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 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 133 | /* |
| 134 | * Add a shrinker callback to be called from the vm |
| 135 | */ |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 136 | void register_shrinker(struct shrinker *shrinker) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 137 | { |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 138 | shrinker->nr = 0; |
| 139 | down_write(&shrinker_rwsem); |
| 140 | list_add_tail(&shrinker->list, &shrinker_list); |
| 141 | up_write(&shrinker_rwsem); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 142 | } |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 143 | EXPORT_SYMBOL(register_shrinker); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 144 | |
| 145 | /* |
| 146 | * Remove one |
| 147 | */ |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 148 | void unregister_shrinker(struct shrinker *shrinker) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 149 | { |
| 150 | down_write(&shrinker_rwsem); |
| 151 | list_del(&shrinker->list); |
| 152 | up_write(&shrinker_rwsem); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 153 | } |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 154 | EXPORT_SYMBOL(unregister_shrinker); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 155 | |
| 156 | #define SHRINK_BATCH 128 |
| 157 | /* |
| 158 | * Call the shrink functions to age shrinkable caches |
| 159 | * |
| 160 | * Here we assume it costs one seek to replace a lru page and that it also |
| 161 | * takes a seek to recreate a cache object. With this in mind we age equal |
| 162 | * percentages of the lru and ageable caches. This should balance the seeks |
| 163 | * generated by these structures. |
| 164 | * |
Simon Arlott | 183ff22 | 2007-10-20 01:27:18 +0200 | [diff] [blame] | 165 | * 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] | 166 | * slab to avoid swapping. |
| 167 | * |
| 168 | * We do weird things to avoid (scanned*seeks*entries) overflowing 32 bits. |
| 169 | * |
| 170 | * `lru_pages' represents the number of on-LRU pages in all the zones which |
| 171 | * are eligible for the caller's allocation attempt. It is used for balancing |
| 172 | * slab reclaim versus page reclaim. |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 173 | * |
| 174 | * Returns the number of slab objects which we shrunk. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 175 | */ |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 176 | unsigned long shrink_slab(unsigned long scanned, gfp_t gfp_mask, |
| 177 | unsigned long lru_pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 178 | { |
| 179 | struct shrinker *shrinker; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 180 | unsigned long ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 181 | |
| 182 | if (scanned == 0) |
| 183 | scanned = SWAP_CLUSTER_MAX; |
| 184 | |
| 185 | if (!down_read_trylock(&shrinker_rwsem)) |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 186 | return 1; /* Assume we'll be able to shrink next time */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 187 | |
| 188 | list_for_each_entry(shrinker, &shrinker_list, list) { |
| 189 | unsigned long long delta; |
| 190 | unsigned long total_scan; |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 191 | unsigned long max_pass = (*shrinker->shrink)(0, gfp_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 192 | |
| 193 | delta = (4 * scanned) / shrinker->seeks; |
Andrea Arcangeli | ea164d7 | 2005-11-28 13:44:15 -0800 | [diff] [blame] | 194 | delta *= max_pass; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 195 | do_div(delta, lru_pages + 1); |
| 196 | shrinker->nr += delta; |
Andrea Arcangeli | ea164d7 | 2005-11-28 13:44:15 -0800 | [diff] [blame] | 197 | if (shrinker->nr < 0) { |
| 198 | printk(KERN_ERR "%s: nr=%ld\n", |
Harvey Harrison | d40cee2 | 2008-04-30 00:55:07 -0700 | [diff] [blame] | 199 | __func__, shrinker->nr); |
Andrea Arcangeli | ea164d7 | 2005-11-28 13:44:15 -0800 | [diff] [blame] | 200 | shrinker->nr = max_pass; |
| 201 | } |
| 202 | |
| 203 | /* |
| 204 | * Avoid risking looping forever due to too large nr value: |
| 205 | * never try to free more than twice the estimate number of |
| 206 | * freeable entries. |
| 207 | */ |
| 208 | if (shrinker->nr > max_pass * 2) |
| 209 | shrinker->nr = max_pass * 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 210 | |
| 211 | total_scan = shrinker->nr; |
| 212 | shrinker->nr = 0; |
| 213 | |
| 214 | while (total_scan >= SHRINK_BATCH) { |
| 215 | long this_scan = SHRINK_BATCH; |
| 216 | int shrink_ret; |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 217 | int nr_before; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 218 | |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 219 | nr_before = (*shrinker->shrink)(0, gfp_mask); |
| 220 | shrink_ret = (*shrinker->shrink)(this_scan, gfp_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 221 | if (shrink_ret == -1) |
| 222 | break; |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 223 | if (shrink_ret < nr_before) |
| 224 | ret += nr_before - shrink_ret; |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 225 | count_vm_events(SLABS_SCANNED, this_scan); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 226 | total_scan -= this_scan; |
| 227 | |
| 228 | cond_resched(); |
| 229 | } |
| 230 | |
| 231 | shrinker->nr += total_scan; |
| 232 | } |
| 233 | up_read(&shrinker_rwsem); |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 234 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 235 | } |
| 236 | |
| 237 | /* Called without lock on whether page is mapped, so answer is unstable */ |
| 238 | static inline int page_mapping_inuse(struct page *page) |
| 239 | { |
| 240 | struct address_space *mapping; |
| 241 | |
| 242 | /* Page is in somebody's page tables. */ |
| 243 | if (page_mapped(page)) |
| 244 | return 1; |
| 245 | |
| 246 | /* Be more reluctant to reclaim swapcache than pagecache */ |
| 247 | if (PageSwapCache(page)) |
| 248 | return 1; |
| 249 | |
| 250 | mapping = page_mapping(page); |
| 251 | if (!mapping) |
| 252 | return 0; |
| 253 | |
| 254 | /* File is mmap'd by somebody? */ |
| 255 | return mapping_mapped(mapping); |
| 256 | } |
| 257 | |
| 258 | static inline int is_page_cache_freeable(struct page *page) |
| 259 | { |
| 260 | return page_count(page) - !!PagePrivate(page) == 2; |
| 261 | } |
| 262 | |
| 263 | static int may_write_to_queue(struct backing_dev_info *bdi) |
| 264 | { |
Christoph Lameter | 930d915 | 2006-01-08 01:00:47 -0800 | [diff] [blame] | 265 | if (current->flags & PF_SWAPWRITE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 266 | return 1; |
| 267 | if (!bdi_write_congested(bdi)) |
| 268 | return 1; |
| 269 | if (bdi == current->backing_dev_info) |
| 270 | return 1; |
| 271 | return 0; |
| 272 | } |
| 273 | |
| 274 | /* |
| 275 | * We detected a synchronous write error writing a page out. Probably |
| 276 | * -ENOSPC. We need to propagate that into the address_space for a subsequent |
| 277 | * fsync(), msync() or close(). |
| 278 | * |
| 279 | * The tricky part is that after writepage we cannot touch the mapping: nothing |
| 280 | * prevents it from being freed up. But we have a ref on the page and once |
| 281 | * that page is locked, the mapping is pinned. |
| 282 | * |
| 283 | * We're allowed to run sleeping lock_page() here because we know the caller has |
| 284 | * __GFP_FS. |
| 285 | */ |
| 286 | static void handle_write_error(struct address_space *mapping, |
| 287 | struct page *page, int error) |
| 288 | { |
| 289 | lock_page(page); |
Guillaume Chazarain | 3e9f45b | 2007-05-08 00:23:25 -0700 | [diff] [blame] | 290 | if (page_mapping(page) == mapping) |
| 291 | mapping_set_error(mapping, error); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 292 | unlock_page(page); |
| 293 | } |
| 294 | |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 295 | /* Request for sync pageout. */ |
| 296 | enum pageout_io { |
| 297 | PAGEOUT_IO_ASYNC, |
| 298 | PAGEOUT_IO_SYNC, |
| 299 | }; |
| 300 | |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 301 | /* possible outcome of pageout() */ |
| 302 | typedef enum { |
| 303 | /* failed to write page out, page is locked */ |
| 304 | PAGE_KEEP, |
| 305 | /* move page to the active list, page is locked */ |
| 306 | PAGE_ACTIVATE, |
| 307 | /* page has been sent to the disk successfully, page is unlocked */ |
| 308 | PAGE_SUCCESS, |
| 309 | /* page is clean and locked */ |
| 310 | PAGE_CLEAN, |
| 311 | } pageout_t; |
| 312 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 313 | /* |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 314 | * pageout is called by shrink_page_list() for each dirty page. |
| 315 | * Calls ->writepage(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 316 | */ |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 317 | static pageout_t pageout(struct page *page, struct address_space *mapping, |
| 318 | enum pageout_io sync_writeback) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 319 | { |
| 320 | /* |
| 321 | * If the page is dirty, only perform writeback if that write |
| 322 | * will be non-blocking. To prevent this allocation from being |
| 323 | * stalled by pagecache activity. But note that there may be |
| 324 | * stalls if we need to run get_block(). We could test |
| 325 | * PagePrivate for that. |
| 326 | * |
| 327 | * If this process is currently in generic_file_write() against |
| 328 | * this page's queue, we can perform writeback even if that |
| 329 | * will block. |
| 330 | * |
| 331 | * If the page is swapcache, write it back even if that would |
| 332 | * block, for some throttling. This happens by accident, because |
| 333 | * swap_backing_dev_info is bust: it doesn't reflect the |
| 334 | * congestion state of the swapdevs. Easy to fix, if needed. |
| 335 | * See swapfile.c:page_queue_congested(). |
| 336 | */ |
| 337 | if (!is_page_cache_freeable(page)) |
| 338 | return PAGE_KEEP; |
| 339 | if (!mapping) { |
| 340 | /* |
| 341 | * Some data journaling orphaned pages can have |
| 342 | * page->mapping == NULL while being dirty with clean buffers. |
| 343 | */ |
akpm@osdl.org | 323aca6 | 2005-04-16 15:24:06 -0700 | [diff] [blame] | 344 | if (PagePrivate(page)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 345 | if (try_to_free_buffers(page)) { |
| 346 | ClearPageDirty(page); |
Harvey Harrison | d40cee2 | 2008-04-30 00:55:07 -0700 | [diff] [blame] | 347 | printk("%s: orphaned page\n", __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 348 | return PAGE_CLEAN; |
| 349 | } |
| 350 | } |
| 351 | return PAGE_KEEP; |
| 352 | } |
| 353 | if (mapping->a_ops->writepage == NULL) |
| 354 | return PAGE_ACTIVATE; |
| 355 | if (!may_write_to_queue(mapping->backing_dev_info)) |
| 356 | return PAGE_KEEP; |
| 357 | |
| 358 | if (clear_page_dirty_for_io(page)) { |
| 359 | int res; |
| 360 | struct writeback_control wbc = { |
| 361 | .sync_mode = WB_SYNC_NONE, |
| 362 | .nr_to_write = SWAP_CLUSTER_MAX, |
OGAWA Hirofumi | 111ebb6 | 2006-06-23 02:03:26 -0700 | [diff] [blame] | 363 | .range_start = 0, |
| 364 | .range_end = LLONG_MAX, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 365 | .nonblocking = 1, |
| 366 | .for_reclaim = 1, |
| 367 | }; |
| 368 | |
| 369 | SetPageReclaim(page); |
| 370 | res = mapping->a_ops->writepage(page, &wbc); |
| 371 | if (res < 0) |
| 372 | handle_write_error(mapping, page, res); |
Zach Brown | 994fc28c | 2005-12-15 14:28:17 -0800 | [diff] [blame] | 373 | if (res == AOP_WRITEPAGE_ACTIVATE) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 374 | ClearPageReclaim(page); |
| 375 | return PAGE_ACTIVATE; |
| 376 | } |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 377 | |
| 378 | /* |
| 379 | * Wait on writeback if requested to. This happens when |
| 380 | * direct reclaiming a large contiguous area and the |
| 381 | * first attempt to free a range of pages fails. |
| 382 | */ |
| 383 | if (PageWriteback(page) && sync_writeback == PAGEOUT_IO_SYNC) |
| 384 | wait_on_page_writeback(page); |
| 385 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 386 | if (!PageWriteback(page)) { |
| 387 | /* synchronous write or broken a_ops? */ |
| 388 | ClearPageReclaim(page); |
| 389 | } |
Andrew Morton | e129b5c | 2006-09-27 01:50:00 -0700 | [diff] [blame] | 390 | inc_zone_page_state(page, NR_VMSCAN_WRITE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 391 | return PAGE_SUCCESS; |
| 392 | } |
| 393 | |
| 394 | return PAGE_CLEAN; |
| 395 | } |
| 396 | |
Andrew Morton | a649fd9 | 2006-10-17 00:09:36 -0700 | [diff] [blame] | 397 | /* |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 398 | * Same as remove_mapping, but if the page is removed from the mapping, it |
| 399 | * gets returned with a refcount of 0. |
Andrew Morton | a649fd9 | 2006-10-17 00:09:36 -0700 | [diff] [blame] | 400 | */ |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 401 | static int __remove_mapping(struct address_space *mapping, struct page *page) |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 402 | { |
Nick Piggin | 28e4d96 | 2006-09-25 23:31:23 -0700 | [diff] [blame] | 403 | BUG_ON(!PageLocked(page)); |
| 404 | BUG_ON(mapping != page_mapping(page)); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 405 | |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 406 | spin_lock_irq(&mapping->tree_lock); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 407 | /* |
Nick Piggin | 0fd0e6b | 2006-09-27 01:50:02 -0700 | [diff] [blame] | 408 | * The non racy check for a busy page. |
| 409 | * |
| 410 | * Must be careful with the order of the tests. When someone has |
| 411 | * a ref to the page, it may be possible that they dirty it then |
| 412 | * drop the reference. So if PageDirty is tested before page_count |
| 413 | * here, then the following race may occur: |
| 414 | * |
| 415 | * get_user_pages(&page); |
| 416 | * [user mapping goes away] |
| 417 | * write_to(page); |
| 418 | * !PageDirty(page) [good] |
| 419 | * SetPageDirty(page); |
| 420 | * put_page(page); |
| 421 | * !page_count(page) [good, discard it] |
| 422 | * |
| 423 | * [oops, our write_to data is lost] |
| 424 | * |
| 425 | * Reversing the order of the tests ensures such a situation cannot |
| 426 | * escape unnoticed. The smp_rmb is needed to ensure the page->flags |
| 427 | * load is not satisfied before that of page->_count. |
| 428 | * |
| 429 | * Note that if SetPageDirty is always performed via set_page_dirty, |
| 430 | * and thus under tree_lock, then this ordering is not required. |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 431 | */ |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 432 | if (!page_freeze_refs(page, 2)) |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 433 | goto cannot_free; |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 434 | /* note: atomic_cmpxchg in page_freeze_refs provides the smp_rmb */ |
| 435 | if (unlikely(PageDirty(page))) { |
| 436 | page_unfreeze_refs(page, 2); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 437 | goto cannot_free; |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 438 | } |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 439 | |
| 440 | if (PageSwapCache(page)) { |
| 441 | swp_entry_t swap = { .val = page_private(page) }; |
| 442 | __delete_from_swap_cache(page); |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 443 | spin_unlock_irq(&mapping->tree_lock); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 444 | swap_free(swap); |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 445 | } else { |
| 446 | __remove_from_page_cache(page); |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 447 | spin_unlock_irq(&mapping->tree_lock); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 448 | } |
| 449 | |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 450 | return 1; |
| 451 | |
| 452 | cannot_free: |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 453 | spin_unlock_irq(&mapping->tree_lock); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 454 | return 0; |
| 455 | } |
| 456 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 457 | /* |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 458 | * Attempt to detach a locked page from its ->mapping. If it is dirty or if |
| 459 | * someone else has a ref on the page, abort and return 0. If it was |
| 460 | * successfully detached, return 1. Assumes the caller has a single ref on |
| 461 | * this page. |
| 462 | */ |
| 463 | int remove_mapping(struct address_space *mapping, struct page *page) |
| 464 | { |
| 465 | if (__remove_mapping(mapping, page)) { |
| 466 | /* |
| 467 | * Unfreezing the refcount with 1 rather than 2 effectively |
| 468 | * drops the pagecache ref for us without requiring another |
| 469 | * atomic operation. |
| 470 | */ |
| 471 | page_unfreeze_refs(page, 1); |
| 472 | return 1; |
| 473 | } |
| 474 | return 0; |
| 475 | } |
| 476 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 477 | /** |
| 478 | * putback_lru_page - put previously isolated page onto appropriate LRU list |
| 479 | * @page: page to be put back to appropriate lru list |
| 480 | * |
| 481 | * Add previously isolated @page to appropriate LRU list. |
| 482 | * Page may still be unevictable for other reasons. |
| 483 | * |
| 484 | * lru_lock must not be held, interrupts must be enabled. |
| 485 | */ |
| 486 | #ifdef CONFIG_UNEVICTABLE_LRU |
| 487 | void putback_lru_page(struct page *page) |
| 488 | { |
| 489 | int lru; |
| 490 | int active = !!TestClearPageActive(page); |
Lee Schermerhorn | bbfd28e | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 491 | int was_unevictable = PageUnevictable(page); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 492 | |
| 493 | VM_BUG_ON(PageLRU(page)); |
| 494 | |
| 495 | redo: |
| 496 | ClearPageUnevictable(page); |
| 497 | |
| 498 | if (page_evictable(page, NULL)) { |
| 499 | /* |
| 500 | * For evictable pages, we can use the cache. |
| 501 | * In event of a race, worst case is we end up with an |
| 502 | * unevictable page on [in]active list. |
| 503 | * We know how to handle that. |
| 504 | */ |
| 505 | lru = active + page_is_file_cache(page); |
| 506 | lru_cache_add_lru(page, lru); |
| 507 | } else { |
| 508 | /* |
| 509 | * Put unevictable pages directly on zone's unevictable |
| 510 | * list. |
| 511 | */ |
| 512 | lru = LRU_UNEVICTABLE; |
| 513 | add_page_to_unevictable_list(page); |
| 514 | } |
| 515 | mem_cgroup_move_lists(page, lru); |
| 516 | |
| 517 | /* |
| 518 | * page's status can change while we move it among lru. If an evictable |
| 519 | * page is on unevictable list, it never be freed. To avoid that, |
| 520 | * check after we added it to the list, again. |
| 521 | */ |
| 522 | if (lru == LRU_UNEVICTABLE && page_evictable(page, NULL)) { |
| 523 | if (!isolate_lru_page(page)) { |
| 524 | put_page(page); |
| 525 | goto redo; |
| 526 | } |
| 527 | /* This means someone else dropped this page from LRU |
| 528 | * So, it will be freed or putback to LRU again. There is |
| 529 | * nothing to do here. |
| 530 | */ |
| 531 | } |
| 532 | |
Lee Schermerhorn | bbfd28e | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 533 | if (was_unevictable && lru != LRU_UNEVICTABLE) |
| 534 | count_vm_event(UNEVICTABLE_PGRESCUED); |
| 535 | else if (!was_unevictable && lru == LRU_UNEVICTABLE) |
| 536 | count_vm_event(UNEVICTABLE_PGCULLED); |
| 537 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 538 | put_page(page); /* drop ref from isolate */ |
| 539 | } |
| 540 | |
| 541 | #else /* CONFIG_UNEVICTABLE_LRU */ |
| 542 | |
| 543 | void putback_lru_page(struct page *page) |
| 544 | { |
| 545 | int lru; |
| 546 | VM_BUG_ON(PageLRU(page)); |
| 547 | |
| 548 | lru = !!TestClearPageActive(page) + page_is_file_cache(page); |
| 549 | lru_cache_add_lru(page, lru); |
| 550 | mem_cgroup_move_lists(page, lru); |
| 551 | put_page(page); |
| 552 | } |
| 553 | #endif /* CONFIG_UNEVICTABLE_LRU */ |
| 554 | |
| 555 | |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 556 | /* |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 557 | * shrink_page_list() returns the number of reclaimed pages |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 558 | */ |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 559 | static unsigned long shrink_page_list(struct list_head *page_list, |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 560 | struct scan_control *sc, |
| 561 | enum pageout_io sync_writeback) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 562 | { |
| 563 | LIST_HEAD(ret_pages); |
| 564 | struct pagevec freed_pvec; |
| 565 | int pgactivate = 0; |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 566 | unsigned long nr_reclaimed = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 567 | |
| 568 | cond_resched(); |
| 569 | |
| 570 | pagevec_init(&freed_pvec, 1); |
| 571 | while (!list_empty(page_list)) { |
| 572 | struct address_space *mapping; |
| 573 | struct page *page; |
| 574 | int may_enter_fs; |
| 575 | int referenced; |
| 576 | |
| 577 | cond_resched(); |
| 578 | |
| 579 | page = lru_to_page(page_list); |
| 580 | list_del(&page->lru); |
| 581 | |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 582 | if (!trylock_page(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 583 | goto keep; |
| 584 | |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 585 | VM_BUG_ON(PageActive(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 586 | |
| 587 | sc->nr_scanned++; |
Christoph Lameter | 80e4342 | 2006-02-11 17:55:53 -0800 | [diff] [blame] | 588 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 589 | if (unlikely(!page_evictable(page, NULL))) |
| 590 | goto cull_mlocked; |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 591 | |
Christoph Lameter | 80e4342 | 2006-02-11 17:55:53 -0800 | [diff] [blame] | 592 | if (!sc->may_swap && page_mapped(page)) |
| 593 | goto keep_locked; |
| 594 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 595 | /* Double the slab pressure for mapped and swapcache pages */ |
| 596 | if (page_mapped(page) || PageSwapCache(page)) |
| 597 | sc->nr_scanned++; |
| 598 | |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 599 | may_enter_fs = (sc->gfp_mask & __GFP_FS) || |
| 600 | (PageSwapCache(page) && (sc->gfp_mask & __GFP_IO)); |
| 601 | |
| 602 | if (PageWriteback(page)) { |
| 603 | /* |
| 604 | * Synchronous reclaim is performed in two passes, |
| 605 | * first an asynchronous pass over the list to |
| 606 | * start parallel writeback, and a second synchronous |
| 607 | * pass to wait for the IO to complete. Wait here |
| 608 | * for any page for which writeback has already |
| 609 | * started. |
| 610 | */ |
| 611 | if (sync_writeback == PAGEOUT_IO_SYNC && may_enter_fs) |
| 612 | wait_on_page_writeback(page); |
Andrew Morton | 4dd4b92 | 2008-03-24 12:29:52 -0700 | [diff] [blame] | 613 | else |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 614 | goto keep_locked; |
| 615 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 616 | |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 617 | referenced = page_referenced(page, 1, sc->mem_cgroup); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 618 | /* In active use or really unfreeable? Activate it. */ |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 619 | if (sc->order <= PAGE_ALLOC_COSTLY_ORDER && |
| 620 | referenced && page_mapping_inuse(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 621 | goto activate_locked; |
| 622 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 623 | /* |
| 624 | * Anonymous process memory has backing store? |
| 625 | * Try to allocate it some swap space here. |
| 626 | */ |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 627 | if (PageAnon(page) && !PageSwapCache(page)) { |
Hugh Dickins | 63eb6b9 | 2008-11-19 15:36:37 -0800 | [diff] [blame] | 628 | if (!(sc->gfp_mask & __GFP_IO)) |
| 629 | goto keep_locked; |
Hugh Dickins | ac47b00 | 2009-01-06 14:39:39 -0800 | [diff] [blame] | 630 | if (!add_to_swap(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 631 | goto activate_locked; |
Hugh Dickins | 63eb6b9 | 2008-11-19 15:36:37 -0800 | [diff] [blame] | 632 | may_enter_fs = 1; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 633 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 634 | |
| 635 | mapping = page_mapping(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 636 | |
| 637 | /* |
| 638 | * The page is mapped into the page tables of one or more |
| 639 | * processes. Try to unmap it here. |
| 640 | */ |
| 641 | if (page_mapped(page) && mapping) { |
Christoph Lameter | a48d07a | 2006-02-01 03:05:38 -0800 | [diff] [blame] | 642 | switch (try_to_unmap(page, 0)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 643 | case SWAP_FAIL: |
| 644 | goto activate_locked; |
| 645 | case SWAP_AGAIN: |
| 646 | goto keep_locked; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 647 | case SWAP_MLOCK: |
| 648 | goto cull_mlocked; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 649 | case SWAP_SUCCESS: |
| 650 | ; /* try to free the page below */ |
| 651 | } |
| 652 | } |
| 653 | |
| 654 | if (PageDirty(page)) { |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 655 | if (sc->order <= PAGE_ALLOC_COSTLY_ORDER && referenced) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 656 | goto keep_locked; |
Andrew Morton | 4dd4b92 | 2008-03-24 12:29:52 -0700 | [diff] [blame] | 657 | if (!may_enter_fs) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 658 | goto keep_locked; |
Christoph Lameter | 52a8363 | 2006-02-01 03:05:28 -0800 | [diff] [blame] | 659 | if (!sc->may_writepage) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 660 | goto keep_locked; |
| 661 | |
| 662 | /* Page is dirty, try to write it out here */ |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 663 | switch (pageout(page, mapping, sync_writeback)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 664 | case PAGE_KEEP: |
| 665 | goto keep_locked; |
| 666 | case PAGE_ACTIVATE: |
| 667 | goto activate_locked; |
| 668 | case PAGE_SUCCESS: |
Andrew Morton | 4dd4b92 | 2008-03-24 12:29:52 -0700 | [diff] [blame] | 669 | if (PageWriteback(page) || PageDirty(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 670 | goto keep; |
| 671 | /* |
| 672 | * A synchronous write - probably a ramdisk. Go |
| 673 | * ahead and try to reclaim the page. |
| 674 | */ |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 675 | if (!trylock_page(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 676 | goto keep; |
| 677 | if (PageDirty(page) || PageWriteback(page)) |
| 678 | goto keep_locked; |
| 679 | mapping = page_mapping(page); |
| 680 | case PAGE_CLEAN: |
| 681 | ; /* try to free the page below */ |
| 682 | } |
| 683 | } |
| 684 | |
| 685 | /* |
| 686 | * If the page has buffers, try to free the buffer mappings |
| 687 | * associated with this page. If we succeed we try to free |
| 688 | * the page as well. |
| 689 | * |
| 690 | * We do this even if the page is PageDirty(). |
| 691 | * try_to_release_page() does not perform I/O, but it is |
| 692 | * possible for a page to have PageDirty set, but it is actually |
| 693 | * clean (all its buffers are clean). This happens if the |
| 694 | * buffers were written out directly, with submit_bh(). ext3 |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 695 | * will do this, as well as the blockdev mapping. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 696 | * try_to_release_page() will discover that cleanness and will |
| 697 | * drop the buffers and mark the page clean - it can be freed. |
| 698 | * |
| 699 | * Rarely, pages can have buffers and no ->mapping. These are |
| 700 | * the pages which were not successfully invalidated in |
| 701 | * truncate_complete_page(). We try to drop those buffers here |
| 702 | * and if that worked, and the page is no longer mapped into |
| 703 | * process address space (page_count == 1) it can be freed. |
| 704 | * Otherwise, leave the page on the LRU so it is swappable. |
| 705 | */ |
| 706 | if (PagePrivate(page)) { |
| 707 | if (!try_to_release_page(page, sc->gfp_mask)) |
| 708 | goto activate_locked; |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 709 | if (!mapping && page_count(page) == 1) { |
| 710 | unlock_page(page); |
| 711 | if (put_page_testzero(page)) |
| 712 | goto free_it; |
| 713 | else { |
| 714 | /* |
| 715 | * rare race with speculative reference. |
| 716 | * the speculative reference will free |
| 717 | * this page shortly, so we may |
| 718 | * increment nr_reclaimed here (and |
| 719 | * leave it off the LRU). |
| 720 | */ |
| 721 | nr_reclaimed++; |
| 722 | continue; |
| 723 | } |
| 724 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 725 | } |
| 726 | |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 727 | if (!mapping || !__remove_mapping(mapping, page)) |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 728 | goto keep_locked; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 729 | |
Nick Piggin | a978d6f | 2008-10-18 20:26:58 -0700 | [diff] [blame] | 730 | /* |
| 731 | * At this point, we have no other references and there is |
| 732 | * no way to pick any more up (removed from LRU, removed |
| 733 | * from pagecache). Can use non-atomic bitops now (and |
| 734 | * we obviously don't have to worry about waking up a process |
| 735 | * waiting on the page lock, because there are no references. |
| 736 | */ |
| 737 | __clear_page_locked(page); |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 738 | free_it: |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 739 | nr_reclaimed++; |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 740 | if (!pagevec_add(&freed_pvec, page)) { |
| 741 | __pagevec_free(&freed_pvec); |
| 742 | pagevec_reinit(&freed_pvec); |
| 743 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 744 | continue; |
| 745 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 746 | cull_mlocked: |
Hugh Dickins | 63d6c5a | 2009-01-06 14:39:38 -0800 | [diff] [blame] | 747 | if (PageSwapCache(page)) |
| 748 | try_to_free_swap(page); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 749 | unlock_page(page); |
| 750 | putback_lru_page(page); |
| 751 | continue; |
| 752 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 753 | activate_locked: |
Rik van Riel | 68a22394 | 2008-10-18 20:26:23 -0700 | [diff] [blame] | 754 | /* Not a candidate for swapping, so reclaim swap space. */ |
| 755 | if (PageSwapCache(page) && vm_swap_full()) |
Hugh Dickins | a2c43ee | 2009-01-06 14:39:36 -0800 | [diff] [blame] | 756 | try_to_free_swap(page); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 757 | VM_BUG_ON(PageActive(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 758 | SetPageActive(page); |
| 759 | pgactivate++; |
| 760 | keep_locked: |
| 761 | unlock_page(page); |
| 762 | keep: |
| 763 | list_add(&page->lru, &ret_pages); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 764 | VM_BUG_ON(PageLRU(page) || PageUnevictable(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 765 | } |
| 766 | list_splice(&ret_pages, page_list); |
| 767 | if (pagevec_count(&freed_pvec)) |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 768 | __pagevec_free(&freed_pvec); |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 769 | count_vm_events(PGACTIVATE, pgactivate); |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 770 | return nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 771 | } |
| 772 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 773 | /* LRU Isolation modes. */ |
| 774 | #define ISOLATE_INACTIVE 0 /* Isolate inactive pages. */ |
| 775 | #define ISOLATE_ACTIVE 1 /* Isolate active pages. */ |
| 776 | #define ISOLATE_BOTH 2 /* Isolate both active and inactive pages. */ |
| 777 | |
| 778 | /* |
| 779 | * Attempt to remove the specified page from its LRU. Only take this page |
| 780 | * if it is of the appropriate PageActive status. Pages which are being |
| 781 | * freed elsewhere are also ignored. |
| 782 | * |
| 783 | * page: page to consider |
| 784 | * mode: one of the LRU isolation modes defined above |
| 785 | * |
| 786 | * returns 0 on success, -ve errno on failure. |
| 787 | */ |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 788 | int __isolate_lru_page(struct page *page, int mode, int file) |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 789 | { |
| 790 | int ret = -EINVAL; |
| 791 | |
| 792 | /* Only take pages on the LRU. */ |
| 793 | if (!PageLRU(page)) |
| 794 | return ret; |
| 795 | |
| 796 | /* |
| 797 | * When checking the active state, we need to be sure we are |
| 798 | * dealing with comparible boolean values. Take the logical not |
| 799 | * of each. |
| 800 | */ |
| 801 | if (mode != ISOLATE_BOTH && (!PageActive(page) != !mode)) |
| 802 | return ret; |
| 803 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 804 | if (mode != ISOLATE_BOTH && (!page_is_file_cache(page) != !file)) |
| 805 | return ret; |
| 806 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 807 | /* |
| 808 | * When this function is being called for lumpy reclaim, we |
| 809 | * initially look into all LRU pages, active, inactive and |
| 810 | * unevictable; only give shrink_page_list evictable pages. |
| 811 | */ |
| 812 | if (PageUnevictable(page)) |
| 813 | return ret; |
| 814 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 815 | ret = -EBUSY; |
| 816 | if (likely(get_page_unless_zero(page))) { |
| 817 | /* |
| 818 | * Be careful not to clear PageLRU until after we're |
| 819 | * sure the page is not being freed elsewhere -- the |
| 820 | * page release code relies on it. |
| 821 | */ |
| 822 | ClearPageLRU(page); |
| 823 | ret = 0; |
| 824 | } |
| 825 | |
| 826 | return ret; |
| 827 | } |
| 828 | |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 829 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 830 | * zone->lru_lock is heavily contended. Some of the functions that |
| 831 | * shrink the lists perform better by taking out a batch of pages |
| 832 | * and working on them outside the LRU lock. |
| 833 | * |
| 834 | * For pagecache intensive workloads, this function is the hottest |
| 835 | * spot in the kernel (apart from copy_*_user functions). |
| 836 | * |
| 837 | * Appropriate locks must be held before calling this function. |
| 838 | * |
| 839 | * @nr_to_scan: The number of pages to look through on the list. |
| 840 | * @src: The LRU list to pull pages off. |
| 841 | * @dst: The temp list to put pages on to. |
| 842 | * @scanned: The number of pages that were scanned. |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 843 | * @order: The caller's attempted allocation order |
| 844 | * @mode: One of the LRU isolation modes |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 845 | * @file: True [1] if isolating file [!anon] pages |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 846 | * |
| 847 | * returns how many pages were moved onto *@dst. |
| 848 | */ |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 849 | static unsigned long isolate_lru_pages(unsigned long nr_to_scan, |
| 850 | struct list_head *src, struct list_head *dst, |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 851 | unsigned long *scanned, int order, int mode, int file) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 852 | { |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 853 | unsigned long nr_taken = 0; |
Wu Fengguang | c9b02d9 | 2006-03-22 00:08:23 -0800 | [diff] [blame] | 854 | unsigned long scan; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 855 | |
Wu Fengguang | c9b02d9 | 2006-03-22 00:08:23 -0800 | [diff] [blame] | 856 | for (scan = 0; scan < nr_to_scan && !list_empty(src); scan++) { |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 857 | struct page *page; |
| 858 | unsigned long pfn; |
| 859 | unsigned long end_pfn; |
| 860 | unsigned long page_pfn; |
| 861 | int zone_id; |
| 862 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 863 | page = lru_to_page(src); |
| 864 | prefetchw_prev_lru_page(page, src, flags); |
| 865 | |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 866 | VM_BUG_ON(!PageLRU(page)); |
Nick Piggin | 8d438f9 | 2006-03-22 00:07:59 -0800 | [diff] [blame] | 867 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 868 | switch (__isolate_lru_page(page, mode, file)) { |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 869 | case 0: |
| 870 | list_move(&page->lru, dst); |
Nick Piggin | 7c8ee9a | 2006-03-22 00:08:03 -0800 | [diff] [blame] | 871 | nr_taken++; |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 872 | break; |
Nick Piggin | 46453a6 | 2006-03-22 00:07:58 -0800 | [diff] [blame] | 873 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 874 | case -EBUSY: |
| 875 | /* else it is being freed elsewhere */ |
| 876 | list_move(&page->lru, src); |
| 877 | continue; |
| 878 | |
| 879 | default: |
| 880 | BUG(); |
| 881 | } |
| 882 | |
| 883 | if (!order) |
| 884 | continue; |
| 885 | |
| 886 | /* |
| 887 | * Attempt to take all pages in the order aligned region |
| 888 | * surrounding the tag page. Only take those pages of |
| 889 | * the same active state as that tag page. We may safely |
| 890 | * round the target page pfn down to the requested order |
| 891 | * as the mem_map is guarenteed valid out to MAX_ORDER, |
| 892 | * where that page is in a different zone we will detect |
| 893 | * it from its zone id and abort this block scan. |
| 894 | */ |
| 895 | zone_id = page_zone_id(page); |
| 896 | page_pfn = page_to_pfn(page); |
| 897 | pfn = page_pfn & ~((1 << order) - 1); |
| 898 | end_pfn = pfn + (1 << order); |
| 899 | for (; pfn < end_pfn; pfn++) { |
| 900 | struct page *cursor_page; |
| 901 | |
| 902 | /* The target page is in the block, ignore it. */ |
| 903 | if (unlikely(pfn == page_pfn)) |
| 904 | continue; |
| 905 | |
| 906 | /* Avoid holes within the zone. */ |
| 907 | if (unlikely(!pfn_valid_within(pfn))) |
| 908 | break; |
| 909 | |
| 910 | cursor_page = pfn_to_page(pfn); |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 911 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 912 | /* Check that we have not crossed a zone boundary. */ |
| 913 | if (unlikely(page_zone_id(cursor_page) != zone_id)) |
| 914 | continue; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 915 | switch (__isolate_lru_page(cursor_page, mode, file)) { |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 916 | case 0: |
| 917 | list_move(&cursor_page->lru, dst); |
| 918 | nr_taken++; |
| 919 | scan++; |
| 920 | break; |
| 921 | |
| 922 | case -EBUSY: |
| 923 | /* else it is being freed elsewhere */ |
| 924 | list_move(&cursor_page->lru, src); |
| 925 | default: |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 926 | break; /* ! on LRU or wrong list */ |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 927 | } |
| 928 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 929 | } |
| 930 | |
| 931 | *scanned = scan; |
| 932 | return nr_taken; |
| 933 | } |
| 934 | |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 935 | static unsigned long isolate_pages_global(unsigned long nr, |
| 936 | struct list_head *dst, |
| 937 | unsigned long *scanned, int order, |
| 938 | int mode, struct zone *z, |
| 939 | struct mem_cgroup *mem_cont, |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 940 | int active, int file) |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 941 | { |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 942 | int lru = LRU_BASE; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 943 | if (active) |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 944 | lru += LRU_ACTIVE; |
| 945 | if (file) |
| 946 | lru += LRU_FILE; |
| 947 | return isolate_lru_pages(nr, &z->lru[lru].list, dst, scanned, order, |
| 948 | mode, !!file); |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 949 | } |
| 950 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 951 | /* |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 952 | * clear_active_flags() is a helper for shrink_active_list(), clearing |
| 953 | * any active bits from the pages in the list. |
| 954 | */ |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 955 | static unsigned long clear_active_flags(struct list_head *page_list, |
| 956 | unsigned int *count) |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 957 | { |
| 958 | int nr_active = 0; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 959 | int lru; |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 960 | struct page *page; |
| 961 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 962 | list_for_each_entry(page, page_list, lru) { |
| 963 | lru = page_is_file_cache(page); |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 964 | if (PageActive(page)) { |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 965 | lru += LRU_ACTIVE; |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 966 | ClearPageActive(page); |
| 967 | nr_active++; |
| 968 | } |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 969 | count[lru]++; |
| 970 | } |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 971 | |
| 972 | return nr_active; |
| 973 | } |
| 974 | |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 975 | /** |
| 976 | * isolate_lru_page - tries to isolate a page from its LRU list |
| 977 | * @page: page to isolate from its LRU list |
| 978 | * |
| 979 | * Isolates a @page from an LRU list, clears PageLRU and adjusts the |
| 980 | * vmstat statistic corresponding to whatever LRU list the page was on. |
| 981 | * |
| 982 | * Returns 0 if the page was removed from an LRU list. |
| 983 | * Returns -EBUSY if the page was not on an LRU list. |
| 984 | * |
| 985 | * The returned page will have PageLRU() cleared. If it was found on |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 986 | * the active list, it will have PageActive set. If it was found on |
| 987 | * the unevictable list, it will have the PageUnevictable bit set. That flag |
| 988 | * 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] | 989 | * |
| 990 | * The vmstat statistic corresponding to the list on which the page was |
| 991 | * found will be decremented. |
| 992 | * |
| 993 | * Restrictions: |
| 994 | * (1) Must be called with an elevated refcount on the page. This is a |
| 995 | * fundamentnal difference from isolate_lru_pages (which is called |
| 996 | * without a stable reference). |
| 997 | * (2) the lru_lock must not be held. |
| 998 | * (3) interrupts must be enabled. |
| 999 | */ |
| 1000 | int isolate_lru_page(struct page *page) |
| 1001 | { |
| 1002 | int ret = -EBUSY; |
| 1003 | |
| 1004 | if (PageLRU(page)) { |
| 1005 | struct zone *zone = page_zone(page); |
| 1006 | |
| 1007 | spin_lock_irq(&zone->lru_lock); |
| 1008 | if (PageLRU(page) && get_page_unless_zero(page)) { |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1009 | int lru = page_lru(page); |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1010 | ret = 0; |
| 1011 | ClearPageLRU(page); |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1012 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1013 | del_page_from_lru_list(zone, page, lru); |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1014 | } |
| 1015 | spin_unlock_irq(&zone->lru_lock); |
| 1016 | } |
| 1017 | return ret; |
| 1018 | } |
| 1019 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1020 | /* |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 1021 | * shrink_inactive_list() is a helper for shrink_zone(). It returns the number |
| 1022 | * of reclaimed pages |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1023 | */ |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 1024 | static unsigned long shrink_inactive_list(unsigned long max_scan, |
Rik van Riel | 33c120e | 2008-10-18 20:26:36 -0700 | [diff] [blame] | 1025 | struct zone *zone, struct scan_control *sc, |
| 1026 | int priority, int file) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1027 | { |
| 1028 | LIST_HEAD(page_list); |
| 1029 | struct pagevec pvec; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1030 | unsigned long nr_scanned = 0; |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 1031 | unsigned long nr_reclaimed = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1032 | |
| 1033 | pagevec_init(&pvec, 1); |
| 1034 | |
| 1035 | lru_add_drain(); |
| 1036 | spin_lock_irq(&zone->lru_lock); |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1037 | do { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1038 | struct page *page; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1039 | unsigned long nr_taken; |
| 1040 | unsigned long nr_scan; |
| 1041 | unsigned long nr_freed; |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1042 | unsigned long nr_active; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1043 | unsigned int count[NR_LRU_LISTS] = { 0, }; |
Rik van Riel | 33c120e | 2008-10-18 20:26:36 -0700 | [diff] [blame] | 1044 | int mode = ISOLATE_INACTIVE; |
| 1045 | |
| 1046 | /* |
| 1047 | * If we need a large contiguous chunk of memory, or have |
| 1048 | * trouble getting a small set of contiguous pages, we |
| 1049 | * will reclaim both active and inactive pages. |
| 1050 | * |
| 1051 | * We use the same threshold as pageout congestion_wait below. |
| 1052 | */ |
| 1053 | if (sc->order > PAGE_ALLOC_COSTLY_ORDER) |
| 1054 | mode = ISOLATE_BOTH; |
| 1055 | else if (sc->order && priority < DEF_PRIORITY - 2) |
| 1056 | mode = ISOLATE_BOTH; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1057 | |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1058 | nr_taken = sc->isolate_pages(sc->swap_cluster_max, |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1059 | &page_list, &nr_scan, sc->order, mode, |
| 1060 | zone, sc->mem_cgroup, 0, file); |
| 1061 | nr_active = clear_active_flags(&page_list, count); |
Andy Whitcroft | e9187bd | 2007-08-22 14:01:25 -0700 | [diff] [blame] | 1062 | __count_vm_events(PGDEACTIVATE, nr_active); |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1063 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1064 | __mod_zone_page_state(zone, NR_ACTIVE_FILE, |
| 1065 | -count[LRU_ACTIVE_FILE]); |
| 1066 | __mod_zone_page_state(zone, NR_INACTIVE_FILE, |
| 1067 | -count[LRU_INACTIVE_FILE]); |
| 1068 | __mod_zone_page_state(zone, NR_ACTIVE_ANON, |
| 1069 | -count[LRU_ACTIVE_ANON]); |
| 1070 | __mod_zone_page_state(zone, NR_INACTIVE_ANON, |
| 1071 | -count[LRU_INACTIVE_ANON]); |
| 1072 | |
| 1073 | if (scan_global_lru(sc)) { |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1074 | zone->pages_scanned += nr_scan; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1075 | zone->recent_scanned[0] += count[LRU_INACTIVE_ANON]; |
| 1076 | zone->recent_scanned[0] += count[LRU_ACTIVE_ANON]; |
| 1077 | zone->recent_scanned[1] += count[LRU_INACTIVE_FILE]; |
| 1078 | zone->recent_scanned[1] += count[LRU_ACTIVE_FILE]; |
| 1079 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1080 | spin_unlock_irq(&zone->lru_lock); |
| 1081 | |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1082 | nr_scanned += nr_scan; |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 1083 | nr_freed = shrink_page_list(&page_list, sc, PAGEOUT_IO_ASYNC); |
| 1084 | |
| 1085 | /* |
| 1086 | * If we are direct reclaiming for contiguous pages and we do |
| 1087 | * not reclaim everything in the list, try again and wait |
| 1088 | * for IO to complete. This will stall high-order allocations |
| 1089 | * but that should be acceptable to the caller |
| 1090 | */ |
| 1091 | if (nr_freed < nr_taken && !current_is_kswapd() && |
| 1092 | sc->order > PAGE_ALLOC_COSTLY_ORDER) { |
| 1093 | congestion_wait(WRITE, HZ/10); |
| 1094 | |
| 1095 | /* |
| 1096 | * The attempt at page out may have made some |
| 1097 | * of the pages active, mark them inactive again. |
| 1098 | */ |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1099 | nr_active = clear_active_flags(&page_list, count); |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 1100 | count_vm_events(PGDEACTIVATE, nr_active); |
| 1101 | |
| 1102 | nr_freed += shrink_page_list(&page_list, sc, |
| 1103 | PAGEOUT_IO_SYNC); |
| 1104 | } |
| 1105 | |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 1106 | nr_reclaimed += nr_freed; |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 1107 | local_irq_disable(); |
| 1108 | if (current_is_kswapd()) { |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 1109 | __count_zone_vm_events(PGSCAN_KSWAPD, zone, nr_scan); |
| 1110 | __count_vm_events(KSWAPD_STEAL, nr_freed); |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1111 | } else if (scan_global_lru(sc)) |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 1112 | __count_zone_vm_events(PGSCAN_DIRECT, zone, nr_scan); |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1113 | |
Shantanu Goel | 918d3f9 | 2006-12-29 16:48:59 -0800 | [diff] [blame] | 1114 | __count_zone_vm_events(PGSTEAL, zone, nr_freed); |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 1115 | |
Wu Fengguang | fb8d14e | 2006-03-22 00:08:28 -0800 | [diff] [blame] | 1116 | if (nr_taken == 0) |
| 1117 | goto done; |
| 1118 | |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 1119 | spin_lock(&zone->lru_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1120 | /* |
| 1121 | * Put back any unfreeable pages. |
| 1122 | */ |
| 1123 | while (!list_empty(&page_list)) { |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1124 | int lru; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1125 | page = lru_to_page(&page_list); |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 1126 | VM_BUG_ON(PageLRU(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1127 | list_del(&page->lru); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1128 | if (unlikely(!page_evictable(page, NULL))) { |
| 1129 | spin_unlock_irq(&zone->lru_lock); |
| 1130 | putback_lru_page(page); |
| 1131 | spin_lock_irq(&zone->lru_lock); |
| 1132 | continue; |
| 1133 | } |
| 1134 | SetPageLRU(page); |
| 1135 | lru = page_lru(page); |
| 1136 | add_page_to_lru_list(zone, page, lru); |
| 1137 | mem_cgroup_move_lists(page, lru); |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1138 | if (PageActive(page) && scan_global_lru(sc)) { |
| 1139 | int file = !!page_is_file_cache(page); |
| 1140 | zone->recent_rotated[file]++; |
| 1141 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1142 | if (!pagevec_add(&pvec, page)) { |
| 1143 | spin_unlock_irq(&zone->lru_lock); |
| 1144 | __pagevec_release(&pvec); |
| 1145 | spin_lock_irq(&zone->lru_lock); |
| 1146 | } |
| 1147 | } |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1148 | } while (nr_scanned < max_scan); |
Wu Fengguang | fb8d14e | 2006-03-22 00:08:28 -0800 | [diff] [blame] | 1149 | spin_unlock(&zone->lru_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1150 | done: |
Wu Fengguang | fb8d14e | 2006-03-22 00:08:28 -0800 | [diff] [blame] | 1151 | local_irq_enable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1152 | pagevec_release(&pvec); |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 1153 | return nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1154 | } |
| 1155 | |
Martin Bligh | 3bb1a85 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 1156 | /* |
| 1157 | * We are about to scan this zone at a certain priority level. If that priority |
| 1158 | * level is smaller (ie: more urgent) than the previous priority, then note |
| 1159 | * that priority level within the zone. This is done so that when the next |
| 1160 | * process comes in to scan this zone, it will immediately start out at this |
| 1161 | * priority level rather than having to build up its own scanning priority. |
| 1162 | * Here, this priority affects only the reclaim-mapped threshold. |
| 1163 | */ |
| 1164 | static inline void note_zone_scanning_priority(struct zone *zone, int priority) |
| 1165 | { |
| 1166 | if (priority < zone->prev_priority) |
| 1167 | zone->prev_priority = priority; |
| 1168 | } |
| 1169 | |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1170 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1171 | * This moves pages from the active list to the inactive list. |
| 1172 | * |
| 1173 | * We move them the other way if the page is referenced by one or more |
| 1174 | * processes, from rmap. |
| 1175 | * |
| 1176 | * If the pages are mostly unmapped, the processing is fast and it is |
| 1177 | * appropriate to hold zone->lru_lock across the whole operation. But if |
| 1178 | * the pages are mapped, the processing is slow (page_referenced()) so we |
| 1179 | * should drop zone->lru_lock around each page. It's impossible to balance |
| 1180 | * this, so instead we remove the pages from the LRU while processing them. |
| 1181 | * It is safe to rely on PG_active against the non-LRU pages in here because |
| 1182 | * nobody will play with that bit on a non-LRU page. |
| 1183 | * |
| 1184 | * The downside is that we have to touch page->_count against each page. |
| 1185 | * But we had to alter page->flags anyway. |
| 1186 | */ |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1187 | |
| 1188 | |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 1189 | 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] | 1190 | struct scan_control *sc, int priority, int file) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1191 | { |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1192 | unsigned long pgmoved; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1193 | int pgdeactivate = 0; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1194 | unsigned long pgscanned; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1195 | LIST_HEAD(l_hold); /* The pages which were snipped off */ |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1196 | LIST_HEAD(l_inactive); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1197 | struct page *page; |
| 1198 | struct pagevec pvec; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1199 | enum lru_list lru; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1200 | |
| 1201 | lru_add_drain(); |
| 1202 | spin_lock_irq(&zone->lru_lock); |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1203 | pgmoved = sc->isolate_pages(nr_pages, &l_hold, &pgscanned, sc->order, |
| 1204 | ISOLATE_ACTIVE, zone, |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1205 | sc->mem_cgroup, 1, file); |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1206 | /* |
| 1207 | * zone->pages_scanned is used for detect zone's oom |
| 1208 | * mem_cgroup remembers nr_scan by itself. |
| 1209 | */ |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1210 | if (scan_global_lru(sc)) { |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1211 | zone->pages_scanned += pgscanned; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1212 | zone->recent_scanned[!!file] += pgmoved; |
| 1213 | } |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1214 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1215 | if (file) |
| 1216 | __mod_zone_page_state(zone, NR_ACTIVE_FILE, -pgmoved); |
| 1217 | else |
| 1218 | __mod_zone_page_state(zone, NR_ACTIVE_ANON, -pgmoved); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1219 | spin_unlock_irq(&zone->lru_lock); |
| 1220 | |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1221 | pgmoved = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1222 | while (!list_empty(&l_hold)) { |
| 1223 | cond_resched(); |
| 1224 | page = lru_to_page(&l_hold); |
| 1225 | list_del(&page->lru); |
Rik van Riel | 7e9cd48 | 2008-10-18 20:26:35 -0700 | [diff] [blame] | 1226 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1227 | if (unlikely(!page_evictable(page, NULL))) { |
| 1228 | putback_lru_page(page); |
| 1229 | continue; |
| 1230 | } |
| 1231 | |
Rik van Riel | 7e9cd48 | 2008-10-18 20:26:35 -0700 | [diff] [blame] | 1232 | /* page_referenced clears PageReferenced */ |
| 1233 | if (page_mapping_inuse(page) && |
| 1234 | page_referenced(page, 0, sc->mem_cgroup)) |
| 1235 | pgmoved++; |
| 1236 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1237 | list_add(&page->lru, &l_inactive); |
| 1238 | } |
| 1239 | |
Andrew Morton | b555749 | 2009-01-06 14:40:13 -0800 | [diff] [blame] | 1240 | /* |
| 1241 | * Move the pages to the [file or anon] inactive list. |
| 1242 | */ |
| 1243 | pagevec_init(&pvec, 1); |
| 1244 | pgmoved = 0; |
| 1245 | lru = LRU_BASE + file * LRU_FILE; |
| 1246 | |
Johannes Weiner | 2a1dc50 | 2008-12-01 03:00:35 +0100 | [diff] [blame] | 1247 | spin_lock_irq(&zone->lru_lock); |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1248 | /* |
Rik van Riel | 7e9cd48 | 2008-10-18 20:26:35 -0700 | [diff] [blame] | 1249 | * Count referenced pages from currently used mappings as |
| 1250 | * rotated, even though they are moved to the inactive list. |
| 1251 | * This helps balance scan pressure between file and anonymous |
| 1252 | * pages in get_scan_ratio. |
| 1253 | */ |
KOSAKI Motohiro | 077cbc5 | 2009-01-06 14:39:42 -0800 | [diff] [blame] | 1254 | if (scan_global_lru(sc)) |
| 1255 | zone->recent_rotated[!!file] += pgmoved; |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1256 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1257 | while (!list_empty(&l_inactive)) { |
| 1258 | page = lru_to_page(&l_inactive); |
| 1259 | prefetchw_prev_lru_page(page, &l_inactive, flags); |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 1260 | VM_BUG_ON(PageLRU(page)); |
Nick Piggin | 8d438f9 | 2006-03-22 00:07:59 -0800 | [diff] [blame] | 1261 | SetPageLRU(page); |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 1262 | VM_BUG_ON(!PageActive(page)); |
Nick Piggin | 4c84cac | 2006-03-22 00:08:00 -0800 | [diff] [blame] | 1263 | ClearPageActive(page); |
| 1264 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1265 | list_move(&page->lru, &zone->lru[lru].list); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1266 | mem_cgroup_move_lists(page, lru); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1267 | pgmoved++; |
| 1268 | if (!pagevec_add(&pvec, page)) { |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1269 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1270 | spin_unlock_irq(&zone->lru_lock); |
| 1271 | pgdeactivate += pgmoved; |
| 1272 | pgmoved = 0; |
| 1273 | if (buffer_heads_over_limit) |
| 1274 | pagevec_strip(&pvec); |
| 1275 | __pagevec_release(&pvec); |
| 1276 | spin_lock_irq(&zone->lru_lock); |
| 1277 | } |
| 1278 | } |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1279 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1280 | pgdeactivate += pgmoved; |
| 1281 | if (buffer_heads_over_limit) { |
| 1282 | spin_unlock_irq(&zone->lru_lock); |
| 1283 | pagevec_strip(&pvec); |
| 1284 | spin_lock_irq(&zone->lru_lock); |
| 1285 | } |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 1286 | __count_zone_vm_events(PGREFILL, zone, pgscanned); |
| 1287 | __count_vm_events(PGDEACTIVATE, pgdeactivate); |
| 1288 | spin_unlock_irq(&zone->lru_lock); |
Rik van Riel | 68a22394 | 2008-10-18 20:26:23 -0700 | [diff] [blame] | 1289 | if (vm_swap_full()) |
| 1290 | pagevec_swap_free(&pvec); |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 1291 | |
| 1292 | pagevec_release(&pvec); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1293 | } |
| 1294 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1295 | 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] | 1296 | struct zone *zone, struct scan_control *sc, int priority) |
| 1297 | { |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1298 | int file = is_file_lru(lru); |
| 1299 | |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1300 | if (lru == LRU_ACTIVE_FILE) { |
| 1301 | shrink_active_list(nr_to_scan, zone, sc, priority, file); |
| 1302 | return 0; |
| 1303 | } |
| 1304 | |
| 1305 | if (lru == LRU_ACTIVE_ANON && |
| 1306 | (!scan_global_lru(sc) || inactive_anon_is_low(zone))) { |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1307 | shrink_active_list(nr_to_scan, zone, sc, priority, file); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1308 | return 0; |
| 1309 | } |
Rik van Riel | 33c120e | 2008-10-18 20:26:36 -0700 | [diff] [blame] | 1310 | return shrink_inactive_list(nr_to_scan, zone, sc, priority, file); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1311 | } |
| 1312 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1313 | /* |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1314 | * Determine how aggressively the anon and file LRU lists should be |
| 1315 | * scanned. The relative value of each set of LRU lists is determined |
| 1316 | * by looking at the fraction of the pages scanned we did rotate back |
| 1317 | * onto the active list instead of evict. |
| 1318 | * |
| 1319 | * percent[0] specifies how much pressure to put on ram/swap backed |
| 1320 | * memory, while percent[1] determines pressure on the file LRUs. |
| 1321 | */ |
| 1322 | static void get_scan_ratio(struct zone *zone, struct scan_control *sc, |
| 1323 | unsigned long *percent) |
| 1324 | { |
| 1325 | unsigned long anon, file, free; |
| 1326 | unsigned long anon_prio, file_prio; |
| 1327 | unsigned long ap, fp; |
| 1328 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1329 | /* If we have no swap space, do not bother scanning anon pages. */ |
| 1330 | if (nr_swap_pages <= 0) { |
| 1331 | percent[0] = 0; |
| 1332 | percent[1] = 100; |
| 1333 | return; |
| 1334 | } |
| 1335 | |
Hugh Dickins | b962716 | 2009-01-06 14:39:41 -0800 | [diff] [blame] | 1336 | anon = zone_page_state(zone, NR_ACTIVE_ANON) + |
| 1337 | zone_page_state(zone, NR_INACTIVE_ANON); |
| 1338 | file = zone_page_state(zone, NR_ACTIVE_FILE) + |
| 1339 | zone_page_state(zone, NR_INACTIVE_FILE); |
| 1340 | free = zone_page_state(zone, NR_FREE_PAGES); |
| 1341 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1342 | /* If we have very few page cache pages, force-scan anon pages. */ |
| 1343 | if (unlikely(file + free <= zone->pages_high)) { |
| 1344 | percent[0] = 100; |
| 1345 | percent[1] = 0; |
| 1346 | return; |
| 1347 | } |
| 1348 | |
| 1349 | /* |
| 1350 | * OK, so we have swap space and a fair amount of page cache |
| 1351 | * pages. We use the recently rotated / recently scanned |
| 1352 | * ratios to determine how valuable each cache is. |
| 1353 | * |
| 1354 | * Because workloads change over time (and to avoid overflow) |
| 1355 | * we keep these statistics as a floating average, which ends |
| 1356 | * up weighing recent references more than old ones. |
| 1357 | * |
| 1358 | * anon in [0], file in [1] |
| 1359 | */ |
| 1360 | if (unlikely(zone->recent_scanned[0] > anon / 4)) { |
| 1361 | spin_lock_irq(&zone->lru_lock); |
| 1362 | zone->recent_scanned[0] /= 2; |
| 1363 | zone->recent_rotated[0] /= 2; |
| 1364 | spin_unlock_irq(&zone->lru_lock); |
| 1365 | } |
| 1366 | |
| 1367 | if (unlikely(zone->recent_scanned[1] > file / 4)) { |
| 1368 | spin_lock_irq(&zone->lru_lock); |
| 1369 | zone->recent_scanned[1] /= 2; |
| 1370 | zone->recent_rotated[1] /= 2; |
| 1371 | spin_unlock_irq(&zone->lru_lock); |
| 1372 | } |
| 1373 | |
| 1374 | /* |
| 1375 | * With swappiness at 100, anonymous and file have the same priority. |
| 1376 | * This scanning priority is essentially the inverse of IO cost. |
| 1377 | */ |
| 1378 | anon_prio = sc->swappiness; |
| 1379 | file_prio = 200 - sc->swappiness; |
| 1380 | |
| 1381 | /* |
Rik van Riel | 00d8089 | 2008-11-19 15:36:44 -0800 | [diff] [blame] | 1382 | * The amount of pressure on anon vs file pages is inversely |
| 1383 | * proportional to the fraction of recently scanned pages on |
| 1384 | * each list that were recently referenced and in active use. |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1385 | */ |
| 1386 | ap = (anon_prio + 1) * (zone->recent_scanned[0] + 1); |
| 1387 | ap /= zone->recent_rotated[0] + 1; |
| 1388 | |
| 1389 | fp = (file_prio + 1) * (zone->recent_scanned[1] + 1); |
| 1390 | fp /= zone->recent_rotated[1] + 1; |
| 1391 | |
| 1392 | /* Normalize to percentages */ |
| 1393 | percent[0] = 100 * ap / (ap + fp + 1); |
| 1394 | percent[1] = 100 - percent[0]; |
| 1395 | } |
| 1396 | |
| 1397 | |
| 1398 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1399 | * This is a basic per-zone page freer. Used by both kswapd and direct reclaim. |
| 1400 | */ |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1401 | static void shrink_zone(int priority, struct zone *zone, |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 1402 | struct scan_control *sc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1403 | { |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1404 | unsigned long nr[NR_LRU_LISTS]; |
Christoph Lameter | 8695949 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 1405 | unsigned long nr_to_scan; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1406 | unsigned long percent[2]; /* anon @ 0; file @ 1 */ |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1407 | enum lru_list l; |
KOSAKI Motohiro | 01dbe5c | 2009-01-06 14:40:02 -0800 | [diff] [blame] | 1408 | unsigned long nr_reclaimed = sc->nr_reclaimed; |
| 1409 | unsigned long swap_cluster_max = sc->swap_cluster_max; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1410 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1411 | get_scan_ratio(zone, sc, percent); |
| 1412 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1413 | for_each_evictable_lru(l) { |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1414 | if (scan_global_lru(sc)) { |
| 1415 | int file = is_file_lru(l); |
| 1416 | int scan; |
Johannes Weiner | e0f79b8 | 2008-10-18 20:26:55 -0700 | [diff] [blame] | 1417 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1418 | scan = zone_page_state(zone, NR_LRU_BASE + l); |
| 1419 | if (priority) { |
| 1420 | scan >>= priority; |
| 1421 | scan = (scan * percent[file]) / 100; |
| 1422 | } |
Johannes Weiner | e0f79b8 | 2008-10-18 20:26:55 -0700 | [diff] [blame] | 1423 | zone->lru[l].nr_scan += scan; |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1424 | nr[l] = zone->lru[l].nr_scan; |
KOSAKI Motohiro | 01dbe5c | 2009-01-06 14:40:02 -0800 | [diff] [blame] | 1425 | if (nr[l] >= swap_cluster_max) |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1426 | zone->lru[l].nr_scan = 0; |
| 1427 | else |
| 1428 | nr[l] = 0; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1429 | } else { |
| 1430 | /* |
| 1431 | * This reclaim occurs not because zone memory shortage |
| 1432 | * but because memory controller hits its limit. |
| 1433 | * Don't modify zone reclaim related data. |
| 1434 | */ |
| 1435 | nr[l] = mem_cgroup_calc_reclaim(sc->mem_cgroup, zone, |
| 1436 | priority, l); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1437 | } |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1438 | } |
| 1439 | |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1440 | while (nr[LRU_INACTIVE_ANON] || nr[LRU_ACTIVE_FILE] || |
| 1441 | nr[LRU_INACTIVE_FILE]) { |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1442 | for_each_evictable_lru(l) { |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1443 | if (nr[l]) { |
KOSAKI Motohiro | 01dbe5c | 2009-01-06 14:40:02 -0800 | [diff] [blame] | 1444 | nr_to_scan = min(nr[l], swap_cluster_max); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1445 | nr[l] -= nr_to_scan; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1446 | |
KOSAKI Motohiro | 01dbe5c | 2009-01-06 14:40:02 -0800 | [diff] [blame] | 1447 | nr_reclaimed += shrink_list(l, nr_to_scan, |
| 1448 | zone, sc, priority); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1449 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1450 | } |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1451 | /* |
| 1452 | * On large memory systems, scan >> priority can become |
| 1453 | * really large. This is fine for the starting priority; |
| 1454 | * we want to put equal scanning pressure on each zone. |
| 1455 | * However, if the VM has a harder time of freeing pages, |
| 1456 | * with multiple processes reclaiming pages, the total |
| 1457 | * freeing target can get unreasonably large. |
| 1458 | */ |
KOSAKI Motohiro | 01dbe5c | 2009-01-06 14:40:02 -0800 | [diff] [blame] | 1459 | if (nr_reclaimed > swap_cluster_max && |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1460 | priority < DEF_PRIORITY && !current_is_kswapd()) |
| 1461 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1462 | } |
| 1463 | |
KOSAKI Motohiro | 01dbe5c | 2009-01-06 14:40:02 -0800 | [diff] [blame] | 1464 | sc->nr_reclaimed = nr_reclaimed; |
| 1465 | |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1466 | /* |
| 1467 | * Even if we did not try to evict anon pages at all, we want to |
| 1468 | * rebalance the anon lru active/inactive ratio. |
| 1469 | */ |
| 1470 | if (!scan_global_lru(sc) || inactive_anon_is_low(zone)) |
| 1471 | shrink_active_list(SWAP_CLUSTER_MAX, zone, sc, priority, 0); |
| 1472 | else if (!scan_global_lru(sc)) |
| 1473 | shrink_active_list(SWAP_CLUSTER_MAX, zone, sc, priority, 0); |
| 1474 | |
Andrew Morton | 232ea4d | 2007-02-28 20:13:21 -0800 | [diff] [blame] | 1475 | throttle_vm_writeout(sc->gfp_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1476 | } |
| 1477 | |
| 1478 | /* |
| 1479 | * This is the direct reclaim path, for page-allocating processes. We only |
| 1480 | * try to reclaim pages from zones which will satisfy the caller's allocation |
| 1481 | * request. |
| 1482 | * |
| 1483 | * We reclaim from a zone even if that zone is over pages_high. Because: |
| 1484 | * a) The caller may be trying to free *extra* pages to satisfy a higher-order |
| 1485 | * allocation or |
| 1486 | * b) The zones may be over pages_high but they must go *over* pages_high to |
| 1487 | * satisfy the `incremental min' zone defense algorithm. |
| 1488 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1489 | * If a zone is deemed to be full of pinned pages then just give it a light |
| 1490 | * scan then give up on it. |
| 1491 | */ |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1492 | static void shrink_zones(int priority, struct zonelist *zonelist, |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 1493 | struct scan_control *sc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1494 | { |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 1495 | enum zone_type high_zoneidx = gfp_zone(sc->gfp_mask); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1496 | struct zoneref *z; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 1497 | struct zone *zone; |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1498 | |
Nick Piggin | 408d854 | 2006-09-25 23:31:27 -0700 | [diff] [blame] | 1499 | sc->all_unreclaimable = 1; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 1500 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
Con Kolivas | f3fe651 | 2006-01-06 00:11:15 -0800 | [diff] [blame] | 1501 | if (!populated_zone(zone)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1502 | continue; |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1503 | /* |
| 1504 | * Take care memory controller reclaiming has small influence |
| 1505 | * to global LRU. |
| 1506 | */ |
| 1507 | if (scan_global_lru(sc)) { |
| 1508 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
| 1509 | continue; |
| 1510 | note_zone_scanning_priority(zone, priority); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1511 | |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1512 | if (zone_is_all_unreclaimable(zone) && |
| 1513 | priority != DEF_PRIORITY) |
| 1514 | continue; /* Let kswapd poll it */ |
| 1515 | sc->all_unreclaimable = 0; |
| 1516 | } else { |
| 1517 | /* |
| 1518 | * Ignore cpuset limitation here. We just want to reduce |
| 1519 | * # of used pages by us regardless of memory shortage. |
| 1520 | */ |
| 1521 | sc->all_unreclaimable = 0; |
| 1522 | mem_cgroup_note_reclaim_priority(sc->mem_cgroup, |
| 1523 | priority); |
| 1524 | } |
Nick Piggin | 408d854 | 2006-09-25 23:31:27 -0700 | [diff] [blame] | 1525 | |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1526 | shrink_zone(priority, zone, sc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1527 | } |
| 1528 | } |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1529 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1530 | /* |
| 1531 | * This is the main entry point to direct page reclaim. |
| 1532 | * |
| 1533 | * If a full scan of the inactive list fails to free enough memory then we |
| 1534 | * are "out of memory" and something needs to be killed. |
| 1535 | * |
| 1536 | * If the caller is !__GFP_FS then the probability of a failure is reasonably |
| 1537 | * high - the zone may be full of dirty or under-writeback pages, which this |
| 1538 | * caller can't do much about. We kick pdflush and take explicit naps in the |
| 1539 | * hope that some of these pages can be written. But if the allocating task |
| 1540 | * holds filesystem locks which prevent writeout this might not work, and the |
| 1541 | * allocation attempt will fail. |
Nishanth Aravamudan | a41f24e | 2008-04-29 00:58:25 -0700 | [diff] [blame] | 1542 | * |
| 1543 | * returns: 0, if no pages reclaimed |
| 1544 | * else, the number of pages reclaimed |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1545 | */ |
Mel Gorman | dac1d27 | 2008-04-28 02:12:12 -0700 | [diff] [blame] | 1546 | static unsigned long do_try_to_free_pages(struct zonelist *zonelist, |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1547 | struct scan_control *sc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1548 | { |
| 1549 | int priority; |
kosaki.motohiro@jp.fujitsu.com | c700be3 | 2008-06-12 15:21:27 -0700 | [diff] [blame] | 1550 | unsigned long ret = 0; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1551 | unsigned long total_scanned = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1552 | struct reclaim_state *reclaim_state = current->reclaim_state; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1553 | unsigned long lru_pages = 0; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1554 | struct zoneref *z; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 1555 | struct zone *zone; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1556 | enum zone_type high_zoneidx = gfp_zone(sc->gfp_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1557 | |
Keika Kobayashi | 873b477 | 2008-07-25 01:48:52 -0700 | [diff] [blame] | 1558 | delayacct_freepages_start(); |
| 1559 | |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1560 | if (scan_global_lru(sc)) |
| 1561 | count_vm_event(ALLOCSTALL); |
| 1562 | /* |
| 1563 | * mem_cgroup will not do shrink_slab. |
| 1564 | */ |
| 1565 | if (scan_global_lru(sc)) { |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 1566 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1567 | |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1568 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
| 1569 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1570 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1571 | lru_pages += zone_lru_pages(zone); |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1572 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1573 | } |
| 1574 | |
| 1575 | for (priority = DEF_PRIORITY; priority >= 0; priority--) { |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1576 | sc->nr_scanned = 0; |
Rik van Riel | f7b7fd8 | 2005-11-28 13:44:07 -0800 | [diff] [blame] | 1577 | if (!priority) |
| 1578 | disable_swap_token(); |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1579 | shrink_zones(priority, zonelist, sc); |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1580 | /* |
| 1581 | * Don't shrink slabs when reclaiming memory from |
| 1582 | * over limit cgroups |
| 1583 | */ |
KAMEZAWA Hiroyuki | 91a4547 | 2008-02-07 00:14:29 -0800 | [diff] [blame] | 1584 | if (scan_global_lru(sc)) { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1585 | shrink_slab(sc->nr_scanned, sc->gfp_mask, lru_pages); |
KAMEZAWA Hiroyuki | 91a4547 | 2008-02-07 00:14:29 -0800 | [diff] [blame] | 1586 | if (reclaim_state) { |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1587 | sc->nr_reclaimed += reclaim_state->reclaimed_slab; |
KAMEZAWA Hiroyuki | 91a4547 | 2008-02-07 00:14:29 -0800 | [diff] [blame] | 1588 | reclaim_state->reclaimed_slab = 0; |
| 1589 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1590 | } |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1591 | total_scanned += sc->nr_scanned; |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1592 | if (sc->nr_reclaimed >= sc->swap_cluster_max) { |
| 1593 | ret = sc->nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1594 | goto out; |
| 1595 | } |
| 1596 | |
| 1597 | /* |
| 1598 | * Try to write back as many pages as we just scanned. This |
| 1599 | * tends to cause slow streaming writers to write data to the |
| 1600 | * disk smoothly, at the dirtying rate, which is nice. But |
| 1601 | * that's undesirable in laptop mode, where we *want* lumpy |
| 1602 | * writeout. So in laptop mode, write out the whole world. |
| 1603 | */ |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1604 | if (total_scanned > sc->swap_cluster_max + |
| 1605 | sc->swap_cluster_max / 2) { |
Pekka J Enberg | 687a21c | 2005-06-28 20:44:55 -0700 | [diff] [blame] | 1606 | wakeup_pdflush(laptop_mode ? 0 : total_scanned); |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1607 | sc->may_writepage = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1608 | } |
| 1609 | |
| 1610 | /* Take a nap, wait for some writeback to complete */ |
Andrew Morton | 4dd4b92 | 2008-03-24 12:29:52 -0700 | [diff] [blame] | 1611 | if (sc->nr_scanned && priority < DEF_PRIORITY - 2) |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 1612 | congestion_wait(WRITE, HZ/10); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1613 | } |
Fernando Luis Vazquez Cao | 87547ee | 2008-07-29 22:33:42 -0700 | [diff] [blame] | 1614 | /* 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] | 1615 | if (!sc->all_unreclaimable && scan_global_lru(sc)) |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1616 | ret = sc->nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1617 | out: |
Martin Bligh | 3bb1a85 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 1618 | /* |
| 1619 | * Now that we've scanned all the zones at this priority level, note |
| 1620 | * that level within the zone so that the next thread which performs |
| 1621 | * scanning of this zone will immediately start out at this priority |
| 1622 | * level. This affects only the decision whether or not to bring |
| 1623 | * mapped pages onto the inactive list. |
| 1624 | */ |
| 1625 | if (priority < 0) |
| 1626 | priority = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1627 | |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1628 | if (scan_global_lru(sc)) { |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 1629 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1630 | |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1631 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
| 1632 | continue; |
| 1633 | |
| 1634 | zone->prev_priority = priority; |
| 1635 | } |
| 1636 | } else |
| 1637 | mem_cgroup_record_reclaim_priority(sc->mem_cgroup, priority); |
| 1638 | |
Keika Kobayashi | 873b477 | 2008-07-25 01:48:52 -0700 | [diff] [blame] | 1639 | delayacct_freepages_end(); |
| 1640 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1641 | return ret; |
| 1642 | } |
| 1643 | |
Mel Gorman | dac1d27 | 2008-04-28 02:12:12 -0700 | [diff] [blame] | 1644 | unsigned long try_to_free_pages(struct zonelist *zonelist, int order, |
| 1645 | gfp_t gfp_mask) |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1646 | { |
| 1647 | struct scan_control sc = { |
| 1648 | .gfp_mask = gfp_mask, |
| 1649 | .may_writepage = !laptop_mode, |
| 1650 | .swap_cluster_max = SWAP_CLUSTER_MAX, |
| 1651 | .may_swap = 1, |
| 1652 | .swappiness = vm_swappiness, |
| 1653 | .order = order, |
| 1654 | .mem_cgroup = NULL, |
| 1655 | .isolate_pages = isolate_pages_global, |
| 1656 | }; |
| 1657 | |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1658 | return do_try_to_free_pages(zonelist, &sc); |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1659 | } |
| 1660 | |
Balbir Singh | 00f0b82 | 2008-03-04 14:28:39 -0800 | [diff] [blame] | 1661 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1662 | |
Balbir Singh | e1a1cd5 | 2008-02-07 00:14:02 -0800 | [diff] [blame] | 1663 | 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^] | 1664 | gfp_t gfp_mask, |
| 1665 | bool noswap) |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1666 | { |
| 1667 | struct scan_control sc = { |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1668 | .may_writepage = !laptop_mode, |
| 1669 | .may_swap = 1, |
| 1670 | .swap_cluster_max = SWAP_CLUSTER_MAX, |
| 1671 | .swappiness = vm_swappiness, |
| 1672 | .order = 0, |
| 1673 | .mem_cgroup = mem_cont, |
| 1674 | .isolate_pages = mem_cgroup_isolate_pages, |
| 1675 | }; |
Mel Gorman | dac1d27 | 2008-04-28 02:12:12 -0700 | [diff] [blame] | 1676 | struct zonelist *zonelist; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1677 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame^] | 1678 | if (noswap) |
| 1679 | sc.may_swap = 0; |
| 1680 | |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1681 | sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | |
| 1682 | (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK); |
| 1683 | zonelist = NODE_DATA(numa_node_id())->node_zonelists; |
| 1684 | return do_try_to_free_pages(zonelist, &sc); |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1685 | } |
| 1686 | #endif |
| 1687 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1688 | /* |
| 1689 | * For kswapd, balance_pgdat() will work across all this node's zones until |
| 1690 | * they are all at pages_high. |
| 1691 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1692 | * Returns the number of pages which were actually freed. |
| 1693 | * |
| 1694 | * There is special handling here for zones which are full of pinned pages. |
| 1695 | * This can happen if the pages are all mlocked, or if they are all used by |
| 1696 | * device drivers (say, ZONE_DMA). Or if they are all in use by hugetlb. |
| 1697 | * What we do is to detect the case where all pages in the zone have been |
| 1698 | * scanned twice and there has been zero successful reclaim. Mark the zone as |
| 1699 | * dead and from now on, only perform a short scan. Basically we're polling |
| 1700 | * the zone for when the problem goes away. |
| 1701 | * |
| 1702 | * kswapd scans the zones in the highmem->normal->dma direction. It skips |
| 1703 | * zones which have free_pages > pages_high, but once a zone is found to have |
| 1704 | * free_pages <= pages_high, we scan that zone and the lower zones regardless |
| 1705 | * of the number of free pages in the lower zones. This interoperates with |
| 1706 | * the page allocator fallback scheme to ensure that aging of pages is balanced |
| 1707 | * across the zones. |
| 1708 | */ |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 1709 | static unsigned long balance_pgdat(pg_data_t *pgdat, int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1710 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1711 | int all_zones_ok; |
| 1712 | int priority; |
| 1713 | int i; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1714 | unsigned long total_scanned; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1715 | struct reclaim_state *reclaim_state = current->reclaim_state; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 1716 | struct scan_control sc = { |
| 1717 | .gfp_mask = GFP_KERNEL, |
| 1718 | .may_swap = 1, |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 1719 | .swap_cluster_max = SWAP_CLUSTER_MAX, |
| 1720 | .swappiness = vm_swappiness, |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1721 | .order = order, |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1722 | .mem_cgroup = NULL, |
| 1723 | .isolate_pages = isolate_pages_global, |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 1724 | }; |
Martin Bligh | 3bb1a85 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 1725 | /* |
| 1726 | * temp_priority is used to remember the scanning priority at which |
| 1727 | * this zone was successfully refilled to free_pages == pages_high. |
| 1728 | */ |
| 1729 | int temp_priority[MAX_NR_ZONES]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1730 | |
| 1731 | loop_again: |
| 1732 | total_scanned = 0; |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1733 | sc.nr_reclaimed = 0; |
Christoph Lameter | c0bbbc7 | 2006-06-11 15:22:26 -0700 | [diff] [blame] | 1734 | sc.may_writepage = !laptop_mode; |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 1735 | count_vm_event(PAGEOUTRUN); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1736 | |
Martin Bligh | 3bb1a85 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 1737 | for (i = 0; i < pgdat->nr_zones; i++) |
| 1738 | temp_priority[i] = DEF_PRIORITY; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1739 | |
| 1740 | for (priority = DEF_PRIORITY; priority >= 0; priority--) { |
| 1741 | int end_zone = 0; /* Inclusive. 0 = ZONE_DMA */ |
| 1742 | unsigned long lru_pages = 0; |
| 1743 | |
Rik van Riel | f7b7fd8 | 2005-11-28 13:44:07 -0800 | [diff] [blame] | 1744 | /* The swap token gets in the way of swapout... */ |
| 1745 | if (!priority) |
| 1746 | disable_swap_token(); |
| 1747 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1748 | all_zones_ok = 1; |
| 1749 | |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 1750 | /* |
| 1751 | * Scan in the highmem->dma direction for the highest |
| 1752 | * zone which needs scanning |
| 1753 | */ |
| 1754 | for (i = pgdat->nr_zones - 1; i >= 0; i--) { |
| 1755 | struct zone *zone = pgdat->node_zones + i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1756 | |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 1757 | if (!populated_zone(zone)) |
| 1758 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1759 | |
David Rientjes | e815af9 | 2007-10-16 23:25:54 -0700 | [diff] [blame] | 1760 | if (zone_is_all_unreclaimable(zone) && |
| 1761 | priority != DEF_PRIORITY) |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 1762 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1763 | |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1764 | /* |
| 1765 | * Do some background aging of the anon list, to give |
| 1766 | * pages a chance to be referenced before reclaiming. |
| 1767 | */ |
| 1768 | if (inactive_anon_is_low(zone)) |
| 1769 | shrink_active_list(SWAP_CLUSTER_MAX, zone, |
| 1770 | &sc, priority, 0); |
| 1771 | |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 1772 | if (!zone_watermark_ok(zone, order, zone->pages_high, |
| 1773 | 0, 0)) { |
| 1774 | end_zone = i; |
Andrew Morton | e1dbeda | 2006-12-06 20:32:01 -0800 | [diff] [blame] | 1775 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1776 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1777 | } |
Andrew Morton | e1dbeda | 2006-12-06 20:32:01 -0800 | [diff] [blame] | 1778 | if (i < 0) |
| 1779 | goto out; |
| 1780 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1781 | for (i = 0; i <= end_zone; i++) { |
| 1782 | struct zone *zone = pgdat->node_zones + i; |
| 1783 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1784 | lru_pages += zone_lru_pages(zone); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1785 | } |
| 1786 | |
| 1787 | /* |
| 1788 | * Now scan the zone in the dma->highmem direction, stopping |
| 1789 | * at the last zone which needs scanning. |
| 1790 | * |
| 1791 | * We do this because the page allocator works in the opposite |
| 1792 | * direction. This prevents the page allocator from allocating |
| 1793 | * pages behind kswapd's direction of progress, which would |
| 1794 | * cause too much scanning of the lower zones. |
| 1795 | */ |
| 1796 | for (i = 0; i <= end_zone; i++) { |
| 1797 | struct zone *zone = pgdat->node_zones + i; |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 1798 | int nr_slab; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1799 | |
Con Kolivas | f3fe651 | 2006-01-06 00:11:15 -0800 | [diff] [blame] | 1800 | if (!populated_zone(zone)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1801 | continue; |
| 1802 | |
David Rientjes | e815af9 | 2007-10-16 23:25:54 -0700 | [diff] [blame] | 1803 | if (zone_is_all_unreclaimable(zone) && |
| 1804 | priority != DEF_PRIORITY) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1805 | continue; |
| 1806 | |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 1807 | if (!zone_watermark_ok(zone, order, zone->pages_high, |
| 1808 | end_zone, 0)) |
| 1809 | all_zones_ok = 0; |
Martin Bligh | 3bb1a85 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 1810 | temp_priority[i] = priority; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1811 | sc.nr_scanned = 0; |
Martin Bligh | 3bb1a85 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 1812 | note_zone_scanning_priority(zone, priority); |
Rik van Riel | 32a4330 | 2007-10-16 01:24:50 -0700 | [diff] [blame] | 1813 | /* |
| 1814 | * We put equal pressure on every zone, unless one |
| 1815 | * zone has way too many pages free already. |
| 1816 | */ |
| 1817 | if (!zone_watermark_ok(zone, order, 8*zone->pages_high, |
| 1818 | end_zone, 0)) |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1819 | shrink_zone(priority, zone, &sc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1820 | reclaim_state->reclaimed_slab = 0; |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 1821 | nr_slab = shrink_slab(sc.nr_scanned, GFP_KERNEL, |
| 1822 | lru_pages); |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1823 | sc.nr_reclaimed += reclaim_state->reclaimed_slab; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1824 | total_scanned += sc.nr_scanned; |
David Rientjes | e815af9 | 2007-10-16 23:25:54 -0700 | [diff] [blame] | 1825 | if (zone_is_all_unreclaimable(zone)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1826 | continue; |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 1827 | if (nr_slab == 0 && zone->pages_scanned >= |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1828 | (zone_lru_pages(zone) * 6)) |
David Rientjes | e815af9 | 2007-10-16 23:25:54 -0700 | [diff] [blame] | 1829 | zone_set_flag(zone, |
| 1830 | ZONE_ALL_UNRECLAIMABLE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1831 | /* |
| 1832 | * If we've done a decent amount of scanning and |
| 1833 | * the reclaim ratio is low, start doing writepage |
| 1834 | * even in laptop mode |
| 1835 | */ |
| 1836 | if (total_scanned > SWAP_CLUSTER_MAX * 2 && |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1837 | total_scanned > sc.nr_reclaimed + sc.nr_reclaimed / 2) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1838 | sc.may_writepage = 1; |
| 1839 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1840 | if (all_zones_ok) |
| 1841 | break; /* kswapd: all done */ |
| 1842 | /* |
| 1843 | * OK, kswapd is getting into trouble. Take a nap, then take |
| 1844 | * another pass across the zones. |
| 1845 | */ |
Andrew Morton | 4dd4b92 | 2008-03-24 12:29:52 -0700 | [diff] [blame] | 1846 | if (total_scanned && priority < DEF_PRIORITY - 2) |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 1847 | congestion_wait(WRITE, HZ/10); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1848 | |
| 1849 | /* |
| 1850 | * We do this so kswapd doesn't build up large priorities for |
| 1851 | * example when it is freeing in parallel with allocators. It |
| 1852 | * matches the direct reclaim path behaviour in terms of impact |
| 1853 | * on zone->*_priority. |
| 1854 | */ |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1855 | if (sc.nr_reclaimed >= SWAP_CLUSTER_MAX) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1856 | break; |
| 1857 | } |
| 1858 | out: |
Martin Bligh | 3bb1a85 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 1859 | /* |
| 1860 | * Note within each zone the priority level at which this zone was |
| 1861 | * brought into a happy state. So that the next thread which scans this |
| 1862 | * zone will start out at that priority level. |
| 1863 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1864 | for (i = 0; i < pgdat->nr_zones; i++) { |
| 1865 | struct zone *zone = pgdat->node_zones + i; |
| 1866 | |
Martin Bligh | 3bb1a85 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 1867 | zone->prev_priority = temp_priority[i]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1868 | } |
| 1869 | if (!all_zones_ok) { |
| 1870 | cond_resched(); |
Rafael J. Wysocki | 8357376 | 2006-12-06 20:34:18 -0800 | [diff] [blame] | 1871 | |
| 1872 | try_to_freeze(); |
| 1873 | |
KOSAKI Motohiro | 73ce02e | 2009-01-06 14:40:33 -0800 | [diff] [blame] | 1874 | /* |
| 1875 | * Fragmentation may mean that the system cannot be |
| 1876 | * rebalanced for high-order allocations in all zones. |
| 1877 | * At this point, if nr_reclaimed < SWAP_CLUSTER_MAX, |
| 1878 | * it means the zones have been fully scanned and are still |
| 1879 | * not balanced. For high-order allocations, there is |
| 1880 | * little point trying all over again as kswapd may |
| 1881 | * infinite loop. |
| 1882 | * |
| 1883 | * Instead, recheck all watermarks at order-0 as they |
| 1884 | * are the most important. If watermarks are ok, kswapd will go |
| 1885 | * back to sleep. High-order users can still perform direct |
| 1886 | * reclaim if they wish. |
| 1887 | */ |
| 1888 | if (sc.nr_reclaimed < SWAP_CLUSTER_MAX) |
| 1889 | order = sc.order = 0; |
| 1890 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1891 | goto loop_again; |
| 1892 | } |
| 1893 | |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1894 | return sc.nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1895 | } |
| 1896 | |
| 1897 | /* |
| 1898 | * The background pageout daemon, started as a kernel thread |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1899 | * from the init process. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1900 | * |
| 1901 | * This basically trickles out pages so that we have _some_ |
| 1902 | * free memory available even if there is no other activity |
| 1903 | * that frees anything up. This is needed for things like routing |
| 1904 | * etc, where we otherwise might have all activity going on in |
| 1905 | * asynchronous contexts that cannot page things out. |
| 1906 | * |
| 1907 | * If there are applications that are active memory-allocators |
| 1908 | * (most normal use), this basically shouldn't matter. |
| 1909 | */ |
| 1910 | static int kswapd(void *p) |
| 1911 | { |
| 1912 | unsigned long order; |
| 1913 | pg_data_t *pgdat = (pg_data_t*)p; |
| 1914 | struct task_struct *tsk = current; |
| 1915 | DEFINE_WAIT(wait); |
| 1916 | struct reclaim_state reclaim_state = { |
| 1917 | .reclaimed_slab = 0, |
| 1918 | }; |
Mike Travis | c5f59f0 | 2008-04-04 18:11:10 -0700 | [diff] [blame] | 1919 | node_to_cpumask_ptr(cpumask, pgdat->node_id); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1920 | |
Rusty Russell | 174596a | 2009-01-01 10:12:29 +1030 | [diff] [blame] | 1921 | if (!cpumask_empty(cpumask)) |
Mike Travis | c5f59f0 | 2008-04-04 18:11:10 -0700 | [diff] [blame] | 1922 | set_cpus_allowed_ptr(tsk, cpumask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1923 | current->reclaim_state = &reclaim_state; |
| 1924 | |
| 1925 | /* |
| 1926 | * Tell the memory management that we're a "memory allocator", |
| 1927 | * and that if we need more memory we should get access to it |
| 1928 | * regardless (see "__alloc_pages()"). "kswapd" should |
| 1929 | * never get caught in the normal page freeing logic. |
| 1930 | * |
| 1931 | * (Kswapd normally doesn't need memory anyway, but sometimes |
| 1932 | * you need a small amount of memory in order to be able to |
| 1933 | * page out something else, and this flag essentially protects |
| 1934 | * us from recursively trying to free more memory as we're |
| 1935 | * trying to free the first piece of memory in the first place). |
| 1936 | */ |
Christoph Lameter | 930d915 | 2006-01-08 01:00:47 -0800 | [diff] [blame] | 1937 | tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD; |
Rafael J. Wysocki | 8314418 | 2007-07-17 04:03:35 -0700 | [diff] [blame] | 1938 | set_freezable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1939 | |
| 1940 | order = 0; |
| 1941 | for ( ; ; ) { |
| 1942 | unsigned long new_order; |
Christoph Lameter | 3e1d1d2 | 2005-06-24 23:13:50 -0700 | [diff] [blame] | 1943 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1944 | prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); |
| 1945 | new_order = pgdat->kswapd_max_order; |
| 1946 | pgdat->kswapd_max_order = 0; |
| 1947 | if (order < new_order) { |
| 1948 | /* |
| 1949 | * Don't sleep if someone wants a larger 'order' |
| 1950 | * allocation |
| 1951 | */ |
| 1952 | order = new_order; |
| 1953 | } else { |
Rafael J. Wysocki | b1296cc | 2007-05-06 14:50:48 -0700 | [diff] [blame] | 1954 | if (!freezing(current)) |
| 1955 | schedule(); |
| 1956 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1957 | order = pgdat->kswapd_max_order; |
| 1958 | } |
| 1959 | finish_wait(&pgdat->kswapd_wait, &wait); |
| 1960 | |
Rafael J. Wysocki | b1296cc | 2007-05-06 14:50:48 -0700 | [diff] [blame] | 1961 | if (!try_to_freeze()) { |
| 1962 | /* We can speed up thawing tasks if we don't call |
| 1963 | * balance_pgdat after returning from the refrigerator |
| 1964 | */ |
| 1965 | balance_pgdat(pgdat, order); |
| 1966 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1967 | } |
| 1968 | return 0; |
| 1969 | } |
| 1970 | |
| 1971 | /* |
| 1972 | * A zone is low on free memory, so wake its kswapd task to service it. |
| 1973 | */ |
| 1974 | void wakeup_kswapd(struct zone *zone, int order) |
| 1975 | { |
| 1976 | pg_data_t *pgdat; |
| 1977 | |
Con Kolivas | f3fe651 | 2006-01-06 00:11:15 -0800 | [diff] [blame] | 1978 | if (!populated_zone(zone)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1979 | return; |
| 1980 | |
| 1981 | pgdat = zone->zone_pgdat; |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1982 | if (zone_watermark_ok(zone, order, zone->pages_low, 0, 0)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1983 | return; |
| 1984 | if (pgdat->kswapd_max_order < order) |
| 1985 | pgdat->kswapd_max_order = order; |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 1986 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1987 | return; |
Con Kolivas | 8d0986e | 2005-09-13 01:25:07 -0700 | [diff] [blame] | 1988 | if (!waitqueue_active(&pgdat->kswapd_wait)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1989 | return; |
Con Kolivas | 8d0986e | 2005-09-13 01:25:07 -0700 | [diff] [blame] | 1990 | wake_up_interruptible(&pgdat->kswapd_wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1991 | } |
| 1992 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1993 | unsigned long global_lru_pages(void) |
| 1994 | { |
| 1995 | return global_page_state(NR_ACTIVE_ANON) |
| 1996 | + global_page_state(NR_ACTIVE_FILE) |
| 1997 | + global_page_state(NR_INACTIVE_ANON) |
| 1998 | + global_page_state(NR_INACTIVE_FILE); |
| 1999 | } |
| 2000 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2001 | #ifdef CONFIG_PM |
| 2002 | /* |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2003 | * Helper function for shrink_all_memory(). Tries to reclaim 'nr_pages' pages |
| 2004 | * from LRU lists system-wide, for given pass and priority, and returns the |
| 2005 | * number of reclaimed pages |
| 2006 | * |
| 2007 | * For pass > 3 we also try to shrink the LRU lists that contain a few pages |
| 2008 | */ |
Nigel Cunningham | e07aa05 | 2006-12-22 01:07:21 -0800 | [diff] [blame] | 2009 | static unsigned long shrink_all_zones(unsigned long nr_pages, int prio, |
| 2010 | int pass, struct scan_control *sc) |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2011 | { |
| 2012 | struct zone *zone; |
| 2013 | unsigned long nr_to_scan, ret = 0; |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 2014 | enum lru_list l; |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2015 | |
| 2016 | for_each_zone(zone) { |
| 2017 | |
| 2018 | if (!populated_zone(zone)) |
| 2019 | continue; |
| 2020 | |
David Rientjes | e815af9 | 2007-10-16 23:25:54 -0700 | [diff] [blame] | 2021 | if (zone_is_all_unreclaimable(zone) && prio != DEF_PRIORITY) |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2022 | continue; |
| 2023 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 2024 | for_each_evictable_lru(l) { |
| 2025 | /* For pass = 0, we don't shrink the active list */ |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2026 | if (pass == 0 && |
| 2027 | (l == LRU_ACTIVE || l == LRU_ACTIVE_FILE)) |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 2028 | continue; |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2029 | |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 2030 | zone->lru[l].nr_scan += |
| 2031 | (zone_page_state(zone, NR_LRU_BASE + l) |
| 2032 | >> prio) + 1; |
| 2033 | if (zone->lru[l].nr_scan >= nr_pages || pass > 3) { |
| 2034 | zone->lru[l].nr_scan = 0; |
| 2035 | nr_to_scan = min(nr_pages, |
| 2036 | zone_page_state(zone, |
| 2037 | NR_LRU_BASE + l)); |
| 2038 | ret += shrink_list(l, nr_to_scan, zone, |
| 2039 | sc, prio); |
| 2040 | if (ret >= nr_pages) |
| 2041 | return ret; |
| 2042 | } |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2043 | } |
| 2044 | } |
| 2045 | |
| 2046 | return ret; |
| 2047 | } |
| 2048 | |
| 2049 | /* |
| 2050 | * Try to free `nr_pages' of memory, system-wide, and return the number of |
| 2051 | * freed pages. |
| 2052 | * |
| 2053 | * Rather than trying to age LRUs the aim is to preserve the overall |
| 2054 | * LRU order by reclaiming preferentially |
| 2055 | * inactive > active > active referenced > active mapped |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2056 | */ |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 2057 | unsigned long shrink_all_memory(unsigned long nr_pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2058 | { |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2059 | unsigned long lru_pages, nr_slab; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 2060 | unsigned long ret = 0; |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2061 | int pass; |
| 2062 | struct reclaim_state reclaim_state; |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2063 | struct scan_control sc = { |
| 2064 | .gfp_mask = GFP_KERNEL, |
| 2065 | .may_swap = 0, |
| 2066 | .swap_cluster_max = nr_pages, |
| 2067 | .may_writepage = 1, |
| 2068 | .swappiness = vm_swappiness, |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2069 | .isolate_pages = isolate_pages_global, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2070 | }; |
| 2071 | |
| 2072 | current->reclaim_state = &reclaim_state; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 2073 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2074 | lru_pages = global_lru_pages(); |
Christoph Lameter | 972d1a7 | 2006-09-25 23:31:51 -0700 | [diff] [blame] | 2075 | nr_slab = global_page_state(NR_SLAB_RECLAIMABLE); |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2076 | /* If slab caches are huge, it's better to hit them first */ |
| 2077 | while (nr_slab >= lru_pages) { |
| 2078 | reclaim_state.reclaimed_slab = 0; |
| 2079 | shrink_slab(nr_pages, sc.gfp_mask, lru_pages); |
| 2080 | if (!reclaim_state.reclaimed_slab) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2081 | break; |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2082 | |
| 2083 | ret += reclaim_state.reclaimed_slab; |
| 2084 | if (ret >= nr_pages) |
| 2085 | goto out; |
| 2086 | |
| 2087 | nr_slab -= reclaim_state.reclaimed_slab; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2088 | } |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2089 | |
| 2090 | /* |
| 2091 | * We try to shrink LRUs in 5 passes: |
| 2092 | * 0 = Reclaim from inactive_list only |
| 2093 | * 1 = Reclaim from active list but don't reclaim mapped |
| 2094 | * 2 = 2nd pass of type 1 |
| 2095 | * 3 = Reclaim mapped (normal reclaim) |
| 2096 | * 4 = 2nd pass of type 3 |
| 2097 | */ |
| 2098 | for (pass = 0; pass < 5; pass++) { |
| 2099 | int prio; |
| 2100 | |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2101 | /* Force reclaiming mapped pages in the passes #3 and #4 */ |
| 2102 | if (pass > 2) { |
| 2103 | sc.may_swap = 1; |
| 2104 | sc.swappiness = 100; |
| 2105 | } |
| 2106 | |
| 2107 | for (prio = DEF_PRIORITY; prio >= 0; prio--) { |
| 2108 | unsigned long nr_to_scan = nr_pages - ret; |
| 2109 | |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2110 | sc.nr_scanned = 0; |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2111 | ret += shrink_all_zones(nr_to_scan, prio, pass, &sc); |
| 2112 | if (ret >= nr_pages) |
| 2113 | goto out; |
| 2114 | |
| 2115 | reclaim_state.reclaimed_slab = 0; |
Andrew Morton | 76395d3 | 2007-01-05 16:37:05 -0800 | [diff] [blame] | 2116 | shrink_slab(sc.nr_scanned, sc.gfp_mask, |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2117 | global_lru_pages()); |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2118 | ret += reclaim_state.reclaimed_slab; |
| 2119 | if (ret >= nr_pages) |
| 2120 | goto out; |
| 2121 | |
| 2122 | if (sc.nr_scanned && prio < DEF_PRIORITY - 2) |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 2123 | congestion_wait(WRITE, HZ / 10); |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2124 | } |
Rafael J. Wysocki | 248a030 | 2006-03-22 00:09:04 -0800 | [diff] [blame] | 2125 | } |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2126 | |
| 2127 | /* |
| 2128 | * If ret = 0, we could not shrink LRUs, but there may be something |
| 2129 | * in slab caches |
| 2130 | */ |
Andrew Morton | 76395d3 | 2007-01-05 16:37:05 -0800 | [diff] [blame] | 2131 | if (!ret) { |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2132 | do { |
| 2133 | reclaim_state.reclaimed_slab = 0; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2134 | shrink_slab(nr_pages, sc.gfp_mask, global_lru_pages()); |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2135 | ret += reclaim_state.reclaimed_slab; |
| 2136 | } while (ret < nr_pages && reclaim_state.reclaimed_slab > 0); |
Andrew Morton | 76395d3 | 2007-01-05 16:37:05 -0800 | [diff] [blame] | 2137 | } |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2138 | |
| 2139 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2140 | current->reclaim_state = NULL; |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2141 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2142 | return ret; |
| 2143 | } |
| 2144 | #endif |
| 2145 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2146 | /* It's optimal to keep kswapds on the same CPUs as their memory, but |
| 2147 | not required for correctness. So if the last cpu in a node goes |
| 2148 | away, we get changed to run anywhere: as the first one comes back, |
| 2149 | restore their cpu bindings. */ |
Chandra Seetharaman | 9c7b216 | 2006-06-27 02:54:07 -0700 | [diff] [blame] | 2150 | static int __devinit cpu_callback(struct notifier_block *nfb, |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 2151 | unsigned long action, void *hcpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2152 | { |
Yasunori Goto | 58c0a4a | 2007-10-16 01:25:40 -0700 | [diff] [blame] | 2153 | int nid; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2154 | |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 2155 | if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN) { |
Yasunori Goto | 58c0a4a | 2007-10-16 01:25:40 -0700 | [diff] [blame] | 2156 | for_each_node_state(nid, N_HIGH_MEMORY) { |
Mike Travis | c5f59f0 | 2008-04-04 18:11:10 -0700 | [diff] [blame] | 2157 | pg_data_t *pgdat = NODE_DATA(nid); |
| 2158 | node_to_cpumask_ptr(mask, pgdat->node_id); |
| 2159 | |
Rusty Russell | 3e59794 | 2009-01-01 10:12:24 +1030 | [diff] [blame] | 2160 | if (cpumask_any_and(cpu_online_mask, mask) < nr_cpu_ids) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2161 | /* One of our CPUs online: restore mask */ |
Mike Travis | c5f59f0 | 2008-04-04 18:11:10 -0700 | [diff] [blame] | 2162 | set_cpus_allowed_ptr(pgdat->kswapd, mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2163 | } |
| 2164 | } |
| 2165 | return NOTIFY_OK; |
| 2166 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2167 | |
Yasunori Goto | 3218ae1 | 2006-06-27 02:53:33 -0700 | [diff] [blame] | 2168 | /* |
| 2169 | * This kswapd start function will be called by init and node-hot-add. |
| 2170 | * On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added. |
| 2171 | */ |
| 2172 | int kswapd_run(int nid) |
| 2173 | { |
| 2174 | pg_data_t *pgdat = NODE_DATA(nid); |
| 2175 | int ret = 0; |
| 2176 | |
| 2177 | if (pgdat->kswapd) |
| 2178 | return 0; |
| 2179 | |
| 2180 | pgdat->kswapd = kthread_run(kswapd, pgdat, "kswapd%d", nid); |
| 2181 | if (IS_ERR(pgdat->kswapd)) { |
| 2182 | /* failure at boot is fatal */ |
| 2183 | BUG_ON(system_state == SYSTEM_BOOTING); |
| 2184 | printk("Failed to start kswapd on node %d\n",nid); |
| 2185 | ret = -1; |
| 2186 | } |
| 2187 | return ret; |
| 2188 | } |
| 2189 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2190 | static int __init kswapd_init(void) |
| 2191 | { |
Yasunori Goto | 3218ae1 | 2006-06-27 02:53:33 -0700 | [diff] [blame] | 2192 | int nid; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 2193 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2194 | swap_setup(); |
Christoph Lameter | 9422ffb | 2007-10-16 01:25:31 -0700 | [diff] [blame] | 2195 | for_each_node_state(nid, N_HIGH_MEMORY) |
Yasunori Goto | 3218ae1 | 2006-06-27 02:53:33 -0700 | [diff] [blame] | 2196 | kswapd_run(nid); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2197 | hotcpu_notifier(cpu_callback, 0); |
| 2198 | return 0; |
| 2199 | } |
| 2200 | |
| 2201 | module_init(kswapd_init) |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2202 | |
| 2203 | #ifdef CONFIG_NUMA |
| 2204 | /* |
| 2205 | * Zone reclaim mode |
| 2206 | * |
| 2207 | * If non-zero call zone_reclaim when the number of free pages falls below |
| 2208 | * the watermarks. |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2209 | */ |
| 2210 | int zone_reclaim_mode __read_mostly; |
| 2211 | |
Christoph Lameter | 1b2ffb7 | 2006-02-01 03:05:34 -0800 | [diff] [blame] | 2212 | #define RECLAIM_OFF 0 |
Fernando Luis Vazquez Cao | 7d03431 | 2008-07-29 22:33:41 -0700 | [diff] [blame] | 2213 | #define RECLAIM_ZONE (1<<0) /* Run shrink_inactive_list on the zone */ |
Christoph Lameter | 1b2ffb7 | 2006-02-01 03:05:34 -0800 | [diff] [blame] | 2214 | #define RECLAIM_WRITE (1<<1) /* Writeout pages during reclaim */ |
| 2215 | #define RECLAIM_SWAP (1<<2) /* Swap pages out during reclaim */ |
| 2216 | |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2217 | /* |
Christoph Lameter | a92f712 | 2006-02-01 03:05:32 -0800 | [diff] [blame] | 2218 | * Priority for ZONE_RECLAIM. This determines the fraction of pages |
| 2219 | * of a node considered for each zone_reclaim. 4 scans 1/16th of |
| 2220 | * a zone. |
| 2221 | */ |
| 2222 | #define ZONE_RECLAIM_PRIORITY 4 |
| 2223 | |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2224 | /* |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 2225 | * Percentage of pages in a zone that must be unmapped for zone_reclaim to |
| 2226 | * occur. |
| 2227 | */ |
| 2228 | int sysctl_min_unmapped_ratio = 1; |
| 2229 | |
| 2230 | /* |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 2231 | * If the number of slab pages in a zone grows beyond this percentage then |
| 2232 | * slab reclaim needs to occur. |
| 2233 | */ |
| 2234 | int sysctl_min_slab_ratio = 5; |
| 2235 | |
| 2236 | /* |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2237 | * Try to free up some pages from this zone through reclaim. |
| 2238 | */ |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2239 | 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] | 2240 | { |
Christoph Lameter | 7fb2d46 | 2006-03-22 00:08:22 -0800 | [diff] [blame] | 2241 | /* Minimum pages needed in order to stay on node */ |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 2242 | const unsigned long nr_pages = 1 << order; |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2243 | struct task_struct *p = current; |
| 2244 | struct reclaim_state reclaim_state; |
Christoph Lameter | 8695949 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2245 | int priority; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2246 | struct scan_control sc = { |
| 2247 | .may_writepage = !!(zone_reclaim_mode & RECLAIM_WRITE), |
| 2248 | .may_swap = !!(zone_reclaim_mode & RECLAIM_SWAP), |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 2249 | .swap_cluster_max = max_t(unsigned long, nr_pages, |
| 2250 | SWAP_CLUSTER_MAX), |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2251 | .gfp_mask = gfp_mask, |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2252 | .swappiness = vm_swappiness, |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2253 | .isolate_pages = isolate_pages_global, |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2254 | }; |
Christoph Lameter | 83e33a4 | 2006-09-25 23:31:53 -0700 | [diff] [blame] | 2255 | unsigned long slab_reclaimable; |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2256 | |
| 2257 | disable_swap_token(); |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2258 | cond_resched(); |
Christoph Lameter | d4f7796 | 2006-02-24 13:04:22 -0800 | [diff] [blame] | 2259 | /* |
| 2260 | * We need to be able to allocate from the reserves for RECLAIM_SWAP |
| 2261 | * and we also need to be able to write out pages for RECLAIM_WRITE |
| 2262 | * and RECLAIM_SWAP. |
| 2263 | */ |
| 2264 | p->flags |= PF_MEMALLOC | PF_SWAPWRITE; |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2265 | reclaim_state.reclaimed_slab = 0; |
| 2266 | p->reclaim_state = &reclaim_state; |
Christoph Lameter | c84db23 | 2006-02-01 03:05:29 -0800 | [diff] [blame] | 2267 | |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 2268 | if (zone_page_state(zone, NR_FILE_PAGES) - |
| 2269 | zone_page_state(zone, NR_FILE_MAPPED) > |
| 2270 | zone->min_unmapped_pages) { |
| 2271 | /* |
| 2272 | * Free memory by calling shrink zone with increasing |
| 2273 | * priorities until we have enough memory freed. |
| 2274 | */ |
| 2275 | priority = ZONE_RECLAIM_PRIORITY; |
| 2276 | do { |
Martin Bligh | 3bb1a85 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 2277 | note_zone_scanning_priority(zone, priority); |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2278 | shrink_zone(priority, zone, &sc); |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 2279 | priority--; |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2280 | } while (priority >= 0 && sc.nr_reclaimed < nr_pages); |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 2281 | } |
Christoph Lameter | c84db23 | 2006-02-01 03:05:29 -0800 | [diff] [blame] | 2282 | |
Christoph Lameter | 83e33a4 | 2006-09-25 23:31:53 -0700 | [diff] [blame] | 2283 | slab_reclaimable = zone_page_state(zone, NR_SLAB_RECLAIMABLE); |
| 2284 | if (slab_reclaimable > zone->min_slab_pages) { |
Christoph Lameter | 2a16e3f | 2006-02-01 03:05:35 -0800 | [diff] [blame] | 2285 | /* |
Christoph Lameter | 7fb2d46 | 2006-03-22 00:08:22 -0800 | [diff] [blame] | 2286 | * shrink_slab() does not currently allow us to determine how |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 2287 | * many pages were freed in this zone. So we take the current |
| 2288 | * number of slab pages and shake the slab until it is reduced |
| 2289 | * by the same nr_pages that we used for reclaiming unmapped |
| 2290 | * pages. |
Christoph Lameter | 2a16e3f | 2006-02-01 03:05:35 -0800 | [diff] [blame] | 2291 | * |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 2292 | * Note that shrink_slab will free memory on all zones and may |
| 2293 | * take a long time. |
Christoph Lameter | 2a16e3f | 2006-02-01 03:05:35 -0800 | [diff] [blame] | 2294 | */ |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 2295 | while (shrink_slab(sc.nr_scanned, gfp_mask, order) && |
Christoph Lameter | 83e33a4 | 2006-09-25 23:31:53 -0700 | [diff] [blame] | 2296 | zone_page_state(zone, NR_SLAB_RECLAIMABLE) > |
| 2297 | slab_reclaimable - nr_pages) |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 2298 | ; |
Christoph Lameter | 83e33a4 | 2006-09-25 23:31:53 -0700 | [diff] [blame] | 2299 | |
| 2300 | /* |
| 2301 | * Update nr_reclaimed by the number of slab pages we |
| 2302 | * reclaimed from this zone. |
| 2303 | */ |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2304 | sc.nr_reclaimed += slab_reclaimable - |
Christoph Lameter | 83e33a4 | 2006-09-25 23:31:53 -0700 | [diff] [blame] | 2305 | zone_page_state(zone, NR_SLAB_RECLAIMABLE); |
Christoph Lameter | 2a16e3f | 2006-02-01 03:05:35 -0800 | [diff] [blame] | 2306 | } |
| 2307 | |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2308 | p->reclaim_state = NULL; |
Christoph Lameter | d4f7796 | 2006-02-24 13:04:22 -0800 | [diff] [blame] | 2309 | current->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE); |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2310 | return sc.nr_reclaimed >= nr_pages; |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2311 | } |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2312 | |
| 2313 | int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) |
| 2314 | { |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2315 | int node_id; |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 2316 | int ret; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2317 | |
| 2318 | /* |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 2319 | * Zone reclaim reclaims unmapped file backed pages and |
| 2320 | * slab pages if we are over the defined limits. |
Christoph Lameter | 34aa133 | 2006-06-30 01:55:37 -0700 | [diff] [blame] | 2321 | * |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 2322 | * A small portion of unmapped file backed pages is needed for |
| 2323 | * file I/O otherwise pages read by file I/O will be immediately |
| 2324 | * thrown out if the zone is overallocated. So we do not reclaim |
| 2325 | * if less than a specified percentage of the zone is used by |
| 2326 | * unmapped file backed pages. |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2327 | */ |
Christoph Lameter | 34aa133 | 2006-06-30 01:55:37 -0700 | [diff] [blame] | 2328 | if (zone_page_state(zone, NR_FILE_PAGES) - |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 2329 | zone_page_state(zone, NR_FILE_MAPPED) <= zone->min_unmapped_pages |
| 2330 | && zone_page_state(zone, NR_SLAB_RECLAIMABLE) |
| 2331 | <= zone->min_slab_pages) |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 2332 | return 0; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2333 | |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 2334 | if (zone_is_all_unreclaimable(zone)) |
| 2335 | return 0; |
| 2336 | |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2337 | /* |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 2338 | * Do not scan if the allocation should not be delayed. |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2339 | */ |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 2340 | if (!(gfp_mask & __GFP_WAIT) || (current->flags & PF_MEMALLOC)) |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2341 | return 0; |
| 2342 | |
| 2343 | /* |
| 2344 | * Only run zone reclaim on the local zone or on zones that do not |
| 2345 | * have associated processors. This will favor the local processor |
| 2346 | * over remote processors and spread off node memory allocations |
| 2347 | * as wide as possible. |
| 2348 | */ |
Christoph Lameter | 89fa302 | 2006-09-25 23:31:55 -0700 | [diff] [blame] | 2349 | node_id = zone_to_nid(zone); |
Christoph Lameter | 37c0708 | 2007-10-16 01:25:36 -0700 | [diff] [blame] | 2350 | if (node_state(node_id, N_CPU) && node_id != numa_node_id()) |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2351 | return 0; |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 2352 | |
| 2353 | if (zone_test_and_set_flag(zone, ZONE_RECLAIM_LOCKED)) |
| 2354 | return 0; |
| 2355 | ret = __zone_reclaim(zone, gfp_mask, order); |
| 2356 | zone_clear_flag(zone, ZONE_RECLAIM_LOCKED); |
| 2357 | |
| 2358 | return ret; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2359 | } |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2360 | #endif |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 2361 | |
| 2362 | #ifdef CONFIG_UNEVICTABLE_LRU |
| 2363 | /* |
| 2364 | * page_evictable - test whether a page is evictable |
| 2365 | * @page: the page to test |
| 2366 | * @vma: the VMA in which the page is or will be mapped, may be NULL |
| 2367 | * |
| 2368 | * 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] | 2369 | * lists vs unevictable list. The vma argument is !NULL when called from the |
| 2370 | * fault path to determine how to instantate a new page. |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 2371 | * |
| 2372 | * Reasons page might not be evictable: |
Lee Schermerhorn | ba9ddf4 | 2008-10-18 20:26:42 -0700 | [diff] [blame] | 2373 | * (1) page's mapping marked unevictable |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 2374 | * (2) page is part of an mlocked VMA |
Lee Schermerhorn | ba9ddf4 | 2008-10-18 20:26:42 -0700 | [diff] [blame] | 2375 | * |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 2376 | */ |
| 2377 | int page_evictable(struct page *page, struct vm_area_struct *vma) |
| 2378 | { |
| 2379 | |
Lee Schermerhorn | ba9ddf4 | 2008-10-18 20:26:42 -0700 | [diff] [blame] | 2380 | if (mapping_unevictable(page_mapping(page))) |
| 2381 | return 0; |
| 2382 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 2383 | if (PageMlocked(page) || (vma && is_mlocked_vma(vma, page))) |
| 2384 | return 0; |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 2385 | |
| 2386 | return 1; |
| 2387 | } |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 2388 | |
| 2389 | /** |
| 2390 | * check_move_unevictable_page - check page for evictability and move to appropriate zone lru list |
| 2391 | * @page: page to check evictability and move to appropriate lru list |
| 2392 | * @zone: zone page is in |
| 2393 | * |
| 2394 | * Checks a page for evictability and moves the page to the appropriate |
| 2395 | * zone lru list. |
| 2396 | * |
| 2397 | * Restrictions: zone->lru_lock must be held, page must be on LRU and must |
| 2398 | * have PageUnevictable set. |
| 2399 | */ |
| 2400 | static void check_move_unevictable_page(struct page *page, struct zone *zone) |
| 2401 | { |
| 2402 | VM_BUG_ON(PageActive(page)); |
| 2403 | |
| 2404 | retry: |
| 2405 | ClearPageUnevictable(page); |
| 2406 | if (page_evictable(page, NULL)) { |
| 2407 | enum lru_list l = LRU_INACTIVE_ANON + page_is_file_cache(page); |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 2408 | |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 2409 | __dec_zone_state(zone, NR_UNEVICTABLE); |
| 2410 | list_move(&page->lru, &zone->lru[l].list); |
| 2411 | __inc_zone_state(zone, NR_INACTIVE_ANON + l); |
| 2412 | __count_vm_event(UNEVICTABLE_PGRESCUED); |
| 2413 | } else { |
| 2414 | /* |
| 2415 | * rotate unevictable list |
| 2416 | */ |
| 2417 | SetPageUnevictable(page); |
| 2418 | list_move(&page->lru, &zone->lru[LRU_UNEVICTABLE].list); |
| 2419 | if (page_evictable(page, NULL)) |
| 2420 | goto retry; |
| 2421 | } |
| 2422 | } |
| 2423 | |
| 2424 | /** |
| 2425 | * scan_mapping_unevictable_pages - scan an address space for evictable pages |
| 2426 | * @mapping: struct address_space to scan for evictable pages |
| 2427 | * |
| 2428 | * Scan all pages in mapping. Check unevictable pages for |
| 2429 | * evictability and move them to the appropriate zone lru list. |
| 2430 | */ |
| 2431 | void scan_mapping_unevictable_pages(struct address_space *mapping) |
| 2432 | { |
| 2433 | pgoff_t next = 0; |
| 2434 | pgoff_t end = (i_size_read(mapping->host) + PAGE_CACHE_SIZE - 1) >> |
| 2435 | PAGE_CACHE_SHIFT; |
| 2436 | struct zone *zone; |
| 2437 | struct pagevec pvec; |
| 2438 | |
| 2439 | if (mapping->nrpages == 0) |
| 2440 | return; |
| 2441 | |
| 2442 | pagevec_init(&pvec, 0); |
| 2443 | while (next < end && |
| 2444 | pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { |
| 2445 | int i; |
| 2446 | int pg_scanned = 0; |
| 2447 | |
| 2448 | zone = NULL; |
| 2449 | |
| 2450 | for (i = 0; i < pagevec_count(&pvec); i++) { |
| 2451 | struct page *page = pvec.pages[i]; |
| 2452 | pgoff_t page_index = page->index; |
| 2453 | struct zone *pagezone = page_zone(page); |
| 2454 | |
| 2455 | pg_scanned++; |
| 2456 | if (page_index > next) |
| 2457 | next = page_index; |
| 2458 | next++; |
| 2459 | |
| 2460 | if (pagezone != zone) { |
| 2461 | if (zone) |
| 2462 | spin_unlock_irq(&zone->lru_lock); |
| 2463 | zone = pagezone; |
| 2464 | spin_lock_irq(&zone->lru_lock); |
| 2465 | } |
| 2466 | |
| 2467 | if (PageLRU(page) && PageUnevictable(page)) |
| 2468 | check_move_unevictable_page(page, zone); |
| 2469 | } |
| 2470 | if (zone) |
| 2471 | spin_unlock_irq(&zone->lru_lock); |
| 2472 | pagevec_release(&pvec); |
| 2473 | |
| 2474 | count_vm_events(UNEVICTABLE_PGSCANNED, pg_scanned); |
| 2475 | } |
| 2476 | |
| 2477 | } |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 2478 | |
| 2479 | /** |
| 2480 | * scan_zone_unevictable_pages - check unevictable list for evictable pages |
| 2481 | * @zone - zone of which to scan the unevictable list |
| 2482 | * |
| 2483 | * Scan @zone's unevictable LRU lists to check for pages that have become |
| 2484 | * evictable. Move those that have to @zone's inactive list where they |
| 2485 | * become candidates for reclaim, unless shrink_inactive_zone() decides |
| 2486 | * to reactivate them. Pages that are still unevictable are rotated |
| 2487 | * back onto @zone's unevictable list. |
| 2488 | */ |
| 2489 | #define SCAN_UNEVICTABLE_BATCH_SIZE 16UL /* arbitrary lock hold batch size */ |
KOSAKI Motohiro | 14b90b2 | 2009-01-06 14:39:45 -0800 | [diff] [blame] | 2490 | static void scan_zone_unevictable_pages(struct zone *zone) |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 2491 | { |
| 2492 | struct list_head *l_unevictable = &zone->lru[LRU_UNEVICTABLE].list; |
| 2493 | unsigned long scan; |
| 2494 | unsigned long nr_to_scan = zone_page_state(zone, NR_UNEVICTABLE); |
| 2495 | |
| 2496 | while (nr_to_scan > 0) { |
| 2497 | unsigned long batch_size = min(nr_to_scan, |
| 2498 | SCAN_UNEVICTABLE_BATCH_SIZE); |
| 2499 | |
| 2500 | spin_lock_irq(&zone->lru_lock); |
| 2501 | for (scan = 0; scan < batch_size; scan++) { |
| 2502 | struct page *page = lru_to_page(l_unevictable); |
| 2503 | |
| 2504 | if (!trylock_page(page)) |
| 2505 | continue; |
| 2506 | |
| 2507 | prefetchw_prev_lru_page(page, l_unevictable, flags); |
| 2508 | |
| 2509 | if (likely(PageLRU(page) && PageUnevictable(page))) |
| 2510 | check_move_unevictable_page(page, zone); |
| 2511 | |
| 2512 | unlock_page(page); |
| 2513 | } |
| 2514 | spin_unlock_irq(&zone->lru_lock); |
| 2515 | |
| 2516 | nr_to_scan -= batch_size; |
| 2517 | } |
| 2518 | } |
| 2519 | |
| 2520 | |
| 2521 | /** |
| 2522 | * scan_all_zones_unevictable_pages - scan all unevictable lists for evictable pages |
| 2523 | * |
| 2524 | * A really big hammer: scan all zones' unevictable LRU lists to check for |
| 2525 | * pages that have become evictable. Move those back to the zones' |
| 2526 | * inactive list where they become candidates for reclaim. |
| 2527 | * This occurs when, e.g., we have unswappable pages on the unevictable lists, |
| 2528 | * and we add swap to the system. As such, it runs in the context of a task |
| 2529 | * that has possibly/probably made some previously unevictable pages |
| 2530 | * evictable. |
| 2531 | */ |
KOSAKI Motohiro | ff30153 | 2009-01-06 14:39:44 -0800 | [diff] [blame] | 2532 | static void scan_all_zones_unevictable_pages(void) |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 2533 | { |
| 2534 | struct zone *zone; |
| 2535 | |
| 2536 | for_each_zone(zone) { |
| 2537 | scan_zone_unevictable_pages(zone); |
| 2538 | } |
| 2539 | } |
| 2540 | |
| 2541 | /* |
| 2542 | * scan_unevictable_pages [vm] sysctl handler. On demand re-scan of |
| 2543 | * all nodes' unevictable lists for evictable pages |
| 2544 | */ |
| 2545 | unsigned long scan_unevictable_pages; |
| 2546 | |
| 2547 | int scan_unevictable_handler(struct ctl_table *table, int write, |
| 2548 | struct file *file, void __user *buffer, |
| 2549 | size_t *length, loff_t *ppos) |
| 2550 | { |
| 2551 | proc_doulongvec_minmax(table, write, file, buffer, length, ppos); |
| 2552 | |
| 2553 | if (write && *(unsigned long *)table->data) |
| 2554 | scan_all_zones_unevictable_pages(); |
| 2555 | |
| 2556 | scan_unevictable_pages = 0; |
| 2557 | return 0; |
| 2558 | } |
| 2559 | |
| 2560 | /* |
| 2561 | * per node 'scan_unevictable_pages' attribute. On demand re-scan of |
| 2562 | * a specified node's per zone unevictable lists for evictable pages. |
| 2563 | */ |
| 2564 | |
| 2565 | static ssize_t read_scan_unevictable_node(struct sys_device *dev, |
| 2566 | struct sysdev_attribute *attr, |
| 2567 | char *buf) |
| 2568 | { |
| 2569 | return sprintf(buf, "0\n"); /* always zero; should fit... */ |
| 2570 | } |
| 2571 | |
| 2572 | static ssize_t write_scan_unevictable_node(struct sys_device *dev, |
| 2573 | struct sysdev_attribute *attr, |
| 2574 | const char *buf, size_t count) |
| 2575 | { |
| 2576 | struct zone *node_zones = NODE_DATA(dev->id)->node_zones; |
| 2577 | struct zone *zone; |
| 2578 | unsigned long res; |
| 2579 | unsigned long req = strict_strtoul(buf, 10, &res); |
| 2580 | |
| 2581 | if (!req) |
| 2582 | return 1; /* zero is no-op */ |
| 2583 | |
| 2584 | for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) { |
| 2585 | if (!populated_zone(zone)) |
| 2586 | continue; |
| 2587 | scan_zone_unevictable_pages(zone); |
| 2588 | } |
| 2589 | return 1; |
| 2590 | } |
| 2591 | |
| 2592 | |
| 2593 | static SYSDEV_ATTR(scan_unevictable_pages, S_IRUGO | S_IWUSR, |
| 2594 | read_scan_unevictable_node, |
| 2595 | write_scan_unevictable_node); |
| 2596 | |
| 2597 | int scan_unevictable_register_node(struct node *node) |
| 2598 | { |
| 2599 | return sysdev_create_file(&node->sysdev, &attr_scan_unevictable_pages); |
| 2600 | } |
| 2601 | |
| 2602 | void scan_unevictable_unregister_node(struct node *node) |
| 2603 | { |
| 2604 | sysdev_remove_file(&node->sysdev, &attr_scan_unevictable_pages); |
| 2605 | } |
| 2606 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 2607 | #endif |