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 | |
Mitchel Humpherys | b1de0d1 | 2014-06-06 14:38:30 -0700 | [diff] [blame] | 14 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
| 15 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 16 | #include <linux/mm.h> |
Ingo Molnar | 5b3cc15 | 2017-02-02 20:43:54 +0100 | [diff] [blame] | 17 | #include <linux/sched/mm.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | #include <linux/module.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 19 | #include <linux/gfp.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 20 | #include <linux/kernel_stat.h> |
| 21 | #include <linux/swap.h> |
| 22 | #include <linux/pagemap.h> |
| 23 | #include <linux/init.h> |
| 24 | #include <linux/highmem.h> |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 25 | #include <linux/vmpressure.h> |
Andrew Morton | e129b5c | 2006-09-27 01:50:00 -0700 | [diff] [blame] | 26 | #include <linux/vmstat.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 27 | #include <linux/file.h> |
| 28 | #include <linux/writeback.h> |
| 29 | #include <linux/blkdev.h> |
| 30 | #include <linux/buffer_head.h> /* for try_to_release_page(), |
| 31 | buffer_heads_over_limit */ |
| 32 | #include <linux/mm_inline.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 33 | #include <linux/backing-dev.h> |
| 34 | #include <linux/rmap.h> |
| 35 | #include <linux/topology.h> |
| 36 | #include <linux/cpu.h> |
| 37 | #include <linux/cpuset.h> |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 38 | #include <linux/compaction.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 39 | #include <linux/notifier.h> |
| 40 | #include <linux/rwsem.h> |
Rafael J. Wysocki | 248a030 | 2006-03-22 00:09:04 -0800 | [diff] [blame] | 41 | #include <linux/delay.h> |
Yasunori Goto | 3218ae1 | 2006-06-27 02:53:33 -0700 | [diff] [blame] | 42 | #include <linux/kthread.h> |
Nigel Cunningham | 7dfb710 | 2006-12-06 20:34:23 -0800 | [diff] [blame] | 43 | #include <linux/freezer.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 44 | #include <linux/memcontrol.h> |
Keika Kobayashi | 873b477 | 2008-07-25 01:48:52 -0700 | [diff] [blame] | 45 | #include <linux/delayacct.h> |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 46 | #include <linux/sysctl.h> |
KOSAKI Motohiro | 929bea7 | 2011-04-14 15:22:12 -0700 | [diff] [blame] | 47 | #include <linux/oom.h> |
Linus Torvalds | 268bb0c | 2011-05-20 12:50:29 -0700 | [diff] [blame] | 48 | #include <linux/prefetch.h> |
Mitchel Humpherys | b1de0d1 | 2014-06-06 14:38:30 -0700 | [diff] [blame] | 49 | #include <linux/printk.h> |
Ross Zwisler | f9fe48b | 2016-01-22 15:10:40 -0800 | [diff] [blame] | 50 | #include <linux/dax.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 51 | |
| 52 | #include <asm/tlbflush.h> |
| 53 | #include <asm/div64.h> |
| 54 | |
| 55 | #include <linux/swapops.h> |
Rafael Aquini | 117aad1 | 2013-09-30 13:45:16 -0700 | [diff] [blame] | 56 | #include <linux/balloon_compaction.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 57 | |
Nick Piggin | 0f8053a | 2006-03-22 00:08:33 -0800 | [diff] [blame] | 58 | #include "internal.h" |
| 59 | |
Mel Gorman | 33906bc | 2010-08-09 17:19:16 -0700 | [diff] [blame] | 60 | #define CREATE_TRACE_POINTS |
| 61 | #include <trace/events/vmscan.h> |
| 62 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 63 | struct scan_control { |
KOSAKI Motohiro | 22fba33 | 2009-12-14 17:59:10 -0800 | [diff] [blame] | 64 | /* How many pages shrink_list() should reclaim */ |
| 65 | unsigned long nr_to_reclaim; |
| 66 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 67 | /* This context's GFP mask */ |
Al Viro | 6daa0e2 | 2005-10-21 03:18:50 -0400 | [diff] [blame] | 68 | gfp_t gfp_mask; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 69 | |
Johannes Weiner | ee814fe | 2014-08-06 16:06:19 -0700 | [diff] [blame] | 70 | /* Allocation order */ |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 71 | int order; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 72 | |
Johannes Weiner | ee814fe | 2014-08-06 16:06:19 -0700 | [diff] [blame] | 73 | /* |
| 74 | * Nodemask of nodes allowed by the caller. If NULL, all nodes |
| 75 | * are scanned. |
| 76 | */ |
| 77 | nodemask_t *nodemask; |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 78 | |
KOSAKI Motohiro | 5f53e76 | 2010-05-24 14:32:37 -0700 | [diff] [blame] | 79 | /* |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 80 | * The memory cgroup that hit its limit and as a result is the |
| 81 | * primary target of this reclaim invocation. |
| 82 | */ |
| 83 | struct mem_cgroup *target_mem_cgroup; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 84 | |
Johannes Weiner | ee814fe | 2014-08-06 16:06:19 -0700 | [diff] [blame] | 85 | /* Scan (total_size >> priority) pages at once */ |
| 86 | int priority; |
| 87 | |
Mel Gorman | b2e1875 | 2016-07-28 15:45:37 -0700 | [diff] [blame] | 88 | /* The highest zone to isolate pages for reclaim from */ |
| 89 | enum zone_type reclaim_idx; |
| 90 | |
Johannes Weiner | 1276ad6 | 2017-02-24 14:56:11 -0800 | [diff] [blame] | 91 | /* Writepage batching in laptop mode; RECLAIM_WRITE */ |
Johannes Weiner | ee814fe | 2014-08-06 16:06:19 -0700 | [diff] [blame] | 92 | unsigned int may_writepage:1; |
| 93 | |
| 94 | /* Can mapped pages be reclaimed? */ |
| 95 | unsigned int may_unmap:1; |
| 96 | |
| 97 | /* Can pages be swapped as part of reclaim? */ |
| 98 | unsigned int may_swap:1; |
| 99 | |
Yisheng Xie | d6622f6 | 2017-05-03 14:53:57 -0700 | [diff] [blame] | 100 | /* |
| 101 | * Cgroups are not reclaimed below their configured memory.low, |
| 102 | * unless we threaten to OOM. If any cgroups are skipped due to |
| 103 | * memory.low and nothing was reclaimed, go back for memory.low. |
| 104 | */ |
| 105 | unsigned int memcg_low_reclaim:1; |
| 106 | unsigned int memcg_low_skipped:1; |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 107 | |
Johannes Weiner | ee814fe | 2014-08-06 16:06:19 -0700 | [diff] [blame] | 108 | unsigned int hibernation_mode:1; |
| 109 | |
| 110 | /* One of the zones is ready for compaction */ |
| 111 | unsigned int compaction_ready:1; |
| 112 | |
| 113 | /* Incremented by the number of inactive pages that were scanned */ |
| 114 | unsigned long nr_scanned; |
| 115 | |
| 116 | /* Number of pages freed so far during a call to shrink_zones() */ |
| 117 | unsigned long nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 118 | }; |
| 119 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 120 | #ifdef ARCH_HAS_PREFETCH |
| 121 | #define prefetch_prev_lru_page(_page, _base, _field) \ |
| 122 | do { \ |
| 123 | if ((_page)->lru.prev != _base) { \ |
| 124 | struct page *prev; \ |
| 125 | \ |
| 126 | prev = lru_to_page(&(_page->lru)); \ |
| 127 | prefetch(&prev->_field); \ |
| 128 | } \ |
| 129 | } while (0) |
| 130 | #else |
| 131 | #define prefetch_prev_lru_page(_page, _base, _field) do { } while (0) |
| 132 | #endif |
| 133 | |
| 134 | #ifdef ARCH_HAS_PREFETCHW |
| 135 | #define prefetchw_prev_lru_page(_page, _base, _field) \ |
| 136 | do { \ |
| 137 | if ((_page)->lru.prev != _base) { \ |
| 138 | struct page *prev; \ |
| 139 | \ |
| 140 | prev = lru_to_page(&(_page->lru)); \ |
| 141 | prefetchw(&prev->_field); \ |
| 142 | } \ |
| 143 | } while (0) |
| 144 | #else |
| 145 | #define prefetchw_prev_lru_page(_page, _base, _field) do { } while (0) |
| 146 | #endif |
| 147 | |
| 148 | /* |
| 149 | * From 0 .. 100. Higher means more swappy. |
| 150 | */ |
| 151 | int vm_swappiness = 60; |
Wang Sheng-Hui | d0480be | 2014-08-06 16:07:07 -0700 | [diff] [blame] | 152 | /* |
| 153 | * The total number of pages which are beyond the high watermark within all |
| 154 | * zones. |
| 155 | */ |
| 156 | unsigned long vm_total_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 157 | |
| 158 | static LIST_HEAD(shrinker_list); |
| 159 | static DECLARE_RWSEM(shrinker_rwsem); |
| 160 | |
Andrew Morton | c255a45 | 2012-07-31 16:43:02 -0700 | [diff] [blame] | 161 | #ifdef CONFIG_MEMCG |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 162 | static bool global_reclaim(struct scan_control *sc) |
| 163 | { |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 164 | return !sc->target_mem_cgroup; |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 165 | } |
Tejun Heo | 97c9341 | 2015-05-22 18:23:36 -0400 | [diff] [blame] | 166 | |
| 167 | /** |
| 168 | * sane_reclaim - is the usual dirty throttling mechanism operational? |
| 169 | * @sc: scan_control in question |
| 170 | * |
| 171 | * The normal page dirty throttling mechanism in balance_dirty_pages() is |
| 172 | * completely broken with the legacy memcg and direct stalling in |
| 173 | * shrink_page_list() is used for throttling instead, which lacks all the |
| 174 | * niceties such as fairness, adaptive pausing, bandwidth proportional |
| 175 | * allocation and configurability. |
| 176 | * |
| 177 | * This function tests whether the vmscan currently in progress can assume |
| 178 | * that the normal dirty throttling mechanism is operational. |
| 179 | */ |
| 180 | static bool sane_reclaim(struct scan_control *sc) |
| 181 | { |
| 182 | struct mem_cgroup *memcg = sc->target_mem_cgroup; |
| 183 | |
| 184 | if (!memcg) |
| 185 | return true; |
| 186 | #ifdef CONFIG_CGROUP_WRITEBACK |
Linus Torvalds | 69234ac | 2015-11-05 14:51:32 -0800 | [diff] [blame] | 187 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
Tejun Heo | 97c9341 | 2015-05-22 18:23:36 -0400 | [diff] [blame] | 188 | return true; |
| 189 | #endif |
| 190 | return false; |
| 191 | } |
KAMEZAWA Hiroyuki | 91a4547 | 2008-02-07 00:14:29 -0800 | [diff] [blame] | 192 | #else |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 193 | static bool global_reclaim(struct scan_control *sc) |
| 194 | { |
| 195 | return true; |
| 196 | } |
Tejun Heo | 97c9341 | 2015-05-22 18:23:36 -0400 | [diff] [blame] | 197 | |
| 198 | static bool sane_reclaim(struct scan_control *sc) |
| 199 | { |
| 200 | return true; |
| 201 | } |
KAMEZAWA Hiroyuki | 91a4547 | 2008-02-07 00:14:29 -0800 | [diff] [blame] | 202 | #endif |
| 203 | |
Mel Gorman | 5a1c84b | 2016-07-28 15:47:31 -0700 | [diff] [blame] | 204 | /* |
| 205 | * This misses isolated pages which are not accounted for to save counters. |
| 206 | * As the data only determines if reclaim or compaction continues, it is |
| 207 | * not expected that isolated pages will be a dominating factor. |
| 208 | */ |
| 209 | unsigned long zone_reclaimable_pages(struct zone *zone) |
| 210 | { |
| 211 | unsigned long nr; |
| 212 | |
| 213 | nr = zone_page_state_snapshot(zone, NR_ZONE_INACTIVE_FILE) + |
| 214 | zone_page_state_snapshot(zone, NR_ZONE_ACTIVE_FILE); |
| 215 | if (get_nr_swap_pages() > 0) |
| 216 | nr += zone_page_state_snapshot(zone, NR_ZONE_INACTIVE_ANON) + |
| 217 | zone_page_state_snapshot(zone, NR_ZONE_ACTIVE_ANON); |
| 218 | |
| 219 | return nr; |
| 220 | } |
| 221 | |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 222 | unsigned long pgdat_reclaimable_pages(struct pglist_data *pgdat) |
Lisa Du | 6e543d5 | 2013-09-11 14:22:36 -0700 | [diff] [blame] | 223 | { |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 224 | unsigned long nr; |
| 225 | |
| 226 | nr = node_page_state_snapshot(pgdat, NR_ACTIVE_FILE) + |
| 227 | node_page_state_snapshot(pgdat, NR_INACTIVE_FILE) + |
| 228 | node_page_state_snapshot(pgdat, NR_ISOLATED_FILE); |
| 229 | |
| 230 | if (get_nr_swap_pages() > 0) |
| 231 | nr += node_page_state_snapshot(pgdat, NR_ACTIVE_ANON) + |
| 232 | node_page_state_snapshot(pgdat, NR_INACTIVE_ANON) + |
| 233 | node_page_state_snapshot(pgdat, NR_ISOLATED_ANON); |
| 234 | |
| 235 | return nr; |
| 236 | } |
| 237 | |
Michal Hocko | fd53880 | 2017-02-22 15:45:58 -0800 | [diff] [blame] | 238 | /** |
| 239 | * lruvec_lru_size - Returns the number of pages on the given LRU list. |
| 240 | * @lruvec: lru vector |
| 241 | * @lru: lru to use |
| 242 | * @zone_idx: zones to consider (use MAX_NR_ZONES for the whole LRU list) |
| 243 | */ |
| 244 | unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx) |
KOSAKI Motohiro | c9f299d | 2009-01-07 18:08:16 -0800 | [diff] [blame] | 245 | { |
Michal Hocko | fd53880 | 2017-02-22 15:45:58 -0800 | [diff] [blame] | 246 | unsigned long lru_size; |
| 247 | int zid; |
| 248 | |
Hugh Dickins | c3c787e | 2012-05-29 15:06:52 -0700 | [diff] [blame] | 249 | if (!mem_cgroup_disabled()) |
Michal Hocko | fd53880 | 2017-02-22 15:45:58 -0800 | [diff] [blame] | 250 | lru_size = mem_cgroup_get_lru_size(lruvec, lru); |
| 251 | else |
| 252 | lru_size = node_page_state(lruvec_pgdat(lruvec), NR_LRU_BASE + lru); |
KOSAKI Motohiro | a3d8e05 | 2009-01-07 18:08:19 -0800 | [diff] [blame] | 253 | |
Michal Hocko | fd53880 | 2017-02-22 15:45:58 -0800 | [diff] [blame] | 254 | for (zid = zone_idx + 1; zid < MAX_NR_ZONES; zid++) { |
| 255 | struct zone *zone = &lruvec_pgdat(lruvec)->node_zones[zid]; |
| 256 | unsigned long size; |
KOSAKI Motohiro | c9f299d | 2009-01-07 18:08:16 -0800 | [diff] [blame] | 257 | |
Michal Hocko | fd53880 | 2017-02-22 15:45:58 -0800 | [diff] [blame] | 258 | if (!managed_zone(zone)) |
| 259 | continue; |
Michal Hocko | b4536f0c8 | 2017-01-10 16:58:04 -0800 | [diff] [blame] | 260 | |
Michal Hocko | fd53880 | 2017-02-22 15:45:58 -0800 | [diff] [blame] | 261 | if (!mem_cgroup_disabled()) |
| 262 | size = mem_cgroup_get_zone_lru_size(lruvec, lru, zid); |
| 263 | else |
| 264 | size = zone_page_state(&lruvec_pgdat(lruvec)->node_zones[zid], |
| 265 | NR_ZONE_LRU_BASE + lru); |
| 266 | lru_size -= min(size, lru_size); |
| 267 | } |
| 268 | |
| 269 | return lru_size; |
| 270 | |
Michal Hocko | b4536f0c8 | 2017-01-10 16:58:04 -0800 | [diff] [blame] | 271 | } |
| 272 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 273 | /* |
Glauber Costa | 1d3d443 | 2013-08-28 10:18:04 +1000 | [diff] [blame] | 274 | * Add a shrinker callback to be called from the vm. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 275 | */ |
Glauber Costa | 1d3d443 | 2013-08-28 10:18:04 +1000 | [diff] [blame] | 276 | int register_shrinker(struct shrinker *shrinker) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 277 | { |
Glauber Costa | 1d3d443 | 2013-08-28 10:18:04 +1000 | [diff] [blame] | 278 | size_t size = sizeof(*shrinker->nr_deferred); |
| 279 | |
Glauber Costa | 1d3d443 | 2013-08-28 10:18:04 +1000 | [diff] [blame] | 280 | if (shrinker->flags & SHRINKER_NUMA_AWARE) |
| 281 | size *= nr_node_ids; |
| 282 | |
| 283 | shrinker->nr_deferred = kzalloc(size, GFP_KERNEL); |
| 284 | if (!shrinker->nr_deferred) |
| 285 | return -ENOMEM; |
| 286 | |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 287 | down_write(&shrinker_rwsem); |
| 288 | list_add_tail(&shrinker->list, &shrinker_list); |
| 289 | up_write(&shrinker_rwsem); |
Glauber Costa | 1d3d443 | 2013-08-28 10:18:04 +1000 | [diff] [blame] | 290 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 291 | } |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 292 | EXPORT_SYMBOL(register_shrinker); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 293 | |
| 294 | /* |
| 295 | * Remove one |
| 296 | */ |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 297 | void unregister_shrinker(struct shrinker *shrinker) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 298 | { |
| 299 | down_write(&shrinker_rwsem); |
| 300 | list_del(&shrinker->list); |
| 301 | up_write(&shrinker_rwsem); |
Andrew Vagin | ae39332 | 2013-10-16 13:46:46 -0700 | [diff] [blame] | 302 | kfree(shrinker->nr_deferred); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 303 | } |
Rusty Russell | 8e1f936 | 2007-07-17 04:03:17 -0700 | [diff] [blame] | 304 | EXPORT_SYMBOL(unregister_shrinker); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 305 | |
| 306 | #define SHRINK_BATCH 128 |
Glauber Costa | 1d3d443 | 2013-08-28 10:18:04 +1000 | [diff] [blame] | 307 | |
Vladimir Davydov | cb731d6 | 2015-02-12 14:58:54 -0800 | [diff] [blame] | 308 | static unsigned long do_shrink_slab(struct shrink_control *shrinkctl, |
| 309 | struct shrinker *shrinker, |
| 310 | unsigned long nr_scanned, |
| 311 | unsigned long nr_eligible) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 312 | { |
Glauber Costa | 1d3d443 | 2013-08-28 10:18:04 +1000 | [diff] [blame] | 313 | unsigned long freed = 0; |
| 314 | unsigned long long delta; |
| 315 | long total_scan; |
Vladimir Davydov | d5bc5fd | 2014-04-03 14:47:32 -0700 | [diff] [blame] | 316 | long freeable; |
Glauber Costa | 1d3d443 | 2013-08-28 10:18:04 +1000 | [diff] [blame] | 317 | long nr; |
| 318 | long new_nr; |
| 319 | int nid = shrinkctl->nid; |
| 320 | long batch_size = shrinker->batch ? shrinker->batch |
| 321 | : SHRINK_BATCH; |
Shaohua Li | 5f33a08 | 2016-12-12 16:41:50 -0800 | [diff] [blame] | 322 | long scanned = 0, next_deferred; |
Glauber Costa | 1d3d443 | 2013-08-28 10:18:04 +1000 | [diff] [blame] | 323 | |
Vladimir Davydov | d5bc5fd | 2014-04-03 14:47:32 -0700 | [diff] [blame] | 324 | freeable = shrinker->count_objects(shrinker, shrinkctl); |
| 325 | if (freeable == 0) |
Glauber Costa | 1d3d443 | 2013-08-28 10:18:04 +1000 | [diff] [blame] | 326 | return 0; |
| 327 | |
| 328 | /* |
| 329 | * copy the current shrinker scan count into a local variable |
| 330 | * and zero it so that other concurrent shrinker invocations |
| 331 | * don't also do this scanning work. |
| 332 | */ |
| 333 | nr = atomic_long_xchg(&shrinker->nr_deferred[nid], 0); |
| 334 | |
| 335 | total_scan = nr; |
Johannes Weiner | 6b4f779 | 2014-12-12 16:56:13 -0800 | [diff] [blame] | 336 | delta = (4 * nr_scanned) / shrinker->seeks; |
Vladimir Davydov | d5bc5fd | 2014-04-03 14:47:32 -0700 | [diff] [blame] | 337 | delta *= freeable; |
Johannes Weiner | 6b4f779 | 2014-12-12 16:56:13 -0800 | [diff] [blame] | 338 | do_div(delta, nr_eligible + 1); |
Glauber Costa | 1d3d443 | 2013-08-28 10:18:04 +1000 | [diff] [blame] | 339 | total_scan += delta; |
| 340 | if (total_scan < 0) { |
Pintu Kumar | 8612c66 | 2014-12-10 15:42:58 -0800 | [diff] [blame] | 341 | pr_err("shrink_slab: %pF negative objects to delete nr=%ld\n", |
Dave Chinner | a0b0213 | 2013-08-28 10:18:16 +1000 | [diff] [blame] | 342 | shrinker->scan_objects, total_scan); |
Vladimir Davydov | d5bc5fd | 2014-04-03 14:47:32 -0700 | [diff] [blame] | 343 | total_scan = freeable; |
Shaohua Li | 5f33a08 | 2016-12-12 16:41:50 -0800 | [diff] [blame] | 344 | next_deferred = nr; |
| 345 | } else |
| 346 | next_deferred = total_scan; |
Glauber Costa | 1d3d443 | 2013-08-28 10:18:04 +1000 | [diff] [blame] | 347 | |
| 348 | /* |
| 349 | * We need to avoid excessive windup on filesystem shrinkers |
| 350 | * due to large numbers of GFP_NOFS allocations causing the |
| 351 | * shrinkers to return -1 all the time. This results in a large |
| 352 | * nr being built up so when a shrink that can do some work |
| 353 | * comes along it empties the entire cache due to nr >>> |
Vladimir Davydov | d5bc5fd | 2014-04-03 14:47:32 -0700 | [diff] [blame] | 354 | * freeable. This is bad for sustaining a working set in |
Glauber Costa | 1d3d443 | 2013-08-28 10:18:04 +1000 | [diff] [blame] | 355 | * memory. |
| 356 | * |
| 357 | * Hence only allow the shrinker to scan the entire cache when |
| 358 | * a large delta change is calculated directly. |
| 359 | */ |
Vladimir Davydov | d5bc5fd | 2014-04-03 14:47:32 -0700 | [diff] [blame] | 360 | if (delta < freeable / 4) |
| 361 | total_scan = min(total_scan, freeable / 2); |
Glauber Costa | 1d3d443 | 2013-08-28 10:18:04 +1000 | [diff] [blame] | 362 | |
| 363 | /* |
| 364 | * Avoid risking looping forever due to too large nr value: |
| 365 | * never try to free more than twice the estimate number of |
| 366 | * freeable entries. |
| 367 | */ |
Vladimir Davydov | d5bc5fd | 2014-04-03 14:47:32 -0700 | [diff] [blame] | 368 | if (total_scan > freeable * 2) |
| 369 | total_scan = freeable * 2; |
Glauber Costa | 1d3d443 | 2013-08-28 10:18:04 +1000 | [diff] [blame] | 370 | |
| 371 | trace_mm_shrink_slab_start(shrinker, shrinkctl, nr, |
Johannes Weiner | 6b4f779 | 2014-12-12 16:56:13 -0800 | [diff] [blame] | 372 | nr_scanned, nr_eligible, |
| 373 | freeable, delta, total_scan); |
Glauber Costa | 1d3d443 | 2013-08-28 10:18:04 +1000 | [diff] [blame] | 374 | |
Vladimir Davydov | 0b1fb40 | 2014-01-23 15:53:22 -0800 | [diff] [blame] | 375 | /* |
| 376 | * Normally, we should not scan less than batch_size objects in one |
| 377 | * pass to avoid too frequent shrinker calls, but if the slab has less |
| 378 | * than batch_size objects in total and we are really tight on memory, |
| 379 | * we will try to reclaim all available objects, otherwise we can end |
| 380 | * up failing allocations although there are plenty of reclaimable |
| 381 | * objects spread over several slabs with usage less than the |
| 382 | * batch_size. |
| 383 | * |
| 384 | * We detect the "tight on memory" situations by looking at the total |
| 385 | * number of objects we want to scan (total_scan). If it is greater |
Vladimir Davydov | d5bc5fd | 2014-04-03 14:47:32 -0700 | [diff] [blame] | 386 | * than the total number of objects on slab (freeable), we must be |
Vladimir Davydov | 0b1fb40 | 2014-01-23 15:53:22 -0800 | [diff] [blame] | 387 | * scanning at high prio and therefore should try to reclaim as much as |
| 388 | * possible. |
| 389 | */ |
| 390 | while (total_scan >= batch_size || |
Vladimir Davydov | d5bc5fd | 2014-04-03 14:47:32 -0700 | [diff] [blame] | 391 | total_scan >= freeable) { |
Dave Chinner | a0b0213 | 2013-08-28 10:18:16 +1000 | [diff] [blame] | 392 | unsigned long ret; |
Vladimir Davydov | 0b1fb40 | 2014-01-23 15:53:22 -0800 | [diff] [blame] | 393 | unsigned long nr_to_scan = min(batch_size, total_scan); |
Glauber Costa | 1d3d443 | 2013-08-28 10:18:04 +1000 | [diff] [blame] | 394 | |
Vladimir Davydov | 0b1fb40 | 2014-01-23 15:53:22 -0800 | [diff] [blame] | 395 | shrinkctl->nr_to_scan = nr_to_scan; |
Dave Chinner | a0b0213 | 2013-08-28 10:18:16 +1000 | [diff] [blame] | 396 | ret = shrinker->scan_objects(shrinker, shrinkctl); |
| 397 | if (ret == SHRINK_STOP) |
| 398 | break; |
| 399 | freed += ret; |
Glauber Costa | 1d3d443 | 2013-08-28 10:18:04 +1000 | [diff] [blame] | 400 | |
Vladimir Davydov | 0b1fb40 | 2014-01-23 15:53:22 -0800 | [diff] [blame] | 401 | count_vm_events(SLABS_SCANNED, nr_to_scan); |
| 402 | total_scan -= nr_to_scan; |
Shaohua Li | 5f33a08 | 2016-12-12 16:41:50 -0800 | [diff] [blame] | 403 | scanned += nr_to_scan; |
Glauber Costa | 1d3d443 | 2013-08-28 10:18:04 +1000 | [diff] [blame] | 404 | |
| 405 | cond_resched(); |
| 406 | } |
| 407 | |
Shaohua Li | 5f33a08 | 2016-12-12 16:41:50 -0800 | [diff] [blame] | 408 | if (next_deferred >= scanned) |
| 409 | next_deferred -= scanned; |
| 410 | else |
| 411 | next_deferred = 0; |
Glauber Costa | 1d3d443 | 2013-08-28 10:18:04 +1000 | [diff] [blame] | 412 | /* |
| 413 | * move the unused scan count back into the shrinker in a |
| 414 | * manner that handles concurrent updates. If we exhausted the |
| 415 | * scan, there is no need to do an update. |
| 416 | */ |
Shaohua Li | 5f33a08 | 2016-12-12 16:41:50 -0800 | [diff] [blame] | 417 | if (next_deferred > 0) |
| 418 | new_nr = atomic_long_add_return(next_deferred, |
Glauber Costa | 1d3d443 | 2013-08-28 10:18:04 +1000 | [diff] [blame] | 419 | &shrinker->nr_deferred[nid]); |
| 420 | else |
| 421 | new_nr = atomic_long_read(&shrinker->nr_deferred[nid]); |
| 422 | |
Dave Hansen | df9024a | 2014-06-04 16:08:07 -0700 | [diff] [blame] | 423 | trace_mm_shrink_slab_end(shrinker, nid, freed, nr, new_nr, total_scan); |
Glauber Costa | 1d3d443 | 2013-08-28 10:18:04 +1000 | [diff] [blame] | 424 | return freed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 425 | } |
| 426 | |
Johannes Weiner | 6b4f779 | 2014-12-12 16:56:13 -0800 | [diff] [blame] | 427 | /** |
Vladimir Davydov | cb731d6 | 2015-02-12 14:58:54 -0800 | [diff] [blame] | 428 | * shrink_slab - shrink slab caches |
Johannes Weiner | 6b4f779 | 2014-12-12 16:56:13 -0800 | [diff] [blame] | 429 | * @gfp_mask: allocation context |
| 430 | * @nid: node whose slab caches to target |
Vladimir Davydov | cb731d6 | 2015-02-12 14:58:54 -0800 | [diff] [blame] | 431 | * @memcg: memory cgroup whose slab caches to target |
Johannes Weiner | 6b4f779 | 2014-12-12 16:56:13 -0800 | [diff] [blame] | 432 | * @nr_scanned: pressure numerator |
| 433 | * @nr_eligible: pressure denominator |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 434 | * |
Johannes Weiner | 6b4f779 | 2014-12-12 16:56:13 -0800 | [diff] [blame] | 435 | * Call the shrink functions to age shrinkable caches. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 436 | * |
Johannes Weiner | 6b4f779 | 2014-12-12 16:56:13 -0800 | [diff] [blame] | 437 | * @nid is passed along to shrinkers with SHRINKER_NUMA_AWARE set, |
| 438 | * unaware shrinkers will receive a node id of 0 instead. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 439 | * |
Vladimir Davydov | cb731d6 | 2015-02-12 14:58:54 -0800 | [diff] [blame] | 440 | * @memcg specifies the memory cgroup to target. If it is not NULL, |
| 441 | * only shrinkers with SHRINKER_MEMCG_AWARE set will be called to scan |
Vladimir Davydov | 0fc9f58 | 2016-03-17 14:18:30 -0700 | [diff] [blame] | 442 | * objects from the memory cgroup specified. Otherwise, only unaware |
| 443 | * shrinkers are called. |
Vladimir Davydov | cb731d6 | 2015-02-12 14:58:54 -0800 | [diff] [blame] | 444 | * |
Johannes Weiner | 6b4f779 | 2014-12-12 16:56:13 -0800 | [diff] [blame] | 445 | * @nr_scanned and @nr_eligible form a ratio that indicate how much of |
| 446 | * the available objects should be scanned. Page reclaim for example |
| 447 | * passes the number of pages scanned and the number of pages on the |
| 448 | * LRU lists that it considered on @nid, plus a bias in @nr_scanned |
| 449 | * when it encountered mapped pages. The ratio is further biased by |
| 450 | * the ->seeks setting of the shrink function, which indicates the |
| 451 | * cost to recreate an object relative to that of an LRU page. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 452 | * |
Johannes Weiner | 6b4f779 | 2014-12-12 16:56:13 -0800 | [diff] [blame] | 453 | * Returns the number of reclaimed slab objects. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 454 | */ |
Vladimir Davydov | cb731d6 | 2015-02-12 14:58:54 -0800 | [diff] [blame] | 455 | static unsigned long shrink_slab(gfp_t gfp_mask, int nid, |
| 456 | struct mem_cgroup *memcg, |
| 457 | unsigned long nr_scanned, |
| 458 | unsigned long nr_eligible) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 459 | { |
| 460 | struct shrinker *shrinker; |
Dave Chinner | 24f7c6b | 2013-08-28 10:17:56 +1000 | [diff] [blame] | 461 | unsigned long freed = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 462 | |
Vladimir Davydov | 0fc9f58 | 2016-03-17 14:18:30 -0700 | [diff] [blame] | 463 | if (memcg && (!memcg_kmem_enabled() || !mem_cgroup_online(memcg))) |
Vladimir Davydov | cb731d6 | 2015-02-12 14:58:54 -0800 | [diff] [blame] | 464 | return 0; |
| 465 | |
Johannes Weiner | 6b4f779 | 2014-12-12 16:56:13 -0800 | [diff] [blame] | 466 | if (nr_scanned == 0) |
| 467 | nr_scanned = SWAP_CLUSTER_MAX; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 468 | |
Minchan Kim | f06590b | 2011-05-24 17:11:11 -0700 | [diff] [blame] | 469 | if (!down_read_trylock(&shrinker_rwsem)) { |
Dave Chinner | 24f7c6b | 2013-08-28 10:17:56 +1000 | [diff] [blame] | 470 | /* |
| 471 | * If we would return 0, our callers would understand that we |
| 472 | * have nothing else to shrink and give up trying. By returning |
| 473 | * 1 we keep it going and assume we'll be able to shrink next |
| 474 | * time. |
| 475 | */ |
| 476 | freed = 1; |
Minchan Kim | f06590b | 2011-05-24 17:11:11 -0700 | [diff] [blame] | 477 | goto out; |
| 478 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 479 | |
| 480 | list_for_each_entry(shrinker, &shrinker_list, list) { |
Johannes Weiner | 6b4f779 | 2014-12-12 16:56:13 -0800 | [diff] [blame] | 481 | struct shrink_control sc = { |
| 482 | .gfp_mask = gfp_mask, |
| 483 | .nid = nid, |
Vladimir Davydov | cb731d6 | 2015-02-12 14:58:54 -0800 | [diff] [blame] | 484 | .memcg = memcg, |
Johannes Weiner | 6b4f779 | 2014-12-12 16:56:13 -0800 | [diff] [blame] | 485 | }; |
Vladimir Davydov | ec97097 | 2014-01-23 15:53:23 -0800 | [diff] [blame] | 486 | |
Vladimir Davydov | 0fc9f58 | 2016-03-17 14:18:30 -0700 | [diff] [blame] | 487 | /* |
| 488 | * If kernel memory accounting is disabled, we ignore |
| 489 | * SHRINKER_MEMCG_AWARE flag and call all shrinkers |
| 490 | * passing NULL for memcg. |
| 491 | */ |
| 492 | if (memcg_kmem_enabled() && |
| 493 | !!memcg != !!(shrinker->flags & SHRINKER_MEMCG_AWARE)) |
Vladimir Davydov | cb731d6 | 2015-02-12 14:58:54 -0800 | [diff] [blame] | 494 | continue; |
| 495 | |
Johannes Weiner | 6b4f779 | 2014-12-12 16:56:13 -0800 | [diff] [blame] | 496 | if (!(shrinker->flags & SHRINKER_NUMA_AWARE)) |
| 497 | sc.nid = 0; |
Dave Chinner | acf92b4 | 2011-07-08 14:14:35 +1000 | [diff] [blame] | 498 | |
Vladimir Davydov | cb731d6 | 2015-02-12 14:58:54 -0800 | [diff] [blame] | 499 | freed += do_shrink_slab(&sc, shrinker, nr_scanned, nr_eligible); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 500 | } |
Johannes Weiner | 6b4f779 | 2014-12-12 16:56:13 -0800 | [diff] [blame] | 501 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 502 | up_read(&shrinker_rwsem); |
Minchan Kim | f06590b | 2011-05-24 17:11:11 -0700 | [diff] [blame] | 503 | out: |
| 504 | cond_resched(); |
Dave Chinner | 24f7c6b | 2013-08-28 10:17:56 +1000 | [diff] [blame] | 505 | return freed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 506 | } |
| 507 | |
Vladimir Davydov | cb731d6 | 2015-02-12 14:58:54 -0800 | [diff] [blame] | 508 | void drop_slab_node(int nid) |
| 509 | { |
| 510 | unsigned long freed; |
| 511 | |
| 512 | do { |
| 513 | struct mem_cgroup *memcg = NULL; |
| 514 | |
| 515 | freed = 0; |
| 516 | do { |
| 517 | freed += shrink_slab(GFP_KERNEL, nid, memcg, |
| 518 | 1000, 1000); |
| 519 | } while ((memcg = mem_cgroup_iter(NULL, memcg, NULL)) != NULL); |
| 520 | } while (freed > 10); |
| 521 | } |
| 522 | |
| 523 | void drop_slab(void) |
| 524 | { |
| 525 | int nid; |
| 526 | |
| 527 | for_each_online_node(nid) |
| 528 | drop_slab_node(nid); |
| 529 | } |
| 530 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 531 | static inline int is_page_cache_freeable(struct page *page) |
| 532 | { |
Johannes Weiner | ceddc3a | 2009-09-21 17:03:00 -0700 | [diff] [blame] | 533 | /* |
| 534 | * A freeable page cache page is referenced only by the caller |
| 535 | * that isolated the page, the page cache radix tree and |
| 536 | * optional buffer heads at page->private. |
| 537 | */ |
Johannes Weiner | edcf474 | 2009-09-21 17:02:59 -0700 | [diff] [blame] | 538 | return page_count(page) - page_has_private(page) == 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 539 | } |
| 540 | |
Tejun Heo | 703c270 | 2015-05-22 17:13:44 -0400 | [diff] [blame] | 541 | static int may_write_to_inode(struct inode *inode, struct scan_control *sc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 542 | { |
Christoph Lameter | 930d915 | 2006-01-08 01:00:47 -0800 | [diff] [blame] | 543 | if (current->flags & PF_SWAPWRITE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 544 | return 1; |
Tejun Heo | 703c270 | 2015-05-22 17:13:44 -0400 | [diff] [blame] | 545 | if (!inode_write_congested(inode)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 546 | return 1; |
Tejun Heo | 703c270 | 2015-05-22 17:13:44 -0400 | [diff] [blame] | 547 | if (inode_to_bdi(inode) == current->backing_dev_info) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 548 | return 1; |
| 549 | return 0; |
| 550 | } |
| 551 | |
| 552 | /* |
| 553 | * We detected a synchronous write error writing a page out. Probably |
| 554 | * -ENOSPC. We need to propagate that into the address_space for a subsequent |
| 555 | * fsync(), msync() or close(). |
| 556 | * |
| 557 | * The tricky part is that after writepage we cannot touch the mapping: nothing |
| 558 | * prevents it from being freed up. But we have a ref on the page and once |
| 559 | * that page is locked, the mapping is pinned. |
| 560 | * |
| 561 | * We're allowed to run sleeping lock_page() here because we know the caller has |
| 562 | * __GFP_FS. |
| 563 | */ |
| 564 | static void handle_write_error(struct address_space *mapping, |
| 565 | struct page *page, int error) |
| 566 | { |
Jens Axboe | 7eaceac | 2011-03-10 08:52:07 +0100 | [diff] [blame] | 567 | lock_page(page); |
Guillaume Chazarain | 3e9f45b | 2007-05-08 00:23:25 -0700 | [diff] [blame] | 568 | if (page_mapping(page) == mapping) |
| 569 | mapping_set_error(mapping, error); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 570 | unlock_page(page); |
| 571 | } |
| 572 | |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 573 | /* possible outcome of pageout() */ |
| 574 | typedef enum { |
| 575 | /* failed to write page out, page is locked */ |
| 576 | PAGE_KEEP, |
| 577 | /* move page to the active list, page is locked */ |
| 578 | PAGE_ACTIVATE, |
| 579 | /* page has been sent to the disk successfully, page is unlocked */ |
| 580 | PAGE_SUCCESS, |
| 581 | /* page is clean and locked */ |
| 582 | PAGE_CLEAN, |
| 583 | } pageout_t; |
| 584 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 585 | /* |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 586 | * pageout is called by shrink_page_list() for each dirty page. |
| 587 | * Calls ->writepage(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 588 | */ |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 589 | static pageout_t pageout(struct page *page, struct address_space *mapping, |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 590 | struct scan_control *sc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 591 | { |
| 592 | /* |
| 593 | * If the page is dirty, only perform writeback if that write |
| 594 | * will be non-blocking. To prevent this allocation from being |
| 595 | * stalled by pagecache activity. But note that there may be |
| 596 | * stalls if we need to run get_block(). We could test |
| 597 | * PagePrivate for that. |
| 598 | * |
Al Viro | 8174202 | 2014-04-03 03:17:43 -0400 | [diff] [blame] | 599 | * If this process is currently in __generic_file_write_iter() against |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 600 | * this page's queue, we can perform writeback even if that |
| 601 | * will block. |
| 602 | * |
| 603 | * If the page is swapcache, write it back even if that would |
| 604 | * block, for some throttling. This happens by accident, because |
| 605 | * swap_backing_dev_info is bust: it doesn't reflect the |
| 606 | * congestion state of the swapdevs. Easy to fix, if needed. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 607 | */ |
| 608 | if (!is_page_cache_freeable(page)) |
| 609 | return PAGE_KEEP; |
| 610 | if (!mapping) { |
| 611 | /* |
| 612 | * Some data journaling orphaned pages can have |
| 613 | * page->mapping == NULL while being dirty with clean buffers. |
| 614 | */ |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 615 | if (page_has_private(page)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 616 | if (try_to_free_buffers(page)) { |
| 617 | ClearPageDirty(page); |
Mitchel Humpherys | b1de0d1 | 2014-06-06 14:38:30 -0700 | [diff] [blame] | 618 | pr_info("%s: orphaned page\n", __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 619 | return PAGE_CLEAN; |
| 620 | } |
| 621 | } |
| 622 | return PAGE_KEEP; |
| 623 | } |
| 624 | if (mapping->a_ops->writepage == NULL) |
| 625 | return PAGE_ACTIVATE; |
Tejun Heo | 703c270 | 2015-05-22 17:13:44 -0400 | [diff] [blame] | 626 | if (!may_write_to_inode(mapping->host, sc)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 627 | return PAGE_KEEP; |
| 628 | |
| 629 | if (clear_page_dirty_for_io(page)) { |
| 630 | int res; |
| 631 | struct writeback_control wbc = { |
| 632 | .sync_mode = WB_SYNC_NONE, |
| 633 | .nr_to_write = SWAP_CLUSTER_MAX, |
OGAWA Hirofumi | 111ebb6 | 2006-06-23 02:03:26 -0700 | [diff] [blame] | 634 | .range_start = 0, |
| 635 | .range_end = LLONG_MAX, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 636 | .for_reclaim = 1, |
| 637 | }; |
| 638 | |
| 639 | SetPageReclaim(page); |
| 640 | res = mapping->a_ops->writepage(page, &wbc); |
| 641 | if (res < 0) |
| 642 | handle_write_error(mapping, page, res); |
Zach Brown | 994fc28c | 2005-12-15 14:28:17 -0800 | [diff] [blame] | 643 | if (res == AOP_WRITEPAGE_ACTIVATE) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 644 | ClearPageReclaim(page); |
| 645 | return PAGE_ACTIVATE; |
| 646 | } |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 647 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 648 | if (!PageWriteback(page)) { |
| 649 | /* synchronous write or broken a_ops? */ |
| 650 | ClearPageReclaim(page); |
| 651 | } |
yalin wang | 3aa2385 | 2016-01-14 15:18:30 -0800 | [diff] [blame] | 652 | trace_mm_vmscan_writepage(page); |
Mel Gorman | c4a2563 | 2016-07-28 15:46:23 -0700 | [diff] [blame] | 653 | inc_node_page_state(page, NR_VMSCAN_WRITE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 654 | return PAGE_SUCCESS; |
| 655 | } |
| 656 | |
| 657 | return PAGE_CLEAN; |
| 658 | } |
| 659 | |
Andrew Morton | a649fd9 | 2006-10-17 00:09:36 -0700 | [diff] [blame] | 660 | /* |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 661 | * Same as remove_mapping, but if the page is removed from the mapping, it |
| 662 | * gets returned with a refcount of 0. |
Andrew Morton | a649fd9 | 2006-10-17 00:09:36 -0700 | [diff] [blame] | 663 | */ |
Johannes Weiner | a528910 | 2014-04-03 14:47:51 -0700 | [diff] [blame] | 664 | static int __remove_mapping(struct address_space *mapping, struct page *page, |
| 665 | bool reclaimed) |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 666 | { |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 667 | unsigned long flags; |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 668 | |
Nick Piggin | 28e4d96 | 2006-09-25 23:31:23 -0700 | [diff] [blame] | 669 | BUG_ON(!PageLocked(page)); |
| 670 | BUG_ON(mapping != page_mapping(page)); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 671 | |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 672 | spin_lock_irqsave(&mapping->tree_lock, flags); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 673 | /* |
Nick Piggin | 0fd0e6b | 2006-09-27 01:50:02 -0700 | [diff] [blame] | 674 | * The non racy check for a busy page. |
| 675 | * |
| 676 | * Must be careful with the order of the tests. When someone has |
| 677 | * a ref to the page, it may be possible that they dirty it then |
| 678 | * drop the reference. So if PageDirty is tested before page_count |
| 679 | * here, then the following race may occur: |
| 680 | * |
| 681 | * get_user_pages(&page); |
| 682 | * [user mapping goes away] |
| 683 | * write_to(page); |
| 684 | * !PageDirty(page) [good] |
| 685 | * SetPageDirty(page); |
| 686 | * put_page(page); |
| 687 | * !page_count(page) [good, discard it] |
| 688 | * |
| 689 | * [oops, our write_to data is lost] |
| 690 | * |
| 691 | * Reversing the order of the tests ensures such a situation cannot |
| 692 | * escape unnoticed. The smp_rmb is needed to ensure the page->flags |
Joonsoo Kim | 0139aa7 | 2016-05-19 17:10:49 -0700 | [diff] [blame] | 693 | * load is not satisfied before that of page->_refcount. |
Nick Piggin | 0fd0e6b | 2006-09-27 01:50:02 -0700 | [diff] [blame] | 694 | * |
| 695 | * Note that if SetPageDirty is always performed via set_page_dirty, |
| 696 | * and thus under tree_lock, then this ordering is not required. |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 697 | */ |
Joonsoo Kim | fe896d1 | 2016-03-17 14:19:26 -0700 | [diff] [blame] | 698 | if (!page_ref_freeze(page, 2)) |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 699 | goto cannot_free; |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 700 | /* note: atomic_cmpxchg in page_freeze_refs provides the smp_rmb */ |
| 701 | if (unlikely(PageDirty(page))) { |
Joonsoo Kim | fe896d1 | 2016-03-17 14:19:26 -0700 | [diff] [blame] | 702 | page_ref_unfreeze(page, 2); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 703 | goto cannot_free; |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 704 | } |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 705 | |
| 706 | if (PageSwapCache(page)) { |
| 707 | swp_entry_t swap = { .val = page_private(page) }; |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 708 | mem_cgroup_swapout(page, swap); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 709 | __delete_from_swap_cache(page); |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 710 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 711 | swapcache_free(swap); |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 712 | } else { |
Linus Torvalds | 6072d13 | 2010-12-01 13:35:19 -0500 | [diff] [blame] | 713 | void (*freepage)(struct page *); |
Johannes Weiner | a528910 | 2014-04-03 14:47:51 -0700 | [diff] [blame] | 714 | void *shadow = NULL; |
Linus Torvalds | 6072d13 | 2010-12-01 13:35:19 -0500 | [diff] [blame] | 715 | |
| 716 | freepage = mapping->a_ops->freepage; |
Johannes Weiner | a528910 | 2014-04-03 14:47:51 -0700 | [diff] [blame] | 717 | /* |
| 718 | * Remember a shadow entry for reclaimed file cache in |
| 719 | * order to detect refaults, thus thrashing, later on. |
| 720 | * |
| 721 | * But don't store shadows in an address space that is |
| 722 | * already exiting. This is not just an optizimation, |
| 723 | * inode reclaim needs to empty out the radix tree or |
| 724 | * the nodes are lost. Don't plant shadows behind its |
| 725 | * back. |
Ross Zwisler | f9fe48b | 2016-01-22 15:10:40 -0800 | [diff] [blame] | 726 | * |
| 727 | * We also don't store shadows for DAX mappings because the |
| 728 | * only page cache pages found in these are zero pages |
| 729 | * covering holes, and because we don't want to mix DAX |
| 730 | * exceptional entries and shadow exceptional entries in the |
| 731 | * same page_tree. |
Johannes Weiner | a528910 | 2014-04-03 14:47:51 -0700 | [diff] [blame] | 732 | */ |
| 733 | if (reclaimed && page_is_file_cache(page) && |
Ross Zwisler | f9fe48b | 2016-01-22 15:10:40 -0800 | [diff] [blame] | 734 | !mapping_exiting(mapping) && !dax_mapping(mapping)) |
Johannes Weiner | a528910 | 2014-04-03 14:47:51 -0700 | [diff] [blame] | 735 | shadow = workingset_eviction(mapping, page); |
Johannes Weiner | 62cccb8 | 2016-03-15 14:57:22 -0700 | [diff] [blame] | 736 | __delete_from_page_cache(page, shadow); |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 737 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
Linus Torvalds | 6072d13 | 2010-12-01 13:35:19 -0500 | [diff] [blame] | 738 | |
| 739 | if (freepage != NULL) |
| 740 | freepage(page); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 741 | } |
| 742 | |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 743 | return 1; |
| 744 | |
| 745 | cannot_free: |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 746 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 747 | return 0; |
| 748 | } |
| 749 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 750 | /* |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 751 | * Attempt to detach a locked page from its ->mapping. If it is dirty or if |
| 752 | * someone else has a ref on the page, abort and return 0. If it was |
| 753 | * successfully detached, return 1. Assumes the caller has a single ref on |
| 754 | * this page. |
| 755 | */ |
| 756 | int remove_mapping(struct address_space *mapping, struct page *page) |
| 757 | { |
Johannes Weiner | a528910 | 2014-04-03 14:47:51 -0700 | [diff] [blame] | 758 | if (__remove_mapping(mapping, page, false)) { |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 759 | /* |
| 760 | * Unfreezing the refcount with 1 rather than 2 effectively |
| 761 | * drops the pagecache ref for us without requiring another |
| 762 | * atomic operation. |
| 763 | */ |
Joonsoo Kim | fe896d1 | 2016-03-17 14:19:26 -0700 | [diff] [blame] | 764 | page_ref_unfreeze(page, 1); |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 765 | return 1; |
| 766 | } |
| 767 | return 0; |
| 768 | } |
| 769 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 770 | /** |
| 771 | * putback_lru_page - put previously isolated page onto appropriate LRU list |
| 772 | * @page: page to be put back to appropriate lru list |
| 773 | * |
| 774 | * Add previously isolated @page to appropriate LRU list. |
| 775 | * Page may still be unevictable for other reasons. |
| 776 | * |
| 777 | * lru_lock must not be held, interrupts must be enabled. |
| 778 | */ |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 779 | void putback_lru_page(struct page *page) |
| 780 | { |
Vlastimil Babka | 0ec3b74 | 2013-09-11 14:22:26 -0700 | [diff] [blame] | 781 | bool is_unevictable; |
Lee Schermerhorn | bbfd28e | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 782 | int was_unevictable = PageUnevictable(page); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 783 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 784 | VM_BUG_ON_PAGE(PageLRU(page), page); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 785 | |
| 786 | redo: |
| 787 | ClearPageUnevictable(page); |
| 788 | |
Hugh Dickins | 39b5f29 | 2012-10-08 16:33:18 -0700 | [diff] [blame] | 789 | if (page_evictable(page)) { |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 790 | /* |
| 791 | * For evictable pages, we can use the cache. |
| 792 | * In event of a race, worst case is we end up with an |
| 793 | * unevictable page on [in]active list. |
| 794 | * We know how to handle that. |
| 795 | */ |
Vlastimil Babka | 0ec3b74 | 2013-09-11 14:22:26 -0700 | [diff] [blame] | 796 | is_unevictable = false; |
Mel Gorman | c53954a | 2013-07-03 15:02:34 -0700 | [diff] [blame] | 797 | lru_cache_add(page); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 798 | } else { |
| 799 | /* |
| 800 | * Put unevictable pages directly on zone's unevictable |
| 801 | * list. |
| 802 | */ |
Vlastimil Babka | 0ec3b74 | 2013-09-11 14:22:26 -0700 | [diff] [blame] | 803 | is_unevictable = true; |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 804 | add_page_to_unevictable_list(page); |
Johannes Weiner | 6a7b954 | 2009-10-26 16:50:00 -0700 | [diff] [blame] | 805 | /* |
Minchan Kim | 21ee9f3 | 2011-10-31 17:09:28 -0700 | [diff] [blame] | 806 | * When racing with an mlock or AS_UNEVICTABLE clearing |
| 807 | * (page is unlocked) make sure that if the other thread |
| 808 | * does not observe our setting of PG_lru and fails |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 809 | * isolation/check_move_unevictable_pages, |
Minchan Kim | 21ee9f3 | 2011-10-31 17:09:28 -0700 | [diff] [blame] | 810 | * we see PG_mlocked/AS_UNEVICTABLE cleared below and move |
Johannes Weiner | 6a7b954 | 2009-10-26 16:50:00 -0700 | [diff] [blame] | 811 | * the page back to the evictable list. |
| 812 | * |
Minchan Kim | 21ee9f3 | 2011-10-31 17:09:28 -0700 | [diff] [blame] | 813 | * The other side is TestClearPageMlocked() or shmem_lock(). |
Johannes Weiner | 6a7b954 | 2009-10-26 16:50:00 -0700 | [diff] [blame] | 814 | */ |
| 815 | smp_mb(); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 816 | } |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 817 | |
| 818 | /* |
| 819 | * page's status can change while we move it among lru. If an evictable |
| 820 | * page is on unevictable list, it never be freed. To avoid that, |
| 821 | * check after we added it to the list, again. |
| 822 | */ |
Vlastimil Babka | 0ec3b74 | 2013-09-11 14:22:26 -0700 | [diff] [blame] | 823 | if (is_unevictable && page_evictable(page)) { |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 824 | if (!isolate_lru_page(page)) { |
| 825 | put_page(page); |
| 826 | goto redo; |
| 827 | } |
| 828 | /* This means someone else dropped this page from LRU |
| 829 | * So, it will be freed or putback to LRU again. There is |
| 830 | * nothing to do here. |
| 831 | */ |
| 832 | } |
| 833 | |
Vlastimil Babka | 0ec3b74 | 2013-09-11 14:22:26 -0700 | [diff] [blame] | 834 | if (was_unevictable && !is_unevictable) |
Lee Schermerhorn | bbfd28e | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 835 | count_vm_event(UNEVICTABLE_PGRESCUED); |
Vlastimil Babka | 0ec3b74 | 2013-09-11 14:22:26 -0700 | [diff] [blame] | 836 | else if (!was_unevictable && is_unevictable) |
Lee Schermerhorn | bbfd28e | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 837 | count_vm_event(UNEVICTABLE_PGCULLED); |
| 838 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 839 | put_page(page); /* drop ref from isolate */ |
| 840 | } |
| 841 | |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 842 | enum page_references { |
| 843 | PAGEREF_RECLAIM, |
| 844 | PAGEREF_RECLAIM_CLEAN, |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 845 | PAGEREF_KEEP, |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 846 | PAGEREF_ACTIVATE, |
| 847 | }; |
| 848 | |
| 849 | static enum page_references page_check_references(struct page *page, |
| 850 | struct scan_control *sc) |
| 851 | { |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 852 | int referenced_ptes, referenced_page; |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 853 | unsigned long vm_flags; |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 854 | |
Johannes Weiner | c3ac9a8 | 2012-05-29 15:06:25 -0700 | [diff] [blame] | 855 | referenced_ptes = page_referenced(page, 1, sc->target_mem_cgroup, |
| 856 | &vm_flags); |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 857 | referenced_page = TestClearPageReferenced(page); |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 858 | |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 859 | /* |
| 860 | * Mlock lost the isolation race with us. Let try_to_unmap() |
| 861 | * move the page to the unevictable list. |
| 862 | */ |
| 863 | if (vm_flags & VM_LOCKED) |
| 864 | return PAGEREF_RECLAIM; |
| 865 | |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 866 | if (referenced_ptes) { |
Michal Hocko | e4898273 | 2012-05-29 15:06:45 -0700 | [diff] [blame] | 867 | if (PageSwapBacked(page)) |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 868 | return PAGEREF_ACTIVATE; |
| 869 | /* |
| 870 | * All mapped pages start out with page table |
| 871 | * references from the instantiating fault, so we need |
| 872 | * to look twice if a mapped file page is used more |
| 873 | * than once. |
| 874 | * |
| 875 | * Mark it and spare it for another trip around the |
| 876 | * inactive list. Another page table reference will |
| 877 | * lead to its activation. |
| 878 | * |
| 879 | * Note: the mark is set for activated pages as well |
| 880 | * so that recently deactivated but used pages are |
| 881 | * quickly recovered. |
| 882 | */ |
| 883 | SetPageReferenced(page); |
| 884 | |
Konstantin Khlebnikov | 34dbc67 | 2012-01-10 15:06:59 -0800 | [diff] [blame] | 885 | if (referenced_page || referenced_ptes > 1) |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 886 | return PAGEREF_ACTIVATE; |
| 887 | |
Konstantin Khlebnikov | c909e99 | 2012-01-10 15:07:03 -0800 | [diff] [blame] | 888 | /* |
| 889 | * Activate file-backed executable pages after first usage. |
| 890 | */ |
| 891 | if (vm_flags & VM_EXEC) |
| 892 | return PAGEREF_ACTIVATE; |
| 893 | |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 894 | return PAGEREF_KEEP; |
| 895 | } |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 896 | |
| 897 | /* Reclaim if clean, defer dirty pages to writeback */ |
KOSAKI Motohiro | 2e30244 | 2010-10-26 14:21:46 -0700 | [diff] [blame] | 898 | if (referenced_page && !PageSwapBacked(page)) |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 899 | return PAGEREF_RECLAIM_CLEAN; |
| 900 | |
| 901 | return PAGEREF_RECLAIM; |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 902 | } |
| 903 | |
Mel Gorman | e2be15f | 2013-07-03 15:01:57 -0700 | [diff] [blame] | 904 | /* Check if a page is dirty or under writeback */ |
| 905 | static void page_check_dirty_writeback(struct page *page, |
| 906 | bool *dirty, bool *writeback) |
| 907 | { |
Mel Gorman | b459722 | 2013-07-03 15:02:05 -0700 | [diff] [blame] | 908 | struct address_space *mapping; |
| 909 | |
Mel Gorman | e2be15f | 2013-07-03 15:01:57 -0700 | [diff] [blame] | 910 | /* |
| 911 | * Anonymous pages are not handled by flushers and must be written |
| 912 | * from reclaim context. Do not stall reclaim based on them |
| 913 | */ |
Shaohua Li | 802a3a9 | 2017-05-03 14:52:32 -0700 | [diff] [blame] | 914 | if (!page_is_file_cache(page) || |
| 915 | (PageAnon(page) && !PageSwapBacked(page))) { |
Mel Gorman | e2be15f | 2013-07-03 15:01:57 -0700 | [diff] [blame] | 916 | *dirty = false; |
| 917 | *writeback = false; |
| 918 | return; |
| 919 | } |
| 920 | |
| 921 | /* By default assume that the page flags are accurate */ |
| 922 | *dirty = PageDirty(page); |
| 923 | *writeback = PageWriteback(page); |
Mel Gorman | b459722 | 2013-07-03 15:02:05 -0700 | [diff] [blame] | 924 | |
| 925 | /* Verify dirty/writeback state if the filesystem supports it */ |
| 926 | if (!page_has_private(page)) |
| 927 | return; |
| 928 | |
| 929 | mapping = page_mapping(page); |
| 930 | if (mapping && mapping->a_ops->is_dirty_writeback) |
| 931 | mapping->a_ops->is_dirty_writeback(page, dirty, writeback); |
Mel Gorman | e2be15f | 2013-07-03 15:01:57 -0700 | [diff] [blame] | 932 | } |
| 933 | |
Michal Hocko | 3c710c1 | 2017-02-22 15:44:27 -0800 | [diff] [blame] | 934 | struct reclaim_stat { |
| 935 | unsigned nr_dirty; |
| 936 | unsigned nr_unqueued_dirty; |
| 937 | unsigned nr_congested; |
| 938 | unsigned nr_writeback; |
| 939 | unsigned nr_immediate; |
Michal Hocko | 5bccd16 | 2017-02-22 15:44:30 -0800 | [diff] [blame] | 940 | unsigned nr_activate; |
| 941 | unsigned nr_ref_keep; |
| 942 | unsigned nr_unmap_fail; |
Michal Hocko | 3c710c1 | 2017-02-22 15:44:27 -0800 | [diff] [blame] | 943 | }; |
| 944 | |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 945 | /* |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 946 | * shrink_page_list() returns the number of reclaimed pages |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 947 | */ |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 948 | static unsigned long shrink_page_list(struct list_head *page_list, |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 949 | struct pglist_data *pgdat, |
Mel Gorman | f84f6e2 | 2011-10-31 17:07:51 -0700 | [diff] [blame] | 950 | struct scan_control *sc, |
Minchan Kim | 02c6de8 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 951 | enum ttu_flags ttu_flags, |
Michal Hocko | 3c710c1 | 2017-02-22 15:44:27 -0800 | [diff] [blame] | 952 | struct reclaim_stat *stat, |
Minchan Kim | 02c6de8 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 953 | bool force_reclaim) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 954 | { |
| 955 | LIST_HEAD(ret_pages); |
Mel Gorman | abe4c3b | 2010-08-09 17:19:31 -0700 | [diff] [blame] | 956 | LIST_HEAD(free_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 957 | int pgactivate = 0; |
Michal Hocko | 3c710c1 | 2017-02-22 15:44:27 -0800 | [diff] [blame] | 958 | unsigned nr_unqueued_dirty = 0; |
| 959 | unsigned nr_dirty = 0; |
| 960 | unsigned nr_congested = 0; |
| 961 | unsigned nr_reclaimed = 0; |
| 962 | unsigned nr_writeback = 0; |
| 963 | unsigned nr_immediate = 0; |
Michal Hocko | 5bccd16 | 2017-02-22 15:44:30 -0800 | [diff] [blame] | 964 | unsigned nr_ref_keep = 0; |
| 965 | unsigned nr_unmap_fail = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 966 | |
| 967 | cond_resched(); |
| 968 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 969 | while (!list_empty(page_list)) { |
| 970 | struct address_space *mapping; |
| 971 | struct page *page; |
| 972 | int may_enter_fs; |
Minchan Kim | 02c6de8 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 973 | enum page_references references = PAGEREF_RECLAIM_CLEAN; |
Mel Gorman | e2be15f | 2013-07-03 15:01:57 -0700 | [diff] [blame] | 974 | bool dirty, writeback; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 975 | |
| 976 | cond_resched(); |
| 977 | |
| 978 | page = lru_to_page(page_list); |
| 979 | list_del(&page->lru); |
| 980 | |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 981 | if (!trylock_page(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 982 | goto keep; |
| 983 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 984 | VM_BUG_ON_PAGE(PageActive(page), page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 985 | |
| 986 | sc->nr_scanned++; |
Christoph Lameter | 80e4342 | 2006-02-11 17:55:53 -0800 | [diff] [blame] | 987 | |
Hugh Dickins | 39b5f29 | 2012-10-08 16:33:18 -0700 | [diff] [blame] | 988 | if (unlikely(!page_evictable(page))) |
Minchan Kim | ad6b670 | 2017-05-03 14:54:13 -0700 | [diff] [blame] | 989 | goto activate_locked; |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 990 | |
Johannes Weiner | a6dc60f8 | 2009-03-31 15:19:30 -0700 | [diff] [blame] | 991 | if (!sc->may_unmap && page_mapped(page)) |
Christoph Lameter | 80e4342 | 2006-02-11 17:55:53 -0800 | [diff] [blame] | 992 | goto keep_locked; |
| 993 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 994 | /* Double the slab pressure for mapped and swapcache pages */ |
Shaohua Li | 802a3a9 | 2017-05-03 14:52:32 -0700 | [diff] [blame] | 995 | if ((page_mapped(page) || PageSwapCache(page)) && |
| 996 | !(PageAnon(page) && !PageSwapBacked(page))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 997 | sc->nr_scanned++; |
| 998 | |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 999 | may_enter_fs = (sc->gfp_mask & __GFP_FS) || |
| 1000 | (PageSwapCache(page) && (sc->gfp_mask & __GFP_IO)); |
| 1001 | |
Mel Gorman | 283aba9 | 2013-07-03 15:01:51 -0700 | [diff] [blame] | 1002 | /* |
Mel Gorman | e2be15f | 2013-07-03 15:01:57 -0700 | [diff] [blame] | 1003 | * The number of dirty pages determines if a zone is marked |
| 1004 | * reclaim_congested which affects wait_iff_congested. kswapd |
| 1005 | * will stall and start writing pages if the tail of the LRU |
| 1006 | * is all dirty unqueued pages. |
| 1007 | */ |
| 1008 | page_check_dirty_writeback(page, &dirty, &writeback); |
| 1009 | if (dirty || writeback) |
| 1010 | nr_dirty++; |
| 1011 | |
| 1012 | if (dirty && !writeback) |
| 1013 | nr_unqueued_dirty++; |
| 1014 | |
Mel Gorman | d04e8ac | 2013-07-03 15:02:03 -0700 | [diff] [blame] | 1015 | /* |
| 1016 | * Treat this page as congested if the underlying BDI is or if |
| 1017 | * pages are cycling through the LRU so quickly that the |
| 1018 | * pages marked for immediate reclaim are making it to the |
| 1019 | * end of the LRU a second time. |
| 1020 | */ |
Mel Gorman | e2be15f | 2013-07-03 15:01:57 -0700 | [diff] [blame] | 1021 | mapping = page_mapping(page); |
Jamie Liu | 1da58ee | 2014-12-10 15:43:20 -0800 | [diff] [blame] | 1022 | if (((dirty || writeback) && mapping && |
Tejun Heo | 703c270 | 2015-05-22 17:13:44 -0400 | [diff] [blame] | 1023 | inode_write_congested(mapping->host)) || |
Mel Gorman | d04e8ac | 2013-07-03 15:02:03 -0700 | [diff] [blame] | 1024 | (writeback && PageReclaim(page))) |
Mel Gorman | e2be15f | 2013-07-03 15:01:57 -0700 | [diff] [blame] | 1025 | nr_congested++; |
| 1026 | |
| 1027 | /* |
Mel Gorman | 283aba9 | 2013-07-03 15:01:51 -0700 | [diff] [blame] | 1028 | * If a page at the tail of the LRU is under writeback, there |
| 1029 | * are three cases to consider. |
| 1030 | * |
| 1031 | * 1) If reclaim is encountering an excessive number of pages |
| 1032 | * under writeback and this page is both under writeback and |
| 1033 | * PageReclaim then it indicates that pages are being queued |
| 1034 | * for IO but are being recycled through the LRU before the |
| 1035 | * IO can complete. Waiting on the page itself risks an |
| 1036 | * indefinite stall if it is impossible to writeback the |
| 1037 | * page due to IO error or disconnected storage so instead |
Mel Gorman | b1a6f21 | 2013-07-03 15:01:58 -0700 | [diff] [blame] | 1038 | * note that the LRU is being scanned too quickly and the |
| 1039 | * caller can stall after page list has been processed. |
Mel Gorman | 283aba9 | 2013-07-03 15:01:51 -0700 | [diff] [blame] | 1040 | * |
Tejun Heo | 97c9341 | 2015-05-22 18:23:36 -0400 | [diff] [blame] | 1041 | * 2) Global or new memcg reclaim encounters a page that is |
Michal Hocko | ecf5fc6 | 2015-08-04 14:36:58 -0700 | [diff] [blame] | 1042 | * not marked for immediate reclaim, or the caller does not |
| 1043 | * have __GFP_FS (or __GFP_IO if it's simply going to swap, |
| 1044 | * not to fs). In this case mark the page for immediate |
Tejun Heo | 97c9341 | 2015-05-22 18:23:36 -0400 | [diff] [blame] | 1045 | * reclaim and continue scanning. |
Mel Gorman | 283aba9 | 2013-07-03 15:01:51 -0700 | [diff] [blame] | 1046 | * |
Michal Hocko | ecf5fc6 | 2015-08-04 14:36:58 -0700 | [diff] [blame] | 1047 | * Require may_enter_fs because we would wait on fs, which |
| 1048 | * may not have submitted IO yet. And the loop driver might |
Mel Gorman | 283aba9 | 2013-07-03 15:01:51 -0700 | [diff] [blame] | 1049 | * enter reclaim, and deadlock if it waits on a page for |
| 1050 | * which it is needed to do the write (loop masks off |
| 1051 | * __GFP_IO|__GFP_FS for this reason); but more thought |
| 1052 | * would probably show more reasons. |
| 1053 | * |
Hugh Dickins | 7fadc82 | 2015-09-08 15:03:46 -0700 | [diff] [blame] | 1054 | * 3) Legacy memcg encounters a page that is already marked |
Mel Gorman | 283aba9 | 2013-07-03 15:01:51 -0700 | [diff] [blame] | 1055 | * PageReclaim. memcg does not have any dirty pages |
| 1056 | * throttling so we could easily OOM just because too many |
| 1057 | * pages are in writeback and there is nothing else to |
| 1058 | * reclaim. Wait for the writeback to complete. |
Johannes Weiner | c55e8d0 | 2017-02-24 14:56:23 -0800 | [diff] [blame] | 1059 | * |
| 1060 | * In cases 1) and 2) we activate the pages to get them out of |
| 1061 | * the way while we continue scanning for clean pages on the |
| 1062 | * inactive list and refilling from the active list. The |
| 1063 | * observation here is that waiting for disk writes is more |
| 1064 | * expensive than potentially causing reloads down the line. |
| 1065 | * Since they're marked for immediate reclaim, they won't put |
| 1066 | * memory pressure on the cache working set any longer than it |
| 1067 | * takes to write them to disk. |
Mel Gorman | 283aba9 | 2013-07-03 15:01:51 -0700 | [diff] [blame] | 1068 | */ |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 1069 | if (PageWriteback(page)) { |
Mel Gorman | 283aba9 | 2013-07-03 15:01:51 -0700 | [diff] [blame] | 1070 | /* Case 1 above */ |
| 1071 | if (current_is_kswapd() && |
| 1072 | PageReclaim(page) && |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1073 | test_bit(PGDAT_WRITEBACK, &pgdat->flags)) { |
Mel Gorman | b1a6f21 | 2013-07-03 15:01:58 -0700 | [diff] [blame] | 1074 | nr_immediate++; |
Johannes Weiner | c55e8d0 | 2017-02-24 14:56:23 -0800 | [diff] [blame] | 1075 | goto activate_locked; |
Mel Gorman | 283aba9 | 2013-07-03 15:01:51 -0700 | [diff] [blame] | 1076 | |
| 1077 | /* Case 2 above */ |
Tejun Heo | 97c9341 | 2015-05-22 18:23:36 -0400 | [diff] [blame] | 1078 | } else if (sane_reclaim(sc) || |
Michal Hocko | ecf5fc6 | 2015-08-04 14:36:58 -0700 | [diff] [blame] | 1079 | !PageReclaim(page) || !may_enter_fs) { |
Hugh Dickins | c3b94f4 | 2012-07-31 16:45:59 -0700 | [diff] [blame] | 1080 | /* |
| 1081 | * This is slightly racy - end_page_writeback() |
| 1082 | * might have just cleared PageReclaim, then |
| 1083 | * setting PageReclaim here end up interpreted |
| 1084 | * as PageReadahead - but that does not matter |
| 1085 | * enough to care. What we do want is for this |
| 1086 | * page to have PageReclaim set next time memcg |
| 1087 | * reclaim reaches the tests above, so it will |
| 1088 | * then wait_on_page_writeback() to avoid OOM; |
| 1089 | * and it's also appropriate in global reclaim. |
| 1090 | */ |
| 1091 | SetPageReclaim(page); |
Michal Hocko | e62e384 | 2012-07-31 16:45:55 -0700 | [diff] [blame] | 1092 | nr_writeback++; |
Johannes Weiner | c55e8d0 | 2017-02-24 14:56:23 -0800 | [diff] [blame] | 1093 | goto activate_locked; |
Mel Gorman | 283aba9 | 2013-07-03 15:01:51 -0700 | [diff] [blame] | 1094 | |
| 1095 | /* Case 3 above */ |
| 1096 | } else { |
Hugh Dickins | 7fadc82 | 2015-09-08 15:03:46 -0700 | [diff] [blame] | 1097 | unlock_page(page); |
Mel Gorman | 283aba9 | 2013-07-03 15:01:51 -0700 | [diff] [blame] | 1098 | wait_on_page_writeback(page); |
Hugh Dickins | 7fadc82 | 2015-09-08 15:03:46 -0700 | [diff] [blame] | 1099 | /* then go back and try same page again */ |
| 1100 | list_add_tail(&page->lru, page_list); |
| 1101 | continue; |
Michal Hocko | e62e384 | 2012-07-31 16:45:55 -0700 | [diff] [blame] | 1102 | } |
Andy Whitcroft | c661b07 | 2007-08-22 14:01:26 -0700 | [diff] [blame] | 1103 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1104 | |
Minchan Kim | 02c6de8 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 1105 | if (!force_reclaim) |
| 1106 | references = page_check_references(page, sc); |
| 1107 | |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 1108 | switch (references) { |
| 1109 | case PAGEREF_ACTIVATE: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1110 | goto activate_locked; |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 1111 | case PAGEREF_KEEP: |
Michal Hocko | 5bccd16 | 2017-02-22 15:44:30 -0800 | [diff] [blame] | 1112 | nr_ref_keep++; |
Johannes Weiner | 64574746 | 2010-03-05 13:42:22 -0800 | [diff] [blame] | 1113 | goto keep_locked; |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 1114 | case PAGEREF_RECLAIM: |
| 1115 | case PAGEREF_RECLAIM_CLEAN: |
| 1116 | ; /* try to reclaim the page below */ |
| 1117 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1118 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1119 | /* |
| 1120 | * Anonymous process memory has backing store? |
| 1121 | * Try to allocate it some swap space here. |
Shaohua Li | 802a3a9 | 2017-05-03 14:52:32 -0700 | [diff] [blame] | 1122 | * Lazyfree page could be freed directly |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1123 | */ |
Shaohua Li | 802a3a9 | 2017-05-03 14:52:32 -0700 | [diff] [blame] | 1124 | if (PageAnon(page) && PageSwapBacked(page) && |
| 1125 | !PageSwapCache(page)) { |
Hugh Dickins | 63eb6b9 | 2008-11-19 15:36:37 -0800 | [diff] [blame] | 1126 | if (!(sc->gfp_mask & __GFP_IO)) |
| 1127 | goto keep_locked; |
Shaohua Li | 5bc7b8a | 2013-04-29 15:08:36 -0700 | [diff] [blame] | 1128 | if (!add_to_swap(page, page_list)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1129 | goto activate_locked; |
Hugh Dickins | 63eb6b9 | 2008-11-19 15:36:37 -0800 | [diff] [blame] | 1130 | may_enter_fs = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1131 | |
Mel Gorman | e2be15f | 2013-07-03 15:01:57 -0700 | [diff] [blame] | 1132 | /* Adding to swap updated mapping */ |
| 1133 | mapping = page_mapping(page); |
Kirill A. Shutemov | 7751b2d | 2016-07-26 15:25:56 -0700 | [diff] [blame] | 1134 | } else if (unlikely(PageTransHuge(page))) { |
| 1135 | /* Split file THP */ |
| 1136 | if (split_huge_page_to_list(page, page_list)) |
| 1137 | goto keep_locked; |
Mel Gorman | e2be15f | 2013-07-03 15:01:57 -0700 | [diff] [blame] | 1138 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1139 | |
Kirill A. Shutemov | 7751b2d | 2016-07-26 15:25:56 -0700 | [diff] [blame] | 1140 | VM_BUG_ON_PAGE(PageTransHuge(page), page); |
| 1141 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1142 | /* |
| 1143 | * The page is mapped into the page tables of one or more |
| 1144 | * processes. Try to unmap it here. |
| 1145 | */ |
Shaohua Li | 802a3a9 | 2017-05-03 14:52:32 -0700 | [diff] [blame] | 1146 | if (page_mapped(page)) { |
Minchan Kim | 666e5a4 | 2017-05-03 14:54:20 -0700 | [diff] [blame] | 1147 | if (!try_to_unmap(page, ttu_flags | TTU_BATCH_FLUSH)) { |
Michal Hocko | 5bccd16 | 2017-02-22 15:44:30 -0800 | [diff] [blame] | 1148 | nr_unmap_fail++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1149 | goto activate_locked; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1150 | } |
| 1151 | } |
| 1152 | |
| 1153 | if (PageDirty(page)) { |
Mel Gorman | ee72886 | 2011-10-31 17:07:38 -0700 | [diff] [blame] | 1154 | /* |
Johannes Weiner | 4eda482 | 2017-02-24 14:56:20 -0800 | [diff] [blame] | 1155 | * Only kswapd can writeback filesystem pages |
| 1156 | * to avoid risk of stack overflow. But avoid |
| 1157 | * injecting inefficient single-page IO into |
| 1158 | * flusher writeback as much as possible: only |
| 1159 | * write pages when we've encountered many |
| 1160 | * dirty pages, and when we've already scanned |
| 1161 | * the rest of the LRU for clean pages and see |
| 1162 | * the same dirty pages again (PageReclaim). |
Mel Gorman | ee72886 | 2011-10-31 17:07:38 -0700 | [diff] [blame] | 1163 | */ |
Mel Gorman | f84f6e2 | 2011-10-31 17:07:51 -0700 | [diff] [blame] | 1164 | if (page_is_file_cache(page) && |
Johannes Weiner | 4eda482 | 2017-02-24 14:56:20 -0800 | [diff] [blame] | 1165 | (!current_is_kswapd() || !PageReclaim(page) || |
| 1166 | !test_bit(PGDAT_DIRTY, &pgdat->flags))) { |
Mel Gorman | 49ea7eb | 2011-10-31 17:07:59 -0700 | [diff] [blame] | 1167 | /* |
| 1168 | * Immediately reclaim when written back. |
| 1169 | * Similar in principal to deactivate_page() |
| 1170 | * except we already have the page isolated |
| 1171 | * and know it's dirty |
| 1172 | */ |
Mel Gorman | c4a2563 | 2016-07-28 15:46:23 -0700 | [diff] [blame] | 1173 | inc_node_page_state(page, NR_VMSCAN_IMMEDIATE); |
Mel Gorman | 49ea7eb | 2011-10-31 17:07:59 -0700 | [diff] [blame] | 1174 | SetPageReclaim(page); |
| 1175 | |
Johannes Weiner | c55e8d0 | 2017-02-24 14:56:23 -0800 | [diff] [blame] | 1176 | goto activate_locked; |
Mel Gorman | ee72886 | 2011-10-31 17:07:38 -0700 | [diff] [blame] | 1177 | } |
| 1178 | |
Johannes Weiner | dfc8d63 | 2010-03-05 13:42:19 -0800 | [diff] [blame] | 1179 | if (references == PAGEREF_RECLAIM_CLEAN) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1180 | goto keep_locked; |
Andrew Morton | 4dd4b92 | 2008-03-24 12:29:52 -0700 | [diff] [blame] | 1181 | if (!may_enter_fs) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1182 | goto keep_locked; |
Christoph Lameter | 52a8363 | 2006-02-01 03:05:28 -0800 | [diff] [blame] | 1183 | if (!sc->may_writepage) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1184 | goto keep_locked; |
| 1185 | |
Mel Gorman | d950c94 | 2015-09-04 15:47:35 -0700 | [diff] [blame] | 1186 | /* |
| 1187 | * Page is dirty. Flush the TLB if a writable entry |
| 1188 | * potentially exists to avoid CPU writes after IO |
| 1189 | * starts and then write it out here. |
| 1190 | */ |
| 1191 | try_to_unmap_flush_dirty(); |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 1192 | switch (pageout(page, mapping, sc)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1193 | case PAGE_KEEP: |
| 1194 | goto keep_locked; |
| 1195 | case PAGE_ACTIVATE: |
| 1196 | goto activate_locked; |
| 1197 | case PAGE_SUCCESS: |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 1198 | if (PageWriteback(page)) |
Mel Gorman | 41ac199 | 2012-05-29 15:06:19 -0700 | [diff] [blame] | 1199 | goto keep; |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 1200 | if (PageDirty(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1201 | goto keep; |
KOSAKI Motohiro | 7d3579e | 2010-10-26 14:21:42 -0700 | [diff] [blame] | 1202 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1203 | /* |
| 1204 | * A synchronous write - probably a ramdisk. Go |
| 1205 | * ahead and try to reclaim the page. |
| 1206 | */ |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 1207 | if (!trylock_page(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1208 | goto keep; |
| 1209 | if (PageDirty(page) || PageWriteback(page)) |
| 1210 | goto keep_locked; |
| 1211 | mapping = page_mapping(page); |
| 1212 | case PAGE_CLEAN: |
| 1213 | ; /* try to free the page below */ |
| 1214 | } |
| 1215 | } |
| 1216 | |
| 1217 | /* |
| 1218 | * If the page has buffers, try to free the buffer mappings |
| 1219 | * associated with this page. If we succeed we try to free |
| 1220 | * the page as well. |
| 1221 | * |
| 1222 | * We do this even if the page is PageDirty(). |
| 1223 | * try_to_release_page() does not perform I/O, but it is |
| 1224 | * possible for a page to have PageDirty set, but it is actually |
| 1225 | * clean (all its buffers are clean). This happens if the |
| 1226 | * buffers were written out directly, with submit_bh(). ext3 |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1227 | * will do this, as well as the blockdev mapping. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1228 | * try_to_release_page() will discover that cleanness and will |
| 1229 | * drop the buffers and mark the page clean - it can be freed. |
| 1230 | * |
| 1231 | * Rarely, pages can have buffers and no ->mapping. These are |
| 1232 | * the pages which were not successfully invalidated in |
| 1233 | * truncate_complete_page(). We try to drop those buffers here |
| 1234 | * and if that worked, and the page is no longer mapped into |
| 1235 | * process address space (page_count == 1) it can be freed. |
| 1236 | * Otherwise, leave the page on the LRU so it is swappable. |
| 1237 | */ |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 1238 | if (page_has_private(page)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1239 | if (!try_to_release_page(page, sc->gfp_mask)) |
| 1240 | goto activate_locked; |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 1241 | if (!mapping && page_count(page) == 1) { |
| 1242 | unlock_page(page); |
| 1243 | if (put_page_testzero(page)) |
| 1244 | goto free_it; |
| 1245 | else { |
| 1246 | /* |
| 1247 | * rare race with speculative reference. |
| 1248 | * the speculative reference will free |
| 1249 | * this page shortly, so we may |
| 1250 | * increment nr_reclaimed here (and |
| 1251 | * leave it off the LRU). |
| 1252 | */ |
| 1253 | nr_reclaimed++; |
| 1254 | continue; |
| 1255 | } |
| 1256 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1257 | } |
| 1258 | |
Shaohua Li | 802a3a9 | 2017-05-03 14:52:32 -0700 | [diff] [blame] | 1259 | if (PageAnon(page) && !PageSwapBacked(page)) { |
| 1260 | /* follow __remove_mapping for reference */ |
| 1261 | if (!page_ref_freeze(page, 1)) |
| 1262 | goto keep_locked; |
| 1263 | if (PageDirty(page)) { |
| 1264 | page_ref_unfreeze(page, 1); |
| 1265 | goto keep_locked; |
| 1266 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1267 | |
Shaohua Li | 802a3a9 | 2017-05-03 14:52:32 -0700 | [diff] [blame] | 1268 | count_vm_event(PGLAZYFREED); |
| 1269 | } else if (!mapping || !__remove_mapping(mapping, page, true)) |
| 1270 | goto keep_locked; |
Nick Piggin | a978d6f | 2008-10-18 20:26:58 -0700 | [diff] [blame] | 1271 | /* |
| 1272 | * At this point, we have no other references and there is |
| 1273 | * no way to pick any more up (removed from LRU, removed |
| 1274 | * from pagecache). Can use non-atomic bitops now (and |
| 1275 | * we obviously don't have to worry about waking up a process |
| 1276 | * waiting on the page lock, because there are no references. |
| 1277 | */ |
Kirill A. Shutemov | 48c935a | 2016-01-15 16:51:24 -0800 | [diff] [blame] | 1278 | __ClearPageLocked(page); |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 1279 | free_it: |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 1280 | nr_reclaimed++; |
Mel Gorman | abe4c3b | 2010-08-09 17:19:31 -0700 | [diff] [blame] | 1281 | |
| 1282 | /* |
| 1283 | * Is there need to periodically free_page_list? It would |
| 1284 | * appear not as the counts should be low |
| 1285 | */ |
| 1286 | list_add(&page->lru, &free_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1287 | continue; |
| 1288 | |
| 1289 | activate_locked: |
Rik van Riel | 68a22394 | 2008-10-18 20:26:23 -0700 | [diff] [blame] | 1290 | /* Not a candidate for swapping, so reclaim swap space. */ |
Minchan Kim | ad6b670 | 2017-05-03 14:54:13 -0700 | [diff] [blame] | 1291 | if (PageSwapCache(page) && (mem_cgroup_swap_full(page) || |
| 1292 | PageMlocked(page))) |
Hugh Dickins | a2c43ee | 2009-01-06 14:39:36 -0800 | [diff] [blame] | 1293 | try_to_free_swap(page); |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 1294 | VM_BUG_ON_PAGE(PageActive(page), page); |
Minchan Kim | ad6b670 | 2017-05-03 14:54:13 -0700 | [diff] [blame] | 1295 | if (!PageMlocked(page)) { |
| 1296 | SetPageActive(page); |
| 1297 | pgactivate++; |
| 1298 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1299 | keep_locked: |
| 1300 | unlock_page(page); |
| 1301 | keep: |
| 1302 | list_add(&page->lru, &ret_pages); |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 1303 | VM_BUG_ON_PAGE(PageLRU(page) || PageUnevictable(page), page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1304 | } |
Mel Gorman | abe4c3b | 2010-08-09 17:19:31 -0700 | [diff] [blame] | 1305 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 1306 | mem_cgroup_uncharge_list(&free_pages); |
Mel Gorman | 72b252a | 2015-09-04 15:47:32 -0700 | [diff] [blame] | 1307 | try_to_unmap_flush(); |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 1308 | free_hot_cold_page_list(&free_pages, true); |
Mel Gorman | abe4c3b | 2010-08-09 17:19:31 -0700 | [diff] [blame] | 1309 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1310 | list_splice(&ret_pages, page_list); |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 1311 | count_vm_events(PGACTIVATE, pgactivate); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 1312 | |
Michal Hocko | 3c710c1 | 2017-02-22 15:44:27 -0800 | [diff] [blame] | 1313 | if (stat) { |
| 1314 | stat->nr_dirty = nr_dirty; |
| 1315 | stat->nr_congested = nr_congested; |
| 1316 | stat->nr_unqueued_dirty = nr_unqueued_dirty; |
| 1317 | stat->nr_writeback = nr_writeback; |
| 1318 | stat->nr_immediate = nr_immediate; |
Michal Hocko | 5bccd16 | 2017-02-22 15:44:30 -0800 | [diff] [blame] | 1319 | stat->nr_activate = pgactivate; |
| 1320 | stat->nr_ref_keep = nr_ref_keep; |
| 1321 | stat->nr_unmap_fail = nr_unmap_fail; |
Michal Hocko | 3c710c1 | 2017-02-22 15:44:27 -0800 | [diff] [blame] | 1322 | } |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 1323 | return nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1324 | } |
| 1325 | |
Minchan Kim | 02c6de8 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 1326 | unsigned long reclaim_clean_pages_from_list(struct zone *zone, |
| 1327 | struct list_head *page_list) |
| 1328 | { |
| 1329 | struct scan_control sc = { |
| 1330 | .gfp_mask = GFP_KERNEL, |
| 1331 | .priority = DEF_PRIORITY, |
| 1332 | .may_unmap = 1, |
| 1333 | }; |
Michal Hocko | 3c710c1 | 2017-02-22 15:44:27 -0800 | [diff] [blame] | 1334 | unsigned long ret; |
Minchan Kim | 02c6de8 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 1335 | struct page *page, *next; |
| 1336 | LIST_HEAD(clean_pages); |
| 1337 | |
| 1338 | list_for_each_entry_safe(page, next, page_list, lru) { |
Rafael Aquini | 117aad1 | 2013-09-30 13:45:16 -0700 | [diff] [blame] | 1339 | if (page_is_file_cache(page) && !PageDirty(page) && |
Minchan Kim | b1123ea6 | 2016-07-26 15:23:09 -0700 | [diff] [blame] | 1340 | !__PageMovable(page)) { |
Minchan Kim | 02c6de8 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 1341 | ClearPageActive(page); |
| 1342 | list_move(&page->lru, &clean_pages); |
| 1343 | } |
| 1344 | } |
| 1345 | |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1346 | ret = shrink_page_list(&clean_pages, zone->zone_pgdat, &sc, |
Shaohua Li | a128ca7 | 2017-05-03 14:52:22 -0700 | [diff] [blame] | 1347 | TTU_IGNORE_ACCESS, NULL, true); |
Minchan Kim | 02c6de8 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 1348 | list_splice(&clean_pages, page_list); |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1349 | mod_node_page_state(zone->zone_pgdat, NR_ISOLATED_FILE, -ret); |
Minchan Kim | 02c6de8 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 1350 | return ret; |
| 1351 | } |
| 1352 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1353 | /* |
| 1354 | * Attempt to remove the specified page from its LRU. Only take this page |
| 1355 | * if it is of the appropriate PageActive status. Pages which are being |
| 1356 | * freed elsewhere are also ignored. |
| 1357 | * |
| 1358 | * page: page to consider |
| 1359 | * mode: one of the LRU isolation modes defined above |
| 1360 | * |
| 1361 | * returns 0 on success, -ve errno on failure. |
| 1362 | */ |
Konstantin Khlebnikov | f3fd4a6 | 2012-05-29 15:06:54 -0700 | [diff] [blame] | 1363 | int __isolate_lru_page(struct page *page, isolate_mode_t mode) |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1364 | { |
| 1365 | int ret = -EINVAL; |
| 1366 | |
| 1367 | /* Only take pages on the LRU. */ |
| 1368 | if (!PageLRU(page)) |
| 1369 | return ret; |
| 1370 | |
Minchan Kim | e46a287 | 2012-10-08 16:33:48 -0700 | [diff] [blame] | 1371 | /* Compaction should not handle unevictable pages but CMA can do so */ |
| 1372 | if (PageUnevictable(page) && !(mode & ISOLATE_UNEVICTABLE)) |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1373 | return ret; |
| 1374 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1375 | ret = -EBUSY; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 1376 | |
Mel Gorman | c824493 | 2012-01-12 17:19:38 -0800 | [diff] [blame] | 1377 | /* |
| 1378 | * To minimise LRU disruption, the caller can indicate that it only |
| 1379 | * wants to isolate pages it will be able to operate on without |
| 1380 | * blocking - clean pages for the most part. |
| 1381 | * |
Mel Gorman | c824493 | 2012-01-12 17:19:38 -0800 | [diff] [blame] | 1382 | * ISOLATE_ASYNC_MIGRATE is used to indicate that it only wants to pages |
| 1383 | * that it is possible to migrate without blocking |
| 1384 | */ |
Johannes Weiner | 1276ad6 | 2017-02-24 14:56:11 -0800 | [diff] [blame] | 1385 | if (mode & ISOLATE_ASYNC_MIGRATE) { |
Mel Gorman | c824493 | 2012-01-12 17:19:38 -0800 | [diff] [blame] | 1386 | /* All the caller can do on PageWriteback is block */ |
| 1387 | if (PageWriteback(page)) |
| 1388 | return ret; |
| 1389 | |
| 1390 | if (PageDirty(page)) { |
| 1391 | struct address_space *mapping; |
| 1392 | |
Mel Gorman | c824493 | 2012-01-12 17:19:38 -0800 | [diff] [blame] | 1393 | /* |
| 1394 | * Only pages without mappings or that have a |
| 1395 | * ->migratepage callback are possible to migrate |
| 1396 | * without blocking |
| 1397 | */ |
| 1398 | mapping = page_mapping(page); |
| 1399 | if (mapping && !mapping->a_ops->migratepage) |
| 1400 | return ret; |
| 1401 | } |
| 1402 | } |
Minchan Kim | 39deaf8 | 2011-10-31 17:06:51 -0700 | [diff] [blame] | 1403 | |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1404 | if ((mode & ISOLATE_UNMAPPED) && page_mapped(page)) |
| 1405 | return ret; |
| 1406 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1407 | if (likely(get_page_unless_zero(page))) { |
| 1408 | /* |
| 1409 | * Be careful not to clear PageLRU until after we're |
| 1410 | * sure the page is not being freed elsewhere -- the |
| 1411 | * page release code relies on it. |
| 1412 | */ |
| 1413 | ClearPageLRU(page); |
| 1414 | ret = 0; |
| 1415 | } |
| 1416 | |
| 1417 | return ret; |
| 1418 | } |
| 1419 | |
Mel Gorman | 7ee36a1 | 2016-07-28 15:47:17 -0700 | [diff] [blame] | 1420 | |
| 1421 | /* |
| 1422 | * Update LRU sizes after isolating pages. The LRU size updates must |
| 1423 | * be complete before mem_cgroup_update_lru_size due to a santity check. |
| 1424 | */ |
| 1425 | static __always_inline void update_lru_sizes(struct lruvec *lruvec, |
Michal Hocko | b4536f0c8 | 2017-01-10 16:58:04 -0800 | [diff] [blame] | 1426 | enum lru_list lru, unsigned long *nr_zone_taken) |
Mel Gorman | 7ee36a1 | 2016-07-28 15:47:17 -0700 | [diff] [blame] | 1427 | { |
Mel Gorman | 7ee36a1 | 2016-07-28 15:47:17 -0700 | [diff] [blame] | 1428 | int zid; |
| 1429 | |
Mel Gorman | 7ee36a1 | 2016-07-28 15:47:17 -0700 | [diff] [blame] | 1430 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
| 1431 | if (!nr_zone_taken[zid]) |
| 1432 | continue; |
| 1433 | |
| 1434 | __update_lru_size(lruvec, lru, zid, -nr_zone_taken[zid]); |
Michal Hocko | b4536f0c8 | 2017-01-10 16:58:04 -0800 | [diff] [blame] | 1435 | #ifdef CONFIG_MEMCG |
| 1436 | mem_cgroup_update_lru_size(lruvec, lru, zid, -nr_zone_taken[zid]); |
| 1437 | #endif |
Mel Gorman | 7ee36a1 | 2016-07-28 15:47:17 -0700 | [diff] [blame] | 1438 | } |
Mel Gorman | 7ee36a1 | 2016-07-28 15:47:17 -0700 | [diff] [blame] | 1439 | |
Mel Gorman | 7ee36a1 | 2016-07-28 15:47:17 -0700 | [diff] [blame] | 1440 | } |
| 1441 | |
Christoph Lameter | 49d2e9c | 2006-01-08 01:00:48 -0800 | [diff] [blame] | 1442 | /* |
Mel Gorman | a52633d | 2016-07-28 15:45:28 -0700 | [diff] [blame] | 1443 | * zone_lru_lock is heavily contended. Some of the functions that |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1444 | * shrink the lists perform better by taking out a batch of pages |
| 1445 | * and working on them outside the LRU lock. |
| 1446 | * |
| 1447 | * For pagecache intensive workloads, this function is the hottest |
| 1448 | * spot in the kernel (apart from copy_*_user functions). |
| 1449 | * |
| 1450 | * Appropriate locks must be held before calling this function. |
| 1451 | * |
| 1452 | * @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] | 1453 | * @lruvec: The LRU vector to pull pages from. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1454 | * @dst: The temp list to put pages on to. |
Hugh Dickins | f626012 | 2012-01-12 17:20:06 -0800 | [diff] [blame] | 1455 | * @nr_scanned: The number of pages that were scanned. |
Rik van Riel | fe2c2a1 | 2012-03-21 16:33:51 -0700 | [diff] [blame] | 1456 | * @sc: The scan_control struct for this reclaim session |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1457 | * @mode: One of the LRU isolation modes |
Konstantin Khlebnikov | 3cb9945 | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 1458 | * @lru: LRU list id for isolating |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1459 | * |
| 1460 | * returns how many pages were moved onto *@dst. |
| 1461 | */ |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1462 | static unsigned long isolate_lru_pages(unsigned long nr_to_scan, |
Konstantin Khlebnikov | 5dc3597 | 2012-05-29 15:06:58 -0700 | [diff] [blame] | 1463 | struct lruvec *lruvec, struct list_head *dst, |
Rik van Riel | fe2c2a1 | 2012-03-21 16:33:51 -0700 | [diff] [blame] | 1464 | unsigned long *nr_scanned, struct scan_control *sc, |
Konstantin Khlebnikov | 3cb9945 | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 1465 | isolate_mode_t mode, enum lru_list lru) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1466 | { |
Hugh Dickins | 75b00af | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1467 | struct list_head *src = &lruvec->lists[lru]; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 1468 | unsigned long nr_taken = 0; |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1469 | unsigned long nr_zone_taken[MAX_NR_ZONES] = { 0 }; |
Mel Gorman | 7cc30fc | 2016-07-28 15:46:59 -0700 | [diff] [blame] | 1470 | unsigned long nr_skipped[MAX_NR_ZONES] = { 0, }; |
Johannes Weiner | 3db6581 | 2017-05-03 14:52:13 -0700 | [diff] [blame] | 1471 | unsigned long skipped = 0; |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1472 | unsigned long scan, nr_pages; |
Mel Gorman | b2e1875 | 2016-07-28 15:45:37 -0700 | [diff] [blame] | 1473 | LIST_HEAD(pages_skipped); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1474 | |
Vladimir Davydov | 0b802f1 | 2015-09-08 15:02:18 -0700 | [diff] [blame] | 1475 | for (scan = 0; scan < nr_to_scan && nr_taken < nr_to_scan && |
Johannes Weiner | 3db6581 | 2017-05-03 14:52:13 -0700 | [diff] [blame] | 1476 | !list_empty(src); scan++) { |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1477 | struct page *page; |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1478 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1479 | page = lru_to_page(src); |
| 1480 | prefetchw_prev_lru_page(page, src, flags); |
| 1481 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 1482 | VM_BUG_ON_PAGE(!PageLRU(page), page); |
Nick Piggin | 8d438f9 | 2006-03-22 00:07:59 -0800 | [diff] [blame] | 1483 | |
Mel Gorman | b2e1875 | 2016-07-28 15:45:37 -0700 | [diff] [blame] | 1484 | if (page_zonenum(page) > sc->reclaim_idx) { |
| 1485 | list_move(&page->lru, &pages_skipped); |
Mel Gorman | 7cc30fc | 2016-07-28 15:46:59 -0700 | [diff] [blame] | 1486 | nr_skipped[page_zonenum(page)]++; |
Mel Gorman | b2e1875 | 2016-07-28 15:45:37 -0700 | [diff] [blame] | 1487 | continue; |
| 1488 | } |
| 1489 | |
Konstantin Khlebnikov | f3fd4a6 | 2012-05-29 15:06:54 -0700 | [diff] [blame] | 1490 | switch (__isolate_lru_page(page, mode)) { |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1491 | case 0: |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1492 | nr_pages = hpage_nr_pages(page); |
| 1493 | nr_taken += nr_pages; |
| 1494 | nr_zone_taken[page_zonenum(page)] += nr_pages; |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1495 | list_move(&page->lru, dst); |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1496 | break; |
Nick Piggin | 46453a6 | 2006-03-22 00:07:58 -0800 | [diff] [blame] | 1497 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1498 | case -EBUSY: |
| 1499 | /* else it is being freed elsewhere */ |
| 1500 | list_move(&page->lru, src); |
| 1501 | continue; |
| 1502 | |
| 1503 | default: |
| 1504 | BUG(); |
| 1505 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1506 | } |
| 1507 | |
Mel Gorman | b2e1875 | 2016-07-28 15:45:37 -0700 | [diff] [blame] | 1508 | /* |
| 1509 | * Splice any skipped pages to the start of the LRU list. Note that |
| 1510 | * this disrupts the LRU order when reclaiming for lower zones but |
| 1511 | * we cannot splice to the tail. If we did then the SWAP_CLUSTER_MAX |
| 1512 | * scanning would soon rescan the same pages to skip and put the |
| 1513 | * system at risk of premature OOM. |
| 1514 | */ |
Mel Gorman | 7cc30fc | 2016-07-28 15:46:59 -0700 | [diff] [blame] | 1515 | if (!list_empty(&pages_skipped)) { |
| 1516 | int zid; |
| 1517 | |
Johannes Weiner | 3db6581 | 2017-05-03 14:52:13 -0700 | [diff] [blame] | 1518 | list_splice(&pages_skipped, src); |
Mel Gorman | 7cc30fc | 2016-07-28 15:46:59 -0700 | [diff] [blame] | 1519 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
| 1520 | if (!nr_skipped[zid]) |
| 1521 | continue; |
| 1522 | |
| 1523 | __count_zid_vm_events(PGSCAN_SKIP, zid, nr_skipped[zid]); |
Michal Hocko | 1265e3a | 2017-02-22 15:44:21 -0800 | [diff] [blame] | 1524 | skipped += nr_skipped[zid]; |
Mel Gorman | 7cc30fc | 2016-07-28 15:46:59 -0700 | [diff] [blame] | 1525 | } |
| 1526 | } |
Johannes Weiner | 3db6581 | 2017-05-03 14:52:13 -0700 | [diff] [blame] | 1527 | *nr_scanned = scan; |
Michal Hocko | 1265e3a | 2017-02-22 15:44:21 -0800 | [diff] [blame] | 1528 | trace_mm_vmscan_lru_isolate(sc->reclaim_idx, sc->order, nr_to_scan, |
Michal Hocko | 32b3f29 | 2017-02-22 15:44:24 -0800 | [diff] [blame] | 1529 | scan, skipped, nr_taken, mode, lru); |
Michal Hocko | b4536f0c8 | 2017-01-10 16:58:04 -0800 | [diff] [blame] | 1530 | update_lru_sizes(lruvec, lru, nr_zone_taken); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1531 | return nr_taken; |
| 1532 | } |
| 1533 | |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1534 | /** |
| 1535 | * isolate_lru_page - tries to isolate a page from its LRU list |
| 1536 | * @page: page to isolate from its LRU list |
| 1537 | * |
| 1538 | * Isolates a @page from an LRU list, clears PageLRU and adjusts the |
| 1539 | * vmstat statistic corresponding to whatever LRU list the page was on. |
| 1540 | * |
| 1541 | * Returns 0 if the page was removed from an LRU list. |
| 1542 | * Returns -EBUSY if the page was not on an LRU list. |
| 1543 | * |
| 1544 | * The returned page will have PageLRU() cleared. If it was found on |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1545 | * the active list, it will have PageActive set. If it was found on |
| 1546 | * the unevictable list, it will have the PageUnevictable bit set. That flag |
| 1547 | * 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] | 1548 | * |
| 1549 | * The vmstat statistic corresponding to the list on which the page was |
| 1550 | * found will be decremented. |
| 1551 | * |
| 1552 | * Restrictions: |
| 1553 | * (1) Must be called with an elevated refcount on the page. This is a |
| 1554 | * fundamentnal difference from isolate_lru_pages (which is called |
| 1555 | * without a stable reference). |
| 1556 | * (2) the lru_lock must not be held. |
| 1557 | * (3) interrupts must be enabled. |
| 1558 | */ |
| 1559 | int isolate_lru_page(struct page *page) |
| 1560 | { |
| 1561 | int ret = -EBUSY; |
| 1562 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 1563 | VM_BUG_ON_PAGE(!page_count(page), page); |
Kirill A. Shutemov | cf2a82e | 2016-02-05 15:36:36 -0800 | [diff] [blame] | 1564 | WARN_RATELIMIT(PageTail(page), "trying to isolate tail page"); |
Konstantin Khlebnikov | 0c91731 | 2011-05-24 17:12:21 -0700 | [diff] [blame] | 1565 | |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1566 | if (PageLRU(page)) { |
| 1567 | struct zone *zone = page_zone(page); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1568 | struct lruvec *lruvec; |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1569 | |
Mel Gorman | a52633d | 2016-07-28 15:45:28 -0700 | [diff] [blame] | 1570 | spin_lock_irq(zone_lru_lock(zone)); |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1571 | lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat); |
Konstantin Khlebnikov | 0c91731 | 2011-05-24 17:12:21 -0700 | [diff] [blame] | 1572 | if (PageLRU(page)) { |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1573 | int lru = page_lru(page); |
Konstantin Khlebnikov | 0c91731 | 2011-05-24 17:12:21 -0700 | [diff] [blame] | 1574 | get_page(page); |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1575 | ClearPageLRU(page); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1576 | del_page_from_lru_list(page, lruvec, lru); |
| 1577 | ret = 0; |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1578 | } |
Mel Gorman | a52633d | 2016-07-28 15:45:28 -0700 | [diff] [blame] | 1579 | spin_unlock_irq(zone_lru_lock(zone)); |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1580 | } |
| 1581 | return ret; |
| 1582 | } |
| 1583 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 1584 | /* |
Fengguang Wu | d37dd5d | 2012-12-18 14:23:28 -0800 | [diff] [blame] | 1585 | * A direct reclaimer may isolate SWAP_CLUSTER_MAX pages from the LRU list and |
| 1586 | * then get resheduled. When there are massive number of tasks doing page |
| 1587 | * allocation, such sleeping direct reclaimers may keep piling up on each CPU, |
| 1588 | * the LRU list will go small and be scanned faster than necessary, leading to |
| 1589 | * unnecessary swapping, thrashing and OOM. |
Rik van Riel | 35cd781 | 2009-09-21 17:01:38 -0700 | [diff] [blame] | 1590 | */ |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1591 | static int too_many_isolated(struct pglist_data *pgdat, int file, |
Rik van Riel | 35cd781 | 2009-09-21 17:01:38 -0700 | [diff] [blame] | 1592 | struct scan_control *sc) |
| 1593 | { |
| 1594 | unsigned long inactive, isolated; |
| 1595 | |
| 1596 | if (current_is_kswapd()) |
| 1597 | return 0; |
| 1598 | |
Tejun Heo | 97c9341 | 2015-05-22 18:23:36 -0400 | [diff] [blame] | 1599 | if (!sane_reclaim(sc)) |
Rik van Riel | 35cd781 | 2009-09-21 17:01:38 -0700 | [diff] [blame] | 1600 | return 0; |
| 1601 | |
| 1602 | if (file) { |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1603 | inactive = node_page_state(pgdat, NR_INACTIVE_FILE); |
| 1604 | isolated = node_page_state(pgdat, NR_ISOLATED_FILE); |
Rik van Riel | 35cd781 | 2009-09-21 17:01:38 -0700 | [diff] [blame] | 1605 | } else { |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1606 | inactive = node_page_state(pgdat, NR_INACTIVE_ANON); |
| 1607 | isolated = node_page_state(pgdat, NR_ISOLATED_ANON); |
Rik van Riel | 35cd781 | 2009-09-21 17:01:38 -0700 | [diff] [blame] | 1608 | } |
| 1609 | |
Fengguang Wu | 3cf2384 | 2012-12-18 14:23:31 -0800 | [diff] [blame] | 1610 | /* |
| 1611 | * GFP_NOIO/GFP_NOFS callers are allowed to isolate more pages, so they |
| 1612 | * won't get blocked by normal direct-reclaimers, forming a circular |
| 1613 | * deadlock. |
| 1614 | */ |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 1615 | if ((sc->gfp_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS)) |
Fengguang Wu | 3cf2384 | 2012-12-18 14:23:31 -0800 | [diff] [blame] | 1616 | inactive >>= 3; |
| 1617 | |
Rik van Riel | 35cd781 | 2009-09-21 17:01:38 -0700 | [diff] [blame] | 1618 | return isolated > inactive; |
| 1619 | } |
| 1620 | |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1621 | static noinline_for_stack void |
Hugh Dickins | 75b00af | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1622 | putback_inactive_pages(struct lruvec *lruvec, struct list_head *page_list) |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1623 | { |
Konstantin Khlebnikov | 27ac81d | 2012-05-29 15:07:00 -0700 | [diff] [blame] | 1624 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1625 | struct pglist_data *pgdat = lruvec_pgdat(lruvec); |
Hugh Dickins | 3f79768 | 2012-01-12 17:20:07 -0800 | [diff] [blame] | 1626 | LIST_HEAD(pages_to_free); |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1627 | |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1628 | /* |
| 1629 | * Put back any unfreeable pages. |
| 1630 | */ |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1631 | while (!list_empty(page_list)) { |
Hugh Dickins | 3f79768 | 2012-01-12 17:20:07 -0800 | [diff] [blame] | 1632 | struct page *page = lru_to_page(page_list); |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1633 | int lru; |
Hugh Dickins | 3f79768 | 2012-01-12 17:20:07 -0800 | [diff] [blame] | 1634 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 1635 | VM_BUG_ON_PAGE(PageLRU(page), page); |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1636 | list_del(&page->lru); |
Hugh Dickins | 39b5f29 | 2012-10-08 16:33:18 -0700 | [diff] [blame] | 1637 | if (unlikely(!page_evictable(page))) { |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1638 | spin_unlock_irq(&pgdat->lru_lock); |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1639 | putback_lru_page(page); |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1640 | spin_lock_irq(&pgdat->lru_lock); |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1641 | continue; |
| 1642 | } |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1643 | |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1644 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1645 | |
Linus Torvalds | 7a60857 | 2011-01-17 14:42:19 -0800 | [diff] [blame] | 1646 | SetPageLRU(page); |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1647 | lru = page_lru(page); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1648 | add_page_to_lru_list(page, lruvec, lru); |
| 1649 | |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1650 | if (is_active_lru(lru)) { |
| 1651 | int file = is_file_lru(lru); |
Rik van Riel | 9992af1 | 2011-01-13 15:47:13 -0800 | [diff] [blame] | 1652 | int numpages = hpage_nr_pages(page); |
| 1653 | reclaim_stat->recent_rotated[file] += numpages; |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1654 | } |
Hugh Dickins | 2bcf887 | 2012-01-12 17:19:56 -0800 | [diff] [blame] | 1655 | if (put_page_testzero(page)) { |
| 1656 | __ClearPageLRU(page); |
| 1657 | __ClearPageActive(page); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1658 | del_page_from_lru_list(page, lruvec, lru); |
Hugh Dickins | 2bcf887 | 2012-01-12 17:19:56 -0800 | [diff] [blame] | 1659 | |
| 1660 | if (unlikely(PageCompound(page))) { |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1661 | spin_unlock_irq(&pgdat->lru_lock); |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 1662 | mem_cgroup_uncharge(page); |
Hugh Dickins | 2bcf887 | 2012-01-12 17:19:56 -0800 | [diff] [blame] | 1663 | (*get_compound_page_dtor(page))(page); |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1664 | spin_lock_irq(&pgdat->lru_lock); |
Hugh Dickins | 2bcf887 | 2012-01-12 17:19:56 -0800 | [diff] [blame] | 1665 | } else |
| 1666 | list_add(&page->lru, &pages_to_free); |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1667 | } |
| 1668 | } |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1669 | |
Hugh Dickins | 3f79768 | 2012-01-12 17:20:07 -0800 | [diff] [blame] | 1670 | /* |
| 1671 | * To save our caller's stack, now use input list for pages to free. |
| 1672 | */ |
| 1673 | list_splice(&pages_to_free, page_list); |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1674 | } |
| 1675 | |
| 1676 | /* |
NeilBrown | 399ba0b | 2014-06-04 16:07:42 -0700 | [diff] [blame] | 1677 | * If a kernel thread (such as nfsd for loop-back mounts) services |
| 1678 | * a backing device by writing to the page cache it sets PF_LESS_THROTTLE. |
| 1679 | * In that case we should only throttle if the backing device it is |
| 1680 | * writing to is congested. In other cases it is safe to throttle. |
| 1681 | */ |
| 1682 | static int current_may_throttle(void) |
| 1683 | { |
| 1684 | return !(current->flags & PF_LESS_THROTTLE) || |
| 1685 | current->backing_dev_info == NULL || |
| 1686 | bdi_write_congested(current->backing_dev_info); |
| 1687 | } |
| 1688 | |
| 1689 | /* |
Mel Gorman | b2e1875 | 2016-07-28 15:45:37 -0700 | [diff] [blame] | 1690 | * shrink_inactive_list() is a helper for shrink_node(). It returns the number |
Andrew Morton | 1742f19 | 2006-03-22 00:08:21 -0800 | [diff] [blame] | 1691 | * of reclaimed pages |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1692 | */ |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1693 | static noinline_for_stack unsigned long |
Konstantin Khlebnikov | 1a93be0 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 1694 | shrink_inactive_list(unsigned long nr_to_scan, struct lruvec *lruvec, |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1695 | struct scan_control *sc, enum lru_list lru) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1696 | { |
| 1697 | LIST_HEAD(page_list); |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1698 | unsigned long nr_scanned; |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 1699 | unsigned long nr_reclaimed = 0; |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1700 | unsigned long nr_taken; |
Michal Hocko | 3c710c1 | 2017-02-22 15:44:27 -0800 | [diff] [blame] | 1701 | struct reclaim_stat stat = {}; |
Konstantin Khlebnikov | f3fd4a6 | 2012-05-29 15:06:54 -0700 | [diff] [blame] | 1702 | isolate_mode_t isolate_mode = 0; |
Konstantin Khlebnikov | 3cb9945 | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 1703 | int file = is_file_lru(lru); |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1704 | struct pglist_data *pgdat = lruvec_pgdat(lruvec); |
Konstantin Khlebnikov | 1a93be0 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 1705 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
KOSAKI Motohiro | 78dc583 | 2009-06-16 15:31:40 -0700 | [diff] [blame] | 1706 | |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1707 | while (unlikely(too_many_isolated(pgdat, file, sc))) { |
KOSAKI Motohiro | 58355c7 | 2009-10-26 16:49:35 -0700 | [diff] [blame] | 1708 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
Rik van Riel | 35cd781 | 2009-09-21 17:01:38 -0700 | [diff] [blame] | 1709 | |
| 1710 | /* We are about to die and free our memory. Return now. */ |
| 1711 | if (fatal_signal_pending(current)) |
| 1712 | return SWAP_CLUSTER_MAX; |
| 1713 | } |
| 1714 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1715 | lru_add_drain(); |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1716 | |
| 1717 | if (!sc->may_unmap) |
Hillf Danton | 6131728 | 2012-03-21 16:33:48 -0700 | [diff] [blame] | 1718 | isolate_mode |= ISOLATE_UNMAPPED; |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1719 | |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1720 | spin_lock_irq(&pgdat->lru_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1721 | |
Konstantin Khlebnikov | 5dc3597 | 2012-05-29 15:06:58 -0700 | [diff] [blame] | 1722 | nr_taken = isolate_lru_pages(nr_to_scan, lruvec, &page_list, |
| 1723 | &nr_scanned, sc, isolate_mode, lru); |
Konstantin Khlebnikov | 95d918f | 2012-05-29 15:06:59 -0700 | [diff] [blame] | 1724 | |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1725 | __mod_node_page_state(pgdat, NR_ISOLATED_ANON + file, nr_taken); |
Hugh Dickins | 9d5e6a9 | 2016-05-19 17:12:38 -0700 | [diff] [blame] | 1726 | reclaim_stat->recent_scanned[file] += nr_taken; |
Konstantin Khlebnikov | 95d918f | 2012-05-29 15:06:59 -0700 | [diff] [blame] | 1727 | |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 1728 | if (global_reclaim(sc)) { |
KOSAKI Motohiro | b35ea17 | 2009-09-21 17:01:36 -0700 | [diff] [blame] | 1729 | if (current_is_kswapd()) |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1730 | __count_vm_events(PGSCAN_KSWAPD, nr_scanned); |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1731 | else |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1732 | __count_vm_events(PGSCAN_DIRECT, nr_scanned); |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1733 | } |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1734 | spin_unlock_irq(&pgdat->lru_lock); |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 1735 | |
Hillf Danton | d563c05 | 2012-03-21 16:34:02 -0700 | [diff] [blame] | 1736 | if (nr_taken == 0) |
Mel Gorman | 6663562 | 2010-08-09 17:19:30 -0700 | [diff] [blame] | 1737 | return 0; |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1738 | |
Shaohua Li | a128ca7 | 2017-05-03 14:52:22 -0700 | [diff] [blame] | 1739 | nr_reclaimed = shrink_page_list(&page_list, pgdat, sc, 0, |
Michal Hocko | 3c710c1 | 2017-02-22 15:44:27 -0800 | [diff] [blame] | 1740 | &stat, false); |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1741 | |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1742 | spin_lock_irq(&pgdat->lru_lock); |
Hugh Dickins | 3f79768 | 2012-01-12 17:20:07 -0800 | [diff] [blame] | 1743 | |
Ying Han | 904249a | 2012-04-25 16:01:48 -0700 | [diff] [blame] | 1744 | if (global_reclaim(sc)) { |
| 1745 | if (current_is_kswapd()) |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1746 | __count_vm_events(PGSTEAL_KSWAPD, nr_reclaimed); |
Ying Han | 904249a | 2012-04-25 16:01:48 -0700 | [diff] [blame] | 1747 | else |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1748 | __count_vm_events(PGSTEAL_DIRECT, nr_reclaimed); |
Ying Han | 904249a | 2012-04-25 16:01:48 -0700 | [diff] [blame] | 1749 | } |
KOSAKI Motohiro | e247dbc | 2010-08-09 17:19:28 -0700 | [diff] [blame] | 1750 | |
Konstantin Khlebnikov | 27ac81d | 2012-05-29 15:07:00 -0700 | [diff] [blame] | 1751 | putback_inactive_pages(lruvec, &page_list); |
Hugh Dickins | 3f79768 | 2012-01-12 17:20:07 -0800 | [diff] [blame] | 1752 | |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1753 | __mod_node_page_state(pgdat, NR_ISOLATED_ANON + file, -nr_taken); |
Hugh Dickins | 3f79768 | 2012-01-12 17:20:07 -0800 | [diff] [blame] | 1754 | |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1755 | spin_unlock_irq(&pgdat->lru_lock); |
Hugh Dickins | 3f79768 | 2012-01-12 17:20:07 -0800 | [diff] [blame] | 1756 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 1757 | mem_cgroup_uncharge_list(&page_list); |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 1758 | free_hot_cold_page_list(&page_list, true); |
Mel Gorman | e11da5b | 2010-10-26 14:21:40 -0700 | [diff] [blame] | 1759 | |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 1760 | /* |
| 1761 | * If reclaim is isolating dirty pages under writeback, it implies |
| 1762 | * that the long-lived page allocation rate is exceeding the page |
| 1763 | * laundering rate. Either the global limits are not being effective |
| 1764 | * at throttling processes due to the page distribution throughout |
| 1765 | * zones or there is heavy usage of a slow backing device. The |
| 1766 | * only option is to throttle from reclaim context which is not ideal |
| 1767 | * as there is no guarantee the dirtying process is throttled in the |
| 1768 | * same way balance_dirty_pages() manages. |
| 1769 | * |
Mel Gorman | 8e95028 | 2013-07-03 15:02:02 -0700 | [diff] [blame] | 1770 | * Once a zone is flagged ZONE_WRITEBACK, kswapd will count the number |
| 1771 | * of pages under pages flagged for immediate reclaim and stall if any |
| 1772 | * are encountered in the nr_immediate check below. |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 1773 | */ |
Michal Hocko | 3c710c1 | 2017-02-22 15:44:27 -0800 | [diff] [blame] | 1774 | if (stat.nr_writeback && stat.nr_writeback == nr_taken) |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1775 | set_bit(PGDAT_WRITEBACK, &pgdat->flags); |
Mel Gorman | 92df3a7 | 2011-10-31 17:07:56 -0700 | [diff] [blame] | 1776 | |
Mel Gorman | d43006d | 2013-07-03 15:01:50 -0700 | [diff] [blame] | 1777 | /* |
Tejun Heo | 97c9341 | 2015-05-22 18:23:36 -0400 | [diff] [blame] | 1778 | * Legacy memcg will stall in page writeback so avoid forcibly |
| 1779 | * stalling here. |
Mel Gorman | d43006d | 2013-07-03 15:01:50 -0700 | [diff] [blame] | 1780 | */ |
Tejun Heo | 97c9341 | 2015-05-22 18:23:36 -0400 | [diff] [blame] | 1781 | if (sane_reclaim(sc)) { |
Mel Gorman | b1a6f21 | 2013-07-03 15:01:58 -0700 | [diff] [blame] | 1782 | /* |
Mel Gorman | 8e95028 | 2013-07-03 15:02:02 -0700 | [diff] [blame] | 1783 | * Tag a zone as congested if all the dirty pages scanned were |
| 1784 | * backed by a congested BDI and wait_iff_congested will stall. |
| 1785 | */ |
Michal Hocko | 3c710c1 | 2017-02-22 15:44:27 -0800 | [diff] [blame] | 1786 | if (stat.nr_dirty && stat.nr_dirty == stat.nr_congested) |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1787 | set_bit(PGDAT_CONGESTED, &pgdat->flags); |
Mel Gorman | 8e95028 | 2013-07-03 15:02:02 -0700 | [diff] [blame] | 1788 | |
| 1789 | /* |
Mel Gorman | b1a6f21 | 2013-07-03 15:01:58 -0700 | [diff] [blame] | 1790 | * If dirty pages are scanned that are not queued for IO, it |
Johannes Weiner | 726d061 | 2017-02-24 14:56:14 -0800 | [diff] [blame] | 1791 | * implies that flushers are not doing their job. This can |
| 1792 | * happen when memory pressure pushes dirty pages to the end of |
| 1793 | * the LRU before the dirty limits are breached and the dirty |
| 1794 | * data has expired. It can also happen when the proportion of |
| 1795 | * dirty pages grows not through writes but through memory |
| 1796 | * pressure reclaiming all the clean cache. And in some cases, |
| 1797 | * the flushers simply cannot keep up with the allocation |
| 1798 | * rate. Nudge the flusher threads in case they are asleep, but |
| 1799 | * also allow kswapd to start writing pages during reclaim. |
Mel Gorman | b1a6f21 | 2013-07-03 15:01:58 -0700 | [diff] [blame] | 1800 | */ |
Johannes Weiner | 726d061 | 2017-02-24 14:56:14 -0800 | [diff] [blame] | 1801 | if (stat.nr_unqueued_dirty == nr_taken) { |
| 1802 | wakeup_flusher_threads(0, WB_REASON_VMSCAN); |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1803 | set_bit(PGDAT_DIRTY, &pgdat->flags); |
Johannes Weiner | 726d061 | 2017-02-24 14:56:14 -0800 | [diff] [blame] | 1804 | } |
Mel Gorman | b1a6f21 | 2013-07-03 15:01:58 -0700 | [diff] [blame] | 1805 | |
| 1806 | /* |
Linus Torvalds | b738d76 | 2014-06-08 14:17:00 -0700 | [diff] [blame] | 1807 | * If kswapd scans pages marked marked for immediate |
| 1808 | * reclaim and under writeback (nr_immediate), it implies |
| 1809 | * that pages are cycling through the LRU faster than |
Mel Gorman | b1a6f21 | 2013-07-03 15:01:58 -0700 | [diff] [blame] | 1810 | * they are written so also forcibly stall. |
| 1811 | */ |
Michal Hocko | 3c710c1 | 2017-02-22 15:44:27 -0800 | [diff] [blame] | 1812 | if (stat.nr_immediate && current_may_throttle()) |
Mel Gorman | b1a6f21 | 2013-07-03 15:01:58 -0700 | [diff] [blame] | 1813 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
Mel Gorman | e2be15f | 2013-07-03 15:01:57 -0700 | [diff] [blame] | 1814 | } |
Mel Gorman | d43006d | 2013-07-03 15:01:50 -0700 | [diff] [blame] | 1815 | |
Mel Gorman | 8e95028 | 2013-07-03 15:02:02 -0700 | [diff] [blame] | 1816 | /* |
| 1817 | * Stall direct reclaim for IO completions if underlying BDIs or zone |
| 1818 | * is congested. Allow kswapd to continue until it starts encountering |
| 1819 | * unqueued dirty pages or cycling through the LRU too quickly. |
| 1820 | */ |
NeilBrown | 399ba0b | 2014-06-04 16:07:42 -0700 | [diff] [blame] | 1821 | if (!sc->hibernation_mode && !current_is_kswapd() && |
| 1822 | current_may_throttle()) |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1823 | wait_iff_congested(pgdat, BLK_RW_ASYNC, HZ/10); |
Mel Gorman | 8e95028 | 2013-07-03 15:02:02 -0700 | [diff] [blame] | 1824 | |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1825 | trace_mm_vmscan_lru_shrink_inactive(pgdat->node_id, |
| 1826 | nr_scanned, nr_reclaimed, |
Michal Hocko | 5bccd16 | 2017-02-22 15:44:30 -0800 | [diff] [blame] | 1827 | stat.nr_dirty, stat.nr_writeback, |
| 1828 | stat.nr_congested, stat.nr_immediate, |
| 1829 | stat.nr_activate, stat.nr_ref_keep, |
| 1830 | stat.nr_unmap_fail, |
yalin wang | ba5e957 | 2016-01-14 15:18:48 -0800 | [diff] [blame] | 1831 | sc->priority, file); |
Andrew Morton | 05ff513 | 2006-03-22 00:08:20 -0800 | [diff] [blame] | 1832 | return nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1833 | } |
| 1834 | |
Martin Bligh | 3bb1a852 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 1835 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1836 | * This moves pages from the active list to the inactive list. |
| 1837 | * |
| 1838 | * We move them the other way if the page is referenced by one or more |
| 1839 | * processes, from rmap. |
| 1840 | * |
| 1841 | * If the pages are mostly unmapped, the processing is fast and it is |
Mel Gorman | a52633d | 2016-07-28 15:45:28 -0700 | [diff] [blame] | 1842 | * appropriate to hold zone_lru_lock across the whole operation. But if |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1843 | * the pages are mapped, the processing is slow (page_referenced()) so we |
Mel Gorman | a52633d | 2016-07-28 15:45:28 -0700 | [diff] [blame] | 1844 | * should drop zone_lru_lock around each page. It's impossible to balance |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1845 | * this, so instead we remove the pages from the LRU while processing them. |
| 1846 | * It is safe to rely on PG_active against the non-LRU pages in here because |
| 1847 | * nobody will play with that bit on a non-LRU page. |
| 1848 | * |
Joonsoo Kim | 0139aa7 | 2016-05-19 17:10:49 -0700 | [diff] [blame] | 1849 | * The downside is that we have to touch page->_refcount against each page. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1850 | * But we had to alter page->flags anyway. |
Michal Hocko | 9d998b4 | 2017-02-22 15:44:18 -0800 | [diff] [blame] | 1851 | * |
| 1852 | * Returns the number of pages moved to the given lru. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1853 | */ |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1854 | |
Michal Hocko | 9d998b4 | 2017-02-22 15:44:18 -0800 | [diff] [blame] | 1855 | static unsigned move_active_pages_to_lru(struct lruvec *lruvec, |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1856 | struct list_head *list, |
Hugh Dickins | 2bcf887 | 2012-01-12 17:19:56 -0800 | [diff] [blame] | 1857 | struct list_head *pages_to_free, |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1858 | enum lru_list lru) |
| 1859 | { |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1860 | struct pglist_data *pgdat = lruvec_pgdat(lruvec); |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1861 | struct page *page; |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1862 | int nr_pages; |
Michal Hocko | 9d998b4 | 2017-02-22 15:44:18 -0800 | [diff] [blame] | 1863 | int nr_moved = 0; |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1864 | |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1865 | while (!list_empty(list)) { |
| 1866 | page = lru_to_page(list); |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1867 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1868 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 1869 | VM_BUG_ON_PAGE(PageLRU(page), page); |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1870 | SetPageLRU(page); |
| 1871 | |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1872 | nr_pages = hpage_nr_pages(page); |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1873 | update_lru_size(lruvec, lru, page_zonenum(page), nr_pages); |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1874 | list_move(&page->lru, &lruvec->lists[lru]); |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1875 | |
Hugh Dickins | 2bcf887 | 2012-01-12 17:19:56 -0800 | [diff] [blame] | 1876 | if (put_page_testzero(page)) { |
| 1877 | __ClearPageLRU(page); |
| 1878 | __ClearPageActive(page); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1879 | del_page_from_lru_list(page, lruvec, lru); |
Hugh Dickins | 2bcf887 | 2012-01-12 17:19:56 -0800 | [diff] [blame] | 1880 | |
| 1881 | if (unlikely(PageCompound(page))) { |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1882 | spin_unlock_irq(&pgdat->lru_lock); |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 1883 | mem_cgroup_uncharge(page); |
Hugh Dickins | 2bcf887 | 2012-01-12 17:19:56 -0800 | [diff] [blame] | 1884 | (*get_compound_page_dtor(page))(page); |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1885 | spin_lock_irq(&pgdat->lru_lock); |
Hugh Dickins | 2bcf887 | 2012-01-12 17:19:56 -0800 | [diff] [blame] | 1886 | } else |
| 1887 | list_add(&page->lru, pages_to_free); |
Michal Hocko | 9d998b4 | 2017-02-22 15:44:18 -0800 | [diff] [blame] | 1888 | } else { |
| 1889 | nr_moved += nr_pages; |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1890 | } |
| 1891 | } |
Hugh Dickins | 9d5e6a9 | 2016-05-19 17:12:38 -0700 | [diff] [blame] | 1892 | |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1893 | if (!is_active_lru(lru)) |
Michal Hocko | f095890 | 2017-02-22 15:45:55 -0800 | [diff] [blame] | 1894 | __count_vm_events(PGDEACTIVATE, nr_moved); |
Michal Hocko | 9d998b4 | 2017-02-22 15:44:18 -0800 | [diff] [blame] | 1895 | |
| 1896 | return nr_moved; |
Wu Fengguang | 3eb4140 | 2009-06-16 15:33:13 -0700 | [diff] [blame] | 1897 | } |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1898 | |
Hugh Dickins | f626012 | 2012-01-12 17:20:06 -0800 | [diff] [blame] | 1899 | static void shrink_active_list(unsigned long nr_to_scan, |
Konstantin Khlebnikov | 1a93be0 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 1900 | struct lruvec *lruvec, |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 1901 | struct scan_control *sc, |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 1902 | enum lru_list lru) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1903 | { |
KOSAKI Motohiro | 44c241f | 2009-09-21 17:01:35 -0700 | [diff] [blame] | 1904 | unsigned long nr_taken; |
Hugh Dickins | f626012 | 2012-01-12 17:20:06 -0800 | [diff] [blame] | 1905 | unsigned long nr_scanned; |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 1906 | unsigned long vm_flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1907 | LIST_HEAD(l_hold); /* The pages which were snipped off */ |
Wu Fengguang | 8cab475 | 2009-06-16 15:33:12 -0700 | [diff] [blame] | 1908 | LIST_HEAD(l_active); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1909 | LIST_HEAD(l_inactive); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1910 | struct page *page; |
Konstantin Khlebnikov | 1a93be0 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 1911 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
Michal Hocko | 9d998b4 | 2017-02-22 15:44:18 -0800 | [diff] [blame] | 1912 | unsigned nr_deactivate, nr_activate; |
| 1913 | unsigned nr_rotated = 0; |
Konstantin Khlebnikov | f3fd4a6 | 2012-05-29 15:06:54 -0700 | [diff] [blame] | 1914 | isolate_mode_t isolate_mode = 0; |
Konstantin Khlebnikov | 3cb9945 | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 1915 | int file = is_file_lru(lru); |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1916 | struct pglist_data *pgdat = lruvec_pgdat(lruvec); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1917 | |
| 1918 | lru_add_drain(); |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1919 | |
| 1920 | if (!sc->may_unmap) |
Hillf Danton | 6131728 | 2012-03-21 16:33:48 -0700 | [diff] [blame] | 1921 | isolate_mode |= ISOLATE_UNMAPPED; |
Minchan Kim | f80c067 | 2011-10-31 17:06:55 -0700 | [diff] [blame] | 1922 | |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1923 | spin_lock_irq(&pgdat->lru_lock); |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1924 | |
Konstantin Khlebnikov | 5dc3597 | 2012-05-29 15:06:58 -0700 | [diff] [blame] | 1925 | nr_taken = isolate_lru_pages(nr_to_scan, lruvec, &l_hold, |
| 1926 | &nr_scanned, sc, isolate_mode, lru); |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 1927 | |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1928 | __mod_node_page_state(pgdat, NR_ISOLATED_ANON + file, nr_taken); |
Johannes Weiner | b7c46d1 | 2009-09-21 17:02:56 -0700 | [diff] [blame] | 1929 | reclaim_stat->recent_scanned[file] += nr_taken; |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 1930 | |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1931 | __count_vm_events(PGREFILL, nr_scanned); |
Hugh Dickins | 9d5e6a9 | 2016-05-19 17:12:38 -0700 | [diff] [blame] | 1932 | |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1933 | spin_unlock_irq(&pgdat->lru_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1934 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1935 | while (!list_empty(&l_hold)) { |
| 1936 | cond_resched(); |
| 1937 | page = lru_to_page(&l_hold); |
| 1938 | list_del(&page->lru); |
Rik van Riel | 7e9cd48 | 2008-10-18 20:26:35 -0700 | [diff] [blame] | 1939 | |
Hugh Dickins | 39b5f29 | 2012-10-08 16:33:18 -0700 | [diff] [blame] | 1940 | if (unlikely(!page_evictable(page))) { |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 1941 | putback_lru_page(page); |
| 1942 | continue; |
| 1943 | } |
| 1944 | |
Mel Gorman | cc715d9 | 2012-03-21 16:34:00 -0700 | [diff] [blame] | 1945 | if (unlikely(buffer_heads_over_limit)) { |
| 1946 | if (page_has_private(page) && trylock_page(page)) { |
| 1947 | if (page_has_private(page)) |
| 1948 | try_to_release_page(page, 0); |
| 1949 | unlock_page(page); |
| 1950 | } |
| 1951 | } |
| 1952 | |
Johannes Weiner | c3ac9a8 | 2012-05-29 15:06:25 -0700 | [diff] [blame] | 1953 | if (page_referenced(page, 0, sc->target_mem_cgroup, |
| 1954 | &vm_flags)) { |
Rik van Riel | 9992af1 | 2011-01-13 15:47:13 -0800 | [diff] [blame] | 1955 | nr_rotated += hpage_nr_pages(page); |
Wu Fengguang | 8cab475 | 2009-06-16 15:33:12 -0700 | [diff] [blame] | 1956 | /* |
| 1957 | * Identify referenced, file-backed active pages and |
| 1958 | * give them one more trip around the active list. So |
| 1959 | * that executable code get better chances to stay in |
| 1960 | * memory under moderate memory pressure. Anon pages |
| 1961 | * are not likely to be evicted by use-once streaming |
| 1962 | * IO, plus JVM can create lots of anon VM_EXEC pages, |
| 1963 | * so we ignore them here. |
| 1964 | */ |
Wu Fengguang | 41e2098 | 2009-10-26 16:49:53 -0700 | [diff] [blame] | 1965 | if ((vm_flags & VM_EXEC) && page_is_file_cache(page)) { |
Wu Fengguang | 8cab475 | 2009-06-16 15:33:12 -0700 | [diff] [blame] | 1966 | list_add(&page->lru, &l_active); |
| 1967 | continue; |
| 1968 | } |
| 1969 | } |
Rik van Riel | 7e9cd48 | 2008-10-18 20:26:35 -0700 | [diff] [blame] | 1970 | |
KOSAKI Motohiro | 5205e56 | 2009-09-21 17:01:44 -0700 | [diff] [blame] | 1971 | ClearPageActive(page); /* we are de-activating */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1972 | list_add(&page->lru, &l_inactive); |
| 1973 | } |
| 1974 | |
Andrew Morton | b555749 | 2009-01-06 14:40:13 -0800 | [diff] [blame] | 1975 | /* |
Wu Fengguang | 8cab475 | 2009-06-16 15:33:12 -0700 | [diff] [blame] | 1976 | * Move pages back to the lru list. |
Andrew Morton | b555749 | 2009-01-06 14:40:13 -0800 | [diff] [blame] | 1977 | */ |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1978 | spin_lock_irq(&pgdat->lru_lock); |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1979 | /* |
Wu Fengguang | 8cab475 | 2009-06-16 15:33:12 -0700 | [diff] [blame] | 1980 | * Count referenced pages from currently used mappings as rotated, |
| 1981 | * even though only some of them are actually re-activated. This |
| 1982 | * helps balance scan pressure between file and anonymous pages in |
Jerome Marchand | 7c0db9e | 2014-08-06 16:08:01 -0700 | [diff] [blame] | 1983 | * get_scan_count. |
Rik van Riel | 7e9cd48 | 2008-10-18 20:26:35 -0700 | [diff] [blame] | 1984 | */ |
Johannes Weiner | b7c46d1 | 2009-09-21 17:02:56 -0700 | [diff] [blame] | 1985 | reclaim_stat->recent_rotated[file] += nr_rotated; |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 1986 | |
Michal Hocko | 9d998b4 | 2017-02-22 15:44:18 -0800 | [diff] [blame] | 1987 | nr_activate = move_active_pages_to_lru(lruvec, &l_active, &l_hold, lru); |
| 1988 | nr_deactivate = move_active_pages_to_lru(lruvec, &l_inactive, &l_hold, lru - LRU_ACTIVE); |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1989 | __mod_node_page_state(pgdat, NR_ISOLATED_ANON + file, -nr_taken); |
| 1990 | spin_unlock_irq(&pgdat->lru_lock); |
Hugh Dickins | 2bcf887 | 2012-01-12 17:19:56 -0800 | [diff] [blame] | 1991 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 1992 | mem_cgroup_uncharge_list(&l_hold); |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 1993 | free_hot_cold_page_list(&l_hold, true); |
Michal Hocko | 9d998b4 | 2017-02-22 15:44:18 -0800 | [diff] [blame] | 1994 | trace_mm_vmscan_lru_shrink_active(pgdat->node_id, nr_taken, nr_activate, |
| 1995 | nr_deactivate, nr_rotated, sc->priority, file); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1996 | } |
| 1997 | |
Rik van Riel | 59dc76b | 2016-05-20 16:56:31 -0700 | [diff] [blame] | 1998 | /* |
| 1999 | * The inactive anon list should be small enough that the VM never has |
| 2000 | * to do too much work. |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 2001 | * |
Rik van Riel | 59dc76b | 2016-05-20 16:56:31 -0700 | [diff] [blame] | 2002 | * The inactive file list should be small enough to leave most memory |
| 2003 | * to the established workingset on the scan-resistant active list, |
| 2004 | * but large enough to avoid thrashing the aggregate readahead window. |
| 2005 | * |
| 2006 | * Both inactive lists should also be large enough that each inactive |
| 2007 | * page has a chance to be referenced again before it is reclaimed. |
| 2008 | * |
Johannes Weiner | 2a2e488 | 2017-05-03 14:55:03 -0700 | [diff] [blame] | 2009 | * If that fails and refaulting is observed, the inactive list grows. |
| 2010 | * |
Rik van Riel | 59dc76b | 2016-05-20 16:56:31 -0700 | [diff] [blame] | 2011 | * The inactive_ratio is the target ratio of ACTIVE to INACTIVE pages |
| 2012 | * on this LRU, maintained by the pageout code. A zone->inactive_ratio |
| 2013 | * of 3 means 3:1 or 25% of the pages are kept on the inactive list. |
| 2014 | * |
| 2015 | * total target max |
| 2016 | * memory ratio inactive |
| 2017 | * ------------------------------------- |
| 2018 | * 10MB 1 5MB |
| 2019 | * 100MB 1 50MB |
| 2020 | * 1GB 3 250MB |
| 2021 | * 10GB 10 0.9GB |
| 2022 | * 100GB 31 3GB |
| 2023 | * 1TB 101 10GB |
| 2024 | * 10TB 320 32GB |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 2025 | */ |
Mel Gorman | f8d1a31 | 2016-07-28 15:47:34 -0700 | [diff] [blame] | 2026 | static bool inactive_list_is_low(struct lruvec *lruvec, bool file, |
Johannes Weiner | 2a2e488 | 2017-05-03 14:55:03 -0700 | [diff] [blame] | 2027 | struct mem_cgroup *memcg, |
| 2028 | struct scan_control *sc, bool actual_reclaim) |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 2029 | { |
Michal Hocko | fd53880 | 2017-02-22 15:45:58 -0800 | [diff] [blame] | 2030 | enum lru_list active_lru = file * LRU_FILE + LRU_ACTIVE; |
Johannes Weiner | 2a2e488 | 2017-05-03 14:55:03 -0700 | [diff] [blame] | 2031 | struct pglist_data *pgdat = lruvec_pgdat(lruvec); |
| 2032 | enum lru_list inactive_lru = file * LRU_FILE; |
| 2033 | unsigned long inactive, active; |
| 2034 | unsigned long inactive_ratio; |
| 2035 | unsigned long refaults; |
Rik van Riel | 59dc76b | 2016-05-20 16:56:31 -0700 | [diff] [blame] | 2036 | unsigned long gb; |
| 2037 | |
Minchan Kim | 74e3f3c | 2010-10-26 14:21:31 -0700 | [diff] [blame] | 2038 | /* |
| 2039 | * If we don't have swap space, anonymous page deactivation |
| 2040 | * is pointless. |
| 2041 | */ |
Rik van Riel | 59dc76b | 2016-05-20 16:56:31 -0700 | [diff] [blame] | 2042 | if (!file && !total_swap_pages) |
Yaowei Bai | 42e2e45 | 2015-11-05 18:47:36 -0800 | [diff] [blame] | 2043 | return false; |
Minchan Kim | 74e3f3c | 2010-10-26 14:21:31 -0700 | [diff] [blame] | 2044 | |
Michal Hocko | fd53880 | 2017-02-22 15:45:58 -0800 | [diff] [blame] | 2045 | inactive = lruvec_lru_size(lruvec, inactive_lru, sc->reclaim_idx); |
| 2046 | active = lruvec_lru_size(lruvec, active_lru, sc->reclaim_idx); |
Mel Gorman | f8d1a31 | 2016-07-28 15:47:34 -0700 | [diff] [blame] | 2047 | |
Johannes Weiner | 2a2e488 | 2017-05-03 14:55:03 -0700 | [diff] [blame] | 2048 | if (memcg) |
Johannes Weiner | ccda7f4 | 2017-05-03 14:55:16 -0700 | [diff] [blame] | 2049 | refaults = memcg_page_state(memcg, WORKINGSET_ACTIVATE); |
Rik van Riel | b39415b | 2009-12-14 17:59:48 -0800 | [diff] [blame] | 2050 | else |
Johannes Weiner | 2a2e488 | 2017-05-03 14:55:03 -0700 | [diff] [blame] | 2051 | refaults = node_page_state(pgdat, WORKINGSET_ACTIVATE); |
Rik van Riel | 59dc76b | 2016-05-20 16:56:31 -0700 | [diff] [blame] | 2052 | |
Johannes Weiner | 2a2e488 | 2017-05-03 14:55:03 -0700 | [diff] [blame] | 2053 | /* |
| 2054 | * When refaults are being observed, it means a new workingset |
| 2055 | * is being established. Disable active list protection to get |
| 2056 | * rid of the stale workingset quickly. |
| 2057 | */ |
| 2058 | if (file && actual_reclaim && lruvec->refaults != refaults) { |
| 2059 | inactive_ratio = 0; |
| 2060 | } else { |
| 2061 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
| 2062 | if (gb) |
| 2063 | inactive_ratio = int_sqrt(10 * gb); |
| 2064 | else |
| 2065 | inactive_ratio = 1; |
| 2066 | } |
| 2067 | |
| 2068 | if (actual_reclaim) |
| 2069 | trace_mm_vmscan_inactive_list_is_low(pgdat->node_id, sc->reclaim_idx, |
| 2070 | lruvec_lru_size(lruvec, inactive_lru, MAX_NR_ZONES), inactive, |
| 2071 | lruvec_lru_size(lruvec, active_lru, MAX_NR_ZONES), active, |
| 2072 | inactive_ratio, file); |
Michal Hocko | fd53880 | 2017-02-22 15:45:58 -0800 | [diff] [blame] | 2073 | |
Rik van Riel | 59dc76b | 2016-05-20 16:56:31 -0700 | [diff] [blame] | 2074 | return inactive * inactive_ratio < active; |
Rik van Riel | b39415b | 2009-12-14 17:59:48 -0800 | [diff] [blame] | 2075 | } |
| 2076 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2077 | static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan, |
Johannes Weiner | 2a2e488 | 2017-05-03 14:55:03 -0700 | [diff] [blame] | 2078 | struct lruvec *lruvec, struct mem_cgroup *memcg, |
| 2079 | struct scan_control *sc) |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 2080 | { |
Rik van Riel | b39415b | 2009-12-14 17:59:48 -0800 | [diff] [blame] | 2081 | if (is_active_lru(lru)) { |
Johannes Weiner | 2a2e488 | 2017-05-03 14:55:03 -0700 | [diff] [blame] | 2082 | if (inactive_list_is_low(lruvec, is_file_lru(lru), |
| 2083 | memcg, sc, true)) |
Konstantin Khlebnikov | 1a93be0 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 2084 | shrink_active_list(nr_to_scan, lruvec, sc, lru); |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 2085 | return 0; |
| 2086 | } |
| 2087 | |
Konstantin Khlebnikov | 1a93be0 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 2088 | return shrink_inactive_list(nr_to_scan, lruvec, sc, lru); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 2089 | } |
| 2090 | |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 2091 | enum scan_balance { |
| 2092 | SCAN_EQUAL, |
| 2093 | SCAN_FRACT, |
| 2094 | SCAN_ANON, |
| 2095 | SCAN_FILE, |
| 2096 | }; |
| 2097 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2098 | /* |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2099 | * Determine how aggressively the anon and file LRU lists should be |
| 2100 | * scanned. The relative value of each set of LRU lists is determined |
| 2101 | * by looking at the fraction of the pages scanned we did rotate back |
| 2102 | * onto the active list instead of evict. |
| 2103 | * |
Wanpeng Li | be7bd59 | 2012-06-14 20:41:02 +0800 | [diff] [blame] | 2104 | * nr[0] = anon inactive pages to scan; nr[1] = anon active pages to scan |
| 2105 | * 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] | 2106 | */ |
Vladimir Davydov | 3337767 | 2016-01-20 15:02:59 -0800 | [diff] [blame] | 2107 | static void get_scan_count(struct lruvec *lruvec, struct mem_cgroup *memcg, |
Johannes Weiner | 6b4f779 | 2014-12-12 16:56:13 -0800 | [diff] [blame] | 2108 | struct scan_control *sc, unsigned long *nr, |
| 2109 | unsigned long *lru_pages) |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2110 | { |
Vladimir Davydov | 3337767 | 2016-01-20 15:02:59 -0800 | [diff] [blame] | 2111 | int swappiness = mem_cgroup_swappiness(memcg); |
Konstantin Khlebnikov | 9012637 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 2112 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 2113 | u64 fraction[2]; |
| 2114 | u64 denominator = 0; /* gcc */ |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 2115 | struct pglist_data *pgdat = lruvec_pgdat(lruvec); |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 2116 | unsigned long anon_prio, file_prio; |
| 2117 | enum scan_balance scan_balance; |
Johannes Weiner | 0bf1457 | 2014-04-08 16:04:10 -0700 | [diff] [blame] | 2118 | unsigned long anon, file; |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 2119 | unsigned long ap, fp; |
| 2120 | enum lru_list lru; |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 2121 | |
| 2122 | /* If we have no swap space, do not bother scanning anon pages. */ |
Vladimir Davydov | d8b3843 | 2016-01-20 15:03:07 -0800 | [diff] [blame] | 2123 | if (!sc->may_swap || mem_cgroup_get_nr_swap_pages(memcg) <= 0) { |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 2124 | scan_balance = SCAN_FILE; |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 2125 | goto out; |
| 2126 | } |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2127 | |
Johannes Weiner | 10316b3 | 2013-02-22 16:32:14 -0800 | [diff] [blame] | 2128 | /* |
| 2129 | * Global reclaim will swap to prevent OOM even with no |
| 2130 | * swappiness, but memcg users want to use this knob to |
| 2131 | * disable swapping for individual groups completely when |
| 2132 | * using the memory controller's swap limit feature would be |
| 2133 | * too expensive. |
| 2134 | */ |
Johannes Weiner | 0269517 | 2014-08-06 16:06:17 -0700 | [diff] [blame] | 2135 | if (!global_reclaim(sc) && !swappiness) { |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 2136 | scan_balance = SCAN_FILE; |
Johannes Weiner | 10316b3 | 2013-02-22 16:32:14 -0800 | [diff] [blame] | 2137 | goto out; |
| 2138 | } |
| 2139 | |
| 2140 | /* |
| 2141 | * Do not apply any pressure balancing cleverness when the |
| 2142 | * system is close to OOM, scan both anon and file equally |
| 2143 | * (unless the swappiness setting disagrees with swapping). |
| 2144 | */ |
Johannes Weiner | 0269517 | 2014-08-06 16:06:17 -0700 | [diff] [blame] | 2145 | if (!sc->priority && swappiness) { |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 2146 | scan_balance = SCAN_EQUAL; |
Johannes Weiner | 10316b3 | 2013-02-22 16:32:14 -0800 | [diff] [blame] | 2147 | goto out; |
| 2148 | } |
| 2149 | |
Johannes Weiner | 11d16c2 | 2013-02-22 16:32:15 -0800 | [diff] [blame] | 2150 | /* |
Johannes Weiner | 6237625 | 2014-05-06 12:50:07 -0700 | [diff] [blame] | 2151 | * Prevent the reclaimer from falling into the cache trap: as |
| 2152 | * cache pages start out inactive, every cache fault will tip |
| 2153 | * the scan balance towards the file LRU. And as the file LRU |
| 2154 | * shrinks, so does the window for rotation from references. |
| 2155 | * This means we have a runaway feedback loop where a tiny |
| 2156 | * thrashing file LRU becomes infinitely more attractive than |
| 2157 | * anon pages. Try to detect this based on file LRU size. |
| 2158 | */ |
| 2159 | if (global_reclaim(sc)) { |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 2160 | unsigned long pgdatfile; |
| 2161 | unsigned long pgdatfree; |
| 2162 | int z; |
| 2163 | unsigned long total_high_wmark = 0; |
Johannes Weiner | 6237625 | 2014-05-06 12:50:07 -0700 | [diff] [blame] | 2164 | |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 2165 | pgdatfree = sum_zone_node_page_state(pgdat->node_id, NR_FREE_PAGES); |
| 2166 | pgdatfile = node_page_state(pgdat, NR_ACTIVE_FILE) + |
| 2167 | node_page_state(pgdat, NR_INACTIVE_FILE); |
Jerome Marchand | 2ab051e | 2014-08-06 16:08:03 -0700 | [diff] [blame] | 2168 | |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 2169 | for (z = 0; z < MAX_NR_ZONES; z++) { |
| 2170 | struct zone *zone = &pgdat->node_zones[z]; |
Mel Gorman | 6aa303d | 2016-09-01 16:14:55 -0700 | [diff] [blame] | 2171 | if (!managed_zone(zone)) |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 2172 | continue; |
| 2173 | |
| 2174 | total_high_wmark += high_wmark_pages(zone); |
| 2175 | } |
| 2176 | |
| 2177 | if (unlikely(pgdatfile + pgdatfree <= total_high_wmark)) { |
Johannes Weiner | 6237625 | 2014-05-06 12:50:07 -0700 | [diff] [blame] | 2178 | scan_balance = SCAN_ANON; |
| 2179 | goto out; |
| 2180 | } |
| 2181 | } |
| 2182 | |
| 2183 | /* |
Vladimir Davydov | 316bda0 | 2016-01-14 15:19:38 -0800 | [diff] [blame] | 2184 | * If there is enough inactive page cache, i.e. if the size of the |
| 2185 | * inactive list is greater than that of the active list *and* the |
| 2186 | * inactive list actually has some pages to scan on this priority, we |
| 2187 | * do not reclaim anything from the anonymous working set right now. |
| 2188 | * Without the second condition we could end up never scanning an |
| 2189 | * lruvec even if it has plenty of old anonymous pages unless the |
| 2190 | * system is under heavy pressure. |
Johannes Weiner | 7c5bd70 | 2013-02-22 16:32:10 -0800 | [diff] [blame] | 2191 | */ |
Johannes Weiner | 2a2e488 | 2017-05-03 14:55:03 -0700 | [diff] [blame] | 2192 | if (!inactive_list_is_low(lruvec, true, memcg, sc, false) && |
Michal Hocko | 71ab6cf | 2017-02-22 15:46:01 -0800 | [diff] [blame] | 2193 | lruvec_lru_size(lruvec, LRU_INACTIVE_FILE, sc->reclaim_idx) >> sc->priority) { |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 2194 | scan_balance = SCAN_FILE; |
Johannes Weiner | 7c5bd70 | 2013-02-22 16:32:10 -0800 | [diff] [blame] | 2195 | goto out; |
| 2196 | } |
| 2197 | |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 2198 | scan_balance = SCAN_FRACT; |
| 2199 | |
Johannes Weiner | 7c5bd70 | 2013-02-22 16:32:10 -0800 | [diff] [blame] | 2200 | /* |
KOSAKI Motohiro | 58c37f6 | 2010-08-09 17:19:51 -0700 | [diff] [blame] | 2201 | * With swappiness at 100, anonymous and file have the same priority. |
| 2202 | * This scanning priority is essentially the inverse of IO cost. |
| 2203 | */ |
Johannes Weiner | 0269517 | 2014-08-06 16:06:17 -0700 | [diff] [blame] | 2204 | anon_prio = swappiness; |
Hugh Dickins | 75b00af | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 2205 | file_prio = 200 - anon_prio; |
KOSAKI Motohiro | 58c37f6 | 2010-08-09 17:19:51 -0700 | [diff] [blame] | 2206 | |
| 2207 | /* |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2208 | * OK, so we have swap space and a fair amount of page cache |
| 2209 | * pages. We use the recently rotated / recently scanned |
| 2210 | * ratios to determine how valuable each cache is. |
| 2211 | * |
| 2212 | * Because workloads change over time (and to avoid overflow) |
| 2213 | * we keep these statistics as a floating average, which ends |
| 2214 | * up weighing recent references more than old ones. |
| 2215 | * |
| 2216 | * anon in [0], file in [1] |
| 2217 | */ |
Jerome Marchand | 2ab051e | 2014-08-06 16:08:03 -0700 | [diff] [blame] | 2218 | |
Michal Hocko | fd53880 | 2017-02-22 15:45:58 -0800 | [diff] [blame] | 2219 | anon = lruvec_lru_size(lruvec, LRU_ACTIVE_ANON, MAX_NR_ZONES) + |
| 2220 | lruvec_lru_size(lruvec, LRU_INACTIVE_ANON, MAX_NR_ZONES); |
| 2221 | file = lruvec_lru_size(lruvec, LRU_ACTIVE_FILE, MAX_NR_ZONES) + |
| 2222 | lruvec_lru_size(lruvec, LRU_INACTIVE_FILE, MAX_NR_ZONES); |
Jerome Marchand | 2ab051e | 2014-08-06 16:08:03 -0700 | [diff] [blame] | 2223 | |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 2224 | spin_lock_irq(&pgdat->lru_lock); |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 2225 | if (unlikely(reclaim_stat->recent_scanned[0] > anon / 4)) { |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 2226 | reclaim_stat->recent_scanned[0] /= 2; |
| 2227 | reclaim_stat->recent_rotated[0] /= 2; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2228 | } |
| 2229 | |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 2230 | if (unlikely(reclaim_stat->recent_scanned[1] > file / 4)) { |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 2231 | reclaim_stat->recent_scanned[1] /= 2; |
| 2232 | reclaim_stat->recent_rotated[1] /= 2; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2233 | } |
| 2234 | |
| 2235 | /* |
Rik van Riel | 00d8089 | 2008-11-19 15:36:44 -0800 | [diff] [blame] | 2236 | * The amount of pressure on anon vs file pages is inversely |
| 2237 | * proportional to the fraction of recently scanned pages on |
| 2238 | * each list that were recently referenced and in active use. |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2239 | */ |
Satoru Moriya | fe35004 | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 2240 | ap = anon_prio * (reclaim_stat->recent_scanned[0] + 1); |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 2241 | ap /= reclaim_stat->recent_rotated[0] + 1; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2242 | |
Satoru Moriya | fe35004 | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 2243 | fp = file_prio * (reclaim_stat->recent_scanned[1] + 1); |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 2244 | fp /= reclaim_stat->recent_rotated[1] + 1; |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 2245 | spin_unlock_irq(&pgdat->lru_lock); |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2246 | |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 2247 | fraction[0] = ap; |
| 2248 | fraction[1] = fp; |
| 2249 | denominator = ap + fp + 1; |
| 2250 | out: |
Johannes Weiner | 688035f | 2017-05-03 14:52:07 -0700 | [diff] [blame] | 2251 | *lru_pages = 0; |
| 2252 | for_each_evictable_lru(lru) { |
| 2253 | int file = is_file_lru(lru); |
| 2254 | unsigned long size; |
| 2255 | unsigned long scan; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2256 | |
Johannes Weiner | 688035f | 2017-05-03 14:52:07 -0700 | [diff] [blame] | 2257 | size = lruvec_lru_size(lruvec, lru, sc->reclaim_idx); |
| 2258 | scan = size >> sc->priority; |
| 2259 | /* |
| 2260 | * If the cgroup's already been deleted, make sure to |
| 2261 | * scrape out the remaining cache. |
| 2262 | */ |
| 2263 | if (!scan && !mem_cgroup_online(memcg)) |
| 2264 | scan = min(size, SWAP_CLUSTER_MAX); |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 2265 | |
Johannes Weiner | 688035f | 2017-05-03 14:52:07 -0700 | [diff] [blame] | 2266 | switch (scan_balance) { |
| 2267 | case SCAN_EQUAL: |
| 2268 | /* Scan lists relative to size */ |
| 2269 | break; |
| 2270 | case SCAN_FRACT: |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 2271 | /* |
Johannes Weiner | 688035f | 2017-05-03 14:52:07 -0700 | [diff] [blame] | 2272 | * Scan types proportional to swappiness and |
| 2273 | * their relative recent reclaim efficiency. |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 2274 | */ |
Johannes Weiner | 688035f | 2017-05-03 14:52:07 -0700 | [diff] [blame] | 2275 | scan = div64_u64(scan * fraction[file], |
| 2276 | denominator); |
| 2277 | break; |
| 2278 | case SCAN_FILE: |
| 2279 | case SCAN_ANON: |
| 2280 | /* Scan one type exclusively */ |
| 2281 | if ((scan_balance == SCAN_FILE) != file) { |
| 2282 | size = 0; |
| 2283 | scan = 0; |
| 2284 | } |
| 2285 | break; |
| 2286 | default: |
| 2287 | /* Look ma, no brain */ |
| 2288 | BUG(); |
Johannes Weiner | 9a26511 | 2013-02-22 16:32:17 -0800 | [diff] [blame] | 2289 | } |
Johannes Weiner | 688035f | 2017-05-03 14:52:07 -0700 | [diff] [blame] | 2290 | |
| 2291 | *lru_pages += size; |
| 2292 | nr[lru] = scan; |
Shaohua Li | 76a33fc | 2010-05-24 14:32:36 -0700 | [diff] [blame] | 2293 | } |
Wu Fengguang | 6e08a36 | 2009-06-16 15:32:29 -0700 | [diff] [blame] | 2294 | } |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2295 | |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2296 | /* |
Mel Gorman | a9dd0a8 | 2016-07-28 15:46:02 -0700 | [diff] [blame] | 2297 | * This is a basic per-node page freer. Used by both kswapd and direct reclaim. |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2298 | */ |
Mel Gorman | a9dd0a8 | 2016-07-28 15:46:02 -0700 | [diff] [blame] | 2299 | static void shrink_node_memcg(struct pglist_data *pgdat, struct mem_cgroup *memcg, |
Vladimir Davydov | 3337767 | 2016-01-20 15:02:59 -0800 | [diff] [blame] | 2300 | struct scan_control *sc, unsigned long *lru_pages) |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2301 | { |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 2302 | struct lruvec *lruvec = mem_cgroup_lruvec(pgdat, memcg); |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2303 | unsigned long nr[NR_LRU_LISTS]; |
Mel Gorman | e82e056 | 2013-07-03 15:01:44 -0700 | [diff] [blame] | 2304 | unsigned long targets[NR_LRU_LISTS]; |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2305 | unsigned long nr_to_scan; |
| 2306 | enum lru_list lru; |
| 2307 | unsigned long nr_reclaimed = 0; |
| 2308 | unsigned long nr_to_reclaim = sc->nr_to_reclaim; |
| 2309 | struct blk_plug plug; |
Mel Gorman | 1a50190 | 2014-06-04 16:10:49 -0700 | [diff] [blame] | 2310 | bool scan_adjusted; |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2311 | |
Vladimir Davydov | 3337767 | 2016-01-20 15:02:59 -0800 | [diff] [blame] | 2312 | get_scan_count(lruvec, memcg, sc, nr, lru_pages); |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2313 | |
Mel Gorman | e82e056 | 2013-07-03 15:01:44 -0700 | [diff] [blame] | 2314 | /* Record the original scan target for proportional adjustments later */ |
| 2315 | memcpy(targets, nr, sizeof(nr)); |
| 2316 | |
Mel Gorman | 1a50190 | 2014-06-04 16:10:49 -0700 | [diff] [blame] | 2317 | /* |
| 2318 | * Global reclaiming within direct reclaim at DEF_PRIORITY is a normal |
| 2319 | * event that can occur when there is little memory pressure e.g. |
| 2320 | * multiple streaming readers/writers. Hence, we do not abort scanning |
| 2321 | * when the requested number of pages are reclaimed when scanning at |
| 2322 | * DEF_PRIORITY on the assumption that the fact we are direct |
| 2323 | * reclaiming implies that kswapd is not keeping up and it is best to |
| 2324 | * do a batch of work at once. For memcg reclaim one check is made to |
| 2325 | * abort proportional reclaim if either the file or anon lru has already |
| 2326 | * dropped to zero at the first pass. |
| 2327 | */ |
| 2328 | scan_adjusted = (global_reclaim(sc) && !current_is_kswapd() && |
| 2329 | sc->priority == DEF_PRIORITY); |
| 2330 | |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2331 | blk_start_plug(&plug); |
| 2332 | while (nr[LRU_INACTIVE_ANON] || nr[LRU_ACTIVE_FILE] || |
| 2333 | nr[LRU_INACTIVE_FILE]) { |
Mel Gorman | e82e056 | 2013-07-03 15:01:44 -0700 | [diff] [blame] | 2334 | unsigned long nr_anon, nr_file, percentage; |
| 2335 | unsigned long nr_scanned; |
| 2336 | |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2337 | for_each_evictable_lru(lru) { |
| 2338 | if (nr[lru]) { |
| 2339 | nr_to_scan = min(nr[lru], SWAP_CLUSTER_MAX); |
| 2340 | nr[lru] -= nr_to_scan; |
| 2341 | |
| 2342 | nr_reclaimed += shrink_list(lru, nr_to_scan, |
Johannes Weiner | 2a2e488 | 2017-05-03 14:55:03 -0700 | [diff] [blame] | 2343 | lruvec, memcg, sc); |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2344 | } |
| 2345 | } |
Mel Gorman | e82e056 | 2013-07-03 15:01:44 -0700 | [diff] [blame] | 2346 | |
Michal Hocko | bd04173 | 2016-12-02 17:26:48 -0800 | [diff] [blame] | 2347 | cond_resched(); |
| 2348 | |
Mel Gorman | e82e056 | 2013-07-03 15:01:44 -0700 | [diff] [blame] | 2349 | if (nr_reclaimed < nr_to_reclaim || scan_adjusted) |
| 2350 | continue; |
| 2351 | |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2352 | /* |
Mel Gorman | e82e056 | 2013-07-03 15:01:44 -0700 | [diff] [blame] | 2353 | * For kswapd and memcg, reclaim at least the number of pages |
Mel Gorman | 1a50190 | 2014-06-04 16:10:49 -0700 | [diff] [blame] | 2354 | * requested. Ensure that the anon and file LRUs are scanned |
Mel Gorman | e82e056 | 2013-07-03 15:01:44 -0700 | [diff] [blame] | 2355 | * proportionally what was requested by get_scan_count(). We |
| 2356 | * stop reclaiming one LRU and reduce the amount scanning |
| 2357 | * proportional to the original scan target. |
| 2358 | */ |
| 2359 | nr_file = nr[LRU_INACTIVE_FILE] + nr[LRU_ACTIVE_FILE]; |
| 2360 | nr_anon = nr[LRU_INACTIVE_ANON] + nr[LRU_ACTIVE_ANON]; |
| 2361 | |
Mel Gorman | 1a50190 | 2014-06-04 16:10:49 -0700 | [diff] [blame] | 2362 | /* |
| 2363 | * It's just vindictive to attack the larger once the smaller |
| 2364 | * has gone to zero. And given the way we stop scanning the |
| 2365 | * smaller below, this makes sure that we only make one nudge |
| 2366 | * towards proportionality once we've got nr_to_reclaim. |
| 2367 | */ |
| 2368 | if (!nr_file || !nr_anon) |
| 2369 | break; |
| 2370 | |
Mel Gorman | e82e056 | 2013-07-03 15:01:44 -0700 | [diff] [blame] | 2371 | if (nr_file > nr_anon) { |
| 2372 | unsigned long scan_target = targets[LRU_INACTIVE_ANON] + |
| 2373 | targets[LRU_ACTIVE_ANON] + 1; |
| 2374 | lru = LRU_BASE; |
| 2375 | percentage = nr_anon * 100 / scan_target; |
| 2376 | } else { |
| 2377 | unsigned long scan_target = targets[LRU_INACTIVE_FILE] + |
| 2378 | targets[LRU_ACTIVE_FILE] + 1; |
| 2379 | lru = LRU_FILE; |
| 2380 | percentage = nr_file * 100 / scan_target; |
| 2381 | } |
| 2382 | |
| 2383 | /* Stop scanning the smaller of the LRU */ |
| 2384 | nr[lru] = 0; |
| 2385 | nr[lru + LRU_ACTIVE] = 0; |
| 2386 | |
| 2387 | /* |
| 2388 | * Recalculate the other LRU scan count based on its original |
| 2389 | * scan target and the percentage scanning already complete |
| 2390 | */ |
| 2391 | lru = (lru == LRU_FILE) ? LRU_BASE : LRU_FILE; |
| 2392 | nr_scanned = targets[lru] - nr[lru]; |
| 2393 | nr[lru] = targets[lru] * (100 - percentage) / 100; |
| 2394 | nr[lru] -= min(nr[lru], nr_scanned); |
| 2395 | |
| 2396 | lru += LRU_ACTIVE; |
| 2397 | nr_scanned = targets[lru] - nr[lru]; |
| 2398 | nr[lru] = targets[lru] * (100 - percentage) / 100; |
| 2399 | nr[lru] -= min(nr[lru], nr_scanned); |
| 2400 | |
| 2401 | scan_adjusted = true; |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2402 | } |
| 2403 | blk_finish_plug(&plug); |
| 2404 | sc->nr_reclaimed += nr_reclaimed; |
| 2405 | |
| 2406 | /* |
| 2407 | * Even if we did not try to evict anon pages at all, we want to |
| 2408 | * rebalance the anon lru active/inactive ratio. |
| 2409 | */ |
Johannes Weiner | 2a2e488 | 2017-05-03 14:55:03 -0700 | [diff] [blame] | 2410 | if (inactive_list_is_low(lruvec, false, memcg, sc, true)) |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2411 | shrink_active_list(SWAP_CLUSTER_MAX, lruvec, |
| 2412 | sc, LRU_ACTIVE_ANON); |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2413 | } |
| 2414 | |
Mel Gorman | 23b9da5 | 2012-05-29 15:06:20 -0700 | [diff] [blame] | 2415 | /* Use reclaim/compaction for costly allocs or under memory pressure */ |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2416 | static bool in_reclaim_compaction(struct scan_control *sc) |
Mel Gorman | 23b9da5 | 2012-05-29 15:06:20 -0700 | [diff] [blame] | 2417 | { |
Kirill A. Shutemov | d84da3f | 2012-12-11 16:00:31 -0800 | [diff] [blame] | 2418 | if (IS_ENABLED(CONFIG_COMPACTION) && sc->order && |
Mel Gorman | 23b9da5 | 2012-05-29 15:06:20 -0700 | [diff] [blame] | 2419 | (sc->order > PAGE_ALLOC_COSTLY_ORDER || |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2420 | sc->priority < DEF_PRIORITY - 2)) |
Mel Gorman | 23b9da5 | 2012-05-29 15:06:20 -0700 | [diff] [blame] | 2421 | return true; |
| 2422 | |
| 2423 | return false; |
| 2424 | } |
| 2425 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2426 | /* |
Mel Gorman | 23b9da5 | 2012-05-29 15:06:20 -0700 | [diff] [blame] | 2427 | * Reclaim/compaction is used for high-order allocation requests. It reclaims |
| 2428 | * order-0 pages before compacting the zone. should_continue_reclaim() returns |
| 2429 | * true if more pages should be reclaimed such that when the page allocator |
| 2430 | * calls try_to_compact_zone() that it will have enough free pages to succeed. |
| 2431 | * 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] | 2432 | */ |
Mel Gorman | a9dd0a8 | 2016-07-28 15:46:02 -0700 | [diff] [blame] | 2433 | static inline bool should_continue_reclaim(struct pglist_data *pgdat, |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 2434 | unsigned long nr_reclaimed, |
| 2435 | unsigned long nr_scanned, |
| 2436 | struct scan_control *sc) |
| 2437 | { |
| 2438 | unsigned long pages_for_compaction; |
| 2439 | unsigned long inactive_lru_pages; |
Mel Gorman | a9dd0a8 | 2016-07-28 15:46:02 -0700 | [diff] [blame] | 2440 | int z; |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 2441 | |
| 2442 | /* If not in reclaim/compaction mode, stop */ |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2443 | if (!in_reclaim_compaction(sc)) |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 2444 | return false; |
| 2445 | |
Mel Gorman | 2876592 | 2011-02-25 14:44:20 -0800 | [diff] [blame] | 2446 | /* Consider stopping depending on scan and reclaim activity */ |
| 2447 | if (sc->gfp_mask & __GFP_REPEAT) { |
| 2448 | /* |
| 2449 | * For __GFP_REPEAT allocations, stop reclaiming if the |
| 2450 | * full LRU list has been scanned and we are still failing |
| 2451 | * to reclaim pages. This full LRU scan is potentially |
| 2452 | * expensive but a __GFP_REPEAT caller really wants to succeed |
| 2453 | */ |
| 2454 | if (!nr_reclaimed && !nr_scanned) |
| 2455 | return false; |
| 2456 | } else { |
| 2457 | /* |
| 2458 | * For non-__GFP_REPEAT allocations which can presumably |
| 2459 | * fail without consequence, stop if we failed to reclaim |
| 2460 | * any pages from the last SWAP_CLUSTER_MAX number of |
| 2461 | * pages that were scanned. This will return to the |
| 2462 | * caller faster at the risk reclaim/compaction and |
| 2463 | * the resulting allocation attempt fails |
| 2464 | */ |
| 2465 | if (!nr_reclaimed) |
| 2466 | return false; |
| 2467 | } |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 2468 | |
| 2469 | /* |
| 2470 | * If we have not reclaimed enough pages for compaction and the |
| 2471 | * inactive lists are large enough, continue reclaiming |
| 2472 | */ |
Vlastimil Babka | 9861a62 | 2016-10-07 16:57:53 -0700 | [diff] [blame] | 2473 | pages_for_compaction = compact_gap(sc->order); |
Mel Gorman | a9dd0a8 | 2016-07-28 15:46:02 -0700 | [diff] [blame] | 2474 | inactive_lru_pages = node_page_state(pgdat, NR_INACTIVE_FILE); |
Shaohua Li | ec8acf2 | 2013-02-22 16:34:38 -0800 | [diff] [blame] | 2475 | if (get_nr_swap_pages() > 0) |
Mel Gorman | a9dd0a8 | 2016-07-28 15:46:02 -0700 | [diff] [blame] | 2476 | inactive_lru_pages += node_page_state(pgdat, NR_INACTIVE_ANON); |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 2477 | if (sc->nr_reclaimed < pages_for_compaction && |
| 2478 | inactive_lru_pages > pages_for_compaction) |
| 2479 | return true; |
| 2480 | |
| 2481 | /* If compaction would go ahead or the allocation would succeed, stop */ |
Mel Gorman | a9dd0a8 | 2016-07-28 15:46:02 -0700 | [diff] [blame] | 2482 | for (z = 0; z <= sc->reclaim_idx; z++) { |
| 2483 | struct zone *zone = &pgdat->node_zones[z]; |
Mel Gorman | 6aa303d | 2016-09-01 16:14:55 -0700 | [diff] [blame] | 2484 | if (!managed_zone(zone)) |
Mel Gorman | a9dd0a8 | 2016-07-28 15:46:02 -0700 | [diff] [blame] | 2485 | continue; |
| 2486 | |
| 2487 | switch (compaction_suitable(zone, sc->order, 0, sc->reclaim_idx)) { |
Vlastimil Babka | cf37831 | 2016-10-07 16:57:41 -0700 | [diff] [blame] | 2488 | case COMPACT_SUCCESS: |
Mel Gorman | a9dd0a8 | 2016-07-28 15:46:02 -0700 | [diff] [blame] | 2489 | case COMPACT_CONTINUE: |
| 2490 | return false; |
| 2491 | default: |
| 2492 | /* check next zone */ |
| 2493 | ; |
| 2494 | } |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 2495 | } |
Mel Gorman | a9dd0a8 | 2016-07-28 15:46:02 -0700 | [diff] [blame] | 2496 | return true; |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 2497 | } |
| 2498 | |
Mel Gorman | 970a39a | 2016-07-28 15:46:35 -0700 | [diff] [blame] | 2499 | static bool shrink_node(pg_data_t *pgdat, struct scan_control *sc) |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 2500 | { |
Vladimir Davydov | cb731d6 | 2015-02-12 14:58:54 -0800 | [diff] [blame] | 2501 | struct reclaim_state *reclaim_state = current->reclaim_state; |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2502 | unsigned long nr_reclaimed, nr_scanned; |
Johannes Weiner | 2344d7e | 2014-08-06 16:06:15 -0700 | [diff] [blame] | 2503 | bool reclaimable = false; |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 2504 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 2505 | do { |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2506 | struct mem_cgroup *root = sc->target_mem_cgroup; |
| 2507 | struct mem_cgroup_reclaim_cookie reclaim = { |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 2508 | .pgdat = pgdat, |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2509 | .priority = sc->priority, |
| 2510 | }; |
Mel Gorman | a9dd0a8 | 2016-07-28 15:46:02 -0700 | [diff] [blame] | 2511 | unsigned long node_lru_pages = 0; |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 2512 | struct mem_cgroup *memcg; |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 2513 | |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2514 | nr_reclaimed = sc->nr_reclaimed; |
| 2515 | nr_scanned = sc->nr_scanned; |
Konstantin Khlebnikov | f9be23d | 2012-05-29 15:07:02 -0700 | [diff] [blame] | 2516 | |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 2517 | memcg = mem_cgroup_iter(root, NULL, &reclaim); |
| 2518 | do { |
Johannes Weiner | 6b4f779 | 2014-12-12 16:56:13 -0800 | [diff] [blame] | 2519 | unsigned long lru_pages; |
Johannes Weiner | 8e8ae64 | 2016-01-14 15:21:32 -0800 | [diff] [blame] | 2520 | unsigned long reclaimed; |
Vladimir Davydov | cb731d6 | 2015-02-12 14:58:54 -0800 | [diff] [blame] | 2521 | unsigned long scanned; |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2522 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2523 | if (mem_cgroup_low(root, memcg)) { |
Yisheng Xie | d6622f6 | 2017-05-03 14:53:57 -0700 | [diff] [blame] | 2524 | if (!sc->memcg_low_reclaim) { |
| 2525 | sc->memcg_low_skipped = 1; |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2526 | continue; |
Yisheng Xie | d6622f6 | 2017-05-03 14:53:57 -0700 | [diff] [blame] | 2527 | } |
Johannes Weiner | 31176c7 | 2017-05-03 14:55:07 -0700 | [diff] [blame] | 2528 | mem_cgroup_event(memcg, MEMCG_LOW); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2529 | } |
| 2530 | |
Johannes Weiner | 8e8ae64 | 2016-01-14 15:21:32 -0800 | [diff] [blame] | 2531 | reclaimed = sc->nr_reclaimed; |
Vladimir Davydov | cb731d6 | 2015-02-12 14:58:54 -0800 | [diff] [blame] | 2532 | scanned = sc->nr_scanned; |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2533 | |
Mel Gorman | a9dd0a8 | 2016-07-28 15:46:02 -0700 | [diff] [blame] | 2534 | shrink_node_memcg(pgdat, memcg, sc, &lru_pages); |
| 2535 | node_lru_pages += lru_pages; |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2536 | |
Vladimir Davydov | b5afba2 | 2016-08-02 14:03:04 -0700 | [diff] [blame] | 2537 | if (memcg) |
Mel Gorman | a9dd0a8 | 2016-07-28 15:46:02 -0700 | [diff] [blame] | 2538 | shrink_slab(sc->gfp_mask, pgdat->node_id, |
Vladimir Davydov | cb731d6 | 2015-02-12 14:58:54 -0800 | [diff] [blame] | 2539 | memcg, sc->nr_scanned - scanned, |
| 2540 | lru_pages); |
| 2541 | |
Johannes Weiner | 8e8ae64 | 2016-01-14 15:21:32 -0800 | [diff] [blame] | 2542 | /* Record the group's reclaim efficiency */ |
| 2543 | vmpressure(sc->gfp_mask, memcg, false, |
| 2544 | sc->nr_scanned - scanned, |
| 2545 | sc->nr_reclaimed - reclaimed); |
| 2546 | |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2547 | /* |
Michal Hocko | a394cb8 | 2013-02-22 16:32:30 -0800 | [diff] [blame] | 2548 | * Direct reclaim and kswapd have to scan all memory |
| 2549 | * cgroups to fulfill the overall scan target for the |
Mel Gorman | a9dd0a8 | 2016-07-28 15:46:02 -0700 | [diff] [blame] | 2550 | * node. |
Michal Hocko | a394cb8 | 2013-02-22 16:32:30 -0800 | [diff] [blame] | 2551 | * |
| 2552 | * Limit reclaim, on the other hand, only cares about |
| 2553 | * nr_to_reclaim pages to be reclaimed and it will |
| 2554 | * retry with decreasing priority if one round over the |
| 2555 | * whole hierarchy is not sufficient. |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2556 | */ |
Michal Hocko | a394cb8 | 2013-02-22 16:32:30 -0800 | [diff] [blame] | 2557 | if (!global_reclaim(sc) && |
| 2558 | sc->nr_reclaimed >= sc->nr_to_reclaim) { |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2559 | mem_cgroup_iter_break(root, memcg); |
| 2560 | break; |
| 2561 | } |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2562 | } while ((memcg = mem_cgroup_iter(root, memcg, &reclaim))); |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 2563 | |
Johannes Weiner | 6b4f779 | 2014-12-12 16:56:13 -0800 | [diff] [blame] | 2564 | /* |
| 2565 | * Shrink the slab caches in the same proportion that |
| 2566 | * the eligible LRU pages were scanned. |
| 2567 | */ |
Mel Gorman | b2e1875 | 2016-07-28 15:45:37 -0700 | [diff] [blame] | 2568 | if (global_reclaim(sc)) |
Mel Gorman | a9dd0a8 | 2016-07-28 15:46:02 -0700 | [diff] [blame] | 2569 | shrink_slab(sc->gfp_mask, pgdat->node_id, NULL, |
Vladimir Davydov | cb731d6 | 2015-02-12 14:58:54 -0800 | [diff] [blame] | 2570 | sc->nr_scanned - nr_scanned, |
Mel Gorman | a9dd0a8 | 2016-07-28 15:46:02 -0700 | [diff] [blame] | 2571 | node_lru_pages); |
Johannes Weiner | 6b4f779 | 2014-12-12 16:56:13 -0800 | [diff] [blame] | 2572 | |
Vladimir Davydov | cb731d6 | 2015-02-12 14:58:54 -0800 | [diff] [blame] | 2573 | if (reclaim_state) { |
| 2574 | sc->nr_reclaimed += reclaim_state->reclaimed_slab; |
| 2575 | reclaim_state->reclaimed_slab = 0; |
Johannes Weiner | 6b4f779 | 2014-12-12 16:56:13 -0800 | [diff] [blame] | 2576 | } |
| 2577 | |
Johannes Weiner | 8e8ae64 | 2016-01-14 15:21:32 -0800 | [diff] [blame] | 2578 | /* Record the subtree's reclaim efficiency */ |
| 2579 | vmpressure(sc->gfp_mask, sc->target_mem_cgroup, true, |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 2580 | sc->nr_scanned - nr_scanned, |
| 2581 | sc->nr_reclaimed - nr_reclaimed); |
| 2582 | |
Johannes Weiner | 2344d7e | 2014-08-06 16:06:15 -0700 | [diff] [blame] | 2583 | if (sc->nr_reclaimed - nr_reclaimed) |
| 2584 | reclaimable = true; |
| 2585 | |
Mel Gorman | a9dd0a8 | 2016-07-28 15:46:02 -0700 | [diff] [blame] | 2586 | } while (should_continue_reclaim(pgdat, sc->nr_reclaimed - nr_reclaimed, |
Johannes Weiner | 9b4f98c | 2013-02-22 16:32:19 -0800 | [diff] [blame] | 2587 | sc->nr_scanned - nr_scanned, sc)); |
Johannes Weiner | 2344d7e | 2014-08-06 16:06:15 -0700 | [diff] [blame] | 2588 | |
Johannes Weiner | c73322d | 2017-05-03 14:51:51 -0700 | [diff] [blame] | 2589 | /* |
| 2590 | * Kswapd gives up on balancing particular nodes after too |
| 2591 | * many failures to reclaim anything from them and goes to |
| 2592 | * sleep. On reclaim progress, reset the failure counter. A |
| 2593 | * successful direct reclaim run will revive a dormant kswapd. |
| 2594 | */ |
| 2595 | if (reclaimable) |
| 2596 | pgdat->kswapd_failures = 0; |
| 2597 | |
Johannes Weiner | 2344d7e | 2014-08-06 16:06:15 -0700 | [diff] [blame] | 2598 | return reclaimable; |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 2599 | } |
| 2600 | |
Vlastimil Babka | 53853e2 | 2014-10-09 15:27:02 -0700 | [diff] [blame] | 2601 | /* |
Vlastimil Babka | fdd4c614 | 2016-10-07 16:58:03 -0700 | [diff] [blame] | 2602 | * Returns true if compaction should go ahead for a costly-order request, or |
| 2603 | * the allocation would already succeed without compaction. Return false if we |
| 2604 | * should reclaim first. |
Vlastimil Babka | 53853e2 | 2014-10-09 15:27:02 -0700 | [diff] [blame] | 2605 | */ |
Mel Gorman | 4f58833 | 2016-07-28 15:46:38 -0700 | [diff] [blame] | 2606 | static inline bool compaction_ready(struct zone *zone, struct scan_control *sc) |
Mel Gorman | fe4b1b2 | 2012-01-12 17:19:45 -0800 | [diff] [blame] | 2607 | { |
Mel Gorman | 31483b6 | 2016-07-28 15:45:46 -0700 | [diff] [blame] | 2608 | unsigned long watermark; |
Vlastimil Babka | fdd4c614 | 2016-10-07 16:58:03 -0700 | [diff] [blame] | 2609 | enum compact_result suitable; |
Mel Gorman | fe4b1b2 | 2012-01-12 17:19:45 -0800 | [diff] [blame] | 2610 | |
Vlastimil Babka | fdd4c614 | 2016-10-07 16:58:03 -0700 | [diff] [blame] | 2611 | suitable = compaction_suitable(zone, sc->order, 0, sc->reclaim_idx); |
| 2612 | if (suitable == COMPACT_SUCCESS) |
| 2613 | /* Allocation should succeed already. Don't reclaim. */ |
| 2614 | return true; |
| 2615 | if (suitable == COMPACT_SKIPPED) |
| 2616 | /* Compaction cannot yet proceed. Do reclaim. */ |
Mel Gorman | fe4b1b2 | 2012-01-12 17:19:45 -0800 | [diff] [blame] | 2617 | return false; |
| 2618 | |
Vlastimil Babka | fdd4c614 | 2016-10-07 16:58:03 -0700 | [diff] [blame] | 2619 | /* |
| 2620 | * Compaction is already possible, but it takes time to run and there |
| 2621 | * are potentially other callers using the pages just freed. So proceed |
| 2622 | * with reclaim to make a buffer of free pages available to give |
| 2623 | * compaction a reasonable chance of completing and allocating the page. |
| 2624 | * Note that we won't actually reclaim the whole buffer in one attempt |
| 2625 | * as the target watermark in should_continue_reclaim() is lower. But if |
| 2626 | * we are already above the high+gap watermark, don't reclaim at all. |
| 2627 | */ |
| 2628 | watermark = high_wmark_pages(zone) + compact_gap(sc->order); |
| 2629 | |
| 2630 | return zone_watermark_ok_safe(zone, 0, watermark, sc->reclaim_idx); |
Mel Gorman | fe4b1b2 | 2012-01-12 17:19:45 -0800 | [diff] [blame] | 2631 | } |
| 2632 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2633 | /* |
| 2634 | * This is the direct reclaim path, for page-allocating processes. We only |
| 2635 | * try to reclaim pages from zones which will satisfy the caller's allocation |
| 2636 | * request. |
| 2637 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2638 | * If a zone is deemed to be full of pinned pages then just give it a light |
| 2639 | * scan then give up on it. |
| 2640 | */ |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 2641 | static void shrink_zones(struct zonelist *zonelist, struct scan_control *sc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2642 | { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 2643 | struct zoneref *z; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2644 | struct zone *zone; |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2645 | unsigned long nr_soft_reclaimed; |
| 2646 | unsigned long nr_soft_scanned; |
Weijie Yang | 619d0d76 | 2014-04-07 15:36:59 -0700 | [diff] [blame] | 2647 | gfp_t orig_mask; |
Mel Gorman | 79dafcd | 2016-07-28 15:45:53 -0700 | [diff] [blame] | 2648 | pg_data_t *last_pgdat = NULL; |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 2649 | |
Mel Gorman | cc715d9 | 2012-03-21 16:34:00 -0700 | [diff] [blame] | 2650 | /* |
| 2651 | * If the number of buffer_heads in the machine exceeds the maximum |
| 2652 | * allowed level, force direct reclaim to scan the highmem zone as |
| 2653 | * highmem pages could be pinning lowmem pages storing buffer_heads |
| 2654 | */ |
Weijie Yang | 619d0d76 | 2014-04-07 15:36:59 -0700 | [diff] [blame] | 2655 | orig_mask = sc->gfp_mask; |
Mel Gorman | b2e1875 | 2016-07-28 15:45:37 -0700 | [diff] [blame] | 2656 | if (buffer_heads_over_limit) { |
Mel Gorman | cc715d9 | 2012-03-21 16:34:00 -0700 | [diff] [blame] | 2657 | sc->gfp_mask |= __GFP_HIGHMEM; |
Mel Gorman | 4f58833 | 2016-07-28 15:46:38 -0700 | [diff] [blame] | 2658 | sc->reclaim_idx = gfp_zone(sc->gfp_mask); |
Mel Gorman | b2e1875 | 2016-07-28 15:45:37 -0700 | [diff] [blame] | 2659 | } |
Mel Gorman | cc715d9 | 2012-03-21 16:34:00 -0700 | [diff] [blame] | 2660 | |
Mel Gorman | d4debc6 | 2010-08-09 17:19:29 -0700 | [diff] [blame] | 2661 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
Mel Gorman | b2e1875 | 2016-07-28 15:45:37 -0700 | [diff] [blame] | 2662 | sc->reclaim_idx, sc->nodemask) { |
Mel Gorman | b2e1875 | 2016-07-28 15:45:37 -0700 | [diff] [blame] | 2663 | /* |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 2664 | * Take care memory controller reclaiming has small influence |
| 2665 | * to global LRU. |
| 2666 | */ |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 2667 | if (global_reclaim(sc)) { |
Vladimir Davydov | 344736f | 2014-10-20 15:50:30 +0400 | [diff] [blame] | 2668 | if (!cpuset_zone_allowed(zone, |
| 2669 | GFP_KERNEL | __GFP_HARDWALL)) |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 2670 | continue; |
Vladimir Davydov | 65ec02c | 2014-04-03 14:47:20 -0700 | [diff] [blame] | 2671 | |
Johannes Weiner | 0b06496 | 2014-08-06 16:06:12 -0700 | [diff] [blame] | 2672 | /* |
| 2673 | * If we already have plenty of memory free for |
| 2674 | * compaction in this zone, don't free any more. |
| 2675 | * Even though compaction is invoked for any |
| 2676 | * non-zero order, only frequent costly order |
| 2677 | * reclamation is disruptive enough to become a |
| 2678 | * noticeable problem, like transparent huge |
| 2679 | * page allocations. |
| 2680 | */ |
| 2681 | if (IS_ENABLED(CONFIG_COMPACTION) && |
| 2682 | sc->order > PAGE_ALLOC_COSTLY_ORDER && |
Mel Gorman | 4f58833 | 2016-07-28 15:46:38 -0700 | [diff] [blame] | 2683 | compaction_ready(zone, sc)) { |
Johannes Weiner | 0b06496 | 2014-08-06 16:06:12 -0700 | [diff] [blame] | 2684 | sc->compaction_ready = true; |
| 2685 | continue; |
Rik van Riel | e0887c1 | 2011-10-31 17:09:31 -0700 | [diff] [blame] | 2686 | } |
Johannes Weiner | 0b06496 | 2014-08-06 16:06:12 -0700 | [diff] [blame] | 2687 | |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2688 | /* |
Mel Gorman | 79dafcd | 2016-07-28 15:45:53 -0700 | [diff] [blame] | 2689 | * Shrink each node in the zonelist once. If the |
| 2690 | * zonelist is ordered by zone (not the default) then a |
| 2691 | * node may be shrunk multiple times but in that case |
| 2692 | * the user prefers lower zones being preserved. |
| 2693 | */ |
| 2694 | if (zone->zone_pgdat == last_pgdat) |
| 2695 | continue; |
| 2696 | |
| 2697 | /* |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2698 | * This steals pages from memory cgroups over softlimit |
| 2699 | * and returns the number of reclaimed pages and |
| 2700 | * scanned pages. This works for global memory pressure |
| 2701 | * and balancing, not for a memcg's limit. |
| 2702 | */ |
| 2703 | nr_soft_scanned = 0; |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 2704 | nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone->zone_pgdat, |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2705 | sc->order, sc->gfp_mask, |
| 2706 | &nr_soft_scanned); |
| 2707 | sc->nr_reclaimed += nr_soft_reclaimed; |
| 2708 | sc->nr_scanned += nr_soft_scanned; |
KAMEZAWA Hiroyuki | ac34a1a | 2011-06-27 16:18:12 -0700 | [diff] [blame] | 2709 | /* need some check for avoid more shrink_zone() */ |
KAMEZAWA Hiroyuki | 1cfb419 | 2008-02-07 00:14:37 -0800 | [diff] [blame] | 2710 | } |
Nick Piggin | 408d854 | 2006-09-25 23:31:27 -0700 | [diff] [blame] | 2711 | |
Mel Gorman | 79dafcd | 2016-07-28 15:45:53 -0700 | [diff] [blame] | 2712 | /* See comment about same check for global reclaim above */ |
| 2713 | if (zone->zone_pgdat == last_pgdat) |
| 2714 | continue; |
| 2715 | last_pgdat = zone->zone_pgdat; |
Mel Gorman | 970a39a | 2016-07-28 15:46:35 -0700 | [diff] [blame] | 2716 | shrink_node(zone->zone_pgdat, sc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2717 | } |
Mel Gorman | e0c2327 | 2011-10-31 17:09:33 -0700 | [diff] [blame] | 2718 | |
Vladimir Davydov | 65ec02c | 2014-04-03 14:47:20 -0700 | [diff] [blame] | 2719 | /* |
Weijie Yang | 619d0d76 | 2014-04-07 15:36:59 -0700 | [diff] [blame] | 2720 | * Restore to original mask to avoid the impact on the caller if we |
| 2721 | * promoted it to __GFP_HIGHMEM. |
| 2722 | */ |
| 2723 | sc->gfp_mask = orig_mask; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2724 | } |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2725 | |
Johannes Weiner | 2a2e488 | 2017-05-03 14:55:03 -0700 | [diff] [blame] | 2726 | static void snapshot_refaults(struct mem_cgroup *root_memcg, pg_data_t *pgdat) |
| 2727 | { |
| 2728 | struct mem_cgroup *memcg; |
| 2729 | |
| 2730 | memcg = mem_cgroup_iter(root_memcg, NULL, NULL); |
| 2731 | do { |
| 2732 | unsigned long refaults; |
| 2733 | struct lruvec *lruvec; |
| 2734 | |
| 2735 | if (memcg) |
Johannes Weiner | ccda7f4 | 2017-05-03 14:55:16 -0700 | [diff] [blame] | 2736 | refaults = memcg_page_state(memcg, WORKINGSET_ACTIVATE); |
Johannes Weiner | 2a2e488 | 2017-05-03 14:55:03 -0700 | [diff] [blame] | 2737 | else |
| 2738 | refaults = node_page_state(pgdat, WORKINGSET_ACTIVATE); |
| 2739 | |
| 2740 | lruvec = mem_cgroup_lruvec(pgdat, memcg); |
| 2741 | lruvec->refaults = refaults; |
| 2742 | } while ((memcg = mem_cgroup_iter(root_memcg, memcg, NULL))); |
| 2743 | } |
| 2744 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2745 | /* |
| 2746 | * This is the main entry point to direct page reclaim. |
| 2747 | * |
| 2748 | * If a full scan of the inactive list fails to free enough memory then we |
| 2749 | * are "out of memory" and something needs to be killed. |
| 2750 | * |
| 2751 | * If the caller is !__GFP_FS then the probability of a failure is reasonably |
| 2752 | * 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] | 2753 | * caller can't do much about. We kick the writeback threads and take explicit |
| 2754 | * naps in the hope that some of these pages can be written. But if the |
| 2755 | * allocating task holds filesystem locks which prevent writeout this might not |
| 2756 | * work, and the allocation attempt will fail. |
Nishanth Aravamudan | a41f24e | 2008-04-29 00:58:25 -0700 | [diff] [blame] | 2757 | * |
| 2758 | * returns: 0, if no pages reclaimed |
| 2759 | * else, the number of pages reclaimed |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2760 | */ |
Mel Gorman | dac1d27 | 2008-04-28 02:12:12 -0700 | [diff] [blame] | 2761 | static unsigned long do_try_to_free_pages(struct zonelist *zonelist, |
Vladimir Davydov | 3115cd9 | 2014-04-03 14:47:22 -0700 | [diff] [blame] | 2762 | struct scan_control *sc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2763 | { |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2764 | int initial_priority = sc->priority; |
Johannes Weiner | 2a2e488 | 2017-05-03 14:55:03 -0700 | [diff] [blame] | 2765 | pg_data_t *last_pgdat; |
| 2766 | struct zoneref *z; |
| 2767 | struct zone *zone; |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2768 | retry: |
Keika Kobayashi | 873b477 | 2008-07-25 01:48:52 -0700 | [diff] [blame] | 2769 | delayacct_freepages_start(); |
| 2770 | |
Johannes Weiner | 89b5fae | 2012-01-12 17:17:50 -0800 | [diff] [blame] | 2771 | if (global_reclaim(sc)) |
Mel Gorman | 7cc30fc | 2016-07-28 15:46:59 -0700 | [diff] [blame] | 2772 | __count_zid_vm_events(ALLOCSTALL, sc->reclaim_idx, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2773 | |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 2774 | do { |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 2775 | vmpressure_prio(sc->gfp_mask, sc->target_mem_cgroup, |
| 2776 | sc->priority); |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2777 | sc->nr_scanned = 0; |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 2778 | shrink_zones(zonelist, sc); |
Mel Gorman | e0c2327 | 2011-10-31 17:09:33 -0700 | [diff] [blame] | 2779 | |
KOSAKI Motohiro | bb21c7c | 2010-06-04 14:15:05 -0700 | [diff] [blame] | 2780 | if (sc->nr_reclaimed >= sc->nr_to_reclaim) |
Johannes Weiner | 0b06496 | 2014-08-06 16:06:12 -0700 | [diff] [blame] | 2781 | break; |
| 2782 | |
| 2783 | if (sc->compaction_ready) |
| 2784 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2785 | |
| 2786 | /* |
Minchan Kim | 0e50ce3 | 2013-02-22 16:35:37 -0800 | [diff] [blame] | 2787 | * If we're getting trouble reclaiming, start doing |
| 2788 | * writepage even in laptop mode. |
| 2789 | */ |
| 2790 | if (sc->priority < DEF_PRIORITY - 2) |
| 2791 | sc->may_writepage = 1; |
Johannes Weiner | 0b06496 | 2014-08-06 16:06:12 -0700 | [diff] [blame] | 2792 | } while (--sc->priority >= 0); |
KOSAKI Motohiro | bb21c7c | 2010-06-04 14:15:05 -0700 | [diff] [blame] | 2793 | |
Johannes Weiner | 2a2e488 | 2017-05-03 14:55:03 -0700 | [diff] [blame] | 2794 | last_pgdat = NULL; |
| 2795 | for_each_zone_zonelist_nodemask(zone, z, zonelist, sc->reclaim_idx, |
| 2796 | sc->nodemask) { |
| 2797 | if (zone->zone_pgdat == last_pgdat) |
| 2798 | continue; |
| 2799 | last_pgdat = zone->zone_pgdat; |
| 2800 | snapshot_refaults(sc->target_mem_cgroup, zone->zone_pgdat); |
| 2801 | } |
| 2802 | |
Keika Kobayashi | 873b477 | 2008-07-25 01:48:52 -0700 | [diff] [blame] | 2803 | delayacct_freepages_end(); |
| 2804 | |
KOSAKI Motohiro | bb21c7c | 2010-06-04 14:15:05 -0700 | [diff] [blame] | 2805 | if (sc->nr_reclaimed) |
| 2806 | return sc->nr_reclaimed; |
| 2807 | |
Mel Gorman | 0cee34f | 2012-01-12 17:19:49 -0800 | [diff] [blame] | 2808 | /* Aborted reclaim to try compaction? don't OOM, then */ |
Johannes Weiner | 0b06496 | 2014-08-06 16:06:12 -0700 | [diff] [blame] | 2809 | if (sc->compaction_ready) |
Mel Gorman | 7335084 | 2012-01-12 17:19:33 -0800 | [diff] [blame] | 2810 | return 1; |
| 2811 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2812 | /* Untapped cgroup reserves? Don't OOM, retry. */ |
Yisheng Xie | d6622f6 | 2017-05-03 14:53:57 -0700 | [diff] [blame] | 2813 | if (sc->memcg_low_skipped) { |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2814 | sc->priority = initial_priority; |
Yisheng Xie | d6622f6 | 2017-05-03 14:53:57 -0700 | [diff] [blame] | 2815 | sc->memcg_low_reclaim = 1; |
| 2816 | sc->memcg_low_skipped = 0; |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2817 | goto retry; |
| 2818 | } |
| 2819 | |
KOSAKI Motohiro | bb21c7c | 2010-06-04 14:15:05 -0700 | [diff] [blame] | 2820 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2821 | } |
| 2822 | |
Johannes Weiner | c73322d | 2017-05-03 14:51:51 -0700 | [diff] [blame] | 2823 | static bool allow_direct_reclaim(pg_data_t *pgdat) |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2824 | { |
| 2825 | struct zone *zone; |
| 2826 | unsigned long pfmemalloc_reserve = 0; |
| 2827 | unsigned long free_pages = 0; |
| 2828 | int i; |
| 2829 | bool wmark_ok; |
| 2830 | |
Johannes Weiner | c73322d | 2017-05-03 14:51:51 -0700 | [diff] [blame] | 2831 | if (pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES) |
| 2832 | return true; |
| 2833 | |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2834 | for (i = 0; i <= ZONE_NORMAL; i++) { |
| 2835 | zone = &pgdat->node_zones[i]; |
Johannes Weiner | d450abd | 2017-05-03 14:51:54 -0700 | [diff] [blame] | 2836 | if (!managed_zone(zone)) |
| 2837 | continue; |
| 2838 | |
| 2839 | if (!zone_reclaimable_pages(zone)) |
Mel Gorman | 675becc | 2014-06-04 16:07:35 -0700 | [diff] [blame] | 2840 | continue; |
| 2841 | |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2842 | pfmemalloc_reserve += min_wmark_pages(zone); |
| 2843 | free_pages += zone_page_state(zone, NR_FREE_PAGES); |
| 2844 | } |
| 2845 | |
Mel Gorman | 675becc | 2014-06-04 16:07:35 -0700 | [diff] [blame] | 2846 | /* If there are no reserves (unexpected config) then do not throttle */ |
| 2847 | if (!pfmemalloc_reserve) |
| 2848 | return true; |
| 2849 | |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2850 | wmark_ok = free_pages > pfmemalloc_reserve / 2; |
| 2851 | |
| 2852 | /* kswapd must be awake if processes are being throttled */ |
| 2853 | if (!wmark_ok && waitqueue_active(&pgdat->kswapd_wait)) { |
Mel Gorman | 38087d9 | 2016-07-28 15:45:49 -0700 | [diff] [blame] | 2854 | pgdat->kswapd_classzone_idx = min(pgdat->kswapd_classzone_idx, |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2855 | (enum zone_type)ZONE_NORMAL); |
| 2856 | wake_up_interruptible(&pgdat->kswapd_wait); |
| 2857 | } |
| 2858 | |
| 2859 | return wmark_ok; |
| 2860 | } |
| 2861 | |
| 2862 | /* |
| 2863 | * Throttle direct reclaimers if backing storage is backed by the network |
| 2864 | * and the PFMEMALLOC reserve for the preferred node is getting dangerously |
| 2865 | * depleted. kswapd will continue to make progress and wake the processes |
Mel Gorman | 50694c2 | 2012-11-26 16:29:48 -0800 | [diff] [blame] | 2866 | * when the low watermark is reached. |
| 2867 | * |
| 2868 | * Returns true if a fatal signal was delivered during throttling. If this |
| 2869 | * happens, the page allocator should not consider triggering the OOM killer. |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2870 | */ |
Mel Gorman | 50694c2 | 2012-11-26 16:29:48 -0800 | [diff] [blame] | 2871 | static bool throttle_direct_reclaim(gfp_t gfp_mask, struct zonelist *zonelist, |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2872 | nodemask_t *nodemask) |
| 2873 | { |
Mel Gorman | 675becc | 2014-06-04 16:07:35 -0700 | [diff] [blame] | 2874 | struct zoneref *z; |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2875 | struct zone *zone; |
Mel Gorman | 675becc | 2014-06-04 16:07:35 -0700 | [diff] [blame] | 2876 | pg_data_t *pgdat = NULL; |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2877 | |
| 2878 | /* |
| 2879 | * Kernel threads should not be throttled as they may be indirectly |
| 2880 | * responsible for cleaning pages necessary for reclaim to make forward |
| 2881 | * progress. kjournald for example may enter direct reclaim while |
| 2882 | * committing a transaction where throttling it could forcing other |
| 2883 | * processes to block on log_wait_commit(). |
| 2884 | */ |
| 2885 | if (current->flags & PF_KTHREAD) |
Mel Gorman | 50694c2 | 2012-11-26 16:29:48 -0800 | [diff] [blame] | 2886 | goto out; |
| 2887 | |
| 2888 | /* |
| 2889 | * If a fatal signal is pending, this process should not throttle. |
| 2890 | * It should return quickly so it can exit and free its memory |
| 2891 | */ |
| 2892 | if (fatal_signal_pending(current)) |
| 2893 | goto out; |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2894 | |
Mel Gorman | 675becc | 2014-06-04 16:07:35 -0700 | [diff] [blame] | 2895 | /* |
| 2896 | * Check if the pfmemalloc reserves are ok by finding the first node |
| 2897 | * with a usable ZONE_NORMAL or lower zone. The expectation is that |
| 2898 | * GFP_KERNEL will be required for allocating network buffers when |
| 2899 | * swapping over the network so ZONE_HIGHMEM is unusable. |
| 2900 | * |
| 2901 | * Throttling is based on the first usable node and throttled processes |
| 2902 | * wait on a queue until kswapd makes progress and wakes them. There |
| 2903 | * is an affinity then between processes waking up and where reclaim |
| 2904 | * progress has been made assuming the process wakes on the same node. |
| 2905 | * More importantly, processes running on remote nodes will not compete |
| 2906 | * for remote pfmemalloc reserves and processes on different nodes |
| 2907 | * should make reasonable progress. |
| 2908 | */ |
| 2909 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
Michael S. Tsirkin | 17636fa | 2015-01-26 12:58:41 -0800 | [diff] [blame] | 2910 | gfp_zone(gfp_mask), nodemask) { |
Mel Gorman | 675becc | 2014-06-04 16:07:35 -0700 | [diff] [blame] | 2911 | if (zone_idx(zone) > ZONE_NORMAL) |
| 2912 | continue; |
| 2913 | |
| 2914 | /* Throttle based on the first usable node */ |
| 2915 | pgdat = zone->zone_pgdat; |
Johannes Weiner | c73322d | 2017-05-03 14:51:51 -0700 | [diff] [blame] | 2916 | if (allow_direct_reclaim(pgdat)) |
Mel Gorman | 675becc | 2014-06-04 16:07:35 -0700 | [diff] [blame] | 2917 | goto out; |
| 2918 | break; |
| 2919 | } |
| 2920 | |
| 2921 | /* If no zone was usable by the allocation flags then do not throttle */ |
| 2922 | if (!pgdat) |
Mel Gorman | 50694c2 | 2012-11-26 16:29:48 -0800 | [diff] [blame] | 2923 | goto out; |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2924 | |
Mel Gorman | 68243e7 | 2012-07-31 16:44:39 -0700 | [diff] [blame] | 2925 | /* Account for the throttling */ |
| 2926 | count_vm_event(PGSCAN_DIRECT_THROTTLE); |
| 2927 | |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2928 | /* |
| 2929 | * If the caller cannot enter the filesystem, it's possible that it |
| 2930 | * is due to the caller holding an FS lock or performing a journal |
| 2931 | * transaction in the case of a filesystem like ext[3|4]. In this case, |
| 2932 | * it is not safe to block on pfmemalloc_wait as kswapd could be |
| 2933 | * blocked waiting on the same lock. Instead, throttle for up to a |
| 2934 | * second before continuing. |
| 2935 | */ |
| 2936 | if (!(gfp_mask & __GFP_FS)) { |
| 2937 | wait_event_interruptible_timeout(pgdat->pfmemalloc_wait, |
Johannes Weiner | c73322d | 2017-05-03 14:51:51 -0700 | [diff] [blame] | 2938 | allow_direct_reclaim(pgdat), HZ); |
Mel Gorman | 50694c2 | 2012-11-26 16:29:48 -0800 | [diff] [blame] | 2939 | |
| 2940 | goto check_pending; |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2941 | } |
| 2942 | |
| 2943 | /* Throttle until kswapd wakes the process */ |
| 2944 | wait_event_killable(zone->zone_pgdat->pfmemalloc_wait, |
Johannes Weiner | c73322d | 2017-05-03 14:51:51 -0700 | [diff] [blame] | 2945 | allow_direct_reclaim(pgdat)); |
Mel Gorman | 50694c2 | 2012-11-26 16:29:48 -0800 | [diff] [blame] | 2946 | |
| 2947 | check_pending: |
| 2948 | if (fatal_signal_pending(current)) |
| 2949 | return true; |
| 2950 | |
| 2951 | out: |
| 2952 | return false; |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2953 | } |
| 2954 | |
Mel Gorman | dac1d27 | 2008-04-28 02:12:12 -0700 | [diff] [blame] | 2955 | unsigned long try_to_free_pages(struct zonelist *zonelist, int order, |
KAMEZAWA Hiroyuki | 327c0e9 | 2009-03-31 15:23:31 -0700 | [diff] [blame] | 2956 | gfp_t gfp_mask, nodemask_t *nodemask) |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2957 | { |
Mel Gorman | 33906bc | 2010-08-09 17:19:16 -0700 | [diff] [blame] | 2958 | unsigned long nr_reclaimed; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2959 | struct scan_control sc = { |
KOSAKI Motohiro | 22fba33 | 2009-12-14 17:59:10 -0800 | [diff] [blame] | 2960 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
Michal Hocko | 7dea19f | 2017-05-03 14:53:15 -0700 | [diff] [blame] | 2961 | .gfp_mask = (gfp_mask = current_gfp_context(gfp_mask)), |
Mel Gorman | b2e1875 | 2016-07-28 15:45:37 -0700 | [diff] [blame] | 2962 | .reclaim_idx = gfp_zone(gfp_mask), |
Johannes Weiner | ee814fe | 2014-08-06 16:06:19 -0700 | [diff] [blame] | 2963 | .order = order, |
| 2964 | .nodemask = nodemask, |
| 2965 | .priority = DEF_PRIORITY, |
| 2966 | .may_writepage = !laptop_mode, |
Johannes Weiner | a6dc60f8 | 2009-03-31 15:19:30 -0700 | [diff] [blame] | 2967 | .may_unmap = 1, |
KOSAKI Motohiro | 2e2e425 | 2009-04-21 12:24:57 -0700 | [diff] [blame] | 2968 | .may_swap = 1, |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2969 | }; |
| 2970 | |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2971 | /* |
Mel Gorman | 50694c2 | 2012-11-26 16:29:48 -0800 | [diff] [blame] | 2972 | * Do not enter reclaim if fatal signal was delivered while throttled. |
| 2973 | * 1 is returned so that the page allocator does not OOM kill at this |
| 2974 | * point. |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2975 | */ |
Mel Gorman | 50694c2 | 2012-11-26 16:29:48 -0800 | [diff] [blame] | 2976 | if (throttle_direct_reclaim(gfp_mask, zonelist, nodemask)) |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 2977 | return 1; |
| 2978 | |
Mel Gorman | 33906bc | 2010-08-09 17:19:16 -0700 | [diff] [blame] | 2979 | trace_mm_vmscan_direct_reclaim_begin(order, |
| 2980 | sc.may_writepage, |
Mel Gorman | e5146b1 | 2016-07-28 15:46:47 -0700 | [diff] [blame] | 2981 | gfp_mask, |
| 2982 | sc.reclaim_idx); |
Mel Gorman | 33906bc | 2010-08-09 17:19:16 -0700 | [diff] [blame] | 2983 | |
Vladimir Davydov | 3115cd9 | 2014-04-03 14:47:22 -0700 | [diff] [blame] | 2984 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc); |
Mel Gorman | 33906bc | 2010-08-09 17:19:16 -0700 | [diff] [blame] | 2985 | |
| 2986 | trace_mm_vmscan_direct_reclaim_end(nr_reclaimed); |
| 2987 | |
| 2988 | return nr_reclaimed; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2989 | } |
| 2990 | |
Andrew Morton | c255a45 | 2012-07-31 16:43:02 -0700 | [diff] [blame] | 2991 | #ifdef CONFIG_MEMCG |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2992 | |
Mel Gorman | a9dd0a8 | 2016-07-28 15:46:02 -0700 | [diff] [blame] | 2993 | unsigned long mem_cgroup_shrink_node(struct mem_cgroup *memcg, |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2994 | gfp_t gfp_mask, bool noswap, |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 2995 | pg_data_t *pgdat, |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 2996 | unsigned long *nr_scanned) |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 2997 | { |
| 2998 | struct scan_control sc = { |
KOSAKI Motohiro | b8f5c56 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2999 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
Johannes Weiner | ee814fe | 2014-08-06 16:06:19 -0700 | [diff] [blame] | 3000 | .target_mem_cgroup = memcg, |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3001 | .may_writepage = !laptop_mode, |
| 3002 | .may_unmap = 1, |
Mel Gorman | b2e1875 | 2016-07-28 15:45:37 -0700 | [diff] [blame] | 3003 | .reclaim_idx = MAX_NR_ZONES - 1, |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3004 | .may_swap = !noswap, |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3005 | }; |
Johannes Weiner | 6b4f779 | 2014-12-12 16:56:13 -0800 | [diff] [blame] | 3006 | unsigned long lru_pages; |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 3007 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3008 | sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | |
| 3009 | (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK); |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 3010 | |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3011 | trace_mm_vmscan_memcg_softlimit_reclaim_begin(sc.order, |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 3012 | sc.may_writepage, |
Mel Gorman | e5146b1 | 2016-07-28 15:46:47 -0700 | [diff] [blame] | 3013 | sc.gfp_mask, |
| 3014 | sc.reclaim_idx); |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 3015 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3016 | /* |
| 3017 | * NOTE: Although we can get the priority field, using it |
| 3018 | * here is not a good idea, since it limits the pages we can scan. |
Mel Gorman | a9dd0a8 | 2016-07-28 15:46:02 -0700 | [diff] [blame] | 3019 | * if we don't reclaim here, the shrink_node from balance_pgdat |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3020 | * will pick up pages from other mem cgroup's as well. We hack |
| 3021 | * the priority and make it zero. |
| 3022 | */ |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 3023 | shrink_node_memcg(pgdat, memcg, &sc, &lru_pages); |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 3024 | |
| 3025 | trace_mm_vmscan_memcg_softlimit_reclaim_end(sc.nr_reclaimed); |
| 3026 | |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 3027 | *nr_scanned = sc.nr_scanned; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3028 | return sc.nr_reclaimed; |
| 3029 | } |
| 3030 | |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 3031 | unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *memcg, |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 3032 | unsigned long nr_pages, |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3033 | gfp_t gfp_mask, |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 3034 | bool may_swap) |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 3035 | { |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3036 | struct zonelist *zonelist; |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 3037 | unsigned long nr_reclaimed; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 3038 | int nid; |
Vlastimil Babka | 499118e | 2017-05-08 15:59:50 -0700 | [diff] [blame^] | 3039 | unsigned int noreclaim_flag; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 3040 | struct scan_control sc = { |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 3041 | .nr_to_reclaim = max(nr_pages, SWAP_CLUSTER_MAX), |
Michal Hocko | 7dea19f | 2017-05-03 14:53:15 -0700 | [diff] [blame] | 3042 | .gfp_mask = (current_gfp_context(gfp_mask) & GFP_RECLAIM_MASK) | |
Johannes Weiner | ee814fe | 2014-08-06 16:06:19 -0700 | [diff] [blame] | 3043 | (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK), |
Mel Gorman | b2e1875 | 2016-07-28 15:45:37 -0700 | [diff] [blame] | 3044 | .reclaim_idx = MAX_NR_ZONES - 1, |
Johannes Weiner | ee814fe | 2014-08-06 16:06:19 -0700 | [diff] [blame] | 3045 | .target_mem_cgroup = memcg, |
| 3046 | .priority = DEF_PRIORITY, |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 3047 | .may_writepage = !laptop_mode, |
Johannes Weiner | a6dc60f8 | 2009-03-31 15:19:30 -0700 | [diff] [blame] | 3048 | .may_unmap = 1, |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 3049 | .may_swap = may_swap, |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 3050 | }; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 3051 | |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 3052 | /* |
| 3053 | * Unlike direct reclaim via alloc_pages(), memcg's reclaim doesn't |
| 3054 | * take care of from where we get pages. So the node where we start the |
| 3055 | * scan does not need to be the current node. |
| 3056 | */ |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 3057 | nid = mem_cgroup_select_victim_node(memcg); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 3058 | |
Aneesh Kumar K.V | c9634cf | 2016-10-07 16:59:12 -0700 | [diff] [blame] | 3059 | zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK]; |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 3060 | |
| 3061 | trace_mm_vmscan_memcg_reclaim_begin(0, |
| 3062 | sc.may_writepage, |
Mel Gorman | e5146b1 | 2016-07-28 15:46:47 -0700 | [diff] [blame] | 3063 | sc.gfp_mask, |
| 3064 | sc.reclaim_idx); |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 3065 | |
Vlastimil Babka | 499118e | 2017-05-08 15:59:50 -0700 | [diff] [blame^] | 3066 | noreclaim_flag = memalloc_noreclaim_save(); |
Vladimir Davydov | 3115cd9 | 2014-04-03 14:47:22 -0700 | [diff] [blame] | 3067 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc); |
Vlastimil Babka | 499118e | 2017-05-08 15:59:50 -0700 | [diff] [blame^] | 3068 | memalloc_noreclaim_restore(noreclaim_flag); |
KOSAKI Motohiro | bdce6d9 | 2010-08-09 17:19:56 -0700 | [diff] [blame] | 3069 | |
| 3070 | trace_mm_vmscan_memcg_reclaim_end(nr_reclaimed); |
| 3071 | |
| 3072 | return nr_reclaimed; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 3073 | } |
| 3074 | #endif |
| 3075 | |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3076 | static void age_active_anon(struct pglist_data *pgdat, |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 3077 | struct scan_control *sc) |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 3078 | { |
Johannes Weiner | b95a2f2 | 2012-01-12 17:18:06 -0800 | [diff] [blame] | 3079 | struct mem_cgroup *memcg; |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 3080 | |
Johannes Weiner | b95a2f2 | 2012-01-12 17:18:06 -0800 | [diff] [blame] | 3081 | if (!total_swap_pages) |
| 3082 | return; |
| 3083 | |
| 3084 | memcg = mem_cgroup_iter(NULL, NULL, NULL); |
| 3085 | do { |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 3086 | struct lruvec *lruvec = mem_cgroup_lruvec(pgdat, memcg); |
Johannes Weiner | b95a2f2 | 2012-01-12 17:18:06 -0800 | [diff] [blame] | 3087 | |
Johannes Weiner | 2a2e488 | 2017-05-03 14:55:03 -0700 | [diff] [blame] | 3088 | if (inactive_list_is_low(lruvec, false, memcg, sc, true)) |
Konstantin Khlebnikov | 1a93be0 | 2012-05-29 15:07:01 -0700 | [diff] [blame] | 3089 | shrink_active_list(SWAP_CLUSTER_MAX, lruvec, |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3090 | sc, LRU_ACTIVE_ANON); |
Johannes Weiner | b95a2f2 | 2012-01-12 17:18:06 -0800 | [diff] [blame] | 3091 | |
| 3092 | memcg = mem_cgroup_iter(NULL, memcg, NULL); |
| 3093 | } while (memcg); |
Johannes Weiner | f16015f | 2012-01-12 17:17:52 -0800 | [diff] [blame] | 3094 | } |
| 3095 | |
Mel Gorman | e716f2e | 2017-05-03 14:53:45 -0700 | [diff] [blame] | 3096 | /* |
| 3097 | * Returns true if there is an eligible zone balanced for the request order |
| 3098 | * and classzone_idx |
| 3099 | */ |
| 3100 | static bool pgdat_balanced(pg_data_t *pgdat, int order, int classzone_idx) |
Johannes Weiner | 60cefed | 2012-11-29 13:54:23 -0800 | [diff] [blame] | 3101 | { |
Mel Gorman | e716f2e | 2017-05-03 14:53:45 -0700 | [diff] [blame] | 3102 | int i; |
| 3103 | unsigned long mark = -1; |
| 3104 | struct zone *zone; |
Johannes Weiner | 60cefed | 2012-11-29 13:54:23 -0800 | [diff] [blame] | 3105 | |
Mel Gorman | e716f2e | 2017-05-03 14:53:45 -0700 | [diff] [blame] | 3106 | for (i = 0; i <= classzone_idx; i++) { |
| 3107 | zone = pgdat->node_zones + i; |
Mel Gorman | 6256c6b | 2016-07-28 15:45:56 -0700 | [diff] [blame] | 3108 | |
Mel Gorman | e716f2e | 2017-05-03 14:53:45 -0700 | [diff] [blame] | 3109 | if (!managed_zone(zone)) |
| 3110 | continue; |
| 3111 | |
| 3112 | mark = high_wmark_pages(zone); |
| 3113 | if (zone_watermark_ok_safe(zone, order, mark, classzone_idx)) |
| 3114 | return true; |
| 3115 | } |
| 3116 | |
| 3117 | /* |
| 3118 | * If a node has no populated zone within classzone_idx, it does not |
| 3119 | * need balancing by definition. This can happen if a zone-restricted |
| 3120 | * allocation tries to wake a remote kswapd. |
| 3121 | */ |
| 3122 | if (mark == -1) |
| 3123 | return true; |
| 3124 | |
| 3125 | return false; |
Johannes Weiner | 60cefed | 2012-11-29 13:54:23 -0800 | [diff] [blame] | 3126 | } |
| 3127 | |
Mel Gorman | 631b6e0 | 2017-05-03 14:53:41 -0700 | [diff] [blame] | 3128 | /* Clear pgdat state for congested, dirty or under writeback. */ |
| 3129 | static void clear_pgdat_congested(pg_data_t *pgdat) |
| 3130 | { |
| 3131 | clear_bit(PGDAT_CONGESTED, &pgdat->flags); |
| 3132 | clear_bit(PGDAT_DIRTY, &pgdat->flags); |
| 3133 | clear_bit(PGDAT_WRITEBACK, &pgdat->flags); |
| 3134 | } |
| 3135 | |
Mel Gorman | 1741c87 | 2011-01-13 15:46:21 -0800 | [diff] [blame] | 3136 | /* |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 3137 | * Prepare kswapd for sleeping. This verifies that there are no processes |
| 3138 | * waiting in throttle_direct_reclaim() and that watermarks have been met. |
| 3139 | * |
| 3140 | * Returns true if kswapd is ready to sleep |
| 3141 | */ |
Mel Gorman | d9f21d4 | 2016-07-28 15:46:41 -0700 | [diff] [blame] | 3142 | static bool prepare_kswapd_sleep(pg_data_t *pgdat, int order, int classzone_idx) |
Mel Gorman | f50de2d | 2009-12-14 17:58:53 -0800 | [diff] [blame] | 3143 | { |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 3144 | /* |
Vlastimil Babka | 9e5e366 | 2015-01-08 14:32:40 -0800 | [diff] [blame] | 3145 | * The throttled processes are normally woken up in balance_pgdat() as |
Johannes Weiner | c73322d | 2017-05-03 14:51:51 -0700 | [diff] [blame] | 3146 | * soon as allow_direct_reclaim() is true. But there is a potential |
Vlastimil Babka | 9e5e366 | 2015-01-08 14:32:40 -0800 | [diff] [blame] | 3147 | * race between when kswapd checks the watermarks and a process gets |
| 3148 | * throttled. There is also a potential race if processes get |
| 3149 | * throttled, kswapd wakes, a large process exits thereby balancing the |
| 3150 | * zones, which causes kswapd to exit balance_pgdat() before reaching |
| 3151 | * the wake up checks. If kswapd is going to sleep, no process should |
| 3152 | * be sleeping on pfmemalloc_wait, so wake them now if necessary. If |
| 3153 | * the wake up is premature, processes will wake kswapd and get |
| 3154 | * throttled again. The difference from wake ups in balance_pgdat() is |
| 3155 | * that here we are under prepare_to_wait(). |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 3156 | */ |
Vlastimil Babka | 9e5e366 | 2015-01-08 14:32:40 -0800 | [diff] [blame] | 3157 | if (waitqueue_active(&pgdat->pfmemalloc_wait)) |
| 3158 | wake_up_all(&pgdat->pfmemalloc_wait); |
Mel Gorman | f50de2d | 2009-12-14 17:58:53 -0800 | [diff] [blame] | 3159 | |
Johannes Weiner | c73322d | 2017-05-03 14:51:51 -0700 | [diff] [blame] | 3160 | /* Hopeless node, leave it to direct reclaim */ |
| 3161 | if (pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES) |
| 3162 | return true; |
| 3163 | |
Mel Gorman | e716f2e | 2017-05-03 14:53:45 -0700 | [diff] [blame] | 3164 | if (pgdat_balanced(pgdat, order, classzone_idx)) { |
| 3165 | clear_pgdat_congested(pgdat); |
| 3166 | return true; |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3167 | } |
| 3168 | |
Shantanu Goel | 333b0a4 | 2017-05-03 14:53:38 -0700 | [diff] [blame] | 3169 | return false; |
Mel Gorman | f50de2d | 2009-12-14 17:58:53 -0800 | [diff] [blame] | 3170 | } |
| 3171 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3172 | /* |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3173 | * kswapd shrinks a node of pages that are at or below the highest usable |
| 3174 | * zone that is currently unbalanced. |
Mel Gorman | b8e83b9 | 2013-07-03 15:01:45 -0700 | [diff] [blame] | 3175 | * |
| 3176 | * Returns true if kswapd scanned at least the requested number of pages to |
Mel Gorman | 283aba9 | 2013-07-03 15:01:51 -0700 | [diff] [blame] | 3177 | * reclaim or if the lack of progress was due to pages under writeback. |
| 3178 | * This is used to determine if the scanning priority needs to be raised. |
Mel Gorman | 7548536 | 2013-07-03 15:01:42 -0700 | [diff] [blame] | 3179 | */ |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3180 | static bool kswapd_shrink_node(pg_data_t *pgdat, |
Vlastimil Babka | accf624 | 2016-03-17 14:18:15 -0700 | [diff] [blame] | 3181 | struct scan_control *sc) |
Mel Gorman | 7548536 | 2013-07-03 15:01:42 -0700 | [diff] [blame] | 3182 | { |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3183 | struct zone *zone; |
| 3184 | int z; |
Mel Gorman | 7548536 | 2013-07-03 15:01:42 -0700 | [diff] [blame] | 3185 | |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3186 | /* Reclaim a number of pages proportional to the number of zones */ |
| 3187 | sc->nr_to_reclaim = 0; |
Mel Gorman | 970a39a | 2016-07-28 15:46:35 -0700 | [diff] [blame] | 3188 | for (z = 0; z <= sc->reclaim_idx; z++) { |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3189 | zone = pgdat->node_zones + z; |
Mel Gorman | 6aa303d | 2016-09-01 16:14:55 -0700 | [diff] [blame] | 3190 | if (!managed_zone(zone)) |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3191 | continue; |
Mel Gorman | 7c954f6 | 2013-07-03 15:01:54 -0700 | [diff] [blame] | 3192 | |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3193 | sc->nr_to_reclaim += max(high_wmark_pages(zone), SWAP_CLUSTER_MAX); |
Mel Gorman | 7c954f6 | 2013-07-03 15:01:54 -0700 | [diff] [blame] | 3194 | } |
| 3195 | |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3196 | /* |
| 3197 | * Historically care was taken to put equal pressure on all zones but |
| 3198 | * now pressure is applied based on node LRU order. |
| 3199 | */ |
Mel Gorman | 970a39a | 2016-07-28 15:46:35 -0700 | [diff] [blame] | 3200 | shrink_node(pgdat, sc); |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3201 | |
| 3202 | /* |
| 3203 | * Fragmentation may mean that the system cannot be rebalanced for |
| 3204 | * high-order allocations. If twice the allocation size has been |
| 3205 | * reclaimed then recheck watermarks only at order-0 to prevent |
| 3206 | * excessive reclaim. Assume that a process requested a high-order |
| 3207 | * can direct reclaim/compact. |
| 3208 | */ |
Vlastimil Babka | 9861a62 | 2016-10-07 16:57:53 -0700 | [diff] [blame] | 3209 | if (sc->order && sc->nr_reclaimed >= compact_gap(sc->order)) |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3210 | sc->order = 0; |
| 3211 | |
Mel Gorman | b8e83b9 | 2013-07-03 15:01:45 -0700 | [diff] [blame] | 3212 | return sc->nr_scanned >= sc->nr_to_reclaim; |
Mel Gorman | 7548536 | 2013-07-03 15:01:42 -0700 | [diff] [blame] | 3213 | } |
| 3214 | |
| 3215 | /* |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3216 | * For kswapd, balance_pgdat() will reclaim pages across a node from zones |
| 3217 | * that are eligible for use by the caller until at least one zone is |
| 3218 | * balanced. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3219 | * |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3220 | * Returns the order kswapd finished reclaiming at. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3221 | * |
| 3222 | * kswapd scans the zones in the highmem->normal->dma direction. It skips |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 3223 | * zones which have free_pages > high_wmark_pages(zone), but once a zone is |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3224 | * found to have free_pages <= high_wmark_pages(zone), any page is that zone |
| 3225 | * or lower is eligible for reclaim until at least one usable zone is |
| 3226 | * balanced. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3227 | */ |
Vlastimil Babka | accf624 | 2016-03-17 14:18:15 -0700 | [diff] [blame] | 3228 | static int balance_pgdat(pg_data_t *pgdat, int order, int classzone_idx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3229 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3230 | int i; |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 3231 | unsigned long nr_soft_reclaimed; |
| 3232 | unsigned long nr_soft_scanned; |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3233 | struct zone *zone; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3234 | struct scan_control sc = { |
| 3235 | .gfp_mask = GFP_KERNEL, |
Johannes Weiner | ee814fe | 2014-08-06 16:06:19 -0700 | [diff] [blame] | 3236 | .order = order, |
Mel Gorman | b8e83b9 | 2013-07-03 15:01:45 -0700 | [diff] [blame] | 3237 | .priority = DEF_PRIORITY, |
Johannes Weiner | ee814fe | 2014-08-06 16:06:19 -0700 | [diff] [blame] | 3238 | .may_writepage = !laptop_mode, |
Johannes Weiner | a6dc60f8 | 2009-03-31 15:19:30 -0700 | [diff] [blame] | 3239 | .may_unmap = 1, |
KOSAKI Motohiro | 2e2e425 | 2009-04-21 12:24:57 -0700 | [diff] [blame] | 3240 | .may_swap = 1, |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3241 | }; |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 3242 | count_vm_event(PAGEOUTRUN); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3243 | |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3244 | do { |
Johannes Weiner | c73322d | 2017-05-03 14:51:51 -0700 | [diff] [blame] | 3245 | unsigned long nr_reclaimed = sc.nr_reclaimed; |
Mel Gorman | b8e83b9 | 2013-07-03 15:01:45 -0700 | [diff] [blame] | 3246 | bool raise_priority = true; |
| 3247 | |
Mel Gorman | 84c7a77 | 2016-07-28 15:46:44 -0700 | [diff] [blame] | 3248 | sc.reclaim_idx = classzone_idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3249 | |
Mel Gorman | 86c79f6 | 2016-07-28 15:45:59 -0700 | [diff] [blame] | 3250 | /* |
Mel Gorman | 84c7a77 | 2016-07-28 15:46:44 -0700 | [diff] [blame] | 3251 | * If the number of buffer_heads exceeds the maximum allowed |
| 3252 | * then consider reclaiming from all zones. This has a dual |
| 3253 | * purpose -- on 64-bit systems it is expected that |
| 3254 | * buffer_heads are stripped during active rotation. On 32-bit |
| 3255 | * systems, highmem pages can pin lowmem memory and shrinking |
| 3256 | * buffers can relieve lowmem pressure. Reclaim may still not |
| 3257 | * go ahead if all eligible zones for the original allocation |
| 3258 | * request are balanced to avoid excessive reclaim from kswapd. |
Mel Gorman | 86c79f6 | 2016-07-28 15:45:59 -0700 | [diff] [blame] | 3259 | */ |
| 3260 | if (buffer_heads_over_limit) { |
| 3261 | for (i = MAX_NR_ZONES - 1; i >= 0; i--) { |
| 3262 | zone = pgdat->node_zones + i; |
Mel Gorman | 6aa303d | 2016-09-01 16:14:55 -0700 | [diff] [blame] | 3263 | if (!managed_zone(zone)) |
Mel Gorman | 86c79f6 | 2016-07-28 15:45:59 -0700 | [diff] [blame] | 3264 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3265 | |
Mel Gorman | 970a39a | 2016-07-28 15:46:35 -0700 | [diff] [blame] | 3266 | sc.reclaim_idx = i; |
Andrew Morton | e1dbeda | 2006-12-06 20:32:01 -0800 | [diff] [blame] | 3267 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3268 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3269 | } |
Zlatko Calusic | dafcb73 | 2013-02-22 16:32:34 -0800 | [diff] [blame] | 3270 | |
Mel Gorman | 86c79f6 | 2016-07-28 15:45:59 -0700 | [diff] [blame] | 3271 | /* |
Mel Gorman | e716f2e | 2017-05-03 14:53:45 -0700 | [diff] [blame] | 3272 | * Only reclaim if there are no eligible zones. Note that |
| 3273 | * sc.reclaim_idx is not used as buffer_heads_over_limit may |
| 3274 | * have adjusted it. |
Mel Gorman | 86c79f6 | 2016-07-28 15:45:59 -0700 | [diff] [blame] | 3275 | */ |
Mel Gorman | e716f2e | 2017-05-03 14:53:45 -0700 | [diff] [blame] | 3276 | if (pgdat_balanced(pgdat, sc.order, classzone_idx)) |
| 3277 | goto out; |
Andrew Morton | e1dbeda | 2006-12-06 20:32:01 -0800 | [diff] [blame] | 3278 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3279 | /* |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3280 | * Do some background aging of the anon list, to give |
| 3281 | * pages a chance to be referenced before reclaiming. All |
| 3282 | * pages are rotated regardless of classzone as this is |
| 3283 | * about consistent aging. |
| 3284 | */ |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 3285 | age_active_anon(pgdat, &sc); |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3286 | |
| 3287 | /* |
Mel Gorman | b7ea3c4 | 2013-07-03 15:01:53 -0700 | [diff] [blame] | 3288 | * If we're getting trouble reclaiming, start doing writepage |
| 3289 | * even in laptop mode. |
| 3290 | */ |
Johannes Weiner | 047d72c | 2017-05-03 14:51:57 -0700 | [diff] [blame] | 3291 | if (sc.priority < DEF_PRIORITY - 2) |
Mel Gorman | b7ea3c4 | 2013-07-03 15:01:53 -0700 | [diff] [blame] | 3292 | sc.may_writepage = 1; |
| 3293 | |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3294 | /* Call soft limit reclaim before calling shrink_node. */ |
| 3295 | sc.nr_scanned = 0; |
| 3296 | nr_soft_scanned = 0; |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 3297 | nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(pgdat, sc.order, |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3298 | sc.gfp_mask, &nr_soft_scanned); |
| 3299 | sc.nr_reclaimed += nr_soft_reclaimed; |
| 3300 | |
Mel Gorman | b7ea3c4 | 2013-07-03 15:01:53 -0700 | [diff] [blame] | 3301 | /* |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3302 | * There should be no need to raise the scanning priority if |
| 3303 | * enough pages are already being scanned that that high |
| 3304 | * watermark would be met at 100% efficiency. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3305 | */ |
Mel Gorman | 970a39a | 2016-07-28 15:46:35 -0700 | [diff] [blame] | 3306 | if (kswapd_shrink_node(pgdat, &sc)) |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3307 | raise_priority = false; |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 3308 | |
| 3309 | /* |
| 3310 | * If the low watermark is met there is no need for processes |
| 3311 | * to be throttled on pfmemalloc_wait as they should not be |
| 3312 | * able to safely make forward progress. Wake them |
| 3313 | */ |
| 3314 | if (waitqueue_active(&pgdat->pfmemalloc_wait) && |
Johannes Weiner | c73322d | 2017-05-03 14:51:51 -0700 | [diff] [blame] | 3315 | allow_direct_reclaim(pgdat)) |
Vlastimil Babka | cfc5115 | 2015-02-11 15:25:12 -0800 | [diff] [blame] | 3316 | wake_up_all(&pgdat->pfmemalloc_wait); |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 3317 | |
Mel Gorman | b8e83b9 | 2013-07-03 15:01:45 -0700 | [diff] [blame] | 3318 | /* Check if kswapd should be suspending */ |
| 3319 | if (try_to_freeze() || kthread_should_stop()) |
| 3320 | break; |
| 3321 | |
| 3322 | /* |
| 3323 | * Raise priority if scanning rate is too low or there was no |
| 3324 | * progress in reclaiming pages |
| 3325 | */ |
Johannes Weiner | c73322d | 2017-05-03 14:51:51 -0700 | [diff] [blame] | 3326 | nr_reclaimed = sc.nr_reclaimed - nr_reclaimed; |
| 3327 | if (raise_priority || !nr_reclaimed) |
Mel Gorman | b8e83b9 | 2013-07-03 15:01:45 -0700 | [diff] [blame] | 3328 | sc.priority--; |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3329 | } while (sc.priority >= 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3330 | |
Johannes Weiner | c73322d | 2017-05-03 14:51:51 -0700 | [diff] [blame] | 3331 | if (!sc.nr_reclaimed) |
| 3332 | pgdat->kswapd_failures++; |
| 3333 | |
Mel Gorman | b8e83b9 | 2013-07-03 15:01:45 -0700 | [diff] [blame] | 3334 | out: |
Johannes Weiner | 2a2e488 | 2017-05-03 14:55:03 -0700 | [diff] [blame] | 3335 | snapshot_refaults(NULL, pgdat); |
Mel Gorman | 0abdee2 | 2011-01-13 15:46:22 -0800 | [diff] [blame] | 3336 | /* |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3337 | * Return the order kswapd stopped reclaiming at as |
| 3338 | * prepare_kswapd_sleep() takes it into account. If another caller |
| 3339 | * entered the allocator slow path while kswapd was awake, order will |
| 3340 | * remain at the higher level. |
Mel Gorman | 0abdee2 | 2011-01-13 15:46:22 -0800 | [diff] [blame] | 3341 | */ |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3342 | return sc.order; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3343 | } |
| 3344 | |
Mel Gorman | e716f2e | 2017-05-03 14:53:45 -0700 | [diff] [blame] | 3345 | /* |
| 3346 | * pgdat->kswapd_classzone_idx is the highest zone index that a recent |
| 3347 | * allocation request woke kswapd for. When kswapd has not woken recently, |
| 3348 | * the value is MAX_NR_ZONES which is not a valid index. This compares a |
| 3349 | * given classzone and returns it or the highest classzone index kswapd |
| 3350 | * was recently woke for. |
| 3351 | */ |
| 3352 | static enum zone_type kswapd_classzone_idx(pg_data_t *pgdat, |
| 3353 | enum zone_type classzone_idx) |
| 3354 | { |
| 3355 | if (pgdat->kswapd_classzone_idx == MAX_NR_ZONES) |
| 3356 | return classzone_idx; |
| 3357 | |
| 3358 | return max(pgdat->kswapd_classzone_idx, classzone_idx); |
| 3359 | } |
| 3360 | |
Mel Gorman | 38087d9 | 2016-07-28 15:45:49 -0700 | [diff] [blame] | 3361 | static void kswapd_try_to_sleep(pg_data_t *pgdat, int alloc_order, int reclaim_order, |
| 3362 | unsigned int classzone_idx) |
KOSAKI Motohiro | f0bc0a6 | 2011-01-13 15:45:50 -0800 | [diff] [blame] | 3363 | { |
| 3364 | long remaining = 0; |
| 3365 | DEFINE_WAIT(wait); |
| 3366 | |
| 3367 | if (freezing(current) || kthread_should_stop()) |
| 3368 | return; |
| 3369 | |
| 3370 | prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); |
| 3371 | |
Shantanu Goel | 333b0a4 | 2017-05-03 14:53:38 -0700 | [diff] [blame] | 3372 | /* |
| 3373 | * Try to sleep for a short interval. Note that kcompactd will only be |
| 3374 | * woken if it is possible to sleep for a short interval. This is |
| 3375 | * deliberate on the assumption that if reclaim cannot keep an |
| 3376 | * eligible zone balanced that it's also unlikely that compaction will |
| 3377 | * succeed. |
| 3378 | */ |
Mel Gorman | d9f21d4 | 2016-07-28 15:46:41 -0700 | [diff] [blame] | 3379 | if (prepare_kswapd_sleep(pgdat, reclaim_order, classzone_idx)) { |
Vlastimil Babka | fd901c9 | 2016-04-28 16:18:49 -0700 | [diff] [blame] | 3380 | /* |
| 3381 | * Compaction records what page blocks it recently failed to |
| 3382 | * isolate pages from and skips them in the future scanning. |
| 3383 | * When kswapd is going to sleep, it is reasonable to assume |
| 3384 | * that pages and compaction may succeed so reset the cache. |
| 3385 | */ |
| 3386 | reset_isolation_suitable(pgdat); |
| 3387 | |
| 3388 | /* |
| 3389 | * We have freed the memory, now we should compact it to make |
| 3390 | * allocation of the requested order possible. |
| 3391 | */ |
Mel Gorman | 38087d9 | 2016-07-28 15:45:49 -0700 | [diff] [blame] | 3392 | wakeup_kcompactd(pgdat, alloc_order, classzone_idx); |
Vlastimil Babka | fd901c9 | 2016-04-28 16:18:49 -0700 | [diff] [blame] | 3393 | |
KOSAKI Motohiro | f0bc0a6 | 2011-01-13 15:45:50 -0800 | [diff] [blame] | 3394 | remaining = schedule_timeout(HZ/10); |
Mel Gorman | 38087d9 | 2016-07-28 15:45:49 -0700 | [diff] [blame] | 3395 | |
| 3396 | /* |
| 3397 | * If woken prematurely then reset kswapd_classzone_idx and |
| 3398 | * order. The values will either be from a wakeup request or |
| 3399 | * the previous request that slept prematurely. |
| 3400 | */ |
| 3401 | if (remaining) { |
Mel Gorman | e716f2e | 2017-05-03 14:53:45 -0700 | [diff] [blame] | 3402 | pgdat->kswapd_classzone_idx = kswapd_classzone_idx(pgdat, classzone_idx); |
Mel Gorman | 38087d9 | 2016-07-28 15:45:49 -0700 | [diff] [blame] | 3403 | pgdat->kswapd_order = max(pgdat->kswapd_order, reclaim_order); |
| 3404 | } |
| 3405 | |
KOSAKI Motohiro | f0bc0a6 | 2011-01-13 15:45:50 -0800 | [diff] [blame] | 3406 | finish_wait(&pgdat->kswapd_wait, &wait); |
| 3407 | prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); |
| 3408 | } |
| 3409 | |
| 3410 | /* |
| 3411 | * After a short sleep, check if it was a premature sleep. If not, then |
| 3412 | * go fully to sleep until explicitly woken up. |
| 3413 | */ |
Mel Gorman | d9f21d4 | 2016-07-28 15:46:41 -0700 | [diff] [blame] | 3414 | if (!remaining && |
| 3415 | prepare_kswapd_sleep(pgdat, reclaim_order, classzone_idx)) { |
KOSAKI Motohiro | f0bc0a6 | 2011-01-13 15:45:50 -0800 | [diff] [blame] | 3416 | trace_mm_vmscan_kswapd_sleep(pgdat->node_id); |
| 3417 | |
| 3418 | /* |
| 3419 | * vmstat counters are not perfectly accurate and the estimated |
| 3420 | * value for counters such as NR_FREE_PAGES can deviate from the |
| 3421 | * true value by nr_online_cpus * threshold. To avoid the zone |
| 3422 | * watermarks being breached while under pressure, we reduce the |
| 3423 | * per-cpu vmstat threshold while kswapd is awake and restore |
| 3424 | * them before going back to sleep. |
| 3425 | */ |
| 3426 | set_pgdat_percpu_threshold(pgdat, calculate_normal_threshold); |
Aaditya Kumar | 1c7e7f6 | 2012-07-17 15:48:07 -0700 | [diff] [blame] | 3427 | |
| 3428 | if (!kthread_should_stop()) |
| 3429 | schedule(); |
| 3430 | |
KOSAKI Motohiro | f0bc0a6 | 2011-01-13 15:45:50 -0800 | [diff] [blame] | 3431 | set_pgdat_percpu_threshold(pgdat, calculate_pressure_threshold); |
| 3432 | } else { |
| 3433 | if (remaining) |
| 3434 | count_vm_event(KSWAPD_LOW_WMARK_HIT_QUICKLY); |
| 3435 | else |
| 3436 | count_vm_event(KSWAPD_HIGH_WMARK_HIT_QUICKLY); |
| 3437 | } |
| 3438 | finish_wait(&pgdat->kswapd_wait, &wait); |
| 3439 | } |
| 3440 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3441 | /* |
| 3442 | * The background pageout daemon, started as a kernel thread |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 3443 | * from the init process. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3444 | * |
| 3445 | * This basically trickles out pages so that we have _some_ |
| 3446 | * free memory available even if there is no other activity |
| 3447 | * that frees anything up. This is needed for things like routing |
| 3448 | * etc, where we otherwise might have all activity going on in |
| 3449 | * asynchronous contexts that cannot page things out. |
| 3450 | * |
| 3451 | * If there are applications that are active memory-allocators |
| 3452 | * (most normal use), this basically shouldn't matter. |
| 3453 | */ |
| 3454 | static int kswapd(void *p) |
| 3455 | { |
Mel Gorman | e716f2e | 2017-05-03 14:53:45 -0700 | [diff] [blame] | 3456 | unsigned int alloc_order, reclaim_order; |
| 3457 | unsigned int classzone_idx = MAX_NR_ZONES - 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3458 | pg_data_t *pgdat = (pg_data_t*)p; |
| 3459 | struct task_struct *tsk = current; |
KOSAKI Motohiro | f0bc0a6 | 2011-01-13 15:45:50 -0800 | [diff] [blame] | 3460 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3461 | struct reclaim_state reclaim_state = { |
| 3462 | .reclaimed_slab = 0, |
| 3463 | }; |
Rusty Russell | a70f730 | 2009-03-13 14:49:46 +1030 | [diff] [blame] | 3464 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3465 | |
Nick Piggin | cf40bd1 | 2009-01-21 08:12:39 +0100 | [diff] [blame] | 3466 | lockdep_set_current_reclaim_state(GFP_KERNEL); |
| 3467 | |
Rusty Russell | 174596a | 2009-01-01 10:12:29 +1030 | [diff] [blame] | 3468 | if (!cpumask_empty(cpumask)) |
Mike Travis | c5f59f0 | 2008-04-04 18:11:10 -0700 | [diff] [blame] | 3469 | set_cpus_allowed_ptr(tsk, cpumask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3470 | current->reclaim_state = &reclaim_state; |
| 3471 | |
| 3472 | /* |
| 3473 | * Tell the memory management that we're a "memory allocator", |
| 3474 | * and that if we need more memory we should get access to it |
| 3475 | * regardless (see "__alloc_pages()"). "kswapd" should |
| 3476 | * never get caught in the normal page freeing logic. |
| 3477 | * |
| 3478 | * (Kswapd normally doesn't need memory anyway, but sometimes |
| 3479 | * you need a small amount of memory in order to be able to |
| 3480 | * page out something else, and this flag essentially protects |
| 3481 | * us from recursively trying to free more memory as we're |
| 3482 | * trying to free the first piece of memory in the first place). |
| 3483 | */ |
Christoph Lameter | 930d915 | 2006-01-08 01:00:47 -0800 | [diff] [blame] | 3484 | tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD; |
Rafael J. Wysocki | 8314418 | 2007-07-17 04:03:35 -0700 | [diff] [blame] | 3485 | set_freezable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3486 | |
Mel Gorman | e716f2e | 2017-05-03 14:53:45 -0700 | [diff] [blame] | 3487 | pgdat->kswapd_order = 0; |
| 3488 | pgdat->kswapd_classzone_idx = MAX_NR_ZONES; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3489 | for ( ; ; ) { |
Jeff Liu | 6f6313d | 2012-12-11 16:02:48 -0800 | [diff] [blame] | 3490 | bool ret; |
Christoph Lameter | 3e1d1d2 | 2005-06-24 23:13:50 -0700 | [diff] [blame] | 3491 | |
Mel Gorman | e716f2e | 2017-05-03 14:53:45 -0700 | [diff] [blame] | 3492 | alloc_order = reclaim_order = pgdat->kswapd_order; |
| 3493 | classzone_idx = kswapd_classzone_idx(pgdat, classzone_idx); |
| 3494 | |
Mel Gorman | 38087d9 | 2016-07-28 15:45:49 -0700 | [diff] [blame] | 3495 | kswapd_try_sleep: |
| 3496 | kswapd_try_to_sleep(pgdat, alloc_order, reclaim_order, |
| 3497 | classzone_idx); |
Mel Gorman | 215ddd6 | 2011-07-08 15:39:40 -0700 | [diff] [blame] | 3498 | |
Mel Gorman | 38087d9 | 2016-07-28 15:45:49 -0700 | [diff] [blame] | 3499 | /* Read the new order and classzone_idx */ |
| 3500 | alloc_order = reclaim_order = pgdat->kswapd_order; |
Mel Gorman | e716f2e | 2017-05-03 14:53:45 -0700 | [diff] [blame] | 3501 | classzone_idx = kswapd_classzone_idx(pgdat, 0); |
Mel Gorman | 38087d9 | 2016-07-28 15:45:49 -0700 | [diff] [blame] | 3502 | pgdat->kswapd_order = 0; |
Mel Gorman | e716f2e | 2017-05-03 14:53:45 -0700 | [diff] [blame] | 3503 | pgdat->kswapd_classzone_idx = MAX_NR_ZONES; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3504 | |
David Rientjes | 8fe23e0 | 2009-12-14 17:58:33 -0800 | [diff] [blame] | 3505 | ret = try_to_freeze(); |
| 3506 | if (kthread_should_stop()) |
| 3507 | break; |
| 3508 | |
| 3509 | /* |
| 3510 | * We can speed up thawing tasks if we don't call balance_pgdat |
| 3511 | * after returning from the refrigerator |
| 3512 | */ |
Mel Gorman | 38087d9 | 2016-07-28 15:45:49 -0700 | [diff] [blame] | 3513 | if (ret) |
| 3514 | continue; |
Mel Gorman | 1d82de6 | 2016-07-28 15:45:43 -0700 | [diff] [blame] | 3515 | |
Mel Gorman | 38087d9 | 2016-07-28 15:45:49 -0700 | [diff] [blame] | 3516 | /* |
| 3517 | * Reclaim begins at the requested order but if a high-order |
| 3518 | * reclaim fails then kswapd falls back to reclaiming for |
| 3519 | * order-0. If that happens, kswapd will consider sleeping |
| 3520 | * for the order it finished reclaiming at (reclaim_order) |
| 3521 | * but kcompactd is woken to compact for the original |
| 3522 | * request (alloc_order). |
| 3523 | */ |
Mel Gorman | e5146b1 | 2016-07-28 15:46:47 -0700 | [diff] [blame] | 3524 | trace_mm_vmscan_kswapd_wake(pgdat->node_id, classzone_idx, |
| 3525 | alloc_order); |
Mel Gorman | 38087d9 | 2016-07-28 15:45:49 -0700 | [diff] [blame] | 3526 | reclaim_order = balance_pgdat(pgdat, alloc_order, classzone_idx); |
| 3527 | if (reclaim_order < alloc_order) |
| 3528 | goto kswapd_try_sleep; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3529 | } |
Takamori Yamaguchi | b0a8cc5 | 2012-11-08 15:53:39 -0800 | [diff] [blame] | 3530 | |
Johannes Weiner | 71abdc1 | 2014-06-06 14:35:35 -0700 | [diff] [blame] | 3531 | tsk->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD); |
Takamori Yamaguchi | b0a8cc5 | 2012-11-08 15:53:39 -0800 | [diff] [blame] | 3532 | current->reclaim_state = NULL; |
Johannes Weiner | 71abdc1 | 2014-06-06 14:35:35 -0700 | [diff] [blame] | 3533 | lockdep_clear_current_reclaim_state(); |
| 3534 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3535 | return 0; |
| 3536 | } |
| 3537 | |
| 3538 | /* |
| 3539 | * A zone is low on free memory, so wake its kswapd task to service it. |
| 3540 | */ |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 3541 | 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] | 3542 | { |
| 3543 | pg_data_t *pgdat; |
| 3544 | |
Mel Gorman | 6aa303d | 2016-09-01 16:14:55 -0700 | [diff] [blame] | 3545 | if (!managed_zone(zone)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3546 | return; |
| 3547 | |
Vladimir Davydov | 344736f | 2014-10-20 15:50:30 +0400 | [diff] [blame] | 3548 | if (!cpuset_zone_allowed(zone, GFP_KERNEL | __GFP_HARDWALL)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3549 | return; |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 3550 | pgdat = zone->zone_pgdat; |
Mel Gorman | e716f2e | 2017-05-03 14:53:45 -0700 | [diff] [blame] | 3551 | pgdat->kswapd_classzone_idx = kswapd_classzone_idx(pgdat, |
| 3552 | classzone_idx); |
Mel Gorman | 38087d9 | 2016-07-28 15:45:49 -0700 | [diff] [blame] | 3553 | pgdat->kswapd_order = max(pgdat->kswapd_order, order); |
Con Kolivas | 8d0986e | 2005-09-13 01:25:07 -0700 | [diff] [blame] | 3554 | if (!waitqueue_active(&pgdat->kswapd_wait)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3555 | return; |
Mel Gorman | e1a5563 | 2016-07-28 15:46:26 -0700 | [diff] [blame] | 3556 | |
Johannes Weiner | c73322d | 2017-05-03 14:51:51 -0700 | [diff] [blame] | 3557 | /* Hopeless node, leave it to direct reclaim */ |
| 3558 | if (pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES) |
| 3559 | return; |
| 3560 | |
Mel Gorman | e716f2e | 2017-05-03 14:53:45 -0700 | [diff] [blame] | 3561 | if (pgdat_balanced(pgdat, order, classzone_idx)) |
| 3562 | return; |
Mel Gorman | e1a5563 | 2016-07-28 15:46:26 -0700 | [diff] [blame] | 3563 | |
Mel Gorman | e716f2e | 2017-05-03 14:53:45 -0700 | [diff] [blame] | 3564 | trace_mm_vmscan_wakeup_kswapd(pgdat->node_id, classzone_idx, order); |
Con Kolivas | 8d0986e | 2005-09-13 01:25:07 -0700 | [diff] [blame] | 3565 | wake_up_interruptible(&pgdat->kswapd_wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3566 | } |
| 3567 | |
Rafael J. Wysocki | c6f37f1 | 2009-05-24 22:16:31 +0200 | [diff] [blame] | 3568 | #ifdef CONFIG_HIBERNATION |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3569 | /* |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3570 | * 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] | 3571 | * freed pages. |
| 3572 | * |
| 3573 | * Rather than trying to age LRUs the aim is to preserve the overall |
| 3574 | * LRU order by reclaiming preferentially |
| 3575 | * inactive > active > active referenced > active mapped |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3576 | */ |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3577 | unsigned long shrink_all_memory(unsigned long nr_to_reclaim) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3578 | { |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 3579 | struct reclaim_state reclaim_state; |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 3580 | struct scan_control sc = { |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3581 | .nr_to_reclaim = nr_to_reclaim, |
Johannes Weiner | ee814fe | 2014-08-06 16:06:19 -0700 | [diff] [blame] | 3582 | .gfp_mask = GFP_HIGHUSER_MOVABLE, |
Mel Gorman | b2e1875 | 2016-07-28 15:45:37 -0700 | [diff] [blame] | 3583 | .reclaim_idx = MAX_NR_ZONES - 1, |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3584 | .priority = DEF_PRIORITY, |
Johannes Weiner | ee814fe | 2014-08-06 16:06:19 -0700 | [diff] [blame] | 3585 | .may_writepage = 1, |
| 3586 | .may_unmap = 1, |
| 3587 | .may_swap = 1, |
| 3588 | .hibernation_mode = 1, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3589 | }; |
Ying Han | a09ed5e | 2011-05-24 17:12:26 -0700 | [diff] [blame] | 3590 | struct zonelist *zonelist = node_zonelist(numa_node_id(), sc.gfp_mask); |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3591 | struct task_struct *p = current; |
| 3592 | unsigned long nr_reclaimed; |
Vlastimil Babka | 499118e | 2017-05-08 15:59:50 -0700 | [diff] [blame^] | 3593 | unsigned int noreclaim_flag; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3594 | |
Vlastimil Babka | 499118e | 2017-05-08 15:59:50 -0700 | [diff] [blame^] | 3595 | noreclaim_flag = memalloc_noreclaim_save(); |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3596 | lockdep_set_current_reclaim_state(sc.gfp_mask); |
| 3597 | reclaim_state.reclaimed_slab = 0; |
| 3598 | p->reclaim_state = &reclaim_state; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 3599 | |
Vladimir Davydov | 3115cd9 | 2014-04-03 14:47:22 -0700 | [diff] [blame] | 3600 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc); |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 3601 | |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3602 | p->reclaim_state = NULL; |
| 3603 | lockdep_clear_current_reclaim_state(); |
Vlastimil Babka | 499118e | 2017-05-08 15:59:50 -0700 | [diff] [blame^] | 3604 | memalloc_noreclaim_restore(noreclaim_flag); |
Rafael J. Wysocki | d6277db | 2006-06-23 02:03:18 -0700 | [diff] [blame] | 3605 | |
KOSAKI Motohiro | 7b51755 | 2009-12-14 17:59:12 -0800 | [diff] [blame] | 3606 | return nr_reclaimed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3607 | } |
Rafael J. Wysocki | c6f37f1 | 2009-05-24 22:16:31 +0200 | [diff] [blame] | 3608 | #endif /* CONFIG_HIBERNATION */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3609 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3610 | /* It's optimal to keep kswapds on the same CPUs as their memory, but |
| 3611 | not required for correctness. So if the last cpu in a node goes |
| 3612 | away, we get changed to run anywhere: as the first one comes back, |
| 3613 | restore their cpu bindings. */ |
Sebastian Andrzej Siewior | 517bbed | 2016-11-03 15:50:03 +0100 | [diff] [blame] | 3614 | static int kswapd_cpu_online(unsigned int cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3615 | { |
Yasunori Goto | 58c0a4a | 2007-10-16 01:25:40 -0700 | [diff] [blame] | 3616 | int nid; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3617 | |
Sebastian Andrzej Siewior | 517bbed | 2016-11-03 15:50:03 +0100 | [diff] [blame] | 3618 | for_each_node_state(nid, N_MEMORY) { |
| 3619 | pg_data_t *pgdat = NODE_DATA(nid); |
| 3620 | const struct cpumask *mask; |
Rusty Russell | a70f730 | 2009-03-13 14:49:46 +1030 | [diff] [blame] | 3621 | |
Sebastian Andrzej Siewior | 517bbed | 2016-11-03 15:50:03 +0100 | [diff] [blame] | 3622 | mask = cpumask_of_node(pgdat->node_id); |
Mike Travis | c5f59f0 | 2008-04-04 18:11:10 -0700 | [diff] [blame] | 3623 | |
Sebastian Andrzej Siewior | 517bbed | 2016-11-03 15:50:03 +0100 | [diff] [blame] | 3624 | if (cpumask_any_and(cpu_online_mask, mask) < nr_cpu_ids) |
| 3625 | /* One of our CPUs online: restore mask */ |
| 3626 | set_cpus_allowed_ptr(pgdat->kswapd, mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3627 | } |
Sebastian Andrzej Siewior | 517bbed | 2016-11-03 15:50:03 +0100 | [diff] [blame] | 3628 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3629 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3630 | |
Yasunori Goto | 3218ae1 | 2006-06-27 02:53:33 -0700 | [diff] [blame] | 3631 | /* |
| 3632 | * This kswapd start function will be called by init and node-hot-add. |
| 3633 | * On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added. |
| 3634 | */ |
| 3635 | int kswapd_run(int nid) |
| 3636 | { |
| 3637 | pg_data_t *pgdat = NODE_DATA(nid); |
| 3638 | int ret = 0; |
| 3639 | |
| 3640 | if (pgdat->kswapd) |
| 3641 | return 0; |
| 3642 | |
| 3643 | pgdat->kswapd = kthread_run(kswapd, pgdat, "kswapd%d", nid); |
| 3644 | if (IS_ERR(pgdat->kswapd)) { |
| 3645 | /* failure at boot is fatal */ |
| 3646 | BUG_ON(system_state == SYSTEM_BOOTING); |
Gavin Shan | d5dc0ad | 2012-10-08 16:29:27 -0700 | [diff] [blame] | 3647 | pr_err("Failed to start kswapd on node %d\n", nid); |
| 3648 | ret = PTR_ERR(pgdat->kswapd); |
Xishi Qiu | d72515b | 2013-04-17 15:58:34 -0700 | [diff] [blame] | 3649 | pgdat->kswapd = NULL; |
Yasunori Goto | 3218ae1 | 2006-06-27 02:53:33 -0700 | [diff] [blame] | 3650 | } |
| 3651 | return ret; |
| 3652 | } |
| 3653 | |
David Rientjes | 8fe23e0 | 2009-12-14 17:58:33 -0800 | [diff] [blame] | 3654 | /* |
Jiang Liu | d8adde1 | 2012-07-11 14:01:52 -0700 | [diff] [blame] | 3655 | * Called by memory hotplug when all memory in a node is offlined. Caller must |
Vladimir Davydov | bfc8c90 | 2014-06-04 16:07:18 -0700 | [diff] [blame] | 3656 | * hold mem_hotplug_begin/end(). |
David Rientjes | 8fe23e0 | 2009-12-14 17:58:33 -0800 | [diff] [blame] | 3657 | */ |
| 3658 | void kswapd_stop(int nid) |
| 3659 | { |
| 3660 | struct task_struct *kswapd = NODE_DATA(nid)->kswapd; |
| 3661 | |
Jiang Liu | d8adde1 | 2012-07-11 14:01:52 -0700 | [diff] [blame] | 3662 | if (kswapd) { |
David Rientjes | 8fe23e0 | 2009-12-14 17:58:33 -0800 | [diff] [blame] | 3663 | kthread_stop(kswapd); |
Jiang Liu | d8adde1 | 2012-07-11 14:01:52 -0700 | [diff] [blame] | 3664 | NODE_DATA(nid)->kswapd = NULL; |
| 3665 | } |
David Rientjes | 8fe23e0 | 2009-12-14 17:58:33 -0800 | [diff] [blame] | 3666 | } |
| 3667 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3668 | static int __init kswapd_init(void) |
| 3669 | { |
Sebastian Andrzej Siewior | 517bbed | 2016-11-03 15:50:03 +0100 | [diff] [blame] | 3670 | int nid, ret; |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 3671 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3672 | swap_setup(); |
Lai Jiangshan | 48fb2e2 | 2012-12-12 13:51:43 -0800 | [diff] [blame] | 3673 | for_each_node_state(nid, N_MEMORY) |
Yasunori Goto | 3218ae1 | 2006-06-27 02:53:33 -0700 | [diff] [blame] | 3674 | kswapd_run(nid); |
Sebastian Andrzej Siewior | 517bbed | 2016-11-03 15:50:03 +0100 | [diff] [blame] | 3675 | ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, |
| 3676 | "mm/vmscan:online", kswapd_cpu_online, |
| 3677 | NULL); |
| 3678 | WARN_ON(ret < 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3679 | return 0; |
| 3680 | } |
| 3681 | |
| 3682 | module_init(kswapd_init) |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3683 | |
| 3684 | #ifdef CONFIG_NUMA |
| 3685 | /* |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3686 | * Node reclaim mode |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3687 | * |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3688 | * If non-zero call node_reclaim when the number of free pages falls below |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3689 | * the watermarks. |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3690 | */ |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3691 | int node_reclaim_mode __read_mostly; |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3692 | |
Christoph Lameter | 1b2ffb7 | 2006-02-01 03:05:34 -0800 | [diff] [blame] | 3693 | #define RECLAIM_OFF 0 |
Fernando Luis Vazquez Cao | 7d03431 | 2008-07-29 22:33:41 -0700 | [diff] [blame] | 3694 | #define RECLAIM_ZONE (1<<0) /* Run shrink_inactive_list on the zone */ |
Christoph Lameter | 1b2ffb7 | 2006-02-01 03:05:34 -0800 | [diff] [blame] | 3695 | #define RECLAIM_WRITE (1<<1) /* Writeout pages during reclaim */ |
Zhihui Zhang | 95bbc0c | 2015-06-24 16:56:42 -0700 | [diff] [blame] | 3696 | #define RECLAIM_UNMAP (1<<2) /* Unmap pages during reclaim */ |
Christoph Lameter | 1b2ffb7 | 2006-02-01 03:05:34 -0800 | [diff] [blame] | 3697 | |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3698 | /* |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3699 | * Priority for NODE_RECLAIM. This determines the fraction of pages |
Christoph Lameter | a92f712 | 2006-02-01 03:05:32 -0800 | [diff] [blame] | 3700 | * of a node considered for each zone_reclaim. 4 scans 1/16th of |
| 3701 | * a zone. |
| 3702 | */ |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3703 | #define NODE_RECLAIM_PRIORITY 4 |
Christoph Lameter | a92f712 | 2006-02-01 03:05:32 -0800 | [diff] [blame] | 3704 | |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3705 | /* |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3706 | * Percentage of pages in a zone that must be unmapped for node_reclaim to |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 3707 | * occur. |
| 3708 | */ |
| 3709 | int sysctl_min_unmapped_ratio = 1; |
| 3710 | |
| 3711 | /* |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3712 | * If the number of slab pages in a zone grows beyond this percentage then |
| 3713 | * slab reclaim needs to occur. |
| 3714 | */ |
| 3715 | int sysctl_min_slab_ratio = 5; |
| 3716 | |
Mel Gorman | 11fb998 | 2016-07-28 15:46:20 -0700 | [diff] [blame] | 3717 | static inline unsigned long node_unmapped_file_pages(struct pglist_data *pgdat) |
Mel Gorman | 90afa5d | 2009-06-16 15:33:20 -0700 | [diff] [blame] | 3718 | { |
Mel Gorman | 11fb998 | 2016-07-28 15:46:20 -0700 | [diff] [blame] | 3719 | unsigned long file_mapped = node_page_state(pgdat, NR_FILE_MAPPED); |
| 3720 | unsigned long file_lru = node_page_state(pgdat, NR_INACTIVE_FILE) + |
| 3721 | node_page_state(pgdat, NR_ACTIVE_FILE); |
Mel Gorman | 90afa5d | 2009-06-16 15:33:20 -0700 | [diff] [blame] | 3722 | |
| 3723 | /* |
| 3724 | * It's possible for there to be more file mapped pages than |
| 3725 | * accounted for by the pages on the file LRU lists because |
| 3726 | * tmpfs pages accounted for as ANON can also be FILE_MAPPED |
| 3727 | */ |
| 3728 | return (file_lru > file_mapped) ? (file_lru - file_mapped) : 0; |
| 3729 | } |
| 3730 | |
| 3731 | /* Work out how many page cache pages we can reclaim in this reclaim_mode */ |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3732 | static unsigned long node_pagecache_reclaimable(struct pglist_data *pgdat) |
Mel Gorman | 90afa5d | 2009-06-16 15:33:20 -0700 | [diff] [blame] | 3733 | { |
Alexandru Moise | d031a15 | 2015-11-05 18:48:08 -0800 | [diff] [blame] | 3734 | unsigned long nr_pagecache_reclaimable; |
| 3735 | unsigned long delta = 0; |
Mel Gorman | 90afa5d | 2009-06-16 15:33:20 -0700 | [diff] [blame] | 3736 | |
| 3737 | /* |
Zhihui Zhang | 95bbc0c | 2015-06-24 16:56:42 -0700 | [diff] [blame] | 3738 | * If RECLAIM_UNMAP is set, then all file pages are considered |
Mel Gorman | 90afa5d | 2009-06-16 15:33:20 -0700 | [diff] [blame] | 3739 | * potentially reclaimable. Otherwise, we have to worry about |
Mel Gorman | 11fb998 | 2016-07-28 15:46:20 -0700 | [diff] [blame] | 3740 | * pages like swapcache and node_unmapped_file_pages() provides |
Mel Gorman | 90afa5d | 2009-06-16 15:33:20 -0700 | [diff] [blame] | 3741 | * a better estimate |
| 3742 | */ |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3743 | if (node_reclaim_mode & RECLAIM_UNMAP) |
| 3744 | nr_pagecache_reclaimable = node_page_state(pgdat, NR_FILE_PAGES); |
Mel Gorman | 90afa5d | 2009-06-16 15:33:20 -0700 | [diff] [blame] | 3745 | else |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3746 | nr_pagecache_reclaimable = node_unmapped_file_pages(pgdat); |
Mel Gorman | 90afa5d | 2009-06-16 15:33:20 -0700 | [diff] [blame] | 3747 | |
| 3748 | /* If we can't clean pages, remove dirty pages from consideration */ |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3749 | if (!(node_reclaim_mode & RECLAIM_WRITE)) |
| 3750 | delta += node_page_state(pgdat, NR_FILE_DIRTY); |
Mel Gorman | 90afa5d | 2009-06-16 15:33:20 -0700 | [diff] [blame] | 3751 | |
| 3752 | /* Watch for any possible underflows due to delta */ |
| 3753 | if (unlikely(delta > nr_pagecache_reclaimable)) |
| 3754 | delta = nr_pagecache_reclaimable; |
| 3755 | |
| 3756 | return nr_pagecache_reclaimable - delta; |
| 3757 | } |
| 3758 | |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3759 | /* |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3760 | * Try to free up some pages from this node through reclaim. |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3761 | */ |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3762 | static int __node_reclaim(struct pglist_data *pgdat, gfp_t gfp_mask, unsigned int order) |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3763 | { |
Christoph Lameter | 7fb2d46 | 2006-03-22 00:08:22 -0800 | [diff] [blame] | 3764 | /* Minimum pages needed in order to stay on node */ |
Andrew Morton | 69e0594 | 2006-03-22 00:08:19 -0800 | [diff] [blame] | 3765 | const unsigned long nr_pages = 1 << order; |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3766 | struct task_struct *p = current; |
| 3767 | struct reclaim_state reclaim_state; |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3768 | int classzone_idx = gfp_zone(gfp_mask); |
Vlastimil Babka | 499118e | 2017-05-08 15:59:50 -0700 | [diff] [blame^] | 3769 | unsigned int noreclaim_flag; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3770 | struct scan_control sc = { |
Andrew Morton | 62b726c | 2013-02-22 16:32:24 -0800 | [diff] [blame] | 3771 | .nr_to_reclaim = max(nr_pages, SWAP_CLUSTER_MAX), |
Michal Hocko | 7dea19f | 2017-05-03 14:53:15 -0700 | [diff] [blame] | 3772 | .gfp_mask = (gfp_mask = current_gfp_context(gfp_mask)), |
Johannes Weiner | bd2f619 | 2009-03-31 15:19:38 -0700 | [diff] [blame] | 3773 | .order = order, |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3774 | .priority = NODE_RECLAIM_PRIORITY, |
| 3775 | .may_writepage = !!(node_reclaim_mode & RECLAIM_WRITE), |
| 3776 | .may_unmap = !!(node_reclaim_mode & RECLAIM_UNMAP), |
Johannes Weiner | ee814fe | 2014-08-06 16:06:19 -0700 | [diff] [blame] | 3777 | .may_swap = 1, |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3778 | .reclaim_idx = classzone_idx, |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3779 | }; |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3780 | |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3781 | cond_resched(); |
Christoph Lameter | d4f7796 | 2006-02-24 13:04:22 -0800 | [diff] [blame] | 3782 | /* |
Zhihui Zhang | 95bbc0c | 2015-06-24 16:56:42 -0700 | [diff] [blame] | 3783 | * We need to be able to allocate from the reserves for RECLAIM_UNMAP |
Christoph Lameter | d4f7796 | 2006-02-24 13:04:22 -0800 | [diff] [blame] | 3784 | * and we also need to be able to write out pages for RECLAIM_WRITE |
Zhihui Zhang | 95bbc0c | 2015-06-24 16:56:42 -0700 | [diff] [blame] | 3785 | * and RECLAIM_UNMAP. |
Christoph Lameter | d4f7796 | 2006-02-24 13:04:22 -0800 | [diff] [blame] | 3786 | */ |
Vlastimil Babka | 499118e | 2017-05-08 15:59:50 -0700 | [diff] [blame^] | 3787 | noreclaim_flag = memalloc_noreclaim_save(); |
| 3788 | p->flags |= PF_SWAPWRITE; |
KOSAKI Motohiro | 76ca542 | 2010-03-05 13:41:47 -0800 | [diff] [blame] | 3789 | lockdep_set_current_reclaim_state(gfp_mask); |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3790 | reclaim_state.reclaimed_slab = 0; |
| 3791 | p->reclaim_state = &reclaim_state; |
Christoph Lameter | c84db23 | 2006-02-01 03:05:29 -0800 | [diff] [blame] | 3792 | |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3793 | if (node_pagecache_reclaimable(pgdat) > pgdat->min_unmapped_pages) { |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3794 | /* |
| 3795 | * Free memory by calling shrink zone with increasing |
| 3796 | * priorities until we have enough memory freed. |
| 3797 | */ |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3798 | do { |
Mel Gorman | 970a39a | 2016-07-28 15:46:35 -0700 | [diff] [blame] | 3799 | shrink_node(pgdat, &sc); |
Konstantin Khlebnikov | 9e3b2f8 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3800 | } while (sc.nr_reclaimed < nr_pages && --sc.priority >= 0); |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3801 | } |
Christoph Lameter | c84db23 | 2006-02-01 03:05:29 -0800 | [diff] [blame] | 3802 | |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3803 | p->reclaim_state = NULL; |
Vlastimil Babka | 499118e | 2017-05-08 15:59:50 -0700 | [diff] [blame^] | 3804 | current->flags &= ~PF_SWAPWRITE; |
| 3805 | memalloc_noreclaim_restore(noreclaim_flag); |
KOSAKI Motohiro | 76ca542 | 2010-03-05 13:41:47 -0800 | [diff] [blame] | 3806 | lockdep_clear_current_reclaim_state(); |
Rik van Riel | a79311c | 2009-01-06 14:40:01 -0800 | [diff] [blame] | 3807 | return sc.nr_reclaimed >= nr_pages; |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3808 | } |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3809 | |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3810 | int node_reclaim(struct pglist_data *pgdat, gfp_t gfp_mask, unsigned int order) |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3811 | { |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3812 | int ret; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3813 | |
| 3814 | /* |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3815 | * Node reclaim reclaims unmapped file backed pages and |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 3816 | * slab pages if we are over the defined limits. |
Christoph Lameter | 34aa133 | 2006-06-30 01:55:37 -0700 | [diff] [blame] | 3817 | * |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 3818 | * A small portion of unmapped file backed pages is needed for |
| 3819 | * file I/O otherwise pages read by file I/O will be immediately |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3820 | * thrown out if the node is overallocated. So we do not reclaim |
| 3821 | * if less than a specified percentage of the node is used by |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 3822 | * unmapped file backed pages. |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3823 | */ |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3824 | if (node_pagecache_reclaimable(pgdat) <= pgdat->min_unmapped_pages && |
| 3825 | sum_zone_node_page_state(pgdat->node_id, NR_SLAB_RECLAIMABLE) <= pgdat->min_slab_pages) |
| 3826 | return NODE_RECLAIM_FULL; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3827 | |
| 3828 | /* |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3829 | * Do not scan if the allocation should not be delayed. |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3830 | */ |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 3831 | if (!gfpflags_allow_blocking(gfp_mask) || (current->flags & PF_MEMALLOC)) |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3832 | return NODE_RECLAIM_NOSCAN; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3833 | |
| 3834 | /* |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3835 | * Only run node reclaim on the local node or on nodes that do not |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3836 | * have associated processors. This will favor the local processor |
| 3837 | * over remote processors and spread off node memory allocations |
| 3838 | * as wide as possible. |
| 3839 | */ |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3840 | if (node_state(pgdat->node_id, N_CPU) && pgdat->node_id != numa_node_id()) |
| 3841 | return NODE_RECLAIM_NOSCAN; |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3842 | |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3843 | if (test_and_set_bit(PGDAT_RECLAIM_LOCKED, &pgdat->flags)) |
| 3844 | return NODE_RECLAIM_NOSCAN; |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 3845 | |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 3846 | ret = __node_reclaim(pgdat, gfp_mask, order); |
| 3847 | clear_bit(PGDAT_RECLAIM_LOCKED, &pgdat->flags); |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3848 | |
Mel Gorman | 24cf72518 | 2009-06-16 15:33:23 -0700 | [diff] [blame] | 3849 | if (!ret) |
| 3850 | count_vm_event(PGSCAN_ZONE_RECLAIM_FAILED); |
| 3851 | |
David Rientjes | d773ed6 | 2007-10-16 23:26:01 -0700 | [diff] [blame] | 3852 | return ret; |
Andrew Morton | 179e963 | 2006-03-22 00:08:18 -0800 | [diff] [blame] | 3853 | } |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3854 | #endif |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3855 | |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3856 | /* |
| 3857 | * page_evictable - test whether a page is evictable |
| 3858 | * @page: the page to test |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3859 | * |
| 3860 | * 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] | 3861 | * lists vs unevictable list. |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3862 | * |
| 3863 | * Reasons page might not be evictable: |
Lee Schermerhorn | ba9ddf4 | 2008-10-18 20:26:42 -0700 | [diff] [blame] | 3864 | * (1) page's mapping marked unevictable |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 3865 | * (2) page is part of an mlocked VMA |
Lee Schermerhorn | ba9ddf4 | 2008-10-18 20:26:42 -0700 | [diff] [blame] | 3866 | * |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3867 | */ |
Hugh Dickins | 39b5f29 | 2012-10-08 16:33:18 -0700 | [diff] [blame] | 3868 | int page_evictable(struct page *page) |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3869 | { |
Hugh Dickins | 39b5f29 | 2012-10-08 16:33:18 -0700 | [diff] [blame] | 3870 | return !mapping_unevictable(page_mapping(page)) && !PageMlocked(page); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 3871 | } |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3872 | |
Hugh Dickins | 85046579 | 2012-01-20 14:34:19 -0800 | [diff] [blame] | 3873 | #ifdef CONFIG_SHMEM |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3874 | /** |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3875 | * check_move_unevictable_pages - check pages for evictability and move to appropriate zone lru list |
| 3876 | * @pages: array of pages to check |
| 3877 | * @nr_pages: number of pages to check |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3878 | * |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3879 | * Checks pages for evictability and moves them to the appropriate lru list. |
Hugh Dickins | 85046579 | 2012-01-20 14:34:19 -0800 | [diff] [blame] | 3880 | * |
| 3881 | * This function is only used for SysV IPC SHM_UNLOCK. |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3882 | */ |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3883 | void check_move_unevictable_pages(struct page **pages, int nr_pages) |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3884 | { |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 3885 | struct lruvec *lruvec; |
Mel Gorman | 785b99f | 2016-07-28 15:47:23 -0700 | [diff] [blame] | 3886 | struct pglist_data *pgdat = NULL; |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3887 | int pgscanned = 0; |
| 3888 | int pgrescued = 0; |
| 3889 | int i; |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3890 | |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3891 | for (i = 0; i < nr_pages; i++) { |
| 3892 | struct page *page = pages[i]; |
Mel Gorman | 785b99f | 2016-07-28 15:47:23 -0700 | [diff] [blame] | 3893 | struct pglist_data *pagepgdat = page_pgdat(page); |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 3894 | |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3895 | pgscanned++; |
Mel Gorman | 785b99f | 2016-07-28 15:47:23 -0700 | [diff] [blame] | 3896 | if (pagepgdat != pgdat) { |
| 3897 | if (pgdat) |
| 3898 | spin_unlock_irq(&pgdat->lru_lock); |
| 3899 | pgdat = pagepgdat; |
| 3900 | spin_lock_irq(&pgdat->lru_lock); |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3901 | } |
Mel Gorman | 785b99f | 2016-07-28 15:47:23 -0700 | [diff] [blame] | 3902 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3903 | |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3904 | if (!PageLRU(page) || !PageUnevictable(page)) |
| 3905 | continue; |
| 3906 | |
Hugh Dickins | 39b5f29 | 2012-10-08 16:33:18 -0700 | [diff] [blame] | 3907 | if (page_evictable(page)) { |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3908 | enum lru_list lru = page_lru_base_type(page); |
| 3909 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 3910 | VM_BUG_ON_PAGE(PageActive(page), page); |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3911 | ClearPageUnevictable(page); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 3912 | del_page_from_lru_list(page, lruvec, LRU_UNEVICTABLE); |
| 3913 | add_page_to_lru_list(page, lruvec, lru); |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3914 | pgrescued++; |
| 3915 | } |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 3916 | } |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3917 | |
Mel Gorman | 785b99f | 2016-07-28 15:47:23 -0700 | [diff] [blame] | 3918 | if (pgdat) { |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3919 | __count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued); |
| 3920 | __count_vm_events(UNEVICTABLE_PGSCANNED, pgscanned); |
Mel Gorman | 785b99f | 2016-07-28 15:47:23 -0700 | [diff] [blame] | 3921 | spin_unlock_irq(&pgdat->lru_lock); |
Hugh Dickins | 2451326 | 2012-01-20 14:34:21 -0800 | [diff] [blame] | 3922 | } |
Hugh Dickins | 85046579 | 2012-01-20 14:34:19 -0800 | [diff] [blame] | 3923 | } |
| 3924 | #endif /* CONFIG_SHMEM */ |