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> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 16 | #include <linux/gfp.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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> |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 35 | #include <linux/compaction.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 36 | #include <linux/notifier.h> |
| 37 | #include <linux/rwsem.h> |
Rafael J. Wysocki | 248a030 | 2006-03-22 00:09:04 -0800 | [diff] [blame] | 38 | #include <linux/delay.h> |
Yasunori Goto | 3218ae1 | 2006-06-27 02:53:33 -0700 | [diff] [blame] | 39 | #include <linux/kthread.h> |
Nigel Cunningham | 7dfb710 | 2006-12-06 20:34:23 -0800 | [diff] [blame] | 40 | #include <linux/freezer.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 41 | #include <linux/memcontrol.h> |
Keika Kobayashi | 873b477 | 2008-07-25 01:48:52 -0700 | [diff] [blame] | 42 | #include <linux/delayacct.h> |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 43 | #include <linux/sysctl.h> |
KOSAKI Motohiro | 929bea7 | 2011-04-14 15:22:12 -0700 | [diff] [blame] | 44 | #include <linux/oom.h> |
Linus Torvalds | 268bb0c | 2011-05-20 12:50:29 -0700 | [diff] [blame] | 45 | #include <linux/prefetch.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 46 | |
| 47 | #include <asm/tlbflush.h> |
| 48 | #include <asm/div64.h> |
| 49 | |
| 50 | #include <linux/swapops.h> |
| 51 | |
Nick Piggin | 0f8053a | 2006-03-22 00:08:33 -0800 | [diff] [blame] | 52 | #include "internal.h" |
| 53 | |
Mel Gorman | 33906bc | 2010-08-09 17:19:16 -0700 | [diff] [blame] | 54 | #define CREATE_TRACE_POINTS |
| 55 | #include <trace/events/vmscan.h> |
| 56 | |
Mel Gorman | ee64fc9 | 2011-01-13 15:45:55 -0800 | [diff] [blame] | 57 | /* |
Mel Gorman | f3a310b | 2011-01-13 15:46:00 -0800 | [diff] [blame] | 58 | * reclaim_mode determines how the inactive list is shrunk |
| 59 | * RECLAIM_MODE_SINGLE: Reclaim only order-0 pages |
| 60 | * RECLAIM_MODE_ASYNC: Do not block |
| 61 | * RECLAIM_MODE_SYNC: Allow blocking e.g. call wait_on_page_writeback |
| 62 | * RECLAIM_MODE_LUMPYRECLAIM: For high-order allocations, take a reference |
Mel Gorman | ee64fc9 | 2011-01-13 15:45:55 -0800 | [diff] [blame] | 63 | * page from the LRU and reclaim all pages within a |
| 64 | * naturally aligned range |
Mel Gorman | f3a310b | 2011-01-13 15:46:00 -0800 | [diff] [blame] | 65 | * RECLAIM_MODE_COMPACTION: For high-order allocations, reclaim a number of |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 66 | * order-0 pages and then compact the zone |
Mel Gorman | ee64fc9 | 2011-01-13 15:45:55 -0800 | [diff] [blame] | 67 | */ |
Mel Gorman | f3a310b | 2011-01-13 15:46:00 -0800 | [diff] [blame] | 68 | typedef unsigned __bitwise__ reclaim_mode_t; |
| 69 | #define RECLAIM_MODE_SINGLE ((__force reclaim_mode_t)0x01u) |
| 70 | #define RECLAIM_MODE_ASYNC ((__force reclaim_mode_t)0x02u) |
| 71 | #define RECLAIM_MODE_SYNC ((__force reclaim_mode_t)0x04u) |
| 72 | #define RECLAIM_MODE_LUMPYRECLAIM ((__force reclaim_mode_t)0x08u) |
| 73 | #define RECLAIM_MODE_COMPACTION ((__force reclaim_mode_t)0x10u) |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 74 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 75 | struct scan_control { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 76 | /* Incremented by the number of inactive pages that were scanned */ |
| 77 | unsigned long nr_scanned; |
| 78 | |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 79 | /* Number of pages freed so far during a call to shrink_zones() */ |
| 80 | unsigned long nr_reclaimed; |
| 81 | |
KOSAKI Motohiro | 22fba33 | 2009-12-14 17:59:10 -0800 | [diff] [blame] | 82 | /* How many pages shrink_list() should reclaim */ |
| 83 | unsigned long nr_to_reclaim; |
| 84 | |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 85 | unsigned long hibernation_mode; |
| 86 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 87 | /* This context's GFP mask */ |
Al Viro | 6daa0e2 | 2005-10-21 03:18:50 -0400 | [diff] [blame] | 88 | gfp_t gfp_mask; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 89 | |
| 90 | int may_writepage; |
| 91 | |
Johannes Weiner | a6dc60f8 | 2009-03-31 15:19:30 -0700 | [diff] [blame] | 92 | /* Can mapped pages be reclaimed? */ |
| 93 | int may_unmap; |
Christoph Lameter | f1fd106 | 2006-01-18 17:42:30 -0800 | [diff] [blame] | 94 | |
KOSAKI Motohiro | 2e2e425 | 2009-04-21 12:24:57 -0700 | [diff] [blame] | 95 | /* Can pages be swapped as part of reclaim? */ |
| 96 | int may_swap; |
| 97 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 98 | int order; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 99 | |
KOSAKI Motohiro | 5f53e76 | 2010-05-24 14:32:37 -0700 | [diff] [blame] | 100 | /* |
Nikanth Karthikesan | 415b54e | 2010-08-17 15:39:09 +0530 | [diff] [blame] | 101 | * Intend to reclaim enough continuous memory rather than reclaim |
| 102 | * enough amount of memory. i.e, mode for high order allocation. |
KOSAKI Motohiro | 5f53e76 | 2010-05-24 14:32:37 -0700 | [diff] [blame] | 103 | */ |
Mel Gorman | f3a310b | 2011-01-13 15:46:00 -0800 | [diff] [blame] | 104 | reclaim_mode_t reclaim_mode; |
KOSAKI Motohiro | 5f53e76 | 2010-05-24 14:32:37 -0700 | [diff] [blame] | 105 | |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 106 | /* Which cgroup do we reclaim from */ |
| 107 | struct mem_cgroup *mem_cgroup; |
| 108 | |
KAMEZAWA Hiroyuki | 327c0e9 | 2009-03-31 15:23:31 -0700 | [diff] [blame] | 109 | /* |
| 110 | * Nodemask of nodes allowed by the caller. If NULL, all nodes |
| 111 | * are scanned. |
| 112 | */ |
| 113 | nodemask_t *nodemask; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 114 | }; |
| 115 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 116 | #define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru)) |
| 117 | |
| 118 | #ifdef ARCH_HAS_PREFETCH |
| 119 | #define prefetch_prev_lru_page(_page, _base, _field) \ |
| 120 | do { \ |
| 121 | if ((_page)->lru.prev != _base) { \ |
| 122 | struct page *prev; \ |
| 123 | \ |
| 124 | prev = lru_to_page(&(_page->lru)); \ |
| 125 | prefetch(&prev->_field); \ |
| 126 | } \ |
| 127 | } while (0) |
| 128 | #else |
| 129 | #define prefetch_prev_lru_page(_page, _base, _field) do { } while (0) |
| 130 | #endif |
| 131 | |
| 132 | #ifdef ARCH_HAS_PREFETCHW |
| 133 | #define prefetchw_prev_lru_page(_page, _base, _field) \ |
| 134 | do { \ |
| 135 | if ((_page)->lru.prev != _base) { \ |
| 136 | struct page *prev; \ |
| 137 | \ |
| 138 | prev = lru_to_page(&(_page->lru)); \ |
| 139 | prefetchw(&prev->_field); \ |
| 140 | } \ |
| 141 | } while (0) |
| 142 | #else |
| 143 | #define prefetchw_prev_lru_page(_page, _base, _field) do { } while (0) |
| 144 | #endif |
| 145 | |
| 146 | /* |
| 147 | * From 0 .. 100. Higher means more swappy. |
| 148 | */ |
| 149 | int vm_swappiness = 60; |
Andrew Morton | bd1e22b | 2006-06-23 02:03:47 -0700 | [diff] [blame] | 150 | 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] | 151 | |
| 152 | static LIST_HEAD(shrinker_list); |
| 153 | static DECLARE_RWSEM(shrinker_rwsem); |
| 154 | |
Balbir Singh | 00f0b82 | 2008-03-04 14:28:39 -0800 | [diff] [blame] | 155 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
KAMEZAWA Hiroyuki | e72e2bd | 2009-01-07 18:08:23 -0800 | [diff] [blame] | 156 | #define scanning_global_lru(sc) (!(sc)->mem_cgroup) |
KAMEZAWA Hiroyuki | 91a4547 | 2008-02-07 00:14:29 -0800 | [diff] [blame] | 157 | #else |
KAMEZAWA Hiroyuki | e72e2bd | 2009-01-07 18:08:23 -0800 | [diff] [blame] | 158 | #define scanning_global_lru(sc) (1) |
KAMEZAWA Hiroyuki | 91a4547 | 2008-02-07 00:14:29 -0800 | [diff] [blame] | 159 | #endif |
| 160 | |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 161 | static struct zone_reclaim_stat *get_reclaim_stat(struct zone *zone, |
| 162 | struct scan_control *sc) |
| 163 | { |
KAMEZAWA Hiroyuki | e72e2bd | 2009-01-07 18:08:23 -0800 | [diff] [blame] | 164 | if (!scanning_global_lru(sc)) |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 165 | return mem_cgroup_get_reclaim_stat(sc->mem_cgroup, zone); |
| 166 | |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 167 | return &zone->reclaim_stat; |
| 168 | } |
| 169 | |
Vincent Li | 0b21767 | 2009-09-21 17:03:09 -0700 | [diff] [blame] | 170 | static unsigned long zone_nr_lru_pages(struct zone *zone, |
| 171 | struct scan_control *sc, enum lru_list lru) |
KOSAKI Motohiro | c9f299d | 2009-01-07 18:08:16 -0800 | [diff] [blame] | 172 | { |
KAMEZAWA Hiroyuki | e72e2bd | 2009-01-07 18:08:23 -0800 | [diff] [blame] | 173 | if (!scanning_global_lru(sc)) |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 174 | return mem_cgroup_zone_nr_lru_pages(sc->mem_cgroup, |
| 175 | zone_to_nid(zone), zone_idx(zone), BIT(lru)); |
KOSAKI Motohiro | a3d8e05 | 2009-01-07 18:08:19 -0800 | [diff] [blame] | 176 | |
KOSAKI Motohiro | c9f299d | 2009-01-07 18:08:16 -0800 | [diff] [blame] | 177 | return zone_page_state(zone, NR_LRU_BASE + lru); |
| 178 | } |
| 179 | |
| 180 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 181 | /* |
| 182 | * Add a shrinker callback to be called from the vm |
| 183 | */ |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 184 | void register_shrinker(struct shrinker *shrinker) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 185 | { |
Konstantin Khlebnikov | 83aeead | 2011-12-08 14:33:54 -0800 | [diff] [blame] | 186 | atomic_long_set(&shrinker->nr_in_batch, 0); |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 187 | down_write(&shrinker_rwsem); |
| 188 | list_add_tail(&shrinker->list, &shrinker_list); |
| 189 | up_write(&shrinker_rwsem); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 190 | } |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 191 | EXPORT_SYMBOL(register_shrinker); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 192 | |
| 193 | /* |
| 194 | * Remove one |
| 195 | */ |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 196 | void unregister_shrinker(struct shrinker *shrinker) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 197 | { |
| 198 | down_write(&shrinker_rwsem); |
| 199 | list_del(&shrinker->list); |
| 200 | up_write(&shrinker_rwsem); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 201 | } |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 202 | EXPORT_SYMBOL(unregister_shrinker); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 203 | |
Ying Han | 1495f23 | 2011-05-24 17:12:27 -0700 | [diff] [blame] | 204 | static inline int do_shrinker_shrink(struct shrinker *shrinker, |
| 205 | struct shrink_control *sc, |
| 206 | unsigned long nr_to_scan) |
| 207 | { |
| 208 | sc->nr_to_scan = nr_to_scan; |
| 209 | return (*shrinker->shrink)(shrinker, sc); |
| 210 | } |
| 211 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 212 | #define SHRINK_BATCH 128 |
| 213 | /* |
| 214 | * Call the shrink functions to age shrinkable caches |
| 215 | * |
| 216 | * Here we assume it costs one seek to replace a lru page and that it also |
| 217 | * takes a seek to recreate a cache object. With this in mind we age equal |
| 218 | * percentages of the lru and ageable caches. This should balance the seeks |
| 219 | * generated by these structures. |
| 220 | * |
Simon Arlott | 183ff22 | 2007-10-20 01:27:18 +0200 | [diff] [blame] | 221 | * 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] | 222 | * slab to avoid swapping. |
| 223 | * |
| 224 | * We do weird things to avoid (scanned*seeks*entries) overflowing 32 bits. |
| 225 | * |
| 226 | * `lru_pages' represents the number of on-LRU pages in all the zones which |
| 227 | * are eligible for the caller's allocation attempt. It is used for balancing |
| 228 | * slab reclaim versus page reclaim. |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 229 | * |
| 230 | * Returns the number of slab objects which we shrunk. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 231 | */ |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 232 | unsigned long shrink_slab(struct shrink_control *shrink, |
Ying Han | 1495f23 | 2011-05-24 17:12:27 -0700 | [diff] [blame] | 233 | unsigned long nr_pages_scanned, |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 234 | unsigned long lru_pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 235 | { |
| 236 | struct shrinker *shrinker; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 237 | unsigned long ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 238 | |
Ying Han | 1495f23 | 2011-05-24 17:12:27 -0700 | [diff] [blame] | 239 | if (nr_pages_scanned == 0) |
| 240 | nr_pages_scanned = SWAP_CLUSTER_MAX; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 241 | |
Minchan Kim | f06590b | 2011-05-24 17:11:11 -0700 | [diff] [blame] | 242 | if (!down_read_trylock(&shrinker_rwsem)) { |
| 243 | /* Assume we'll be able to shrink next time */ |
| 244 | ret = 1; |
| 245 | goto out; |
| 246 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 247 | |
| 248 | list_for_each_entry(shrinker, &shrinker_list, list) { |
| 249 | unsigned long long delta; |
Konstantin Khlebnikov | 635697c | 2011-12-08 14:33:51 -0800 | [diff] [blame] | 250 | long total_scan; |
| 251 | long max_pass; |
Dave Chinner | 0957607 | 2011-07-08 14:14:34 +1000 | [diff] [blame] | 252 | int shrink_ret = 0; |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 253 | long nr; |
| 254 | long new_nr; |
Dave Chinner | e9299f5 | 2011-07-08 14:14:37 +1000 | [diff] [blame] | 255 | long batch_size = shrinker->batch ? shrinker->batch |
| 256 | : SHRINK_BATCH; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 257 | |
Konstantin Khlebnikov | 635697c | 2011-12-08 14:33:51 -0800 | [diff] [blame] | 258 | max_pass = do_shrinker_shrink(shrinker, shrink, 0); |
| 259 | if (max_pass <= 0) |
| 260 | continue; |
| 261 | |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 262 | /* |
| 263 | * copy the current shrinker scan count into a local variable |
| 264 | * and zero it so that other concurrent shrinker invocations |
| 265 | * don't also do this scanning work. |
| 266 | */ |
Konstantin Khlebnikov | 83aeead | 2011-12-08 14:33:54 -0800 | [diff] [blame] | 267 | nr = atomic_long_xchg(&shrinker->nr_in_batch, 0); |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 268 | |
| 269 | total_scan = nr; |
Ying Han | 1495f23 | 2011-05-24 17:12:27 -0700 | [diff] [blame] | 270 | delta = (4 * nr_pages_scanned) / shrinker->seeks; |
Andrea Arcangeli | ea164d7 | 2005-11-28 13:44:15 -0800 | [diff] [blame] | 271 | delta *= max_pass; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 272 | do_div(delta, lru_pages + 1); |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 273 | total_scan += delta; |
| 274 | if (total_scan < 0) { |
David Rientjes | 88c3bd7 | 2009-03-31 15:23:29 -0700 | [diff] [blame] | 275 | printk(KERN_ERR "shrink_slab: %pF negative objects to " |
| 276 | "delete nr=%ld\n", |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 277 | shrinker->shrink, total_scan); |
| 278 | total_scan = max_pass; |
Andrea Arcangeli | ea164d7 | 2005-11-28 13:44:15 -0800 | [diff] [blame] | 279 | } |
| 280 | |
| 281 | /* |
Dave Chinner | 3567b59 | 2011-07-08 14:14:36 +1000 | [diff] [blame] | 282 | * We need to avoid excessive windup on filesystem shrinkers |
| 283 | * due to large numbers of GFP_NOFS allocations causing the |
| 284 | * shrinkers to return -1 all the time. This results in a large |
| 285 | * nr being built up so when a shrink that can do some work |
| 286 | * comes along it empties the entire cache due to nr >>> |
| 287 | * max_pass. This is bad for sustaining a working set in |
| 288 | * memory. |
| 289 | * |
| 290 | * Hence only allow the shrinker to scan the entire cache when |
| 291 | * a large delta change is calculated directly. |
| 292 | */ |
| 293 | if (delta < max_pass / 4) |
| 294 | total_scan = min(total_scan, max_pass / 2); |
| 295 | |
| 296 | /* |
Andrea Arcangeli | ea164d7 | 2005-11-28 13:44:15 -0800 | [diff] [blame] | 297 | * Avoid risking looping forever due to too large nr value: |
| 298 | * never try to free more than twice the estimate number of |
| 299 | * freeable entries. |
| 300 | */ |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 301 | if (total_scan > max_pass * 2) |
| 302 | total_scan = max_pass * 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 303 | |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 304 | trace_mm_shrink_slab_start(shrinker, shrink, nr, |
Dave Chinner | 0957607 | 2011-07-08 14:14:34 +1000 | [diff] [blame] | 305 | nr_pages_scanned, lru_pages, |
| 306 | max_pass, delta, total_scan); |
| 307 | |
Dave Chinner | e9299f5 | 2011-07-08 14:14:37 +1000 | [diff] [blame] | 308 | while (total_scan >= batch_size) { |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 309 | int nr_before; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 310 | |
Ying Han | 1495f23 | 2011-05-24 17:12:27 -0700 | [diff] [blame] | 311 | nr_before = do_shrinker_shrink(shrinker, shrink, 0); |
| 312 | shrink_ret = do_shrinker_shrink(shrinker, shrink, |
Dave Chinner | e9299f5 | 2011-07-08 14:14:37 +1000 | [diff] [blame] | 313 | batch_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 314 | if (shrink_ret == -1) |
| 315 | break; |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 316 | if (shrink_ret < nr_before) |
| 317 | ret += nr_before - shrink_ret; |
Dave Chinner | e9299f5 | 2011-07-08 14:14:37 +1000 | [diff] [blame] | 318 | count_vm_events(SLABS_SCANNED, batch_size); |
| 319 | total_scan -= batch_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 320 | |
| 321 | cond_resched(); |
| 322 | } |
| 323 | |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 324 | /* |
| 325 | * move the unused scan count back into the shrinker in a |
| 326 | * manner that handles concurrent updates. If we exhausted the |
| 327 | * scan, there is no need to do an update. |
| 328 | */ |
Konstantin Khlebnikov | 83aeead | 2011-12-08 14:33:54 -0800 | [diff] [blame] | 329 | if (total_scan > 0) |
| 330 | new_nr = atomic_long_add_return(total_scan, |
| 331 | &shrinker->nr_in_batch); |
| 332 | else |
| 333 | new_nr = atomic_long_read(&shrinker->nr_in_batch); |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 334 | |
| 335 | trace_mm_shrink_slab_end(shrinker, shrink_ret, nr, new_nr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 336 | } |
| 337 | up_read(&shrinker_rwsem); |
Minchan Kim | f06590b | 2011-05-24 17:11:11 -0700 | [diff] [blame] | 338 | out: |
| 339 | cond_resched(); |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 340 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 341 | } |
| 342 | |
Mel Gorman | f3a310b | 2011-01-13 15:46:00 -0800 | [diff] [blame] | 343 | static void set_reclaim_mode(int priority, struct scan_control *sc, |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 344 | bool sync) |
| 345 | { |
Mel Gorman | f3a310b | 2011-01-13 15:46:00 -0800 | [diff] [blame] | 346 | reclaim_mode_t syncmode = sync ? RECLAIM_MODE_SYNC : RECLAIM_MODE_ASYNC; |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 347 | |
| 348 | /* |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 349 | * Initially assume we are entering either lumpy reclaim or |
| 350 | * reclaim/compaction.Depending on the order, we will either set the |
| 351 | * sync mode or just reclaim order-0 pages later. |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 352 | */ |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 353 | if (COMPACTION_BUILD) |
Mel Gorman | f3a310b | 2011-01-13 15:46:00 -0800 | [diff] [blame] | 354 | sc->reclaim_mode = RECLAIM_MODE_COMPACTION; |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 355 | else |
Mel Gorman | f3a310b | 2011-01-13 15:46:00 -0800 | [diff] [blame] | 356 | sc->reclaim_mode = RECLAIM_MODE_LUMPYRECLAIM; |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 357 | |
| 358 | /* |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 359 | * Avoid using lumpy reclaim or reclaim/compaction if possible by |
| 360 | * restricting when its set to either costly allocations or when |
| 361 | * under memory pressure |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 362 | */ |
| 363 | if (sc->order > PAGE_ALLOC_COSTLY_ORDER) |
Mel Gorman | f3a310b | 2011-01-13 15:46:00 -0800 | [diff] [blame] | 364 | sc->reclaim_mode |= syncmode; |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 365 | else if (sc->order && priority < DEF_PRIORITY - 2) |
Mel Gorman | f3a310b | 2011-01-13 15:46:00 -0800 | [diff] [blame] | 366 | sc->reclaim_mode |= syncmode; |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 367 | else |
Mel Gorman | f3a310b | 2011-01-13 15:46:00 -0800 | [diff] [blame] | 368 | sc->reclaim_mode = RECLAIM_MODE_SINGLE | RECLAIM_MODE_ASYNC; |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 369 | } |
| 370 | |
Mel Gorman | f3a310b | 2011-01-13 15:46:00 -0800 | [diff] [blame] | 371 | static void reset_reclaim_mode(struct scan_control *sc) |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 372 | { |
Mel Gorman | f3a310b | 2011-01-13 15:46:00 -0800 | [diff] [blame] | 373 | sc->reclaim_mode = RECLAIM_MODE_SINGLE | RECLAIM_MODE_ASYNC; |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 374 | } |
| 375 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 376 | static inline int is_page_cache_freeable(struct page *page) |
| 377 | { |
Johannes Weiner | ceddc3a | 2009-09-21 17:03:00 -0700 | [diff] [blame] | 378 | /* |
| 379 | * A freeable page cache page is referenced only by the caller |
| 380 | * that isolated the page, the page cache radix tree and |
| 381 | * optional buffer heads at page->private. |
| 382 | */ |
Johannes Weiner | edcf474 | 2009-09-21 17:02:59 -0700 | [diff] [blame] | 383 | return page_count(page) - page_has_private(page) == 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 384 | } |
| 385 | |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 386 | static int may_write_to_queue(struct backing_dev_info *bdi, |
| 387 | struct scan_control *sc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 388 | { |
Christoph Lameter | 930d915 | 2006-01-08 01:00:47 -0800 | [diff] [blame] | 389 | if (current->flags & PF_SWAPWRITE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 390 | return 1; |
| 391 | if (!bdi_write_congested(bdi)) |
| 392 | return 1; |
| 393 | if (bdi == current->backing_dev_info) |
| 394 | return 1; |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 395 | |
| 396 | /* lumpy reclaim for hugepage often need a lot of write */ |
| 397 | if (sc->order > PAGE_ALLOC_COSTLY_ORDER) |
| 398 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 399 | return 0; |
| 400 | } |
| 401 | |
| 402 | /* |
| 403 | * We detected a synchronous write error writing a page out. Probably |
| 404 | * -ENOSPC. We need to propagate that into the address_space for a subsequent |
| 405 | * fsync(), msync() or close(). |
| 406 | * |
| 407 | * The tricky part is that after writepage we cannot touch the mapping: nothing |
| 408 | * prevents it from being freed up. But we have a ref on the page and once |
| 409 | * that page is locked, the mapping is pinned. |
| 410 | * |
| 411 | * We're allowed to run sleeping lock_page() here because we know the caller has |
| 412 | * __GFP_FS. |
| 413 | */ |
| 414 | static void handle_write_error(struct address_space *mapping, |
| 415 | struct page *page, int error) |
| 416 | { |
Jens Axboe | 7eaceac | 2011-03-10 08:52:07 +0100 | [diff] [blame] | 417 | lock_page(page); |
Guillaume Chazarain | 3e9f45b | 2007-05-08 00:23:25 -0700 | [diff] [blame] | 418 | if (page_mapping(page) == mapping) |
| 419 | mapping_set_error(mapping, error); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 420 | unlock_page(page); |
| 421 | } |
| 422 | |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 423 | /* possible outcome of pageout() */ |
| 424 | typedef enum { |
| 425 | /* failed to write page out, page is locked */ |
| 426 | PAGE_KEEP, |
| 427 | /* move page to the active list, page is locked */ |
| 428 | PAGE_ACTIVATE, |
| 429 | /* page has been sent to the disk successfully, page is unlocked */ |
| 430 | PAGE_SUCCESS, |
| 431 | /* page is clean and locked */ |
| 432 | PAGE_CLEAN, |
| 433 | } pageout_t; |
| 434 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 435 | /* |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 436 | * pageout is called by shrink_page_list() for each dirty page. |
| 437 | * Calls ->writepage(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 438 | */ |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 439 | static pageout_t pageout(struct page *page, struct address_space *mapping, |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 440 | struct scan_control *sc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 441 | { |
| 442 | /* |
| 443 | * If the page is dirty, only perform writeback if that write |
| 444 | * will be non-blocking. To prevent this allocation from being |
| 445 | * stalled by pagecache activity. But note that there may be |
| 446 | * stalls if we need to run get_block(). We could test |
| 447 | * PagePrivate for that. |
| 448 | * |
Vincent Li | 6aceb53 | 2009-12-14 17:58:49 -0800 | [diff] [blame] | 449 | * If this process is currently in __generic_file_aio_write() against |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 450 | * this page's queue, we can perform writeback even if that |
| 451 | * will block. |
| 452 | * |
| 453 | * If the page is swapcache, write it back even if that would |
| 454 | * block, for some throttling. This happens by accident, because |
| 455 | * swap_backing_dev_info is bust: it doesn't reflect the |
| 456 | * congestion state of the swapdevs. Easy to fix, if needed. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 457 | */ |
| 458 | if (!is_page_cache_freeable(page)) |
| 459 | return PAGE_KEEP; |
| 460 | if (!mapping) { |
| 461 | /* |
| 462 | * Some data journaling orphaned pages can have |
| 463 | * page->mapping == NULL while being dirty with clean buffers. |
| 464 | */ |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 465 | if (page_has_private(page)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 466 | if (try_to_free_buffers(page)) { |
| 467 | ClearPageDirty(page); |
Harvey Harrison | d40cee2 | 2008-04-30 00:55:07 -0700 | [diff] [blame] | 468 | printk("%s: orphaned page\n", __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 469 | return PAGE_CLEAN; |
| 470 | } |
| 471 | } |
| 472 | return PAGE_KEEP; |
| 473 | } |
| 474 | if (mapping->a_ops->writepage == NULL) |
| 475 | return PAGE_ACTIVATE; |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 476 | if (!may_write_to_queue(mapping->backing_dev_info, sc)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 477 | return PAGE_KEEP; |
| 478 | |
| 479 | if (clear_page_dirty_for_io(page)) { |
| 480 | int res; |
| 481 | struct writeback_control wbc = { |
| 482 | .sync_mode = WB_SYNC_NONE, |
| 483 | .nr_to_write = SWAP_CLUSTER_MAX, |
OGAWA Hirofumi | 111ebb6 | 2006-06-23 02:03:26 -0700 | [diff] [blame] | 484 | .range_start = 0, |
| 485 | .range_end = LLONG_MAX, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 486 | .for_reclaim = 1, |
| 487 | }; |
| 488 | |
| 489 | SetPageReclaim(page); |
| 490 | res = mapping->a_ops->writepage(page, &wbc); |
| 491 | if (res < 0) |
| 492 | handle_write_error(mapping, page, res); |
Zach Brown | 994fc28c | 2005-12-15 14:28:17 -0800 | [diff] [blame] | 493 | if (res == AOP_WRITEPAGE_ACTIVATE) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 494 | ClearPageReclaim(page); |
| 495 | return PAGE_ACTIVATE; |
| 496 | } |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 497 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 498 | if (!PageWriteback(page)) { |
| 499 | /* synchronous write or broken a_ops? */ |
| 500 | ClearPageReclaim(page); |
| 501 | } |
Mel Gorman | 755f022 | 2010-08-09 17:19:18 -0700 | [diff] [blame] | 502 | trace_mm_vmscan_writepage(page, |
Mel Gorman | f3a310b | 2011-01-13 15:46:00 -0800 | [diff] [blame] | 503 | trace_reclaim_flags(page, sc->reclaim_mode)); |
Andrew Morton | e129b5c | 2006-09-27 01:50:00 -0700 | [diff] [blame] | 504 | inc_zone_page_state(page, NR_VMSCAN_WRITE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 505 | return PAGE_SUCCESS; |
| 506 | } |
| 507 | |
| 508 | return PAGE_CLEAN; |
| 509 | } |
| 510 | |
Andrew Morton | a649fd9 | 2006-10-17 00:09:36 -0700 | [diff] [blame] | 511 | /* |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 512 | * Same as remove_mapping, but if the page is removed from the mapping, it |
| 513 | * gets returned with a refcount of 0. |
Andrew Morton | a649fd9 | 2006-10-17 00:09:36 -0700 | [diff] [blame] | 514 | */ |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 515 | static int __remove_mapping(struct address_space *mapping, struct page *page) |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 516 | { |
Nick Piggin | 28e4d96 | 2006-09-25 23:31:23 -0700 | [diff] [blame] | 517 | BUG_ON(!PageLocked(page)); |
| 518 | BUG_ON(mapping != page_mapping(page)); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 519 | |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 520 | spin_lock_irq(&mapping->tree_lock); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 521 | /* |
Nick Piggin | 0fd0e6b | 2006-09-27 01:50:02 -0700 | [diff] [blame] | 522 | * The non racy check for a busy page. |
| 523 | * |
| 524 | * Must be careful with the order of the tests. When someone has |
| 525 | * a ref to the page, it may be possible that they dirty it then |
| 526 | * drop the reference. So if PageDirty is tested before page_count |
| 527 | * here, then the following race may occur: |
| 528 | * |
| 529 | * get_user_pages(&page); |
| 530 | * [user mapping goes away] |
| 531 | * write_to(page); |
| 532 | * !PageDirty(page) [good] |
| 533 | * SetPageDirty(page); |
| 534 | * put_page(page); |
| 535 | * !page_count(page) [good, discard it] |
| 536 | * |
| 537 | * [oops, our write_to data is lost] |
| 538 | * |
| 539 | * Reversing the order of the tests ensures such a situation cannot |
| 540 | * escape unnoticed. The smp_rmb is needed to ensure the page->flags |
| 541 | * load is not satisfied before that of page->_count. |
| 542 | * |
| 543 | * Note that if SetPageDirty is always performed via set_page_dirty, |
| 544 | * and thus under tree_lock, then this ordering is not required. |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 545 | */ |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 546 | if (!page_freeze_refs(page, 2)) |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 547 | goto cannot_free; |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 548 | /* note: atomic_cmpxchg in page_freeze_refs provides the smp_rmb */ |
| 549 | if (unlikely(PageDirty(page))) { |
| 550 | page_unfreeze_refs(page, 2); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 551 | goto cannot_free; |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 552 | } |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 553 | |
| 554 | if (PageSwapCache(page)) { |
| 555 | swp_entry_t swap = { .val = page_private(page) }; |
| 556 | __delete_from_swap_cache(page); |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 557 | spin_unlock_irq(&mapping->tree_lock); |
KAMEZAWA Hiroyuki | cb4b86b | 2009-06-16 15:32:52 -0700 | [diff] [blame] | 558 | swapcache_free(swap, page); |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 559 | } else { |
Linus Torvalds | 6072d13 | 2010-12-01 13:35:19 -0500 | [diff] [blame] | 560 | void (*freepage)(struct page *); |
| 561 | |
| 562 | freepage = mapping->a_ops->freepage; |
| 563 | |
Minchan Kim | e64a782 | 2011-03-22 16:32:44 -0700 | [diff] [blame] | 564 | __delete_from_page_cache(page); |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 565 | spin_unlock_irq(&mapping->tree_lock); |
Daisuke Nishimura | e767e05 | 2009-05-28 14:34:28 -0700 | [diff] [blame] | 566 | mem_cgroup_uncharge_cache_page(page); |
Linus Torvalds | 6072d13 | 2010-12-01 13:35:19 -0500 | [diff] [blame] | 567 | |
| 568 | if (freepage != NULL) |
| 569 | freepage(page); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 570 | } |
| 571 | |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 572 | return 1; |
| 573 | |
| 574 | cannot_free: |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 575 | spin_unlock_irq(&mapping->tree_lock); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 576 | return 0; |
| 577 | } |
| 578 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 579 | /* |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 580 | * Attempt to detach a locked page from its ->mapping. If it is dirty or if |
| 581 | * someone else has a ref on the page, abort and return 0. If it was |
| 582 | * successfully detached, return 1. Assumes the caller has a single ref on |
| 583 | * this page. |
| 584 | */ |
| 585 | int remove_mapping(struct address_space *mapping, struct page *page) |
| 586 | { |
| 587 | if (__remove_mapping(mapping, page)) { |
| 588 | /* |
| 589 | * Unfreezing the refcount with 1 rather than 2 effectively |
| 590 | * drops the pagecache ref for us without requiring another |
| 591 | * atomic operation. |
| 592 | */ |
| 593 | page_unfreeze_refs(page, 1); |
| 594 | return 1; |
| 595 | } |
| 596 | return 0; |
| 597 | } |
| 598 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 599 | /** |
| 600 | * putback_lru_page - put previously isolated page onto appropriate LRU list |
| 601 | * @page: page to be put back to appropriate lru list |
| 602 | * |
| 603 | * Add previously isolated @page to appropriate LRU list. |
| 604 | * Page may still be unevictable for other reasons. |
| 605 | * |
| 606 | * lru_lock must not be held, interrupts must be enabled. |
| 607 | */ |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 608 | void putback_lru_page(struct page *page) |
| 609 | { |
| 610 | int lru; |
| 611 | int active = !!TestClearPageActive(page); |
Lee Schermerhorn | bbfd28e | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 612 | int was_unevictable = PageUnevictable(page); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 613 | |
| 614 | VM_BUG_ON(PageLRU(page)); |
| 615 | |
| 616 | redo: |
| 617 | ClearPageUnevictable(page); |
| 618 | |
| 619 | if (page_evictable(page, NULL)) { |
| 620 | /* |
| 621 | * For evictable pages, we can use the cache. |
| 622 | * In event of a race, worst case is we end up with an |
| 623 | * unevictable page on [in]active list. |
| 624 | * We know how to handle that. |
| 625 | */ |
Johannes Weiner | 401a8e1 | 2009-09-21 17:02:58 -0700 | [diff] [blame] | 626 | lru = active + page_lru_base_type(page); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 627 | lru_cache_add_lru(page, lru); |
| 628 | } else { |
| 629 | /* |
| 630 | * Put unevictable pages directly on zone's unevictable |
| 631 | * list. |
| 632 | */ |
| 633 | lru = LRU_UNEVICTABLE; |
| 634 | add_page_to_unevictable_list(page); |
Johannes Weiner | 6a7b954 | 2009-10-26 16:50:00 -0700 | [diff] [blame] | 635 | /* |
Minchan Kim | 21ee9f3 | 2011-10-31 17:09:28 -0700 | [diff] [blame] | 636 | * When racing with an mlock or AS_UNEVICTABLE clearing |
| 637 | * (page is unlocked) make sure that if the other thread |
| 638 | * does not observe our setting of PG_lru and fails |
| 639 | * isolation/check_move_unevictable_page, |
| 640 | * we see PG_mlocked/AS_UNEVICTABLE cleared below and move |
Johannes Weiner | 6a7b954 | 2009-10-26 16:50:00 -0700 | [diff] [blame] | 641 | * the page back to the evictable list. |
| 642 | * |
Minchan Kim | 21ee9f3 | 2011-10-31 17:09:28 -0700 | [diff] [blame] | 643 | * The other side is TestClearPageMlocked() or shmem_lock(). |
Johannes Weiner | 6a7b954 | 2009-10-26 16:50:00 -0700 | [diff] [blame] | 644 | */ |
| 645 | smp_mb(); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 646 | } |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 647 | |
| 648 | /* |
| 649 | * page's status can change while we move it among lru. If an evictable |
| 650 | * page is on unevictable list, it never be freed. To avoid that, |
| 651 | * check after we added it to the list, again. |
| 652 | */ |
| 653 | if (lru == LRU_UNEVICTABLE && page_evictable(page, NULL)) { |
| 654 | if (!isolate_lru_page(page)) { |
| 655 | put_page(page); |
| 656 | goto redo; |
| 657 | } |
| 658 | /* This means someone else dropped this page from LRU |
| 659 | * So, it will be freed or putback to LRU again. There is |
| 660 | * nothing to do here. |
| 661 | */ |
| 662 | } |
| 663 | |
Lee Schermerhorn | bbfd28e | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 664 | if (was_unevictable && lru != LRU_UNEVICTABLE) |
| 665 | count_vm_event(UNEVICTABLE_PGRESCUED); |
| 666 | else if (!was_unevictable && lru == LRU_UNEVICTABLE) |
| 667 | count_vm_event(UNEVICTABLE_PGCULLED); |
| 668 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 669 | put_page(page); /* drop ref from isolate */ |
| 670 | } |
| 671 | |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 672 | enum page_references { |
| 673 | PAGEREF_RECLAIM, |
| 674 | PAGEREF_RECLAIM_CLEAN, |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 675 | PAGEREF_KEEP, |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 676 | PAGEREF_ACTIVATE, |
| 677 | }; |
| 678 | |
| 679 | static enum page_references page_check_references(struct page *page, |
| 680 | struct scan_control *sc) |
| 681 | { |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 682 | int referenced_ptes, referenced_page; |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 683 | unsigned long vm_flags; |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 684 | |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 685 | referenced_ptes = page_referenced(page, 1, sc->mem_cgroup, &vm_flags); |
| 686 | referenced_page = TestClearPageReferenced(page); |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 687 | |
| 688 | /* Lumpy reclaim - ignore references */ |
Mel Gorman | f3a310b | 2011-01-13 15:46:00 -0800 | [diff] [blame] | 689 | if (sc->reclaim_mode & RECLAIM_MODE_LUMPYRECLAIM) |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 690 | return PAGEREF_RECLAIM; |
| 691 | |
| 692 | /* |
| 693 | * Mlock lost the isolation race with us. Let try_to_unmap() |
| 694 | * move the page to the unevictable list. |
| 695 | */ |
| 696 | if (vm_flags & VM_LOCKED) |
| 697 | return PAGEREF_RECLAIM; |
| 698 | |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 699 | if (referenced_ptes) { |
| 700 | if (PageAnon(page)) |
| 701 | return PAGEREF_ACTIVATE; |
| 702 | /* |
| 703 | * All mapped pages start out with page table |
| 704 | * references from the instantiating fault, so we need |
| 705 | * to look twice if a mapped file page is used more |
| 706 | * than once. |
| 707 | * |
| 708 | * Mark it and spare it for another trip around the |
| 709 | * inactive list. Another page table reference will |
| 710 | * lead to its activation. |
| 711 | * |
| 712 | * Note: the mark is set for activated pages as well |
| 713 | * so that recently deactivated but used pages are |
| 714 | * quickly recovered. |
| 715 | */ |
| 716 | SetPageReferenced(page); |
| 717 | |
Konstantin Khlebnikov | 34dbc67 | 2012-01-10 15:06:59 -0800 | [diff] [blame] | 718 | if (referenced_page || referenced_ptes > 1) |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 719 | return PAGEREF_ACTIVATE; |
| 720 | |
Konstantin Khlebnikov | c909e99 | 2012-01-10 15:07:03 -0800 | [diff] [blame^] | 721 | /* |
| 722 | * Activate file-backed executable pages after first usage. |
| 723 | */ |
| 724 | if (vm_flags & VM_EXEC) |
| 725 | return PAGEREF_ACTIVATE; |
| 726 | |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 727 | return PAGEREF_KEEP; |
| 728 | } |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 729 | |
| 730 | /* Reclaim if clean, defer dirty pages to writeback */ |
KOSAKI Motohiro | 2e30244 | 2010-10-26 14:21:46 -0700 | [diff] [blame] | 731 | if (referenced_page && !PageSwapBacked(page)) |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 732 | return PAGEREF_RECLAIM_CLEAN; |
| 733 | |
| 734 | return PAGEREF_RECLAIM; |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 735 | } |
| 736 | |
Mel Gorman | abe4c3b | 2010-08-09 17:19:31 -0700 | [diff] [blame] | 737 | static noinline_for_stack void free_page_list(struct list_head *free_pages) |
| 738 | { |
| 739 | struct pagevec freed_pvec; |
| 740 | struct page *page, *tmp; |
| 741 | |
| 742 | pagevec_init(&freed_pvec, 1); |
| 743 | |
| 744 | list_for_each_entry_safe(page, tmp, free_pages, lru) { |
| 745 | list_del(&page->lru); |
| 746 | if (!pagevec_add(&freed_pvec, page)) { |
| 747 | __pagevec_free(&freed_pvec); |
| 748 | pagevec_reinit(&freed_pvec); |
| 749 | } |
| 750 | } |
| 751 | |
| 752 | pagevec_free(&freed_pvec); |
| 753 | } |
| 754 | |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 755 | /* |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 756 | * shrink_page_list() returns the number of reclaimed pages |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 757 | */ |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 758 | static unsigned long shrink_page_list(struct list_head *page_list, |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 759 | struct zone *zone, |
Mel Gorman | f84f6e2 | 2011-10-31 17:07:51 -0700 | [diff] [blame] | 760 | struct scan_control *sc, |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 761 | int priority, |
| 762 | unsigned long *ret_nr_dirty, |
| 763 | unsigned long *ret_nr_writeback) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 764 | { |
| 765 | LIST_HEAD(ret_pages); |
Mel Gorman | abe4c3b | 2010-08-09 17:19:31 -0700 | [diff] [blame] | 766 | LIST_HEAD(free_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 767 | int pgactivate = 0; |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 768 | unsigned long nr_dirty = 0; |
| 769 | unsigned long nr_congested = 0; |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 770 | unsigned long nr_reclaimed = 0; |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 771 | unsigned long nr_writeback = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 772 | |
| 773 | cond_resched(); |
| 774 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 775 | while (!list_empty(page_list)) { |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 776 | enum page_references references; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 777 | struct address_space *mapping; |
| 778 | struct page *page; |
| 779 | int may_enter_fs; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 780 | |
| 781 | cond_resched(); |
| 782 | |
| 783 | page = lru_to_page(page_list); |
| 784 | list_del(&page->lru); |
| 785 | |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 786 | if (!trylock_page(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 787 | goto keep; |
| 788 | |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 789 | VM_BUG_ON(PageActive(page)); |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 790 | VM_BUG_ON(page_zone(page) != zone); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 791 | |
| 792 | sc->nr_scanned++; |
Christoph Lameter | 80e4342 | 2006-02-11 17:55:53 -0800 | [diff] [blame] | 793 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 794 | if (unlikely(!page_evictable(page, NULL))) |
| 795 | goto cull_mlocked; |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 796 | |
Johannes Weiner | a6dc60f8 | 2009-03-31 15:19:30 -0700 | [diff] [blame] | 797 | if (!sc->may_unmap && page_mapped(page)) |
Christoph Lameter | 80e4342 | 2006-02-11 17:55:53 -0800 | [diff] [blame] | 798 | goto keep_locked; |
| 799 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 800 | /* Double the slab pressure for mapped and swapcache pages */ |
| 801 | if (page_mapped(page) || PageSwapCache(page)) |
| 802 | sc->nr_scanned++; |
| 803 | |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 804 | may_enter_fs = (sc->gfp_mask & __GFP_FS) || |
| 805 | (PageSwapCache(page) && (sc->gfp_mask & __GFP_IO)); |
| 806 | |
| 807 | if (PageWriteback(page)) { |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 808 | nr_writeback++; |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 809 | /* |
Mel Gorman | a18bba0 | 2011-10-31 17:07:42 -0700 | [diff] [blame] | 810 | * Synchronous reclaim cannot queue pages for |
| 811 | * writeback due to the possibility of stack overflow |
| 812 | * but if it encounters a page under writeback, wait |
| 813 | * for the IO to complete. |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 814 | */ |
Mel Gorman | f3a310b | 2011-01-13 15:46:00 -0800 | [diff] [blame] | 815 | if ((sc->reclaim_mode & RECLAIM_MODE_SYNC) && |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 816 | may_enter_fs) |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 817 | wait_on_page_writeback(page); |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 818 | else { |
| 819 | unlock_page(page); |
| 820 | goto keep_lumpy; |
| 821 | } |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 822 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 823 | |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 824 | references = page_check_references(page, sc); |
| 825 | switch (references) { |
| 826 | case PAGEREF_ACTIVATE: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 827 | goto activate_locked; |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 828 | case PAGEREF_KEEP: |
| 829 | goto keep_locked; |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 830 | case PAGEREF_RECLAIM: |
| 831 | case PAGEREF_RECLAIM_CLEAN: |
| 832 | ; /* try to reclaim the page below */ |
| 833 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 834 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 835 | /* |
| 836 | * Anonymous process memory has backing store? |
| 837 | * Try to allocate it some swap space here. |
| 838 | */ |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 839 | if (PageAnon(page) && !PageSwapCache(page)) { |
Hugh Dickins | 63eb6b9 | 2008-11-19 15:36:37 -0800 | [diff] [blame] | 840 | if (!(sc->gfp_mask & __GFP_IO)) |
| 841 | goto keep_locked; |
Hugh Dickins | ac47b00 | 2009-01-06 14:39:39 -0800 | [diff] [blame] | 842 | if (!add_to_swap(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 843 | goto activate_locked; |
Hugh Dickins | 63eb6b9 | 2008-11-19 15:36:37 -0800 | [diff] [blame] | 844 | may_enter_fs = 1; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 845 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 846 | |
| 847 | mapping = page_mapping(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 848 | |
| 849 | /* |
| 850 | * The page is mapped into the page tables of one or more |
| 851 | * processes. Try to unmap it here. |
| 852 | */ |
| 853 | if (page_mapped(page) && mapping) { |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 854 | switch (try_to_unmap(page, TTU_UNMAP)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 855 | case SWAP_FAIL: |
| 856 | goto activate_locked; |
| 857 | case SWAP_AGAIN: |
| 858 | goto keep_locked; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 859 | case SWAP_MLOCK: |
| 860 | goto cull_mlocked; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 861 | case SWAP_SUCCESS: |
| 862 | ; /* try to free the page below */ |
| 863 | } |
| 864 | } |
| 865 | |
| 866 | if (PageDirty(page)) { |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 867 | nr_dirty++; |
| 868 | |
Mel Gorman | ee72886 | 2011-10-31 17:07:38 -0700 | [diff] [blame] | 869 | /* |
| 870 | * Only kswapd can writeback filesystem pages to |
Mel Gorman | f84f6e2 | 2011-10-31 17:07:51 -0700 | [diff] [blame] | 871 | * avoid risk of stack overflow but do not writeback |
| 872 | * unless under significant pressure. |
Mel Gorman | ee72886 | 2011-10-31 17:07:38 -0700 | [diff] [blame] | 873 | */ |
Mel Gorman | f84f6e2 | 2011-10-31 17:07:51 -0700 | [diff] [blame] | 874 | if (page_is_file_cache(page) && |
| 875 | (!current_is_kswapd() || priority >= DEF_PRIORITY - 2)) { |
Mel Gorman | 49ea7eb | 2011-10-31 17:07:59 -0700 | [diff] [blame] | 876 | /* |
| 877 | * Immediately reclaim when written back. |
| 878 | * Similar in principal to deactivate_page() |
| 879 | * except we already have the page isolated |
| 880 | * and know it's dirty |
| 881 | */ |
| 882 | inc_zone_page_state(page, NR_VMSCAN_IMMEDIATE); |
| 883 | SetPageReclaim(page); |
| 884 | |
Mel Gorman | ee72886 | 2011-10-31 17:07:38 -0700 | [diff] [blame] | 885 | goto keep_locked; |
| 886 | } |
| 887 | |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 888 | if (references == PAGEREF_RECLAIM_CLEAN) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 889 | goto keep_locked; |
Andrew Morton | 4dd4b92 | 2008-03-24 12:29:52 -0700 | [diff] [blame] | 890 | if (!may_enter_fs) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 891 | goto keep_locked; |
Christoph Lameter | 52a8363 | 2006-02-01 03:05:28 -0800 | [diff] [blame] | 892 | if (!sc->may_writepage) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 893 | goto keep_locked; |
| 894 | |
| 895 | /* Page is dirty, try to write it out here */ |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 896 | switch (pageout(page, mapping, sc)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 897 | case PAGE_KEEP: |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 898 | nr_congested++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 899 | goto keep_locked; |
| 900 | case PAGE_ACTIVATE: |
| 901 | goto activate_locked; |
| 902 | case PAGE_SUCCESS: |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 903 | if (PageWriteback(page)) |
| 904 | goto keep_lumpy; |
| 905 | if (PageDirty(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 906 | goto keep; |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 907 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 908 | /* |
| 909 | * A synchronous write - probably a ramdisk. Go |
| 910 | * ahead and try to reclaim the page. |
| 911 | */ |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 912 | if (!trylock_page(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 913 | goto keep; |
| 914 | if (PageDirty(page) || PageWriteback(page)) |
| 915 | goto keep_locked; |
| 916 | mapping = page_mapping(page); |
| 917 | case PAGE_CLEAN: |
| 918 | ; /* try to free the page below */ |
| 919 | } |
| 920 | } |
| 921 | |
| 922 | /* |
| 923 | * If the page has buffers, try to free the buffer mappings |
| 924 | * associated with this page. If we succeed we try to free |
| 925 | * the page as well. |
| 926 | * |
| 927 | * We do this even if the page is PageDirty(). |
| 928 | * try_to_release_page() does not perform I/O, but it is |
| 929 | * possible for a page to have PageDirty set, but it is actually |
| 930 | * clean (all its buffers are clean). This happens if the |
| 931 | * buffers were written out directly, with submit_bh(). ext3 |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 932 | * will do this, as well as the blockdev mapping. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 933 | * try_to_release_page() will discover that cleanness and will |
| 934 | * drop the buffers and mark the page clean - it can be freed. |
| 935 | * |
| 936 | * Rarely, pages can have buffers and no ->mapping. These are |
| 937 | * the pages which were not successfully invalidated in |
| 938 | * truncate_complete_page(). We try to drop those buffers here |
| 939 | * and if that worked, and the page is no longer mapped into |
| 940 | * process address space (page_count == 1) it can be freed. |
| 941 | * Otherwise, leave the page on the LRU so it is swappable. |
| 942 | */ |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 943 | if (page_has_private(page)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 944 | if (!try_to_release_page(page, sc->gfp_mask)) |
| 945 | goto activate_locked; |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 946 | if (!mapping && page_count(page) == 1) { |
| 947 | unlock_page(page); |
| 948 | if (put_page_testzero(page)) |
| 949 | goto free_it; |
| 950 | else { |
| 951 | /* |
| 952 | * rare race with speculative reference. |
| 953 | * the speculative reference will free |
| 954 | * this page shortly, so we may |
| 955 | * increment nr_reclaimed here (and |
| 956 | * leave it off the LRU). |
| 957 | */ |
| 958 | nr_reclaimed++; |
| 959 | continue; |
| 960 | } |
| 961 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 962 | } |
| 963 | |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 964 | if (!mapping || !__remove_mapping(mapping, page)) |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 965 | goto keep_locked; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 966 | |
Nick Piggin | a978d6f | 2008-10-18 20:26:58 -0700 | [diff] [blame] | 967 | /* |
| 968 | * At this point, we have no other references and there is |
| 969 | * no way to pick any more up (removed from LRU, removed |
| 970 | * from pagecache). Can use non-atomic bitops now (and |
| 971 | * we obviously don't have to worry about waking up a process |
| 972 | * waiting on the page lock, because there are no references. |
| 973 | */ |
| 974 | __clear_page_locked(page); |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 975 | free_it: |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 976 | nr_reclaimed++; |
Mel Gorman | abe4c3b | 2010-08-09 17:19:31 -0700 | [diff] [blame] | 977 | |
| 978 | /* |
| 979 | * Is there need to periodically free_page_list? It would |
| 980 | * appear not as the counts should be low |
| 981 | */ |
| 982 | list_add(&page->lru, &free_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 983 | continue; |
| 984 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 985 | cull_mlocked: |
Hugh Dickins | 63d6c5a | 2009-01-06 14:39:38 -0800 | [diff] [blame] | 986 | if (PageSwapCache(page)) |
| 987 | try_to_free_swap(page); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 988 | unlock_page(page); |
| 989 | putback_lru_page(page); |
Mel Gorman | f3a310b | 2011-01-13 15:46:00 -0800 | [diff] [blame] | 990 | reset_reclaim_mode(sc); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 991 | continue; |
| 992 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 993 | activate_locked: |
Rik van Riel | 68a22394 | 2008-10-18 20:26:23 -0700 | [diff] [blame] | 994 | /* Not a candidate for swapping, so reclaim swap space. */ |
| 995 | if (PageSwapCache(page) && vm_swap_full()) |
Hugh Dickins | a2c43ee | 2009-01-06 14:39:36 -0800 | [diff] [blame] | 996 | try_to_free_swap(page); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 997 | VM_BUG_ON(PageActive(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 998 | SetPageActive(page); |
| 999 | pgactivate++; |
| 1000 | keep_locked: |
| 1001 | unlock_page(page); |
| 1002 | keep: |
Mel Gorman | f3a310b | 2011-01-13 15:46:00 -0800 | [diff] [blame] | 1003 | reset_reclaim_mode(sc); |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 1004 | keep_lumpy: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1005 | list_add(&page->lru, &ret_pages); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1006 | VM_BUG_ON(PageLRU(page) || PageUnevictable(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1007 | } |
Mel Gorman | abe4c3b | 2010-08-09 17:19:31 -0700 | [diff] [blame] | 1008 | |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 1009 | /* |
| 1010 | * Tag a zone as congested if all the dirty pages encountered were |
| 1011 | * backed by a congested BDI. In this case, reclaimers should just |
| 1012 | * back off and wait for congestion to clear because further reclaim |
| 1013 | * will encounter the same problem |
| 1014 | */ |
KAMEZAWA Hiroyuki | d6c438b | 2011-05-17 15:44:10 -0700 | [diff] [blame] | 1015 | if (nr_dirty && nr_dirty == nr_congested && scanning_global_lru(sc)) |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 1016 | zone_set_flag(zone, ZONE_CONGESTED); |
| 1017 | |
Mel Gorman | abe4c3b | 2010-08-09 17:19:31 -0700 | [diff] [blame] | 1018 | free_page_list(&free_pages); |
| 1019 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1020 | list_splice(&ret_pages, page_list); |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 1021 | count_vm_events(PGACTIVATE, pgactivate); |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 1022 | *ret_nr_dirty += nr_dirty; |
| 1023 | *ret_nr_writeback += nr_writeback; |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 1024 | return nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1025 | } |
| 1026 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1027 | /* |
| 1028 | * Attempt to remove the specified page from its LRU. Only take this page |
| 1029 | * if it is of the appropriate PageActive status. Pages which are being |
| 1030 | * freed elsewhere are also ignored. |
| 1031 | * |
| 1032 | * page: page to consider |
| 1033 | * mode: one of the LRU isolation modes defined above |
| 1034 | * |
| 1035 | * returns 0 on success, -ve errno on failure. |
| 1036 | */ |
Minchan Kim | 4356f21 | 2011-10-31 17:06:47 -0700 | [diff] [blame] | 1037 | int __isolate_lru_page(struct page *page, isolate_mode_t mode, int file) |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1038 | { |
Minchan Kim | 4356f21 | 2011-10-31 17:06:47 -0700 | [diff] [blame] | 1039 | bool all_lru_mode; |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1040 | int ret = -EINVAL; |
| 1041 | |
| 1042 | /* Only take pages on the LRU. */ |
| 1043 | if (!PageLRU(page)) |
| 1044 | return ret; |
| 1045 | |
Minchan Kim | 4356f21 | 2011-10-31 17:06:47 -0700 | [diff] [blame] | 1046 | all_lru_mode = (mode & (ISOLATE_ACTIVE|ISOLATE_INACTIVE)) == |
| 1047 | (ISOLATE_ACTIVE|ISOLATE_INACTIVE); |
| 1048 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1049 | /* |
| 1050 | * When checking the active state, we need to be sure we are |
| 1051 | * dealing with comparible boolean values. Take the logical not |
| 1052 | * of each. |
| 1053 | */ |
Minchan Kim | 4356f21 | 2011-10-31 17:06:47 -0700 | [diff] [blame] | 1054 | if (!all_lru_mode && !PageActive(page) != !(mode & ISOLATE_ACTIVE)) |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1055 | return ret; |
| 1056 | |
Minchan Kim | 4356f21 | 2011-10-31 17:06:47 -0700 | [diff] [blame] | 1057 | if (!all_lru_mode && !!page_is_file_cache(page) != file) |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1058 | return ret; |
| 1059 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1060 | /* |
| 1061 | * When this function is being called for lumpy reclaim, we |
| 1062 | * initially look into all LRU pages, active, inactive and |
| 1063 | * unevictable; only give shrink_page_list evictable pages. |
| 1064 | */ |
| 1065 | if (PageUnevictable(page)) |
| 1066 | return ret; |
| 1067 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1068 | ret = -EBUSY; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 1069 | |
Minchan Kim | 39deaf8 | 2011-10-31 17:06:51 -0700 | [diff] [blame] | 1070 | if ((mode & ISOLATE_CLEAN) && (PageDirty(page) || PageWriteback(page))) |
| 1071 | return ret; |
| 1072 | |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1073 | if ((mode & ISOLATE_UNMAPPED) && page_mapped(page)) |
| 1074 | return ret; |
| 1075 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1076 | if (likely(get_page_unless_zero(page))) { |
| 1077 | /* |
| 1078 | * Be careful not to clear PageLRU until after we're |
| 1079 | * sure the page is not being freed elsewhere -- the |
| 1080 | * page release code relies on it. |
| 1081 | */ |
| 1082 | ClearPageLRU(page); |
| 1083 | ret = 0; |
| 1084 | } |
| 1085 | |
| 1086 | return ret; |
| 1087 | } |
| 1088 | |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 1089 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1090 | * zone->lru_lock is heavily contended. Some of the functions that |
| 1091 | * shrink the lists perform better by taking out a batch of pages |
| 1092 | * and working on them outside the LRU lock. |
| 1093 | * |
| 1094 | * For pagecache intensive workloads, this function is the hottest |
| 1095 | * spot in the kernel (apart from copy_*_user functions). |
| 1096 | * |
| 1097 | * Appropriate locks must be held before calling this function. |
| 1098 | * |
| 1099 | * @nr_to_scan: The number of pages to look through on the list. |
| 1100 | * @src: The LRU list to pull pages off. |
| 1101 | * @dst: The temp list to put pages on to. |
| 1102 | * @scanned: The number of pages that were scanned. |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1103 | * @order: The caller's attempted allocation order |
| 1104 | * @mode: One of the LRU isolation modes |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1105 | * @file: True [1] if isolating file [!anon] pages |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1106 | * |
| 1107 | * returns how many pages were moved onto *@dst. |
| 1108 | */ |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1109 | static unsigned long isolate_lru_pages(unsigned long nr_to_scan, |
| 1110 | struct list_head *src, struct list_head *dst, |
Minchan Kim | 4356f21 | 2011-10-31 17:06:47 -0700 | [diff] [blame] | 1111 | unsigned long *scanned, int order, isolate_mode_t mode, |
| 1112 | int file) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1113 | { |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1114 | unsigned long nr_taken = 0; |
Mel Gorman | a8a94d1 | 2010-08-09 17:19:17 -0700 | [diff] [blame] | 1115 | unsigned long nr_lumpy_taken = 0; |
| 1116 | unsigned long nr_lumpy_dirty = 0; |
| 1117 | unsigned long nr_lumpy_failed = 0; |
Wu Fengguang | c9b02d9 | 2006-03-22 00:08:23 -0800 | [diff] [blame] | 1118 | unsigned long scan; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1119 | |
Wu Fengguang | c9b02d9 | 2006-03-22 00:08:23 -0800 | [diff] [blame] | 1120 | for (scan = 0; scan < nr_to_scan && !list_empty(src); scan++) { |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1121 | struct page *page; |
| 1122 | unsigned long pfn; |
| 1123 | unsigned long end_pfn; |
| 1124 | unsigned long page_pfn; |
| 1125 | int zone_id; |
| 1126 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1127 | page = lru_to_page(src); |
| 1128 | prefetchw_prev_lru_page(page, src, flags); |
| 1129 | |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 1130 | VM_BUG_ON(!PageLRU(page)); |
Nick Piggin | 8d438f9 | 2006-03-22 00:07:59 -0800 | [diff] [blame] | 1131 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1132 | switch (__isolate_lru_page(page, mode, file)) { |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1133 | case 0: |
| 1134 | list_move(&page->lru, dst); |
KAMEZAWA Hiroyuki | 2ffebca | 2009-06-17 16:27:21 -0700 | [diff] [blame] | 1135 | mem_cgroup_del_lru(page); |
Rik van Riel | 2c888cf | 2011-01-13 15:47:13 -0800 | [diff] [blame] | 1136 | nr_taken += hpage_nr_pages(page); |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1137 | break; |
Nick Piggin | 46453a6 | 2006-03-22 00:07:58 -0800 | [diff] [blame] | 1138 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1139 | case -EBUSY: |
| 1140 | /* else it is being freed elsewhere */ |
| 1141 | list_move(&page->lru, src); |
KAMEZAWA Hiroyuki | 2ffebca | 2009-06-17 16:27:21 -0700 | [diff] [blame] | 1142 | mem_cgroup_rotate_lru_list(page, page_lru(page)); |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1143 | continue; |
| 1144 | |
| 1145 | default: |
| 1146 | BUG(); |
| 1147 | } |
| 1148 | |
| 1149 | if (!order) |
| 1150 | continue; |
| 1151 | |
| 1152 | /* |
| 1153 | * Attempt to take all pages in the order aligned region |
| 1154 | * surrounding the tag page. Only take those pages of |
| 1155 | * the same active state as that tag page. We may safely |
| 1156 | * round the target page pfn down to the requested order |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 1157 | * as the mem_map is guaranteed valid out to MAX_ORDER, |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1158 | * where that page is in a different zone we will detect |
| 1159 | * it from its zone id and abort this block scan. |
| 1160 | */ |
| 1161 | zone_id = page_zone_id(page); |
| 1162 | page_pfn = page_to_pfn(page); |
| 1163 | pfn = page_pfn & ~((1 << order) - 1); |
| 1164 | end_pfn = pfn + (1 << order); |
| 1165 | for (; pfn < end_pfn; pfn++) { |
| 1166 | struct page *cursor_page; |
| 1167 | |
| 1168 | /* The target page is in the block, ignore it. */ |
| 1169 | if (unlikely(pfn == page_pfn)) |
| 1170 | continue; |
| 1171 | |
| 1172 | /* Avoid holes within the zone. */ |
| 1173 | if (unlikely(!pfn_valid_within(pfn))) |
| 1174 | break; |
| 1175 | |
| 1176 | cursor_page = pfn_to_page(pfn); |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1177 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1178 | /* Check that we have not crossed a zone boundary. */ |
| 1179 | if (unlikely(page_zone_id(cursor_page) != zone_id)) |
KOSAKI Motohiro | 08fc468 | 2010-10-26 14:21:44 -0700 | [diff] [blame] | 1180 | break; |
Minchan Kim | de2e756 | 2009-09-21 17:01:43 -0700 | [diff] [blame] | 1181 | |
| 1182 | /* |
| 1183 | * If we don't have enough swap space, reclaiming of |
| 1184 | * anon page which don't already have a swap slot is |
| 1185 | * pointless. |
| 1186 | */ |
| 1187 | if (nr_swap_pages <= 0 && PageAnon(cursor_page) && |
KOSAKI Motohiro | 08fc468 | 2010-10-26 14:21:44 -0700 | [diff] [blame] | 1188 | !PageSwapCache(cursor_page)) |
| 1189 | break; |
Minchan Kim | de2e756 | 2009-09-21 17:01:43 -0700 | [diff] [blame] | 1190 | |
KAMEZAWA Hiroyuki | ee993b1 | 2009-06-16 15:33:24 -0700 | [diff] [blame] | 1191 | if (__isolate_lru_page(cursor_page, mode, file) == 0) { |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1192 | list_move(&cursor_page->lru, dst); |
KAMEZAWA Hiroyuki | cb4cbcf | 2009-06-23 08:57:55 +0900 | [diff] [blame] | 1193 | mem_cgroup_del_lru(cursor_page); |
Rik van Riel | 2c888cf | 2011-01-13 15:47:13 -0800 | [diff] [blame] | 1194 | nr_taken += hpage_nr_pages(page); |
Mel Gorman | a8a94d1 | 2010-08-09 17:19:17 -0700 | [diff] [blame] | 1195 | nr_lumpy_taken++; |
| 1196 | if (PageDirty(cursor_page)) |
| 1197 | nr_lumpy_dirty++; |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1198 | scan++; |
Mel Gorman | a8a94d1 | 2010-08-09 17:19:17 -0700 | [diff] [blame] | 1199 | } else { |
Andrea Arcangeli | d179e84 | 2011-06-15 15:08:51 -0700 | [diff] [blame] | 1200 | /* |
| 1201 | * Check if the page is freed already. |
| 1202 | * |
| 1203 | * We can't use page_count() as that |
| 1204 | * requires compound_head and we don't |
| 1205 | * have a pin on the page here. If a |
| 1206 | * page is tail, we may or may not |
| 1207 | * have isolated the head, so assume |
| 1208 | * it's not free, it'd be tricky to |
| 1209 | * track the head status without a |
| 1210 | * page pin. |
| 1211 | */ |
| 1212 | if (!PageTail(cursor_page) && |
| 1213 | !atomic_read(&cursor_page->_count)) |
KOSAKI Motohiro | 08fc468 | 2010-10-26 14:21:44 -0700 | [diff] [blame] | 1214 | continue; |
| 1215 | break; |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1216 | } |
| 1217 | } |
KOSAKI Motohiro | 08fc468 | 2010-10-26 14:21:44 -0700 | [diff] [blame] | 1218 | |
| 1219 | /* If we break out of the loop above, lumpy reclaim failed */ |
| 1220 | if (pfn < end_pfn) |
| 1221 | nr_lumpy_failed++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1222 | } |
| 1223 | |
| 1224 | *scanned = scan; |
Mel Gorman | a8a94d1 | 2010-08-09 17:19:17 -0700 | [diff] [blame] | 1225 | |
| 1226 | trace_mm_vmscan_lru_isolate(order, |
| 1227 | nr_to_scan, scan, |
| 1228 | nr_taken, |
| 1229 | nr_lumpy_taken, nr_lumpy_dirty, nr_lumpy_failed, |
| 1230 | mode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1231 | return nr_taken; |
| 1232 | } |
| 1233 | |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1234 | static unsigned long isolate_pages_global(unsigned long nr, |
| 1235 | struct list_head *dst, |
| 1236 | unsigned long *scanned, int order, |
Minchan Kim | 4356f21 | 2011-10-31 17:06:47 -0700 | [diff] [blame] | 1237 | isolate_mode_t mode, |
| 1238 | struct zone *z, int active, int file) |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1239 | { |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1240 | int lru = LRU_BASE; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1241 | if (active) |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1242 | lru += LRU_ACTIVE; |
| 1243 | if (file) |
| 1244 | lru += LRU_FILE; |
| 1245 | return isolate_lru_pages(nr, &z->lru[lru].list, dst, scanned, order, |
Johannes Weiner | b7c46d1 | 2009-09-21 17:02:56 -0700 | [diff] [blame] | 1246 | mode, file); |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1247 | } |
| 1248 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1249 | /* |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1250 | * clear_active_flags() is a helper for shrink_active_list(), clearing |
| 1251 | * any active bits from the pages in the list. |
| 1252 | */ |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1253 | static unsigned long clear_active_flags(struct list_head *page_list, |
| 1254 | unsigned int *count) |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1255 | { |
| 1256 | int nr_active = 0; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1257 | int lru; |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1258 | struct page *page; |
| 1259 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1260 | list_for_each_entry(page, page_list, lru) { |
Rik van Riel | 2c888cf | 2011-01-13 15:47:13 -0800 | [diff] [blame] | 1261 | int numpages = hpage_nr_pages(page); |
Johannes Weiner | 401a8e1 | 2009-09-21 17:02:58 -0700 | [diff] [blame] | 1262 | lru = page_lru_base_type(page); |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1263 | if (PageActive(page)) { |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1264 | lru += LRU_ACTIVE; |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1265 | ClearPageActive(page); |
Rik van Riel | 2c888cf | 2011-01-13 15:47:13 -0800 | [diff] [blame] | 1266 | nr_active += numpages; |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1267 | } |
Mel Gorman | 1489fa1 | 2010-08-09 17:19:33 -0700 | [diff] [blame] | 1268 | if (count) |
Rik van Riel | 2c888cf | 2011-01-13 15:47:13 -0800 | [diff] [blame] | 1269 | count[lru] += numpages; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1270 | } |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1271 | |
| 1272 | return nr_active; |
| 1273 | } |
| 1274 | |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1275 | /** |
| 1276 | * isolate_lru_page - tries to isolate a page from its LRU list |
| 1277 | * @page: page to isolate from its LRU list |
| 1278 | * |
| 1279 | * Isolates a @page from an LRU list, clears PageLRU and adjusts the |
| 1280 | * vmstat statistic corresponding to whatever LRU list the page was on. |
| 1281 | * |
| 1282 | * Returns 0 if the page was removed from an LRU list. |
| 1283 | * Returns -EBUSY if the page was not on an LRU list. |
| 1284 | * |
| 1285 | * The returned page will have PageLRU() cleared. If it was found on |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1286 | * the active list, it will have PageActive set. If it was found on |
| 1287 | * the unevictable list, it will have the PageUnevictable bit set. That flag |
| 1288 | * 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] | 1289 | * |
| 1290 | * The vmstat statistic corresponding to the list on which the page was |
| 1291 | * found will be decremented. |
| 1292 | * |
| 1293 | * Restrictions: |
| 1294 | * (1) Must be called with an elevated refcount on the page. This is a |
| 1295 | * fundamentnal difference from isolate_lru_pages (which is called |
| 1296 | * without a stable reference). |
| 1297 | * (2) the lru_lock must not be held. |
| 1298 | * (3) interrupts must be enabled. |
| 1299 | */ |
| 1300 | int isolate_lru_page(struct page *page) |
| 1301 | { |
| 1302 | int ret = -EBUSY; |
| 1303 | |
Konstantin Khlebnikov | 0c91731 | 2011-05-24 17:12:21 -0700 | [diff] [blame] | 1304 | VM_BUG_ON(!page_count(page)); |
| 1305 | |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1306 | if (PageLRU(page)) { |
| 1307 | struct zone *zone = page_zone(page); |
| 1308 | |
| 1309 | spin_lock_irq(&zone->lru_lock); |
Konstantin Khlebnikov | 0c91731 | 2011-05-24 17:12:21 -0700 | [diff] [blame] | 1310 | if (PageLRU(page)) { |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1311 | int lru = page_lru(page); |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1312 | ret = 0; |
Konstantin Khlebnikov | 0c91731 | 2011-05-24 17:12:21 -0700 | [diff] [blame] | 1313 | get_page(page); |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1314 | ClearPageLRU(page); |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1315 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1316 | del_page_from_lru_list(zone, page, lru); |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1317 | } |
| 1318 | spin_unlock_irq(&zone->lru_lock); |
| 1319 | } |
| 1320 | return ret; |
| 1321 | } |
| 1322 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1323 | /* |
Rik van Riel | 35cd781 | 2009-09-21 17:01:38 -0700 | [diff] [blame] | 1324 | * Are there way too many processes in the direct reclaim path already? |
| 1325 | */ |
| 1326 | static int too_many_isolated(struct zone *zone, int file, |
| 1327 | struct scan_control *sc) |
| 1328 | { |
| 1329 | unsigned long inactive, isolated; |
| 1330 | |
| 1331 | if (current_is_kswapd()) |
| 1332 | return 0; |
| 1333 | |
| 1334 | if (!scanning_global_lru(sc)) |
| 1335 | return 0; |
| 1336 | |
| 1337 | if (file) { |
| 1338 | inactive = zone_page_state(zone, NR_INACTIVE_FILE); |
| 1339 | isolated = zone_page_state(zone, NR_ISOLATED_FILE); |
| 1340 | } else { |
| 1341 | inactive = zone_page_state(zone, NR_INACTIVE_ANON); |
| 1342 | isolated = zone_page_state(zone, NR_ISOLATED_ANON); |
| 1343 | } |
| 1344 | |
| 1345 | return isolated > inactive; |
| 1346 | } |
| 1347 | |
| 1348 | /* |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1349 | * TODO: Try merging with migrations version of putback_lru_pages |
| 1350 | */ |
| 1351 | static noinline_for_stack void |
Mel Gorman | 1489fa1 | 2010-08-09 17:19:33 -0700 | [diff] [blame] | 1352 | putback_lru_pages(struct zone *zone, struct scan_control *sc, |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1353 | unsigned long nr_anon, unsigned long nr_file, |
| 1354 | struct list_head *page_list) |
| 1355 | { |
| 1356 | struct page *page; |
| 1357 | struct pagevec pvec; |
Mel Gorman | 1489fa1 | 2010-08-09 17:19:33 -0700 | [diff] [blame] | 1358 | struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc); |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1359 | |
| 1360 | pagevec_init(&pvec, 1); |
| 1361 | |
| 1362 | /* |
| 1363 | * Put back any unfreeable pages. |
| 1364 | */ |
| 1365 | spin_lock(&zone->lru_lock); |
| 1366 | while (!list_empty(page_list)) { |
| 1367 | int lru; |
| 1368 | page = lru_to_page(page_list); |
| 1369 | VM_BUG_ON(PageLRU(page)); |
| 1370 | list_del(&page->lru); |
| 1371 | if (unlikely(!page_evictable(page, NULL))) { |
| 1372 | spin_unlock_irq(&zone->lru_lock); |
| 1373 | putback_lru_page(page); |
| 1374 | spin_lock_irq(&zone->lru_lock); |
| 1375 | continue; |
| 1376 | } |
Linus Torvalds | 7a60857 | 2011-01-17 14:42:19 -0800 | [diff] [blame] | 1377 | SetPageLRU(page); |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1378 | lru = page_lru(page); |
Linus Torvalds | 7a60857 | 2011-01-17 14:42:19 -0800 | [diff] [blame] | 1379 | add_page_to_lru_list(zone, page, lru); |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1380 | if (is_active_lru(lru)) { |
| 1381 | int file = is_file_lru(lru); |
Rik van Riel | 9992af1 | 2011-01-13 15:47:13 -0800 | [diff] [blame] | 1382 | int numpages = hpage_nr_pages(page); |
| 1383 | reclaim_stat->recent_rotated[file] += numpages; |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1384 | } |
| 1385 | if (!pagevec_add(&pvec, page)) { |
| 1386 | spin_unlock_irq(&zone->lru_lock); |
| 1387 | __pagevec_release(&pvec); |
| 1388 | spin_lock_irq(&zone->lru_lock); |
| 1389 | } |
| 1390 | } |
| 1391 | __mod_zone_page_state(zone, NR_ISOLATED_ANON, -nr_anon); |
| 1392 | __mod_zone_page_state(zone, NR_ISOLATED_FILE, -nr_file); |
| 1393 | |
| 1394 | spin_unlock_irq(&zone->lru_lock); |
| 1395 | pagevec_release(&pvec); |
| 1396 | } |
| 1397 | |
Mel Gorman | 1489fa1 | 2010-08-09 17:19:33 -0700 | [diff] [blame] | 1398 | static noinline_for_stack void update_isolated_counts(struct zone *zone, |
| 1399 | struct scan_control *sc, |
| 1400 | unsigned long *nr_anon, |
| 1401 | unsigned long *nr_file, |
| 1402 | struct list_head *isolated_list) |
| 1403 | { |
| 1404 | unsigned long nr_active; |
| 1405 | unsigned int count[NR_LRU_LISTS] = { 0, }; |
| 1406 | struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc); |
| 1407 | |
| 1408 | nr_active = clear_active_flags(isolated_list, count); |
| 1409 | __count_vm_events(PGDEACTIVATE, nr_active); |
| 1410 | |
| 1411 | __mod_zone_page_state(zone, NR_ACTIVE_FILE, |
| 1412 | -count[LRU_ACTIVE_FILE]); |
| 1413 | __mod_zone_page_state(zone, NR_INACTIVE_FILE, |
| 1414 | -count[LRU_INACTIVE_FILE]); |
| 1415 | __mod_zone_page_state(zone, NR_ACTIVE_ANON, |
| 1416 | -count[LRU_ACTIVE_ANON]); |
| 1417 | __mod_zone_page_state(zone, NR_INACTIVE_ANON, |
| 1418 | -count[LRU_INACTIVE_ANON]); |
| 1419 | |
| 1420 | *nr_anon = count[LRU_ACTIVE_ANON] + count[LRU_INACTIVE_ANON]; |
| 1421 | *nr_file = count[LRU_ACTIVE_FILE] + count[LRU_INACTIVE_FILE]; |
| 1422 | __mod_zone_page_state(zone, NR_ISOLATED_ANON, *nr_anon); |
| 1423 | __mod_zone_page_state(zone, NR_ISOLATED_FILE, *nr_file); |
| 1424 | |
| 1425 | reclaim_stat->recent_scanned[0] += *nr_anon; |
| 1426 | reclaim_stat->recent_scanned[1] += *nr_file; |
| 1427 | } |
| 1428 | |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1429 | /* |
Mel Gorman | a18bba0 | 2011-10-31 17:07:42 -0700 | [diff] [blame] | 1430 | * Returns true if a direct reclaim should wait on pages under writeback. |
Wu Fengguang | e31f3698 | 2010-08-09 17:20:01 -0700 | [diff] [blame] | 1431 | * |
| 1432 | * If we are direct reclaiming for contiguous pages and we do not reclaim |
| 1433 | * everything in the list, try again and wait for writeback IO to complete. |
| 1434 | * This will stall high-order allocations noticeably. Only do that when really |
| 1435 | * need to free the pages under high memory pressure. |
| 1436 | */ |
| 1437 | static inline bool should_reclaim_stall(unsigned long nr_taken, |
| 1438 | unsigned long nr_freed, |
| 1439 | int priority, |
| 1440 | struct scan_control *sc) |
| 1441 | { |
| 1442 | int lumpy_stall_priority; |
| 1443 | |
| 1444 | /* kswapd should not stall on sync IO */ |
| 1445 | if (current_is_kswapd()) |
| 1446 | return false; |
| 1447 | |
| 1448 | /* Only stall on lumpy reclaim */ |
Mel Gorman | f3a310b | 2011-01-13 15:46:00 -0800 | [diff] [blame] | 1449 | if (sc->reclaim_mode & RECLAIM_MODE_SINGLE) |
Wu Fengguang | e31f3698 | 2010-08-09 17:20:01 -0700 | [diff] [blame] | 1450 | return false; |
| 1451 | |
Justin P. Mattock | 81d66c7 | 2011-08-23 09:28:02 -0700 | [diff] [blame] | 1452 | /* If we have reclaimed everything on the isolated list, no stall */ |
Wu Fengguang | e31f3698 | 2010-08-09 17:20:01 -0700 | [diff] [blame] | 1453 | if (nr_freed == nr_taken) |
| 1454 | return false; |
| 1455 | |
| 1456 | /* |
| 1457 | * For high-order allocations, there are two stall thresholds. |
| 1458 | * High-cost allocations stall immediately where as lower |
| 1459 | * order allocations such as stacks require the scanning |
| 1460 | * priority to be much higher before stalling. |
| 1461 | */ |
| 1462 | if (sc->order > PAGE_ALLOC_COSTLY_ORDER) |
| 1463 | lumpy_stall_priority = DEF_PRIORITY; |
| 1464 | else |
| 1465 | lumpy_stall_priority = DEF_PRIORITY / 3; |
| 1466 | |
| 1467 | return priority <= lumpy_stall_priority; |
| 1468 | } |
| 1469 | |
| 1470 | /* |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 1471 | * shrink_inactive_list() is a helper for shrink_zone(). It returns the number |
| 1472 | * of reclaimed pages |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1473 | */ |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1474 | static noinline_for_stack unsigned long |
| 1475 | shrink_inactive_list(unsigned long nr_to_scan, struct zone *zone, |
| 1476 | struct scan_control *sc, int priority, int file) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1477 | { |
| 1478 | LIST_HEAD(page_list); |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1479 | unsigned long nr_scanned; |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 1480 | unsigned long nr_reclaimed = 0; |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1481 | unsigned long nr_taken; |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1482 | unsigned long nr_anon; |
| 1483 | unsigned long nr_file; |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 1484 | unsigned long nr_dirty = 0; |
| 1485 | unsigned long nr_writeback = 0; |
Minchan Kim | 4356f21 | 2011-10-31 17:06:47 -0700 | [diff] [blame] | 1486 | isolate_mode_t reclaim_mode = ISOLATE_INACTIVE; |
KOSAKI Motohiro | 78dc583 | 2009-06-16 15:31:40 -0700 | [diff] [blame] | 1487 | |
Rik van Riel | 35cd781 | 2009-09-21 17:01:38 -0700 | [diff] [blame] | 1488 | while (unlikely(too_many_isolated(zone, file, sc))) { |
KOSAKI Motohiro | 58355c7 | 2009-10-26 16:49:35 -0700 | [diff] [blame] | 1489 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
Rik van Riel | 35cd781 | 2009-09-21 17:01:38 -0700 | [diff] [blame] | 1490 | |
| 1491 | /* We are about to die and free our memory. Return now. */ |
| 1492 | if (fatal_signal_pending(current)) |
| 1493 | return SWAP_CLUSTER_MAX; |
| 1494 | } |
| 1495 | |
Mel Gorman | f3a310b | 2011-01-13 15:46:00 -0800 | [diff] [blame] | 1496 | set_reclaim_mode(priority, sc, false); |
Minchan Kim | 4356f21 | 2011-10-31 17:06:47 -0700 | [diff] [blame] | 1497 | if (sc->reclaim_mode & RECLAIM_MODE_LUMPYRECLAIM) |
| 1498 | reclaim_mode |= ISOLATE_ACTIVE; |
| 1499 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1500 | lru_add_drain(); |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1501 | |
| 1502 | if (!sc->may_unmap) |
| 1503 | reclaim_mode |= ISOLATE_UNMAPPED; |
| 1504 | if (!sc->may_writepage) |
| 1505 | reclaim_mode |= ISOLATE_CLEAN; |
| 1506 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1507 | spin_lock_irq(&zone->lru_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1508 | |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1509 | if (scanning_global_lru(sc)) { |
Minchan Kim | 4356f21 | 2011-10-31 17:06:47 -0700 | [diff] [blame] | 1510 | nr_taken = isolate_pages_global(nr_to_scan, &page_list, |
| 1511 | &nr_scanned, sc->order, reclaim_mode, zone, 0, file); |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1512 | zone->pages_scanned += nr_scanned; |
KOSAKI Motohiro | b35ea17 | 2009-09-21 17:01:36 -0700 | [diff] [blame] | 1513 | if (current_is_kswapd()) |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1514 | __count_zone_vm_events(PGSCAN_KSWAPD, zone, |
| 1515 | nr_scanned); |
| 1516 | else |
| 1517 | __count_zone_vm_events(PGSCAN_DIRECT, zone, |
| 1518 | nr_scanned); |
| 1519 | } else { |
Minchan Kim | 4356f21 | 2011-10-31 17:06:47 -0700 | [diff] [blame] | 1520 | nr_taken = mem_cgroup_isolate_pages(nr_to_scan, &page_list, |
| 1521 | &nr_scanned, sc->order, reclaim_mode, zone, |
| 1522 | sc->mem_cgroup, 0, file); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1523 | /* |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1524 | * mem_cgroup_isolate_pages() keeps track of |
| 1525 | * scanned pages on its own. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1526 | */ |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1527 | } |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 1528 | |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1529 | if (nr_taken == 0) { |
| 1530 | spin_unlock_irq(&zone->lru_lock); |
| 1531 | return 0; |
| 1532 | } |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1533 | |
Mel Gorman | 1489fa1 | 2010-08-09 17:19:33 -0700 | [diff] [blame] | 1534 | update_isolated_counts(zone, sc, &nr_anon, &nr_file, &page_list); |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1535 | |
| 1536 | spin_unlock_irq(&zone->lru_lock); |
| 1537 | |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 1538 | nr_reclaimed = shrink_page_list(&page_list, zone, sc, priority, |
| 1539 | &nr_dirty, &nr_writeback); |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1540 | |
Wu Fengguang | e31f3698 | 2010-08-09 17:20:01 -0700 | [diff] [blame] | 1541 | /* Check if we should syncronously wait for writeback */ |
| 1542 | if (should_reclaim_stall(nr_taken, nr_reclaimed, priority, sc)) { |
Mel Gorman | f3a310b | 2011-01-13 15:46:00 -0800 | [diff] [blame] | 1543 | set_reclaim_mode(priority, sc, true); |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 1544 | nr_reclaimed += shrink_page_list(&page_list, zone, sc, |
| 1545 | priority, &nr_dirty, &nr_writeback); |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1546 | } |
| 1547 | |
| 1548 | local_irq_disable(); |
| 1549 | if (current_is_kswapd()) |
| 1550 | __count_vm_events(KSWAPD_STEAL, nr_reclaimed); |
| 1551 | __count_zone_vm_events(PGSTEAL, zone, nr_reclaimed); |
| 1552 | |
Mel Gorman | 1489fa1 | 2010-08-09 17:19:33 -0700 | [diff] [blame] | 1553 | putback_lru_pages(zone, sc, nr_anon, nr_file, &page_list); |
Mel Gorman | e11da5b | 2010-10-26 14:21:40 -0700 | [diff] [blame] | 1554 | |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 1555 | /* |
| 1556 | * If reclaim is isolating dirty pages under writeback, it implies |
| 1557 | * that the long-lived page allocation rate is exceeding the page |
| 1558 | * laundering rate. Either the global limits are not being effective |
| 1559 | * at throttling processes due to the page distribution throughout |
| 1560 | * zones or there is heavy usage of a slow backing device. The |
| 1561 | * only option is to throttle from reclaim context which is not ideal |
| 1562 | * as there is no guarantee the dirtying process is throttled in the |
| 1563 | * same way balance_dirty_pages() manages. |
| 1564 | * |
| 1565 | * This scales the number of dirty pages that must be under writeback |
| 1566 | * before throttling depending on priority. It is a simple backoff |
| 1567 | * function that has the most effect in the range DEF_PRIORITY to |
| 1568 | * DEF_PRIORITY-2 which is the priority reclaim is considered to be |
| 1569 | * in trouble and reclaim is considered to be in trouble. |
| 1570 | * |
| 1571 | * DEF_PRIORITY 100% isolated pages must be PageWriteback to throttle |
| 1572 | * DEF_PRIORITY-1 50% must be PageWriteback |
| 1573 | * DEF_PRIORITY-2 25% must be PageWriteback, kswapd in trouble |
| 1574 | * ... |
| 1575 | * DEF_PRIORITY-6 For SWAP_CLUSTER_MAX isolated pages, throttle if any |
| 1576 | * isolated page is PageWriteback |
| 1577 | */ |
| 1578 | if (nr_writeback && nr_writeback >= (nr_taken >> (DEF_PRIORITY-priority))) |
| 1579 | wait_iff_congested(zone, BLK_RW_ASYNC, HZ/10); |
| 1580 | |
Mel Gorman | e11da5b | 2010-10-26 14:21:40 -0700 | [diff] [blame] | 1581 | trace_mm_vmscan_lru_shrink_inactive(zone->zone_pgdat->node_id, |
| 1582 | zone_idx(zone), |
| 1583 | nr_scanned, nr_reclaimed, |
| 1584 | priority, |
Mel Gorman | f3a310b | 2011-01-13 15:46:00 -0800 | [diff] [blame] | 1585 | trace_shrink_flags(file, sc->reclaim_mode)); |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 1586 | return nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1587 | } |
| 1588 | |
Martin Bligh | 3bb1a852 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 1589 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1590 | * This moves pages from the active list to the inactive list. |
| 1591 | * |
| 1592 | * We move them the other way if the page is referenced by one or more |
| 1593 | * processes, from rmap. |
| 1594 | * |
| 1595 | * If the pages are mostly unmapped, the processing is fast and it is |
| 1596 | * appropriate to hold zone->lru_lock across the whole operation. But if |
| 1597 | * the pages are mapped, the processing is slow (page_referenced()) so we |
| 1598 | * should drop zone->lru_lock around each page. It's impossible to balance |
| 1599 | * this, so instead we remove the pages from the LRU while processing them. |
| 1600 | * It is safe to rely on PG_active against the non-LRU pages in here because |
| 1601 | * nobody will play with that bit on a non-LRU page. |
| 1602 | * |
| 1603 | * The downside is that we have to touch page->_count against each page. |
| 1604 | * But we had to alter page->flags anyway. |
| 1605 | */ |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1606 | |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1607 | static void move_active_pages_to_lru(struct zone *zone, |
| 1608 | struct list_head *list, |
| 1609 | enum lru_list lru) |
| 1610 | { |
| 1611 | unsigned long pgmoved = 0; |
| 1612 | struct pagevec pvec; |
| 1613 | struct page *page; |
| 1614 | |
| 1615 | pagevec_init(&pvec, 1); |
| 1616 | |
| 1617 | while (!list_empty(list)) { |
| 1618 | page = lru_to_page(list); |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1619 | |
| 1620 | VM_BUG_ON(PageLRU(page)); |
| 1621 | SetPageLRU(page); |
| 1622 | |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1623 | list_move(&page->lru, &zone->lru[lru].list); |
| 1624 | mem_cgroup_add_lru_list(page, lru); |
Rik van Riel | 2c888cf | 2011-01-13 15:47:13 -0800 | [diff] [blame] | 1625 | pgmoved += hpage_nr_pages(page); |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1626 | |
| 1627 | if (!pagevec_add(&pvec, page) || list_empty(list)) { |
| 1628 | spin_unlock_irq(&zone->lru_lock); |
| 1629 | if (buffer_heads_over_limit) |
| 1630 | pagevec_strip(&pvec); |
| 1631 | __pagevec_release(&pvec); |
| 1632 | spin_lock_irq(&zone->lru_lock); |
| 1633 | } |
| 1634 | } |
| 1635 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved); |
| 1636 | if (!is_active_lru(lru)) |
| 1637 | __count_vm_events(PGDEACTIVATE, pgmoved); |
| 1638 | } |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1639 | |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 1640 | 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] | 1641 | struct scan_control *sc, int priority, int file) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1642 | { |
KOSAKI Motohiro | 44c241f | 2009-09-21 17:01:35 -0700 | [diff] [blame] | 1643 | unsigned long nr_taken; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1644 | unsigned long pgscanned; |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 1645 | unsigned long vm_flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1646 | LIST_HEAD(l_hold); /* The pages which were snipped off */ |
Wu Fengguang | 8cab475 | 2009-06-16 15:33:12 -0700 | [diff] [blame] | 1647 | LIST_HEAD(l_active); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1648 | LIST_HEAD(l_inactive); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1649 | struct page *page; |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1650 | struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc); |
KOSAKI Motohiro | 44c241f | 2009-09-21 17:01:35 -0700 | [diff] [blame] | 1651 | unsigned long nr_rotated = 0; |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1652 | isolate_mode_t reclaim_mode = ISOLATE_ACTIVE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1653 | |
| 1654 | lru_add_drain(); |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1655 | |
| 1656 | if (!sc->may_unmap) |
| 1657 | reclaim_mode |= ISOLATE_UNMAPPED; |
| 1658 | if (!sc->may_writepage) |
| 1659 | reclaim_mode |= ISOLATE_CLEAN; |
| 1660 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1661 | spin_lock_irq(&zone->lru_lock); |
KAMEZAWA Hiroyuki | e72e2bd | 2009-01-07 18:08:23 -0800 | [diff] [blame] | 1662 | if (scanning_global_lru(sc)) { |
Johannes Weiner | 8b25c6d | 2010-05-24 14:32:40 -0700 | [diff] [blame] | 1663 | nr_taken = isolate_pages_global(nr_pages, &l_hold, |
| 1664 | &pgscanned, sc->order, |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1665 | reclaim_mode, zone, |
Johannes Weiner | 8b25c6d | 2010-05-24 14:32:40 -0700 | [diff] [blame] | 1666 | 1, file); |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1667 | zone->pages_scanned += pgscanned; |
Johannes Weiner | 8b25c6d | 2010-05-24 14:32:40 -0700 | [diff] [blame] | 1668 | } else { |
| 1669 | nr_taken = mem_cgroup_isolate_pages(nr_pages, &l_hold, |
| 1670 | &pgscanned, sc->order, |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1671 | reclaim_mode, zone, |
Johannes Weiner | 8b25c6d | 2010-05-24 14:32:40 -0700 | [diff] [blame] | 1672 | sc->mem_cgroup, 1, file); |
| 1673 | /* |
| 1674 | * mem_cgroup_isolate_pages() keeps track of |
| 1675 | * scanned pages on its own. |
| 1676 | */ |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1677 | } |
Johannes Weiner | 8b25c6d | 2010-05-24 14:32:40 -0700 | [diff] [blame] | 1678 | |
Johannes Weiner | b7c46d1 | 2009-09-21 17:02:56 -0700 | [diff] [blame] | 1679 | reclaim_stat->recent_scanned[file] += nr_taken; |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1680 | |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1681 | __count_zone_vm_events(PGREFILL, zone, pgscanned); |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1682 | if (file) |
KOSAKI Motohiro | 44c241f | 2009-09-21 17:01:35 -0700 | [diff] [blame] | 1683 | __mod_zone_page_state(zone, NR_ACTIVE_FILE, -nr_taken); |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1684 | else |
KOSAKI Motohiro | 44c241f | 2009-09-21 17:01:35 -0700 | [diff] [blame] | 1685 | __mod_zone_page_state(zone, NR_ACTIVE_ANON, -nr_taken); |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 1686 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, nr_taken); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1687 | spin_unlock_irq(&zone->lru_lock); |
| 1688 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1689 | while (!list_empty(&l_hold)) { |
| 1690 | cond_resched(); |
| 1691 | page = lru_to_page(&l_hold); |
| 1692 | list_del(&page->lru); |
Rik van Riel | 7e9cd48 | 2008-10-18 20:26:35 -0700 | [diff] [blame] | 1693 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1694 | if (unlikely(!page_evictable(page, NULL))) { |
| 1695 | putback_lru_page(page); |
| 1696 | continue; |
| 1697 | } |
| 1698 | |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 1699 | if (page_referenced(page, 0, sc->mem_cgroup, &vm_flags)) { |
Rik van Riel | 9992af1 | 2011-01-13 15:47:13 -0800 | [diff] [blame] | 1700 | nr_rotated += hpage_nr_pages(page); |
Wu Fengguang | 8cab475 | 2009-06-16 15:33:12 -0700 | [diff] [blame] | 1701 | /* |
| 1702 | * Identify referenced, file-backed active pages and |
| 1703 | * give them one more trip around the active list. So |
| 1704 | * that executable code get better chances to stay in |
| 1705 | * memory under moderate memory pressure. Anon pages |
| 1706 | * are not likely to be evicted by use-once streaming |
| 1707 | * IO, plus JVM can create lots of anon VM_EXEC pages, |
| 1708 | * so we ignore them here. |
| 1709 | */ |
Wu Fengguang | 41e2098 | 2009-10-26 16:49:53 -0700 | [diff] [blame] | 1710 | if ((vm_flags & VM_EXEC) && page_is_file_cache(page)) { |
Wu Fengguang | 8cab475 | 2009-06-16 15:33:12 -0700 | [diff] [blame] | 1711 | list_add(&page->lru, &l_active); |
| 1712 | continue; |
| 1713 | } |
| 1714 | } |
Rik van Riel | 7e9cd48 | 2008-10-18 20:26:35 -0700 | [diff] [blame] | 1715 | |
KOSAKI Motohiro | 5205e56 | 2009-09-21 17:01:44 -0700 | [diff] [blame] | 1716 | ClearPageActive(page); /* we are de-activating */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1717 | list_add(&page->lru, &l_inactive); |
| 1718 | } |
| 1719 | |
Andrew Morton | b555749 | 2009-01-06 14:40:13 -0800 | [diff] [blame] | 1720 | /* |
Wu Fengguang | 8cab475 | 2009-06-16 15:33:12 -0700 | [diff] [blame] | 1721 | * Move pages back to the lru list. |
Andrew Morton | b555749 | 2009-01-06 14:40:13 -0800 | [diff] [blame] | 1722 | */ |
Johannes Weiner | 2a1dc50 | 2008-12-01 03:00:35 +0100 | [diff] [blame] | 1723 | spin_lock_irq(&zone->lru_lock); |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1724 | /* |
Wu Fengguang | 8cab475 | 2009-06-16 15:33:12 -0700 | [diff] [blame] | 1725 | * Count referenced pages from currently used mappings as rotated, |
| 1726 | * even though only some of them are actually re-activated. This |
| 1727 | * helps balance scan pressure between file and anonymous pages in |
| 1728 | * get_scan_ratio. |
Rik van Riel | 7e9cd48 | 2008-10-18 20:26:35 -0700 | [diff] [blame] | 1729 | */ |
Johannes Weiner | b7c46d1 | 2009-09-21 17:02:56 -0700 | [diff] [blame] | 1730 | reclaim_stat->recent_rotated[file] += nr_rotated; |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1731 | |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1732 | move_active_pages_to_lru(zone, &l_active, |
| 1733 | LRU_ACTIVE + file * LRU_FILE); |
| 1734 | move_active_pages_to_lru(zone, &l_inactive, |
| 1735 | LRU_BASE + file * LRU_FILE); |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 1736 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, -nr_taken); |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 1737 | spin_unlock_irq(&zone->lru_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1738 | } |
| 1739 | |
Minchan Kim | 74e3f3c | 2010-10-26 14:21:31 -0700 | [diff] [blame] | 1740 | #ifdef CONFIG_SWAP |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1741 | static int inactive_anon_is_low_global(struct zone *zone) |
KOSAKI Motohiro | f89eb90 | 2009-01-07 18:08:14 -0800 | [diff] [blame] | 1742 | { |
| 1743 | unsigned long active, inactive; |
| 1744 | |
| 1745 | active = zone_page_state(zone, NR_ACTIVE_ANON); |
| 1746 | inactive = zone_page_state(zone, NR_INACTIVE_ANON); |
| 1747 | |
| 1748 | if (inactive * zone->inactive_ratio < active) |
| 1749 | return 1; |
| 1750 | |
| 1751 | return 0; |
| 1752 | } |
| 1753 | |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1754 | /** |
| 1755 | * inactive_anon_is_low - check if anonymous pages need to be deactivated |
| 1756 | * @zone: zone to check |
| 1757 | * @sc: scan control of this context |
| 1758 | * |
| 1759 | * Returns true if the zone does not have enough inactive anon pages, |
| 1760 | * meaning some active anon pages need to be deactivated. |
| 1761 | */ |
| 1762 | static int inactive_anon_is_low(struct zone *zone, struct scan_control *sc) |
| 1763 | { |
| 1764 | int low; |
| 1765 | |
Minchan Kim | 74e3f3c | 2010-10-26 14:21:31 -0700 | [diff] [blame] | 1766 | /* |
| 1767 | * If we don't have swap space, anonymous page deactivation |
| 1768 | * is pointless. |
| 1769 | */ |
| 1770 | if (!total_swap_pages) |
| 1771 | return 0; |
| 1772 | |
KAMEZAWA Hiroyuki | e72e2bd | 2009-01-07 18:08:23 -0800 | [diff] [blame] | 1773 | if (scanning_global_lru(sc)) |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1774 | low = inactive_anon_is_low_global(zone); |
| 1775 | else |
Johannes Weiner | 9b27297 | 2011-11-02 13:38:23 -0700 | [diff] [blame] | 1776 | low = mem_cgroup_inactive_anon_is_low(sc->mem_cgroup, zone); |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1777 | return low; |
| 1778 | } |
Minchan Kim | 74e3f3c | 2010-10-26 14:21:31 -0700 | [diff] [blame] | 1779 | #else |
| 1780 | static inline int inactive_anon_is_low(struct zone *zone, |
| 1781 | struct scan_control *sc) |
| 1782 | { |
| 1783 | return 0; |
| 1784 | } |
| 1785 | #endif |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1786 | |
Rik van Riel | 56e49d2 | 2009-06-16 15:32:28 -0700 | [diff] [blame] | 1787 | static int inactive_file_is_low_global(struct zone *zone) |
| 1788 | { |
| 1789 | unsigned long active, inactive; |
| 1790 | |
| 1791 | active = zone_page_state(zone, NR_ACTIVE_FILE); |
| 1792 | inactive = zone_page_state(zone, NR_INACTIVE_FILE); |
| 1793 | |
| 1794 | return (active > inactive); |
| 1795 | } |
| 1796 | |
| 1797 | /** |
| 1798 | * inactive_file_is_low - check if file pages need to be deactivated |
| 1799 | * @zone: zone to check |
| 1800 | * @sc: scan control of this context |
| 1801 | * |
| 1802 | * When the system is doing streaming IO, memory pressure here |
| 1803 | * ensures that active file pages get deactivated, until more |
| 1804 | * than half of the file pages are on the inactive list. |
| 1805 | * |
| 1806 | * Once we get to that situation, protect the system's working |
| 1807 | * set from being evicted by disabling active file page aging. |
| 1808 | * |
| 1809 | * This uses a different ratio than the anonymous pages, because |
| 1810 | * the page cache uses a use-once replacement algorithm. |
| 1811 | */ |
| 1812 | static int inactive_file_is_low(struct zone *zone, struct scan_control *sc) |
| 1813 | { |
| 1814 | int low; |
| 1815 | |
| 1816 | if (scanning_global_lru(sc)) |
| 1817 | low = inactive_file_is_low_global(zone); |
| 1818 | else |
Johannes Weiner | 9b27297 | 2011-11-02 13:38:23 -0700 | [diff] [blame] | 1819 | low = mem_cgroup_inactive_file_is_low(sc->mem_cgroup, zone); |
Rik van Riel | 56e49d2 | 2009-06-16 15:32:28 -0700 | [diff] [blame] | 1820 | return low; |
| 1821 | } |
| 1822 | |
Rik van Riel | b39415b | 2009-12-14 17:59:48 -0800 | [diff] [blame] | 1823 | static int inactive_list_is_low(struct zone *zone, struct scan_control *sc, |
| 1824 | int file) |
| 1825 | { |
| 1826 | if (file) |
| 1827 | return inactive_file_is_low(zone, sc); |
| 1828 | else |
| 1829 | return inactive_anon_is_low(zone, sc); |
| 1830 | } |
| 1831 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1832 | 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] | 1833 | struct zone *zone, struct scan_control *sc, int priority) |
| 1834 | { |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1835 | int file = is_file_lru(lru); |
| 1836 | |
Rik van Riel | b39415b | 2009-12-14 17:59:48 -0800 | [diff] [blame] | 1837 | if (is_active_lru(lru)) { |
| 1838 | if (inactive_list_is_low(zone, sc, file)) |
| 1839 | shrink_active_list(nr_to_scan, zone, sc, priority, file); |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1840 | return 0; |
| 1841 | } |
| 1842 | |
Rik van Riel | 33c120e | 2008-10-18 20:26:36 -0700 | [diff] [blame] | 1843 | return shrink_inactive_list(nr_to_scan, zone, sc, priority, file); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1844 | } |
| 1845 | |
KAMEZAWA Hiroyuki | 1f4c025 | 2011-07-26 16:08:21 -0700 | [diff] [blame] | 1846 | static int vmscan_swappiness(struct scan_control *sc) |
| 1847 | { |
| 1848 | if (scanning_global_lru(sc)) |
| 1849 | return vm_swappiness; |
| 1850 | return mem_cgroup_swappiness(sc->mem_cgroup); |
| 1851 | } |
| 1852 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1853 | /* |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1854 | * Determine how aggressively the anon and file LRU lists should be |
| 1855 | * scanned. The relative value of each set of LRU lists is determined |
| 1856 | * by looking at the fraction of the pages scanned we did rotate back |
| 1857 | * onto the active list instead of evict. |
| 1858 | * |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1859 | * nr[0] = anon pages to scan; nr[1] = file pages to scan |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1860 | */ |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1861 | static void get_scan_count(struct zone *zone, struct scan_control *sc, |
| 1862 | unsigned long *nr, int priority) |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1863 | { |
| 1864 | unsigned long anon, file, free; |
| 1865 | unsigned long anon_prio, file_prio; |
| 1866 | unsigned long ap, fp; |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1867 | struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc); |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1868 | u64 fraction[2], denominator; |
| 1869 | enum lru_list l; |
| 1870 | int noswap = 0; |
Johannes Weiner | a4d3e9e | 2011-09-14 16:21:52 -0700 | [diff] [blame] | 1871 | bool force_scan = false; |
KAMEZAWA Hiroyuki | 246e87a | 2011-05-26 16:25:34 -0700 | [diff] [blame] | 1872 | |
Johannes Weiner | f11c0ca | 2011-10-31 17:07:27 -0700 | [diff] [blame] | 1873 | /* |
| 1874 | * If the zone or memcg is small, nr[l] can be 0. This |
| 1875 | * results in no scanning on this priority and a potential |
| 1876 | * priority drop. Global direct reclaim can go to the next |
| 1877 | * zone and tends to have no problems. Global kswapd is for |
| 1878 | * zone balancing and it needs to scan a minimum amount. When |
| 1879 | * reclaiming for a memcg, a priority drop can cause high |
| 1880 | * latencies, so it's better to scan a minimum amount there as |
| 1881 | * well. |
| 1882 | */ |
Johannes Weiner | a4d3e9e | 2011-09-14 16:21:52 -0700 | [diff] [blame] | 1883 | if (scanning_global_lru(sc) && current_is_kswapd()) |
| 1884 | force_scan = true; |
Johannes Weiner | a4d3e9e | 2011-09-14 16:21:52 -0700 | [diff] [blame] | 1885 | if (!scanning_global_lru(sc)) |
| 1886 | force_scan = true; |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1887 | |
| 1888 | /* If we have no swap space, do not bother scanning anon pages. */ |
| 1889 | if (!sc->may_swap || (nr_swap_pages <= 0)) { |
| 1890 | noswap = 1; |
| 1891 | fraction[0] = 0; |
| 1892 | fraction[1] = 1; |
| 1893 | denominator = 1; |
| 1894 | goto out; |
| 1895 | } |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1896 | |
Johannes Weiner | a4d3e9e | 2011-09-14 16:21:52 -0700 | [diff] [blame] | 1897 | anon = zone_nr_lru_pages(zone, sc, LRU_ACTIVE_ANON) + |
| 1898 | zone_nr_lru_pages(zone, sc, LRU_INACTIVE_ANON); |
| 1899 | file = zone_nr_lru_pages(zone, sc, LRU_ACTIVE_FILE) + |
| 1900 | zone_nr_lru_pages(zone, sc, LRU_INACTIVE_FILE); |
| 1901 | |
KAMEZAWA Hiroyuki | e72e2bd | 2009-01-07 18:08:23 -0800 | [diff] [blame] | 1902 | if (scanning_global_lru(sc)) { |
KOSAKI Motohiro | eeee9a8 | 2009-01-07 18:08:17 -0800 | [diff] [blame] | 1903 | free = zone_page_state(zone, NR_FREE_PAGES); |
| 1904 | /* If we have very few page cache pages, |
| 1905 | force-scan anon pages. */ |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 1906 | if (unlikely(file + free <= high_wmark_pages(zone))) { |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1907 | fraction[0] = 1; |
| 1908 | fraction[1] = 0; |
| 1909 | denominator = 1; |
| 1910 | goto out; |
KOSAKI Motohiro | eeee9a8 | 2009-01-07 18:08:17 -0800 | [diff] [blame] | 1911 | } |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1912 | } |
| 1913 | |
| 1914 | /* |
KOSAKI Motohiro | 58c37f6 | 2010-08-09 17:19:51 -0700 | [diff] [blame] | 1915 | * With swappiness at 100, anonymous and file have the same priority. |
| 1916 | * This scanning priority is essentially the inverse of IO cost. |
| 1917 | */ |
KAMEZAWA Hiroyuki | 1f4c025 | 2011-07-26 16:08:21 -0700 | [diff] [blame] | 1918 | anon_prio = vmscan_swappiness(sc); |
| 1919 | file_prio = 200 - vmscan_swappiness(sc); |
KOSAKI Motohiro | 58c37f6 | 2010-08-09 17:19:51 -0700 | [diff] [blame] | 1920 | |
| 1921 | /* |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1922 | * OK, so we have swap space and a fair amount of page cache |
| 1923 | * pages. We use the recently rotated / recently scanned |
| 1924 | * ratios to determine how valuable each cache is. |
| 1925 | * |
| 1926 | * Because workloads change over time (and to avoid overflow) |
| 1927 | * we keep these statistics as a floating average, which ends |
| 1928 | * up weighing recent references more than old ones. |
| 1929 | * |
| 1930 | * anon in [0], file in [1] |
| 1931 | */ |
KOSAKI Motohiro | 58c37f6 | 2010-08-09 17:19:51 -0700 | [diff] [blame] | 1932 | spin_lock_irq(&zone->lru_lock); |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1933 | if (unlikely(reclaim_stat->recent_scanned[0] > anon / 4)) { |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1934 | reclaim_stat->recent_scanned[0] /= 2; |
| 1935 | reclaim_stat->recent_rotated[0] /= 2; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1936 | } |
| 1937 | |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1938 | if (unlikely(reclaim_stat->recent_scanned[1] > file / 4)) { |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1939 | reclaim_stat->recent_scanned[1] /= 2; |
| 1940 | reclaim_stat->recent_rotated[1] /= 2; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1941 | } |
| 1942 | |
| 1943 | /* |
Rik van Riel | 00d8089 | 2008-11-19 15:36:44 -0800 | [diff] [blame] | 1944 | * The amount of pressure on anon vs file pages is inversely |
| 1945 | * proportional to the fraction of recently scanned pages on |
| 1946 | * each list that were recently referenced and in active use. |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1947 | */ |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1948 | ap = (anon_prio + 1) * (reclaim_stat->recent_scanned[0] + 1); |
| 1949 | ap /= reclaim_stat->recent_rotated[0] + 1; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1950 | |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1951 | fp = (file_prio + 1) * (reclaim_stat->recent_scanned[1] + 1); |
| 1952 | fp /= reclaim_stat->recent_rotated[1] + 1; |
KOSAKI Motohiro | 58c37f6 | 2010-08-09 17:19:51 -0700 | [diff] [blame] | 1953 | spin_unlock_irq(&zone->lru_lock); |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1954 | |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1955 | fraction[0] = ap; |
| 1956 | fraction[1] = fp; |
| 1957 | denominator = ap + fp + 1; |
| 1958 | out: |
| 1959 | for_each_evictable_lru(l) { |
| 1960 | int file = is_file_lru(l); |
| 1961 | unsigned long scan; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1962 | |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1963 | scan = zone_nr_lru_pages(zone, sc, l); |
| 1964 | if (priority || noswap) { |
| 1965 | scan >>= priority; |
Johannes Weiner | f11c0ca | 2011-10-31 17:07:27 -0700 | [diff] [blame] | 1966 | if (!scan && force_scan) |
| 1967 | scan = SWAP_CLUSTER_MAX; |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1968 | scan = div64_u64(scan * fraction[file], denominator); |
| 1969 | } |
KAMEZAWA Hiroyuki | 246e87a | 2011-05-26 16:25:34 -0700 | [diff] [blame] | 1970 | nr[l] = scan; |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1971 | } |
Wu Fengguang | 6e08a36 | 2009-06-16 15:32:29 -0700 | [diff] [blame] | 1972 | } |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1973 | |
| 1974 | /* |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1975 | * Reclaim/compaction depends on a number of pages being freed. To avoid |
| 1976 | * disruption to the system, a small number of order-0 pages continue to be |
| 1977 | * rotated and reclaimed in the normal fashion. However, by the time we get |
| 1978 | * back to the allocator and call try_to_compact_zone(), we ensure that |
| 1979 | * there are enough free pages for it to be likely successful |
| 1980 | */ |
| 1981 | static inline bool should_continue_reclaim(struct zone *zone, |
| 1982 | unsigned long nr_reclaimed, |
| 1983 | unsigned long nr_scanned, |
| 1984 | struct scan_control *sc) |
| 1985 | { |
| 1986 | unsigned long pages_for_compaction; |
| 1987 | unsigned long inactive_lru_pages; |
| 1988 | |
| 1989 | /* If not in reclaim/compaction mode, stop */ |
Mel Gorman | f3a310b | 2011-01-13 15:46:00 -0800 | [diff] [blame] | 1990 | if (!(sc->reclaim_mode & RECLAIM_MODE_COMPACTION)) |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1991 | return false; |
| 1992 | |
Mel Gorman | 2876592 | 2011-02-25 14:44:20 -0800 | [diff] [blame] | 1993 | /* Consider stopping depending on scan and reclaim activity */ |
| 1994 | if (sc->gfp_mask & __GFP_REPEAT) { |
| 1995 | /* |
| 1996 | * For __GFP_REPEAT allocations, stop reclaiming if the |
| 1997 | * full LRU list has been scanned and we are still failing |
| 1998 | * to reclaim pages. This full LRU scan is potentially |
| 1999 | * expensive but a __GFP_REPEAT caller really wants to succeed |
| 2000 | */ |
| 2001 | if (!nr_reclaimed && !nr_scanned) |
| 2002 | return false; |
| 2003 | } else { |
| 2004 | /* |
| 2005 | * For non-__GFP_REPEAT allocations which can presumably |
| 2006 | * fail without consequence, stop if we failed to reclaim |
| 2007 | * any pages from the last SWAP_CLUSTER_MAX number of |
| 2008 | * pages that were scanned. This will return to the |
| 2009 | * caller faster at the risk reclaim/compaction and |
| 2010 | * the resulting allocation attempt fails |
| 2011 | */ |
| 2012 | if (!nr_reclaimed) |
| 2013 | return false; |
| 2014 | } |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 2015 | |
| 2016 | /* |
| 2017 | * If we have not reclaimed enough pages for compaction and the |
| 2018 | * inactive lists are large enough, continue reclaiming |
| 2019 | */ |
| 2020 | pages_for_compaction = (2UL << sc->order); |
| 2021 | inactive_lru_pages = zone_nr_lru_pages(zone, sc, LRU_INACTIVE_ANON) + |
| 2022 | zone_nr_lru_pages(zone, sc, LRU_INACTIVE_FILE); |
| 2023 | if (sc->nr_reclaimed < pages_for_compaction && |
| 2024 | inactive_lru_pages > pages_for_compaction) |
| 2025 | return true; |
| 2026 | |
| 2027 | /* If compaction would go ahead or the allocation would succeed, stop */ |
| 2028 | switch (compaction_suitable(zone, sc->order)) { |
| 2029 | case COMPACT_PARTIAL: |
| 2030 | case COMPACT_CONTINUE: |
| 2031 | return false; |
| 2032 | default: |
| 2033 | return true; |
| 2034 | } |
| 2035 | } |
| 2036 | |
| 2037 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2038 | * This is a basic per-zone page freer. Used by both kswapd and direct reclaim. |
| 2039 | */ |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2040 | static void shrink_zone(int priority, struct zone *zone, |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 2041 | struct scan_control *sc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2042 | { |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 2043 | unsigned long nr[NR_LRU_LISTS]; |
Christoph Lameter | 8695949 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2044 | unsigned long nr_to_scan; |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 2045 | enum lru_list l; |
Johannes Weiner | f0fdc5e | 2011-02-10 15:01:34 -0800 | [diff] [blame] | 2046 | unsigned long nr_reclaimed, nr_scanned; |
KOSAKI Motohiro | 22fba33 | 2009-12-14 17:59:10 -0800 | [diff] [blame] | 2047 | unsigned long nr_to_reclaim = sc->nr_to_reclaim; |
Shaohua Li | 3da367c | 2011-10-31 17:07:03 -0700 | [diff] [blame] | 2048 | struct blk_plug plug; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2049 | |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 2050 | restart: |
| 2051 | nr_reclaimed = 0; |
Johannes Weiner | f0fdc5e | 2011-02-10 15:01:34 -0800 | [diff] [blame] | 2052 | nr_scanned = sc->nr_scanned; |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 2053 | get_scan_count(zone, sc, nr, priority); |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 2054 | |
Shaohua Li | 3da367c | 2011-10-31 17:07:03 -0700 | [diff] [blame] | 2055 | blk_start_plug(&plug); |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 2056 | while (nr[LRU_INACTIVE_ANON] || nr[LRU_ACTIVE_FILE] || |
| 2057 | nr[LRU_INACTIVE_FILE]) { |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 2058 | for_each_evictable_lru(l) { |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 2059 | if (nr[l]) { |
KOSAKI Motohiro | ece74b2 | 2009-12-14 17:59:14 -0800 | [diff] [blame] | 2060 | nr_to_scan = min_t(unsigned long, |
| 2061 | nr[l], SWAP_CLUSTER_MAX); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 2062 | nr[l] -= nr_to_scan; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2063 | |
KOSAKI Motohiro | 01dbe5c | 2009-01-06 14:40:02 -0800 | [diff] [blame] | 2064 | nr_reclaimed += shrink_list(l, nr_to_scan, |
| 2065 | zone, sc, priority); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 2066 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2067 | } |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2068 | /* |
| 2069 | * On large memory systems, scan >> priority can become |
| 2070 | * really large. This is fine for the starting priority; |
| 2071 | * we want to put equal scanning pressure on each zone. |
| 2072 | * However, if the VM has a harder time of freeing pages, |
| 2073 | * with multiple processes reclaiming pages, the total |
| 2074 | * freeing target can get unreasonably large. |
| 2075 | */ |
KOSAKI Motohiro | 338fde9 | 2009-12-14 17:59:15 -0800 | [diff] [blame] | 2076 | if (nr_reclaimed >= nr_to_reclaim && priority < DEF_PRIORITY) |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2077 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2078 | } |
Shaohua Li | 3da367c | 2011-10-31 17:07:03 -0700 | [diff] [blame] | 2079 | blk_finish_plug(&plug); |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 2080 | sc->nr_reclaimed += nr_reclaimed; |
KOSAKI Motohiro | 01dbe5c | 2009-01-06 14:40:02 -0800 | [diff] [blame] | 2081 | |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 2082 | /* |
| 2083 | * Even if we did not try to evict anon pages at all, we want to |
| 2084 | * rebalance the anon lru active/inactive ratio. |
| 2085 | */ |
Minchan Kim | 74e3f3c | 2010-10-26 14:21:31 -0700 | [diff] [blame] | 2086 | if (inactive_anon_is_low(zone, sc)) |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 2087 | shrink_active_list(SWAP_CLUSTER_MAX, zone, sc, priority, 0); |
| 2088 | |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 2089 | /* reclaim/compaction might need reclaim to continue */ |
| 2090 | if (should_continue_reclaim(zone, nr_reclaimed, |
| 2091 | sc->nr_scanned - nr_scanned, sc)) |
| 2092 | goto restart; |
| 2093 | |
Andrew Morton | 232ea4d | 2007-02-28 20:13:21 -0800 | [diff] [blame] | 2094 | throttle_vm_writeout(sc->gfp_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2095 | } |
| 2096 | |
| 2097 | /* |
| 2098 | * This is the direct reclaim path, for page-allocating processes. We only |
| 2099 | * try to reclaim pages from zones which will satisfy the caller's allocation |
| 2100 | * request. |
| 2101 | * |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 2102 | * We reclaim from a zone even if that zone is over high_wmark_pages(zone). |
| 2103 | * Because: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2104 | * a) The caller may be trying to free *extra* pages to satisfy a higher-order |
| 2105 | * allocation or |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 2106 | * b) The target zone may be at high_wmark_pages(zone) but the lower zones |
| 2107 | * must go *over* high_wmark_pages(zone) to satisfy the `incremental min' |
| 2108 | * zone defense algorithm. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2109 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2110 | * If a zone is deemed to be full of pinned pages then just give it a light |
| 2111 | * scan then give up on it. |
Mel Gorman | e0c2327 | 2011-10-31 17:09:33 -0700 | [diff] [blame] | 2112 | * |
| 2113 | * This function returns true if a zone is being reclaimed for a costly |
| 2114 | * high-order allocation and compaction is either ready to begin or deferred. |
| 2115 | * This indicates to the caller that it should retry the allocation or fail. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2116 | */ |
Mel Gorman | e0c2327 | 2011-10-31 17:09:33 -0700 | [diff] [blame] | 2117 | static bool shrink_zones(int priority, struct zonelist *zonelist, |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 2118 | struct scan_control *sc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2119 | { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 2120 | struct zoneref *z; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2121 | struct zone *zone; |
Ying Han | d149e3b | 2011-05-26 16:25:27 -0700 | [diff] [blame] | 2122 | unsigned long nr_soft_reclaimed; |
| 2123 | unsigned long nr_soft_scanned; |
Mel Gorman | e0c2327 | 2011-10-31 17:09:33 -0700 | [diff] [blame] | 2124 | bool should_abort_reclaim = false; |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 2125 | |
Mel Gorman | d4debc6 | 2010-08-09 17:19:29 -0700 | [diff] [blame] | 2126 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
| 2127 | gfp_zone(sc->gfp_mask), sc->nodemask) { |
Con Kolivas | f3fe651 | 2006-01-06 00:11:15 -0800 | [diff] [blame] | 2128 | if (!populated_zone(zone)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2129 | continue; |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 2130 | /* |
| 2131 | * Take care memory controller reclaiming has small influence |
| 2132 | * to global LRU. |
| 2133 | */ |
KAMEZAWA Hiroyuki | e72e2bd | 2009-01-07 18:08:23 -0800 | [diff] [blame] | 2134 | if (scanning_global_lru(sc)) { |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 2135 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
| 2136 | continue; |
KOSAKI Motohiro | 93e4a89 | 2010-03-05 13:41:55 -0800 | [diff] [blame] | 2137 | if (zone->all_unreclaimable && priority != DEF_PRIORITY) |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 2138 | continue; /* Let kswapd poll it */ |
Rik van Riel | e0887c1 | 2011-10-31 17:09:31 -0700 | [diff] [blame] | 2139 | if (COMPACTION_BUILD) { |
| 2140 | /* |
Mel Gorman | e0c2327 | 2011-10-31 17:09:33 -0700 | [diff] [blame] | 2141 | * If we already have plenty of memory free for |
| 2142 | * compaction in this zone, don't free any more. |
| 2143 | * Even though compaction is invoked for any |
| 2144 | * non-zero order, only frequent costly order |
| 2145 | * reclamation is disruptive enough to become a |
| 2146 | * noticable problem, like transparent huge page |
| 2147 | * allocations. |
Rik van Riel | e0887c1 | 2011-10-31 17:09:31 -0700 | [diff] [blame] | 2148 | */ |
| 2149 | if (sc->order > PAGE_ALLOC_COSTLY_ORDER && |
| 2150 | (compaction_suitable(zone, sc->order) || |
Mel Gorman | e0c2327 | 2011-10-31 17:09:33 -0700 | [diff] [blame] | 2151 | compaction_deferred(zone))) { |
| 2152 | should_abort_reclaim = true; |
Rik van Riel | e0887c1 | 2011-10-31 17:09:31 -0700 | [diff] [blame] | 2153 | continue; |
Mel Gorman | e0c2327 | 2011-10-31 17:09:33 -0700 | [diff] [blame] | 2154 | } |
Rik van Riel | e0887c1 | 2011-10-31 17:09:31 -0700 | [diff] [blame] | 2155 | } |
KAMEZAWA Hiroyuki | ac34a1a | 2011-06-27 16:18:12 -0700 | [diff] [blame] | 2156 | /* |
| 2157 | * This steals pages from memory cgroups over softlimit |
| 2158 | * and returns the number of reclaimed pages and |
| 2159 | * scanned pages. This works for global memory pressure |
| 2160 | * and balancing, not for a memcg's limit. |
| 2161 | */ |
| 2162 | nr_soft_scanned = 0; |
| 2163 | nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone, |
| 2164 | sc->order, sc->gfp_mask, |
| 2165 | &nr_soft_scanned); |
| 2166 | sc->nr_reclaimed += nr_soft_reclaimed; |
| 2167 | sc->nr_scanned += nr_soft_scanned; |
| 2168 | /* need some check for avoid more shrink_zone() */ |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 2169 | } |
Nick Piggin | 408d854 | 2006-09-25 23:31:27 -0700 | [diff] [blame] | 2170 | |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2171 | shrink_zone(priority, zone, sc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2172 | } |
Mel Gorman | e0c2327 | 2011-10-31 17:09:33 -0700 | [diff] [blame] | 2173 | |
| 2174 | return should_abort_reclaim; |
Minchan Kim | d190836 | 2010-09-22 13:05:01 -0700 | [diff] [blame] | 2175 | } |
| 2176 | |
| 2177 | static bool zone_reclaimable(struct zone *zone) |
| 2178 | { |
| 2179 | return zone->pages_scanned < zone_reclaimable_pages(zone) * 6; |
| 2180 | } |
| 2181 | |
KOSAKI Motohiro | 929bea7 | 2011-04-14 15:22:12 -0700 | [diff] [blame] | 2182 | /* All zones in zonelist are unreclaimable? */ |
Minchan Kim | d190836 | 2010-09-22 13:05:01 -0700 | [diff] [blame] | 2183 | static bool all_unreclaimable(struct zonelist *zonelist, |
| 2184 | struct scan_control *sc) |
| 2185 | { |
| 2186 | struct zoneref *z; |
| 2187 | struct zone *zone; |
Minchan Kim | d190836 | 2010-09-22 13:05:01 -0700 | [diff] [blame] | 2188 | |
| 2189 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
| 2190 | gfp_zone(sc->gfp_mask), sc->nodemask) { |
| 2191 | if (!populated_zone(zone)) |
| 2192 | continue; |
| 2193 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
| 2194 | continue; |
KOSAKI Motohiro | 929bea7 | 2011-04-14 15:22:12 -0700 | [diff] [blame] | 2195 | if (!zone->all_unreclaimable) |
| 2196 | return false; |
Minchan Kim | d190836 | 2010-09-22 13:05:01 -0700 | [diff] [blame] | 2197 | } |
| 2198 | |
KOSAKI Motohiro | 929bea7 | 2011-04-14 15:22:12 -0700 | [diff] [blame] | 2199 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2200 | } |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2201 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2202 | /* |
| 2203 | * This is the main entry point to direct page reclaim. |
| 2204 | * |
| 2205 | * If a full scan of the inactive list fails to free enough memory then we |
| 2206 | * are "out of memory" and something needs to be killed. |
| 2207 | * |
| 2208 | * If the caller is !__GFP_FS then the probability of a failure is reasonably |
| 2209 | * high - the zone may be full of dirty or under-writeback pages, which this |
Jens Axboe | 5b0830c | 2009-09-23 19:37:09 +0200 | [diff] [blame] | 2210 | * caller can't do much about. We kick the writeback threads and take explicit |
| 2211 | * naps in the hope that some of these pages can be written. But if the |
| 2212 | * allocating task holds filesystem locks which prevent writeout this might not |
| 2213 | * work, and the allocation attempt will fail. |
Nishanth Aravamudan | a41f24e | 2008-04-29 00:58:25 -0700 | [diff] [blame] | 2214 | * |
| 2215 | * returns: 0, if no pages reclaimed |
| 2216 | * else, the number of pages reclaimed |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2217 | */ |
Mel Gorman | dac1d27 | 2008-04-28 02:12:12 -0700 | [diff] [blame] | 2218 | static unsigned long do_try_to_free_pages(struct zonelist *zonelist, |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 2219 | struct scan_control *sc, |
| 2220 | struct shrink_control *shrink) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2221 | { |
| 2222 | int priority; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 2223 | unsigned long total_scanned = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2224 | struct reclaim_state *reclaim_state = current->reclaim_state; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 2225 | struct zoneref *z; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2226 | struct zone *zone; |
KOSAKI Motohiro | 22fba33 | 2009-12-14 17:59:10 -0800 | [diff] [blame] | 2227 | unsigned long writeback_threshold; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2228 | |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 2229 | get_mems_allowed(); |
Keika Kobayashi | 873b477 | 2008-07-25 01:48:52 -0700 | [diff] [blame] | 2230 | delayacct_freepages_start(); |
| 2231 | |
KAMEZAWA Hiroyuki | e72e2bd | 2009-01-07 18:08:23 -0800 | [diff] [blame] | 2232 | if (scanning_global_lru(sc)) |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 2233 | count_vm_event(ALLOCSTALL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2234 | |
| 2235 | for (priority = DEF_PRIORITY; priority >= 0; priority--) { |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2236 | sc->nr_scanned = 0; |
Rik van Riel | f7b7fd8 | 2005-11-28 13:44:07 -0800 | [diff] [blame] | 2237 | if (!priority) |
KOSAKI Motohiro | a433658 | 2011-06-15 15:08:13 -0700 | [diff] [blame] | 2238 | disable_swap_token(sc->mem_cgroup); |
Mel Gorman | e0c2327 | 2011-10-31 17:09:33 -0700 | [diff] [blame] | 2239 | if (shrink_zones(priority, zonelist, sc)) |
| 2240 | break; |
| 2241 | |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2242 | /* |
| 2243 | * Don't shrink slabs when reclaiming memory from |
| 2244 | * over limit cgroups |
| 2245 | */ |
KAMEZAWA Hiroyuki | e72e2bd | 2009-01-07 18:08:23 -0800 | [diff] [blame] | 2246 | if (scanning_global_lru(sc)) { |
KOSAKI Motohiro | c6a8a8c | 2010-08-09 17:19:14 -0700 | [diff] [blame] | 2247 | unsigned long lru_pages = 0; |
Mel Gorman | d4debc6 | 2010-08-09 17:19:29 -0700 | [diff] [blame] | 2248 | for_each_zone_zonelist(zone, z, zonelist, |
| 2249 | gfp_zone(sc->gfp_mask)) { |
KOSAKI Motohiro | c6a8a8c | 2010-08-09 17:19:14 -0700 | [diff] [blame] | 2250 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
| 2251 | continue; |
| 2252 | |
| 2253 | lru_pages += zone_reclaimable_pages(zone); |
| 2254 | } |
| 2255 | |
Ying Han | 1495f23 | 2011-05-24 17:12:27 -0700 | [diff] [blame] | 2256 | shrink_slab(shrink, sc->nr_scanned, lru_pages); |
KAMEZAWA Hiroyuki | 91a4547 | 2008-02-07 00:14:29 -0800 | [diff] [blame] | 2257 | if (reclaim_state) { |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2258 | sc->nr_reclaimed += reclaim_state->reclaimed_slab; |
KAMEZAWA Hiroyuki | 91a4547 | 2008-02-07 00:14:29 -0800 | [diff] [blame] | 2259 | reclaim_state->reclaimed_slab = 0; |
| 2260 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2261 | } |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2262 | total_scanned += sc->nr_scanned; |
KOSAKI Motohiro | bb21c7c | 2010-06-04 14:15:05 -0700 | [diff] [blame] | 2263 | if (sc->nr_reclaimed >= sc->nr_to_reclaim) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2264 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2265 | |
| 2266 | /* |
| 2267 | * Try to write back as many pages as we just scanned. This |
| 2268 | * tends to cause slow streaming writers to write data to the |
| 2269 | * disk smoothly, at the dirtying rate, which is nice. But |
| 2270 | * that's undesirable in laptop mode, where we *want* lumpy |
| 2271 | * writeout. So in laptop mode, write out the whole world. |
| 2272 | */ |
KOSAKI Motohiro | 22fba33 | 2009-12-14 17:59:10 -0800 | [diff] [blame] | 2273 | writeback_threshold = sc->nr_to_reclaim + sc->nr_to_reclaim / 2; |
| 2274 | if (total_scanned > writeback_threshold) { |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 2275 | wakeup_flusher_threads(laptop_mode ? 0 : total_scanned, |
| 2276 | WB_REASON_TRY_TO_FREE_PAGES); |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2277 | sc->may_writepage = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2278 | } |
| 2279 | |
| 2280 | /* Take a nap, wait for some writeback to complete */ |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 2281 | if (!sc->hibernation_mode && sc->nr_scanned && |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 2282 | priority < DEF_PRIORITY - 2) { |
| 2283 | struct zone *preferred_zone; |
| 2284 | |
| 2285 | first_zones_zonelist(zonelist, gfp_zone(sc->gfp_mask), |
David Rientjes | f33261d | 2011-01-25 15:07:20 -0800 | [diff] [blame] | 2286 | &cpuset_current_mems_allowed, |
| 2287 | &preferred_zone); |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 2288 | wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/10); |
| 2289 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2290 | } |
KOSAKI Motohiro | bb21c7c | 2010-06-04 14:15:05 -0700 | [diff] [blame] | 2291 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2292 | out: |
Keika Kobayashi | 873b477 | 2008-07-25 01:48:52 -0700 | [diff] [blame] | 2293 | delayacct_freepages_end(); |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 2294 | put_mems_allowed(); |
Keika Kobayashi | 873b477 | 2008-07-25 01:48:52 -0700 | [diff] [blame] | 2295 | |
KOSAKI Motohiro | bb21c7c | 2010-06-04 14:15:05 -0700 | [diff] [blame] | 2296 | if (sc->nr_reclaimed) |
| 2297 | return sc->nr_reclaimed; |
| 2298 | |
KOSAKI Motohiro | 929bea7 | 2011-04-14 15:22:12 -0700 | [diff] [blame] | 2299 | /* |
| 2300 | * As hibernation is going on, kswapd is freezed so that it can't mark |
| 2301 | * the zone into all_unreclaimable. Thus bypassing all_unreclaimable |
| 2302 | * check. |
| 2303 | */ |
| 2304 | if (oom_killer_disabled) |
| 2305 | return 0; |
| 2306 | |
KOSAKI Motohiro | bb21c7c | 2010-06-04 14:15:05 -0700 | [diff] [blame] | 2307 | /* top priority shrink_zones still had more to do? don't OOM, then */ |
Minchan Kim | d190836 | 2010-09-22 13:05:01 -0700 | [diff] [blame] | 2308 | if (scanning_global_lru(sc) && !all_unreclaimable(zonelist, sc)) |
KOSAKI Motohiro | bb21c7c | 2010-06-04 14:15:05 -0700 | [diff] [blame] | 2309 | return 1; |
| 2310 | |
| 2311 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2312 | } |
| 2313 | |
Mel Gorman | dac1d27 | 2008-04-28 02:12:12 -0700 | [diff] [blame] | 2314 | unsigned long try_to_free_pages(struct zonelist *zonelist, int order, |
KAMEZAWA Hiroyuki | 327c0e9 | 2009-03-31 15:23:31 -0700 | [diff] [blame] | 2315 | gfp_t gfp_mask, nodemask_t *nodemask) |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2316 | { |
Mel Gorman | 33906bc | 2010-08-09 17:19:16 -0700 | [diff] [blame] | 2317 | unsigned long nr_reclaimed; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2318 | struct scan_control sc = { |
| 2319 | .gfp_mask = gfp_mask, |
| 2320 | .may_writepage = !laptop_mode, |
KOSAKI Motohiro | 22fba33 | 2009-12-14 17:59:10 -0800 | [diff] [blame] | 2321 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
Johannes Weiner | a6dc60f8 | 2009-03-31 15:19:30 -0700 | [diff] [blame] | 2322 | .may_unmap = 1, |
KOSAKI Motohiro | 2e2e425 | 2009-04-21 12:24:57 -0700 | [diff] [blame] | 2323 | .may_swap = 1, |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2324 | .order = order, |
| 2325 | .mem_cgroup = NULL, |
KAMEZAWA Hiroyuki | 327c0e9 | 2009-03-31 15:23:31 -0700 | [diff] [blame] | 2326 | .nodemask = nodemask, |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2327 | }; |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 2328 | struct shrink_control shrink = { |
| 2329 | .gfp_mask = sc.gfp_mask, |
| 2330 | }; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2331 | |
Mel Gorman | 33906bc | 2010-08-09 17:19:16 -0700 | [diff] [blame] | 2332 | trace_mm_vmscan_direct_reclaim_begin(order, |
| 2333 | sc.may_writepage, |
| 2334 | gfp_mask); |
| 2335 | |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 2336 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink); |
Mel Gorman | 33906bc | 2010-08-09 17:19:16 -0700 | [diff] [blame] | 2337 | |
| 2338 | trace_mm_vmscan_direct_reclaim_end(nr_reclaimed); |
| 2339 | |
| 2340 | return nr_reclaimed; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2341 | } |
| 2342 | |
Balbir Singh | 00f0b82 | 2008-03-04 14:28:39 -0800 | [diff] [blame] | 2343 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2344 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2345 | unsigned long mem_cgroup_shrink_node_zone(struct mem_cgroup *mem, |
| 2346 | gfp_t gfp_mask, bool noswap, |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 2347 | struct zone *zone, |
| 2348 | unsigned long *nr_scanned) |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2349 | { |
| 2350 | struct scan_control sc = { |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 2351 | .nr_scanned = 0, |
KOSAKI Motohiro | b8f5c56 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2352 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2353 | .may_writepage = !laptop_mode, |
| 2354 | .may_unmap = 1, |
| 2355 | .may_swap = !noswap, |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2356 | .order = 0, |
| 2357 | .mem_cgroup = mem, |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2358 | }; |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 2359 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2360 | sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | |
| 2361 | (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK); |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 2362 | |
| 2363 | trace_mm_vmscan_memcg_softlimit_reclaim_begin(0, |
| 2364 | sc.may_writepage, |
| 2365 | sc.gfp_mask); |
| 2366 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2367 | /* |
| 2368 | * NOTE: Although we can get the priority field, using it |
| 2369 | * here is not a good idea, since it limits the pages we can scan. |
| 2370 | * if we don't reclaim here, the shrink_zone from balance_pgdat |
| 2371 | * will pick up pages from other mem cgroup's as well. We hack |
| 2372 | * the priority and make it zero. |
| 2373 | */ |
| 2374 | shrink_zone(0, zone, &sc); |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 2375 | |
| 2376 | trace_mm_vmscan_memcg_softlimit_reclaim_end(sc.nr_reclaimed); |
| 2377 | |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 2378 | *nr_scanned = sc.nr_scanned; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2379 | return sc.nr_reclaimed; |
| 2380 | } |
| 2381 | |
Balbir Singh | e1a1cd5 | 2008-02-07 00:14:02 -0800 | [diff] [blame] | 2382 | unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *mem_cont, |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 2383 | gfp_t gfp_mask, |
Johannes Weiner | 185efc0 | 2011-09-14 16:21:58 -0700 | [diff] [blame] | 2384 | bool noswap) |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2385 | { |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2386 | struct zonelist *zonelist; |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 2387 | unsigned long nr_reclaimed; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 2388 | int nid; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2389 | struct scan_control sc = { |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2390 | .may_writepage = !laptop_mode, |
Johannes Weiner | a6dc60f8 | 2009-03-31 15:19:30 -0700 | [diff] [blame] | 2391 | .may_unmap = 1, |
KOSAKI Motohiro | 2e2e425 | 2009-04-21 12:24:57 -0700 | [diff] [blame] | 2392 | .may_swap = !noswap, |
KOSAKI Motohiro | 22fba33 | 2009-12-14 17:59:10 -0800 | [diff] [blame] | 2393 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2394 | .order = 0, |
| 2395 | .mem_cgroup = mem_cont, |
KAMEZAWA Hiroyuki | 327c0e9 | 2009-03-31 15:23:31 -0700 | [diff] [blame] | 2396 | .nodemask = NULL, /* we don't care the placement */ |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 2397 | .gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | |
| 2398 | (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK), |
| 2399 | }; |
| 2400 | struct shrink_control shrink = { |
| 2401 | .gfp_mask = sc.gfp_mask, |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2402 | }; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2403 | |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 2404 | /* |
| 2405 | * Unlike direct reclaim via alloc_pages(), memcg's reclaim doesn't |
| 2406 | * take care of from where we get pages. So the node where we start the |
| 2407 | * scan does not need to be the current node. |
| 2408 | */ |
| 2409 | nid = mem_cgroup_select_victim_node(mem_cont); |
| 2410 | |
| 2411 | zonelist = NODE_DATA(nid)->node_zonelists; |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 2412 | |
| 2413 | trace_mm_vmscan_memcg_reclaim_begin(0, |
| 2414 | sc.may_writepage, |
| 2415 | sc.gfp_mask); |
| 2416 | |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 2417 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink); |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 2418 | |
| 2419 | trace_mm_vmscan_memcg_reclaim_end(nr_reclaimed); |
| 2420 | |
| 2421 | return nr_reclaimed; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2422 | } |
| 2423 | #endif |
| 2424 | |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2425 | /* |
| 2426 | * pgdat_balanced is used when checking if a node is balanced for high-order |
| 2427 | * allocations. Only zones that meet watermarks and are in a zone allowed |
| 2428 | * by the callers classzone_idx are added to balanced_pages. The total of |
| 2429 | * balanced pages must be at least 25% of the zones allowed by classzone_idx |
| 2430 | * for the node to be considered balanced. Forcing all zones to be balanced |
| 2431 | * for high orders can cause excessive reclaim when there are imbalanced zones. |
| 2432 | * The choice of 25% is due to |
| 2433 | * o a 16M DMA zone that is balanced will not balance a zone on any |
| 2434 | * reasonable sized machine |
| 2435 | * o On all other machines, the top zone must be at least a reasonable |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 2436 | * percentage of the middle zones. For example, on 32-bit x86, highmem |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2437 | * would need to be at least 256M for it to be balance a whole node. |
| 2438 | * Similarly, on x86-64 the Normal zone would need to be at least 1G |
| 2439 | * to balance a node on its own. These seemed like reasonable ratios. |
| 2440 | */ |
| 2441 | static bool pgdat_balanced(pg_data_t *pgdat, unsigned long balanced_pages, |
| 2442 | int classzone_idx) |
| 2443 | { |
| 2444 | unsigned long present_pages = 0; |
| 2445 | int i; |
| 2446 | |
| 2447 | for (i = 0; i <= classzone_idx; i++) |
| 2448 | present_pages += pgdat->node_zones[i].present_pages; |
| 2449 | |
Shaohua Li | 4746efd | 2011-07-19 08:49:26 -0700 | [diff] [blame] | 2450 | /* A special case here: if zone has no page, we think it's balanced */ |
| 2451 | return balanced_pages >= (present_pages >> 2); |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2452 | } |
| 2453 | |
Mel Gorman | f50de2d | 2009-12-14 17:58:53 -0800 | [diff] [blame] | 2454 | /* is kswapd sleeping prematurely? */ |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2455 | static bool sleeping_prematurely(pg_data_t *pgdat, int order, long remaining, |
| 2456 | int classzone_idx) |
Mel Gorman | f50de2d | 2009-12-14 17:58:53 -0800 | [diff] [blame] | 2457 | { |
KOSAKI Motohiro | bb3ab59 | 2009-12-14 17:58:55 -0800 | [diff] [blame] | 2458 | int i; |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2459 | unsigned long balanced = 0; |
| 2460 | bool all_zones_ok = true; |
Mel Gorman | f50de2d | 2009-12-14 17:58:53 -0800 | [diff] [blame] | 2461 | |
| 2462 | /* If a direct reclaimer woke kswapd within HZ/10, it's premature */ |
| 2463 | if (remaining) |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2464 | return true; |
Mel Gorman | f50de2d | 2009-12-14 17:58:53 -0800 | [diff] [blame] | 2465 | |
Mel Gorman | 0abdee2 | 2011-01-13 15:46:22 -0800 | [diff] [blame] | 2466 | /* Check the watermark levels */ |
Mel Gorman | 08951e5 | 2011-07-08 15:39:36 -0700 | [diff] [blame] | 2467 | for (i = 0; i <= classzone_idx; i++) { |
KOSAKI Motohiro | bb3ab59 | 2009-12-14 17:58:55 -0800 | [diff] [blame] | 2468 | struct zone *zone = pgdat->node_zones + i; |
| 2469 | |
| 2470 | if (!populated_zone(zone)) |
| 2471 | continue; |
| 2472 | |
Mel Gorman | 355b09c | 2011-01-13 15:46:24 -0800 | [diff] [blame] | 2473 | /* |
| 2474 | * balance_pgdat() skips over all_unreclaimable after |
| 2475 | * DEF_PRIORITY. Effectively, it considers them balanced so |
| 2476 | * they must be considered balanced here as well if kswapd |
| 2477 | * is to sleep |
| 2478 | */ |
| 2479 | if (zone->all_unreclaimable) { |
| 2480 | balanced += zone->present_pages; |
KOSAKI Motohiro | de3fab3 | 2010-01-15 17:01:25 -0800 | [diff] [blame] | 2481 | continue; |
Mel Gorman | 355b09c | 2011-01-13 15:46:24 -0800 | [diff] [blame] | 2482 | } |
KOSAKI Motohiro | de3fab3 | 2010-01-15 17:01:25 -0800 | [diff] [blame] | 2483 | |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 2484 | if (!zone_watermark_ok_safe(zone, order, high_wmark_pages(zone), |
Mel Gorman | da175d0 | 2011-07-08 15:39:39 -0700 | [diff] [blame] | 2485 | i, 0)) |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2486 | all_zones_ok = false; |
| 2487 | else |
| 2488 | balanced += zone->present_pages; |
KOSAKI Motohiro | bb3ab59 | 2009-12-14 17:58:55 -0800 | [diff] [blame] | 2489 | } |
Mel Gorman | f50de2d | 2009-12-14 17:58:53 -0800 | [diff] [blame] | 2490 | |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2491 | /* |
| 2492 | * For high-order requests, the balanced zones must contain at least |
| 2493 | * 25% of the nodes pages for kswapd to sleep. For order-0, all zones |
| 2494 | * must be balanced |
| 2495 | */ |
| 2496 | if (order) |
Johannes Weiner | afc7e326 | 2011-05-24 17:11:09 -0700 | [diff] [blame] | 2497 | return !pgdat_balanced(pgdat, balanced, classzone_idx); |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2498 | else |
| 2499 | return !all_zones_ok; |
Mel Gorman | f50de2d | 2009-12-14 17:58:53 -0800 | [diff] [blame] | 2500 | } |
| 2501 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2502 | /* |
| 2503 | * For kswapd, balance_pgdat() will work across all this node's zones until |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 2504 | * they are all at high_wmark_pages(zone). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2505 | * |
Mel Gorman | 0abdee2 | 2011-01-13 15:46:22 -0800 | [diff] [blame] | 2506 | * Returns the final order kswapd was reclaiming at |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2507 | * |
| 2508 | * There is special handling here for zones which are full of pinned pages. |
| 2509 | * This can happen if the pages are all mlocked, or if they are all used by |
| 2510 | * device drivers (say, ZONE_DMA). Or if they are all in use by hugetlb. |
| 2511 | * What we do is to detect the case where all pages in the zone have been |
| 2512 | * scanned twice and there has been zero successful reclaim. Mark the zone as |
| 2513 | * dead and from now on, only perform a short scan. Basically we're polling |
| 2514 | * the zone for when the problem goes away. |
| 2515 | * |
| 2516 | * kswapd scans the zones in the highmem->normal->dma direction. It skips |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 2517 | * zones which have free_pages > high_wmark_pages(zone), but once a zone is |
| 2518 | * found to have free_pages <= high_wmark_pages(zone), we scan that zone and the |
| 2519 | * lower zones regardless of the number of free pages in the lower zones. This |
| 2520 | * interoperates with the page allocator fallback scheme to ensure that aging |
| 2521 | * of pages is balanced across the zones. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2522 | */ |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2523 | static unsigned long balance_pgdat(pg_data_t *pgdat, int order, |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2524 | int *classzone_idx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2525 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2526 | int all_zones_ok; |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2527 | unsigned long balanced; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2528 | int priority; |
| 2529 | int i; |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2530 | int end_zone = 0; /* Inclusive. 0 = ZONE_DMA */ |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 2531 | unsigned long total_scanned; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2532 | struct reclaim_state *reclaim_state = current->reclaim_state; |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 2533 | unsigned long nr_soft_reclaimed; |
| 2534 | unsigned long nr_soft_scanned; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2535 | struct scan_control sc = { |
| 2536 | .gfp_mask = GFP_KERNEL, |
Johannes Weiner | a6dc60f8 | 2009-03-31 15:19:30 -0700 | [diff] [blame] | 2537 | .may_unmap = 1, |
KOSAKI Motohiro | 2e2e425 | 2009-04-21 12:24:57 -0700 | [diff] [blame] | 2538 | .may_swap = 1, |
KOSAKI Motohiro | 22fba33 | 2009-12-14 17:59:10 -0800 | [diff] [blame] | 2539 | /* |
| 2540 | * kswapd doesn't want to be bailed out while reclaim. because |
| 2541 | * we want to put equal scanning pressure on each zone. |
| 2542 | */ |
| 2543 | .nr_to_reclaim = ULONG_MAX, |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 2544 | .order = order, |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2545 | .mem_cgroup = NULL, |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2546 | }; |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 2547 | struct shrink_control shrink = { |
| 2548 | .gfp_mask = sc.gfp_mask, |
| 2549 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2550 | loop_again: |
| 2551 | total_scanned = 0; |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2552 | sc.nr_reclaimed = 0; |
Christoph Lameter | c0bbbc7 | 2006-06-11 15:22:26 -0700 | [diff] [blame] | 2553 | sc.may_writepage = !laptop_mode; |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 2554 | count_vm_event(PAGEOUTRUN); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2555 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2556 | for (priority = DEF_PRIORITY; priority >= 0; priority--) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2557 | unsigned long lru_pages = 0; |
KOSAKI Motohiro | bb3ab59 | 2009-12-14 17:58:55 -0800 | [diff] [blame] | 2558 | int has_under_min_watermark_zone = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2559 | |
Rik van Riel | f7b7fd8 | 2005-11-28 13:44:07 -0800 | [diff] [blame] | 2560 | /* The swap token gets in the way of swapout... */ |
| 2561 | if (!priority) |
KOSAKI Motohiro | a433658 | 2011-06-15 15:08:13 -0700 | [diff] [blame] | 2562 | disable_swap_token(NULL); |
Rik van Riel | f7b7fd8 | 2005-11-28 13:44:07 -0800 | [diff] [blame] | 2563 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2564 | all_zones_ok = 1; |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2565 | balanced = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2566 | |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2567 | /* |
| 2568 | * Scan in the highmem->dma direction for the highest |
| 2569 | * zone which needs scanning |
| 2570 | */ |
| 2571 | for (i = pgdat->nr_zones - 1; i >= 0; i--) { |
| 2572 | struct zone *zone = pgdat->node_zones + i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2573 | |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2574 | if (!populated_zone(zone)) |
| 2575 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2576 | |
KOSAKI Motohiro | 93e4a89 | 2010-03-05 13:41:55 -0800 | [diff] [blame] | 2577 | if (zone->all_unreclaimable && priority != DEF_PRIORITY) |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2578 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2579 | |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 2580 | /* |
| 2581 | * Do some background aging of the anon list, to give |
| 2582 | * pages a chance to be referenced before reclaiming. |
| 2583 | */ |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 2584 | if (inactive_anon_is_low(zone, &sc)) |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 2585 | shrink_active_list(SWAP_CLUSTER_MAX, zone, |
| 2586 | &sc, priority, 0); |
| 2587 | |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 2588 | if (!zone_watermark_ok_safe(zone, order, |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 2589 | high_wmark_pages(zone), 0, 0)) { |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2590 | end_zone = i; |
Andrew Morton | e1dbeda | 2006-12-06 20:32:01 -0800 | [diff] [blame] | 2591 | break; |
Shaohua Li | 439423f | 2011-08-25 15:59:12 -0700 | [diff] [blame] | 2592 | } else { |
| 2593 | /* If balanced, clear the congested flag */ |
| 2594 | zone_clear_flag(zone, ZONE_CONGESTED); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2595 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2596 | } |
Andrew Morton | e1dbeda | 2006-12-06 20:32:01 -0800 | [diff] [blame] | 2597 | if (i < 0) |
| 2598 | goto out; |
| 2599 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2600 | for (i = 0; i <= end_zone; i++) { |
| 2601 | struct zone *zone = pgdat->node_zones + i; |
| 2602 | |
Wu Fengguang | adea02a | 2009-09-21 17:01:42 -0700 | [diff] [blame] | 2603 | lru_pages += zone_reclaimable_pages(zone); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2604 | } |
| 2605 | |
| 2606 | /* |
| 2607 | * Now scan the zone in the dma->highmem direction, stopping |
| 2608 | * at the last zone which needs scanning. |
| 2609 | * |
| 2610 | * We do this because the page allocator works in the opposite |
| 2611 | * direction. This prevents the page allocator from allocating |
| 2612 | * pages behind kswapd's direction of progress, which would |
| 2613 | * cause too much scanning of the lower zones. |
| 2614 | */ |
| 2615 | for (i = 0; i <= end_zone; i++) { |
| 2616 | struct zone *zone = pgdat->node_zones + i; |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 2617 | int nr_slab; |
Mel Gorman | 8afdcec | 2011-03-22 16:33:04 -0700 | [diff] [blame] | 2618 | unsigned long balance_gap; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2619 | |
Con Kolivas | f3fe651 | 2006-01-06 00:11:15 -0800 | [diff] [blame] | 2620 | if (!populated_zone(zone)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2621 | continue; |
| 2622 | |
KOSAKI Motohiro | 93e4a89 | 2010-03-05 13:41:55 -0800 | [diff] [blame] | 2623 | if (zone->all_unreclaimable && priority != DEF_PRIORITY) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2624 | continue; |
| 2625 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2626 | sc.nr_scanned = 0; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2627 | |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 2628 | nr_soft_scanned = 0; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2629 | /* |
| 2630 | * Call soft limit reclaim before calling shrink_zone. |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2631 | */ |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 2632 | nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone, |
| 2633 | order, sc.gfp_mask, |
| 2634 | &nr_soft_scanned); |
| 2635 | sc.nr_reclaimed += nr_soft_reclaimed; |
| 2636 | total_scanned += nr_soft_scanned; |
KOSAKI Motohiro | 00918b6 | 2010-08-10 18:03:05 -0700 | [diff] [blame] | 2637 | |
Rik van Riel | 32a4330 | 2007-10-16 01:24:50 -0700 | [diff] [blame] | 2638 | /* |
Mel Gorman | 8afdcec | 2011-03-22 16:33:04 -0700 | [diff] [blame] | 2639 | * We put equal pressure on every zone, unless |
| 2640 | * one zone has way too many pages free |
| 2641 | * already. The "too many pages" is defined |
| 2642 | * as the high wmark plus a "gap" where the |
| 2643 | * gap is either the low watermark or 1% |
| 2644 | * of the zone, whichever is smaller. |
Rik van Riel | 32a4330 | 2007-10-16 01:24:50 -0700 | [diff] [blame] | 2645 | */ |
Mel Gorman | 8afdcec | 2011-03-22 16:33:04 -0700 | [diff] [blame] | 2646 | balance_gap = min(low_wmark_pages(zone), |
| 2647 | (zone->present_pages + |
| 2648 | KSWAPD_ZONE_BALANCE_GAP_RATIO-1) / |
| 2649 | KSWAPD_ZONE_BALANCE_GAP_RATIO); |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 2650 | if (!zone_watermark_ok_safe(zone, order, |
Mel Gorman | 8afdcec | 2011-03-22 16:33:04 -0700 | [diff] [blame] | 2651 | high_wmark_pages(zone) + balance_gap, |
Mel Gorman | d7868da | 2011-07-08 15:39:38 -0700 | [diff] [blame] | 2652 | end_zone, 0)) { |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2653 | shrink_zone(priority, zone, &sc); |
Andrea Arcangeli | 5a03b05 | 2011-01-13 15:47:11 -0800 | [diff] [blame] | 2654 | |
Mel Gorman | d7868da | 2011-07-08 15:39:38 -0700 | [diff] [blame] | 2655 | reclaim_state->reclaimed_slab = 0; |
| 2656 | nr_slab = shrink_slab(&shrink, sc.nr_scanned, lru_pages); |
| 2657 | sc.nr_reclaimed += reclaim_state->reclaimed_slab; |
| 2658 | total_scanned += sc.nr_scanned; |
| 2659 | |
| 2660 | if (nr_slab == 0 && !zone_reclaimable(zone)) |
| 2661 | zone->all_unreclaimable = 1; |
| 2662 | } |
| 2663 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2664 | /* |
| 2665 | * If we've done a decent amount of scanning and |
| 2666 | * the reclaim ratio is low, start doing writepage |
| 2667 | * even in laptop mode |
| 2668 | */ |
| 2669 | if (total_scanned > SWAP_CLUSTER_MAX * 2 && |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2670 | total_scanned > sc.nr_reclaimed + sc.nr_reclaimed / 2) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2671 | sc.may_writepage = 1; |
KOSAKI Motohiro | bb3ab59 | 2009-12-14 17:58:55 -0800 | [diff] [blame] | 2672 | |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2673 | if (zone->all_unreclaimable) { |
| 2674 | if (end_zone && end_zone == i) |
| 2675 | end_zone--; |
Mel Gorman | d7868da | 2011-07-08 15:39:38 -0700 | [diff] [blame] | 2676 | continue; |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2677 | } |
Mel Gorman | d7868da | 2011-07-08 15:39:38 -0700 | [diff] [blame] | 2678 | |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 2679 | if (!zone_watermark_ok_safe(zone, order, |
Minchan Kim | 45973d7 | 2010-03-05 13:41:45 -0800 | [diff] [blame] | 2680 | high_wmark_pages(zone), end_zone, 0)) { |
| 2681 | all_zones_ok = 0; |
| 2682 | /* |
| 2683 | * We are still under min water mark. This |
| 2684 | * means that we have a GFP_ATOMIC allocation |
| 2685 | * failure risk. Hurry up! |
| 2686 | */ |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 2687 | if (!zone_watermark_ok_safe(zone, order, |
Minchan Kim | 45973d7 | 2010-03-05 13:41:45 -0800 | [diff] [blame] | 2688 | min_wmark_pages(zone), end_zone, 0)) |
| 2689 | has_under_min_watermark_zone = 1; |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 2690 | } else { |
| 2691 | /* |
| 2692 | * If a zone reaches its high watermark, |
| 2693 | * consider it to be no longer congested. It's |
| 2694 | * possible there are dirty pages backed by |
| 2695 | * congested BDIs but as pressure is relieved, |
| 2696 | * spectulatively avoid congestion waits |
| 2697 | */ |
| 2698 | zone_clear_flag(zone, ZONE_CONGESTED); |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2699 | if (i <= *classzone_idx) |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2700 | balanced += zone->present_pages; |
Minchan Kim | 45973d7 | 2010-03-05 13:41:45 -0800 | [diff] [blame] | 2701 | } |
KOSAKI Motohiro | bb3ab59 | 2009-12-14 17:58:55 -0800 | [diff] [blame] | 2702 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2703 | } |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2704 | if (all_zones_ok || (order && pgdat_balanced(pgdat, balanced, *classzone_idx))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2705 | break; /* kswapd: all done */ |
| 2706 | /* |
| 2707 | * OK, kswapd is getting into trouble. Take a nap, then take |
| 2708 | * another pass across the zones. |
| 2709 | */ |
KOSAKI Motohiro | bb3ab59 | 2009-12-14 17:58:55 -0800 | [diff] [blame] | 2710 | if (total_scanned && (priority < DEF_PRIORITY - 2)) { |
| 2711 | if (has_under_min_watermark_zone) |
| 2712 | count_vm_event(KSWAPD_SKIP_CONGESTION_WAIT); |
| 2713 | else |
| 2714 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
| 2715 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2716 | |
| 2717 | /* |
| 2718 | * We do this so kswapd doesn't build up large priorities for |
| 2719 | * example when it is freeing in parallel with allocators. It |
| 2720 | * matches the direct reclaim path behaviour in terms of impact |
| 2721 | * on zone->*_priority. |
| 2722 | */ |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2723 | if (sc.nr_reclaimed >= SWAP_CLUSTER_MAX) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2724 | break; |
| 2725 | } |
| 2726 | out: |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2727 | |
| 2728 | /* |
| 2729 | * order-0: All zones must meet high watermark for a balanced node |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2730 | * high-order: Balanced zones must make up at least 25% of the node |
| 2731 | * for the node to be balanced |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2732 | */ |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2733 | if (!(all_zones_ok || (order && pgdat_balanced(pgdat, balanced, *classzone_idx)))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2734 | cond_resched(); |
Rafael J. Wysocki | 8357376 | 2006-12-06 20:34:18 -0800 | [diff] [blame] | 2735 | |
| 2736 | try_to_freeze(); |
| 2737 | |
KOSAKI Motohiro | 73ce02e | 2009-01-06 14:40:33 -0800 | [diff] [blame] | 2738 | /* |
| 2739 | * Fragmentation may mean that the system cannot be |
| 2740 | * rebalanced for high-order allocations in all zones. |
| 2741 | * At this point, if nr_reclaimed < SWAP_CLUSTER_MAX, |
| 2742 | * it means the zones have been fully scanned and are still |
| 2743 | * not balanced. For high-order allocations, there is |
| 2744 | * little point trying all over again as kswapd may |
| 2745 | * infinite loop. |
| 2746 | * |
| 2747 | * Instead, recheck all watermarks at order-0 as they |
| 2748 | * are the most important. If watermarks are ok, kswapd will go |
| 2749 | * back to sleep. High-order users can still perform direct |
| 2750 | * reclaim if they wish. |
| 2751 | */ |
| 2752 | if (sc.nr_reclaimed < SWAP_CLUSTER_MAX) |
| 2753 | order = sc.order = 0; |
| 2754 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2755 | goto loop_again; |
| 2756 | } |
| 2757 | |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2758 | /* |
| 2759 | * If kswapd was reclaiming at a higher order, it has the option of |
| 2760 | * sleeping without all zones being balanced. Before it does, it must |
| 2761 | * ensure that the watermarks for order-0 on *all* zones are met and |
| 2762 | * that the congestion flags are cleared. The congestion flag must |
| 2763 | * be cleared as kswapd is the only mechanism that clears the flag |
| 2764 | * and it is potentially going to sleep here. |
| 2765 | */ |
| 2766 | if (order) { |
| 2767 | for (i = 0; i <= end_zone; i++) { |
| 2768 | struct zone *zone = pgdat->node_zones + i; |
| 2769 | |
| 2770 | if (!populated_zone(zone)) |
| 2771 | continue; |
| 2772 | |
| 2773 | if (zone->all_unreclaimable && priority != DEF_PRIORITY) |
| 2774 | continue; |
| 2775 | |
| 2776 | /* Confirm the zone is balanced for order-0 */ |
| 2777 | if (!zone_watermark_ok(zone, 0, |
| 2778 | high_wmark_pages(zone), 0, 0)) { |
| 2779 | order = sc.order = 0; |
| 2780 | goto loop_again; |
| 2781 | } |
| 2782 | |
| 2783 | /* If balanced, clear the congested flag */ |
| 2784 | zone_clear_flag(zone, ZONE_CONGESTED); |
Shaohua Li | 16fb951 | 2011-10-31 17:08:02 -0700 | [diff] [blame] | 2785 | if (i <= *classzone_idx) |
| 2786 | balanced += zone->present_pages; |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2787 | } |
| 2788 | } |
| 2789 | |
Mel Gorman | 0abdee2 | 2011-01-13 15:46:22 -0800 | [diff] [blame] | 2790 | /* |
| 2791 | * Return the order we were reclaiming at so sleeping_prematurely() |
| 2792 | * makes a decision on the order we were last reclaiming at. However, |
| 2793 | * if another caller entered the allocator slow path while kswapd |
| 2794 | * was awake, order will remain at the higher level |
| 2795 | */ |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2796 | *classzone_idx = end_zone; |
Mel Gorman | 0abdee2 | 2011-01-13 15:46:22 -0800 | [diff] [blame] | 2797 | return order; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2798 | } |
| 2799 | |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2800 | static void kswapd_try_to_sleep(pg_data_t *pgdat, int order, int classzone_idx) |
KOSAKI Motohiro | f0bc0a6 | 2011-01-13 15:45:50 -0800 | [diff] [blame] | 2801 | { |
| 2802 | long remaining = 0; |
| 2803 | DEFINE_WAIT(wait); |
| 2804 | |
| 2805 | if (freezing(current) || kthread_should_stop()) |
| 2806 | return; |
| 2807 | |
| 2808 | prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); |
| 2809 | |
| 2810 | /* Try to sleep for a short interval */ |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2811 | if (!sleeping_prematurely(pgdat, order, remaining, classzone_idx)) { |
KOSAKI Motohiro | f0bc0a6 | 2011-01-13 15:45:50 -0800 | [diff] [blame] | 2812 | remaining = schedule_timeout(HZ/10); |
| 2813 | finish_wait(&pgdat->kswapd_wait, &wait); |
| 2814 | prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); |
| 2815 | } |
| 2816 | |
| 2817 | /* |
| 2818 | * After a short sleep, check if it was a premature sleep. If not, then |
| 2819 | * go fully to sleep until explicitly woken up. |
| 2820 | */ |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2821 | if (!sleeping_prematurely(pgdat, order, remaining, classzone_idx)) { |
KOSAKI Motohiro | f0bc0a6 | 2011-01-13 15:45:50 -0800 | [diff] [blame] | 2822 | trace_mm_vmscan_kswapd_sleep(pgdat->node_id); |
| 2823 | |
| 2824 | /* |
| 2825 | * vmstat counters are not perfectly accurate and the estimated |
| 2826 | * value for counters such as NR_FREE_PAGES can deviate from the |
| 2827 | * true value by nr_online_cpus * threshold. To avoid the zone |
| 2828 | * watermarks being breached while under pressure, we reduce the |
| 2829 | * per-cpu vmstat threshold while kswapd is awake and restore |
| 2830 | * them before going back to sleep. |
| 2831 | */ |
| 2832 | set_pgdat_percpu_threshold(pgdat, calculate_normal_threshold); |
| 2833 | schedule(); |
| 2834 | set_pgdat_percpu_threshold(pgdat, calculate_pressure_threshold); |
| 2835 | } else { |
| 2836 | if (remaining) |
| 2837 | count_vm_event(KSWAPD_LOW_WMARK_HIT_QUICKLY); |
| 2838 | else |
| 2839 | count_vm_event(KSWAPD_HIGH_WMARK_HIT_QUICKLY); |
| 2840 | } |
| 2841 | finish_wait(&pgdat->kswapd_wait, &wait); |
| 2842 | } |
| 2843 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2844 | /* |
| 2845 | * The background pageout daemon, started as a kernel thread |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2846 | * from the init process. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2847 | * |
| 2848 | * This basically trickles out pages so that we have _some_ |
| 2849 | * free memory available even if there is no other activity |
| 2850 | * that frees anything up. This is needed for things like routing |
| 2851 | * etc, where we otherwise might have all activity going on in |
| 2852 | * asynchronous contexts that cannot page things out. |
| 2853 | * |
| 2854 | * If there are applications that are active memory-allocators |
| 2855 | * (most normal use), this basically shouldn't matter. |
| 2856 | */ |
| 2857 | static int kswapd(void *p) |
| 2858 | { |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2859 | unsigned long order, new_order; |
Alex,Shi | d2ebd0f6 | 2011-10-31 17:08:39 -0700 | [diff] [blame] | 2860 | unsigned balanced_order; |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2861 | int classzone_idx, new_classzone_idx; |
Alex,Shi | d2ebd0f6 | 2011-10-31 17:08:39 -0700 | [diff] [blame] | 2862 | int balanced_classzone_idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2863 | pg_data_t *pgdat = (pg_data_t*)p; |
| 2864 | struct task_struct *tsk = current; |
KOSAKI Motohiro | f0bc0a6 | 2011-01-13 15:45:50 -0800 | [diff] [blame] | 2865 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2866 | struct reclaim_state reclaim_state = { |
| 2867 | .reclaimed_slab = 0, |
| 2868 | }; |
Rusty Russell | a70f730 | 2009-03-13 14:49:46 +1030 | [diff] [blame] | 2869 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2870 | |
Nick Piggin | cf40bd1 | 2009-01-21 08:12:39 +0100 | [diff] [blame] | 2871 | lockdep_set_current_reclaim_state(GFP_KERNEL); |
| 2872 | |
Rusty Russell | 174596a | 2009-01-01 10:12:29 +1030 | [diff] [blame] | 2873 | if (!cpumask_empty(cpumask)) |
Mike Travis | c5f59f0 | 2008-04-04 18:11:10 -0700 | [diff] [blame] | 2874 | set_cpus_allowed_ptr(tsk, cpumask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2875 | current->reclaim_state = &reclaim_state; |
| 2876 | |
| 2877 | /* |
| 2878 | * Tell the memory management that we're a "memory allocator", |
| 2879 | * and that if we need more memory we should get access to it |
| 2880 | * regardless (see "__alloc_pages()"). "kswapd" should |
| 2881 | * never get caught in the normal page freeing logic. |
| 2882 | * |
| 2883 | * (Kswapd normally doesn't need memory anyway, but sometimes |
| 2884 | * you need a small amount of memory in order to be able to |
| 2885 | * page out something else, and this flag essentially protects |
| 2886 | * us from recursively trying to free more memory as we're |
| 2887 | * trying to free the first piece of memory in the first place). |
| 2888 | */ |
Christoph Lameter | 930d915 | 2006-01-08 01:00:47 -0800 | [diff] [blame] | 2889 | tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD; |
Rafael J. Wysocki | 8314418 | 2007-07-17 04:03:35 -0700 | [diff] [blame] | 2890 | set_freezable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2891 | |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2892 | order = new_order = 0; |
Alex,Shi | d2ebd0f6 | 2011-10-31 17:08:39 -0700 | [diff] [blame] | 2893 | balanced_order = 0; |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2894 | classzone_idx = new_classzone_idx = pgdat->nr_zones - 1; |
Alex,Shi | d2ebd0f6 | 2011-10-31 17:08:39 -0700 | [diff] [blame] | 2895 | balanced_classzone_idx = classzone_idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2896 | for ( ; ; ) { |
David Rientjes | 8fe23e0 | 2009-12-14 17:58:33 -0800 | [diff] [blame] | 2897 | int ret; |
Christoph Lameter | 3e1d1d2 | 2005-06-24 23:13:50 -0700 | [diff] [blame] | 2898 | |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2899 | /* |
| 2900 | * If the last balance_pgdat was unsuccessful it's unlikely a |
| 2901 | * new request of a similar or harder type will succeed soon |
| 2902 | * so consider going to sleep on the basis we reclaimed at |
| 2903 | */ |
Alex,Shi | d2ebd0f6 | 2011-10-31 17:08:39 -0700 | [diff] [blame] | 2904 | if (balanced_classzone_idx >= new_classzone_idx && |
| 2905 | balanced_order == new_order) { |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2906 | new_order = pgdat->kswapd_max_order; |
| 2907 | new_classzone_idx = pgdat->classzone_idx; |
| 2908 | pgdat->kswapd_max_order = 0; |
| 2909 | pgdat->classzone_idx = pgdat->nr_zones - 1; |
| 2910 | } |
| 2911 | |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2912 | if (order < new_order || classzone_idx > new_classzone_idx) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2913 | /* |
| 2914 | * Don't sleep if someone wants a larger 'order' |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2915 | * allocation or has tigher zone constraints |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2916 | */ |
| 2917 | order = new_order; |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2918 | classzone_idx = new_classzone_idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2919 | } else { |
Alex,Shi | d2ebd0f6 | 2011-10-31 17:08:39 -0700 | [diff] [blame] | 2920 | kswapd_try_to_sleep(pgdat, balanced_order, |
| 2921 | balanced_classzone_idx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2922 | order = pgdat->kswapd_max_order; |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2923 | classzone_idx = pgdat->classzone_idx; |
Alex,Shi | f0dfcde | 2011-10-31 17:08:45 -0700 | [diff] [blame] | 2924 | new_order = order; |
| 2925 | new_classzone_idx = classzone_idx; |
Mel Gorman | 4d40502 | 2011-01-13 15:46:23 -0800 | [diff] [blame] | 2926 | pgdat->kswapd_max_order = 0; |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2927 | pgdat->classzone_idx = pgdat->nr_zones - 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2928 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2929 | |
David Rientjes | 8fe23e0 | 2009-12-14 17:58:33 -0800 | [diff] [blame] | 2930 | ret = try_to_freeze(); |
| 2931 | if (kthread_should_stop()) |
| 2932 | break; |
| 2933 | |
| 2934 | /* |
| 2935 | * We can speed up thawing tasks if we don't call balance_pgdat |
| 2936 | * after returning from the refrigerator |
| 2937 | */ |
Mel Gorman | 33906bc | 2010-08-09 17:19:16 -0700 | [diff] [blame] | 2938 | if (!ret) { |
| 2939 | trace_mm_vmscan_kswapd_wake(pgdat->node_id, order); |
Alex,Shi | d2ebd0f6 | 2011-10-31 17:08:39 -0700 | [diff] [blame] | 2940 | balanced_classzone_idx = classzone_idx; |
| 2941 | balanced_order = balance_pgdat(pgdat, order, |
| 2942 | &balanced_classzone_idx); |
Mel Gorman | 33906bc | 2010-08-09 17:19:16 -0700 | [diff] [blame] | 2943 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2944 | } |
| 2945 | return 0; |
| 2946 | } |
| 2947 | |
| 2948 | /* |
| 2949 | * A zone is low on free memory, so wake its kswapd task to service it. |
| 2950 | */ |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2951 | void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2952 | { |
| 2953 | pg_data_t *pgdat; |
| 2954 | |
Con Kolivas | f3fe651 | 2006-01-06 00:11:15 -0800 | [diff] [blame] | 2955 | if (!populated_zone(zone)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2956 | return; |
| 2957 | |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2958 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2959 | return; |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 2960 | pgdat = zone->zone_pgdat; |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2961 | if (pgdat->kswapd_max_order < order) { |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 2962 | pgdat->kswapd_max_order = order; |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2963 | pgdat->classzone_idx = min(pgdat->classzone_idx, classzone_idx); |
| 2964 | } |
Con Kolivas | 8d0986e | 2005-09-13 01:25:07 -0700 | [diff] [blame] | 2965 | if (!waitqueue_active(&pgdat->kswapd_wait)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2966 | return; |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 2967 | if (zone_watermark_ok_safe(zone, order, low_wmark_pages(zone), 0, 0)) |
| 2968 | return; |
| 2969 | |
| 2970 | trace_mm_vmscan_wakeup_kswapd(pgdat->node_id, zone_idx(zone), order); |
Con Kolivas | 8d0986e | 2005-09-13 01:25:07 -0700 | [diff] [blame] | 2971 | wake_up_interruptible(&pgdat->kswapd_wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2972 | } |
| 2973 | |
Wu Fengguang | adea02a | 2009-09-21 17:01:42 -0700 | [diff] [blame] | 2974 | /* |
| 2975 | * The reclaimable count would be mostly accurate. |
| 2976 | * The less reclaimable pages may be |
| 2977 | * - mlocked pages, which will be moved to unevictable list when encountered |
| 2978 | * - mapped pages, which may require several travels to be reclaimed |
| 2979 | * - dirty pages, which is not "instantly" reclaimable |
| 2980 | */ |
| 2981 | unsigned long global_reclaimable_pages(void) |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2982 | { |
Wu Fengguang | adea02a | 2009-09-21 17:01:42 -0700 | [diff] [blame] | 2983 | int nr; |
| 2984 | |
| 2985 | nr = global_page_state(NR_ACTIVE_FILE) + |
| 2986 | global_page_state(NR_INACTIVE_FILE); |
| 2987 | |
| 2988 | if (nr_swap_pages > 0) |
| 2989 | nr += global_page_state(NR_ACTIVE_ANON) + |
| 2990 | global_page_state(NR_INACTIVE_ANON); |
| 2991 | |
| 2992 | return nr; |
| 2993 | } |
| 2994 | |
| 2995 | unsigned long zone_reclaimable_pages(struct zone *zone) |
| 2996 | { |
| 2997 | int nr; |
| 2998 | |
| 2999 | nr = zone_page_state(zone, NR_ACTIVE_FILE) + |
| 3000 | zone_page_state(zone, NR_INACTIVE_FILE); |
| 3001 | |
| 3002 | if (nr_swap_pages > 0) |
| 3003 | nr += zone_page_state(zone, NR_ACTIVE_ANON) + |
| 3004 | zone_page_state(zone, NR_INACTIVE_ANON); |
| 3005 | |
| 3006 | return nr; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 3007 | } |
| 3008 | |
Rafael J. Wysocki | c6f37f1 | 2009-05-24 22:16:31 +0200 | [diff] [blame] | 3009 | #ifdef CONFIG_HIBERNATION |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3010 | /* |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3011 | * Try to free `nr_to_reclaim' of memory, system-wide, and return the number of |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 3012 | * freed pages. |
| 3013 | * |
| 3014 | * Rather than trying to age LRUs the aim is to preserve the overall |
| 3015 | * LRU order by reclaiming preferentially |
| 3016 | * inactive > active > active referenced > active mapped |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3017 | */ |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3018 | unsigned long shrink_all_memory(unsigned long nr_to_reclaim) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3019 | { |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 3020 | struct reclaim_state reclaim_state; |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 3021 | struct scan_control sc = { |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3022 | .gfp_mask = GFP_HIGHUSER_MOVABLE, |
| 3023 | .may_swap = 1, |
| 3024 | .may_unmap = 1, |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 3025 | .may_writepage = 1, |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3026 | .nr_to_reclaim = nr_to_reclaim, |
| 3027 | .hibernation_mode = 1, |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3028 | .order = 0, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3029 | }; |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 3030 | struct shrink_control shrink = { |
| 3031 | .gfp_mask = sc.gfp_mask, |
| 3032 | }; |
| 3033 | struct zonelist *zonelist = node_zonelist(numa_node_id(), sc.gfp_mask); |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3034 | struct task_struct *p = current; |
| 3035 | unsigned long nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3036 | |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3037 | p->flags |= PF_MEMALLOC; |
| 3038 | lockdep_set_current_reclaim_state(sc.gfp_mask); |
| 3039 | reclaim_state.reclaimed_slab = 0; |
| 3040 | p->reclaim_state = &reclaim_state; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 3041 | |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 3042 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink); |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 3043 | |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3044 | p->reclaim_state = NULL; |
| 3045 | lockdep_clear_current_reclaim_state(); |
| 3046 | p->flags &= ~PF_MEMALLOC; |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 3047 | |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3048 | return nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3049 | } |
Rafael J. Wysocki | c6f37f1 | 2009-05-24 22:16:31 +0200 | [diff] [blame] | 3050 | #endif /* CONFIG_HIBERNATION */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3051 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3052 | /* It's optimal to keep kswapds on the same CPUs as their memory, but |
| 3053 | not required for correctness. So if the last cpu in a node goes |
| 3054 | away, we get changed to run anywhere: as the first one comes back, |
| 3055 | restore their cpu bindings. */ |
Chandra Seetharaman | 9c7b216 | 2006-06-27 02:54:07 -0700 | [diff] [blame] | 3056 | static int __devinit cpu_callback(struct notifier_block *nfb, |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 3057 | unsigned long action, void *hcpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3058 | { |
Yasunori Goto | 58c0a4a | 2007-10-16 01:25:40 -0700 | [diff] [blame] | 3059 | int nid; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3060 | |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 3061 | if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN) { |
Yasunori Goto | 58c0a4a | 2007-10-16 01:25:40 -0700 | [diff] [blame] | 3062 | for_each_node_state(nid, N_HIGH_MEMORY) { |
Mike Travis | c5f59f0 | 2008-04-04 18:11:10 -0700 | [diff] [blame] | 3063 | pg_data_t *pgdat = NODE_DATA(nid); |
Rusty Russell | a70f730 | 2009-03-13 14:49:46 +1030 | [diff] [blame] | 3064 | const struct cpumask *mask; |
| 3065 | |
| 3066 | mask = cpumask_of_node(pgdat->node_id); |
Mike Travis | c5f59f0 | 2008-04-04 18:11:10 -0700 | [diff] [blame] | 3067 | |
Rusty Russell | 3e59794 | 2009-01-01 10:12:24 +1030 | [diff] [blame] | 3068 | if (cpumask_any_and(cpu_online_mask, mask) < nr_cpu_ids) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3069 | /* One of our CPUs online: restore mask */ |
Mike Travis | c5f59f0 | 2008-04-04 18:11:10 -0700 | [diff] [blame] | 3070 | set_cpus_allowed_ptr(pgdat->kswapd, mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3071 | } |
| 3072 | } |
| 3073 | return NOTIFY_OK; |
| 3074 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3075 | |
Yasunori Goto | 3218ae1 | 2006-06-27 02:53:33 -0700 | [diff] [blame] | 3076 | /* |
| 3077 | * This kswapd start function will be called by init and node-hot-add. |
| 3078 | * On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added. |
| 3079 | */ |
| 3080 | int kswapd_run(int nid) |
| 3081 | { |
| 3082 | pg_data_t *pgdat = NODE_DATA(nid); |
| 3083 | int ret = 0; |
| 3084 | |
| 3085 | if (pgdat->kswapd) |
| 3086 | return 0; |
| 3087 | |
| 3088 | pgdat->kswapd = kthread_run(kswapd, pgdat, "kswapd%d", nid); |
| 3089 | if (IS_ERR(pgdat->kswapd)) { |
| 3090 | /* failure at boot is fatal */ |
| 3091 | BUG_ON(system_state == SYSTEM_BOOTING); |
| 3092 | printk("Failed to start kswapd on node %d\n",nid); |
| 3093 | ret = -1; |
| 3094 | } |
| 3095 | return ret; |
| 3096 | } |
| 3097 | |
David Rientjes | 8fe23e0 | 2009-12-14 17:58:33 -0800 | [diff] [blame] | 3098 | /* |
| 3099 | * Called by memory hotplug when all memory in a node is offlined. |
| 3100 | */ |
| 3101 | void kswapd_stop(int nid) |
| 3102 | { |
| 3103 | struct task_struct *kswapd = NODE_DATA(nid)->kswapd; |
| 3104 | |
| 3105 | if (kswapd) |
| 3106 | kthread_stop(kswapd); |
| 3107 | } |
| 3108 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3109 | static int __init kswapd_init(void) |
| 3110 | { |
Yasunori Goto | 3218ae1 | 2006-06-27 02:53:33 -0700 | [diff] [blame] | 3111 | int nid; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 3112 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3113 | swap_setup(); |
Christoph Lameter | 9422ffb | 2007-10-16 01:25:31 -0700 | [diff] [blame] | 3114 | for_each_node_state(nid, N_HIGH_MEMORY) |
Yasunori Goto | 3218ae1 | 2006-06-27 02:53:33 -0700 | [diff] [blame] | 3115 | kswapd_run(nid); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3116 | hotcpu_notifier(cpu_callback, 0); |
| 3117 | return 0; |
| 3118 | } |
| 3119 | |
| 3120 | module_init(kswapd_init) |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3121 | |
| 3122 | #ifdef CONFIG_NUMA |
| 3123 | /* |
| 3124 | * Zone reclaim mode |
| 3125 | * |
| 3126 | * If non-zero call zone_reclaim when the number of free pages falls below |
| 3127 | * the watermarks. |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3128 | */ |
| 3129 | int zone_reclaim_mode __read_mostly; |
| 3130 | |
Christoph Lameter | 1b2ffb7 | 2006-02-01 03:05:34 -0800 | [diff] [blame] | 3131 | #define RECLAIM_OFF 0 |
Fernando Luis Vazquez Cao | 7d03431 | 2008-07-29 22:33:41 -0700 | [diff] [blame] | 3132 | #define RECLAIM_ZONE (1<<0) /* Run shrink_inactive_list on the zone */ |
Christoph Lameter | 1b2ffb7 | 2006-02-01 03:05:34 -0800 | [diff] [blame] | 3133 | #define RECLAIM_WRITE (1<<1) /* Writeout pages during reclaim */ |
| 3134 | #define RECLAIM_SWAP (1<<2) /* Swap pages out during reclaim */ |
| 3135 | |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3136 | /* |
Christoph Lameter | a92f712 | 2006-02-01 03:05:32 -0800 | [diff] [blame] | 3137 | * Priority for ZONE_RECLAIM. This determines the fraction of pages |
| 3138 | * of a node considered for each zone_reclaim. 4 scans 1/16th of |
| 3139 | * a zone. |
| 3140 | */ |
| 3141 | #define ZONE_RECLAIM_PRIORITY 4 |
| 3142 | |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3143 | /* |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 3144 | * Percentage of pages in a zone that must be unmapped for zone_reclaim to |
| 3145 | * occur. |
| 3146 | */ |
| 3147 | int sysctl_min_unmapped_ratio = 1; |
| 3148 | |
| 3149 | /* |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3150 | * If the number of slab pages in a zone grows beyond this percentage then |
| 3151 | * slab reclaim needs to occur. |
| 3152 | */ |
| 3153 | int sysctl_min_slab_ratio = 5; |
| 3154 | |
Mel Gorman | 90afa5d | 2009-06-16 15:33:20 -0700 | [diff] [blame] | 3155 | static inline unsigned long zone_unmapped_file_pages(struct zone *zone) |
| 3156 | { |
| 3157 | unsigned long file_mapped = zone_page_state(zone, NR_FILE_MAPPED); |
| 3158 | unsigned long file_lru = zone_page_state(zone, NR_INACTIVE_FILE) + |
| 3159 | zone_page_state(zone, NR_ACTIVE_FILE); |
| 3160 | |
| 3161 | /* |
| 3162 | * It's possible for there to be more file mapped pages than |
| 3163 | * accounted for by the pages on the file LRU lists because |
| 3164 | * tmpfs pages accounted for as ANON can also be FILE_MAPPED |
| 3165 | */ |
| 3166 | return (file_lru > file_mapped) ? (file_lru - file_mapped) : 0; |
| 3167 | } |
| 3168 | |
| 3169 | /* Work out how many page cache pages we can reclaim in this reclaim_mode */ |
| 3170 | static long zone_pagecache_reclaimable(struct zone *zone) |
| 3171 | { |
| 3172 | long nr_pagecache_reclaimable; |
| 3173 | long delta = 0; |
| 3174 | |
| 3175 | /* |
| 3176 | * If RECLAIM_SWAP is set, then all file pages are considered |
| 3177 | * potentially reclaimable. Otherwise, we have to worry about |
| 3178 | * pages like swapcache and zone_unmapped_file_pages() provides |
| 3179 | * a better estimate |
| 3180 | */ |
| 3181 | if (zone_reclaim_mode & RECLAIM_SWAP) |
| 3182 | nr_pagecache_reclaimable = zone_page_state(zone, NR_FILE_PAGES); |
| 3183 | else |
| 3184 | nr_pagecache_reclaimable = zone_unmapped_file_pages(zone); |
| 3185 | |
| 3186 | /* If we can't clean pages, remove dirty pages from consideration */ |
| 3187 | if (!(zone_reclaim_mode & RECLAIM_WRITE)) |
| 3188 | delta += zone_page_state(zone, NR_FILE_DIRTY); |
| 3189 | |
| 3190 | /* Watch for any possible underflows due to delta */ |
| 3191 | if (unlikely(delta > nr_pagecache_reclaimable)) |
| 3192 | delta = nr_pagecache_reclaimable; |
| 3193 | |
| 3194 | return nr_pagecache_reclaimable - delta; |
| 3195 | } |
| 3196 | |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3197 | /* |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3198 | * Try to free up some pages from this zone through reclaim. |
| 3199 | */ |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3200 | 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] | 3201 | { |
Christoph Lameter | 7fb2d46 | 2006-03-22 00:08:22 -0800 | [diff] [blame] | 3202 | /* Minimum pages needed in order to stay on node */ |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 3203 | const unsigned long nr_pages = 1 << order; |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3204 | struct task_struct *p = current; |
| 3205 | struct reclaim_state reclaim_state; |
Christoph Lameter | 8695949 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3206 | int priority; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3207 | struct scan_control sc = { |
| 3208 | .may_writepage = !!(zone_reclaim_mode & RECLAIM_WRITE), |
Johannes Weiner | a6dc60f8 | 2009-03-31 15:19:30 -0700 | [diff] [blame] | 3209 | .may_unmap = !!(zone_reclaim_mode & RECLAIM_SWAP), |
KOSAKI Motohiro | 2e2e425 | 2009-04-21 12:24:57 -0700 | [diff] [blame] | 3210 | .may_swap = 1, |
KOSAKI Motohiro | 22fba33 | 2009-12-14 17:59:10 -0800 | [diff] [blame] | 3211 | .nr_to_reclaim = max_t(unsigned long, nr_pages, |
| 3212 | SWAP_CLUSTER_MAX), |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3213 | .gfp_mask = gfp_mask, |
Johannes Weiner | bd2f619 | 2009-03-31 15:19:38 -0700 | [diff] [blame] | 3214 | .order = order, |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3215 | }; |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 3216 | struct shrink_control shrink = { |
| 3217 | .gfp_mask = sc.gfp_mask, |
| 3218 | }; |
KOSAKI Motohiro | 1574804 | 2010-08-09 17:19:50 -0700 | [diff] [blame] | 3219 | unsigned long nr_slab_pages0, nr_slab_pages1; |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3220 | |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3221 | cond_resched(); |
Christoph Lameter | d4f7796 | 2006-02-24 13:04:22 -0800 | [diff] [blame] | 3222 | /* |
| 3223 | * We need to be able to allocate from the reserves for RECLAIM_SWAP |
| 3224 | * and we also need to be able to write out pages for RECLAIM_WRITE |
| 3225 | * and RECLAIM_SWAP. |
| 3226 | */ |
| 3227 | p->flags |= PF_MEMALLOC | PF_SWAPWRITE; |
KOSAKI Motohiro | 76ca542 | 2010-03-05 13:41:47 -0800 | [diff] [blame] | 3228 | lockdep_set_current_reclaim_state(gfp_mask); |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3229 | reclaim_state.reclaimed_slab = 0; |
| 3230 | p->reclaim_state = &reclaim_state; |
Christoph Lameter | c84db23 | 2006-02-01 03:05:29 -0800 | [diff] [blame] | 3231 | |
Mel Gorman | 90afa5d | 2009-06-16 15:33:20 -0700 | [diff] [blame] | 3232 | if (zone_pagecache_reclaimable(zone) > zone->min_unmapped_pages) { |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3233 | /* |
| 3234 | * Free memory by calling shrink zone with increasing |
| 3235 | * priorities until we have enough memory freed. |
| 3236 | */ |
| 3237 | priority = ZONE_RECLAIM_PRIORITY; |
| 3238 | do { |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 3239 | shrink_zone(priority, zone, &sc); |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3240 | priority--; |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 3241 | } while (priority >= 0 && sc.nr_reclaimed < nr_pages); |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3242 | } |
Christoph Lameter | c84db23 | 2006-02-01 03:05:29 -0800 | [diff] [blame] | 3243 | |
KOSAKI Motohiro | 1574804 | 2010-08-09 17:19:50 -0700 | [diff] [blame] | 3244 | nr_slab_pages0 = zone_page_state(zone, NR_SLAB_RECLAIMABLE); |
| 3245 | if (nr_slab_pages0 > zone->min_slab_pages) { |
Christoph Lameter | 2a16e3f | 2006-02-01 03:05:35 -0800 | [diff] [blame] | 3246 | /* |
Christoph Lameter | 7fb2d46 | 2006-03-22 00:08:22 -0800 | [diff] [blame] | 3247 | * shrink_slab() does not currently allow us to determine how |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3248 | * many pages were freed in this zone. So we take the current |
| 3249 | * number of slab pages and shake the slab until it is reduced |
| 3250 | * by the same nr_pages that we used for reclaiming unmapped |
| 3251 | * pages. |
Christoph Lameter | 2a16e3f | 2006-02-01 03:05:35 -0800 | [diff] [blame] | 3252 | * |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3253 | * Note that shrink_slab will free memory on all zones and may |
| 3254 | * take a long time. |
Christoph Lameter | 2a16e3f | 2006-02-01 03:05:35 -0800 | [diff] [blame] | 3255 | */ |
KOSAKI Motohiro | 4dc4b3d | 2010-08-09 17:19:54 -0700 | [diff] [blame] | 3256 | for (;;) { |
| 3257 | unsigned long lru_pages = zone_reclaimable_pages(zone); |
| 3258 | |
| 3259 | /* No reclaimable slab or very low memory pressure */ |
Ying Han | 1495f23 | 2011-05-24 17:12:27 -0700 | [diff] [blame] | 3260 | if (!shrink_slab(&shrink, sc.nr_scanned, lru_pages)) |
KOSAKI Motohiro | 4dc4b3d | 2010-08-09 17:19:54 -0700 | [diff] [blame] | 3261 | break; |
| 3262 | |
| 3263 | /* Freed enough memory */ |
| 3264 | nr_slab_pages1 = zone_page_state(zone, |
| 3265 | NR_SLAB_RECLAIMABLE); |
| 3266 | if (nr_slab_pages1 + nr_pages <= nr_slab_pages0) |
| 3267 | break; |
| 3268 | } |
Christoph Lameter | 83e33a4 | 2006-09-25 23:31:53 -0700 | [diff] [blame] | 3269 | |
| 3270 | /* |
| 3271 | * Update nr_reclaimed by the number of slab pages we |
| 3272 | * reclaimed from this zone. |
| 3273 | */ |
KOSAKI Motohiro | 1574804 | 2010-08-09 17:19:50 -0700 | [diff] [blame] | 3274 | nr_slab_pages1 = zone_page_state(zone, NR_SLAB_RECLAIMABLE); |
| 3275 | if (nr_slab_pages1 < nr_slab_pages0) |
| 3276 | sc.nr_reclaimed += nr_slab_pages0 - nr_slab_pages1; |
Christoph Lameter | 2a16e3f | 2006-02-01 03:05:35 -0800 | [diff] [blame] | 3277 | } |
| 3278 | |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3279 | p->reclaim_state = NULL; |
Christoph Lameter | d4f7796 | 2006-02-24 13:04:22 -0800 | [diff] [blame] | 3280 | current->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE); |
KOSAKI Motohiro | 76ca542 | 2010-03-05 13:41:47 -0800 | [diff] [blame] | 3281 | lockdep_clear_current_reclaim_state(); |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 3282 | return sc.nr_reclaimed >= nr_pages; |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3283 | } |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3284 | |
| 3285 | int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) |
| 3286 | { |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3287 | int node_id; |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3288 | int ret; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3289 | |
| 3290 | /* |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3291 | * Zone reclaim reclaims unmapped file backed pages and |
| 3292 | * slab pages if we are over the defined limits. |
Christoph Lameter | 34aa133 | 2006-06-30 01:55:37 -0700 | [diff] [blame] | 3293 | * |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 3294 | * A small portion of unmapped file backed pages is needed for |
| 3295 | * file I/O otherwise pages read by file I/O will be immediately |
| 3296 | * thrown out if the zone is overallocated. So we do not reclaim |
| 3297 | * if less than a specified percentage of the zone is used by |
| 3298 | * unmapped file backed pages. |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3299 | */ |
Mel Gorman | 90afa5d | 2009-06-16 15:33:20 -0700 | [diff] [blame] | 3300 | if (zone_pagecache_reclaimable(zone) <= zone->min_unmapped_pages && |
| 3301 | zone_page_state(zone, NR_SLAB_RECLAIMABLE) <= zone->min_slab_pages) |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 3302 | return ZONE_RECLAIM_FULL; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3303 | |
KOSAKI Motohiro | 93e4a89 | 2010-03-05 13:41:55 -0800 | [diff] [blame] | 3304 | if (zone->all_unreclaimable) |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 3305 | return ZONE_RECLAIM_FULL; |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3306 | |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3307 | /* |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3308 | * Do not scan if the allocation should not be delayed. |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3309 | */ |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3310 | if (!(gfp_mask & __GFP_WAIT) || (current->flags & PF_MEMALLOC)) |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 3311 | return ZONE_RECLAIM_NOSCAN; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3312 | |
| 3313 | /* |
| 3314 | * Only run zone reclaim on the local zone or on zones that do not |
| 3315 | * have associated processors. This will favor the local processor |
| 3316 | * over remote processors and spread off node memory allocations |
| 3317 | * as wide as possible. |
| 3318 | */ |
Christoph Lameter | 89fa302 | 2006-09-25 23:31:55 -0700 | [diff] [blame] | 3319 | node_id = zone_to_nid(zone); |
Christoph Lameter | 37c0708 | 2007-10-16 01:25:36 -0700 | [diff] [blame] | 3320 | if (node_state(node_id, N_CPU) && node_id != numa_node_id()) |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 3321 | return ZONE_RECLAIM_NOSCAN; |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3322 | |
| 3323 | if (zone_test_and_set_flag(zone, ZONE_RECLAIM_LOCKED)) |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 3324 | return ZONE_RECLAIM_NOSCAN; |
| 3325 | |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3326 | ret = __zone_reclaim(zone, gfp_mask, order); |
| 3327 | zone_clear_flag(zone, ZONE_RECLAIM_LOCKED); |
| 3328 | |
Mel Gorman | 24cf72518 | 2009-06-16 15:33:23 -0700 | [diff] [blame] | 3329 | if (!ret) |
| 3330 | count_vm_event(PGSCAN_ZONE_RECLAIM_FAILED); |
| 3331 | |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3332 | return ret; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3333 | } |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3334 | #endif |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3335 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3336 | /* |
| 3337 | * page_evictable - test whether a page is evictable |
| 3338 | * @page: the page to test |
| 3339 | * @vma: the VMA in which the page is or will be mapped, may be NULL |
| 3340 | * |
| 3341 | * 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] | 3342 | * lists vs unevictable list. The vma argument is !NULL when called from the |
| 3343 | * fault path to determine how to instantate a new page. |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3344 | * |
| 3345 | * Reasons page might not be evictable: |
Lee Schermerhorn | ba9ddf4 | 2008-10-18 20:26:42 -0700 | [diff] [blame] | 3346 | * (1) page's mapping marked unevictable |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 3347 | * (2) page is part of an mlocked VMA |
Lee Schermerhorn | ba9ddf4 | 2008-10-18 20:26:42 -0700 | [diff] [blame] | 3348 | * |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3349 | */ |
| 3350 | int page_evictable(struct page *page, struct vm_area_struct *vma) |
| 3351 | { |
| 3352 | |
Lee Schermerhorn | ba9ddf4 | 2008-10-18 20:26:42 -0700 | [diff] [blame] | 3353 | if (mapping_unevictable(page_mapping(page))) |
| 3354 | return 0; |
| 3355 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 3356 | if (PageMlocked(page) || (vma && is_mlocked_vma(vma, page))) |
| 3357 | return 0; |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3358 | |
| 3359 | return 1; |
| 3360 | } |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3361 | |
| 3362 | /** |
| 3363 | * check_move_unevictable_page - check page for evictability and move to appropriate zone lru list |
| 3364 | * @page: page to check evictability and move to appropriate lru list |
| 3365 | * @zone: zone page is in |
| 3366 | * |
| 3367 | * Checks a page for evictability and moves the page to the appropriate |
| 3368 | * zone lru list. |
| 3369 | * |
| 3370 | * Restrictions: zone->lru_lock must be held, page must be on LRU and must |
| 3371 | * have PageUnevictable set. |
| 3372 | */ |
| 3373 | static void check_move_unevictable_page(struct page *page, struct zone *zone) |
| 3374 | { |
| 3375 | VM_BUG_ON(PageActive(page)); |
| 3376 | |
| 3377 | retry: |
| 3378 | ClearPageUnevictable(page); |
| 3379 | if (page_evictable(page, NULL)) { |
Johannes Weiner | 401a8e1 | 2009-09-21 17:02:58 -0700 | [diff] [blame] | 3380 | enum lru_list l = page_lru_base_type(page); |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3381 | |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3382 | __dec_zone_state(zone, NR_UNEVICTABLE); |
| 3383 | list_move(&page->lru, &zone->lru[l].list); |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3384 | mem_cgroup_move_lists(page, LRU_UNEVICTABLE, l); |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3385 | __inc_zone_state(zone, NR_INACTIVE_ANON + l); |
| 3386 | __count_vm_event(UNEVICTABLE_PGRESCUED); |
| 3387 | } else { |
| 3388 | /* |
| 3389 | * rotate unevictable list |
| 3390 | */ |
| 3391 | SetPageUnevictable(page); |
| 3392 | list_move(&page->lru, &zone->lru[LRU_UNEVICTABLE].list); |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3393 | mem_cgroup_rotate_lru_list(page, LRU_UNEVICTABLE); |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3394 | if (page_evictable(page, NULL)) |
| 3395 | goto retry; |
| 3396 | } |
| 3397 | } |
| 3398 | |
| 3399 | /** |
| 3400 | * scan_mapping_unevictable_pages - scan an address space for evictable pages |
| 3401 | * @mapping: struct address_space to scan for evictable pages |
| 3402 | * |
| 3403 | * Scan all pages in mapping. Check unevictable pages for |
| 3404 | * evictability and move them to the appropriate zone lru list. |
| 3405 | */ |
| 3406 | void scan_mapping_unevictable_pages(struct address_space *mapping) |
| 3407 | { |
| 3408 | pgoff_t next = 0; |
| 3409 | pgoff_t end = (i_size_read(mapping->host) + PAGE_CACHE_SIZE - 1) >> |
| 3410 | PAGE_CACHE_SHIFT; |
| 3411 | struct zone *zone; |
| 3412 | struct pagevec pvec; |
| 3413 | |
| 3414 | if (mapping->nrpages == 0) |
| 3415 | return; |
| 3416 | |
| 3417 | pagevec_init(&pvec, 0); |
| 3418 | while (next < end && |
| 3419 | pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { |
| 3420 | int i; |
| 3421 | int pg_scanned = 0; |
| 3422 | |
| 3423 | zone = NULL; |
| 3424 | |
| 3425 | for (i = 0; i < pagevec_count(&pvec); i++) { |
| 3426 | struct page *page = pvec.pages[i]; |
| 3427 | pgoff_t page_index = page->index; |
| 3428 | struct zone *pagezone = page_zone(page); |
| 3429 | |
| 3430 | pg_scanned++; |
| 3431 | if (page_index > next) |
| 3432 | next = page_index; |
| 3433 | next++; |
| 3434 | |
| 3435 | if (pagezone != zone) { |
| 3436 | if (zone) |
| 3437 | spin_unlock_irq(&zone->lru_lock); |
| 3438 | zone = pagezone; |
| 3439 | spin_lock_irq(&zone->lru_lock); |
| 3440 | } |
| 3441 | |
| 3442 | if (PageLRU(page) && PageUnevictable(page)) |
| 3443 | check_move_unevictable_page(page, zone); |
| 3444 | } |
| 3445 | if (zone) |
| 3446 | spin_unlock_irq(&zone->lru_lock); |
| 3447 | pagevec_release(&pvec); |
| 3448 | |
| 3449 | count_vm_events(UNEVICTABLE_PGSCANNED, pg_scanned); |
| 3450 | } |
| 3451 | |
| 3452 | } |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3453 | |
Johannes Weiner | 264e56d | 2011-10-31 17:09:13 -0700 | [diff] [blame] | 3454 | static void warn_scan_unevictable_pages(void) |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3455 | { |
Johannes Weiner | 264e56d | 2011-10-31 17:09:13 -0700 | [diff] [blame] | 3456 | printk_once(KERN_WARNING |
| 3457 | "The scan_unevictable_pages sysctl/node-interface has been " |
| 3458 | "disabled for lack of a legitimate use case. If you have " |
| 3459 | "one, please send an email to linux-mm@kvack.org.\n"); |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3460 | } |
| 3461 | |
| 3462 | /* |
| 3463 | * scan_unevictable_pages [vm] sysctl handler. On demand re-scan of |
| 3464 | * all nodes' unevictable lists for evictable pages |
| 3465 | */ |
| 3466 | unsigned long scan_unevictable_pages; |
| 3467 | |
| 3468 | int scan_unevictable_handler(struct ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 3469 | void __user *buffer, |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3470 | size_t *length, loff_t *ppos) |
| 3471 | { |
Johannes Weiner | 264e56d | 2011-10-31 17:09:13 -0700 | [diff] [blame] | 3472 | warn_scan_unevictable_pages(); |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 3473 | proc_doulongvec_minmax(table, write, buffer, length, ppos); |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3474 | scan_unevictable_pages = 0; |
| 3475 | return 0; |
| 3476 | } |
| 3477 | |
Thadeu Lima de Souza Cascardo | e4455ab | 2010-10-26 14:21:28 -0700 | [diff] [blame] | 3478 | #ifdef CONFIG_NUMA |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3479 | /* |
| 3480 | * per node 'scan_unevictable_pages' attribute. On demand re-scan of |
| 3481 | * a specified node's per zone unevictable lists for evictable pages. |
| 3482 | */ |
| 3483 | |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 3484 | static ssize_t read_scan_unevictable_node(struct device *dev, |
| 3485 | struct device_attribute *attr, |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3486 | char *buf) |
| 3487 | { |
Johannes Weiner | 264e56d | 2011-10-31 17:09:13 -0700 | [diff] [blame] | 3488 | warn_scan_unevictable_pages(); |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3489 | return sprintf(buf, "0\n"); /* always zero; should fit... */ |
| 3490 | } |
| 3491 | |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 3492 | static ssize_t write_scan_unevictable_node(struct device *dev, |
| 3493 | struct device_attribute *attr, |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3494 | const char *buf, size_t count) |
| 3495 | { |
Johannes Weiner | 264e56d | 2011-10-31 17:09:13 -0700 | [diff] [blame] | 3496 | warn_scan_unevictable_pages(); |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3497 | return 1; |
| 3498 | } |
| 3499 | |
| 3500 | |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 3501 | static DEVICE_ATTR(scan_unevictable_pages, S_IRUGO | S_IWUSR, |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3502 | read_scan_unevictable_node, |
| 3503 | write_scan_unevictable_node); |
| 3504 | |
| 3505 | int scan_unevictable_register_node(struct node *node) |
| 3506 | { |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 3507 | return device_create_file(&node->dev, &dev_attr_scan_unevictable_pages); |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3508 | } |
| 3509 | |
| 3510 | void scan_unevictable_unregister_node(struct node *node) |
| 3511 | { |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 3512 | device_remove_file(&node->dev, &dev_attr_scan_unevictable_pages); |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3513 | } |
Thadeu Lima de Souza Cascardo | e4455ab | 2010-10-26 14:21:28 -0700 | [diff] [blame] | 3514 | #endif |