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> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 29 | #include <linux/backing-dev.h> |
| 30 | #include <linux/rmap.h> |
| 31 | #include <linux/topology.h> |
| 32 | #include <linux/cpu.h> |
| 33 | #include <linux/cpuset.h> |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 34 | #include <linux/compaction.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 35 | #include <linux/notifier.h> |
| 36 | #include <linux/rwsem.h> |
Rafael J. Wysocki | 248a030 | 2006-03-22 00:09:04 -0800 | [diff] [blame] | 37 | #include <linux/delay.h> |
Yasunori Goto | 3218ae1 | 2006-06-27 02:53:33 -0700 | [diff] [blame] | 38 | #include <linux/kthread.h> |
Nigel Cunningham | 7dfb710 | 2006-12-06 20:34:23 -0800 | [diff] [blame] | 39 | #include <linux/freezer.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 40 | #include <linux/memcontrol.h> |
Keika Kobayashi | 873b477 | 2008-07-25 01:48:52 -0700 | [diff] [blame] | 41 | #include <linux/delayacct.h> |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 42 | #include <linux/sysctl.h> |
KOSAKI Motohiro | 929bea7 | 2011-04-14 15:22:12 -0700 | [diff] [blame] | 43 | #include <linux/oom.h> |
Linus Torvalds | 268bb0c | 2011-05-20 12:50:29 -0700 | [diff] [blame] | 44 | #include <linux/prefetch.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 45 | |
| 46 | #include <asm/tlbflush.h> |
| 47 | #include <asm/div64.h> |
| 48 | |
| 49 | #include <linux/swapops.h> |
| 50 | |
Nick Piggin | 0f8053a | 2006-03-22 00:08:33 -0800 | [diff] [blame] | 51 | #include "internal.h" |
| 52 | |
Mel Gorman | 33906bc | 2010-08-09 17:19:16 -0700 | [diff] [blame] | 53 | #define CREATE_TRACE_POINTS |
| 54 | #include <trace/events/vmscan.h> |
| 55 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 56 | struct scan_control { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 57 | /* Incremented by the number of inactive pages that were scanned */ |
| 58 | unsigned long nr_scanned; |
| 59 | |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 60 | /* Number of pages freed so far during a call to shrink_zones() */ |
| 61 | unsigned long nr_reclaimed; |
| 62 | |
KOSAKI Motohiro | 22fba33 | 2009-12-14 17:59:10 -0800 | [diff] [blame] | 63 | /* How many pages shrink_list() should reclaim */ |
| 64 | unsigned long nr_to_reclaim; |
| 65 | |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 66 | unsigned long hibernation_mode; |
| 67 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 68 | /* This context's GFP mask */ |
Al Viro | 6daa0e2 | 2005-10-21 03:18:50 -0400 | [diff] [blame] | 69 | gfp_t gfp_mask; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 70 | |
| 71 | int may_writepage; |
| 72 | |
Johannes Weiner | a6dc60f | 2009-03-31 15:19:30 -0700 | [diff] [blame] | 73 | /* Can mapped pages be reclaimed? */ |
| 74 | int may_unmap; |
Christoph Lameter | f1fd106 | 2006-01-18 17:42:30 -0800 | [diff] [blame] | 75 | |
KOSAKI Motohiro | 2e2e425 | 2009-04-21 12:24:57 -0700 | [diff] [blame] | 76 | /* Can pages be swapped as part of reclaim? */ |
| 77 | int may_swap; |
| 78 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 79 | int order; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 80 | |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 81 | /* Scan (total_size >> priority) pages at once */ |
| 82 | int priority; |
| 83 | |
KOSAKI Motohiro | 5f53e76 | 2010-05-24 14:32:37 -0700 | [diff] [blame] | 84 | /* |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 85 | * The memory cgroup that hit its limit and as a result is the |
| 86 | * primary target of this reclaim invocation. |
| 87 | */ |
| 88 | struct mem_cgroup *target_mem_cgroup; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 89 | |
KAMEZAWA Hiroyuki | 327c0e9 | 2009-03-31 15:23:31 -0700 | [diff] [blame] | 90 | /* |
| 91 | * Nodemask of nodes allowed by the caller. If NULL, all nodes |
| 92 | * are scanned. |
| 93 | */ |
| 94 | nodemask_t *nodemask; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 95 | }; |
| 96 | |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 97 | struct mem_cgroup_zone { |
| 98 | struct mem_cgroup *mem_cgroup; |
| 99 | struct zone *zone; |
| 100 | }; |
| 101 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 102 | #define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru)) |
| 103 | |
| 104 | #ifdef ARCH_HAS_PREFETCH |
| 105 | #define prefetch_prev_lru_page(_page, _base, _field) \ |
| 106 | do { \ |
| 107 | if ((_page)->lru.prev != _base) { \ |
| 108 | struct page *prev; \ |
| 109 | \ |
| 110 | prev = lru_to_page(&(_page->lru)); \ |
| 111 | prefetch(&prev->_field); \ |
| 112 | } \ |
| 113 | } while (0) |
| 114 | #else |
| 115 | #define prefetch_prev_lru_page(_page, _base, _field) do { } while (0) |
| 116 | #endif |
| 117 | |
| 118 | #ifdef ARCH_HAS_PREFETCHW |
| 119 | #define prefetchw_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 | prefetchw(&prev->_field); \ |
| 126 | } \ |
| 127 | } while (0) |
| 128 | #else |
| 129 | #define prefetchw_prev_lru_page(_page, _base, _field) do { } while (0) |
| 130 | #endif |
| 131 | |
| 132 | /* |
| 133 | * From 0 .. 100. Higher means more swappy. |
| 134 | */ |
| 135 | int vm_swappiness = 60; |
Andrew Morton | bd1e22b | 2006-06-23 02:03:47 -0700 | [diff] [blame] | 136 | 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] | 137 | |
| 138 | static LIST_HEAD(shrinker_list); |
| 139 | static DECLARE_RWSEM(shrinker_rwsem); |
| 140 | |
Balbir Singh | 00f0b82 | 2008-03-04 14:28:39 -0800 | [diff] [blame] | 141 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 142 | static bool global_reclaim(struct scan_control *sc) |
| 143 | { |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 144 | return !sc->target_mem_cgroup; |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 145 | } |
KAMEZAWA Hiroyuki | 91a4547 | 2008-02-07 00:14:29 -0800 | [diff] [blame] | 146 | #else |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 147 | static bool global_reclaim(struct scan_control *sc) |
| 148 | { |
| 149 | return true; |
| 150 | } |
KAMEZAWA Hiroyuki | 91a4547 | 2008-02-07 00:14:29 -0800 | [diff] [blame] | 151 | #endif |
| 152 | |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 153 | static struct zone_reclaim_stat *get_reclaim_stat(struct mem_cgroup_zone *mz) |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 154 | { |
Hugh Dickins | c12af16 | 2012-05-29 15:06:52 -0700 | [diff] [blame] | 155 | if (!mem_cgroup_disabled()) |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 156 | return mem_cgroup_get_reclaim_stat(mz->mem_cgroup, mz->zone); |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 157 | |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 158 | return &mz->zone->reclaim_stat; |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 159 | } |
| 160 | |
Lisa Du | e093521 | 2013-09-11 14:22:36 -0700 | [diff] [blame] | 161 | unsigned long zone_reclaimable_pages(struct zone *zone) |
| 162 | { |
| 163 | int nr; |
| 164 | |
| 165 | nr = zone_page_state(zone, NR_ACTIVE_FILE) + |
| 166 | zone_page_state(zone, NR_INACTIVE_FILE); |
| 167 | |
| 168 | if (get_nr_swap_pages() > 0) |
| 169 | nr += zone_page_state(zone, NR_ACTIVE_ANON) + |
| 170 | zone_page_state(zone, NR_INACTIVE_ANON); |
| 171 | |
| 172 | return nr; |
| 173 | } |
| 174 | |
| 175 | bool zone_reclaimable(struct zone *zone) |
| 176 | { |
| 177 | return zone->pages_scanned < zone_reclaimable_pages(zone) * 6; |
| 178 | } |
| 179 | |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 180 | static unsigned long zone_nr_lru_pages(struct mem_cgroup_zone *mz, |
| 181 | enum lru_list lru) |
Lisa Du | e093521 | 2013-09-11 14:22:36 -0700 | [diff] [blame] | 182 | |
KOSAKI Motohiro | c9f299d | 2009-01-07 18:08:16 -0800 | [diff] [blame] | 183 | { |
Hugh Dickins | c12af16 | 2012-05-29 15:06:52 -0700 | [diff] [blame] | 184 | if (!mem_cgroup_disabled()) |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 185 | return mem_cgroup_zone_nr_lru_pages(mz->mem_cgroup, |
| 186 | zone_to_nid(mz->zone), |
| 187 | zone_idx(mz->zone), |
| 188 | BIT(lru)); |
KOSAKI Motohiro | a3d8e05 | 2009-01-07 18:08:19 -0800 | [diff] [blame] | 189 | |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 190 | return zone_page_state(mz->zone, NR_LRU_BASE + lru); |
KOSAKI Motohiro | c9f299d | 2009-01-07 18:08:16 -0800 | [diff] [blame] | 191 | } |
| 192 | |
| 193 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 194 | /* |
| 195 | * Add a shrinker callback to be called from the vm |
| 196 | */ |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 197 | void register_shrinker(struct shrinker *shrinker) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 198 | { |
Konstantin Khlebnikov | 83aeead | 2011-12-08 14:33:54 -0800 | [diff] [blame] | 199 | atomic_long_set(&shrinker->nr_in_batch, 0); |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 200 | down_write(&shrinker_rwsem); |
| 201 | list_add_tail(&shrinker->list, &shrinker_list); |
| 202 | up_write(&shrinker_rwsem); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 203 | } |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 204 | EXPORT_SYMBOL(register_shrinker); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 205 | |
| 206 | /* |
| 207 | * Remove one |
| 208 | */ |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 209 | void unregister_shrinker(struct shrinker *shrinker) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 210 | { |
| 211 | down_write(&shrinker_rwsem); |
| 212 | list_del(&shrinker->list); |
| 213 | up_write(&shrinker_rwsem); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 214 | } |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 215 | EXPORT_SYMBOL(unregister_shrinker); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 216 | |
Ying Han | 1495f23 | 2011-05-24 17:12:27 -0700 | [diff] [blame] | 217 | static inline int do_shrinker_shrink(struct shrinker *shrinker, |
| 218 | struct shrink_control *sc, |
| 219 | unsigned long nr_to_scan) |
| 220 | { |
| 221 | sc->nr_to_scan = nr_to_scan; |
| 222 | return (*shrinker->shrink)(shrinker, sc); |
| 223 | } |
| 224 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 225 | #define SHRINK_BATCH 128 |
| 226 | /* |
| 227 | * Call the shrink functions to age shrinkable caches |
| 228 | * |
| 229 | * Here we assume it costs one seek to replace a lru page and that it also |
| 230 | * takes a seek to recreate a cache object. With this in mind we age equal |
| 231 | * percentages of the lru and ageable caches. This should balance the seeks |
| 232 | * generated by these structures. |
| 233 | * |
Simon Arlott | 183ff22 | 2007-10-20 01:27:18 +0200 | [diff] [blame] | 234 | * 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] | 235 | * slab to avoid swapping. |
| 236 | * |
| 237 | * We do weird things to avoid (scanned*seeks*entries) overflowing 32 bits. |
| 238 | * |
| 239 | * `lru_pages' represents the number of on-LRU pages in all the zones which |
| 240 | * are eligible for the caller's allocation attempt. It is used for balancing |
| 241 | * slab reclaim versus page reclaim. |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 242 | * |
| 243 | * Returns the number of slab objects which we shrunk. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 244 | */ |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 245 | unsigned long shrink_slab(struct shrink_control *shrink, |
Ying Han | 1495f23 | 2011-05-24 17:12:27 -0700 | [diff] [blame] | 246 | unsigned long nr_pages_scanned, |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 247 | unsigned long lru_pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 248 | { |
| 249 | struct shrinker *shrinker; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 250 | unsigned long ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 251 | |
Ying Han | 1495f23 | 2011-05-24 17:12:27 -0700 | [diff] [blame] | 252 | if (nr_pages_scanned == 0) |
| 253 | nr_pages_scanned = SWAP_CLUSTER_MAX; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 254 | |
Minchan Kim | f06590b | 2011-05-24 17:11:11 -0700 | [diff] [blame] | 255 | if (!down_read_trylock(&shrinker_rwsem)) { |
| 256 | /* Assume we'll be able to shrink next time */ |
| 257 | ret = 1; |
| 258 | goto out; |
| 259 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 260 | |
| 261 | list_for_each_entry(shrinker, &shrinker_list, list) { |
| 262 | unsigned long long delta; |
Konstantin Khlebnikov | 635697c | 2011-12-08 14:33:51 -0800 | [diff] [blame] | 263 | long total_scan; |
| 264 | long max_pass; |
Dave Chinner | 0957607 | 2011-07-08 14:14:34 +1000 | [diff] [blame] | 265 | int shrink_ret = 0; |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 266 | long nr; |
| 267 | long new_nr; |
Dave Chinner | e9299f5 | 2011-07-08 14:14:37 +1000 | [diff] [blame] | 268 | long batch_size = shrinker->batch ? shrinker->batch |
| 269 | : SHRINK_BATCH; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 270 | |
Konstantin Khlebnikov | 635697c | 2011-12-08 14:33:51 -0800 | [diff] [blame] | 271 | max_pass = do_shrinker_shrink(shrinker, shrink, 0); |
| 272 | if (max_pass <= 0) |
| 273 | continue; |
| 274 | |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 275 | /* |
| 276 | * copy the current shrinker scan count into a local variable |
| 277 | * and zero it so that other concurrent shrinker invocations |
| 278 | * don't also do this scanning work. |
| 279 | */ |
Konstantin Khlebnikov | 83aeead | 2011-12-08 14:33:54 -0800 | [diff] [blame] | 280 | nr = atomic_long_xchg(&shrinker->nr_in_batch, 0); |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 281 | |
| 282 | total_scan = nr; |
Ying Han | 1495f23 | 2011-05-24 17:12:27 -0700 | [diff] [blame] | 283 | delta = (4 * nr_pages_scanned) / shrinker->seeks; |
Andrea Arcangeli | ea164d7 | 2005-11-28 13:44:15 -0800 | [diff] [blame] | 284 | delta *= max_pass; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 285 | do_div(delta, lru_pages + 1); |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 286 | total_scan += delta; |
| 287 | if (total_scan < 0) { |
David Rientjes | 88c3bd7 | 2009-03-31 15:23:29 -0700 | [diff] [blame] | 288 | printk(KERN_ERR "shrink_slab: %pF negative objects to " |
| 289 | "delete nr=%ld\n", |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 290 | shrinker->shrink, total_scan); |
| 291 | total_scan = max_pass; |
Andrea Arcangeli | ea164d7 | 2005-11-28 13:44:15 -0800 | [diff] [blame] | 292 | } |
| 293 | |
| 294 | /* |
Dave Chinner | 3567b59 | 2011-07-08 14:14:36 +1000 | [diff] [blame] | 295 | * We need to avoid excessive windup on filesystem shrinkers |
| 296 | * due to large numbers of GFP_NOFS allocations causing the |
| 297 | * shrinkers to return -1 all the time. This results in a large |
| 298 | * nr being built up so when a shrink that can do some work |
| 299 | * comes along it empties the entire cache due to nr >>> |
| 300 | * max_pass. This is bad for sustaining a working set in |
| 301 | * memory. |
| 302 | * |
| 303 | * Hence only allow the shrinker to scan the entire cache when |
| 304 | * a large delta change is calculated directly. |
| 305 | */ |
| 306 | if (delta < max_pass / 4) |
| 307 | total_scan = min(total_scan, max_pass / 2); |
| 308 | |
| 309 | /* |
Andrea Arcangeli | ea164d7 | 2005-11-28 13:44:15 -0800 | [diff] [blame] | 310 | * Avoid risking looping forever due to too large nr value: |
| 311 | * never try to free more than twice the estimate number of |
| 312 | * freeable entries. |
| 313 | */ |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 314 | if (total_scan > max_pass * 2) |
| 315 | total_scan = max_pass * 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 316 | |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 317 | trace_mm_shrink_slab_start(shrinker, shrink, nr, |
Dave Chinner | 0957607 | 2011-07-08 14:14:34 +1000 | [diff] [blame] | 318 | nr_pages_scanned, lru_pages, |
| 319 | max_pass, delta, total_scan); |
| 320 | |
Dave Chinner | e9299f5 | 2011-07-08 14:14:37 +1000 | [diff] [blame] | 321 | while (total_scan >= batch_size) { |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 322 | int nr_before; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 323 | |
Ying Han | 1495f23 | 2011-05-24 17:12:27 -0700 | [diff] [blame] | 324 | nr_before = do_shrinker_shrink(shrinker, shrink, 0); |
| 325 | shrink_ret = do_shrinker_shrink(shrinker, shrink, |
Dave Chinner | e9299f5 | 2011-07-08 14:14:37 +1000 | [diff] [blame] | 326 | batch_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 327 | if (shrink_ret == -1) |
| 328 | break; |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 329 | if (shrink_ret < nr_before) |
| 330 | ret += nr_before - shrink_ret; |
Dave Chinner | e9299f5 | 2011-07-08 14:14:37 +1000 | [diff] [blame] | 331 | count_vm_events(SLABS_SCANNED, batch_size); |
| 332 | total_scan -= batch_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 333 | |
| 334 | cond_resched(); |
| 335 | } |
| 336 | |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 337 | /* |
| 338 | * move the unused scan count back into the shrinker in a |
| 339 | * manner that handles concurrent updates. If we exhausted the |
| 340 | * scan, there is no need to do an update. |
| 341 | */ |
Konstantin Khlebnikov | 83aeead | 2011-12-08 14:33:54 -0800 | [diff] [blame] | 342 | if (total_scan > 0) |
| 343 | new_nr = atomic_long_add_return(total_scan, |
| 344 | &shrinker->nr_in_batch); |
| 345 | else |
| 346 | new_nr = atomic_long_read(&shrinker->nr_in_batch); |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 347 | |
| 348 | trace_mm_shrink_slab_end(shrinker, shrink_ret, nr, new_nr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 349 | } |
| 350 | up_read(&shrinker_rwsem); |
Minchan Kim | f06590b | 2011-05-24 17:11:11 -0700 | [diff] [blame] | 351 | out: |
| 352 | cond_resched(); |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 353 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 354 | } |
| 355 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 356 | static inline int is_page_cache_freeable(struct page *page) |
| 357 | { |
Johannes Weiner | ceddc3a | 2009-09-21 17:03:00 -0700 | [diff] [blame] | 358 | /* |
| 359 | * A freeable page cache page is referenced only by the caller |
| 360 | * that isolated the page, the page cache radix tree and |
| 361 | * optional buffer heads at page->private. |
| 362 | */ |
Johannes Weiner | edcf474 | 2009-09-21 17:02:59 -0700 | [diff] [blame] | 363 | return page_count(page) - page_has_private(page) == 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 364 | } |
| 365 | |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 366 | static int may_write_to_queue(struct backing_dev_info *bdi, |
| 367 | struct scan_control *sc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 368 | { |
Christoph Lameter | 930d915 | 2006-01-08 01:00:47 -0800 | [diff] [blame] | 369 | if (current->flags & PF_SWAPWRITE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 370 | return 1; |
| 371 | if (!bdi_write_congested(bdi)) |
| 372 | return 1; |
| 373 | if (bdi == current->backing_dev_info) |
| 374 | return 1; |
| 375 | return 0; |
| 376 | } |
| 377 | |
| 378 | /* |
| 379 | * We detected a synchronous write error writing a page out. Probably |
| 380 | * -ENOSPC. We need to propagate that into the address_space for a subsequent |
| 381 | * fsync(), msync() or close(). |
| 382 | * |
| 383 | * The tricky part is that after writepage we cannot touch the mapping: nothing |
| 384 | * prevents it from being freed up. But we have a ref on the page and once |
| 385 | * that page is locked, the mapping is pinned. |
| 386 | * |
| 387 | * We're allowed to run sleeping lock_page() here because we know the caller has |
| 388 | * __GFP_FS. |
| 389 | */ |
| 390 | static void handle_write_error(struct address_space *mapping, |
| 391 | struct page *page, int error) |
| 392 | { |
Jens Axboe | 7eaceac | 2011-03-10 08:52:07 +0100 | [diff] [blame] | 393 | lock_page(page); |
Guillaume Chazarain | 3e9f45b | 2007-05-08 00:23:25 -0700 | [diff] [blame] | 394 | if (page_mapping(page) == mapping) |
| 395 | mapping_set_error(mapping, error); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 396 | unlock_page(page); |
| 397 | } |
| 398 | |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 399 | /* possible outcome of pageout() */ |
| 400 | typedef enum { |
| 401 | /* failed to write page out, page is locked */ |
| 402 | PAGE_KEEP, |
| 403 | /* move page to the active list, page is locked */ |
| 404 | PAGE_ACTIVATE, |
| 405 | /* page has been sent to the disk successfully, page is unlocked */ |
| 406 | PAGE_SUCCESS, |
| 407 | /* page is clean and locked */ |
| 408 | PAGE_CLEAN, |
| 409 | } pageout_t; |
| 410 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 411 | /* |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 412 | * pageout is called by shrink_page_list() for each dirty page. |
| 413 | * Calls ->writepage(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 414 | */ |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 415 | static pageout_t pageout(struct page *page, struct address_space *mapping, |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 416 | struct scan_control *sc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 417 | { |
| 418 | /* |
| 419 | * If the page is dirty, only perform writeback if that write |
| 420 | * will be non-blocking. To prevent this allocation from being |
| 421 | * stalled by pagecache activity. But note that there may be |
| 422 | * stalls if we need to run get_block(). We could test |
| 423 | * PagePrivate for that. |
| 424 | * |
Vincent Li | 6aceb53 | 2009-12-14 17:58:49 -0800 | [diff] [blame] | 425 | * If this process is currently in __generic_file_aio_write() against |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 426 | * this page's queue, we can perform writeback even if that |
| 427 | * will block. |
| 428 | * |
| 429 | * If the page is swapcache, write it back even if that would |
| 430 | * block, for some throttling. This happens by accident, because |
| 431 | * swap_backing_dev_info is bust: it doesn't reflect the |
| 432 | * congestion state of the swapdevs. Easy to fix, if needed. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 433 | */ |
| 434 | if (!is_page_cache_freeable(page)) |
| 435 | return PAGE_KEEP; |
| 436 | if (!mapping) { |
| 437 | /* |
| 438 | * Some data journaling orphaned pages can have |
| 439 | * page->mapping == NULL while being dirty with clean buffers. |
| 440 | */ |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 441 | if (page_has_private(page)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 442 | if (try_to_free_buffers(page)) { |
| 443 | ClearPageDirty(page); |
Harvey Harrison | d40cee2 | 2008-04-30 00:55:07 -0700 | [diff] [blame] | 444 | printk("%s: orphaned page\n", __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 445 | return PAGE_CLEAN; |
| 446 | } |
| 447 | } |
| 448 | return PAGE_KEEP; |
| 449 | } |
| 450 | if (mapping->a_ops->writepage == NULL) |
| 451 | return PAGE_ACTIVATE; |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 452 | if (!may_write_to_queue(mapping->backing_dev_info, sc)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 453 | return PAGE_KEEP; |
| 454 | |
| 455 | if (clear_page_dirty_for_io(page)) { |
| 456 | int res; |
| 457 | struct writeback_control wbc = { |
| 458 | .sync_mode = WB_SYNC_NONE, |
| 459 | .nr_to_write = SWAP_CLUSTER_MAX, |
OGAWA Hirofumi | 111ebb6 | 2006-06-23 02:03:26 -0700 | [diff] [blame] | 460 | .range_start = 0, |
| 461 | .range_end = LLONG_MAX, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 462 | .for_reclaim = 1, |
| 463 | }; |
| 464 | |
| 465 | SetPageReclaim(page); |
| 466 | res = mapping->a_ops->writepage(page, &wbc); |
| 467 | if (res < 0) |
| 468 | handle_write_error(mapping, page, res); |
Zach Brown | 994fc28c | 2005-12-15 14:28:17 -0800 | [diff] [blame] | 469 | if (res == AOP_WRITEPAGE_ACTIVATE) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 470 | ClearPageReclaim(page); |
| 471 | return PAGE_ACTIVATE; |
| 472 | } |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 473 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 474 | if (!PageWriteback(page)) { |
| 475 | /* synchronous write or broken a_ops? */ |
| 476 | ClearPageReclaim(page); |
Vinayak Menon | 8ba7f62 | 2014-11-17 20:26:19 +0530 | [diff] [blame] | 477 | if (PageError(page) && PageSwapCache(page)) { |
| 478 | ClearPageError(page); |
| 479 | /* |
| 480 | * We lock the page here because it is required |
| 481 | * to free the swp space later in |
| 482 | * shrink_page_list. But the page may be |
| 483 | * unclocked by functions like |
| 484 | * handle_write_error. |
| 485 | */ |
| 486 | __set_page_locked(page); |
Olav Haugan | d0a5856 | 2013-11-04 10:46:16 -0800 | [diff] [blame] | 487 | return PAGE_ACTIVATE; |
Vinayak Menon | 8ba7f62 | 2014-11-17 20:26:19 +0530 | [diff] [blame] | 488 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 489 | } |
Mel Gorman | 626ebe7 | 2012-05-29 15:06:20 -0700 | [diff] [blame] | 490 | trace_mm_vmscan_writepage(page, trace_reclaim_flags(page)); |
Andrew Morton | e129b5c | 2006-09-27 01:50:00 -0700 | [diff] [blame] | 491 | inc_zone_page_state(page, NR_VMSCAN_WRITE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 492 | return PAGE_SUCCESS; |
| 493 | } |
| 494 | |
| 495 | return PAGE_CLEAN; |
| 496 | } |
| 497 | |
Andrew Morton | a649fd9 | 2006-10-17 00:09:36 -0700 | [diff] [blame] | 498 | /* |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 499 | * Same as remove_mapping, but if the page is removed from the mapping, it |
| 500 | * gets returned with a refcount of 0. |
Andrew Morton | a649fd9 | 2006-10-17 00:09:36 -0700 | [diff] [blame] | 501 | */ |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 502 | static int __remove_mapping(struct address_space *mapping, struct page *page) |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 503 | { |
Nick Piggin | 28e4d96 | 2006-09-25 23:31:23 -0700 | [diff] [blame] | 504 | BUG_ON(!PageLocked(page)); |
| 505 | BUG_ON(mapping != page_mapping(page)); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 506 | |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 507 | spin_lock_irq(&mapping->tree_lock); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 508 | /* |
Nick Piggin | 0fd0e6b | 2006-09-27 01:50:02 -0700 | [diff] [blame] | 509 | * The non racy check for a busy page. |
| 510 | * |
| 511 | * Must be careful with the order of the tests. When someone has |
| 512 | * a ref to the page, it may be possible that they dirty it then |
| 513 | * drop the reference. So if PageDirty is tested before page_count |
| 514 | * here, then the following race may occur: |
| 515 | * |
| 516 | * get_user_pages(&page); |
| 517 | * [user mapping goes away] |
| 518 | * write_to(page); |
| 519 | * !PageDirty(page) [good] |
| 520 | * SetPageDirty(page); |
| 521 | * put_page(page); |
| 522 | * !page_count(page) [good, discard it] |
| 523 | * |
| 524 | * [oops, our write_to data is lost] |
| 525 | * |
| 526 | * Reversing the order of the tests ensures such a situation cannot |
| 527 | * escape unnoticed. The smp_rmb is needed to ensure the page->flags |
| 528 | * load is not satisfied before that of page->_count. |
| 529 | * |
| 530 | * Note that if SetPageDirty is always performed via set_page_dirty, |
| 531 | * and thus under tree_lock, then this ordering is not required. |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 532 | */ |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 533 | if (!page_freeze_refs(page, 2)) |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 534 | goto cannot_free; |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 535 | /* note: atomic_cmpxchg in page_freeze_refs provides the smp_rmb */ |
| 536 | if (unlikely(PageDirty(page))) { |
| 537 | page_unfreeze_refs(page, 2); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 538 | goto cannot_free; |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 539 | } |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 540 | |
| 541 | if (PageSwapCache(page)) { |
| 542 | swp_entry_t swap = { .val = page_private(page) }; |
| 543 | __delete_from_swap_cache(page); |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 544 | spin_unlock_irq(&mapping->tree_lock); |
KAMEZAWA Hiroyuki | cb4b86b | 2009-06-16 15:32:52 -0700 | [diff] [blame] | 545 | swapcache_free(swap, page); |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 546 | } else { |
Linus Torvalds | 6072d13 | 2010-12-01 13:35:19 -0500 | [diff] [blame] | 547 | void (*freepage)(struct page *); |
| 548 | |
| 549 | freepage = mapping->a_ops->freepage; |
| 550 | |
Minchan Kim | e64a782 | 2011-03-22 16:32:44 -0700 | [diff] [blame] | 551 | __delete_from_page_cache(page); |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 552 | spin_unlock_irq(&mapping->tree_lock); |
Daisuke Nishimura | e767e05 | 2009-05-28 14:34:28 -0700 | [diff] [blame] | 553 | mem_cgroup_uncharge_cache_page(page); |
Linus Torvalds | 6072d13 | 2010-12-01 13:35:19 -0500 | [diff] [blame] | 554 | |
| 555 | if (freepage != NULL) |
| 556 | freepage(page); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 557 | } |
| 558 | |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 559 | return 1; |
| 560 | |
| 561 | cannot_free: |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 562 | spin_unlock_irq(&mapping->tree_lock); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 563 | return 0; |
| 564 | } |
| 565 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 566 | /* |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 567 | * Attempt to detach a locked page from its ->mapping. If it is dirty or if |
| 568 | * someone else has a ref on the page, abort and return 0. If it was |
| 569 | * successfully detached, return 1. Assumes the caller has a single ref on |
| 570 | * this page. |
| 571 | */ |
| 572 | int remove_mapping(struct address_space *mapping, struct page *page) |
| 573 | { |
| 574 | if (__remove_mapping(mapping, page)) { |
| 575 | /* |
| 576 | * Unfreezing the refcount with 1 rather than 2 effectively |
| 577 | * drops the pagecache ref for us without requiring another |
| 578 | * atomic operation. |
| 579 | */ |
| 580 | page_unfreeze_refs(page, 1); |
| 581 | return 1; |
| 582 | } |
| 583 | return 0; |
| 584 | } |
| 585 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 586 | /** |
| 587 | * putback_lru_page - put previously isolated page onto appropriate LRU list |
| 588 | * @page: page to be put back to appropriate lru list |
| 589 | * |
| 590 | * Add previously isolated @page to appropriate LRU list. |
| 591 | * Page may still be unevictable for other reasons. |
| 592 | * |
| 593 | * lru_lock must not be held, interrupts must be enabled. |
| 594 | */ |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 595 | void putback_lru_page(struct page *page) |
| 596 | { |
| 597 | int lru; |
| 598 | int active = !!TestClearPageActive(page); |
Lee Schermerhorn | bbfd28e | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 599 | int was_unevictable = PageUnevictable(page); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 600 | |
| 601 | VM_BUG_ON(PageLRU(page)); |
| 602 | |
| 603 | redo: |
| 604 | ClearPageUnevictable(page); |
| 605 | |
| 606 | if (page_evictable(page, NULL)) { |
| 607 | /* |
| 608 | * For evictable pages, we can use the cache. |
| 609 | * In event of a race, worst case is we end up with an |
| 610 | * unevictable page on [in]active list. |
| 611 | * We know how to handle that. |
| 612 | */ |
Johannes Weiner | 401a8e1 | 2009-09-21 17:02:58 -0700 | [diff] [blame] | 613 | lru = active + page_lru_base_type(page); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 614 | lru_cache_add_lru(page, lru); |
| 615 | } else { |
| 616 | /* |
| 617 | * Put unevictable pages directly on zone's unevictable |
| 618 | * list. |
| 619 | */ |
| 620 | lru = LRU_UNEVICTABLE; |
| 621 | add_page_to_unevictable_list(page); |
Johannes Weiner | 6a7b954 | 2009-10-26 16:50:00 -0700 | [diff] [blame] | 622 | /* |
Minchan Kim | 21ee9f3 | 2011-10-31 17:09:28 -0700 | [diff] [blame] | 623 | * When racing with an mlock or AS_UNEVICTABLE clearing |
| 624 | * (page is unlocked) make sure that if the other thread |
| 625 | * does not observe our setting of PG_lru and fails |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 626 | * isolation/check_move_unevictable_pages, |
Minchan Kim | 21ee9f3 | 2011-10-31 17:09:28 -0700 | [diff] [blame] | 627 | * we see PG_mlocked/AS_UNEVICTABLE cleared below and move |
Johannes Weiner | 6a7b954 | 2009-10-26 16:50:00 -0700 | [diff] [blame] | 628 | * the page back to the evictable list. |
| 629 | * |
Minchan Kim | 21ee9f3 | 2011-10-31 17:09:28 -0700 | [diff] [blame] | 630 | * The other side is TestClearPageMlocked() or shmem_lock(). |
Johannes Weiner | 6a7b954 | 2009-10-26 16:50:00 -0700 | [diff] [blame] | 631 | */ |
| 632 | smp_mb(); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 633 | } |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 634 | |
| 635 | /* |
| 636 | * page's status can change while we move it among lru. If an evictable |
| 637 | * page is on unevictable list, it never be freed. To avoid that, |
| 638 | * check after we added it to the list, again. |
| 639 | */ |
| 640 | if (lru == LRU_UNEVICTABLE && page_evictable(page, NULL)) { |
| 641 | if (!isolate_lru_page(page)) { |
| 642 | put_page(page); |
| 643 | goto redo; |
| 644 | } |
| 645 | /* This means someone else dropped this page from LRU |
| 646 | * So, it will be freed or putback to LRU again. There is |
| 647 | * nothing to do here. |
| 648 | */ |
| 649 | } |
| 650 | |
Lee Schermerhorn | bbfd28e | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 651 | if (was_unevictable && lru != LRU_UNEVICTABLE) |
| 652 | count_vm_event(UNEVICTABLE_PGRESCUED); |
| 653 | else if (!was_unevictable && lru == LRU_UNEVICTABLE) |
| 654 | count_vm_event(UNEVICTABLE_PGCULLED); |
| 655 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 656 | put_page(page); /* drop ref from isolate */ |
| 657 | } |
| 658 | |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 659 | enum page_references { |
| 660 | PAGEREF_RECLAIM, |
| 661 | PAGEREF_RECLAIM_CLEAN, |
Johannes Weiner | 6457474 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 662 | PAGEREF_KEEP, |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 663 | PAGEREF_ACTIVATE, |
| 664 | }; |
| 665 | |
| 666 | static enum page_references page_check_references(struct page *page, |
| 667 | struct scan_control *sc) |
| 668 | { |
Johannes Weiner | 6457474 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 669 | int referenced_ptes, referenced_page; |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 670 | unsigned long vm_flags; |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 671 | |
Johannes Weiner | 2e3d11f | 2012-05-29 15:06:25 -0700 | [diff] [blame] | 672 | referenced_ptes = page_referenced(page, 1, sc->target_mem_cgroup, |
| 673 | &vm_flags); |
Johannes Weiner | 6457474 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 674 | referenced_page = TestClearPageReferenced(page); |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 675 | |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 676 | /* |
| 677 | * Mlock lost the isolation race with us. Let try_to_unmap() |
| 678 | * move the page to the unevictable list. |
| 679 | */ |
| 680 | if (vm_flags & VM_LOCKED) |
| 681 | return PAGEREF_RECLAIM; |
| 682 | |
Johannes Weiner | 6457474 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 683 | if (referenced_ptes) { |
Michal Hocko | e369377 | 2012-05-29 15:06:45 -0700 | [diff] [blame] | 684 | if (PageSwapBacked(page)) |
Johannes Weiner | 6457474 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 685 | return PAGEREF_ACTIVATE; |
| 686 | /* |
| 687 | * All mapped pages start out with page table |
| 688 | * references from the instantiating fault, so we need |
| 689 | * to look twice if a mapped file page is used more |
| 690 | * than once. |
| 691 | * |
| 692 | * Mark it and spare it for another trip around the |
| 693 | * inactive list. Another page table reference will |
| 694 | * lead to its activation. |
| 695 | * |
| 696 | * Note: the mark is set for activated pages as well |
| 697 | * so that recently deactivated but used pages are |
| 698 | * quickly recovered. |
| 699 | */ |
| 700 | SetPageReferenced(page); |
| 701 | |
Konstantin Khlebnikov | 34dbc67 | 2012-01-10 15:06:59 -0800 | [diff] [blame] | 702 | if (referenced_page || referenced_ptes > 1) |
Johannes Weiner | 6457474 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 703 | return PAGEREF_ACTIVATE; |
| 704 | |
Konstantin Khlebnikov | c909e99 | 2012-01-10 15:07:03 -0800 | [diff] [blame] | 705 | /* |
| 706 | * Activate file-backed executable pages after first usage. |
| 707 | */ |
| 708 | if (vm_flags & VM_EXEC) |
| 709 | return PAGEREF_ACTIVATE; |
| 710 | |
Johannes Weiner | 6457474 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 711 | return PAGEREF_KEEP; |
| 712 | } |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 713 | |
| 714 | /* Reclaim if clean, defer dirty pages to writeback */ |
KOSAKI Motohiro | 2e30244 | 2010-10-26 14:21:46 -0700 | [diff] [blame] | 715 | if (referenced_page && !PageSwapBacked(page)) |
Johannes Weiner | 6457474 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 716 | return PAGEREF_RECLAIM_CLEAN; |
| 717 | |
| 718 | return PAGEREF_RECLAIM; |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 719 | } |
| 720 | |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 721 | /* |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 722 | * shrink_page_list() returns the number of reclaimed pages |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 723 | */ |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 724 | static unsigned long shrink_page_list(struct list_head *page_list, |
Konstantin Khlebnikov | 65f4b2c | 2012-05-29 15:06:59 -0700 | [diff] [blame] | 725 | struct zone *zone, |
Mel Gorman | f84f6e2 | 2011-10-31 17:07:51 -0700 | [diff] [blame] | 726 | struct scan_control *sc, |
Minchan Kim | aab50f1 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 727 | enum ttu_flags ttu_flags, |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 728 | unsigned long *ret_nr_dirty, |
Minchan Kim | aab50f1 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 729 | unsigned long *ret_nr_writeback, |
| 730 | bool force_reclaim) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 731 | { |
| 732 | LIST_HEAD(ret_pages); |
Mel Gorman | abe4c3b | 2010-08-09 17:19:31 -0700 | [diff] [blame] | 733 | LIST_HEAD(free_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 734 | int pgactivate = 0; |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 735 | unsigned long nr_dirty = 0; |
| 736 | unsigned long nr_congested = 0; |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 737 | unsigned long nr_reclaimed = 0; |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 738 | unsigned long nr_writeback = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 739 | |
| 740 | cond_resched(); |
| 741 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 742 | while (!list_empty(page_list)) { |
| 743 | struct address_space *mapping; |
| 744 | struct page *page; |
| 745 | int may_enter_fs; |
Minchan Kim | aab50f1 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 746 | enum page_references references = PAGEREF_RECLAIM_CLEAN; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 747 | |
| 748 | cond_resched(); |
| 749 | |
| 750 | page = lru_to_page(page_list); |
| 751 | list_del(&page->lru); |
| 752 | |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 753 | if (!trylock_page(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 754 | goto keep; |
| 755 | |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 756 | VM_BUG_ON(PageActive(page)); |
Konstantin Khlebnikov | 65f4b2c | 2012-05-29 15:06:59 -0700 | [diff] [blame] | 757 | VM_BUG_ON(page_zone(page) != zone); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 758 | |
| 759 | sc->nr_scanned++; |
Christoph Lameter | 80e4342 | 2006-02-11 17:55:53 -0800 | [diff] [blame] | 760 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 761 | if (unlikely(!page_evictable(page, NULL))) |
| 762 | goto cull_mlocked; |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 763 | |
Johannes Weiner | a6dc60f | 2009-03-31 15:19:30 -0700 | [diff] [blame] | 764 | if (!sc->may_unmap && page_mapped(page)) |
Christoph Lameter | 80e4342 | 2006-02-11 17:55:53 -0800 | [diff] [blame] | 765 | goto keep_locked; |
| 766 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 767 | /* Double the slab pressure for mapped and swapcache pages */ |
| 768 | if (page_mapped(page) || PageSwapCache(page)) |
| 769 | sc->nr_scanned++; |
| 770 | |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 771 | may_enter_fs = (sc->gfp_mask & __GFP_FS) || |
| 772 | (PageSwapCache(page) && (sc->gfp_mask & __GFP_IO)); |
| 773 | |
| 774 | if (PageWriteback(page)) { |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 775 | nr_writeback++; |
Mel Gorman | f65a6ed | 2012-05-29 15:06:19 -0700 | [diff] [blame] | 776 | unlock_page(page); |
| 777 | goto keep; |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 778 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 779 | |
Minchan Kim | aab50f1 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 780 | if (!force_reclaim) |
| 781 | references = page_check_references(page, sc); |
| 782 | |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 783 | switch (references) { |
| 784 | case PAGEREF_ACTIVATE: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 785 | goto activate_locked; |
Johannes Weiner | 6457474 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 786 | case PAGEREF_KEEP: |
| 787 | goto keep_locked; |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 788 | case PAGEREF_RECLAIM: |
| 789 | case PAGEREF_RECLAIM_CLEAN: |
| 790 | ; /* try to reclaim the page below */ |
| 791 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 792 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 793 | /* |
| 794 | * Anonymous process memory has backing store? |
| 795 | * Try to allocate it some swap space here. |
| 796 | */ |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 797 | if (PageAnon(page) && !PageSwapCache(page)) { |
Hugh Dickins | 63eb6b9 | 2008-11-19 15:36:37 -0800 | [diff] [blame] | 798 | if (!(sc->gfp_mask & __GFP_IO)) |
| 799 | goto keep_locked; |
Hugh Dickins | ac47b00 | 2009-01-06 14:39:39 -0800 | [diff] [blame] | 800 | if (!add_to_swap(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 801 | goto activate_locked; |
Hugh Dickins | 63eb6b9 | 2008-11-19 15:36:37 -0800 | [diff] [blame] | 802 | may_enter_fs = 1; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 803 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 804 | |
| 805 | mapping = page_mapping(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 806 | |
| 807 | /* |
| 808 | * The page is mapped into the page tables of one or more |
| 809 | * processes. Try to unmap it here. |
| 810 | */ |
| 811 | if (page_mapped(page) && mapping) { |
Minchan Kim | aab50f1 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 812 | switch (try_to_unmap(page, ttu_flags)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 813 | case SWAP_FAIL: |
| 814 | goto activate_locked; |
| 815 | case SWAP_AGAIN: |
| 816 | goto keep_locked; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 817 | case SWAP_MLOCK: |
| 818 | goto cull_mlocked; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 819 | case SWAP_SUCCESS: |
| 820 | ; /* try to free the page below */ |
| 821 | } |
| 822 | } |
| 823 | |
| 824 | if (PageDirty(page)) { |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 825 | nr_dirty++; |
| 826 | |
Mel Gorman | ee72886 | 2011-10-31 17:07:38 -0700 | [diff] [blame] | 827 | /* |
| 828 | * Only kswapd can writeback filesystem pages to |
Mel Gorman | f84f6e2 | 2011-10-31 17:07:51 -0700 | [diff] [blame] | 829 | * avoid risk of stack overflow but do not writeback |
| 830 | * unless under significant pressure. |
Mel Gorman | ee72886 | 2011-10-31 17:07:38 -0700 | [diff] [blame] | 831 | */ |
Mel Gorman | f84f6e2 | 2011-10-31 17:07:51 -0700 | [diff] [blame] | 832 | if (page_is_file_cache(page) && |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 833 | (!current_is_kswapd() || |
| 834 | sc->priority >= DEF_PRIORITY - 2)) { |
Mel Gorman | 49ea7eb | 2011-10-31 17:07:59 -0700 | [diff] [blame] | 835 | /* |
| 836 | * Immediately reclaim when written back. |
| 837 | * Similar in principal to deactivate_page() |
| 838 | * except we already have the page isolated |
| 839 | * and know it's dirty |
| 840 | */ |
| 841 | inc_zone_page_state(page, NR_VMSCAN_IMMEDIATE); |
| 842 | SetPageReclaim(page); |
| 843 | |
Mel Gorman | ee72886 | 2011-10-31 17:07:38 -0700 | [diff] [blame] | 844 | goto keep_locked; |
| 845 | } |
| 846 | |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 847 | if (references == PAGEREF_RECLAIM_CLEAN) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 848 | goto keep_locked; |
Andrew Morton | 4dd4b92 | 2008-03-24 12:29:52 -0700 | [diff] [blame] | 849 | if (!may_enter_fs) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 850 | goto keep_locked; |
Christoph Lameter | 52a8363 | 2006-02-01 03:05:28 -0800 | [diff] [blame] | 851 | if (!sc->may_writepage) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 852 | goto keep_locked; |
| 853 | |
| 854 | /* Page is dirty, try to write it out here */ |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 855 | switch (pageout(page, mapping, sc)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 856 | case PAGE_KEEP: |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 857 | nr_congested++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 858 | goto keep_locked; |
| 859 | case PAGE_ACTIVATE: |
| 860 | goto activate_locked; |
| 861 | case PAGE_SUCCESS: |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 862 | if (PageWriteback(page)) |
Mel Gorman | f65a6ed | 2012-05-29 15:06:19 -0700 | [diff] [blame] | 863 | goto keep; |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 864 | if (PageDirty(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 865 | goto keep; |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 866 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 867 | /* |
| 868 | * A synchronous write - probably a ramdisk. Go |
| 869 | * ahead and try to reclaim the page. |
| 870 | */ |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 871 | if (!trylock_page(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 872 | goto keep; |
| 873 | if (PageDirty(page) || PageWriteback(page)) |
| 874 | goto keep_locked; |
| 875 | mapping = page_mapping(page); |
| 876 | case PAGE_CLEAN: |
| 877 | ; /* try to free the page below */ |
| 878 | } |
| 879 | } |
| 880 | |
| 881 | /* |
| 882 | * If the page has buffers, try to free the buffer mappings |
| 883 | * associated with this page. If we succeed we try to free |
| 884 | * the page as well. |
| 885 | * |
| 886 | * We do this even if the page is PageDirty(). |
| 887 | * try_to_release_page() does not perform I/O, but it is |
| 888 | * possible for a page to have PageDirty set, but it is actually |
| 889 | * clean (all its buffers are clean). This happens if the |
| 890 | * buffers were written out directly, with submit_bh(). ext3 |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 891 | * will do this, as well as the blockdev mapping. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 892 | * try_to_release_page() will discover that cleanness and will |
| 893 | * drop the buffers and mark the page clean - it can be freed. |
| 894 | * |
| 895 | * Rarely, pages can have buffers and no ->mapping. These are |
| 896 | * the pages which were not successfully invalidated in |
| 897 | * truncate_complete_page(). We try to drop those buffers here |
| 898 | * and if that worked, and the page is no longer mapped into |
| 899 | * process address space (page_count == 1) it can be freed. |
| 900 | * Otherwise, leave the page on the LRU so it is swappable. |
| 901 | */ |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 902 | if (page_has_private(page)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 903 | if (!try_to_release_page(page, sc->gfp_mask)) |
| 904 | goto activate_locked; |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 905 | if (!mapping && page_count(page) == 1) { |
| 906 | unlock_page(page); |
| 907 | if (put_page_testzero(page)) |
| 908 | goto free_it; |
| 909 | else { |
| 910 | /* |
| 911 | * rare race with speculative reference. |
| 912 | * the speculative reference will free |
| 913 | * this page shortly, so we may |
| 914 | * increment nr_reclaimed here (and |
| 915 | * leave it off the LRU). |
| 916 | */ |
| 917 | nr_reclaimed++; |
| 918 | continue; |
| 919 | } |
| 920 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 921 | } |
| 922 | |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 923 | if (!mapping || !__remove_mapping(mapping, page)) |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 924 | goto keep_locked; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 925 | |
Nick Piggin | a978d6f | 2008-10-18 20:26:58 -0700 | [diff] [blame] | 926 | /* |
| 927 | * At this point, we have no other references and there is |
| 928 | * no way to pick any more up (removed from LRU, removed |
| 929 | * from pagecache). Can use non-atomic bitops now (and |
| 930 | * we obviously don't have to worry about waking up a process |
| 931 | * waiting on the page lock, because there are no references. |
| 932 | */ |
| 933 | __clear_page_locked(page); |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 934 | free_it: |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 935 | nr_reclaimed++; |
Mel Gorman | abe4c3b | 2010-08-09 17:19:31 -0700 | [diff] [blame] | 936 | |
| 937 | /* |
| 938 | * Is there need to periodically free_page_list? It would |
| 939 | * appear not as the counts should be low |
| 940 | */ |
| 941 | list_add(&page->lru, &free_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 942 | continue; |
| 943 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 944 | cull_mlocked: |
Hugh Dickins | 63d6c5a | 2009-01-06 14:39:38 -0800 | [diff] [blame] | 945 | if (PageSwapCache(page)) |
| 946 | try_to_free_swap(page); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 947 | unlock_page(page); |
| 948 | putback_lru_page(page); |
| 949 | continue; |
| 950 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 951 | activate_locked: |
Rik van Riel | 68a22394 | 2008-10-18 20:26:23 -0700 | [diff] [blame] | 952 | /* Not a candidate for swapping, so reclaim swap space. */ |
| 953 | if (PageSwapCache(page) && vm_swap_full()) |
Hugh Dickins | a2c43ee | 2009-01-06 14:39:36 -0800 | [diff] [blame] | 954 | try_to_free_swap(page); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 955 | VM_BUG_ON(PageActive(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 956 | SetPageActive(page); |
| 957 | pgactivate++; |
| 958 | keep_locked: |
| 959 | unlock_page(page); |
| 960 | keep: |
| 961 | list_add(&page->lru, &ret_pages); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 962 | VM_BUG_ON(PageLRU(page) || PageUnevictable(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 963 | } |
Mel Gorman | abe4c3b | 2010-08-09 17:19:31 -0700 | [diff] [blame] | 964 | |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 965 | /* |
| 966 | * Tag a zone as congested if all the dirty pages encountered were |
| 967 | * backed by a congested BDI. In this case, reclaimers should just |
| 968 | * back off and wait for congestion to clear because further reclaim |
| 969 | * will encounter the same problem |
| 970 | */ |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 971 | if (nr_dirty && nr_dirty == nr_congested && global_reclaim(sc)) |
Konstantin Khlebnikov | 65f4b2c | 2012-05-29 15:06:59 -0700 | [diff] [blame] | 972 | zone_set_flag(zone, ZONE_CONGESTED); |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 973 | |
Konstantin Khlebnikov | cc59850 | 2012-01-10 15:07:04 -0800 | [diff] [blame] | 974 | free_hot_cold_page_list(&free_pages, 1); |
Mel Gorman | abe4c3b | 2010-08-09 17:19:31 -0700 | [diff] [blame] | 975 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 976 | list_splice(&ret_pages, page_list); |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 977 | count_vm_events(PGACTIVATE, pgactivate); |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 978 | *ret_nr_dirty += nr_dirty; |
| 979 | *ret_nr_writeback += nr_writeback; |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 980 | return nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 981 | } |
| 982 | |
Minchan Kim | aab50f1 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 983 | unsigned long reclaim_clean_pages_from_list(struct zone *zone, |
| 984 | struct list_head *page_list) |
| 985 | { |
| 986 | struct scan_control sc = { |
| 987 | .gfp_mask = GFP_KERNEL, |
| 988 | .priority = DEF_PRIORITY, |
| 989 | .may_unmap = 1, |
| 990 | }; |
| 991 | unsigned long ret, dummy1, dummy2; |
| 992 | struct page *page, *next; |
| 993 | LIST_HEAD(clean_pages); |
| 994 | |
| 995 | list_for_each_entry_safe(page, next, page_list, lru) { |
| 996 | if (page_is_file_cache(page) && !PageDirty(page)) { |
| 997 | ClearPageActive(page); |
| 998 | list_move(&page->lru, &clean_pages); |
| 999 | } |
| 1000 | } |
| 1001 | |
| 1002 | ret = shrink_page_list(&clean_pages, zone, &sc, |
| 1003 | TTU_UNMAP|TTU_IGNORE_ACCESS, |
| 1004 | &dummy1, &dummy2, true); |
| 1005 | list_splice(&clean_pages, page_list); |
| 1006 | __mod_zone_page_state(zone, NR_ISOLATED_FILE, -ret); |
| 1007 | return ret; |
| 1008 | } |
| 1009 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1010 | /* |
| 1011 | * Attempt to remove the specified page from its LRU. Only take this page |
| 1012 | * if it is of the appropriate PageActive status. Pages which are being |
| 1013 | * freed elsewhere are also ignored. |
| 1014 | * |
| 1015 | * page: page to consider |
| 1016 | * mode: one of the LRU isolation modes defined above |
| 1017 | * |
| 1018 | * returns 0 on success, -ve errno on failure. |
| 1019 | */ |
Konstantin Khlebnikov | 6d8a50e | 2012-05-29 15:06:54 -0700 | [diff] [blame] | 1020 | int __isolate_lru_page(struct page *page, isolate_mode_t mode) |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1021 | { |
| 1022 | int ret = -EINVAL; |
| 1023 | |
| 1024 | /* Only take pages on the LRU. */ |
| 1025 | if (!PageLRU(page)) |
| 1026 | return ret; |
| 1027 | |
Minchan Kim | d05b452 | 2012-10-08 16:33:48 -0700 | [diff] [blame] | 1028 | /* Compaction should not handle unevictable pages but CMA can do so */ |
| 1029 | if (PageUnevictable(page) && !(mode & ISOLATE_UNEVICTABLE)) |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1030 | return ret; |
| 1031 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1032 | ret = -EBUSY; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 1033 | |
Mel Gorman | c824493 | 2012-01-12 17:19:38 -0800 | [diff] [blame] | 1034 | /* |
| 1035 | * To minimise LRU disruption, the caller can indicate that it only |
| 1036 | * wants to isolate pages it will be able to operate on without |
| 1037 | * blocking - clean pages for the most part. |
| 1038 | * |
| 1039 | * ISOLATE_CLEAN means that only clean pages should be isolated. This |
| 1040 | * is used by reclaim when it is cannot write to backing storage |
| 1041 | * |
| 1042 | * ISOLATE_ASYNC_MIGRATE is used to indicate that it only wants to pages |
| 1043 | * that it is possible to migrate without blocking |
| 1044 | */ |
| 1045 | if (mode & (ISOLATE_CLEAN|ISOLATE_ASYNC_MIGRATE)) { |
| 1046 | /* All the caller can do on PageWriteback is block */ |
| 1047 | if (PageWriteback(page)) |
| 1048 | return ret; |
| 1049 | |
| 1050 | if (PageDirty(page)) { |
| 1051 | struct address_space *mapping; |
| 1052 | |
| 1053 | /* ISOLATE_CLEAN means only clean pages */ |
| 1054 | if (mode & ISOLATE_CLEAN) |
| 1055 | return ret; |
| 1056 | |
| 1057 | /* |
| 1058 | * Only pages without mappings or that have a |
| 1059 | * ->migratepage callback are possible to migrate |
| 1060 | * without blocking |
| 1061 | */ |
| 1062 | mapping = page_mapping(page); |
| 1063 | if (mapping && !mapping->a_ops->migratepage) |
| 1064 | return ret; |
| 1065 | } |
| 1066 | } |
Minchan Kim | 39deaf8 | 2011-10-31 17:06:51 -0700 | [diff] [blame] | 1067 | |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1068 | if ((mode & ISOLATE_UNMAPPED) && page_mapped(page)) |
| 1069 | return ret; |
| 1070 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1071 | if (likely(get_page_unless_zero(page))) { |
| 1072 | /* |
| 1073 | * Be careful not to clear PageLRU until after we're |
| 1074 | * sure the page is not being freed elsewhere -- the |
| 1075 | * page release code relies on it. |
| 1076 | */ |
| 1077 | ClearPageLRU(page); |
| 1078 | ret = 0; |
| 1079 | } |
| 1080 | |
| 1081 | return ret; |
| 1082 | } |
| 1083 | |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 1084 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1085 | * zone->lru_lock is heavily contended. Some of the functions that |
| 1086 | * shrink the lists perform better by taking out a batch of pages |
| 1087 | * and working on them outside the LRU lock. |
| 1088 | * |
| 1089 | * For pagecache intensive workloads, this function is the hottest |
| 1090 | * spot in the kernel (apart from copy_*_user functions). |
| 1091 | * |
| 1092 | * Appropriate locks must be held before calling this function. |
| 1093 | * |
| 1094 | * @nr_to_scan: The number of pages to look through on the list. |
Konstantin Khlebnikov | 3a15fe8 | 2012-05-29 15:06:58 -0700 | [diff] [blame] | 1095 | * @lruvec: The LRU vector to pull pages from. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1096 | * @dst: The temp list to put pages on to. |
Hugh Dickins | f626012 | 2012-01-12 17:20:06 -0800 | [diff] [blame] | 1097 | * @nr_scanned: The number of pages that were scanned. |
Rik van Riel | fe2c2a1 | 2012-03-21 16:33:51 -0700 | [diff] [blame] | 1098 | * @sc: The scan_control struct for this reclaim session |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1099 | * @mode: One of the LRU isolation modes |
Konstantin Khlebnikov | 7f8a920 | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 1100 | * @lru: LRU list id for isolating |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1101 | * |
| 1102 | * returns how many pages were moved onto *@dst. |
| 1103 | */ |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1104 | static unsigned long isolate_lru_pages(unsigned long nr_to_scan, |
Konstantin Khlebnikov | 3a15fe8 | 2012-05-29 15:06:58 -0700 | [diff] [blame] | 1105 | struct lruvec *lruvec, struct list_head *dst, |
Rik van Riel | fe2c2a1 | 2012-03-21 16:33:51 -0700 | [diff] [blame] | 1106 | unsigned long *nr_scanned, struct scan_control *sc, |
Konstantin Khlebnikov | 7f8a920 | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 1107 | isolate_mode_t mode, enum lru_list lru) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1108 | { |
Hugh Dickins | f626012 | 2012-01-12 17:20:06 -0800 | [diff] [blame] | 1109 | struct list_head *src; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1110 | unsigned long nr_taken = 0; |
Wu Fengguang | c9b02d9 | 2006-03-22 00:08:23 -0800 | [diff] [blame] | 1111 | unsigned long scan; |
Konstantin Khlebnikov | 7f8a920 | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 1112 | int file = is_file_lru(lru); |
Hugh Dickins | f626012 | 2012-01-12 17:20:06 -0800 | [diff] [blame] | 1113 | |
Hugh Dickins | f626012 | 2012-01-12 17:20:06 -0800 | [diff] [blame] | 1114 | src = &lruvec->lists[lru]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1115 | |
Wu Fengguang | c9b02d9 | 2006-03-22 00:08:23 -0800 | [diff] [blame] | 1116 | for (scan = 0; scan < nr_to_scan && !list_empty(src); scan++) { |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1117 | struct page *page; |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1118 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1119 | page = lru_to_page(src); |
| 1120 | prefetchw_prev_lru_page(page, src, flags); |
| 1121 | |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 1122 | VM_BUG_ON(!PageLRU(page)); |
Nick Piggin | 8d438f9 | 2006-03-22 00:07:59 -0800 | [diff] [blame] | 1123 | |
Konstantin Khlebnikov | 6d8a50e | 2012-05-29 15:06:54 -0700 | [diff] [blame] | 1124 | switch (__isolate_lru_page(page, mode)) { |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1125 | case 0: |
Konstantin Khlebnikov | 35b8473 | 2012-05-29 15:06:54 -0700 | [diff] [blame] | 1126 | mem_cgroup_lru_del_list(page, lru); |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1127 | list_move(&page->lru, dst); |
Rik van Riel | 2c888cf | 2011-01-13 15:47:13 -0800 | [diff] [blame] | 1128 | nr_taken += hpage_nr_pages(page); |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1129 | break; |
Nick Piggin | 46453a6 | 2006-03-22 00:07:58 -0800 | [diff] [blame] | 1130 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1131 | case -EBUSY: |
| 1132 | /* else it is being freed elsewhere */ |
| 1133 | list_move(&page->lru, src); |
| 1134 | continue; |
| 1135 | |
| 1136 | default: |
| 1137 | BUG(); |
| 1138 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1139 | } |
| 1140 | |
Hugh Dickins | f626012 | 2012-01-12 17:20:06 -0800 | [diff] [blame] | 1141 | *nr_scanned = scan; |
Mel Gorman | a8a94d1 | 2010-08-09 17:19:17 -0700 | [diff] [blame] | 1142 | |
Rik van Riel | fe2c2a1 | 2012-03-21 16:33:51 -0700 | [diff] [blame] | 1143 | trace_mm_vmscan_lru_isolate(sc->order, |
Mel Gorman | a8a94d1 | 2010-08-09 17:19:17 -0700 | [diff] [blame] | 1144 | nr_to_scan, scan, |
| 1145 | nr_taken, |
Tao Ma | ea4d349 | 2012-01-12 17:19:20 -0800 | [diff] [blame] | 1146 | mode, file); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1147 | return nr_taken; |
| 1148 | } |
| 1149 | |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1150 | /** |
| 1151 | * isolate_lru_page - tries to isolate a page from its LRU list |
| 1152 | * @page: page to isolate from its LRU list |
| 1153 | * |
| 1154 | * Isolates a @page from an LRU list, clears PageLRU and adjusts the |
| 1155 | * vmstat statistic corresponding to whatever LRU list the page was on. |
| 1156 | * |
| 1157 | * Returns 0 if the page was removed from an LRU list. |
| 1158 | * Returns -EBUSY if the page was not on an LRU list. |
| 1159 | * |
| 1160 | * The returned page will have PageLRU() cleared. If it was found on |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1161 | * the active list, it will have PageActive set. If it was found on |
| 1162 | * the unevictable list, it will have the PageUnevictable bit set. That flag |
| 1163 | * 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] | 1164 | * |
| 1165 | * The vmstat statistic corresponding to the list on which the page was |
| 1166 | * found will be decremented. |
| 1167 | * |
| 1168 | * Restrictions: |
| 1169 | * (1) Must be called with an elevated refcount on the page. This is a |
| 1170 | * fundamentnal difference from isolate_lru_pages (which is called |
| 1171 | * without a stable reference). |
| 1172 | * (2) the lru_lock must not be held. |
| 1173 | * (3) interrupts must be enabled. |
| 1174 | */ |
| 1175 | int isolate_lru_page(struct page *page) |
| 1176 | { |
| 1177 | int ret = -EBUSY; |
| 1178 | |
Konstantin Khlebnikov | 0c91731 | 2011-05-24 17:12:21 -0700 | [diff] [blame] | 1179 | VM_BUG_ON(!page_count(page)); |
| 1180 | |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1181 | if (PageLRU(page)) { |
| 1182 | struct zone *zone = page_zone(page); |
| 1183 | |
| 1184 | spin_lock_irq(&zone->lru_lock); |
Konstantin Khlebnikov | 0c91731 | 2011-05-24 17:12:21 -0700 | [diff] [blame] | 1185 | if (PageLRU(page)) { |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1186 | int lru = page_lru(page); |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1187 | ret = 0; |
Konstantin Khlebnikov | 0c91731 | 2011-05-24 17:12:21 -0700 | [diff] [blame] | 1188 | get_page(page); |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1189 | ClearPageLRU(page); |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1190 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1191 | del_page_from_lru_list(zone, page, lru); |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1192 | } |
| 1193 | spin_unlock_irq(&zone->lru_lock); |
| 1194 | } |
| 1195 | return ret; |
| 1196 | } |
| 1197 | |
Vinayak Menon | c702aaa | 2014-12-26 19:29:41 +0530 | [diff] [blame] | 1198 | static int __too_many_isolated(struct zone *zone, int file, |
| 1199 | struct scan_control *sc, int safe) |
Rik van Riel | 35cd781 | 2009-09-21 17:01:38 -0700 | [diff] [blame] | 1200 | { |
| 1201 | unsigned long inactive, isolated; |
| 1202 | |
Rik van Riel | 35cd781 | 2009-09-21 17:01:38 -0700 | [diff] [blame] | 1203 | if (file) { |
Vinayak Menon | c702aaa | 2014-12-26 19:29:41 +0530 | [diff] [blame] | 1204 | if (safe) { |
| 1205 | inactive = zone_page_state_snapshot(zone, |
| 1206 | NR_INACTIVE_FILE); |
| 1207 | isolated = zone_page_state_snapshot(zone, |
| 1208 | NR_ISOLATED_FILE); |
| 1209 | } else { |
| 1210 | inactive = zone_page_state(zone, NR_INACTIVE_FILE); |
| 1211 | isolated = zone_page_state(zone, NR_ISOLATED_FILE); |
| 1212 | } |
Rik van Riel | 35cd781 | 2009-09-21 17:01:38 -0700 | [diff] [blame] | 1213 | } else { |
Vinayak Menon | c702aaa | 2014-12-26 19:29:41 +0530 | [diff] [blame] | 1214 | if (safe) { |
| 1215 | inactive = zone_page_state_snapshot(zone, |
| 1216 | NR_INACTIVE_ANON); |
| 1217 | isolated = zone_page_state_snapshot(zone, |
| 1218 | NR_ISOLATED_ANON); |
| 1219 | } else { |
| 1220 | inactive = zone_page_state(zone, NR_INACTIVE_ANON); |
| 1221 | isolated = zone_page_state(zone, NR_ISOLATED_ANON); |
| 1222 | } |
Rik van Riel | 35cd781 | 2009-09-21 17:01:38 -0700 | [diff] [blame] | 1223 | } |
| 1224 | |
Fengguang Wu | 2f25e5a | 2012-12-18 14:23:31 -0800 | [diff] [blame] | 1225 | /* |
| 1226 | * GFP_NOIO/GFP_NOFS callers are allowed to isolate more pages, so they |
| 1227 | * won't get blocked by normal direct-reclaimers, forming a circular |
| 1228 | * deadlock. |
| 1229 | */ |
| 1230 | if ((sc->gfp_mask & GFP_IOFS) == GFP_IOFS) |
| 1231 | inactive >>= 3; |
| 1232 | |
Rik van Riel | 35cd781 | 2009-09-21 17:01:38 -0700 | [diff] [blame] | 1233 | return isolated > inactive; |
| 1234 | } |
| 1235 | |
Vinayak Menon | c702aaa | 2014-12-26 19:29:41 +0530 | [diff] [blame] | 1236 | /* |
| 1237 | * Are there way too many processes in the direct reclaim path already? |
| 1238 | */ |
| 1239 | static int too_many_isolated(struct zone *zone, int file, |
| 1240 | struct scan_control *sc, int safe) |
| 1241 | { |
| 1242 | if (current_is_kswapd()) |
| 1243 | return 0; |
| 1244 | |
| 1245 | if (!global_reclaim(sc)) |
| 1246 | return 0; |
| 1247 | |
| 1248 | if (unlikely(__too_many_isolated(zone, file, sc, 0))) { |
| 1249 | if (safe) |
| 1250 | return __too_many_isolated(zone, file, sc, safe); |
| 1251 | else |
| 1252 | return 1; |
| 1253 | } |
| 1254 | |
| 1255 | return 0; |
| 1256 | } |
| 1257 | |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1258 | static noinline_for_stack void |
Hugh Dickins | 3f79768 | 2012-01-12 17:20:07 -0800 | [diff] [blame] | 1259 | putback_inactive_pages(struct mem_cgroup_zone *mz, |
| 1260 | struct list_head *page_list) |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1261 | { |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1262 | struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(mz); |
Hugh Dickins | 3f79768 | 2012-01-12 17:20:07 -0800 | [diff] [blame] | 1263 | struct zone *zone = mz->zone; |
| 1264 | LIST_HEAD(pages_to_free); |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1265 | |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1266 | /* |
| 1267 | * Put back any unfreeable pages. |
| 1268 | */ |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1269 | while (!list_empty(page_list)) { |
Hugh Dickins | 3f79768 | 2012-01-12 17:20:07 -0800 | [diff] [blame] | 1270 | struct page *page = lru_to_page(page_list); |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1271 | int lru; |
Hugh Dickins | 3f79768 | 2012-01-12 17:20:07 -0800 | [diff] [blame] | 1272 | |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1273 | VM_BUG_ON(PageLRU(page)); |
| 1274 | list_del(&page->lru); |
| 1275 | if (unlikely(!page_evictable(page, NULL))) { |
| 1276 | spin_unlock_irq(&zone->lru_lock); |
| 1277 | putback_lru_page(page); |
| 1278 | spin_lock_irq(&zone->lru_lock); |
| 1279 | continue; |
| 1280 | } |
Linus Torvalds | 7a60857 | 2011-01-17 14:42:19 -0800 | [diff] [blame] | 1281 | SetPageLRU(page); |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1282 | lru = page_lru(page); |
Linus Torvalds | 7a60857 | 2011-01-17 14:42:19 -0800 | [diff] [blame] | 1283 | add_page_to_lru_list(zone, page, lru); |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1284 | if (is_active_lru(lru)) { |
| 1285 | int file = is_file_lru(lru); |
Rik van Riel | 9992af1 | 2011-01-13 15:47:13 -0800 | [diff] [blame] | 1286 | int numpages = hpage_nr_pages(page); |
| 1287 | reclaim_stat->recent_rotated[file] += numpages; |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1288 | } |
Hugh Dickins | 2bcf887 | 2012-01-12 17:19:56 -0800 | [diff] [blame] | 1289 | if (put_page_testzero(page)) { |
| 1290 | __ClearPageLRU(page); |
| 1291 | __ClearPageActive(page); |
| 1292 | del_page_from_lru_list(zone, page, lru); |
| 1293 | |
| 1294 | if (unlikely(PageCompound(page))) { |
| 1295 | spin_unlock_irq(&zone->lru_lock); |
| 1296 | (*get_compound_page_dtor(page))(page); |
| 1297 | spin_lock_irq(&zone->lru_lock); |
| 1298 | } else |
| 1299 | list_add(&page->lru, &pages_to_free); |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1300 | } |
| 1301 | } |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1302 | |
Hugh Dickins | 3f79768 | 2012-01-12 17:20:07 -0800 | [diff] [blame] | 1303 | /* |
| 1304 | * To save our caller's stack, now use input list for pages to free. |
| 1305 | */ |
| 1306 | list_splice(&pages_to_free, page_list); |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1307 | } |
| 1308 | |
| 1309 | /* |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 1310 | * shrink_inactive_list() is a helper for shrink_zone(). It returns the number |
| 1311 | * of reclaimed pages |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1312 | */ |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1313 | static noinline_for_stack unsigned long |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1314 | shrink_inactive_list(unsigned long nr_to_scan, struct mem_cgroup_zone *mz, |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1315 | struct scan_control *sc, enum lru_list lru) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1316 | { |
| 1317 | LIST_HEAD(page_list); |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1318 | unsigned long nr_scanned; |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 1319 | unsigned long nr_reclaimed = 0; |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1320 | unsigned long nr_taken; |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 1321 | unsigned long nr_dirty = 0; |
| 1322 | unsigned long nr_writeback = 0; |
Konstantin Khlebnikov | 6d8a50e | 2012-05-29 15:06:54 -0700 | [diff] [blame] | 1323 | isolate_mode_t isolate_mode = 0; |
Konstantin Khlebnikov | 7f8a920 | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 1324 | int file = is_file_lru(lru); |
Vinayak Menon | c702aaa | 2014-12-26 19:29:41 +0530 | [diff] [blame] | 1325 | int safe = 0; |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1326 | struct zone *zone = mz->zone; |
Hillf Danton | d563c05 | 2012-03-21 16:34:02 -0700 | [diff] [blame] | 1327 | struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(mz); |
Konstantin Khlebnikov | 3a15fe8 | 2012-05-29 15:06:58 -0700 | [diff] [blame] | 1328 | struct lruvec *lruvec = mem_cgroup_zone_lruvec(zone, mz->mem_cgroup); |
KOSAKI Motohiro | 78dc583 | 2009-06-16 15:31:40 -0700 | [diff] [blame] | 1329 | |
Vinayak Menon | c702aaa | 2014-12-26 19:29:41 +0530 | [diff] [blame] | 1330 | while (unlikely(too_many_isolated(zone, file, sc, safe))) { |
KOSAKI Motohiro | 58355c7 | 2009-10-26 16:49:35 -0700 | [diff] [blame] | 1331 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
Rik van Riel | 35cd781 | 2009-09-21 17:01:38 -0700 | [diff] [blame] | 1332 | |
| 1333 | /* We are about to die and free our memory. Return now. */ |
| 1334 | if (fatal_signal_pending(current)) |
| 1335 | return SWAP_CLUSTER_MAX; |
Vinayak Menon | c702aaa | 2014-12-26 19:29:41 +0530 | [diff] [blame] | 1336 | |
| 1337 | safe = 1; |
Rik van Riel | 35cd781 | 2009-09-21 17:01:38 -0700 | [diff] [blame] | 1338 | } |
| 1339 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1340 | lru_add_drain(); |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1341 | |
| 1342 | if (!sc->may_unmap) |
Hillf Danton | 6131728 | 2012-03-21 16:33:48 -0700 | [diff] [blame] | 1343 | isolate_mode |= ISOLATE_UNMAPPED; |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1344 | if (!sc->may_writepage) |
Hillf Danton | 6131728 | 2012-03-21 16:33:48 -0700 | [diff] [blame] | 1345 | isolate_mode |= ISOLATE_CLEAN; |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1346 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1347 | spin_lock_irq(&zone->lru_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1348 | |
Konstantin Khlebnikov | 3a15fe8 | 2012-05-29 15:06:58 -0700 | [diff] [blame] | 1349 | nr_taken = isolate_lru_pages(nr_to_scan, lruvec, &page_list, |
| 1350 | &nr_scanned, sc, isolate_mode, lru); |
Konstantin Khlebnikov | 650e095 | 2012-05-29 15:06:59 -0700 | [diff] [blame] | 1351 | |
| 1352 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, -nr_taken); |
| 1353 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, nr_taken); |
| 1354 | |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 1355 | if (global_reclaim(sc)) { |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1356 | zone->pages_scanned += nr_scanned; |
KOSAKI Motohiro | b35ea17 | 2009-09-21 17:01:36 -0700 | [diff] [blame] | 1357 | if (current_is_kswapd()) |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1358 | __count_zone_vm_events(PGSCAN_KSWAPD, zone, |
| 1359 | nr_scanned); |
| 1360 | else |
| 1361 | __count_zone_vm_events(PGSCAN_DIRECT, zone, |
| 1362 | nr_scanned); |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1363 | } |
Hillf Danton | d563c05 | 2012-03-21 16:34:02 -0700 | [diff] [blame] | 1364 | spin_unlock_irq(&zone->lru_lock); |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 1365 | |
Hillf Danton | d563c05 | 2012-03-21 16:34:02 -0700 | [diff] [blame] | 1366 | if (nr_taken == 0) |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1367 | return 0; |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1368 | |
Minchan Kim | aab50f1 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 1369 | nr_reclaimed = shrink_page_list(&page_list, zone, sc, TTU_UNMAP, |
| 1370 | &nr_dirty, &nr_writeback, false); |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1371 | |
Hugh Dickins | 3f79768 | 2012-01-12 17:20:07 -0800 | [diff] [blame] | 1372 | spin_lock_irq(&zone->lru_lock); |
| 1373 | |
Konstantin Khlebnikov | 650e095 | 2012-05-29 15:06:59 -0700 | [diff] [blame] | 1374 | reclaim_stat->recent_scanned[file] += nr_taken; |
Hillf Danton | d563c05 | 2012-03-21 16:34:02 -0700 | [diff] [blame] | 1375 | |
Ying Han | 904249a | 2012-04-25 16:01:48 -0700 | [diff] [blame] | 1376 | if (global_reclaim(sc)) { |
| 1377 | if (current_is_kswapd()) |
| 1378 | __count_zone_vm_events(PGSTEAL_KSWAPD, zone, |
| 1379 | nr_reclaimed); |
| 1380 | else |
| 1381 | __count_zone_vm_events(PGSTEAL_DIRECT, zone, |
| 1382 | nr_reclaimed); |
| 1383 | } |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1384 | |
Hugh Dickins | 3f79768 | 2012-01-12 17:20:07 -0800 | [diff] [blame] | 1385 | putback_inactive_pages(mz, &page_list); |
| 1386 | |
Konstantin Khlebnikov | 650e095 | 2012-05-29 15:06:59 -0700 | [diff] [blame] | 1387 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, -nr_taken); |
Hugh Dickins | 3f79768 | 2012-01-12 17:20:07 -0800 | [diff] [blame] | 1388 | |
| 1389 | spin_unlock_irq(&zone->lru_lock); |
| 1390 | |
| 1391 | free_hot_cold_page_list(&page_list, 1); |
Mel Gorman | e11da5b | 2010-10-26 14:21:40 -0700 | [diff] [blame] | 1392 | |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 1393 | /* |
| 1394 | * If reclaim is isolating dirty pages under writeback, it implies |
| 1395 | * that the long-lived page allocation rate is exceeding the page |
| 1396 | * laundering rate. Either the global limits are not being effective |
| 1397 | * at throttling processes due to the page distribution throughout |
| 1398 | * zones or there is heavy usage of a slow backing device. The |
| 1399 | * only option is to throttle from reclaim context which is not ideal |
| 1400 | * as there is no guarantee the dirtying process is throttled in the |
| 1401 | * same way balance_dirty_pages() manages. |
| 1402 | * |
| 1403 | * This scales the number of dirty pages that must be under writeback |
| 1404 | * before throttling depending on priority. It is a simple backoff |
| 1405 | * function that has the most effect in the range DEF_PRIORITY to |
| 1406 | * DEF_PRIORITY-2 which is the priority reclaim is considered to be |
| 1407 | * in trouble and reclaim is considered to be in trouble. |
| 1408 | * |
| 1409 | * DEF_PRIORITY 100% isolated pages must be PageWriteback to throttle |
| 1410 | * DEF_PRIORITY-1 50% must be PageWriteback |
| 1411 | * DEF_PRIORITY-2 25% must be PageWriteback, kswapd in trouble |
| 1412 | * ... |
| 1413 | * DEF_PRIORITY-6 For SWAP_CLUSTER_MAX isolated pages, throttle if any |
| 1414 | * isolated page is PageWriteback |
| 1415 | */ |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1416 | if (nr_writeback && nr_writeback >= |
| 1417 | (nr_taken >> (DEF_PRIORITY - sc->priority))) |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 1418 | wait_iff_congested(zone, BLK_RW_ASYNC, HZ/10); |
| 1419 | |
Mel Gorman | e11da5b | 2010-10-26 14:21:40 -0700 | [diff] [blame] | 1420 | trace_mm_vmscan_lru_shrink_inactive(zone->zone_pgdat->node_id, |
| 1421 | zone_idx(zone), |
| 1422 | nr_scanned, nr_reclaimed, |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1423 | sc->priority, |
Mel Gorman | 626ebe7 | 2012-05-29 15:06:20 -0700 | [diff] [blame] | 1424 | trace_shrink_flags(file)); |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 1425 | return nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1426 | } |
| 1427 | |
Martin Bligh | 3bb1a85 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 1428 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1429 | * This moves pages from the active list to the inactive list. |
| 1430 | * |
| 1431 | * We move them the other way if the page is referenced by one or more |
| 1432 | * processes, from rmap. |
| 1433 | * |
| 1434 | * If the pages are mostly unmapped, the processing is fast and it is |
| 1435 | * appropriate to hold zone->lru_lock across the whole operation. But if |
| 1436 | * the pages are mapped, the processing is slow (page_referenced()) so we |
| 1437 | * should drop zone->lru_lock around each page. It's impossible to balance |
| 1438 | * this, so instead we remove the pages from the LRU while processing them. |
| 1439 | * It is safe to rely on PG_active against the non-LRU pages in here because |
| 1440 | * nobody will play with that bit on a non-LRU page. |
| 1441 | * |
| 1442 | * The downside is that we have to touch page->_count against each page. |
| 1443 | * But we had to alter page->flags anyway. |
| 1444 | */ |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1445 | |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1446 | static void move_active_pages_to_lru(struct zone *zone, |
| 1447 | struct list_head *list, |
Hugh Dickins | 2bcf887 | 2012-01-12 17:19:56 -0800 | [diff] [blame] | 1448 | struct list_head *pages_to_free, |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1449 | enum lru_list lru) |
| 1450 | { |
| 1451 | unsigned long pgmoved = 0; |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1452 | struct page *page; |
| 1453 | |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1454 | while (!list_empty(list)) { |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1455 | struct lruvec *lruvec; |
| 1456 | |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1457 | page = lru_to_page(list); |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1458 | |
| 1459 | VM_BUG_ON(PageLRU(page)); |
| 1460 | SetPageLRU(page); |
| 1461 | |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1462 | lruvec = mem_cgroup_lru_add_list(zone, page, lru); |
| 1463 | list_move(&page->lru, &lruvec->lists[lru]); |
Rik van Riel | 2c888cf | 2011-01-13 15:47:13 -0800 | [diff] [blame] | 1464 | pgmoved += hpage_nr_pages(page); |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1465 | |
Hugh Dickins | 2bcf887 | 2012-01-12 17:19:56 -0800 | [diff] [blame] | 1466 | if (put_page_testzero(page)) { |
| 1467 | __ClearPageLRU(page); |
| 1468 | __ClearPageActive(page); |
| 1469 | del_page_from_lru_list(zone, page, lru); |
| 1470 | |
| 1471 | if (unlikely(PageCompound(page))) { |
| 1472 | spin_unlock_irq(&zone->lru_lock); |
| 1473 | (*get_compound_page_dtor(page))(page); |
| 1474 | spin_lock_irq(&zone->lru_lock); |
| 1475 | } else |
| 1476 | list_add(&page->lru, pages_to_free); |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1477 | } |
| 1478 | } |
| 1479 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved); |
| 1480 | if (!is_active_lru(lru)) |
| 1481 | __count_vm_events(PGDEACTIVATE, pgmoved); |
| 1482 | } |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1483 | |
Hugh Dickins | f626012 | 2012-01-12 17:20:06 -0800 | [diff] [blame] | 1484 | static void shrink_active_list(unsigned long nr_to_scan, |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1485 | struct mem_cgroup_zone *mz, |
| 1486 | struct scan_control *sc, |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1487 | enum lru_list lru) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1488 | { |
KOSAKI Motohiro | 44c241f | 2009-09-21 17:01:35 -0700 | [diff] [blame] | 1489 | unsigned long nr_taken; |
Hugh Dickins | f626012 | 2012-01-12 17:20:06 -0800 | [diff] [blame] | 1490 | unsigned long nr_scanned; |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 1491 | unsigned long vm_flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1492 | LIST_HEAD(l_hold); /* The pages which were snipped off */ |
Wu Fengguang | 8cab475 | 2009-06-16 15:33:12 -0700 | [diff] [blame] | 1493 | LIST_HEAD(l_active); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1494 | LIST_HEAD(l_inactive); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1495 | struct page *page; |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1496 | struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(mz); |
KOSAKI Motohiro | 44c241f | 2009-09-21 17:01:35 -0700 | [diff] [blame] | 1497 | unsigned long nr_rotated = 0; |
Konstantin Khlebnikov | 6d8a50e | 2012-05-29 15:06:54 -0700 | [diff] [blame] | 1498 | isolate_mode_t isolate_mode = 0; |
Konstantin Khlebnikov | 7f8a920 | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 1499 | int file = is_file_lru(lru); |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1500 | struct zone *zone = mz->zone; |
Konstantin Khlebnikov | 3a15fe8 | 2012-05-29 15:06:58 -0700 | [diff] [blame] | 1501 | struct lruvec *lruvec = mem_cgroup_zone_lruvec(zone, mz->mem_cgroup); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1502 | |
| 1503 | lru_add_drain(); |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1504 | |
| 1505 | if (!sc->may_unmap) |
Hillf Danton | 6131728 | 2012-03-21 16:33:48 -0700 | [diff] [blame] | 1506 | isolate_mode |= ISOLATE_UNMAPPED; |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1507 | if (!sc->may_writepage) |
Hillf Danton | 6131728 | 2012-03-21 16:33:48 -0700 | [diff] [blame] | 1508 | isolate_mode |= ISOLATE_CLEAN; |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1509 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1510 | spin_lock_irq(&zone->lru_lock); |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1511 | |
Konstantin Khlebnikov | 3a15fe8 | 2012-05-29 15:06:58 -0700 | [diff] [blame] | 1512 | nr_taken = isolate_lru_pages(nr_to_scan, lruvec, &l_hold, |
| 1513 | &nr_scanned, sc, isolate_mode, lru); |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 1514 | if (global_reclaim(sc)) |
Hugh Dickins | f626012 | 2012-01-12 17:20:06 -0800 | [diff] [blame] | 1515 | zone->pages_scanned += nr_scanned; |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 1516 | |
Johannes Weiner | b7c46d1 | 2009-09-21 17:02:56 -0700 | [diff] [blame] | 1517 | reclaim_stat->recent_scanned[file] += nr_taken; |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1518 | |
Hugh Dickins | f626012 | 2012-01-12 17:20:06 -0800 | [diff] [blame] | 1519 | __count_zone_vm_events(PGREFILL, zone, nr_scanned); |
Konstantin Khlebnikov | 7f8a920 | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 1520 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, -nr_taken); |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 1521 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, nr_taken); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1522 | spin_unlock_irq(&zone->lru_lock); |
| 1523 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1524 | while (!list_empty(&l_hold)) { |
| 1525 | cond_resched(); |
| 1526 | page = lru_to_page(&l_hold); |
| 1527 | list_del(&page->lru); |
Rik van Riel | 7e9cd48 | 2008-10-18 20:26:35 -0700 | [diff] [blame] | 1528 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1529 | if (unlikely(!page_evictable(page, NULL))) { |
| 1530 | putback_lru_page(page); |
| 1531 | continue; |
| 1532 | } |
| 1533 | |
Mel Gorman | cc715d9 | 2012-03-21 16:34:00 -0700 | [diff] [blame] | 1534 | if (unlikely(buffer_heads_over_limit)) { |
| 1535 | if (page_has_private(page) && trylock_page(page)) { |
| 1536 | if (page_has_private(page)) |
| 1537 | try_to_release_page(page, 0); |
| 1538 | unlock_page(page); |
| 1539 | } |
| 1540 | } |
| 1541 | |
Johannes Weiner | 2e3d11f | 2012-05-29 15:06:25 -0700 | [diff] [blame] | 1542 | if (page_referenced(page, 0, sc->target_mem_cgroup, |
| 1543 | &vm_flags)) { |
Rik van Riel | 9992af1 | 2011-01-13 15:47:13 -0800 | [diff] [blame] | 1544 | nr_rotated += hpage_nr_pages(page); |
Wu Fengguang | 8cab475 | 2009-06-16 15:33:12 -0700 | [diff] [blame] | 1545 | /* |
| 1546 | * Identify referenced, file-backed active pages and |
| 1547 | * give them one more trip around the active list. So |
| 1548 | * that executable code get better chances to stay in |
| 1549 | * memory under moderate memory pressure. Anon pages |
| 1550 | * are not likely to be evicted by use-once streaming |
| 1551 | * IO, plus JVM can create lots of anon VM_EXEC pages, |
| 1552 | * so we ignore them here. |
| 1553 | */ |
Wu Fengguang | 41e2098 | 2009-10-26 16:49:53 -0700 | [diff] [blame] | 1554 | if ((vm_flags & VM_EXEC) && page_is_file_cache(page)) { |
Wu Fengguang | 8cab475 | 2009-06-16 15:33:12 -0700 | [diff] [blame] | 1555 | list_add(&page->lru, &l_active); |
| 1556 | continue; |
| 1557 | } |
| 1558 | } |
Rik van Riel | 7e9cd48 | 2008-10-18 20:26:35 -0700 | [diff] [blame] | 1559 | |
KOSAKI Motohiro | 5205e56 | 2009-09-21 17:01:44 -0700 | [diff] [blame] | 1560 | ClearPageActive(page); /* we are de-activating */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1561 | list_add(&page->lru, &l_inactive); |
| 1562 | } |
| 1563 | |
Andrew Morton | b555749 | 2009-01-06 14:40:13 -0800 | [diff] [blame] | 1564 | /* |
Wu Fengguang | 8cab475 | 2009-06-16 15:33:12 -0700 | [diff] [blame] | 1565 | * Move pages back to the lru list. |
Andrew Morton | b555749 | 2009-01-06 14:40:13 -0800 | [diff] [blame] | 1566 | */ |
Johannes Weiner | 2a1dc50 | 2008-12-01 03:00:35 +0100 | [diff] [blame] | 1567 | spin_lock_irq(&zone->lru_lock); |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1568 | /* |
Wu Fengguang | 8cab475 | 2009-06-16 15:33:12 -0700 | [diff] [blame] | 1569 | * Count referenced pages from currently used mappings as rotated, |
| 1570 | * even though only some of them are actually re-activated. This |
| 1571 | * helps balance scan pressure between file and anonymous pages in |
| 1572 | * get_scan_ratio. |
Rik van Riel | 7e9cd48 | 2008-10-18 20:26:35 -0700 | [diff] [blame] | 1573 | */ |
Johannes Weiner | b7c46d1 | 2009-09-21 17:02:56 -0700 | [diff] [blame] | 1574 | reclaim_stat->recent_rotated[file] += nr_rotated; |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1575 | |
Konstantin Khlebnikov | 7f8a920 | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 1576 | move_active_pages_to_lru(zone, &l_active, &l_hold, lru); |
| 1577 | move_active_pages_to_lru(zone, &l_inactive, &l_hold, lru - LRU_ACTIVE); |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 1578 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, -nr_taken); |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 1579 | spin_unlock_irq(&zone->lru_lock); |
Hugh Dickins | 2bcf887 | 2012-01-12 17:19:56 -0800 | [diff] [blame] | 1580 | |
| 1581 | free_hot_cold_page_list(&l_hold, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1582 | } |
| 1583 | |
Minchan Kim | 74e3f3c | 2010-10-26 14:21:31 -0700 | [diff] [blame] | 1584 | #ifdef CONFIG_SWAP |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1585 | static int inactive_anon_is_low_global(struct zone *zone) |
KOSAKI Motohiro | f89eb90 | 2009-01-07 18:08:14 -0800 | [diff] [blame] | 1586 | { |
| 1587 | unsigned long active, inactive; |
| 1588 | |
| 1589 | active = zone_page_state(zone, NR_ACTIVE_ANON); |
| 1590 | inactive = zone_page_state(zone, NR_INACTIVE_ANON); |
| 1591 | |
| 1592 | if (inactive * zone->inactive_ratio < active) |
| 1593 | return 1; |
| 1594 | |
| 1595 | return 0; |
| 1596 | } |
| 1597 | |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1598 | /** |
| 1599 | * inactive_anon_is_low - check if anonymous pages need to be deactivated |
| 1600 | * @zone: zone to check |
| 1601 | * @sc: scan control of this context |
| 1602 | * |
| 1603 | * Returns true if the zone does not have enough inactive anon pages, |
| 1604 | * meaning some active anon pages need to be deactivated. |
| 1605 | */ |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1606 | static int inactive_anon_is_low(struct mem_cgroup_zone *mz) |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1607 | { |
Minchan Kim | 74e3f3c | 2010-10-26 14:21:31 -0700 | [diff] [blame] | 1608 | /* |
| 1609 | * If we don't have swap space, anonymous page deactivation |
| 1610 | * is pointless. |
| 1611 | */ |
| 1612 | if (!total_swap_pages) |
| 1613 | return 0; |
| 1614 | |
Hugh Dickins | c12af16 | 2012-05-29 15:06:52 -0700 | [diff] [blame] | 1615 | if (!mem_cgroup_disabled()) |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1616 | return mem_cgroup_inactive_anon_is_low(mz->mem_cgroup, |
| 1617 | mz->zone); |
| 1618 | |
| 1619 | return inactive_anon_is_low_global(mz->zone); |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1620 | } |
Minchan Kim | 74e3f3c | 2010-10-26 14:21:31 -0700 | [diff] [blame] | 1621 | #else |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1622 | static inline int inactive_anon_is_low(struct mem_cgroup_zone *mz) |
Minchan Kim | 74e3f3c | 2010-10-26 14:21:31 -0700 | [diff] [blame] | 1623 | { |
| 1624 | return 0; |
| 1625 | } |
| 1626 | #endif |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1627 | |
Rik van Riel | 56e49d2 | 2009-06-16 15:32:28 -0700 | [diff] [blame] | 1628 | static int inactive_file_is_low_global(struct zone *zone) |
| 1629 | { |
| 1630 | unsigned long active, inactive; |
| 1631 | |
| 1632 | active = zone_page_state(zone, NR_ACTIVE_FILE); |
| 1633 | inactive = zone_page_state(zone, NR_INACTIVE_FILE); |
| 1634 | |
| 1635 | return (active > inactive); |
| 1636 | } |
| 1637 | |
| 1638 | /** |
| 1639 | * inactive_file_is_low - check if file pages need to be deactivated |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1640 | * @mz: memory cgroup and zone to check |
Rik van Riel | 56e49d2 | 2009-06-16 15:32:28 -0700 | [diff] [blame] | 1641 | * |
| 1642 | * When the system is doing streaming IO, memory pressure here |
| 1643 | * ensures that active file pages get deactivated, until more |
| 1644 | * than half of the file pages are on the inactive list. |
| 1645 | * |
| 1646 | * Once we get to that situation, protect the system's working |
| 1647 | * set from being evicted by disabling active file page aging. |
| 1648 | * |
| 1649 | * This uses a different ratio than the anonymous pages, because |
| 1650 | * the page cache uses a use-once replacement algorithm. |
| 1651 | */ |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1652 | static int inactive_file_is_low(struct mem_cgroup_zone *mz) |
Rik van Riel | 56e49d2 | 2009-06-16 15:32:28 -0700 | [diff] [blame] | 1653 | { |
Hugh Dickins | c12af16 | 2012-05-29 15:06:52 -0700 | [diff] [blame] | 1654 | if (!mem_cgroup_disabled()) |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1655 | return mem_cgroup_inactive_file_is_low(mz->mem_cgroup, |
| 1656 | mz->zone); |
Rik van Riel | 56e49d2 | 2009-06-16 15:32:28 -0700 | [diff] [blame] | 1657 | |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1658 | return inactive_file_is_low_global(mz->zone); |
Rik van Riel | 56e49d2 | 2009-06-16 15:32:28 -0700 | [diff] [blame] | 1659 | } |
| 1660 | |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1661 | static int inactive_list_is_low(struct mem_cgroup_zone *mz, int file) |
Rik van Riel | b39415b | 2009-12-14 17:59:48 -0800 | [diff] [blame] | 1662 | { |
| 1663 | if (file) |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1664 | return inactive_file_is_low(mz); |
Rik van Riel | b39415b | 2009-12-14 17:59:48 -0800 | [diff] [blame] | 1665 | else |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1666 | return inactive_anon_is_low(mz); |
Rik van Riel | b39415b | 2009-12-14 17:59:48 -0800 | [diff] [blame] | 1667 | } |
| 1668 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1669 | static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan, |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1670 | struct mem_cgroup_zone *mz, |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1671 | struct scan_control *sc) |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1672 | { |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1673 | int file = is_file_lru(lru); |
| 1674 | |
Rik van Riel | b39415b | 2009-12-14 17:59:48 -0800 | [diff] [blame] | 1675 | if (is_active_lru(lru)) { |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1676 | if (inactive_list_is_low(mz, file)) |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1677 | shrink_active_list(nr_to_scan, mz, sc, lru); |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1678 | return 0; |
| 1679 | } |
| 1680 | |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1681 | return shrink_inactive_list(nr_to_scan, mz, sc, lru); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1682 | } |
| 1683 | |
Konstantin Khlebnikov | 1e09e00 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1684 | static int vmscan_swappiness(struct scan_control *sc) |
KAMEZAWA Hiroyuki | 1f4c025 | 2011-07-26 16:08:21 -0700 | [diff] [blame] | 1685 | { |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 1686 | if (global_reclaim(sc)) |
KAMEZAWA Hiroyuki | 1f4c025 | 2011-07-26 16:08:21 -0700 | [diff] [blame] | 1687 | return vm_swappiness; |
Konstantin Khlebnikov | 1e09e00 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1688 | return mem_cgroup_swappiness(sc->target_mem_cgroup); |
KAMEZAWA Hiroyuki | 1f4c025 | 2011-07-26 16:08:21 -0700 | [diff] [blame] | 1689 | } |
| 1690 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1691 | /* |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1692 | * Determine how aggressively the anon and file LRU lists should be |
| 1693 | * scanned. The relative value of each set of LRU lists is determined |
| 1694 | * by looking at the fraction of the pages scanned we did rotate back |
| 1695 | * onto the active list instead of evict. |
| 1696 | * |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1697 | * 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] | 1698 | */ |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1699 | static void get_scan_count(struct mem_cgroup_zone *mz, struct scan_control *sc, |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1700 | unsigned long *nr) |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1701 | { |
| 1702 | unsigned long anon, file, free; |
| 1703 | unsigned long anon_prio, file_prio; |
| 1704 | unsigned long ap, fp; |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1705 | struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(mz); |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1706 | u64 fraction[2], denominator; |
Hugh Dickins | 4111304 | 2012-01-12 17:20:01 -0800 | [diff] [blame] | 1707 | enum lru_list lru; |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1708 | int noswap = 0; |
Johannes Weiner | a4d3e9e | 2011-09-14 16:21:52 -0700 | [diff] [blame] | 1709 | bool force_scan = false; |
KAMEZAWA Hiroyuki | 246e87a | 2011-05-26 16:25:34 -0700 | [diff] [blame] | 1710 | |
Johannes Weiner | f11c0ca | 2011-10-31 17:07:27 -0700 | [diff] [blame] | 1711 | /* |
| 1712 | * If the zone or memcg is small, nr[l] can be 0. This |
| 1713 | * results in no scanning on this priority and a potential |
| 1714 | * priority drop. Global direct reclaim can go to the next |
| 1715 | * zone and tends to have no problems. Global kswapd is for |
| 1716 | * zone balancing and it needs to scan a minimum amount. When |
| 1717 | * reclaiming for a memcg, a priority drop can cause high |
| 1718 | * latencies, so it's better to scan a minimum amount there as |
| 1719 | * well. |
| 1720 | */ |
Lisa Du | e093521 | 2013-09-11 14:22:36 -0700 | [diff] [blame] | 1721 | if (current_is_kswapd() && !zone_reclaimable(mz->zone)) |
Johannes Weiner | a4d3e9e | 2011-09-14 16:21:52 -0700 | [diff] [blame] | 1722 | force_scan = true; |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 1723 | if (!global_reclaim(sc)) |
Johannes Weiner | a4d3e9e | 2011-09-14 16:21:52 -0700 | [diff] [blame] | 1724 | force_scan = true; |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1725 | |
| 1726 | /* If we have no swap space, do not bother scanning anon pages. */ |
Shaohua Li | 1abbca7 | 2013-02-22 16:34:38 -0800 | [diff] [blame] | 1727 | if (!sc->may_swap || (get_nr_swap_pages() <= 0)) { |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1728 | noswap = 1; |
| 1729 | fraction[0] = 0; |
| 1730 | fraction[1] = 1; |
| 1731 | denominator = 1; |
| 1732 | goto out; |
| 1733 | } |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1734 | |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1735 | anon = zone_nr_lru_pages(mz, LRU_ACTIVE_ANON) + |
| 1736 | zone_nr_lru_pages(mz, LRU_INACTIVE_ANON); |
| 1737 | file = zone_nr_lru_pages(mz, LRU_ACTIVE_FILE) + |
| 1738 | zone_nr_lru_pages(mz, LRU_INACTIVE_FILE); |
Johannes Weiner | a4d3e9e | 2011-09-14 16:21:52 -0700 | [diff] [blame] | 1739 | |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 1740 | if (global_reclaim(sc)) { |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1741 | free = zone_page_state(mz->zone, NR_FREE_PAGES); |
KOSAKI Motohiro | eeee9a8 | 2009-01-07 18:08:17 -0800 | [diff] [blame] | 1742 | /* If we have very few page cache pages, |
| 1743 | force-scan anon pages. */ |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1744 | if (unlikely(file + free <= high_wmark_pages(mz->zone))) { |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1745 | fraction[0] = 1; |
| 1746 | fraction[1] = 0; |
| 1747 | denominator = 1; |
| 1748 | goto out; |
KOSAKI Motohiro | eeee9a8 | 2009-01-07 18:08:17 -0800 | [diff] [blame] | 1749 | } |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1750 | } |
| 1751 | |
| 1752 | /* |
KOSAKI Motohiro | 58c37f6 | 2010-08-09 17:19:51 -0700 | [diff] [blame] | 1753 | * With swappiness at 100, anonymous and file have the same priority. |
| 1754 | * This scanning priority is essentially the inverse of IO cost. |
| 1755 | */ |
Konstantin Khlebnikov | 1e09e00 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1756 | anon_prio = vmscan_swappiness(sc); |
| 1757 | file_prio = 200 - vmscan_swappiness(sc); |
KOSAKI Motohiro | 58c37f6 | 2010-08-09 17:19:51 -0700 | [diff] [blame] | 1758 | |
| 1759 | /* |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1760 | * OK, so we have swap space and a fair amount of page cache |
| 1761 | * pages. We use the recently rotated / recently scanned |
| 1762 | * ratios to determine how valuable each cache is. |
| 1763 | * |
| 1764 | * Because workloads change over time (and to avoid overflow) |
| 1765 | * we keep these statistics as a floating average, which ends |
| 1766 | * up weighing recent references more than old ones. |
| 1767 | * |
| 1768 | * anon in [0], file in [1] |
| 1769 | */ |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1770 | spin_lock_irq(&mz->zone->lru_lock); |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1771 | if (unlikely(reclaim_stat->recent_scanned[0] > anon / 4)) { |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1772 | reclaim_stat->recent_scanned[0] /= 2; |
| 1773 | reclaim_stat->recent_rotated[0] /= 2; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1774 | } |
| 1775 | |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1776 | if (unlikely(reclaim_stat->recent_scanned[1] > file / 4)) { |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1777 | reclaim_stat->recent_scanned[1] /= 2; |
| 1778 | reclaim_stat->recent_rotated[1] /= 2; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1779 | } |
| 1780 | |
| 1781 | /* |
Rik van Riel | 00d8089 | 2008-11-19 15:36:44 -0800 | [diff] [blame] | 1782 | * The amount of pressure on anon vs file pages is inversely |
| 1783 | * proportional to the fraction of recently scanned pages on |
| 1784 | * each list that were recently referenced and in active use. |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1785 | */ |
Satoru Moriya | 1cc0039 | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 1786 | ap = anon_prio * (reclaim_stat->recent_scanned[0] + 1); |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1787 | ap /= reclaim_stat->recent_rotated[0] + 1; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1788 | |
Satoru Moriya | 1cc0039 | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 1789 | fp = file_prio * (reclaim_stat->recent_scanned[1] + 1); |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1790 | fp /= reclaim_stat->recent_rotated[1] + 1; |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1791 | spin_unlock_irq(&mz->zone->lru_lock); |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1792 | |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1793 | fraction[0] = ap; |
| 1794 | fraction[1] = fp; |
| 1795 | denominator = ap + fp + 1; |
| 1796 | out: |
Hugh Dickins | 4111304 | 2012-01-12 17:20:01 -0800 | [diff] [blame] | 1797 | for_each_evictable_lru(lru) { |
| 1798 | int file = is_file_lru(lru); |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1799 | unsigned long scan; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1800 | |
Hugh Dickins | 4111304 | 2012-01-12 17:20:01 -0800 | [diff] [blame] | 1801 | scan = zone_nr_lru_pages(mz, lru); |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1802 | if (sc->priority || noswap || !vmscan_swappiness(sc)) { |
| 1803 | scan >>= sc->priority; |
Johannes Weiner | f11c0ca | 2011-10-31 17:07:27 -0700 | [diff] [blame] | 1804 | if (!scan && force_scan) |
| 1805 | scan = SWAP_CLUSTER_MAX; |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1806 | scan = div64_u64(scan * fraction[file], denominator); |
| 1807 | } |
Hugh Dickins | 4111304 | 2012-01-12 17:20:01 -0800 | [diff] [blame] | 1808 | nr[lru] = scan; |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1809 | } |
Wu Fengguang | 6e08a36 | 2009-06-16 15:32:29 -0700 | [diff] [blame] | 1810 | } |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1811 | |
Mel Gorman | 626ebe7 | 2012-05-29 15:06:20 -0700 | [diff] [blame] | 1812 | /* Use reclaim/compaction for costly allocs or under memory pressure */ |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1813 | static bool in_reclaim_compaction(struct scan_control *sc) |
Mel Gorman | 626ebe7 | 2012-05-29 15:06:20 -0700 | [diff] [blame] | 1814 | { |
| 1815 | if (COMPACTION_BUILD && sc->order && |
| 1816 | (sc->order > PAGE_ALLOC_COSTLY_ORDER || |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1817 | sc->priority < DEF_PRIORITY - 2)) |
Mel Gorman | 626ebe7 | 2012-05-29 15:06:20 -0700 | [diff] [blame] | 1818 | return true; |
| 1819 | |
| 1820 | return false; |
| 1821 | } |
| 1822 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1823 | /* |
Mel Gorman | 626ebe7 | 2012-05-29 15:06:20 -0700 | [diff] [blame] | 1824 | * Reclaim/compaction is used for high-order allocation requests. It reclaims |
| 1825 | * order-0 pages before compacting the zone. should_continue_reclaim() returns |
| 1826 | * true if more pages should be reclaimed such that when the page allocator |
| 1827 | * calls try_to_compact_zone() that it will have enough free pages to succeed. |
| 1828 | * It will give up earlier than that if there is difficulty reclaiming pages. |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1829 | */ |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1830 | static inline bool should_continue_reclaim(struct mem_cgroup_zone *mz, |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1831 | unsigned long nr_reclaimed, |
| 1832 | unsigned long nr_scanned, |
| 1833 | struct scan_control *sc) |
| 1834 | { |
| 1835 | unsigned long pages_for_compaction; |
| 1836 | unsigned long inactive_lru_pages; |
| 1837 | |
| 1838 | /* If not in reclaim/compaction mode, stop */ |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1839 | if (!in_reclaim_compaction(sc)) |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1840 | return false; |
| 1841 | |
Mel Gorman | 2876592 | 2011-02-25 14:44:20 -0800 | [diff] [blame] | 1842 | /* Consider stopping depending on scan and reclaim activity */ |
| 1843 | if (sc->gfp_mask & __GFP_REPEAT) { |
| 1844 | /* |
| 1845 | * For __GFP_REPEAT allocations, stop reclaiming if the |
| 1846 | * full LRU list has been scanned and we are still failing |
| 1847 | * to reclaim pages. This full LRU scan is potentially |
| 1848 | * expensive but a __GFP_REPEAT caller really wants to succeed |
| 1849 | */ |
| 1850 | if (!nr_reclaimed && !nr_scanned) |
| 1851 | return false; |
| 1852 | } else { |
| 1853 | /* |
| 1854 | * For non-__GFP_REPEAT allocations which can presumably |
| 1855 | * fail without consequence, stop if we failed to reclaim |
| 1856 | * any pages from the last SWAP_CLUSTER_MAX number of |
| 1857 | * pages that were scanned. This will return to the |
| 1858 | * caller faster at the risk reclaim/compaction and |
| 1859 | * the resulting allocation attempt fails |
| 1860 | */ |
| 1861 | if (!nr_reclaimed) |
| 1862 | return false; |
| 1863 | } |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1864 | |
| 1865 | /* |
| 1866 | * If we have not reclaimed enough pages for compaction and the |
| 1867 | * inactive lists are large enough, continue reclaiming |
| 1868 | */ |
| 1869 | pages_for_compaction = (2UL << sc->order); |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1870 | inactive_lru_pages = zone_nr_lru_pages(mz, LRU_INACTIVE_FILE); |
Shaohua Li | 1abbca7 | 2013-02-22 16:34:38 -0800 | [diff] [blame] | 1871 | if (get_nr_swap_pages() > 0) |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1872 | inactive_lru_pages += zone_nr_lru_pages(mz, LRU_INACTIVE_ANON); |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1873 | if (sc->nr_reclaimed < pages_for_compaction && |
| 1874 | inactive_lru_pages > pages_for_compaction) |
| 1875 | return true; |
| 1876 | |
| 1877 | /* If compaction would go ahead or the allocation would succeed, stop */ |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1878 | switch (compaction_suitable(mz->zone, sc->order)) { |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1879 | case COMPACT_PARTIAL: |
| 1880 | case COMPACT_CONTINUE: |
| 1881 | return false; |
| 1882 | default: |
| 1883 | return true; |
| 1884 | } |
| 1885 | } |
| 1886 | |
| 1887 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1888 | * This is a basic per-zone page freer. Used by both kswapd and direct reclaim. |
| 1889 | */ |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1890 | static void shrink_mem_cgroup_zone(struct mem_cgroup_zone *mz, |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1891 | struct scan_control *sc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1892 | { |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1893 | unsigned long nr[NR_LRU_LISTS]; |
Christoph Lameter | 8695949 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 1894 | unsigned long nr_to_scan; |
Hugh Dickins | 4111304 | 2012-01-12 17:20:01 -0800 | [diff] [blame] | 1895 | enum lru_list lru; |
Johannes Weiner | f0fdc5e | 2011-02-10 15:01:34 -0800 | [diff] [blame] | 1896 | unsigned long nr_reclaimed, nr_scanned; |
KOSAKI Motohiro | 22fba33 | 2009-12-14 17:59:10 -0800 | [diff] [blame] | 1897 | unsigned long nr_to_reclaim = sc->nr_to_reclaim; |
Shaohua Li | 3da367c | 2011-10-31 17:07:03 -0700 | [diff] [blame] | 1898 | struct blk_plug plug; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1899 | |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1900 | restart: |
| 1901 | nr_reclaimed = 0; |
Johannes Weiner | f0fdc5e | 2011-02-10 15:01:34 -0800 | [diff] [blame] | 1902 | nr_scanned = sc->nr_scanned; |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1903 | get_scan_count(mz, sc, nr); |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1904 | |
Shaohua Li | 3da367c | 2011-10-31 17:07:03 -0700 | [diff] [blame] | 1905 | blk_start_plug(&plug); |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1906 | while (nr[LRU_INACTIVE_ANON] || nr[LRU_ACTIVE_FILE] || |
| 1907 | nr[LRU_INACTIVE_FILE]) { |
Hugh Dickins | 4111304 | 2012-01-12 17:20:01 -0800 | [diff] [blame] | 1908 | for_each_evictable_lru(lru) { |
| 1909 | if (nr[lru]) { |
KOSAKI Motohiro | ece74b2 | 2009-12-14 17:59:14 -0800 | [diff] [blame] | 1910 | nr_to_scan = min_t(unsigned long, |
Hugh Dickins | 4111304 | 2012-01-12 17:20:01 -0800 | [diff] [blame] | 1911 | nr[lru], SWAP_CLUSTER_MAX); |
| 1912 | nr[lru] -= nr_to_scan; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1913 | |
Hugh Dickins | 4111304 | 2012-01-12 17:20:01 -0800 | [diff] [blame] | 1914 | nr_reclaimed += shrink_list(lru, nr_to_scan, |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1915 | mz, sc); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1916 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1917 | } |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1918 | /* |
| 1919 | * On large memory systems, scan >> priority can become |
| 1920 | * really large. This is fine for the starting priority; |
| 1921 | * we want to put equal scanning pressure on each zone. |
| 1922 | * However, if the VM has a harder time of freeing pages, |
| 1923 | * with multiple processes reclaiming pages, the total |
| 1924 | * freeing target can get unreasonably large. |
| 1925 | */ |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1926 | if (nr_reclaimed >= nr_to_reclaim && |
| 1927 | sc->priority < DEF_PRIORITY) |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 1928 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1929 | } |
Shaohua Li | 3da367c | 2011-10-31 17:07:03 -0700 | [diff] [blame] | 1930 | blk_finish_plug(&plug); |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1931 | sc->nr_reclaimed += nr_reclaimed; |
KOSAKI Motohiro | 01dbe5c | 2009-01-06 14:40:02 -0800 | [diff] [blame] | 1932 | |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1933 | /* |
| 1934 | * Even if we did not try to evict anon pages at all, we want to |
| 1935 | * rebalance the anon lru active/inactive ratio. |
| 1936 | */ |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1937 | if (inactive_anon_is_low(mz)) |
Konstantin Khlebnikov | 7f8a920 | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 1938 | shrink_active_list(SWAP_CLUSTER_MAX, mz, |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1939 | sc, LRU_ACTIVE_ANON); |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1940 | |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1941 | /* reclaim/compaction might need reclaim to continue */ |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1942 | if (should_continue_reclaim(mz, nr_reclaimed, |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1943 | sc->nr_scanned - nr_scanned, sc)) |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1944 | goto restart; |
| 1945 | |
Andrew Morton | 232ea4d | 2007-02-28 20:13:21 -0800 | [diff] [blame] | 1946 | throttle_vm_writeout(sc->gfp_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1947 | } |
| 1948 | |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1949 | static void shrink_zone(struct zone *zone, struct scan_control *sc) |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1950 | { |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1951 | struct mem_cgroup *root = sc->target_mem_cgroup; |
| 1952 | struct mem_cgroup_reclaim_cookie reclaim = { |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1953 | .zone = zone, |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1954 | .priority = sc->priority, |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1955 | }; |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1956 | struct mem_cgroup *memcg; |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1957 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1958 | memcg = mem_cgroup_iter(root, NULL, &reclaim); |
| 1959 | do { |
| 1960 | struct mem_cgroup_zone mz = { |
| 1961 | .mem_cgroup = memcg, |
| 1962 | .zone = zone, |
| 1963 | }; |
| 1964 | |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1965 | shrink_mem_cgroup_zone(&mz, sc); |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1966 | /* |
| 1967 | * Limit reclaim has historically picked one memcg and |
| 1968 | * scanned it with decreasing priority levels until |
| 1969 | * nr_to_reclaim had been reclaimed. This priority |
| 1970 | * cycle is thus over after a single memcg. |
Johannes Weiner | b95a2f2 | 2012-01-12 17:18:06 -0800 | [diff] [blame] | 1971 | * |
| 1972 | * Direct reclaim and kswapd, on the other hand, have |
| 1973 | * to scan all memory cgroups to fulfill the overall |
| 1974 | * scan target for the zone. |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1975 | */ |
| 1976 | if (!global_reclaim(sc)) { |
| 1977 | mem_cgroup_iter_break(root, memcg); |
| 1978 | break; |
| 1979 | } |
| 1980 | memcg = mem_cgroup_iter(root, memcg, &reclaim); |
| 1981 | } while (memcg); |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1982 | } |
| 1983 | |
Mel Gorman | fe4b1b2 | 2012-01-12 17:19:45 -0800 | [diff] [blame] | 1984 | /* Returns true if compaction should go ahead for a high-order request */ |
| 1985 | static inline bool compaction_ready(struct zone *zone, struct scan_control *sc) |
| 1986 | { |
| 1987 | unsigned long balance_gap, watermark; |
| 1988 | bool watermark_ok; |
| 1989 | |
| 1990 | /* Do not consider compaction for orders reclaim is meant to satisfy */ |
| 1991 | if (sc->order <= PAGE_ALLOC_COSTLY_ORDER) |
| 1992 | return false; |
| 1993 | |
| 1994 | /* |
| 1995 | * Compaction takes time to run and there are potentially other |
| 1996 | * callers using the pages just freed. Continue reclaiming until |
| 1997 | * there is a buffer of free pages available to give compaction |
| 1998 | * a reasonable chance of completing and allocating the page |
| 1999 | */ |
| 2000 | balance_gap = min(low_wmark_pages(zone), |
| 2001 | (zone->present_pages + KSWAPD_ZONE_BALANCE_GAP_RATIO-1) / |
| 2002 | KSWAPD_ZONE_BALANCE_GAP_RATIO); |
| 2003 | watermark = high_wmark_pages(zone) + balance_gap + (2UL << sc->order); |
| 2004 | watermark_ok = zone_watermark_ok_safe(zone, 0, watermark, 0, 0); |
| 2005 | |
| 2006 | /* |
| 2007 | * If compaction is deferred, reclaim up to a point where |
| 2008 | * compaction will have a chance of success when re-enabled |
| 2009 | */ |
Rik van Riel | aff6224 | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 2010 | if (compaction_deferred(zone, sc->order)) |
Mel Gorman | fe4b1b2 | 2012-01-12 17:19:45 -0800 | [diff] [blame] | 2011 | return watermark_ok; |
| 2012 | |
| 2013 | /* If compaction is not ready to start, keep reclaiming */ |
| 2014 | if (!compaction_suitable(zone, sc->order)) |
| 2015 | return false; |
| 2016 | |
| 2017 | return watermark_ok; |
| 2018 | } |
| 2019 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2020 | /* |
| 2021 | * This is the direct reclaim path, for page-allocating processes. We only |
| 2022 | * try to reclaim pages from zones which will satisfy the caller's allocation |
| 2023 | * request. |
| 2024 | * |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 2025 | * We reclaim from a zone even if that zone is over high_wmark_pages(zone). |
| 2026 | * Because: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2027 | * a) The caller may be trying to free *extra* pages to satisfy a higher-order |
| 2028 | * allocation or |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 2029 | * b) The target zone may be at high_wmark_pages(zone) but the lower zones |
| 2030 | * must go *over* high_wmark_pages(zone) to satisfy the `incremental min' |
| 2031 | * zone defense algorithm. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2032 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2033 | * If a zone is deemed to be full of pinned pages then just give it a light |
| 2034 | * scan then give up on it. |
Mel Gorman | e0c2327 | 2011-10-31 17:09:33 -0700 | [diff] [blame] | 2035 | * |
| 2036 | * This function returns true if a zone is being reclaimed for a costly |
Mel Gorman | fe4b1b2 | 2012-01-12 17:19:45 -0800 | [diff] [blame] | 2037 | * high-order allocation and compaction is ready to begin. This indicates to |
Mel Gorman | 0cee34f | 2012-01-12 17:19:49 -0800 | [diff] [blame] | 2038 | * the caller that it should consider retrying the allocation instead of |
| 2039 | * further reclaim. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2040 | */ |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2041 | static bool shrink_zones(struct zonelist *zonelist, struct scan_control *sc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2042 | { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 2043 | struct zoneref *z; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2044 | struct zone *zone; |
Ying Han | d149e3b | 2011-05-26 16:25:27 -0700 | [diff] [blame] | 2045 | unsigned long nr_soft_reclaimed; |
| 2046 | unsigned long nr_soft_scanned; |
Mel Gorman | 0cee34f | 2012-01-12 17:19:49 -0800 | [diff] [blame] | 2047 | bool aborted_reclaim = false; |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 2048 | |
Mel Gorman | cc715d9 | 2012-03-21 16:34:00 -0700 | [diff] [blame] | 2049 | /* |
| 2050 | * If the number of buffer_heads in the machine exceeds the maximum |
| 2051 | * allowed level, force direct reclaim to scan the highmem zone as |
| 2052 | * highmem pages could be pinning lowmem pages storing buffer_heads |
| 2053 | */ |
| 2054 | if (buffer_heads_over_limit) |
| 2055 | sc->gfp_mask |= __GFP_HIGHMEM; |
| 2056 | |
Mel Gorman | d4debc6 | 2010-08-09 17:19:29 -0700 | [diff] [blame] | 2057 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
| 2058 | gfp_zone(sc->gfp_mask), sc->nodemask) { |
Con Kolivas | f3fe651 | 2006-01-06 00:11:15 -0800 | [diff] [blame] | 2059 | if (!populated_zone(zone)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2060 | continue; |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 2061 | /* |
| 2062 | * Take care memory controller reclaiming has small influence |
| 2063 | * to global LRU. |
| 2064 | */ |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 2065 | if (global_reclaim(sc)) { |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 2066 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
| 2067 | continue; |
Lisa Du | e093521 | 2013-09-11 14:22:36 -0700 | [diff] [blame] | 2068 | if (sc->priority != DEF_PRIORITY && |
| 2069 | !zone_reclaimable(zone)) |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 2070 | continue; /* Let kswapd poll it */ |
Rik van Riel | e0887c1 | 2011-10-31 17:09:31 -0700 | [diff] [blame] | 2071 | if (COMPACTION_BUILD) { |
| 2072 | /* |
Mel Gorman | e0c2327 | 2011-10-31 17:09:33 -0700 | [diff] [blame] | 2073 | * If we already have plenty of memory free for |
| 2074 | * compaction in this zone, don't free any more. |
| 2075 | * Even though compaction is invoked for any |
| 2076 | * non-zero order, only frequent costly order |
| 2077 | * reclamation is disruptive enough to become a |
Copot Alexandru | c7cfa37 | 2012-03-21 16:34:10 -0700 | [diff] [blame] | 2078 | * noticeable problem, like transparent huge |
| 2079 | * page allocations. |
Rik van Riel | e0887c1 | 2011-10-31 17:09:31 -0700 | [diff] [blame] | 2080 | */ |
Mel Gorman | fe4b1b2 | 2012-01-12 17:19:45 -0800 | [diff] [blame] | 2081 | if (compaction_ready(zone, sc)) { |
Mel Gorman | 0cee34f | 2012-01-12 17:19:49 -0800 | [diff] [blame] | 2082 | aborted_reclaim = true; |
Rik van Riel | e0887c1 | 2011-10-31 17:09:31 -0700 | [diff] [blame] | 2083 | continue; |
Mel Gorman | e0c2327 | 2011-10-31 17:09:33 -0700 | [diff] [blame] | 2084 | } |
Rik van Riel | e0887c1 | 2011-10-31 17:09:31 -0700 | [diff] [blame] | 2085 | } |
KAMEZAWA Hiroyuki | ac34a1a | 2011-06-27 16:18:12 -0700 | [diff] [blame] | 2086 | /* |
| 2087 | * This steals pages from memory cgroups over softlimit |
| 2088 | * and returns the number of reclaimed pages and |
| 2089 | * scanned pages. This works for global memory pressure |
| 2090 | * and balancing, not for a memcg's limit. |
| 2091 | */ |
| 2092 | nr_soft_scanned = 0; |
| 2093 | nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone, |
| 2094 | sc->order, sc->gfp_mask, |
| 2095 | &nr_soft_scanned); |
| 2096 | sc->nr_reclaimed += nr_soft_reclaimed; |
| 2097 | sc->nr_scanned += nr_soft_scanned; |
| 2098 | /* need some check for avoid more shrink_zone() */ |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 2099 | } |
Nick Piggin | 408d854 | 2006-09-25 23:31:27 -0700 | [diff] [blame] | 2100 | |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2101 | shrink_zone(zone, sc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2102 | } |
Mel Gorman | e0c2327 | 2011-10-31 17:09:33 -0700 | [diff] [blame] | 2103 | |
Mel Gorman | 0cee34f | 2012-01-12 17:19:49 -0800 | [diff] [blame] | 2104 | return aborted_reclaim; |
Minchan Kim | d190836 | 2010-09-22 13:05:01 -0700 | [diff] [blame] | 2105 | } |
| 2106 | |
KOSAKI Motohiro | 929bea7 | 2011-04-14 15:22:12 -0700 | [diff] [blame] | 2107 | /* All zones in zonelist are unreclaimable? */ |
Minchan Kim | d190836 | 2010-09-22 13:05:01 -0700 | [diff] [blame] | 2108 | static bool all_unreclaimable(struct zonelist *zonelist, |
| 2109 | struct scan_control *sc) |
| 2110 | { |
| 2111 | struct zoneref *z; |
| 2112 | struct zone *zone; |
Minchan Kim | d190836 | 2010-09-22 13:05:01 -0700 | [diff] [blame] | 2113 | |
| 2114 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
| 2115 | gfp_zone(sc->gfp_mask), sc->nodemask) { |
| 2116 | if (!populated_zone(zone)) |
| 2117 | continue; |
| 2118 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
| 2119 | continue; |
Lisa Du | e093521 | 2013-09-11 14:22:36 -0700 | [diff] [blame] | 2120 | if (zone_reclaimable(zone)) |
KOSAKI Motohiro | 929bea7 | 2011-04-14 15:22:12 -0700 | [diff] [blame] | 2121 | return false; |
Minchan Kim | d190836 | 2010-09-22 13:05:01 -0700 | [diff] [blame] | 2122 | } |
| 2123 | |
KOSAKI Motohiro | 929bea7 | 2011-04-14 15:22:12 -0700 | [diff] [blame] | 2124 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2125 | } |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2126 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2127 | /* |
| 2128 | * This is the main entry point to direct page reclaim. |
| 2129 | * |
| 2130 | * If a full scan of the inactive list fails to free enough memory then we |
| 2131 | * are "out of memory" and something needs to be killed. |
| 2132 | * |
| 2133 | * If the caller is !__GFP_FS then the probability of a failure is reasonably |
| 2134 | * 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] | 2135 | * caller can't do much about. We kick the writeback threads and take explicit |
| 2136 | * naps in the hope that some of these pages can be written. But if the |
| 2137 | * allocating task holds filesystem locks which prevent writeout this might not |
| 2138 | * work, and the allocation attempt will fail. |
Nishanth Aravamudan | a41f24e | 2008-04-29 00:58:25 -0700 | [diff] [blame] | 2139 | * |
| 2140 | * returns: 0, if no pages reclaimed |
| 2141 | * else, the number of pages reclaimed |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2142 | */ |
Mel Gorman | dac1d27 | 2008-04-28 02:12:12 -0700 | [diff] [blame] | 2143 | static unsigned long do_try_to_free_pages(struct zonelist *zonelist, |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 2144 | struct scan_control *sc, |
| 2145 | struct shrink_control *shrink) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2146 | { |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 2147 | unsigned long total_scanned = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2148 | struct reclaim_state *reclaim_state = current->reclaim_state; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 2149 | struct zoneref *z; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2150 | struct zone *zone; |
KOSAKI Motohiro | 22fba33 | 2009-12-14 17:59:10 -0800 | [diff] [blame] | 2151 | unsigned long writeback_threshold; |
Mel Gorman | 0cee34f | 2012-01-12 17:19:49 -0800 | [diff] [blame] | 2152 | bool aborted_reclaim; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2153 | |
Keika Kobayashi | 873b477 | 2008-07-25 01:48:52 -0700 | [diff] [blame] | 2154 | delayacct_freepages_start(); |
| 2155 | |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 2156 | if (global_reclaim(sc)) |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 2157 | count_vm_event(ALLOCSTALL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2158 | |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2159 | do { |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2160 | sc->nr_scanned = 0; |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2161 | aborted_reclaim = shrink_zones(zonelist, sc); |
Mel Gorman | e0c2327 | 2011-10-31 17:09:33 -0700 | [diff] [blame] | 2162 | |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2163 | /* |
| 2164 | * Don't shrink slabs when reclaiming memory from |
| 2165 | * over limit cgroups |
| 2166 | */ |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 2167 | if (global_reclaim(sc)) { |
KOSAKI Motohiro | c6a8a8c | 2010-08-09 17:19:14 -0700 | [diff] [blame] | 2168 | unsigned long lru_pages = 0; |
Mel Gorman | d4debc6 | 2010-08-09 17:19:29 -0700 | [diff] [blame] | 2169 | for_each_zone_zonelist(zone, z, zonelist, |
| 2170 | gfp_zone(sc->gfp_mask)) { |
KOSAKI Motohiro | c6a8a8c | 2010-08-09 17:19:14 -0700 | [diff] [blame] | 2171 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
| 2172 | continue; |
| 2173 | |
| 2174 | lru_pages += zone_reclaimable_pages(zone); |
| 2175 | } |
| 2176 | |
Ying Han | 1495f23 | 2011-05-24 17:12:27 -0700 | [diff] [blame] | 2177 | shrink_slab(shrink, sc->nr_scanned, lru_pages); |
KAMEZAWA Hiroyuki | 91a4547 | 2008-02-07 00:14:29 -0800 | [diff] [blame] | 2178 | if (reclaim_state) { |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2179 | sc->nr_reclaimed += reclaim_state->reclaimed_slab; |
KAMEZAWA Hiroyuki | 91a4547 | 2008-02-07 00:14:29 -0800 | [diff] [blame] | 2180 | reclaim_state->reclaimed_slab = 0; |
| 2181 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2182 | } |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2183 | total_scanned += sc->nr_scanned; |
KOSAKI Motohiro | bb21c7c | 2010-06-04 14:15:05 -0700 | [diff] [blame] | 2184 | if (sc->nr_reclaimed >= sc->nr_to_reclaim) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2185 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2186 | |
| 2187 | /* |
| 2188 | * Try to write back as many pages as we just scanned. This |
| 2189 | * tends to cause slow streaming writers to write data to the |
| 2190 | * disk smoothly, at the dirtying rate, which is nice. But |
| 2191 | * that's undesirable in laptop mode, where we *want* lumpy |
| 2192 | * writeout. So in laptop mode, write out the whole world. |
| 2193 | */ |
KOSAKI Motohiro | 22fba33 | 2009-12-14 17:59:10 -0800 | [diff] [blame] | 2194 | writeback_threshold = sc->nr_to_reclaim + sc->nr_to_reclaim / 2; |
| 2195 | if (total_scanned > writeback_threshold) { |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 2196 | wakeup_flusher_threads(laptop_mode ? 0 : total_scanned, |
| 2197 | WB_REASON_TRY_TO_FREE_PAGES); |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2198 | sc->may_writepage = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2199 | } |
| 2200 | |
| 2201 | /* Take a nap, wait for some writeback to complete */ |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 2202 | if (!sc->hibernation_mode && sc->nr_scanned && |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2203 | sc->priority < DEF_PRIORITY - 2) { |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 2204 | struct zone *preferred_zone; |
| 2205 | |
| 2206 | first_zones_zonelist(zonelist, gfp_zone(sc->gfp_mask), |
David Rientjes | f33261d | 2011-01-25 15:07:20 -0800 | [diff] [blame] | 2207 | &cpuset_current_mems_allowed, |
| 2208 | &preferred_zone); |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 2209 | wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/10); |
| 2210 | } |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2211 | } while (--sc->priority >= 0); |
KOSAKI Motohiro | bb21c7c | 2010-06-04 14:15:05 -0700 | [diff] [blame] | 2212 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2213 | out: |
Keika Kobayashi | 873b477 | 2008-07-25 01:48:52 -0700 | [diff] [blame] | 2214 | delayacct_freepages_end(); |
| 2215 | |
KOSAKI Motohiro | bb21c7c | 2010-06-04 14:15:05 -0700 | [diff] [blame] | 2216 | if (sc->nr_reclaimed) |
| 2217 | return sc->nr_reclaimed; |
| 2218 | |
KOSAKI Motohiro | 929bea7 | 2011-04-14 15:22:12 -0700 | [diff] [blame] | 2219 | /* |
| 2220 | * As hibernation is going on, kswapd is freezed so that it can't mark |
| 2221 | * the zone into all_unreclaimable. Thus bypassing all_unreclaimable |
| 2222 | * check. |
| 2223 | */ |
| 2224 | if (oom_killer_disabled) |
| 2225 | return 0; |
| 2226 | |
Mel Gorman | 0cee34f | 2012-01-12 17:19:49 -0800 | [diff] [blame] | 2227 | /* Aborted reclaim to try compaction? don't OOM, then */ |
| 2228 | if (aborted_reclaim) |
Mel Gorman | 7335084 | 2012-01-12 17:19:33 -0800 | [diff] [blame] | 2229 | return 1; |
| 2230 | |
KOSAKI Motohiro | bb21c7c | 2010-06-04 14:15:05 -0700 | [diff] [blame] | 2231 | /* top priority shrink_zones still had more to do? don't OOM, then */ |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 2232 | if (global_reclaim(sc) && !all_unreclaimable(zonelist, sc)) |
KOSAKI Motohiro | bb21c7c | 2010-06-04 14:15:05 -0700 | [diff] [blame] | 2233 | return 1; |
| 2234 | |
| 2235 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2236 | } |
| 2237 | |
Mel Gorman | dac1d27 | 2008-04-28 02:12:12 -0700 | [diff] [blame] | 2238 | unsigned long try_to_free_pages(struct zonelist *zonelist, int order, |
KAMEZAWA Hiroyuki | 327c0e9 | 2009-03-31 15:23:31 -0700 | [diff] [blame] | 2239 | gfp_t gfp_mask, nodemask_t *nodemask) |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2240 | { |
Mel Gorman | 33906bc | 2010-08-09 17:19:16 -0700 | [diff] [blame] | 2241 | unsigned long nr_reclaimed; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2242 | struct scan_control sc = { |
| 2243 | .gfp_mask = gfp_mask, |
| 2244 | .may_writepage = !laptop_mode, |
KOSAKI Motohiro | 22fba33 | 2009-12-14 17:59:10 -0800 | [diff] [blame] | 2245 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
Johannes Weiner | a6dc60f | 2009-03-31 15:19:30 -0700 | [diff] [blame] | 2246 | .may_unmap = 1, |
KOSAKI Motohiro | 2e2e425 | 2009-04-21 12:24:57 -0700 | [diff] [blame] | 2247 | .may_swap = 1, |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2248 | .order = order, |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2249 | .priority = DEF_PRIORITY, |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 2250 | .target_mem_cgroup = NULL, |
KAMEZAWA Hiroyuki | 327c0e9 | 2009-03-31 15:23:31 -0700 | [diff] [blame] | 2251 | .nodemask = nodemask, |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2252 | }; |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 2253 | struct shrink_control shrink = { |
| 2254 | .gfp_mask = sc.gfp_mask, |
| 2255 | }; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2256 | |
Mel Gorman | 33906bc | 2010-08-09 17:19:16 -0700 | [diff] [blame] | 2257 | trace_mm_vmscan_direct_reclaim_begin(order, |
| 2258 | sc.may_writepage, |
| 2259 | gfp_mask); |
| 2260 | |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 2261 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink); |
Mel Gorman | 33906bc | 2010-08-09 17:19:16 -0700 | [diff] [blame] | 2262 | |
| 2263 | trace_mm_vmscan_direct_reclaim_end(nr_reclaimed); |
| 2264 | |
| 2265 | return nr_reclaimed; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2266 | } |
| 2267 | |
Balbir Singh | 00f0b82 | 2008-03-04 14:28:39 -0800 | [diff] [blame] | 2268 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2269 | |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 2270 | unsigned long mem_cgroup_shrink_node_zone(struct mem_cgroup *memcg, |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2271 | gfp_t gfp_mask, bool noswap, |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 2272 | struct zone *zone, |
| 2273 | unsigned long *nr_scanned) |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2274 | { |
| 2275 | struct scan_control sc = { |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 2276 | .nr_scanned = 0, |
KOSAKI Motohiro | b8f5c56 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2277 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2278 | .may_writepage = !laptop_mode, |
| 2279 | .may_unmap = 1, |
| 2280 | .may_swap = !noswap, |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2281 | .order = 0, |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2282 | .priority = 0, |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 2283 | .target_mem_cgroup = memcg, |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2284 | }; |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 2285 | struct mem_cgroup_zone mz = { |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 2286 | .mem_cgroup = memcg, |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 2287 | .zone = zone, |
| 2288 | }; |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 2289 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2290 | sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | |
| 2291 | (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK); |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 2292 | |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2293 | trace_mm_vmscan_memcg_softlimit_reclaim_begin(sc.order, |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 2294 | sc.may_writepage, |
| 2295 | sc.gfp_mask); |
| 2296 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2297 | /* |
| 2298 | * NOTE: Although we can get the priority field, using it |
| 2299 | * here is not a good idea, since it limits the pages we can scan. |
| 2300 | * if we don't reclaim here, the shrink_zone from balance_pgdat |
| 2301 | * will pick up pages from other mem cgroup's as well. We hack |
| 2302 | * the priority and make it zero. |
| 2303 | */ |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2304 | shrink_mem_cgroup_zone(&mz, &sc); |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 2305 | |
| 2306 | trace_mm_vmscan_memcg_softlimit_reclaim_end(sc.nr_reclaimed); |
| 2307 | |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 2308 | *nr_scanned = sc.nr_scanned; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2309 | return sc.nr_reclaimed; |
| 2310 | } |
| 2311 | |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 2312 | unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *memcg, |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 2313 | gfp_t gfp_mask, |
Johannes Weiner | 185efc0 | 2011-09-14 16:21:58 -0700 | [diff] [blame] | 2314 | bool noswap) |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2315 | { |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2316 | struct zonelist *zonelist; |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 2317 | unsigned long nr_reclaimed; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 2318 | int nid; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2319 | struct scan_control sc = { |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2320 | .may_writepage = !laptop_mode, |
Johannes Weiner | a6dc60f | 2009-03-31 15:19:30 -0700 | [diff] [blame] | 2321 | .may_unmap = 1, |
KOSAKI Motohiro | 2e2e425 | 2009-04-21 12:24:57 -0700 | [diff] [blame] | 2322 | .may_swap = !noswap, |
KOSAKI Motohiro | 22fba33 | 2009-12-14 17:59:10 -0800 | [diff] [blame] | 2323 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2324 | .order = 0, |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2325 | .priority = DEF_PRIORITY, |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 2326 | .target_mem_cgroup = memcg, |
KAMEZAWA Hiroyuki | 327c0e9 | 2009-03-31 15:23:31 -0700 | [diff] [blame] | 2327 | .nodemask = NULL, /* we don't care the placement */ |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 2328 | .gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | |
| 2329 | (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK), |
| 2330 | }; |
| 2331 | struct shrink_control shrink = { |
| 2332 | .gfp_mask = sc.gfp_mask, |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2333 | }; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2334 | |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 2335 | /* |
| 2336 | * Unlike direct reclaim via alloc_pages(), memcg's reclaim doesn't |
| 2337 | * take care of from where we get pages. So the node where we start the |
| 2338 | * scan does not need to be the current node. |
| 2339 | */ |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 2340 | nid = mem_cgroup_select_victim_node(memcg); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 2341 | |
| 2342 | zonelist = NODE_DATA(nid)->node_zonelists; |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 2343 | |
| 2344 | trace_mm_vmscan_memcg_reclaim_begin(0, |
| 2345 | sc.may_writepage, |
| 2346 | sc.gfp_mask); |
| 2347 | |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 2348 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink); |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 2349 | |
| 2350 | trace_mm_vmscan_memcg_reclaim_end(nr_reclaimed); |
| 2351 | |
| 2352 | return nr_reclaimed; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2353 | } |
| 2354 | #endif |
| 2355 | |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2356 | static void age_active_anon(struct zone *zone, struct scan_control *sc) |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 2357 | { |
Johannes Weiner | b95a2f2 | 2012-01-12 17:18:06 -0800 | [diff] [blame] | 2358 | struct mem_cgroup *memcg; |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 2359 | |
Johannes Weiner | b95a2f2 | 2012-01-12 17:18:06 -0800 | [diff] [blame] | 2360 | if (!total_swap_pages) |
| 2361 | return; |
| 2362 | |
| 2363 | memcg = mem_cgroup_iter(NULL, NULL, NULL); |
| 2364 | do { |
| 2365 | struct mem_cgroup_zone mz = { |
| 2366 | .mem_cgroup = memcg, |
| 2367 | .zone = zone, |
| 2368 | }; |
| 2369 | |
| 2370 | if (inactive_anon_is_low(&mz)) |
| 2371 | shrink_active_list(SWAP_CLUSTER_MAX, &mz, |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2372 | sc, LRU_ACTIVE_ANON); |
Johannes Weiner | b95a2f2 | 2012-01-12 17:18:06 -0800 | [diff] [blame] | 2373 | |
| 2374 | memcg = mem_cgroup_iter(NULL, memcg, NULL); |
| 2375 | } while (memcg); |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 2376 | } |
| 2377 | |
Johannes Weiner | 791355e | 2012-11-29 13:54:23 -0800 | [diff] [blame] | 2378 | static bool zone_balanced(struct zone *zone, int order, |
| 2379 | unsigned long balance_gap, int classzone_idx) |
| 2380 | { |
| 2381 | if (!zone_watermark_ok_safe(zone, order, high_wmark_pages(zone) + |
| 2382 | balance_gap, classzone_idx, 0)) |
| 2383 | return false; |
| 2384 | |
| 2385 | if (COMPACTION_BUILD && order && !compaction_suitable(zone, order)) |
| 2386 | return false; |
| 2387 | |
| 2388 | return true; |
| 2389 | } |
| 2390 | |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2391 | /* |
| 2392 | * pgdat_balanced is used when checking if a node is balanced for high-order |
| 2393 | * allocations. Only zones that meet watermarks and are in a zone allowed |
| 2394 | * by the callers classzone_idx are added to balanced_pages. The total of |
| 2395 | * balanced pages must be at least 25% of the zones allowed by classzone_idx |
| 2396 | * for the node to be considered balanced. Forcing all zones to be balanced |
| 2397 | * for high orders can cause excessive reclaim when there are imbalanced zones. |
| 2398 | * The choice of 25% is due to |
| 2399 | * o a 16M DMA zone that is balanced will not balance a zone on any |
| 2400 | * reasonable sized machine |
| 2401 | * 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] | 2402 | * percentage of the middle zones. For example, on 32-bit x86, highmem |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2403 | * would need to be at least 256M for it to be balance a whole node. |
| 2404 | * Similarly, on x86-64 the Normal zone would need to be at least 1G |
| 2405 | * to balance a node on its own. These seemed like reasonable ratios. |
| 2406 | */ |
| 2407 | static bool pgdat_balanced(pg_data_t *pgdat, unsigned long balanced_pages, |
| 2408 | int classzone_idx) |
| 2409 | { |
| 2410 | unsigned long present_pages = 0; |
| 2411 | int i; |
| 2412 | |
| 2413 | for (i = 0; i <= classzone_idx; i++) |
| 2414 | present_pages += pgdat->node_zones[i].present_pages; |
| 2415 | |
Shaohua Li | 4746efd | 2011-07-19 08:49:26 -0700 | [diff] [blame] | 2416 | /* A special case here: if zone has no page, we think it's balanced */ |
| 2417 | return balanced_pages >= (present_pages >> 2); |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2418 | } |
| 2419 | |
Mel Gorman | f50de2d | 2009-12-14 17:58:53 -0800 | [diff] [blame] | 2420 | /* is kswapd sleeping prematurely? */ |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2421 | static bool sleeping_prematurely(pg_data_t *pgdat, int order, long remaining, |
| 2422 | int classzone_idx) |
Mel Gorman | f50de2d | 2009-12-14 17:58:53 -0800 | [diff] [blame] | 2423 | { |
KOSAKI Motohiro | bb3ab59 | 2009-12-14 17:58:55 -0800 | [diff] [blame] | 2424 | int i; |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2425 | unsigned long balanced = 0; |
| 2426 | bool all_zones_ok = true; |
Mel Gorman | f50de2d | 2009-12-14 17:58:53 -0800 | [diff] [blame] | 2427 | |
| 2428 | /* If a direct reclaimer woke kswapd within HZ/10, it's premature */ |
| 2429 | if (remaining) |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2430 | return true; |
Mel Gorman | f50de2d | 2009-12-14 17:58:53 -0800 | [diff] [blame] | 2431 | |
Mel Gorman | 0abdee2 | 2011-01-13 15:46:22 -0800 | [diff] [blame] | 2432 | /* Check the watermark levels */ |
Mel Gorman | 08951e5 | 2011-07-08 15:39:36 -0700 | [diff] [blame] | 2433 | for (i = 0; i <= classzone_idx; i++) { |
KOSAKI Motohiro | bb3ab59 | 2009-12-14 17:58:55 -0800 | [diff] [blame] | 2434 | struct zone *zone = pgdat->node_zones + i; |
| 2435 | |
| 2436 | if (!populated_zone(zone)) |
| 2437 | continue; |
| 2438 | |
Mel Gorman | 355b09c | 2011-01-13 15:46:24 -0800 | [diff] [blame] | 2439 | /* |
| 2440 | * balance_pgdat() skips over all_unreclaimable after |
| 2441 | * DEF_PRIORITY. Effectively, it considers them balanced so |
| 2442 | * they must be considered balanced here as well if kswapd |
| 2443 | * is to sleep |
| 2444 | */ |
Lisa Du | e093521 | 2013-09-11 14:22:36 -0700 | [diff] [blame] | 2445 | if (!zone_reclaimable(zone)) { |
Mel Gorman | 355b09c | 2011-01-13 15:46:24 -0800 | [diff] [blame] | 2446 | balanced += zone->present_pages; |
KOSAKI Motohiro | de3fab3 | 2010-01-15 17:01:25 -0800 | [diff] [blame] | 2447 | continue; |
Mel Gorman | 355b09c | 2011-01-13 15:46:24 -0800 | [diff] [blame] | 2448 | } |
KOSAKI Motohiro | de3fab3 | 2010-01-15 17:01:25 -0800 | [diff] [blame] | 2449 | |
Johannes Weiner | 791355e | 2012-11-29 13:54:23 -0800 | [diff] [blame] | 2450 | if (!zone_balanced(zone, order, 0, i)) |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2451 | all_zones_ok = false; |
| 2452 | else |
| 2453 | balanced += zone->present_pages; |
KOSAKI Motohiro | bb3ab59 | 2009-12-14 17:58:55 -0800 | [diff] [blame] | 2454 | } |
Mel Gorman | f50de2d | 2009-12-14 17:58:53 -0800 | [diff] [blame] | 2455 | |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2456 | /* |
| 2457 | * For high-order requests, the balanced zones must contain at least |
| 2458 | * 25% of the nodes pages for kswapd to sleep. For order-0, all zones |
| 2459 | * must be balanced |
| 2460 | */ |
| 2461 | if (order) |
Johannes Weiner | afc7e32 | 2011-05-24 17:11:09 -0700 | [diff] [blame] | 2462 | return !pgdat_balanced(pgdat, balanced, classzone_idx); |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2463 | else |
| 2464 | return !all_zones_ok; |
Mel Gorman | f50de2d | 2009-12-14 17:58:53 -0800 | [diff] [blame] | 2465 | } |
| 2466 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2467 | /* |
| 2468 | * 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] | 2469 | * they are all at high_wmark_pages(zone). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2470 | * |
Mel Gorman | 0abdee2 | 2011-01-13 15:46:22 -0800 | [diff] [blame] | 2471 | * Returns the final order kswapd was reclaiming at |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2472 | * |
| 2473 | * There is special handling here for zones which are full of pinned pages. |
| 2474 | * This can happen if the pages are all mlocked, or if they are all used by |
| 2475 | * device drivers (say, ZONE_DMA). Or if they are all in use by hugetlb. |
| 2476 | * What we do is to detect the case where all pages in the zone have been |
| 2477 | * scanned twice and there has been zero successful reclaim. Mark the zone as |
| 2478 | * dead and from now on, only perform a short scan. Basically we're polling |
| 2479 | * the zone for when the problem goes away. |
| 2480 | * |
| 2481 | * kswapd scans the zones in the highmem->normal->dma direction. It skips |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 2482 | * zones which have free_pages > high_wmark_pages(zone), but once a zone is |
| 2483 | * found to have free_pages <= high_wmark_pages(zone), we scan that zone and the |
| 2484 | * lower zones regardless of the number of free pages in the lower zones. This |
| 2485 | * interoperates with the page allocator fallback scheme to ensure that aging |
| 2486 | * of pages is balanced across the zones. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2487 | */ |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2488 | static unsigned long balance_pgdat(pg_data_t *pgdat, int order, |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2489 | int *classzone_idx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2490 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2491 | int all_zones_ok; |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2492 | unsigned long balanced; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2493 | int i; |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2494 | int end_zone = 0; /* Inclusive. 0 = ZONE_DMA */ |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 2495 | unsigned long total_scanned; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2496 | struct reclaim_state *reclaim_state = current->reclaim_state; |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 2497 | unsigned long nr_soft_reclaimed; |
| 2498 | unsigned long nr_soft_scanned; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2499 | struct scan_control sc = { |
| 2500 | .gfp_mask = GFP_KERNEL, |
Johannes Weiner | a6dc60f | 2009-03-31 15:19:30 -0700 | [diff] [blame] | 2501 | .may_unmap = 1, |
KOSAKI Motohiro | 2e2e425 | 2009-04-21 12:24:57 -0700 | [diff] [blame] | 2502 | .may_swap = 1, |
KOSAKI Motohiro | 22fba33 | 2009-12-14 17:59:10 -0800 | [diff] [blame] | 2503 | /* |
| 2504 | * kswapd doesn't want to be bailed out while reclaim. because |
| 2505 | * we want to put equal scanning pressure on each zone. |
| 2506 | */ |
| 2507 | .nr_to_reclaim = ULONG_MAX, |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 2508 | .order = order, |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 2509 | .target_mem_cgroup = NULL, |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2510 | }; |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 2511 | struct shrink_control shrink = { |
| 2512 | .gfp_mask = sc.gfp_mask, |
| 2513 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2514 | loop_again: |
| 2515 | total_scanned = 0; |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2516 | sc.priority = DEF_PRIORITY; |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2517 | sc.nr_reclaimed = 0; |
Christoph Lameter | c0bbbc7 | 2006-06-11 15:22:26 -0700 | [diff] [blame] | 2518 | sc.may_writepage = !laptop_mode; |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 2519 | count_vm_event(PAGEOUTRUN); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2520 | |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2521 | do { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2522 | unsigned long lru_pages = 0; |
KOSAKI Motohiro | bb3ab59 | 2009-12-14 17:58:55 -0800 | [diff] [blame] | 2523 | int has_under_min_watermark_zone = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2524 | |
| 2525 | all_zones_ok = 1; |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2526 | balanced = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2527 | |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2528 | /* |
| 2529 | * Scan in the highmem->dma direction for the highest |
| 2530 | * zone which needs scanning |
| 2531 | */ |
| 2532 | for (i = pgdat->nr_zones - 1; i >= 0; i--) { |
| 2533 | struct zone *zone = pgdat->node_zones + i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2534 | |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2535 | if (!populated_zone(zone)) |
| 2536 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2537 | |
Lisa Du | e093521 | 2013-09-11 14:22:36 -0700 | [diff] [blame] | 2538 | if (sc.priority != DEF_PRIORITY && |
| 2539 | !zone_reclaimable(zone)) |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2540 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2541 | |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 2542 | /* |
| 2543 | * Do some background aging of the anon list, to give |
| 2544 | * pages a chance to be referenced before reclaiming. |
| 2545 | */ |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2546 | age_active_anon(zone, &sc); |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 2547 | |
Mel Gorman | cc715d9 | 2012-03-21 16:34:00 -0700 | [diff] [blame] | 2548 | /* |
| 2549 | * If the number of buffer_heads in the machine |
| 2550 | * exceeds the maximum allowed level and this node |
| 2551 | * has a highmem zone, force kswapd to reclaim from |
| 2552 | * it to relieve lowmem pressure. |
| 2553 | */ |
| 2554 | if (buffer_heads_over_limit && is_highmem_idx(i)) { |
| 2555 | end_zone = i; |
| 2556 | break; |
| 2557 | } |
| 2558 | |
Johannes Weiner | 791355e | 2012-11-29 13:54:23 -0800 | [diff] [blame] | 2559 | if (!zone_balanced(zone, order, 0, 0)) { |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2560 | end_zone = i; |
Andrew Morton | e1dbeda | 2006-12-06 20:32:01 -0800 | [diff] [blame] | 2561 | break; |
Shaohua Li | 439423f | 2011-08-25 15:59:12 -0700 | [diff] [blame] | 2562 | } else { |
| 2563 | /* If balanced, clear the congested flag */ |
| 2564 | zone_clear_flag(zone, ZONE_CONGESTED); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2565 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2566 | } |
Andrew Morton | e1dbeda | 2006-12-06 20:32:01 -0800 | [diff] [blame] | 2567 | if (i < 0) |
| 2568 | goto out; |
| 2569 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2570 | for (i = 0; i <= end_zone; i++) { |
| 2571 | struct zone *zone = pgdat->node_zones + i; |
| 2572 | |
Wu Fengguang | adea02a | 2009-09-21 17:01:42 -0700 | [diff] [blame] | 2573 | lru_pages += zone_reclaimable_pages(zone); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2574 | } |
| 2575 | |
| 2576 | /* |
| 2577 | * Now scan the zone in the dma->highmem direction, stopping |
| 2578 | * at the last zone which needs scanning. |
| 2579 | * |
| 2580 | * We do this because the page allocator works in the opposite |
| 2581 | * direction. This prevents the page allocator from allocating |
| 2582 | * pages behind kswapd's direction of progress, which would |
| 2583 | * cause too much scanning of the lower zones. |
| 2584 | */ |
| 2585 | for (i = 0; i <= end_zone; i++) { |
| 2586 | struct zone *zone = pgdat->node_zones + i; |
Rik van Riel | fe2c2a1 | 2012-03-21 16:33:51 -0700 | [diff] [blame] | 2587 | int nr_slab, testorder; |
Mel Gorman | 8afdcec | 2011-03-22 16:33:04 -0700 | [diff] [blame] | 2588 | unsigned long balance_gap; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2589 | |
Con Kolivas | f3fe651 | 2006-01-06 00:11:15 -0800 | [diff] [blame] | 2590 | if (!populated_zone(zone)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2591 | continue; |
| 2592 | |
Lisa Du | e093521 | 2013-09-11 14:22:36 -0700 | [diff] [blame] | 2593 | if (sc.priority != DEF_PRIORITY && |
| 2594 | !zone_reclaimable(zone)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2595 | continue; |
| 2596 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2597 | sc.nr_scanned = 0; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2598 | |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 2599 | nr_soft_scanned = 0; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2600 | /* |
| 2601 | * Call soft limit reclaim before calling shrink_zone. |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2602 | */ |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 2603 | nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone, |
| 2604 | order, sc.gfp_mask, |
| 2605 | &nr_soft_scanned); |
| 2606 | sc.nr_reclaimed += nr_soft_reclaimed; |
| 2607 | total_scanned += nr_soft_scanned; |
KOSAKI Motohiro | 00918b6 | 2010-08-10 18:03:05 -0700 | [diff] [blame] | 2608 | |
Rik van Riel | 32a4330 | 2007-10-16 01:24:50 -0700 | [diff] [blame] | 2609 | /* |
Mel Gorman | 8afdcec | 2011-03-22 16:33:04 -0700 | [diff] [blame] | 2610 | * We put equal pressure on every zone, unless |
| 2611 | * one zone has way too many pages free |
| 2612 | * already. The "too many pages" is defined |
| 2613 | * as the high wmark plus a "gap" where the |
| 2614 | * gap is either the low watermark or 1% |
| 2615 | * of the zone, whichever is smaller. |
Rik van Riel | 32a4330 | 2007-10-16 01:24:50 -0700 | [diff] [blame] | 2616 | */ |
Mel Gorman | 8afdcec | 2011-03-22 16:33:04 -0700 | [diff] [blame] | 2617 | balance_gap = min(low_wmark_pages(zone), |
| 2618 | (zone->present_pages + |
| 2619 | KSWAPD_ZONE_BALANCE_GAP_RATIO-1) / |
| 2620 | KSWAPD_ZONE_BALANCE_GAP_RATIO); |
Rik van Riel | fe2c2a1 | 2012-03-21 16:33:51 -0700 | [diff] [blame] | 2621 | /* |
| 2622 | * Kswapd reclaims only single pages with compaction |
| 2623 | * enabled. Trying too hard to reclaim until contiguous |
| 2624 | * free pages have become available can hurt performance |
| 2625 | * by evicting too much useful data from memory. |
| 2626 | * Do not reclaim more than needed for compaction. |
| 2627 | */ |
| 2628 | testorder = order; |
| 2629 | if (COMPACTION_BUILD && order && |
| 2630 | compaction_suitable(zone, order) != |
| 2631 | COMPACT_SKIPPED) |
| 2632 | testorder = 0; |
| 2633 | |
Mel Gorman | cc715d9 | 2012-03-21 16:34:00 -0700 | [diff] [blame] | 2634 | if ((buffer_heads_over_limit && is_highmem_idx(i)) || |
Johannes Weiner | 791355e | 2012-11-29 13:54:23 -0800 | [diff] [blame] | 2635 | !zone_balanced(zone, testorder, |
| 2636 | balance_gap, end_zone)) { |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2637 | shrink_zone(zone, &sc); |
Andrea Arcangeli | 5a03b05 | 2011-01-13 15:47:11 -0800 | [diff] [blame] | 2638 | |
Mel Gorman | d7868da | 2011-07-08 15:39:38 -0700 | [diff] [blame] | 2639 | reclaim_state->reclaimed_slab = 0; |
| 2640 | nr_slab = shrink_slab(&shrink, sc.nr_scanned, lru_pages); |
| 2641 | sc.nr_reclaimed += reclaim_state->reclaimed_slab; |
| 2642 | total_scanned += sc.nr_scanned; |
| 2643 | |
Mel Gorman | d7868da | 2011-07-08 15:39:38 -0700 | [diff] [blame] | 2644 | } |
| 2645 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2646 | /* |
| 2647 | * If we've done a decent amount of scanning and |
| 2648 | * the reclaim ratio is low, start doing writepage |
| 2649 | * even in laptop mode |
| 2650 | */ |
| 2651 | if (total_scanned > SWAP_CLUSTER_MAX * 2 && |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2652 | total_scanned > sc.nr_reclaimed + sc.nr_reclaimed / 2) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2653 | sc.may_writepage = 1; |
KOSAKI Motohiro | bb3ab59 | 2009-12-14 17:58:55 -0800 | [diff] [blame] | 2654 | |
Lisa Du | e093521 | 2013-09-11 14:22:36 -0700 | [diff] [blame] | 2655 | if (!zone_reclaimable(zone)) { |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2656 | if (end_zone && end_zone == i) |
| 2657 | end_zone--; |
Mel Gorman | d7868da | 2011-07-08 15:39:38 -0700 | [diff] [blame] | 2658 | continue; |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2659 | } |
Mel Gorman | d7868da | 2011-07-08 15:39:38 -0700 | [diff] [blame] | 2660 | |
Johannes Weiner | 791355e | 2012-11-29 13:54:23 -0800 | [diff] [blame] | 2661 | if (!zone_balanced(zone, testorder, 0, end_zone)) { |
Minchan Kim | 45973d7 | 2010-03-05 13:41:45 -0800 | [diff] [blame] | 2662 | all_zones_ok = 0; |
| 2663 | /* |
| 2664 | * We are still under min water mark. This |
| 2665 | * means that we have a GFP_ATOMIC allocation |
| 2666 | * failure risk. Hurry up! |
| 2667 | */ |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 2668 | if (!zone_watermark_ok_safe(zone, order, |
Minchan Kim | 45973d7 | 2010-03-05 13:41:45 -0800 | [diff] [blame] | 2669 | min_wmark_pages(zone), end_zone, 0)) |
| 2670 | has_under_min_watermark_zone = 1; |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 2671 | } else { |
| 2672 | /* |
| 2673 | * If a zone reaches its high watermark, |
| 2674 | * consider it to be no longer congested. It's |
| 2675 | * possible there are dirty pages backed by |
| 2676 | * congested BDIs but as pressure is relieved, |
| 2677 | * spectulatively avoid congestion waits |
| 2678 | */ |
| 2679 | zone_clear_flag(zone, ZONE_CONGESTED); |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2680 | if (i <= *classzone_idx) |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2681 | balanced += zone->present_pages; |
Minchan Kim | 45973d7 | 2010-03-05 13:41:45 -0800 | [diff] [blame] | 2682 | } |
KOSAKI Motohiro | bb3ab59 | 2009-12-14 17:58:55 -0800 | [diff] [blame] | 2683 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2684 | } |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2685 | if (all_zones_ok || (order && pgdat_balanced(pgdat, balanced, *classzone_idx))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2686 | break; /* kswapd: all done */ |
| 2687 | /* |
| 2688 | * OK, kswapd is getting into trouble. Take a nap, then take |
| 2689 | * another pass across the zones. |
| 2690 | */ |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2691 | if (total_scanned && (sc.priority < DEF_PRIORITY - 2)) { |
KOSAKI Motohiro | bb3ab59 | 2009-12-14 17:58:55 -0800 | [diff] [blame] | 2692 | if (has_under_min_watermark_zone) |
| 2693 | count_vm_event(KSWAPD_SKIP_CONGESTION_WAIT); |
| 2694 | else |
| 2695 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
| 2696 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2697 | |
| 2698 | /* |
| 2699 | * We do this so kswapd doesn't build up large priorities for |
| 2700 | * example when it is freeing in parallel with allocators. It |
| 2701 | * matches the direct reclaim path behaviour in terms of impact |
| 2702 | * on zone->*_priority. |
| 2703 | */ |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2704 | if (sc.nr_reclaimed >= SWAP_CLUSTER_MAX) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2705 | break; |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2706 | } while (--sc.priority >= 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2707 | out: |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2708 | |
| 2709 | /* |
| 2710 | * order-0: All zones must meet high watermark for a balanced node |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2711 | * high-order: Balanced zones must make up at least 25% of the node |
| 2712 | * for the node to be balanced |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2713 | */ |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2714 | if (!(all_zones_ok || (order && pgdat_balanced(pgdat, balanced, *classzone_idx)))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2715 | cond_resched(); |
Rafael J. Wysocki | 8357376 | 2006-12-06 20:34:18 -0800 | [diff] [blame] | 2716 | |
| 2717 | try_to_freeze(); |
| 2718 | |
KOSAKI Motohiro | 73ce02e | 2009-01-06 14:40:33 -0800 | [diff] [blame] | 2719 | /* |
| 2720 | * Fragmentation may mean that the system cannot be |
| 2721 | * rebalanced for high-order allocations in all zones. |
| 2722 | * At this point, if nr_reclaimed < SWAP_CLUSTER_MAX, |
| 2723 | * it means the zones have been fully scanned and are still |
| 2724 | * not balanced. For high-order allocations, there is |
| 2725 | * little point trying all over again as kswapd may |
| 2726 | * infinite loop. |
| 2727 | * |
| 2728 | * Instead, recheck all watermarks at order-0 as they |
| 2729 | * are the most important. If watermarks are ok, kswapd will go |
| 2730 | * back to sleep. High-order users can still perform direct |
| 2731 | * reclaim if they wish. |
| 2732 | */ |
| 2733 | if (sc.nr_reclaimed < SWAP_CLUSTER_MAX) |
| 2734 | order = sc.order = 0; |
| 2735 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2736 | goto loop_again; |
| 2737 | } |
| 2738 | |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2739 | /* |
| 2740 | * If kswapd was reclaiming at a higher order, it has the option of |
| 2741 | * sleeping without all zones being balanced. Before it does, it must |
| 2742 | * ensure that the watermarks for order-0 on *all* zones are met and |
| 2743 | * that the congestion flags are cleared. The congestion flag must |
| 2744 | * be cleared as kswapd is the only mechanism that clears the flag |
| 2745 | * and it is potentially going to sleep here. |
| 2746 | */ |
| 2747 | if (order) { |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 2748 | int zones_need_compaction = 1; |
| 2749 | |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2750 | for (i = 0; i <= end_zone; i++) { |
| 2751 | struct zone *zone = pgdat->node_zones + i; |
| 2752 | |
| 2753 | if (!populated_zone(zone)) |
| 2754 | continue; |
| 2755 | |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 2756 | /* Check if the memory needs to be defragmented. */ |
| 2757 | if (zone_watermark_ok(zone, order, |
| 2758 | low_wmark_pages(zone), *classzone_idx, 0)) |
| 2759 | zones_need_compaction = 0; |
| 2760 | |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2761 | /* If balanced, clear the congested flag */ |
| 2762 | zone_clear_flag(zone, ZONE_CONGESTED); |
| 2763 | } |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 2764 | |
| 2765 | if (zones_need_compaction) |
| 2766 | compact_pgdat(pgdat, order); |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2767 | } |
| 2768 | |
Mel Gorman | 0abdee2 | 2011-01-13 15:46:22 -0800 | [diff] [blame] | 2769 | /* |
| 2770 | * Return the order we were reclaiming at so sleeping_prematurely() |
| 2771 | * makes a decision on the order we were last reclaiming at. However, |
| 2772 | * if another caller entered the allocator slow path while kswapd |
| 2773 | * was awake, order will remain at the higher level |
| 2774 | */ |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2775 | *classzone_idx = end_zone; |
Mel Gorman | 0abdee2 | 2011-01-13 15:46:22 -0800 | [diff] [blame] | 2776 | return order; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2777 | } |
| 2778 | |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2779 | 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] | 2780 | { |
| 2781 | long remaining = 0; |
| 2782 | DEFINE_WAIT(wait); |
| 2783 | |
| 2784 | if (freezing(current) || kthread_should_stop()) |
| 2785 | return; |
| 2786 | |
| 2787 | prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); |
| 2788 | |
| 2789 | /* Try to sleep for a short interval */ |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2790 | if (!sleeping_prematurely(pgdat, order, remaining, classzone_idx)) { |
KOSAKI Motohiro | f0bc0a6 | 2011-01-13 15:45:50 -0800 | [diff] [blame] | 2791 | remaining = schedule_timeout(HZ/10); |
| 2792 | finish_wait(&pgdat->kswapd_wait, &wait); |
| 2793 | prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); |
| 2794 | } |
| 2795 | |
| 2796 | /* |
| 2797 | * After a short sleep, check if it was a premature sleep. If not, then |
| 2798 | * go fully to sleep until explicitly woken up. |
| 2799 | */ |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2800 | if (!sleeping_prematurely(pgdat, order, remaining, classzone_idx)) { |
KOSAKI Motohiro | f0bc0a6 | 2011-01-13 15:45:50 -0800 | [diff] [blame] | 2801 | trace_mm_vmscan_kswapd_sleep(pgdat->node_id); |
| 2802 | |
| 2803 | /* |
| 2804 | * vmstat counters are not perfectly accurate and the estimated |
| 2805 | * value for counters such as NR_FREE_PAGES can deviate from the |
| 2806 | * true value by nr_online_cpus * threshold. To avoid the zone |
| 2807 | * watermarks being breached while under pressure, we reduce the |
| 2808 | * per-cpu vmstat threshold while kswapd is awake and restore |
| 2809 | * them before going back to sleep. |
| 2810 | */ |
| 2811 | set_pgdat_percpu_threshold(pgdat, calculate_normal_threshold); |
Aaditya Kumar | 8af6845 | 2012-07-17 15:48:07 -0700 | [diff] [blame] | 2812 | |
Mel Gorman | 9317a2b | 2012-10-08 16:32:47 -0700 | [diff] [blame] | 2813 | /* |
| 2814 | * Compaction records what page blocks it recently failed to |
| 2815 | * isolate pages from and skips them in the future scanning. |
| 2816 | * When kswapd is going to sleep, it is reasonable to assume |
| 2817 | * that pages and compaction may succeed so reset the cache. |
| 2818 | */ |
| 2819 | reset_isolation_suitable(pgdat); |
| 2820 | |
Aaditya Kumar | 8af6845 | 2012-07-17 15:48:07 -0700 | [diff] [blame] | 2821 | if (!kthread_should_stop()) |
| 2822 | schedule(); |
| 2823 | |
KOSAKI Motohiro | f0bc0a6 | 2011-01-13 15:45:50 -0800 | [diff] [blame] | 2824 | set_pgdat_percpu_threshold(pgdat, calculate_pressure_threshold); |
| 2825 | } else { |
| 2826 | if (remaining) |
| 2827 | count_vm_event(KSWAPD_LOW_WMARK_HIT_QUICKLY); |
| 2828 | else |
| 2829 | count_vm_event(KSWAPD_HIGH_WMARK_HIT_QUICKLY); |
| 2830 | } |
| 2831 | finish_wait(&pgdat->kswapd_wait, &wait); |
| 2832 | } |
| 2833 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2834 | /* |
| 2835 | * The background pageout daemon, started as a kernel thread |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2836 | * from the init process. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2837 | * |
| 2838 | * This basically trickles out pages so that we have _some_ |
| 2839 | * free memory available even if there is no other activity |
| 2840 | * that frees anything up. This is needed for things like routing |
| 2841 | * etc, where we otherwise might have all activity going on in |
| 2842 | * asynchronous contexts that cannot page things out. |
| 2843 | * |
| 2844 | * If there are applications that are active memory-allocators |
| 2845 | * (most normal use), this basically shouldn't matter. |
| 2846 | */ |
| 2847 | static int kswapd(void *p) |
| 2848 | { |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2849 | unsigned long order, new_order; |
Alex,Shi | d2ebd0f6 | 2011-10-31 17:08:39 -0700 | [diff] [blame] | 2850 | unsigned balanced_order; |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2851 | int classzone_idx, new_classzone_idx; |
Alex,Shi | d2ebd0f6 | 2011-10-31 17:08:39 -0700 | [diff] [blame] | 2852 | int balanced_classzone_idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2853 | pg_data_t *pgdat = (pg_data_t*)p; |
| 2854 | struct task_struct *tsk = current; |
KOSAKI Motohiro | f0bc0a6 | 2011-01-13 15:45:50 -0800 | [diff] [blame] | 2855 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2856 | struct reclaim_state reclaim_state = { |
| 2857 | .reclaimed_slab = 0, |
| 2858 | }; |
Rusty Russell | a70f730 | 2009-03-13 14:49:46 +1030 | [diff] [blame] | 2859 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2860 | |
Nick Piggin | cf40bd1 | 2009-01-21 08:12:39 +0100 | [diff] [blame] | 2861 | lockdep_set_current_reclaim_state(GFP_KERNEL); |
| 2862 | |
Rusty Russell | 174596a | 2009-01-01 10:12:29 +1030 | [diff] [blame] | 2863 | if (!cpumask_empty(cpumask)) |
Mike Travis | c5f59f0 | 2008-04-04 18:11:10 -0700 | [diff] [blame] | 2864 | set_cpus_allowed_ptr(tsk, cpumask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2865 | current->reclaim_state = &reclaim_state; |
| 2866 | |
| 2867 | /* |
| 2868 | * Tell the memory management that we're a "memory allocator", |
| 2869 | * and that if we need more memory we should get access to it |
| 2870 | * regardless (see "__alloc_pages()"). "kswapd" should |
| 2871 | * never get caught in the normal page freeing logic. |
| 2872 | * |
| 2873 | * (Kswapd normally doesn't need memory anyway, but sometimes |
| 2874 | * you need a small amount of memory in order to be able to |
| 2875 | * page out something else, and this flag essentially protects |
| 2876 | * us from recursively trying to free more memory as we're |
| 2877 | * trying to free the first piece of memory in the first place). |
| 2878 | */ |
Christoph Lameter | 930d915 | 2006-01-08 01:00:47 -0800 | [diff] [blame] | 2879 | tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD; |
Rafael J. Wysocki | 8314418 | 2007-07-17 04:03:35 -0700 | [diff] [blame] | 2880 | set_freezable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2881 | |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2882 | order = new_order = 0; |
Alex,Shi | d2ebd0f6 | 2011-10-31 17:08:39 -0700 | [diff] [blame] | 2883 | balanced_order = 0; |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2884 | classzone_idx = new_classzone_idx = pgdat->nr_zones - 1; |
Alex,Shi | d2ebd0f6 | 2011-10-31 17:08:39 -0700 | [diff] [blame] | 2885 | balanced_classzone_idx = classzone_idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2886 | for ( ; ; ) { |
David Rientjes | 8fe23e0 | 2009-12-14 17:58:33 -0800 | [diff] [blame] | 2887 | int ret; |
Christoph Lameter | 3e1d1d2 | 2005-06-24 23:13:50 -0700 | [diff] [blame] | 2888 | |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2889 | /* |
| 2890 | * If the last balance_pgdat was unsuccessful it's unlikely a |
| 2891 | * new request of a similar or harder type will succeed soon |
| 2892 | * so consider going to sleep on the basis we reclaimed at |
| 2893 | */ |
Alex,Shi | d2ebd0f6 | 2011-10-31 17:08:39 -0700 | [diff] [blame] | 2894 | if (balanced_classzone_idx >= new_classzone_idx && |
| 2895 | balanced_order == new_order) { |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2896 | new_order = pgdat->kswapd_max_order; |
| 2897 | new_classzone_idx = pgdat->classzone_idx; |
| 2898 | pgdat->kswapd_max_order = 0; |
| 2899 | pgdat->classzone_idx = pgdat->nr_zones - 1; |
| 2900 | } |
| 2901 | |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2902 | if (order < new_order || classzone_idx > new_classzone_idx) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2903 | /* |
| 2904 | * Don't sleep if someone wants a larger 'order' |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2905 | * allocation or has tigher zone constraints |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2906 | */ |
| 2907 | order = new_order; |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2908 | classzone_idx = new_classzone_idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2909 | } else { |
Alex,Shi | d2ebd0f6 | 2011-10-31 17:08:39 -0700 | [diff] [blame] | 2910 | kswapd_try_to_sleep(pgdat, balanced_order, |
| 2911 | balanced_classzone_idx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2912 | order = pgdat->kswapd_max_order; |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2913 | classzone_idx = pgdat->classzone_idx; |
Alex,Shi | f0dfcde | 2011-10-31 17:08:45 -0700 | [diff] [blame] | 2914 | new_order = order; |
| 2915 | new_classzone_idx = classzone_idx; |
Mel Gorman | 4d40502 | 2011-01-13 15:46:23 -0800 | [diff] [blame] | 2916 | pgdat->kswapd_max_order = 0; |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2917 | pgdat->classzone_idx = pgdat->nr_zones - 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2918 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2919 | |
David Rientjes | 8fe23e0 | 2009-12-14 17:58:33 -0800 | [diff] [blame] | 2920 | ret = try_to_freeze(); |
| 2921 | if (kthread_should_stop()) |
| 2922 | break; |
| 2923 | |
| 2924 | /* |
| 2925 | * We can speed up thawing tasks if we don't call balance_pgdat |
| 2926 | * after returning from the refrigerator |
| 2927 | */ |
Mel Gorman | 33906bc | 2010-08-09 17:19:16 -0700 | [diff] [blame] | 2928 | if (!ret) { |
| 2929 | trace_mm_vmscan_kswapd_wake(pgdat->node_id, order); |
Alex,Shi | d2ebd0f6 | 2011-10-31 17:08:39 -0700 | [diff] [blame] | 2930 | balanced_classzone_idx = classzone_idx; |
| 2931 | balanced_order = balance_pgdat(pgdat, order, |
| 2932 | &balanced_classzone_idx); |
Mel Gorman | 33906bc | 2010-08-09 17:19:16 -0700 | [diff] [blame] | 2933 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2934 | } |
Takamori Yamaguchi | b49d5c6 | 2012-11-08 15:53:39 -0800 | [diff] [blame] | 2935 | |
Johannes Weiner | 2abdb32a | 2014-06-06 14:35:35 -0700 | [diff] [blame] | 2936 | tsk->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD); |
Takamori Yamaguchi | b49d5c6 | 2012-11-08 15:53:39 -0800 | [diff] [blame] | 2937 | current->reclaim_state = NULL; |
Johannes Weiner | 2abdb32a | 2014-06-06 14:35:35 -0700 | [diff] [blame] | 2938 | lockdep_clear_current_reclaim_state(); |
| 2939 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2940 | return 0; |
| 2941 | } |
| 2942 | |
| 2943 | /* |
| 2944 | * A zone is low on free memory, so wake its kswapd task to service it. |
| 2945 | */ |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2946 | 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] | 2947 | { |
| 2948 | pg_data_t *pgdat; |
| 2949 | |
Con Kolivas | f3fe651 | 2006-01-06 00:11:15 -0800 | [diff] [blame] | 2950 | if (!populated_zone(zone)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2951 | return; |
| 2952 | |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2953 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2954 | return; |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 2955 | pgdat = zone->zone_pgdat; |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2956 | if (pgdat->kswapd_max_order < order) { |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 2957 | pgdat->kswapd_max_order = order; |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2958 | pgdat->classzone_idx = min(pgdat->classzone_idx, classzone_idx); |
| 2959 | } |
Con Kolivas | 8d0986e | 2005-09-13 01:25:07 -0700 | [diff] [blame] | 2960 | if (!waitqueue_active(&pgdat->kswapd_wait)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2961 | return; |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 2962 | if (zone_watermark_ok_safe(zone, order, low_wmark_pages(zone), 0, 0)) |
| 2963 | return; |
| 2964 | |
| 2965 | trace_mm_vmscan_wakeup_kswapd(pgdat->node_id, zone_idx(zone), order); |
Con Kolivas | 8d0986e | 2005-09-13 01:25:07 -0700 | [diff] [blame] | 2966 | wake_up_interruptible(&pgdat->kswapd_wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2967 | } |
| 2968 | |
Wu Fengguang | adea02a | 2009-09-21 17:01:42 -0700 | [diff] [blame] | 2969 | /* |
| 2970 | * The reclaimable count would be mostly accurate. |
| 2971 | * The less reclaimable pages may be |
| 2972 | * - mlocked pages, which will be moved to unevictable list when encountered |
| 2973 | * - mapped pages, which may require several travels to be reclaimed |
| 2974 | * - dirty pages, which is not "instantly" reclaimable |
| 2975 | */ |
| 2976 | unsigned long global_reclaimable_pages(void) |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2977 | { |
Wu Fengguang | adea02a | 2009-09-21 17:01:42 -0700 | [diff] [blame] | 2978 | int nr; |
| 2979 | |
| 2980 | nr = global_page_state(NR_ACTIVE_FILE) + |
| 2981 | global_page_state(NR_INACTIVE_FILE); |
| 2982 | |
Shaohua Li | 1abbca7 | 2013-02-22 16:34:38 -0800 | [diff] [blame] | 2983 | if (get_nr_swap_pages() > 0) |
Wu Fengguang | adea02a | 2009-09-21 17:01:42 -0700 | [diff] [blame] | 2984 | nr += global_page_state(NR_ACTIVE_ANON) + |
| 2985 | global_page_state(NR_INACTIVE_ANON); |
| 2986 | |
| 2987 | return nr; |
| 2988 | } |
| 2989 | |
Rafael J. Wysocki | c6f37f1 | 2009-05-24 22:16:31 +0200 | [diff] [blame] | 2990 | #ifdef CONFIG_HIBERNATION |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2991 | /* |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 2992 | * 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] | 2993 | * freed pages. |
| 2994 | * |
| 2995 | * Rather than trying to age LRUs the aim is to preserve the overall |
| 2996 | * LRU order by reclaiming preferentially |
| 2997 | * inactive > active > active referenced > active mapped |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2998 | */ |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 2999 | unsigned long shrink_all_memory(unsigned long nr_to_reclaim) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3000 | { |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 3001 | struct reclaim_state reclaim_state; |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 3002 | struct scan_control sc = { |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3003 | .gfp_mask = GFP_HIGHUSER_MOVABLE, |
| 3004 | .may_swap = 1, |
| 3005 | .may_unmap = 1, |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 3006 | .may_writepage = 1, |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3007 | .nr_to_reclaim = nr_to_reclaim, |
| 3008 | .hibernation_mode = 1, |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3009 | .order = 0, |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3010 | .priority = DEF_PRIORITY, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3011 | }; |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 3012 | struct shrink_control shrink = { |
| 3013 | .gfp_mask = sc.gfp_mask, |
| 3014 | }; |
| 3015 | struct zonelist *zonelist = node_zonelist(numa_node_id(), sc.gfp_mask); |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3016 | struct task_struct *p = current; |
| 3017 | unsigned long nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3018 | |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3019 | p->flags |= PF_MEMALLOC; |
| 3020 | lockdep_set_current_reclaim_state(sc.gfp_mask); |
| 3021 | reclaim_state.reclaimed_slab = 0; |
| 3022 | p->reclaim_state = &reclaim_state; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 3023 | |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 3024 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink); |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 3025 | |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3026 | p->reclaim_state = NULL; |
| 3027 | lockdep_clear_current_reclaim_state(); |
| 3028 | p->flags &= ~PF_MEMALLOC; |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 3029 | |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3030 | return nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3031 | } |
Rafael J. Wysocki | c6f37f1 | 2009-05-24 22:16:31 +0200 | [diff] [blame] | 3032 | #endif /* CONFIG_HIBERNATION */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3033 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3034 | /* It's optimal to keep kswapds on the same CPUs as their memory, but |
| 3035 | not required for correctness. So if the last cpu in a node goes |
| 3036 | away, we get changed to run anywhere: as the first one comes back, |
| 3037 | restore their cpu bindings. */ |
Chandra Seetharaman | 9c7b216 | 2006-06-27 02:54:07 -0700 | [diff] [blame] | 3038 | static int __devinit cpu_callback(struct notifier_block *nfb, |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 3039 | unsigned long action, void *hcpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3040 | { |
Yasunori Goto | 58c0a4a | 2007-10-16 01:25:40 -0700 | [diff] [blame] | 3041 | int nid; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3042 | |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 3043 | if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN) { |
Yasunori Goto | 58c0a4a | 2007-10-16 01:25:40 -0700 | [diff] [blame] | 3044 | for_each_node_state(nid, N_HIGH_MEMORY) { |
Mike Travis | c5f59f0 | 2008-04-04 18:11:10 -0700 | [diff] [blame] | 3045 | pg_data_t *pgdat = NODE_DATA(nid); |
Rusty Russell | a70f730 | 2009-03-13 14:49:46 +1030 | [diff] [blame] | 3046 | const struct cpumask *mask; |
| 3047 | |
| 3048 | mask = cpumask_of_node(pgdat->node_id); |
Mike Travis | c5f59f0 | 2008-04-04 18:11:10 -0700 | [diff] [blame] | 3049 | |
Rusty Russell | 3e59794 | 2009-01-01 10:12:24 +1030 | [diff] [blame] | 3050 | if (cpumask_any_and(cpu_online_mask, mask) < nr_cpu_ids) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3051 | /* One of our CPUs online: restore mask */ |
Mike Travis | c5f59f0 | 2008-04-04 18:11:10 -0700 | [diff] [blame] | 3052 | set_cpus_allowed_ptr(pgdat->kswapd, mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3053 | } |
| 3054 | } |
| 3055 | return NOTIFY_OK; |
| 3056 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3057 | |
Yasunori Goto | 3218ae1 | 2006-06-27 02:53:33 -0700 | [diff] [blame] | 3058 | /* |
| 3059 | * This kswapd start function will be called by init and node-hot-add. |
| 3060 | * On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added. |
| 3061 | */ |
| 3062 | int kswapd_run(int nid) |
| 3063 | { |
| 3064 | pg_data_t *pgdat = NODE_DATA(nid); |
| 3065 | int ret = 0; |
| 3066 | |
| 3067 | if (pgdat->kswapd) |
| 3068 | return 0; |
| 3069 | |
| 3070 | pgdat->kswapd = kthread_run(kswapd, pgdat, "kswapd%d", nid); |
| 3071 | if (IS_ERR(pgdat->kswapd)) { |
| 3072 | /* failure at boot is fatal */ |
| 3073 | BUG_ON(system_state == SYSTEM_BOOTING); |
| 3074 | printk("Failed to start kswapd on node %d\n",nid); |
| 3075 | ret = -1; |
| 3076 | } |
| 3077 | return ret; |
| 3078 | } |
| 3079 | |
David Rientjes | 8fe23e0 | 2009-12-14 17:58:33 -0800 | [diff] [blame] | 3080 | /* |
Jiang Liu | ec75cfe | 2012-07-11 14:01:52 -0700 | [diff] [blame] | 3081 | * Called by memory hotplug when all memory in a node is offlined. Caller must |
| 3082 | * hold lock_memory_hotplug(). |
David Rientjes | 8fe23e0 | 2009-12-14 17:58:33 -0800 | [diff] [blame] | 3083 | */ |
| 3084 | void kswapd_stop(int nid) |
| 3085 | { |
| 3086 | struct task_struct *kswapd = NODE_DATA(nid)->kswapd; |
| 3087 | |
Jiang Liu | ec75cfe | 2012-07-11 14:01:52 -0700 | [diff] [blame] | 3088 | if (kswapd) { |
David Rientjes | 8fe23e0 | 2009-12-14 17:58:33 -0800 | [diff] [blame] | 3089 | kthread_stop(kswapd); |
Jiang Liu | ec75cfe | 2012-07-11 14:01:52 -0700 | [diff] [blame] | 3090 | NODE_DATA(nid)->kswapd = NULL; |
| 3091 | } |
David Rientjes | 8fe23e0 | 2009-12-14 17:58:33 -0800 | [diff] [blame] | 3092 | } |
| 3093 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3094 | static int __init kswapd_init(void) |
| 3095 | { |
Yasunori Goto | 3218ae1 | 2006-06-27 02:53:33 -0700 | [diff] [blame] | 3096 | int nid; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 3097 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3098 | swap_setup(); |
Christoph Lameter | 9422ffb | 2007-10-16 01:25:31 -0700 | [diff] [blame] | 3099 | for_each_node_state(nid, N_HIGH_MEMORY) |
Yasunori Goto | 3218ae1 | 2006-06-27 02:53:33 -0700 | [diff] [blame] | 3100 | kswapd_run(nid); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3101 | hotcpu_notifier(cpu_callback, 0); |
| 3102 | return 0; |
| 3103 | } |
| 3104 | |
| 3105 | module_init(kswapd_init) |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3106 | |
| 3107 | #ifdef CONFIG_NUMA |
| 3108 | /* |
| 3109 | * Zone reclaim mode |
| 3110 | * |
| 3111 | * If non-zero call zone_reclaim when the number of free pages falls below |
| 3112 | * the watermarks. |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3113 | */ |
| 3114 | int zone_reclaim_mode __read_mostly; |
| 3115 | |
Christoph Lameter | 1b2ffb7 | 2006-02-01 03:05:34 -0800 | [diff] [blame] | 3116 | #define RECLAIM_OFF 0 |
Fernando Luis Vazquez Cao | 7d03431 | 2008-07-29 22:33:41 -0700 | [diff] [blame] | 3117 | #define RECLAIM_ZONE (1<<0) /* Run shrink_inactive_list on the zone */ |
Christoph Lameter | 1b2ffb7 | 2006-02-01 03:05:34 -0800 | [diff] [blame] | 3118 | #define RECLAIM_WRITE (1<<1) /* Writeout pages during reclaim */ |
| 3119 | #define RECLAIM_SWAP (1<<2) /* Swap pages out during reclaim */ |
| 3120 | |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3121 | /* |
Christoph Lameter | a92f712 | 2006-02-01 03:05:32 -0800 | [diff] [blame] | 3122 | * Priority for ZONE_RECLAIM. This determines the fraction of pages |
| 3123 | * of a node considered for each zone_reclaim. 4 scans 1/16th of |
| 3124 | * a zone. |
| 3125 | */ |
| 3126 | #define ZONE_RECLAIM_PRIORITY 4 |
| 3127 | |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3128 | /* |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 3129 | * Percentage of pages in a zone that must be unmapped for zone_reclaim to |
| 3130 | * occur. |
| 3131 | */ |
| 3132 | int sysctl_min_unmapped_ratio = 1; |
| 3133 | |
| 3134 | /* |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3135 | * If the number of slab pages in a zone grows beyond this percentage then |
| 3136 | * slab reclaim needs to occur. |
| 3137 | */ |
| 3138 | int sysctl_min_slab_ratio = 5; |
| 3139 | |
Mel Gorman | 90afa5d | 2009-06-16 15:33:20 -0700 | [diff] [blame] | 3140 | static inline unsigned long zone_unmapped_file_pages(struct zone *zone) |
| 3141 | { |
| 3142 | unsigned long file_mapped = zone_page_state(zone, NR_FILE_MAPPED); |
| 3143 | unsigned long file_lru = zone_page_state(zone, NR_INACTIVE_FILE) + |
| 3144 | zone_page_state(zone, NR_ACTIVE_FILE); |
| 3145 | |
| 3146 | /* |
| 3147 | * It's possible for there to be more file mapped pages than |
| 3148 | * accounted for by the pages on the file LRU lists because |
| 3149 | * tmpfs pages accounted for as ANON can also be FILE_MAPPED |
| 3150 | */ |
| 3151 | return (file_lru > file_mapped) ? (file_lru - file_mapped) : 0; |
| 3152 | } |
| 3153 | |
| 3154 | /* Work out how many page cache pages we can reclaim in this reclaim_mode */ |
| 3155 | static long zone_pagecache_reclaimable(struct zone *zone) |
| 3156 | { |
| 3157 | long nr_pagecache_reclaimable; |
| 3158 | long delta = 0; |
| 3159 | |
| 3160 | /* |
| 3161 | * If RECLAIM_SWAP is set, then all file pages are considered |
| 3162 | * potentially reclaimable. Otherwise, we have to worry about |
| 3163 | * pages like swapcache and zone_unmapped_file_pages() provides |
| 3164 | * a better estimate |
| 3165 | */ |
| 3166 | if (zone_reclaim_mode & RECLAIM_SWAP) |
| 3167 | nr_pagecache_reclaimable = zone_page_state(zone, NR_FILE_PAGES); |
| 3168 | else |
| 3169 | nr_pagecache_reclaimable = zone_unmapped_file_pages(zone); |
| 3170 | |
| 3171 | /* If we can't clean pages, remove dirty pages from consideration */ |
| 3172 | if (!(zone_reclaim_mode & RECLAIM_WRITE)) |
| 3173 | delta += zone_page_state(zone, NR_FILE_DIRTY); |
| 3174 | |
| 3175 | /* Watch for any possible underflows due to delta */ |
| 3176 | if (unlikely(delta > nr_pagecache_reclaimable)) |
| 3177 | delta = nr_pagecache_reclaimable; |
| 3178 | |
| 3179 | return nr_pagecache_reclaimable - delta; |
| 3180 | } |
| 3181 | |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3182 | /* |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3183 | * Try to free up some pages from this zone through reclaim. |
| 3184 | */ |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3185 | 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] | 3186 | { |
Christoph Lameter | 7fb2d46 | 2006-03-22 00:08:22 -0800 | [diff] [blame] | 3187 | /* Minimum pages needed in order to stay on node */ |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 3188 | const unsigned long nr_pages = 1 << order; |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3189 | struct task_struct *p = current; |
| 3190 | struct reclaim_state reclaim_state; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3191 | struct scan_control sc = { |
| 3192 | .may_writepage = !!(zone_reclaim_mode & RECLAIM_WRITE), |
Johannes Weiner | a6dc60f | 2009-03-31 15:19:30 -0700 | [diff] [blame] | 3193 | .may_unmap = !!(zone_reclaim_mode & RECLAIM_SWAP), |
KOSAKI Motohiro | 2e2e425 | 2009-04-21 12:24:57 -0700 | [diff] [blame] | 3194 | .may_swap = 1, |
KOSAKI Motohiro | 22fba33 | 2009-12-14 17:59:10 -0800 | [diff] [blame] | 3195 | .nr_to_reclaim = max_t(unsigned long, nr_pages, |
| 3196 | SWAP_CLUSTER_MAX), |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3197 | .gfp_mask = gfp_mask, |
Johannes Weiner | bd2f619 | 2009-03-31 15:19:38 -0700 | [diff] [blame] | 3198 | .order = order, |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3199 | .priority = ZONE_RECLAIM_PRIORITY, |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3200 | }; |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 3201 | struct shrink_control shrink = { |
| 3202 | .gfp_mask = sc.gfp_mask, |
| 3203 | }; |
KOSAKI Motohiro | 1574804 | 2010-08-09 17:19:50 -0700 | [diff] [blame] | 3204 | unsigned long nr_slab_pages0, nr_slab_pages1; |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3205 | |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3206 | cond_resched(); |
Christoph Lameter | d4f7796 | 2006-02-24 13:04:22 -0800 | [diff] [blame] | 3207 | /* |
| 3208 | * We need to be able to allocate from the reserves for RECLAIM_SWAP |
| 3209 | * and we also need to be able to write out pages for RECLAIM_WRITE |
| 3210 | * and RECLAIM_SWAP. |
| 3211 | */ |
| 3212 | p->flags |= PF_MEMALLOC | PF_SWAPWRITE; |
KOSAKI Motohiro | 76ca542 | 2010-03-05 13:41:47 -0800 | [diff] [blame] | 3213 | lockdep_set_current_reclaim_state(gfp_mask); |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3214 | reclaim_state.reclaimed_slab = 0; |
| 3215 | p->reclaim_state = &reclaim_state; |
Christoph Lameter | c84db23 | 2006-02-01 03:05:29 -0800 | [diff] [blame] | 3216 | |
Mel Gorman | 90afa5d | 2009-06-16 15:33:20 -0700 | [diff] [blame] | 3217 | if (zone_pagecache_reclaimable(zone) > zone->min_unmapped_pages) { |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3218 | /* |
| 3219 | * Free memory by calling shrink zone with increasing |
| 3220 | * priorities until we have enough memory freed. |
| 3221 | */ |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3222 | do { |
Konstantin Khlebnikov | 038ada9 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3223 | shrink_zone(zone, &sc); |
| 3224 | } while (sc.nr_reclaimed < nr_pages && --sc.priority >= 0); |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3225 | } |
Christoph Lameter | c84db23 | 2006-02-01 03:05:29 -0800 | [diff] [blame] | 3226 | |
KOSAKI Motohiro | 1574804 | 2010-08-09 17:19:50 -0700 | [diff] [blame] | 3227 | nr_slab_pages0 = zone_page_state(zone, NR_SLAB_RECLAIMABLE); |
| 3228 | if (nr_slab_pages0 > zone->min_slab_pages) { |
Christoph Lameter | 2a16e3f | 2006-02-01 03:05:35 -0800 | [diff] [blame] | 3229 | /* |
Christoph Lameter | 7fb2d46 | 2006-03-22 00:08:22 -0800 | [diff] [blame] | 3230 | * shrink_slab() does not currently allow us to determine how |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3231 | * many pages were freed in this zone. So we take the current |
| 3232 | * number of slab pages and shake the slab until it is reduced |
| 3233 | * by the same nr_pages that we used for reclaiming unmapped |
| 3234 | * pages. |
Christoph Lameter | 2a16e3f | 2006-02-01 03:05:35 -0800 | [diff] [blame] | 3235 | * |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3236 | * Note that shrink_slab will free memory on all zones and may |
| 3237 | * take a long time. |
Christoph Lameter | 2a16e3f | 2006-02-01 03:05:35 -0800 | [diff] [blame] | 3238 | */ |
KOSAKI Motohiro | 4dc4b3d | 2010-08-09 17:19:54 -0700 | [diff] [blame] | 3239 | for (;;) { |
| 3240 | unsigned long lru_pages = zone_reclaimable_pages(zone); |
| 3241 | |
| 3242 | /* No reclaimable slab or very low memory pressure */ |
Ying Han | 1495f23 | 2011-05-24 17:12:27 -0700 | [diff] [blame] | 3243 | if (!shrink_slab(&shrink, sc.nr_scanned, lru_pages)) |
KOSAKI Motohiro | 4dc4b3d | 2010-08-09 17:19:54 -0700 | [diff] [blame] | 3244 | break; |
| 3245 | |
| 3246 | /* Freed enough memory */ |
| 3247 | nr_slab_pages1 = zone_page_state(zone, |
| 3248 | NR_SLAB_RECLAIMABLE); |
| 3249 | if (nr_slab_pages1 + nr_pages <= nr_slab_pages0) |
| 3250 | break; |
| 3251 | } |
Christoph Lameter | 83e33a4 | 2006-09-25 23:31:53 -0700 | [diff] [blame] | 3252 | |
| 3253 | /* |
| 3254 | * Update nr_reclaimed by the number of slab pages we |
| 3255 | * reclaimed from this zone. |
| 3256 | */ |
KOSAKI Motohiro | 1574804 | 2010-08-09 17:19:50 -0700 | [diff] [blame] | 3257 | nr_slab_pages1 = zone_page_state(zone, NR_SLAB_RECLAIMABLE); |
| 3258 | if (nr_slab_pages1 < nr_slab_pages0) |
| 3259 | sc.nr_reclaimed += nr_slab_pages0 - nr_slab_pages1; |
Christoph Lameter | 2a16e3f | 2006-02-01 03:05:35 -0800 | [diff] [blame] | 3260 | } |
| 3261 | |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3262 | p->reclaim_state = NULL; |
Christoph Lameter | d4f7796 | 2006-02-24 13:04:22 -0800 | [diff] [blame] | 3263 | current->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE); |
KOSAKI Motohiro | 76ca542 | 2010-03-05 13:41:47 -0800 | [diff] [blame] | 3264 | lockdep_clear_current_reclaim_state(); |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 3265 | return sc.nr_reclaimed >= nr_pages; |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3266 | } |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3267 | |
| 3268 | int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) |
| 3269 | { |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3270 | int node_id; |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3271 | int ret; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3272 | |
| 3273 | /* |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3274 | * Zone reclaim reclaims unmapped file backed pages and |
| 3275 | * slab pages if we are over the defined limits. |
Christoph Lameter | 34aa133 | 2006-06-30 01:55:37 -0700 | [diff] [blame] | 3276 | * |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 3277 | * A small portion of unmapped file backed pages is needed for |
| 3278 | * file I/O otherwise pages read by file I/O will be immediately |
| 3279 | * thrown out if the zone is overallocated. So we do not reclaim |
| 3280 | * if less than a specified percentage of the zone is used by |
| 3281 | * unmapped file backed pages. |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3282 | */ |
Mel Gorman | 90afa5d | 2009-06-16 15:33:20 -0700 | [diff] [blame] | 3283 | if (zone_pagecache_reclaimable(zone) <= zone->min_unmapped_pages && |
| 3284 | zone_page_state(zone, NR_SLAB_RECLAIMABLE) <= zone->min_slab_pages) |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 3285 | return ZONE_RECLAIM_FULL; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3286 | |
Lisa Du | e093521 | 2013-09-11 14:22:36 -0700 | [diff] [blame] | 3287 | if (!zone_reclaimable(zone)) |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 3288 | return ZONE_RECLAIM_FULL; |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3289 | |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3290 | /* |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3291 | * Do not scan if the allocation should not be delayed. |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3292 | */ |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3293 | if (!(gfp_mask & __GFP_WAIT) || (current->flags & PF_MEMALLOC)) |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 3294 | return ZONE_RECLAIM_NOSCAN; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3295 | |
| 3296 | /* |
| 3297 | * Only run zone reclaim on the local zone or on zones that do not |
| 3298 | * have associated processors. This will favor the local processor |
| 3299 | * over remote processors and spread off node memory allocations |
| 3300 | * as wide as possible. |
| 3301 | */ |
Christoph Lameter | 89fa302 | 2006-09-25 23:31:55 -0700 | [diff] [blame] | 3302 | node_id = zone_to_nid(zone); |
Christoph Lameter | 37c0708 | 2007-10-16 01:25:36 -0700 | [diff] [blame] | 3303 | if (node_state(node_id, N_CPU) && node_id != numa_node_id()) |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 3304 | return ZONE_RECLAIM_NOSCAN; |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3305 | |
| 3306 | if (zone_test_and_set_flag(zone, ZONE_RECLAIM_LOCKED)) |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 3307 | return ZONE_RECLAIM_NOSCAN; |
| 3308 | |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3309 | ret = __zone_reclaim(zone, gfp_mask, order); |
| 3310 | zone_clear_flag(zone, ZONE_RECLAIM_LOCKED); |
| 3311 | |
Mel Gorman | 24cf72518 | 2009-06-16 15:33:23 -0700 | [diff] [blame] | 3312 | if (!ret) |
| 3313 | count_vm_event(PGSCAN_ZONE_RECLAIM_FAILED); |
| 3314 | |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3315 | return ret; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3316 | } |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3317 | #endif |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3318 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3319 | /* |
| 3320 | * page_evictable - test whether a page is evictable |
| 3321 | * @page: the page to test |
| 3322 | * @vma: the VMA in which the page is or will be mapped, may be NULL |
| 3323 | * |
| 3324 | * 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] | 3325 | * lists vs unevictable list. The vma argument is !NULL when called from the |
| 3326 | * fault path to determine how to instantate a new page. |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3327 | * |
| 3328 | * Reasons page might not be evictable: |
Lee Schermerhorn | ba9ddf4 | 2008-10-18 20:26:42 -0700 | [diff] [blame] | 3329 | * (1) page's mapping marked unevictable |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 3330 | * (2) page is part of an mlocked VMA |
Lee Schermerhorn | ba9ddf4 | 2008-10-18 20:26:42 -0700 | [diff] [blame] | 3331 | * |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3332 | */ |
| 3333 | int page_evictable(struct page *page, struct vm_area_struct *vma) |
| 3334 | { |
| 3335 | |
Lee Schermerhorn | ba9ddf4 | 2008-10-18 20:26:42 -0700 | [diff] [blame] | 3336 | if (mapping_unevictable(page_mapping(page))) |
| 3337 | return 0; |
| 3338 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 3339 | if (PageMlocked(page) || (vma && is_mlocked_vma(vma, page))) |
| 3340 | return 0; |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3341 | |
| 3342 | return 1; |
| 3343 | } |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3344 | |
Hugh Dickins | 85046579 | 2012-01-20 14:34:19 -0800 | [diff] [blame] | 3345 | #ifdef CONFIG_SHMEM |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3346 | /** |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3347 | * check_move_unevictable_pages - check pages for evictability and move to appropriate zone lru list |
| 3348 | * @pages: array of pages to check |
| 3349 | * @nr_pages: number of pages to check |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3350 | * |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3351 | * Checks pages for evictability and moves them to the appropriate lru list. |
Hugh Dickins | 85046579 | 2012-01-20 14:34:19 -0800 | [diff] [blame] | 3352 | * |
| 3353 | * This function is only used for SysV IPC SHM_UNLOCK. |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3354 | */ |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3355 | void check_move_unevictable_pages(struct page **pages, int nr_pages) |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3356 | { |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 3357 | struct lruvec *lruvec; |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3358 | struct zone *zone = NULL; |
| 3359 | int pgscanned = 0; |
| 3360 | int pgrescued = 0; |
| 3361 | int i; |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3362 | |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3363 | for (i = 0; i < nr_pages; i++) { |
| 3364 | struct page *page = pages[i]; |
| 3365 | struct zone *pagezone; |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3366 | |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3367 | pgscanned++; |
| 3368 | pagezone = page_zone(page); |
| 3369 | if (pagezone != zone) { |
| 3370 | if (zone) |
| 3371 | spin_unlock_irq(&zone->lru_lock); |
| 3372 | zone = pagezone; |
| 3373 | spin_lock_irq(&zone->lru_lock); |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3374 | } |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3375 | |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3376 | if (!PageLRU(page) || !PageUnevictable(page)) |
| 3377 | continue; |
| 3378 | |
| 3379 | if (page_evictable(page, NULL)) { |
| 3380 | enum lru_list lru = page_lru_base_type(page); |
| 3381 | |
| 3382 | VM_BUG_ON(PageActive(page)); |
| 3383 | ClearPageUnevictable(page); |
| 3384 | __dec_zone_state(zone, NR_UNEVICTABLE); |
| 3385 | lruvec = mem_cgroup_lru_move_lists(zone, page, |
| 3386 | LRU_UNEVICTABLE, lru); |
| 3387 | list_move(&page->lru, &lruvec->lists[lru]); |
| 3388 | __inc_zone_state(zone, NR_INACTIVE_ANON + lru); |
| 3389 | pgrescued++; |
| 3390 | } |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3391 | } |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3392 | |
| 3393 | if (zone) { |
| 3394 | __count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued); |
| 3395 | __count_vm_events(UNEVICTABLE_PGSCANNED, pgscanned); |
| 3396 | spin_unlock_irq(&zone->lru_lock); |
| 3397 | } |
Hugh Dickins | 85046579 | 2012-01-20 14:34:19 -0800 | [diff] [blame] | 3398 | } |
| 3399 | #endif /* CONFIG_SHMEM */ |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3400 | |
Johannes Weiner | 264e56d | 2011-10-31 17:09:13 -0700 | [diff] [blame] | 3401 | static void warn_scan_unevictable_pages(void) |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3402 | { |
Johannes Weiner | 264e56d | 2011-10-31 17:09:13 -0700 | [diff] [blame] | 3403 | printk_once(KERN_WARNING |
KOSAKI Motohiro | 25bd91b | 2012-01-10 15:07:40 -0800 | [diff] [blame] | 3404 | "%s: The scan_unevictable_pages sysctl/node-interface has been " |
Johannes Weiner | 264e56d | 2011-10-31 17:09:13 -0700 | [diff] [blame] | 3405 | "disabled for lack of a legitimate use case. If you have " |
KOSAKI Motohiro | 25bd91b | 2012-01-10 15:07:40 -0800 | [diff] [blame] | 3406 | "one, please send an email to linux-mm@kvack.org.\n", |
| 3407 | current->comm); |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3408 | } |
| 3409 | |
| 3410 | /* |
| 3411 | * scan_unevictable_pages [vm] sysctl handler. On demand re-scan of |
| 3412 | * all nodes' unevictable lists for evictable pages |
| 3413 | */ |
| 3414 | unsigned long scan_unevictable_pages; |
| 3415 | |
| 3416 | int scan_unevictable_handler(struct ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 3417 | void __user *buffer, |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3418 | size_t *length, loff_t *ppos) |
| 3419 | { |
Johannes Weiner | 264e56d | 2011-10-31 17:09:13 -0700 | [diff] [blame] | 3420 | warn_scan_unevictable_pages(); |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 3421 | proc_doulongvec_minmax(table, write, buffer, length, ppos); |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3422 | scan_unevictable_pages = 0; |
| 3423 | return 0; |
| 3424 | } |
| 3425 | |
Thadeu Lima de Souza Cascardo | e4455ab | 2010-10-26 14:21:28 -0700 | [diff] [blame] | 3426 | #ifdef CONFIG_NUMA |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3427 | /* |
| 3428 | * per node 'scan_unevictable_pages' attribute. On demand re-scan of |
| 3429 | * a specified node's per zone unevictable lists for evictable pages. |
| 3430 | */ |
| 3431 | |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 3432 | static ssize_t read_scan_unevictable_node(struct device *dev, |
| 3433 | struct device_attribute *attr, |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3434 | char *buf) |
| 3435 | { |
Johannes Weiner | 264e56d | 2011-10-31 17:09:13 -0700 | [diff] [blame] | 3436 | warn_scan_unevictable_pages(); |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3437 | return sprintf(buf, "0\n"); /* always zero; should fit... */ |
| 3438 | } |
| 3439 | |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 3440 | static ssize_t write_scan_unevictable_node(struct device *dev, |
| 3441 | struct device_attribute *attr, |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3442 | const char *buf, size_t count) |
| 3443 | { |
Johannes Weiner | 264e56d | 2011-10-31 17:09:13 -0700 | [diff] [blame] | 3444 | warn_scan_unevictable_pages(); |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3445 | return 1; |
| 3446 | } |
| 3447 | |
| 3448 | |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 3449 | static DEVICE_ATTR(scan_unevictable_pages, S_IRUGO | S_IWUSR, |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3450 | read_scan_unevictable_node, |
| 3451 | write_scan_unevictable_node); |
| 3452 | |
| 3453 | int scan_unevictable_register_node(struct node *node) |
| 3454 | { |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 3455 | return device_create_file(&node->dev, &dev_attr_scan_unevictable_pages); |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3456 | } |
| 3457 | |
| 3458 | void scan_unevictable_unregister_node(struct node *node) |
| 3459 | { |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 3460 | device_remove_file(&node->dev, &dev_attr_scan_unevictable_pages); |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3461 | } |
Thadeu Lima de Souza Cascardo | e4455ab | 2010-10-26 14:21:28 -0700 | [diff] [blame] | 3462 | #endif |