Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 1 | /* memcontrol.c - Memory Controller |
| 2 | * |
| 3 | * Copyright IBM Corporation, 2007 |
| 4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> |
| 5 | * |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 6 | * Copyright 2007 OpenVZ SWsoft Inc |
| 7 | * Author: Pavel Emelianov <xemul@openvz.org> |
| 8 | * |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 9 | * Memory thresholds |
| 10 | * Copyright (C) 2009 Nokia Corporation |
| 11 | * Author: Kirill A. Shutemov |
| 12 | * |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 13 | * This program is free software; you can redistribute it and/or modify |
| 14 | * it under the terms of the GNU General Public License as published by |
| 15 | * the Free Software Foundation; either version 2 of the License, or |
| 16 | * (at your option) any later version. |
| 17 | * |
| 18 | * This program is distributed in the hope that it will be useful, |
| 19 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 20 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 21 | * GNU General Public License for more details. |
| 22 | */ |
| 23 | |
| 24 | #include <linux/res_counter.h> |
| 25 | #include <linux/memcontrol.h> |
| 26 | #include <linux/cgroup.h> |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 27 | #include <linux/mm.h> |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 28 | #include <linux/hugetlb.h> |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 29 | #include <linux/pagemap.h> |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 30 | #include <linux/smp.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 31 | #include <linux/page-flags.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 32 | #include <linux/backing-dev.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 33 | #include <linux/bit_spinlock.h> |
| 34 | #include <linux/rcupdate.h> |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 35 | #include <linux/limits.h> |
Paul Gortmaker | b9e15ba | 2011-05-26 16:00:52 -0400 | [diff] [blame] | 36 | #include <linux/export.h> |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 37 | #include <linux/mutex.h> |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 38 | #include <linux/rbtree.h> |
Balbir Singh | b6ac57d | 2008-04-29 01:00:19 -0700 | [diff] [blame] | 39 | #include <linux/slab.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 40 | #include <linux/swap.h> |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 41 | #include <linux/swapops.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 42 | #include <linux/spinlock.h> |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 43 | #include <linux/eventfd.h> |
| 44 | #include <linux/sort.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 45 | #include <linux/fs.h> |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 46 | #include <linux/seq_file.h> |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 47 | #include <linux/vmalloc.h> |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 48 | #include <linux/mm_inline.h> |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 49 | #include <linux/page_cgroup.h> |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 50 | #include <linux/cpu.h> |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 51 | #include <linux/oom.h> |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 52 | #include "internal.h" |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 53 | #include <net/sock.h> |
| 54 | #include <net/tcp_memcontrol.h> |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 55 | |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 56 | #include <asm/uaccess.h> |
| 57 | |
KOSAKI Motohiro | cc8e970 | 2010-08-09 17:19:57 -0700 | [diff] [blame] | 58 | #include <trace/events/vmscan.h> |
| 59 | |
KAMEZAWA Hiroyuki | a181b0e | 2008-07-25 01:47:08 -0700 | [diff] [blame] | 60 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
KAMEZAWA Hiroyuki | a181b0e | 2008-07-25 01:47:08 -0700 | [diff] [blame] | 61 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 62 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 63 | |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 64 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
Li Zefan | 338c843 | 2009-06-17 16:27:15 -0700 | [diff] [blame] | 65 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 66 | int do_swap_account __read_mostly; |
Michal Hocko | a42c390 | 2010-11-24 12:57:08 -0800 | [diff] [blame] | 67 | |
| 68 | /* for remember boot option*/ |
| 69 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED |
| 70 | static int really_do_swap_account __initdata = 1; |
| 71 | #else |
| 72 | static int really_do_swap_account __initdata = 0; |
| 73 | #endif |
| 74 | |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 75 | #else |
| 76 | #define do_swap_account (0) |
| 77 | #endif |
| 78 | |
| 79 | |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 80 | /* |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 81 | * Statistics for memory cgroup. |
| 82 | */ |
| 83 | enum mem_cgroup_stat_index { |
| 84 | /* |
| 85 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. |
| 86 | */ |
| 87 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 88 | MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ |
KAMEZAWA Hiroyuki | d804658 | 2009-12-15 16:47:09 -0800 | [diff] [blame] | 89 | MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */ |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 90 | MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */ |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 91 | MEM_CGROUP_STAT_DATA, /* end of data requires synchronization */ |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 92 | MEM_CGROUP_STAT_NSTATS, |
| 93 | }; |
| 94 | |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 95 | enum mem_cgroup_events_index { |
| 96 | MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */ |
| 97 | MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */ |
| 98 | MEM_CGROUP_EVENTS_COUNT, /* # of pages paged in/out */ |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 99 | MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */ |
| 100 | MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */ |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 101 | MEM_CGROUP_EVENTS_NSTATS, |
| 102 | }; |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 103 | /* |
| 104 | * Per memcg event counter is incremented at every pagein/pageout. With THP, |
| 105 | * it will be incremated by the number of pages. This counter is used for |
| 106 | * for trigger some periodic events. This is straightforward and better |
| 107 | * than using jiffies etc. to handle periodic memcg event. |
| 108 | */ |
| 109 | enum mem_cgroup_events_target { |
| 110 | MEM_CGROUP_TARGET_THRESH, |
| 111 | MEM_CGROUP_TARGET_SOFTLIMIT, |
KAMEZAWA Hiroyuki | 453a9bf3 | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 112 | MEM_CGROUP_TARGET_NUMAINFO, |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 113 | MEM_CGROUP_NTARGETS, |
| 114 | }; |
| 115 | #define THRESHOLDS_EVENTS_TARGET (128) |
| 116 | #define SOFTLIMIT_EVENTS_TARGET (1024) |
KAMEZAWA Hiroyuki | 453a9bf3 | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 117 | #define NUMAINFO_EVENTS_TARGET (1024) |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 118 | |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 119 | struct mem_cgroup_stat_cpu { |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 120 | long count[MEM_CGROUP_STAT_NSTATS]; |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 121 | unsigned long events[MEM_CGROUP_EVENTS_NSTATS]; |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 122 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 123 | }; |
| 124 | |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 125 | struct mem_cgroup_reclaim_iter { |
| 126 | /* css_id of the last scanned hierarchy member */ |
| 127 | int position; |
| 128 | /* scan generation, increased every round-trip */ |
| 129 | unsigned int generation; |
| 130 | }; |
| 131 | |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 132 | /* |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 133 | * per-zone information in memory controller. |
| 134 | */ |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 135 | struct mem_cgroup_per_zone { |
Johannes Weiner | 6290df5 | 2012-01-12 17:18:10 -0800 | [diff] [blame] | 136 | struct lruvec lruvec; |
Hugh Dickins | 1eb4927 | 2012-03-21 16:34:19 -0700 | [diff] [blame] | 137 | unsigned long lru_size[NR_LRU_LISTS]; |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 138 | |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 139 | struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1]; |
| 140 | |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 141 | struct zone_reclaim_stat reclaim_stat; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 142 | struct rb_node tree_node; /* RB tree node */ |
| 143 | unsigned long long usage_in_excess;/* Set to the value by which */ |
| 144 | /* the soft limit is exceeded*/ |
| 145 | bool on_tree; |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 146 | struct mem_cgroup *memcg; /* Back pointer, we cannot */ |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 147 | /* use container_of */ |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 148 | }; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 149 | |
| 150 | struct mem_cgroup_per_node { |
| 151 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; |
| 152 | }; |
| 153 | |
| 154 | struct mem_cgroup_lru_info { |
| 155 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; |
| 156 | }; |
| 157 | |
| 158 | /* |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 159 | * Cgroups above their limits are maintained in a RB-Tree, independent of |
| 160 | * their hierarchy representation |
| 161 | */ |
| 162 | |
| 163 | struct mem_cgroup_tree_per_zone { |
| 164 | struct rb_root rb_root; |
| 165 | spinlock_t lock; |
| 166 | }; |
| 167 | |
| 168 | struct mem_cgroup_tree_per_node { |
| 169 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; |
| 170 | }; |
| 171 | |
| 172 | struct mem_cgroup_tree { |
| 173 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; |
| 174 | }; |
| 175 | |
| 176 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; |
| 177 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 178 | struct mem_cgroup_threshold { |
| 179 | struct eventfd_ctx *eventfd; |
| 180 | u64 threshold; |
| 181 | }; |
| 182 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 183 | /* For threshold */ |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 184 | struct mem_cgroup_threshold_ary { |
| 185 | /* An array index points to threshold just below usage. */ |
Phil Carmody | 5407a56 | 2010-05-26 14:42:42 -0700 | [diff] [blame] | 186 | int current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 187 | /* Size of entries[] */ |
| 188 | unsigned int size; |
| 189 | /* Array of thresholds */ |
| 190 | struct mem_cgroup_threshold entries[0]; |
| 191 | }; |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 192 | |
| 193 | struct mem_cgroup_thresholds { |
| 194 | /* Primary thresholds array */ |
| 195 | struct mem_cgroup_threshold_ary *primary; |
| 196 | /* |
| 197 | * Spare threshold array. |
| 198 | * This is needed to make mem_cgroup_unregister_event() "never fail". |
| 199 | * It must be able to store at least primary->size - 1 entries. |
| 200 | */ |
| 201 | struct mem_cgroup_threshold_ary *spare; |
| 202 | }; |
| 203 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 204 | /* for OOM */ |
| 205 | struct mem_cgroup_eventfd_list { |
| 206 | struct list_head list; |
| 207 | struct eventfd_ctx *eventfd; |
| 208 | }; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 209 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 210 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
| 211 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 212 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 213 | /* |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 214 | * The memory controller data structure. The memory controller controls both |
| 215 | * page cache and RSS per cgroup. We would eventually like to provide |
| 216 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, |
| 217 | * to help the administrator determine what knobs to tune. |
| 218 | * |
| 219 | * TODO: Add a water mark for the memory controller. Reclaim will begin when |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 220 | * we hit the water mark. May be even add a low water mark, such that |
| 221 | * no reclaim occurs from a cgroup at it's low water mark, this is |
| 222 | * a feature that will be implemented much later in the future. |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 223 | */ |
| 224 | struct mem_cgroup { |
| 225 | struct cgroup_subsys_state css; |
| 226 | /* |
| 227 | * the counter to account for memory usage |
| 228 | */ |
| 229 | struct res_counter res; |
Hugh Dickins | 59927fb | 2012-03-15 15:17:07 -0700 | [diff] [blame] | 230 | |
| 231 | union { |
| 232 | /* |
| 233 | * the counter to account for mem+swap usage. |
| 234 | */ |
| 235 | struct res_counter memsw; |
| 236 | |
| 237 | /* |
| 238 | * rcu_freeing is used only when freeing struct mem_cgroup, |
| 239 | * so put it into a union to avoid wasting more memory. |
| 240 | * It must be disjoint from the css field. It could be |
| 241 | * in a union with the res field, but res plays a much |
| 242 | * larger part in mem_cgroup life than memsw, and might |
| 243 | * be of interest, even at time of free, when debugging. |
| 244 | * So share rcu_head with the less interesting memsw. |
| 245 | */ |
| 246 | struct rcu_head rcu_freeing; |
| 247 | /* |
| 248 | * But when using vfree(), that cannot be done at |
| 249 | * interrupt time, so we must then queue the work. |
| 250 | */ |
| 251 | struct work_struct work_freeing; |
| 252 | }; |
| 253 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 254 | /* |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 255 | * Per cgroup active and inactive list, similar to the |
| 256 | * per zone LRU lists. |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 257 | */ |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 258 | struct mem_cgroup_lru_info info; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 259 | int last_scanned_node; |
| 260 | #if MAX_NUMNODES > 1 |
| 261 | nodemask_t scan_nodes; |
KAMEZAWA Hiroyuki | 453a9bf3 | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 262 | atomic_t numainfo_events; |
| 263 | atomic_t numainfo_updating; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 264 | #endif |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 265 | /* |
| 266 | * Should the accounting and control be hierarchical, per subtree? |
| 267 | */ |
| 268 | bool use_hierarchy; |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 269 | |
| 270 | bool oom_lock; |
| 271 | atomic_t under_oom; |
| 272 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 273 | atomic_t refcnt; |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 274 | |
KAMEZAWA Hiroyuki | 1f4c025 | 2011-07-26 16:08:21 -0700 | [diff] [blame] | 275 | int swappiness; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 276 | /* OOM-Killer disable */ |
| 277 | int oom_kill_disable; |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 278 | |
KAMEZAWA Hiroyuki | 22a668d | 2009-06-17 16:27:19 -0700 | [diff] [blame] | 279 | /* set when res.limit == memsw.limit */ |
| 280 | bool memsw_is_minimum; |
| 281 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 282 | /* protect arrays of thresholds */ |
| 283 | struct mutex thresholds_lock; |
| 284 | |
| 285 | /* thresholds for memory usage. RCU-protected */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 286 | struct mem_cgroup_thresholds thresholds; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 287 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 288 | /* thresholds for mem+swap usage. RCU-protected */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 289 | struct mem_cgroup_thresholds memsw_thresholds; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 290 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 291 | /* For oom notifier event fd */ |
| 292 | struct list_head oom_notify; |
Johannes Weiner | 185efc0 | 2011-09-14 16:21:58 -0700 | [diff] [blame] | 293 | |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 294 | /* |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 295 | * Should we move charges of a task when a task is moved into this |
| 296 | * mem_cgroup ? And what type of charges should we move ? |
| 297 | */ |
| 298 | unsigned long move_charge_at_immigrate; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 299 | /* |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 300 | * set > 0 if pages under this cgroup are moving to other cgroup. |
| 301 | */ |
| 302 | atomic_t moving_account; |
KAMEZAWA Hiroyuki | 312734c0 | 2012-03-21 16:34:24 -0700 | [diff] [blame] | 303 | /* taken only while moving_account > 0 */ |
| 304 | spinlock_t move_lock; |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 305 | /* |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 306 | * percpu counter. |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 307 | */ |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 308 | struct mem_cgroup_stat_cpu *stat; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 309 | /* |
| 310 | * used when a cpu is offlined or other synchronizations |
| 311 | * See mem_cgroup_read_stat(). |
| 312 | */ |
| 313 | struct mem_cgroup_stat_cpu nocpu_base; |
| 314 | spinlock_t pcp_counter_lock; |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 315 | |
| 316 | #ifdef CONFIG_INET |
| 317 | struct tcp_memcontrol tcp_mem; |
| 318 | #endif |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 319 | }; |
| 320 | |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 321 | /* Stuffs for move charges at task migration. */ |
| 322 | /* |
| 323 | * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a |
| 324 | * left-shifted bitmap of these types. |
| 325 | */ |
| 326 | enum move_type { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 327 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 328 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 329 | NR_MOVE_TYPE, |
| 330 | }; |
| 331 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 332 | /* "mc" and its members are protected by cgroup_mutex */ |
| 333 | static struct move_charge_struct { |
Daisuke Nishimura | b1dd693 | 2010-11-24 12:57:06 -0800 | [diff] [blame] | 334 | spinlock_t lock; /* for from, to */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 335 | struct mem_cgroup *from; |
| 336 | struct mem_cgroup *to; |
| 337 | unsigned long precharge; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 338 | unsigned long moved_charge; |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 339 | unsigned long moved_swap; |
Daisuke Nishimura | 8033b97 | 2010-03-10 15:22:16 -0800 | [diff] [blame] | 340 | struct task_struct *moving_task; /* a task moving charges */ |
| 341 | wait_queue_head_t waitq; /* a waitq for other context */ |
| 342 | } mc = { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 343 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
Daisuke Nishimura | 8033b97 | 2010-03-10 15:22:16 -0800 | [diff] [blame] | 344 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
| 345 | }; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 346 | |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 347 | static bool move_anon(void) |
| 348 | { |
| 349 | return test_bit(MOVE_CHARGE_TYPE_ANON, |
| 350 | &mc.to->move_charge_at_immigrate); |
| 351 | } |
| 352 | |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 353 | static bool move_file(void) |
| 354 | { |
| 355 | return test_bit(MOVE_CHARGE_TYPE_FILE, |
| 356 | &mc.to->move_charge_at_immigrate); |
| 357 | } |
| 358 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 359 | /* |
| 360 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft |
| 361 | * limit reclaim to prevent infinite loops, if they ever occur. |
| 362 | */ |
| 363 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100) |
| 364 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2) |
| 365 | |
KAMEZAWA Hiroyuki | 217bc31 | 2008-02-07 00:14:17 -0800 | [diff] [blame] | 366 | enum charge_type { |
| 367 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, |
| 368 | MEM_CGROUP_CHARGE_TYPE_MAPPED, |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 369 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
KAMEZAWA Hiroyuki | c05555b | 2008-10-18 20:28:11 -0700 | [diff] [blame] | 370 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 371 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
KAMEZAWA Hiroyuki | 8a9478ca | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 372 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
KAMEZAWA Hiroyuki | c05555b | 2008-10-18 20:28:11 -0700 | [diff] [blame] | 373 | NR_CHARGE_TYPE, |
| 374 | }; |
| 375 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 376 | /* for encoding cft->private value on file */ |
Glauber Costa | 65c64ce | 2011-12-22 01:02:27 +0000 | [diff] [blame] | 377 | #define _MEM (0) |
| 378 | #define _MEMSWAP (1) |
| 379 | #define _OOM_TYPE (2) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 380 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) |
| 381 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) |
| 382 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 383 | /* Used for OOM nofiier */ |
| 384 | #define OOM_CONTROL (0) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 385 | |
Balbir Singh | 75822b4 | 2009-09-23 15:56:38 -0700 | [diff] [blame] | 386 | /* |
| 387 | * Reclaim flags for mem_cgroup_hierarchical_reclaim |
| 388 | */ |
| 389 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 |
| 390 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) |
| 391 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 |
| 392 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) |
| 393 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 394 | static void mem_cgroup_get(struct mem_cgroup *memcg); |
| 395 | static void mem_cgroup_put(struct mem_cgroup *memcg); |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 396 | |
| 397 | /* Writing them here to avoid exposing memcg's inner layout */ |
| 398 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 399 | #include <net/sock.h> |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 400 | #include <net/ip.h> |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 401 | |
| 402 | static bool mem_cgroup_is_root(struct mem_cgroup *memcg); |
| 403 | void sock_update_memcg(struct sock *sk) |
| 404 | { |
Glauber Costa | 376be5f | 2012-01-20 04:57:14 +0000 | [diff] [blame] | 405 | if (mem_cgroup_sockets_enabled) { |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 406 | struct mem_cgroup *memcg; |
| 407 | |
| 408 | BUG_ON(!sk->sk_prot->proto_cgroup); |
| 409 | |
Glauber Costa | f3f511e | 2012-01-05 20:16:39 +0000 | [diff] [blame] | 410 | /* Socket cloning can throw us here with sk_cgrp already |
| 411 | * filled. It won't however, necessarily happen from |
| 412 | * process context. So the test for root memcg given |
| 413 | * the current task's memcg won't help us in this case. |
| 414 | * |
| 415 | * Respecting the original socket's memcg is a better |
| 416 | * decision in this case. |
| 417 | */ |
| 418 | if (sk->sk_cgrp) { |
| 419 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); |
| 420 | mem_cgroup_get(sk->sk_cgrp->memcg); |
| 421 | return; |
| 422 | } |
| 423 | |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 424 | rcu_read_lock(); |
| 425 | memcg = mem_cgroup_from_task(current); |
| 426 | if (!mem_cgroup_is_root(memcg)) { |
| 427 | mem_cgroup_get(memcg); |
| 428 | sk->sk_cgrp = sk->sk_prot->proto_cgroup(memcg); |
| 429 | } |
| 430 | rcu_read_unlock(); |
| 431 | } |
| 432 | } |
| 433 | EXPORT_SYMBOL(sock_update_memcg); |
| 434 | |
| 435 | void sock_release_memcg(struct sock *sk) |
| 436 | { |
Glauber Costa | 376be5f | 2012-01-20 04:57:14 +0000 | [diff] [blame] | 437 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 438 | struct mem_cgroup *memcg; |
| 439 | WARN_ON(!sk->sk_cgrp->memcg); |
| 440 | memcg = sk->sk_cgrp->memcg; |
| 441 | mem_cgroup_put(memcg); |
| 442 | } |
| 443 | } |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 444 | |
Glauber Costa | 319d3b9 | 2012-01-15 22:04:39 +0000 | [diff] [blame] | 445 | #ifdef CONFIG_INET |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 446 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) |
| 447 | { |
| 448 | if (!memcg || mem_cgroup_is_root(memcg)) |
| 449 | return NULL; |
| 450 | |
| 451 | return &memcg->tcp_mem.cg_proto; |
| 452 | } |
| 453 | EXPORT_SYMBOL(tcp_proto_cgroup); |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 454 | #endif /* CONFIG_INET */ |
| 455 | #endif /* CONFIG_CGROUP_MEM_RES_CTLR_KMEM */ |
| 456 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 457 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 458 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 459 | static struct mem_cgroup_per_zone * |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 460 | mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid) |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 461 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 462 | return &memcg->info.nodeinfo[nid]->zoneinfo[zid]; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 463 | } |
| 464 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 465 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
Wu Fengguang | d324236 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 466 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 467 | return &memcg->css; |
Wu Fengguang | d324236 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 468 | } |
| 469 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 470 | static struct mem_cgroup_per_zone * |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 471 | page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 472 | { |
Johannes Weiner | 97a6c37 | 2011-03-23 16:42:27 -0700 | [diff] [blame] | 473 | int nid = page_to_nid(page); |
| 474 | int zid = page_zonenum(page); |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 475 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 476 | return mem_cgroup_zoneinfo(memcg, nid, zid); |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 477 | } |
| 478 | |
| 479 | static struct mem_cgroup_tree_per_zone * |
| 480 | soft_limit_tree_node_zone(int nid, int zid) |
| 481 | { |
| 482 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; |
| 483 | } |
| 484 | |
| 485 | static struct mem_cgroup_tree_per_zone * |
| 486 | soft_limit_tree_from_page(struct page *page) |
| 487 | { |
| 488 | int nid = page_to_nid(page); |
| 489 | int zid = page_zonenum(page); |
| 490 | |
| 491 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; |
| 492 | } |
| 493 | |
| 494 | static void |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 495 | __mem_cgroup_insert_exceeded(struct mem_cgroup *memcg, |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 496 | struct mem_cgroup_per_zone *mz, |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 497 | struct mem_cgroup_tree_per_zone *mctz, |
| 498 | unsigned long long new_usage_in_excess) |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 499 | { |
| 500 | struct rb_node **p = &mctz->rb_root.rb_node; |
| 501 | struct rb_node *parent = NULL; |
| 502 | struct mem_cgroup_per_zone *mz_node; |
| 503 | |
| 504 | if (mz->on_tree) |
| 505 | return; |
| 506 | |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 507 | mz->usage_in_excess = new_usage_in_excess; |
| 508 | if (!mz->usage_in_excess) |
| 509 | return; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 510 | while (*p) { |
| 511 | parent = *p; |
| 512 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, |
| 513 | tree_node); |
| 514 | if (mz->usage_in_excess < mz_node->usage_in_excess) |
| 515 | p = &(*p)->rb_left; |
| 516 | /* |
| 517 | * We can't avoid mem cgroups that are over their soft |
| 518 | * limit by the same amount |
| 519 | */ |
| 520 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) |
| 521 | p = &(*p)->rb_right; |
| 522 | } |
| 523 | rb_link_node(&mz->tree_node, parent, p); |
| 524 | rb_insert_color(&mz->tree_node, &mctz->rb_root); |
| 525 | mz->on_tree = true; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 526 | } |
| 527 | |
| 528 | static void |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 529 | __mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 530 | struct mem_cgroup_per_zone *mz, |
| 531 | struct mem_cgroup_tree_per_zone *mctz) |
| 532 | { |
| 533 | if (!mz->on_tree) |
| 534 | return; |
| 535 | rb_erase(&mz->tree_node, &mctz->rb_root); |
| 536 | mz->on_tree = false; |
| 537 | } |
| 538 | |
| 539 | static void |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 540 | mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 541 | struct mem_cgroup_per_zone *mz, |
| 542 | struct mem_cgroup_tree_per_zone *mctz) |
| 543 | { |
| 544 | spin_lock(&mctz->lock); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 545 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 546 | spin_unlock(&mctz->lock); |
| 547 | } |
| 548 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 549 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 550 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 551 | { |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 552 | unsigned long long excess; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 553 | struct mem_cgroup_per_zone *mz; |
| 554 | struct mem_cgroup_tree_per_zone *mctz; |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 555 | int nid = page_to_nid(page); |
| 556 | int zid = page_zonenum(page); |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 557 | mctz = soft_limit_tree_from_page(page); |
| 558 | |
| 559 | /* |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 560 | * Necessary to update all ancestors when hierarchy is used. |
| 561 | * because their event counter is not touched. |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 562 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 563 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
| 564 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
| 565 | excess = res_counter_soft_limit_excess(&memcg->res); |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 566 | /* |
| 567 | * We have to update the tree if mz is on RB-tree or |
| 568 | * mem is over its softlimit. |
| 569 | */ |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 570 | if (excess || mz->on_tree) { |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 571 | spin_lock(&mctz->lock); |
| 572 | /* if on-tree, remove it */ |
| 573 | if (mz->on_tree) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 574 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 575 | /* |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 576 | * Insert again. mz->usage_in_excess will be updated. |
| 577 | * If excess is 0, no tree ops. |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 578 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 579 | __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess); |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 580 | spin_unlock(&mctz->lock); |
| 581 | } |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 582 | } |
| 583 | } |
| 584 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 585 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 586 | { |
| 587 | int node, zone; |
| 588 | struct mem_cgroup_per_zone *mz; |
| 589 | struct mem_cgroup_tree_per_zone *mctz; |
| 590 | |
Bob Liu | 3ed28fa | 2012-01-12 17:19:04 -0800 | [diff] [blame] | 591 | for_each_node(node) { |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 592 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 593 | mz = mem_cgroup_zoneinfo(memcg, node, zone); |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 594 | mctz = soft_limit_tree_node_zone(node, zone); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 595 | mem_cgroup_remove_exceeded(memcg, mz, mctz); |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 596 | } |
| 597 | } |
| 598 | } |
| 599 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 600 | static struct mem_cgroup_per_zone * |
| 601 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) |
| 602 | { |
| 603 | struct rb_node *rightmost = NULL; |
KAMEZAWA Hiroyuki | 26251ea | 2009-10-01 15:44:08 -0700 | [diff] [blame] | 604 | struct mem_cgroup_per_zone *mz; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 605 | |
| 606 | retry: |
KAMEZAWA Hiroyuki | 26251ea | 2009-10-01 15:44:08 -0700 | [diff] [blame] | 607 | mz = NULL; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 608 | rightmost = rb_last(&mctz->rb_root); |
| 609 | if (!rightmost) |
| 610 | goto done; /* Nothing to reclaim from */ |
| 611 | |
| 612 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); |
| 613 | /* |
| 614 | * Remove the node now but someone else can add it back, |
| 615 | * we will to add it back at the end of reclaim to its correct |
| 616 | * position in the tree. |
| 617 | */ |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 618 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
| 619 | if (!res_counter_soft_limit_excess(&mz->memcg->res) || |
| 620 | !css_tryget(&mz->memcg->css)) |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 621 | goto retry; |
| 622 | done: |
| 623 | return mz; |
| 624 | } |
| 625 | |
| 626 | static struct mem_cgroup_per_zone * |
| 627 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) |
| 628 | { |
| 629 | struct mem_cgroup_per_zone *mz; |
| 630 | |
| 631 | spin_lock(&mctz->lock); |
| 632 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
| 633 | spin_unlock(&mctz->lock); |
| 634 | return mz; |
| 635 | } |
| 636 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 637 | /* |
| 638 | * Implementation Note: reading percpu statistics for memcg. |
| 639 | * |
| 640 | * Both of vmstat[] and percpu_counter has threshold and do periodic |
| 641 | * synchronization to implement "quick" read. There are trade-off between |
| 642 | * reading cost and precision of value. Then, we may have a chance to implement |
| 643 | * a periodic synchronizion of counter in memcg's counter. |
| 644 | * |
| 645 | * But this _read() function is used for user interface now. The user accounts |
| 646 | * memory usage by memory cgroup and he _always_ requires exact value because |
| 647 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always |
| 648 | * have to visit all online cpus and make sum. So, for now, unnecessary |
| 649 | * synchronization is not implemented. (just implemented for cpu hotplug) |
| 650 | * |
| 651 | * If there are kernel internal actions which can make use of some not-exact |
| 652 | * value, and reading all cpu value can be performance bottleneck in some |
| 653 | * common workload, threashold and synchonization as vmstat[] should be |
| 654 | * implemented. |
| 655 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 656 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 657 | enum mem_cgroup_stat_index idx) |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 658 | { |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 659 | long val = 0; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 660 | int cpu; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 661 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 662 | get_online_cpus(); |
| 663 | for_each_online_cpu(cpu) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 664 | val += per_cpu(memcg->stat->count[idx], cpu); |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 665 | #ifdef CONFIG_HOTPLUG_CPU |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 666 | spin_lock(&memcg->pcp_counter_lock); |
| 667 | val += memcg->nocpu_base.count[idx]; |
| 668 | spin_unlock(&memcg->pcp_counter_lock); |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 669 | #endif |
| 670 | put_online_cpus(); |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 671 | return val; |
| 672 | } |
| 673 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 674 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 675 | bool charge) |
| 676 | { |
| 677 | int val = (charge) ? 1 : -1; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 678 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAPOUT], val); |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 679 | } |
| 680 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 681 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 682 | enum mem_cgroup_events_index idx) |
| 683 | { |
| 684 | unsigned long val = 0; |
| 685 | int cpu; |
| 686 | |
| 687 | for_each_online_cpu(cpu) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 688 | val += per_cpu(memcg->stat->events[idx], cpu); |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 689 | #ifdef CONFIG_HOTPLUG_CPU |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 690 | spin_lock(&memcg->pcp_counter_lock); |
| 691 | val += memcg->nocpu_base.events[idx]; |
| 692 | spin_unlock(&memcg->pcp_counter_lock); |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 693 | #endif |
| 694 | return val; |
| 695 | } |
| 696 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 697 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 698 | bool anon, int nr_pages) |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 699 | { |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 700 | preempt_disable(); |
| 701 | |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 702 | /* |
| 703 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is |
| 704 | * counted as CACHE even if it's on ANON LRU. |
| 705 | */ |
| 706 | if (anon) |
| 707 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 708 | nr_pages); |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 709 | else |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 710 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 711 | nr_pages); |
Balaji Rao | 55e462b | 2008-05-01 04:35:12 -0700 | [diff] [blame] | 712 | |
KAMEZAWA Hiroyuki | e401f17 | 2011-01-20 14:44:23 -0800 | [diff] [blame] | 713 | /* pagein of a big page is an event. So, ignore page size */ |
| 714 | if (nr_pages > 0) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 715 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
KAMEZAWA Hiroyuki | 3751d60 | 2011-02-01 15:52:45 -0800 | [diff] [blame] | 716 | else { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 717 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
KAMEZAWA Hiroyuki | 3751d60 | 2011-02-01 15:52:45 -0800 | [diff] [blame] | 718 | nr_pages = -nr_pages; /* for event */ |
| 719 | } |
KAMEZAWA Hiroyuki | e401f17 | 2011-01-20 14:44:23 -0800 | [diff] [blame] | 720 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 721 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT], nr_pages); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 722 | |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 723 | preempt_enable(); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 724 | } |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 725 | |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 726 | unsigned long |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 727 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid, |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 728 | unsigned int lru_mask) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 729 | { |
| 730 | struct mem_cgroup_per_zone *mz; |
Hugh Dickins | f156ab93 | 2012-03-21 16:34:19 -0700 | [diff] [blame] | 731 | enum lru_list lru; |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 732 | unsigned long ret = 0; |
| 733 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 734 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 735 | |
Hugh Dickins | f156ab93 | 2012-03-21 16:34:19 -0700 | [diff] [blame] | 736 | for_each_lru(lru) { |
| 737 | if (BIT(lru) & lru_mask) |
| 738 | ret += mz->lru_size[lru]; |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 739 | } |
| 740 | return ret; |
| 741 | } |
| 742 | |
| 743 | static unsigned long |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 744 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 745 | int nid, unsigned int lru_mask) |
| 746 | { |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 747 | u64 total = 0; |
| 748 | int zid; |
| 749 | |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 750 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 751 | total += mem_cgroup_zone_nr_lru_pages(memcg, |
| 752 | nid, zid, lru_mask); |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 753 | |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 754 | return total; |
| 755 | } |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 756 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 757 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 758 | unsigned int lru_mask) |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 759 | { |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 760 | int nid; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 761 | u64 total = 0; |
| 762 | |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 763 | for_each_node_state(nid, N_HIGH_MEMORY) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 764 | total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 765 | return total; |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 766 | } |
| 767 | |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 768 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
| 769 | enum mem_cgroup_events_target target) |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 770 | { |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 771 | unsigned long val, next; |
| 772 | |
Steven Rostedt | 4799401 | 2011-11-02 13:38:33 -0700 | [diff] [blame] | 773 | val = __this_cpu_read(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT]); |
| 774 | next = __this_cpu_read(memcg->stat->targets[target]); |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 775 | /* from time_after() in jiffies.h */ |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 776 | if ((long)next - (long)val < 0) { |
| 777 | switch (target) { |
| 778 | case MEM_CGROUP_TARGET_THRESH: |
| 779 | next = val + THRESHOLDS_EVENTS_TARGET; |
| 780 | break; |
| 781 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
| 782 | next = val + SOFTLIMIT_EVENTS_TARGET; |
| 783 | break; |
| 784 | case MEM_CGROUP_TARGET_NUMAINFO: |
| 785 | next = val + NUMAINFO_EVENTS_TARGET; |
| 786 | break; |
| 787 | default: |
| 788 | break; |
| 789 | } |
| 790 | __this_cpu_write(memcg->stat->targets[target], next); |
| 791 | return true; |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 792 | } |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 793 | return false; |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 794 | } |
| 795 | |
| 796 | /* |
| 797 | * Check events in order. |
| 798 | * |
| 799 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 800 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 801 | { |
Steven Rostedt | 4799401 | 2011-11-02 13:38:33 -0700 | [diff] [blame] | 802 | preempt_disable(); |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 803 | /* threshold event is triggered in finer grain than soft limit */ |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 804 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
| 805 | MEM_CGROUP_TARGET_THRESH))) { |
Andrew Morton | 82b3f2a | 2012-02-03 15:37:14 -0800 | [diff] [blame] | 806 | bool do_softlimit; |
| 807 | bool do_numainfo __maybe_unused; |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 808 | |
| 809 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
| 810 | MEM_CGROUP_TARGET_SOFTLIMIT); |
KAMEZAWA Hiroyuki | 453a9bf3 | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 811 | #if MAX_NUMNODES > 1 |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 812 | do_numainfo = mem_cgroup_event_ratelimit(memcg, |
| 813 | MEM_CGROUP_TARGET_NUMAINFO); |
KAMEZAWA Hiroyuki | 453a9bf3 | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 814 | #endif |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 815 | preempt_enable(); |
| 816 | |
| 817 | mem_cgroup_threshold(memcg); |
| 818 | if (unlikely(do_softlimit)) |
| 819 | mem_cgroup_update_tree(memcg, page); |
| 820 | #if MAX_NUMNODES > 1 |
| 821 | if (unlikely(do_numainfo)) |
| 822 | atomic_inc(&memcg->numainfo_events); |
| 823 | #endif |
| 824 | } else |
| 825 | preempt_enable(); |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 826 | } |
| 827 | |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 828 | struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 829 | { |
| 830 | return container_of(cgroup_subsys_state(cont, |
| 831 | mem_cgroup_subsys_id), struct mem_cgroup, |
| 832 | css); |
| 833 | } |
| 834 | |
Balbir Singh | cf475ad | 2008-04-29 01:00:16 -0700 | [diff] [blame] | 835 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 836 | { |
Balbir Singh | 31a78f2 | 2008-09-28 23:09:31 +0100 | [diff] [blame] | 837 | /* |
| 838 | * mm_update_next_owner() may clear mm->owner to NULL |
| 839 | * if it races with swapoff, page migration, etc. |
| 840 | * So this can be called with p == NULL. |
| 841 | */ |
| 842 | if (unlikely(!p)) |
| 843 | return NULL; |
| 844 | |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 845 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
| 846 | struct mem_cgroup, css); |
| 847 | } |
| 848 | |
KOSAKI Motohiro | a433658 | 2011-06-15 15:08:13 -0700 | [diff] [blame] | 849 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 850 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 851 | struct mem_cgroup *memcg = NULL; |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 852 | |
| 853 | if (!mm) |
| 854 | return NULL; |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 855 | /* |
| 856 | * Because we have no locks, mm->owner's may be being moved to other |
| 857 | * cgroup. We use css_tryget() here even if this looks |
| 858 | * pessimistic (rather than adding locks here). |
| 859 | */ |
| 860 | rcu_read_lock(); |
| 861 | do { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 862 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
| 863 | if (unlikely(!memcg)) |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 864 | break; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 865 | } while (!css_tryget(&memcg->css)); |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 866 | rcu_read_unlock(); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 867 | return memcg; |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 868 | } |
| 869 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 870 | /** |
| 871 | * mem_cgroup_iter - iterate over memory cgroup hierarchy |
| 872 | * @root: hierarchy root |
| 873 | * @prev: previously returned memcg, NULL on first invocation |
| 874 | * @reclaim: cookie for shared reclaim walks, NULL for full walks |
| 875 | * |
| 876 | * Returns references to children of the hierarchy below @root, or |
| 877 | * @root itself, or %NULL after a full round-trip. |
| 878 | * |
| 879 | * Caller must pass the return value in @prev on subsequent |
| 880 | * invocations for reference counting, or use mem_cgroup_iter_break() |
| 881 | * to cancel a hierarchy walk before the round-trip is complete. |
| 882 | * |
| 883 | * Reclaimers can specify a zone and a priority level in @reclaim to |
| 884 | * divide up the memcgs in the hierarchy among all concurrent |
| 885 | * reclaimers operating on the same zone and priority. |
| 886 | */ |
| 887 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
| 888 | struct mem_cgroup *prev, |
| 889 | struct mem_cgroup_reclaim_cookie *reclaim) |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 890 | { |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 891 | struct mem_cgroup *memcg = NULL; |
| 892 | int id = 0; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 893 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 894 | if (mem_cgroup_disabled()) |
| 895 | return NULL; |
| 896 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 897 | if (!root) |
| 898 | root = root_mem_cgroup; |
| 899 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 900 | if (prev && !reclaim) |
| 901 | id = css_id(&prev->css); |
| 902 | |
| 903 | if (prev && prev != root) |
| 904 | css_put(&prev->css); |
| 905 | |
| 906 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
| 907 | if (prev) |
| 908 | return NULL; |
| 909 | return root; |
| 910 | } |
| 911 | |
| 912 | while (!memcg) { |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 913 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 914 | struct cgroup_subsys_state *css; |
| 915 | |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 916 | if (reclaim) { |
| 917 | int nid = zone_to_nid(reclaim->zone); |
| 918 | int zid = zone_idx(reclaim->zone); |
| 919 | struct mem_cgroup_per_zone *mz; |
| 920 | |
| 921 | mz = mem_cgroup_zoneinfo(root, nid, zid); |
| 922 | iter = &mz->reclaim_iter[reclaim->priority]; |
| 923 | if (prev && reclaim->generation != iter->generation) |
| 924 | return NULL; |
| 925 | id = iter->position; |
| 926 | } |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 927 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 928 | rcu_read_lock(); |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 929 | css = css_get_next(&mem_cgroup_subsys, id + 1, &root->css, &id); |
| 930 | if (css) { |
| 931 | if (css == &root->css || css_tryget(css)) |
| 932 | memcg = container_of(css, |
| 933 | struct mem_cgroup, css); |
| 934 | } else |
| 935 | id = 0; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 936 | rcu_read_unlock(); |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 937 | |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 938 | if (reclaim) { |
| 939 | iter->position = id; |
| 940 | if (!css) |
| 941 | iter->generation++; |
| 942 | else if (!prev && memcg) |
| 943 | reclaim->generation = iter->generation; |
| 944 | } |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 945 | |
| 946 | if (prev && !css) |
| 947 | return NULL; |
| 948 | } |
| 949 | return memcg; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 950 | } |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 951 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 952 | /** |
| 953 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely |
| 954 | * @root: hierarchy root |
| 955 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() |
| 956 | */ |
| 957 | void mem_cgroup_iter_break(struct mem_cgroup *root, |
| 958 | struct mem_cgroup *prev) |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 959 | { |
| 960 | if (!root) |
| 961 | root = root_mem_cgroup; |
| 962 | if (prev && prev != root) |
| 963 | css_put(&prev->css); |
| 964 | } |
| 965 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 966 | /* |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 967 | * Iteration constructs for visiting all cgroups (under a tree). If |
| 968 | * loops are exited prematurely (break), mem_cgroup_iter_break() must |
| 969 | * be used for reference counting. |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 970 | */ |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 971 | #define for_each_mem_cgroup_tree(iter, root) \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 972 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 973 | iter != NULL; \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 974 | iter = mem_cgroup_iter(root, iter, NULL)) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 975 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 976 | #define for_each_mem_cgroup(iter) \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 977 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 978 | iter != NULL; \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 979 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 980 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 981 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 982 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 983 | return (memcg == root_mem_cgroup); |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 984 | } |
| 985 | |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 986 | void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
| 987 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 988 | struct mem_cgroup *memcg; |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 989 | |
| 990 | if (!mm) |
| 991 | return; |
| 992 | |
| 993 | rcu_read_lock(); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 994 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
| 995 | if (unlikely(!memcg)) |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 996 | goto out; |
| 997 | |
| 998 | switch (idx) { |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 999 | case PGFAULT: |
Johannes Weiner | 0e574a9 | 2012-01-12 17:18:35 -0800 | [diff] [blame] | 1000 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
| 1001 | break; |
| 1002 | case PGMAJFAULT: |
| 1003 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 1004 | break; |
| 1005 | default: |
| 1006 | BUG(); |
| 1007 | } |
| 1008 | out: |
| 1009 | rcu_read_unlock(); |
| 1010 | } |
| 1011 | EXPORT_SYMBOL(mem_cgroup_count_vm_event); |
| 1012 | |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1013 | /** |
| 1014 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg |
| 1015 | * @zone: zone of the wanted lruvec |
| 1016 | * @mem: memcg of the wanted lruvec |
| 1017 | * |
| 1018 | * Returns the lru list vector holding pages for the given @zone and |
| 1019 | * @mem. This can be the global zone lruvec, if the memory controller |
| 1020 | * is disabled. |
| 1021 | */ |
| 1022 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, |
| 1023 | struct mem_cgroup *memcg) |
| 1024 | { |
| 1025 | struct mem_cgroup_per_zone *mz; |
| 1026 | |
| 1027 | if (mem_cgroup_disabled()) |
| 1028 | return &zone->lruvec; |
| 1029 | |
| 1030 | mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone)); |
| 1031 | return &mz->lruvec; |
| 1032 | } |
| 1033 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 1034 | /* |
| 1035 | * Following LRU functions are allowed to be used without PCG_LOCK. |
| 1036 | * Operations are called by routine of global LRU independently from memcg. |
| 1037 | * What we have to take care of here is validness of pc->mem_cgroup. |
| 1038 | * |
| 1039 | * Changes to pc->mem_cgroup happens when |
| 1040 | * 1. charge |
| 1041 | * 2. moving account |
| 1042 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. |
| 1043 | * It is added to LRU before charge. |
| 1044 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. |
| 1045 | * When moving account, the page is not on LRU. It's isolated. |
| 1046 | */ |
| 1047 | |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1048 | /** |
| 1049 | * mem_cgroup_lru_add_list - account for adding an lru page and return lruvec |
| 1050 | * @zone: zone of the page |
| 1051 | * @page: the page |
| 1052 | * @lru: current lru |
| 1053 | * |
| 1054 | * This function accounts for @page being added to @lru, and returns |
| 1055 | * the lruvec for the given @zone and the memcg @page is charged to. |
| 1056 | * |
| 1057 | * The callsite is then responsible for physically linking the page to |
| 1058 | * the returned lruvec->lists[@lru]. |
Minchan Kim | 3f58a82 | 2011-03-22 16:32:53 -0700 | [diff] [blame] | 1059 | */ |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1060 | struct lruvec *mem_cgroup_lru_add_list(struct zone *zone, struct page *page, |
| 1061 | enum lru_list lru) |
Minchan Kim | 3f58a82 | 2011-03-22 16:32:53 -0700 | [diff] [blame] | 1062 | { |
| 1063 | struct mem_cgroup_per_zone *mz; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1064 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 1065 | struct page_cgroup *pc; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 1066 | |
Hirokazu Takahashi | f8d66542 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 1067 | if (mem_cgroup_disabled()) |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1068 | return &zone->lruvec; |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1069 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 1070 | pc = lookup_page_cgroup(page); |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 1071 | memcg = pc->mem_cgroup; |
Hugh Dickins | 7512102 | 2012-03-05 14:59:18 -0800 | [diff] [blame] | 1072 | |
| 1073 | /* |
| 1074 | * Surreptitiously switch any uncharged page to root: |
| 1075 | * an uncharged page off lru does nothing to secure |
| 1076 | * its former mem_cgroup from sudden removal. |
| 1077 | * |
| 1078 | * Our caller holds lru_lock, and PageCgroupUsed is updated |
| 1079 | * under page_cgroup lock: between them, they make all uses |
| 1080 | * of pc->mem_cgroup safe. |
| 1081 | */ |
| 1082 | if (!PageCgroupUsed(pc) && memcg != root_mem_cgroup) |
| 1083 | pc->mem_cgroup = memcg = root_mem_cgroup; |
| 1084 | |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1085 | mz = page_cgroup_zoneinfo(memcg, page); |
| 1086 | /* compound_order() is stabilized through lru_lock */ |
Hugh Dickins | 1eb4927 | 2012-03-21 16:34:19 -0700 | [diff] [blame] | 1087 | mz->lru_size[lru] += 1 << compound_order(page); |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1088 | return &mz->lruvec; |
| 1089 | } |
| 1090 | |
| 1091 | /** |
| 1092 | * mem_cgroup_lru_del_list - account for removing an lru page |
| 1093 | * @page: the page |
| 1094 | * @lru: target lru |
| 1095 | * |
| 1096 | * This function accounts for @page being removed from @lru. |
| 1097 | * |
| 1098 | * The callsite is then responsible for physically unlinking |
| 1099 | * @page->lru. |
| 1100 | */ |
| 1101 | void mem_cgroup_lru_del_list(struct page *page, enum lru_list lru) |
| 1102 | { |
| 1103 | struct mem_cgroup_per_zone *mz; |
| 1104 | struct mem_cgroup *memcg; |
| 1105 | struct page_cgroup *pc; |
| 1106 | |
| 1107 | if (mem_cgroup_disabled()) |
| 1108 | return; |
| 1109 | |
| 1110 | pc = lookup_page_cgroup(page); |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 1111 | memcg = pc->mem_cgroup; |
| 1112 | VM_BUG_ON(!memcg); |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1113 | mz = page_cgroup_zoneinfo(memcg, page); |
| 1114 | /* huge page split is done under lru_lock. so, we have no races. */ |
Hugh Dickins | 1eb4927 | 2012-03-21 16:34:19 -0700 | [diff] [blame] | 1115 | VM_BUG_ON(mz->lru_size[lru] < (1 << compound_order(page))); |
| 1116 | mz->lru_size[lru] -= 1 << compound_order(page); |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1117 | } |
| 1118 | |
| 1119 | void mem_cgroup_lru_del(struct page *page) |
| 1120 | { |
| 1121 | mem_cgroup_lru_del_list(page, page_lru(page)); |
| 1122 | } |
| 1123 | |
| 1124 | /** |
| 1125 | * mem_cgroup_lru_move_lists - account for moving a page between lrus |
| 1126 | * @zone: zone of the page |
| 1127 | * @page: the page |
| 1128 | * @from: current lru |
| 1129 | * @to: target lru |
| 1130 | * |
| 1131 | * This function accounts for @page being moved between the lrus @from |
| 1132 | * and @to, and returns the lruvec for the given @zone and the memcg |
| 1133 | * @page is charged to. |
| 1134 | * |
| 1135 | * The callsite is then responsible for physically relinking |
| 1136 | * @page->lru to the returned lruvec->lists[@to]. |
| 1137 | */ |
| 1138 | struct lruvec *mem_cgroup_lru_move_lists(struct zone *zone, |
| 1139 | struct page *page, |
| 1140 | enum lru_list from, |
| 1141 | enum lru_list to) |
| 1142 | { |
| 1143 | /* XXX: Optimize this, especially for @from == @to */ |
| 1144 | mem_cgroup_lru_del_list(page, from); |
| 1145 | return mem_cgroup_lru_add_list(zone, page, to); |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 1146 | } |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 1147 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 1148 | /* |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1149 | * Checks whether given mem is same or in the root_mem_cgroup's |
Michal Hocko | 3e92041 | 2011-07-26 16:08:29 -0700 | [diff] [blame] | 1150 | * hierarchy subtree |
| 1151 | */ |
Johannes Weiner | c3ac9a8 | 2012-05-29 15:06:25 -0700 | [diff] [blame^] | 1152 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
| 1153 | struct mem_cgroup *memcg) |
Michal Hocko | 3e92041 | 2011-07-26 16:08:29 -0700 | [diff] [blame] | 1154 | { |
Johannes Weiner | 91c63734 | 2012-05-29 15:06:24 -0700 | [diff] [blame] | 1155 | if (root_memcg == memcg) |
| 1156 | return true; |
| 1157 | if (!root_memcg->use_hierarchy) |
| 1158 | return false; |
Johannes Weiner | c3ac9a8 | 2012-05-29 15:06:25 -0700 | [diff] [blame^] | 1159 | return css_is_ancestor(&memcg->css, &root_memcg->css); |
| 1160 | } |
| 1161 | |
| 1162 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
| 1163 | struct mem_cgroup *memcg) |
| 1164 | { |
| 1165 | bool ret; |
| 1166 | |
Johannes Weiner | 91c63734 | 2012-05-29 15:06:24 -0700 | [diff] [blame] | 1167 | rcu_read_lock(); |
Johannes Weiner | c3ac9a8 | 2012-05-29 15:06:25 -0700 | [diff] [blame^] | 1168 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
Johannes Weiner | 91c63734 | 2012-05-29 15:06:24 -0700 | [diff] [blame] | 1169 | rcu_read_unlock(); |
| 1170 | return ret; |
Michal Hocko | 3e92041 | 2011-07-26 16:08:29 -0700 | [diff] [blame] | 1171 | } |
| 1172 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1173 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg) |
David Rientjes | 4c4a221 | 2008-02-07 00:14:06 -0800 | [diff] [blame] | 1174 | { |
| 1175 | int ret; |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1176 | struct mem_cgroup *curr = NULL; |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 1177 | struct task_struct *p; |
David Rientjes | 4c4a221 | 2008-02-07 00:14:06 -0800 | [diff] [blame] | 1178 | |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 1179 | p = find_lock_task_mm(task); |
David Rientjes | de077d2 | 2012-01-12 17:18:52 -0800 | [diff] [blame] | 1180 | if (p) { |
| 1181 | curr = try_get_mem_cgroup_from_mm(p->mm); |
| 1182 | task_unlock(p); |
| 1183 | } else { |
| 1184 | /* |
| 1185 | * All threads may have already detached their mm's, but the oom |
| 1186 | * killer still needs to detect if they have already been oom |
| 1187 | * killed to prevent needlessly killing additional tasks. |
| 1188 | */ |
| 1189 | task_lock(task); |
| 1190 | curr = mem_cgroup_from_task(task); |
| 1191 | if (curr) |
| 1192 | css_get(&curr->css); |
| 1193 | task_unlock(task); |
| 1194 | } |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1195 | if (!curr) |
| 1196 | return 0; |
Daisuke Nishimura | d31f56d | 2009-12-15 16:47:12 -0800 | [diff] [blame] | 1197 | /* |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1198 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
Daisuke Nishimura | d31f56d | 2009-12-15 16:47:12 -0800 | [diff] [blame] | 1199 | * use_hierarchy of "curr" here make this function true if hierarchy is |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1200 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
| 1201 | * hierarchy(even if use_hierarchy is disabled in "memcg"). |
Daisuke Nishimura | d31f56d | 2009-12-15 16:47:12 -0800 | [diff] [blame] | 1202 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1203 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1204 | css_put(&curr->css); |
David Rientjes | 4c4a221 | 2008-02-07 00:14:06 -0800 | [diff] [blame] | 1205 | return ret; |
| 1206 | } |
| 1207 | |
Johannes Weiner | 9b27297 | 2011-11-02 13:38:23 -0700 | [diff] [blame] | 1208 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg, struct zone *zone) |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1209 | { |
KOSAKI Motohiro | c772be9 | 2009-01-07 18:08:25 -0800 | [diff] [blame] | 1210 | unsigned long inactive_ratio; |
Johannes Weiner | 9b27297 | 2011-11-02 13:38:23 -0700 | [diff] [blame] | 1211 | int nid = zone_to_nid(zone); |
| 1212 | int zid = zone_idx(zone); |
| 1213 | unsigned long inactive; |
| 1214 | unsigned long active; |
| 1215 | unsigned long gb; |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1216 | |
Johannes Weiner | 9b27297 | 2011-11-02 13:38:23 -0700 | [diff] [blame] | 1217 | inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, |
| 1218 | BIT(LRU_INACTIVE_ANON)); |
| 1219 | active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, |
| 1220 | BIT(LRU_ACTIVE_ANON)); |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1221 | |
KOSAKI Motohiro | c772be9 | 2009-01-07 18:08:25 -0800 | [diff] [blame] | 1222 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
| 1223 | if (gb) |
| 1224 | inactive_ratio = int_sqrt(10 * gb); |
| 1225 | else |
| 1226 | inactive_ratio = 1; |
| 1227 | |
Johannes Weiner | 9b27297 | 2011-11-02 13:38:23 -0700 | [diff] [blame] | 1228 | return inactive * inactive_ratio < active; |
KOSAKI Motohiro | c772be9 | 2009-01-07 18:08:25 -0800 | [diff] [blame] | 1229 | } |
| 1230 | |
Johannes Weiner | 9b27297 | 2011-11-02 13:38:23 -0700 | [diff] [blame] | 1231 | int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg, struct zone *zone) |
KOSAKI Motohiro | c772be9 | 2009-01-07 18:08:25 -0800 | [diff] [blame] | 1232 | { |
| 1233 | unsigned long active; |
| 1234 | unsigned long inactive; |
Johannes Weiner | 9b27297 | 2011-11-02 13:38:23 -0700 | [diff] [blame] | 1235 | int zid = zone_idx(zone); |
| 1236 | int nid = zone_to_nid(zone); |
KOSAKI Motohiro | c772be9 | 2009-01-07 18:08:25 -0800 | [diff] [blame] | 1237 | |
Johannes Weiner | 9b27297 | 2011-11-02 13:38:23 -0700 | [diff] [blame] | 1238 | inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, |
| 1239 | BIT(LRU_INACTIVE_FILE)); |
| 1240 | active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, |
| 1241 | BIT(LRU_ACTIVE_FILE)); |
Rik van Riel | 56e49d2 | 2009-06-16 15:32:28 -0700 | [diff] [blame] | 1242 | |
| 1243 | return (active > inactive); |
| 1244 | } |
| 1245 | |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 1246 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, |
| 1247 | struct zone *zone) |
| 1248 | { |
KOSAKI Motohiro | 13d7e3a | 2010-08-10 18:03:06 -0700 | [diff] [blame] | 1249 | int nid = zone_to_nid(zone); |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 1250 | int zid = zone_idx(zone); |
| 1251 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
| 1252 | |
| 1253 | return &mz->reclaim_stat; |
| 1254 | } |
| 1255 | |
| 1256 | struct zone_reclaim_stat * |
| 1257 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) |
| 1258 | { |
| 1259 | struct page_cgroup *pc; |
| 1260 | struct mem_cgroup_per_zone *mz; |
| 1261 | |
| 1262 | if (mem_cgroup_disabled()) |
| 1263 | return NULL; |
| 1264 | |
| 1265 | pc = lookup_page_cgroup(page); |
Daisuke Nishimura | bd112db | 2009-01-15 13:51:11 -0800 | [diff] [blame] | 1266 | if (!PageCgroupUsed(pc)) |
| 1267 | return NULL; |
Johannes Weiner | 713735b | 2011-01-20 14:44:31 -0800 | [diff] [blame] | 1268 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ |
| 1269 | smp_rmb(); |
Johannes Weiner | 97a6c37 | 2011-03-23 16:42:27 -0700 | [diff] [blame] | 1270 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 1271 | return &mz->reclaim_stat; |
| 1272 | } |
| 1273 | |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1274 | #define mem_cgroup_from_res_counter(counter, member) \ |
| 1275 | container_of(counter, struct mem_cgroup, member) |
| 1276 | |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1277 | /** |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1278 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
| 1279 | * @mem: the memory cgroup |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1280 | * |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1281 | * Returns the maximum amount of memory @mem can be charged with, in |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 1282 | * pages. |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1283 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1284 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1285 | { |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1286 | unsigned long long margin; |
| 1287 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1288 | margin = res_counter_margin(&memcg->res); |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1289 | if (do_swap_account) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1290 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 1291 | return margin >> PAGE_SHIFT; |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1292 | } |
| 1293 | |
KAMEZAWA Hiroyuki | 1f4c025 | 2011-07-26 16:08:21 -0700 | [diff] [blame] | 1294 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 1295 | { |
| 1296 | struct cgroup *cgrp = memcg->css.cgroup; |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 1297 | |
| 1298 | /* root ? */ |
| 1299 | if (cgrp->parent == NULL) |
| 1300 | return vm_swappiness; |
| 1301 | |
Johannes Weiner | bf1ff26 | 2011-03-23 16:42:32 -0700 | [diff] [blame] | 1302 | return memcg->swappiness; |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 1303 | } |
| 1304 | |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 1305 | /* |
| 1306 | * memcg->moving_account is used for checking possibility that some thread is |
| 1307 | * calling move_account(). When a thread on CPU-A starts moving pages under |
| 1308 | * a memcg, other threads should check memcg->moving_account under |
| 1309 | * rcu_read_lock(), like this: |
| 1310 | * |
| 1311 | * CPU-A CPU-B |
| 1312 | * rcu_read_lock() |
| 1313 | * memcg->moving_account+1 if (memcg->mocing_account) |
| 1314 | * take heavy locks. |
| 1315 | * synchronize_rcu() update something. |
| 1316 | * rcu_read_unlock() |
| 1317 | * start move here. |
| 1318 | */ |
KAMEZAWA Hiroyuki | 4331f7d | 2012-03-21 16:34:26 -0700 | [diff] [blame] | 1319 | |
| 1320 | /* for quick checking without looking up memcg */ |
| 1321 | atomic_t memcg_moving __read_mostly; |
| 1322 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1323 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1324 | { |
KAMEZAWA Hiroyuki | 4331f7d | 2012-03-21 16:34:26 -0700 | [diff] [blame] | 1325 | atomic_inc(&memcg_moving); |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 1326 | atomic_inc(&memcg->moving_account); |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1327 | synchronize_rcu(); |
| 1328 | } |
| 1329 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1330 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1331 | { |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 1332 | /* |
| 1333 | * Now, mem_cgroup_clear_mc() may call this function with NULL. |
| 1334 | * We check NULL in callee rather than caller. |
| 1335 | */ |
KAMEZAWA Hiroyuki | 4331f7d | 2012-03-21 16:34:26 -0700 | [diff] [blame] | 1336 | if (memcg) { |
| 1337 | atomic_dec(&memcg_moving); |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 1338 | atomic_dec(&memcg->moving_account); |
KAMEZAWA Hiroyuki | 4331f7d | 2012-03-21 16:34:26 -0700 | [diff] [blame] | 1339 | } |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1340 | } |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 1341 | |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1342 | /* |
| 1343 | * 2 routines for checking "mem" is under move_account() or not. |
| 1344 | * |
Andrew Morton | 13fd1dd9 | 2012-03-21 16:34:26 -0700 | [diff] [blame] | 1345 | * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This |
| 1346 | * is used for avoiding races in accounting. If true, |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1347 | * pc->mem_cgroup may be overwritten. |
| 1348 | * |
| 1349 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or |
| 1350 | * under hierarchy of moving cgroups. This is for |
| 1351 | * waiting at hith-memory prressure caused by "move". |
| 1352 | */ |
| 1353 | |
Andrew Morton | 13fd1dd9 | 2012-03-21 16:34:26 -0700 | [diff] [blame] | 1354 | static bool mem_cgroup_stolen(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1355 | { |
| 1356 | VM_BUG_ON(!rcu_read_lock_held()); |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 1357 | return atomic_read(&memcg->moving_account) > 0; |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1358 | } |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1359 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1360 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1361 | { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1362 | struct mem_cgroup *from; |
| 1363 | struct mem_cgroup *to; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1364 | bool ret = false; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1365 | /* |
| 1366 | * Unlike task_move routines, we access mc.to, mc.from not under |
| 1367 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. |
| 1368 | */ |
| 1369 | spin_lock(&mc.lock); |
| 1370 | from = mc.from; |
| 1371 | to = mc.to; |
| 1372 | if (!from) |
| 1373 | goto unlock; |
Michal Hocko | 3e92041 | 2011-07-26 16:08:29 -0700 | [diff] [blame] | 1374 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1375 | ret = mem_cgroup_same_or_subtree(memcg, from) |
| 1376 | || mem_cgroup_same_or_subtree(memcg, to); |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1377 | unlock: |
| 1378 | spin_unlock(&mc.lock); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1379 | return ret; |
| 1380 | } |
| 1381 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1382 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1383 | { |
| 1384 | if (mc.moving_task && current != mc.moving_task) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1385 | if (mem_cgroup_under_move(memcg)) { |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1386 | DEFINE_WAIT(wait); |
| 1387 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); |
| 1388 | /* moving charge context might have finished. */ |
| 1389 | if (mc.moving_task) |
| 1390 | schedule(); |
| 1391 | finish_wait(&mc.waitq, &wait); |
| 1392 | return true; |
| 1393 | } |
| 1394 | } |
| 1395 | return false; |
| 1396 | } |
| 1397 | |
KAMEZAWA Hiroyuki | 312734c0 | 2012-03-21 16:34:24 -0700 | [diff] [blame] | 1398 | /* |
| 1399 | * Take this lock when |
| 1400 | * - a code tries to modify page's memcg while it's USED. |
| 1401 | * - a code tries to modify page state accounting in a memcg. |
Andrew Morton | 13fd1dd9 | 2012-03-21 16:34:26 -0700 | [diff] [blame] | 1402 | * see mem_cgroup_stolen(), too. |
KAMEZAWA Hiroyuki | 312734c0 | 2012-03-21 16:34:24 -0700 | [diff] [blame] | 1403 | */ |
| 1404 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, |
| 1405 | unsigned long *flags) |
| 1406 | { |
| 1407 | spin_lock_irqsave(&memcg->move_lock, *flags); |
| 1408 | } |
| 1409 | |
| 1410 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, |
| 1411 | unsigned long *flags) |
| 1412 | { |
| 1413 | spin_unlock_irqrestore(&memcg->move_lock, *flags); |
| 1414 | } |
| 1415 | |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1416 | /** |
Kirill A. Shutemov | 6a6135b | 2010-03-10 15:22:25 -0800 | [diff] [blame] | 1417 | * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode. |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1418 | * @memcg: The memory cgroup that went over limit |
| 1419 | * @p: Task that is going to be killed |
| 1420 | * |
| 1421 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is |
| 1422 | * enabled |
| 1423 | */ |
| 1424 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) |
| 1425 | { |
| 1426 | struct cgroup *task_cgrp; |
| 1427 | struct cgroup *mem_cgrp; |
| 1428 | /* |
| 1429 | * Need a buffer in BSS, can't rely on allocations. The code relies |
| 1430 | * on the assumption that OOM is serialized for memory controller. |
| 1431 | * If this assumption is broken, revisit this code. |
| 1432 | */ |
| 1433 | static char memcg_name[PATH_MAX]; |
| 1434 | int ret; |
| 1435 | |
Daisuke Nishimura | d31f56d | 2009-12-15 16:47:12 -0800 | [diff] [blame] | 1436 | if (!memcg || !p) |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1437 | return; |
| 1438 | |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1439 | rcu_read_lock(); |
| 1440 | |
| 1441 | mem_cgrp = memcg->css.cgroup; |
| 1442 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); |
| 1443 | |
| 1444 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); |
| 1445 | if (ret < 0) { |
| 1446 | /* |
| 1447 | * Unfortunately, we are unable to convert to a useful name |
| 1448 | * But we'll still print out the usage information |
| 1449 | */ |
| 1450 | rcu_read_unlock(); |
| 1451 | goto done; |
| 1452 | } |
| 1453 | rcu_read_unlock(); |
| 1454 | |
| 1455 | printk(KERN_INFO "Task in %s killed", memcg_name); |
| 1456 | |
| 1457 | rcu_read_lock(); |
| 1458 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); |
| 1459 | if (ret < 0) { |
| 1460 | rcu_read_unlock(); |
| 1461 | goto done; |
| 1462 | } |
| 1463 | rcu_read_unlock(); |
| 1464 | |
| 1465 | /* |
| 1466 | * Continues from above, so we don't need an KERN_ level |
| 1467 | */ |
| 1468 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); |
| 1469 | done: |
| 1470 | |
| 1471 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", |
| 1472 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, |
| 1473 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, |
| 1474 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); |
| 1475 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " |
| 1476 | "failcnt %llu\n", |
| 1477 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, |
| 1478 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, |
| 1479 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); |
| 1480 | } |
| 1481 | |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1482 | /* |
| 1483 | * This function returns the number of memcg under hierarchy tree. Returns |
| 1484 | * 1(self count) if no children. |
| 1485 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1486 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1487 | { |
| 1488 | int num = 0; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1489 | struct mem_cgroup *iter; |
| 1490 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1491 | for_each_mem_cgroup_tree(iter, memcg) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1492 | num++; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1493 | return num; |
| 1494 | } |
| 1495 | |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1496 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1497 | * Return the memory (and swap, if configured) limit for a memcg. |
| 1498 | */ |
| 1499 | u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) |
| 1500 | { |
| 1501 | u64 limit; |
| 1502 | u64 memsw; |
| 1503 | |
Johannes Weiner | f3e8eb7 | 2011-01-13 15:47:39 -0800 | [diff] [blame] | 1504 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
| 1505 | limit += total_swap_pages << PAGE_SHIFT; |
| 1506 | |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1507 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
| 1508 | /* |
| 1509 | * If memsw is finite and limits the amount of swap space available |
| 1510 | * to this memcg, return that limit. |
| 1511 | */ |
| 1512 | return min(limit, memsw); |
| 1513 | } |
| 1514 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1515 | static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg, |
| 1516 | gfp_t gfp_mask, |
| 1517 | unsigned long flags) |
| 1518 | { |
| 1519 | unsigned long total = 0; |
| 1520 | bool noswap = false; |
| 1521 | int loop; |
| 1522 | |
| 1523 | if (flags & MEM_CGROUP_RECLAIM_NOSWAP) |
| 1524 | noswap = true; |
| 1525 | if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum) |
| 1526 | noswap = true; |
| 1527 | |
| 1528 | for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) { |
| 1529 | if (loop) |
| 1530 | drain_all_stock_async(memcg); |
| 1531 | total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap); |
| 1532 | /* |
| 1533 | * Allow limit shrinkers, which are triggered directly |
| 1534 | * by userspace, to catch signals and stop reclaim |
| 1535 | * after minimal progress, regardless of the margin. |
| 1536 | */ |
| 1537 | if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK)) |
| 1538 | break; |
| 1539 | if (mem_cgroup_margin(memcg)) |
| 1540 | break; |
| 1541 | /* |
| 1542 | * If nothing was reclaimed after two attempts, there |
| 1543 | * may be no reclaimable pages in this hierarchy. |
| 1544 | */ |
| 1545 | if (loop && !total) |
| 1546 | break; |
| 1547 | } |
| 1548 | return total; |
| 1549 | } |
| 1550 | |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1551 | /** |
| 1552 | * test_mem_cgroup_node_reclaimable |
| 1553 | * @mem: the target memcg |
| 1554 | * @nid: the node ID to be checked. |
| 1555 | * @noswap : specify true here if the user wants flle only information. |
| 1556 | * |
| 1557 | * This function returns whether the specified memcg contains any |
| 1558 | * reclaimable pages on a node. Returns true if there are any reclaimable |
| 1559 | * pages in the node. |
| 1560 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1561 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1562 | int nid, bool noswap) |
| 1563 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1564 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1565 | return true; |
| 1566 | if (noswap || !total_swap_pages) |
| 1567 | return false; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1568 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1569 | return true; |
| 1570 | return false; |
| 1571 | |
| 1572 | } |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1573 | #if MAX_NUMNODES > 1 |
| 1574 | |
| 1575 | /* |
| 1576 | * Always updating the nodemask is not very good - even if we have an empty |
| 1577 | * list or the wrong list here, we can start from some node and traverse all |
| 1578 | * nodes based on the zonelist. So update the list loosely once per 10 secs. |
| 1579 | * |
| 1580 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1581 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1582 | { |
| 1583 | int nid; |
KAMEZAWA Hiroyuki | 453a9bf3 | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 1584 | /* |
| 1585 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET |
| 1586 | * pagein/pageout changes since the last update. |
| 1587 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1588 | if (!atomic_read(&memcg->numainfo_events)) |
KAMEZAWA Hiroyuki | 453a9bf3 | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 1589 | return; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1590 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1591 | return; |
| 1592 | |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1593 | /* make a nodemask where this memcg uses memory from */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1594 | memcg->scan_nodes = node_states[N_HIGH_MEMORY]; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1595 | |
| 1596 | for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) { |
| 1597 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1598 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
| 1599 | node_clear(nid, memcg->scan_nodes); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1600 | } |
KAMEZAWA Hiroyuki | 453a9bf3 | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 1601 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1602 | atomic_set(&memcg->numainfo_events, 0); |
| 1603 | atomic_set(&memcg->numainfo_updating, 0); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1604 | } |
| 1605 | |
| 1606 | /* |
| 1607 | * Selecting a node where we start reclaim from. Because what we need is just |
| 1608 | * reducing usage counter, start from anywhere is O,K. Considering |
| 1609 | * memory reclaim from current node, there are pros. and cons. |
| 1610 | * |
| 1611 | * Freeing memory from current node means freeing memory from a node which |
| 1612 | * we'll use or we've used. So, it may make LRU bad. And if several threads |
| 1613 | * hit limits, it will see a contention on a node. But freeing from remote |
| 1614 | * node means more costs for memory reclaim because of memory latency. |
| 1615 | * |
| 1616 | * Now, we use round-robin. Better algorithm is welcomed. |
| 1617 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1618 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1619 | { |
| 1620 | int node; |
| 1621 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1622 | mem_cgroup_may_update_nodemask(memcg); |
| 1623 | node = memcg->last_scanned_node; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1624 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1625 | node = next_node(node, memcg->scan_nodes); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1626 | if (node == MAX_NUMNODES) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1627 | node = first_node(memcg->scan_nodes); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1628 | /* |
| 1629 | * We call this when we hit limit, not when pages are added to LRU. |
| 1630 | * No LRU may hold pages because all pages are UNEVICTABLE or |
| 1631 | * memcg is too small and all pages are not on LRU. In that case, |
| 1632 | * we use curret node. |
| 1633 | */ |
| 1634 | if (unlikely(node == MAX_NUMNODES)) |
| 1635 | node = numa_node_id(); |
| 1636 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1637 | memcg->last_scanned_node = node; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1638 | return node; |
| 1639 | } |
| 1640 | |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1641 | /* |
| 1642 | * Check all nodes whether it contains reclaimable pages or not. |
| 1643 | * For quick scan, we make use of scan_nodes. This will allow us to skip |
| 1644 | * unused nodes. But scan_nodes is lazily updated and may not cotain |
| 1645 | * enough new information. We need to do double check. |
| 1646 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1647 | bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1648 | { |
| 1649 | int nid; |
| 1650 | |
| 1651 | /* |
| 1652 | * quick check...making use of scan_node. |
| 1653 | * We can skip unused nodes. |
| 1654 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1655 | if (!nodes_empty(memcg->scan_nodes)) { |
| 1656 | for (nid = first_node(memcg->scan_nodes); |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1657 | nid < MAX_NUMNODES; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1658 | nid = next_node(nid, memcg->scan_nodes)) { |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1659 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1660 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1661 | return true; |
| 1662 | } |
| 1663 | } |
| 1664 | /* |
| 1665 | * Check rest of nodes. |
| 1666 | */ |
| 1667 | for_each_node_state(nid, N_HIGH_MEMORY) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1668 | if (node_isset(nid, memcg->scan_nodes)) |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1669 | continue; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1670 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1671 | return true; |
| 1672 | } |
| 1673 | return false; |
| 1674 | } |
| 1675 | |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1676 | #else |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1677 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1678 | { |
| 1679 | return 0; |
| 1680 | } |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1681 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1682 | bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1683 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1684 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1685 | } |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1686 | #endif |
| 1687 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1688 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
| 1689 | struct zone *zone, |
| 1690 | gfp_t gfp_mask, |
| 1691 | unsigned long *total_scanned) |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1692 | { |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1693 | struct mem_cgroup *victim = NULL; |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1694 | int total = 0; |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1695 | int loop = 0; |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1696 | unsigned long excess; |
Johannes Weiner | 185efc0 | 2011-09-14 16:21:58 -0700 | [diff] [blame] | 1697 | unsigned long nr_scanned; |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 1698 | struct mem_cgroup_reclaim_cookie reclaim = { |
| 1699 | .zone = zone, |
| 1700 | .priority = 0, |
| 1701 | }; |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1702 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1703 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1704 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1705 | while (1) { |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 1706 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1707 | if (!victim) { |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1708 | loop++; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1709 | if (loop >= 2) { |
| 1710 | /* |
| 1711 | * If we have not been able to reclaim |
| 1712 | * anything, it might because there are |
| 1713 | * no reclaimable pages under this hierarchy |
| 1714 | */ |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1715 | if (!total) |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1716 | break; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1717 | /* |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 1718 | * We want to do more targeted reclaim. |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1719 | * excess >> 2 is not to excessive so as to |
| 1720 | * reclaim too much, nor too less that we keep |
| 1721 | * coming back to reclaim from this cgroup |
| 1722 | */ |
| 1723 | if (total >= (excess >> 2) || |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1724 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1725 | break; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1726 | } |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1727 | continue; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1728 | } |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1729 | if (!mem_cgroup_reclaimable(victim, false)) |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1730 | continue; |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1731 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
| 1732 | zone, &nr_scanned); |
| 1733 | *total_scanned += nr_scanned; |
| 1734 | if (!res_counter_soft_limit_excess(&root_memcg->res)) |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1735 | break; |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1736 | } |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1737 | mem_cgroup_iter_break(root_memcg, victim); |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1738 | return total; |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1739 | } |
| 1740 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1741 | /* |
| 1742 | * Check OOM-Killer is already running under our hierarchy. |
| 1743 | * If someone is running, return false. |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 1744 | * Has to be called with memcg_oom_lock |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1745 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1746 | static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1747 | { |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1748 | struct mem_cgroup *iter, *failed = NULL; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 1749 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1750 | for_each_mem_cgroup_tree(iter, memcg) { |
Johannes Weiner | 23751be | 2011-08-25 15:59:16 -0700 | [diff] [blame] | 1751 | if (iter->oom_lock) { |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1752 | /* |
| 1753 | * this subtree of our hierarchy is already locked |
| 1754 | * so we cannot give a lock. |
| 1755 | */ |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1756 | failed = iter; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1757 | mem_cgroup_iter_break(memcg, iter); |
| 1758 | break; |
Johannes Weiner | 23751be | 2011-08-25 15:59:16 -0700 | [diff] [blame] | 1759 | } else |
| 1760 | iter->oom_lock = true; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1761 | } |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1762 | |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1763 | if (!failed) |
Johannes Weiner | 23751be | 2011-08-25 15:59:16 -0700 | [diff] [blame] | 1764 | return true; |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1765 | |
| 1766 | /* |
| 1767 | * OK, we failed to lock the whole subtree so we have to clean up |
| 1768 | * what we set up to the failing subtree |
| 1769 | */ |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1770 | for_each_mem_cgroup_tree(iter, memcg) { |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1771 | if (iter == failed) { |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1772 | mem_cgroup_iter_break(memcg, iter); |
| 1773 | break; |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1774 | } |
| 1775 | iter->oom_lock = false; |
| 1776 | } |
Johannes Weiner | 23751be | 2011-08-25 15:59:16 -0700 | [diff] [blame] | 1777 | return false; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 1778 | } |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1779 | |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1780 | /* |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 1781 | * Has to be called with memcg_oom_lock |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1782 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1783 | static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1784 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1785 | struct mem_cgroup *iter; |
| 1786 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1787 | for_each_mem_cgroup_tree(iter, memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1788 | iter->oom_lock = false; |
| 1789 | return 0; |
| 1790 | } |
| 1791 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1792 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1793 | { |
| 1794 | struct mem_cgroup *iter; |
| 1795 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1796 | for_each_mem_cgroup_tree(iter, memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1797 | atomic_inc(&iter->under_oom); |
| 1798 | } |
| 1799 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1800 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1801 | { |
| 1802 | struct mem_cgroup *iter; |
| 1803 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1804 | /* |
| 1805 | * When a new child is created while the hierarchy is under oom, |
| 1806 | * mem_cgroup_oom_lock() may not be called. We have to use |
| 1807 | * atomic_add_unless() here. |
| 1808 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1809 | for_each_mem_cgroup_tree(iter, memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1810 | atomic_add_unless(&iter->under_oom, -1, 0); |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1811 | } |
| 1812 | |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 1813 | static DEFINE_SPINLOCK(memcg_oom_lock); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1814 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
| 1815 | |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1816 | struct oom_wait_info { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 1817 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1818 | wait_queue_t wait; |
| 1819 | }; |
| 1820 | |
| 1821 | static int memcg_oom_wake_function(wait_queue_t *wait, |
| 1822 | unsigned mode, int sync, void *arg) |
| 1823 | { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 1824 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
| 1825 | struct mem_cgroup *oom_wait_memcg; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1826 | struct oom_wait_info *oom_wait_info; |
| 1827 | |
| 1828 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 1829 | oom_wait_memcg = oom_wait_info->memcg; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1830 | |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1831 | /* |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 1832 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1833 | * Then we can use css_is_ancestor without taking care of RCU. |
| 1834 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1835 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
| 1836 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1837 | return 0; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1838 | return autoremove_wake_function(wait, mode, sync, arg); |
| 1839 | } |
| 1840 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1841 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1842 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1843 | /* for filtering, pass "memcg" as argument. */ |
| 1844 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1845 | } |
| 1846 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1847 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1848 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1849 | if (memcg && atomic_read(&memcg->under_oom)) |
| 1850 | memcg_wakeup_oom(memcg); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1851 | } |
| 1852 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1853 | /* |
| 1854 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. |
| 1855 | */ |
David Rientjes | e845e19 | 2012-03-21 16:34:10 -0700 | [diff] [blame] | 1856 | bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1857 | { |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1858 | struct oom_wait_info owait; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1859 | bool locked, need_to_kill; |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1860 | |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 1861 | owait.memcg = memcg; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1862 | owait.wait.flags = 0; |
| 1863 | owait.wait.func = memcg_oom_wake_function; |
| 1864 | owait.wait.private = current; |
| 1865 | INIT_LIST_HEAD(&owait.wait.task_list); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1866 | need_to_kill = true; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1867 | mem_cgroup_mark_under_oom(memcg); |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1868 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1869 | /* At first, try to OOM lock hierarchy under memcg.*/ |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 1870 | spin_lock(&memcg_oom_lock); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1871 | locked = mem_cgroup_oom_lock(memcg); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1872 | /* |
| 1873 | * Even if signal_pending(), we can't quit charge() loop without |
| 1874 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL |
| 1875 | * under OOM is always welcomed, use TASK_KILLABLE here. |
| 1876 | */ |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1877 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1878 | if (!locked || memcg->oom_kill_disable) |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1879 | need_to_kill = false; |
| 1880 | if (locked) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1881 | mem_cgroup_oom_notify(memcg); |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 1882 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1883 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1884 | if (need_to_kill) { |
| 1885 | finish_wait(&memcg_oom_waitq, &owait.wait); |
David Rientjes | e845e19 | 2012-03-21 16:34:10 -0700 | [diff] [blame] | 1886 | mem_cgroup_out_of_memory(memcg, mask, order); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1887 | } else { |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1888 | schedule(); |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1889 | finish_wait(&memcg_oom_waitq, &owait.wait); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1890 | } |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 1891 | spin_lock(&memcg_oom_lock); |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1892 | if (locked) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1893 | mem_cgroup_oom_unlock(memcg); |
| 1894 | memcg_wakeup_oom(memcg); |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 1895 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1896 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1897 | mem_cgroup_unmark_under_oom(memcg); |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1898 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1899 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) |
| 1900 | return false; |
| 1901 | /* Give chance to dying process */ |
KAMEZAWA Hiroyuki | 715a5ee | 2011-11-02 13:38:18 -0700 | [diff] [blame] | 1902 | schedule_timeout_uninterruptible(1); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1903 | return true; |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1904 | } |
| 1905 | |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1906 | /* |
| 1907 | * Currently used to update mapped file statistics, but the routine can be |
| 1908 | * generalized to update other statistics as well. |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1909 | * |
| 1910 | * Notes: Race condition |
| 1911 | * |
| 1912 | * We usually use page_cgroup_lock() for accessing page_cgroup member but |
| 1913 | * it tends to be costly. But considering some conditions, we doesn't need |
| 1914 | * to do so _always_. |
| 1915 | * |
| 1916 | * Considering "charge", lock_page_cgroup() is not required because all |
| 1917 | * file-stat operations happen after a page is attached to radix-tree. There |
| 1918 | * are no race with "charge". |
| 1919 | * |
| 1920 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup |
| 1921 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even |
| 1922 | * if there are race with "uncharge". Statistics itself is properly handled |
| 1923 | * by flags. |
| 1924 | * |
| 1925 | * Considering "move", this is an only case we see a race. To make the race |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 1926 | * small, we check mm->moving_account and detect there are possibility of race |
| 1927 | * If there is, we take a lock. |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1928 | */ |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 1929 | |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1930 | void __mem_cgroup_begin_update_page_stat(struct page *page, |
| 1931 | bool *locked, unsigned long *flags) |
| 1932 | { |
| 1933 | struct mem_cgroup *memcg; |
| 1934 | struct page_cgroup *pc; |
| 1935 | |
| 1936 | pc = lookup_page_cgroup(page); |
| 1937 | again: |
| 1938 | memcg = pc->mem_cgroup; |
| 1939 | if (unlikely(!memcg || !PageCgroupUsed(pc))) |
| 1940 | return; |
| 1941 | /* |
| 1942 | * If this memory cgroup is not under account moving, we don't |
| 1943 | * need to take move_lock_page_cgroup(). Because we already hold |
| 1944 | * rcu_read_lock(), any calls to move_account will be delayed until |
Andrew Morton | 13fd1dd9 | 2012-03-21 16:34:26 -0700 | [diff] [blame] | 1945 | * rcu_read_unlock() if mem_cgroup_stolen() == true. |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1946 | */ |
Andrew Morton | 13fd1dd9 | 2012-03-21 16:34:26 -0700 | [diff] [blame] | 1947 | if (!mem_cgroup_stolen(memcg)) |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1948 | return; |
| 1949 | |
| 1950 | move_lock_mem_cgroup(memcg, flags); |
| 1951 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { |
| 1952 | move_unlock_mem_cgroup(memcg, flags); |
| 1953 | goto again; |
| 1954 | } |
| 1955 | *locked = true; |
| 1956 | } |
| 1957 | |
| 1958 | void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags) |
| 1959 | { |
| 1960 | struct page_cgroup *pc = lookup_page_cgroup(page); |
| 1961 | |
| 1962 | /* |
| 1963 | * It's guaranteed that pc->mem_cgroup never changes while |
| 1964 | * lock is held because a routine modifies pc->mem_cgroup |
| 1965 | * should take move_lock_page_cgroup(). |
| 1966 | */ |
| 1967 | move_unlock_mem_cgroup(pc->mem_cgroup, flags); |
| 1968 | } |
| 1969 | |
Greg Thelen | 2a7106f | 2011-01-13 15:47:37 -0800 | [diff] [blame] | 1970 | void mem_cgroup_update_page_stat(struct page *page, |
| 1971 | enum mem_cgroup_page_stat_item idx, int val) |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1972 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1973 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1974 | struct page_cgroup *pc = lookup_page_cgroup(page); |
KAMEZAWA Hiroyuki | dbd4ea7 | 2011-01-13 15:47:38 -0800 | [diff] [blame] | 1975 | unsigned long uninitialized_var(flags); |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1976 | |
Johannes Weiner | cfa4494 | 2012-01-12 17:18:38 -0800 | [diff] [blame] | 1977 | if (mem_cgroup_disabled()) |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1978 | return; |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1979 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1980 | memcg = pc->mem_cgroup; |
| 1981 | if (unlikely(!memcg || !PageCgroupUsed(pc))) |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1982 | return; |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 1983 | |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 1984 | switch (idx) { |
Greg Thelen | 2a7106f | 2011-01-13 15:47:37 -0800 | [diff] [blame] | 1985 | case MEMCG_NR_FILE_MAPPED: |
Greg Thelen | 2a7106f | 2011-01-13 15:47:37 -0800 | [diff] [blame] | 1986 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 1987 | break; |
| 1988 | default: |
| 1989 | BUG(); |
KAMEZAWA Hiroyuki | 8725d54 | 2010-04-06 14:35:05 -0700 | [diff] [blame] | 1990 | } |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1991 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1992 | this_cpu_add(memcg->stat->count[idx], val); |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1993 | } |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 1994 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 1995 | /* |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1996 | * size of first charge trial. "32" comes from vmscan.c's magic value. |
| 1997 | * TODO: maybe necessary to use big numbers in big irons. |
| 1998 | */ |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 1999 | #define CHARGE_BATCH 32U |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2000 | struct memcg_stock_pcp { |
| 2001 | struct mem_cgroup *cached; /* this never be root cgroup */ |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 2002 | unsigned int nr_pages; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2003 | struct work_struct work; |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2004 | unsigned long flags; |
| 2005 | #define FLUSHING_CACHED_CHARGE (0) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2006 | }; |
| 2007 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); |
Michal Hocko | 9f50fad | 2011-08-09 11:56:26 +0200 | [diff] [blame] | 2008 | static DEFINE_MUTEX(percpu_charge_mutex); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2009 | |
| 2010 | /* |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 2011 | * Try to consume stocked charge on this cpu. If success, one page is consumed |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2012 | * from local stock and true is returned. If the stock is 0 or charges from a |
| 2013 | * cgroup which is not current target, returns false. This stock will be |
| 2014 | * refilled. |
| 2015 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2016 | static bool consume_stock(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2017 | { |
| 2018 | struct memcg_stock_pcp *stock; |
| 2019 | bool ret = true; |
| 2020 | |
| 2021 | stock = &get_cpu_var(memcg_stock); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2022 | if (memcg == stock->cached && stock->nr_pages) |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 2023 | stock->nr_pages--; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2024 | else /* need to call res_counter_charge */ |
| 2025 | ret = false; |
| 2026 | put_cpu_var(memcg_stock); |
| 2027 | return ret; |
| 2028 | } |
| 2029 | |
| 2030 | /* |
| 2031 | * Returns stocks cached in percpu to res_counter and reset cached information. |
| 2032 | */ |
| 2033 | static void drain_stock(struct memcg_stock_pcp *stock) |
| 2034 | { |
| 2035 | struct mem_cgroup *old = stock->cached; |
| 2036 | |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 2037 | if (stock->nr_pages) { |
| 2038 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; |
| 2039 | |
| 2040 | res_counter_uncharge(&old->res, bytes); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2041 | if (do_swap_account) |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 2042 | res_counter_uncharge(&old->memsw, bytes); |
| 2043 | stock->nr_pages = 0; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2044 | } |
| 2045 | stock->cached = NULL; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2046 | } |
| 2047 | |
| 2048 | /* |
| 2049 | * This must be called under preempt disabled or must be called by |
| 2050 | * a thread which is pinned to local cpu. |
| 2051 | */ |
| 2052 | static void drain_local_stock(struct work_struct *dummy) |
| 2053 | { |
| 2054 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); |
| 2055 | drain_stock(stock); |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2056 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2057 | } |
| 2058 | |
| 2059 | /* |
| 2060 | * Cache charges(val) which is from res_counter, to local per_cpu area. |
Greg Thelen | 320cc51 | 2010-03-15 15:27:28 +0100 | [diff] [blame] | 2061 | * This will be consumed by consume_stock() function, later. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2062 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2063 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2064 | { |
| 2065 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); |
| 2066 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2067 | if (stock->cached != memcg) { /* reset if necessary */ |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2068 | drain_stock(stock); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2069 | stock->cached = memcg; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2070 | } |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 2071 | stock->nr_pages += nr_pages; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2072 | put_cpu_var(memcg_stock); |
| 2073 | } |
| 2074 | |
| 2075 | /* |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2076 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
Michal Hocko | d38144b | 2011-07-26 16:08:28 -0700 | [diff] [blame] | 2077 | * of the hierarchy under it. sync flag says whether we should block |
| 2078 | * until the work is done. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2079 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2080 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2081 | { |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2082 | int cpu, curcpu; |
Michal Hocko | d38144b | 2011-07-26 16:08:28 -0700 | [diff] [blame] | 2083 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2084 | /* Notify other cpus that system-wide "drain" is running */ |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2085 | get_online_cpus(); |
Johannes Weiner | 5af12d0 | 2011-08-25 15:59:07 -0700 | [diff] [blame] | 2086 | curcpu = get_cpu(); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2087 | for_each_online_cpu(cpu) { |
| 2088 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2089 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2090 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2091 | memcg = stock->cached; |
| 2092 | if (!memcg || !stock->nr_pages) |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2093 | continue; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2094 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
Michal Hocko | 3e92041 | 2011-07-26 16:08:29 -0700 | [diff] [blame] | 2095 | continue; |
Michal Hocko | d1a05b6 | 2011-07-26 16:08:27 -0700 | [diff] [blame] | 2096 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
| 2097 | if (cpu == curcpu) |
| 2098 | drain_local_stock(&stock->work); |
| 2099 | else |
| 2100 | schedule_work_on(cpu, &stock->work); |
| 2101 | } |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2102 | } |
Johannes Weiner | 5af12d0 | 2011-08-25 15:59:07 -0700 | [diff] [blame] | 2103 | put_cpu(); |
Michal Hocko | d38144b | 2011-07-26 16:08:28 -0700 | [diff] [blame] | 2104 | |
| 2105 | if (!sync) |
| 2106 | goto out; |
| 2107 | |
| 2108 | for_each_online_cpu(cpu) { |
| 2109 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); |
Michal Hocko | 9f50fad | 2011-08-09 11:56:26 +0200 | [diff] [blame] | 2110 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
Michal Hocko | d38144b | 2011-07-26 16:08:28 -0700 | [diff] [blame] | 2111 | flush_work(&stock->work); |
| 2112 | } |
| 2113 | out: |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2114 | put_online_cpus(); |
Michal Hocko | d38144b | 2011-07-26 16:08:28 -0700 | [diff] [blame] | 2115 | } |
| 2116 | |
| 2117 | /* |
| 2118 | * Tries to drain stocked charges in other cpus. This function is asynchronous |
| 2119 | * and just put a work per cpu for draining localy on each cpu. Caller can |
| 2120 | * expects some charges will be back to res_counter later but cannot wait for |
| 2121 | * it. |
| 2122 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2123 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
Michal Hocko | d38144b | 2011-07-26 16:08:28 -0700 | [diff] [blame] | 2124 | { |
Michal Hocko | 9f50fad | 2011-08-09 11:56:26 +0200 | [diff] [blame] | 2125 | /* |
| 2126 | * If someone calls draining, avoid adding more kworker runs. |
| 2127 | */ |
| 2128 | if (!mutex_trylock(&percpu_charge_mutex)) |
| 2129 | return; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2130 | drain_all_stock(root_memcg, false); |
Michal Hocko | 9f50fad | 2011-08-09 11:56:26 +0200 | [diff] [blame] | 2131 | mutex_unlock(&percpu_charge_mutex); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2132 | } |
| 2133 | |
| 2134 | /* This is a synchronous drain interface. */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2135 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2136 | { |
| 2137 | /* called when force_empty is called */ |
Michal Hocko | 9f50fad | 2011-08-09 11:56:26 +0200 | [diff] [blame] | 2138 | mutex_lock(&percpu_charge_mutex); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2139 | drain_all_stock(root_memcg, true); |
Michal Hocko | 9f50fad | 2011-08-09 11:56:26 +0200 | [diff] [blame] | 2140 | mutex_unlock(&percpu_charge_mutex); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2141 | } |
| 2142 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2143 | /* |
| 2144 | * This function drains percpu counter value from DEAD cpu and |
| 2145 | * move it to local cpu. Note that this function can be preempted. |
| 2146 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2147 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2148 | { |
| 2149 | int i; |
| 2150 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2151 | spin_lock(&memcg->pcp_counter_lock); |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2152 | for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2153 | long x = per_cpu(memcg->stat->count[i], cpu); |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2154 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2155 | per_cpu(memcg->stat->count[i], cpu) = 0; |
| 2156 | memcg->nocpu_base.count[i] += x; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2157 | } |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 2158 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2159 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 2160 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2161 | per_cpu(memcg->stat->events[i], cpu) = 0; |
| 2162 | memcg->nocpu_base.events[i] += x; |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 2163 | } |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2164 | spin_unlock(&memcg->pcp_counter_lock); |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2165 | } |
| 2166 | |
| 2167 | static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb, |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2168 | unsigned long action, |
| 2169 | void *hcpu) |
| 2170 | { |
| 2171 | int cpu = (unsigned long)hcpu; |
| 2172 | struct memcg_stock_pcp *stock; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2173 | struct mem_cgroup *iter; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2174 | |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 2175 | if (action == CPU_ONLINE) |
KAMEZAWA Hiroyuki | 1489eba | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2176 | return NOTIFY_OK; |
KAMEZAWA Hiroyuki | 1489eba | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2177 | |
Kirill A. Shutemov | d833049 | 2012-04-12 12:49:11 -0700 | [diff] [blame] | 2178 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2179 | return NOTIFY_OK; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2180 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 2181 | for_each_mem_cgroup(iter) |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2182 | mem_cgroup_drain_pcp_counter(iter, cpu); |
| 2183 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2184 | stock = &per_cpu(memcg_stock, cpu); |
| 2185 | drain_stock(stock); |
| 2186 | return NOTIFY_OK; |
| 2187 | } |
| 2188 | |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2189 | |
| 2190 | /* See __mem_cgroup_try_charge() for details */ |
| 2191 | enum { |
| 2192 | CHARGE_OK, /* success */ |
| 2193 | CHARGE_RETRY, /* need to retry but retry is not bad */ |
| 2194 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ |
| 2195 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ |
| 2196 | CHARGE_OOM_DIE, /* the current is killed because of OOM */ |
| 2197 | }; |
| 2198 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2199 | static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2200 | unsigned int nr_pages, bool oom_check) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2201 | { |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2202 | unsigned long csize = nr_pages * PAGE_SIZE; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2203 | struct mem_cgroup *mem_over_limit; |
| 2204 | struct res_counter *fail_res; |
| 2205 | unsigned long flags = 0; |
| 2206 | int ret; |
| 2207 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2208 | ret = res_counter_charge(&memcg->res, csize, &fail_res); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2209 | |
| 2210 | if (likely(!ret)) { |
| 2211 | if (!do_swap_account) |
| 2212 | return CHARGE_OK; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2213 | ret = res_counter_charge(&memcg->memsw, csize, &fail_res); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2214 | if (likely(!ret)) |
| 2215 | return CHARGE_OK; |
| 2216 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2217 | res_counter_uncharge(&memcg->res, csize); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2218 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
| 2219 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; |
| 2220 | } else |
| 2221 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); |
Johannes Weiner | 9221edb | 2011-02-01 15:52:42 -0800 | [diff] [blame] | 2222 | /* |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2223 | * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch |
| 2224 | * of regular pages (CHARGE_BATCH), or a single regular page (1). |
Johannes Weiner | 9221edb | 2011-02-01 15:52:42 -0800 | [diff] [blame] | 2225 | * |
| 2226 | * Never reclaim on behalf of optional batching, retry with a |
| 2227 | * single page instead. |
| 2228 | */ |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2229 | if (nr_pages == CHARGE_BATCH) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2230 | return CHARGE_RETRY; |
| 2231 | |
| 2232 | if (!(gfp_mask & __GFP_WAIT)) |
| 2233 | return CHARGE_WOULDBLOCK; |
| 2234 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 2235 | ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags); |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2236 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 2237 | return CHARGE_RETRY; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2238 | /* |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 2239 | * Even though the limit is exceeded at this point, reclaim |
| 2240 | * may have been able to free some pages. Retry the charge |
| 2241 | * before killing the task. |
| 2242 | * |
| 2243 | * Only for regular pages, though: huge pages are rather |
| 2244 | * unlikely to succeed so close to the limit, and we fall back |
| 2245 | * to regular pages anyway in case of failure. |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2246 | */ |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2247 | if (nr_pages == 1 && ret) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2248 | return CHARGE_RETRY; |
| 2249 | |
| 2250 | /* |
| 2251 | * At task move, charge accounts can be doubly counted. So, it's |
| 2252 | * better to wait until the end of task_move if something is going on. |
| 2253 | */ |
| 2254 | if (mem_cgroup_wait_acct_move(mem_over_limit)) |
| 2255 | return CHARGE_RETRY; |
| 2256 | |
| 2257 | /* If we don't need to call oom-killer at el, return immediately */ |
| 2258 | if (!oom_check) |
| 2259 | return CHARGE_NOMEM; |
| 2260 | /* check OOM */ |
David Rientjes | e845e19 | 2012-03-21 16:34:10 -0700 | [diff] [blame] | 2261 | if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize))) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2262 | return CHARGE_OOM_DIE; |
| 2263 | |
| 2264 | return CHARGE_RETRY; |
| 2265 | } |
| 2266 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2267 | /* |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 2268 | * __mem_cgroup_try_charge() does |
| 2269 | * 1. detect memcg to be charged against from passed *mm and *ptr, |
| 2270 | * 2. update res_counter |
| 2271 | * 3. call memory reclaim if necessary. |
| 2272 | * |
| 2273 | * In some special case, if the task is fatal, fatal_signal_pending() or |
| 2274 | * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup |
| 2275 | * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon |
| 2276 | * as possible without any hazards. 2: all pages should have a valid |
| 2277 | * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg |
| 2278 | * pointer, that is treated as a charge to root_mem_cgroup. |
| 2279 | * |
| 2280 | * So __mem_cgroup_try_charge() will return |
| 2281 | * 0 ... on success, filling *ptr with a valid memcg pointer. |
| 2282 | * -ENOMEM ... charge failure because of resource limits. |
| 2283 | * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup. |
| 2284 | * |
| 2285 | * Unlike the exported interface, an "oom" parameter is added. if oom==true, |
| 2286 | * the oom-killer can be invoked. |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2287 | */ |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2288 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2289 | gfp_t gfp_mask, |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2290 | unsigned int nr_pages, |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2291 | struct mem_cgroup **ptr, |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2292 | bool oom) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2293 | { |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2294 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2295 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2296 | struct mem_cgroup *memcg = NULL; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2297 | int ret; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 2298 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2299 | /* |
| 2300 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage |
| 2301 | * in system level. So, allow to go ahead dying process in addition to |
| 2302 | * MEMDIE process. |
| 2303 | */ |
| 2304 | if (unlikely(test_thread_flag(TIF_MEMDIE) |
| 2305 | || fatal_signal_pending(current))) |
| 2306 | goto bypass; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 2307 | |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2308 | /* |
Hugh Dickins | 3be9127 | 2008-02-07 00:14:19 -0800 | [diff] [blame] | 2309 | * We always charge the cgroup the mm_struct belongs to. |
| 2310 | * The mm_struct's mem_cgroup changes on task migration if the |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2311 | * thread group leader migrates. It's possible that mm is not |
| 2312 | * set, if so charge the init_mm (happens for pagecache usage). |
| 2313 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2314 | if (!*ptr && !mm) |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 2315 | *ptr = root_mem_cgroup; |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2316 | again: |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2317 | if (*ptr) { /* css should be a valid one */ |
| 2318 | memcg = *ptr; |
| 2319 | VM_BUG_ON(css_is_removed(&memcg->css)); |
| 2320 | if (mem_cgroup_is_root(memcg)) |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2321 | goto done; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2322 | if (nr_pages == 1 && consume_stock(memcg)) |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2323 | goto done; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2324 | css_get(&memcg->css); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2325 | } else { |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2326 | struct task_struct *p; |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 2327 | |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2328 | rcu_read_lock(); |
| 2329 | p = rcu_dereference(mm->owner); |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2330 | /* |
KAMEZAWA Hiroyuki | ebb76ce | 2010-12-29 14:07:11 -0800 | [diff] [blame] | 2331 | * Because we don't have task_lock(), "p" can exit. |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2332 | * In that case, "memcg" can point to root or p can be NULL with |
KAMEZAWA Hiroyuki | ebb76ce | 2010-12-29 14:07:11 -0800 | [diff] [blame] | 2333 | * race with swapoff. Then, we have small risk of mis-accouning. |
| 2334 | * But such kind of mis-account by race always happens because |
| 2335 | * we don't have cgroup_mutex(). It's overkill and we allo that |
| 2336 | * small race, here. |
| 2337 | * (*) swapoff at el will charge against mm-struct not against |
| 2338 | * task-struct. So, mm->owner can be NULL. |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2339 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2340 | memcg = mem_cgroup_from_task(p); |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 2341 | if (!memcg) |
| 2342 | memcg = root_mem_cgroup; |
| 2343 | if (mem_cgroup_is_root(memcg)) { |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2344 | rcu_read_unlock(); |
| 2345 | goto done; |
| 2346 | } |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2347 | if (nr_pages == 1 && consume_stock(memcg)) { |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2348 | /* |
| 2349 | * It seems dagerous to access memcg without css_get(). |
| 2350 | * But considering how consume_stok works, it's not |
| 2351 | * necessary. If consume_stock success, some charges |
| 2352 | * from this memcg are cached on this cpu. So, we |
| 2353 | * don't need to call css_get()/css_tryget() before |
| 2354 | * calling consume_stock(). |
| 2355 | */ |
| 2356 | rcu_read_unlock(); |
| 2357 | goto done; |
| 2358 | } |
| 2359 | /* after here, we may be blocked. we need to get refcnt */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2360 | if (!css_tryget(&memcg->css)) { |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2361 | rcu_read_unlock(); |
| 2362 | goto again; |
| 2363 | } |
| 2364 | rcu_read_unlock(); |
| 2365 | } |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2366 | |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2367 | do { |
| 2368 | bool oom_check; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2369 | |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2370 | /* If killed, bypass charge */ |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2371 | if (fatal_signal_pending(current)) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2372 | css_put(&memcg->css); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2373 | goto bypass; |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2374 | } |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2375 | |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2376 | oom_check = false; |
| 2377 | if (oom && !nr_oom_retries) { |
| 2378 | oom_check = true; |
| 2379 | nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
| 2380 | } |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 2381 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2382 | ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, oom_check); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2383 | switch (ret) { |
| 2384 | case CHARGE_OK: |
| 2385 | break; |
| 2386 | case CHARGE_RETRY: /* not in OOM situation but retry */ |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2387 | batch = nr_pages; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2388 | css_put(&memcg->css); |
| 2389 | memcg = NULL; |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2390 | goto again; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2391 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2392 | css_put(&memcg->css); |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2393 | goto nomem; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2394 | case CHARGE_NOMEM: /* OOM routine works */ |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2395 | if (!oom) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2396 | css_put(&memcg->css); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2397 | goto nomem; |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2398 | } |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2399 | /* If oom, we never return -ENOMEM */ |
| 2400 | nr_oom_retries--; |
| 2401 | break; |
| 2402 | case CHARGE_OOM_DIE: /* Killed by OOM Killer */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2403 | css_put(&memcg->css); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2404 | goto bypass; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2405 | } |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2406 | } while (ret != CHARGE_OK); |
| 2407 | |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2408 | if (batch > nr_pages) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2409 | refill_stock(memcg, batch - nr_pages); |
| 2410 | css_put(&memcg->css); |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 2411 | done: |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2412 | *ptr = memcg; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2413 | return 0; |
| 2414 | nomem: |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2415 | *ptr = NULL; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2416 | return -ENOMEM; |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2417 | bypass: |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 2418 | *ptr = root_mem_cgroup; |
| 2419 | return -EINTR; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2420 | } |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2421 | |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2422 | /* |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2423 | * Somemtimes we have to undo a charge we got by try_charge(). |
| 2424 | * This function is for that and do uncharge, put css's refcnt. |
| 2425 | * gotten by try_charge(). |
| 2426 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2427 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
Johannes Weiner | e7018b8d | 2011-03-23 16:42:33 -0700 | [diff] [blame] | 2428 | unsigned int nr_pages) |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2429 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2430 | if (!mem_cgroup_is_root(memcg)) { |
Johannes Weiner | e7018b8d | 2011-03-23 16:42:33 -0700 | [diff] [blame] | 2431 | unsigned long bytes = nr_pages * PAGE_SIZE; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2432 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2433 | res_counter_uncharge(&memcg->res, bytes); |
Johannes Weiner | e7018b8d | 2011-03-23 16:42:33 -0700 | [diff] [blame] | 2434 | if (do_swap_account) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2435 | res_counter_uncharge(&memcg->memsw, bytes); |
Johannes Weiner | e7018b8d | 2011-03-23 16:42:33 -0700 | [diff] [blame] | 2436 | } |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2437 | } |
| 2438 | |
| 2439 | /* |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2440 | * A helper function to get mem_cgroup from ID. must be called under |
| 2441 | * rcu_read_lock(). The caller must check css_is_removed() or some if |
| 2442 | * it's concern. (dropping refcnt from swap can be called against removed |
| 2443 | * memcg.) |
| 2444 | */ |
| 2445 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) |
| 2446 | { |
| 2447 | struct cgroup_subsys_state *css; |
| 2448 | |
| 2449 | /* ID 0 is unused ID */ |
| 2450 | if (!id) |
| 2451 | return NULL; |
| 2452 | css = css_lookup(&mem_cgroup_subsys, id); |
| 2453 | if (!css) |
| 2454 | return NULL; |
| 2455 | return container_of(css, struct mem_cgroup, css); |
| 2456 | } |
| 2457 | |
Wu Fengguang | e42d9d5 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 2458 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2459 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2460 | struct mem_cgroup *memcg = NULL; |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2461 | struct page_cgroup *pc; |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2462 | unsigned short id; |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2463 | swp_entry_t ent; |
| 2464 | |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2465 | VM_BUG_ON(!PageLocked(page)); |
| 2466 | |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2467 | pc = lookup_page_cgroup(page); |
Daisuke Nishimura | c0bd3f6 | 2009-04-30 15:08:11 -0700 | [diff] [blame] | 2468 | lock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2469 | if (PageCgroupUsed(pc)) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2470 | memcg = pc->mem_cgroup; |
| 2471 | if (memcg && !css_tryget(&memcg->css)) |
| 2472 | memcg = NULL; |
Wu Fengguang | e42d9d5 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 2473 | } else if (PageSwapCache(page)) { |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2474 | ent.val = page_private(page); |
Bob Liu | 9fb4b7c | 2012-01-12 17:18:48 -0800 | [diff] [blame] | 2475 | id = lookup_swap_cgroup_id(ent); |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2476 | rcu_read_lock(); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2477 | memcg = mem_cgroup_lookup(id); |
| 2478 | if (memcg && !css_tryget(&memcg->css)) |
| 2479 | memcg = NULL; |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2480 | rcu_read_unlock(); |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2481 | } |
Daisuke Nishimura | c0bd3f6 | 2009-04-30 15:08:11 -0700 | [diff] [blame] | 2482 | unlock_page_cgroup(pc); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2483 | return memcg; |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2484 | } |
| 2485 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2486 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
Johannes Weiner | 5564e88 | 2011-03-23 16:42:29 -0700 | [diff] [blame] | 2487 | struct page *page, |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2488 | unsigned int nr_pages, |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2489 | enum charge_type ctype, |
| 2490 | bool lrucare) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2491 | { |
Johannes Weiner | ce587e6 | 2012-04-24 20:22:33 +0200 | [diff] [blame] | 2492 | struct page_cgroup *pc = lookup_page_cgroup(page); |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2493 | struct zone *uninitialized_var(zone); |
| 2494 | bool was_on_lru = false; |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 2495 | bool anon; |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2496 | |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2497 | lock_page_cgroup(pc); |
| 2498 | if (unlikely(PageCgroupUsed(pc))) { |
| 2499 | unlock_page_cgroup(pc); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2500 | __mem_cgroup_cancel_charge(memcg, nr_pages); |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2501 | return; |
| 2502 | } |
| 2503 | /* |
| 2504 | * we don't need page_cgroup_lock about tail pages, becase they are not |
| 2505 | * accessed by any other context at this point. |
| 2506 | */ |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2507 | |
| 2508 | /* |
| 2509 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page |
| 2510 | * may already be on some other mem_cgroup's LRU. Take care of it. |
| 2511 | */ |
| 2512 | if (lrucare) { |
| 2513 | zone = page_zone(page); |
| 2514 | spin_lock_irq(&zone->lru_lock); |
| 2515 | if (PageLRU(page)) { |
| 2516 | ClearPageLRU(page); |
| 2517 | del_page_from_lru_list(zone, page, page_lru(page)); |
| 2518 | was_on_lru = true; |
| 2519 | } |
| 2520 | } |
| 2521 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2522 | pc->mem_cgroup = memcg; |
KAMEZAWA Hiroyuki | 261fb61 | 2009-09-23 15:56:33 -0700 | [diff] [blame] | 2523 | /* |
| 2524 | * We access a page_cgroup asynchronously without lock_page_cgroup(). |
| 2525 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup |
| 2526 | * is accessed after testing USED bit. To make pc->mem_cgroup visible |
| 2527 | * before USED bit, we need memory barrier here. |
| 2528 | * See mem_cgroup_add_lru_list(), etc. |
| 2529 | */ |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2530 | smp_wmb(); |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 2531 | SetPageCgroupUsed(pc); |
Hugh Dickins | 3be9127 | 2008-02-07 00:14:19 -0800 | [diff] [blame] | 2532 | |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2533 | if (lrucare) { |
| 2534 | if (was_on_lru) { |
| 2535 | VM_BUG_ON(PageLRU(page)); |
| 2536 | SetPageLRU(page); |
| 2537 | add_page_to_lru_list(zone, page, page_lru(page)); |
| 2538 | } |
| 2539 | spin_unlock_irq(&zone->lru_lock); |
| 2540 | } |
| 2541 | |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 2542 | if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) |
| 2543 | anon = true; |
| 2544 | else |
| 2545 | anon = false; |
| 2546 | |
| 2547 | mem_cgroup_charge_statistics(memcg, anon, nr_pages); |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2548 | unlock_page_cgroup(pc); |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2549 | |
KAMEZAWA Hiroyuki | 430e4863 | 2010-03-10 15:22:30 -0800 | [diff] [blame] | 2550 | /* |
| 2551 | * "charge_statistics" updated event counter. Then, check it. |
| 2552 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. |
| 2553 | * if they exceeds softlimit. |
| 2554 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2555 | memcg_check_events(memcg, page); |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2556 | } |
| 2557 | |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2558 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 2559 | |
KAMEZAWA Hiroyuki | 312734c0 | 2012-03-21 16:34:24 -0700 | [diff] [blame] | 2560 | #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MIGRATION)) |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2561 | /* |
| 2562 | * Because tail pages are not marked as "used", set it. We're under |
KAMEZAWA Hiroyuki | e94c8a9 | 2012-01-12 17:18:20 -0800 | [diff] [blame] | 2563 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
| 2564 | * charge/uncharge will be never happen and move_account() is done under |
| 2565 | * compound_lock(), so we don't have to take care of races. |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2566 | */ |
KAMEZAWA Hiroyuki | e94c8a9 | 2012-01-12 17:18:20 -0800 | [diff] [blame] | 2567 | void mem_cgroup_split_huge_fixup(struct page *head) |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2568 | { |
| 2569 | struct page_cgroup *head_pc = lookup_page_cgroup(head); |
KAMEZAWA Hiroyuki | e94c8a9 | 2012-01-12 17:18:20 -0800 | [diff] [blame] | 2570 | struct page_cgroup *pc; |
| 2571 | int i; |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2572 | |
KAMEZAWA Hiroyuki | 3d37c4a | 2011-01-25 15:07:28 -0800 | [diff] [blame] | 2573 | if (mem_cgroup_disabled()) |
| 2574 | return; |
KAMEZAWA Hiroyuki | e94c8a9 | 2012-01-12 17:18:20 -0800 | [diff] [blame] | 2575 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
| 2576 | pc = head_pc + i; |
| 2577 | pc->mem_cgroup = head_pc->mem_cgroup; |
| 2578 | smp_wmb();/* see __commit_charge() */ |
KAMEZAWA Hiroyuki | e94c8a9 | 2012-01-12 17:18:20 -0800 | [diff] [blame] | 2579 | pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
| 2580 | } |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2581 | } |
Hugh Dickins | 12d2710 | 2012-01-12 17:19:52 -0800 | [diff] [blame] | 2582 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2583 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2584 | /** |
Johannes Weiner | de3638d | 2011-03-23 16:42:28 -0700 | [diff] [blame] | 2585 | * mem_cgroup_move_account - move account of the page |
Johannes Weiner | 5564e88 | 2011-03-23 16:42:29 -0700 | [diff] [blame] | 2586 | * @page: the page |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2587 | * @nr_pages: number of regular pages (>1 for huge pages) |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2588 | * @pc: page_cgroup of the page. |
| 2589 | * @from: mem_cgroup which the page is moved from. |
| 2590 | * @to: mem_cgroup which the page is moved to. @from != @to. |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2591 | * @uncharge: whether we should call uncharge and css_put against @from. |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2592 | * |
| 2593 | * The caller must confirm following. |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2594 | * - page is not on LRU (isolate_page() is useful.) |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2595 | * - compound_lock is held when nr_pages > 1 |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2596 | * |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2597 | * This function doesn't do "charge" nor css_get to new cgroup. It should be |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 2598 | * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2599 | * true, this function does "uncharge" from old cgroup, but it doesn't if |
| 2600 | * @uncharge is false, so a caller should do "uncharge". |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2601 | */ |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2602 | static int mem_cgroup_move_account(struct page *page, |
| 2603 | unsigned int nr_pages, |
| 2604 | struct page_cgroup *pc, |
| 2605 | struct mem_cgroup *from, |
| 2606 | struct mem_cgroup *to, |
| 2607 | bool uncharge) |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2608 | { |
Johannes Weiner | de3638d | 2011-03-23 16:42:28 -0700 | [diff] [blame] | 2609 | unsigned long flags; |
| 2610 | int ret; |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 2611 | bool anon = PageAnon(page); |
KAMEZAWA Hiroyuki | 987eba6 | 2011-01-20 14:44:25 -0800 | [diff] [blame] | 2612 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2613 | VM_BUG_ON(from == to); |
Johannes Weiner | 5564e88 | 2011-03-23 16:42:29 -0700 | [diff] [blame] | 2614 | VM_BUG_ON(PageLRU(page)); |
Johannes Weiner | de3638d | 2011-03-23 16:42:28 -0700 | [diff] [blame] | 2615 | /* |
| 2616 | * The page is isolated from LRU. So, collapse function |
| 2617 | * will not handle this page. But page splitting can happen. |
| 2618 | * Do this check under compound_page_lock(). The caller should |
| 2619 | * hold it. |
| 2620 | */ |
| 2621 | ret = -EBUSY; |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2622 | if (nr_pages > 1 && !PageTransHuge(page)) |
Johannes Weiner | de3638d | 2011-03-23 16:42:28 -0700 | [diff] [blame] | 2623 | goto out; |
| 2624 | |
| 2625 | lock_page_cgroup(pc); |
| 2626 | |
| 2627 | ret = -EINVAL; |
| 2628 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) |
| 2629 | goto unlock; |
| 2630 | |
KAMEZAWA Hiroyuki | 312734c0 | 2012-03-21 16:34:24 -0700 | [diff] [blame] | 2631 | move_lock_mem_cgroup(from, &flags); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2632 | |
KAMEZAWA Hiroyuki | 2ff76f1 | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2633 | if (!anon && page_mapped(page)) { |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 2634 | /* Update mapped_file data for mem_cgroup */ |
| 2635 | preempt_disable(); |
| 2636 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); |
| 2637 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); |
| 2638 | preempt_enable(); |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 2639 | } |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 2640 | mem_cgroup_charge_statistics(from, anon, -nr_pages); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2641 | if (uncharge) |
| 2642 | /* This is not "cancel", but cancel_charge does all we need. */ |
Johannes Weiner | e7018b8d | 2011-03-23 16:42:33 -0700 | [diff] [blame] | 2643 | __mem_cgroup_cancel_charge(from, nr_pages); |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 2644 | |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2645 | /* caller should have done css_get */ |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2646 | pc->mem_cgroup = to; |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 2647 | mem_cgroup_charge_statistics(to, anon, nr_pages); |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 2648 | /* |
| 2649 | * We charges against "to" which may not have any tasks. Then, "to" |
| 2650 | * can be under rmdir(). But in current implementation, caller of |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 2651 | * this function is just force_empty() and move charge, so it's |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 2652 | * guaranteed that "to" is never removed. So, we don't check rmdir |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 2653 | * status here. |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 2654 | */ |
KAMEZAWA Hiroyuki | 312734c0 | 2012-03-21 16:34:24 -0700 | [diff] [blame] | 2655 | move_unlock_mem_cgroup(from, &flags); |
Johannes Weiner | de3638d | 2011-03-23 16:42:28 -0700 | [diff] [blame] | 2656 | ret = 0; |
| 2657 | unlock: |
Daisuke Nishimura | 57f9fd7d | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2658 | unlock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 2659 | /* |
| 2660 | * check events |
| 2661 | */ |
Johannes Weiner | 5564e88 | 2011-03-23 16:42:29 -0700 | [diff] [blame] | 2662 | memcg_check_events(to, page); |
| 2663 | memcg_check_events(from, page); |
Johannes Weiner | de3638d | 2011-03-23 16:42:28 -0700 | [diff] [blame] | 2664 | out: |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2665 | return ret; |
| 2666 | } |
| 2667 | |
| 2668 | /* |
| 2669 | * move charges to its parent. |
| 2670 | */ |
| 2671 | |
Johannes Weiner | 5564e88 | 2011-03-23 16:42:29 -0700 | [diff] [blame] | 2672 | static int mem_cgroup_move_parent(struct page *page, |
| 2673 | struct page_cgroup *pc, |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2674 | struct mem_cgroup *child, |
| 2675 | gfp_t gfp_mask) |
| 2676 | { |
| 2677 | struct cgroup *cg = child->css.cgroup; |
| 2678 | struct cgroup *pcg = cg->parent; |
| 2679 | struct mem_cgroup *parent; |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2680 | unsigned int nr_pages; |
Andrew Morton | 4be4489 | 2011-03-23 16:42:39 -0700 | [diff] [blame] | 2681 | unsigned long uninitialized_var(flags); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2682 | int ret; |
| 2683 | |
| 2684 | /* Is ROOT ? */ |
| 2685 | if (!pcg) |
| 2686 | return -EINVAL; |
| 2687 | |
Daisuke Nishimura | 57f9fd7d | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2688 | ret = -EBUSY; |
| 2689 | if (!get_page_unless_zero(page)) |
| 2690 | goto out; |
| 2691 | if (isolate_lru_page(page)) |
| 2692 | goto put; |
KAMEZAWA Hiroyuki | 52dbb90 | 2011-01-25 15:07:29 -0800 | [diff] [blame] | 2693 | |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2694 | nr_pages = hpage_nr_pages(page); |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2695 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2696 | parent = mem_cgroup_from_cont(pcg); |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2697 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, nr_pages, &parent, false); |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 2698 | if (ret) |
Daisuke Nishimura | 57f9fd7d | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2699 | goto put_back; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2700 | |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2701 | if (nr_pages > 1) |
KAMEZAWA Hiroyuki | 987eba6 | 2011-01-20 14:44:25 -0800 | [diff] [blame] | 2702 | flags = compound_lock_irqsave(page); |
| 2703 | |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2704 | ret = mem_cgroup_move_account(page, nr_pages, pc, child, parent, true); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2705 | if (ret) |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2706 | __mem_cgroup_cancel_charge(parent, nr_pages); |
Jesper Juhl | 8dba474 | 2011-01-25 15:07:24 -0800 | [diff] [blame] | 2707 | |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2708 | if (nr_pages > 1) |
KAMEZAWA Hiroyuki | 987eba6 | 2011-01-20 14:44:25 -0800 | [diff] [blame] | 2709 | compound_unlock_irqrestore(page, flags); |
Jesper Juhl | 8dba474 | 2011-01-25 15:07:24 -0800 | [diff] [blame] | 2710 | put_back: |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2711 | putback_lru_page(page); |
Daisuke Nishimura | 57f9fd7d | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2712 | put: |
Daisuke Nishimura | 40d5813 | 2009-01-15 13:51:12 -0800 | [diff] [blame] | 2713 | put_page(page); |
Daisuke Nishimura | 57f9fd7d | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2714 | out: |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2715 | return ret; |
| 2716 | } |
| 2717 | |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2718 | /* |
| 2719 | * Charge the memory controller for page usage. |
| 2720 | * Return |
| 2721 | * 0 if the charge was successful |
| 2722 | * < 0 if the cgroup is over its limit |
| 2723 | */ |
| 2724 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, |
Daisuke Nishimura | 73045c4 | 2010-08-10 18:02:59 -0700 | [diff] [blame] | 2725 | gfp_t gfp_mask, enum charge_type ctype) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2726 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2727 | struct mem_cgroup *memcg = NULL; |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2728 | unsigned int nr_pages = 1; |
Johannes Weiner | 8493ae4 | 2011-02-01 15:52:44 -0800 | [diff] [blame] | 2729 | bool oom = true; |
| 2730 | int ret; |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2731 | |
Andrea Arcangeli | 37c2ac7 | 2011-01-13 15:47:16 -0800 | [diff] [blame] | 2732 | if (PageTransHuge(page)) { |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2733 | nr_pages <<= compound_order(page); |
Andrea Arcangeli | 37c2ac7 | 2011-01-13 15:47:16 -0800 | [diff] [blame] | 2734 | VM_BUG_ON(!PageTransHuge(page)); |
Johannes Weiner | 8493ae4 | 2011-02-01 15:52:44 -0800 | [diff] [blame] | 2735 | /* |
| 2736 | * Never OOM-kill a process for a huge page. The |
| 2737 | * fault handler will fall back to regular pages. |
| 2738 | */ |
| 2739 | oom = false; |
Andrea Arcangeli | 37c2ac7 | 2011-01-13 15:47:16 -0800 | [diff] [blame] | 2740 | } |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2741 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2742 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom); |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 2743 | if (ret == -ENOMEM) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2744 | return ret; |
Johannes Weiner | ce587e6 | 2012-04-24 20:22:33 +0200 | [diff] [blame] | 2745 | __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false); |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2746 | return 0; |
| 2747 | } |
| 2748 | |
| 2749 | int mem_cgroup_newpage_charge(struct page *page, |
| 2750 | struct mm_struct *mm, gfp_t gfp_mask) |
KAMEZAWA Hiroyuki | 217bc31 | 2008-02-07 00:14:17 -0800 | [diff] [blame] | 2751 | { |
Hirokazu Takahashi | f8d66542 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 2752 | if (mem_cgroup_disabled()) |
Li Zefan | cede86a | 2008-07-25 01:47:18 -0700 | [diff] [blame] | 2753 | return 0; |
Johannes Weiner | 7a0524c | 2012-01-12 17:18:43 -0800 | [diff] [blame] | 2754 | VM_BUG_ON(page_mapped(page)); |
| 2755 | VM_BUG_ON(page->mapping && !PageAnon(page)); |
| 2756 | VM_BUG_ON(!mm); |
KAMEZAWA Hiroyuki | 217bc31 | 2008-02-07 00:14:17 -0800 | [diff] [blame] | 2757 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
Johannes Weiner | 7a0524c | 2012-01-12 17:18:43 -0800 | [diff] [blame] | 2758 | MEM_CGROUP_CHARGE_TYPE_MAPPED); |
KAMEZAWA Hiroyuki | 217bc31 | 2008-02-07 00:14:17 -0800 | [diff] [blame] | 2759 | } |
| 2760 | |
Daisuke Nishimura | 83aae4c | 2009-04-02 16:57:48 -0700 | [diff] [blame] | 2761 | static void |
| 2762 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, |
| 2763 | enum charge_type ctype); |
| 2764 | |
Balbir Singh | e1a1cd5 | 2008-02-07 00:14:02 -0800 | [diff] [blame] | 2765 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
| 2766 | gfp_t gfp_mask) |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 2767 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2768 | struct mem_cgroup *memcg = NULL; |
KAMEZAWA Hiroyuki | dc67d50 | 2012-01-12 17:18:55 -0800 | [diff] [blame] | 2769 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2770 | int ret; |
| 2771 | |
Hirokazu Takahashi | f8d66542 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 2772 | if (mem_cgroup_disabled()) |
Li Zefan | cede86a | 2008-07-25 01:47:18 -0700 | [diff] [blame] | 2773 | return 0; |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2774 | if (PageCompound(page)) |
| 2775 | return 0; |
KAMEZAWA Hiroyuki | accf163 | 2008-07-25 01:47:17 -0700 | [diff] [blame] | 2776 | |
Daisuke Nishimura | 73045c4 | 2010-08-10 18:02:59 -0700 | [diff] [blame] | 2777 | if (unlikely(!mm)) |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 2778 | mm = &init_mm; |
KAMEZAWA Hiroyuki | dc67d50 | 2012-01-12 17:18:55 -0800 | [diff] [blame] | 2779 | if (!page_is_file_cache(page)) |
| 2780 | type = MEM_CGROUP_CHARGE_TYPE_SHMEM; |
KAMEZAWA Hiroyuki | accf163 | 2008-07-25 01:47:17 -0700 | [diff] [blame] | 2781 | |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 2782 | if (!PageSwapCache(page)) |
KAMEZAWA Hiroyuki | dc67d50 | 2012-01-12 17:18:55 -0800 | [diff] [blame] | 2783 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, type); |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 2784 | else { /* page is swapcache/shmem */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2785 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &memcg); |
Daisuke Nishimura | 83aae4c | 2009-04-02 16:57:48 -0700 | [diff] [blame] | 2786 | if (!ret) |
KAMEZAWA Hiroyuki | dc67d50 | 2012-01-12 17:18:55 -0800 | [diff] [blame] | 2787 | __mem_cgroup_commit_charge_swapin(page, memcg, type); |
| 2788 | } |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2789 | return ret; |
KAMEZAWA Hiroyuki | e8589cc | 2008-07-25 01:47:10 -0700 | [diff] [blame] | 2790 | } |
| 2791 | |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 2792 | /* |
| 2793 | * While swap-in, try_charge -> commit or cancel, the page is locked. |
| 2794 | * And when try_charge() successfully returns, one refcnt to memcg without |
Uwe Kleine-König | 21ae295 | 2009-10-07 15:21:09 +0200 | [diff] [blame] | 2795 | * struct page_cgroup is acquired. This refcnt will be consumed by |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 2796 | * "commit()" or removed by "cancel()" |
| 2797 | */ |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2798 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
| 2799 | struct page *page, |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 2800 | gfp_t mask, struct mem_cgroup **memcgp) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2801 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2802 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 2803 | int ret; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2804 | |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 2805 | *memcgp = NULL; |
KAMEZAWA Hiroyuki | 56039ef | 2011-03-23 16:42:19 -0700 | [diff] [blame] | 2806 | |
Hirokazu Takahashi | f8d66542 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 2807 | if (mem_cgroup_disabled()) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2808 | return 0; |
| 2809 | |
| 2810 | if (!do_swap_account) |
| 2811 | goto charge_cur_mm; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2812 | /* |
| 2813 | * A racing thread's fault, or swapoff, may have already updated |
Hugh Dickins | 407f9c8 | 2009-12-14 17:59:30 -0800 | [diff] [blame] | 2814 | * the pte, and even removed page from swap cache: in those cases |
| 2815 | * do_swap_page()'s pte_same() test will fail; but there's also a |
| 2816 | * KSM case which does need to charge the page. |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2817 | */ |
| 2818 | if (!PageSwapCache(page)) |
Hugh Dickins | 407f9c8 | 2009-12-14 17:59:30 -0800 | [diff] [blame] | 2819 | goto charge_cur_mm; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2820 | memcg = try_get_mem_cgroup_from_page(page); |
| 2821 | if (!memcg) |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 2822 | goto charge_cur_mm; |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 2823 | *memcgp = memcg; |
| 2824 | ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2825 | css_put(&memcg->css); |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 2826 | if (ret == -EINTR) |
| 2827 | ret = 0; |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 2828 | return ret; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2829 | charge_cur_mm: |
| 2830 | if (unlikely(!mm)) |
| 2831 | mm = &init_mm; |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 2832 | ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true); |
| 2833 | if (ret == -EINTR) |
| 2834 | ret = 0; |
| 2835 | return ret; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2836 | } |
| 2837 | |
Daisuke Nishimura | 83aae4c | 2009-04-02 16:57:48 -0700 | [diff] [blame] | 2838 | static void |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 2839 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg, |
Daisuke Nishimura | 83aae4c | 2009-04-02 16:57:48 -0700 | [diff] [blame] | 2840 | enum charge_type ctype) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2841 | { |
Hirokazu Takahashi | f8d66542 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 2842 | if (mem_cgroup_disabled()) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2843 | return; |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 2844 | if (!memcg) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2845 | return; |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 2846 | cgroup_exclude_rmdir(&memcg->css); |
KAMEZAWA Hiroyuki | 5a6475a | 2011-03-23 16:42:42 -0700 | [diff] [blame] | 2847 | |
Johannes Weiner | ce587e6 | 2012-04-24 20:22:33 +0200 | [diff] [blame] | 2848 | __mem_cgroup_commit_charge(memcg, page, 1, ctype, true); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2849 | /* |
| 2850 | * Now swap is on-memory. This means this page may be |
| 2851 | * counted both as mem and swap....double count. |
KAMEZAWA Hiroyuki | 03f3c43 | 2009-01-07 18:08:31 -0800 | [diff] [blame] | 2852 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
| 2853 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() |
| 2854 | * may call delete_from_swap_cache() before reach here. |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2855 | */ |
KAMEZAWA Hiroyuki | 03f3c43 | 2009-01-07 18:08:31 -0800 | [diff] [blame] | 2856 | if (do_swap_account && PageSwapCache(page)) { |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2857 | swp_entry_t ent = {.val = page_private(page)}; |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 2858 | struct mem_cgroup *swap_memcg; |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2859 | unsigned short id; |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2860 | |
| 2861 | id = swap_cgroup_record(ent, 0); |
| 2862 | rcu_read_lock(); |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 2863 | swap_memcg = mem_cgroup_lookup(id); |
| 2864 | if (swap_memcg) { |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2865 | /* |
| 2866 | * This recorded memcg can be obsolete one. So, avoid |
| 2867 | * calling css_tryget |
| 2868 | */ |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 2869 | if (!mem_cgroup_is_root(swap_memcg)) |
| 2870 | res_counter_uncharge(&swap_memcg->memsw, |
| 2871 | PAGE_SIZE); |
| 2872 | mem_cgroup_swap_statistics(swap_memcg, false); |
| 2873 | mem_cgroup_put(swap_memcg); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2874 | } |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2875 | rcu_read_unlock(); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2876 | } |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 2877 | /* |
| 2878 | * At swapin, we may charge account against cgroup which has no tasks. |
| 2879 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
| 2880 | * In that case, we need to call pre_destroy() again. check it here. |
| 2881 | */ |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 2882 | cgroup_release_and_wakeup_rmdir(&memcg->css); |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2883 | } |
| 2884 | |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 2885 | void mem_cgroup_commit_charge_swapin(struct page *page, |
| 2886 | struct mem_cgroup *memcg) |
Daisuke Nishimura | 83aae4c | 2009-04-02 16:57:48 -0700 | [diff] [blame] | 2887 | { |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 2888 | __mem_cgroup_commit_charge_swapin(page, memcg, |
| 2889 | MEM_CGROUP_CHARGE_TYPE_MAPPED); |
Daisuke Nishimura | 83aae4c | 2009-04-02 16:57:48 -0700 | [diff] [blame] | 2890 | } |
| 2891 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2892 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2893 | { |
Hirokazu Takahashi | f8d66542 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 2894 | if (mem_cgroup_disabled()) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2895 | return; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2896 | if (!memcg) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2897 | return; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2898 | __mem_cgroup_cancel_charge(memcg, 1); |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2899 | } |
| 2900 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2901 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2902 | unsigned int nr_pages, |
| 2903 | const enum charge_type ctype) |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2904 | { |
| 2905 | struct memcg_batch_info *batch = NULL; |
| 2906 | bool uncharge_memsw = true; |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2907 | |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2908 | /* If swapout, usage of swap doesn't decrease */ |
| 2909 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) |
| 2910 | uncharge_memsw = false; |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2911 | |
| 2912 | batch = ¤t->memcg_batch; |
| 2913 | /* |
| 2914 | * In usual, we do css_get() when we remember memcg pointer. |
| 2915 | * But in this case, we keep res->usage until end of a series of |
| 2916 | * uncharges. Then, it's ok to ignore memcg's refcnt. |
| 2917 | */ |
| 2918 | if (!batch->memcg) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2919 | batch->memcg = memcg; |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2920 | /* |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2921 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 2922 | * In those cases, all pages freed continuously can be expected to be in |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2923 | * the same cgroup and we have chance to coalesce uncharges. |
| 2924 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) |
| 2925 | * because we want to do uncharge as soon as possible. |
| 2926 | */ |
| 2927 | |
| 2928 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) |
| 2929 | goto direct_uncharge; |
| 2930 | |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2931 | if (nr_pages > 1) |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2932 | goto direct_uncharge; |
| 2933 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2934 | /* |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2935 | * In typical case, batch->memcg == mem. This means we can |
| 2936 | * merge a series of uncharges to an uncharge of res_counter. |
| 2937 | * If not, we uncharge res_counter ony by one. |
| 2938 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2939 | if (batch->memcg != memcg) |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2940 | goto direct_uncharge; |
| 2941 | /* remember freed charge and uncharge it later */ |
Johannes Weiner | 7ffd4ca | 2011-03-23 16:42:35 -0700 | [diff] [blame] | 2942 | batch->nr_pages++; |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2943 | if (uncharge_memsw) |
Johannes Weiner | 7ffd4ca | 2011-03-23 16:42:35 -0700 | [diff] [blame] | 2944 | batch->memsw_nr_pages++; |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2945 | return; |
| 2946 | direct_uncharge: |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2947 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2948 | if (uncharge_memsw) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2949 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
| 2950 | if (unlikely(batch->memcg != memcg)) |
| 2951 | memcg_oom_recover(memcg); |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2952 | } |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2953 | |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 2954 | /* |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 2955 | * uncharge if !page_mapped(page) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2956 | */ |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2957 | static struct mem_cgroup * |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 2958 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2959 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2960 | struct mem_cgroup *memcg = NULL; |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2961 | unsigned int nr_pages = 1; |
| 2962 | struct page_cgroup *pc; |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 2963 | bool anon; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2964 | |
Hirokazu Takahashi | f8d66542 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 2965 | if (mem_cgroup_disabled()) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2966 | return NULL; |
Balbir Singh | 4077960 | 2008-04-04 14:29:59 -0700 | [diff] [blame] | 2967 | |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 2968 | if (PageSwapCache(page)) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2969 | return NULL; |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 2970 | |
Andrea Arcangeli | 37c2ac7 | 2011-01-13 15:47:16 -0800 | [diff] [blame] | 2971 | if (PageTransHuge(page)) { |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2972 | nr_pages <<= compound_order(page); |
Andrea Arcangeli | 37c2ac7 | 2011-01-13 15:47:16 -0800 | [diff] [blame] | 2973 | VM_BUG_ON(!PageTransHuge(page)); |
| 2974 | } |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 2975 | /* |
Balbir Singh | 3c541e1 | 2008-02-07 00:14:41 -0800 | [diff] [blame] | 2976 | * Check if our page_cgroup is valid |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 2977 | */ |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2978 | pc = lookup_page_cgroup(page); |
Johannes Weiner | cfa4494 | 2012-01-12 17:18:38 -0800 | [diff] [blame] | 2979 | if (unlikely(!PageCgroupUsed(pc))) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2980 | return NULL; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2981 | |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2982 | lock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 2983 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2984 | memcg = pc->mem_cgroup; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2985 | |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 2986 | if (!PageCgroupUsed(pc)) |
| 2987 | goto unlock_out; |
| 2988 | |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 2989 | anon = PageAnon(page); |
| 2990 | |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 2991 | switch (ctype) { |
| 2992 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: |
KAMEZAWA Hiroyuki | 2ff76f1 | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2993 | /* |
| 2994 | * Generally PageAnon tells if it's the anon statistics to be |
| 2995 | * updated; but sometimes e.g. mem_cgroup_uncharge_page() is |
| 2996 | * used before page reached the stage of being marked PageAnon. |
| 2997 | */ |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 2998 | anon = true; |
| 2999 | /* fallthrough */ |
KAMEZAWA Hiroyuki | 8a9478ca | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 3000 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3001 | /* See mem_cgroup_prepare_migration() */ |
| 3002 | if (page_mapped(page) || PageCgroupMigration(pc)) |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 3003 | goto unlock_out; |
| 3004 | break; |
| 3005 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: |
| 3006 | if (!PageAnon(page)) { /* Shared memory */ |
| 3007 | if (page->mapping && !page_is_file_cache(page)) |
| 3008 | goto unlock_out; |
| 3009 | } else if (page_mapped(page)) /* Anon */ |
| 3010 | goto unlock_out; |
| 3011 | break; |
| 3012 | default: |
| 3013 | break; |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 3014 | } |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 3015 | |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 3016 | mem_cgroup_charge_statistics(memcg, anon, -nr_pages); |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 3017 | |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 3018 | ClearPageCgroupUsed(pc); |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 3019 | /* |
| 3020 | * pc->mem_cgroup is not cleared here. It will be accessed when it's |
| 3021 | * freed from LRU. This is safe because uncharged page is expected not |
| 3022 | * to be reused (freed soon). Exception is SwapCache, it's handled by |
| 3023 | * special functions. |
| 3024 | */ |
Hugh Dickins | b9c565d | 2008-03-04 14:29:11 -0800 | [diff] [blame] | 3025 | |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 3026 | unlock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 3027 | /* |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3028 | * even after unlock, we have memcg->res.usage here and this memcg |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 3029 | * will never be freed. |
| 3030 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3031 | memcg_check_events(memcg, page); |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 3032 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3033 | mem_cgroup_swap_statistics(memcg, true); |
| 3034 | mem_cgroup_get(memcg); |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 3035 | } |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3036 | if (!mem_cgroup_is_root(memcg)) |
| 3037 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 3038 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3039 | return memcg; |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 3040 | |
| 3041 | unlock_out: |
| 3042 | unlock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3043 | return NULL; |
Balbir Singh | 3c541e1 | 2008-02-07 00:14:41 -0800 | [diff] [blame] | 3044 | } |
| 3045 | |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 3046 | void mem_cgroup_uncharge_page(struct page *page) |
| 3047 | { |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 3048 | /* early check. */ |
| 3049 | if (page_mapped(page)) |
| 3050 | return; |
Johannes Weiner | 40f23a2 | 2012-01-12 17:18:45 -0800 | [diff] [blame] | 3051 | VM_BUG_ON(page->mapping && !PageAnon(page)); |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 3052 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
| 3053 | } |
| 3054 | |
| 3055 | void mem_cgroup_uncharge_cache_page(struct page *page) |
| 3056 | { |
| 3057 | VM_BUG_ON(page_mapped(page)); |
KAMEZAWA Hiroyuki | b7abea9 | 2008-10-18 20:28:09 -0700 | [diff] [blame] | 3058 | VM_BUG_ON(page->mapping); |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 3059 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
| 3060 | } |
| 3061 | |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 3062 | /* |
| 3063 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. |
| 3064 | * In that cases, pages are freed continuously and we can expect pages |
| 3065 | * are in the same memcg. All these calls itself limits the number of |
| 3066 | * pages freed at once, then uncharge_start/end() is called properly. |
| 3067 | * This may be called prural(2) times in a context, |
| 3068 | */ |
| 3069 | |
| 3070 | void mem_cgroup_uncharge_start(void) |
| 3071 | { |
| 3072 | current->memcg_batch.do_batch++; |
| 3073 | /* We can do nest. */ |
| 3074 | if (current->memcg_batch.do_batch == 1) { |
| 3075 | current->memcg_batch.memcg = NULL; |
Johannes Weiner | 7ffd4ca | 2011-03-23 16:42:35 -0700 | [diff] [blame] | 3076 | current->memcg_batch.nr_pages = 0; |
| 3077 | current->memcg_batch.memsw_nr_pages = 0; |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 3078 | } |
| 3079 | } |
| 3080 | |
| 3081 | void mem_cgroup_uncharge_end(void) |
| 3082 | { |
| 3083 | struct memcg_batch_info *batch = ¤t->memcg_batch; |
| 3084 | |
| 3085 | if (!batch->do_batch) |
| 3086 | return; |
| 3087 | |
| 3088 | batch->do_batch--; |
| 3089 | if (batch->do_batch) /* If stacked, do nothing. */ |
| 3090 | return; |
| 3091 | |
| 3092 | if (!batch->memcg) |
| 3093 | return; |
| 3094 | /* |
| 3095 | * This "batch->memcg" is valid without any css_get/put etc... |
| 3096 | * bacause we hide charges behind us. |
| 3097 | */ |
Johannes Weiner | 7ffd4ca | 2011-03-23 16:42:35 -0700 | [diff] [blame] | 3098 | if (batch->nr_pages) |
| 3099 | res_counter_uncharge(&batch->memcg->res, |
| 3100 | batch->nr_pages * PAGE_SIZE); |
| 3101 | if (batch->memsw_nr_pages) |
| 3102 | res_counter_uncharge(&batch->memcg->memsw, |
| 3103 | batch->memsw_nr_pages * PAGE_SIZE); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3104 | memcg_oom_recover(batch->memcg); |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 3105 | /* forget this pointer (for sanity check) */ |
| 3106 | batch->memcg = NULL; |
| 3107 | } |
| 3108 | |
Daisuke Nishimura | e767e05 | 2009-05-28 14:34:28 -0700 | [diff] [blame] | 3109 | #ifdef CONFIG_SWAP |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3110 | /* |
Daisuke Nishimura | e767e05 | 2009-05-28 14:34:28 -0700 | [diff] [blame] | 3111 | * called after __delete_from_swap_cache() and drop "page" account. |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3112 | * memcg information is recorded to swap_cgroup of "ent" |
| 3113 | */ |
KAMEZAWA Hiroyuki | 8a9478ca | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 3114 | void |
| 3115 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 3116 | { |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3117 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | 8a9478ca | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 3118 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3119 | |
KAMEZAWA Hiroyuki | 8a9478ca | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 3120 | if (!swapout) /* this was a swap cache but the swap is unused ! */ |
| 3121 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; |
| 3122 | |
| 3123 | memcg = __mem_cgroup_uncharge_common(page, ctype); |
| 3124 | |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 3125 | /* |
| 3126 | * record memcg information, if swapout && memcg != NULL, |
| 3127 | * mem_cgroup_get() was called in uncharge(). |
| 3128 | */ |
| 3129 | if (do_swap_account && swapout && memcg) |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 3130 | swap_cgroup_record(ent, css_id(&memcg->css)); |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 3131 | } |
Daisuke Nishimura | e767e05 | 2009-05-28 14:34:28 -0700 | [diff] [blame] | 3132 | #endif |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 3133 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3134 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
| 3135 | /* |
| 3136 | * called from swap_entry_free(). remove record in swap_cgroup and |
| 3137 | * uncharge "memsw" account. |
| 3138 | */ |
| 3139 | void mem_cgroup_uncharge_swap(swp_entry_t ent) |
| 3140 | { |
| 3141 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 3142 | unsigned short id; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3143 | |
| 3144 | if (!do_swap_account) |
| 3145 | return; |
| 3146 | |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 3147 | id = swap_cgroup_record(ent, 0); |
| 3148 | rcu_read_lock(); |
| 3149 | memcg = mem_cgroup_lookup(id); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3150 | if (memcg) { |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 3151 | /* |
| 3152 | * We uncharge this because swap is freed. |
| 3153 | * This memcg can be obsolete one. We avoid calling css_tryget |
| 3154 | */ |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 3155 | if (!mem_cgroup_is_root(memcg)) |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 3156 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 3157 | mem_cgroup_swap_statistics(memcg, false); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3158 | mem_cgroup_put(memcg); |
| 3159 | } |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 3160 | rcu_read_unlock(); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3161 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3162 | |
| 3163 | /** |
| 3164 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. |
| 3165 | * @entry: swap entry to be moved |
| 3166 | * @from: mem_cgroup which the entry is moved from |
| 3167 | * @to: mem_cgroup which the entry is moved to |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 3168 | * @need_fixup: whether we should fixup res_counters and refcounts. |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3169 | * |
| 3170 | * It succeeds only when the swap_cgroup's record for this entry is the same |
| 3171 | * as the mem_cgroup's id of @from. |
| 3172 | * |
| 3173 | * Returns 0 on success, -EINVAL on failure. |
| 3174 | * |
| 3175 | * The caller must have charged to @to, IOW, called res_counter_charge() about |
| 3176 | * both res and memsw, and called css_get(). |
| 3177 | */ |
| 3178 | static int mem_cgroup_move_swap_account(swp_entry_t entry, |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 3179 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3180 | { |
| 3181 | unsigned short old_id, new_id; |
| 3182 | |
| 3183 | old_id = css_id(&from->css); |
| 3184 | new_id = css_id(&to->css); |
| 3185 | |
| 3186 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3187 | mem_cgroup_swap_statistics(from, false); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3188 | mem_cgroup_swap_statistics(to, true); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 3189 | /* |
| 3190 | * This function is only called from task migration context now. |
| 3191 | * It postpones res_counter and refcount handling till the end |
| 3192 | * of task migration(mem_cgroup_clear_mc()) for performance |
| 3193 | * improvement. But we cannot postpone mem_cgroup_get(to) |
| 3194 | * because if the process that has been moved to @to does |
| 3195 | * swap-in, the refcount of @to might be decreased to 0. |
| 3196 | */ |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3197 | mem_cgroup_get(to); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 3198 | if (need_fixup) { |
| 3199 | if (!mem_cgroup_is_root(from)) |
| 3200 | res_counter_uncharge(&from->memsw, PAGE_SIZE); |
| 3201 | mem_cgroup_put(from); |
| 3202 | /* |
| 3203 | * we charged both to->res and to->memsw, so we should |
| 3204 | * uncharge to->res. |
| 3205 | */ |
| 3206 | if (!mem_cgroup_is_root(to)) |
| 3207 | res_counter_uncharge(&to->res, PAGE_SIZE); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 3208 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3209 | return 0; |
| 3210 | } |
| 3211 | return -EINVAL; |
| 3212 | } |
| 3213 | #else |
| 3214 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 3215 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3216 | { |
| 3217 | return -EINVAL; |
| 3218 | } |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3219 | #endif |
| 3220 | |
KAMEZAWA Hiroyuki | ae41be3 | 2008-02-07 00:14:10 -0800 | [diff] [blame] | 3221 | /* |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 3222 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
| 3223 | * page belongs to. |
KAMEZAWA Hiroyuki | ae41be3 | 2008-02-07 00:14:10 -0800 | [diff] [blame] | 3224 | */ |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3225 | int mem_cgroup_prepare_migration(struct page *page, |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 3226 | struct page *newpage, struct mem_cgroup **memcgp, gfp_t gfp_mask) |
KAMEZAWA Hiroyuki | ae41be3 | 2008-02-07 00:14:10 -0800 | [diff] [blame] | 3227 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3228 | struct mem_cgroup *memcg = NULL; |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 3229 | struct page_cgroup *pc; |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3230 | enum charge_type ctype; |
KAMEZAWA Hiroyuki | e8589cc | 2008-07-25 01:47:10 -0700 | [diff] [blame] | 3231 | int ret = 0; |
Hugh Dickins | 8869b8f | 2008-03-04 14:29:09 -0800 | [diff] [blame] | 3232 | |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 3233 | *memcgp = NULL; |
KAMEZAWA Hiroyuki | 56039ef | 2011-03-23 16:42:19 -0700 | [diff] [blame] | 3234 | |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 3235 | VM_BUG_ON(PageTransHuge(page)); |
Hirokazu Takahashi | f8d66542 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 3236 | if (mem_cgroup_disabled()) |
Balbir Singh | 4077960 | 2008-04-04 14:29:59 -0700 | [diff] [blame] | 3237 | return 0; |
| 3238 | |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 3239 | pc = lookup_page_cgroup(page); |
| 3240 | lock_page_cgroup(pc); |
| 3241 | if (PageCgroupUsed(pc)) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3242 | memcg = pc->mem_cgroup; |
| 3243 | css_get(&memcg->css); |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3244 | /* |
| 3245 | * At migrating an anonymous page, its mapcount goes down |
| 3246 | * to 0 and uncharge() will be called. But, even if it's fully |
| 3247 | * unmapped, migration may fail and this page has to be |
| 3248 | * charged again. We set MIGRATION flag here and delay uncharge |
| 3249 | * until end_migration() is called |
| 3250 | * |
| 3251 | * Corner Case Thinking |
| 3252 | * A) |
| 3253 | * When the old page was mapped as Anon and it's unmap-and-freed |
| 3254 | * while migration was ongoing. |
| 3255 | * If unmap finds the old page, uncharge() of it will be delayed |
| 3256 | * until end_migration(). If unmap finds a new page, it's |
| 3257 | * uncharged when it make mapcount to be 1->0. If unmap code |
| 3258 | * finds swap_migration_entry, the new page will not be mapped |
| 3259 | * and end_migration() will find it(mapcount==0). |
| 3260 | * |
| 3261 | * B) |
| 3262 | * When the old page was mapped but migraion fails, the kernel |
| 3263 | * remaps it. A charge for it is kept by MIGRATION flag even |
| 3264 | * if mapcount goes down to 0. We can do remap successfully |
| 3265 | * without charging it again. |
| 3266 | * |
| 3267 | * C) |
| 3268 | * The "old" page is under lock_page() until the end of |
| 3269 | * migration, so, the old page itself will not be swapped-out. |
| 3270 | * If the new page is swapped out before end_migraton, our |
| 3271 | * hook to usual swap-out path will catch the event. |
| 3272 | */ |
| 3273 | if (PageAnon(page)) |
| 3274 | SetPageCgroupMigration(pc); |
Hugh Dickins | b9c565d | 2008-03-04 14:29:11 -0800 | [diff] [blame] | 3275 | } |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 3276 | unlock_page_cgroup(pc); |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3277 | /* |
| 3278 | * If the page is not charged at this point, |
| 3279 | * we return here. |
| 3280 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3281 | if (!memcg) |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3282 | return 0; |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 3283 | |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 3284 | *memcgp = memcg; |
| 3285 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, 1, memcgp, false); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3286 | css_put(&memcg->css);/* drop extra refcnt */ |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 3287 | if (ret) { |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3288 | if (PageAnon(page)) { |
| 3289 | lock_page_cgroup(pc); |
| 3290 | ClearPageCgroupMigration(pc); |
| 3291 | unlock_page_cgroup(pc); |
| 3292 | /* |
| 3293 | * The old page may be fully unmapped while we kept it. |
| 3294 | */ |
| 3295 | mem_cgroup_uncharge_page(page); |
| 3296 | } |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 3297 | /* we'll need to revisit this error code (we have -EINTR) */ |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3298 | return -ENOMEM; |
KAMEZAWA Hiroyuki | e8589cc | 2008-07-25 01:47:10 -0700 | [diff] [blame] | 3299 | } |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3300 | /* |
| 3301 | * We charge new page before it's used/mapped. So, even if unlock_page() |
| 3302 | * is called before end_migration, we can catch all events on this new |
| 3303 | * page. In the case new page is migrated but not remapped, new page's |
| 3304 | * mapcount will be finally 0 and we call uncharge in end_migration(). |
| 3305 | */ |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3306 | if (PageAnon(page)) |
| 3307 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; |
| 3308 | else if (page_is_file_cache(page)) |
| 3309 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; |
| 3310 | else |
| 3311 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; |
Johannes Weiner | ce587e6 | 2012-04-24 20:22:33 +0200 | [diff] [blame] | 3312 | __mem_cgroup_commit_charge(memcg, newpage, 1, ctype, false); |
KAMEZAWA Hiroyuki | e8589cc | 2008-07-25 01:47:10 -0700 | [diff] [blame] | 3313 | return ret; |
| 3314 | } |
Hugh Dickins | fb59e9f | 2008-03-04 14:29:16 -0800 | [diff] [blame] | 3315 | |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 3316 | /* remove redundant charge if migration failed*/ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3317 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
Daisuke Nishimura | 50de1dd | 2011-01-13 15:47:43 -0800 | [diff] [blame] | 3318 | struct page *oldpage, struct page *newpage, bool migration_ok) |
KAMEZAWA Hiroyuki | e8589cc | 2008-07-25 01:47:10 -0700 | [diff] [blame] | 3319 | { |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3320 | struct page *used, *unused; |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 3321 | struct page_cgroup *pc; |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 3322 | bool anon; |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 3323 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3324 | if (!memcg) |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 3325 | return; |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3326 | /* blocks rmdir() */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3327 | cgroup_exclude_rmdir(&memcg->css); |
Daisuke Nishimura | 50de1dd | 2011-01-13 15:47:43 -0800 | [diff] [blame] | 3328 | if (!migration_ok) { |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3329 | used = oldpage; |
| 3330 | unused = newpage; |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 3331 | } else { |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3332 | used = newpage; |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 3333 | unused = oldpage; |
| 3334 | } |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 3335 | /* |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3336 | * We disallowed uncharge of pages under migration because mapcount |
| 3337 | * of the page goes down to zero, temporarly. |
| 3338 | * Clear the flag and check the page should be charged. |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 3339 | */ |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3340 | pc = lookup_page_cgroup(oldpage); |
| 3341 | lock_page_cgroup(pc); |
| 3342 | ClearPageCgroupMigration(pc); |
| 3343 | unlock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 3344 | anon = PageAnon(used); |
| 3345 | __mem_cgroup_uncharge_common(unused, |
| 3346 | anon ? MEM_CGROUP_CHARGE_TYPE_MAPPED |
| 3347 | : MEM_CGROUP_CHARGE_TYPE_CACHE); |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3348 | |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 3349 | /* |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3350 | * If a page is a file cache, radix-tree replacement is very atomic |
| 3351 | * and we can skip this check. When it was an Anon page, its mapcount |
| 3352 | * goes down to 0. But because we added MIGRATION flage, it's not |
| 3353 | * uncharged yet. There are several case but page->mapcount check |
| 3354 | * and USED bit check in mem_cgroup_uncharge_page() will do enough |
| 3355 | * check. (see prepare_charge() also) |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 3356 | */ |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 3357 | if (anon) |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3358 | mem_cgroup_uncharge_page(used); |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 3359 | /* |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3360 | * At migration, we may charge account against cgroup which has no |
| 3361 | * tasks. |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 3362 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
| 3363 | * In that case, we need to call pre_destroy() again. check it here. |
| 3364 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3365 | cgroup_release_and_wakeup_rmdir(&memcg->css); |
KAMEZAWA Hiroyuki | ae41be3 | 2008-02-07 00:14:10 -0800 | [diff] [blame] | 3366 | } |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 3367 | |
KAMEZAWA Hiroyuki | ab936cb | 2012-01-12 17:17:44 -0800 | [diff] [blame] | 3368 | /* |
| 3369 | * At replace page cache, newpage is not under any memcg but it's on |
| 3370 | * LRU. So, this function doesn't touch res_counter but handles LRU |
| 3371 | * in correct way. Both pages are locked so we cannot race with uncharge. |
| 3372 | */ |
| 3373 | void mem_cgroup_replace_page_cache(struct page *oldpage, |
| 3374 | struct page *newpage) |
| 3375 | { |
| 3376 | struct mem_cgroup *memcg; |
| 3377 | struct page_cgroup *pc; |
KAMEZAWA Hiroyuki | ab936cb | 2012-01-12 17:17:44 -0800 | [diff] [blame] | 3378 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
KAMEZAWA Hiroyuki | ab936cb | 2012-01-12 17:17:44 -0800 | [diff] [blame] | 3379 | |
| 3380 | if (mem_cgroup_disabled()) |
| 3381 | return; |
| 3382 | |
| 3383 | pc = lookup_page_cgroup(oldpage); |
| 3384 | /* fix accounting on old pages */ |
| 3385 | lock_page_cgroup(pc); |
| 3386 | memcg = pc->mem_cgroup; |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 3387 | mem_cgroup_charge_statistics(memcg, false, -1); |
KAMEZAWA Hiroyuki | ab936cb | 2012-01-12 17:17:44 -0800 | [diff] [blame] | 3388 | ClearPageCgroupUsed(pc); |
| 3389 | unlock_page_cgroup(pc); |
| 3390 | |
| 3391 | if (PageSwapBacked(oldpage)) |
| 3392 | type = MEM_CGROUP_CHARGE_TYPE_SHMEM; |
| 3393 | |
KAMEZAWA Hiroyuki | ab936cb | 2012-01-12 17:17:44 -0800 | [diff] [blame] | 3394 | /* |
| 3395 | * Even if newpage->mapping was NULL before starting replacement, |
| 3396 | * the newpage may be on LRU(or pagevec for LRU) already. We lock |
| 3397 | * LRU while we overwrite pc->mem_cgroup. |
| 3398 | */ |
Johannes Weiner | ce587e6 | 2012-04-24 20:22:33 +0200 | [diff] [blame] | 3399 | __mem_cgroup_commit_charge(memcg, newpage, 1, type, true); |
KAMEZAWA Hiroyuki | ab936cb | 2012-01-12 17:17:44 -0800 | [diff] [blame] | 3400 | } |
| 3401 | |
Daisuke Nishimura | f212ad7 | 2011-03-23 16:42:25 -0700 | [diff] [blame] | 3402 | #ifdef CONFIG_DEBUG_VM |
| 3403 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) |
| 3404 | { |
| 3405 | struct page_cgroup *pc; |
| 3406 | |
| 3407 | pc = lookup_page_cgroup(page); |
Johannes Weiner | cfa4494 | 2012-01-12 17:18:38 -0800 | [diff] [blame] | 3408 | /* |
| 3409 | * Can be NULL while feeding pages into the page allocator for |
| 3410 | * the first time, i.e. during boot or memory hotplug; |
| 3411 | * or when mem_cgroup_disabled(). |
| 3412 | */ |
Daisuke Nishimura | f212ad7 | 2011-03-23 16:42:25 -0700 | [diff] [blame] | 3413 | if (likely(pc) && PageCgroupUsed(pc)) |
| 3414 | return pc; |
| 3415 | return NULL; |
| 3416 | } |
| 3417 | |
| 3418 | bool mem_cgroup_bad_page_check(struct page *page) |
| 3419 | { |
| 3420 | if (mem_cgroup_disabled()) |
| 3421 | return false; |
| 3422 | |
| 3423 | return lookup_page_cgroup_used(page) != NULL; |
| 3424 | } |
| 3425 | |
| 3426 | void mem_cgroup_print_bad_page(struct page *page) |
| 3427 | { |
| 3428 | struct page_cgroup *pc; |
| 3429 | |
| 3430 | pc = lookup_page_cgroup_used(page); |
| 3431 | if (pc) { |
Hugh Dickins | 90b3fea | 2012-01-12 17:19:54 -0800 | [diff] [blame] | 3432 | printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
Daisuke Nishimura | f212ad7 | 2011-03-23 16:42:25 -0700 | [diff] [blame] | 3433 | pc, pc->flags, pc->mem_cgroup); |
Daisuke Nishimura | f212ad7 | 2011-03-23 16:42:25 -0700 | [diff] [blame] | 3434 | } |
| 3435 | } |
| 3436 | #endif |
| 3437 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3438 | static DEFINE_MUTEX(set_limit_mutex); |
| 3439 | |
KOSAKI Motohiro | d38d2a7 | 2009-01-06 14:39:44 -0800 | [diff] [blame] | 3440 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3441 | unsigned long long val) |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3442 | { |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3443 | int retry_count; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3444 | u64 memswlimit, memlimit; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3445 | int ret = 0; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3446 | int children = mem_cgroup_count_children(memcg); |
| 3447 | u64 curusage, oldusage; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3448 | int enlarge; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3449 | |
| 3450 | /* |
| 3451 | * For keeping hierarchical_reclaim simple, how long we should retry |
| 3452 | * is depends on callers. We set our retry-count to be function |
| 3453 | * of # of children which we should visit in this loop. |
| 3454 | */ |
| 3455 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; |
| 3456 | |
| 3457 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3458 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3459 | enlarge = 0; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3460 | while (retry_count) { |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3461 | if (signal_pending(current)) { |
| 3462 | ret = -EINTR; |
| 3463 | break; |
| 3464 | } |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3465 | /* |
| 3466 | * Rather than hide all in some function, I do this in |
| 3467 | * open coded manner. You see what this really does. |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3468 | * We have to guarantee memcg->res.limit < memcg->memsw.limit. |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3469 | */ |
| 3470 | mutex_lock(&set_limit_mutex); |
| 3471 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
| 3472 | if (memswlimit < val) { |
| 3473 | ret = -EINVAL; |
| 3474 | mutex_unlock(&set_limit_mutex); |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3475 | break; |
| 3476 | } |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3477 | |
| 3478 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
| 3479 | if (memlimit < val) |
| 3480 | enlarge = 1; |
| 3481 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3482 | ret = res_counter_set_limit(&memcg->res, val); |
KAMEZAWA Hiroyuki | 22a668d | 2009-06-17 16:27:19 -0700 | [diff] [blame] | 3483 | if (!ret) { |
| 3484 | if (memswlimit == val) |
| 3485 | memcg->memsw_is_minimum = true; |
| 3486 | else |
| 3487 | memcg->memsw_is_minimum = false; |
| 3488 | } |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3489 | mutex_unlock(&set_limit_mutex); |
| 3490 | |
| 3491 | if (!ret) |
| 3492 | break; |
| 3493 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 3494 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
| 3495 | MEM_CGROUP_RECLAIM_SHRINK); |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3496 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
| 3497 | /* Usage is reduced ? */ |
| 3498 | if (curusage >= oldusage) |
| 3499 | retry_count--; |
| 3500 | else |
| 3501 | oldusage = curusage; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3502 | } |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3503 | if (!ret && enlarge) |
| 3504 | memcg_oom_recover(memcg); |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 3505 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3506 | return ret; |
| 3507 | } |
| 3508 | |
Li Zefan | 338c843 | 2009-06-17 16:27:15 -0700 | [diff] [blame] | 3509 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
| 3510 | unsigned long long val) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3511 | { |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3512 | int retry_count; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3513 | u64 memlimit, memswlimit, oldusage, curusage; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3514 | int children = mem_cgroup_count_children(memcg); |
| 3515 | int ret = -EBUSY; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3516 | int enlarge = 0; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3517 | |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3518 | /* see mem_cgroup_resize_res_limit */ |
| 3519 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; |
| 3520 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3521 | while (retry_count) { |
| 3522 | if (signal_pending(current)) { |
| 3523 | ret = -EINTR; |
| 3524 | break; |
| 3525 | } |
| 3526 | /* |
| 3527 | * Rather than hide all in some function, I do this in |
| 3528 | * open coded manner. You see what this really does. |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3529 | * We have to guarantee memcg->res.limit < memcg->memsw.limit. |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3530 | */ |
| 3531 | mutex_lock(&set_limit_mutex); |
| 3532 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
| 3533 | if (memlimit > val) { |
| 3534 | ret = -EINVAL; |
| 3535 | mutex_unlock(&set_limit_mutex); |
| 3536 | break; |
| 3537 | } |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3538 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
| 3539 | if (memswlimit < val) |
| 3540 | enlarge = 1; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3541 | ret = res_counter_set_limit(&memcg->memsw, val); |
KAMEZAWA Hiroyuki | 22a668d | 2009-06-17 16:27:19 -0700 | [diff] [blame] | 3542 | if (!ret) { |
| 3543 | if (memlimit == val) |
| 3544 | memcg->memsw_is_minimum = true; |
| 3545 | else |
| 3546 | memcg->memsw_is_minimum = false; |
| 3547 | } |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3548 | mutex_unlock(&set_limit_mutex); |
| 3549 | |
| 3550 | if (!ret) |
| 3551 | break; |
| 3552 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 3553 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
| 3554 | MEM_CGROUP_RECLAIM_NOSWAP | |
| 3555 | MEM_CGROUP_RECLAIM_SHRINK); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3556 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3557 | /* Usage is reduced ? */ |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3558 | if (curusage >= oldusage) |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3559 | retry_count--; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3560 | else |
| 3561 | oldusage = curusage; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3562 | } |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3563 | if (!ret && enlarge) |
| 3564 | memcg_oom_recover(memcg); |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3565 | return ret; |
| 3566 | } |
| 3567 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3568 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 3569 | gfp_t gfp_mask, |
| 3570 | unsigned long *total_scanned) |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3571 | { |
| 3572 | unsigned long nr_reclaimed = 0; |
| 3573 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; |
| 3574 | unsigned long reclaimed; |
| 3575 | int loop = 0; |
| 3576 | struct mem_cgroup_tree_per_zone *mctz; |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 3577 | unsigned long long excess; |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 3578 | unsigned long nr_scanned; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3579 | |
| 3580 | if (order > 0) |
| 3581 | return 0; |
| 3582 | |
KOSAKI Motohiro | 00918b6 | 2010-08-10 18:03:05 -0700 | [diff] [blame] | 3583 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3584 | /* |
| 3585 | * This loop can run a while, specially if mem_cgroup's continuously |
| 3586 | * keep exceeding their soft limit and putting the system under |
| 3587 | * pressure |
| 3588 | */ |
| 3589 | do { |
| 3590 | if (next_mz) |
| 3591 | mz = next_mz; |
| 3592 | else |
| 3593 | mz = mem_cgroup_largest_soft_limit_node(mctz); |
| 3594 | if (!mz) |
| 3595 | break; |
| 3596 | |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 3597 | nr_scanned = 0; |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 3598 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 3599 | gfp_mask, &nr_scanned); |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3600 | nr_reclaimed += reclaimed; |
Ying Han | 0ae5e89 | 2011-05-26 16:25:25 -0700 | [diff] [blame] | 3601 | *total_scanned += nr_scanned; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3602 | spin_lock(&mctz->lock); |
| 3603 | |
| 3604 | /* |
| 3605 | * If we failed to reclaim anything from this memory cgroup |
| 3606 | * it is time to move on to the next cgroup |
| 3607 | */ |
| 3608 | next_mz = NULL; |
| 3609 | if (!reclaimed) { |
| 3610 | do { |
| 3611 | /* |
| 3612 | * Loop until we find yet another one. |
| 3613 | * |
| 3614 | * By the time we get the soft_limit lock |
| 3615 | * again, someone might have aded the |
| 3616 | * group back on the RB tree. Iterate to |
| 3617 | * make sure we get a different mem. |
| 3618 | * mem_cgroup_largest_soft_limit_node returns |
| 3619 | * NULL if no other cgroup is present on |
| 3620 | * the tree |
| 3621 | */ |
| 3622 | next_mz = |
| 3623 | __mem_cgroup_largest_soft_limit_node(mctz); |
Michal Hocko | 39cc98f | 2011-05-26 16:25:28 -0700 | [diff] [blame] | 3624 | if (next_mz == mz) |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 3625 | css_put(&next_mz->memcg->css); |
Michal Hocko | 39cc98f | 2011-05-26 16:25:28 -0700 | [diff] [blame] | 3626 | else /* next_mz == NULL or other memcg */ |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3627 | break; |
| 3628 | } while (1); |
| 3629 | } |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 3630 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
| 3631 | excess = res_counter_soft_limit_excess(&mz->memcg->res); |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3632 | /* |
| 3633 | * One school of thought says that we should not add |
| 3634 | * back the node to the tree if reclaim returns 0. |
| 3635 | * But our reclaim could return 0, simply because due |
| 3636 | * to priority we are exposing a smaller subset of |
| 3637 | * memory to reclaim from. Consider this as a longer |
| 3638 | * term TODO. |
| 3639 | */ |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 3640 | /* If excess == 0, no tree ops */ |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 3641 | __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess); |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3642 | spin_unlock(&mctz->lock); |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 3643 | css_put(&mz->memcg->css); |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3644 | loop++; |
| 3645 | /* |
| 3646 | * Could not reclaim anything and there are no more |
| 3647 | * mem cgroups to try or we seem to be looping without |
| 3648 | * reclaiming anything. |
| 3649 | */ |
| 3650 | if (!nr_reclaimed && |
| 3651 | (next_mz == NULL || |
| 3652 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) |
| 3653 | break; |
| 3654 | } while (!nr_reclaimed); |
| 3655 | if (next_mz) |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 3656 | css_put(&next_mz->memcg->css); |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3657 | return nr_reclaimed; |
| 3658 | } |
| 3659 | |
KAMEZAWA Hiroyuki | c9b0ed5 | 2008-07-25 01:47:15 -0700 | [diff] [blame] | 3660 | /* |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3661 | * This routine traverse page_cgroup in given list and drop them all. |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3662 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
| 3663 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3664 | static int mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3665 | int node, int zid, enum lru_list lru) |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3666 | { |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3667 | struct mem_cgroup_per_zone *mz; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3668 | unsigned long flags, loop; |
KAMEZAWA Hiroyuki | 072c56c1 | 2008-02-07 00:14:39 -0800 | [diff] [blame] | 3669 | struct list_head *list; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 3670 | struct page *busy; |
| 3671 | struct zone *zone; |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3672 | int ret = 0; |
KAMEZAWA Hiroyuki | 072c56c1 | 2008-02-07 00:14:39 -0800 | [diff] [blame] | 3673 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3674 | zone = &NODE_DATA(node)->node_zones[zid]; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3675 | mz = mem_cgroup_zoneinfo(memcg, node, zid); |
Johannes Weiner | 6290df5 | 2012-01-12 17:18:10 -0800 | [diff] [blame] | 3676 | list = &mz->lruvec.lists[lru]; |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3677 | |
Hugh Dickins | 1eb4927 | 2012-03-21 16:34:19 -0700 | [diff] [blame] | 3678 | loop = mz->lru_size[lru]; |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3679 | /* give some margin against EBUSY etc...*/ |
| 3680 | loop += 256; |
| 3681 | busy = NULL; |
| 3682 | while (loop--) { |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 3683 | struct page_cgroup *pc; |
Johannes Weiner | 5564e88 | 2011-03-23 16:42:29 -0700 | [diff] [blame] | 3684 | struct page *page; |
| 3685 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3686 | ret = 0; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3687 | spin_lock_irqsave(&zone->lru_lock, flags); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3688 | if (list_empty(list)) { |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3689 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 3690 | break; |
| 3691 | } |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 3692 | page = list_entry(list->prev, struct page, lru); |
| 3693 | if (busy == page) { |
| 3694 | list_move(&page->lru, list); |
Thiago Farina | 648bcc7 | 2010-03-05 13:42:04 -0800 | [diff] [blame] | 3695 | busy = NULL; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3696 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3697 | continue; |
| 3698 | } |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3699 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3700 | |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 3701 | pc = lookup_page_cgroup(page); |
Johannes Weiner | 5564e88 | 2011-03-23 16:42:29 -0700 | [diff] [blame] | 3702 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3703 | ret = mem_cgroup_move_parent(page, pc, memcg, GFP_KERNEL); |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 3704 | if (ret == -ENOMEM || ret == -EINTR) |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3705 | break; |
| 3706 | |
| 3707 | if (ret == -EBUSY || ret == -EINVAL) { |
| 3708 | /* found lock contention or "pc" is obsolete. */ |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 3709 | busy = page; |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3710 | cond_resched(); |
| 3711 | } else |
| 3712 | busy = NULL; |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3713 | } |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3714 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3715 | if (!ret && !list_empty(list)) |
| 3716 | return -EBUSY; |
| 3717 | return ret; |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3718 | } |
| 3719 | |
| 3720 | /* |
| 3721 | * make mem_cgroup's charge to be 0 if there is no task. |
| 3722 | * This enables deleting this mem_cgroup. |
| 3723 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3724 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg, bool free_all) |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3725 | { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3726 | int ret; |
| 3727 | int node, zid, shrink; |
| 3728 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3729 | struct cgroup *cgrp = memcg->css.cgroup; |
Hugh Dickins | 8869b8f | 2008-03-04 14:29:09 -0800 | [diff] [blame] | 3730 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3731 | css_get(&memcg->css); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3732 | |
| 3733 | shrink = 0; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3734 | /* should free all ? */ |
| 3735 | if (free_all) |
| 3736 | goto try_to_free; |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3737 | move_account: |
Daisuke Nishimura | fce6647 | 2010-01-15 17:01:30 -0800 | [diff] [blame] | 3738 | do { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3739 | ret = -EBUSY; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3740 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3741 | goto out; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3742 | ret = -EINTR; |
| 3743 | if (signal_pending(current)) |
| 3744 | goto out; |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 3745 | /* This is for making all *used* pages to be on LRU. */ |
| 3746 | lru_add_drain_all(); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3747 | drain_all_stock_sync(memcg); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3748 | ret = 0; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3749 | mem_cgroup_start_move(memcg); |
KAMEZAWA Hiroyuki | 299b4ea | 2009-01-29 14:25:17 -0800 | [diff] [blame] | 3750 | for_each_node_state(node, N_HIGH_MEMORY) { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3751 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
Hugh Dickins | f156ab93 | 2012-03-21 16:34:19 -0700 | [diff] [blame] | 3752 | enum lru_list lru; |
| 3753 | for_each_lru(lru) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3754 | ret = mem_cgroup_force_empty_list(memcg, |
Hugh Dickins | f156ab93 | 2012-03-21 16:34:19 -0700 | [diff] [blame] | 3755 | node, zid, lru); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3756 | if (ret) |
| 3757 | break; |
| 3758 | } |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 3759 | } |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3760 | if (ret) |
| 3761 | break; |
| 3762 | } |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3763 | mem_cgroup_end_move(memcg); |
| 3764 | memcg_oom_recover(memcg); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3765 | /* it seems parent cgroup doesn't have enough mem */ |
| 3766 | if (ret == -ENOMEM) |
| 3767 | goto try_to_free; |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 3768 | cond_resched(); |
Daisuke Nishimura | fce6647 | 2010-01-15 17:01:30 -0800 | [diff] [blame] | 3769 | /* "ret" should also be checked to ensure all lists are empty. */ |
Glauber Costa | 569530f | 2012-04-12 12:49:13 -0700 | [diff] [blame] | 3770 | } while (res_counter_read_u64(&memcg->res, RES_USAGE) > 0 || ret); |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3771 | out: |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3772 | css_put(&memcg->css); |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3773 | return ret; |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3774 | |
| 3775 | try_to_free: |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3776 | /* returns EBUSY if there is a task or if we come here twice. */ |
| 3777 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3778 | ret = -EBUSY; |
| 3779 | goto out; |
| 3780 | } |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3781 | /* we call try-to-free pages for make this cgroup empty */ |
| 3782 | lru_add_drain_all(); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3783 | /* try to free all pages in this cgroup */ |
| 3784 | shrink = 1; |
Glauber Costa | 569530f | 2012-04-12 12:49:13 -0700 | [diff] [blame] | 3785 | while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3786 | int progress; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3787 | |
| 3788 | if (signal_pending(current)) { |
| 3789 | ret = -EINTR; |
| 3790 | goto out; |
| 3791 | } |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3792 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
Johannes Weiner | 185efc0 | 2011-09-14 16:21:58 -0700 | [diff] [blame] | 3793 | false); |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3794 | if (!progress) { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3795 | nr_retries--; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3796 | /* maybe some writeback is necessary */ |
Jens Axboe | 8aa7e84 | 2009-07-09 14:52:32 +0200 | [diff] [blame] | 3797 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3798 | } |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3799 | |
| 3800 | } |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3801 | lru_add_drain(); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3802 | /* try move_account...there may be some *locked* pages. */ |
Daisuke Nishimura | fce6647 | 2010-01-15 17:01:30 -0800 | [diff] [blame] | 3803 | goto move_account; |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3804 | } |
| 3805 | |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3806 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
| 3807 | { |
| 3808 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); |
| 3809 | } |
| 3810 | |
| 3811 | |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3812 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
| 3813 | { |
| 3814 | return mem_cgroup_from_cont(cont)->use_hierarchy; |
| 3815 | } |
| 3816 | |
| 3817 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, |
| 3818 | u64 val) |
| 3819 | { |
| 3820 | int retval = 0; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3821 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3822 | struct cgroup *parent = cont->parent; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3823 | struct mem_cgroup *parent_memcg = NULL; |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3824 | |
| 3825 | if (parent) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3826 | parent_memcg = mem_cgroup_from_cont(parent); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3827 | |
| 3828 | cgroup_lock(); |
| 3829 | /* |
André Goddard Rosa | af901ca | 2009-11-14 13:09:05 -0200 | [diff] [blame] | 3830 | * If parent's use_hierarchy is set, we can't make any modifications |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3831 | * in the child subtrees. If it is unset, then the change can |
| 3832 | * occur, provided the current cgroup has no children. |
| 3833 | * |
| 3834 | * For the root cgroup, parent_mem is NULL, we allow value to be |
| 3835 | * set if there are no children. |
| 3836 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3837 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3838 | (val == 1 || val == 0)) { |
| 3839 | if (list_empty(&cont->children)) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3840 | memcg->use_hierarchy = val; |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3841 | else |
| 3842 | retval = -EBUSY; |
| 3843 | } else |
| 3844 | retval = -EINVAL; |
| 3845 | cgroup_unlock(); |
| 3846 | |
| 3847 | return retval; |
| 3848 | } |
| 3849 | |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 3850 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3851 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 3852 | enum mem_cgroup_stat_index idx) |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 3853 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3854 | struct mem_cgroup *iter; |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 3855 | long val = 0; |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 3856 | |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 3857 | /* Per-cpu values can be negative, use a signed accumulator */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3858 | for_each_mem_cgroup_tree(iter, memcg) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3859 | val += mem_cgroup_read_stat(iter, idx); |
| 3860 | |
| 3861 | if (val < 0) /* race ? */ |
| 3862 | val = 0; |
| 3863 | return val; |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 3864 | } |
| 3865 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3866 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3867 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3868 | u64 val; |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3869 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3870 | if (!mem_cgroup_is_root(memcg)) { |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3871 | if (!swap) |
Glauber Costa | 65c64ce | 2011-12-22 01:02:27 +0000 | [diff] [blame] | 3872 | return res_counter_read_u64(&memcg->res, RES_USAGE); |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3873 | else |
Glauber Costa | 65c64ce | 2011-12-22 01:02:27 +0000 | [diff] [blame] | 3874 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3875 | } |
| 3876 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3877 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); |
| 3878 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3879 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3880 | if (swap) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3881 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAPOUT); |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3882 | |
| 3883 | return val << PAGE_SHIFT; |
| 3884 | } |
| 3885 | |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3886 | static ssize_t mem_cgroup_read(struct cgroup *cont, struct cftype *cft, |
| 3887 | struct file *file, char __user *buf, |
| 3888 | size_t nbytes, loff_t *ppos) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3889 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3890 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3891 | char str[64]; |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3892 | u64 val; |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3893 | int type, name, len; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3894 | |
| 3895 | type = MEMFILE_TYPE(cft->private); |
| 3896 | name = MEMFILE_ATTR(cft->private); |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3897 | |
| 3898 | if (!do_swap_account && type == _MEMSWAP) |
| 3899 | return -EOPNOTSUPP; |
| 3900 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3901 | switch (type) { |
| 3902 | case _MEM: |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3903 | if (name == RES_USAGE) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3904 | val = mem_cgroup_usage(memcg, false); |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3905 | else |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3906 | val = res_counter_read_u64(&memcg->res, name); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3907 | break; |
| 3908 | case _MEMSWAP: |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3909 | if (name == RES_USAGE) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3910 | val = mem_cgroup_usage(memcg, true); |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3911 | else |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3912 | val = res_counter_read_u64(&memcg->memsw, name); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3913 | break; |
| 3914 | default: |
| 3915 | BUG(); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3916 | } |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3917 | |
| 3918 | len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val); |
| 3919 | return simple_read_from_buffer(buf, nbytes, ppos, str, len); |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3920 | } |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3921 | /* |
| 3922 | * The user of this function is... |
| 3923 | * RES_LIMIT. |
| 3924 | */ |
Paul Menage | 856c13a | 2008-07-25 01:47:04 -0700 | [diff] [blame] | 3925 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
| 3926 | const char *buffer) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3927 | { |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3928 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3929 | int type, name; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3930 | unsigned long long val; |
| 3931 | int ret; |
| 3932 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3933 | type = MEMFILE_TYPE(cft->private); |
| 3934 | name = MEMFILE_ATTR(cft->private); |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3935 | |
| 3936 | if (!do_swap_account && type == _MEMSWAP) |
| 3937 | return -EOPNOTSUPP; |
| 3938 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3939 | switch (name) { |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3940 | case RES_LIMIT: |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 3941 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
| 3942 | ret = -EINVAL; |
| 3943 | break; |
| 3944 | } |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3945 | /* This function does all necessary parse...reuse it */ |
| 3946 | ret = res_counter_memparse_write_strategy(buffer, &val); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3947 | if (ret) |
| 3948 | break; |
| 3949 | if (type == _MEM) |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3950 | ret = mem_cgroup_resize_limit(memcg, val); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3951 | else |
| 3952 | ret = mem_cgroup_resize_memsw_limit(memcg, val); |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3953 | break; |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 3954 | case RES_SOFT_LIMIT: |
| 3955 | ret = res_counter_memparse_write_strategy(buffer, &val); |
| 3956 | if (ret) |
| 3957 | break; |
| 3958 | /* |
| 3959 | * For memsw, soft limits are hard to implement in terms |
| 3960 | * of semantics, for now, we support soft limits for |
| 3961 | * control without swap |
| 3962 | */ |
| 3963 | if (type == _MEM) |
| 3964 | ret = res_counter_set_soft_limit(&memcg->res, val); |
| 3965 | else |
| 3966 | ret = -EINVAL; |
| 3967 | break; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3968 | default: |
| 3969 | ret = -EINVAL; /* should be BUG() ? */ |
| 3970 | break; |
| 3971 | } |
| 3972 | return ret; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3973 | } |
| 3974 | |
KAMEZAWA Hiroyuki | fee7b54 | 2009-01-07 18:08:26 -0800 | [diff] [blame] | 3975 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
| 3976 | unsigned long long *mem_limit, unsigned long long *memsw_limit) |
| 3977 | { |
| 3978 | struct cgroup *cgroup; |
| 3979 | unsigned long long min_limit, min_memsw_limit, tmp; |
| 3980 | |
| 3981 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
| 3982 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
| 3983 | cgroup = memcg->css.cgroup; |
| 3984 | if (!memcg->use_hierarchy) |
| 3985 | goto out; |
| 3986 | |
| 3987 | while (cgroup->parent) { |
| 3988 | cgroup = cgroup->parent; |
| 3989 | memcg = mem_cgroup_from_cont(cgroup); |
| 3990 | if (!memcg->use_hierarchy) |
| 3991 | break; |
| 3992 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); |
| 3993 | min_limit = min(min_limit, tmp); |
| 3994 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
| 3995 | min_memsw_limit = min(min_memsw_limit, tmp); |
| 3996 | } |
| 3997 | out: |
| 3998 | *mem_limit = min_limit; |
| 3999 | *memsw_limit = min_memsw_limit; |
KAMEZAWA Hiroyuki | fee7b54 | 2009-01-07 18:08:26 -0800 | [diff] [blame] | 4000 | } |
| 4001 | |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 4002 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 4003 | { |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4004 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4005 | int type, name; |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 4006 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4007 | type = MEMFILE_TYPE(event); |
| 4008 | name = MEMFILE_ATTR(event); |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4009 | |
| 4010 | if (!do_swap_account && type == _MEMSWAP) |
| 4011 | return -EOPNOTSUPP; |
| 4012 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4013 | switch (name) { |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 4014 | case RES_MAX_USAGE: |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4015 | if (type == _MEM) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4016 | res_counter_reset_max(&memcg->res); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4017 | else |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4018 | res_counter_reset_max(&memcg->memsw); |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 4019 | break; |
| 4020 | case RES_FAILCNT: |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4021 | if (type == _MEM) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4022 | res_counter_reset_failcnt(&memcg->res); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4023 | else |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4024 | res_counter_reset_failcnt(&memcg->memsw); |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 4025 | break; |
| 4026 | } |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 4027 | |
Pavel Emelyanov | 85cc59d | 2008-04-29 01:00:20 -0700 | [diff] [blame] | 4028 | return 0; |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 4029 | } |
| 4030 | |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4031 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
| 4032 | struct cftype *cft) |
| 4033 | { |
| 4034 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; |
| 4035 | } |
| 4036 | |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4037 | #ifdef CONFIG_MMU |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4038 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
| 4039 | struct cftype *cft, u64 val) |
| 4040 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4041 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4042 | |
| 4043 | if (val >= (1 << NR_MOVE_TYPE)) |
| 4044 | return -EINVAL; |
| 4045 | /* |
| 4046 | * We check this value several times in both in can_attach() and |
| 4047 | * attach(), so we need cgroup lock to prevent this value from being |
| 4048 | * inconsistent. |
| 4049 | */ |
| 4050 | cgroup_lock(); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4051 | memcg->move_charge_at_immigrate = val; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4052 | cgroup_unlock(); |
| 4053 | |
| 4054 | return 0; |
| 4055 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4056 | #else |
| 4057 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
| 4058 | struct cftype *cft, u64 val) |
| 4059 | { |
| 4060 | return -ENOSYS; |
| 4061 | } |
| 4062 | #endif |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4063 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4064 | |
| 4065 | /* For read statistics */ |
| 4066 | enum { |
| 4067 | MCS_CACHE, |
| 4068 | MCS_RSS, |
KAMEZAWA Hiroyuki | d804658 | 2009-12-15 16:47:09 -0800 | [diff] [blame] | 4069 | MCS_FILE_MAPPED, |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4070 | MCS_PGPGIN, |
| 4071 | MCS_PGPGOUT, |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 4072 | MCS_SWAP, |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 4073 | MCS_PGFAULT, |
| 4074 | MCS_PGMAJFAULT, |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4075 | MCS_INACTIVE_ANON, |
| 4076 | MCS_ACTIVE_ANON, |
| 4077 | MCS_INACTIVE_FILE, |
| 4078 | MCS_ACTIVE_FILE, |
| 4079 | MCS_UNEVICTABLE, |
| 4080 | NR_MCS_STAT, |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 4081 | }; |
| 4082 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4083 | struct mcs_total_stat { |
| 4084 | s64 stat[NR_MCS_STAT]; |
| 4085 | }; |
| 4086 | |
| 4087 | struct { |
| 4088 | char *local_name; |
| 4089 | char *total_name; |
| 4090 | } memcg_stat_strings[NR_MCS_STAT] = { |
| 4091 | {"cache", "total_cache"}, |
| 4092 | {"rss", "total_rss"}, |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 4093 | {"mapped_file", "total_mapped_file"}, |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4094 | {"pgpgin", "total_pgpgin"}, |
| 4095 | {"pgpgout", "total_pgpgout"}, |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 4096 | {"swap", "total_swap"}, |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 4097 | {"pgfault", "total_pgfault"}, |
| 4098 | {"pgmajfault", "total_pgmajfault"}, |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4099 | {"inactive_anon", "total_inactive_anon"}, |
| 4100 | {"active_anon", "total_active_anon"}, |
| 4101 | {"inactive_file", "total_inactive_file"}, |
| 4102 | {"active_file", "total_active_file"}, |
| 4103 | {"unevictable", "total_unevictable"} |
| 4104 | }; |
| 4105 | |
| 4106 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 4107 | static void |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4108 | mem_cgroup_get_local_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s) |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4109 | { |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4110 | s64 val; |
| 4111 | |
| 4112 | /* per cpu stat */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4113 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_CACHE); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4114 | s->stat[MCS_CACHE] += val * PAGE_SIZE; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4115 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_RSS); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4116 | s->stat[MCS_RSS] += val * PAGE_SIZE; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4117 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED); |
KAMEZAWA Hiroyuki | d804658 | 2009-12-15 16:47:09 -0800 | [diff] [blame] | 4118 | s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4119 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGIN); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4120 | s->stat[MCS_PGPGIN] += val; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4121 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGOUT); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4122 | s->stat[MCS_PGPGOUT] += val; |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 4123 | if (do_swap_account) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4124 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_SWAPOUT); |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 4125 | s->stat[MCS_SWAP] += val * PAGE_SIZE; |
| 4126 | } |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4127 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGFAULT); |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 4128 | s->stat[MCS_PGFAULT] += val; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4129 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGMAJFAULT); |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 4130 | s->stat[MCS_PGMAJFAULT] += val; |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4131 | |
| 4132 | /* per zone stat */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4133 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_ANON)); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4134 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4135 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_ANON)); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4136 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4137 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_FILE)); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4138 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4139 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_FILE)); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4140 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4141 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE)); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4142 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4143 | } |
| 4144 | |
| 4145 | static void |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4146 | mem_cgroup_get_total_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s) |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4147 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 4148 | struct mem_cgroup *iter; |
| 4149 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4150 | for_each_mem_cgroup_tree(iter, memcg) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 4151 | mem_cgroup_get_local_stat(iter, s); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4152 | } |
| 4153 | |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 4154 | #ifdef CONFIG_NUMA |
| 4155 | static int mem_control_numa_stat_show(struct seq_file *m, void *arg) |
| 4156 | { |
| 4157 | int nid; |
| 4158 | unsigned long total_nr, file_nr, anon_nr, unevictable_nr; |
| 4159 | unsigned long node_nr; |
| 4160 | struct cgroup *cont = m->private; |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4161 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 4162 | |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4163 | total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 4164 | seq_printf(m, "total=%lu", total_nr); |
| 4165 | for_each_node_state(nid, N_HIGH_MEMORY) { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4166 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 4167 | seq_printf(m, " N%d=%lu", nid, node_nr); |
| 4168 | } |
| 4169 | seq_putc(m, '\n'); |
| 4170 | |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4171 | file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 4172 | seq_printf(m, "file=%lu", file_nr); |
| 4173 | for_each_node_state(nid, N_HIGH_MEMORY) { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4174 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 4175 | LRU_ALL_FILE); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 4176 | seq_printf(m, " N%d=%lu", nid, node_nr); |
| 4177 | } |
| 4178 | seq_putc(m, '\n'); |
| 4179 | |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4180 | anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 4181 | seq_printf(m, "anon=%lu", anon_nr); |
| 4182 | for_each_node_state(nid, N_HIGH_MEMORY) { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4183 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 4184 | LRU_ALL_ANON); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 4185 | seq_printf(m, " N%d=%lu", nid, node_nr); |
| 4186 | } |
| 4187 | seq_putc(m, '\n'); |
| 4188 | |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4189 | unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE)); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 4190 | seq_printf(m, "unevictable=%lu", unevictable_nr); |
| 4191 | for_each_node_state(nid, N_HIGH_MEMORY) { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4192 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 4193 | BIT(LRU_UNEVICTABLE)); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 4194 | seq_printf(m, " N%d=%lu", nid, node_nr); |
| 4195 | } |
| 4196 | seq_putc(m, '\n'); |
| 4197 | return 0; |
| 4198 | } |
| 4199 | #endif /* CONFIG_NUMA */ |
| 4200 | |
Paul Menage | c64745c | 2008-04-29 01:00:02 -0700 | [diff] [blame] | 4201 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
| 4202 | struct cgroup_map_cb *cb) |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 4203 | { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4204 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4205 | struct mcs_total_stat mystat; |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 4206 | int i; |
| 4207 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4208 | memset(&mystat, 0, sizeof(mystat)); |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4209 | mem_cgroup_get_local_stat(memcg, &mystat); |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 4210 | |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 4211 | |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 4212 | for (i = 0; i < NR_MCS_STAT; i++) { |
| 4213 | if (i == MCS_SWAP && !do_swap_account) |
| 4214 | continue; |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4215 | cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 4216 | } |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4217 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4218 | /* Hierarchical information */ |
KAMEZAWA Hiroyuki | fee7b54 | 2009-01-07 18:08:26 -0800 | [diff] [blame] | 4219 | { |
| 4220 | unsigned long long limit, memsw_limit; |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4221 | memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit); |
KAMEZAWA Hiroyuki | fee7b54 | 2009-01-07 18:08:26 -0800 | [diff] [blame] | 4222 | cb->fill(cb, "hierarchical_memory_limit", limit); |
| 4223 | if (do_swap_account) |
| 4224 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); |
| 4225 | } |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 4226 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4227 | memset(&mystat, 0, sizeof(mystat)); |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4228 | mem_cgroup_get_total_stat(memcg, &mystat); |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 4229 | for (i = 0; i < NR_MCS_STAT; i++) { |
| 4230 | if (i == MCS_SWAP && !do_swap_account) |
| 4231 | continue; |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4232 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 4233 | } |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4234 | |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 4235 | #ifdef CONFIG_DEBUG_VM |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 4236 | { |
| 4237 | int nid, zid; |
| 4238 | struct mem_cgroup_per_zone *mz; |
| 4239 | unsigned long recent_rotated[2] = {0, 0}; |
| 4240 | unsigned long recent_scanned[2] = {0, 0}; |
| 4241 | |
| 4242 | for_each_online_node(nid) |
| 4243 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4244 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 4245 | |
| 4246 | recent_rotated[0] += |
| 4247 | mz->reclaim_stat.recent_rotated[0]; |
| 4248 | recent_rotated[1] += |
| 4249 | mz->reclaim_stat.recent_rotated[1]; |
| 4250 | recent_scanned[0] += |
| 4251 | mz->reclaim_stat.recent_scanned[0]; |
| 4252 | recent_scanned[1] += |
| 4253 | mz->reclaim_stat.recent_scanned[1]; |
| 4254 | } |
| 4255 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); |
| 4256 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); |
| 4257 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); |
| 4258 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); |
| 4259 | } |
| 4260 | #endif |
| 4261 | |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 4262 | return 0; |
| 4263 | } |
| 4264 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4265 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
| 4266 | { |
| 4267 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
| 4268 | |
KAMEZAWA Hiroyuki | 1f4c025 | 2011-07-26 16:08:21 -0700 | [diff] [blame] | 4269 | return mem_cgroup_swappiness(memcg); |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4270 | } |
| 4271 | |
| 4272 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, |
| 4273 | u64 val) |
| 4274 | { |
| 4275 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
| 4276 | struct mem_cgroup *parent; |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 4277 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4278 | if (val > 100) |
| 4279 | return -EINVAL; |
| 4280 | |
| 4281 | if (cgrp->parent == NULL) |
| 4282 | return -EINVAL; |
| 4283 | |
| 4284 | parent = mem_cgroup_from_cont(cgrp->parent); |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 4285 | |
| 4286 | cgroup_lock(); |
| 4287 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4288 | /* If under hierarchy, only empty-root can set this value */ |
| 4289 | if ((parent->use_hierarchy) || |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 4290 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
| 4291 | cgroup_unlock(); |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4292 | return -EINVAL; |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 4293 | } |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4294 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4295 | memcg->swappiness = val; |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4296 | |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 4297 | cgroup_unlock(); |
| 4298 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4299 | return 0; |
| 4300 | } |
| 4301 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4302 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
| 4303 | { |
| 4304 | struct mem_cgroup_threshold_ary *t; |
| 4305 | u64 usage; |
| 4306 | int i; |
| 4307 | |
| 4308 | rcu_read_lock(); |
| 4309 | if (!swap) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4310 | t = rcu_dereference(memcg->thresholds.primary); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4311 | else |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4312 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4313 | |
| 4314 | if (!t) |
| 4315 | goto unlock; |
| 4316 | |
| 4317 | usage = mem_cgroup_usage(memcg, swap); |
| 4318 | |
| 4319 | /* |
| 4320 | * current_threshold points to threshold just below usage. |
| 4321 | * If it's not true, a threshold was crossed after last |
| 4322 | * call of __mem_cgroup_threshold(). |
| 4323 | */ |
Phil Carmody | 5407a56 | 2010-05-26 14:42:42 -0700 | [diff] [blame] | 4324 | i = t->current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4325 | |
| 4326 | /* |
| 4327 | * Iterate backward over array of thresholds starting from |
| 4328 | * current_threshold and check if a threshold is crossed. |
| 4329 | * If none of thresholds below usage is crossed, we read |
| 4330 | * only one element of the array here. |
| 4331 | */ |
| 4332 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) |
| 4333 | eventfd_signal(t->entries[i].eventfd, 1); |
| 4334 | |
| 4335 | /* i = current_threshold + 1 */ |
| 4336 | i++; |
| 4337 | |
| 4338 | /* |
| 4339 | * Iterate forward over array of thresholds starting from |
| 4340 | * current_threshold+1 and check if a threshold is crossed. |
| 4341 | * If none of thresholds above usage is crossed, we read |
| 4342 | * only one element of the array here. |
| 4343 | */ |
| 4344 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) |
| 4345 | eventfd_signal(t->entries[i].eventfd, 1); |
| 4346 | |
| 4347 | /* Update current_threshold */ |
Phil Carmody | 5407a56 | 2010-05-26 14:42:42 -0700 | [diff] [blame] | 4348 | t->current_threshold = i - 1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4349 | unlock: |
| 4350 | rcu_read_unlock(); |
| 4351 | } |
| 4352 | |
| 4353 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) |
| 4354 | { |
Kirill A. Shutemov | ad4ca5f | 2010-10-07 12:59:27 -0700 | [diff] [blame] | 4355 | while (memcg) { |
| 4356 | __mem_cgroup_threshold(memcg, false); |
| 4357 | if (do_swap_account) |
| 4358 | __mem_cgroup_threshold(memcg, true); |
| 4359 | |
| 4360 | memcg = parent_mem_cgroup(memcg); |
| 4361 | } |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4362 | } |
| 4363 | |
| 4364 | static int compare_thresholds(const void *a, const void *b) |
| 4365 | { |
| 4366 | const struct mem_cgroup_threshold *_a = a; |
| 4367 | const struct mem_cgroup_threshold *_b = b; |
| 4368 | |
| 4369 | return _a->threshold - _b->threshold; |
| 4370 | } |
| 4371 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4372 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4373 | { |
| 4374 | struct mem_cgroup_eventfd_list *ev; |
| 4375 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4376 | list_for_each_entry(ev, &memcg->oom_notify, list) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4377 | eventfd_signal(ev->eventfd, 1); |
| 4378 | return 0; |
| 4379 | } |
| 4380 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4381 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4382 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 4383 | struct mem_cgroup *iter; |
| 4384 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4385 | for_each_mem_cgroup_tree(iter, memcg) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 4386 | mem_cgroup_oom_notify_cb(iter); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4387 | } |
| 4388 | |
| 4389 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, |
| 4390 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4391 | { |
| 4392 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4393 | struct mem_cgroup_thresholds *thresholds; |
| 4394 | struct mem_cgroup_threshold_ary *new; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4395 | int type = MEMFILE_TYPE(cft->private); |
| 4396 | u64 threshold, usage; |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4397 | int i, size, ret; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4398 | |
| 4399 | ret = res_counter_memparse_write_strategy(args, &threshold); |
| 4400 | if (ret) |
| 4401 | return ret; |
| 4402 | |
| 4403 | mutex_lock(&memcg->thresholds_lock); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4404 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4405 | if (type == _MEM) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4406 | thresholds = &memcg->thresholds; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4407 | else if (type == _MEMSWAP) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4408 | thresholds = &memcg->memsw_thresholds; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4409 | else |
| 4410 | BUG(); |
| 4411 | |
| 4412 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
| 4413 | |
| 4414 | /* Check if a threshold crossed before adding a new one */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4415 | if (thresholds->primary) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4416 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
| 4417 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4418 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4419 | |
| 4420 | /* Allocate memory for new array of thresholds */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4421 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4422 | GFP_KERNEL); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4423 | if (!new) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4424 | ret = -ENOMEM; |
| 4425 | goto unlock; |
| 4426 | } |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4427 | new->size = size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4428 | |
| 4429 | /* Copy thresholds (if any) to new array */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4430 | if (thresholds->primary) { |
| 4431 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4432 | sizeof(struct mem_cgroup_threshold)); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4433 | } |
| 4434 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4435 | /* Add new threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4436 | new->entries[size - 1].eventfd = eventfd; |
| 4437 | new->entries[size - 1].threshold = threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4438 | |
| 4439 | /* Sort thresholds. Registering of new threshold isn't time-critical */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4440 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4441 | compare_thresholds, NULL); |
| 4442 | |
| 4443 | /* Find current threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4444 | new->current_threshold = -1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4445 | for (i = 0; i < size; i++) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4446 | if (new->entries[i].threshold < usage) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4447 | /* |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4448 | * new->current_threshold will not be used until |
| 4449 | * rcu_assign_pointer(), so it's safe to increment |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4450 | * it here. |
| 4451 | */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4452 | ++new->current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4453 | } |
| 4454 | } |
| 4455 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4456 | /* Free old spare buffer and save old primary buffer as spare */ |
| 4457 | kfree(thresholds->spare); |
| 4458 | thresholds->spare = thresholds->primary; |
| 4459 | |
| 4460 | rcu_assign_pointer(thresholds->primary, new); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4461 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4462 | /* To be sure that nobody uses thresholds */ |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4463 | synchronize_rcu(); |
| 4464 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4465 | unlock: |
| 4466 | mutex_unlock(&memcg->thresholds_lock); |
| 4467 | |
| 4468 | return ret; |
| 4469 | } |
| 4470 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4471 | static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4472 | struct cftype *cft, struct eventfd_ctx *eventfd) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4473 | { |
| 4474 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4475 | struct mem_cgroup_thresholds *thresholds; |
| 4476 | struct mem_cgroup_threshold_ary *new; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4477 | int type = MEMFILE_TYPE(cft->private); |
| 4478 | u64 usage; |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4479 | int i, j, size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4480 | |
| 4481 | mutex_lock(&memcg->thresholds_lock); |
| 4482 | if (type == _MEM) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4483 | thresholds = &memcg->thresholds; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4484 | else if (type == _MEMSWAP) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4485 | thresholds = &memcg->memsw_thresholds; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4486 | else |
| 4487 | BUG(); |
| 4488 | |
Anton Vorontsov | 371528c | 2012-02-24 05:14:46 +0400 | [diff] [blame] | 4489 | if (!thresholds->primary) |
| 4490 | goto unlock; |
| 4491 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4492 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
| 4493 | |
| 4494 | /* Check if a threshold crossed before removing */ |
| 4495 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
| 4496 | |
| 4497 | /* Calculate new number of threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4498 | size = 0; |
| 4499 | for (i = 0; i < thresholds->primary->size; i++) { |
| 4500 | if (thresholds->primary->entries[i].eventfd != eventfd) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4501 | size++; |
| 4502 | } |
| 4503 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4504 | new = thresholds->spare; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4505 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4506 | /* Set thresholds array to NULL if we don't have thresholds */ |
| 4507 | if (!size) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4508 | kfree(new); |
| 4509 | new = NULL; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4510 | goto swap_buffers; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4511 | } |
| 4512 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4513 | new->size = size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4514 | |
| 4515 | /* Copy thresholds and find current threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4516 | new->current_threshold = -1; |
| 4517 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { |
| 4518 | if (thresholds->primary->entries[i].eventfd == eventfd) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4519 | continue; |
| 4520 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4521 | new->entries[j] = thresholds->primary->entries[i]; |
| 4522 | if (new->entries[j].threshold < usage) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4523 | /* |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4524 | * new->current_threshold will not be used |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4525 | * until rcu_assign_pointer(), so it's safe to increment |
| 4526 | * it here. |
| 4527 | */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4528 | ++new->current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4529 | } |
| 4530 | j++; |
| 4531 | } |
| 4532 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4533 | swap_buffers: |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4534 | /* Swap primary and spare array */ |
| 4535 | thresholds->spare = thresholds->primary; |
Sha Zhengju | 8c75776 | 2012-05-10 13:01:45 -0700 | [diff] [blame] | 4536 | /* If all events are unregistered, free the spare array */ |
| 4537 | if (!new) { |
| 4538 | kfree(thresholds->spare); |
| 4539 | thresholds->spare = NULL; |
| 4540 | } |
| 4541 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4542 | rcu_assign_pointer(thresholds->primary, new); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4543 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4544 | /* To be sure that nobody uses thresholds */ |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4545 | synchronize_rcu(); |
Anton Vorontsov | 371528c | 2012-02-24 05:14:46 +0400 | [diff] [blame] | 4546 | unlock: |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4547 | mutex_unlock(&memcg->thresholds_lock); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4548 | } |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 4549 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4550 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
| 4551 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) |
| 4552 | { |
| 4553 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
| 4554 | struct mem_cgroup_eventfd_list *event; |
| 4555 | int type = MEMFILE_TYPE(cft->private); |
| 4556 | |
| 4557 | BUG_ON(type != _OOM_TYPE); |
| 4558 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
| 4559 | if (!event) |
| 4560 | return -ENOMEM; |
| 4561 | |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 4562 | spin_lock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4563 | |
| 4564 | event->eventfd = eventfd; |
| 4565 | list_add(&event->list, &memcg->oom_notify); |
| 4566 | |
| 4567 | /* already in OOM ? */ |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 4568 | if (atomic_read(&memcg->under_oom)) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4569 | eventfd_signal(eventfd, 1); |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 4570 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4571 | |
| 4572 | return 0; |
| 4573 | } |
| 4574 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4575 | static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4576 | struct cftype *cft, struct eventfd_ctx *eventfd) |
| 4577 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4578 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4579 | struct mem_cgroup_eventfd_list *ev, *tmp; |
| 4580 | int type = MEMFILE_TYPE(cft->private); |
| 4581 | |
| 4582 | BUG_ON(type != _OOM_TYPE); |
| 4583 | |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 4584 | spin_lock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4585 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4586 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4587 | if (ev->eventfd == eventfd) { |
| 4588 | list_del(&ev->list); |
| 4589 | kfree(ev); |
| 4590 | } |
| 4591 | } |
| 4592 | |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 4593 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4594 | } |
| 4595 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4596 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
| 4597 | struct cftype *cft, struct cgroup_map_cb *cb) |
| 4598 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4599 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4600 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4601 | cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4602 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4603 | if (atomic_read(&memcg->under_oom)) |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4604 | cb->fill(cb, "under_oom", 1); |
| 4605 | else |
| 4606 | cb->fill(cb, "under_oom", 0); |
| 4607 | return 0; |
| 4608 | } |
| 4609 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4610 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, |
| 4611 | struct cftype *cft, u64 val) |
| 4612 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4613 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4614 | struct mem_cgroup *parent; |
| 4615 | |
| 4616 | /* cannot set to root cgroup and only 0 and 1 are allowed */ |
| 4617 | if (!cgrp->parent || !((val == 0) || (val == 1))) |
| 4618 | return -EINVAL; |
| 4619 | |
| 4620 | parent = mem_cgroup_from_cont(cgrp->parent); |
| 4621 | |
| 4622 | cgroup_lock(); |
| 4623 | /* oom-kill-disable is a flag for subhierarchy. */ |
| 4624 | if ((parent->use_hierarchy) || |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4625 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4626 | cgroup_unlock(); |
| 4627 | return -EINVAL; |
| 4628 | } |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4629 | memcg->oom_kill_disable = val; |
KAMEZAWA Hiroyuki | 4d845eb | 2010-06-29 15:05:18 -0700 | [diff] [blame] | 4630 | if (!val) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4631 | memcg_oom_recover(memcg); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4632 | cgroup_unlock(); |
| 4633 | return 0; |
| 4634 | } |
| 4635 | |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 4636 | #ifdef CONFIG_NUMA |
| 4637 | static const struct file_operations mem_control_numa_stat_file_operations = { |
| 4638 | .read = seq_read, |
| 4639 | .llseek = seq_lseek, |
| 4640 | .release = single_release, |
| 4641 | }; |
| 4642 | |
| 4643 | static int mem_control_numa_stat_open(struct inode *unused, struct file *file) |
| 4644 | { |
| 4645 | struct cgroup *cont = file->f_dentry->d_parent->d_fsdata; |
| 4646 | |
| 4647 | file->f_op = &mem_control_numa_stat_file_operations; |
| 4648 | return single_open(file, mem_control_numa_stat_show, cont); |
| 4649 | } |
| 4650 | #endif /* CONFIG_NUMA */ |
| 4651 | |
Glauber Costa | e5671df | 2011-12-11 21:47:01 +0000 | [diff] [blame] | 4652 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM |
Glauber Costa | cbe128e3 | 2012-04-09 19:36:34 -0300 | [diff] [blame] | 4653 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
Glauber Costa | e5671df | 2011-12-11 21:47:01 +0000 | [diff] [blame] | 4654 | { |
Glauber Costa | 1d62e43 | 2012-04-09 19:36:33 -0300 | [diff] [blame] | 4655 | return mem_cgroup_sockets_init(memcg, ss); |
Glauber Costa | e5671df | 2011-12-11 21:47:01 +0000 | [diff] [blame] | 4656 | }; |
| 4657 | |
Glauber Costa | 1d62e43 | 2012-04-09 19:36:33 -0300 | [diff] [blame] | 4658 | static void kmem_cgroup_destroy(struct mem_cgroup *memcg) |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 4659 | { |
Glauber Costa | 1d62e43 | 2012-04-09 19:36:33 -0300 | [diff] [blame] | 4660 | mem_cgroup_sockets_destroy(memcg); |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 4661 | } |
Glauber Costa | e5671df | 2011-12-11 21:47:01 +0000 | [diff] [blame] | 4662 | #else |
Glauber Costa | cbe128e3 | 2012-04-09 19:36:34 -0300 | [diff] [blame] | 4663 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
Glauber Costa | e5671df | 2011-12-11 21:47:01 +0000 | [diff] [blame] | 4664 | { |
| 4665 | return 0; |
| 4666 | } |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 4667 | |
Glauber Costa | 1d62e43 | 2012-04-09 19:36:33 -0300 | [diff] [blame] | 4668 | static void kmem_cgroup_destroy(struct mem_cgroup *memcg) |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 4669 | { |
| 4670 | } |
Glauber Costa | e5671df | 2011-12-11 21:47:01 +0000 | [diff] [blame] | 4671 | #endif |
| 4672 | |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4673 | static struct cftype mem_cgroup_files[] = { |
| 4674 | { |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 4675 | .name = "usage_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4676 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4677 | .read = mem_cgroup_read, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4678 | .register_event = mem_cgroup_usage_register_event, |
| 4679 | .unregister_event = mem_cgroup_usage_unregister_event, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4680 | }, |
| 4681 | { |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 4682 | .name = "max_usage_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4683 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 4684 | .trigger = mem_cgroup_reset, |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4685 | .read = mem_cgroup_read, |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 4686 | }, |
| 4687 | { |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 4688 | .name = "limit_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4689 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
Paul Menage | 856c13a | 2008-07-25 01:47:04 -0700 | [diff] [blame] | 4690 | .write_string = mem_cgroup_write, |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4691 | .read = mem_cgroup_read, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4692 | }, |
| 4693 | { |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 4694 | .name = "soft_limit_in_bytes", |
| 4695 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), |
| 4696 | .write_string = mem_cgroup_write, |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4697 | .read = mem_cgroup_read, |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 4698 | }, |
| 4699 | { |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4700 | .name = "failcnt", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4701 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 4702 | .trigger = mem_cgroup_reset, |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4703 | .read = mem_cgroup_read, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4704 | }, |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 4705 | { |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 4706 | .name = "stat", |
Paul Menage | c64745c | 2008-04-29 01:00:02 -0700 | [diff] [blame] | 4707 | .read_map = mem_control_stat_show, |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 4708 | }, |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 4709 | { |
| 4710 | .name = "force_empty", |
| 4711 | .trigger = mem_cgroup_force_empty_write, |
| 4712 | }, |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4713 | { |
| 4714 | .name = "use_hierarchy", |
| 4715 | .write_u64 = mem_cgroup_hierarchy_write, |
| 4716 | .read_u64 = mem_cgroup_hierarchy_read, |
| 4717 | }, |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4718 | { |
| 4719 | .name = "swappiness", |
| 4720 | .read_u64 = mem_cgroup_swappiness_read, |
| 4721 | .write_u64 = mem_cgroup_swappiness_write, |
| 4722 | }, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4723 | { |
| 4724 | .name = "move_charge_at_immigrate", |
| 4725 | .read_u64 = mem_cgroup_move_charge_read, |
| 4726 | .write_u64 = mem_cgroup_move_charge_write, |
| 4727 | }, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4728 | { |
| 4729 | .name = "oom_control", |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4730 | .read_map = mem_cgroup_oom_control_read, |
| 4731 | .write_u64 = mem_cgroup_oom_control_write, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4732 | .register_event = mem_cgroup_oom_register_event, |
| 4733 | .unregister_event = mem_cgroup_oom_unregister_event, |
| 4734 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
| 4735 | }, |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 4736 | #ifdef CONFIG_NUMA |
| 4737 | { |
| 4738 | .name = "numa_stat", |
| 4739 | .open = mem_control_numa_stat_open, |
KAMEZAWA Hiroyuki | 8957712 | 2011-06-15 15:08:41 -0700 | [diff] [blame] | 4740 | .mode = S_IRUGO, |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 4741 | }, |
| 4742 | #endif |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4743 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4744 | { |
| 4745 | .name = "memsw.usage_in_bytes", |
| 4746 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4747 | .read = mem_cgroup_read, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4748 | .register_event = mem_cgroup_usage_register_event, |
| 4749 | .unregister_event = mem_cgroup_usage_unregister_event, |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4750 | }, |
| 4751 | { |
| 4752 | .name = "memsw.max_usage_in_bytes", |
| 4753 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), |
| 4754 | .trigger = mem_cgroup_reset, |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4755 | .read = mem_cgroup_read, |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4756 | }, |
| 4757 | { |
| 4758 | .name = "memsw.limit_in_bytes", |
| 4759 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), |
| 4760 | .write_string = mem_cgroup_write, |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4761 | .read = mem_cgroup_read, |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4762 | }, |
| 4763 | { |
| 4764 | .name = "memsw.failcnt", |
| 4765 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), |
| 4766 | .trigger = mem_cgroup_reset, |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4767 | .read = mem_cgroup_read, |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4768 | }, |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4769 | #endif |
Tejun Heo | 6bc1034 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4770 | { }, /* terminate */ |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4771 | }; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4772 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4773 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4774 | { |
| 4775 | struct mem_cgroup_per_node *pn; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4776 | struct mem_cgroup_per_zone *mz; |
Hugh Dickins | f156ab93 | 2012-03-21 16:34:19 -0700 | [diff] [blame] | 4777 | enum lru_list lru; |
KAMEZAWA Hiroyuki | 41e3355 | 2008-04-08 17:41:54 -0700 | [diff] [blame] | 4778 | int zone, tmp = node; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4779 | /* |
| 4780 | * This routine is called against possible nodes. |
| 4781 | * But it's BUG to call kmalloc() against offline node. |
| 4782 | * |
| 4783 | * TODO: this routine can waste much memory for nodes which will |
| 4784 | * never be onlined. It's better to use memory hotplug callback |
| 4785 | * function. |
| 4786 | */ |
KAMEZAWA Hiroyuki | 41e3355 | 2008-04-08 17:41:54 -0700 | [diff] [blame] | 4787 | if (!node_state(node, N_NORMAL_MEMORY)) |
| 4788 | tmp = -1; |
Jesper Juhl | 17295c8 | 2011-01-13 15:47:42 -0800 | [diff] [blame] | 4789 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4790 | if (!pn) |
| 4791 | return 1; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4792 | |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4793 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
| 4794 | mz = &pn->zoneinfo[zone]; |
Hugh Dickins | f156ab93 | 2012-03-21 16:34:19 -0700 | [diff] [blame] | 4795 | for_each_lru(lru) |
| 4796 | INIT_LIST_HEAD(&mz->lruvec.lists[lru]); |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 4797 | mz->usage_in_excess = 0; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 4798 | mz->on_tree = false; |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4799 | mz->memcg = memcg; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4800 | } |
Igor Mammedov | 0a619e5 | 2011-11-02 13:38:21 -0700 | [diff] [blame] | 4801 | memcg->info.nodeinfo[node] = pn; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4802 | return 0; |
| 4803 | } |
| 4804 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4805 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4806 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4807 | kfree(memcg->info.nodeinfo[node]); |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4808 | } |
| 4809 | |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4810 | static struct mem_cgroup *mem_cgroup_alloc(void) |
| 4811 | { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4812 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 4813 | int size = sizeof(struct mem_cgroup); |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4814 | |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 4815 | /* Can be very big if MAX_NUMNODES is very big */ |
Jan Blunck | c8dad2b | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 4816 | if (size < PAGE_SIZE) |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4817 | memcg = kzalloc(size, GFP_KERNEL); |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4818 | else |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4819 | memcg = vzalloc(size); |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4820 | |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4821 | if (!memcg) |
Dan Carpenter | e7bbcdf | 2010-03-23 13:35:12 -0700 | [diff] [blame] | 4822 | return NULL; |
| 4823 | |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4824 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
| 4825 | if (!memcg->stat) |
Dan Carpenter | d2e61b8 | 2010-11-11 14:05:12 -0800 | [diff] [blame] | 4826 | goto out_free; |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4827 | spin_lock_init(&memcg->pcp_counter_lock); |
| 4828 | return memcg; |
Dan Carpenter | d2e61b8 | 2010-11-11 14:05:12 -0800 | [diff] [blame] | 4829 | |
| 4830 | out_free: |
| 4831 | if (size < PAGE_SIZE) |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4832 | kfree(memcg); |
Dan Carpenter | d2e61b8 | 2010-11-11 14:05:12 -0800 | [diff] [blame] | 4833 | else |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4834 | vfree(memcg); |
Dan Carpenter | d2e61b8 | 2010-11-11 14:05:12 -0800 | [diff] [blame] | 4835 | return NULL; |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4836 | } |
| 4837 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4838 | /* |
Hugh Dickins | 59927fb | 2012-03-15 15:17:07 -0700 | [diff] [blame] | 4839 | * Helpers for freeing a vzalloc()ed mem_cgroup by RCU, |
| 4840 | * but in process context. The work_freeing structure is overlaid |
| 4841 | * on the rcu_freeing structure, which itself is overlaid on memsw. |
| 4842 | */ |
| 4843 | static void vfree_work(struct work_struct *work) |
| 4844 | { |
| 4845 | struct mem_cgroup *memcg; |
| 4846 | |
| 4847 | memcg = container_of(work, struct mem_cgroup, work_freeing); |
| 4848 | vfree(memcg); |
| 4849 | } |
| 4850 | static void vfree_rcu(struct rcu_head *rcu_head) |
| 4851 | { |
| 4852 | struct mem_cgroup *memcg; |
| 4853 | |
| 4854 | memcg = container_of(rcu_head, struct mem_cgroup, rcu_freeing); |
| 4855 | INIT_WORK(&memcg->work_freeing, vfree_work); |
| 4856 | schedule_work(&memcg->work_freeing); |
| 4857 | } |
| 4858 | |
| 4859 | /* |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4860 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
| 4861 | * (scanning all at force_empty is too costly...) |
| 4862 | * |
| 4863 | * Instead of clearing all references at force_empty, we remember |
| 4864 | * the number of reference from swap_cgroup and free mem_cgroup when |
| 4865 | * it goes down to 0. |
| 4866 | * |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4867 | * Removal of cgroup itself succeeds regardless of refs from swap. |
| 4868 | */ |
| 4869 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4870 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4871 | { |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 4872 | int node; |
| 4873 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4874 | mem_cgroup_remove_from_trees(memcg); |
| 4875 | free_css_id(&mem_cgroup_subsys, &memcg->css); |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 4876 | |
Bob Liu | 3ed28fa | 2012-01-12 17:19:04 -0800 | [diff] [blame] | 4877 | for_each_node(node) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4878 | free_mem_cgroup_per_zone_info(memcg, node); |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 4879 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4880 | free_percpu(memcg->stat); |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 4881 | if (sizeof(struct mem_cgroup) < PAGE_SIZE) |
Hugh Dickins | 59927fb | 2012-03-15 15:17:07 -0700 | [diff] [blame] | 4882 | kfree_rcu(memcg, rcu_freeing); |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4883 | else |
Hugh Dickins | 59927fb | 2012-03-15 15:17:07 -0700 | [diff] [blame] | 4884 | call_rcu(&memcg->rcu_freeing, vfree_rcu); |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4885 | } |
| 4886 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4887 | static void mem_cgroup_get(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4888 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4889 | atomic_inc(&memcg->refcnt); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4890 | } |
| 4891 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4892 | static void __mem_cgroup_put(struct mem_cgroup *memcg, int count) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4893 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4894 | if (atomic_sub_and_test(count, &memcg->refcnt)) { |
| 4895 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
| 4896 | __mem_cgroup_free(memcg); |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4897 | if (parent) |
| 4898 | mem_cgroup_put(parent); |
| 4899 | } |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4900 | } |
| 4901 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4902 | static void mem_cgroup_put(struct mem_cgroup *memcg) |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4903 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4904 | __mem_cgroup_put(memcg, 1); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4905 | } |
| 4906 | |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4907 | /* |
| 4908 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. |
| 4909 | */ |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 4910 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4911 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4912 | if (!memcg->res.parent) |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4913 | return NULL; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4914 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4915 | } |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 4916 | EXPORT_SYMBOL(parent_mem_cgroup); |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4917 | |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 4918 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
| 4919 | static void __init enable_swap_cgroup(void) |
| 4920 | { |
Hirokazu Takahashi | f8d66542 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 4921 | if (!mem_cgroup_disabled() && really_do_swap_account) |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 4922 | do_swap_account = 1; |
| 4923 | } |
| 4924 | #else |
| 4925 | static void __init enable_swap_cgroup(void) |
| 4926 | { |
| 4927 | } |
| 4928 | #endif |
| 4929 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 4930 | static int mem_cgroup_soft_limit_tree_init(void) |
| 4931 | { |
| 4932 | struct mem_cgroup_tree_per_node *rtpn; |
| 4933 | struct mem_cgroup_tree_per_zone *rtpz; |
| 4934 | int tmp, node, zone; |
| 4935 | |
Bob Liu | 3ed28fa | 2012-01-12 17:19:04 -0800 | [diff] [blame] | 4936 | for_each_node(node) { |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 4937 | tmp = node; |
| 4938 | if (!node_state(node, N_NORMAL_MEMORY)) |
| 4939 | tmp = -1; |
| 4940 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); |
| 4941 | if (!rtpn) |
Michal Hocko | c3cecc6 | 2012-01-12 17:18:50 -0800 | [diff] [blame] | 4942 | goto err_cleanup; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 4943 | |
| 4944 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
| 4945 | |
| 4946 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
| 4947 | rtpz = &rtpn->rb_tree_per_zone[zone]; |
| 4948 | rtpz->rb_root = RB_ROOT; |
| 4949 | spin_lock_init(&rtpz->lock); |
| 4950 | } |
| 4951 | } |
| 4952 | return 0; |
Michal Hocko | c3cecc6 | 2012-01-12 17:18:50 -0800 | [diff] [blame] | 4953 | |
| 4954 | err_cleanup: |
Bob Liu | 3ed28fa | 2012-01-12 17:19:04 -0800 | [diff] [blame] | 4955 | for_each_node(node) { |
Michal Hocko | c3cecc6 | 2012-01-12 17:18:50 -0800 | [diff] [blame] | 4956 | if (!soft_limit_tree.rb_tree_per_node[node]) |
| 4957 | break; |
| 4958 | kfree(soft_limit_tree.rb_tree_per_node[node]); |
| 4959 | soft_limit_tree.rb_tree_per_node[node] = NULL; |
| 4960 | } |
| 4961 | return 1; |
| 4962 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 4963 | } |
| 4964 | |
Li Zefan | 0eb253e | 2009-01-15 13:51:25 -0800 | [diff] [blame] | 4965 | static struct cgroup_subsys_state * __ref |
Li Zefan | 761b3ef5 | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 4966 | mem_cgroup_create(struct cgroup *cont) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4967 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4968 | struct mem_cgroup *memcg, *parent; |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 4969 | long error = -ENOMEM; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4970 | int node; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4971 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4972 | memcg = mem_cgroup_alloc(); |
| 4973 | if (!memcg) |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 4974 | return ERR_PTR(error); |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 4975 | |
Bob Liu | 3ed28fa | 2012-01-12 17:19:04 -0800 | [diff] [blame] | 4976 | for_each_node(node) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4977 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4978 | goto free_out; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 4979 | |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 4980 | /* root ? */ |
Balbir Singh | 28dbc4b | 2009-01-07 18:08:05 -0800 | [diff] [blame] | 4981 | if (cont->parent == NULL) { |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 4982 | int cpu; |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 4983 | enable_swap_cgroup(); |
Balbir Singh | 28dbc4b | 2009-01-07 18:08:05 -0800 | [diff] [blame] | 4984 | parent = NULL; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 4985 | if (mem_cgroup_soft_limit_tree_init()) |
| 4986 | goto free_out; |
Hillf Danton | a41c58a | 2011-12-19 17:11:57 -0800 | [diff] [blame] | 4987 | root_mem_cgroup = memcg; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 4988 | for_each_possible_cpu(cpu) { |
| 4989 | struct memcg_stock_pcp *stock = |
| 4990 | &per_cpu(memcg_stock, cpu); |
| 4991 | INIT_WORK(&stock->work, drain_local_stock); |
| 4992 | } |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 4993 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4994 | } else { |
Balbir Singh | 28dbc4b | 2009-01-07 18:08:05 -0800 | [diff] [blame] | 4995 | parent = mem_cgroup_from_cont(cont->parent); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4996 | memcg->use_hierarchy = parent->use_hierarchy; |
| 4997 | memcg->oom_kill_disable = parent->oom_kill_disable; |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4998 | } |
Balbir Singh | 28dbc4b | 2009-01-07 18:08:05 -0800 | [diff] [blame] | 4999 | |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 5000 | if (parent && parent->use_hierarchy) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5001 | res_counter_init(&memcg->res, &parent->res); |
| 5002 | res_counter_init(&memcg->memsw, &parent->memsw); |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 5003 | /* |
| 5004 | * We increment refcnt of the parent to ensure that we can |
| 5005 | * safely access it on res_counter_charge/uncharge. |
| 5006 | * This refcnt will be decremented when freeing this |
| 5007 | * mem_cgroup(see mem_cgroup_put). |
| 5008 | */ |
| 5009 | mem_cgroup_get(parent); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 5010 | } else { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5011 | res_counter_init(&memcg->res, NULL); |
| 5012 | res_counter_init(&memcg->memsw, NULL); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 5013 | } |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5014 | memcg->last_scanned_node = MAX_NUMNODES; |
| 5015 | INIT_LIST_HEAD(&memcg->oom_notify); |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 5016 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 5017 | if (parent) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5018 | memcg->swappiness = mem_cgroup_swappiness(parent); |
| 5019 | atomic_set(&memcg->refcnt, 1); |
| 5020 | memcg->move_charge_at_immigrate = 0; |
| 5021 | mutex_init(&memcg->thresholds_lock); |
KAMEZAWA Hiroyuki | 312734c0 | 2012-03-21 16:34:24 -0700 | [diff] [blame] | 5022 | spin_lock_init(&memcg->move_lock); |
Glauber Costa | cbe128e3 | 2012-04-09 19:36:34 -0300 | [diff] [blame] | 5023 | |
| 5024 | error = memcg_init_kmem(memcg, &mem_cgroup_subsys); |
| 5025 | if (error) { |
| 5026 | /* |
| 5027 | * We call put now because our (and parent's) refcnts |
| 5028 | * are already in place. mem_cgroup_put() will internally |
| 5029 | * call __mem_cgroup_free, so return directly |
| 5030 | */ |
| 5031 | mem_cgroup_put(memcg); |
| 5032 | return ERR_PTR(error); |
| 5033 | } |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5034 | return &memcg->css; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 5035 | free_out: |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5036 | __mem_cgroup_free(memcg); |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 5037 | return ERR_PTR(error); |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 5038 | } |
| 5039 | |
Li Zefan | 761b3ef5 | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 5040 | static int mem_cgroup_pre_destroy(struct cgroup *cont) |
KAMEZAWA Hiroyuki | df878fb | 2008-02-07 00:14:28 -0800 | [diff] [blame] | 5041 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5042 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 5043 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5044 | return mem_cgroup_force_empty(memcg, false); |
KAMEZAWA Hiroyuki | df878fb | 2008-02-07 00:14:28 -0800 | [diff] [blame] | 5045 | } |
| 5046 | |
Li Zefan | 761b3ef5 | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 5047 | static void mem_cgroup_destroy(struct cgroup *cont) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 5048 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5049 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
Daisuke Nishimura | c268e99 | 2009-01-15 13:51:13 -0800 | [diff] [blame] | 5050 | |
Glauber Costa | 1d62e43 | 2012-04-09 19:36:33 -0300 | [diff] [blame] | 5051 | kmem_cgroup_destroy(memcg); |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 5052 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5053 | mem_cgroup_put(memcg); |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 5054 | } |
| 5055 | |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5056 | #ifdef CONFIG_MMU |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5057 | /* Handlers for move charge at task migration. */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5058 | #define PRECHARGE_COUNT_AT_ONCE 256 |
| 5059 | static int mem_cgroup_do_precharge(unsigned long count) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5060 | { |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5061 | int ret = 0; |
| 5062 | int batch_count = PRECHARGE_COUNT_AT_ONCE; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5063 | struct mem_cgroup *memcg = mc.to; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5064 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5065 | if (mem_cgroup_is_root(memcg)) { |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5066 | mc.precharge += count; |
| 5067 | /* we don't need css_get for root */ |
| 5068 | return ret; |
| 5069 | } |
| 5070 | /* try to charge at once */ |
| 5071 | if (count > 1) { |
| 5072 | struct res_counter *dummy; |
| 5073 | /* |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5074 | * "memcg" cannot be under rmdir() because we've already checked |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5075 | * by cgroup_lock_live_cgroup() that it is not removed and we |
| 5076 | * are still under the same cgroup_mutex. So we can postpone |
| 5077 | * css_get(). |
| 5078 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5079 | if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy)) |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5080 | goto one_by_one; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5081 | if (do_swap_account && res_counter_charge(&memcg->memsw, |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5082 | PAGE_SIZE * count, &dummy)) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5083 | res_counter_uncharge(&memcg->res, PAGE_SIZE * count); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5084 | goto one_by_one; |
| 5085 | } |
| 5086 | mc.precharge += count; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5087 | return ret; |
| 5088 | } |
| 5089 | one_by_one: |
| 5090 | /* fall back to one by one charge */ |
| 5091 | while (count--) { |
| 5092 | if (signal_pending(current)) { |
| 5093 | ret = -EINTR; |
| 5094 | break; |
| 5095 | } |
| 5096 | if (!batch_count--) { |
| 5097 | batch_count = PRECHARGE_COUNT_AT_ONCE; |
| 5098 | cond_resched(); |
| 5099 | } |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5100 | ret = __mem_cgroup_try_charge(NULL, |
| 5101 | GFP_KERNEL, 1, &memcg, false); |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 5102 | if (ret) |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5103 | /* mem_cgroup_clear_mc() will do uncharge later */ |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 5104 | return ret; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5105 | mc.precharge++; |
| 5106 | } |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5107 | return ret; |
| 5108 | } |
| 5109 | |
| 5110 | /** |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 5111 | * get_mctgt_type - get target type of moving charge |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5112 | * @vma: the vma the pte to be checked belongs |
| 5113 | * @addr: the address corresponding to the pte to be checked |
| 5114 | * @ptent: the pte to be checked |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5115 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5116 | * |
| 5117 | * Returns |
| 5118 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. |
| 5119 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for |
| 5120 | * move charge. if @target is not NULL, the page is stored in target->page |
| 5121 | * with extra refcnt got(Callers should handle it). |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5122 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
| 5123 | * target for charge migration. if @target is not NULL, the entry is stored |
| 5124 | * in target->ent. |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5125 | * |
| 5126 | * Called with pte lock held. |
| 5127 | */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5128 | union mc_target { |
| 5129 | struct page *page; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5130 | swp_entry_t ent; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5131 | }; |
| 5132 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5133 | enum mc_target_type { |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 5134 | MC_TARGET_NONE = 0, |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5135 | MC_TARGET_PAGE, |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5136 | MC_TARGET_SWAP, |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5137 | }; |
| 5138 | |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 5139 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
| 5140 | unsigned long addr, pte_t ptent) |
| 5141 | { |
| 5142 | struct page *page = vm_normal_page(vma, addr, ptent); |
| 5143 | |
| 5144 | if (!page || !page_mapped(page)) |
| 5145 | return NULL; |
| 5146 | if (PageAnon(page)) { |
| 5147 | /* we don't move shared anon */ |
Hugh Dickins | be22aec | 2012-03-09 13:37:32 -0800 | [diff] [blame] | 5148 | if (!move_anon() || page_mapcount(page) > 2) |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 5149 | return NULL; |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 5150 | } else if (!move_file()) |
| 5151 | /* we ignore mapcount for file pages */ |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 5152 | return NULL; |
| 5153 | if (!get_page_unless_zero(page)) |
| 5154 | return NULL; |
| 5155 | |
| 5156 | return page; |
| 5157 | } |
| 5158 | |
| 5159 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
| 5160 | unsigned long addr, pte_t ptent, swp_entry_t *entry) |
| 5161 | { |
| 5162 | int usage_count; |
| 5163 | struct page *page = NULL; |
| 5164 | swp_entry_t ent = pte_to_swp_entry(ptent); |
| 5165 | |
| 5166 | if (!move_anon() || non_swap_entry(ent)) |
| 5167 | return NULL; |
| 5168 | usage_count = mem_cgroup_count_swap_user(ent, &page); |
| 5169 | if (usage_count > 1) { /* we don't move shared anon */ |
| 5170 | if (page) |
| 5171 | put_page(page); |
| 5172 | return NULL; |
| 5173 | } |
| 5174 | if (do_swap_account) |
| 5175 | entry->val = ent.val; |
| 5176 | |
| 5177 | return page; |
| 5178 | } |
| 5179 | |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 5180 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
| 5181 | unsigned long addr, pte_t ptent, swp_entry_t *entry) |
| 5182 | { |
| 5183 | struct page *page = NULL; |
| 5184 | struct inode *inode; |
| 5185 | struct address_space *mapping; |
| 5186 | pgoff_t pgoff; |
| 5187 | |
| 5188 | if (!vma->vm_file) /* anonymous vma */ |
| 5189 | return NULL; |
| 5190 | if (!move_file()) |
| 5191 | return NULL; |
| 5192 | |
| 5193 | inode = vma->vm_file->f_path.dentry->d_inode; |
| 5194 | mapping = vma->vm_file->f_mapping; |
| 5195 | if (pte_none(ptent)) |
| 5196 | pgoff = linear_page_index(vma, addr); |
| 5197 | else /* pte_file(ptent) is true */ |
| 5198 | pgoff = pte_to_pgoff(ptent); |
| 5199 | |
| 5200 | /* page is moved even if it's not RSS of this task(page-faulted). */ |
Hugh Dickins | aa3b189 | 2011-08-03 16:21:24 -0700 | [diff] [blame] | 5201 | page = find_get_page(mapping, pgoff); |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 5202 | |
Hugh Dickins | aa3b189 | 2011-08-03 16:21:24 -0700 | [diff] [blame] | 5203 | #ifdef CONFIG_SWAP |
| 5204 | /* shmem/tmpfs may report page out on swap: account for that too. */ |
| 5205 | if (radix_tree_exceptional_entry(page)) { |
| 5206 | swp_entry_t swap = radix_to_swp_entry(page); |
| 5207 | if (do_swap_account) |
| 5208 | *entry = swap; |
| 5209 | page = find_get_page(&swapper_space, swap.val); |
| 5210 | } |
| 5211 | #endif |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 5212 | return page; |
| 5213 | } |
| 5214 | |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 5215 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5216 | unsigned long addr, pte_t ptent, union mc_target *target) |
| 5217 | { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5218 | struct page *page = NULL; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5219 | struct page_cgroup *pc; |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 5220 | enum mc_target_type ret = MC_TARGET_NONE; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5221 | swp_entry_t ent = { .val = 0 }; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5222 | |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 5223 | if (pte_present(ptent)) |
| 5224 | page = mc_handle_present_pte(vma, addr, ptent); |
| 5225 | else if (is_swap_pte(ptent)) |
| 5226 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 5227 | else if (pte_none(ptent) || pte_file(ptent)) |
| 5228 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 5229 | |
| 5230 | if (!page && !ent.val) |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 5231 | return ret; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5232 | if (page) { |
| 5233 | pc = lookup_page_cgroup(page); |
| 5234 | /* |
| 5235 | * Do only loose check w/o page_cgroup lock. |
| 5236 | * mem_cgroup_move_account() checks the pc is valid or not under |
| 5237 | * the lock. |
| 5238 | */ |
| 5239 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { |
| 5240 | ret = MC_TARGET_PAGE; |
| 5241 | if (target) |
| 5242 | target->page = page; |
| 5243 | } |
| 5244 | if (!ret || !target) |
| 5245 | put_page(page); |
| 5246 | } |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 5247 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
| 5248 | if (ent.val && !ret && |
Bob Liu | 9fb4b7c | 2012-01-12 17:18:48 -0800 | [diff] [blame] | 5249 | css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) { |
KAMEZAWA Hiroyuki | 7f0f154 | 2010-05-11 14:06:58 -0700 | [diff] [blame] | 5250 | ret = MC_TARGET_SWAP; |
| 5251 | if (target) |
| 5252 | target->ent = ent; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5253 | } |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5254 | return ret; |
| 5255 | } |
| 5256 | |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5257 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 5258 | /* |
| 5259 | * We don't consider swapping or file mapped pages because THP does not |
| 5260 | * support them for now. |
| 5261 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
| 5262 | */ |
| 5263 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, |
| 5264 | unsigned long addr, pmd_t pmd, union mc_target *target) |
| 5265 | { |
| 5266 | struct page *page = NULL; |
| 5267 | struct page_cgroup *pc; |
| 5268 | enum mc_target_type ret = MC_TARGET_NONE; |
| 5269 | |
| 5270 | page = pmd_page(pmd); |
| 5271 | VM_BUG_ON(!page || !PageHead(page)); |
| 5272 | if (!move_anon()) |
| 5273 | return ret; |
| 5274 | pc = lookup_page_cgroup(page); |
| 5275 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { |
| 5276 | ret = MC_TARGET_PAGE; |
| 5277 | if (target) { |
| 5278 | get_page(page); |
| 5279 | target->page = page; |
| 5280 | } |
| 5281 | } |
| 5282 | return ret; |
| 5283 | } |
| 5284 | #else |
| 5285 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, |
| 5286 | unsigned long addr, pmd_t pmd, union mc_target *target) |
| 5287 | { |
| 5288 | return MC_TARGET_NONE; |
| 5289 | } |
| 5290 | #endif |
| 5291 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5292 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
| 5293 | unsigned long addr, unsigned long end, |
| 5294 | struct mm_walk *walk) |
| 5295 | { |
| 5296 | struct vm_area_struct *vma = walk->private; |
| 5297 | pte_t *pte; |
| 5298 | spinlock_t *ptl; |
| 5299 | |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5300 | if (pmd_trans_huge_lock(pmd, vma) == 1) { |
| 5301 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
| 5302 | mc.precharge += HPAGE_PMD_NR; |
| 5303 | spin_unlock(&vma->vm_mm->page_table_lock); |
Andrea Arcangeli | 1a5a990 | 2012-03-21 16:33:42 -0700 | [diff] [blame] | 5304 | return 0; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5305 | } |
Dave Hansen | 0331932 | 2011-03-22 16:32:56 -0700 | [diff] [blame] | 5306 | |
Andrea Arcangeli | 45f83ce | 2012-03-28 14:42:40 -0700 | [diff] [blame] | 5307 | if (pmd_trans_unstable(pmd)) |
| 5308 | return 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5309 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
| 5310 | for (; addr != end; pte++, addr += PAGE_SIZE) |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 5311 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5312 | mc.precharge++; /* increment precharge temporarily */ |
| 5313 | pte_unmap_unlock(pte - 1, ptl); |
| 5314 | cond_resched(); |
| 5315 | |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5316 | return 0; |
| 5317 | } |
| 5318 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5319 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
| 5320 | { |
| 5321 | unsigned long precharge; |
| 5322 | struct vm_area_struct *vma; |
| 5323 | |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5324 | down_read(&mm->mmap_sem); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5325 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
| 5326 | struct mm_walk mem_cgroup_count_precharge_walk = { |
| 5327 | .pmd_entry = mem_cgroup_count_precharge_pte_range, |
| 5328 | .mm = mm, |
| 5329 | .private = vma, |
| 5330 | }; |
| 5331 | if (is_vm_hugetlb_page(vma)) |
| 5332 | continue; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5333 | walk_page_range(vma->vm_start, vma->vm_end, |
| 5334 | &mem_cgroup_count_precharge_walk); |
| 5335 | } |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5336 | up_read(&mm->mmap_sem); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5337 | |
| 5338 | precharge = mc.precharge; |
| 5339 | mc.precharge = 0; |
| 5340 | |
| 5341 | return precharge; |
| 5342 | } |
| 5343 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5344 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
| 5345 | { |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5346 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
| 5347 | |
| 5348 | VM_BUG_ON(mc.moving_task); |
| 5349 | mc.moving_task = current; |
| 5350 | return mem_cgroup_do_precharge(precharge); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5351 | } |
| 5352 | |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5353 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
| 5354 | static void __mem_cgroup_clear_mc(void) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5355 | { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 5356 | struct mem_cgroup *from = mc.from; |
| 5357 | struct mem_cgroup *to = mc.to; |
| 5358 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5359 | /* we must uncharge all the leftover precharges from mc.to */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5360 | if (mc.precharge) { |
| 5361 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); |
| 5362 | mc.precharge = 0; |
| 5363 | } |
| 5364 | /* |
| 5365 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so |
| 5366 | * we must uncharge here. |
| 5367 | */ |
| 5368 | if (mc.moved_charge) { |
| 5369 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); |
| 5370 | mc.moved_charge = 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5371 | } |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 5372 | /* we must fixup refcnts and charges */ |
| 5373 | if (mc.moved_swap) { |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 5374 | /* uncharge swap account from the old cgroup */ |
| 5375 | if (!mem_cgroup_is_root(mc.from)) |
| 5376 | res_counter_uncharge(&mc.from->memsw, |
| 5377 | PAGE_SIZE * mc.moved_swap); |
| 5378 | __mem_cgroup_put(mc.from, mc.moved_swap); |
| 5379 | |
| 5380 | if (!mem_cgroup_is_root(mc.to)) { |
| 5381 | /* |
| 5382 | * we charged both to->res and to->memsw, so we should |
| 5383 | * uncharge to->res. |
| 5384 | */ |
| 5385 | res_counter_uncharge(&mc.to->res, |
| 5386 | PAGE_SIZE * mc.moved_swap); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 5387 | } |
| 5388 | /* we've already done mem_cgroup_get(mc.to) */ |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 5389 | mc.moved_swap = 0; |
| 5390 | } |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5391 | memcg_oom_recover(from); |
| 5392 | memcg_oom_recover(to); |
| 5393 | wake_up_all(&mc.waitq); |
| 5394 | } |
| 5395 | |
| 5396 | static void mem_cgroup_clear_mc(void) |
| 5397 | { |
| 5398 | struct mem_cgroup *from = mc.from; |
| 5399 | |
| 5400 | /* |
| 5401 | * we must clear moving_task before waking up waiters at the end of |
| 5402 | * task migration. |
| 5403 | */ |
| 5404 | mc.moving_task = NULL; |
| 5405 | __mem_cgroup_clear_mc(); |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 5406 | spin_lock(&mc.lock); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5407 | mc.from = NULL; |
| 5408 | mc.to = NULL; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 5409 | spin_unlock(&mc.lock); |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 5410 | mem_cgroup_end_move(from); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5411 | } |
| 5412 | |
Li Zefan | 761b3ef5 | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 5413 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
| 5414 | struct cgroup_taskset *tset) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5415 | { |
Tejun Heo | 2f7ee56 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 5416 | struct task_struct *p = cgroup_taskset_first(tset); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5417 | int ret = 0; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5418 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5419 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5420 | if (memcg->move_charge_at_immigrate) { |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5421 | struct mm_struct *mm; |
| 5422 | struct mem_cgroup *from = mem_cgroup_from_task(p); |
| 5423 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5424 | VM_BUG_ON(from == memcg); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5425 | |
| 5426 | mm = get_task_mm(p); |
| 5427 | if (!mm) |
| 5428 | return 0; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5429 | /* We move charges only when we move a owner of the mm */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5430 | if (mm->owner == p) { |
| 5431 | VM_BUG_ON(mc.from); |
| 5432 | VM_BUG_ON(mc.to); |
| 5433 | VM_BUG_ON(mc.precharge); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5434 | VM_BUG_ON(mc.moved_charge); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 5435 | VM_BUG_ON(mc.moved_swap); |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 5436 | mem_cgroup_start_move(from); |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 5437 | spin_lock(&mc.lock); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5438 | mc.from = from; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5439 | mc.to = memcg; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 5440 | spin_unlock(&mc.lock); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5441 | /* We set mc.moving_task later */ |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5442 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5443 | ret = mem_cgroup_precharge_mc(mm); |
| 5444 | if (ret) |
| 5445 | mem_cgroup_clear_mc(); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5446 | } |
| 5447 | mmput(mm); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5448 | } |
| 5449 | return ret; |
| 5450 | } |
| 5451 | |
Li Zefan | 761b3ef5 | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 5452 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
| 5453 | struct cgroup_taskset *tset) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5454 | { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5455 | mem_cgroup_clear_mc(); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5456 | } |
| 5457 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5458 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
| 5459 | unsigned long addr, unsigned long end, |
| 5460 | struct mm_walk *walk) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5461 | { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5462 | int ret = 0; |
| 5463 | struct vm_area_struct *vma = walk->private; |
| 5464 | pte_t *pte; |
| 5465 | spinlock_t *ptl; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5466 | enum mc_target_type target_type; |
| 5467 | union mc_target target; |
| 5468 | struct page *page; |
| 5469 | struct page_cgroup *pc; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5470 | |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5471 | /* |
| 5472 | * We don't take compound_lock() here but no race with splitting thp |
| 5473 | * happens because: |
| 5474 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not |
| 5475 | * under splitting, which means there's no concurrent thp split, |
| 5476 | * - if another thread runs into split_huge_page() just after we |
| 5477 | * entered this if-block, the thread must wait for page table lock |
| 5478 | * to be unlocked in __split_huge_page_splitting(), where the main |
| 5479 | * part of thp split is not executed yet. |
| 5480 | */ |
| 5481 | if (pmd_trans_huge_lock(pmd, vma) == 1) { |
Hugh Dickins | 62ade86 | 2012-05-18 11:28:34 -0700 | [diff] [blame] | 5482 | if (mc.precharge < HPAGE_PMD_NR) { |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5483 | spin_unlock(&vma->vm_mm->page_table_lock); |
| 5484 | return 0; |
| 5485 | } |
| 5486 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); |
| 5487 | if (target_type == MC_TARGET_PAGE) { |
| 5488 | page = target.page; |
| 5489 | if (!isolate_lru_page(page)) { |
| 5490 | pc = lookup_page_cgroup(page); |
| 5491 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, |
| 5492 | pc, mc.from, mc.to, |
| 5493 | false)) { |
| 5494 | mc.precharge -= HPAGE_PMD_NR; |
| 5495 | mc.moved_charge += HPAGE_PMD_NR; |
| 5496 | } |
| 5497 | putback_lru_page(page); |
| 5498 | } |
| 5499 | put_page(page); |
| 5500 | } |
| 5501 | spin_unlock(&vma->vm_mm->page_table_lock); |
Andrea Arcangeli | 1a5a990 | 2012-03-21 16:33:42 -0700 | [diff] [blame] | 5502 | return 0; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5503 | } |
| 5504 | |
Andrea Arcangeli | 45f83ce | 2012-03-28 14:42:40 -0700 | [diff] [blame] | 5505 | if (pmd_trans_unstable(pmd)) |
| 5506 | return 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5507 | retry: |
| 5508 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
| 5509 | for (; addr != end; addr += PAGE_SIZE) { |
| 5510 | pte_t ptent = *(pte++); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5511 | swp_entry_t ent; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5512 | |
| 5513 | if (!mc.precharge) |
| 5514 | break; |
| 5515 | |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 5516 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5517 | case MC_TARGET_PAGE: |
| 5518 | page = target.page; |
| 5519 | if (isolate_lru_page(page)) |
| 5520 | goto put; |
| 5521 | pc = lookup_page_cgroup(page); |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 5522 | if (!mem_cgroup_move_account(page, 1, pc, |
| 5523 | mc.from, mc.to, false)) { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5524 | mc.precharge--; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5525 | /* we uncharge from mc.from later. */ |
| 5526 | mc.moved_charge++; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5527 | } |
| 5528 | putback_lru_page(page); |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 5529 | put: /* get_mctgt_type() gets the page */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5530 | put_page(page); |
| 5531 | break; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5532 | case MC_TARGET_SWAP: |
| 5533 | ent = target.ent; |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 5534 | if (!mem_cgroup_move_swap_account(ent, |
| 5535 | mc.from, mc.to, false)) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5536 | mc.precharge--; |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 5537 | /* we fixup refcnts and charges later. */ |
| 5538 | mc.moved_swap++; |
| 5539 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5540 | break; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5541 | default: |
| 5542 | break; |
| 5543 | } |
| 5544 | } |
| 5545 | pte_unmap_unlock(pte - 1, ptl); |
| 5546 | cond_resched(); |
| 5547 | |
| 5548 | if (addr != end) { |
| 5549 | /* |
| 5550 | * We have consumed all precharges we got in can_attach(). |
| 5551 | * We try charge one by one, but don't do any additional |
| 5552 | * charges to mc.to if we have failed in charge once in attach() |
| 5553 | * phase. |
| 5554 | */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5555 | ret = mem_cgroup_do_precharge(1); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5556 | if (!ret) |
| 5557 | goto retry; |
| 5558 | } |
| 5559 | |
| 5560 | return ret; |
| 5561 | } |
| 5562 | |
| 5563 | static void mem_cgroup_move_charge(struct mm_struct *mm) |
| 5564 | { |
| 5565 | struct vm_area_struct *vma; |
| 5566 | |
| 5567 | lru_add_drain_all(); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5568 | retry: |
| 5569 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { |
| 5570 | /* |
| 5571 | * Someone who are holding the mmap_sem might be waiting in |
| 5572 | * waitq. So we cancel all extra charges, wake up all waiters, |
| 5573 | * and retry. Because we cancel precharges, we might not be able |
| 5574 | * to move enough charges, but moving charge is a best-effort |
| 5575 | * feature anyway, so it wouldn't be a big problem. |
| 5576 | */ |
| 5577 | __mem_cgroup_clear_mc(); |
| 5578 | cond_resched(); |
| 5579 | goto retry; |
| 5580 | } |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5581 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
| 5582 | int ret; |
| 5583 | struct mm_walk mem_cgroup_move_charge_walk = { |
| 5584 | .pmd_entry = mem_cgroup_move_charge_pte_range, |
| 5585 | .mm = mm, |
| 5586 | .private = vma, |
| 5587 | }; |
| 5588 | if (is_vm_hugetlb_page(vma)) |
| 5589 | continue; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5590 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
| 5591 | &mem_cgroup_move_charge_walk); |
| 5592 | if (ret) |
| 5593 | /* |
| 5594 | * means we have consumed all precharges and failed in |
| 5595 | * doing additional charge. Just abandon here. |
| 5596 | */ |
| 5597 | break; |
| 5598 | } |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5599 | up_read(&mm->mmap_sem); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5600 | } |
| 5601 | |
Li Zefan | 761b3ef5 | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 5602 | static void mem_cgroup_move_task(struct cgroup *cont, |
| 5603 | struct cgroup_taskset *tset) |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 5604 | { |
Tejun Heo | 2f7ee56 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 5605 | struct task_struct *p = cgroup_taskset_first(tset); |
KOSAKI Motohiro | a433658 | 2011-06-15 15:08:13 -0700 | [diff] [blame] | 5606 | struct mm_struct *mm = get_task_mm(p); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5607 | |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5608 | if (mm) { |
KOSAKI Motohiro | a433658 | 2011-06-15 15:08:13 -0700 | [diff] [blame] | 5609 | if (mc.to) |
| 5610 | mem_cgroup_move_charge(mm); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5611 | mmput(mm); |
| 5612 | } |
KOSAKI Motohiro | a433658 | 2011-06-15 15:08:13 -0700 | [diff] [blame] | 5613 | if (mc.to) |
| 5614 | mem_cgroup_clear_mc(); |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 5615 | } |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 5616 | #else /* !CONFIG_MMU */ |
Li Zefan | 761b3ef5 | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 5617 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
| 5618 | struct cgroup_taskset *tset) |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 5619 | { |
| 5620 | return 0; |
| 5621 | } |
Li Zefan | 761b3ef5 | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 5622 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
| 5623 | struct cgroup_taskset *tset) |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 5624 | { |
| 5625 | } |
Li Zefan | 761b3ef5 | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 5626 | static void mem_cgroup_move_task(struct cgroup *cont, |
| 5627 | struct cgroup_taskset *tset) |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 5628 | { |
| 5629 | } |
| 5630 | #endif |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 5631 | |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 5632 | struct cgroup_subsys mem_cgroup_subsys = { |
| 5633 | .name = "memory", |
| 5634 | .subsys_id = mem_cgroup_subsys_id, |
| 5635 | .create = mem_cgroup_create, |
KAMEZAWA Hiroyuki | df878fb | 2008-02-07 00:14:28 -0800 | [diff] [blame] | 5636 | .pre_destroy = mem_cgroup_pre_destroy, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 5637 | .destroy = mem_cgroup_destroy, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5638 | .can_attach = mem_cgroup_can_attach, |
| 5639 | .cancel_attach = mem_cgroup_cancel_attach, |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 5640 | .attach = mem_cgroup_move_task, |
Tejun Heo | 6bc1034 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 5641 | .base_cftypes = mem_cgroup_files, |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 5642 | .early_init = 0, |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 5643 | .use_id = 1, |
Tejun Heo | 48ddbe1 | 2012-04-01 12:09:56 -0700 | [diff] [blame] | 5644 | .__DEPRECATED_clear_css_refs = true, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 5645 | }; |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 5646 | |
| 5647 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
Michal Hocko | a42c390 | 2010-11-24 12:57:08 -0800 | [diff] [blame] | 5648 | static int __init enable_swap_account(char *s) |
| 5649 | { |
| 5650 | /* consider enabled if no parameter or 1 is given */ |
Michal Hocko | a2c8990 | 2011-05-24 17:12:50 -0700 | [diff] [blame] | 5651 | if (!strcmp(s, "1")) |
Michal Hocko | a42c390 | 2010-11-24 12:57:08 -0800 | [diff] [blame] | 5652 | really_do_swap_account = 1; |
Michal Hocko | a2c8990 | 2011-05-24 17:12:50 -0700 | [diff] [blame] | 5653 | else if (!strcmp(s, "0")) |
Michal Hocko | a42c390 | 2010-11-24 12:57:08 -0800 | [diff] [blame] | 5654 | really_do_swap_account = 0; |
| 5655 | return 1; |
| 5656 | } |
Michal Hocko | a2c8990 | 2011-05-24 17:12:50 -0700 | [diff] [blame] | 5657 | __setup("swapaccount=", enable_swap_account); |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 5658 | |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 5659 | #endif |