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 | a6dc60f8 | 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 | 9e3b2f8 | 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 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 97 | #define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru)) |
| 98 | |
| 99 | #ifdef ARCH_HAS_PREFETCH |
| 100 | #define prefetch_prev_lru_page(_page, _base, _field) \ |
| 101 | do { \ |
| 102 | if ((_page)->lru.prev != _base) { \ |
| 103 | struct page *prev; \ |
| 104 | \ |
| 105 | prev = lru_to_page(&(_page->lru)); \ |
| 106 | prefetch(&prev->_field); \ |
| 107 | } \ |
| 108 | } while (0) |
| 109 | #else |
| 110 | #define prefetch_prev_lru_page(_page, _base, _field) do { } while (0) |
| 111 | #endif |
| 112 | |
| 113 | #ifdef ARCH_HAS_PREFETCHW |
| 114 | #define prefetchw_prev_lru_page(_page, _base, _field) \ |
| 115 | do { \ |
| 116 | if ((_page)->lru.prev != _base) { \ |
| 117 | struct page *prev; \ |
| 118 | \ |
| 119 | prev = lru_to_page(&(_page->lru)); \ |
| 120 | prefetchw(&prev->_field); \ |
| 121 | } \ |
| 122 | } while (0) |
| 123 | #else |
| 124 | #define prefetchw_prev_lru_page(_page, _base, _field) do { } while (0) |
| 125 | #endif |
| 126 | |
| 127 | /* |
| 128 | * From 0 .. 100. Higher means more swappy. |
| 129 | */ |
| 130 | int vm_swappiness = 60; |
Zhang Yanfei | b21e0b9 | 2013-02-22 16:35:48 -0800 | [diff] [blame] | 131 | unsigned 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] | 132 | |
| 133 | static LIST_HEAD(shrinker_list); |
| 134 | static DECLARE_RWSEM(shrinker_rwsem); |
| 135 | |
Andrew Morton | c255a45 | 2012-07-31 16:43:02 -0700 | [diff] [blame] | 136 | #ifdef CONFIG_MEMCG |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 137 | static bool global_reclaim(struct scan_control *sc) |
| 138 | { |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 139 | return !sc->target_mem_cgroup; |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 140 | } |
KAMEZAWA Hiroyuki | 91a4547 | 2008-02-07 00:14:29 -0800 | [diff] [blame] | 141 | #else |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 142 | static bool global_reclaim(struct scan_control *sc) |
| 143 | { |
| 144 | return true; |
| 145 | } |
KAMEZAWA Hiroyuki | 91a4547 | 2008-02-07 00:14:29 -0800 | [diff] [blame] | 146 | #endif |
| 147 | |
Hugh Dickins | 4d7dcca | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 148 | static unsigned long get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
KOSAKI Motohiro | c9f299d | 2009-01-07 18:08:16 -0800 | [diff] [blame] | 149 | { |
Hugh Dickins | c3c787e | 2012-05-29 15:06:52 -0700 | [diff] [blame] | 150 | if (!mem_cgroup_disabled()) |
Hugh Dickins | 4d7dcca | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 151 | return mem_cgroup_get_lru_size(lruvec, lru); |
KOSAKI Motohiro | a3d8e05 | 2009-01-07 18:08:19 -0800 | [diff] [blame] | 152 | |
Konstantin Khlebnikov | 074291f | 2012-05-29 15:07:00 -0700 | [diff] [blame] | 153 | return zone_page_state(lruvec_zone(lruvec), NR_LRU_BASE + lru); |
KOSAKI Motohiro | c9f299d | 2009-01-07 18:08:16 -0800 | [diff] [blame] | 154 | } |
| 155 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 156 | /* |
| 157 | * Add a shrinker callback to be called from the vm |
| 158 | */ |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 159 | void register_shrinker(struct shrinker *shrinker) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 160 | { |
Konstantin Khlebnikov | 83aeead | 2011-12-08 14:33:54 -0800 | [diff] [blame] | 161 | atomic_long_set(&shrinker->nr_in_batch, 0); |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 162 | down_write(&shrinker_rwsem); |
| 163 | list_add_tail(&shrinker->list, &shrinker_list); |
| 164 | up_write(&shrinker_rwsem); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 165 | } |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 166 | EXPORT_SYMBOL(register_shrinker); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 167 | |
| 168 | /* |
| 169 | * Remove one |
| 170 | */ |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 171 | void unregister_shrinker(struct shrinker *shrinker) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 172 | { |
| 173 | down_write(&shrinker_rwsem); |
| 174 | list_del(&shrinker->list); |
| 175 | up_write(&shrinker_rwsem); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 176 | } |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 177 | EXPORT_SYMBOL(unregister_shrinker); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 178 | |
Ying Han | 1495f23 | 2011-05-24 17:12:27 -0700 | [diff] [blame] | 179 | static inline int do_shrinker_shrink(struct shrinker *shrinker, |
| 180 | struct shrink_control *sc, |
| 181 | unsigned long nr_to_scan) |
| 182 | { |
| 183 | sc->nr_to_scan = nr_to_scan; |
| 184 | return (*shrinker->shrink)(shrinker, sc); |
| 185 | } |
| 186 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 187 | #define SHRINK_BATCH 128 |
| 188 | /* |
| 189 | * Call the shrink functions to age shrinkable caches |
| 190 | * |
| 191 | * Here we assume it costs one seek to replace a lru page and that it also |
| 192 | * takes a seek to recreate a cache object. With this in mind we age equal |
| 193 | * percentages of the lru and ageable caches. This should balance the seeks |
| 194 | * generated by these structures. |
| 195 | * |
Simon Arlott | 183ff22 | 2007-10-20 01:27:18 +0200 | [diff] [blame] | 196 | * 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] | 197 | * slab to avoid swapping. |
| 198 | * |
| 199 | * We do weird things to avoid (scanned*seeks*entries) overflowing 32 bits. |
| 200 | * |
| 201 | * `lru_pages' represents the number of on-LRU pages in all the zones which |
| 202 | * are eligible for the caller's allocation attempt. It is used for balancing |
| 203 | * slab reclaim versus page reclaim. |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 204 | * |
| 205 | * Returns the number of slab objects which we shrunk. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 206 | */ |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 207 | unsigned long shrink_slab(struct shrink_control *shrink, |
Ying Han | 1495f23 | 2011-05-24 17:12:27 -0700 | [diff] [blame] | 208 | unsigned long nr_pages_scanned, |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 209 | unsigned long lru_pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 210 | { |
| 211 | struct shrinker *shrinker; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 212 | unsigned long ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 213 | |
Ying Han | 1495f23 | 2011-05-24 17:12:27 -0700 | [diff] [blame] | 214 | if (nr_pages_scanned == 0) |
| 215 | nr_pages_scanned = SWAP_CLUSTER_MAX; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 216 | |
Minchan Kim | f06590b | 2011-05-24 17:11:11 -0700 | [diff] [blame] | 217 | if (!down_read_trylock(&shrinker_rwsem)) { |
| 218 | /* Assume we'll be able to shrink next time */ |
| 219 | ret = 1; |
| 220 | goto out; |
| 221 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 222 | |
| 223 | list_for_each_entry(shrinker, &shrinker_list, list) { |
| 224 | unsigned long long delta; |
Konstantin Khlebnikov | 635697c | 2011-12-08 14:33:51 -0800 | [diff] [blame] | 225 | long total_scan; |
| 226 | long max_pass; |
Dave Chinner | 0957607 | 2011-07-08 14:14:34 +1000 | [diff] [blame] | 227 | int shrink_ret = 0; |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 228 | long nr; |
| 229 | long new_nr; |
Dave Chinner | e9299f5 | 2011-07-08 14:14:37 +1000 | [diff] [blame] | 230 | long batch_size = shrinker->batch ? shrinker->batch |
| 231 | : SHRINK_BATCH; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 232 | |
Konstantin Khlebnikov | 635697c | 2011-12-08 14:33:51 -0800 | [diff] [blame] | 233 | max_pass = do_shrinker_shrink(shrinker, shrink, 0); |
| 234 | if (max_pass <= 0) |
| 235 | continue; |
| 236 | |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 237 | /* |
| 238 | * copy the current shrinker scan count into a local variable |
| 239 | * and zero it so that other concurrent shrinker invocations |
| 240 | * don't also do this scanning work. |
| 241 | */ |
Konstantin Khlebnikov | 83aeead | 2011-12-08 14:33:54 -0800 | [diff] [blame] | 242 | nr = atomic_long_xchg(&shrinker->nr_in_batch, 0); |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 243 | |
| 244 | total_scan = nr; |
Ying Han | 1495f23 | 2011-05-24 17:12:27 -0700 | [diff] [blame] | 245 | delta = (4 * nr_pages_scanned) / shrinker->seeks; |
Andrea Arcangeli | ea164d7 | 2005-11-28 13:44:15 -0800 | [diff] [blame] | 246 | delta *= max_pass; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 247 | do_div(delta, lru_pages + 1); |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 248 | total_scan += delta; |
| 249 | if (total_scan < 0) { |
David Rientjes | 88c3bd7 | 2009-03-31 15:23:29 -0700 | [diff] [blame] | 250 | printk(KERN_ERR "shrink_slab: %pF negative objects to " |
| 251 | "delete nr=%ld\n", |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 252 | shrinker->shrink, total_scan); |
| 253 | total_scan = max_pass; |
Andrea Arcangeli | ea164d7 | 2005-11-28 13:44:15 -0800 | [diff] [blame] | 254 | } |
| 255 | |
| 256 | /* |
Dave Chinner | 3567b59 | 2011-07-08 14:14:36 +1000 | [diff] [blame] | 257 | * We need to avoid excessive windup on filesystem shrinkers |
| 258 | * due to large numbers of GFP_NOFS allocations causing the |
| 259 | * shrinkers to return -1 all the time. This results in a large |
| 260 | * nr being built up so when a shrink that can do some work |
| 261 | * comes along it empties the entire cache due to nr >>> |
| 262 | * max_pass. This is bad for sustaining a working set in |
| 263 | * memory. |
| 264 | * |
| 265 | * Hence only allow the shrinker to scan the entire cache when |
| 266 | * a large delta change is calculated directly. |
| 267 | */ |
| 268 | if (delta < max_pass / 4) |
| 269 | total_scan = min(total_scan, max_pass / 2); |
| 270 | |
| 271 | /* |
Andrea Arcangeli | ea164d7 | 2005-11-28 13:44:15 -0800 | [diff] [blame] | 272 | * Avoid risking looping forever due to too large nr value: |
| 273 | * never try to free more than twice the estimate number of |
| 274 | * freeable entries. |
| 275 | */ |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 276 | if (total_scan > max_pass * 2) |
| 277 | total_scan = max_pass * 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 278 | |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 279 | trace_mm_shrink_slab_start(shrinker, shrink, nr, |
Dave Chinner | 0957607 | 2011-07-08 14:14:34 +1000 | [diff] [blame] | 280 | nr_pages_scanned, lru_pages, |
| 281 | max_pass, delta, total_scan); |
| 282 | |
Dave Chinner | e9299f5 | 2011-07-08 14:14:37 +1000 | [diff] [blame] | 283 | while (total_scan >= batch_size) { |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 284 | int nr_before; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 285 | |
Ying Han | 1495f23 | 2011-05-24 17:12:27 -0700 | [diff] [blame] | 286 | nr_before = do_shrinker_shrink(shrinker, shrink, 0); |
| 287 | shrink_ret = do_shrinker_shrink(shrinker, shrink, |
Dave Chinner | e9299f5 | 2011-07-08 14:14:37 +1000 | [diff] [blame] | 288 | batch_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 289 | if (shrink_ret == -1) |
| 290 | break; |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 291 | if (shrink_ret < nr_before) |
| 292 | ret += nr_before - shrink_ret; |
Dave Chinner | e9299f5 | 2011-07-08 14:14:37 +1000 | [diff] [blame] | 293 | count_vm_events(SLABS_SCANNED, batch_size); |
| 294 | total_scan -= batch_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 295 | |
| 296 | cond_resched(); |
| 297 | } |
| 298 | |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 299 | /* |
| 300 | * move the unused scan count back into the shrinker in a |
| 301 | * manner that handles concurrent updates. If we exhausted the |
| 302 | * scan, there is no need to do an update. |
| 303 | */ |
Konstantin Khlebnikov | 83aeead | 2011-12-08 14:33:54 -0800 | [diff] [blame] | 304 | if (total_scan > 0) |
| 305 | new_nr = atomic_long_add_return(total_scan, |
| 306 | &shrinker->nr_in_batch); |
| 307 | else |
| 308 | new_nr = atomic_long_read(&shrinker->nr_in_batch); |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 309 | |
| 310 | trace_mm_shrink_slab_end(shrinker, shrink_ret, nr, new_nr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 311 | } |
| 312 | up_read(&shrinker_rwsem); |
Minchan Kim | f06590b | 2011-05-24 17:11:11 -0700 | [diff] [blame] | 313 | out: |
| 314 | cond_resched(); |
akpm@osdl.org | b15e090 | 2005-06-21 17:14:35 -0700 | [diff] [blame] | 315 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 316 | } |
| 317 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 318 | static inline int is_page_cache_freeable(struct page *page) |
| 319 | { |
Johannes Weiner | ceddc3a | 2009-09-21 17:03:00 -0700 | [diff] [blame] | 320 | /* |
| 321 | * A freeable page cache page is referenced only by the caller |
| 322 | * that isolated the page, the page cache radix tree and |
| 323 | * optional buffer heads at page->private. |
| 324 | */ |
Johannes Weiner | edcf474 | 2009-09-21 17:02:59 -0700 | [diff] [blame] | 325 | return page_count(page) - page_has_private(page) == 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 326 | } |
| 327 | |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 328 | static int may_write_to_queue(struct backing_dev_info *bdi, |
| 329 | struct scan_control *sc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 330 | { |
Christoph Lameter | 930d915 | 2006-01-08 01:00:47 -0800 | [diff] [blame] | 331 | if (current->flags & PF_SWAPWRITE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 332 | return 1; |
| 333 | if (!bdi_write_congested(bdi)) |
| 334 | return 1; |
| 335 | if (bdi == current->backing_dev_info) |
| 336 | return 1; |
| 337 | return 0; |
| 338 | } |
| 339 | |
| 340 | /* |
| 341 | * We detected a synchronous write error writing a page out. Probably |
| 342 | * -ENOSPC. We need to propagate that into the address_space for a subsequent |
| 343 | * fsync(), msync() or close(). |
| 344 | * |
| 345 | * The tricky part is that after writepage we cannot touch the mapping: nothing |
| 346 | * prevents it from being freed up. But we have a ref on the page and once |
| 347 | * that page is locked, the mapping is pinned. |
| 348 | * |
| 349 | * We're allowed to run sleeping lock_page() here because we know the caller has |
| 350 | * __GFP_FS. |
| 351 | */ |
| 352 | static void handle_write_error(struct address_space *mapping, |
| 353 | struct page *page, int error) |
| 354 | { |
Jens Axboe | 7eaceac | 2011-03-10 08:52:07 +0100 | [diff] [blame] | 355 | lock_page(page); |
Guillaume Chazarain | 3e9f45b | 2007-05-08 00:23:25 -0700 | [diff] [blame] | 356 | if (page_mapping(page) == mapping) |
| 357 | mapping_set_error(mapping, error); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 358 | unlock_page(page); |
| 359 | } |
| 360 | |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 361 | /* possible outcome of pageout() */ |
| 362 | typedef enum { |
| 363 | /* failed to write page out, page is locked */ |
| 364 | PAGE_KEEP, |
| 365 | /* move page to the active list, page is locked */ |
| 366 | PAGE_ACTIVATE, |
| 367 | /* page has been sent to the disk successfully, page is unlocked */ |
| 368 | PAGE_SUCCESS, |
| 369 | /* page is clean and locked */ |
| 370 | PAGE_CLEAN, |
| 371 | } pageout_t; |
| 372 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 373 | /* |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 374 | * pageout is called by shrink_page_list() for each dirty page. |
| 375 | * Calls ->writepage(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 376 | */ |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 377 | static pageout_t pageout(struct page *page, struct address_space *mapping, |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 378 | struct scan_control *sc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 379 | { |
| 380 | /* |
| 381 | * If the page is dirty, only perform writeback if that write |
| 382 | * will be non-blocking. To prevent this allocation from being |
| 383 | * stalled by pagecache activity. But note that there may be |
| 384 | * stalls if we need to run get_block(). We could test |
| 385 | * PagePrivate for that. |
| 386 | * |
Vincent Li | 6aceb53 | 2009-12-14 17:58:49 -0800 | [diff] [blame] | 387 | * If this process is currently in __generic_file_aio_write() against |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 388 | * this page's queue, we can perform writeback even if that |
| 389 | * will block. |
| 390 | * |
| 391 | * If the page is swapcache, write it back even if that would |
| 392 | * block, for some throttling. This happens by accident, because |
| 393 | * swap_backing_dev_info is bust: it doesn't reflect the |
| 394 | * congestion state of the swapdevs. Easy to fix, if needed. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 395 | */ |
| 396 | if (!is_page_cache_freeable(page)) |
| 397 | return PAGE_KEEP; |
| 398 | if (!mapping) { |
| 399 | /* |
| 400 | * Some data journaling orphaned pages can have |
| 401 | * page->mapping == NULL while being dirty with clean buffers. |
| 402 | */ |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 403 | if (page_has_private(page)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 404 | if (try_to_free_buffers(page)) { |
| 405 | ClearPageDirty(page); |
Harvey Harrison | d40cee2 | 2008-04-30 00:55:07 -0700 | [diff] [blame] | 406 | printk("%s: orphaned page\n", __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 407 | return PAGE_CLEAN; |
| 408 | } |
| 409 | } |
| 410 | return PAGE_KEEP; |
| 411 | } |
| 412 | if (mapping->a_ops->writepage == NULL) |
| 413 | return PAGE_ACTIVATE; |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 414 | if (!may_write_to_queue(mapping->backing_dev_info, sc)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 415 | return PAGE_KEEP; |
| 416 | |
| 417 | if (clear_page_dirty_for_io(page)) { |
| 418 | int res; |
| 419 | struct writeback_control wbc = { |
| 420 | .sync_mode = WB_SYNC_NONE, |
| 421 | .nr_to_write = SWAP_CLUSTER_MAX, |
OGAWA Hirofumi | 111ebb6 | 2006-06-23 02:03:26 -0700 | [diff] [blame] | 422 | .range_start = 0, |
| 423 | .range_end = LLONG_MAX, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 424 | .for_reclaim = 1, |
| 425 | }; |
| 426 | |
| 427 | SetPageReclaim(page); |
| 428 | res = mapping->a_ops->writepage(page, &wbc); |
| 429 | if (res < 0) |
| 430 | handle_write_error(mapping, page, res); |
Zach Brown | 994fc28c | 2005-12-15 14:28:17 -0800 | [diff] [blame] | 431 | if (res == AOP_WRITEPAGE_ACTIVATE) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 432 | ClearPageReclaim(page); |
| 433 | return PAGE_ACTIVATE; |
| 434 | } |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 435 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 436 | if (!PageWriteback(page)) { |
| 437 | /* synchronous write or broken a_ops? */ |
| 438 | ClearPageReclaim(page); |
| 439 | } |
Mel Gorman | 23b9da5 | 2012-05-29 15:06:20 -0700 | [diff] [blame] | 440 | trace_mm_vmscan_writepage(page, trace_reclaim_flags(page)); |
Andrew Morton | e129b5c | 2006-09-27 01:50:00 -0700 | [diff] [blame] | 441 | inc_zone_page_state(page, NR_VMSCAN_WRITE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 442 | return PAGE_SUCCESS; |
| 443 | } |
| 444 | |
| 445 | return PAGE_CLEAN; |
| 446 | } |
| 447 | |
Andrew Morton | a649fd9 | 2006-10-17 00:09:36 -0700 | [diff] [blame] | 448 | /* |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 449 | * Same as remove_mapping, but if the page is removed from the mapping, it |
| 450 | * gets returned with a refcount of 0. |
Andrew Morton | a649fd9 | 2006-10-17 00:09:36 -0700 | [diff] [blame] | 451 | */ |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 452 | static int __remove_mapping(struct address_space *mapping, struct page *page) |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 453 | { |
Nick Piggin | 28e4d96 | 2006-09-25 23:31:23 -0700 | [diff] [blame] | 454 | BUG_ON(!PageLocked(page)); |
| 455 | BUG_ON(mapping != page_mapping(page)); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 456 | |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 457 | spin_lock_irq(&mapping->tree_lock); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 458 | /* |
Nick Piggin | 0fd0e6b | 2006-09-27 01:50:02 -0700 | [diff] [blame] | 459 | * The non racy check for a busy page. |
| 460 | * |
| 461 | * Must be careful with the order of the tests. When someone has |
| 462 | * a ref to the page, it may be possible that they dirty it then |
| 463 | * drop the reference. So if PageDirty is tested before page_count |
| 464 | * here, then the following race may occur: |
| 465 | * |
| 466 | * get_user_pages(&page); |
| 467 | * [user mapping goes away] |
| 468 | * write_to(page); |
| 469 | * !PageDirty(page) [good] |
| 470 | * SetPageDirty(page); |
| 471 | * put_page(page); |
| 472 | * !page_count(page) [good, discard it] |
| 473 | * |
| 474 | * [oops, our write_to data is lost] |
| 475 | * |
| 476 | * Reversing the order of the tests ensures such a situation cannot |
| 477 | * escape unnoticed. The smp_rmb is needed to ensure the page->flags |
| 478 | * load is not satisfied before that of page->_count. |
| 479 | * |
| 480 | * Note that if SetPageDirty is always performed via set_page_dirty, |
| 481 | * and thus under tree_lock, then this ordering is not required. |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 482 | */ |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 483 | if (!page_freeze_refs(page, 2)) |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 484 | goto cannot_free; |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 485 | /* note: atomic_cmpxchg in page_freeze_refs provides the smp_rmb */ |
| 486 | if (unlikely(PageDirty(page))) { |
| 487 | page_unfreeze_refs(page, 2); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 488 | goto cannot_free; |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 489 | } |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 490 | |
| 491 | if (PageSwapCache(page)) { |
| 492 | swp_entry_t swap = { .val = page_private(page) }; |
| 493 | __delete_from_swap_cache(page); |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 494 | spin_unlock_irq(&mapping->tree_lock); |
KAMEZAWA Hiroyuki | cb4b86b | 2009-06-16 15:32:52 -0700 | [diff] [blame] | 495 | swapcache_free(swap, page); |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 496 | } else { |
Linus Torvalds | 6072d13 | 2010-12-01 13:35:19 -0500 | [diff] [blame] | 497 | void (*freepage)(struct page *); |
| 498 | |
| 499 | freepage = mapping->a_ops->freepage; |
| 500 | |
Minchan Kim | e64a782 | 2011-03-22 16:32:44 -0700 | [diff] [blame] | 501 | __delete_from_page_cache(page); |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 502 | spin_unlock_irq(&mapping->tree_lock); |
Daisuke Nishimura | e767e05 | 2009-05-28 14:34:28 -0700 | [diff] [blame] | 503 | mem_cgroup_uncharge_cache_page(page); |
Linus Torvalds | 6072d13 | 2010-12-01 13:35:19 -0500 | [diff] [blame] | 504 | |
| 505 | if (freepage != NULL) |
| 506 | freepage(page); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 507 | } |
| 508 | |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 509 | return 1; |
| 510 | |
| 511 | cannot_free: |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 512 | spin_unlock_irq(&mapping->tree_lock); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 513 | return 0; |
| 514 | } |
| 515 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 516 | /* |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 517 | * Attempt to detach a locked page from its ->mapping. If it is dirty or if |
| 518 | * someone else has a ref on the page, abort and return 0. If it was |
| 519 | * successfully detached, return 1. Assumes the caller has a single ref on |
| 520 | * this page. |
| 521 | */ |
| 522 | int remove_mapping(struct address_space *mapping, struct page *page) |
| 523 | { |
| 524 | if (__remove_mapping(mapping, page)) { |
| 525 | /* |
| 526 | * Unfreezing the refcount with 1 rather than 2 effectively |
| 527 | * drops the pagecache ref for us without requiring another |
| 528 | * atomic operation. |
| 529 | */ |
| 530 | page_unfreeze_refs(page, 1); |
| 531 | return 1; |
| 532 | } |
| 533 | return 0; |
| 534 | } |
| 535 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 536 | /** |
| 537 | * putback_lru_page - put previously isolated page onto appropriate LRU list |
| 538 | * @page: page to be put back to appropriate lru list |
| 539 | * |
| 540 | * Add previously isolated @page to appropriate LRU list. |
| 541 | * Page may still be unevictable for other reasons. |
| 542 | * |
| 543 | * lru_lock must not be held, interrupts must be enabled. |
| 544 | */ |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 545 | void putback_lru_page(struct page *page) |
| 546 | { |
| 547 | int lru; |
| 548 | int active = !!TestClearPageActive(page); |
Lee Schermerhorn | bbfd28e | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 549 | int was_unevictable = PageUnevictable(page); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 550 | |
| 551 | VM_BUG_ON(PageLRU(page)); |
| 552 | |
| 553 | redo: |
| 554 | ClearPageUnevictable(page); |
| 555 | |
Hugh Dickins | 39b5f29 | 2012-10-08 16:33:18 -0700 | [diff] [blame] | 556 | if (page_evictable(page)) { |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 557 | /* |
| 558 | * For evictable pages, we can use the cache. |
| 559 | * In event of a race, worst case is we end up with an |
| 560 | * unevictable page on [in]active list. |
| 561 | * We know how to handle that. |
| 562 | */ |
Johannes Weiner | 401a8e1 | 2009-09-21 17:02:58 -0700 | [diff] [blame] | 563 | lru = active + page_lru_base_type(page); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 564 | lru_cache_add_lru(page, lru); |
| 565 | } else { |
| 566 | /* |
| 567 | * Put unevictable pages directly on zone's unevictable |
| 568 | * list. |
| 569 | */ |
| 570 | lru = LRU_UNEVICTABLE; |
| 571 | add_page_to_unevictable_list(page); |
Johannes Weiner | 6a7b954 | 2009-10-26 16:50:00 -0700 | [diff] [blame] | 572 | /* |
Minchan Kim | 21ee9f3 | 2011-10-31 17:09:28 -0700 | [diff] [blame] | 573 | * When racing with an mlock or AS_UNEVICTABLE clearing |
| 574 | * (page is unlocked) make sure that if the other thread |
| 575 | * does not observe our setting of PG_lru and fails |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 576 | * isolation/check_move_unevictable_pages, |
Minchan Kim | 21ee9f3 | 2011-10-31 17:09:28 -0700 | [diff] [blame] | 577 | * we see PG_mlocked/AS_UNEVICTABLE cleared below and move |
Johannes Weiner | 6a7b954 | 2009-10-26 16:50:00 -0700 | [diff] [blame] | 578 | * the page back to the evictable list. |
| 579 | * |
Minchan Kim | 21ee9f3 | 2011-10-31 17:09:28 -0700 | [diff] [blame] | 580 | * The other side is TestClearPageMlocked() or shmem_lock(). |
Johannes Weiner | 6a7b954 | 2009-10-26 16:50:00 -0700 | [diff] [blame] | 581 | */ |
| 582 | smp_mb(); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 583 | } |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 584 | |
| 585 | /* |
| 586 | * page's status can change while we move it among lru. If an evictable |
| 587 | * page is on unevictable list, it never be freed. To avoid that, |
| 588 | * check after we added it to the list, again. |
| 589 | */ |
Hugh Dickins | 39b5f29 | 2012-10-08 16:33:18 -0700 | [diff] [blame] | 590 | if (lru == LRU_UNEVICTABLE && page_evictable(page)) { |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 591 | if (!isolate_lru_page(page)) { |
| 592 | put_page(page); |
| 593 | goto redo; |
| 594 | } |
| 595 | /* This means someone else dropped this page from LRU |
| 596 | * So, it will be freed or putback to LRU again. There is |
| 597 | * nothing to do here. |
| 598 | */ |
| 599 | } |
| 600 | |
Lee Schermerhorn | bbfd28e | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 601 | if (was_unevictable && lru != LRU_UNEVICTABLE) |
| 602 | count_vm_event(UNEVICTABLE_PGRESCUED); |
| 603 | else if (!was_unevictable && lru == LRU_UNEVICTABLE) |
| 604 | count_vm_event(UNEVICTABLE_PGCULLED); |
| 605 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 606 | put_page(page); /* drop ref from isolate */ |
| 607 | } |
| 608 | |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 609 | enum page_references { |
| 610 | PAGEREF_RECLAIM, |
| 611 | PAGEREF_RECLAIM_CLEAN, |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 612 | PAGEREF_KEEP, |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 613 | PAGEREF_ACTIVATE, |
| 614 | }; |
| 615 | |
| 616 | static enum page_references page_check_references(struct page *page, |
| 617 | struct scan_control *sc) |
| 618 | { |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 619 | int referenced_ptes, referenced_page; |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 620 | unsigned long vm_flags; |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 621 | |
Johannes Weiner | c3ac9a8 | 2012-05-29 15:06:25 -0700 | [diff] [blame] | 622 | referenced_ptes = page_referenced(page, 1, sc->target_mem_cgroup, |
| 623 | &vm_flags); |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 624 | referenced_page = TestClearPageReferenced(page); |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 625 | |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 626 | /* |
| 627 | * Mlock lost the isolation race with us. Let try_to_unmap() |
| 628 | * move the page to the unevictable list. |
| 629 | */ |
| 630 | if (vm_flags & VM_LOCKED) |
| 631 | return PAGEREF_RECLAIM; |
| 632 | |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 633 | if (referenced_ptes) { |
Michal Hocko | e4898273 | 2012-05-29 15:06:45 -0700 | [diff] [blame] | 634 | if (PageSwapBacked(page)) |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 635 | return PAGEREF_ACTIVATE; |
| 636 | /* |
| 637 | * All mapped pages start out with page table |
| 638 | * references from the instantiating fault, so we need |
| 639 | * to look twice if a mapped file page is used more |
| 640 | * than once. |
| 641 | * |
| 642 | * Mark it and spare it for another trip around the |
| 643 | * inactive list. Another page table reference will |
| 644 | * lead to its activation. |
| 645 | * |
| 646 | * Note: the mark is set for activated pages as well |
| 647 | * so that recently deactivated but used pages are |
| 648 | * quickly recovered. |
| 649 | */ |
| 650 | SetPageReferenced(page); |
| 651 | |
Konstantin Khlebnikov | 34dbc67 | 2012-01-10 15:06:59 -0800 | [diff] [blame] | 652 | if (referenced_page || referenced_ptes > 1) |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 653 | return PAGEREF_ACTIVATE; |
| 654 | |
Konstantin Khlebnikov | c909e99 | 2012-01-10 15:07:03 -0800 | [diff] [blame] | 655 | /* |
| 656 | * Activate file-backed executable pages after first usage. |
| 657 | */ |
| 658 | if (vm_flags & VM_EXEC) |
| 659 | return PAGEREF_ACTIVATE; |
| 660 | |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 661 | return PAGEREF_KEEP; |
| 662 | } |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 663 | |
| 664 | /* Reclaim if clean, defer dirty pages to writeback */ |
KOSAKI Motohiro | 2e30244 | 2010-10-26 14:21:46 -0700 | [diff] [blame] | 665 | if (referenced_page && !PageSwapBacked(page)) |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 666 | return PAGEREF_RECLAIM_CLEAN; |
| 667 | |
| 668 | return PAGEREF_RECLAIM; |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 669 | } |
| 670 | |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 671 | /* |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 672 | * shrink_page_list() returns the number of reclaimed pages |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 673 | */ |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 674 | static unsigned long shrink_page_list(struct list_head *page_list, |
Konstantin Khlebnikov | 6a18adb | 2012-05-29 15:06:59 -0700 | [diff] [blame] | 675 | struct zone *zone, |
Mel Gorman | f84f6e2 | 2011-10-31 17:07:51 -0700 | [diff] [blame] | 676 | struct scan_control *sc, |
Minchan Kim | 02c6de8 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 677 | enum ttu_flags ttu_flags, |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 678 | unsigned long *ret_nr_dirty, |
Minchan Kim | 02c6de8 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 679 | unsigned long *ret_nr_writeback, |
| 680 | bool force_reclaim) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 681 | { |
| 682 | LIST_HEAD(ret_pages); |
Mel Gorman | abe4c3b | 2010-08-09 17:19:31 -0700 | [diff] [blame] | 683 | LIST_HEAD(free_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 684 | int pgactivate = 0; |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 685 | unsigned long nr_dirty = 0; |
| 686 | unsigned long nr_congested = 0; |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 687 | unsigned long nr_reclaimed = 0; |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 688 | unsigned long nr_writeback = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 689 | |
| 690 | cond_resched(); |
| 691 | |
Tim Chen | 69980e3 | 2012-07-31 16:46:08 -0700 | [diff] [blame] | 692 | mem_cgroup_uncharge_start(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 693 | while (!list_empty(page_list)) { |
| 694 | struct address_space *mapping; |
| 695 | struct page *page; |
| 696 | int may_enter_fs; |
Minchan Kim | 02c6de8 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 697 | enum page_references references = PAGEREF_RECLAIM_CLEAN; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 698 | |
| 699 | cond_resched(); |
| 700 | |
| 701 | page = lru_to_page(page_list); |
| 702 | list_del(&page->lru); |
| 703 | |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 704 | if (!trylock_page(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 705 | goto keep; |
| 706 | |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 707 | VM_BUG_ON(PageActive(page)); |
Konstantin Khlebnikov | 6a18adb | 2012-05-29 15:06:59 -0700 | [diff] [blame] | 708 | VM_BUG_ON(page_zone(page) != zone); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 709 | |
| 710 | sc->nr_scanned++; |
Christoph Lameter | 80e4342 | 2006-02-11 17:55:53 -0800 | [diff] [blame] | 711 | |
Hugh Dickins | 39b5f29 | 2012-10-08 16:33:18 -0700 | [diff] [blame] | 712 | if (unlikely(!page_evictable(page))) |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 713 | goto cull_mlocked; |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 714 | |
Johannes Weiner | a6dc60f8 | 2009-03-31 15:19:30 -0700 | [diff] [blame] | 715 | if (!sc->may_unmap && page_mapped(page)) |
Christoph Lameter | 80e4342 | 2006-02-11 17:55:53 -0800 | [diff] [blame] | 716 | goto keep_locked; |
| 717 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 718 | /* Double the slab pressure for mapped and swapcache pages */ |
| 719 | if (page_mapped(page) || PageSwapCache(page)) |
| 720 | sc->nr_scanned++; |
| 721 | |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 722 | may_enter_fs = (sc->gfp_mask & __GFP_FS) || |
| 723 | (PageSwapCache(page) && (sc->gfp_mask & __GFP_IO)); |
| 724 | |
| 725 | if (PageWriteback(page)) { |
Michal Hocko | e62e384 | 2012-07-31 16:45:55 -0700 | [diff] [blame] | 726 | /* |
| 727 | * memcg doesn't have any dirty pages throttling so we |
| 728 | * could easily OOM just because too many pages are in |
Hugh Dickins | c3b94f4 | 2012-07-31 16:45:59 -0700 | [diff] [blame] | 729 | * writeback and there is nothing else to reclaim. |
Michal Hocko | e62e384 | 2012-07-31 16:45:55 -0700 | [diff] [blame] | 730 | * |
Hugh Dickins | c3b94f4 | 2012-07-31 16:45:59 -0700 | [diff] [blame] | 731 | * Check __GFP_IO, certainly because a loop driver |
Michal Hocko | e62e384 | 2012-07-31 16:45:55 -0700 | [diff] [blame] | 732 | * thread might enter reclaim, and deadlock if it waits |
| 733 | * on a page for which it is needed to do the write |
| 734 | * (loop masks off __GFP_IO|__GFP_FS for this reason); |
| 735 | * but more thought would probably show more reasons. |
Hugh Dickins | c3b94f4 | 2012-07-31 16:45:59 -0700 | [diff] [blame] | 736 | * |
| 737 | * Don't require __GFP_FS, since we're not going into |
| 738 | * the FS, just waiting on its writeback completion. |
| 739 | * Worryingly, ext4 gfs2 and xfs allocate pages with |
| 740 | * grab_cache_page_write_begin(,,AOP_FLAG_NOFS), so |
| 741 | * testing may_enter_fs here is liable to OOM on them. |
Michal Hocko | e62e384 | 2012-07-31 16:45:55 -0700 | [diff] [blame] | 742 | */ |
Hugh Dickins | c3b94f4 | 2012-07-31 16:45:59 -0700 | [diff] [blame] | 743 | if (global_reclaim(sc) || |
| 744 | !PageReclaim(page) || !(sc->gfp_mask & __GFP_IO)) { |
| 745 | /* |
| 746 | * This is slightly racy - end_page_writeback() |
| 747 | * might have just cleared PageReclaim, then |
| 748 | * setting PageReclaim here end up interpreted |
| 749 | * as PageReadahead - but that does not matter |
| 750 | * enough to care. What we do want is for this |
| 751 | * page to have PageReclaim set next time memcg |
| 752 | * reclaim reaches the tests above, so it will |
| 753 | * then wait_on_page_writeback() to avoid OOM; |
| 754 | * and it's also appropriate in global reclaim. |
| 755 | */ |
| 756 | SetPageReclaim(page); |
Michal Hocko | e62e384 | 2012-07-31 16:45:55 -0700 | [diff] [blame] | 757 | nr_writeback++; |
Hugh Dickins | c3b94f4 | 2012-07-31 16:45:59 -0700 | [diff] [blame] | 758 | goto keep_locked; |
Michal Hocko | e62e384 | 2012-07-31 16:45:55 -0700 | [diff] [blame] | 759 | } |
Hugh Dickins | c3b94f4 | 2012-07-31 16:45:59 -0700 | [diff] [blame] | 760 | wait_on_page_writeback(page); |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 761 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 762 | |
Minchan Kim | 02c6de8 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 763 | if (!force_reclaim) |
| 764 | references = page_check_references(page, sc); |
| 765 | |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 766 | switch (references) { |
| 767 | case PAGEREF_ACTIVATE: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 768 | goto activate_locked; |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 769 | case PAGEREF_KEEP: |
| 770 | goto keep_locked; |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 771 | case PAGEREF_RECLAIM: |
| 772 | case PAGEREF_RECLAIM_CLEAN: |
| 773 | ; /* try to reclaim the page below */ |
| 774 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 775 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 776 | /* |
| 777 | * Anonymous process memory has backing store? |
| 778 | * Try to allocate it some swap space here. |
| 779 | */ |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 780 | if (PageAnon(page) && !PageSwapCache(page)) { |
Hugh Dickins | 63eb6b9 | 2008-11-19 15:36:37 -0800 | [diff] [blame] | 781 | if (!(sc->gfp_mask & __GFP_IO)) |
| 782 | goto keep_locked; |
Hugh Dickins | ac47b00 | 2009-01-06 14:39:39 -0800 | [diff] [blame] | 783 | if (!add_to_swap(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 784 | goto activate_locked; |
Hugh Dickins | 63eb6b9 | 2008-11-19 15:36:37 -0800 | [diff] [blame] | 785 | may_enter_fs = 1; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 786 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 787 | |
| 788 | mapping = page_mapping(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 789 | |
| 790 | /* |
| 791 | * The page is mapped into the page tables of one or more |
| 792 | * processes. Try to unmap it here. |
| 793 | */ |
| 794 | if (page_mapped(page) && mapping) { |
Minchan Kim | 02c6de8 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 795 | switch (try_to_unmap(page, ttu_flags)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 796 | case SWAP_FAIL: |
| 797 | goto activate_locked; |
| 798 | case SWAP_AGAIN: |
| 799 | goto keep_locked; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 800 | case SWAP_MLOCK: |
| 801 | goto cull_mlocked; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 802 | case SWAP_SUCCESS: |
| 803 | ; /* try to free the page below */ |
| 804 | } |
| 805 | } |
| 806 | |
| 807 | if (PageDirty(page)) { |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 808 | nr_dirty++; |
| 809 | |
Mel Gorman | ee72886 | 2011-10-31 17:07:38 -0700 | [diff] [blame] | 810 | /* |
| 811 | * Only kswapd can writeback filesystem pages to |
Mel Gorman | f84f6e2 | 2011-10-31 17:07:51 -0700 | [diff] [blame] | 812 | * avoid risk of stack overflow but do not writeback |
| 813 | * unless under significant pressure. |
Mel Gorman | ee72886 | 2011-10-31 17:07:38 -0700 | [diff] [blame] | 814 | */ |
Mel Gorman | f84f6e2 | 2011-10-31 17:07:51 -0700 | [diff] [blame] | 815 | if (page_is_file_cache(page) && |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 816 | (!current_is_kswapd() || |
| 817 | sc->priority >= DEF_PRIORITY - 2)) { |
Mel Gorman | 49ea7eb | 2011-10-31 17:07:59 -0700 | [diff] [blame] | 818 | /* |
| 819 | * Immediately reclaim when written back. |
| 820 | * Similar in principal to deactivate_page() |
| 821 | * except we already have the page isolated |
| 822 | * and know it's dirty |
| 823 | */ |
| 824 | inc_zone_page_state(page, NR_VMSCAN_IMMEDIATE); |
| 825 | SetPageReclaim(page); |
| 826 | |
Mel Gorman | ee72886 | 2011-10-31 17:07:38 -0700 | [diff] [blame] | 827 | goto keep_locked; |
| 828 | } |
| 829 | |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 830 | if (references == PAGEREF_RECLAIM_CLEAN) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 831 | goto keep_locked; |
Andrew Morton | 4dd4b92 | 2008-03-24 12:29:52 -0700 | [diff] [blame] | 832 | if (!may_enter_fs) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 833 | goto keep_locked; |
Christoph Lameter | 52a8363 | 2006-02-01 03:05:28 -0800 | [diff] [blame] | 834 | if (!sc->may_writepage) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 835 | goto keep_locked; |
| 836 | |
| 837 | /* Page is dirty, try to write it out here */ |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 838 | switch (pageout(page, mapping, sc)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 839 | case PAGE_KEEP: |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 840 | nr_congested++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 841 | goto keep_locked; |
| 842 | case PAGE_ACTIVATE: |
| 843 | goto activate_locked; |
| 844 | case PAGE_SUCCESS: |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 845 | if (PageWriteback(page)) |
Mel Gorman | 41ac199 | 2012-05-29 15:06:19 -0700 | [diff] [blame] | 846 | goto keep; |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 847 | if (PageDirty(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 848 | goto keep; |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 849 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 850 | /* |
| 851 | * A synchronous write - probably a ramdisk. Go |
| 852 | * ahead and try to reclaim the page. |
| 853 | */ |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 854 | if (!trylock_page(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 855 | goto keep; |
| 856 | if (PageDirty(page) || PageWriteback(page)) |
| 857 | goto keep_locked; |
| 858 | mapping = page_mapping(page); |
| 859 | case PAGE_CLEAN: |
| 860 | ; /* try to free the page below */ |
| 861 | } |
| 862 | } |
| 863 | |
| 864 | /* |
| 865 | * If the page has buffers, try to free the buffer mappings |
| 866 | * associated with this page. If we succeed we try to free |
| 867 | * the page as well. |
| 868 | * |
| 869 | * We do this even if the page is PageDirty(). |
| 870 | * try_to_release_page() does not perform I/O, but it is |
| 871 | * possible for a page to have PageDirty set, but it is actually |
| 872 | * clean (all its buffers are clean). This happens if the |
| 873 | * buffers were written out directly, with submit_bh(). ext3 |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 874 | * will do this, as well as the blockdev mapping. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 875 | * try_to_release_page() will discover that cleanness and will |
| 876 | * drop the buffers and mark the page clean - it can be freed. |
| 877 | * |
| 878 | * Rarely, pages can have buffers and no ->mapping. These are |
| 879 | * the pages which were not successfully invalidated in |
| 880 | * truncate_complete_page(). We try to drop those buffers here |
| 881 | * and if that worked, and the page is no longer mapped into |
| 882 | * process address space (page_count == 1) it can be freed. |
| 883 | * Otherwise, leave the page on the LRU so it is swappable. |
| 884 | */ |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 885 | if (page_has_private(page)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 886 | if (!try_to_release_page(page, sc->gfp_mask)) |
| 887 | goto activate_locked; |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 888 | if (!mapping && page_count(page) == 1) { |
| 889 | unlock_page(page); |
| 890 | if (put_page_testzero(page)) |
| 891 | goto free_it; |
| 892 | else { |
| 893 | /* |
| 894 | * rare race with speculative reference. |
| 895 | * the speculative reference will free |
| 896 | * this page shortly, so we may |
| 897 | * increment nr_reclaimed here (and |
| 898 | * leave it off the LRU). |
| 899 | */ |
| 900 | nr_reclaimed++; |
| 901 | continue; |
| 902 | } |
| 903 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 904 | } |
| 905 | |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 906 | if (!mapping || !__remove_mapping(mapping, page)) |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 907 | goto keep_locked; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 908 | |
Nick Piggin | a978d6f | 2008-10-18 20:26:58 -0700 | [diff] [blame] | 909 | /* |
| 910 | * At this point, we have no other references and there is |
| 911 | * no way to pick any more up (removed from LRU, removed |
| 912 | * from pagecache). Can use non-atomic bitops now (and |
| 913 | * we obviously don't have to worry about waking up a process |
| 914 | * waiting on the page lock, because there are no references. |
| 915 | */ |
| 916 | __clear_page_locked(page); |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 917 | free_it: |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 918 | nr_reclaimed++; |
Mel Gorman | abe4c3b | 2010-08-09 17:19:31 -0700 | [diff] [blame] | 919 | |
| 920 | /* |
| 921 | * Is there need to periodically free_page_list? It would |
| 922 | * appear not as the counts should be low |
| 923 | */ |
| 924 | list_add(&page->lru, &free_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 925 | continue; |
| 926 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 927 | cull_mlocked: |
Hugh Dickins | 63d6c5a | 2009-01-06 14:39:38 -0800 | [diff] [blame] | 928 | if (PageSwapCache(page)) |
| 929 | try_to_free_swap(page); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 930 | unlock_page(page); |
| 931 | putback_lru_page(page); |
| 932 | continue; |
| 933 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 934 | activate_locked: |
Rik van Riel | 68a22394 | 2008-10-18 20:26:23 -0700 | [diff] [blame] | 935 | /* Not a candidate for swapping, so reclaim swap space. */ |
| 936 | if (PageSwapCache(page) && vm_swap_full()) |
Hugh Dickins | a2c43ee | 2009-01-06 14:39:36 -0800 | [diff] [blame] | 937 | try_to_free_swap(page); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 938 | VM_BUG_ON(PageActive(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 939 | SetPageActive(page); |
| 940 | pgactivate++; |
| 941 | keep_locked: |
| 942 | unlock_page(page); |
| 943 | keep: |
| 944 | list_add(&page->lru, &ret_pages); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 945 | VM_BUG_ON(PageLRU(page) || PageUnevictable(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 946 | } |
Mel Gorman | abe4c3b | 2010-08-09 17:19:31 -0700 | [diff] [blame] | 947 | |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 948 | /* |
| 949 | * Tag a zone as congested if all the dirty pages encountered were |
| 950 | * backed by a congested BDI. In this case, reclaimers should just |
| 951 | * back off and wait for congestion to clear because further reclaim |
| 952 | * will encounter the same problem |
| 953 | */ |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 954 | if (nr_dirty && nr_dirty == nr_congested && global_reclaim(sc)) |
Konstantin Khlebnikov | 6a18adb | 2012-05-29 15:06:59 -0700 | [diff] [blame] | 955 | zone_set_flag(zone, ZONE_CONGESTED); |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 956 | |
Konstantin Khlebnikov | cc59850 | 2012-01-10 15:07:04 -0800 | [diff] [blame] | 957 | free_hot_cold_page_list(&free_pages, 1); |
Mel Gorman | abe4c3b | 2010-08-09 17:19:31 -0700 | [diff] [blame] | 958 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 959 | list_splice(&ret_pages, page_list); |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 960 | count_vm_events(PGACTIVATE, pgactivate); |
Tim Chen | 69980e3 | 2012-07-31 16:46:08 -0700 | [diff] [blame] | 961 | mem_cgroup_uncharge_end(); |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 962 | *ret_nr_dirty += nr_dirty; |
| 963 | *ret_nr_writeback += nr_writeback; |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 964 | return nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 965 | } |
| 966 | |
Minchan Kim | 02c6de8 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 967 | unsigned long reclaim_clean_pages_from_list(struct zone *zone, |
| 968 | struct list_head *page_list) |
| 969 | { |
| 970 | struct scan_control sc = { |
| 971 | .gfp_mask = GFP_KERNEL, |
| 972 | .priority = DEF_PRIORITY, |
| 973 | .may_unmap = 1, |
| 974 | }; |
| 975 | unsigned long ret, dummy1, dummy2; |
| 976 | struct page *page, *next; |
| 977 | LIST_HEAD(clean_pages); |
| 978 | |
| 979 | list_for_each_entry_safe(page, next, page_list, lru) { |
| 980 | if (page_is_file_cache(page) && !PageDirty(page)) { |
| 981 | ClearPageActive(page); |
| 982 | list_move(&page->lru, &clean_pages); |
| 983 | } |
| 984 | } |
| 985 | |
| 986 | ret = shrink_page_list(&clean_pages, zone, &sc, |
| 987 | TTU_UNMAP|TTU_IGNORE_ACCESS, |
| 988 | &dummy1, &dummy2, true); |
| 989 | list_splice(&clean_pages, page_list); |
| 990 | __mod_zone_page_state(zone, NR_ISOLATED_FILE, -ret); |
| 991 | return ret; |
| 992 | } |
| 993 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 994 | /* |
| 995 | * Attempt to remove the specified page from its LRU. Only take this page |
| 996 | * if it is of the appropriate PageActive status. Pages which are being |
| 997 | * freed elsewhere are also ignored. |
| 998 | * |
| 999 | * page: page to consider |
| 1000 | * mode: one of the LRU isolation modes defined above |
| 1001 | * |
| 1002 | * returns 0 on success, -ve errno on failure. |
| 1003 | */ |
Konstantin Khlebnikov | f3fd4a6 | 2012-05-29 15:06:54 -0700 | [diff] [blame] | 1004 | int __isolate_lru_page(struct page *page, isolate_mode_t mode) |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1005 | { |
| 1006 | int ret = -EINVAL; |
| 1007 | |
| 1008 | /* Only take pages on the LRU. */ |
| 1009 | if (!PageLRU(page)) |
| 1010 | return ret; |
| 1011 | |
Minchan Kim | e46a287 | 2012-10-08 16:33:48 -0700 | [diff] [blame] | 1012 | /* Compaction should not handle unevictable pages but CMA can do so */ |
| 1013 | if (PageUnevictable(page) && !(mode & ISOLATE_UNEVICTABLE)) |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1014 | return ret; |
| 1015 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1016 | ret = -EBUSY; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 1017 | |
Mel Gorman | c824493 | 2012-01-12 17:19:38 -0800 | [diff] [blame] | 1018 | /* |
| 1019 | * To minimise LRU disruption, the caller can indicate that it only |
| 1020 | * wants to isolate pages it will be able to operate on without |
| 1021 | * blocking - clean pages for the most part. |
| 1022 | * |
| 1023 | * ISOLATE_CLEAN means that only clean pages should be isolated. This |
| 1024 | * is used by reclaim when it is cannot write to backing storage |
| 1025 | * |
| 1026 | * ISOLATE_ASYNC_MIGRATE is used to indicate that it only wants to pages |
| 1027 | * that it is possible to migrate without blocking |
| 1028 | */ |
| 1029 | if (mode & (ISOLATE_CLEAN|ISOLATE_ASYNC_MIGRATE)) { |
| 1030 | /* All the caller can do on PageWriteback is block */ |
| 1031 | if (PageWriteback(page)) |
| 1032 | return ret; |
| 1033 | |
| 1034 | if (PageDirty(page)) { |
| 1035 | struct address_space *mapping; |
| 1036 | |
| 1037 | /* ISOLATE_CLEAN means only clean pages */ |
| 1038 | if (mode & ISOLATE_CLEAN) |
| 1039 | return ret; |
| 1040 | |
| 1041 | /* |
| 1042 | * Only pages without mappings or that have a |
| 1043 | * ->migratepage callback are possible to migrate |
| 1044 | * without blocking |
| 1045 | */ |
| 1046 | mapping = page_mapping(page); |
| 1047 | if (mapping && !mapping->a_ops->migratepage) |
| 1048 | return ret; |
| 1049 | } |
| 1050 | } |
Minchan Kim | 39deaf8 | 2011-10-31 17:06:51 -0700 | [diff] [blame] | 1051 | |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1052 | if ((mode & ISOLATE_UNMAPPED) && page_mapped(page)) |
| 1053 | return ret; |
| 1054 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1055 | if (likely(get_page_unless_zero(page))) { |
| 1056 | /* |
| 1057 | * Be careful not to clear PageLRU until after we're |
| 1058 | * sure the page is not being freed elsewhere -- the |
| 1059 | * page release code relies on it. |
| 1060 | */ |
| 1061 | ClearPageLRU(page); |
| 1062 | ret = 0; |
| 1063 | } |
| 1064 | |
| 1065 | return ret; |
| 1066 | } |
| 1067 | |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 1068 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1069 | * zone->lru_lock is heavily contended. Some of the functions that |
| 1070 | * shrink the lists perform better by taking out a batch of pages |
| 1071 | * and working on them outside the LRU lock. |
| 1072 | * |
| 1073 | * For pagecache intensive workloads, this function is the hottest |
| 1074 | * spot in the kernel (apart from copy_*_user functions). |
| 1075 | * |
| 1076 | * Appropriate locks must be held before calling this function. |
| 1077 | * |
| 1078 | * @nr_to_scan: The number of pages to look through on the list. |
Konstantin Khlebnikov | 5dc3597 | 2012-05-29 15:06:58 -0700 | [diff] [blame] | 1079 | * @lruvec: The LRU vector to pull pages from. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1080 | * @dst: The temp list to put pages on to. |
Hugh Dickins | f626012 | 2012-01-12 17:20:06 -0800 | [diff] [blame] | 1081 | * @nr_scanned: The number of pages that were scanned. |
Rik van Riel | fe2c2a1 | 2012-03-21 16:33:51 -0700 | [diff] [blame] | 1082 | * @sc: The scan_control struct for this reclaim session |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1083 | * @mode: One of the LRU isolation modes |
Konstantin Khlebnikov | 3cb9945 | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 1084 | * @lru: LRU list id for isolating |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1085 | * |
| 1086 | * returns how many pages were moved onto *@dst. |
| 1087 | */ |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1088 | static unsigned long isolate_lru_pages(unsigned long nr_to_scan, |
Konstantin Khlebnikov | 5dc3597 | 2012-05-29 15:06:58 -0700 | [diff] [blame] | 1089 | struct lruvec *lruvec, struct list_head *dst, |
Rik van Riel | fe2c2a1 | 2012-03-21 16:33:51 -0700 | [diff] [blame] | 1090 | unsigned long *nr_scanned, struct scan_control *sc, |
Konstantin Khlebnikov | 3cb9945 | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 1091 | isolate_mode_t mode, enum lru_list lru) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1092 | { |
Hugh Dickins | 75b00af | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1093 | struct list_head *src = &lruvec->lists[lru]; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1094 | unsigned long nr_taken = 0; |
Wu Fengguang | c9b02d9 | 2006-03-22 00:08:23 -0800 | [diff] [blame] | 1095 | unsigned long scan; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1096 | |
Wu Fengguang | c9b02d9 | 2006-03-22 00:08:23 -0800 | [diff] [blame] | 1097 | for (scan = 0; scan < nr_to_scan && !list_empty(src); scan++) { |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1098 | struct page *page; |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1099 | int nr_pages; |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1100 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1101 | page = lru_to_page(src); |
| 1102 | prefetchw_prev_lru_page(page, src, flags); |
| 1103 | |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 1104 | VM_BUG_ON(!PageLRU(page)); |
Nick Piggin | 8d438f9 | 2006-03-22 00:07:59 -0800 | [diff] [blame] | 1105 | |
Konstantin Khlebnikov | f3fd4a6 | 2012-05-29 15:06:54 -0700 | [diff] [blame] | 1106 | switch (__isolate_lru_page(page, mode)) { |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1107 | case 0: |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1108 | nr_pages = hpage_nr_pages(page); |
| 1109 | mem_cgroup_update_lru_size(lruvec, lru, -nr_pages); |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1110 | list_move(&page->lru, dst); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1111 | nr_taken += nr_pages; |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1112 | break; |
Nick Piggin | 46453a6 | 2006-03-22 00:07:58 -0800 | [diff] [blame] | 1113 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1114 | case -EBUSY: |
| 1115 | /* else it is being freed elsewhere */ |
| 1116 | list_move(&page->lru, src); |
| 1117 | continue; |
| 1118 | |
| 1119 | default: |
| 1120 | BUG(); |
| 1121 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1122 | } |
| 1123 | |
Hugh Dickins | f626012 | 2012-01-12 17:20:06 -0800 | [diff] [blame] | 1124 | *nr_scanned = scan; |
Hugh Dickins | 75b00af | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1125 | trace_mm_vmscan_lru_isolate(sc->order, nr_to_scan, scan, |
| 1126 | nr_taken, mode, is_file_lru(lru)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1127 | return nr_taken; |
| 1128 | } |
| 1129 | |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1130 | /** |
| 1131 | * isolate_lru_page - tries to isolate a page from its LRU list |
| 1132 | * @page: page to isolate from its LRU list |
| 1133 | * |
| 1134 | * Isolates a @page from an LRU list, clears PageLRU and adjusts the |
| 1135 | * vmstat statistic corresponding to whatever LRU list the page was on. |
| 1136 | * |
| 1137 | * Returns 0 if the page was removed from an LRU list. |
| 1138 | * Returns -EBUSY if the page was not on an LRU list. |
| 1139 | * |
| 1140 | * The returned page will have PageLRU() cleared. If it was found on |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1141 | * the active list, it will have PageActive set. If it was found on |
| 1142 | * the unevictable list, it will have the PageUnevictable bit set. That flag |
| 1143 | * 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] | 1144 | * |
| 1145 | * The vmstat statistic corresponding to the list on which the page was |
| 1146 | * found will be decremented. |
| 1147 | * |
| 1148 | * Restrictions: |
| 1149 | * (1) Must be called with an elevated refcount on the page. This is a |
| 1150 | * fundamentnal difference from isolate_lru_pages (which is called |
| 1151 | * without a stable reference). |
| 1152 | * (2) the lru_lock must not be held. |
| 1153 | * (3) interrupts must be enabled. |
| 1154 | */ |
| 1155 | int isolate_lru_page(struct page *page) |
| 1156 | { |
| 1157 | int ret = -EBUSY; |
| 1158 | |
Konstantin Khlebnikov | 0c91731 | 2011-05-24 17:12:21 -0700 | [diff] [blame] | 1159 | VM_BUG_ON(!page_count(page)); |
| 1160 | |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1161 | if (PageLRU(page)) { |
| 1162 | struct zone *zone = page_zone(page); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1163 | struct lruvec *lruvec; |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1164 | |
| 1165 | spin_lock_irq(&zone->lru_lock); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1166 | lruvec = mem_cgroup_page_lruvec(page, zone); |
Konstantin Khlebnikov | 0c91731 | 2011-05-24 17:12:21 -0700 | [diff] [blame] | 1167 | if (PageLRU(page)) { |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1168 | int lru = page_lru(page); |
Konstantin Khlebnikov | 0c91731 | 2011-05-24 17:12:21 -0700 | [diff] [blame] | 1169 | get_page(page); |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1170 | ClearPageLRU(page); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1171 | del_page_from_lru_list(page, lruvec, lru); |
| 1172 | ret = 0; |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1173 | } |
| 1174 | spin_unlock_irq(&zone->lru_lock); |
| 1175 | } |
| 1176 | return ret; |
| 1177 | } |
| 1178 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1179 | /* |
Fengguang Wu | d37dd5d | 2012-12-18 14:23:28 -0800 | [diff] [blame] | 1180 | * A direct reclaimer may isolate SWAP_CLUSTER_MAX pages from the LRU list and |
| 1181 | * then get resheduled. When there are massive number of tasks doing page |
| 1182 | * allocation, such sleeping direct reclaimers may keep piling up on each CPU, |
| 1183 | * the LRU list will go small and be scanned faster than necessary, leading to |
| 1184 | * unnecessary swapping, thrashing and OOM. |
Rik van Riel | 35cd781 | 2009-09-21 17:01:38 -0700 | [diff] [blame] | 1185 | */ |
| 1186 | static int too_many_isolated(struct zone *zone, int file, |
| 1187 | struct scan_control *sc) |
| 1188 | { |
| 1189 | unsigned long inactive, isolated; |
| 1190 | |
| 1191 | if (current_is_kswapd()) |
| 1192 | return 0; |
| 1193 | |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 1194 | if (!global_reclaim(sc)) |
Rik van Riel | 35cd781 | 2009-09-21 17:01:38 -0700 | [diff] [blame] | 1195 | return 0; |
| 1196 | |
| 1197 | if (file) { |
| 1198 | inactive = zone_page_state(zone, NR_INACTIVE_FILE); |
| 1199 | isolated = zone_page_state(zone, NR_ISOLATED_FILE); |
| 1200 | } else { |
| 1201 | inactive = zone_page_state(zone, NR_INACTIVE_ANON); |
| 1202 | isolated = zone_page_state(zone, NR_ISOLATED_ANON); |
| 1203 | } |
| 1204 | |
Fengguang Wu | 3cf2384 | 2012-12-18 14:23:31 -0800 | [diff] [blame] | 1205 | /* |
| 1206 | * GFP_NOIO/GFP_NOFS callers are allowed to isolate more pages, so they |
| 1207 | * won't get blocked by normal direct-reclaimers, forming a circular |
| 1208 | * deadlock. |
| 1209 | */ |
| 1210 | if ((sc->gfp_mask & GFP_IOFS) == GFP_IOFS) |
| 1211 | inactive >>= 3; |
| 1212 | |
Rik van Riel | 35cd781 | 2009-09-21 17:01:38 -0700 | [diff] [blame] | 1213 | return isolated > inactive; |
| 1214 | } |
| 1215 | |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1216 | static noinline_for_stack void |
Hugh Dickins | 75b00af | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1217 | putback_inactive_pages(struct lruvec *lruvec, struct list_head *page_list) |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1218 | { |
Konstantin Khlebnikov | 27ac81d | 2012-05-29 15:07:00 -0700 | [diff] [blame] | 1219 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
| 1220 | struct zone *zone = lruvec_zone(lruvec); |
Hugh Dickins | 3f79768 | 2012-01-12 17:20:07 -0800 | [diff] [blame] | 1221 | LIST_HEAD(pages_to_free); |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1222 | |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1223 | /* |
| 1224 | * Put back any unfreeable pages. |
| 1225 | */ |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1226 | while (!list_empty(page_list)) { |
Hugh Dickins | 3f79768 | 2012-01-12 17:20:07 -0800 | [diff] [blame] | 1227 | struct page *page = lru_to_page(page_list); |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1228 | int lru; |
Hugh Dickins | 3f79768 | 2012-01-12 17:20:07 -0800 | [diff] [blame] | 1229 | |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1230 | VM_BUG_ON(PageLRU(page)); |
| 1231 | list_del(&page->lru); |
Hugh Dickins | 39b5f29 | 2012-10-08 16:33:18 -0700 | [diff] [blame] | 1232 | if (unlikely(!page_evictable(page))) { |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1233 | spin_unlock_irq(&zone->lru_lock); |
| 1234 | putback_lru_page(page); |
| 1235 | spin_lock_irq(&zone->lru_lock); |
| 1236 | continue; |
| 1237 | } |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1238 | |
| 1239 | lruvec = mem_cgroup_page_lruvec(page, zone); |
| 1240 | |
Linus Torvalds | 7a60857 | 2011-01-17 14:42:19 -0800 | [diff] [blame] | 1241 | SetPageLRU(page); |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1242 | lru = page_lru(page); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1243 | add_page_to_lru_list(page, lruvec, lru); |
| 1244 | |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1245 | if (is_active_lru(lru)) { |
| 1246 | int file = is_file_lru(lru); |
Rik van Riel | 9992af1 | 2011-01-13 15:47:13 -0800 | [diff] [blame] | 1247 | int numpages = hpage_nr_pages(page); |
| 1248 | reclaim_stat->recent_rotated[file] += numpages; |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1249 | } |
Hugh Dickins | 2bcf887 | 2012-01-12 17:19:56 -0800 | [diff] [blame] | 1250 | if (put_page_testzero(page)) { |
| 1251 | __ClearPageLRU(page); |
| 1252 | __ClearPageActive(page); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1253 | del_page_from_lru_list(page, lruvec, lru); |
Hugh Dickins | 2bcf887 | 2012-01-12 17:19:56 -0800 | [diff] [blame] | 1254 | |
| 1255 | if (unlikely(PageCompound(page))) { |
| 1256 | spin_unlock_irq(&zone->lru_lock); |
| 1257 | (*get_compound_page_dtor(page))(page); |
| 1258 | spin_lock_irq(&zone->lru_lock); |
| 1259 | } else |
| 1260 | list_add(&page->lru, &pages_to_free); |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1261 | } |
| 1262 | } |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1263 | |
Hugh Dickins | 3f79768 | 2012-01-12 17:20:07 -0800 | [diff] [blame] | 1264 | /* |
| 1265 | * To save our caller's stack, now use input list for pages to free. |
| 1266 | */ |
| 1267 | list_splice(&pages_to_free, page_list); |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1268 | } |
| 1269 | |
| 1270 | /* |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 1271 | * shrink_inactive_list() is a helper for shrink_zone(). It returns the number |
| 1272 | * of reclaimed pages |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1273 | */ |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1274 | static noinline_for_stack unsigned long |
Konstantin Khlebnikov | 1a93be0 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 1275 | shrink_inactive_list(unsigned long nr_to_scan, struct lruvec *lruvec, |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1276 | struct scan_control *sc, enum lru_list lru) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1277 | { |
| 1278 | LIST_HEAD(page_list); |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1279 | unsigned long nr_scanned; |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 1280 | unsigned long nr_reclaimed = 0; |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1281 | unsigned long nr_taken; |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 1282 | unsigned long nr_dirty = 0; |
| 1283 | unsigned long nr_writeback = 0; |
Konstantin Khlebnikov | f3fd4a6 | 2012-05-29 15:06:54 -0700 | [diff] [blame] | 1284 | isolate_mode_t isolate_mode = 0; |
Konstantin Khlebnikov | 3cb9945 | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 1285 | int file = is_file_lru(lru); |
Konstantin Khlebnikov | 1a93be0 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 1286 | struct zone *zone = lruvec_zone(lruvec); |
| 1287 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
KOSAKI Motohiro | 78dc583 | 2009-06-16 15:31:40 -0700 | [diff] [blame] | 1288 | |
Rik van Riel | 35cd781 | 2009-09-21 17:01:38 -0700 | [diff] [blame] | 1289 | while (unlikely(too_many_isolated(zone, file, sc))) { |
KOSAKI Motohiro | 58355c7 | 2009-10-26 16:49:35 -0700 | [diff] [blame] | 1290 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
Rik van Riel | 35cd781 | 2009-09-21 17:01:38 -0700 | [diff] [blame] | 1291 | |
| 1292 | /* We are about to die and free our memory. Return now. */ |
| 1293 | if (fatal_signal_pending(current)) |
| 1294 | return SWAP_CLUSTER_MAX; |
| 1295 | } |
| 1296 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1297 | lru_add_drain(); |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1298 | |
| 1299 | if (!sc->may_unmap) |
Hillf Danton | 6131728 | 2012-03-21 16:33:48 -0700 | [diff] [blame] | 1300 | isolate_mode |= ISOLATE_UNMAPPED; |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1301 | if (!sc->may_writepage) |
Hillf Danton | 6131728 | 2012-03-21 16:33:48 -0700 | [diff] [blame] | 1302 | isolate_mode |= ISOLATE_CLEAN; |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1303 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1304 | spin_lock_irq(&zone->lru_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1305 | |
Konstantin Khlebnikov | 5dc3597 | 2012-05-29 15:06:58 -0700 | [diff] [blame] | 1306 | nr_taken = isolate_lru_pages(nr_to_scan, lruvec, &page_list, |
| 1307 | &nr_scanned, sc, isolate_mode, lru); |
Konstantin Khlebnikov | 95d918f | 2012-05-29 15:06:59 -0700 | [diff] [blame] | 1308 | |
| 1309 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, -nr_taken); |
| 1310 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, nr_taken); |
| 1311 | |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 1312 | if (global_reclaim(sc)) { |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1313 | zone->pages_scanned += nr_scanned; |
KOSAKI Motohiro | b35ea17 | 2009-09-21 17:01:36 -0700 | [diff] [blame] | 1314 | if (current_is_kswapd()) |
Hugh Dickins | 75b00af | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1315 | __count_zone_vm_events(PGSCAN_KSWAPD, zone, nr_scanned); |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1316 | else |
Hugh Dickins | 75b00af | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1317 | __count_zone_vm_events(PGSCAN_DIRECT, zone, nr_scanned); |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1318 | } |
Hillf Danton | d563c05 | 2012-03-21 16:34:02 -0700 | [diff] [blame] | 1319 | spin_unlock_irq(&zone->lru_lock); |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 1320 | |
Hillf Danton | d563c05 | 2012-03-21 16:34:02 -0700 | [diff] [blame] | 1321 | if (nr_taken == 0) |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1322 | return 0; |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1323 | |
Minchan Kim | 02c6de8 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 1324 | nr_reclaimed = shrink_page_list(&page_list, zone, sc, TTU_UNMAP, |
| 1325 | &nr_dirty, &nr_writeback, false); |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1326 | |
Hugh Dickins | 3f79768 | 2012-01-12 17:20:07 -0800 | [diff] [blame] | 1327 | spin_lock_irq(&zone->lru_lock); |
| 1328 | |
Konstantin Khlebnikov | 95d918f | 2012-05-29 15:06:59 -0700 | [diff] [blame] | 1329 | reclaim_stat->recent_scanned[file] += nr_taken; |
Hillf Danton | d563c05 | 2012-03-21 16:34:02 -0700 | [diff] [blame] | 1330 | |
Ying Han | 904249a | 2012-04-25 16:01:48 -0700 | [diff] [blame] | 1331 | if (global_reclaim(sc)) { |
| 1332 | if (current_is_kswapd()) |
| 1333 | __count_zone_vm_events(PGSTEAL_KSWAPD, zone, |
| 1334 | nr_reclaimed); |
| 1335 | else |
| 1336 | __count_zone_vm_events(PGSTEAL_DIRECT, zone, |
| 1337 | nr_reclaimed); |
| 1338 | } |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1339 | |
Konstantin Khlebnikov | 27ac81d | 2012-05-29 15:07:00 -0700 | [diff] [blame] | 1340 | putback_inactive_pages(lruvec, &page_list); |
Hugh Dickins | 3f79768 | 2012-01-12 17:20:07 -0800 | [diff] [blame] | 1341 | |
Konstantin Khlebnikov | 95d918f | 2012-05-29 15:06:59 -0700 | [diff] [blame] | 1342 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, -nr_taken); |
Hugh Dickins | 3f79768 | 2012-01-12 17:20:07 -0800 | [diff] [blame] | 1343 | |
| 1344 | spin_unlock_irq(&zone->lru_lock); |
| 1345 | |
| 1346 | free_hot_cold_page_list(&page_list, 1); |
Mel Gorman | e11da5b | 2010-10-26 14:21:40 -0700 | [diff] [blame] | 1347 | |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 1348 | /* |
| 1349 | * If reclaim is isolating dirty pages under writeback, it implies |
| 1350 | * that the long-lived page allocation rate is exceeding the page |
| 1351 | * laundering rate. Either the global limits are not being effective |
| 1352 | * at throttling processes due to the page distribution throughout |
| 1353 | * zones or there is heavy usage of a slow backing device. The |
| 1354 | * only option is to throttle from reclaim context which is not ideal |
| 1355 | * as there is no guarantee the dirtying process is throttled in the |
| 1356 | * same way balance_dirty_pages() manages. |
| 1357 | * |
| 1358 | * This scales the number of dirty pages that must be under writeback |
| 1359 | * before throttling depending on priority. It is a simple backoff |
| 1360 | * function that has the most effect in the range DEF_PRIORITY to |
| 1361 | * DEF_PRIORITY-2 which is the priority reclaim is considered to be |
| 1362 | * in trouble and reclaim is considered to be in trouble. |
| 1363 | * |
| 1364 | * DEF_PRIORITY 100% isolated pages must be PageWriteback to throttle |
| 1365 | * DEF_PRIORITY-1 50% must be PageWriteback |
| 1366 | * DEF_PRIORITY-2 25% must be PageWriteback, kswapd in trouble |
| 1367 | * ... |
| 1368 | * DEF_PRIORITY-6 For SWAP_CLUSTER_MAX isolated pages, throttle if any |
| 1369 | * isolated page is PageWriteback |
| 1370 | */ |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1371 | if (nr_writeback && nr_writeback >= |
| 1372 | (nr_taken >> (DEF_PRIORITY - sc->priority))) |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 1373 | wait_iff_congested(zone, BLK_RW_ASYNC, HZ/10); |
| 1374 | |
Mel Gorman | e11da5b | 2010-10-26 14:21:40 -0700 | [diff] [blame] | 1375 | trace_mm_vmscan_lru_shrink_inactive(zone->zone_pgdat->node_id, |
| 1376 | zone_idx(zone), |
| 1377 | nr_scanned, nr_reclaimed, |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1378 | sc->priority, |
Mel Gorman | 23b9da5 | 2012-05-29 15:06:20 -0700 | [diff] [blame] | 1379 | trace_shrink_flags(file)); |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 1380 | return nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1381 | } |
| 1382 | |
Martin Bligh | 3bb1a852 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 1383 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1384 | * This moves pages from the active list to the inactive list. |
| 1385 | * |
| 1386 | * We move them the other way if the page is referenced by one or more |
| 1387 | * processes, from rmap. |
| 1388 | * |
| 1389 | * If the pages are mostly unmapped, the processing is fast and it is |
| 1390 | * appropriate to hold zone->lru_lock across the whole operation. But if |
| 1391 | * the pages are mapped, the processing is slow (page_referenced()) so we |
| 1392 | * should drop zone->lru_lock around each page. It's impossible to balance |
| 1393 | * this, so instead we remove the pages from the LRU while processing them. |
| 1394 | * It is safe to rely on PG_active against the non-LRU pages in here because |
| 1395 | * nobody will play with that bit on a non-LRU page. |
| 1396 | * |
| 1397 | * The downside is that we have to touch page->_count against each page. |
| 1398 | * But we had to alter page->flags anyway. |
| 1399 | */ |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1400 | |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1401 | static void move_active_pages_to_lru(struct lruvec *lruvec, |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1402 | struct list_head *list, |
Hugh Dickins | 2bcf887 | 2012-01-12 17:19:56 -0800 | [diff] [blame] | 1403 | struct list_head *pages_to_free, |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1404 | enum lru_list lru) |
| 1405 | { |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1406 | struct zone *zone = lruvec_zone(lruvec); |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1407 | unsigned long pgmoved = 0; |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1408 | struct page *page; |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1409 | int nr_pages; |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1410 | |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1411 | while (!list_empty(list)) { |
| 1412 | page = lru_to_page(list); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1413 | lruvec = mem_cgroup_page_lruvec(page, zone); |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1414 | |
| 1415 | VM_BUG_ON(PageLRU(page)); |
| 1416 | SetPageLRU(page); |
| 1417 | |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1418 | nr_pages = hpage_nr_pages(page); |
| 1419 | mem_cgroup_update_lru_size(lruvec, lru, nr_pages); |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1420 | list_move(&page->lru, &lruvec->lists[lru]); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1421 | pgmoved += nr_pages; |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1422 | |
Hugh Dickins | 2bcf887 | 2012-01-12 17:19:56 -0800 | [diff] [blame] | 1423 | if (put_page_testzero(page)) { |
| 1424 | __ClearPageLRU(page); |
| 1425 | __ClearPageActive(page); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1426 | del_page_from_lru_list(page, lruvec, lru); |
Hugh Dickins | 2bcf887 | 2012-01-12 17:19:56 -0800 | [diff] [blame] | 1427 | |
| 1428 | if (unlikely(PageCompound(page))) { |
| 1429 | spin_unlock_irq(&zone->lru_lock); |
| 1430 | (*get_compound_page_dtor(page))(page); |
| 1431 | spin_lock_irq(&zone->lru_lock); |
| 1432 | } else |
| 1433 | list_add(&page->lru, pages_to_free); |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1434 | } |
| 1435 | } |
| 1436 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved); |
| 1437 | if (!is_active_lru(lru)) |
| 1438 | __count_vm_events(PGDEACTIVATE, pgmoved); |
| 1439 | } |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1440 | |
Hugh Dickins | f626012 | 2012-01-12 17:20:06 -0800 | [diff] [blame] | 1441 | static void shrink_active_list(unsigned long nr_to_scan, |
Konstantin Khlebnikov | 1a93be0 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 1442 | struct lruvec *lruvec, |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1443 | struct scan_control *sc, |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1444 | enum lru_list lru) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1445 | { |
KOSAKI Motohiro | 44c241f | 2009-09-21 17:01:35 -0700 | [diff] [blame] | 1446 | unsigned long nr_taken; |
Hugh Dickins | f626012 | 2012-01-12 17:20:06 -0800 | [diff] [blame] | 1447 | unsigned long nr_scanned; |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 1448 | unsigned long vm_flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1449 | LIST_HEAD(l_hold); /* The pages which were snipped off */ |
Wu Fengguang | 8cab475 | 2009-06-16 15:33:12 -0700 | [diff] [blame] | 1450 | LIST_HEAD(l_active); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1451 | LIST_HEAD(l_inactive); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1452 | struct page *page; |
Konstantin Khlebnikov | 1a93be0 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 1453 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
KOSAKI Motohiro | 44c241f | 2009-09-21 17:01:35 -0700 | [diff] [blame] | 1454 | unsigned long nr_rotated = 0; |
Konstantin Khlebnikov | f3fd4a6 | 2012-05-29 15:06:54 -0700 | [diff] [blame] | 1455 | isolate_mode_t isolate_mode = 0; |
Konstantin Khlebnikov | 3cb9945 | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 1456 | int file = is_file_lru(lru); |
Konstantin Khlebnikov | 1a93be0 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 1457 | struct zone *zone = lruvec_zone(lruvec); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1458 | |
| 1459 | lru_add_drain(); |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1460 | |
| 1461 | if (!sc->may_unmap) |
Hillf Danton | 6131728 | 2012-03-21 16:33:48 -0700 | [diff] [blame] | 1462 | isolate_mode |= ISOLATE_UNMAPPED; |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1463 | if (!sc->may_writepage) |
Hillf Danton | 6131728 | 2012-03-21 16:33:48 -0700 | [diff] [blame] | 1464 | isolate_mode |= ISOLATE_CLEAN; |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1465 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1466 | spin_lock_irq(&zone->lru_lock); |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1467 | |
Konstantin Khlebnikov | 5dc3597 | 2012-05-29 15:06:58 -0700 | [diff] [blame] | 1468 | nr_taken = isolate_lru_pages(nr_to_scan, lruvec, &l_hold, |
| 1469 | &nr_scanned, sc, isolate_mode, lru); |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 1470 | if (global_reclaim(sc)) |
Hugh Dickins | f626012 | 2012-01-12 17:20:06 -0800 | [diff] [blame] | 1471 | zone->pages_scanned += nr_scanned; |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 1472 | |
Johannes Weiner | b7c46d1 | 2009-09-21 17:02:56 -0700 | [diff] [blame] | 1473 | reclaim_stat->recent_scanned[file] += nr_taken; |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1474 | |
Hugh Dickins | f626012 | 2012-01-12 17:20:06 -0800 | [diff] [blame] | 1475 | __count_zone_vm_events(PGREFILL, zone, nr_scanned); |
Konstantin Khlebnikov | 3cb9945 | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 1476 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, -nr_taken); |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 1477 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, nr_taken); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1478 | spin_unlock_irq(&zone->lru_lock); |
| 1479 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1480 | while (!list_empty(&l_hold)) { |
| 1481 | cond_resched(); |
| 1482 | page = lru_to_page(&l_hold); |
| 1483 | list_del(&page->lru); |
Rik van Riel | 7e9cd48 | 2008-10-18 20:26:35 -0700 | [diff] [blame] | 1484 | |
Hugh Dickins | 39b5f29 | 2012-10-08 16:33:18 -0700 | [diff] [blame] | 1485 | if (unlikely(!page_evictable(page))) { |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1486 | putback_lru_page(page); |
| 1487 | continue; |
| 1488 | } |
| 1489 | |
Mel Gorman | cc715d9 | 2012-03-21 16:34:00 -0700 | [diff] [blame] | 1490 | if (unlikely(buffer_heads_over_limit)) { |
| 1491 | if (page_has_private(page) && trylock_page(page)) { |
| 1492 | if (page_has_private(page)) |
| 1493 | try_to_release_page(page, 0); |
| 1494 | unlock_page(page); |
| 1495 | } |
| 1496 | } |
| 1497 | |
Johannes Weiner | c3ac9a8 | 2012-05-29 15:06:25 -0700 | [diff] [blame] | 1498 | if (page_referenced(page, 0, sc->target_mem_cgroup, |
| 1499 | &vm_flags)) { |
Rik van Riel | 9992af1 | 2011-01-13 15:47:13 -0800 | [diff] [blame] | 1500 | nr_rotated += hpage_nr_pages(page); |
Wu Fengguang | 8cab475 | 2009-06-16 15:33:12 -0700 | [diff] [blame] | 1501 | /* |
| 1502 | * Identify referenced, file-backed active pages and |
| 1503 | * give them one more trip around the active list. So |
| 1504 | * that executable code get better chances to stay in |
| 1505 | * memory under moderate memory pressure. Anon pages |
| 1506 | * are not likely to be evicted by use-once streaming |
| 1507 | * IO, plus JVM can create lots of anon VM_EXEC pages, |
| 1508 | * so we ignore them here. |
| 1509 | */ |
Wu Fengguang | 41e2098 | 2009-10-26 16:49:53 -0700 | [diff] [blame] | 1510 | if ((vm_flags & VM_EXEC) && page_is_file_cache(page)) { |
Wu Fengguang | 8cab475 | 2009-06-16 15:33:12 -0700 | [diff] [blame] | 1511 | list_add(&page->lru, &l_active); |
| 1512 | continue; |
| 1513 | } |
| 1514 | } |
Rik van Riel | 7e9cd48 | 2008-10-18 20:26:35 -0700 | [diff] [blame] | 1515 | |
KOSAKI Motohiro | 5205e56 | 2009-09-21 17:01:44 -0700 | [diff] [blame] | 1516 | ClearPageActive(page); /* we are de-activating */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1517 | list_add(&page->lru, &l_inactive); |
| 1518 | } |
| 1519 | |
Andrew Morton | b555749 | 2009-01-06 14:40:13 -0800 | [diff] [blame] | 1520 | /* |
Wu Fengguang | 8cab475 | 2009-06-16 15:33:12 -0700 | [diff] [blame] | 1521 | * Move pages back to the lru list. |
Andrew Morton | b555749 | 2009-01-06 14:40:13 -0800 | [diff] [blame] | 1522 | */ |
Johannes Weiner | 2a1dc50 | 2008-12-01 03:00:35 +0100 | [diff] [blame] | 1523 | spin_lock_irq(&zone->lru_lock); |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1524 | /* |
Wu Fengguang | 8cab475 | 2009-06-16 15:33:12 -0700 | [diff] [blame] | 1525 | * Count referenced pages from currently used mappings as rotated, |
| 1526 | * even though only some of them are actually re-activated. This |
| 1527 | * helps balance scan pressure between file and anonymous pages in |
| 1528 | * get_scan_ratio. |
Rik van Riel | 7e9cd48 | 2008-10-18 20:26:35 -0700 | [diff] [blame] | 1529 | */ |
Johannes Weiner | b7c46d1 | 2009-09-21 17:02:56 -0700 | [diff] [blame] | 1530 | reclaim_stat->recent_rotated[file] += nr_rotated; |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1531 | |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1532 | move_active_pages_to_lru(lruvec, &l_active, &l_hold, lru); |
| 1533 | move_active_pages_to_lru(lruvec, &l_inactive, &l_hold, lru - LRU_ACTIVE); |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 1534 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, -nr_taken); |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 1535 | spin_unlock_irq(&zone->lru_lock); |
Hugh Dickins | 2bcf887 | 2012-01-12 17:19:56 -0800 | [diff] [blame] | 1536 | |
| 1537 | free_hot_cold_page_list(&l_hold, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1538 | } |
| 1539 | |
Minchan Kim | 74e3f3c | 2010-10-26 14:21:31 -0700 | [diff] [blame] | 1540 | #ifdef CONFIG_SWAP |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1541 | static int inactive_anon_is_low_global(struct zone *zone) |
KOSAKI Motohiro | f89eb90 | 2009-01-07 18:08:14 -0800 | [diff] [blame] | 1542 | { |
| 1543 | unsigned long active, inactive; |
| 1544 | |
| 1545 | active = zone_page_state(zone, NR_ACTIVE_ANON); |
| 1546 | inactive = zone_page_state(zone, NR_INACTIVE_ANON); |
| 1547 | |
| 1548 | if (inactive * zone->inactive_ratio < active) |
| 1549 | return 1; |
| 1550 | |
| 1551 | return 0; |
| 1552 | } |
| 1553 | |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1554 | /** |
| 1555 | * inactive_anon_is_low - check if anonymous pages need to be deactivated |
Konstantin Khlebnikov | c56d5c7 | 2012-05-29 15:07:00 -0700 | [diff] [blame] | 1556 | * @lruvec: LRU vector to check |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1557 | * |
| 1558 | * Returns true if the zone does not have enough inactive anon pages, |
| 1559 | * meaning some active anon pages need to be deactivated. |
| 1560 | */ |
Konstantin Khlebnikov | c56d5c7 | 2012-05-29 15:07:00 -0700 | [diff] [blame] | 1561 | static int inactive_anon_is_low(struct lruvec *lruvec) |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1562 | { |
Minchan Kim | 74e3f3c | 2010-10-26 14:21:31 -0700 | [diff] [blame] | 1563 | /* |
| 1564 | * If we don't have swap space, anonymous page deactivation |
| 1565 | * is pointless. |
| 1566 | */ |
| 1567 | if (!total_swap_pages) |
| 1568 | return 0; |
| 1569 | |
Hugh Dickins | c3c787e | 2012-05-29 15:06:52 -0700 | [diff] [blame] | 1570 | if (!mem_cgroup_disabled()) |
Konstantin Khlebnikov | c56d5c7 | 2012-05-29 15:07:00 -0700 | [diff] [blame] | 1571 | return mem_cgroup_inactive_anon_is_low(lruvec); |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1572 | |
Konstantin Khlebnikov | c56d5c7 | 2012-05-29 15:07:00 -0700 | [diff] [blame] | 1573 | return inactive_anon_is_low_global(lruvec_zone(lruvec)); |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1574 | } |
Minchan Kim | 74e3f3c | 2010-10-26 14:21:31 -0700 | [diff] [blame] | 1575 | #else |
Konstantin Khlebnikov | c56d5c7 | 2012-05-29 15:07:00 -0700 | [diff] [blame] | 1576 | static inline int inactive_anon_is_low(struct lruvec *lruvec) |
Minchan Kim | 74e3f3c | 2010-10-26 14:21:31 -0700 | [diff] [blame] | 1577 | { |
| 1578 | return 0; |
| 1579 | } |
| 1580 | #endif |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1581 | |
Rik van Riel | 56e49d2 | 2009-06-16 15:32:28 -0700 | [diff] [blame] | 1582 | /** |
| 1583 | * inactive_file_is_low - check if file pages need to be deactivated |
Konstantin Khlebnikov | c56d5c7 | 2012-05-29 15:07:00 -0700 | [diff] [blame] | 1584 | * @lruvec: LRU vector to check |
Rik van Riel | 56e49d2 | 2009-06-16 15:32:28 -0700 | [diff] [blame] | 1585 | * |
| 1586 | * When the system is doing streaming IO, memory pressure here |
| 1587 | * ensures that active file pages get deactivated, until more |
| 1588 | * than half of the file pages are on the inactive list. |
| 1589 | * |
| 1590 | * Once we get to that situation, protect the system's working |
| 1591 | * set from being evicted by disabling active file page aging. |
| 1592 | * |
| 1593 | * This uses a different ratio than the anonymous pages, because |
| 1594 | * the page cache uses a use-once replacement algorithm. |
| 1595 | */ |
Konstantin Khlebnikov | c56d5c7 | 2012-05-29 15:07:00 -0700 | [diff] [blame] | 1596 | static int inactive_file_is_low(struct lruvec *lruvec) |
Rik van Riel | 56e49d2 | 2009-06-16 15:32:28 -0700 | [diff] [blame] | 1597 | { |
Johannes Weiner | e379014 | 2013-02-22 16:35:19 -0800 | [diff] [blame] | 1598 | unsigned long inactive; |
| 1599 | unsigned long active; |
Rik van Riel | 56e49d2 | 2009-06-16 15:32:28 -0700 | [diff] [blame] | 1600 | |
Johannes Weiner | e379014 | 2013-02-22 16:35:19 -0800 | [diff] [blame] | 1601 | inactive = get_lru_size(lruvec, LRU_INACTIVE_FILE); |
| 1602 | active = get_lru_size(lruvec, LRU_ACTIVE_FILE); |
| 1603 | |
| 1604 | return active > inactive; |
Rik van Riel | 56e49d2 | 2009-06-16 15:32:28 -0700 | [diff] [blame] | 1605 | } |
| 1606 | |
Hugh Dickins | 75b00af | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1607 | static int inactive_list_is_low(struct lruvec *lruvec, enum lru_list lru) |
Rik van Riel | b39415b | 2009-12-14 17:59:48 -0800 | [diff] [blame] | 1608 | { |
Hugh Dickins | 75b00af | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1609 | if (is_file_lru(lru)) |
Konstantin Khlebnikov | c56d5c7 | 2012-05-29 15:07:00 -0700 | [diff] [blame] | 1610 | return inactive_file_is_low(lruvec); |
Rik van Riel | b39415b | 2009-12-14 17:59:48 -0800 | [diff] [blame] | 1611 | else |
Konstantin Khlebnikov | c56d5c7 | 2012-05-29 15:07:00 -0700 | [diff] [blame] | 1612 | return inactive_anon_is_low(lruvec); |
Rik van Riel | b39415b | 2009-12-14 17:59:48 -0800 | [diff] [blame] | 1613 | } |
| 1614 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1615 | static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan, |
Konstantin Khlebnikov | 1a93be0 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 1616 | struct lruvec *lruvec, struct scan_control *sc) |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1617 | { |
Rik van Riel | b39415b | 2009-12-14 17:59:48 -0800 | [diff] [blame] | 1618 | if (is_active_lru(lru)) { |
Hugh Dickins | 75b00af | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1619 | if (inactive_list_is_low(lruvec, lru)) |
Konstantin Khlebnikov | 1a93be0 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 1620 | shrink_active_list(nr_to_scan, lruvec, sc, lru); |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1621 | return 0; |
| 1622 | } |
| 1623 | |
Konstantin Khlebnikov | 1a93be0 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 1624 | return shrink_inactive_list(nr_to_scan, lruvec, sc, lru); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1625 | } |
| 1626 | |
Konstantin Khlebnikov | 3d58ab5 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1627 | static int vmscan_swappiness(struct scan_control *sc) |
KAMEZAWA Hiroyuki | 1f4c025 | 2011-07-26 16:08:21 -0700 | [diff] [blame] | 1628 | { |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 1629 | if (global_reclaim(sc)) |
KAMEZAWA Hiroyuki | 1f4c025 | 2011-07-26 16:08:21 -0700 | [diff] [blame] | 1630 | return vm_swappiness; |
Konstantin Khlebnikov | 3d58ab5 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1631 | return mem_cgroup_swappiness(sc->target_mem_cgroup); |
KAMEZAWA Hiroyuki | 1f4c025 | 2011-07-26 16:08:21 -0700 | [diff] [blame] | 1632 | } |
| 1633 | |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 1634 | enum scan_balance { |
| 1635 | SCAN_EQUAL, |
| 1636 | SCAN_FRACT, |
| 1637 | SCAN_ANON, |
| 1638 | SCAN_FILE, |
| 1639 | }; |
| 1640 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1641 | /* |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1642 | * Determine how aggressively the anon and file LRU lists should be |
| 1643 | * scanned. The relative value of each set of LRU lists is determined |
| 1644 | * by looking at the fraction of the pages scanned we did rotate back |
| 1645 | * onto the active list instead of evict. |
| 1646 | * |
Wanpeng Li | be7bd59 | 2012-06-14 20:41:02 +0800 | [diff] [blame] | 1647 | * nr[0] = anon inactive pages to scan; nr[1] = anon active pages to scan |
| 1648 | * nr[2] = file inactive pages to scan; nr[3] = file active pages to scan |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1649 | */ |
Konstantin Khlebnikov | 9012637 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 1650 | static void get_scan_count(struct lruvec *lruvec, struct scan_control *sc, |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1651 | unsigned long *nr) |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1652 | { |
Konstantin Khlebnikov | 9012637 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 1653 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 1654 | u64 fraction[2]; |
| 1655 | u64 denominator = 0; /* gcc */ |
Konstantin Khlebnikov | 9012637 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 1656 | struct zone *zone = lruvec_zone(lruvec); |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 1657 | unsigned long anon_prio, file_prio; |
| 1658 | enum scan_balance scan_balance; |
| 1659 | unsigned long anon, file, free; |
| 1660 | bool force_scan = false; |
| 1661 | unsigned long ap, fp; |
| 1662 | enum lru_list lru; |
KAMEZAWA Hiroyuki | 246e87a | 2011-05-26 16:25:34 -0700 | [diff] [blame] | 1663 | |
Johannes Weiner | f11c0ca | 2011-10-31 17:07:27 -0700 | [diff] [blame] | 1664 | /* |
| 1665 | * If the zone or memcg is small, nr[l] can be 0. This |
| 1666 | * results in no scanning on this priority and a potential |
| 1667 | * priority drop. Global direct reclaim can go to the next |
| 1668 | * zone and tends to have no problems. Global kswapd is for |
| 1669 | * zone balancing and it needs to scan a minimum amount. When |
| 1670 | * reclaiming for a memcg, a priority drop can cause high |
| 1671 | * latencies, so it's better to scan a minimum amount there as |
| 1672 | * well. |
| 1673 | */ |
Konstantin Khlebnikov | 9012637 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 1674 | if (current_is_kswapd() && zone->all_unreclaimable) |
Johannes Weiner | a4d3e9e | 2011-09-14 16:21:52 -0700 | [diff] [blame] | 1675 | force_scan = true; |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 1676 | if (!global_reclaim(sc)) |
Johannes Weiner | a4d3e9e | 2011-09-14 16:21:52 -0700 | [diff] [blame] | 1677 | force_scan = true; |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1678 | |
| 1679 | /* If we have no swap space, do not bother scanning anon pages. */ |
Shaohua Li | ec8acf2 | 2013-02-22 16:34:38 -0800 | [diff] [blame] | 1680 | if (!sc->may_swap || (get_nr_swap_pages() <= 0)) { |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 1681 | scan_balance = SCAN_FILE; |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1682 | goto out; |
| 1683 | } |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1684 | |
Johannes Weiner | 10316b3 | 2013-02-22 16:32:14 -0800 | [diff] [blame] | 1685 | /* |
| 1686 | * Global reclaim will swap to prevent OOM even with no |
| 1687 | * swappiness, but memcg users want to use this knob to |
| 1688 | * disable swapping for individual groups completely when |
| 1689 | * using the memory controller's swap limit feature would be |
| 1690 | * too expensive. |
| 1691 | */ |
| 1692 | if (!global_reclaim(sc) && !vmscan_swappiness(sc)) { |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 1693 | scan_balance = SCAN_FILE; |
Johannes Weiner | 10316b3 | 2013-02-22 16:32:14 -0800 | [diff] [blame] | 1694 | goto out; |
| 1695 | } |
| 1696 | |
| 1697 | /* |
| 1698 | * Do not apply any pressure balancing cleverness when the |
| 1699 | * system is close to OOM, scan both anon and file equally |
| 1700 | * (unless the swappiness setting disagrees with swapping). |
| 1701 | */ |
| 1702 | if (!sc->priority && vmscan_swappiness(sc)) { |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 1703 | scan_balance = SCAN_EQUAL; |
Johannes Weiner | 10316b3 | 2013-02-22 16:32:14 -0800 | [diff] [blame] | 1704 | goto out; |
| 1705 | } |
| 1706 | |
Hugh Dickins | 4d7dcca | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 1707 | anon = get_lru_size(lruvec, LRU_ACTIVE_ANON) + |
| 1708 | get_lru_size(lruvec, LRU_INACTIVE_ANON); |
| 1709 | file = get_lru_size(lruvec, LRU_ACTIVE_FILE) + |
| 1710 | get_lru_size(lruvec, LRU_INACTIVE_FILE); |
Johannes Weiner | a4d3e9e | 2011-09-14 16:21:52 -0700 | [diff] [blame] | 1711 | |
Johannes Weiner | 11d16c2 | 2013-02-22 16:32:15 -0800 | [diff] [blame] | 1712 | /* |
| 1713 | * If it's foreseeable that reclaiming the file cache won't be |
| 1714 | * enough to get the zone back into a desirable shape, we have |
| 1715 | * to swap. Better start now and leave the - probably heavily |
| 1716 | * thrashing - remaining file pages alone. |
| 1717 | */ |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 1718 | if (global_reclaim(sc)) { |
Johannes Weiner | 11d16c2 | 2013-02-22 16:32:15 -0800 | [diff] [blame] | 1719 | free = zone_page_state(zone, NR_FREE_PAGES); |
Konstantin Khlebnikov | 9012637 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 1720 | if (unlikely(file + free <= high_wmark_pages(zone))) { |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 1721 | scan_balance = SCAN_ANON; |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1722 | goto out; |
KOSAKI Motohiro | eeee9a8 | 2009-01-07 18:08:17 -0800 | [diff] [blame] | 1723 | } |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1724 | } |
| 1725 | |
| 1726 | /* |
Johannes Weiner | 7c5bd70 | 2013-02-22 16:32:10 -0800 | [diff] [blame] | 1727 | * There is enough inactive page cache, do not reclaim |
| 1728 | * anything from the anonymous working set right now. |
| 1729 | */ |
| 1730 | if (!inactive_file_is_low(lruvec)) { |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 1731 | scan_balance = SCAN_FILE; |
Johannes Weiner | 7c5bd70 | 2013-02-22 16:32:10 -0800 | [diff] [blame] | 1732 | goto out; |
| 1733 | } |
| 1734 | |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 1735 | scan_balance = SCAN_FRACT; |
| 1736 | |
Johannes Weiner | 7c5bd70 | 2013-02-22 16:32:10 -0800 | [diff] [blame] | 1737 | /* |
KOSAKI Motohiro | 58c37f6 | 2010-08-09 17:19:51 -0700 | [diff] [blame] | 1738 | * With swappiness at 100, anonymous and file have the same priority. |
| 1739 | * This scanning priority is essentially the inverse of IO cost. |
| 1740 | */ |
Konstantin Khlebnikov | 3d58ab5 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1741 | anon_prio = vmscan_swappiness(sc); |
Hugh Dickins | 75b00af | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1742 | file_prio = 200 - anon_prio; |
KOSAKI Motohiro | 58c37f6 | 2010-08-09 17:19:51 -0700 | [diff] [blame] | 1743 | |
| 1744 | /* |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1745 | * OK, so we have swap space and a fair amount of page cache |
| 1746 | * pages. We use the recently rotated / recently scanned |
| 1747 | * ratios to determine how valuable each cache is. |
| 1748 | * |
| 1749 | * Because workloads change over time (and to avoid overflow) |
| 1750 | * we keep these statistics as a floating average, which ends |
| 1751 | * up weighing recent references more than old ones. |
| 1752 | * |
| 1753 | * anon in [0], file in [1] |
| 1754 | */ |
Konstantin Khlebnikov | 9012637 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 1755 | spin_lock_irq(&zone->lru_lock); |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1756 | if (unlikely(reclaim_stat->recent_scanned[0] > anon / 4)) { |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1757 | reclaim_stat->recent_scanned[0] /= 2; |
| 1758 | reclaim_stat->recent_rotated[0] /= 2; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1759 | } |
| 1760 | |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1761 | if (unlikely(reclaim_stat->recent_scanned[1] > file / 4)) { |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1762 | reclaim_stat->recent_scanned[1] /= 2; |
| 1763 | reclaim_stat->recent_rotated[1] /= 2; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1764 | } |
| 1765 | |
| 1766 | /* |
Rik van Riel | 00d8089 | 2008-11-19 15:36:44 -0800 | [diff] [blame] | 1767 | * The amount of pressure on anon vs file pages is inversely |
| 1768 | * proportional to the fraction of recently scanned pages on |
| 1769 | * each list that were recently referenced and in active use. |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1770 | */ |
Satoru Moriya | fe35004 | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 1771 | ap = anon_prio * (reclaim_stat->recent_scanned[0] + 1); |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1772 | ap /= reclaim_stat->recent_rotated[0] + 1; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1773 | |
Satoru Moriya | fe35004 | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 1774 | fp = file_prio * (reclaim_stat->recent_scanned[1] + 1); |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 1775 | fp /= reclaim_stat->recent_rotated[1] + 1; |
Konstantin Khlebnikov | 9012637 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 1776 | spin_unlock_irq(&zone->lru_lock); |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1777 | |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1778 | fraction[0] = ap; |
| 1779 | fraction[1] = fp; |
| 1780 | denominator = ap + fp + 1; |
| 1781 | out: |
Hugh Dickins | 4111304 | 2012-01-12 17:20:01 -0800 | [diff] [blame] | 1782 | for_each_evictable_lru(lru) { |
| 1783 | int file = is_file_lru(lru); |
Johannes Weiner | d778df5 | 2013-02-22 16:32:12 -0800 | [diff] [blame] | 1784 | unsigned long size; |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1785 | unsigned long scan; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1786 | |
Johannes Weiner | d778df5 | 2013-02-22 16:32:12 -0800 | [diff] [blame] | 1787 | size = get_lru_size(lruvec, lru); |
Johannes Weiner | 10316b3 | 2013-02-22 16:32:14 -0800 | [diff] [blame] | 1788 | scan = size >> sc->priority; |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 1789 | |
Johannes Weiner | 10316b3 | 2013-02-22 16:32:14 -0800 | [diff] [blame] | 1790 | if (!scan && force_scan) |
| 1791 | scan = min(size, SWAP_CLUSTER_MAX); |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 1792 | |
| 1793 | switch (scan_balance) { |
| 1794 | case SCAN_EQUAL: |
| 1795 | /* Scan lists relative to size */ |
| 1796 | break; |
| 1797 | case SCAN_FRACT: |
| 1798 | /* |
| 1799 | * Scan types proportional to swappiness and |
| 1800 | * their relative recent reclaim efficiency. |
| 1801 | */ |
| 1802 | scan = div64_u64(scan * fraction[file], denominator); |
| 1803 | break; |
| 1804 | case SCAN_FILE: |
| 1805 | case SCAN_ANON: |
| 1806 | /* Scan one type exclusively */ |
| 1807 | if ((scan_balance == SCAN_FILE) != file) |
| 1808 | scan = 0; |
| 1809 | break; |
| 1810 | default: |
| 1811 | /* Look ma, no brain */ |
| 1812 | BUG(); |
| 1813 | } |
Hugh Dickins | 4111304 | 2012-01-12 17:20:01 -0800 | [diff] [blame] | 1814 | nr[lru] = scan; |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 1815 | } |
Wu Fengguang | 6e08a36 | 2009-06-16 15:32:29 -0700 | [diff] [blame] | 1816 | } |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1817 | |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 1818 | /* |
| 1819 | * This is a basic per-zone page freer. Used by both kswapd and direct reclaim. |
| 1820 | */ |
| 1821 | static void shrink_lruvec(struct lruvec *lruvec, struct scan_control *sc) |
| 1822 | { |
| 1823 | unsigned long nr[NR_LRU_LISTS]; |
| 1824 | unsigned long nr_to_scan; |
| 1825 | enum lru_list lru; |
| 1826 | unsigned long nr_reclaimed = 0; |
| 1827 | unsigned long nr_to_reclaim = sc->nr_to_reclaim; |
| 1828 | struct blk_plug plug; |
| 1829 | |
| 1830 | get_scan_count(lruvec, sc, nr); |
| 1831 | |
| 1832 | blk_start_plug(&plug); |
| 1833 | while (nr[LRU_INACTIVE_ANON] || nr[LRU_ACTIVE_FILE] || |
| 1834 | nr[LRU_INACTIVE_FILE]) { |
| 1835 | for_each_evictable_lru(lru) { |
| 1836 | if (nr[lru]) { |
| 1837 | nr_to_scan = min(nr[lru], SWAP_CLUSTER_MAX); |
| 1838 | nr[lru] -= nr_to_scan; |
| 1839 | |
| 1840 | nr_reclaimed += shrink_list(lru, nr_to_scan, |
| 1841 | lruvec, sc); |
| 1842 | } |
| 1843 | } |
| 1844 | /* |
| 1845 | * On large memory systems, scan >> priority can become |
| 1846 | * really large. This is fine for the starting priority; |
| 1847 | * we want to put equal scanning pressure on each zone. |
| 1848 | * However, if the VM has a harder time of freeing pages, |
| 1849 | * with multiple processes reclaiming pages, the total |
| 1850 | * freeing target can get unreasonably large. |
| 1851 | */ |
| 1852 | if (nr_reclaimed >= nr_to_reclaim && |
| 1853 | sc->priority < DEF_PRIORITY) |
| 1854 | break; |
| 1855 | } |
| 1856 | blk_finish_plug(&plug); |
| 1857 | sc->nr_reclaimed += nr_reclaimed; |
| 1858 | |
| 1859 | /* |
| 1860 | * Even if we did not try to evict anon pages at all, we want to |
| 1861 | * rebalance the anon lru active/inactive ratio. |
| 1862 | */ |
| 1863 | if (inactive_anon_is_low(lruvec)) |
| 1864 | shrink_active_list(SWAP_CLUSTER_MAX, lruvec, |
| 1865 | sc, LRU_ACTIVE_ANON); |
| 1866 | |
| 1867 | throttle_vm_writeout(sc->gfp_mask); |
| 1868 | } |
| 1869 | |
Mel Gorman | 23b9da5 | 2012-05-29 15:06:20 -0700 | [diff] [blame] | 1870 | /* Use reclaim/compaction for costly allocs or under memory pressure */ |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1871 | static bool in_reclaim_compaction(struct scan_control *sc) |
Mel Gorman | 23b9da5 | 2012-05-29 15:06:20 -0700 | [diff] [blame] | 1872 | { |
Kirill A. Shutemov | d84da3f | 2012-12-11 16:00:31 -0800 | [diff] [blame] | 1873 | if (IS_ENABLED(CONFIG_COMPACTION) && sc->order && |
Mel Gorman | 23b9da5 | 2012-05-29 15:06:20 -0700 | [diff] [blame] | 1874 | (sc->order > PAGE_ALLOC_COSTLY_ORDER || |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1875 | sc->priority < DEF_PRIORITY - 2)) |
Mel Gorman | 23b9da5 | 2012-05-29 15:06:20 -0700 | [diff] [blame] | 1876 | return true; |
| 1877 | |
| 1878 | return false; |
| 1879 | } |
| 1880 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1881 | /* |
Mel Gorman | 23b9da5 | 2012-05-29 15:06:20 -0700 | [diff] [blame] | 1882 | * Reclaim/compaction is used for high-order allocation requests. It reclaims |
| 1883 | * order-0 pages before compacting the zone. should_continue_reclaim() returns |
| 1884 | * true if more pages should be reclaimed such that when the page allocator |
| 1885 | * calls try_to_compact_zone() that it will have enough free pages to succeed. |
| 1886 | * 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] | 1887 | */ |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 1888 | static inline bool should_continue_reclaim(struct zone *zone, |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1889 | unsigned long nr_reclaimed, |
| 1890 | unsigned long nr_scanned, |
| 1891 | struct scan_control *sc) |
| 1892 | { |
| 1893 | unsigned long pages_for_compaction; |
| 1894 | unsigned long inactive_lru_pages; |
| 1895 | |
| 1896 | /* If not in reclaim/compaction mode, stop */ |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1897 | if (!in_reclaim_compaction(sc)) |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1898 | return false; |
| 1899 | |
Mel Gorman | 2876592 | 2011-02-25 14:44:20 -0800 | [diff] [blame] | 1900 | /* Consider stopping depending on scan and reclaim activity */ |
| 1901 | if (sc->gfp_mask & __GFP_REPEAT) { |
| 1902 | /* |
| 1903 | * For __GFP_REPEAT allocations, stop reclaiming if the |
| 1904 | * full LRU list has been scanned and we are still failing |
| 1905 | * to reclaim pages. This full LRU scan is potentially |
| 1906 | * expensive but a __GFP_REPEAT caller really wants to succeed |
| 1907 | */ |
| 1908 | if (!nr_reclaimed && !nr_scanned) |
| 1909 | return false; |
| 1910 | } else { |
| 1911 | /* |
| 1912 | * For non-__GFP_REPEAT allocations which can presumably |
| 1913 | * fail without consequence, stop if we failed to reclaim |
| 1914 | * any pages from the last SWAP_CLUSTER_MAX number of |
| 1915 | * pages that were scanned. This will return to the |
| 1916 | * caller faster at the risk reclaim/compaction and |
| 1917 | * the resulting allocation attempt fails |
| 1918 | */ |
| 1919 | if (!nr_reclaimed) |
| 1920 | return false; |
| 1921 | } |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1922 | |
| 1923 | /* |
| 1924 | * If we have not reclaimed enough pages for compaction and the |
| 1925 | * inactive lists are large enough, continue reclaiming |
| 1926 | */ |
| 1927 | pages_for_compaction = (2UL << sc->order); |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 1928 | inactive_lru_pages = zone_page_state(zone, NR_INACTIVE_FILE); |
Shaohua Li | ec8acf2 | 2013-02-22 16:34:38 -0800 | [diff] [blame] | 1929 | if (get_nr_swap_pages() > 0) |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 1930 | inactive_lru_pages += zone_page_state(zone, NR_INACTIVE_ANON); |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1931 | if (sc->nr_reclaimed < pages_for_compaction && |
| 1932 | inactive_lru_pages > pages_for_compaction) |
| 1933 | return true; |
| 1934 | |
| 1935 | /* If compaction would go ahead or the allocation would succeed, stop */ |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 1936 | switch (compaction_suitable(zone, sc->order)) { |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 1937 | case COMPACT_PARTIAL: |
| 1938 | case COMPACT_CONTINUE: |
| 1939 | return false; |
| 1940 | default: |
| 1941 | return true; |
| 1942 | } |
| 1943 | } |
| 1944 | |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1945 | static void shrink_zone(struct zone *zone, struct scan_control *sc) |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1946 | { |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 1947 | unsigned long nr_reclaimed, nr_scanned; |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1948 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1949 | do { |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 1950 | struct mem_cgroup *root = sc->target_mem_cgroup; |
| 1951 | struct mem_cgroup_reclaim_cookie reclaim = { |
| 1952 | .zone = zone, |
| 1953 | .priority = sc->priority, |
| 1954 | }; |
| 1955 | struct mem_cgroup *memcg; |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1956 | |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 1957 | nr_reclaimed = sc->nr_reclaimed; |
| 1958 | nr_scanned = sc->nr_scanned; |
Konstantin Khlebnikov | f9be23d | 2012-05-29 15:07:02 -0700 | [diff] [blame] | 1959 | |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 1960 | memcg = mem_cgroup_iter(root, NULL, &reclaim); |
| 1961 | do { |
| 1962 | struct lruvec *lruvec; |
| 1963 | |
| 1964 | lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
| 1965 | |
| 1966 | shrink_lruvec(lruvec, sc); |
| 1967 | |
| 1968 | /* |
Michal Hocko | a394cb8 | 2013-02-22 16:32:30 -0800 | [diff] [blame] | 1969 | * Direct reclaim and kswapd have to scan all memory |
| 1970 | * cgroups to fulfill the overall scan target for the |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 1971 | * zone. |
Michal Hocko | a394cb8 | 2013-02-22 16:32:30 -0800 | [diff] [blame] | 1972 | * |
| 1973 | * Limit reclaim, on the other hand, only cares about |
| 1974 | * nr_to_reclaim pages to be reclaimed and it will |
| 1975 | * retry with decreasing priority if one round over the |
| 1976 | * whole hierarchy is not sufficient. |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 1977 | */ |
Michal Hocko | a394cb8 | 2013-02-22 16:32:30 -0800 | [diff] [blame] | 1978 | if (!global_reclaim(sc) && |
| 1979 | sc->nr_reclaimed >= sc->nr_to_reclaim) { |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 1980 | mem_cgroup_iter_break(root, memcg); |
| 1981 | break; |
| 1982 | } |
| 1983 | memcg = mem_cgroup_iter(root, memcg, &reclaim); |
| 1984 | } while (memcg); |
| 1985 | } while (should_continue_reclaim(zone, sc->nr_reclaimed - nr_reclaimed, |
| 1986 | sc->nr_scanned - nr_scanned, sc)); |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1987 | } |
| 1988 | |
Mel Gorman | fe4b1b2 | 2012-01-12 17:19:45 -0800 | [diff] [blame] | 1989 | /* Returns true if compaction should go ahead for a high-order request */ |
| 1990 | static inline bool compaction_ready(struct zone *zone, struct scan_control *sc) |
| 1991 | { |
| 1992 | unsigned long balance_gap, watermark; |
| 1993 | bool watermark_ok; |
| 1994 | |
| 1995 | /* Do not consider compaction for orders reclaim is meant to satisfy */ |
| 1996 | if (sc->order <= PAGE_ALLOC_COSTLY_ORDER) |
| 1997 | return false; |
| 1998 | |
| 1999 | /* |
| 2000 | * Compaction takes time to run and there are potentially other |
| 2001 | * callers using the pages just freed. Continue reclaiming until |
| 2002 | * there is a buffer of free pages available to give compaction |
| 2003 | * a reasonable chance of completing and allocating the page |
| 2004 | */ |
| 2005 | balance_gap = min(low_wmark_pages(zone), |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 2006 | (zone->managed_pages + KSWAPD_ZONE_BALANCE_GAP_RATIO-1) / |
Mel Gorman | fe4b1b2 | 2012-01-12 17:19:45 -0800 | [diff] [blame] | 2007 | KSWAPD_ZONE_BALANCE_GAP_RATIO); |
| 2008 | watermark = high_wmark_pages(zone) + balance_gap + (2UL << sc->order); |
| 2009 | watermark_ok = zone_watermark_ok_safe(zone, 0, watermark, 0, 0); |
| 2010 | |
| 2011 | /* |
| 2012 | * If compaction is deferred, reclaim up to a point where |
| 2013 | * compaction will have a chance of success when re-enabled |
| 2014 | */ |
Rik van Riel | aff6224 | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 2015 | if (compaction_deferred(zone, sc->order)) |
Mel Gorman | fe4b1b2 | 2012-01-12 17:19:45 -0800 | [diff] [blame] | 2016 | return watermark_ok; |
| 2017 | |
| 2018 | /* If compaction is not ready to start, keep reclaiming */ |
| 2019 | if (!compaction_suitable(zone, sc->order)) |
| 2020 | return false; |
| 2021 | |
| 2022 | return watermark_ok; |
| 2023 | } |
| 2024 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2025 | /* |
| 2026 | * This is the direct reclaim path, for page-allocating processes. We only |
| 2027 | * try to reclaim pages from zones which will satisfy the caller's allocation |
| 2028 | * request. |
| 2029 | * |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 2030 | * We reclaim from a zone even if that zone is over high_wmark_pages(zone). |
| 2031 | * Because: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2032 | * a) The caller may be trying to free *extra* pages to satisfy a higher-order |
| 2033 | * allocation or |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 2034 | * b) The target zone may be at high_wmark_pages(zone) but the lower zones |
| 2035 | * must go *over* high_wmark_pages(zone) to satisfy the `incremental min' |
| 2036 | * zone defense algorithm. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2037 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2038 | * If a zone is deemed to be full of pinned pages then just give it a light |
| 2039 | * scan then give up on it. |
Mel Gorman | e0c2327 | 2011-10-31 17:09:33 -0700 | [diff] [blame] | 2040 | * |
| 2041 | * 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] | 2042 | * high-order allocation and compaction is ready to begin. This indicates to |
Mel Gorman | 0cee34f | 2012-01-12 17:19:49 -0800 | [diff] [blame] | 2043 | * the caller that it should consider retrying the allocation instead of |
| 2044 | * further reclaim. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2045 | */ |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2046 | static bool shrink_zones(struct zonelist *zonelist, struct scan_control *sc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2047 | { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 2048 | struct zoneref *z; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2049 | struct zone *zone; |
Ying Han | d149e3b | 2011-05-26 16:25:27 -0700 | [diff] [blame] | 2050 | unsigned long nr_soft_reclaimed; |
| 2051 | unsigned long nr_soft_scanned; |
Mel Gorman | 0cee34f | 2012-01-12 17:19:49 -0800 | [diff] [blame] | 2052 | bool aborted_reclaim = false; |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 2053 | |
Mel Gorman | cc715d9 | 2012-03-21 16:34:00 -0700 | [diff] [blame] | 2054 | /* |
| 2055 | * If the number of buffer_heads in the machine exceeds the maximum |
| 2056 | * allowed level, force direct reclaim to scan the highmem zone as |
| 2057 | * highmem pages could be pinning lowmem pages storing buffer_heads |
| 2058 | */ |
| 2059 | if (buffer_heads_over_limit) |
| 2060 | sc->gfp_mask |= __GFP_HIGHMEM; |
| 2061 | |
Mel Gorman | d4debc6 | 2010-08-09 17:19:29 -0700 | [diff] [blame] | 2062 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
| 2063 | gfp_zone(sc->gfp_mask), sc->nodemask) { |
Con Kolivas | f3fe651 | 2006-01-06 00:11:15 -0800 | [diff] [blame] | 2064 | if (!populated_zone(zone)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2065 | continue; |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 2066 | /* |
| 2067 | * Take care memory controller reclaiming has small influence |
| 2068 | * to global LRU. |
| 2069 | */ |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 2070 | if (global_reclaim(sc)) { |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 2071 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
| 2072 | continue; |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2073 | if (zone->all_unreclaimable && |
| 2074 | sc->priority != DEF_PRIORITY) |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 2075 | continue; /* Let kswapd poll it */ |
Kirill A. Shutemov | d84da3f | 2012-12-11 16:00:31 -0800 | [diff] [blame] | 2076 | if (IS_ENABLED(CONFIG_COMPACTION)) { |
Rik van Riel | e0887c1 | 2011-10-31 17:09:31 -0700 | [diff] [blame] | 2077 | /* |
Mel Gorman | e0c2327 | 2011-10-31 17:09:33 -0700 | [diff] [blame] | 2078 | * If we already have plenty of memory free for |
| 2079 | * compaction in this zone, don't free any more. |
| 2080 | * Even though compaction is invoked for any |
| 2081 | * non-zero order, only frequent costly order |
| 2082 | * reclamation is disruptive enough to become a |
Copot Alexandru | c7cfa37 | 2012-03-21 16:34:10 -0700 | [diff] [blame] | 2083 | * noticeable problem, like transparent huge |
| 2084 | * page allocations. |
Rik van Riel | e0887c1 | 2011-10-31 17:09:31 -0700 | [diff] [blame] | 2085 | */ |
Mel Gorman | fe4b1b2 | 2012-01-12 17:19:45 -0800 | [diff] [blame] | 2086 | if (compaction_ready(zone, sc)) { |
Mel Gorman | 0cee34f | 2012-01-12 17:19:49 -0800 | [diff] [blame] | 2087 | aborted_reclaim = true; |
Rik van Riel | e0887c1 | 2011-10-31 17:09:31 -0700 | [diff] [blame] | 2088 | continue; |
Mel Gorman | e0c2327 | 2011-10-31 17:09:33 -0700 | [diff] [blame] | 2089 | } |
Rik van Riel | e0887c1 | 2011-10-31 17:09:31 -0700 | [diff] [blame] | 2090 | } |
KAMEZAWA Hiroyuki | ac34a1a | 2011-06-27 16:18:12 -0700 | [diff] [blame] | 2091 | /* |
| 2092 | * This steals pages from memory cgroups over softlimit |
| 2093 | * and returns the number of reclaimed pages and |
| 2094 | * scanned pages. This works for global memory pressure |
| 2095 | * and balancing, not for a memcg's limit. |
| 2096 | */ |
| 2097 | nr_soft_scanned = 0; |
| 2098 | nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone, |
| 2099 | sc->order, sc->gfp_mask, |
| 2100 | &nr_soft_scanned); |
| 2101 | sc->nr_reclaimed += nr_soft_reclaimed; |
| 2102 | sc->nr_scanned += nr_soft_scanned; |
| 2103 | /* need some check for avoid more shrink_zone() */ |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 2104 | } |
Nick Piggin | 408d854 | 2006-09-25 23:31:27 -0700 | [diff] [blame] | 2105 | |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2106 | shrink_zone(zone, sc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2107 | } |
Mel Gorman | e0c2327 | 2011-10-31 17:09:33 -0700 | [diff] [blame] | 2108 | |
Mel Gorman | 0cee34f | 2012-01-12 17:19:49 -0800 | [diff] [blame] | 2109 | return aborted_reclaim; |
Minchan Kim | d190836 | 2010-09-22 13:05:01 -0700 | [diff] [blame] | 2110 | } |
| 2111 | |
| 2112 | static bool zone_reclaimable(struct zone *zone) |
| 2113 | { |
| 2114 | return zone->pages_scanned < zone_reclaimable_pages(zone) * 6; |
| 2115 | } |
| 2116 | |
KOSAKI Motohiro | 929bea7 | 2011-04-14 15:22:12 -0700 | [diff] [blame] | 2117 | /* All zones in zonelist are unreclaimable? */ |
Minchan Kim | d190836 | 2010-09-22 13:05:01 -0700 | [diff] [blame] | 2118 | static bool all_unreclaimable(struct zonelist *zonelist, |
| 2119 | struct scan_control *sc) |
| 2120 | { |
| 2121 | struct zoneref *z; |
| 2122 | struct zone *zone; |
Minchan Kim | d190836 | 2010-09-22 13:05:01 -0700 | [diff] [blame] | 2123 | |
| 2124 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
| 2125 | gfp_zone(sc->gfp_mask), sc->nodemask) { |
| 2126 | if (!populated_zone(zone)) |
| 2127 | continue; |
| 2128 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
| 2129 | continue; |
KOSAKI Motohiro | 929bea7 | 2011-04-14 15:22:12 -0700 | [diff] [blame] | 2130 | if (!zone->all_unreclaimable) |
| 2131 | return false; |
Minchan Kim | d190836 | 2010-09-22 13:05:01 -0700 | [diff] [blame] | 2132 | } |
| 2133 | |
KOSAKI Motohiro | 929bea7 | 2011-04-14 15:22:12 -0700 | [diff] [blame] | 2134 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2135 | } |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2136 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2137 | /* |
| 2138 | * This is the main entry point to direct page reclaim. |
| 2139 | * |
| 2140 | * If a full scan of the inactive list fails to free enough memory then we |
| 2141 | * are "out of memory" and something needs to be killed. |
| 2142 | * |
| 2143 | * If the caller is !__GFP_FS then the probability of a failure is reasonably |
| 2144 | * 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] | 2145 | * caller can't do much about. We kick the writeback threads and take explicit |
| 2146 | * naps in the hope that some of these pages can be written. But if the |
| 2147 | * allocating task holds filesystem locks which prevent writeout this might not |
| 2148 | * work, and the allocation attempt will fail. |
Nishanth Aravamudan | a41f24e | 2008-04-29 00:58:25 -0700 | [diff] [blame] | 2149 | * |
| 2150 | * returns: 0, if no pages reclaimed |
| 2151 | * else, the number of pages reclaimed |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2152 | */ |
Mel Gorman | dac1d27 | 2008-04-28 02:12:12 -0700 | [diff] [blame] | 2153 | static unsigned long do_try_to_free_pages(struct zonelist *zonelist, |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 2154 | struct scan_control *sc, |
| 2155 | struct shrink_control *shrink) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2156 | { |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 2157 | unsigned long total_scanned = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2158 | struct reclaim_state *reclaim_state = current->reclaim_state; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 2159 | struct zoneref *z; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2160 | struct zone *zone; |
KOSAKI Motohiro | 22fba33 | 2009-12-14 17:59:10 -0800 | [diff] [blame] | 2161 | unsigned long writeback_threshold; |
Mel Gorman | 0cee34f | 2012-01-12 17:19:49 -0800 | [diff] [blame] | 2162 | bool aborted_reclaim; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2163 | |
Keika Kobayashi | 873b477 | 2008-07-25 01:48:52 -0700 | [diff] [blame] | 2164 | delayacct_freepages_start(); |
| 2165 | |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 2166 | if (global_reclaim(sc)) |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 2167 | count_vm_event(ALLOCSTALL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2168 | |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2169 | do { |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2170 | sc->nr_scanned = 0; |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2171 | aborted_reclaim = shrink_zones(zonelist, sc); |
Mel Gorman | e0c2327 | 2011-10-31 17:09:33 -0700 | [diff] [blame] | 2172 | |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2173 | /* |
| 2174 | * Don't shrink slabs when reclaiming memory from |
| 2175 | * over limit cgroups |
| 2176 | */ |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 2177 | if (global_reclaim(sc)) { |
KOSAKI Motohiro | c6a8a8c | 2010-08-09 17:19:14 -0700 | [diff] [blame] | 2178 | unsigned long lru_pages = 0; |
Mel Gorman | d4debc6 | 2010-08-09 17:19:29 -0700 | [diff] [blame] | 2179 | for_each_zone_zonelist(zone, z, zonelist, |
| 2180 | gfp_zone(sc->gfp_mask)) { |
KOSAKI Motohiro | c6a8a8c | 2010-08-09 17:19:14 -0700 | [diff] [blame] | 2181 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
| 2182 | continue; |
| 2183 | |
| 2184 | lru_pages += zone_reclaimable_pages(zone); |
| 2185 | } |
| 2186 | |
Ying Han | 1495f23 | 2011-05-24 17:12:27 -0700 | [diff] [blame] | 2187 | shrink_slab(shrink, sc->nr_scanned, lru_pages); |
KAMEZAWA Hiroyuki | 91a4547 | 2008-02-07 00:14:29 -0800 | [diff] [blame] | 2188 | if (reclaim_state) { |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2189 | sc->nr_reclaimed += reclaim_state->reclaimed_slab; |
KAMEZAWA Hiroyuki | 91a4547 | 2008-02-07 00:14:29 -0800 | [diff] [blame] | 2190 | reclaim_state->reclaimed_slab = 0; |
| 2191 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2192 | } |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2193 | total_scanned += sc->nr_scanned; |
KOSAKI Motohiro | bb21c7c | 2010-06-04 14:15:05 -0700 | [diff] [blame] | 2194 | if (sc->nr_reclaimed >= sc->nr_to_reclaim) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2195 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2196 | |
| 2197 | /* |
Minchan Kim | 0e50ce3 | 2013-02-22 16:35:37 -0800 | [diff] [blame] | 2198 | * If we're getting trouble reclaiming, start doing |
| 2199 | * writepage even in laptop mode. |
| 2200 | */ |
| 2201 | if (sc->priority < DEF_PRIORITY - 2) |
| 2202 | sc->may_writepage = 1; |
| 2203 | |
| 2204 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2205 | * Try to write back as many pages as we just scanned. This |
| 2206 | * tends to cause slow streaming writers to write data to the |
| 2207 | * disk smoothly, at the dirtying rate, which is nice. But |
| 2208 | * that's undesirable in laptop mode, where we *want* lumpy |
| 2209 | * writeout. So in laptop mode, write out the whole world. |
| 2210 | */ |
KOSAKI Motohiro | 22fba33 | 2009-12-14 17:59:10 -0800 | [diff] [blame] | 2211 | writeback_threshold = sc->nr_to_reclaim + sc->nr_to_reclaim / 2; |
| 2212 | if (total_scanned > writeback_threshold) { |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 2213 | wakeup_flusher_threads(laptop_mode ? 0 : total_scanned, |
| 2214 | WB_REASON_TRY_TO_FREE_PAGES); |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2215 | sc->may_writepage = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2216 | } |
| 2217 | |
| 2218 | /* Take a nap, wait for some writeback to complete */ |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 2219 | if (!sc->hibernation_mode && sc->nr_scanned && |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2220 | sc->priority < DEF_PRIORITY - 2) { |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 2221 | struct zone *preferred_zone; |
| 2222 | |
| 2223 | first_zones_zonelist(zonelist, gfp_zone(sc->gfp_mask), |
David Rientjes | f33261d | 2011-01-25 15:07:20 -0800 | [diff] [blame] | 2224 | &cpuset_current_mems_allowed, |
| 2225 | &preferred_zone); |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 2226 | wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/10); |
| 2227 | } |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2228 | } while (--sc->priority >= 0); |
KOSAKI Motohiro | bb21c7c | 2010-06-04 14:15:05 -0700 | [diff] [blame] | 2229 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2230 | out: |
Keika Kobayashi | 873b477 | 2008-07-25 01:48:52 -0700 | [diff] [blame] | 2231 | delayacct_freepages_end(); |
| 2232 | |
KOSAKI Motohiro | bb21c7c | 2010-06-04 14:15:05 -0700 | [diff] [blame] | 2233 | if (sc->nr_reclaimed) |
| 2234 | return sc->nr_reclaimed; |
| 2235 | |
KOSAKI Motohiro | 929bea7 | 2011-04-14 15:22:12 -0700 | [diff] [blame] | 2236 | /* |
| 2237 | * As hibernation is going on, kswapd is freezed so that it can't mark |
| 2238 | * the zone into all_unreclaimable. Thus bypassing all_unreclaimable |
| 2239 | * check. |
| 2240 | */ |
| 2241 | if (oom_killer_disabled) |
| 2242 | return 0; |
| 2243 | |
Mel Gorman | 0cee34f | 2012-01-12 17:19:49 -0800 | [diff] [blame] | 2244 | /* Aborted reclaim to try compaction? don't OOM, then */ |
| 2245 | if (aborted_reclaim) |
Mel Gorman | 7335084 | 2012-01-12 17:19:33 -0800 | [diff] [blame] | 2246 | return 1; |
| 2247 | |
KOSAKI Motohiro | bb21c7c | 2010-06-04 14:15:05 -0700 | [diff] [blame] | 2248 | /* 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] | 2249 | if (global_reclaim(sc) && !all_unreclaimable(zonelist, sc)) |
KOSAKI Motohiro | bb21c7c | 2010-06-04 14:15:05 -0700 | [diff] [blame] | 2250 | return 1; |
| 2251 | |
| 2252 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2253 | } |
| 2254 | |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2255 | static bool pfmemalloc_watermark_ok(pg_data_t *pgdat) |
| 2256 | { |
| 2257 | struct zone *zone; |
| 2258 | unsigned long pfmemalloc_reserve = 0; |
| 2259 | unsigned long free_pages = 0; |
| 2260 | int i; |
| 2261 | bool wmark_ok; |
| 2262 | |
| 2263 | for (i = 0; i <= ZONE_NORMAL; i++) { |
| 2264 | zone = &pgdat->node_zones[i]; |
| 2265 | pfmemalloc_reserve += min_wmark_pages(zone); |
| 2266 | free_pages += zone_page_state(zone, NR_FREE_PAGES); |
| 2267 | } |
| 2268 | |
| 2269 | wmark_ok = free_pages > pfmemalloc_reserve / 2; |
| 2270 | |
| 2271 | /* kswapd must be awake if processes are being throttled */ |
| 2272 | if (!wmark_ok && waitqueue_active(&pgdat->kswapd_wait)) { |
| 2273 | pgdat->classzone_idx = min(pgdat->classzone_idx, |
| 2274 | (enum zone_type)ZONE_NORMAL); |
| 2275 | wake_up_interruptible(&pgdat->kswapd_wait); |
| 2276 | } |
| 2277 | |
| 2278 | return wmark_ok; |
| 2279 | } |
| 2280 | |
| 2281 | /* |
| 2282 | * Throttle direct reclaimers if backing storage is backed by the network |
| 2283 | * and the PFMEMALLOC reserve for the preferred node is getting dangerously |
| 2284 | * depleted. kswapd will continue to make progress and wake the processes |
Mel Gorman | 50694c2 | 2012-11-26 16:29:48 -0800 | [diff] [blame] | 2285 | * when the low watermark is reached. |
| 2286 | * |
| 2287 | * Returns true if a fatal signal was delivered during throttling. If this |
| 2288 | * happens, the page allocator should not consider triggering the OOM killer. |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2289 | */ |
Mel Gorman | 50694c2 | 2012-11-26 16:29:48 -0800 | [diff] [blame] | 2290 | static bool throttle_direct_reclaim(gfp_t gfp_mask, struct zonelist *zonelist, |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2291 | nodemask_t *nodemask) |
| 2292 | { |
| 2293 | struct zone *zone; |
| 2294 | int high_zoneidx = gfp_zone(gfp_mask); |
| 2295 | pg_data_t *pgdat; |
| 2296 | |
| 2297 | /* |
| 2298 | * Kernel threads should not be throttled as they may be indirectly |
| 2299 | * responsible for cleaning pages necessary for reclaim to make forward |
| 2300 | * progress. kjournald for example may enter direct reclaim while |
| 2301 | * committing a transaction where throttling it could forcing other |
| 2302 | * processes to block on log_wait_commit(). |
| 2303 | */ |
| 2304 | if (current->flags & PF_KTHREAD) |
Mel Gorman | 50694c2 | 2012-11-26 16:29:48 -0800 | [diff] [blame] | 2305 | goto out; |
| 2306 | |
| 2307 | /* |
| 2308 | * If a fatal signal is pending, this process should not throttle. |
| 2309 | * It should return quickly so it can exit and free its memory |
| 2310 | */ |
| 2311 | if (fatal_signal_pending(current)) |
| 2312 | goto out; |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2313 | |
| 2314 | /* Check if the pfmemalloc reserves are ok */ |
| 2315 | first_zones_zonelist(zonelist, high_zoneidx, NULL, &zone); |
| 2316 | pgdat = zone->zone_pgdat; |
| 2317 | if (pfmemalloc_watermark_ok(pgdat)) |
Mel Gorman | 50694c2 | 2012-11-26 16:29:48 -0800 | [diff] [blame] | 2318 | goto out; |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2319 | |
Mel Gorman | 68243e7 | 2012-07-31 16:44:39 -0700 | [diff] [blame] | 2320 | /* Account for the throttling */ |
| 2321 | count_vm_event(PGSCAN_DIRECT_THROTTLE); |
| 2322 | |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2323 | /* |
| 2324 | * If the caller cannot enter the filesystem, it's possible that it |
| 2325 | * is due to the caller holding an FS lock or performing a journal |
| 2326 | * transaction in the case of a filesystem like ext[3|4]. In this case, |
| 2327 | * it is not safe to block on pfmemalloc_wait as kswapd could be |
| 2328 | * blocked waiting on the same lock. Instead, throttle for up to a |
| 2329 | * second before continuing. |
| 2330 | */ |
| 2331 | if (!(gfp_mask & __GFP_FS)) { |
| 2332 | wait_event_interruptible_timeout(pgdat->pfmemalloc_wait, |
| 2333 | pfmemalloc_watermark_ok(pgdat), HZ); |
Mel Gorman | 50694c2 | 2012-11-26 16:29:48 -0800 | [diff] [blame] | 2334 | |
| 2335 | goto check_pending; |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2336 | } |
| 2337 | |
| 2338 | /* Throttle until kswapd wakes the process */ |
| 2339 | wait_event_killable(zone->zone_pgdat->pfmemalloc_wait, |
| 2340 | pfmemalloc_watermark_ok(pgdat)); |
Mel Gorman | 50694c2 | 2012-11-26 16:29:48 -0800 | [diff] [blame] | 2341 | |
| 2342 | check_pending: |
| 2343 | if (fatal_signal_pending(current)) |
| 2344 | return true; |
| 2345 | |
| 2346 | out: |
| 2347 | return false; |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2348 | } |
| 2349 | |
Mel Gorman | dac1d27 | 2008-04-28 02:12:12 -0700 | [diff] [blame] | 2350 | unsigned long try_to_free_pages(struct zonelist *zonelist, int order, |
KAMEZAWA Hiroyuki | 327c0e9 | 2009-03-31 15:23:31 -0700 | [diff] [blame] | 2351 | gfp_t gfp_mask, nodemask_t *nodemask) |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2352 | { |
Mel Gorman | 33906bc | 2010-08-09 17:19:16 -0700 | [diff] [blame] | 2353 | unsigned long nr_reclaimed; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2354 | struct scan_control sc = { |
Ming Lei | 21caf2f | 2013-02-22 16:34:08 -0800 | [diff] [blame] | 2355 | .gfp_mask = (gfp_mask = memalloc_noio_flags(gfp_mask)), |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2356 | .may_writepage = !laptop_mode, |
KOSAKI Motohiro | 22fba33 | 2009-12-14 17:59:10 -0800 | [diff] [blame] | 2357 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
Johannes Weiner | a6dc60f8 | 2009-03-31 15:19:30 -0700 | [diff] [blame] | 2358 | .may_unmap = 1, |
KOSAKI Motohiro | 2e2e425 | 2009-04-21 12:24:57 -0700 | [diff] [blame] | 2359 | .may_swap = 1, |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2360 | .order = order, |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2361 | .priority = DEF_PRIORITY, |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 2362 | .target_mem_cgroup = NULL, |
KAMEZAWA Hiroyuki | 327c0e9 | 2009-03-31 15:23:31 -0700 | [diff] [blame] | 2363 | .nodemask = nodemask, |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2364 | }; |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 2365 | struct shrink_control shrink = { |
| 2366 | .gfp_mask = sc.gfp_mask, |
| 2367 | }; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2368 | |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2369 | /* |
Mel Gorman | 50694c2 | 2012-11-26 16:29:48 -0800 | [diff] [blame] | 2370 | * Do not enter reclaim if fatal signal was delivered while throttled. |
| 2371 | * 1 is returned so that the page allocator does not OOM kill at this |
| 2372 | * point. |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2373 | */ |
Mel Gorman | 50694c2 | 2012-11-26 16:29:48 -0800 | [diff] [blame] | 2374 | if (throttle_direct_reclaim(gfp_mask, zonelist, nodemask)) |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2375 | return 1; |
| 2376 | |
Mel Gorman | 33906bc | 2010-08-09 17:19:16 -0700 | [diff] [blame] | 2377 | trace_mm_vmscan_direct_reclaim_begin(order, |
| 2378 | sc.may_writepage, |
| 2379 | gfp_mask); |
| 2380 | |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 2381 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink); |
Mel Gorman | 33906bc | 2010-08-09 17:19:16 -0700 | [diff] [blame] | 2382 | |
| 2383 | trace_mm_vmscan_direct_reclaim_end(nr_reclaimed); |
| 2384 | |
| 2385 | return nr_reclaimed; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2386 | } |
| 2387 | |
Andrew Morton | c255a45 | 2012-07-31 16:43:02 -0700 | [diff] [blame] | 2388 | #ifdef CONFIG_MEMCG |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2389 | |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 2390 | unsigned long mem_cgroup_shrink_node_zone(struct mem_cgroup *memcg, |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2391 | gfp_t gfp_mask, bool noswap, |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 2392 | struct zone *zone, |
| 2393 | unsigned long *nr_scanned) |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2394 | { |
| 2395 | struct scan_control sc = { |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 2396 | .nr_scanned = 0, |
KOSAKI Motohiro | b8f5c56 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2397 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2398 | .may_writepage = !laptop_mode, |
| 2399 | .may_unmap = 1, |
| 2400 | .may_swap = !noswap, |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2401 | .order = 0, |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2402 | .priority = 0, |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 2403 | .target_mem_cgroup = memcg, |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2404 | }; |
Konstantin Khlebnikov | f9be23d | 2012-05-29 15:07:02 -0700 | [diff] [blame] | 2405 | struct lruvec *lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 2406 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2407 | sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | |
| 2408 | (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK); |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 2409 | |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2410 | trace_mm_vmscan_memcg_softlimit_reclaim_begin(sc.order, |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 2411 | sc.may_writepage, |
| 2412 | sc.gfp_mask); |
| 2413 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2414 | /* |
| 2415 | * NOTE: Although we can get the priority field, using it |
| 2416 | * here is not a good idea, since it limits the pages we can scan. |
| 2417 | * if we don't reclaim here, the shrink_zone from balance_pgdat |
| 2418 | * will pick up pages from other mem cgroup's as well. We hack |
| 2419 | * the priority and make it zero. |
| 2420 | */ |
Konstantin Khlebnikov | f9be23d | 2012-05-29 15:07:02 -0700 | [diff] [blame] | 2421 | shrink_lruvec(lruvec, &sc); |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 2422 | |
| 2423 | trace_mm_vmscan_memcg_softlimit_reclaim_end(sc.nr_reclaimed); |
| 2424 | |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 2425 | *nr_scanned = sc.nr_scanned; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2426 | return sc.nr_reclaimed; |
| 2427 | } |
| 2428 | |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 2429 | unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *memcg, |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 2430 | gfp_t gfp_mask, |
Johannes Weiner | 185efc0 | 2011-09-14 16:21:58 -0700 | [diff] [blame] | 2431 | bool noswap) |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2432 | { |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2433 | struct zonelist *zonelist; |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 2434 | unsigned long nr_reclaimed; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 2435 | int nid; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2436 | struct scan_control sc = { |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2437 | .may_writepage = !laptop_mode, |
Johannes Weiner | a6dc60f8 | 2009-03-31 15:19:30 -0700 | [diff] [blame] | 2438 | .may_unmap = 1, |
KOSAKI Motohiro | 2e2e425 | 2009-04-21 12:24:57 -0700 | [diff] [blame] | 2439 | .may_swap = !noswap, |
KOSAKI Motohiro | 22fba33 | 2009-12-14 17:59:10 -0800 | [diff] [blame] | 2440 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2441 | .order = 0, |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2442 | .priority = DEF_PRIORITY, |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 2443 | .target_mem_cgroup = memcg, |
KAMEZAWA Hiroyuki | 327c0e9 | 2009-03-31 15:23:31 -0700 | [diff] [blame] | 2444 | .nodemask = NULL, /* we don't care the placement */ |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 2445 | .gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | |
| 2446 | (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK), |
| 2447 | }; |
| 2448 | struct shrink_control shrink = { |
| 2449 | .gfp_mask = sc.gfp_mask, |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2450 | }; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2451 | |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 2452 | /* |
| 2453 | * Unlike direct reclaim via alloc_pages(), memcg's reclaim doesn't |
| 2454 | * take care of from where we get pages. So the node where we start the |
| 2455 | * scan does not need to be the current node. |
| 2456 | */ |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 2457 | nid = mem_cgroup_select_victim_node(memcg); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 2458 | |
| 2459 | zonelist = NODE_DATA(nid)->node_zonelists; |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 2460 | |
| 2461 | trace_mm_vmscan_memcg_reclaim_begin(0, |
| 2462 | sc.may_writepage, |
| 2463 | sc.gfp_mask); |
| 2464 | |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 2465 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink); |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 2466 | |
| 2467 | trace_mm_vmscan_memcg_reclaim_end(nr_reclaimed); |
| 2468 | |
| 2469 | return nr_reclaimed; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2470 | } |
| 2471 | #endif |
| 2472 | |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2473 | static void age_active_anon(struct zone *zone, struct scan_control *sc) |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 2474 | { |
Johannes Weiner | b95a2f2 | 2012-01-12 17:18:06 -0800 | [diff] [blame] | 2475 | struct mem_cgroup *memcg; |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 2476 | |
Johannes Weiner | b95a2f2 | 2012-01-12 17:18:06 -0800 | [diff] [blame] | 2477 | if (!total_swap_pages) |
| 2478 | return; |
| 2479 | |
| 2480 | memcg = mem_cgroup_iter(NULL, NULL, NULL); |
| 2481 | do { |
Konstantin Khlebnikov | c56d5c7 | 2012-05-29 15:07:00 -0700 | [diff] [blame] | 2482 | struct lruvec *lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
Johannes Weiner | b95a2f2 | 2012-01-12 17:18:06 -0800 | [diff] [blame] | 2483 | |
Konstantin Khlebnikov | c56d5c7 | 2012-05-29 15:07:00 -0700 | [diff] [blame] | 2484 | if (inactive_anon_is_low(lruvec)) |
Konstantin Khlebnikov | 1a93be0 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 2485 | shrink_active_list(SWAP_CLUSTER_MAX, lruvec, |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2486 | sc, LRU_ACTIVE_ANON); |
Johannes Weiner | b95a2f2 | 2012-01-12 17:18:06 -0800 | [diff] [blame] | 2487 | |
| 2488 | memcg = mem_cgroup_iter(NULL, memcg, NULL); |
| 2489 | } while (memcg); |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 2490 | } |
| 2491 | |
Johannes Weiner | 60cefed | 2012-11-29 13:54:23 -0800 | [diff] [blame] | 2492 | static bool zone_balanced(struct zone *zone, int order, |
| 2493 | unsigned long balance_gap, int classzone_idx) |
| 2494 | { |
| 2495 | if (!zone_watermark_ok_safe(zone, order, high_wmark_pages(zone) + |
| 2496 | balance_gap, classzone_idx, 0)) |
| 2497 | return false; |
| 2498 | |
Kirill A. Shutemov | d84da3f | 2012-12-11 16:00:31 -0800 | [diff] [blame] | 2499 | if (IS_ENABLED(CONFIG_COMPACTION) && order && |
| 2500 | !compaction_suitable(zone, order)) |
Johannes Weiner | 60cefed | 2012-11-29 13:54:23 -0800 | [diff] [blame] | 2501 | return false; |
| 2502 | |
| 2503 | return true; |
| 2504 | } |
| 2505 | |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2506 | /* |
Zlatko Calusic | 4ae0a48 | 2012-12-23 15:12:54 +0100 | [diff] [blame] | 2507 | * pgdat_balanced() is used when checking if a node is balanced. |
| 2508 | * |
| 2509 | * For order-0, all zones must be balanced! |
| 2510 | * |
| 2511 | * For high-order allocations only zones that meet watermarks and are in a |
| 2512 | * zone allowed by the callers classzone_idx are added to balanced_pages. The |
| 2513 | * total of balanced pages must be at least 25% of the zones allowed by |
| 2514 | * classzone_idx for the node to be considered balanced. Forcing all zones to |
| 2515 | * be balanced for high orders can cause excessive reclaim when there are |
| 2516 | * imbalanced zones. |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2517 | * The choice of 25% is due to |
| 2518 | * o a 16M DMA zone that is balanced will not balance a zone on any |
| 2519 | * reasonable sized machine |
| 2520 | * 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] | 2521 | * percentage of the middle zones. For example, on 32-bit x86, highmem |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2522 | * would need to be at least 256M for it to be balance a whole node. |
| 2523 | * Similarly, on x86-64 the Normal zone would need to be at least 1G |
| 2524 | * to balance a node on its own. These seemed like reasonable ratios. |
| 2525 | */ |
Zlatko Calusic | 4ae0a48 | 2012-12-23 15:12:54 +0100 | [diff] [blame] | 2526 | static bool pgdat_balanced(pg_data_t *pgdat, int order, int classzone_idx) |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2527 | { |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 2528 | unsigned long managed_pages = 0; |
Zlatko Calusic | 4ae0a48 | 2012-12-23 15:12:54 +0100 | [diff] [blame] | 2529 | unsigned long balanced_pages = 0; |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2530 | int i; |
| 2531 | |
Zlatko Calusic | 4ae0a48 | 2012-12-23 15:12:54 +0100 | [diff] [blame] | 2532 | /* Check the watermark levels */ |
| 2533 | for (i = 0; i <= classzone_idx; i++) { |
| 2534 | struct zone *zone = pgdat->node_zones + i; |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2535 | |
Zlatko Calusic | 4ae0a48 | 2012-12-23 15:12:54 +0100 | [diff] [blame] | 2536 | if (!populated_zone(zone)) |
| 2537 | continue; |
| 2538 | |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 2539 | managed_pages += zone->managed_pages; |
Zlatko Calusic | 4ae0a48 | 2012-12-23 15:12:54 +0100 | [diff] [blame] | 2540 | |
| 2541 | /* |
| 2542 | * A special case here: |
| 2543 | * |
| 2544 | * balance_pgdat() skips over all_unreclaimable after |
| 2545 | * DEF_PRIORITY. Effectively, it considers them balanced so |
| 2546 | * they must be considered balanced here as well! |
| 2547 | */ |
| 2548 | if (zone->all_unreclaimable) { |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 2549 | balanced_pages += zone->managed_pages; |
Zlatko Calusic | 4ae0a48 | 2012-12-23 15:12:54 +0100 | [diff] [blame] | 2550 | continue; |
| 2551 | } |
| 2552 | |
| 2553 | if (zone_balanced(zone, order, 0, i)) |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 2554 | balanced_pages += zone->managed_pages; |
Zlatko Calusic | 4ae0a48 | 2012-12-23 15:12:54 +0100 | [diff] [blame] | 2555 | else if (!order) |
| 2556 | return false; |
| 2557 | } |
| 2558 | |
| 2559 | if (order) |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 2560 | return balanced_pages >= (managed_pages >> 2); |
Zlatko Calusic | 4ae0a48 | 2012-12-23 15:12:54 +0100 | [diff] [blame] | 2561 | else |
| 2562 | return true; |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 2563 | } |
| 2564 | |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2565 | /* |
| 2566 | * Prepare kswapd for sleeping. This verifies that there are no processes |
| 2567 | * waiting in throttle_direct_reclaim() and that watermarks have been met. |
| 2568 | * |
| 2569 | * Returns true if kswapd is ready to sleep |
| 2570 | */ |
| 2571 | static bool prepare_kswapd_sleep(pg_data_t *pgdat, int order, long remaining, |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2572 | int classzone_idx) |
Mel Gorman | f50de2d | 2009-12-14 17:58:53 -0800 | [diff] [blame] | 2573 | { |
Mel Gorman | f50de2d | 2009-12-14 17:58:53 -0800 | [diff] [blame] | 2574 | /* If a direct reclaimer woke kswapd within HZ/10, it's premature */ |
| 2575 | if (remaining) |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2576 | return false; |
| 2577 | |
| 2578 | /* |
| 2579 | * There is a potential race between when kswapd checks its watermarks |
| 2580 | * and a process gets throttled. There is also a potential race if |
| 2581 | * processes get throttled, kswapd wakes, a large process exits therby |
| 2582 | * balancing the zones that causes kswapd to miss a wakeup. If kswapd |
| 2583 | * is going to sleep, no process should be sleeping on pfmemalloc_wait |
| 2584 | * so wake them now if necessary. If necessary, processes will wake |
| 2585 | * kswapd and get throttled again |
| 2586 | */ |
| 2587 | if (waitqueue_active(&pgdat->pfmemalloc_wait)) { |
| 2588 | wake_up(&pgdat->pfmemalloc_wait); |
| 2589 | return false; |
| 2590 | } |
Mel Gorman | f50de2d | 2009-12-14 17:58:53 -0800 | [diff] [blame] | 2591 | |
Zlatko Calusic | 4ae0a48 | 2012-12-23 15:12:54 +0100 | [diff] [blame] | 2592 | return pgdat_balanced(pgdat, order, classzone_idx); |
Mel Gorman | f50de2d | 2009-12-14 17:58:53 -0800 | [diff] [blame] | 2593 | } |
| 2594 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2595 | /* |
| 2596 | * 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] | 2597 | * they are all at high_wmark_pages(zone). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2598 | * |
Mel Gorman | 0abdee2 | 2011-01-13 15:46:22 -0800 | [diff] [blame] | 2599 | * Returns the final order kswapd was reclaiming at |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2600 | * |
| 2601 | * There is special handling here for zones which are full of pinned pages. |
| 2602 | * This can happen if the pages are all mlocked, or if they are all used by |
| 2603 | * device drivers (say, ZONE_DMA). Or if they are all in use by hugetlb. |
| 2604 | * What we do is to detect the case where all pages in the zone have been |
| 2605 | * scanned twice and there has been zero successful reclaim. Mark the zone as |
| 2606 | * dead and from now on, only perform a short scan. Basically we're polling |
| 2607 | * the zone for when the problem goes away. |
| 2608 | * |
| 2609 | * kswapd scans the zones in the highmem->normal->dma direction. It skips |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 2610 | * zones which have free_pages > high_wmark_pages(zone), but once a zone is |
| 2611 | * found to have free_pages <= high_wmark_pages(zone), we scan that zone and the |
| 2612 | * lower zones regardless of the number of free pages in the lower zones. This |
| 2613 | * interoperates with the page allocator fallback scheme to ensure that aging |
| 2614 | * of pages is balanced across the zones. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2615 | */ |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2616 | static unsigned long balance_pgdat(pg_data_t *pgdat, int order, |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2617 | int *classzone_idx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2618 | { |
Zlatko Calusic | dafcb73 | 2013-02-22 16:32:34 -0800 | [diff] [blame] | 2619 | bool pgdat_is_balanced = false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2620 | int i; |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2621 | int end_zone = 0; /* Inclusive. 0 = ZONE_DMA */ |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 2622 | unsigned long total_scanned; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2623 | struct reclaim_state *reclaim_state = current->reclaim_state; |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 2624 | unsigned long nr_soft_reclaimed; |
| 2625 | unsigned long nr_soft_scanned; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2626 | struct scan_control sc = { |
| 2627 | .gfp_mask = GFP_KERNEL, |
Johannes Weiner | a6dc60f8 | 2009-03-31 15:19:30 -0700 | [diff] [blame] | 2628 | .may_unmap = 1, |
KOSAKI Motohiro | 2e2e425 | 2009-04-21 12:24:57 -0700 | [diff] [blame] | 2629 | .may_swap = 1, |
KOSAKI Motohiro | 22fba33 | 2009-12-14 17:59:10 -0800 | [diff] [blame] | 2630 | /* |
| 2631 | * kswapd doesn't want to be bailed out while reclaim. because |
| 2632 | * we want to put equal scanning pressure on each zone. |
| 2633 | */ |
| 2634 | .nr_to_reclaim = ULONG_MAX, |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 2635 | .order = order, |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 2636 | .target_mem_cgroup = NULL, |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 2637 | }; |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 2638 | struct shrink_control shrink = { |
| 2639 | .gfp_mask = sc.gfp_mask, |
| 2640 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2641 | loop_again: |
| 2642 | total_scanned = 0; |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2643 | sc.priority = DEF_PRIORITY; |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2644 | sc.nr_reclaimed = 0; |
Christoph Lameter | c0bbbc7 | 2006-06-11 15:22:26 -0700 | [diff] [blame] | 2645 | sc.may_writepage = !laptop_mode; |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 2646 | count_vm_event(PAGEOUTRUN); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2647 | |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2648 | do { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2649 | unsigned long lru_pages = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2650 | |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2651 | /* |
| 2652 | * Scan in the highmem->dma direction for the highest |
| 2653 | * zone which needs scanning |
| 2654 | */ |
| 2655 | for (i = pgdat->nr_zones - 1; i >= 0; i--) { |
| 2656 | struct zone *zone = pgdat->node_zones + i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2657 | |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2658 | if (!populated_zone(zone)) |
| 2659 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2660 | |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2661 | if (zone->all_unreclaimable && |
| 2662 | sc.priority != DEF_PRIORITY) |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2663 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2664 | |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 2665 | /* |
| 2666 | * Do some background aging of the anon list, to give |
| 2667 | * pages a chance to be referenced before reclaiming. |
| 2668 | */ |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2669 | age_active_anon(zone, &sc); |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 2670 | |
Mel Gorman | cc715d9 | 2012-03-21 16:34:00 -0700 | [diff] [blame] | 2671 | /* |
| 2672 | * If the number of buffer_heads in the machine |
| 2673 | * exceeds the maximum allowed level and this node |
| 2674 | * has a highmem zone, force kswapd to reclaim from |
| 2675 | * it to relieve lowmem pressure. |
| 2676 | */ |
| 2677 | if (buffer_heads_over_limit && is_highmem_idx(i)) { |
| 2678 | end_zone = i; |
| 2679 | break; |
| 2680 | } |
| 2681 | |
Johannes Weiner | 60cefed | 2012-11-29 13:54:23 -0800 | [diff] [blame] | 2682 | if (!zone_balanced(zone, order, 0, 0)) { |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 2683 | end_zone = i; |
Andrew Morton | e1dbeda | 2006-12-06 20:32:01 -0800 | [diff] [blame] | 2684 | break; |
Shaohua Li | 439423f | 2011-08-25 15:59:12 -0700 | [diff] [blame] | 2685 | } else { |
| 2686 | /* If balanced, clear the congested flag */ |
| 2687 | zone_clear_flag(zone, ZONE_CONGESTED); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2688 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2689 | } |
Zlatko Calusic | dafcb73 | 2013-02-22 16:32:34 -0800 | [diff] [blame] | 2690 | |
| 2691 | if (i < 0) { |
| 2692 | pgdat_is_balanced = true; |
Andrew Morton | e1dbeda | 2006-12-06 20:32:01 -0800 | [diff] [blame] | 2693 | goto out; |
Zlatko Calusic | dafcb73 | 2013-02-22 16:32:34 -0800 | [diff] [blame] | 2694 | } |
Andrew Morton | e1dbeda | 2006-12-06 20:32:01 -0800 | [diff] [blame] | 2695 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2696 | for (i = 0; i <= end_zone; i++) { |
| 2697 | struct zone *zone = pgdat->node_zones + i; |
| 2698 | |
Wu Fengguang | adea02a | 2009-09-21 17:01:42 -0700 | [diff] [blame] | 2699 | lru_pages += zone_reclaimable_pages(zone); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2700 | } |
| 2701 | |
| 2702 | /* |
| 2703 | * Now scan the zone in the dma->highmem direction, stopping |
| 2704 | * at the last zone which needs scanning. |
| 2705 | * |
| 2706 | * We do this because the page allocator works in the opposite |
| 2707 | * direction. This prevents the page allocator from allocating |
| 2708 | * pages behind kswapd's direction of progress, which would |
| 2709 | * cause too much scanning of the lower zones. |
| 2710 | */ |
| 2711 | for (i = 0; i <= end_zone; i++) { |
| 2712 | struct zone *zone = pgdat->node_zones + i; |
Rik van Riel | fe2c2a1 | 2012-03-21 16:33:51 -0700 | [diff] [blame] | 2713 | int nr_slab, testorder; |
Mel Gorman | 8afdcec | 2011-03-22 16:33:04 -0700 | [diff] [blame] | 2714 | unsigned long balance_gap; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2715 | |
Con Kolivas | f3fe651 | 2006-01-06 00:11:15 -0800 | [diff] [blame] | 2716 | if (!populated_zone(zone)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2717 | continue; |
| 2718 | |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2719 | if (zone->all_unreclaimable && |
| 2720 | sc.priority != DEF_PRIORITY) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2721 | continue; |
| 2722 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2723 | sc.nr_scanned = 0; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2724 | |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 2725 | nr_soft_scanned = 0; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2726 | /* |
| 2727 | * Call soft limit reclaim before calling shrink_zone. |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2728 | */ |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 2729 | nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone, |
| 2730 | order, sc.gfp_mask, |
| 2731 | &nr_soft_scanned); |
| 2732 | sc.nr_reclaimed += nr_soft_reclaimed; |
| 2733 | total_scanned += nr_soft_scanned; |
KOSAKI Motohiro | 00918b6 | 2010-08-10 18:03:05 -0700 | [diff] [blame] | 2734 | |
Rik van Riel | 32a4330 | 2007-10-16 01:24:50 -0700 | [diff] [blame] | 2735 | /* |
Mel Gorman | 8afdcec | 2011-03-22 16:33:04 -0700 | [diff] [blame] | 2736 | * We put equal pressure on every zone, unless |
| 2737 | * one zone has way too many pages free |
| 2738 | * already. The "too many pages" is defined |
| 2739 | * as the high wmark plus a "gap" where the |
| 2740 | * gap is either the low watermark or 1% |
| 2741 | * of the zone, whichever is smaller. |
Rik van Riel | 32a4330 | 2007-10-16 01:24:50 -0700 | [diff] [blame] | 2742 | */ |
Mel Gorman | 8afdcec | 2011-03-22 16:33:04 -0700 | [diff] [blame] | 2743 | balance_gap = min(low_wmark_pages(zone), |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 2744 | (zone->managed_pages + |
Mel Gorman | 8afdcec | 2011-03-22 16:33:04 -0700 | [diff] [blame] | 2745 | KSWAPD_ZONE_BALANCE_GAP_RATIO-1) / |
| 2746 | KSWAPD_ZONE_BALANCE_GAP_RATIO); |
Rik van Riel | fe2c2a1 | 2012-03-21 16:33:51 -0700 | [diff] [blame] | 2747 | /* |
| 2748 | * Kswapd reclaims only single pages with compaction |
| 2749 | * enabled. Trying too hard to reclaim until contiguous |
| 2750 | * free pages have become available can hurt performance |
| 2751 | * by evicting too much useful data from memory. |
| 2752 | * Do not reclaim more than needed for compaction. |
| 2753 | */ |
| 2754 | testorder = order; |
Kirill A. Shutemov | d84da3f | 2012-12-11 16:00:31 -0800 | [diff] [blame] | 2755 | if (IS_ENABLED(CONFIG_COMPACTION) && order && |
Rik van Riel | fe2c2a1 | 2012-03-21 16:33:51 -0700 | [diff] [blame] | 2756 | compaction_suitable(zone, order) != |
| 2757 | COMPACT_SKIPPED) |
| 2758 | testorder = 0; |
| 2759 | |
Mel Gorman | cc715d9 | 2012-03-21 16:34:00 -0700 | [diff] [blame] | 2760 | if ((buffer_heads_over_limit && is_highmem_idx(i)) || |
Johannes Weiner | 60cefed | 2012-11-29 13:54:23 -0800 | [diff] [blame] | 2761 | !zone_balanced(zone, testorder, |
| 2762 | balance_gap, end_zone)) { |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2763 | shrink_zone(zone, &sc); |
Andrea Arcangeli | 5a03b05 | 2011-01-13 15:47:11 -0800 | [diff] [blame] | 2764 | |
Mel Gorman | d7868da | 2011-07-08 15:39:38 -0700 | [diff] [blame] | 2765 | reclaim_state->reclaimed_slab = 0; |
| 2766 | nr_slab = shrink_slab(&shrink, sc.nr_scanned, lru_pages); |
| 2767 | sc.nr_reclaimed += reclaim_state->reclaimed_slab; |
| 2768 | total_scanned += sc.nr_scanned; |
| 2769 | |
| 2770 | if (nr_slab == 0 && !zone_reclaimable(zone)) |
| 2771 | zone->all_unreclaimable = 1; |
| 2772 | } |
| 2773 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2774 | /* |
Minchan Kim | 0e50ce3 | 2013-02-22 16:35:37 -0800 | [diff] [blame] | 2775 | * If we're getting trouble reclaiming, start doing |
| 2776 | * writepage even in laptop mode. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2777 | */ |
Minchan Kim | 0e50ce3 | 2013-02-22 16:35:37 -0800 | [diff] [blame] | 2778 | if (sc.priority < DEF_PRIORITY - 2) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2779 | sc.may_writepage = 1; |
KOSAKI Motohiro | bb3ab59 | 2009-12-14 17:58:55 -0800 | [diff] [blame] | 2780 | |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2781 | if (zone->all_unreclaimable) { |
| 2782 | if (end_zone && end_zone == i) |
| 2783 | end_zone--; |
Mel Gorman | d7868da | 2011-07-08 15:39:38 -0700 | [diff] [blame] | 2784 | continue; |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2785 | } |
Mel Gorman | d7868da | 2011-07-08 15:39:38 -0700 | [diff] [blame] | 2786 | |
Zlatko Calusic | 258401a | 2013-02-22 16:34:06 -0800 | [diff] [blame] | 2787 | if (zone_balanced(zone, testorder, 0, end_zone)) |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 2788 | /* |
| 2789 | * If a zone reaches its high watermark, |
| 2790 | * consider it to be no longer congested. It's |
| 2791 | * possible there are dirty pages backed by |
| 2792 | * congested BDIs but as pressure is relieved, |
Wanpeng Li | ab8704b | 2012-06-17 09:27:18 +0800 | [diff] [blame] | 2793 | * speculatively avoid congestion waits |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 2794 | */ |
| 2795 | zone_clear_flag(zone, ZONE_CONGESTED); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2796 | } |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2797 | |
| 2798 | /* |
| 2799 | * If the low watermark is met there is no need for processes |
| 2800 | * to be throttled on pfmemalloc_wait as they should not be |
| 2801 | * able to safely make forward progress. Wake them |
| 2802 | */ |
| 2803 | if (waitqueue_active(&pgdat->pfmemalloc_wait) && |
| 2804 | pfmemalloc_watermark_ok(pgdat)) |
| 2805 | wake_up(&pgdat->pfmemalloc_wait); |
| 2806 | |
Zlatko Calusic | dafcb73 | 2013-02-22 16:32:34 -0800 | [diff] [blame] | 2807 | if (pgdat_balanced(pgdat, order, *classzone_idx)) { |
| 2808 | pgdat_is_balanced = true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2809 | break; /* kswapd: all done */ |
Zlatko Calusic | dafcb73 | 2013-02-22 16:32:34 -0800 | [diff] [blame] | 2810 | } |
| 2811 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2812 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2813 | * We do this so kswapd doesn't build up large priorities for |
| 2814 | * example when it is freeing in parallel with allocators. It |
| 2815 | * matches the direct reclaim path behaviour in terms of impact |
| 2816 | * on zone->*_priority. |
| 2817 | */ |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 2818 | if (sc.nr_reclaimed >= SWAP_CLUSTER_MAX) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2819 | break; |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2820 | } while (--sc.priority >= 0); |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2821 | |
Zlatko Calusic | dafcb73 | 2013-02-22 16:32:34 -0800 | [diff] [blame] | 2822 | out: |
| 2823 | if (!pgdat_is_balanced) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2824 | cond_resched(); |
Rafael J. Wysocki | 8357376 | 2006-12-06 20:34:18 -0800 | [diff] [blame] | 2825 | |
| 2826 | try_to_freeze(); |
| 2827 | |
KOSAKI Motohiro | 73ce02e | 2009-01-06 14:40:33 -0800 | [diff] [blame] | 2828 | /* |
| 2829 | * Fragmentation may mean that the system cannot be |
| 2830 | * rebalanced for high-order allocations in all zones. |
| 2831 | * At this point, if nr_reclaimed < SWAP_CLUSTER_MAX, |
| 2832 | * it means the zones have been fully scanned and are still |
| 2833 | * not balanced. For high-order allocations, there is |
| 2834 | * little point trying all over again as kswapd may |
| 2835 | * infinite loop. |
| 2836 | * |
| 2837 | * Instead, recheck all watermarks at order-0 as they |
| 2838 | * are the most important. If watermarks are ok, kswapd will go |
| 2839 | * back to sleep. High-order users can still perform direct |
| 2840 | * reclaim if they wish. |
| 2841 | */ |
| 2842 | if (sc.nr_reclaimed < SWAP_CLUSTER_MAX) |
| 2843 | order = sc.order = 0; |
| 2844 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2845 | goto loop_again; |
| 2846 | } |
| 2847 | |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2848 | /* |
| 2849 | * If kswapd was reclaiming at a higher order, it has the option of |
| 2850 | * sleeping without all zones being balanced. Before it does, it must |
| 2851 | * ensure that the watermarks for order-0 on *all* zones are met and |
| 2852 | * that the congestion flags are cleared. The congestion flag must |
| 2853 | * be cleared as kswapd is the only mechanism that clears the flag |
| 2854 | * and it is potentially going to sleep here. |
| 2855 | */ |
| 2856 | if (order) { |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 2857 | int zones_need_compaction = 1; |
| 2858 | |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2859 | for (i = 0; i <= end_zone; i++) { |
| 2860 | struct zone *zone = pgdat->node_zones + i; |
| 2861 | |
| 2862 | if (!populated_zone(zone)) |
| 2863 | continue; |
| 2864 | |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 2865 | /* Check if the memory needs to be defragmented. */ |
| 2866 | if (zone_watermark_ok(zone, order, |
| 2867 | low_wmark_pages(zone), *classzone_idx, 0)) |
| 2868 | zones_need_compaction = 0; |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2869 | } |
Rik van Riel | 7be62de | 2012-03-21 16:33:52 -0700 | [diff] [blame] | 2870 | |
| 2871 | if (zones_need_compaction) |
| 2872 | compact_pgdat(pgdat, order); |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2873 | } |
| 2874 | |
Mel Gorman | 0abdee2 | 2011-01-13 15:46:22 -0800 | [diff] [blame] | 2875 | /* |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2876 | * Return the order we were reclaiming at so prepare_kswapd_sleep() |
Mel Gorman | 0abdee2 | 2011-01-13 15:46:22 -0800 | [diff] [blame] | 2877 | * makes a decision on the order we were last reclaiming at. However, |
| 2878 | * if another caller entered the allocator slow path while kswapd |
| 2879 | * was awake, order will remain at the higher level |
| 2880 | */ |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2881 | *classzone_idx = end_zone; |
Mel Gorman | 0abdee2 | 2011-01-13 15:46:22 -0800 | [diff] [blame] | 2882 | return order; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2883 | } |
| 2884 | |
Mel Gorman | dc83edd | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2885 | 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] | 2886 | { |
| 2887 | long remaining = 0; |
| 2888 | DEFINE_WAIT(wait); |
| 2889 | |
| 2890 | if (freezing(current) || kthread_should_stop()) |
| 2891 | return; |
| 2892 | |
| 2893 | prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); |
| 2894 | |
| 2895 | /* Try to sleep for a short interval */ |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2896 | if (prepare_kswapd_sleep(pgdat, order, remaining, classzone_idx)) { |
KOSAKI Motohiro | f0bc0a6 | 2011-01-13 15:45:50 -0800 | [diff] [blame] | 2897 | remaining = schedule_timeout(HZ/10); |
| 2898 | finish_wait(&pgdat->kswapd_wait, &wait); |
| 2899 | prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); |
| 2900 | } |
| 2901 | |
| 2902 | /* |
| 2903 | * After a short sleep, check if it was a premature sleep. If not, then |
| 2904 | * go fully to sleep until explicitly woken up. |
| 2905 | */ |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2906 | if (prepare_kswapd_sleep(pgdat, order, remaining, classzone_idx)) { |
KOSAKI Motohiro | f0bc0a6 | 2011-01-13 15:45:50 -0800 | [diff] [blame] | 2907 | trace_mm_vmscan_kswapd_sleep(pgdat->node_id); |
| 2908 | |
| 2909 | /* |
| 2910 | * vmstat counters are not perfectly accurate and the estimated |
| 2911 | * value for counters such as NR_FREE_PAGES can deviate from the |
| 2912 | * true value by nr_online_cpus * threshold. To avoid the zone |
| 2913 | * watermarks being breached while under pressure, we reduce the |
| 2914 | * per-cpu vmstat threshold while kswapd is awake and restore |
| 2915 | * them before going back to sleep. |
| 2916 | */ |
| 2917 | set_pgdat_percpu_threshold(pgdat, calculate_normal_threshold); |
Aaditya Kumar | 1c7e7f6 | 2012-07-17 15:48:07 -0700 | [diff] [blame] | 2918 | |
Mel Gorman | 6299702 | 2012-10-08 16:32:47 -0700 | [diff] [blame] | 2919 | /* |
| 2920 | * Compaction records what page blocks it recently failed to |
| 2921 | * isolate pages from and skips them in the future scanning. |
| 2922 | * When kswapd is going to sleep, it is reasonable to assume |
| 2923 | * that pages and compaction may succeed so reset the cache. |
| 2924 | */ |
| 2925 | reset_isolation_suitable(pgdat); |
| 2926 | |
Aaditya Kumar | 1c7e7f6 | 2012-07-17 15:48:07 -0700 | [diff] [blame] | 2927 | if (!kthread_should_stop()) |
| 2928 | schedule(); |
| 2929 | |
KOSAKI Motohiro | f0bc0a6 | 2011-01-13 15:45:50 -0800 | [diff] [blame] | 2930 | set_pgdat_percpu_threshold(pgdat, calculate_pressure_threshold); |
| 2931 | } else { |
| 2932 | if (remaining) |
| 2933 | count_vm_event(KSWAPD_LOW_WMARK_HIT_QUICKLY); |
| 2934 | else |
| 2935 | count_vm_event(KSWAPD_HIGH_WMARK_HIT_QUICKLY); |
| 2936 | } |
| 2937 | finish_wait(&pgdat->kswapd_wait, &wait); |
| 2938 | } |
| 2939 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2940 | /* |
| 2941 | * The background pageout daemon, started as a kernel thread |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2942 | * from the init process. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2943 | * |
| 2944 | * This basically trickles out pages so that we have _some_ |
| 2945 | * free memory available even if there is no other activity |
| 2946 | * that frees anything up. This is needed for things like routing |
| 2947 | * etc, where we otherwise might have all activity going on in |
| 2948 | * asynchronous contexts that cannot page things out. |
| 2949 | * |
| 2950 | * If there are applications that are active memory-allocators |
| 2951 | * (most normal use), this basically shouldn't matter. |
| 2952 | */ |
| 2953 | static int kswapd(void *p) |
| 2954 | { |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2955 | unsigned long order, new_order; |
Alex,Shi | d2ebd0f6 | 2011-10-31 17:08:39 -0700 | [diff] [blame] | 2956 | unsigned balanced_order; |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2957 | int classzone_idx, new_classzone_idx; |
Alex,Shi | d2ebd0f6 | 2011-10-31 17:08:39 -0700 | [diff] [blame] | 2958 | int balanced_classzone_idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2959 | pg_data_t *pgdat = (pg_data_t*)p; |
| 2960 | struct task_struct *tsk = current; |
KOSAKI Motohiro | f0bc0a6 | 2011-01-13 15:45:50 -0800 | [diff] [blame] | 2961 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2962 | struct reclaim_state reclaim_state = { |
| 2963 | .reclaimed_slab = 0, |
| 2964 | }; |
Rusty Russell | a70f730 | 2009-03-13 14:49:46 +1030 | [diff] [blame] | 2965 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2966 | |
Nick Piggin | cf40bd1 | 2009-01-21 08:12:39 +0100 | [diff] [blame] | 2967 | lockdep_set_current_reclaim_state(GFP_KERNEL); |
| 2968 | |
Rusty Russell | 174596a | 2009-01-01 10:12:29 +1030 | [diff] [blame] | 2969 | if (!cpumask_empty(cpumask)) |
Mike Travis | c5f59f0 | 2008-04-04 18:11:10 -0700 | [diff] [blame] | 2970 | set_cpus_allowed_ptr(tsk, cpumask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2971 | current->reclaim_state = &reclaim_state; |
| 2972 | |
| 2973 | /* |
| 2974 | * Tell the memory management that we're a "memory allocator", |
| 2975 | * and that if we need more memory we should get access to it |
| 2976 | * regardless (see "__alloc_pages()"). "kswapd" should |
| 2977 | * never get caught in the normal page freeing logic. |
| 2978 | * |
| 2979 | * (Kswapd normally doesn't need memory anyway, but sometimes |
| 2980 | * you need a small amount of memory in order to be able to |
| 2981 | * page out something else, and this flag essentially protects |
| 2982 | * us from recursively trying to free more memory as we're |
| 2983 | * trying to free the first piece of memory in the first place). |
| 2984 | */ |
Christoph Lameter | 930d915 | 2006-01-08 01:00:47 -0800 | [diff] [blame] | 2985 | tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD; |
Rafael J. Wysocki | 8314418 | 2007-07-17 04:03:35 -0700 | [diff] [blame] | 2986 | set_freezable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2987 | |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2988 | order = new_order = 0; |
Alex,Shi | d2ebd0f6 | 2011-10-31 17:08:39 -0700 | [diff] [blame] | 2989 | balanced_order = 0; |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2990 | classzone_idx = new_classzone_idx = pgdat->nr_zones - 1; |
Alex,Shi | d2ebd0f6 | 2011-10-31 17:08:39 -0700 | [diff] [blame] | 2991 | balanced_classzone_idx = classzone_idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2992 | for ( ; ; ) { |
Jeff Liu | 6f6313d | 2012-12-11 16:02:48 -0800 | [diff] [blame] | 2993 | bool ret; |
Christoph Lameter | 3e1d1d2 | 2005-06-24 23:13:50 -0700 | [diff] [blame] | 2994 | |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 2995 | /* |
| 2996 | * If the last balance_pgdat was unsuccessful it's unlikely a |
| 2997 | * new request of a similar or harder type will succeed soon |
| 2998 | * so consider going to sleep on the basis we reclaimed at |
| 2999 | */ |
Alex,Shi | d2ebd0f6 | 2011-10-31 17:08:39 -0700 | [diff] [blame] | 3000 | if (balanced_classzone_idx >= new_classzone_idx && |
| 3001 | balanced_order == new_order) { |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 3002 | new_order = pgdat->kswapd_max_order; |
| 3003 | new_classzone_idx = pgdat->classzone_idx; |
| 3004 | pgdat->kswapd_max_order = 0; |
| 3005 | pgdat->classzone_idx = pgdat->nr_zones - 1; |
| 3006 | } |
| 3007 | |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 3008 | if (order < new_order || classzone_idx > new_classzone_idx) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3009 | /* |
| 3010 | * Don't sleep if someone wants a larger 'order' |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 3011 | * allocation or has tigher zone constraints |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3012 | */ |
| 3013 | order = new_order; |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 3014 | classzone_idx = new_classzone_idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3015 | } else { |
Alex,Shi | d2ebd0f6 | 2011-10-31 17:08:39 -0700 | [diff] [blame] | 3016 | kswapd_try_to_sleep(pgdat, balanced_order, |
| 3017 | balanced_classzone_idx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3018 | order = pgdat->kswapd_max_order; |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 3019 | classzone_idx = pgdat->classzone_idx; |
Alex,Shi | f0dfcde | 2011-10-31 17:08:45 -0700 | [diff] [blame] | 3020 | new_order = order; |
| 3021 | new_classzone_idx = classzone_idx; |
Mel Gorman | 4d40502 | 2011-01-13 15:46:23 -0800 | [diff] [blame] | 3022 | pgdat->kswapd_max_order = 0; |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 3023 | pgdat->classzone_idx = pgdat->nr_zones - 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3024 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3025 | |
David Rientjes | 8fe23e0 | 2009-12-14 17:58:33 -0800 | [diff] [blame] | 3026 | ret = try_to_freeze(); |
| 3027 | if (kthread_should_stop()) |
| 3028 | break; |
| 3029 | |
| 3030 | /* |
| 3031 | * We can speed up thawing tasks if we don't call balance_pgdat |
| 3032 | * after returning from the refrigerator |
| 3033 | */ |
Mel Gorman | 33906bc | 2010-08-09 17:19:16 -0700 | [diff] [blame] | 3034 | if (!ret) { |
| 3035 | trace_mm_vmscan_kswapd_wake(pgdat->node_id, order); |
Alex,Shi | d2ebd0f6 | 2011-10-31 17:08:39 -0700 | [diff] [blame] | 3036 | balanced_classzone_idx = classzone_idx; |
| 3037 | balanced_order = balance_pgdat(pgdat, order, |
| 3038 | &balanced_classzone_idx); |
Mel Gorman | 33906bc | 2010-08-09 17:19:16 -0700 | [diff] [blame] | 3039 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3040 | } |
Takamori Yamaguchi | b0a8cc5 | 2012-11-08 15:53:39 -0800 | [diff] [blame] | 3041 | |
| 3042 | current->reclaim_state = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3043 | return 0; |
| 3044 | } |
| 3045 | |
| 3046 | /* |
| 3047 | * A zone is low on free memory, so wake its kswapd task to service it. |
| 3048 | */ |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 3049 | 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] | 3050 | { |
| 3051 | pg_data_t *pgdat; |
| 3052 | |
Con Kolivas | f3fe651 | 2006-01-06 00:11:15 -0800 | [diff] [blame] | 3053 | if (!populated_zone(zone)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3054 | return; |
| 3055 | |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 3056 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3057 | return; |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 3058 | pgdat = zone->zone_pgdat; |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 3059 | if (pgdat->kswapd_max_order < order) { |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 3060 | pgdat->kswapd_max_order = order; |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 3061 | pgdat->classzone_idx = min(pgdat->classzone_idx, classzone_idx); |
| 3062 | } |
Con Kolivas | 8d0986e | 2005-09-13 01:25:07 -0700 | [diff] [blame] | 3063 | if (!waitqueue_active(&pgdat->kswapd_wait)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3064 | return; |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 3065 | if (zone_watermark_ok_safe(zone, order, low_wmark_pages(zone), 0, 0)) |
| 3066 | return; |
| 3067 | |
| 3068 | trace_mm_vmscan_wakeup_kswapd(pgdat->node_id, zone_idx(zone), order); |
Con Kolivas | 8d0986e | 2005-09-13 01:25:07 -0700 | [diff] [blame] | 3069 | wake_up_interruptible(&pgdat->kswapd_wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3070 | } |
| 3071 | |
Wu Fengguang | adea02a | 2009-09-21 17:01:42 -0700 | [diff] [blame] | 3072 | /* |
| 3073 | * The reclaimable count would be mostly accurate. |
| 3074 | * The less reclaimable pages may be |
| 3075 | * - mlocked pages, which will be moved to unevictable list when encountered |
| 3076 | * - mapped pages, which may require several travels to be reclaimed |
| 3077 | * - dirty pages, which is not "instantly" reclaimable |
| 3078 | */ |
| 3079 | unsigned long global_reclaimable_pages(void) |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 3080 | { |
Wu Fengguang | adea02a | 2009-09-21 17:01:42 -0700 | [diff] [blame] | 3081 | int nr; |
| 3082 | |
| 3083 | nr = global_page_state(NR_ACTIVE_FILE) + |
| 3084 | global_page_state(NR_INACTIVE_FILE); |
| 3085 | |
Shaohua Li | ec8acf2 | 2013-02-22 16:34:38 -0800 | [diff] [blame] | 3086 | if (get_nr_swap_pages() > 0) |
Wu Fengguang | adea02a | 2009-09-21 17:01:42 -0700 | [diff] [blame] | 3087 | nr += global_page_state(NR_ACTIVE_ANON) + |
| 3088 | global_page_state(NR_INACTIVE_ANON); |
| 3089 | |
| 3090 | return nr; |
| 3091 | } |
| 3092 | |
| 3093 | unsigned long zone_reclaimable_pages(struct zone *zone) |
| 3094 | { |
| 3095 | int nr; |
| 3096 | |
| 3097 | nr = zone_page_state(zone, NR_ACTIVE_FILE) + |
| 3098 | zone_page_state(zone, NR_INACTIVE_FILE); |
| 3099 | |
Shaohua Li | ec8acf2 | 2013-02-22 16:34:38 -0800 | [diff] [blame] | 3100 | if (get_nr_swap_pages() > 0) |
Wu Fengguang | adea02a | 2009-09-21 17:01:42 -0700 | [diff] [blame] | 3101 | nr += zone_page_state(zone, NR_ACTIVE_ANON) + |
| 3102 | zone_page_state(zone, NR_INACTIVE_ANON); |
| 3103 | |
| 3104 | return nr; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 3105 | } |
| 3106 | |
Rafael J. Wysocki | c6f37f1 | 2009-05-24 22:16:31 +0200 | [diff] [blame] | 3107 | #ifdef CONFIG_HIBERNATION |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3108 | /* |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3109 | * 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] | 3110 | * freed pages. |
| 3111 | * |
| 3112 | * Rather than trying to age LRUs the aim is to preserve the overall |
| 3113 | * LRU order by reclaiming preferentially |
| 3114 | * inactive > active > active referenced > active mapped |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3115 | */ |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3116 | unsigned long shrink_all_memory(unsigned long nr_to_reclaim) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3117 | { |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 3118 | struct reclaim_state reclaim_state; |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 3119 | struct scan_control sc = { |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3120 | .gfp_mask = GFP_HIGHUSER_MOVABLE, |
| 3121 | .may_swap = 1, |
| 3122 | .may_unmap = 1, |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 3123 | .may_writepage = 1, |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3124 | .nr_to_reclaim = nr_to_reclaim, |
| 3125 | .hibernation_mode = 1, |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3126 | .order = 0, |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3127 | .priority = DEF_PRIORITY, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3128 | }; |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 3129 | struct shrink_control shrink = { |
| 3130 | .gfp_mask = sc.gfp_mask, |
| 3131 | }; |
| 3132 | struct zonelist *zonelist = node_zonelist(numa_node_id(), sc.gfp_mask); |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3133 | struct task_struct *p = current; |
| 3134 | unsigned long nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3135 | |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3136 | p->flags |= PF_MEMALLOC; |
| 3137 | lockdep_set_current_reclaim_state(sc.gfp_mask); |
| 3138 | reclaim_state.reclaimed_slab = 0; |
| 3139 | p->reclaim_state = &reclaim_state; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 3140 | |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 3141 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink); |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 3142 | |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3143 | p->reclaim_state = NULL; |
| 3144 | lockdep_clear_current_reclaim_state(); |
| 3145 | p->flags &= ~PF_MEMALLOC; |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 3146 | |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3147 | return nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3148 | } |
Rafael J. Wysocki | c6f37f1 | 2009-05-24 22:16:31 +0200 | [diff] [blame] | 3149 | #endif /* CONFIG_HIBERNATION */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3150 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3151 | /* It's optimal to keep kswapds on the same CPUs as their memory, but |
| 3152 | not required for correctness. So if the last cpu in a node goes |
| 3153 | away, we get changed to run anywhere: as the first one comes back, |
| 3154 | restore their cpu bindings. */ |
Greg Kroah-Hartman | fcb35a9 | 2012-12-21 15:01:06 -0800 | [diff] [blame] | 3155 | static int cpu_callback(struct notifier_block *nfb, unsigned long action, |
| 3156 | void *hcpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3157 | { |
Yasunori Goto | 58c0a4a | 2007-10-16 01:25:40 -0700 | [diff] [blame] | 3158 | int nid; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3159 | |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 3160 | if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN) { |
Lai Jiangshan | 48fb2e2 | 2012-12-12 13:51:43 -0800 | [diff] [blame] | 3161 | for_each_node_state(nid, N_MEMORY) { |
Mike Travis | c5f59f0 | 2008-04-04 18:11:10 -0700 | [diff] [blame] | 3162 | pg_data_t *pgdat = NODE_DATA(nid); |
Rusty Russell | a70f730 | 2009-03-13 14:49:46 +1030 | [diff] [blame] | 3163 | const struct cpumask *mask; |
| 3164 | |
| 3165 | mask = cpumask_of_node(pgdat->node_id); |
Mike Travis | c5f59f0 | 2008-04-04 18:11:10 -0700 | [diff] [blame] | 3166 | |
Rusty Russell | 3e59794 | 2009-01-01 10:12:24 +1030 | [diff] [blame] | 3167 | if (cpumask_any_and(cpu_online_mask, mask) < nr_cpu_ids) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3168 | /* One of our CPUs online: restore mask */ |
Mike Travis | c5f59f0 | 2008-04-04 18:11:10 -0700 | [diff] [blame] | 3169 | set_cpus_allowed_ptr(pgdat->kswapd, mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3170 | } |
| 3171 | } |
| 3172 | return NOTIFY_OK; |
| 3173 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3174 | |
Yasunori Goto | 3218ae1 | 2006-06-27 02:53:33 -0700 | [diff] [blame] | 3175 | /* |
| 3176 | * This kswapd start function will be called by init and node-hot-add. |
| 3177 | * On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added. |
| 3178 | */ |
| 3179 | int kswapd_run(int nid) |
| 3180 | { |
| 3181 | pg_data_t *pgdat = NODE_DATA(nid); |
| 3182 | int ret = 0; |
| 3183 | |
| 3184 | if (pgdat->kswapd) |
| 3185 | return 0; |
| 3186 | |
| 3187 | pgdat->kswapd = kthread_run(kswapd, pgdat, "kswapd%d", nid); |
| 3188 | if (IS_ERR(pgdat->kswapd)) { |
| 3189 | /* failure at boot is fatal */ |
| 3190 | BUG_ON(system_state == SYSTEM_BOOTING); |
Gavin Shan | d5dc0ad | 2012-10-08 16:29:27 -0700 | [diff] [blame] | 3191 | pr_err("Failed to start kswapd on node %d\n", nid); |
| 3192 | ret = PTR_ERR(pgdat->kswapd); |
Xishi Qiu | d72515b | 2013-04-17 15:58:34 -0700 | [diff] [blame^] | 3193 | pgdat->kswapd = NULL; |
Yasunori Goto | 3218ae1 | 2006-06-27 02:53:33 -0700 | [diff] [blame] | 3194 | } |
| 3195 | return ret; |
| 3196 | } |
| 3197 | |
David Rientjes | 8fe23e0 | 2009-12-14 17:58:33 -0800 | [diff] [blame] | 3198 | /* |
Jiang Liu | d8adde1 | 2012-07-11 14:01:52 -0700 | [diff] [blame] | 3199 | * Called by memory hotplug when all memory in a node is offlined. Caller must |
| 3200 | * hold lock_memory_hotplug(). |
David Rientjes | 8fe23e0 | 2009-12-14 17:58:33 -0800 | [diff] [blame] | 3201 | */ |
| 3202 | void kswapd_stop(int nid) |
| 3203 | { |
| 3204 | struct task_struct *kswapd = NODE_DATA(nid)->kswapd; |
| 3205 | |
Jiang Liu | d8adde1 | 2012-07-11 14:01:52 -0700 | [diff] [blame] | 3206 | if (kswapd) { |
David Rientjes | 8fe23e0 | 2009-12-14 17:58:33 -0800 | [diff] [blame] | 3207 | kthread_stop(kswapd); |
Jiang Liu | d8adde1 | 2012-07-11 14:01:52 -0700 | [diff] [blame] | 3208 | NODE_DATA(nid)->kswapd = NULL; |
| 3209 | } |
David Rientjes | 8fe23e0 | 2009-12-14 17:58:33 -0800 | [diff] [blame] | 3210 | } |
| 3211 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3212 | static int __init kswapd_init(void) |
| 3213 | { |
Yasunori Goto | 3218ae1 | 2006-06-27 02:53:33 -0700 | [diff] [blame] | 3214 | int nid; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 3215 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3216 | swap_setup(); |
Lai Jiangshan | 48fb2e2 | 2012-12-12 13:51:43 -0800 | [diff] [blame] | 3217 | for_each_node_state(nid, N_MEMORY) |
Yasunori Goto | 3218ae1 | 2006-06-27 02:53:33 -0700 | [diff] [blame] | 3218 | kswapd_run(nid); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3219 | hotcpu_notifier(cpu_callback, 0); |
| 3220 | return 0; |
| 3221 | } |
| 3222 | |
| 3223 | module_init(kswapd_init) |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3224 | |
| 3225 | #ifdef CONFIG_NUMA |
| 3226 | /* |
| 3227 | * Zone reclaim mode |
| 3228 | * |
| 3229 | * If non-zero call zone_reclaim when the number of free pages falls below |
| 3230 | * the watermarks. |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3231 | */ |
| 3232 | int zone_reclaim_mode __read_mostly; |
| 3233 | |
Christoph Lameter | 1b2ffb7 | 2006-02-01 03:05:34 -0800 | [diff] [blame] | 3234 | #define RECLAIM_OFF 0 |
Fernando Luis Vazquez Cao | 7d03431 | 2008-07-29 22:33:41 -0700 | [diff] [blame] | 3235 | #define RECLAIM_ZONE (1<<0) /* Run shrink_inactive_list on the zone */ |
Christoph Lameter | 1b2ffb7 | 2006-02-01 03:05:34 -0800 | [diff] [blame] | 3236 | #define RECLAIM_WRITE (1<<1) /* Writeout pages during reclaim */ |
| 3237 | #define RECLAIM_SWAP (1<<2) /* Swap pages out during reclaim */ |
| 3238 | |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3239 | /* |
Christoph Lameter | a92f712 | 2006-02-01 03:05:32 -0800 | [diff] [blame] | 3240 | * Priority for ZONE_RECLAIM. This determines the fraction of pages |
| 3241 | * of a node considered for each zone_reclaim. 4 scans 1/16th of |
| 3242 | * a zone. |
| 3243 | */ |
| 3244 | #define ZONE_RECLAIM_PRIORITY 4 |
| 3245 | |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3246 | /* |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 3247 | * Percentage of pages in a zone that must be unmapped for zone_reclaim to |
| 3248 | * occur. |
| 3249 | */ |
| 3250 | int sysctl_min_unmapped_ratio = 1; |
| 3251 | |
| 3252 | /* |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3253 | * If the number of slab pages in a zone grows beyond this percentage then |
| 3254 | * slab reclaim needs to occur. |
| 3255 | */ |
| 3256 | int sysctl_min_slab_ratio = 5; |
| 3257 | |
Mel Gorman | 90afa5d | 2009-06-16 15:33:20 -0700 | [diff] [blame] | 3258 | static inline unsigned long zone_unmapped_file_pages(struct zone *zone) |
| 3259 | { |
| 3260 | unsigned long file_mapped = zone_page_state(zone, NR_FILE_MAPPED); |
| 3261 | unsigned long file_lru = zone_page_state(zone, NR_INACTIVE_FILE) + |
| 3262 | zone_page_state(zone, NR_ACTIVE_FILE); |
| 3263 | |
| 3264 | /* |
| 3265 | * It's possible for there to be more file mapped pages than |
| 3266 | * accounted for by the pages on the file LRU lists because |
| 3267 | * tmpfs pages accounted for as ANON can also be FILE_MAPPED |
| 3268 | */ |
| 3269 | return (file_lru > file_mapped) ? (file_lru - file_mapped) : 0; |
| 3270 | } |
| 3271 | |
| 3272 | /* Work out how many page cache pages we can reclaim in this reclaim_mode */ |
| 3273 | static long zone_pagecache_reclaimable(struct zone *zone) |
| 3274 | { |
| 3275 | long nr_pagecache_reclaimable; |
| 3276 | long delta = 0; |
| 3277 | |
| 3278 | /* |
| 3279 | * If RECLAIM_SWAP is set, then all file pages are considered |
| 3280 | * potentially reclaimable. Otherwise, we have to worry about |
| 3281 | * pages like swapcache and zone_unmapped_file_pages() provides |
| 3282 | * a better estimate |
| 3283 | */ |
| 3284 | if (zone_reclaim_mode & RECLAIM_SWAP) |
| 3285 | nr_pagecache_reclaimable = zone_page_state(zone, NR_FILE_PAGES); |
| 3286 | else |
| 3287 | nr_pagecache_reclaimable = zone_unmapped_file_pages(zone); |
| 3288 | |
| 3289 | /* If we can't clean pages, remove dirty pages from consideration */ |
| 3290 | if (!(zone_reclaim_mode & RECLAIM_WRITE)) |
| 3291 | delta += zone_page_state(zone, NR_FILE_DIRTY); |
| 3292 | |
| 3293 | /* Watch for any possible underflows due to delta */ |
| 3294 | if (unlikely(delta > nr_pagecache_reclaimable)) |
| 3295 | delta = nr_pagecache_reclaimable; |
| 3296 | |
| 3297 | return nr_pagecache_reclaimable - delta; |
| 3298 | } |
| 3299 | |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3300 | /* |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3301 | * Try to free up some pages from this zone through reclaim. |
| 3302 | */ |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3303 | 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] | 3304 | { |
Christoph Lameter | 7fb2d46 | 2006-03-22 00:08:22 -0800 | [diff] [blame] | 3305 | /* Minimum pages needed in order to stay on node */ |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 3306 | const unsigned long nr_pages = 1 << order; |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3307 | struct task_struct *p = current; |
| 3308 | struct reclaim_state reclaim_state; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3309 | struct scan_control sc = { |
| 3310 | .may_writepage = !!(zone_reclaim_mode & RECLAIM_WRITE), |
Johannes Weiner | a6dc60f8 | 2009-03-31 15:19:30 -0700 | [diff] [blame] | 3311 | .may_unmap = !!(zone_reclaim_mode & RECLAIM_SWAP), |
KOSAKI Motohiro | 2e2e425 | 2009-04-21 12:24:57 -0700 | [diff] [blame] | 3312 | .may_swap = 1, |
Andrew Morton | 62b726c | 2013-02-22 16:32:24 -0800 | [diff] [blame] | 3313 | .nr_to_reclaim = max(nr_pages, SWAP_CLUSTER_MAX), |
Ming Lei | 21caf2f | 2013-02-22 16:34:08 -0800 | [diff] [blame] | 3314 | .gfp_mask = (gfp_mask = memalloc_noio_flags(gfp_mask)), |
Johannes Weiner | bd2f619 | 2009-03-31 15:19:38 -0700 | [diff] [blame] | 3315 | .order = order, |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3316 | .priority = ZONE_RECLAIM_PRIORITY, |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3317 | }; |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 3318 | struct shrink_control shrink = { |
| 3319 | .gfp_mask = sc.gfp_mask, |
| 3320 | }; |
KOSAKI Motohiro | 1574804 | 2010-08-09 17:19:50 -0700 | [diff] [blame] | 3321 | unsigned long nr_slab_pages0, nr_slab_pages1; |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3322 | |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3323 | cond_resched(); |
Christoph Lameter | d4f7796 | 2006-02-24 13:04:22 -0800 | [diff] [blame] | 3324 | /* |
| 3325 | * We need to be able to allocate from the reserves for RECLAIM_SWAP |
| 3326 | * and we also need to be able to write out pages for RECLAIM_WRITE |
| 3327 | * and RECLAIM_SWAP. |
| 3328 | */ |
| 3329 | p->flags |= PF_MEMALLOC | PF_SWAPWRITE; |
KOSAKI Motohiro | 76ca542 | 2010-03-05 13:41:47 -0800 | [diff] [blame] | 3330 | lockdep_set_current_reclaim_state(gfp_mask); |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3331 | reclaim_state.reclaimed_slab = 0; |
| 3332 | p->reclaim_state = &reclaim_state; |
Christoph Lameter | c84db23 | 2006-02-01 03:05:29 -0800 | [diff] [blame] | 3333 | |
Mel Gorman | 90afa5d | 2009-06-16 15:33:20 -0700 | [diff] [blame] | 3334 | if (zone_pagecache_reclaimable(zone) > zone->min_unmapped_pages) { |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3335 | /* |
| 3336 | * Free memory by calling shrink zone with increasing |
| 3337 | * priorities until we have enough memory freed. |
| 3338 | */ |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3339 | do { |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3340 | shrink_zone(zone, &sc); |
| 3341 | } while (sc.nr_reclaimed < nr_pages && --sc.priority >= 0); |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3342 | } |
Christoph Lameter | c84db23 | 2006-02-01 03:05:29 -0800 | [diff] [blame] | 3343 | |
KOSAKI Motohiro | 1574804 | 2010-08-09 17:19:50 -0700 | [diff] [blame] | 3344 | nr_slab_pages0 = zone_page_state(zone, NR_SLAB_RECLAIMABLE); |
| 3345 | if (nr_slab_pages0 > zone->min_slab_pages) { |
Christoph Lameter | 2a16e3f | 2006-02-01 03:05:35 -0800 | [diff] [blame] | 3346 | /* |
Christoph Lameter | 7fb2d46 | 2006-03-22 00:08:22 -0800 | [diff] [blame] | 3347 | * shrink_slab() does not currently allow us to determine how |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3348 | * many pages were freed in this zone. So we take the current |
| 3349 | * number of slab pages and shake the slab until it is reduced |
| 3350 | * by the same nr_pages that we used for reclaiming unmapped |
| 3351 | * pages. |
Christoph Lameter | 2a16e3f | 2006-02-01 03:05:35 -0800 | [diff] [blame] | 3352 | * |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3353 | * Note that shrink_slab will free memory on all zones and may |
| 3354 | * take a long time. |
Christoph Lameter | 2a16e3f | 2006-02-01 03:05:35 -0800 | [diff] [blame] | 3355 | */ |
KOSAKI Motohiro | 4dc4b3d | 2010-08-09 17:19:54 -0700 | [diff] [blame] | 3356 | for (;;) { |
| 3357 | unsigned long lru_pages = zone_reclaimable_pages(zone); |
| 3358 | |
| 3359 | /* No reclaimable slab or very low memory pressure */ |
Ying Han | 1495f23 | 2011-05-24 17:12:27 -0700 | [diff] [blame] | 3360 | if (!shrink_slab(&shrink, sc.nr_scanned, lru_pages)) |
KOSAKI Motohiro | 4dc4b3d | 2010-08-09 17:19:54 -0700 | [diff] [blame] | 3361 | break; |
| 3362 | |
| 3363 | /* Freed enough memory */ |
| 3364 | nr_slab_pages1 = zone_page_state(zone, |
| 3365 | NR_SLAB_RECLAIMABLE); |
| 3366 | if (nr_slab_pages1 + nr_pages <= nr_slab_pages0) |
| 3367 | break; |
| 3368 | } |
Christoph Lameter | 83e33a4 | 2006-09-25 23:31:53 -0700 | [diff] [blame] | 3369 | |
| 3370 | /* |
| 3371 | * Update nr_reclaimed by the number of slab pages we |
| 3372 | * reclaimed from this zone. |
| 3373 | */ |
KOSAKI Motohiro | 1574804 | 2010-08-09 17:19:50 -0700 | [diff] [blame] | 3374 | nr_slab_pages1 = zone_page_state(zone, NR_SLAB_RECLAIMABLE); |
| 3375 | if (nr_slab_pages1 < nr_slab_pages0) |
| 3376 | sc.nr_reclaimed += nr_slab_pages0 - nr_slab_pages1; |
Christoph Lameter | 2a16e3f | 2006-02-01 03:05:35 -0800 | [diff] [blame] | 3377 | } |
| 3378 | |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3379 | p->reclaim_state = NULL; |
Christoph Lameter | d4f7796 | 2006-02-24 13:04:22 -0800 | [diff] [blame] | 3380 | current->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE); |
KOSAKI Motohiro | 76ca542 | 2010-03-05 13:41:47 -0800 | [diff] [blame] | 3381 | lockdep_clear_current_reclaim_state(); |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 3382 | return sc.nr_reclaimed >= nr_pages; |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3383 | } |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3384 | |
| 3385 | int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) |
| 3386 | { |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3387 | int node_id; |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3388 | int ret; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3389 | |
| 3390 | /* |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3391 | * Zone reclaim reclaims unmapped file backed pages and |
| 3392 | * slab pages if we are over the defined limits. |
Christoph Lameter | 34aa133 | 2006-06-30 01:55:37 -0700 | [diff] [blame] | 3393 | * |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 3394 | * A small portion of unmapped file backed pages is needed for |
| 3395 | * file I/O otherwise pages read by file I/O will be immediately |
| 3396 | * thrown out if the zone is overallocated. So we do not reclaim |
| 3397 | * if less than a specified percentage of the zone is used by |
| 3398 | * unmapped file backed pages. |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3399 | */ |
Mel Gorman | 90afa5d | 2009-06-16 15:33:20 -0700 | [diff] [blame] | 3400 | if (zone_pagecache_reclaimable(zone) <= zone->min_unmapped_pages && |
| 3401 | zone_page_state(zone, NR_SLAB_RECLAIMABLE) <= zone->min_slab_pages) |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 3402 | return ZONE_RECLAIM_FULL; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3403 | |
KOSAKI Motohiro | 93e4a89 | 2010-03-05 13:41:55 -0800 | [diff] [blame] | 3404 | if (zone->all_unreclaimable) |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 3405 | return ZONE_RECLAIM_FULL; |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3406 | |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3407 | /* |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3408 | * Do not scan if the allocation should not be delayed. |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3409 | */ |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3410 | if (!(gfp_mask & __GFP_WAIT) || (current->flags & PF_MEMALLOC)) |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 3411 | return ZONE_RECLAIM_NOSCAN; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3412 | |
| 3413 | /* |
| 3414 | * Only run zone reclaim on the local zone or on zones that do not |
| 3415 | * have associated processors. This will favor the local processor |
| 3416 | * over remote processors and spread off node memory allocations |
| 3417 | * as wide as possible. |
| 3418 | */ |
Christoph Lameter | 89fa302 | 2006-09-25 23:31:55 -0700 | [diff] [blame] | 3419 | node_id = zone_to_nid(zone); |
Christoph Lameter | 37c0708 | 2007-10-16 01:25:36 -0700 | [diff] [blame] | 3420 | if (node_state(node_id, N_CPU) && node_id != numa_node_id()) |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 3421 | return ZONE_RECLAIM_NOSCAN; |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3422 | |
| 3423 | if (zone_test_and_set_flag(zone, ZONE_RECLAIM_LOCKED)) |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 3424 | return ZONE_RECLAIM_NOSCAN; |
| 3425 | |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3426 | ret = __zone_reclaim(zone, gfp_mask, order); |
| 3427 | zone_clear_flag(zone, ZONE_RECLAIM_LOCKED); |
| 3428 | |
Mel Gorman | 24cf72518 | 2009-06-16 15:33:23 -0700 | [diff] [blame] | 3429 | if (!ret) |
| 3430 | count_vm_event(PGSCAN_ZONE_RECLAIM_FAILED); |
| 3431 | |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3432 | return ret; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3433 | } |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3434 | #endif |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3435 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3436 | /* |
| 3437 | * page_evictable - test whether a page is evictable |
| 3438 | * @page: the page to test |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3439 | * |
| 3440 | * Test whether page is evictable--i.e., should be placed on active/inactive |
Hugh Dickins | 39b5f29 | 2012-10-08 16:33:18 -0700 | [diff] [blame] | 3441 | * lists vs unevictable list. |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3442 | * |
| 3443 | * Reasons page might not be evictable: |
Lee Schermerhorn | ba9ddf4 | 2008-10-18 20:26:42 -0700 | [diff] [blame] | 3444 | * (1) page's mapping marked unevictable |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 3445 | * (2) page is part of an mlocked VMA |
Lee Schermerhorn | ba9ddf4 | 2008-10-18 20:26:42 -0700 | [diff] [blame] | 3446 | * |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3447 | */ |
Hugh Dickins | 39b5f29 | 2012-10-08 16:33:18 -0700 | [diff] [blame] | 3448 | int page_evictable(struct page *page) |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3449 | { |
Hugh Dickins | 39b5f29 | 2012-10-08 16:33:18 -0700 | [diff] [blame] | 3450 | return !mapping_unevictable(page_mapping(page)) && !PageMlocked(page); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3451 | } |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3452 | |
Hugh Dickins | 85046579 | 2012-01-20 14:34:19 -0800 | [diff] [blame] | 3453 | #ifdef CONFIG_SHMEM |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3454 | /** |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3455 | * check_move_unevictable_pages - check pages for evictability and move to appropriate zone lru list |
| 3456 | * @pages: array of pages to check |
| 3457 | * @nr_pages: number of pages to check |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3458 | * |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3459 | * Checks pages for evictability and moves them to the appropriate lru list. |
Hugh Dickins | 85046579 | 2012-01-20 14:34:19 -0800 | [diff] [blame] | 3460 | * |
| 3461 | * This function is only used for SysV IPC SHM_UNLOCK. |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3462 | */ |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3463 | void check_move_unevictable_pages(struct page **pages, int nr_pages) |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3464 | { |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 3465 | struct lruvec *lruvec; |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3466 | struct zone *zone = NULL; |
| 3467 | int pgscanned = 0; |
| 3468 | int pgrescued = 0; |
| 3469 | int i; |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3470 | |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3471 | for (i = 0; i < nr_pages; i++) { |
| 3472 | struct page *page = pages[i]; |
| 3473 | struct zone *pagezone; |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3474 | |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3475 | pgscanned++; |
| 3476 | pagezone = page_zone(page); |
| 3477 | if (pagezone != zone) { |
| 3478 | if (zone) |
| 3479 | spin_unlock_irq(&zone->lru_lock); |
| 3480 | zone = pagezone; |
| 3481 | spin_lock_irq(&zone->lru_lock); |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3482 | } |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 3483 | lruvec = mem_cgroup_page_lruvec(page, zone); |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3484 | |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3485 | if (!PageLRU(page) || !PageUnevictable(page)) |
| 3486 | continue; |
| 3487 | |
Hugh Dickins | 39b5f29 | 2012-10-08 16:33:18 -0700 | [diff] [blame] | 3488 | if (page_evictable(page)) { |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3489 | enum lru_list lru = page_lru_base_type(page); |
| 3490 | |
| 3491 | VM_BUG_ON(PageActive(page)); |
| 3492 | ClearPageUnevictable(page); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 3493 | del_page_from_lru_list(page, lruvec, LRU_UNEVICTABLE); |
| 3494 | add_page_to_lru_list(page, lruvec, lru); |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3495 | pgrescued++; |
| 3496 | } |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3497 | } |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3498 | |
| 3499 | if (zone) { |
| 3500 | __count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued); |
| 3501 | __count_vm_events(UNEVICTABLE_PGSCANNED, pgscanned); |
| 3502 | spin_unlock_irq(&zone->lru_lock); |
| 3503 | } |
Hugh Dickins | 85046579 | 2012-01-20 14:34:19 -0800 | [diff] [blame] | 3504 | } |
| 3505 | #endif /* CONFIG_SHMEM */ |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3506 | |
Johannes Weiner | 264e56d | 2011-10-31 17:09:13 -0700 | [diff] [blame] | 3507 | static void warn_scan_unevictable_pages(void) |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3508 | { |
Johannes Weiner | 264e56d | 2011-10-31 17:09:13 -0700 | [diff] [blame] | 3509 | printk_once(KERN_WARNING |
KOSAKI Motohiro | 25bd91b | 2012-01-10 15:07:40 -0800 | [diff] [blame] | 3510 | "%s: The scan_unevictable_pages sysctl/node-interface has been " |
Johannes Weiner | 264e56d | 2011-10-31 17:09:13 -0700 | [diff] [blame] | 3511 | "disabled for lack of a legitimate use case. If you have " |
KOSAKI Motohiro | 25bd91b | 2012-01-10 15:07:40 -0800 | [diff] [blame] | 3512 | "one, please send an email to linux-mm@kvack.org.\n", |
| 3513 | current->comm); |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3514 | } |
| 3515 | |
| 3516 | /* |
| 3517 | * scan_unevictable_pages [vm] sysctl handler. On demand re-scan of |
| 3518 | * all nodes' unevictable lists for evictable pages |
| 3519 | */ |
| 3520 | unsigned long scan_unevictable_pages; |
| 3521 | |
| 3522 | int scan_unevictable_handler(struct ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 3523 | void __user *buffer, |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3524 | size_t *length, loff_t *ppos) |
| 3525 | { |
Johannes Weiner | 264e56d | 2011-10-31 17:09:13 -0700 | [diff] [blame] | 3526 | warn_scan_unevictable_pages(); |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 3527 | proc_doulongvec_minmax(table, write, buffer, length, ppos); |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3528 | scan_unevictable_pages = 0; |
| 3529 | return 0; |
| 3530 | } |
| 3531 | |
Thadeu Lima de Souza Cascardo | e4455ab | 2010-10-26 14:21:28 -0700 | [diff] [blame] | 3532 | #ifdef CONFIG_NUMA |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3533 | /* |
| 3534 | * per node 'scan_unevictable_pages' attribute. On demand re-scan of |
| 3535 | * a specified node's per zone unevictable lists for evictable pages. |
| 3536 | */ |
| 3537 | |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 3538 | static ssize_t read_scan_unevictable_node(struct device *dev, |
| 3539 | struct device_attribute *attr, |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3540 | char *buf) |
| 3541 | { |
Johannes Weiner | 264e56d | 2011-10-31 17:09:13 -0700 | [diff] [blame] | 3542 | warn_scan_unevictable_pages(); |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3543 | return sprintf(buf, "0\n"); /* always zero; should fit... */ |
| 3544 | } |
| 3545 | |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 3546 | static ssize_t write_scan_unevictable_node(struct device *dev, |
| 3547 | struct device_attribute *attr, |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3548 | const char *buf, size_t count) |
| 3549 | { |
Johannes Weiner | 264e56d | 2011-10-31 17:09:13 -0700 | [diff] [blame] | 3550 | warn_scan_unevictable_pages(); |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3551 | return 1; |
| 3552 | } |
| 3553 | |
| 3554 | |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 3555 | static DEVICE_ATTR(scan_unevictable_pages, S_IRUGO | S_IWUSR, |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3556 | read_scan_unevictable_node, |
| 3557 | write_scan_unevictable_node); |
| 3558 | |
| 3559 | int scan_unevictable_register_node(struct node *node) |
| 3560 | { |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 3561 | return device_create_file(&node->dev, &dev_attr_scan_unevictable_pages); |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3562 | } |
| 3563 | |
| 3564 | void scan_unevictable_unregister_node(struct node *node) |
| 3565 | { |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 3566 | device_remove_file(&node->dev, &dev_attr_scan_unevictable_pages); |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3567 | } |
Thadeu Lima de Souza Cascardo | e4455ab | 2010-10-26 14:21:28 -0700 | [diff] [blame] | 3568 | #endif |