We could test |
| 344 | * PagePrivate for that. |
| 345 | * |
| 346 | * If this process is currently in generic_file_write() against |
| 347 | * this page's queue, we can perform writeback even if that |
| 348 | * will block. |
| 349 | * |
| 350 | * If the page is swapcache, write it back even if that would |
| 351 | * block, for some throttling. This happens by accident, because |
| 352 | * swap_backing_dev_info is bust: it doesn't reflect the |
| 353 | * congestion state of the swapdevs. Easy to fix, if needed. |
| 354 | * See swapfile.c:page_queue_congested(). |
| 355 | */ |
| 356 | if (!is_page_cache_freeable(page)) |
| 357 | return PAGE_KEEP; |
| 358 | if (!mapping) { |
| 359 | /* |
| 360 | * Some data journaling orphaned pages can have |
| 361 | * page->mapping == NULL while being dirty with clean buffers. |
| 362 | */ |
akpm@osdl.org | 323aca6 | 2005-04-16 15:24:06 -0700 | [diff] [blame] | 363 | if (PagePrivate(page)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 364 | if (try_to_free_buffers(page)) { |
| 365 | ClearPageDirty(page); |
Harvey Harrison | d40cee2 | 2008-04-30 00:55:07 -0700 | [diff] [blame] | 366 | printk("%s: orphaned page\n", __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 367 | return PAGE_CLEAN; |
| 368 | } |
| 369 | } |
| 370 | return PAGE_KEEP; |
| 371 | } |
| 372 | if (mapping->a_ops->writepage == NULL) |
| 373 | return PAGE_ACTIVATE; |
| 374 | if (!may_write_to_queue(mapping->backing_dev_info)) |
| 375 | return PAGE_KEEP; |
| 376 | |
| 377 | if (clear_page_dirty_for_io(page)) { |
| 378 | int res; |
| 379 | struct writeback_control wbc = { |
| 380 | .sync_mode = WB_SYNC_NONE, |
| 381 | .nr_to_write = SWAP_CLUSTER_MAX, |
OGAWA Hirofumi | 111ebb6 | 2006-06-23 02:03:26 -0700 | [diff] [blame] | 382 | .range_start = 0, |
| 383 | .range_end = LLONG_MAX, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 384 | .nonblocking = 1, |
| 385 | .for_reclaim = 1, |
| 386 | }; |
| 387 | |
| 388 | SetPageReclaim(page); |
| 389 | res = mapping->a_ops->writepage(page, &wbc); |
| 390 | if (res < 0) |
| 391 | handle_write_error(mapping, page, res); |
Zach Brown | 994fc28c | 2005-12-15 14:28:17 -0800 | [diff] [blame] | 392 | if (res == AOP_WRITEPAGE_ACTIVATE) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 393 | ClearPageReclaim(page); |
| 394 | return PAGE_ACTIVATE; |
| 395 | } |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 396 | |
| 397 | /* |
| 398 | * Wait on writeback if requested to. This happens when |
| 399 | * direct reclaiming a large contiguous area and the |
| 400 | * first attempt to free a range of pages fails. |
| 401 | */ |
| 402 | if (PageWriteback(page) && sync_writeback == PAGEOUT_IO_SYNC) |
| 403 | wait_on_page_writeback(page); |
| 404 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 405 | if (!PageWriteback(page)) { |
| 406 | /* synchronous write or broken a_ops? */ |
| 407 | ClearPageReclaim(page); |
| 408 | } |
Andrew Morton | e129b5c | 2006-09-27 01:50:00 -0700 | [diff] [blame] | 409 | inc_zone_page_state(page, NR_VMSCAN_WRITE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 410 | return PAGE_SUCCESS; |
| 411 | } |
| 412 | |
| 413 | return PAGE_CLEAN; |
| 414 | } |
| 415 | |
Andrew Morton | a649fd9 | 2006-10-17 00:09:36 -0700 | [diff] [blame] | 416 | /* |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 417 | * Same as remove_mapping, but if the page is removed from the mapping, it |
| 418 | * gets returned with a refcount of 0. |
Andrew Morton | a649fd9 | 2006-10-17 00:09:36 -0700 | [diff] [blame] | 419 | */ |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 420 | static int __remove_mapping(struct address_space *mapping, struct page *page) |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 421 | { |
Nick Piggin | 28e4d96 | 2006-09-25 23:31:23 -0700 | [diff] [blame] | 422 | BUG_ON(!PageLocked(page)); |
| 423 | BUG_ON(mapping != page_mapping(page)); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 424 | |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 425 | spin_lock_irq(&mapping->tree_lock); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 426 | /* |
Nick Piggin | 0fd0e6b | 2006-09-27 01:50:02 -0700 | [diff] [blame] | 427 | * The non racy check for a busy page. |
| 428 | * |
| 429 | * Must be careful with the order of the tests. When someone has |
| 430 | * a ref to the page, it may be possible that they dirty it then |
| 431 | * drop the reference. So if PageDirty is tested before page_count |
| 432 | * here, then the following race may occur: |
| 433 | * |
| 434 | * get_user_pages(&page); |
| 435 | * [user mapping goes away] |
| 436 | * write_to(page); |
| 437 | * !PageDirty(page) [good] |
| 438 | * SetPageDirty(page); |
| 439 | * put_page(page); |
| 440 | * !page_count(page) [good, discard it] |
| 441 | * |
| 442 | * [oops, our write_to data is lost] |
| 443 | * |
| 444 | * Reversing the order of the tests ensures such a situation cannot |
| 445 | * escape unnoticed. The smp_rmb is needed to ensure the page->flags |
| 446 | * load is not satisfied before that of page->_count. |
| 447 | * |
| 448 | * Note that if SetPageDirty is always performed via set_page_dirty, |
| 449 | * and thus under tree_lock, then this ordering is not required. |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 450 | */ |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 451 | if (!page_freeze_refs(page, 2)) |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 452 | goto cannot_free; |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 453 | /* note: atomic_cmpxchg in page_freeze_refs provides the smp_rmb */ |
| 454 | if (unlikely(PageDirty(page))) { |
| 455 | page_unfreeze_refs(page, 2); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 456 | goto cannot_free; |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 457 | } |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 458 | |
| 459 | if (PageSwapCache(page)) { |
| 460 | swp_entry_t swap = { .val = page_private(page) }; |
| 461 | __delete_from_swap_cache(page); |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 462 | spin_unlock_irq(&mapping->tree_lock); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 463 | swap_free(swap); |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 464 | } else { |
| 465 | __remove_from_page_cache(page); |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 466 | spin_unlock_irq(&mapping->tree_lock); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 467 | } |
| 468 | |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 469 | return 1; |
| 470 | |
| 471 | cannot_free: |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 472 | spin_unlock_irq(&mapping->tree_lock); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 473 | return 0; |
| 474 | } |
| 475 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 476 | /* |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 477 | * Attempt to detach a locked page from its ->mapping. If it is dirty or if |
| 478 | * someone else has a ref on the page, abort and return 0. If it was |
| 479 | * successfully detached, return 1. Assumes the caller has a single ref on |
| 480 | * this page. |
| 481 | */ |
| 482 | int remove_mapping(struct address_space *mapping, struct page *page) |
| 483 | { |
| 484 | if (__remove_mapping(mapping, page)) { |
| 485 | /* |
| 486 | * Unfreezing the refcount with 1 rather than 2 effectively |
| 487 | * drops the pagecache ref for us without requiring another |
| 488 | * atomic operation. |
| 489 | */ |
| 490 | page_unfreeze_refs(page, 1); |
| 491 | return 1; |
| 492 | } |
| 493 | return 0; |
| 494 | } |
| 495 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 496 | /** |
| 497 | * putback_lru_page - put previously isolated page onto appropriate LRU list |
| 498 | * @page: page to be put back to appropriate lru list |
| 499 | * |
| 500 | * Add previously isolated @page to appropriate LRU list. |
| 501 | * Page may still be unevictable for other reasons. |
| 502 | * |
| 503 | * lru_lock must not be held, interrupts must be enabled. |
| 504 | */ |
| 505 | #ifdef CONFIG_UNEVICTABLE_LRU |
| 506 | void putback_lru_page(struct page *page) |
| 507 | { |
| 508 | int lru; |
| 509 | int active = !!TestClearPageActive(page); |
Lee Schermerhorn | bbfd28e | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 510 | int was_unevictable = PageUnevictable(page); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 511 | |
| 512 | VM_BUG_ON(PageLRU(page)); |
| 513 | |
| 514 | redo: |
| 515 | ClearPageUnevictable(page); |
| 516 | |
| 517 | if (page_evictable(page, NULL)) { |
| 518 | /* |
| 519 | * For evictable pages, we can use the cache. |
| 520 | * In event of a race, worst case is we end up with an |
| 521 | * unevictable page on [in]active list. |
| 522 | * We know how to handle that. |
| 523 | */ |
| 524 | lru = active + page_is_file_cache(page); |
| 525 | lru_cache_add_lru(page, lru); |
| 526 | } else { |
| 527 | /* |
| 528 | * Put unevictable pages directly on zone's unevictable |
| 529 | * list. |
| 530 | */ |
| 531 | lru = LRU_UNEVICTABLE; |
| 532 | add_page_to_unevictable_list(page); |
| 533 | } |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 534 | |
| 535 | /* |
| 536 | * page's status can change while we move it among lru. If an evictable |
| 537 | * page is on unevictable list, it never be freed. To avoid that, |
| 538 | * check after we added it to the list, again. |
| 539 | */ |
| 540 | if (lru == LRU_UNEVICTABLE && page_evictable(page, NULL)) { |
| 541 | if (!isolate_lru_page(page)) { |
| 542 | put_page(page); |
| 543 | goto redo; |
| 544 | } |
| 545 | /* This means someone else dropped this page from LRU |
| 546 | * So, it will be freed or putback to LRU again. There is |
| 547 | * nothing to do here. |
| 548 | */ |
| 549 | } |
| 550 | |
Lee Schermerhorn | bbfd28e | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 551 | if (was_unevictable && lru != LRU_UNEVICTABLE) |
| 552 | count_vm_event(UNEVICTABLE_PGRESCUED); |
| 553 | else if (!was_unevictable && lru == LRU_UNEVICTABLE) |
| 554 | count_vm_event(UNEVICTABLE_PGCULLED); |
| 555 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 556 | put_page(page); /* drop ref from isolate */ |
| 557 | } |
| 558 | |
| 559 | #else /* CONFIG_UNEVICTABLE_LRU */ |
| 560 | |
| 561 | void putback_lru_page(struct page *page) |
| 562 | { |
| 563 | int lru; |
| 564 | VM_BUG_ON(PageLRU(page)); |
| 565 | |
| 566 | lru = !!TestClearPageActive(page) + page_is_file_cache(page); |
| 567 | lru_cache_add_lru(page, lru); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 568 | put_page(page); |
| 569 | } |
| 570 | #endif /* CONFIG_UNEVICTABLE_LRU */ |
| 571 | |
| 572 | |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 573 | /* |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 574 | * shrink_page_list() returns the number of reclaimed pages |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 575 | */ |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 576 | static unsigned long shrink_page_list(struct list_head *page_list, |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 577 | struct scan_control *sc, |
| 578 | enum pageout_io sync_writeback) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 579 | { |
| 580 | LIST_HEAD(ret_pages); |
| 581 | struct pagevec freed_pvec; |
| 582 | int pgactivate = 0; |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 583 | unsigned long nr_reclaimed = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 584 | |
| 585 | cond_resched(); |
| 586 | |
| 587 | pagevec_init(&freed_pvec, 1); |
| 588 | while (!list_empty(page_list)) { |
| 589 | struct address_space *mapping; |
| 590 | struct page *page; |
| 591 | int may_enter_fs; |
| 592 | int referenced; |
| 593 | |
| 594 | cond_resched(); |
| 595 | |
| 596 | page = lru_to_page(page_list); |
| 597 | list_del(&page->lru); |
| 598 | |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 599 | if (!trylock_page(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 600 | goto keep; |
| 601 | |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 602 | VM_BUG_ON(PageActive(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 603 | |
| 604 | sc->nr_scanned++; |
Christoph Lameter | 80e4342 | 2006-02-11 17:55:53 -0800 | [diff] [blame] | 605 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 606 | if (unlikely(!page_evictable(page, NULL))) |
| 607 | goto cull_mlocked; |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 608 | |
Christoph Lameter | 80e4342 | 2006-02-11 17:55:53 -0800 | [diff] [blame] | 609 | if (!sc->may_swap && page_mapped(page)) |
| 610 | goto keep_locked; |
| 611 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 612 | /* Double the slab pressure for mapped and swapcache pages */ |
| 613 | if (page_mapped(page) || PageSwapCache(page)) |
| 614 | sc->nr_scanned++; |
| 615 | |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 616 | may_enter_fs = (sc->gfp_mask & __GFP_FS) || |
| 617 | (PageSwapCache(page) && (sc->gfp_mask & __GFP_IO)); |
| 618 | |
| 619 | if (PageWriteback(page)) { |
| 620 | /* |
| 621 | * Synchronous reclaim is performed in two passes, |
| 622 | * first an asynchronous pass over the list to |
| 623 | * start parallel writeback, and a second synchronous |
| 624 | * pass to wait for the IO to complete. Wait here |
| 625 | * for any page for which writeback has already |
| 626 | * started. |
| 627 | */ |
| 628 | if (sync_writeback == PAGEOUT_IO_SYNC && may_enter_fs) |
| 629 | wait_on_page_writeback(page); |
Andrew Morton | 4dd4b92 | 2008-03-24 12:29:52 -0700 | [diff] [blame] | 630 | else |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 631 | goto keep_locked; |
| 632 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 633 | |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 634 | referenced = page_referenced(page, 1, sc->mem_cgroup); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 635 | /* In active use or really unfreeable? Activate it. */ |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 636 | if (sc->order <= PAGE_ALLOC_COSTLY_ORDER && |
| 637 | referenced && page_mapping_inuse(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 638 | goto activate_locked; |
| 639 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 640 | /* |
| 641 | * Anonymous process memory has backing store? |
| 642 | * Try to allocate it some swap space here. |
| 643 | */ |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 644 | if (PageAnon(page) && !PageSwapCache(page)) { |
Hugh Dickins | 63eb6b9 | 2008-11-19 15:36:37 -0800 | [diff] [blame] | 645 | if (!(sc->gfp_mask & __GFP_IO)) |
| 646 | goto keep_locked; |
Hugh Dickins | ac47b00 | 2009-01-06 14:39:39 -0800 | [diff] [blame] | 647 | if (!add_to_swap(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 648 | goto activate_locked; |
Hugh Dickins | 63eb6b9 | 2008-11-19 15:36:37 -0800 | [diff] [blame] | 649 | may_enter_fs = 1; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 650 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 651 | |
| 652 | mapping = page_mapping(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 653 | |
| 654 | /* |
| 655 | * The page is mapped into the page tables of one or more |
| 656 | * processes. Try to unmap it here. |
| 657 | */ |
| 658 | if (page_mapped(page) && mapping) { |
Christoph Lameter | a48d07a | 2006-02-01 03:05:38 -0800 | [diff] [blame] | 659 | switch (try_to_unmap(page, 0)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 660 | case SWAP_FAIL: |
| 661 | goto activate_locked; |
| 662 | case SWAP_AGAIN: |
| 663 | goto keep_locked; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 664 | case SWAP_MLOCK: |
| 665 | goto cull_mlocked; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 666 | case SWAP_SUCCESS: |
| 667 | ; /* try to free the page below */ |
| 668 | } |
| 669 | } |
| 670 | |
| 671 | if (PageDirty(page)) { |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 672 | if (sc->order <= PAGE_ALLOC_COSTLY_ORDER && referenced) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 673 | goto keep_locked; |
Andrew Morton | 4dd4b92 | 2008-03-24 12:29:52 -0700 | [diff] [blame] | 674 | if (!may_enter_fs) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 675 | goto keep_locked; |
Christoph Lameter | 52a8363 | 2006-02-01 03:05:28 -0800 | [diff] [blame] | 676 | if (!sc->may_writepage) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 677 | goto keep_locked; |
| 678 | |
| 679 | /* Page is dirty, try to write it out here */ |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 680 | switch (pageout(page, mapping, sync_writeback)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 681 | case PAGE_KEEP: |
| 682 | goto keep_locked; |
| 683 | case PAGE_ACTIVATE: |
| 684 | goto activate_locked; |
| 685 | case PAGE_SUCCESS: |
Andrew Morton | 4dd4b92 | 2008-03-24 12:29:52 -0700 | [diff] [blame] | 686 | if (PageWriteback(page) || PageDirty(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 687 | goto keep; |
| 688 | /* |
| 689 | * A synchronous write - probably a ramdisk. Go |
| 690 | * ahead and try to reclaim the page. |
| 691 | */ |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 692 | if (!trylock_page(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 693 | goto keep; |
| 694 | if (PageDirty(page) || PageWriteback(page)) |
| 695 | goto keep_locked; |
| 696 | mapping = page_mapping(page); |
| 697 | case PAGE_CLEAN: |
| 698 | ; /* try to free the page below */ |
| 699 | } |
| 700 | } |
| 701 | |
| 702 | /* |
| 703 | * If the page has buffers, try to free the buffer mappings |
| 704 | * associated with this page. If we succeed we try to free |
| 705 | * the page as well. |
| 706 | * |
| 707 | * We do this even if the page is PageDirty(). |
| 708 | * try_to_release_page() does not perform I/O, but it is |
| 709 | * possible for a page to have PageDirty set, but it is actually |
| 710 | * clean (all its buffers are clean). This happens if the |
| 711 | * buffers were written out directly, with submit_bh(). ext3 |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 712 | * will do this, as well as the blockdev mapping. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 713 | * try_to_release_page() will discover that cleanness and will |
| 714 | * drop the buffers and mark the page clean - it can be freed. |
| 715 | * |
| 716 | * Rarely, pages can have buffers and no ->mapping. These are |
| 717 | * the pages which were not successfully invalidated in |
| 718 | * truncate_complete_page(). We try to drop those buffers here |
| 719 | * and if that worked, and the page is no longer mapped into |
| 720 | * process address space (page_count == 1) it can be freed. |
| 721 | * Otherwise, leave the page on the LRU so it is swappable. |
| 722 | */ |
| 723 | if (PagePrivate(page)) { |
| 724 | if (!try_to_release_page(page, sc->gfp_mask)) |
| 725 | goto activate_locked; |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 726 | if (!mapping && page_count(page) == 1) { |
| 727 | unlock_page(page); |
| 728 | if (put_page_testzero(page)) |
| 729 | goto free_it; |
| 730 | else { |
| 731 | /* |
| 732 | * rare race with speculative reference. |
| 733 | * the speculative reference will free |
| 734 | * this page shortly, so we may |
| 735 | * increment nr_reclaimed here (and |
| 736 | * leave it off the LRU). |
| 737 | */ |
| 738 | nr_reclaimed++; |
| 739 | continue; |
| 740 | } |
| 741 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 742 | } |
| 743 | |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 744 | if (!mapping || !__remove_mapping(mapping, page)) |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 745 | goto keep_locked; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 746 | |
Nick Piggin | a978d6f | 2008-10-18 20:26:58 -0700 | [diff] [blame] | 747 | /* |
| 748 | * At this point, we have no other references and there is |
| 749 | * no way to pick any more up (removed from LRU, removed |
| 750 | * from pagecache). Can use non-atomic bitops now (and |
| 751 | * we obviously don't have to worry about waking up a process |
| 752 | * waiting on the page lock, because there are no references. |
| 753 | */ |
| 754 | __clear_page_locked(page); |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 755 | free_it: |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 756 | nr_reclaimed++; |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 757 | if (!pagevec_add(&freed_pvec, page)) { |
| 758 | __pagevec_free(&freed_pvec); |
| 759 | pagevec_reinit(&freed_pvec); |
| 760 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 761 | continue; |
| 762 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 763 | cull_mlocked: |
Hugh Dickins | 63d6c5a | 2009-01-06 14:39:38 -0800 | [diff] [blame] | 764 | if (PageSwapCache(page)) |
| 765 | try_to_free_swap(page); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 766 | unlock_page(page); |
| 767 | putback_lru_page(page); |
| 768 | continue; |
| 769 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 770 | activate_locked: |
Rik van Riel | 68a22394 | 2008-10-18 20:26:23 -0700 | [diff] [blame] | 771 | /* Not a candidate for swapping, so reclaim swap space. */ |
| 772 | if (PageSwapCache(page) && vm_swap_full()) |
Hugh Dickins | a2c43ee | 2009-01-06 14:39:36 -0800 | [diff] [blame] | 773 | try_to_free_swap(page); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 774 | VM_BUG_ON(PageActive(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 775 | SetPageActive(page); |
| 776 | pgactivate++; |
| 777 | keep_locked: |
| 778 | unlock_page(page); |
| 779 | keep: |
| 780 | list_add(&page->lru, &ret_pages); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 781 | VM_BUG_ON(PageLRU(page) || PageUnevictable(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 782 | } |
| 783 | list_splice(&ret_pages, page_list); |
| 784 | if (pagevec_count(&freed_pvec)) |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 785 | __pagevec_free(&freed_pvec); |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 786 | count_vm_events(PGACTIVATE, pgactivate); |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 787 | return nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 788 | } |
| 789 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 790 | /* LRU Isolation modes. */ |
| 791 | #define ISOLATE_INACTIVE 0 /* Isolate inactive pages. */ |
| 792 | #define ISOLATE_ACTIVE 1 /* Isolate active pages. */ |
| 793 | #define ISOLATE_BOTH 2 /* Isolate both active and inactive pages. */ |
| 794 | |
| 795 | /* |
| 796 | * Attempt to remove the specified page from its LRU. Only take this page |
| 797 | * if it is of the appropriate PageActive status. Pages which are being |
| 798 | * freed elsewhere are also ignored. |
| 799 | * |
| 800 | * page: page to consider |
| 801 | * mode: one of the LRU isolation modes defined above |
| 802 | * |
| 803 | * returns 0 on success, -ve errno on failure. |
| 804 | */ |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 805 | int __isolate_lru_page(struct page *page, int mode, int file) |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 806 | { |
| 807 | int ret = -EINVAL; |
| 808 | |
| 809 | /* Only take pages on the LRU. */ |
| 810 | if (!PageLRU(page)) |
| 811 | return ret; |
| 812 | |
| 813 | /* |
| 814 | * When checking the active state, we need to be sure we are |
| 815 | * dealing with comparible boolean values. Take the logical not |
| 816 | * of each. |
| 817 | */ |
| 818 | if (mode != ISOLATE_BOTH && (!PageActive(page) != !mode)) |
| 819 | return ret; |
| 820 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 821 | if (mode != ISOLATE_BOTH && (!page_is_file_cache(page) != !file)) |
| 822 | return ret; |
| 823 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 824 | /* |
| 825 | * When this function is being called for lumpy reclaim, we |
| 826 | * initially look into all LRU pages, active, inactive and |
| 827 | * unevictable; only give shrink_page_list evictable pages. |
| 828 | */ |
| 829 | if (PageUnevictable(page)) |
| 830 | return ret; |
| 831 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 832 | ret = -EBUSY; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 833 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 834 | if (likely(get_page_unless_zero(page))) { |
| 835 | /* |
| 836 | * Be careful not to clear PageLRU until after we're |
| 837 | * sure the page is not being freed elsewhere -- the |
| 838 | * page release code relies on it. |
| 839 | */ |
| 840 | ClearPageLRU(page); |
| 841 | ret = 0; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 842 | mem_cgroup_del_lru(page); |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 843 | } |
| 844 | |
| 845 | return ret; |
| 846 | } |
| 847 | |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 848 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 849 | * zone->lru_lock is heavily contended. Some of the functions that |
| 850 | * shrink the lists perform better by taking out a batch of pages |
| 851 | * and working on them outside the LRU lock. |
| 852 | * |
| 853 | * For pagecache intensive workloads, this function is the hottest |
| 854 | * spot in the kernel (apart from copy_*_user functions). |
| 855 | * |
| 856 | * Appropriate locks must be held before calling this function. |
| 857 | * |
| 858 | * @nr_to_scan: The number of pages to look through on the list. |
| 859 | * @src: The LRU list to pull pages off. |
| 860 | * @dst: The temp list to put pages on to. |
| 861 | * @scanned: The number of pages that were scanned. |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 862 | * @order: The caller's attempted allocation order |
| 863 | * @mode: One of the LRU isolation modes |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 864 | * @file: True [1] if isolating file [!anon] pages |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 865 | * |
| 866 | * returns how many pages were moved onto *@dst. |
| 867 | */ |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 868 | static unsigned long isolate_lru_pages(unsigned long nr_to_scan, |
| 869 | struct list_head *src, struct list_head *dst, |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 870 | unsigned long *scanned, int order, int mode, int file) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 871 | { |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 872 | unsigned long nr_taken = 0; |
Wu Fengguang | c9b02d9 | 2006-03-22 00:08:23 -0800 | [diff] [blame] | 873 | unsigned long scan; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 874 | |
Wu Fengguang | c9b02d9 | 2006-03-22 00:08:23 -0800 | [diff] [blame] | 875 | for (scan = 0; scan < nr_to_scan && !list_empty(src); scan++) { |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 876 | struct page *page; |
| 877 | unsigned long pfn; |
| 878 | unsigned long end_pfn; |
| 879 | unsigned long page_pfn; |
| 880 | int zone_id; |
| 881 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 882 | page = lru_to_page(src); |
| 883 | prefetchw_prev_lru_page(page, src, flags); |
| 884 | |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 885 | VM_BUG_ON(!PageLRU(page)); |
Nick Piggin | 8d438f9 | 2006-03-22 00:07:59 -0800 | [diff] [blame] | 886 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 887 | switch (__isolate_lru_page(page, mode, file)) { |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 888 | case 0: |
| 889 | list_move(&page->lru, dst); |
Nick Piggin | 7c8ee9a | 2006-03-22 00:08:03 -0800 | [diff] [blame] | 890 | nr_taken++; |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 891 | break; |
Nick Piggin | 46453a6 | 2006-03-22 00:07:58 -0800 | [diff] [blame] | 892 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 893 | case -EBUSY: |
| 894 | /* else it is being freed elsewhere */ |
| 895 | list_move(&page->lru, src); |
| 896 | continue; |
| 897 | |
| 898 | default: |
| 899 | BUG(); |
| 900 | } |
| 901 | |
| 902 | if (!order) |
| 903 | continue; |
| 904 | |
| 905 | /* |
| 906 | * Attempt to take all pages in the order aligned region |
| 907 | * surrounding the tag page. Only take those pages of |
| 908 | * the same active state as that tag page. We may safely |
| 909 | * round the target page pfn down to the requested order |
| 910 | * as the mem_map is guarenteed valid out to MAX_ORDER, |
| 911 | * where that page is in a different zone we will detect |
| 912 | * it from its zone id and abort this block scan. |
| 913 | */ |
| 914 | zone_id = page_zone_id(page); |
| 915 | page_pfn = page_to_pfn(page); |
| 916 | pfn = page_pfn & ~((1 << order) - 1); |
| 917 | end_pfn = pfn + (1 << order); |
| 918 | for (; pfn < end_pfn; pfn++) { |
| 919 | struct page *cursor_page; |
| 920 | |
| 921 | /* The target page is in the block, ignore it. */ |
| 922 | if (unlikely(pfn == page_pfn)) |
| 923 | continue; |
| 924 | |
| 925 | /* Avoid holes within the zone. */ |
| 926 | if (unlikely(!pfn_valid_within(pfn))) |
| 927 | break; |
| 928 | |
| 929 | cursor_page = pfn_to_page(pfn); |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 930 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 931 | /* Check that we have not crossed a zone boundary. */ |
| 932 | if (unlikely(page_zone_id(cursor_page) != zone_id)) |
| 933 | continue; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 934 | switch (__isolate_lru_page(cursor_page, mode, file)) { |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 935 | case 0: |
| 936 | list_move(&cursor_page->lru, dst); |
| 937 | nr_taken++; |
| 938 | scan++; |
| 939 | break; |
| 940 | |
| 941 | case -EBUSY: |
| 942 | /* else it is being freed elsewhere */ |
| 943 | list_move(&cursor_page->lru, src); |
| 944 | default: |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 945 | break; /* ! on LRU or wrong list */ |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 946 | } |
| 947 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 948 | } |
| 949 | |
| 950 | *scanned = scan; |
| 951 | return nr_taken; |
| 952 | } |
| 953 | |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 954 | static unsigned long isolate_pages_global(unsigned long nr, |
| 955 | struct list_head *dst, |
| 956 | unsigned long *scanned, int order, |
| 957 | int mode, struct zone *z, |
| 958 | struct mem_cgroup *mem_cont, |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 959 | int active, int file) |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 960 | { |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 961 | int lru = LRU_BASE; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 962 | if (active) |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 963 | lru += LRU_ACTIVE; |
| 964 | if (file) |
| 965 | lru += LRU_FILE; |
| 966 | return isolate_lru_pages(nr, &z->lru[lru].list, dst, scanned, order, |
| 967 | mode, !!file); |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 968 | } |
| 969 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 970 | /* |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 971 | * clear_active_flags() is a helper for shrink_active_list(), clearing |
| 972 | * any active bits from the pages in the list. |
| 973 | */ |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 974 | static unsigned long clear_active_flags(struct list_head *page_list, |
| 975 | unsigned int *count) |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 976 | { |
| 977 | int nr_active = 0; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 978 | int lru; |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 979 | struct page *page; |
| 980 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 981 | list_for_each_entry(page, page_list, lru) { |
| 982 | lru = page_is_file_cache(page); |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 983 | if (PageActive(page)) { |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 984 | lru += LRU_ACTIVE; |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 985 | ClearPageActive(page); |
| 986 | nr_active++; |
| 987 | } |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 988 | count[lru]++; |
| 989 | } |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 990 | |
| 991 | return nr_active; |
| 992 | } |
| 993 | |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 994 | /** |
| 995 | * isolate_lru_page - tries to isolate a page from its LRU list |
| 996 | * @page: page to isolate from its LRU list |
| 997 | * |
| 998 | * Isolates a @page from an LRU list, clears PageLRU and adjusts the |
| 999 | * vmstat statistic corresponding to whatever LRU list the page was on. |
| 1000 | * |
| 1001 | * Returns 0 if the page was removed from an LRU list. |
| 1002 | * Returns -EBUSY if the page was not on an LRU list. |
| 1003 | * |
| 1004 | * The returned page will have PageLRU() cleared. If it was found on |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1005 | * the active list, it will have PageActive set. If it was found on |
| 1006 | * the unevictable list, it will have the PageUnevictable bit set. That flag |
| 1007 | * may need to be cleared by the caller before letting the page go. |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1008 | * |
| 1009 | * The vmstat statistic corresponding to the list on which the page was |
| 1010 | * found will be decremented. |
| 1011 | * |
| 1012 | * Restrictions: |
| 1013 | * (1) Must be called with an elevated refcount on the page. This is a |
| 1014 | * fundamentnal difference from isolate_lru_pages (which is called |
| 1015 | * without a stable reference). |
| 1016 | * (2) the lru_lock must not be held. |
| 1017 | * (3) interrupts must be enabled. |
| 1018 | */ |
| 1019 | int isolate_lru_page(struct page *page) |
| 1020 | { |
| 1021 | int ret = -EBUSY; |
| 1022 | |
| 1023 | if (PageLRU(page)) { |
| 1024 | struct zone *zone = page_zone(page); |
| 1025 | |
| 1026 | spin_lock_irq(&zone->lru_lock); |
| 1027 | if (PageLRU(page) && get_page_unless_zero(page)) { |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1028 | int lru = page_lru(page); |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1029 | ret = 0; |
| 1030 | ClearPageLRU(page); |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1031 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1032 | del_page_from_lru_list(zone, page, lru); |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1033 | } |
| 1034 | spin_unlock_irq(&zone->lru_lock); |
| 1035 | } |
| 1036 | return ret; |
| 1037 | } |
| 1038 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1039 | /* |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 1040 | * shrink_inactive_list() is a helper for shrink_zone(). It returns the number |
| 1041 | * of reclaimed pages |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1042 | */ |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 1043 | static unsigned long shrink_inactive_list(unsigned long max_scan, |
Rik van Riel | 33c120e | 2008-10-18 20:26:36 -0700 | [diff] [blame] | 1044 | struct zone *zone, struct scan_control *sc, |
| 1045 | int priority, int file) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1046 | { |
| 1047 | LIST_HEAD(page_list); |
| 1048 | struct pagevec pvec; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1049 | unsigned long nr_scanned = 0; |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 1050 | unsigned long nr_reclaimed = 0; |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1051 | struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1052 | |
| 1053 | pagevec_init(&pvec, 1); |
| 1054 | |
| 1055 | lru_add_drain(); |
| 1056 | spin_lock_irq(&zone->lru_lock); |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1057 | do { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1058 | struct page *page; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1059 | unsigned long nr_taken; |
| 1060 | unsigned long nr_scan; |
| 1061 | unsigned long nr_freed; |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1062 | unsigned long nr_active; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1063 | unsigned int count[NR_LRU_LISTS] = { 0, }; |
Rik van Riel | 33c120e | 2008-10-18 20:26:36 -0700 | [diff] [blame] | 1064 | int mode = ISOLATE_INACTIVE; |
| 1065 | |
| 1066 | /* |
| 1067 | * If we need a large contiguous chunk of memory, or have |
| 1068 | * trouble getting a small set of contiguous pages, we |
| 1069 | * will reclaim both active and inactive pages. |
| 1070 | * |
| 1071 | * We use the same threshold as pageout congestion_wait below. |
| 1072 | */ |
| 1073 | if (sc->order > PAGE_ALLOC_COSTLY_ORDER) |
| 1074 | mode = ISOLATE_BOTH; |
| 1075 | else if (sc->order && priority < DEF_PRIORITY - 2) |
| 1076 | mode = ISOLATE_BOTH; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1077 | |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1078 | nr_taken = sc->isolate_pages(sc->swap_cluster_max, |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1079 | &page_list, &nr_scan, sc->order, mode, |
| 1080 | zone, sc->mem_cgroup, 0, file); |
| 1081 | nr_active = clear_active_flags(&page_list, count); |
Andy Whitcroft | e9187bd | 2007-08-22 14:01:25 -0700 | [diff] [blame] | 1082 | __count_vm_events(PGDEACTIVATE, nr_active); |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1083 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1084 | __mod_zone_page_state(zone, NR_ACTIVE_FILE, |
| 1085 | -count[LRU_ACTIVE_FILE]); |
| 1086 | __mod_zone_page_state(zone, NR_INACTIVE_FILE, |
| 1087 | -count[LRU_INACTIVE_FILE]); |
| 1088 | __mod_zone_page_state(zone, NR_ACTIVE_ANON, |
| 1089 | -count[LRU_ACTIVE_ANON]); |
| 1090 | __mod_zone_page_state(zone, NR_INACTIVE_ANON, |
| 1091 | -count[LRU_INACTIVE_ANON]); |
| 1092 | |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 1093 | if (scan_global_lru(sc)) |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1094 | zone->pages_scanned += nr_scan; |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 1095 | |
| 1096 | reclaim_stat->recent_scanned[0] += count[LRU_INACTIVE_ANON]; |
| 1097 | reclaim_stat->recent_scanned[0] += count[LRU_ACTIVE_ANON]; |
| 1098 | reclaim_stat->recent_scanned[1] += count[LRU_INACTIVE_FILE]; |
| 1099 | reclaim_stat->recent_scanned[1] += count[LRU_ACTIVE_FILE]; |
| 1100 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1101 | spin_unlock_irq(&zone->lru_lock); |
| 1102 | |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1103 | nr_scanned += nr_scan; |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 1104 | nr_freed = shrink_page_list(&page_list, sc, PAGEOUT_IO_ASYNC); |
| 1105 | |
| 1106 | /* |
| 1107 | * If we are direct reclaiming for contiguous pages and we do |
| 1108 | * not reclaim everything in the list, try again and wait |
| 1109 | * for IO to complete. This will stall high-order allocations |
| 1110 | * but that should be acceptable to the caller |
| 1111 | */ |
| 1112 | if (nr_freed < nr_taken && !current_is_kswapd() && |
| 1113 | sc->order > PAGE_ALLOC_COSTLY_ORDER) { |
| 1114 | congestion_wait(WRITE, HZ/10); |
| 1115 | |
| 1116 | /* |
| 1117 | * The attempt at page out may have made some |
| 1118 | * of the pages active, mark them inactive again. |
| 1119 | */ |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1120 | nr_active = clear_active_flags(&page_list, count); |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 1121 | count_vm_events(PGDEACTIVATE, nr_active); |
| 1122 | |
| 1123 | nr_freed += shrink_page_list(&page_list, sc, |
| 1124 | PAGEOUT_IO_SYNC); |
| 1125 | } |
| 1126 | |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 1127 | nr_reclaimed += nr_freed; |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 1128 | local_irq_disable(); |
| 1129 | if (current_is_kswapd()) { |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 1130 | __count_zone_vm_events(PGSCAN_KSWAPD, zone, nr_scan); |
| 1131 | __count_vm_events(KSWAPD_STEAL, nr_freed); |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1132 | } else if (scan_global_lru(sc)) |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 1133 | __count_zone_vm_events(PGSCAN_DIRECT, zone, nr_scan); |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1134 | |
Shantanu Goel | 918d3f9 | 2006-12-29 16:48:59 -0800 | [diff] [blame] | 1135 | __count_zone_vm_events(PGSTEAL, zone, nr_freed); |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 1136 | |
Wu Fengguang | fb8d14e | 2006-03-22 00:08:28 -0800 | [diff] [blame] | 1137 | if (nr_taken == 0) |
| 1138 | goto done; |
| 1139 | |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 1140 | spin_lock(&zone->lru_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1141 | /* |
| 1142 | * Put back any unfreeable pages. |
| 1143 | */ |
| 1144 | while (!list_empty(&page_list)) { |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1145 | int lru; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1146 | page = lru_to_page(&page_list); |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 1147 | VM_BUG_ON(PageLRU(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1148 | list_del(&page->lru); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1149 | if (unlikely(!page_evictable(page, NULL))) { |
| 1150 | spin_unlock_irq(&zone->lru_lock); |
| 1151 | putback_lru_page(page); |
| 1152 | spin_lock_irq(&zone->lru_lock); |
| 1153 | continue; |
| 1154 | } |
| 1155 | SetPageLRU(page); |
| 1156 | lru = page_lru(page); |
| 1157 | add_page_to_lru_list(zone, page, lru); |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 1158 | if (PageActive(page)) { |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1159 | int file = !!page_is_file_cache(page); |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1160 | reclaim_stat->recent_rotated[file]++; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1161 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1162 | if (!pagevec_add(&pvec, page)) { |
| 1163 | spin_unlock_irq(&zone->lru_lock); |
| 1164 | __pagevec_release(&pvec); |
| 1165 | spin_lock_irq(&zone->lru_lock); |
| 1166 | } |
| 1167 | } |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1168 | } while (nr_scanned < max_scan); |
Wu Fengguang | fb8d14e | 2006-03-22 00:08:28 -0800 | [diff] [blame] | 1169 | spin_unlock(&zone->lru_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1170 | done: |
Wu Fengguang | fb8d14e | 2006-03-22 00:08:28 -0800 | [diff] [blame] | 1171 | local_irq_enable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1172 | pagevec_release(&pvec); |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 1173 | return nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1174 | } |
| 1175 | |
Martin Bligh | 3bb1a85 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 1176 | /* |
| 1177 | * We are about to scan this zone at a certain priority level. If that priority |
| 1178 | * level is smaller (ie: more urgent) than the previous priority, then note |
| 1179 | * that priority level within the zone. This is done so that when the next |
| 1180 | * process comes in to scan this zone, it will immediately start out at this |
| 1181 | * priority level rather than having to build up its own scanning priority. |
| 1182 | * Here, this priority affects only the reclaim-mapped threshold. |
| 1183 | */ |
| 1184 | static inline void note_zone_scanning_priority(struct zone *zone, int priority) |
| 1185 | { |
| 1186 | if (priority < zone->prev_priority) |
| 1187 | zone->prev_priority = priority; |
| 1188 | } |
| 1189 | |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1190 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1191 | * This moves pages from the active list to the inactive list. |
| 1192 | * |
| 1193 | * We move them the other way if the page is referenced by one or more |
| 1194 | * processes, from rmap. |
| 1195 | * |
| 1196 | * If the pages are mostly unmapped, the processing is fast and it is |
| 1197 | * appropriate to hold zone->lru_lock across the whole operation. But if |
| 1198 | * the pages are mapped, the processing is slow (page_referenced()) so we |
| 1199 | * should drop zone->lru_lock around each page. It's impossible to balance |
| 1200 | * this, so instead we remove the pages from the LRU while processing them. |
| 1201 | * It is safe to rely on PG_active against the non-LRU pages in here because |
| 1202 | * nobody will play with that bit on a non-LRU page. |
| 1203 | * |
| 1204 | * The downside is that we have to touch page->_count against each page. |
| 1205 | * But we had to alter page->flags anyway. |
| 1206 | */ |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1207 | |
| 1208 | |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 1209 | static void shrink_active_list(unsigned long nr_pages, struct zone *zone, |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1210 | struct scan_control *sc, int priority, int file) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1211 | { |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1212 | unsigned long pgmoved; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1213 | int pgdeactivate = 0; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1214 | unsigned long pgscanned; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1215 | LIST_HEAD(l_hold); /* The pages which were snipped off */ |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1216 | LIST_HEAD(l_inactive); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1217 | struct page *page; |
| 1218 | struct pagevec pvec; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1219 | enum lru_list lru; |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1220 | struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1221 | |
| 1222 | lru_add_drain(); |
| 1223 | spin_lock_irq(&zone->lru_lock); |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1224 | pgmoved = sc->isolate_pages(nr_pages, &l_hold, &pgscanned, sc->order, |
| 1225 | ISOLATE_ACTIVE, zone, |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1226 | sc->mem_cgroup, 1, file); |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1227 | /* |
| 1228 | * zone->pages_scanned is used for detect zone's oom |
| 1229 | * mem_cgroup remembers nr_scan by itself. |
| 1230 | */ |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1231 | if (scan_global_lru(sc)) { |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1232 | zone->pages_scanned += pgscanned; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1233 | } |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 1234 | reclaim_stat->recent_scanned[!!file] += pgmoved; |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1235 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1236 | if (file) |
| 1237 | __mod_zone_page_state(zone, NR_ACTIVE_FILE, -pgmoved); |
| 1238 | else |
| 1239 | __mod_zone_page_state(zone, NR_ACTIVE_ANON, -pgmoved); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1240 | spin_unlock_irq(&zone->lru_lock); |
| 1241 | |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1242 | pgmoved = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1243 | while (!list_empty(&l_hold)) { |
| 1244 | cond_resched(); |
| 1245 | page = lru_to_page(&l_hold); |
| 1246 | list_del(&page->lru); |
Rik van Riel | 7e9cd48 | 2008-10-18 20:26:35 -0700 | [diff] [blame] | 1247 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1248 | if (unlikely(!page_evictable(page, NULL))) { |
| 1249 | putback_lru_page(page); |
| 1250 | continue; |
| 1251 | } |
| 1252 | |
Rik van Riel | 7e9cd48 | 2008-10-18 20:26:35 -0700 | [diff] [blame] | 1253 | /* page_referenced clears PageReferenced */ |
| 1254 | if (page_mapping_inuse(page) && |
| 1255 | page_referenced(page, 0, sc->mem_cgroup)) |
| 1256 | pgmoved++; |
| 1257 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1258 | list_add(&page->lru, &l_inactive); |
| 1259 | } |
| 1260 | |
Andrew Morton | b555749 | 2009-01-06 14:40:13 -0800 | [diff] [blame] | 1261 | /* |
| 1262 | * Move the pages to the [file or anon] inactive list. |
| 1263 | */ |
| 1264 | pagevec_init(&pvec, 1); |
| 1265 | pgmoved = 0; |
| 1266 | lru = LRU_BASE + file * LRU_FILE; |
| 1267 | |
Johannes Weiner | 2a1dc50 | 2008-12-01 03:00:35 +0100 | [diff] [blame] | 1268 | spin_lock_irq(&zone->lru_lock); |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1269 | /* |
Rik van Riel | 7e9cd48 | 2008-10-18 20:26:35 -0700 | [diff] [blame] | 1270 | * Count referenced pages from currently used mappings as |
| 1271 | * rotated, even though they are moved to the inactive list. |
| 1272 | * This helps balance scan pressure between file and anonymous |
| 1273 | * pages in get_scan_ratio. |
| 1274 | */ |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 1275 | reclaim_stat->recent_rotated[!!file] += pgmoved; |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1276 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1277 | while (!list_empty(&l_inactive)) { |
| 1278 | page = lru_to_page(&l_inactive); |
| 1279 | prefetchw_prev_lru_page(page, &l_inactive, flags); |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 1280 | VM_BUG_ON(PageLRU(page)); |
Nick Piggin | 8d438f9 | 2006-03-22 00:07:59 -0800 | [diff] [blame] | 1281 | SetPageLRU(page); |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 1282 | VM_BUG_ON(!PageActive(page)); |
Nick Piggin | 4c84cac | 2006-03-22 00:08:00 -0800 | [diff] [blame] | 1283 | ClearPageActive(page); |
| 1284 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1285 | list_move(&page->lru, &zone->lru[lru].list); |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 1286 | mem_cgroup_add_lru_list(page, lru); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1287 | pgmoved++; |
| 1288 | if (!pagevec_add(&pvec, page)) { |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1289 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1290 | spin_unlock_irq(&zone->lru_lock); |
| 1291 | pgdeactivate += pgmoved; |
| 1292 | pgmoved = 0; |
| 1293 | if (buffer_heads_over_limit) |
| 1294 | pagevec_strip(&pvec); |
| 1295 | __pagevec_release(&pvec); |
| 1296 | spin_lock_irq(&zone->lru_lock); |
| 1297 | } |
| 1298 | } |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1299 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1300 | pgdeactivate += pgmoved; |
| 1301 | if (buffer_heads_over_limit) { |
| 1302 | spin_unlock_irq(&zone->lru_lock); |
| 1303 | pagevec_strip(&pvec); |
| 1304 | spin_lock_irq(&zone->lru_lock); |
| 1305 | } |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 1306 | __count_zone_vm_events(PGREFILL, zone, pgscanned); |
| 1307 | __count_vm_events(PGDEACTIVATE, pgdeactivate); |
| 1308 | spin_unlock_irq(&zone->lru_lock); |
Rik van Riel | 68a22394 | 2008-10-18 20:26:23 -0700 | [diff] [blame] | 1309 | if (vm_swap_full()) |
| 1310 | pagevec_swap_free(&pvec); |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 1311 | |
| 1312 | pagevec_release(&pvec); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1313 | } |
| 1314 | |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1315 | static int inactive_anon_is_low_global(struct zone *zone) |
KOSAKI Motohiro | f89eb90 | 2009-01-07 18:08:14 -0800 | [diff] [blame] | 1316 | { |
| 1317 | unsigned long active, inactive; |
| 1318 | |
| 1319 | active = zone_page_state(zone, NR_ACTIVE_ANON); |
| 1320 | inactive = zone_page_state(zone, NR_INACTIVE_ANON); |
| 1321 | |
| 1322 | if (inactive * zone->inactive_ratio < active) |
| 1323 | return 1; |
| 1324 | |
| 1325 | return 0; |
| 1326 | } |
| 1327 | |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1328 | /** |
| 1329 | * inactive_anon_is_low - check if anonymous pages need to be deactivated |
| 1330 | * @zone: zone to check |
| 1331 | * @sc: scan control of this context |
| 1332 | * |
| 1333 | * Returns true if the zone does not have enough inactive anon pages, |
| 1334 | * meaning some active anon pages need to be deactivated. |
| 1335 | */ |
| 1336 | static int inactive_anon_is_low(struct zone *zone, struct scan_control *sc) |
| 1337 | { |
| 1338 | int low; |
| 1339 | |
| 1340 | if (scan_global_lru(sc)) |
| 1341 | low = inactive_anon_is_low_global(zone); |
| 1342 | else |
| 1343 | low = mem_cgroup_inactive_anon_is_low(sc->mem_cgroup, zone); |
| 1344 | return low; |
| 1345 | } |
| 1346 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1347 | static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan, |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1348 | struct zone *zone, struct scan_control *sc, int priority) |
| 1349 | { |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1350 | int file = is_file_lru(lru); |
| 1351 | |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1352 | if (lru == LRU_ACTIVE_FILE) { |
| 1353 | shrink_active_list(nr_to_scan, zone, sc, priority, file); |
| 1354 | return 0; |
| 1355 | } |
| 1356 | |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1357 | if (lru == LRU_ACTIVE_ANON && inactive_anon_is_low(zone, sc)) { |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1358 | shrink_active_list(nr_to_scan, zone, sc, priority, file); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1359 | return 0; |
| 1360 | } |
Rik van Riel | 33c120e | 2008-10-18 20:26:36 -0700 | [diff] [blame] | 1361 | return shrink_inactive_list(nr_to_scan, zone, sc, priority, file); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1362 | } |
| 1363 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1364 | /* |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1365 | * Determine how aggressively the anon and file LRU lists should be |
| 1366 | * scanned. The relative value of each set of LRU lists is determined |
| 1367 | * by looking at the fraction of the pages scanned we did rotate back |
| 1368 | * onto the active list instead of evict. |
| 1369 | * |
| 1370 | * percent[0] specifies how much pressure to put on ram/swap backed |
| 1371 | * memory, while percent[1] determines pressure on the file LRUs. |
| 1372 | */ |
| 1373 | static void get_scan_ratio(struct zone *zone, struct scan_control *sc, |
| 1374 | unsigned long *percent) |
| 1375 | { |
| 1376 | unsigned long anon, file, free; |
| 1377 | unsigned long anon_prio, file_prio; |
| 1378 | unsigned long ap, fp; |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1379 | struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc); |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1380 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1381 | /* If we have no swap space, do not bother scanning anon pages. */ |
| 1382 | if (nr_swap_pages <= 0) { |
| 1383 | percent[0] = 0; |
| 1384 | percent[1] = 100; |
| 1385 | return; |
| 1386 | } |
| 1387 | |
KOSAKI Motohiro | c9f299d | 2009-01-07 18:08:16 -0800 | [diff] [blame] | 1388 | anon = zone_nr_pages(zone, sc, LRU_ACTIVE_ANON) + |
| 1389 | zone_nr_pages(zone, sc, LRU_INACTIVE_ANON); |
| 1390 | file = zone_nr_pages(zone, sc, LRU_ACTIVE_FILE) + |
| 1391 | zone_nr_pages(zone, sc, LRU_INACTIVE_FILE); |
Hugh Dickins | b962716 | 2009-01-06 14:39:41 -0800 | [diff] [blame] | 1392 | |
KOSAKI Motohiro | eeee9a8 | 2009-01-07 18:08:17 -0800 | [diff] [blame] | 1393 | if (scan_global_lru(sc)) { |
| 1394 | free = zone_page_state(zone, NR_FREE_PAGES); |
| 1395 | /* If we have very few page cache pages, |
| 1396 | force-scan anon pages. */ |
| 1397 | if (unlikely(file + free <= zone->pages_high)) { |
| 1398 | percent[0] = 100; |
| 1399 | percent[1] = 0; |
| 1400 | return; |
| 1401 | } |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1402 | } |
| 1403 | |
| 1404 | /* |
| 1405 | * OK, so we have swap space and a fair amount of page cache |
| 1406 | * pages. We use the recently rotated / recently scanned |
| 1407 | * ratios to determine how valuable each cache is. |
| 1408 | * |
| 1409 | * Because workloads change over time (and to avoid overflow) |
| 1410 | * we keep these statistics as a floating average, which ends |
| 1411 | * up weighing recent references more than old ones. |
| 1412 | * |
| 1413 | * anon in [0], file in [1] |
| 1414 | */ |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1415 | if (unlikely(reclaim_stat->recent_scanned[0] > anon / 4)) { |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1416 | spin_lock_irq(&zone->lru_lock); |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1417 | reclaim_stat->recent_scanned[0] /= 2; |
| 1418 | reclaim_stat->recent_rotated[0] /= 2; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1419 | spin_unlock_irq(&zone->lru_lock); |
| 1420 | } |
| 1421 | |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1422 | if (unlikely(reclaim_stat->recent_scanned[1] > file / 4)) { |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1423 | spin_lock_irq(&zone->lru_lock); |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1424 | reclaim_stat->recent_scanned[1] /= 2; |
| 1425 | reclaim_stat->recent_rotated[1] /= 2; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1426 | spin_unlock_irq(&zone->lru_lock); |
| 1427 | } |
| 1428 | |
| 1429 | /* |
| 1430 | * With swappiness at 100, anonymous and file have the same priority. |
| 1431 | * This scanning priority is essentially the inverse of IO cost. |
| 1432 | */ |
| 1433 | anon_prio = sc->swappiness; |
| 1434 | file_prio = 200 - sc->swappiness; |
| 1435 | |
| 1436 | /* |
Rik van Riel | 00d8089 | 2008-11-19 15:36:44 -0800 | [diff] [blame] | 1437 | * The amount of pressure on anon vs file pages is inversely |
| 1438 | * proportional to the fraction of recently scanned pages on |
| 1439 | * each list that were recently referenced and in active use. |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1440 | */ |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1441 | ap = (anon_prio + 1) * (reclaim_stat->recent_scanned[0] + 1); |
| 1442 | ap /= reclaim_stat->recent_rotated[0] + 1; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1443 | |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1444 | fp = (file_prio + 1) * (reclaim_stat->recent_scanned[1] + 1); |
| 1445 | fp /= reclaim_stat->recent_rotated[1] + 1; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1446 | |
| 1447 | /* Normalize to percentages */ |
| 1448 | percent[0] = 100 * ap / (ap + fp + 1); |
| 1449 | percent[1] = 100 - percent[0]; |
| 1450 | } |
| 1451 | |
| 1452 | |
| 1453 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1454 | * This is a basic per-zone page freer. Used by both kswapd and direct reclaim. |
| 1455 | */ |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1456 | static void shrink_zone(int priority, struct zone *zone, |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 1457 | struct scan_control *sc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1458 | { |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1459 | unsigned long nr[NR_LRU_LISTS]; |
Christoph Lameter | 8695949 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 1460 | unsigned long nr_to_scan; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1461 | unsigned long percent[2]; /* anon @ 0; file @ 1 */ |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1462 | enum lru_list l; |
KOSAKI Motohiro | 01dbe5c | 2009-01-06 14:40:02 -0800 | [diff] [blame] | 1463 | unsigned long nr_reclaimed = sc->nr_reclaimed; |
| 1464 | unsigned long swap_cluster_max = sc->swap_cluster_max; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1465 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1466 | get_scan_ratio(zone, sc, percent); |
| 1467 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1468 | for_each_evictable_lru(l) { |
KOSAKI Motohiro | 9439c1c | 2009-01-07 18:08:21 -0800 | [diff] [blame^] | 1469 | int file = is_file_lru(l); |
| 1470 | int scan; |
Johannes Weiner | e0f79b8 | 2008-10-18 20:26:55 -0700 | [diff] [blame] | 1471 | |
KOSAKI Motohiro | 9439c1c | 2009-01-07 18:08:21 -0800 | [diff] [blame^] | 1472 | scan = zone_page_state(zone, NR_LRU_BASE + l); |
| 1473 | if (priority) { |
| 1474 | scan >>= priority; |
| 1475 | scan = (scan * percent[file]) / 100; |
| 1476 | } |
| 1477 | if (scan_global_lru(sc)) { |
Johannes Weiner | e0f79b8 | 2008-10-18 20:26:55 -0700 | [diff] [blame] | 1478 | zone->lru[l].nr_scan += scan; |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1479 | nr[l] = zone->lru[l].nr_scan; |
KOSAKI Motohiro | 01dbe5c | 2009-01-06 14:40:02 -0800 | [diff] [blame] | 1480 | if (nr[l] >= swap_cluster_max) |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1481 | zone->lru[l].nr_scan = 0; |
| 1482 | else |
| 1483 | nr[l] = 0; |
KOSAKI Motohiro | 9439c1c | 2009-01-07 18:08:21 -0800 | [diff] [blame^] | 1484 | } else |
| 1485 | nr[l] = scan; |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1486 | } |
| 1487 | |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1488 | while (nr[LRU_INACTIVE_ANON] || nr[LRU_ACTIVE_FILE] || |
| 1489 | nr[LRU_INACTIVE_FILE]) { |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1490 | for_each_evictable_lru(l) { |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1491 | if (nr[l]) { |
KOSAKI Motohiro | 01dbe5c | 2009-01-06 14:40:02 -0800 | [diff] [blame] | 1492 | nr_to_scan = min(nr[l], swap_cluster_max); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1493 | nr[l] -= nr_to_scan; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1494 | |
KOSAKI Motohiro | 01dbe5c | 2009-01-06 14:40:02 -0800 | [diff] [blame] | 1495 | nr_reclaimed += shrink_list(l, nr_to_scan, |
| 1496 | zone, sc, priority); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1497 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1498 | } |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1499 | /* |
| 1500 | * On large memory systems, scan >> priority can become |
| 1501 | * really large. This is fine for the starting priority; |
| 1502 | * we want to put equal scanning pressure on each zone. |
| 1503 | * However, if the VM has a harder time of freeing pages, |
| 1504 | * with multiple processes reclaiming pages, the total |
| 1505 | * freeing target can get unreasonably large. |
| 1506 | */ |
KOSAKI Motohiro | 01dbe5c | 2009-01-06 14:40:02 -0800 | [diff] [blame] | 1507 | if (nr_reclaimed > swap_cluster_max && |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1508 | priority < DEF_PRIORITY && !current_is_kswapd()) |
| 1509 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1510 | } |
| 1511 | |
KOSAKI Motohiro | 01dbe5c | 2009-01-06 14:40:02 -0800 | [diff] [blame] | 1512 | sc->nr_reclaimed = nr_reclaimed; |
| 1513 | |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1514 | /* |
| 1515 | * Even if we did not try to evict anon pages at all, we want to |
| 1516 | * rebalance the anon lru active/inactive ratio. |
| 1517 | */ |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1518 | if (inactive_anon_is_low(zone, sc)) |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1519 | shrink_active_list(SWAP_CLUSTER_MAX, zone, sc, priority, 0); |
| 1520 | |
Andrew Morton | 232ea4d | 2007-02-28 20:13:21 -0800 | [diff] [blame] | 1521 | throttle_vm_writeout(sc->gfp_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1522 | } |
| 1523 | |
| 1524 | /* |
| 1525 | * This is the direct reclaim path, for page-allocating processes. We only |
| 1526 | * try to reclaim pages from zones which will satisfy the caller's allocation |
| 1527 | * request. |
| 1528 | * |
| 1529 | * We reclaim from a zone even if that zone is over pages_high. Because: |
| 1530 | * a) The caller may be trying to free *extra* pages to satisfy a higher-order |
| 1531 | * allocation or |
| 1532 | * b) The zones may be over pages_high but they must go *over* pages_high to |
| 1533 | * satisfy the `incremental min' zone defense algorithm. |
| 1534 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1535 | * If a zone is deemed to be full of pinned pages then just give it a light |
| 1536 | * scan then give up on it. |
| 1537 | */ |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1538 | static void shrink_zones(int priority, struct zonelist *zonelist, |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 1539 | struct scan_control *sc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1540 | { |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 1541 | enum zone_type high_zoneidx = gfp_zone(sc->gfp_mask); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1542 | struct zoneref *z; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 1543 | struct zone *zone; |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1544 | |
Nick Piggin | 408d854 | 2006-09-25 23:31:27 -0700 | [diff] [blame] | 1545 | sc->all_unreclaimable = 1; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 1546 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
Con Kolivas | f3fe651 | 2006-01-06 00:11:15 -0800 | [diff] [blame] | 1547 | if (!populated_zone(zone)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1548 | continue; |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1549 | /* |
| 1550 | * Take care memory controller reclaiming has small influence |
| 1551 | * to global LRU. |
| 1552 | */ |
| 1553 | if (scan_global_lru(sc)) { |
| 1554 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
| 1555 | continue; |
| 1556 | note_zone_scanning_priority(zone, priority); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1557 | |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1558 | if (zone_is_all_unreclaimable(zone) && |
| 1559 | priority != DEF_PRIORITY) |
| 1560 | continue; /* Let kswapd poll it */ |
| 1561 | sc->all_unreclaimable = 0; |
| 1562 | } else { |
| 1563 | /* |
| 1564 | * Ignore cpuset limitation here. We just want to reduce |
| 1565 | * # of used pages by us regardless of memory shortage. |
| 1566 | */ |
| 1567 | sc->all_unreclaimable = 0; |
| 1568 | mem_cgroup_note_reclaim_priority(sc->mem_cgroup, |
| 1569 | priority); |
| 1570 | } |
Nick Piggin | 408d854 | 2006-09-25 23:31:27 -0700 | [diff] [blame] | 1571 | |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1572 | shrink_zone(priority, zone, sc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1573 | } |
| 1574 | } |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1575 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1576 | /* |
| 1577 | * This is the main entry point to direct page reclaim. |
| 1578 | * |
| 1579 | * If a full scan of the inactive list fails to free enough memory then we |
| 1580 | * are "out of memory" and something needs to be killed. |
| 1581 | * |
| 1582 | * If the caller is !__GFP_FS then the probability of a failure is reasonably |
| 1583 | * high - the zone may be full of dirty or under-writeback pages, which this |
| 1584 | * caller can't do much about. We kick pdflush and take explicit naps in the |
| 1585 | * hope that some of these pages can be written. But if the allocating task |
| 1586 | * holds filesystem locks which prevent writeout this might not work, and the |
| 1587 | * allocation attempt will fail. |
Nishanth Aravamudan | a41f24e | 2008-04-29 00:58:25 -0700 | [diff] [blame] | 1588 | * |
| 1589 | * returns: 0, if no pages reclaimed |
| 1590 | * else, the number of pages reclaimed |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1591 | */ |
Mel Gorman | dac1d27 | 2008-04-28 02:12:12 -0700 | [diff] [blame] | 1592 | static unsigned long do_try_to_free_pages(struct zonelist *zonelist, |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1593 | struct scan_control *sc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1594 | { |
| 1595 | int priority; |
kosaki.motohiro@jp.fujitsu.com | c700be3 | 2008-06-12 15:21:27 -0700 | [diff] [blame] | 1596 | unsigned long ret = 0; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1597 | unsigned long total_scanned = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1598 | struct reclaim_state *reclaim_state = current->reclaim_state; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1599 | unsigned long lru_pages = 0; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1600 | struct zoneref *z; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 1601 | struct zone *zone; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1602 | enum zone_type high_zoneidx = gfp_zone(sc->gfp_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1603 | |
Keika Kobayashi | 873b477 | 2008-07-25 01:48:52 -0700 | [diff] [blame] | 1604 | delayacct_freepages_start(); |
| 1605 | |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1606 | if (scan_global_lru(sc)) |
| 1607 | count_vm_event(ALLOCSTALL); |
| 1608 | /* |
| 1609 | * mem_cgroup will not do shrink_slab. |
| 1610 | */ |
| 1611 | if (scan_global_lru(sc)) { |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 1612 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1613 | |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1614 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
| 1615 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1616 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1617 | lru_pages += zone_lru_pages(zone); |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1618 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1619 | } |
| 1620 | |
| 1621 | for (priority = DEF_PRIORITY; priority >= 0; priority--) { |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1622 | sc->nr_scanned = 0; |
Rik van Riel | f7b7fd8 | 2005-11-28 13:44:07 -0800 | [diff] [blame] | 1623 | if (!priority) |
| 1624 | disable_swap_token(); |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1625 | shrink_zones(priority, zonelist, sc); |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1626 | /* |
| 1627 | * Don't shrink slabs when reclaiming memory from |
| 1628 | * over limit cgroups |
| 1629 | */ |
KAMEZAWA Hiroyuki | 91a4547 | 2008-02-07 00:14:29 -0800 | [diff] [blame] | 1630 | if (scan_global_lru(sc)) { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1631 | shrink_slab(sc->nr_scanned, sc->gfp_mask, lru_pages); |
KAMEZAWA Hiroyuki | 91a4547 | 2008-02-07 00:14:29 -0800 | [diff] [blame] | 1632 | if (reclaim_state) { |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1633 | sc->nr_reclaimed += reclaim_state->reclaimed_slab; |
KAMEZAWA Hiroyuki | 91a4547 | 2008-02-07 00:14:29 -0800 | [diff] [blame] | 1634 | reclaim_state->reclaimed_slab = 0; |
| 1635 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1636 | } |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1637 | total_scanned += sc->nr_scanned; |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1638 | if (sc->nr_reclaimed >= sc->swap_cluster_max) { |
| 1639 | ret = sc->nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1640 | goto out; |
| 1641 | } |
| 1642 | |
| 1643 | /* |
| 1644 | * Try to write back as many pages as we just scanned. This |
| 1645 | * tends to cause slow streaming writers to write data to the |
| 1646 | * disk smoothly, at the dirtying rate, which is nice. But |
| 1647 | * that's undesirable in laptop mode, where we *want* lumpy |
| 1648 | * writeout. So in laptop mode, write out the whole world. |
| 1649 | */ |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1650 | if (total_scanned > sc->swap_cluster_max + |
| 1651 | sc->swap_cluster_max / 2) { |
Pekka J Enberg | 687a21c | 2005-06-28 20:44:55 -0700 | [diff] [blame] | 1652 | wakeup_pdflush(laptop_mode ? 0 : total_scanned); |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1653 | sc->may_writepage = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1654 | } |
| 1655 | |
| 1656 | /* Take a nap, wait for some writeback to complete */ |
Andrew Morton | 4dd4b92 | 2008-03-24 12:29:52 -0700 | [diff] [blame] | 1657 | if (sc->nr_scanned && priority < DEF_PRIORITY - 2) |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 1658 | congestion_wait(WRITE, HZ/10); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1659 | } |
Fernando Luis Vazquez Cao | 87547ee | 2008-07-29 22:33:42 -0700 | [diff] [blame] | 1660 | /* top priority shrink_zones still had more to do? don't OOM, then */ |
KAMEZAWA Hiroyuki | 91a4547 | 2008-02-07 00:14:29 -0800 | [diff] [blame] | 1661 | if (!sc->all_unreclaimable && scan_global_lru(sc)) |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1662 | ret = sc->nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1663 | out: |
Martin Bligh | 3bb1a85 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 1664 | /* |
| 1665 | * Now that we've scanned all the zones at this priority level, note |
| 1666 | * that level within the zone so that the next thread which performs |
| 1667 | * scanning of this zone will immediately start out at this priority |
| 1668 | * level. This affects only the decision whether or not to bring |
| 1669 | * mapped pages onto the inactive list. |
| 1670 | */ |
| 1671 | if (priority < 0) |
| 1672 | priority = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1673 | |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1674 | if (scan_global_lru(sc)) { |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 1675 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1676 | |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1677 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
| 1678 | continue; |
| 1679 | |
| 1680 | zone->prev_priority = priority; |
| 1681 | } |
| 1682 | } else |
| 1683 | mem_cgroup_record_reclaim_priority(sc->mem_cgroup, priority); |
| 1684 | |
Keika Kobayashi | 873b477 | 2008-07-25 01:48:52 -0700 | [diff] [blame] | 1685 | delayacct_freepages_end(); |
| 1686 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1687 | return ret; |
| 1688 | } |
| 1689 | |
Mel Gorman | dac1d27 | 2008-04-28 02:12:12 -0700 | [diff] [blame] | 1690 | unsigned long try_to_free_pages(struct zonelist *zonelist, int order, |
| 1691 | gfp_t gfp_mask) |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1692 | { |
| 1693 | struct scan_control sc = { |
| 1694 | .gfp_mask = gfp_mask, |
| 1695 | .may_writepage = !laptop_mode, |
| 1696 | .swap_cluster_max = SWAP_CLUSTER_MAX, |
| 1697 | .may_swap = 1, |
| 1698 | .swappiness = vm_swappiness, |
| 1699 | .order = order, |
| 1700 | .mem_cgroup = NULL, |
| 1701 | .isolate_pages = isolate_pages_global, |
| 1702 | }; |
| 1703 | |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1704 | return do_try_to_free_pages(zonelist, &sc); |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1705 | } |
| 1706 | |
Balbir Singh | 00f0b82 | 2008-03-04 14:28:39 -0800 | [diff] [blame] | 1707 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1708 | |
Balbir Singh | e1a1cd5 | 2008-02-07 00:14:02 -0800 | [diff] [blame] | 1709 | unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *mem_cont, |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 1710 | gfp_t gfp_mask, |
| 1711 | bool noswap) |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1712 | { |
| 1713 | struct scan_control sc = { |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1714 | .may_writepage = !laptop_mode, |
| 1715 | .may_swap = 1, |
| 1716 | .swap_cluster_max = SWAP_CLUSTER_MAX, |
| 1717 | .swappiness = vm_swappiness, |
| 1718 | .order = 0, |
| 1719 | .mem_cgroup = mem_cont, |
| 1720 | .isolate_pages = mem_cgroup_isolate_pages, |
| 1721 | }; |
Mel Gorman | dac1d27 | 2008-04-28 02:12:12 -0700 | [diff] [blame] | 1722 | struct zonelist *zonelist; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1723 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 1724 | if (noswap) |
| 1725 | sc.may_swap = 0; |
| 1726 | |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1727 | sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | |
| 1728 | (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK); |
| 1729 | zonelist = NODE_DATA(numa_node_id())->node_zonelists; |
| 1730 | return do_try_to_free_pages(zonelist, &sc); |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1731 | } |
| 1732 | #endif |
| 1733 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1734 | /* |
| 1735 | * For kswapd, balance_pgdat() will work across all this node's zones until |
| 1736 | * they are all at pages_high. |
| 1737 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1738 | * Returns the number of pages which were actually freed. |
| 1739 | * |
| 1740 | * There is special handling here for zones which are full of pinned pages. |
| 1741 | * This can happen if the pages are all mlocked, or if they are all used by |
| 1742 | * device drivers (say, ZONE_DMA). Or if they are all in use by hugetlb. |
| 1743 | * What we do is to detect the case where all pages in the zone have been |
| 1744 | * scanned twice and there has been zero successful reclaim. Mark the zone as |
| 1745 | * dead and from now on, only perform a short scan. Basically we're polling |
| 1746 | * the zone for when the problem goes away. |
| 1747 | * |
| 1748 | * kswapd scans the zones in the highmem->normal->dma direction. It skips |
| 1749 | * zones which have free_pages > pages_high, but once a zone is found to have |
| 1750 | * free_pages <= pages_high, we scan that zone and the lower zones regardless |
| 1751 | * of the number of free pages in the lower zones. This interoperates with |
| 1752 | * the page allocator fallback scheme to ensure that aging of pages is balanced |
| 1753 | * across the zones. |
| 1754 | */ |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 1755 | static unsigned long balance_pgdat(pg_data_t *pgdat, int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1756 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1757 | int all_zones_ok; |
| 1758 | int priority; |
| 1759 | int i; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1760 | unsigned long total_scanned; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1761 | struct reclaim_state *reclaim_state = current->reclaim_state; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 1762 | struct scan_control sc = { |
| 1763 | .gfp_mask = GFP_KERNEL, |
| 1764 | .may_swap = 1, |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 1765 | .swap_cluster_max = SWAP_CLUSTER_MAX, |
| 1766 | .swappiness = vm_swappiness, |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1767 | .order = order, |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1768 | .mem_cgroup = NULL, |
| 1769 | .isolate_pages = isolate_pages_global, |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 1770 | }; |
Martin Bligh | 3bb1a85 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 1771 | /* |
| 1772 | * temp_priority is used to remember the scanning priority at which |
| 1773 | * this zone was successfully refilled to free_pages == pages_high. |
| 1774 | */ |
| 1775 | int temp_priority[MAX_NR_ZONES]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1776 | |
| 1777 | loop_again: |
| 1778 | total_scanned = 0; |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1779 | sc.nr_reclaimed = 0; |
Christoph Lameter | c0bbbc7 | 2006-06-11 15:22:26 -0700 | [diff] [blame] | 1780 | sc.may_writepage = !laptop_mode; |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 1781 | count_vm_event(PAGEOUTRUN); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1782 | |
Martin Bligh | 3bb1a85 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 1783 | for (i = 0; i < pgdat->nr_zones; i++) |
| 1784 | temp_priority[i] = DEF_PRIORITY; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1785 | |
| 1786 | for (priority = DEF_PRIORITY; priority >= 0; priority--) { |
| 1787 | int end_zone = 0; /* Inclusive. 0 = ZONE_DMA */ |
| 1788 | unsigned long lru_pages = 0; |
| 1789 | |
Rik van Riel | f7b7fd8 | 2005-11-28 13:44:07 -0800 | [diff] [blame] | 1790 | /* The swap token gets in the way of swapout... */ |
| 1791 | if (!priority) |
| 1792 | disable_swap_token(); |
| 1793 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1794 | all_zones_ok = 1; |
| 1795 | |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 1796 | /* |
| 1797 | * Scan in the highmem->dma direction for the highest |
| 1798 | * zone which needs scanning |
| 1799 | */ |
| 1800 | for (i = pgdat->nr_zones - 1; i >= 0; i--) { |
| 1801 | struct zone *zone = pgdat->node_zones + i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1802 | |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 1803 | if (!populated_zone(zone)) |
| 1804 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1805 | |
David Rientjes | e815af9 | 2007-10-16 23:25:54 -0700 | [diff] [blame] | 1806 | if (zone_is_all_unreclaimable(zone) && |
| 1807 | priority != DEF_PRIORITY) |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 1808 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1809 | |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1810 | /* |
| 1811 | * Do some background aging of the anon list, to give |
| 1812 | * pages a chance to be referenced before reclaiming. |
| 1813 | */ |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1814 | if (inactive_anon_is_low(zone, &sc)) |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1815 | shrink_active_list(SWAP_CLUSTER_MAX, zone, |
| 1816 | &sc, priority, 0); |
| 1817 | |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 1818 | if (!zone_watermark_ok(zone, order, zone->pages_high, |
| 1819 | 0, 0)) { |
| 1820 | end_zone = i; |
Andrew Morton | e1dbeda | 2006-12-06 20:32:01 -0800 | [diff] [blame] | 1821 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1822 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1823 | } |
Andrew Morton | e1dbeda | 2006-12-06 20:32:01 -0800 | [diff] [blame] | 1824 | if (i < 0) |
| 1825 | goto out; |
| 1826 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1827 | for (i = 0; i <= end_zone; i++) { |
| 1828 | struct zone *zone = pgdat->node_zones + i; |
| 1829 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1830 | lru_pages += zone_lru_pages(zone); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1831 | } |
| 1832 | |
| 1833 | /* |
| 1834 | * Now scan the zone in the dma->highmem direction, stopping |
| 1835 | * at the last zone which needs scanning. |
| 1836 | * |
| 1837 | * We do this because the page allocator works in the opposite |
| 1838 | * direction. This prevents the page allocator from allocating |
| 1839 | * pages behind kswapd's direction of progress, which would |
| 1840 | * cause too much scanning of the lower zones. |
| 1841 | */ |
| 1842 | for (i = 0; i <= end_zone; i++) { |
| 1843 | struct zone *zone = pgdat->node_zones + i; |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 1844 | int nr_slab; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1845 | |
Con Kolivas | f3fe651 | 2006-01-06 00:11:15 -0800 | [diff] [blame] | 1846 | if (!populated_zone(zone)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1847 | continue; |
| 1848 | |
David Rientjes | e815af9 | 2007-10-16 23:25:54 -0700 | [diff] [blame] | 1849 | if (zone_is_all_unreclaimable(zone) && |
| 1850 | priority != DEF_PRIORITY) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1851 | continue; |
| 1852 | |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 1853 | if (!zone_watermark_ok(zone, order, zone->pages_high, |
| 1854 | end_zone, 0)) |
| 1855 | all_zones_ok = 0; |
Martin Bligh | 3bb1a85 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 1856 | temp_priority[i] = priority; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1857 | sc.nr_scanned = 0; |
Martin Bligh | 3bb1a85 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 1858 | note_zone_scanning_priority(zone, priority); |
Rik van Riel | 32a4330 | 2007-10-16 01:24:50 -0700 | [diff] [blame] | 1859 | /* |
| 1860 | * We put equal pressure on every zone, unless one |
| 1861 | * zone has way too many pages free already. |
| 1862 | */ |
| 1863 | if (!zone_watermark_ok(zone, order, 8*zone->pages_high, |
| 1864 | end_zone, 0)) |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1865 | shrink_zone(priority, zone, &sc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1866 | reclaim_state->reclaimed_slab = 0; |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 1867 | nr_slab = shrink_slab(sc.nr_scanned, GFP_KERNEL, |
| 1868 | lru_pages); |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1869 | sc.nr_reclaimed += reclaim_state->reclaimed_slab; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1870 | total_scanned += sc.nr_scanned; |
David Rientjes | e815af9 | 2007-10-16 23:25:54 -0700 | [diff] [blame] | 1871 | if (zone_is_all_unreclaimable(zone)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1872 | continue; |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 1873 | if (nr_slab == 0 && zone->pages_scanned >= |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1874 | (zone_lru_pages(zone) * 6)) |
David Rientjes | e815af9 | 2007-10-16 23:25:54 -0700 | [diff] [blame] | 1875 | zone_set_flag(zone, |
| 1876 | ZONE_ALL_UNRECLAIMABLE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1877 | /* |
| 1878 | * If we've done a decent amount of scanning and |
| 1879 | * the reclaim ratio is low, start doing writepage |
| 1880 | * even in laptop mode |
| 1881 | */ |
| 1882 | if (total_scanned > SWAP_CLUSTER_MAX * 2 && |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1883 | total_scanned > sc.nr_reclaimed + sc.nr_reclaimed / 2) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1884 | sc.may_writepage = 1; |
| 1885 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1886 | if (all_zones_ok) |
| 1887 | break; /* kswapd: all done */ |
| 1888 | /* |
| 1889 | * OK, kswapd is getting into trouble. Take a nap, then take |
| 1890 | * another pass across the zones. |
| 1891 | */ |
Andrew Morton | 4dd4b92 | 2008-03-24 12:29:52 -0700 | [diff] [blame] | 1892 | if (total_scanned && priority < DEF_PRIORITY - 2) |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 1893 | congestion_wait(WRITE, HZ/10); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1894 | |
| 1895 | /* |
| 1896 | * We do this so kswapd doesn't build up large priorities for |
| 1897 | * example when it is freeing in parallel with allocators. It |
| 1898 | * matches the direct reclaim path behaviour in terms of impact |
| 1899 | * on zone->*_priority. |
| 1900 | */ |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1901 | if (sc.nr_reclaimed >= SWAP_CLUSTER_MAX) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1902 | break; |
| 1903 | } |
| 1904 | out: |
Martin Bligh | 3bb1a85 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 1905 | /* |
| 1906 | * Note within each zone the priority level at which this zone was |
| 1907 | * brought into a happy state. So that the next thread which scans this |
| 1908 | * zone will start out at that priority level. |
| 1909 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1910 | for (i = 0; i < pgdat->nr_zones; i++) { |
| 1911 | struct zone *zone = pgdat->node_zones + i; |
| 1912 | |
Martin Bligh | 3bb1a85 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 1913 | zone->prev_priority = temp_priority[i]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1914 | } |
| 1915 | if (!all_zones_ok) { |
| 1916 | cond_resched(); |
Rafael J. Wysocki | 8357376 | 2006-12-06 20:34:18 -0800 | [diff] [blame] | 1917 | |
| 1918 | try_to_freeze(); |
| 1919 | |
KOSAKI Motohiro | 73ce02e | 2009-01-06 14:40:33 -0800 | [diff] [blame] | 1920 | /* |
| 1921 | * Fragmentation may mean that the system cannot be |
| 1922 | * rebalanced for high-order allocations in all zones. |
| 1923 | * At this point, if nr_reclaimed < SWAP_CLUSTER_MAX, |
| 1924 | * it means the zones have been fully scanned and are still |
| 1925 | * not balanced. For high-order allocations, there is |
| 1926 | * little point trying all over again as kswapd may |
| 1927 | * infinite loop. |
| 1928 | * |
| 1929 | * Instead, recheck all watermarks at order-0 as they |
| 1930 | * are the most important. If watermarks are ok, kswapd will go |
| 1931 | * back to sleep. High-order users can still perform direct |
| 1932 | * reclaim if they wish. |
| 1933 | */ |
| 1934 | if (sc.nr_reclaimed < SWAP_CLUSTER_MAX) |
| 1935 | order = sc.order = 0; |
| 1936 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1937 | goto loop_again; |
| 1938 | } |
| 1939 | |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1940 | return sc.nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1941 | } |
| 1942 | |
| 1943 | /* |
| 1944 | * The background pageout daemon, started as a kernel thread |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1945 | * from the init process. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1946 | * |
| 1947 | * This basically trickles out pages so that we have _some_ |
| 1948 | * free memory available even if there is no other activity |
| 1949 | * that frees anything up. This is needed for things like routing |
| 1950 | * etc, where we otherwise might have all activity going on in |
| 1951 | * asynchronous contexts that cannot page things out. |
| 1952 | * |
| 1953 | * If there are applications that are active memory-allocators |
| 1954 | * (most normal use), this basically shouldn't matter. |
| 1955 | */ |
| 1956 | static int kswapd(void *p) |
| 1957 | { |
| 1958 | unsigned long order; |
| 1959 | pg_data_t *pgdat = (pg_data_t*)p; |
| 1960 | struct task_struct *tsk = current; |
| 1961 | DEFINE_WAIT(wait); |
| 1962 | struct reclaim_state reclaim_state = { |
| 1963 | .reclaimed_slab = 0, |
| 1964 | }; |
Mike Travis | c5f59f0 | 2008-04-04 18:11:10 -0700 | [diff] [blame] | 1965 | node_to_cpumask_ptr(cpumask, pgdat->node_id); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1966 | |
Rusty Russell | 174596a | 2009-01-01 10:12:29 +1030 | [diff] [blame] | 1967 | if (!cpumask_empty(cpumask)) |
Mike Travis | c5f59f0 | 2008-04-04 18:11:10 -0700 | [diff] [blame] | 1968 | set_cpus_allowed_ptr(tsk, cpumask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1969 | current->reclaim_state = &reclaim_state; |
| 1970 | |
| 1971 | /* |
| 1972 | * Tell the memory management that we're a "memory allocator", |
| 1973 | * and that if we need more memory we should get access to it |
| 1974 | * regardless (see "__alloc_pages()"). "kswapd" should |
| 1975 | * never get caught in the normal page freeing logic. |
| 1976 | * |
| 1977 | * (Kswapd normally doesn't need memory anyway, but sometimes |
| 1978 | * you need a small amount of memory in order to be able to |
| 1979 | * page out something else, and this flag essentially protects |
| 1980 | * us from recursively trying to free more memory as we're |
| 1981 | * trying to free the first piece of memory in the first place). |
| 1982 | */ |
Christoph Lameter | 930d915 | 2006-01-08 01:00:47 -0800 | [diff] [blame] | 1983 | tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD; |
Rafael J. Wysocki | 8314418 | 2007-07-17 04:03:35 -0700 | [diff] [blame] | 1984 | set_freezable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1985 | |
| 1986 | order = 0; |
| 1987 | for ( ; ; ) { |
| 1988 | unsigned long new_order; |
Christoph Lameter | 3e1d1d2 | 2005-06-24 23:13:50 -0700 | [diff] [blame] | 1989 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1990 | prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); |
| 1991 | new_order = pgdat->kswapd_max_order; |
| 1992 | pgdat->kswapd_max_order = 0; |
| 1993 | if (order < new_order) { |
| 1994 | /* |
| 1995 | * Don't sleep if someone wants a larger 'order' |
| 1996 | * allocation |
| 1997 | */ |
| 1998 | order = new_order; |
| 1999 | } else { |
Rafael J. Wysocki | b1296cc | 2007-05-06 14:50:48 -0700 | [diff] [blame] | 2000 | if (!freezing(current)) |
| 2001 | schedule(); |
| 2002 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2003 | order = pgdat->kswapd_max_order; |
| 2004 | } |
| 2005 | finish_wait(&pgdat->kswapd_wait, &wait); |
| 2006 | |
Rafael J. Wysocki | b1296cc | 2007-05-06 14:50:48 -0700 | [diff] [blame] | 2007 | if (!try_to_freeze()) { |
| 2008 | /* We can speed up thawing tasks if we don't call |
| 2009 | * balance_pgdat after returning from the refrigerator |
| 2010 | */ |
| 2011 | balance_pgdat(pgdat, order); |
| 2012 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2013 | } |
| 2014 | return 0; |
| 2015 | } |
| 2016 | |
| 2017 | /* |
| 2018 | * A zone is low on free memory, so wake its kswapd task to service it. |
| 2019 | */ |
| 2020 | void wakeup_kswapd(struct zone *zone, int order) |
| 2021 | { |
| 2022 | pg_data_t *pgdat; |
| 2023 | |
Con Kolivas | f3fe651 | 2006-01-06 00:11:15 -0800 | [diff] [blame] | 2024 | if (!populated_zone(zone)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2025 | return; |
| 2026 | |
| 2027 | pgdat = zone->zone_pgdat; |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 2028 | if (zone_watermark_ok(zone, order, zone->pages_low, 0, 0)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2029 | return; |
| 2030 | if (pgdat->kswapd_max_order < order) |
| 2031 | pgdat->kswapd_max_order = order; |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2032 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2033 | return; |
Con Kolivas | 8d0986e | 2005-09-13 01:25:07 -0700 | [diff] [blame] | 2034 | if (!waitqueue_active(&pgdat->kswapd_wait)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2035 | return; |
Con Kolivas | 8d0986e | 2005-09-13 01:25:07 -0700 | [diff] [blame] | 2036 | wake_up_interruptible(&pgdat->kswapd_wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2037 | } |
| 2038 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2039 | unsigned long global_lru_pages(void) |
| 2040 | { |
| 2041 | return global_page_state(NR_ACTIVE_ANON) |
| 2042 | + global_page_state(NR_ACTIVE_FILE) |
| 2043 | + global_page_state(NR_INACTIVE_ANON) |
| 2044 | + global_page_state(NR_INACTIVE_FILE); |
| 2045 | } |
| 2046 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2047 | #ifdef CONFIG_PM |
| 2048 | /* |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2049 | * Helper function for shrink_all_memory(). Tries to reclaim 'nr_pages' pages |
| 2050 | * from LRU lists system-wide, for given pass and priority, and returns the |
| 2051 | * number of reclaimed pages |
| 2052 | * |
| 2053 | * For pass > 3 we also try to shrink the LRU lists that contain a few pages |
| 2054 | */ |
Nigel Cunningham | e07aa05 | 2006-12-22 01:07:21 -0800 | [diff] [blame] | 2055 | static unsigned long shrink_all_zones(unsigned long nr_pages, int prio, |
| 2056 | int pass, struct scan_control *sc) |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2057 | { |
| 2058 | struct zone *zone; |
| 2059 | unsigned long nr_to_scan, ret = 0; |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 2060 | enum lru_list l; |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2061 | |
| 2062 | for_each_zone(zone) { |
| 2063 | |
| 2064 | if (!populated_zone(zone)) |
| 2065 | continue; |
| 2066 | |
David Rientjes | e815af9 | 2007-10-16 23:25:54 -0700 | [diff] [blame] | 2067 | if (zone_is_all_unreclaimable(zone) && prio != DEF_PRIORITY) |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2068 | continue; |
| 2069 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 2070 | for_each_evictable_lru(l) { |
| 2071 | /* For pass = 0, we don't shrink the active list */ |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2072 | if (pass == 0 && |
| 2073 | (l == LRU_ACTIVE || l == LRU_ACTIVE_FILE)) |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 2074 | continue; |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2075 | |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 2076 | zone->lru[l].nr_scan += |
| 2077 | (zone_page_state(zone, NR_LRU_BASE + l) |
| 2078 | >> prio) + 1; |
| 2079 | if (zone->lru[l].nr_scan >= nr_pages || pass > 3) { |
| 2080 | zone->lru[l].nr_scan = 0; |
| 2081 | nr_to_scan = min(nr_pages, |
| 2082 | zone_page_state(zone, |
| 2083 | NR_LRU_BASE + l)); |
| 2084 | ret += shrink_list(l, nr_to_scan, zone, |
| 2085 | sc, prio); |
| 2086 | if (ret >= nr_pages) |
| 2087 | return ret; |
| 2088 | } |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2089 | } |
| 2090 | } |
| 2091 | |
| 2092 | return ret; |
| 2093 | } |
| 2094 | |
| 2095 | /* |
| 2096 | * Try to free `nr_pages' of memory, system-wide, and return the number of |
| 2097 | * freed pages. |
| 2098 | * |
| 2099 | * Rather than trying to age LRUs the aim is to preserve the overall |
| 2100 | * LRU order by reclaiming preferentially |
| 2101 | * inactive > active > active referenced > active mapped |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2102 | */ |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 2103 | unsigned long shrink_all_memory(unsigned long nr_pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2104 | { |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2105 | unsigned long lru_pages, nr_slab; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 2106 | unsigned long ret = 0; |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2107 | int pass; |
| 2108 | struct reclaim_state reclaim_state; |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2109 | struct scan_control sc = { |
| 2110 | .gfp_mask = GFP_KERNEL, |
| 2111 | .may_swap = 0, |
| 2112 | .swap_cluster_max = nr_pages, |
| 2113 | .may_writepage = 1, |
| 2114 | .swappiness = vm_swappiness, |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2115 | .isolate_pages = isolate_pages_global, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2116 | }; |
| 2117 | |
| 2118 | current->reclaim_state = &reclaim_state; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 2119 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2120 | lru_pages = global_lru_pages(); |
Christoph Lameter | 972d1a7 | 2006-09-25 23:31:51 -0700 | [diff] [blame] | 2121 | nr_slab = global_page_state(NR_SLAB_RECLAIMABLE); |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2122 | /* If slab caches are huge, it's better to hit them first */ |
| 2123 | while (nr_slab >= lru_pages) { |
| 2124 | reclaim_state.reclaimed_slab = 0; |
| 2125 | shrink_slab(nr_pages, sc.gfp_mask, lru_pages); |
| 2126 | if (!reclaim_state.reclaimed_slab) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2127 | break; |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2128 | |
| 2129 | ret += reclaim_state.reclaimed_slab; |
| 2130 | if (ret >= nr_pages) |
| 2131 | goto out; |
| 2132 | |
| 2133 | nr_slab -= reclaim_state.reclaimed_slab; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2134 | } |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2135 | |
| 2136 | /* |
| 2137 | * We try to shrink LRUs in 5 passes: |
| 2138 | * 0 = Reclaim from inactive_list only |
| 2139 | * 1 = Reclaim from active list but don't reclaim mapped |
| 2140 | * 2 = 2nd pass of type 1 |
| 2141 | * 3 = Reclaim mapped (normal reclaim) |
| 2142 | * 4 = 2nd pass of type 3 |
| 2143 | */ |
| 2144 | for (pass = 0; pass < 5; pass++) { |
| 2145 | int prio; |
| 2146 | |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2147 | /* Force reclaiming mapped pages in the passes #3 and #4 */ |
| 2148 | if (pass > 2) { |
| 2149 | sc.may_swap = 1; |
| 2150 | sc.swappiness = 100; |
| 2151 | } |
| 2152 | |
| 2153 | for (prio = DEF_PRIORITY; prio >= 0; prio--) { |
| 2154 | unsigned long nr_to_scan = nr_pages - ret; |
| 2155 | |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2156 | sc.nr_scanned = 0; |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2157 | ret += shrink_all_zones(nr_to_scan, prio, pass, &sc); |
| 2158 | if (ret >= nr_pages) |
| 2159 | goto out; |
| 2160 | |
| 2161 | reclaim_state.reclaimed_slab = 0; |
Andrew Morton | 76395d3 | 2007-01-05 16:37:05 -0800 | [diff] [blame] | 2162 | shrink_slab(sc.nr_scanned, sc.gfp_mask, |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2163 | global_lru_pages()); |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2164 | ret += reclaim_state.reclaimed_slab; |
| 2165 | if (ret >= nr_pages) |
| 2166 | goto out; |
| 2167 | |
| 2168 | if (sc.nr_scanned && prio < DEF_PRIORITY - 2) |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 2169 | congestion_wait(WRITE, HZ / 10); |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2170 | } |
Rafael J. Wysocki | 248a030 | 2006-03-22 00:09:04 -0800 | [diff] [blame] | 2171 | } |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2172 | |
| 2173 | /* |
| 2174 | * If ret = 0, we could not shrink LRUs, but there may be something |
| 2175 | * in slab caches |
| 2176 | */ |
Andrew Morton | 76395d3 | 2007-01-05 16:37:05 -0800 | [diff] [blame] | 2177 | if (!ret) { |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2178 | do { |
| 2179 | reclaim_state.reclaimed_slab = 0; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2180 | shrink_slab(nr_pages, sc.gfp_mask, global_lru_pages()); |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2181 | ret += reclaim_state.reclaimed_slab; |
| 2182 | } while (ret < nr_pages && reclaim_state.reclaimed_slab > 0); |
Andrew Morton | 76395d3 | 2007-01-05 16:37:05 -0800 | [diff] [blame] | 2183 | } |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2184 | |
| 2185 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2186 | current->reclaim_state = NULL; |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2187 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2188 | return ret; |
| 2189 | } |
| 2190 | #endif |
| 2191 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2192 | /* It's optimal to keep kswapds on the same CPUs as their memory, but |
| 2193 | not required for correctness. So if the last cpu in a node goes |
| 2194 | away, we get changed to run anywhere: as the first one comes back, |
| 2195 | restore their cpu bindings. */ |
Chandra Seetharaman | 9c7b216 | 2006-06-27 02:54:07 -0700 | [diff] [blame] | 2196 | static int __devinit cpu_callback(struct notifier_block *nfb, |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 2197 | unsigned long action, void *hcpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2198 | { |
Yasunori Goto | 58c0a4a | 2007-10-16 01:25:40 -0700 | [diff] [blame] | 2199 | int nid; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2200 | |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 2201 | if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN) { |
Yasunori Goto | 58c0a4a | 2007-10-16 01:25:40 -0700 | [diff] [blame] | 2202 | for_each_node_state(nid, N_HIGH_MEMORY) { |
Mike Travis | c5f59f0 | 2008-04-04 18:11:10 -0700 | [diff] [blame] | 2203 | pg_data_t *pgdat = NODE_DATA(nid); |
| 2204 | node_to_cpumask_ptr(mask, pgdat->node_id); |
| 2205 | |
Rusty Russell | 3e59794 | 2009-01-01 10:12:24 +1030 | [diff] [blame] | 2206 | if (cpumask_any_and(cpu_online_mask, mask) < nr_cpu_ids) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2207 | /* One of our CPUs online: restore mask */ |
Mike Travis | c5f59f0 | 2008-04-04 18:11:10 -0700 | [diff] [blame] | 2208 | set_cpus_allowed_ptr(pgdat->kswapd, mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2209 | } |
| 2210 | } |
| 2211 | return NOTIFY_OK; |
| 2212 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2213 | |
Yasunori Goto | 3218ae1 | 2006-06-27 02:53:33 -0700 | [diff] [blame] | 2214 | /* |
| 2215 | * This kswapd start function will be called by init and node-hot-add. |
| 2216 | * On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added. |
| 2217 | */ |
| 2218 | int kswapd_run(int nid) |
| 2219 | { |
| 2220 | pg_data_t *pgdat = NODE_DATA(nid); |
| 2221 | int ret = 0; |
| 2222 | |
| 2223 | if (pgdat->kswapd) |
| 2224 | return 0; |
| 2225 | |
| 2226 | pgdat->kswapd = kthread_run(kswapd, pgdat, "kswapd%d", nid); |
| 2227 | if (IS_ERR(pgdat->kswapd)) { |
| 2228 | /* failure at boot is fatal */ |
| 2229 | BUG_ON(system_state == SYSTEM_BOOTING); |
| 2230 | printk("Failed to start kswapd on node %d\n",nid); |
| 2231 | ret = -1; |
| 2232 | } |
| 2233 | return ret; |
| 2234 | } |
| 2235 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2236 | static int __init kswapd_init(void) |
| 2237 | { |
Yasunori Goto | 3218ae1 | 2006-06-27 02:53:33 -0700 | [diff] [blame] | 2238 | int nid; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 2239 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2240 | swap_setup(); |
Christoph Lameter | 9422ffb | 2007-10-16 01:25:31 -0700 | [diff] [blame] | 2241 | for_each_node_state(nid, N_HIGH_MEMORY) |
Yasunori Goto | 3218ae1 | 2006-06-27 02:53:33 -0700 | [diff] [blame] | 2242 | kswapd_run(nid); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2243 | hotcpu_notifier(cpu_callback, 0); |
| 2244 | return 0; |
| 2245 | } |
| 2246 | |
| 2247 | module_init(kswapd_init) |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2248 | |
| 2249 | #ifdef CONFIG_NUMA |
| 2250 | /* |
| 2251 | * Zone reclaim mode |
| 2252 | * |
| 2253 | * If non-zero call zone_reclaim when the number of free pages falls below |
| 2254 | * the watermarks. |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2255 | */ |
| 2256 | int zone_reclaim_mode __read_mostly; |
| 2257 | |
Christoph Lameter | 1b2ffb7 | 2006-02-01 03:05:34 -0800 | [diff] [blame] | 2258 | #define RECLAIM_OFF 0 |
Fernando Luis Vazquez Cao | 7d03431 | 2008-07-29 22:33:41 -0700 | [diff] [blame] | 2259 | #define RECLAIM_ZONE (1<<0) /* Run shrink_inactive_list on the zone */ |
Christoph Lameter | 1b2ffb7 | 2006-02-01 03:05:34 -0800 | [diff] [blame] | 2260 | #define RECLAIM_WRITE (1<<1) /* Writeout pages during reclaim */ |
| 2261 | #define RECLAIM_SWAP (1<<2) /* Swap pages out during reclaim */ |
| 2262 | |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2263 | /* |
Christoph Lameter | a92f712 | 2006-02-01 03:05:32 -0800 | [diff] [blame] | 2264 | * Priority for ZONE_RECLAIM. This determines the fraction of pages |
| 2265 | * of a node considered for each zone_reclaim. 4 scans 1/16th of |
| 2266 | * a zone. |
| 2267 | */ |
| 2268 | #define ZONE_RECLAIM_PRIORITY 4 |
| 2269 | |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2270 | /* |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 2271 | * Percentage of pages in a zone that must be unmapped for zone_reclaim to |
| 2272 | * occur. |
| 2273 | */ |
| 2274 | int sysctl_min_unmapped_ratio = 1; |
| 2275 | |
| 2276 | /* |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 2277 | * If the number of slab pages in a zone grows beyond this percentage then |
| 2278 | * slab reclaim needs to occur. |
| 2279 | */ |
| 2280 | int sysctl_min_slab_ratio = 5; |
| 2281 | |
| 2282 | /* |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2283 | * Try to free up some pages from this zone through reclaim. |
| 2284 | */ |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2285 | static int __zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2286 | { |
Christoph Lameter | 7fb2d46 | 2006-03-22 00:08:22 -0800 | [diff] [blame] | 2287 | /* Minimum pages needed in order to stay on node */ |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 2288 | const unsigned long nr_pages = 1 << order; |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2289 | struct task_struct *p = current; |
| 2290 | struct reclaim_state reclaim_state; |
Christoph Lameter | 8695949 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2291 | int priority; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2292 | struct scan_control sc = { |
| 2293 | .may_writepage = !!(zone_reclaim_mode & RECLAIM_WRITE), |
| 2294 | .may_swap = !!(zone_reclaim_mode & RECLAIM_SWAP), |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 2295 | .swap_cluster_max = max_t(unsigned long, nr_pages, |
| 2296 | SWAP_CLUSTER_MAX), |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2297 | .gfp_mask = gfp_mask, |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2298 | .swappiness = vm_swappiness, |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2299 | .isolate_pages = isolate_pages_global, |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2300 | }; |
Christoph Lameter | 83e33a4 | 2006-09-25 23:31:53 -0700 | [diff] [blame] | 2301 | unsigned long slab_reclaimable; |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2302 | |
| 2303 | disable_swap_token(); |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2304 | cond_resched(); |
Christoph Lameter | d4f7796 | 2006-02-24 13:04:22 -0800 | [diff] [blame] | 2305 | /* |
| 2306 | * We need to be able to allocate from the reserves for RECLAIM_SWAP |
| 2307 | * and we also need to be able to write out pages for RECLAIM_WRITE |
| 2308 | * and RECLAIM_SWAP. |
| 2309 | */ |
| 2310 | p->flags |= PF_MEMALLOC | PF_SWAPWRITE; |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2311 | reclaim_state.reclaimed_slab = 0; |
| 2312 | p->reclaim_state = &reclaim_state; |
Christoph Lameter | c84db23 | 2006-02-01 03:05:29 -0800 | [diff] [blame] | 2313 | |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 2314 | if (zone_page_state(zone, NR_FILE_PAGES) - |
| 2315 | zone_page_state(zone, NR_FILE_MAPPED) > |
| 2316 | zone->min_unmapped_pages) { |
| 2317 | /* |
| 2318 | * Free memory by calling shrink zone with increasing |
| 2319 | * priorities until we have enough memory freed. |
| 2320 | */ |
| 2321 | priority = ZONE_RECLAIM_PRIORITY; |
| 2322 | do { |
Martin Bligh | 3bb1a85 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 2323 | note_zone_scanning_priority(zone, priority); |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2324 | shrink_zone(priority, zone, &sc); |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 2325 | priority--; |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2326 | } while (priority >= 0 && sc.nr_reclaimed < nr_pages); |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 2327 | } |
Christoph Lameter | c84db23 | 2006-02-01 03:05:29 -0800 | [diff] [blame] | 2328 | |
Christoph Lameter | 83e33a4 | 2006-09-25 23:31:53 -0700 | [diff] [blame] | 2329 | slab_reclaimable = zone_page_state(zone, NR_SLAB_RECLAIMABLE); |
| 2330 | if (slab_reclaimable > zone->min_slab_pages) { |
Christoph Lameter | 2a16e3f | 2006-02-01 03:05:35 -0800 | [diff] [blame] | 2331 | /* |
Christoph Lameter | 7fb2d46 | 2006-03-22 00:08:22 -0800 | [diff] [blame] | 2332 | * shrink_slab() does not currently allow us to determine how |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 2333 | * many pages were freed in this zone. So we take the current |
| 2334 | * number of slab pages and shake the slab until it is reduced |
| 2335 | * by the same nr_pages that we used for reclaiming unmapped |
| 2336 | * pages. |
Christoph Lameter | 2a16e3f | 2006-02-01 03:05:35 -0800 | [diff] [blame] | 2337 | * |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 2338 | * Note that shrink_slab will free memory on all zones and may |
| 2339 | * take a long time. |
Christoph Lameter | 2a16e3f | 2006-02-01 03:05:35 -0800 | [diff] [blame] | 2340 | */ |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 2341 | while (shrink_slab(sc.nr_scanned, gfp_mask, order) && |
Christoph Lameter | 83e33a4 | 2006-09-25 23:31:53 -0700 | [diff] [blame] | 2342 | zone_page_state(zone, NR_SLAB_RECLAIMABLE) > |
| 2343 | slab_reclaimable - nr_pages) |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 2344 | ; |
Christoph Lameter | 83e33a4 | 2006-09-25 23:31:53 -0700 | [diff] [blame] | 2345 | |
| 2346 | /* |
| 2347 | * Update nr_reclaimed by the number of slab pages we |
| 2348 | * reclaimed from this zone. |
| 2349 | */ |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2350 | sc.nr_reclaimed += slab_reclaimable - |
Christoph Lameter | 83e33a4 | 2006-09-25 23:31:53 -0700 | [diff] [blame] | 2351 | zone_page_state(zone, NR_SLAB_RECLAIMABLE); |
Christoph Lameter | 2a16e3f | 2006-02-01 03:05:35 -0800 | [diff] [blame] | 2352 | } |
| 2353 | |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2354 | p->reclaim_state = NULL; |
Christoph Lameter | d4f7796 | 2006-02-24 13:04:22 -0800 | [diff] [blame] | 2355 | current->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE); |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2356 | return sc.nr_reclaimed >= nr_pages; |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2357 | } |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2358 | |
| 2359 | int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) |
| 2360 | { |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2361 | int node_id; |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 2362 | int ret; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2363 | |
| 2364 | /* |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 2365 | * Zone reclaim reclaims unmapped file backed pages and |
| 2366 | * slab pages if we are over the defined limits. |
Christoph Lameter | 34aa133 | 2006-06-30 01:55:37 -0700 | [diff] [blame] | 2367 | * |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 2368 | * A small portion of unmapped file backed pages is needed for |
| 2369 | * file I/O otherwise pages read by file I/O will be immediately |
| 2370 | * thrown out if the zone is overallocated. So we do not reclaim |
| 2371 | * if less than a specified percentage of the zone is used by |
| 2372 | * unmapped file backed pages. |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2373 | */ |
Christoph Lameter | 34aa133 | 2006-06-30 01:55:37 -0700 | [diff] [blame] | 2374 | if (zone_page_state(zone, NR_FILE_PAGES) - |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 2375 | zone_page_state(zone, NR_FILE_MAPPED) <= zone->min_unmapped_pages |
| 2376 | && zone_page_state(zone, NR_SLAB_RECLAIMABLE) |
| 2377 | <= zone->min_slab_pages) |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 2378 | return 0; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2379 | |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 2380 | if (zone_is_all_unreclaimable(zone)) |
| 2381 | return 0; |
| 2382 | |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2383 | /* |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 2384 | * Do not scan if the allocation should not be delayed. |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2385 | */ |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 2386 | if (!(gfp_mask & __GFP_WAIT) || (current->flags & PF_MEMALLOC)) |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2387 | return 0; |
| 2388 | |
| 2389 | /* |
| 2390 | * Only run zone reclaim on the local zone or on zones that do not |
| 2391 | * have associated processors. This will favor the local processor |
| 2392 | * over remote processors and spread off node memory allocations |
| 2393 | * as wide as possible. |
| 2394 | */ |
Christoph Lameter | 89fa302 | 2006-09-25 23:31:55 -0700 | [diff] [blame] | 2395 | node_id = zone_to_nid(zone); |
Christoph Lameter | 37c0708 | 2007-10-16 01:25:36 -0700 | [diff] [blame] | 2396 | if (node_state(node_id, N_CPU) && node_id != numa_node_id()) |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2397 | return 0; |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 2398 | |
| 2399 | if (zone_test_and_set_flag(zone, ZONE_RECLAIM_LOCKED)) |
| 2400 | return 0; |
| 2401 | ret = __zone_reclaim(zone, gfp_mask, order); |
| 2402 | zone_clear_flag(zone, ZONE_RECLAIM_LOCKED); |
| 2403 | |
| 2404 | return ret; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2405 | } |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2406 | #endif |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 2407 | |
| 2408 | #ifdef CONFIG_UNEVICTABLE_LRU |
| 2409 | /* |
| 2410 | * page_evictable - test whether a page is evictable |
| 2411 | * @page: the page to test |
| 2412 | * @vma: the VMA in which the page is or will be mapped, may be NULL |
| 2413 | * |
| 2414 | * Test whether page is evictable--i.e., should be placed on active/inactive |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 2415 | * lists vs unevictable list. The vma argument is !NULL when called from the |
| 2416 | * fault path to determine how to instantate a new page. |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 2417 | * |
| 2418 | * Reasons page might not be evictable: |
Lee Schermerhorn | ba9ddf4 | 2008-10-18 20:26:42 -0700 | [diff] [blame] | 2419 | * (1) page's mapping marked unevictable |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 2420 | * (2) page is part of an mlocked VMA |
Lee Schermerhorn | ba9ddf4 | 2008-10-18 20:26:42 -0700 | [diff] [blame] | 2421 | * |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 2422 | */ |
| 2423 | int page_evictable(struct page *page, struct vm_area_struct *vma) |
| 2424 | { |
| 2425 | |
Lee Schermerhorn | ba9ddf4 | 2008-10-18 20:26:42 -0700 | [diff] [blame] | 2426 | if (mapping_unevictable(page_mapping(page))) |
| 2427 | return 0; |
| 2428 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 2429 | if (PageMlocked(page) || (vma && is_mlocked_vma(vma, page))) |
| 2430 | return 0; |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 2431 | |
| 2432 | return 1; |
| 2433 | } |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 2434 | |
| 2435 | /** |
| 2436 | * check_move_unevictable_page - check page for evictability and move to appropriate zone lru list |
| 2437 | * @page: page to check evictability and move to appropriate lru list |
| 2438 | * @zone: zone page is in |
| 2439 | * |
| 2440 | * Checks a page for evictability and moves the page to the appropriate |
| 2441 | * zone lru list. |
| 2442 | * |
| 2443 | * Restrictions: zone->lru_lock must be held, page must be on LRU and must |
| 2444 | * have PageUnevictable set. |
| 2445 | */ |
| 2446 | static void check_move_unevictable_page(struct page *page, struct zone *zone) |
| 2447 | { |
| 2448 | VM_BUG_ON(PageActive(page)); |
| 2449 | |
| 2450 | retry: |
| 2451 | ClearPageUnevictable(page); |
| 2452 | if (page_evictable(page, NULL)) { |
| 2453 | enum lru_list l = LRU_INACTIVE_ANON + page_is_file_cache(page); |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 2454 | |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 2455 | __dec_zone_state(zone, NR_UNEVICTABLE); |
| 2456 | list_move(&page->lru, &zone->lru[l].list); |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2457 | mem_cgroup_move_lists(page, LRU_UNEVICTABLE, l); |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 2458 | __inc_zone_state(zone, NR_INACTIVE_ANON + l); |
| 2459 | __count_vm_event(UNEVICTABLE_PGRESCUED); |
| 2460 | } else { |
| 2461 | /* |
| 2462 | * rotate unevictable list |
| 2463 | */ |
| 2464 | SetPageUnevictable(page); |
| 2465 | list_move(&page->lru, &zone->lru[LRU_UNEVICTABLE].list); |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2466 | mem_cgroup_rotate_lru_list(page, LRU_UNEVICTABLE); |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 2467 | if (page_evictable(page, NULL)) |
| 2468 | goto retry; |
| 2469 | } |
| 2470 | } |
| 2471 | |
| 2472 | /** |
| 2473 | * scan_mapping_unevictable_pages - scan an address space for evictable pages |
| 2474 | * @mapping: struct address_space to scan for evictable pages |
| 2475 | * |
| 2476 | * Scan all pages in mapping. Check unevictable pages for |
| 2477 | * evictability and move them to the appropriate zone lru list. |
| 2478 | */ |
| 2479 | void scan_mapping_unevictable_pages(struct address_space *mapping) |
| 2480 | { |
| 2481 | pgoff_t next = 0; |
| 2482 | pgoff_t end = (i_size_read(mapping->host) + PAGE_CACHE_SIZE - 1) >> |
| 2483 | PAGE_CACHE_SHIFT; |
| 2484 | struct zone *zone; |
| 2485 | struct pagevec pvec; |
| 2486 | |
| 2487 | if (mapping->nrpages == 0) |
| 2488 | return; |
| 2489 | |
| 2490 | pagevec_init(&pvec, 0); |
| 2491 | while (next < end && |
| 2492 | pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { |
| 2493 | int i; |
| 2494 | int pg_scanned = 0; |
| 2495 | |
| 2496 | zone = NULL; |
| 2497 | |
| 2498 | for (i = 0; i < pagevec_count(&pvec); i++) { |
| 2499 | struct page *page = pvec.pages[i]; |
| 2500 | pgoff_t page_index = page->index; |
| 2501 | struct zone *pagezone = page_zone(page); |
| 2502 | |
| 2503 | pg_scanned++; |
| 2504 | if (page_index > next) |
| 2505 | next = page_index; |
| 2506 | next++; |
| 2507 | |
| 2508 | if (pagezone != zone) { |
| 2509 | if (zone) |
| 2510 | spin_unlock_irq(&zone->lru_lock); |
| 2511 | zone = pagezone; |
| 2512 | spin_lock_irq(&zone->lru_lock); |
| 2513 | } |
| 2514 | |
| 2515 | if (PageLRU(page) && PageUnevictable(page)) |
| 2516 | check_move_unevictable_page(page, zone); |
| 2517 | } |
| 2518 | if (zone) |
| 2519 | spin_unlock_irq(&zone->lru_lock); |
| 2520 | pagevec_release(&pvec); |
| 2521 | |
| 2522 | count_vm_events(UNEVICTABLE_PGSCANNED, pg_scanned); |
| 2523 | } |
| 2524 | |
| 2525 | } |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 2526 | |
| 2527 | /** |
| 2528 | * scan_zone_unevictable_pages - check unevictable list for evictable pages |
| 2529 | * @zone - zone of which to scan the unevictable list |
| 2530 | * |
| 2531 | * Scan @zone's unevictable LRU lists to check for pages that have become |
| 2532 | * evictable. Move those that have to @zone's inactive list where they |
| 2533 | * become candidates for reclaim, unless shrink_inactive_zone() decides |
| 2534 | * to reactivate them. Pages that are still unevictable are rotated |
| 2535 | * back onto @zone's unevictable list. |
| 2536 | */ |
| 2537 | #define SCAN_UNEVICTABLE_BATCH_SIZE 16UL /* arbitrary lock hold batch size */ |
KOSAKI Motohiro | 14b90b2 | 2009-01-06 14:39:45 -0800 | [diff] [blame] | 2538 | static void scan_zone_unevictable_pages(struct zone *zone) |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 2539 | { |
| 2540 | struct list_head *l_unevictable = &zone->lru[LRU_UNEVICTABLE].list; |
| 2541 | unsigned long scan; |
| 2542 | unsigned long nr_to_scan = zone_page_state(zone, NR_UNEVICTABLE); |
| 2543 | |
| 2544 | while (nr_to_scan > 0) { |
| 2545 | unsigned long batch_size = min(nr_to_scan, |
| 2546 | SCAN_UNEVICTABLE_BATCH_SIZE); |
| 2547 | |
| 2548 | spin_lock_irq(&zone->lru_lock); |
| 2549 | for (scan = 0; scan < batch_size; scan++) { |
| 2550 | struct page *page = lru_to_page(l_unevictable); |
| 2551 | |
| 2552 | if (!trylock_page(page)) |
| 2553 | continue; |
| 2554 | |
| 2555 | prefetchw_prev_lru_page(page, l_unevictable, flags); |
| 2556 | |
| 2557 | if (likely(PageLRU(page) && PageUnevictable(page))) |
| 2558 | check_move_unevictable_page(page, zone); |
| 2559 | |
| 2560 | unlock_page(page); |
| 2561 | } |
| 2562 | spin_unlock_irq(&zone->lru_lock); |
| 2563 | |
| 2564 | nr_to_scan -= batch_size; |
| 2565 | } |
| 2566 | } |
| 2567 | |
| 2568 | |
| 2569 | /** |
| 2570 | * scan_all_zones_unevictable_pages - scan all unevictable lists for evictable pages |
| 2571 | * |
| 2572 | * A really big hammer: scan all zones' unevictable LRU lists to check for |
| 2573 | * pages that have become evictable. Move those back to the zones' |
| 2574 | * inactive list where they become candidates for reclaim. |
| 2575 | * This occurs when, e.g., we have unswappable pages on the unevictable lists, |
| 2576 | * and we add swap to the system. As such, it runs in the context of a task |
| 2577 | * that has possibly/probably made some previously unevictable pages |
| 2578 | * evictable. |
| 2579 | */ |
KOSAKI Motohiro | ff30153 | 2009-01-06 14:39:44 -0800 | [diff] [blame] | 2580 | static void scan_all_zones_unevictable_pages(void) |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 2581 | { |
| 2582 | struct zone *zone; |
| 2583 | |
| 2584 | for_each_zone(zone) { |
| 2585 | scan_zone_unevictable_pages(zone); |
| 2586 | } |
| 2587 | } |
| 2588 | |
| 2589 | /* |
| 2590 | * scan_unevictable_pages [vm] sysctl handler. On demand re-scan of |
| 2591 | * all nodes' unevictable lists for evictable pages |
| 2592 | */ |
| 2593 | unsigned long scan_unevictable_pages; |
| 2594 | |
| 2595 | int scan_unevictable_handler(struct ctl_table *table, int write, |
| 2596 | struct file *file, void __user *buffer, |
| 2597 | size_t *length, loff_t *ppos) |
| 2598 | { |
| 2599 | proc_doulongvec_minmax(table, write, file, buffer, length, ppos); |
| 2600 | |
| 2601 | if (write && *(unsigned long *)table->data) |
| 2602 | scan_all_zones_unevictable_pages(); |
| 2603 | |
| 2604 | scan_unevictable_pages = 0; |
| 2605 | return 0; |
| 2606 | } |
| 2607 | |
| 2608 | /* |
| 2609 | * per node 'scan_unevictable_pages' attribute. On demand re-scan of |
| 2610 | * a specified node's per zone unevictable lists for evictable pages. |
| 2611 | */ |
| 2612 | |
| 2613 | static ssize_t read_scan_unevictable_node(struct sys_device *dev, |
| 2614 | struct sysdev_attribute *attr, |
| 2615 | char *buf) |
| 2616 | { |
| 2617 | return sprintf(buf, "0\n"); /* always zero; should fit... */ |
| 2618 | } |
| 2619 | |
| 2620 | static ssize_t write_scan_unevictable_node(struct sys_device *dev, |
| 2621 | struct sysdev_attribute *attr, |
| 2622 | const char *buf, size_t count) |
| 2623 | { |
| 2624 | struct zone *node_zones = NODE_DATA(dev->id)->node_zones; |
| 2625 | struct zone *zone; |
| 2626 | unsigned long res; |
| 2627 | unsigned long req = strict_strtoul(buf, 10, &res); |
| 2628 | |
| 2629 | if (!req) |
| 2630 | return 1; /* zero is no-op */ |
| 2631 | |
| 2632 | for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) { |
| 2633 | if (!populated_zone(zone)) |
| 2634 | continue; |
| 2635 | scan_zone_unevictable_pages(zone); |
| 2636 | } |
| 2637 | return 1; |
| 2638 | } |
| 2639 | |
| 2640 | |
| 2641 | static SYSDEV_ATTR(scan_unevictable_pages, S_IRUGO | S_IWUSR, |
| 2642 | read_scan_unevictable_node, |
| 2643 | write_scan_unevictable_node); |
| 2644 | |
| 2645 | int scan_unevictable_register_node(struct node *node) |
| 2646 | { |
| 2647 | return sysdev_create_file(&node->sysdev, &attr_scan_unevictable_pages); |
| 2648 | } |
| 2649 | |
| 2650 | void scan_unevictable_unregister_node(struct node *node) |
| 2651 | { |
| 2652 | sysdev_remove_file(&node->sysdev, &attr_scan_unevictable_pages); |
| 2653 | } |
| 2654 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 2655 | #endif |